Google Git
Sign in
llvm / llvm-project / b580829c2b16de0b22716b3f9b9a9b2e79c55af5 / . / llvm / lib / Target / RISCV / RISCVISelLowering.cpp
blob: 21ef630e0f69209ffbbe4fdc1174636dbbea1386 [file] [log] [blame]
//===-- RISCVISelLowering.cpp - RISC-V DAG Lowering Implementation -------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file defines the interfaces that RISC-V uses to lower LLVM code into a
// selection DAG.
//
//===----------------------------------------------------------------------===//
#include "RISCVISelLowering.h"
#include "MCTargetDesc/RISCVMatInt.h"
#include "RISCV.h"
#include "RISCVConstantPoolValue.h"
#include "RISCVMachineFunctionInfo.h"
#include "RISCVRegisterInfo.h"
#include "RISCVSelectionDAGInfo.h"
#include "RISCVSubtarget.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/MemoryLocation.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Analysis/VectorUtils.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/SDPatternMatch.h"
#include "llvm/CodeGen/SelectionDAGAddressAnalysis.h"
#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/DiagnosticInfo.h"
#include "llvm/IR/DiagnosticPrinter.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicsRISCV.h"
#include "llvm/MC/MCCodeEmitter.h"
#include "llvm/MC/MCInstBuilder.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/InstructionCost.h"
#include "llvm/Support/KnownBits.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include <optional>
using namespace llvm;
#define DEBUG_TYPE "riscv-lower"
STATISTIC(NumTailCalls, "Number of tail calls");
static cl::opt<unsigned> ExtensionMaxWebSize(
DEBUG_TYPE "-ext-max-web-size", cl::Hidden,
cl::desc("Give the maximum size (in number of nodes) of the web of "
"instructions that we will consider for VW expansion"),
cl::init(18));
static cl::opt<bool>
AllowSplatInVW_W(DEBUG_TYPE "-form-vw-w-with-splat", cl::Hidden,
cl::desc("Allow the formation of VW_W operations (e.g., "
"VWADD_W) with splat constants"),
cl::init(false));
static cl::opt<unsigned> NumRepeatedDivisors(
DEBUG_TYPE "-fp-repeated-divisors", cl::Hidden,
cl::desc("Set the minimum number of repetitions of a divisor to allow "
"transformation to multiplications by the reciprocal"),
cl::init(2));
static cl::opt<int>
FPImmCost(DEBUG_TYPE "-fpimm-cost", cl::Hidden,
cl::desc("Give the maximum number of instructions that we will "
"use for creating a floating-point immediate value"),
cl::init(2));
static cl::opt<bool>
ReassocShlAddiAdd("reassoc-shl-addi-add", cl::Hidden,
cl::desc("Swap add and addi in cases where the add may "
"be combined with a shift"),
cl::init(true));
RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
const RISCVSubtarget &STI)
: TargetLowering(TM), Subtarget(STI) {
RISCVABI::ABI ABI = Subtarget.getTargetABI();
assert(ABI != RISCVABI::ABI_Unknown && "Improperly initialised target ABI");
if ((ABI == RISCVABI::ABI_ILP32F || ABI == RISCVABI::ABI_LP64F) &&
!Subtarget.hasStdExtF()) {
errs() << "Hard-float 'f' ABI can't be used for a target that "
"doesn't support the F instruction set extension (ignoring "
"target-abi)\n";
ABI = Subtarget.is64Bit() ? RISCVABI::ABI_LP64 : RISCVABI::ABI_ILP32;
} else if ((ABI == RISCVABI::ABI_ILP32D || ABI == RISCVABI::ABI_LP64D) &&
!Subtarget.hasStdExtD()) {
errs() << "Hard-float 'd' ABI can't be used for a target that "
"doesn't support the D instruction set extension (ignoring "
"target-abi)\n";
ABI = Subtarget.is64Bit() ? RISCVABI::ABI_LP64 : RISCVABI::ABI_ILP32;
}
switch (ABI) {
default:
reportFatalUsageError("Don't know how to lower this ABI");
case RISCVABI::ABI_ILP32:
case RISCVABI::ABI_ILP32E:
case RISCVABI::ABI_LP64E:
case RISCVABI::ABI_ILP32F:
case RISCVABI::ABI_ILP32D:
case RISCVABI::ABI_LP64:
case RISCVABI::ABI_LP64F:
case RISCVABI::ABI_LP64D:
break;
}
MVT XLenVT = Subtarget.getXLenVT();
// Set up the register classes.
addRegisterClass(XLenVT, &RISCV::GPRRegClass);
if (Subtarget.hasStdExtZfhmin())
addRegisterClass(MVT::f16, &RISCV::FPR16RegClass);
if (Subtarget.hasStdExtZfbfmin() || Subtarget.hasVendorXAndesBFHCvt())
addRegisterClass(MVT::bf16, &RISCV::FPR16RegClass);
if (Subtarget.hasStdExtF())
addRegisterClass(MVT::f32, &RISCV::FPR32RegClass);
if (Subtarget.hasStdExtD())
addRegisterClass(MVT::f64, &RISCV::FPR64RegClass);
if (Subtarget.hasStdExtZhinxmin())
addRegisterClass(MVT::f16, &RISCV::GPRF16RegClass);
if (Subtarget.hasStdExtZfinx())
addRegisterClass(MVT::f32, &RISCV::GPRF32RegClass);
if (Subtarget.hasStdExtZdinx()) {
if (Subtarget.is64Bit())
addRegisterClass(MVT::f64, &RISCV::GPRRegClass);
else
addRegisterClass(MVT::f64, &RISCV::GPRPairRegClass);
}
static const MVT::SimpleValueType BoolVecVTs[] = {
MVT::nxv1i1, MVT::nxv2i1, MVT::nxv4i1, MVT::nxv8i1,
MVT::nxv16i1, MVT::nxv32i1, MVT::nxv64i1};
static const MVT::SimpleValueType IntVecVTs[] = {
MVT::nxv1i8, MVT::nxv2i8, MVT::nxv4i8, MVT::nxv8i8, MVT::nxv16i8,
MVT::nxv32i8, MVT::nxv64i8, MVT::nxv1i16, MVT::nxv2i16, MVT::nxv4i16,
MVT::nxv8i16, MVT::nxv16i16, MVT::nxv32i16, MVT::nxv1i32, MVT::nxv2i32,
MVT::nxv4i32, MVT::nxv8i32, MVT::nxv16i32, MVT::nxv1i64, MVT::nxv2i64,
MVT::nxv4i64, MVT::nxv8i64};
static const MVT::SimpleValueType F16VecVTs[] = {
MVT::nxv1f16, MVT::nxv2f16, MVT::nxv4f16,
MVT::nxv8f16, MVT::nxv16f16, MVT::nxv32f16};
static const MVT::SimpleValueType BF16VecVTs[] = {
MVT::nxv1bf16, MVT::nxv2bf16, MVT::nxv4bf16,
MVT::nxv8bf16, MVT::nxv16bf16, MVT::nxv32bf16};
static const MVT::SimpleValueType F32VecVTs[] = {
MVT::nxv1f32, MVT::nxv2f32, MVT::nxv4f32, MVT::nxv8f32, MVT::nxv16f32};
static const MVT::SimpleValueType F64VecVTs[] = {
MVT::nxv1f64, MVT::nxv2f64, MVT::nxv4f64, MVT::nxv8f64};
static const MVT::SimpleValueType VecTupleVTs[] = {
MVT::riscv_nxv1i8x2, MVT::riscv_nxv1i8x3, MVT::riscv_nxv1i8x4,
MVT::riscv_nxv1i8x5, MVT::riscv_nxv1i8x6, MVT::riscv_nxv1i8x7,
MVT::riscv_nxv1i8x8, MVT::riscv_nxv2i8x2, MVT::riscv_nxv2i8x3,
MVT::riscv_nxv2i8x4, MVT::riscv_nxv2i8x5, MVT::riscv_nxv2i8x6,
MVT::riscv_nxv2i8x7, MVT::riscv_nxv2i8x8, MVT::riscv_nxv4i8x2,
MVT::riscv_nxv4i8x3, MVT::riscv_nxv4i8x4, MVT::riscv_nxv4i8x5,
MVT::riscv_nxv4i8x6, MVT::riscv_nxv4i8x7, MVT::riscv_nxv4i8x8,
MVT::riscv_nxv8i8x2, MVT::riscv_nxv8i8x3, MVT::riscv_nxv8i8x4,
MVT::riscv_nxv8i8x5, MVT::riscv_nxv8i8x6, MVT::riscv_nxv8i8x7,
MVT::riscv_nxv8i8x8, MVT::riscv_nxv16i8x2, MVT::riscv_nxv16i8x3,
MVT::riscv_nxv16i8x4, MVT::riscv_nxv32i8x2};
if (Subtarget.hasVInstructions()) {
auto addRegClassForRVV = [this](MVT VT) {
// Disable the smallest fractional LMUL types if ELEN is less than
// RVVBitsPerBlock.
unsigned MinElts = RISCV::RVVBitsPerBlock / Subtarget.getELen();
if (VT.getVectorMinNumElements() < MinElts)
return;
unsigned Size = VT.getSizeInBits().getKnownMinValue();
const TargetRegisterClass *RC;
if (Size <= RISCV::RVVBitsPerBlock)
RC = &RISCV::VRRegClass;
else if (Size == 2 * RISCV::RVVBitsPerBlock)
RC = &RISCV::VRM2RegClass;
else if (Size == 4 * RISCV::RVVBitsPerBlock)
RC = &RISCV::VRM4RegClass;
else if (Size == 8 * RISCV::RVVBitsPerBlock)
RC = &RISCV::VRM8RegClass;
else
llvm_unreachable("Unexpected size");
addRegisterClass(VT, RC);
};
for (MVT VT : BoolVecVTs)
addRegClassForRVV(VT);
for (MVT VT : IntVecVTs) {
if (VT.getVectorElementType() == MVT::i64 &&
!Subtarget.hasVInstructionsI64())
continue;
addRegClassForRVV(VT);
}
if (Subtarget.hasVInstructionsF16Minimal() ||
Subtarget.hasVendorXAndesVPackFPH())
for (MVT VT : F16VecVTs)
addRegClassForRVV(VT);
if (Subtarget.hasVInstructionsBF16Minimal() ||
Subtarget.hasVendorXAndesVBFHCvt())
for (MVT VT : BF16VecVTs)
addRegClassForRVV(VT);
if (Subtarget.hasVInstructionsF32())
for (MVT VT : F32VecVTs)
addRegClassForRVV(VT);
if (Subtarget.hasVInstructionsF64())
for (MVT VT : F64VecVTs)
addRegClassForRVV(VT);
if (Subtarget.useRVVForFixedLengthVectors()) {
auto addRegClassForFixedVectors = [this](MVT VT) {
MVT ContainerVT = getContainerForFixedLengthVector(VT);
unsigned RCID = getRegClassIDForVecVT(ContainerVT);
const RISCVRegisterInfo &TRI = *Subtarget.getRegisterInfo();
addRegisterClass(VT, TRI.getRegClass(RCID));
};
for (MVT VT : MVT::integer_fixedlen_vector_valuetypes())
if (useRVVForFixedLengthVectorVT(VT))
addRegClassForFixedVectors(VT);
for (MVT VT : MVT::fp_fixedlen_vector_valuetypes())
if (useRVVForFixedLengthVectorVT(VT))
addRegClassForFixedVectors(VT);
}
addRegisterClass(MVT::riscv_nxv1i8x2, &RISCV::VRN2M1RegClass);
addRegisterClass(MVT::riscv_nxv1i8x3, &RISCV::VRN3M1RegClass);
addRegisterClass(MVT::riscv_nxv1i8x4, &RISCV::VRN4M1RegClass);
addRegisterClass(MVT::riscv_nxv1i8x5, &RISCV::VRN5M1RegClass);
addRegisterClass(MVT::riscv_nxv1i8x6, &RISCV::VRN6M1RegClass);
addRegisterClass(MVT::riscv_nxv1i8x7, &RISCV::VRN7M1RegClass);
addRegisterClass(MVT::riscv_nxv1i8x8, &RISCV::VRN8M1RegClass);
addRegisterClass(MVT::riscv_nxv2i8x2, &RISCV::VRN2M1RegClass);
addRegisterClass(MVT::riscv_nxv2i8x3, &RISCV::VRN3M1RegClass);
addRegisterClass(MVT::riscv_nxv2i8x4, &RISCV::VRN4M1RegClass);
addRegisterClass(MVT::riscv_nxv2i8x5, &RISCV::VRN5M1RegClass);
addRegisterClass(MVT::riscv_nxv2i8x6, &RISCV::VRN6M1RegClass);
addRegisterClass(MVT::riscv_nxv2i8x7, &RISCV::VRN7M1RegClass);
addRegisterClass(MVT::riscv_nxv2i8x8, &RISCV::VRN8M1RegClass);
addRegisterClass(MVT::riscv_nxv4i8x2, &RISCV::VRN2M1RegClass);
addRegisterClass(MVT::riscv_nxv4i8x3, &RISCV::VRN3M1RegClass);
addRegisterClass(MVT::riscv_nxv4i8x4, &RISCV::VRN4M1RegClass);
addRegisterClass(MVT::riscv_nxv4i8x5, &RISCV::VRN5M1RegClass);
addRegisterClass(MVT::riscv_nxv4i8x6, &RISCV::VRN6M1RegClass);
addRegisterClass(MVT::riscv_nxv4i8x7, &RISCV::VRN7M1RegClass);
addRegisterClass(MVT::riscv_nxv4i8x8, &RISCV::VRN8M1RegClass);
addRegisterClass(MVT::riscv_nxv8i8x2, &RISCV::VRN2M1RegClass);
addRegisterClass(MVT::riscv_nxv8i8x3, &RISCV::VRN3M1RegClass);
addRegisterClass(MVT::riscv_nxv8i8x4, &RISCV::VRN4M1RegClass);
addRegisterClass(MVT::riscv_nxv8i8x5, &RISCV::VRN5M1RegClass);
addRegisterClass(MVT::riscv_nxv8i8x6, &RISCV::VRN6M1RegClass);
addRegisterClass(MVT::riscv_nxv8i8x7, &RISCV::VRN7M1RegClass);
addRegisterClass(MVT::riscv_nxv8i8x8, &RISCV::VRN8M1RegClass);
addRegisterClass(MVT::riscv_nxv16i8x2, &RISCV::VRN2M2RegClass);
addRegisterClass(MVT::riscv_nxv16i8x3, &RISCV::VRN3M2RegClass);
addRegisterClass(MVT::riscv_nxv16i8x4, &RISCV::VRN4M2RegClass);
addRegisterClass(MVT::riscv_nxv32i8x2, &RISCV::VRN2M4RegClass);
}
// Compute derived properties from the register classes.
computeRegisterProperties(STI.getRegisterInfo());
setStackPointerRegisterToSaveRestore(RISCV::X2);
setLoadExtAction({ISD::EXTLOAD, ISD::SEXTLOAD, ISD::ZEXTLOAD}, XLenVT,
MVT::i1, Promote);
// DAGCombiner can call isLoadExtLegal for types that aren't legal.
setLoadExtAction({ISD::EXTLOAD, ISD::SEXTLOAD, ISD::ZEXTLOAD}, MVT::i32,
MVT::i1, Promote);
// TODO: add all necessary setOperationAction calls.
setOperationAction(ISD::DYNAMIC_STACKALLOC, XLenVT, Custom);
setOperationAction(ISD::BR_JT, MVT::Other, Expand);
setOperationAction(ISD::BR_CC, XLenVT, Expand);
setOperationAction(ISD::BRCOND, MVT::Other, Custom);
setOperationAction(ISD::SELECT_CC, XLenVT, Expand);
setCondCodeAction(ISD::SETGT, XLenVT, Custom);
setCondCodeAction(ISD::SETGE, XLenVT, Expand);
setCondCodeAction(ISD::SETUGT, XLenVT, Custom);
setCondCodeAction(ISD::SETUGE, XLenVT, Expand);
if (!(Subtarget.hasVendorXCValu() && !Subtarget.is64Bit())) {
setCondCodeAction(ISD::SETULE, XLenVT, Expand);
setCondCodeAction(ISD::SETLE, XLenVT, Expand);
}
setOperationAction({ISD::STACKSAVE, ISD::STACKRESTORE}, MVT::Other, Expand);
setOperationAction(ISD::VASTART, MVT::Other, Custom);
setOperationAction({ISD::VAARG, ISD::VACOPY, ISD::VAEND}, MVT::Other, Expand);
if (!Subtarget.hasVendorXTHeadBb() && !Subtarget.hasVendorXqcibm() &&
!Subtarget.hasVendorXAndesPerf())
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
setOperationAction(ISD::EH_DWARF_CFA, MVT::i32, Custom);
if (!Subtarget.hasStdExtZbb() && !Subtarget.hasVendorXTHeadBb() &&
!Subtarget.hasVendorXqcibm() && !Subtarget.hasVendorXAndesPerf() &&
!(Subtarget.hasVendorXCValu() && !Subtarget.is64Bit()))
setOperationAction(ISD::SIGN_EXTEND_INREG, {MVT::i8, MVT::i16}, Expand);
if (Subtarget.hasStdExtZilsd() && !Subtarget.is64Bit()) {
setOperationAction(ISD::LOAD, MVT::i64, Custom);
setOperationAction(ISD::STORE, MVT::i64, Custom);
}
if (Subtarget.is64Bit()) {
setOperationAction(ISD::EH_DWARF_CFA, MVT::i64, Custom);
setOperationAction(ISD::LOAD, MVT::i32, Custom);
setOperationAction({ISD::ADD, ISD::SUB, ISD::SHL, ISD::SRA, ISD::SRL},
MVT::i32, Custom);
setOperationAction({ISD::UADDO, ISD::USUBO}, MVT::i32, Custom);
if (!Subtarget.hasStdExtZbb())
setOperationAction(
{ISD::SADDSAT, ISD::SSUBSAT, ISD::UADDSAT, ISD::USUBSAT}, MVT::i32,
Custom);
setOperationAction(ISD::SADDO, MVT::i32, Custom);
}
if (!Subtarget.hasStdExtZmmul()) {
setOperationAction({ISD::MUL, ISD::MULHS, ISD::MULHU}, XLenVT, Expand);
} else if (Subtarget.is64Bit()) {
setOperationAction(ISD::MUL, MVT::i128, Custom);
setOperationAction(ISD::MUL, MVT::i32, Custom);
} else {
setOperationAction(ISD::MUL, MVT::i64, Custom);
}
if (!Subtarget.hasStdExtM()) {
setOperationAction({ISD::SDIV, ISD::UDIV, ISD::SREM, ISD::UREM}, XLenVT,
Expand);
} else if (Subtarget.is64Bit()) {
setOperationAction({ISD::SDIV, ISD::UDIV, ISD::UREM},
{MVT::i8, MVT::i16, MVT::i32}, Custom);
}
setOperationAction(
{ISD::SDIVREM, ISD::UDIVREM, ISD::SMUL_LOHI, ISD::UMUL_LOHI}, XLenVT,
Expand);
setOperationAction({ISD::SHL_PARTS, ISD::SRL_PARTS, ISD::SRA_PARTS}, XLenVT,
Custom);
if (Subtarget.hasStdExtZbb() || Subtarget.hasStdExtZbkb()) {
if (Subtarget.is64Bit())
setOperationAction({ISD::ROTL, ISD::ROTR}, MVT::i32, Custom);
} else if (Subtarget.hasVendorXTHeadBb()) {
if (Subtarget.is64Bit())
setOperationAction({ISD::ROTL, ISD::ROTR}, MVT::i32, Custom);
setOperationAction({ISD::ROTL, ISD::ROTR}, XLenVT, Custom);
} else if (Subtarget.hasVendorXCVbitmanip() && !Subtarget.is64Bit()) {
setOperationAction(ISD::ROTL, XLenVT, Expand);
} else {
setOperationAction({ISD::ROTL, ISD::ROTR}, XLenVT, Expand);
}
// With Zbb we have an XLen rev8 instruction, but not GREVI. So we'll
// pattern match it directly in isel.
setOperationAction(ISD::BSWAP, XLenVT,
(Subtarget.hasStdExtZbb() || Subtarget.hasStdExtZbkb() ||
Subtarget.hasVendorXTHeadBb())
? Legal
: Expand);
if ((Subtarget.hasVendorXCVbitmanip() || Subtarget.hasVendorXqcibm()) &&
!Subtarget.is64Bit()) {
setOperationAction(ISD::BITREVERSE, XLenVT, Legal);
} else {
// Zbkb can use rev8+brev8 to implement bitreverse.
setOperationAction(ISD::BITREVERSE, XLenVT,
Subtarget.hasStdExtZbkb() ? Custom : Expand);
if (Subtarget.hasStdExtZbkb())
setOperationAction(ISD::BITREVERSE, MVT::i8, Custom);
}
if (Subtarget.hasStdExtZbb() ||
(Subtarget.hasVendorXCValu() && !Subtarget.is64Bit())) {
setOperationAction({ISD::SMIN, ISD::SMAX, ISD::UMIN, ISD::UMAX}, XLenVT,
Legal);
}
if (Subtarget.hasStdExtZbb() ||
(Subtarget.hasVendorXCVbitmanip() && !Subtarget.is64Bit())) {
if (Subtarget.is64Bit())
setOperationAction({ISD::CTTZ, ISD::CTTZ_ZERO_UNDEF}, MVT::i32, Custom);
} else {
setOperationAction(ISD::CTTZ, XLenVT, Expand);
// TODO: These should be set to LibCall, but this currently breaks
// the Linux kernel build. See #101786. Lacks i128 tests, too.
if (Subtarget.is64Bit())
setOperationAction(ISD::CTPOP, MVT::i128, Expand);
else
setOperationAction(ISD::CTPOP, MVT::i32, Expand);
setOperationAction(ISD::CTPOP, MVT::i64, Expand);
}
if (Subtarget.hasStdExtZbb() || Subtarget.hasVendorXTHeadBb() ||
(Subtarget.hasVendorXCVbitmanip() && !Subtarget.is64Bit())) {
// We need the custom lowering to make sure that the resulting sequence
// for the 32bit case is efficient on 64bit targets.
if (Subtarget.is64Bit())
setOperationAction({ISD::CTLZ, ISD::CTLZ_ZERO_UNDEF}, MVT::i32, Custom);
} else {
setOperationAction(ISD::CTLZ, XLenVT, Expand);
}
if (Subtarget.hasVendorXCValu() && !Subtarget.is64Bit()) {
setOperationAction(ISD::ABS, XLenVT, Legal);
} else if (Subtarget.hasShortForwardBranchOpt()) {
// We can use PseudoCCSUB to implement ABS.
setOperationAction(ISD::ABS, XLenVT, Legal);
} else if (Subtarget.is64Bit()) {
setOperationAction(ISD::ABS, MVT::i32, Custom);
}
if (!Subtarget.useCCMovInsn() && !Subtarget.hasVendorXTHeadCondMov() &&
!Subtarget.hasVendorXqcicm() && !Subtarget.hasVendorXqcics())
setOperationAction(ISD::SELECT, XLenVT, Custom);
if (Subtarget.hasVendorXqcia() && !Subtarget.is64Bit()) {
setOperationAction(ISD::UADDSAT, MVT::i32, Legal);
setOperationAction(ISD::SADDSAT, MVT::i32, Legal);
setOperationAction(ISD::USUBSAT, MVT::i32, Legal);
setOperationAction(ISD::SSUBSAT, MVT::i32, Legal);
setOperationAction(ISD::SSHLSAT, MVT::i32, Legal);
setOperationAction(ISD::USHLSAT, MVT::i32, Legal);
}
static const unsigned FPLegalNodeTypes[] = {
ISD::FMINNUM, ISD::FMAXNUM, ISD::FMINIMUMNUM,
ISD::FMAXIMUMNUM, ISD::LRINT, ISD::LLRINT,
ISD::LROUND, ISD::LLROUND, ISD::STRICT_LRINT,
ISD::STRICT_LLRINT, ISD::STRICT_LROUND, ISD::STRICT_LLROUND,
ISD::STRICT_FMA, ISD::STRICT_FADD, ISD::STRICT_FSUB,
ISD::STRICT_FMUL, ISD::STRICT_FDIV, ISD::STRICT_FSQRT,
ISD::STRICT_FSETCC, ISD::STRICT_FSETCCS, ISD::FCANONICALIZE};
static const ISD::CondCode FPCCToExpand[] = {
ISD::SETOGT, ISD::SETOGE, ISD::SETONE, ISD::SETUEQ, ISD::SETUGT,
ISD::SETUGE, ISD::SETULT, ISD::SETULE, ISD::SETUNE, ISD::SETGT,
ISD::SETGE, ISD::SETNE, ISD::SETO, ISD::SETUO};
static const unsigned FPOpToExpand[] = {
ISD::FSIN, ISD::FCOS, ISD::FSINCOS, ISD::FPOW,
ISD::FREM};
static const unsigned FPRndMode[] = {
ISD::FCEIL, ISD::FFLOOR, ISD::FTRUNC, ISD::FRINT, ISD::FROUND,
ISD::FROUNDEVEN};
static const unsigned ZfhminZfbfminPromoteOps[] = {
ISD::FMINNUM, ISD::FMAXNUM, ISD::FMAXIMUMNUM,
ISD::FMINIMUMNUM, ISD::FADD, ISD::FSUB,
ISD::FMUL, ISD::FMA, ISD::FDIV,
ISD::FSQRT, ISD::STRICT_FMA, ISD::STRICT_FADD,
ISD::STRICT_FSUB, ISD::STRICT_FMUL, ISD::STRICT_FDIV,
ISD::STRICT_FSQRT, ISD::STRICT_FSETCC, ISD::STRICT_FSETCCS,
ISD::SETCC, ISD::FCEIL, ISD::FFLOOR,
ISD::FTRUNC, ISD::FRINT, ISD::FROUND,
ISD::FROUNDEVEN, ISD::FCANONICALIZE};
if (Subtarget.hasStdExtZfbfmin()) {
setOperationAction(ISD::BITCAST, MVT::i16, Custom);
setOperationAction(ISD::ConstantFP, MVT::bf16, Expand);
setOperationAction(ISD::SELECT_CC, MVT::bf16, Expand);
setOperationAction(ISD::SELECT, MVT::bf16, Custom);
setOperationAction(ISD::BR_CC, MVT::bf16, Expand);
setOperationAction(ZfhminZfbfminPromoteOps, MVT::bf16, Promote);
setOperationAction(ISD::FREM, MVT::bf16, Promote);
setOperationAction(ISD::FABS, MVT::bf16, Custom);
setOperationAction(ISD::FNEG, MVT::bf16, Custom);
setOperationAction(ISD::FCOPYSIGN, MVT::bf16, Custom);
setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT}, XLenVT, Custom);
setOperationAction({ISD::SINT_TO_FP, ISD::UINT_TO_FP}, XLenVT, Custom);
}
if (Subtarget.hasStdExtZfhminOrZhinxmin()) {
if (Subtarget.hasStdExtZfhOrZhinx()) {
setOperationAction(FPLegalNodeTypes, MVT::f16, Legal);
setOperationAction(FPRndMode, MVT::f16,
Subtarget.hasStdExtZfa() ? Legal : Custom);
setOperationAction(ISD::IS_FPCLASS, MVT::f16, Custom);
setOperationAction({ISD::FMAXIMUM, ISD::FMINIMUM}, MVT::f16,
Subtarget.hasStdExtZfa() ? Legal : Custom);
if (Subtarget.hasStdExtZfa())
setOperationAction(ISD::ConstantFP, MVT::f16, Custom);
} else {
setOperationAction(ZfhminZfbfminPromoteOps, MVT::f16, Promote);
setOperationAction({ISD::FMAXIMUM, ISD::FMINIMUM}, MVT::f16, Promote);
for (auto Op : {ISD::LROUND, ISD::LLROUND, ISD::LRINT, ISD::LLRINT,
ISD::STRICT_LROUND, ISD::STRICT_LLROUND,
ISD::STRICT_LRINT, ISD::STRICT_LLRINT})
setOperationAction(Op, MVT::f16, Custom);
setOperationAction(ISD::FABS, MVT::f16, Custom);
setOperationAction(ISD::FNEG, MVT::f16, Custom);
setOperationAction(ISD::FCOPYSIGN, MVT::f16, Custom);
setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT}, XLenVT, Custom);
setOperationAction({ISD::SINT_TO_FP, ISD::UINT_TO_FP}, XLenVT, Custom);
}
setOperationAction(ISD::BITCAST, MVT::i16, Custom);
setOperationAction(ISD::STRICT_FP_ROUND, MVT::f16, Legal);
setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f32, Legal);
setCondCodeAction(FPCCToExpand, MVT::f16, Expand);
setOperationAction(ISD::SELECT_CC, MVT::f16, Expand);
setOperationAction(ISD::SELECT, MVT::f16, Custom);
setOperationAction(ISD::BR_CC, MVT::f16, Expand);
setOperationAction(
ISD::FNEARBYINT, MVT::f16,
Subtarget.hasStdExtZfh() && Subtarget.hasStdExtZfa() ? Legal : Promote);
setOperationAction({ISD::FREM, ISD::FPOW, ISD::FPOWI,
ISD::FCOS, ISD::FSIN, ISD::FSINCOS, ISD::FEXP,
ISD::FEXP2, ISD::FEXP10, ISD::FLOG, ISD::FLOG2,
ISD::FLOG10, ISD::FLDEXP, ISD::FFREXP},
MVT::f16, Promote);
// FIXME: Need to promote f16 STRICT_* to f32 libcalls, but we don't have
// complete support for all operations in LegalizeDAG.
setOperationAction({ISD::STRICT_FCEIL, ISD::STRICT_FFLOOR,
ISD::STRICT_FNEARBYINT, ISD::STRICT_FRINT,
ISD::STRICT_FROUND, ISD::STRICT_FROUNDEVEN,
ISD::STRICT_FTRUNC, ISD::STRICT_FLDEXP},
MVT::f16, Promote);
// We need to custom promote this.
if (Subtarget.is64Bit())
setOperationAction(ISD::FPOWI, MVT::i32, Custom);
}
if (Subtarget.hasStdExtFOrZfinx()) {
setOperationAction(FPLegalNodeTypes, MVT::f32, Legal);
setOperationAction(FPRndMode, MVT::f32,
Subtarget.hasStdExtZfa() ? Legal : Custom);
setCondCodeAction(FPCCToExpand, MVT::f32, Expand);
setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
setOperationAction(ISD::SELECT, MVT::f32, Custom);
setOperationAction(ISD::BR_CC, MVT::f32, Expand);
setOperationAction(FPOpToExpand, MVT::f32, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
setTruncStoreAction(MVT::f32, MVT::f16, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::bf16, Expand);
setTruncStoreAction(MVT::f32, MVT::bf16, Expand);
setOperationAction(ISD::IS_FPCLASS, MVT::f32, Custom);
setOperationAction(ISD::BF16_TO_FP, MVT::f32, Custom);
setOperationAction(ISD::FP_TO_BF16, MVT::f32,
Subtarget.isSoftFPABI() ? LibCall : Custom);
setOperationAction(ISD::FP_TO_FP16, MVT::f32, Custom);
setOperationAction(ISD::FP16_TO_FP, MVT::f32, Custom);
setOperationAction(ISD::STRICT_FP_TO_FP16, MVT::f32, Custom);
setOperationAction(ISD::STRICT_FP16_TO_FP, MVT::f32, Custom);
if (Subtarget.hasStdExtZfa()) {
setOperationAction(ISD::ConstantFP, MVT::f32, Custom);
setOperationAction(ISD::FNEARBYINT, MVT::f32, Legal);
setOperationAction({ISD::FMAXIMUM, ISD::FMINIMUM}, MVT::f32, Legal);
} else {
setOperationAction({ISD::FMAXIMUM, ISD::FMINIMUM}, MVT::f32, Custom);
}
}
if (Subtarget.hasStdExtFOrZfinx() && Subtarget.is64Bit())
setOperationAction(ISD::BITCAST, MVT::i32, Custom);
if (Subtarget.hasStdExtDOrZdinx()) {
setOperationAction(FPLegalNodeTypes, MVT::f64, Legal);
if (!Subtarget.is64Bit())
setOperationAction(ISD::BITCAST, MVT::i64, Custom);
if (Subtarget.hasStdExtZdinx() && !Subtarget.hasStdExtZilsd() &&
!Subtarget.is64Bit()) {
setOperationAction(ISD::LOAD, MVT::f64, Custom);
setOperationAction(ISD::STORE, MVT::f64, Custom);
}
if (Subtarget.hasStdExtZfa()) {
setOperationAction(ISD::ConstantFP, MVT::f64, Custom);
setOperationAction(FPRndMode, MVT::f64, Legal);
setOperationAction(ISD::FNEARBYINT, MVT::f64, Legal);
setOperationAction({ISD::FMAXIMUM, ISD::FMINIMUM}, MVT::f64, Legal);
} else {
if (Subtarget.is64Bit())
setOperationAction(FPRndMode, MVT::f64, Custom);
setOperationAction({ISD::FMAXIMUM, ISD::FMINIMUM}, MVT::f64, Custom);
}
setOperationAction(ISD::STRICT_FP_ROUND, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f64, Legal);
setCondCodeAction(FPCCToExpand, MVT::f64, Expand);
setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
setOperationAction(ISD::SELECT, MVT::f64, Custom);
setOperationAction(ISD::BR_CC, MVT::f64, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
setOperationAction(FPOpToExpand, MVT::f64, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
setTruncStoreAction(MVT::f64, MVT::f16, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::bf16, Expand);
setTruncStoreAction(MVT::f64, MVT::bf16, Expand);
setOperationAction(ISD::IS_FPCLASS, MVT::f64, Custom);
setOperationAction(ISD::BF16_TO_FP, MVT::f64, Custom);
setOperationAction(ISD::FP_TO_BF16, MVT::f64,
Subtarget.isSoftFPABI() ? LibCall : Custom);
setOperationAction(ISD::FP_TO_FP16, MVT::f64, Custom);
setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
setOperationAction(ISD::STRICT_FP_TO_FP16, MVT::f64, Custom);
setOperationAction(ISD::STRICT_FP16_TO_FP, MVT::f64, Expand);
}
if (Subtarget.is64Bit()) {
setOperationAction({ISD::FP_TO_UINT, ISD::FP_TO_SINT,
ISD::STRICT_FP_TO_UINT, ISD::STRICT_FP_TO_SINT},
MVT::i32, Custom);
setOperationAction(ISD::LROUND, MVT::i32, Custom);
}
if (Subtarget.hasStdExtFOrZfinx()) {
setOperationAction({ISD::FP_TO_UINT_SAT, ISD::FP_TO_SINT_SAT}, XLenVT,
Custom);
// f16/bf16 require custom handling.
setOperationAction({ISD::STRICT_FP_TO_UINT, ISD::STRICT_FP_TO_SINT}, XLenVT,
Custom);
setOperationAction({ISD::STRICT_UINT_TO_FP, ISD::STRICT_SINT_TO_FP}, XLenVT,
Custom);
setOperationAction(ISD::GET_ROUNDING, XLenVT, Custom);
setOperationAction(ISD::SET_ROUNDING, MVT::Other, Custom);
setOperationAction(ISD::GET_FPENV, XLenVT, Custom);
setOperationAction(ISD::SET_FPENV, XLenVT, Custom);
setOperationAction(ISD::RESET_FPENV, MVT::Other, Custom);
setOperationAction(ISD::GET_FPMODE, XLenVT, Custom);
setOperationAction(ISD::SET_FPMODE, XLenVT, Custom);
setOperationAction(ISD::RESET_FPMODE, MVT::Other, Custom);
}
setOperationAction({ISD::GlobalAddress, ISD::BlockAddress, ISD::ConstantPool,
ISD::JumpTable},
XLenVT, Custom);
setOperationAction(ISD::GlobalTLSAddress, XLenVT, Custom);
if (Subtarget.is64Bit())
setOperationAction(ISD::Constant, MVT::i64, Custom);
// TODO: On M-mode only targets, the cycle[h]/time[h] CSR may not be present.
// Unfortunately this can't be determined just from the ISA naming string.
setOperationAction(ISD::READCYCLECOUNTER, MVT::i64,
Subtarget.is64Bit() ? Legal : Custom);
setOperationAction(ISD::READSTEADYCOUNTER, MVT::i64,
Subtarget.is64Bit() ? Legal : Custom);
if (Subtarget.is64Bit()) {
setOperationAction(ISD::INIT_TRAMPOLINE, MVT::Other, Custom);
setOperationAction(ISD::ADJUST_TRAMPOLINE, MVT::Other, Custom);
}
setOperationAction({ISD::TRAP, ISD::DEBUGTRAP}, MVT::Other, Legal);
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
if (Subtarget.is64Bit())
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::i32, Custom);
if (Subtarget.hasVendorXMIPSCBOP())
setOperationAction(ISD::PREFETCH, MVT::Other, Custom);
else if (Subtarget.hasStdExtZicbop())
setOperationAction(ISD::PREFETCH, MVT::Other, Legal);
if (Subtarget.hasStdExtA()) {
setMaxAtomicSizeInBitsSupported(Subtarget.getXLen());
if (Subtarget.hasStdExtZabha() && Subtarget.hasStdExtZacas())
setMinCmpXchgSizeInBits(8);
else
setMinCmpXchgSizeInBits(32);
} else if (Subtarget.hasForcedAtomics()) {
setMaxAtomicSizeInBitsSupported(Subtarget.getXLen());
} else {
setMaxAtomicSizeInBitsSupported(0);
}
setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom);
setBooleanContents(ZeroOrOneBooleanContent);
if (getTargetMachine().getTargetTriple().isOSLinux()) {
// Custom lowering of llvm.clear_cache.
setOperationAction(ISD::CLEAR_CACHE, MVT::Other, Custom);
}
if (Subtarget.hasVInstructions()) {
setBooleanVectorContents(ZeroOrOneBooleanContent);
setOperationAction(ISD::VSCALE, XLenVT, Custom);
// RVV intrinsics may have illegal operands.
// We also need to custom legalize vmv.x.s.
setOperationAction({ISD::INTRINSIC_WO_CHAIN, ISD::INTRINSIC_W_CHAIN,
ISD::INTRINSIC_VOID},
{MVT::i8, MVT::i16}, Custom);
if (Subtarget.is64Bit())
setOperationAction({ISD::INTRINSIC_W_CHAIN, ISD::INTRINSIC_VOID},
MVT::i32, Custom);
else
setOperationAction({ISD::INTRINSIC_WO_CHAIN, ISD::INTRINSIC_W_CHAIN},
MVT::i64, Custom);
setOperationAction({ISD::INTRINSIC_W_CHAIN, ISD::INTRINSIC_VOID},
MVT::Other, Custom);
static const unsigned IntegerVPOps[] = {
ISD::VP_ADD, ISD::VP_SUB, ISD::VP_MUL,
ISD::VP_SDIV, ISD::VP_UDIV, ISD::VP_SREM,
ISD::VP_UREM, ISD::VP_AND, ISD::VP_OR,
ISD::VP_XOR, ISD::VP_SRA, ISD::VP_SRL,
ISD::VP_SHL, ISD::VP_REDUCE_ADD, ISD::VP_REDUCE_AND,
ISD::VP_REDUCE_OR, ISD::VP_REDUCE_XOR, ISD::VP_REDUCE_SMAX,
ISD::VP_REDUCE_SMIN, ISD::VP_REDUCE_UMAX, ISD::VP_REDUCE_UMIN,
ISD::VP_MERGE, ISD::VP_SELECT, ISD::VP_FP_TO_SINT,
ISD::VP_FP_TO_UINT, ISD::VP_SETCC, ISD::VP_SIGN_EXTEND,
ISD::VP_ZERO_EXTEND, ISD::VP_TRUNCATE, ISD::VP_SMIN,
ISD::VP_SMAX, ISD::VP_UMIN, ISD::VP_UMAX,
ISD::VP_ABS, ISD::EXPERIMENTAL_VP_REVERSE, ISD::EXPERIMENTAL_VP_SPLICE,
ISD::VP_SADDSAT, ISD::VP_UADDSAT, ISD::VP_SSUBSAT,
ISD::VP_USUBSAT, ISD::VP_CTTZ_ELTS, ISD::VP_CTTZ_ELTS_ZERO_UNDEF,
ISD::EXPERIMENTAL_VP_SPLAT};
static const unsigned FloatingPointVPOps[] = {
ISD::VP_FADD, ISD::VP_FSUB, ISD::VP_FMUL,
ISD::VP_FDIV, ISD::VP_FNEG, ISD::VP_FABS,
ISD::VP_FMA, ISD::VP_REDUCE_FADD, ISD::VP_REDUCE_SEQ_FADD,
ISD::VP_REDUCE_FMIN, ISD::VP_REDUCE_FMAX, ISD::VP_MERGE,
ISD::VP_SELECT, ISD::VP_SINT_TO_FP, ISD::VP_UINT_TO_FP,
ISD::VP_SETCC, ISD::VP_FP_ROUND, ISD::VP_FP_EXTEND,
ISD::VP_SQRT, ISD::VP_FMINNUM, ISD::VP_FMAXNUM,
ISD::VP_FCEIL, ISD::VP_FFLOOR, ISD::VP_FROUND,
ISD::VP_FROUNDEVEN, ISD::VP_FCOPYSIGN, ISD::VP_FROUNDTOZERO,
ISD::VP_FRINT, ISD::VP_FNEARBYINT, ISD::VP_IS_FPCLASS,
ISD::VP_FMINIMUM, ISD::VP_FMAXIMUM, ISD::VP_LRINT,
ISD::VP_LLRINT, ISD::VP_REDUCE_FMINIMUM,
ISD::VP_REDUCE_FMAXIMUM, ISD::EXPERIMENTAL_VP_SPLAT};
static const unsigned IntegerVecReduceOps[] = {
ISD::VECREDUCE_ADD, ISD::VECREDUCE_AND, ISD::VECREDUCE_OR,
ISD::VECREDUCE_XOR, ISD::VECREDUCE_SMAX, ISD::VECREDUCE_SMIN,
ISD::VECREDUCE_UMAX, ISD::VECREDUCE_UMIN};
static const unsigned FloatingPointVecReduceOps[] = {
ISD::VECREDUCE_FADD, ISD::VECREDUCE_SEQ_FADD, ISD::VECREDUCE_FMIN,
ISD::VECREDUCE_FMAX, ISD::VECREDUCE_FMINIMUM, ISD::VECREDUCE_FMAXIMUM};
static const unsigned FloatingPointLibCallOps[] = {
ISD::FREM, ISD::FPOW, ISD::FCOS, ISD::FSIN, ISD::FSINCOS, ISD::FEXP,
ISD::FEXP2, ISD::FEXP10, ISD::FLOG, ISD::FLOG2, ISD::FLOG10};
if (!Subtarget.is64Bit()) {
// We must custom-lower certain vXi64 operations on RV32 due to the vector
// element type being illegal.
setOperationAction({ISD::INSERT_VECTOR_ELT, ISD::EXTRACT_VECTOR_ELT},
MVT::i64, Custom);
setOperationAction(IntegerVecReduceOps, MVT::i64, Custom);
setOperationAction({ISD::VP_REDUCE_ADD, ISD::VP_REDUCE_AND,
ISD::VP_REDUCE_OR, ISD::VP_REDUCE_XOR,
ISD::VP_REDUCE_SMAX, ISD::VP_REDUCE_SMIN,
ISD::VP_REDUCE_UMAX, ISD::VP_REDUCE_UMIN},
MVT::i64, Custom);
}
for (MVT VT : BoolVecVTs) {
if (!isTypeLegal(VT))
continue;
setOperationAction(ISD::SPLAT_VECTOR, VT, Custom);
// Mask VTs are custom-expanded into a series of standard nodes
setOperationAction({ISD::TRUNCATE, ISD::CONCAT_VECTORS,
ISD::INSERT_SUBVECTOR, ISD::EXTRACT_SUBVECTOR,
ISD::SCALAR_TO_VECTOR},
VT, Custom);
setOperationAction({ISD::INSERT_VECTOR_ELT, ISD::EXTRACT_VECTOR_ELT}, VT,
Custom);
setOperationAction(ISD::SELECT, VT, Custom);
setOperationAction({ISD::SELECT_CC, ISD::VSELECT, ISD::VP_SELECT}, VT,
Expand);
setOperationAction(ISD::VP_MERGE, VT, Custom);
setOperationAction({ISD::VP_CTTZ_ELTS, ISD::VP_CTTZ_ELTS_ZERO_UNDEF}, VT,
Custom);
setOperationAction({ISD::VP_AND, ISD::VP_OR, ISD::VP_XOR}, VT, Custom);
setOperationAction(
{ISD::VECREDUCE_AND, ISD::VECREDUCE_OR, ISD::VECREDUCE_XOR}, VT,
Custom);
setOperationAction(
{ISD::VP_REDUCE_AND, ISD::VP_REDUCE_OR, ISD::VP_REDUCE_XOR}, VT,
Custom);
// RVV has native int->float & float->int conversions where the
// element type sizes are within one power-of-two of each other. Any
// wider distances between type sizes have to be lowered as sequences
// which progressively narrow the gap in stages.
setOperationAction({ISD::SINT_TO_FP, ISD::UINT_TO_FP, ISD::FP_TO_SINT,
ISD::FP_TO_UINT, ISD::STRICT_SINT_TO_FP,
ISD::STRICT_UINT_TO_FP, ISD::STRICT_FP_TO_SINT,
ISD::STRICT_FP_TO_UINT},
VT, Custom);
setOperationAction({ISD::FP_TO_SINT_SAT, ISD::FP_TO_UINT_SAT}, VT,
Custom);
// Expand all extending loads to types larger than this, and truncating
// stores from types larger than this.
for (MVT OtherVT : MVT::integer_scalable_vector_valuetypes()) {
setTruncStoreAction(VT, OtherVT, Expand);
setLoadExtAction({ISD::EXTLOAD, ISD::SEXTLOAD, ISD::ZEXTLOAD}, VT,
OtherVT, Expand);
}
setOperationAction({ISD::VP_FP_TO_SINT, ISD::VP_FP_TO_UINT,
ISD::VP_TRUNCATE, ISD::VP_SETCC},
VT, Custom);
setOperationAction(ISD::VECTOR_DEINTERLEAVE, VT, Custom);
setOperationAction(ISD::VECTOR_INTERLEAVE, VT, Custom);
setOperationAction(ISD::VECTOR_REVERSE, VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_SPLICE, VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_REVERSE, VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_SPLAT, VT, Custom);
setOperationPromotedToType(
ISD::VECTOR_SPLICE, VT,
MVT::getVectorVT(MVT::i8, VT.getVectorElementCount()));
}
for (MVT VT : IntVecVTs) {
if (!isTypeLegal(VT))
continue;
setOperationAction(ISD::SPLAT_VECTOR, VT, Legal);
setOperationAction(ISD::SPLAT_VECTOR_PARTS, VT, Custom);
// Vectors implement MULHS/MULHU.
setOperationAction({ISD::SMUL_LOHI, ISD::UMUL_LOHI}, VT, Expand);
// nxvXi64 MULHS/MULHU requires the V extension instead of Zve64*.
if (VT.getVectorElementType() == MVT::i64 && !Subtarget.hasStdExtV())
setOperationAction({ISD::MULHU, ISD::MULHS}, VT, Expand);
setOperationAction({ISD::SMIN, ISD::SMAX, ISD::UMIN, ISD::UMAX}, VT,
Legal);
setOperationAction({ISD::ABDS, ISD::ABDU}, VT, Custom);
// Custom-lower extensions and truncations from/to mask types.
setOperationAction({ISD::ANY_EXTEND, ISD::SIGN_EXTEND, ISD::ZERO_EXTEND},
VT, Custom);
// RVV has native int->float & float->int conversions where the
// element type sizes are within one power-of-two of each other. Any
// wider distances between type sizes have to be lowered as sequences
// which progressively narrow the gap in stages.
setOperationAction({ISD::SINT_TO_FP, ISD::UINT_TO_FP, ISD::FP_TO_SINT,
ISD::FP_TO_UINT, ISD::STRICT_SINT_TO_FP,
ISD::STRICT_UINT_TO_FP, ISD::STRICT_FP_TO_SINT,
ISD::STRICT_FP_TO_UINT},
VT, Custom);
setOperationAction({ISD::FP_TO_SINT_SAT, ISD::FP_TO_UINT_SAT}, VT,
Custom);
setOperationAction({ISD::AVGFLOORS, ISD::AVGFLOORU, ISD::AVGCEILS,
ISD::AVGCEILU, ISD::SADDSAT, ISD::UADDSAT,
ISD::SSUBSAT, ISD::USUBSAT},
VT, Legal);
// Integer VTs are lowered as a series of "RISCVISD::TRUNCATE_VECTOR_VL"
// nodes which truncate by one power of two at a time.
setOperationAction(
{ISD::TRUNCATE, ISD::TRUNCATE_SSAT_S, ISD::TRUNCATE_USAT_U}, VT,
Custom);
// Custom-lower insert/extract operations to simplify patterns.
setOperationAction({ISD::INSERT_VECTOR_ELT, ISD::EXTRACT_VECTOR_ELT}, VT,
Custom);
// Custom-lower reduction operations to set up the corresponding custom
// nodes' operands.
setOperationAction(IntegerVecReduceOps, VT, Custom);
setOperationAction(IntegerVPOps, VT, Custom);
setOperationAction({ISD::LOAD, ISD::STORE}, VT, Custom);
setOperationAction({ISD::MLOAD, ISD::MSTORE, ISD::MGATHER, ISD::MSCATTER},
VT, Custom);
setOperationAction(
{ISD::VP_LOAD, ISD::VP_STORE, ISD::EXPERIMENTAL_VP_STRIDED_LOAD,
ISD::EXPERIMENTAL_VP_STRIDED_STORE, ISD::VP_GATHER, ISD::VP_SCATTER},
VT, Custom);
setOperationAction(ISD::VP_LOAD_FF, VT, Custom);
setOperationAction({ISD::CONCAT_VECTORS, ISD::INSERT_SUBVECTOR,
ISD::EXTRACT_SUBVECTOR, ISD::SCALAR_TO_VECTOR},
VT, Custom);
setOperationAction(ISD::SELECT, VT, Custom);
setOperationAction(ISD::SELECT_CC, VT, Expand);
setOperationAction({ISD::STEP_VECTOR, ISD::VECTOR_REVERSE}, VT, Custom);
for (MVT OtherVT : MVT::integer_scalable_vector_valuetypes()) {
setTruncStoreAction(VT, OtherVT, Expand);
setLoadExtAction({ISD::EXTLOAD, ISD::SEXTLOAD, ISD::ZEXTLOAD}, VT,
OtherVT, Expand);
}
setOperationAction(ISD::VECTOR_DEINTERLEAVE, VT, Custom);
setOperationAction(ISD::VECTOR_INTERLEAVE, VT, Custom);
// Splice
setOperationAction(ISD::VECTOR_SPLICE, VT, Custom);
if (Subtarget.hasStdExtZvkb()) {
setOperationAction(ISD::BSWAP, VT, Legal);
setOperationAction(ISD::VP_BSWAP, VT, Custom);
} else {
setOperationAction({ISD::BSWAP, ISD::VP_BSWAP}, VT, Expand);
setOperationAction({ISD::ROTL, ISD::ROTR}, VT, Expand);
}
if (Subtarget.hasStdExtZvbb()) {
setOperationAction(ISD::BITREVERSE, VT, Legal);
setOperationAction(ISD::VP_BITREVERSE, VT, Custom);
setOperationAction({ISD::VP_CTLZ, ISD::VP_CTLZ_ZERO_UNDEF, ISD::VP_CTTZ,
ISD::VP_CTTZ_ZERO_UNDEF, ISD::VP_CTPOP},
VT, Custom);
} else {
setOperationAction({ISD::BITREVERSE, ISD::VP_BITREVERSE}, VT, Expand);
setOperationAction({ISD::CTLZ, ISD::CTTZ, ISD::CTPOP}, VT, Expand);
setOperationAction({ISD::VP_CTLZ, ISD::VP_CTLZ_ZERO_UNDEF, ISD::VP_CTTZ,
ISD::VP_CTTZ_ZERO_UNDEF, ISD::VP_CTPOP},
VT, Expand);
// Lower CTLZ_ZERO_UNDEF and CTTZ_ZERO_UNDEF if element of VT in the
// range of f32.
EVT FloatVT = MVT::getVectorVT(MVT::f32, VT.getVectorElementCount());
if (isTypeLegal(FloatVT)) {
setOperationAction({ISD::CTLZ, ISD::CTLZ_ZERO_UNDEF,
ISD::CTTZ_ZERO_UNDEF, ISD::VP_CTLZ,
ISD::VP_CTLZ_ZERO_UNDEF, ISD::VP_CTTZ_ZERO_UNDEF},
VT, Custom);
}
}
setOperationAction(ISD::VECTOR_COMPRESS, VT, Custom);
}
for (MVT VT : VecTupleVTs) {
if (!isTypeLegal(VT))
continue;
setOperationAction({ISD::LOAD, ISD::STORE}, VT, Custom);
}
// Expand various CCs to best match the RVV ISA, which natively supports UNE
// but no other unordered comparisons, and supports all ordered comparisons
// except ONE. Additionally, we expand GT,OGT,GE,OGE for optimization
// purposes; they are expanded to their swapped-operand CCs (LT,OLT,LE,OLE),
// and we pattern-match those back to the "original", swapping operands once
// more. This way we catch both operations and both "vf" and "fv" forms with
// fewer patterns.
static const ISD::CondCode VFPCCToExpand[] = {
ISD::SETO, ISD::SETONE, ISD::SETUEQ, ISD::SETUGT,
ISD::SETUGE, ISD::SETULT, ISD::SETULE, ISD::SETUO,
ISD::SETGT, ISD::SETOGT, ISD::SETGE, ISD::SETOGE,
};
// TODO: support more ops.
static const unsigned ZvfhminZvfbfminPromoteOps[] = {
ISD::FMINNUM,
ISD::FMAXNUM,
ISD::FMINIMUMNUM,
ISD::FMAXIMUMNUM,
ISD::FADD,
ISD::FSUB,
ISD::FMUL,
ISD::FMA,
ISD::FDIV,
ISD::FSQRT,
ISD::FCEIL,
ISD::FTRUNC,
ISD::FFLOOR,
ISD::FROUND,
ISD::FROUNDEVEN,
ISD::FRINT,
ISD::FNEARBYINT,
ISD::IS_FPCLASS,
ISD::SETCC,
ISD::FMAXIMUM,
ISD::FMINIMUM,
ISD::STRICT_FADD,
ISD::STRICT_FSUB,
ISD::STRICT_FMUL,
ISD::STRICT_FDIV,
ISD::STRICT_FSQRT,
ISD::STRICT_FMA,
ISD::VECREDUCE_FMIN,
ISD::VECREDUCE_FMAX,
ISD::VECREDUCE_FMINIMUM,
ISD::VECREDUCE_FMAXIMUM};
// TODO: support more vp ops.
static const unsigned ZvfhminZvfbfminPromoteVPOps[] = {
ISD::VP_FADD,
ISD::VP_FSUB,
ISD::VP_FMUL,
ISD::VP_FDIV,
ISD::VP_FMA,
ISD::VP_REDUCE_FMIN,
ISD::VP_REDUCE_FMAX,
ISD::VP_SQRT,
ISD::VP_FMINNUM,
ISD::VP_FMAXNUM,
ISD::VP_FCEIL,
ISD::VP_FFLOOR,
ISD::VP_FROUND,
ISD::VP_FROUNDEVEN,
ISD::VP_FROUNDTOZERO,
ISD::VP_FRINT,
ISD::VP_FNEARBYINT,
ISD::VP_SETCC,
ISD::VP_FMINIMUM,
ISD::VP_FMAXIMUM,
ISD::VP_REDUCE_FMINIMUM,
ISD::VP_REDUCE_FMAXIMUM};
// Sets common operation actions on RVV floating-point vector types.
const auto SetCommonVFPActions = [&](MVT VT) {
setOperationAction(ISD::SPLAT_VECTOR, VT, Legal);
// RVV has native FP_ROUND & FP_EXTEND conversions where the element type
// sizes are within one power-of-two of each other. Therefore conversions
// between vXf16 and vXf64 must be lowered as sequences which convert via
// vXf32.
setOperationAction({ISD::FP_ROUND, ISD::FP_EXTEND}, VT, Custom);
setOperationAction({ISD::LRINT, ISD::LLRINT}, VT, Custom);
setOperationAction({ISD::LROUND, ISD::LLROUND}, VT, Custom);
// Custom-lower insert/extract operations to simplify patterns.
setOperationAction({ISD::INSERT_VECTOR_ELT, ISD::EXTRACT_VECTOR_ELT}, VT,
Custom);
// Expand various condition codes (explained above).
setCondCodeAction(VFPCCToExpand, VT, Expand);
setOperationAction(
{ISD::FMINNUM, ISD::FMAXNUM, ISD::FMAXIMUMNUM, ISD::FMINIMUMNUM}, VT,
Legal);
setOperationAction({ISD::FMAXIMUM, ISD::FMINIMUM}, VT, Custom);
setOperationAction({ISD::FTRUNC, ISD::FCEIL, ISD::FFLOOR, ISD::FROUND,
ISD::FROUNDEVEN, ISD::FRINT, ISD::FNEARBYINT,
ISD::IS_FPCLASS},
VT, Custom);
setOperationAction(FloatingPointVecReduceOps, VT, Custom);
// Expand FP operations that need libcalls.
setOperationAction(FloatingPointLibCallOps, VT, Expand);
setOperationAction(ISD::FCOPYSIGN, VT, Legal);
setOperationAction({ISD::LOAD, ISD::STORE}, VT, Custom);
setOperationAction({ISD::MLOAD, ISD::MSTORE, ISD::MGATHER, ISD::MSCATTER},
VT, Custom);
setOperationAction(
{ISD::VP_LOAD, ISD::VP_STORE, ISD::EXPERIMENTAL_VP_STRIDED_LOAD,
ISD::EXPERIMENTAL_VP_STRIDED_STORE, ISD::VP_GATHER, ISD::VP_SCATTER},
VT, Custom);
setOperationAction(ISD::VP_LOAD_FF, VT, Custom);
setOperationAction(ISD::SELECT, VT, Custom);
setOperationAction(ISD::SELECT_CC, VT, Expand);
setOperationAction({ISD::CONCAT_VECTORS, ISD::INSERT_SUBVECTOR,
ISD::EXTRACT_SUBVECTOR, ISD::SCALAR_TO_VECTOR},
VT, Custom);
setOperationAction(ISD::VECTOR_DEINTERLEAVE, VT, Custom);
setOperationAction(ISD::VECTOR_INTERLEAVE, VT, Custom);
setOperationAction({ISD::VECTOR_REVERSE, ISD::VECTOR_SPLICE}, VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_SPLICE, VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_REVERSE, VT, Custom);
setOperationAction(FloatingPointVPOps, VT, Custom);
setOperationAction({ISD::STRICT_FP_EXTEND, ISD::STRICT_FP_ROUND}, VT,
Custom);
setOperationAction({ISD::STRICT_FADD, ISD::STRICT_FSUB, ISD::STRICT_FMUL,
ISD::STRICT_FDIV, ISD::STRICT_FSQRT, ISD::STRICT_FMA},
VT, Legal);
setOperationAction({ISD::STRICT_FSETCC, ISD::STRICT_FSETCCS,
ISD::STRICT_FTRUNC, ISD::STRICT_FCEIL,
ISD::STRICT_FFLOOR, ISD::STRICT_FROUND,
ISD::STRICT_FROUNDEVEN, ISD::STRICT_FNEARBYINT},
VT, Custom);
setOperationAction(ISD::VECTOR_COMPRESS, VT, Custom);
};
// Sets common extload/truncstore actions on RVV floating-point vector
// types.
const auto SetCommonVFPExtLoadTruncStoreActions =
[&](MVT VT, ArrayRef<MVT::SimpleValueType> SmallerVTs) {
for (auto SmallVT : SmallerVTs) {
setTruncStoreAction(VT, SmallVT, Expand);
setLoadExtAction(ISD::EXTLOAD, VT, SmallVT, Expand);
}
};
// Sets common actions for f16 and bf16 for when there's only
// zvfhmin/zvfbfmin and we need to promote to f32 for most operations.
const auto SetCommonPromoteToF32Actions = [&](MVT VT) {
setOperationAction({ISD::FP_ROUND, ISD::FP_EXTEND}, VT, Custom);
setOperationAction({ISD::STRICT_FP_ROUND, ISD::STRICT_FP_EXTEND}, VT,
Custom);
setOperationAction({ISD::VP_FP_ROUND, ISD::VP_FP_EXTEND}, VT, Custom);
setOperationAction({ISD::LRINT, ISD::LLRINT}, VT, Custom);
setOperationAction({ISD::LROUND, ISD::LLROUND}, VT, Custom);
setOperationAction({ISD::VP_MERGE, ISD::VP_SELECT, ISD::SELECT}, VT,
Custom);
setOperationAction(ISD::SELECT_CC, VT, Expand);
setOperationAction({ISD::VP_SINT_TO_FP, ISD::VP_UINT_TO_FP}, VT, Custom);
setOperationAction({ISD::INSERT_VECTOR_ELT, ISD::CONCAT_VECTORS,
ISD::INSERT_SUBVECTOR, ISD::EXTRACT_SUBVECTOR,
ISD::VECTOR_DEINTERLEAVE, ISD::VECTOR_INTERLEAVE,
ISD::VECTOR_REVERSE, ISD::VECTOR_SPLICE,
ISD::VECTOR_COMPRESS},
VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_SPLICE, VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_REVERSE, VT, Custom);
MVT EltVT = VT.getVectorElementType();
if (isTypeLegal(EltVT))
setOperationAction({ISD::SPLAT_VECTOR, ISD::EXPERIMENTAL_VP_SPLAT,
ISD::EXTRACT_VECTOR_ELT},
VT, Custom);
else
setOperationAction({ISD::SPLAT_VECTOR, ISD::EXPERIMENTAL_VP_SPLAT},
EltVT, Custom);
setOperationAction({ISD::LOAD, ISD::STORE, ISD::MLOAD, ISD::MSTORE,
ISD::MGATHER, ISD::MSCATTER, ISD::VP_LOAD,
ISD::VP_STORE, ISD::EXPERIMENTAL_VP_STRIDED_LOAD,
ISD::EXPERIMENTAL_VP_STRIDED_STORE, ISD::VP_GATHER,
ISD::VP_SCATTER},
VT, Custom);
setOperationAction(ISD::VP_LOAD_FF, VT, Custom);
setOperationAction(ISD::FNEG, VT, Expand);
setOperationAction(ISD::FABS, VT, Expand);
setOperationAction(ISD::FCOPYSIGN, VT, Expand);
// Expand FP operations that need libcalls.
setOperationAction(FloatingPointLibCallOps, VT, Expand);
// Custom split nxv32[b]f16 since nxv32[b]f32 is not legal.
if (getLMUL(VT) == RISCVVType::LMUL_8) {
setOperationAction(ZvfhminZvfbfminPromoteOps, VT, Custom);
setOperationAction(ZvfhminZvfbfminPromoteVPOps, VT, Custom);
} else {
MVT F32VecVT = MVT::getVectorVT(MVT::f32, VT.getVectorElementCount());
setOperationPromotedToType(ZvfhminZvfbfminPromoteOps, VT, F32VecVT);
setOperationPromotedToType(ZvfhminZvfbfminPromoteVPOps, VT, F32VecVT);
}
};
if (Subtarget.hasVInstructionsF16()) {
for (MVT VT : F16VecVTs) {
if (!isTypeLegal(VT))
continue;
SetCommonVFPActions(VT);
}
} else if (Subtarget.hasVInstructionsF16Minimal()) {
for (MVT VT : F16VecVTs) {
if (!isTypeLegal(VT))
continue;
SetCommonPromoteToF32Actions(VT);
}
}
if (Subtarget.hasVInstructionsBF16Minimal()) {
for (MVT VT : BF16VecVTs) {
if (!isTypeLegal(VT))
continue;
SetCommonPromoteToF32Actions(VT);
}
}
if (Subtarget.hasVInstructionsF32()) {
for (MVT VT : F32VecVTs) {
if (!isTypeLegal(VT))
continue;
SetCommonVFPActions(VT);
SetCommonVFPExtLoadTruncStoreActions(VT, F16VecVTs);
SetCommonVFPExtLoadTruncStoreActions(VT, BF16VecVTs);
}
}
if (Subtarget.hasVInstructionsF64()) {
for (MVT VT : F64VecVTs) {
if (!isTypeLegal(VT))
continue;
SetCommonVFPActions(VT);
SetCommonVFPExtLoadTruncStoreActions(VT, F16VecVTs);
SetCommonVFPExtLoadTruncStoreActions(VT, BF16VecVTs);
SetCommonVFPExtLoadTruncStoreActions(VT, F32VecVTs);
}
}
if (Subtarget.useRVVForFixedLengthVectors()) {
for (MVT VT : MVT::integer_fixedlen_vector_valuetypes()) {
if (!useRVVForFixedLengthVectorVT(VT))
continue;
// By default everything must be expanded.
for (unsigned Op = 0; Op < ISD::BUILTIN_OP_END; ++Op)
setOperationAction(Op, VT, Expand);
for (MVT OtherVT : MVT::integer_fixedlen_vector_valuetypes()) {
setTruncStoreAction(VT, OtherVT, Expand);
setLoadExtAction({ISD::EXTLOAD, ISD::SEXTLOAD, ISD::ZEXTLOAD}, VT,
OtherVT, Expand);
}
// Custom lower fixed vector undefs to scalable vector undefs to avoid
// expansion to a build_vector of 0s.
setOperationAction(ISD::UNDEF, VT, Custom);
// We use EXTRACT_SUBVECTOR as a "cast" from scalable to fixed.
setOperationAction({ISD::INSERT_SUBVECTOR, ISD::EXTRACT_SUBVECTOR}, VT,
Custom);
setOperationAction(
{ISD::BUILD_VECTOR, ISD::CONCAT_VECTORS, ISD::VECTOR_REVERSE}, VT,
Custom);
setOperationAction({ISD::VECTOR_INTERLEAVE, ISD::VECTOR_DEINTERLEAVE},
VT, Custom);
setOperationAction({ISD::INSERT_VECTOR_ELT, ISD::EXTRACT_VECTOR_ELT},
VT, Custom);
setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Custom);
setOperationAction({ISD::LOAD, ISD::STORE}, VT, Custom);
setOperationAction(ISD::SETCC, VT, Custom);
setOperationAction(ISD::SELECT, VT, Custom);
setOperationAction(
{ISD::TRUNCATE, ISD::TRUNCATE_SSAT_S, ISD::TRUNCATE_USAT_U}, VT,
Custom);
setOperationAction(ISD::BITCAST, VT, Custom);
setOperationAction(
{ISD::VECREDUCE_AND, ISD::VECREDUCE_OR, ISD::VECREDUCE_XOR}, VT,
Custom);
setOperationAction(
{ISD::VP_REDUCE_AND, ISD::VP_REDUCE_OR, ISD::VP_REDUCE_XOR}, VT,
Custom);
setOperationAction(
{
ISD::SINT_TO_FP,
ISD::UINT_TO_FP,
ISD::FP_TO_SINT,
ISD::FP_TO_UINT,
ISD::STRICT_SINT_TO_FP,
ISD::STRICT_UINT_TO_FP,
ISD::STRICT_FP_TO_SINT,
ISD::STRICT_FP_TO_UINT,
},
VT, Custom);
setOperationAction({ISD::FP_TO_SINT_SAT, ISD::FP_TO_UINT_SAT}, VT,
Custom);
setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
// Operations below are different for between masks and other vectors.
if (VT.getVectorElementType() == MVT::i1) {
setOperationAction({ISD::VP_AND, ISD::VP_OR, ISD::VP_XOR, ISD::AND,
ISD::OR, ISD::XOR},
VT, Custom);
setOperationAction({ISD::VP_FP_TO_SINT, ISD::VP_FP_TO_UINT,
ISD::VP_SETCC, ISD::VP_TRUNCATE},
VT, Custom);
setOperationAction(ISD::VP_MERGE, VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_SPLICE, VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_REVERSE, VT, Custom);
continue;
}
// Make SPLAT_VECTOR Legal so DAGCombine will convert splat vectors to
// it before type legalization for i64 vectors on RV32. It will then be
// type legalized to SPLAT_VECTOR_PARTS which we need to Custom handle.
// FIXME: Use SPLAT_VECTOR for all types? DAGCombine probably needs
// improvements first.
if (!Subtarget.is64Bit() && VT.getVectorElementType() == MVT::i64) {
setOperationAction(ISD::SPLAT_VECTOR, VT, Legal);
setOperationAction(ISD::SPLAT_VECTOR_PARTS, VT, Custom);
// Lower BUILD_VECTOR with i64 type to VID on RV32 if possible.
setOperationAction(ISD::BUILD_VECTOR, MVT::i64, Custom);
}
setOperationAction(
{ISD::MLOAD, ISD::MSTORE, ISD::MGATHER, ISD::MSCATTER}, VT, Custom);
setOperationAction({ISD::VP_LOAD, ISD::VP_STORE,
ISD::EXPERIMENTAL_VP_STRIDED_LOAD,
ISD::EXPERIMENTAL_VP_STRIDED_STORE, ISD::VP_GATHER,
ISD::VP_SCATTER},
VT, Custom);
setOperationAction(ISD::VP_LOAD_FF, VT, Custom);
setOperationAction({ISD::ADD, ISD::MUL, ISD::SUB, ISD::AND, ISD::OR,
ISD::XOR, ISD::SDIV, ISD::SREM, ISD::UDIV,
ISD::UREM, ISD::SHL, ISD::SRA, ISD::SRL},
VT, Custom);
setOperationAction(
{ISD::SMIN, ISD::SMAX, ISD::UMIN, ISD::UMAX, ISD::ABS}, VT, Custom);
setOperationAction({ISD::ABDS, ISD::ABDU}, VT, Custom);
// vXi64 MULHS/MULHU requires the V extension instead of Zve64*.
if (VT.getVectorElementType() != MVT::i64 || Subtarget.hasStdExtV())
setOperationAction({ISD::MULHS, ISD::MULHU}, VT, Custom);
setOperationAction({ISD::AVGFLOORS, ISD::AVGFLOORU, ISD::AVGCEILS,
ISD::AVGCEILU, ISD::SADDSAT, ISD::UADDSAT,
ISD::SSUBSAT, ISD::USUBSAT},
VT, Custom);
setOperationAction(ISD::VSELECT, VT, Custom);
setOperationAction(
{ISD::ANY_EXTEND, ISD::SIGN_EXTEND, ISD::ZERO_EXTEND}, VT, Custom);
// Custom-lower reduction operations to set up the corresponding custom
// nodes' operands.
setOperationAction({ISD::VECREDUCE_ADD, ISD::VECREDUCE_SMAX,
ISD::VECREDUCE_SMIN, ISD::VECREDUCE_UMAX,
ISD::VECREDUCE_UMIN},
VT, Custom);
setOperationAction(IntegerVPOps, VT, Custom);
if (Subtarget.hasStdExtZvkb())
setOperationAction({ISD::BSWAP, ISD::ROTL, ISD::ROTR}, VT, Custom);
if (Subtarget.hasStdExtZvbb()) {
setOperationAction({ISD::BITREVERSE, ISD::CTLZ, ISD::CTLZ_ZERO_UNDEF,
ISD::CTTZ, ISD::CTTZ_ZERO_UNDEF, ISD::CTPOP},
VT, Custom);
} else {
// Lower CTLZ_ZERO_UNDEF and CTTZ_ZERO_UNDEF if element of VT in the
// range of f32.
EVT FloatVT = MVT::getVectorVT(MVT::f32, VT.getVectorElementCount());
if (isTypeLegal(FloatVT))
setOperationAction(
{ISD::CTLZ, ISD::CTLZ_ZERO_UNDEF, ISD::CTTZ_ZERO_UNDEF}, VT,
Custom);
}
setOperationAction(ISD::VECTOR_COMPRESS, VT, Custom);
}
for (MVT VT : MVT::fp_fixedlen_vector_valuetypes()) {
// There are no extending loads or truncating stores.
for (MVT InnerVT : MVT::fp_fixedlen_vector_valuetypes()) {
setLoadExtAction(ISD::EXTLOAD, VT, InnerVT, Expand);
setTruncStoreAction(VT, InnerVT, Expand);
}
if (!useRVVForFixedLengthVectorVT(VT))
continue;
// By default everything must be expanded.
for (unsigned Op = 0; Op < ISD::BUILTIN_OP_END; ++Op)
setOperationAction(Op, VT, Expand);
// Custom lower fixed vector undefs to scalable vector undefs to avoid
// expansion to a build_vector of 0s.
setOperationAction(ISD::UNDEF, VT, Custom);
setOperationAction({ISD::INSERT_VECTOR_ELT, ISD::EXTRACT_VECTOR_ELT,
ISD::CONCAT_VECTORS, ISD::INSERT_SUBVECTOR,
ISD::EXTRACT_SUBVECTOR, ISD::VECTOR_REVERSE,
ISD::VECTOR_SHUFFLE, ISD::VECTOR_COMPRESS},
VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_SPLICE, VT, Custom);
setOperationAction(ISD::EXPERIMENTAL_VP_REVERSE, VT, Custom);
setOperationAction({ISD::VECTOR_INTERLEAVE, ISD::VECTOR_DEINTERLEAVE},
VT, Custom);
setOperationAction({ISD::LOAD, ISD::STORE, ISD::MLOAD, ISD::MSTORE,
ISD::MGATHER, ISD::MSCATTER},
VT, Custom);
setOperationAction({ISD::VP_LOAD, ISD::VP_STORE, ISD::VP_GATHER,
ISD::VP_SCATTER, ISD::EXPERIMENTAL_VP_STRIDED_LOAD,
ISD::EXPERIMENTAL_VP_STRIDED_STORE},
VT, Custom);
setOperationAction(ISD::VP_LOAD_FF, VT, Custom);
setOperationAction({ISD::FP_ROUND, ISD::FP_EXTEND}, VT, Custom);
setOperationAction({ISD::STRICT_FP_ROUND, ISD::STRICT_FP_EXTEND}, VT,
Custom);
if (VT.getVectorElementType() == MVT::f16 &&
!Subtarget.hasVInstructionsF16()) {
setOperationAction(ISD::BITCAST, VT, Custom);
setOperationAction({ISD::VP_FP_ROUND, ISD::VP_FP_EXTEND}, VT, Custom);
setOperationAction(
{ISD::VP_MERGE, ISD::VP_SELECT, ISD::VSELECT, ISD::SELECT}, VT,
Custom);
setOperationAction({ISD::VP_SINT_TO_FP, ISD::VP_UINT_TO_FP}, VT,
Custom);
setOperationAction({ISD::LRINT, ISD::LLRINT}, VT, Custom);
setOperationAction({ISD::LROUND, ISD::LLROUND}, VT, Custom);
if (Subtarget.hasStdExtZfhmin()) {
setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
} else {
// We need to custom legalize f16 build vectors if Zfhmin isn't
// available.
setOperationAction(ISD::BUILD_VECTOR, MVT::f16, Custom);
}
setOperationAction(ISD::FNEG, VT, Expand);
setOperationAction(ISD::FABS, VT, Expand);
setOperationAction(ISD::FCOPYSIGN, VT, Expand);
MVT F32VecVT = MVT::getVectorVT(MVT::f32, VT.getVectorElementCount());
// Don't promote f16 vector operations to f32 if f32 vector type is
// not legal.
// TODO: could split the f16 vector into two vectors and do promotion.
if (!isTypeLegal(F32VecVT))
continue;
setOperationPromotedToType(ZvfhminZvfbfminPromoteOps, VT, F32VecVT);
setOperationPromotedToType(ZvfhminZvfbfminPromoteVPOps, VT, F32VecVT);
continue;
}
if (VT.getVectorElementType() == MVT::bf16) {
setOperationAction(ISD::BITCAST, VT, Custom);
setOperationAction({ISD::VP_FP_ROUND, ISD::VP_FP_EXTEND}, VT, Custom);
setOperationAction({ISD::LRINT, ISD::LLRINT}, VT, Custom);
setOperationAction({ISD::LROUND, ISD::LLROUND}, VT, Custom);
if (Subtarget.hasStdExtZfbfmin()) {
setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
} else {
// We need to custom legalize bf16 build vectors if Zfbfmin isn't
// available.
setOperationAction(ISD::BUILD_VECTOR, MVT::bf16, Custom);
}
setOperationAction(
{ISD::VP_MERGE, ISD::VP_SELECT, ISD::VSELECT, ISD::SELECT}, VT,
Custom);
MVT F32VecVT = MVT::getVectorVT(MVT::f32, VT.getVectorElementCount());
// Don't promote f16 vector operations to f32 if f32 vector type is
// not legal.
// TODO: could split the f16 vector into two vectors and do promotion.
if (!isTypeLegal(F32VecVT))
continue;
setOperationPromotedToType(ZvfhminZvfbfminPromoteOps, VT, F32VecVT);
// TODO: Promote VP ops to fp32.
continue;
}
setOperationAction({ISD::BUILD_VECTOR, ISD::SCALAR_TO_VECTOR}, VT,
Custom);
setOperationAction({ISD::FADD, ISD::FSUB, ISD::FMUL, ISD::FDIV,
ISD::FNEG, ISD::FABS, ISD::FCOPYSIGN, ISD::FSQRT,
ISD::FMA, ISD::FMINNUM, ISD::FMAXNUM,
ISD::FMINIMUMNUM, ISD::FMAXIMUMNUM, ISD::IS_FPCLASS,
ISD::FMAXIMUM, ISD::FMINIMUM},
VT, Custom);
setOperationAction({ISD::FTRUNC, ISD::FCEIL, ISD::FFLOOR, ISD::FROUND,
ISD::FROUNDEVEN, ISD::FRINT, ISD::LRINT,
ISD::LLRINT, ISD::LROUND, ISD::LLROUND,
ISD::FNEARBYINT},
VT, Custom);
setCondCodeAction(VFPCCToExpand, VT, Expand);
setOperationAction(ISD::SETCC, VT, Custom);
setOperationAction({ISD::VSELECT, ISD::SELECT}, VT, Custom);
setOperationAction(ISD::BITCAST, VT, Custom);
setOperationAction(FloatingPointVecReduceOps, VT, Custom);
setOperationAction(FloatingPointVPOps, VT, Custom);
setOperationAction(
{ISD::STRICT_FADD, ISD::STRICT_FSUB, ISD::STRICT_FMUL,
ISD::STRICT_FDIV, ISD::STRICT_FSQRT, ISD::STRICT_FMA,
ISD::STRICT_FSETCC, ISD::STRICT_FSETCCS, ISD::STRICT_FTRUNC,
ISD::STRICT_FCEIL, ISD::STRICT_FFLOOR, ISD::STRICT_FROUND,
ISD::STRICT_FROUNDEVEN, ISD::STRICT_FNEARBYINT},
VT, Custom);
}
// Custom-legalize bitcasts from fixed-length vectors to scalar types.
setOperationAction(ISD::BITCAST, {MVT::i8, MVT::i16, MVT::i32}, Custom);
if (Subtarget.is64Bit())
setOperationAction(ISD::BITCAST, MVT::i64, Custom);
if (Subtarget.hasStdExtZfhminOrZhinxmin())
setOperationAction(ISD::BITCAST, MVT::f16, Custom);
if (Subtarget.hasStdExtZfbfmin())
setOperationAction(ISD::BITCAST, MVT::bf16, Custom);
if (Subtarget.hasStdExtFOrZfinx())
setOperationAction(ISD::BITCAST, MVT::f32, Custom);
if (Subtarget.hasStdExtDOrZdinx())
setOperationAction(ISD::BITCAST, MVT::f64, Custom);
}
}
if (Subtarget.hasStdExtA())
setOperationAction(ISD::ATOMIC_LOAD_SUB, XLenVT, Expand);
if (Subtarget.hasForcedAtomics()) {
// Force __sync libcalls to be emitted for atomic rmw/cas operations.
setOperationAction(
{ISD::ATOMIC_CMP_SWAP, ISD::ATOMIC_SWAP, ISD::ATOMIC_LOAD_ADD,
ISD::ATOMIC_LOAD_SUB, ISD::ATOMIC_LOAD_AND, ISD::ATOMIC_LOAD_OR,
ISD::ATOMIC_LOAD_XOR, ISD::ATOMIC_LOAD_NAND, ISD::ATOMIC_LOAD_MIN,
ISD::ATOMIC_LOAD_MAX, ISD::ATOMIC_LOAD_UMIN, ISD::ATOMIC_LOAD_UMAX},
XLenVT, LibCall);
}
if (Subtarget.hasVendorXTHeadMemIdx()) {
for (unsigned im : {ISD::PRE_INC, ISD::POST_INC}) {
setIndexedLoadAction(im, MVT::i8, Legal);
setIndexedStoreAction(im, MVT::i8, Legal);
setIndexedLoadAction(im, MVT::i16, Legal);
setIndexedStoreAction(im, MVT::i16, Legal);
setIndexedLoadAction(im, MVT::i32, Legal);
setIndexedStoreAction(im, MVT::i32, Legal);
if (Subtarget.is64Bit()) {
setIndexedLoadAction(im, MVT::i64, Legal);
setIndexedStoreAction(im, MVT::i64, Legal);
}
}
}
if (Subtarget.hasVendorXCVmem() && !Subtarget.is64Bit()) {
setIndexedLoadAction(ISD::POST_INC, MVT::i8, Legal);
setIndexedLoadAction(ISD::POST_INC, MVT::i16, Legal);
setIndexedLoadAction(ISD::POST_INC, MVT::i32, Legal);
setIndexedStoreAction(ISD::POST_INC, MVT::i8, Legal);
setIndexedStoreAction(ISD::POST_INC, MVT::i16, Legal);
setIndexedStoreAction(ISD::POST_INC, MVT::i32, Legal);
}
// zve32x is broken for partial_reduce_umla, but let's not make it worse.
if (Subtarget.hasStdExtZvqdotq() && Subtarget.getELen() >= 64) {
static const unsigned MLAOps[] = {ISD::PARTIAL_REDUCE_SMLA,
ISD::PARTIAL_REDUCE_UMLA,
ISD::PARTIAL_REDUCE_SUMLA};
setPartialReduceMLAAction(MLAOps, MVT::nxv1i32, MVT::nxv4i8, Custom);
setPartialReduceMLAAction(MLAOps, MVT::nxv2i32, MVT::nxv8i8, Custom);
setPartialReduceMLAAction(MLAOps, MVT::nxv4i32, MVT::nxv16i8, Custom);
setPartialReduceMLAAction(MLAOps, MVT::nxv8i32, MVT::nxv32i8, Custom);
setPartialReduceMLAAction(MLAOps, MVT::nxv16i32, MVT::nxv64i8, Custom);
if (Subtarget.useRVVForFixedLengthVectors()) {
for (MVT VT : MVT::integer_fixedlen_vector_valuetypes()) {
if (VT.getVectorElementType() != MVT::i32 ||
!useRVVForFixedLengthVectorVT(VT))
continue;
ElementCount EC = VT.getVectorElementCount();
MVT ArgVT = MVT::getVectorVT(MVT::i8, EC.multiplyCoefficientBy(4));
setPartialReduceMLAAction(MLAOps, VT, ArgVT, Custom);
}
}
}
// Customize load and store operation for bf16 if zfh isn't enabled.
if (Subtarget.hasVendorXAndesBFHCvt() && !Subtarget.hasStdExtZfh()) {
setOperationAction(ISD::LOAD, MVT::bf16, Custom);
setOperationAction(ISD::STORE, MVT::bf16, Custom);
}
// Function alignments.
const Align FunctionAlignment(Subtarget.hasStdExtZca() ? 2 : 4);
setMinFunctionAlignment(FunctionAlignment);
// Set preferred alignments.
setPrefFunctionAlignment(Subtarget.getPrefFunctionAlignment());
setPrefLoopAlignment(Subtarget.getPrefLoopAlignment());
setTargetDAGCombine({ISD::INTRINSIC_VOID, ISD::INTRINSIC_W_CHAIN,
ISD::INTRINSIC_WO_CHAIN, ISD::ADD, ISD::SUB, ISD::MUL,
ISD::AND, ISD::OR, ISD::XOR, ISD::SETCC, ISD::SELECT});
setTargetDAGCombine(ISD::SRA);
setTargetDAGCombine(ISD::SIGN_EXTEND_INREG);
if (Subtarget.hasStdExtFOrZfinx())
setTargetDAGCombine({ISD::FADD, ISD::FMAXNUM, ISD::FMINNUM, ISD::FMUL});
if (Subtarget.hasStdExtZbb())
setTargetDAGCombine({ISD::UMAX, ISD::UMIN, ISD::SMAX, ISD::SMIN});
if ((Subtarget.hasStdExtZbs() && Subtarget.is64Bit()) ||
Subtarget.hasVInstructions())
setTargetDAGCombine(ISD::TRUNCATE);
if (Subtarget.hasStdExtZbkb())
setTargetDAGCombine(ISD::BITREVERSE);
if (Subtarget.hasStdExtFOrZfinx())
setTargetDAGCombine({ISD::ZERO_EXTEND, ISD::FP_TO_SINT, ISD::FP_TO_UINT,
ISD::FP_TO_SINT_SAT, ISD::FP_TO_UINT_SAT});
if (Subtarget.hasVInstructions())
setTargetDAGCombine(
{ISD::FCOPYSIGN, ISD::MGATHER, ISD::MSCATTER,
ISD::VP_GATHER, ISD::VP_SCATTER, ISD::SRA,
ISD::SRL, ISD::SHL, ISD::STORE,
ISD::SPLAT_VECTOR, ISD::BUILD_VECTOR, ISD::CONCAT_VECTORS,
ISD::VP_STORE, ISD::VP_TRUNCATE, ISD::EXPERIMENTAL_VP_REVERSE,
ISD::MUL, ISD::SDIV, ISD::UDIV,
ISD::SREM, ISD::UREM, ISD::INSERT_VECTOR_ELT,
ISD::ABS, ISD::CTPOP, ISD::VECTOR_SHUFFLE,
ISD::VSELECT, ISD::VECREDUCE_ADD});
if (Subtarget.hasVendorXTHeadMemPair())
setTargetDAGCombine({ISD::LOAD, ISD::STORE});
if (Subtarget.useRVVForFixedLengthVectors())
setTargetDAGCombine(ISD::BITCAST);
// Disable strict node mutation.
IsStrictFPEnabled = true;
EnableExtLdPromotion = true;
// Let the subtarget decide if a predictable select is more expensive than the
// corresponding branch. This information is used in CGP/SelectOpt to decide
// when to convert selects into branches.
PredictableSelectIsExpensive = Subtarget.predictableSelectIsExpensive();
MaxStoresPerMemsetOptSize = Subtarget.getMaxStoresPerMemset(/*OptSize=*/true);
MaxStoresPerMemset = Subtarget.getMaxStoresPerMemset(/*OptSize=*/false);
MaxGluedStoresPerMemcpy = Subtarget.getMaxGluedStoresPerMemcpy();
MaxStoresPerMemcpyOptSize = Subtarget.getMaxStoresPerMemcpy(/*OptSize=*/true);
MaxStoresPerMemcpy = Subtarget.getMaxStoresPerMemcpy(/*OptSize=*/false);
MaxStoresPerMemmoveOptSize =
Subtarget.getMaxStoresPerMemmove(/*OptSize=*/true);
MaxStoresPerMemmove = Subtarget.getMaxStoresPerMemmove(/*OptSize=*/false);
MaxLoadsPerMemcmpOptSize = Subtarget.getMaxLoadsPerMemcmp(/*OptSize=*/true);
MaxLoadsPerMemcmp = Subtarget.getMaxLoadsPerMemcmp(/*OptSize=*/false);
}
EVT RISCVTargetLowering::getSetCCResultType(const DataLayout &DL,
LLVMContext &Context,
EVT VT) const {
if (!VT.isVector())
return getPointerTy(DL);
if (Subtarget.hasVInstructions() &&
(VT.isScalableVector() || Subtarget.useRVVForFixedLengthVectors()))
return EVT::getVectorVT(Context, MVT::i1, VT.getVectorElementCount());
return VT.changeVectorElementTypeToInteger();
}
MVT RISCVTargetLowering::getVPExplicitVectorLengthTy() const {
return Subtarget.getXLenVT();
}
// Return false if we can lower get_vector_length to a vsetvli intrinsic.
bool RISCVTargetLowering::shouldExpandGetVectorLength(EVT TripCountVT,
unsigned VF,
bool IsScalable) const {
if (!Subtarget.hasVInstructions())
return true;
if (!IsScalable)
return true;
if (TripCountVT != MVT::i32 && TripCountVT != Subtarget.getXLenVT())
return true;
// Don't allow VF=1 if those types are't legal.
if (VF < RISCV::RVVBitsPerBlock / Subtarget.getELen())
return true;
// VLEN=32 support is incomplete.
if (Subtarget.getRealMinVLen() < RISCV::RVVBitsPerBlock)
return true;
// The maximum VF is for the smallest element width with LMUL=8.
// VF must be a power of 2.
unsigned MaxVF = RISCV::RVVBytesPerBlock * 8;
return VF > MaxVF || !isPowerOf2_32(VF);
}
bool RISCVTargetLowering::shouldExpandCttzElements(EVT VT) const {
return !Subtarget.hasVInstructions() ||
VT.getVectorElementType() != MVT::i1 || !isTypeLegal(VT);
}
bool RISCVTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
const CallInst &I,
MachineFunction &MF,
unsigned Intrinsic) const {
auto &DL = I.getDataLayout();
auto SetRVVLoadStoreInfo = [&](unsigned PtrOp, bool IsStore,
bool IsUnitStrided, bool UsePtrVal = false) {
Info.opc = IsStore ? ISD::INTRINSIC_VOID : ISD::INTRINSIC_W_CHAIN;
// We can't use ptrVal if the intrinsic can access memory before the
// pointer. This means we can't use it for strided or indexed intrinsics.
if (UsePtrVal)
Info.ptrVal = I.getArgOperand(PtrOp);
else
Info.fallbackAddressSpace =
I.getArgOperand(PtrOp)->getType()->getPointerAddressSpace();
Type *MemTy;
if (IsStore) {
// Store value is the first operand.
MemTy = I.getArgOperand(0)->getType();
} else {
// Use return type. If it's segment load, return type is a struct.
MemTy = I.getType();
if (MemTy->isStructTy())
MemTy = MemTy->getStructElementType(0);
}
if (!IsUnitStrided)
MemTy = MemTy->getScalarType();
Info.memVT = getValueType(DL, MemTy);
if (MemTy->isTargetExtTy()) {
// RISC-V vector tuple type's alignment type should be its element type.
if (cast<TargetExtType>(MemTy)->getName() == "riscv.vector.tuple")
MemTy = Type::getIntNTy(
MemTy->getContext(),
1 << cast<ConstantInt>(I.getArgOperand(I.arg_size() - 1))
->getZExtValue());
Info.align = DL.getABITypeAlign(MemTy);
} else {
Info.align = Align(DL.getTypeStoreSize(MemTy->getScalarType()));
}
Info.size = MemoryLocation::UnknownSize;
Info.flags |=
IsStore ? MachineMemOperand::MOStore : MachineMemOperand::MOLoad;
return true;
};
if (I.hasMetadata(LLVMContext::MD_nontemporal))
Info.flags |= MachineMemOperand::MONonTemporal;
Info.flags |= RISCVTargetLowering::getTargetMMOFlags(I);
switch (Intrinsic) {
default:
return false;
case Intrinsic::riscv_masked_atomicrmw_xchg:
case Intrinsic::riscv_masked_atomicrmw_add:
case Intrinsic::riscv_masked_atomicrmw_sub:
case Intrinsic::riscv_masked_atomicrmw_nand:
case Intrinsic::riscv_masked_atomicrmw_max:
case Intrinsic::riscv_masked_atomicrmw_min:
case Intrinsic::riscv_masked_atomicrmw_umax:
case Intrinsic::riscv_masked_atomicrmw_umin:
case Intrinsic::riscv_masked_cmpxchg:
// riscv_masked_{atomicrmw_*,cmpxchg} intrinsics represent an emulated
// narrow atomic operation. These will be expanded to an LR/SC loop that
// reads/writes to/from an aligned 4 byte location. And, or, shift, etc.
// will be used to modify the appropriate part of the 4 byte data and
// preserve the rest.
Info.opc = ISD::INTRINSIC_W_CHAIN;
Info.memVT = MVT::i32;
Info.ptrVal = I.getArgOperand(0);
Info.offset = 0;
Info.align = Align(4);
Info.flags = MachineMemOperand::MOLoad | MachineMemOperand::MOStore |
MachineMemOperand::MOVolatile;
return true;
case Intrinsic::riscv_seg2_load_mask:
case Intrinsic::riscv_seg3_load_mask:
case Intrinsic::riscv_seg4_load_mask:
case Intrinsic::riscv_seg5_load_mask:
case Intrinsic::riscv_seg6_load_mask:
case Intrinsic::riscv_seg7_load_mask:
case Intrinsic::riscv_seg8_load_mask:
case Intrinsic::riscv_sseg2_load_mask:
case Intrinsic::riscv_sseg3_load_mask:
case Intrinsic::riscv_sseg4_load_mask:
case Intrinsic::riscv_sseg5_load_mask:
case Intrinsic::riscv_sseg6_load_mask:
case Intrinsic::riscv_sseg7_load_mask:
case Intrinsic::riscv_sseg8_load_mask:
return SetRVVLoadStoreInfo(/*PtrOp*/ 0, /*IsStore*/ false,
/*IsUnitStrided*/ false, /*UsePtrVal*/ true);
case Intrinsic::riscv_seg2_store_mask:
case Intrinsic::riscv_seg3_store_mask:
case Intrinsic::riscv_seg4_store_mask:
case Intrinsic::riscv_seg5_store_mask:
case Intrinsic::riscv_seg6_store_mask:
case Intrinsic::riscv_seg7_store_mask:
case Intrinsic::riscv_seg8_store_mask:
// Operands are (vec, ..., vec, ptr, mask, vl)
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 3,
/*IsStore*/ true,
/*IsUnitStrided*/ false, /*UsePtrVal*/ true);
case Intrinsic::riscv_sseg2_store_mask:
case Intrinsic::riscv_sseg3_store_mask:
case Intrinsic::riscv_sseg4_store_mask:
case Intrinsic::riscv_sseg5_store_mask:
case Intrinsic::riscv_sseg6_store_mask:
case Intrinsic::riscv_sseg7_store_mask:
case Intrinsic::riscv_sseg8_store_mask:
// Operands are (vec, ..., vec, ptr, offset, mask, vl)
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 4,
/*IsStore*/ true,
/*IsUnitStrided*/ false, /*UsePtrVal*/ true);
case Intrinsic::riscv_vlm:
return SetRVVLoadStoreInfo(/*PtrOp*/ 0,
/*IsStore*/ false,
/*IsUnitStrided*/ true,
/*UsePtrVal*/ true);
case Intrinsic::riscv_vle:
case Intrinsic::riscv_vle_mask:
case Intrinsic::riscv_vleff:
case Intrinsic::riscv_vleff_mask:
return SetRVVLoadStoreInfo(/*PtrOp*/ 1,
/*IsStore*/ false,
/*IsUnitStrided*/ true,
/*UsePtrVal*/ true);
case Intrinsic::riscv_vsm:
case Intrinsic::riscv_vse:
case Intrinsic::riscv_vse_mask:
return SetRVVLoadStoreInfo(/*PtrOp*/ 1,
/*IsStore*/ true,
/*IsUnitStrided*/ true,
/*UsePtrVal*/ true);
case Intrinsic::riscv_vlse:
case Intrinsic::riscv_vlse_mask:
case Intrinsic::riscv_vloxei:
case Intrinsic::riscv_vloxei_mask:
case Intrinsic::riscv_vluxei:
case Intrinsic::riscv_vluxei_mask:
return SetRVVLoadStoreInfo(/*PtrOp*/ 1,
/*IsStore*/ false,
/*IsUnitStrided*/ false);
case Intrinsic::riscv_vsse:
case Intrinsic::riscv_vsse_mask:
case Intrinsic::riscv_vsoxei:
case Intrinsic::riscv_vsoxei_mask:
case Intrinsic::riscv_vsuxei:
case Intrinsic::riscv_vsuxei_mask:
return SetRVVLoadStoreInfo(/*PtrOp*/ 1,
/*IsStore*/ true,
/*IsUnitStrided*/ false);
case Intrinsic::riscv_vlseg2:
case Intrinsic::riscv_vlseg3:
case Intrinsic::riscv_vlseg4:
case Intrinsic::riscv_vlseg5:
case Intrinsic::riscv_vlseg6:
case Intrinsic::riscv_vlseg7:
case Intrinsic::riscv_vlseg8:
case Intrinsic::riscv_vlseg2ff:
case Intrinsic::riscv_vlseg3ff:
case Intrinsic::riscv_vlseg4ff:
case Intrinsic::riscv_vlseg5ff:
case Intrinsic::riscv_vlseg6ff:
case Intrinsic::riscv_vlseg7ff:
case Intrinsic::riscv_vlseg8ff:
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 3,
/*IsStore*/ false,
/*IsUnitStrided*/ false, /*UsePtrVal*/ true);
case Intrinsic::riscv_vlseg2_mask:
case Intrinsic::riscv_vlseg3_mask:
case Intrinsic::riscv_vlseg4_mask:
case Intrinsic::riscv_vlseg5_mask:
case Intrinsic::riscv_vlseg6_mask:
case Intrinsic::riscv_vlseg7_mask:
case Intrinsic::riscv_vlseg8_mask:
case Intrinsic::riscv_vlseg2ff_mask:
case Intrinsic::riscv_vlseg3ff_mask:
case Intrinsic::riscv_vlseg4ff_mask:
case Intrinsic::riscv_vlseg5ff_mask:
case Intrinsic::riscv_vlseg6ff_mask:
case Intrinsic::riscv_vlseg7ff_mask:
case Intrinsic::riscv_vlseg8ff_mask:
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 5,
/*IsStore*/ false,
/*IsUnitStrided*/ false, /*UsePtrVal*/ true);
case Intrinsic::riscv_vlsseg2:
case Intrinsic::riscv_vlsseg3:
case Intrinsic::riscv_vlsseg4:
case Intrinsic::riscv_vlsseg5:
case Intrinsic::riscv_vlsseg6:
case Intrinsic::riscv_vlsseg7:
case Intrinsic::riscv_vlsseg8:
case Intrinsic::riscv_vloxseg2:
case Intrinsic::riscv_vloxseg3:
case Intrinsic::riscv_vloxseg4:
case Intrinsic::riscv_vloxseg5:
case Intrinsic::riscv_vloxseg6:
case Intrinsic::riscv_vloxseg7:
case Intrinsic::riscv_vloxseg8:
case Intrinsic::riscv_vluxseg2:
case Intrinsic::riscv_vluxseg3:
case Intrinsic::riscv_vluxseg4:
case Intrinsic::riscv_vluxseg5:
case Intrinsic::riscv_vluxseg6:
case Intrinsic::riscv_vluxseg7:
case Intrinsic::riscv_vluxseg8:
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 4,
/*IsStore*/ false,
/*IsUnitStrided*/ false);
case Intrinsic::riscv_vlsseg2_mask:
case Intrinsic::riscv_vlsseg3_mask:
case Intrinsic::riscv_vlsseg4_mask:
case Intrinsic::riscv_vlsseg5_mask:
case Intrinsic::riscv_vlsseg6_mask:
case Intrinsic::riscv_vlsseg7_mask:
case Intrinsic::riscv_vlsseg8_mask:
case Intrinsic::riscv_vloxseg2_mask:
case Intrinsic::riscv_vloxseg3_mask:
case Intrinsic::riscv_vloxseg4_mask:
case Intrinsic::riscv_vloxseg5_mask:
case Intrinsic::riscv_vloxseg6_mask:
case Intrinsic::riscv_vloxseg7_mask:
case Intrinsic::riscv_vloxseg8_mask:
case Intrinsic::riscv_vluxseg2_mask:
case Intrinsic::riscv_vluxseg3_mask:
case Intrinsic::riscv_vluxseg4_mask:
case Intrinsic::riscv_vluxseg5_mask:
case Intrinsic::riscv_vluxseg6_mask:
case Intrinsic::riscv_vluxseg7_mask:
case Intrinsic::riscv_vluxseg8_mask:
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 6,
/*IsStore*/ false,
/*IsUnitStrided*/ false);
case Intrinsic::riscv_vsseg2:
case Intrinsic::riscv_vsseg3:
case Intrinsic::riscv_vsseg4:
case Intrinsic::riscv_vsseg5:
case Intrinsic::riscv_vsseg6:
case Intrinsic::riscv_vsseg7:
case Intrinsic::riscv_vsseg8:
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 3,
/*IsStore*/ true,
/*IsUnitStrided*/ false);
case Intrinsic::riscv_vsseg2_mask:
case Intrinsic::riscv_vsseg3_mask:
case Intrinsic::riscv_vsseg4_mask:
case Intrinsic::riscv_vsseg5_mask:
case Intrinsic::riscv_vsseg6_mask:
case Intrinsic::riscv_vsseg7_mask:
case Intrinsic::riscv_vsseg8_mask:
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 4,
/*IsStore*/ true,
/*IsUnitStrided*/ false);
case Intrinsic::riscv_vssseg2:
case Intrinsic::riscv_vssseg3:
case Intrinsic::riscv_vssseg4:
case Intrinsic::riscv_vssseg5:
case Intrinsic::riscv_vssseg6:
case Intrinsic::riscv_vssseg7:
case Intrinsic::riscv_vssseg8:
case Intrinsic::riscv_vsoxseg2:
case Intrinsic::riscv_vsoxseg3:
case Intrinsic::riscv_vsoxseg4:
case Intrinsic::riscv_vsoxseg5:
case Intrinsic::riscv_vsoxseg6:
case Intrinsic::riscv_vsoxseg7:
case Intrinsic::riscv_vsoxseg8:
case Intrinsic::riscv_vsuxseg2:
case Intrinsic::riscv_vsuxseg3:
case Intrinsic::riscv_vsuxseg4:
case Intrinsic::riscv_vsuxseg5:
case Intrinsic::riscv_vsuxseg6:
case Intrinsic::riscv_vsuxseg7:
case Intrinsic::riscv_vsuxseg8:
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 4,
/*IsStore*/ true,
/*IsUnitStrided*/ false);
case Intrinsic::riscv_vssseg2_mask:
case Intrinsic::riscv_vssseg3_mask:
case Intrinsic::riscv_vssseg4_mask:
case Intrinsic::riscv_vssseg5_mask:
case Intrinsic::riscv_vssseg6_mask:
case Intrinsic::riscv_vssseg7_mask:
case Intrinsic::riscv_vssseg8_mask:
case Intrinsic::riscv_vsoxseg2_mask:
case Intrinsic::riscv_vsoxseg3_mask:
case Intrinsic::riscv_vsoxseg4_mask:
case Intrinsic::riscv_vsoxseg5_mask:
case Intrinsic::riscv_vsoxseg6_mask:
case Intrinsic::riscv_vsoxseg7_mask:
case Intrinsic::riscv_vsoxseg8_mask:
case Intrinsic::riscv_vsuxseg2_mask:
case Intrinsic::riscv_vsuxseg3_mask:
case Intrinsic::riscv_vsuxseg4_mask:
case Intrinsic::riscv_vsuxseg5_mask:
case Intrinsic::riscv_vsuxseg6_mask:
case Intrinsic::riscv_vsuxseg7_mask:
case Intrinsic::riscv_vsuxseg8_mask:
return SetRVVLoadStoreInfo(/*PtrOp*/ I.arg_size() - 5,
/*IsStore*/ true,
/*IsUnitStrided*/ false);
}
}
bool RISCVTargetLowering::isLegalAddressingMode(const DataLayout &DL,
const AddrMode &AM, Type *Ty,
unsigned AS,
Instruction *I) const {
// No global is ever allowed as a base.
if (AM.BaseGV)
return false;
// None of our addressing modes allows a scalable offset
if (AM.ScalableOffset)
return false;
// RVV instructions only support register addressing.
if (Subtarget.hasVInstructions() && isa<VectorType>(Ty))
return AM.HasBaseReg && AM.Scale == 0 && !AM.BaseOffs;
// Require a 12-bit signed offset.
if (!isInt<12>(AM.BaseOffs))
return false;
switch (AM.Scale) {
case 0: // "r+i" or just "i", depending on HasBaseReg.
break;
case 1:
if (!AM.HasBaseReg) // allow "r+i".
break;
return false; // disallow "r+r" or "r+r+i".
default:
return false;
}
return true;
}
bool RISCVTargetLowering::isLegalICmpImmediate(int64_t Imm) const {
return isInt<12>(Imm);
}
bool RISCVTargetLowering::isLegalAddImmediate(int64_t Imm) const {
return isInt<12>(Imm);
}
// On RV32, 64-bit integers are split into their high and low parts and held
// in two different registers, so the trunc is free since the low register can
// just be used.
// FIXME: Should we consider i64->i32 free on RV64 to match the EVT version of
// isTruncateFree?
bool RISCVTargetLowering::isTruncateFree(Type *SrcTy, Type *DstTy) const {
if (Subtarget.is64Bit() || !SrcTy->isIntegerTy() || !DstTy->isIntegerTy())
return false;
unsigned SrcBits = SrcTy->getPrimitiveSizeInBits();
unsigned DestBits = DstTy->getPrimitiveSizeInBits();
return (SrcBits == 64 && DestBits == 32);
}
bool RISCVTargetLowering::isTruncateFree(EVT SrcVT, EVT DstVT) const {
// We consider i64->i32 free on RV64 since we have good selection of W
// instructions that make promoting operations back to i64 free in many cases.
if (SrcVT.isVector() || DstVT.isVector() || !SrcVT.isInteger() ||
!DstVT.isInteger())
return false;
unsigned SrcBits = SrcVT.getSizeInBits();
unsigned DestBits = DstVT.getSizeInBits();
return (SrcBits == 64 && DestBits == 32);
}
bool RISCVTargetLowering::isTruncateFree(SDValue Val, EVT VT2) const {
EVT SrcVT = Val.getValueType();
// free truncate from vnsrl and vnsra
if (Subtarget.hasVInstructions() &&
(Val.getOpcode() == ISD::SRL || Val.getOpcode() == ISD::SRA) &&
SrcVT.isVector() && VT2.isVector()) {
unsigned SrcBits = SrcVT.getVectorElementType().getSizeInBits();
unsigned DestBits = VT2.getVectorElementType().getSizeInBits();
if (SrcBits == DestBits * 2) {
return true;
}
}
return TargetLowering::isTruncateFree(Val, VT2);
}
bool RISCVTargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
// Zexts are free if they can be combined with a load.
// Don't advertise i32->i64 zextload as being free for RV64. It interacts
// poorly with type legalization of compares preferring sext.
if (auto *LD = dyn_cast<LoadSDNode>(Val)) {
EVT MemVT = LD->getMemoryVT();
if ((MemVT == MVT::i8 || MemVT == MVT::i16) &&
(LD->getExtensionType() == ISD::NON_EXTLOAD ||
LD->getExtensionType() == ISD::ZEXTLOAD))
return true;
}
return TargetLowering::isZExtFree(Val, VT2);
}
bool RISCVTargetLowering::isSExtCheaperThanZExt(EVT SrcVT, EVT DstVT) const {
return Subtarget.is64Bit() && SrcVT == MVT::i32 && DstVT == MVT::i64;
}
bool RISCVTargetLowering::signExtendConstant(const ConstantInt *CI) const {
return Subtarget.is64Bit() && CI->getType()->isIntegerTy(32);
}
bool RISCVTargetLowering::isCheapToSpeculateCttz(Type *Ty) const {
return Subtarget.hasStdExtZbb() ||
(Subtarget.hasVendorXCVbitmanip() && !Subtarget.is64Bit());
}
bool RISCVTargetLowering::isCheapToSpeculateCtlz(Type *Ty) const {
return Subtarget.hasStdExtZbb() || Subtarget.hasVendorXTHeadBb() ||
(Subtarget.hasVendorXCVbitmanip() && !Subtarget.is64Bit());
}
bool RISCVTargetLowering::isMaskAndCmp0FoldingBeneficial(
const Instruction &AndI) const {
// We expect to be able to match a bit extraction instruction if the Zbs
// extension is supported and the mask is a power of two. However, we
// conservatively return false if the mask would fit in an ANDI instruction,
// on the basis that it's possible the sinking+duplication of the AND in
// CodeGenPrepare triggered by this hook wouldn't decrease the instruction
// count and would increase code size (e.g. ANDI+BNEZ => BEXTI+BNEZ).
if (!Subtarget.hasStdExtZbs() && !Subtarget.hasVendorXTHeadBs())
return false;
ConstantInt *Mask = dyn_cast<ConstantInt>(AndI.getOperand(1));
if (!Mask)
return false;
return !Mask->getValue().isSignedIntN(12) && Mask->getValue().isPowerOf2();
}
bool RISCVTargetLowering::hasAndNotCompare(SDValue Y) const {
EVT VT = Y.getValueType();
if (VT.isVector())
return false;
return (Subtarget.hasStdExtZbb() || Subtarget.hasStdExtZbkb()) &&
(!isa<ConstantSDNode>(Y) || cast<ConstantSDNode>(Y)->isOpaque());
}
bool RISCVTargetLowering::hasAndNot(SDValue Y) const {
EVT VT = Y.getValueType();
if (!VT.isVector())
return hasAndNotCompare(Y);
return Subtarget.hasStdExtZvkb();
}
bool RISCVTargetLowering::hasBitTest(SDValue X, SDValue Y) const {
// Zbs provides BEXT[_I], which can be used with SEQZ/SNEZ as a bit test.
if (Subtarget.hasStdExtZbs())
return X.getValueType().isScalarInteger();
auto *C = dyn_cast<ConstantSDNode>(Y);
// XTheadBs provides th.tst (similar to bexti), if Y is a constant
if (Subtarget.hasVendorXTHeadBs())
return C != nullptr;
// We can use ANDI+SEQZ/SNEZ as a bit test. Y contains the bit position.
return C && C->getAPIntValue().ule(10);
}
bool RISCVTargetLowering::shouldFoldSelectWithIdentityConstant(
unsigned BinOpcode, EVT VT, unsigned SelectOpcode, SDValue X,
SDValue Y) const {
if (SelectOpcode != ISD::VSELECT)
return false;
// Only enable for rvv.
if (!VT.isVector() || !Subtarget.hasVInstructions())
return false;
if (VT.isFixedLengthVector() && !isTypeLegal(VT))
return false;
return true;
}
bool RISCVTargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getIntegerBitWidth();
if (BitSize > Subtarget.getXLen())
return false;
// Fast path, assume 32-bit immediates are cheap.
int64_t Val = Imm.getSExtValue();
if (isInt<32>(Val))
return true;
// A constant pool entry may be more aligned than the load we're trying to
// replace. If we don't support unaligned scalar mem, prefer the constant
// pool.
// TODO: Can the caller pass down the alignment?
if (!Subtarget.enableUnalignedScalarMem())
return true;
// Prefer to keep the load if it would require many instructions.
// This uses the same threshold we use for constant pools but doesn't
// check useConstantPoolForLargeInts.
// TODO: Should we keep the load only when we're definitely going to emit a
// constant pool?
RISCVMatInt::InstSeq Seq = RISCVMatInt::generateInstSeq(Val, Subtarget);
return Seq.size() <= Subtarget.getMaxBuildIntsCost();
}
bool RISCVTargetLowering::
shouldProduceAndByConstByHoistingConstFromShiftsLHSOfAnd(
SDValue X, ConstantSDNode *XC, ConstantSDNode *CC, SDValue Y,
unsigned OldShiftOpcode, unsigned NewShiftOpcode,
SelectionDAG &DAG) const {
// One interesting pattern that we'd want to form is 'bit extract':
// ((1 >> Y) & 1) ==/!= 0
// But we also need to be careful not to try to reverse that fold.
// Is this '((1 >> Y) & 1)'?
if (XC && OldShiftOpcode == ISD::SRL && XC->isOne())
return false; // Keep the 'bit extract' pattern.
// Will this be '((1 >> Y) & 1)' after the transform?
if (NewShiftOpcode == ISD::SRL && CC->isOne())
return true; // Do form the 'bit extract' pattern.
// If 'X' is a constant, and we transform, then we will immediately
// try to undo the fold, thus causing endless combine loop.
// So only do the transform if X is not a constant. This matches the default
// implementation of this function.
return !XC;
}
bool RISCVTargetLowering::shouldScalarizeBinop(SDValue VecOp) const {
unsigned Opc = VecOp.getOpcode();
// Assume target opcodes can't be scalarized.
// TODO - do we have any exceptions?
if (Opc >= ISD::BUILTIN_OP_END || !isBinOp(Opc))
return false;
// If the vector op is not supported, try to convert to scalar.
EVT VecVT = VecOp.getValueType();
if (!isOperationLegalOrCustomOrPromote(Opc, VecVT))
return true;
// If the vector op is supported, but the scalar op is not, the transform may
// not be worthwhile.
// Permit a vector binary operation can be converted to scalar binary
// operation which is custom lowered with illegal type.
EVT ScalarVT = VecVT.getScalarType();
return isOperationLegalOrCustomOrPromote(Opc, ScalarVT) ||
isOperationCustom(Opc, ScalarVT);
}
bool RISCVTargetLowering::isOffsetFoldingLegal(
const GlobalAddressSDNode *GA) const {
// In order to maximise the opportunity for common subexpression elimination,
// keep a separate ADD node for the global address offset instead of folding
// it in the global address node. Later peephole optimisations may choose to
// fold it back in when profitable.
return false;
}
// Returns 0-31 if the fli instruction is available for the type and this is
// legal FP immediate for the type. Returns -1 otherwise.
int RISCVTargetLowering::getLegalZfaFPImm(const APFloat &Imm, EVT VT) const {
if (!Subtarget.hasStdExtZfa())
return -1;
bool IsSupportedVT = false;
if (VT == MVT::f16) {
IsSupportedVT = Subtarget.hasStdExtZfh() || Subtarget.hasStdExtZvfh();
} else if (VT == MVT::f32) {
IsSupportedVT = true;
} else if (VT == MVT::f64) {
assert(Subtarget.hasStdExtD() && "Expect D extension");
IsSupportedVT = true;
}
if (!IsSupportedVT)
return -1;
return RISCVLoadFPImm::getLoadFPImm(Imm);
}
bool RISCVTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT,
bool ForCodeSize) const {
bool IsLegalVT = false;
if (VT == MVT::f16)
IsLegalVT = Subtarget.hasStdExtZfhminOrZhinxmin();
else if (VT == MVT::f32)
IsLegalVT = Subtarget.hasStdExtFOrZfinx();
else if (VT == MVT::f64)
IsLegalVT = Subtarget.hasStdExtDOrZdinx();
else if (VT == MVT::bf16)
IsLegalVT = Subtarget.hasStdExtZfbfmin();
if (!IsLegalVT)
return false;
if (getLegalZfaFPImm(Imm, VT) >= 0)
return true;
// Some constants can be produced by fli+fneg.
if (Imm.isNegative() && getLegalZfaFPImm(-Imm, VT) >= 0)
return true;
// Cannot create a 64 bit floating-point immediate value for rv32.
if (Subtarget.getXLen() < VT.getScalarSizeInBits()) {
// td can handle +0.0 or -0.0 already.
// -0.0 can be created by fmv + fneg.
return Imm.isZero();
}
// Special case: fmv + fneg
if (Imm.isNegZero())
return true;
// Building an integer and then converting requires a fmv at the end of
// the integer sequence. The fmv is not required for Zfinx.
const int FmvCost = Subtarget.hasStdExtZfinx() ? 0 : 1;
const int Cost =
FmvCost + RISCVMatInt::getIntMatCost(Imm.bitcastToAPInt(),
Subtarget.getXLen(), Subtarget);
return Cost <= FPImmCost;
}
// TODO: This is very conservative.
bool RISCVTargetLowering::isExtractSubvectorCheap(EVT ResVT, EVT SrcVT,
unsigned Index) const {
if (!isOperationLegalOrCustom(ISD::EXTRACT_SUBVECTOR, ResVT))
return false;
// Extracts from index 0 are just subreg extracts.
if (Index == 0)
return true;
// Only support extracting a fixed from a fixed vector for now.
if (ResVT.isScalableVector() || SrcVT.isScalableVector())
return false;
EVT EltVT = ResVT.getVectorElementType();
assert(EltVT == SrcVT.getVectorElementType() && "Should hold for node");
// The smallest type we can slide is i8.
// TODO: We can extract index 0 from a mask vector without a slide.
if (EltVT == MVT::i1)
return false;
unsigned ResElts = ResVT.getVectorNumElements();
unsigned SrcElts = SrcVT.getVectorNumElements();
unsigned MinVLen = Subtarget.getRealMinVLen();
unsigned MinVLMAX = MinVLen / EltVT.getSizeInBits();
// If we're extracting only data from the first VLEN bits of the source
// then we can always do this with an m1 vslidedown.vx. Restricting the
// Index ensures we can use a vslidedown.vi.
// TODO: We can generalize this when the exact VLEN is known.
if (Index + ResElts <= MinVLMAX && Index < 31)
return true;
// Convervatively only handle extracting half of a vector.
// TODO: We can do arbitrary slidedowns, but for now only support extracting
// the upper half of a vector until we have more test coverage.
// TODO: For sizes which aren't multiples of VLEN sizes, this may not be
// a cheap extract. However, this case is important in practice for
// shuffled extracts of longer vectors. How resolve?
return (ResElts * 2) == SrcElts && (Index == 0 || Index == ResElts);
}
MVT RISCVTargetLowering::getRegisterTypeForCallingConv(LLVMContext &Context,
CallingConv::ID CC,
EVT VT) const {
// Use f32 to pass f16 if it is legal and Zfh/Zfhmin is not enabled.
// We might still end up using a GPR but that will be decided based on ABI.
if (VT == MVT::f16 && Subtarget.hasStdExtFOrZfinx() &&
!Subtarget.hasStdExtZfhminOrZhinxmin())
return MVT::f32;
MVT PartVT = TargetLowering::getRegisterTypeForCallingConv(Context, CC, VT);
return PartVT;
}
unsigned
RISCVTargetLowering::getNumRegisters(LLVMContext &Context, EVT VT,
std::optional<MVT> RegisterVT) const {
// Pair inline assembly operand
if (VT == (Subtarget.is64Bit() ? MVT::i128 : MVT::i64) && RegisterVT &&
*RegisterVT == MVT::Untyped)
return 1;
return TargetLowering::getNumRegisters(Context, VT, RegisterVT);
}
unsigned RISCVTargetLowering::getNumRegistersForCallingConv(LLVMContext &Context,
CallingConv::ID CC,
EVT VT) const {
// Use f32 to pass f16 if it is legal and Zfh/Zfhmin is not enabled.
// We might still end up using a GPR but that will be decided based on ABI.
if (VT == MVT::f16 && Subtarget.hasStdExtFOrZfinx() &&
!Subtarget.hasStdExtZfhminOrZhinxmin())
return 1;
return TargetLowering::getNumRegistersForCallingConv(Context, CC, VT);
}
unsigned RISCVTargetLowering::getVectorTypeBreakdownForCallingConv(
LLVMContext &Context, CallingConv::ID CC, EVT VT, EVT &IntermediateVT,
unsigned &NumIntermediates, MVT &RegisterVT) const {
unsigned NumRegs = TargetLowering::getVectorTypeBreakdownForCallingConv(
Context, CC, VT, IntermediateVT, NumIntermediates, RegisterVT);
return NumRegs;
}
// Changes the condition code and swaps operands if necessary, so the SetCC
// operation matches one of the comparisons supported directly by branches
// in the RISC-V ISA. May adjust compares to favor compare with 0 over compare
// with 1/-1.
static void translateSetCCForBranch(const SDLoc &DL, SDValue &LHS, SDValue &RHS,
ISD::CondCode &CC, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
// If this is a single bit test that can't be handled by ANDI, shift the
// bit to be tested to the MSB and perform a signed compare with 0.
if (isIntEqualitySetCC(CC) && isNullConstant(RHS) &&
LHS.getOpcode() == ISD::AND && LHS.hasOneUse() &&
isa<ConstantSDNode>(LHS.getOperand(1)) &&
// XAndesPerf supports branch on test bit.
!Subtarget.hasVendorXAndesPerf()) {
uint64_t Mask = LHS.getConstantOperandVal(1);
if ((isPowerOf2_64(Mask) || isMask_64(Mask)) && !isInt<12>(Mask)) {
unsigned ShAmt = 0;
if (isPowerOf2_64(Mask)) {
CC = CC == ISD::SETEQ ? ISD::SETGE : ISD::SETLT;
ShAmt = LHS.getValueSizeInBits() - 1 - Log2_64(Mask);
} else {
ShAmt = LHS.getValueSizeInBits() - llvm::bit_width(Mask);
}
LHS = LHS.getOperand(0);
if (ShAmt != 0)
LHS = DAG.getNode(ISD::SHL, DL, LHS.getValueType(), LHS,
DAG.getConstant(ShAmt, DL, LHS.getValueType()));
return;
}
}
if (auto *RHSC = dyn_cast<ConstantSDNode>(RHS)) {
int64_t C = RHSC->getSExtValue();
switch (CC) {
default: break;
case ISD::SETGT:
// Convert X > -1 to X >= 0.
if (C == -1) {
RHS = DAG.getConstant(0, DL, RHS.getValueType());
CC = ISD::SETGE;
return;
}
if (Subtarget.hasVendorXqcibi() && C != INT64_MAX && isInt<16>(C + 1)) {
// We have a branch immediate instruction for SETGE but not SETGT.
// Convert X > C to X >= C + 1, if (C + 1) is a 16-bit signed immediate.
RHS = DAG.getSignedConstant(C + 1, DL, RHS.getValueType());
CC = ISD::SETGE;
return;
}
break;
case ISD::SETLT:
// Convert X < 1 to 0 >= X.
if (C == 1) {
RHS = LHS;
LHS = DAG.getConstant(0, DL, RHS.getValueType());
CC = ISD::SETGE;
return;
}
break;
case ISD::SETUGT:
if (Subtarget.hasVendorXqcibi() && C != INT64_MAX && isUInt<16>(C + 1)) {
// We have a branch immediate instruction for SETUGE but not SETUGT.
// Convert X > C to X >= C + 1, if (C + 1) is a 16-bit unsigned
// immediate.
RHS = DAG.getConstant(C + 1, DL, RHS.getValueType());
CC = ISD::SETUGE;
return;
}
break;
}
}
switch (CC) {
default:
break;
case ISD::SETGT:
case ISD::SETLE:
case ISD::SETUGT:
case ISD::SETULE:
CC = ISD::getSetCCSwappedOperands(CC);
std::swap(LHS, RHS);
break;
}
}
RISCVVType::VLMUL RISCVTargetLowering::getLMUL(MVT VT) {
if (VT.isRISCVVectorTuple()) {
if (VT.SimpleTy >= MVT::riscv_nxv1i8x2 &&
VT.SimpleTy <= MVT::riscv_nxv1i8x8)
return RISCVVType::LMUL_F8;
if (VT.SimpleTy >= MVT::riscv_nxv2i8x2 &&
VT.SimpleTy <= MVT::riscv_nxv2i8x8)
return RISCVVType::LMUL_F4;
if (VT.SimpleTy >= MVT::riscv_nxv4i8x2 &&
VT.SimpleTy <= MVT::riscv_nxv4i8x8)
return RISCVVType::LMUL_F2;
if (VT.SimpleTy >= MVT::riscv_nxv8i8x2 &&
VT.SimpleTy <= MVT::riscv_nxv8i8x8)
return RISCVVType::LMUL_1;
if (VT.SimpleTy >= MVT::riscv_nxv16i8x2 &&
VT.SimpleTy <= MVT::riscv_nxv16i8x4)
return RISCVVType::LMUL_2;
if (VT.SimpleTy == MVT::riscv_nxv32i8x2)
return RISCVVType::LMUL_4;
llvm_unreachable("Invalid vector tuple type LMUL.");
}
assert(VT.isScalableVector() && "Expecting a scalable vector type");
unsigned KnownSize = VT.getSizeInBits().getKnownMinValue();
if (VT.getVectorElementType() == MVT::i1)
KnownSize *= 8;
switch (KnownSize) {
default:
llvm_unreachable("Invalid LMUL.");
case 8:
return RISCVVType::LMUL_F8;
case 16:
return RISCVVType::LMUL_F4;
case 32:
return RISCVVType::LMUL_F2;
case 64:
return RISCVVType::LMUL_1;
case 128:
return RISCVVType::LMUL_2;
case 256:
return RISCVVType::LMUL_4;
case 512:
return RISCVVType::LMUL_8;
}
}
unsigned RISCVTargetLowering::getRegClassIDForLMUL(RISCVVType::VLMUL LMul) {
switch (LMul) {
default:
llvm_unreachable("Invalid LMUL.");
case RISCVVType::LMUL_F8:
case RISCVVType::LMUL_F4:
case RISCVVType::LMUL_F2:
case RISCVVType::LMUL_1:
return RISCV::VRRegClassID;
case RISCVVType::LMUL_2:
return RISCV::VRM2RegClassID;
case RISCVVType::LMUL_4:
return RISCV::VRM4RegClassID;
case RISCVVType::LMUL_8:
return RISCV::VRM8RegClassID;
}
}
unsigned RISCVTargetLowering::getSubregIndexByMVT(MVT VT, unsigned Index) {
RISCVVType::VLMUL LMUL = getLMUL(VT);
if (LMUL == RISCVVType::LMUL_F8 || LMUL == RISCVVType::LMUL_F4 ||
LMUL == RISCVVType::LMUL_F2 || LMUL == RISCVVType::LMUL_1) {
static_assert(RISCV::sub_vrm1_7 == RISCV::sub_vrm1_0 + 7,
"Unexpected subreg numbering");
return RISCV::sub_vrm1_0 + Index;
}
if (LMUL == RISCVVType::LMUL_2) {
static_assert(RISCV::sub_vrm2_3 == RISCV::sub_vrm2_0 + 3,
"Unexpected subreg numbering");
return RISCV::sub_vrm2_0 + Index;
}
if (LMUL == RISCVVType::LMUL_4) {
static_assert(RISCV::sub_vrm4_1 == RISCV::sub_vrm4_0 + 1,
"Unexpected subreg numbering");
return RISCV::sub_vrm4_0 + Index;
}
llvm_unreachable("Invalid vector type.");
}
unsigned RISCVTargetLowering::getRegClassIDForVecVT(MVT VT) {
if (VT.isRISCVVectorTuple()) {
unsigned NF = VT.getRISCVVectorTupleNumFields();
unsigned RegsPerField =
std::max(1U, (unsigned)VT.getSizeInBits().getKnownMinValue() /
(NF * RISCV::RVVBitsPerBlock));
switch (RegsPerField) {
case 1:
if (NF == 2)
return RISCV::VRN2M1RegClassID;
if (NF == 3)
return RISCV::VRN3M1RegClassID;
if (NF == 4)
return RISCV::VRN4M1RegClassID;
if (NF == 5)
return RISCV::VRN5M1RegClassID;
if (NF == 6)
return RISCV::VRN6M1RegClassID;
if (NF == 7)
return RISCV::VRN7M1RegClassID;
if (NF == 8)
return RISCV::VRN8M1RegClassID;
break;
case 2:
if (NF == 2)
return RISCV::VRN2M2RegClassID;
if (NF == 3)
return RISCV::VRN3M2RegClassID;
if (NF == 4)
return RISCV::VRN4M2RegClassID;
break;
case 4:
assert(NF == 2);
return RISCV::VRN2M4RegClassID;
default:
break;
}
llvm_unreachable("Invalid vector tuple type RegClass.");
}
if (VT.getVectorElementType() == MVT::i1)
return RISCV::VRRegClassID;
return getRegClassIDForLMUL(getLMUL(VT));
}
// Attempt to decompose a subvector insert/extract between VecVT and
// SubVecVT via subregister indices. Returns the subregister index that
// can perform the subvector insert/extract with the given element index, as
// well as the index corresponding to any leftover subvectors that must be
// further inserted/extracted within the register class for SubVecVT.
std::pair<unsigned, unsigned>
RISCVTargetLowering::decomposeSubvectorInsertExtractToSubRegs(
MVT VecVT, MVT SubVecVT, unsigned InsertExtractIdx,
const RISCVRegisterInfo *TRI) {
static_assert((RISCV::VRM8RegClassID > RISCV::VRM4RegClassID &&
RISCV::VRM4RegClassID > RISCV::VRM2RegClassID &&
RISCV::VRM2RegClassID > RISCV::VRRegClassID),
"Register classes not ordered");
unsigned VecRegClassID = getRegClassIDForVecVT(VecVT);
unsigned SubRegClassID = getRegClassIDForVecVT(SubVecVT);
// If VecVT is a vector tuple type, either it's the tuple type with same
// RegClass with SubVecVT or SubVecVT is a actually a subvector of the VecVT.
if (VecVT.isRISCVVectorTuple()) {
if (VecRegClassID == SubRegClassID)
return {RISCV::NoSubRegister, 0};
assert(SubVecVT.isScalableVector() &&
"Only allow scalable vector subvector.");
assert(getLMUL(VecVT) == getLMUL(SubVecVT) &&
"Invalid vector tuple insert/extract for vector and subvector with "
"different LMUL.");
return {getSubregIndexByMVT(VecVT, InsertExtractIdx), 0};
}
// Try to compose a subregister index that takes us from the incoming
// LMUL>1 register class down to the outgoing one. At each step we half
// the LMUL:
// nxv16i32@12 -> nxv2i32: sub_vrm4_1_then_sub_vrm2_1_then_sub_vrm1_0
// Note that this is not guaranteed to find a subregister index, such as
// when we are extracting from one VR type to another.
unsigned SubRegIdx = RISCV::NoSubRegister;
for (const unsigned RCID :
{RISCV::VRM4RegClassID, RISCV::VRM2RegClassID, RISCV::VRRegClassID})
if (VecRegClassID > RCID && SubRegClassID <= RCID) {
VecVT = VecVT.getHalfNumVectorElementsVT();
bool IsHi =
InsertExtractIdx >= VecVT.getVectorElementCount().getKnownMinValue();
SubRegIdx = TRI->composeSubRegIndices(SubRegIdx,
getSubregIndexByMVT(VecVT, IsHi));
if (IsHi)
InsertExtractIdx -= VecVT.getVectorElementCount().getKnownMinValue();
}
return {SubRegIdx, InsertExtractIdx};
}
// Permit combining of mask vectors as BUILD_VECTOR never expands to scalar
// stores for those types.
bool RISCVTargetLowering::mergeStoresAfterLegalization(EVT VT) const {
return !Subtarget.useRVVForFixedLengthVectors() ||
(VT.isFixedLengthVector() && VT.getVectorElementType() == MVT::i1);
}
bool RISCVTargetLowering::isLegalElementTypeForRVV(EVT ScalarTy) const {
if (!ScalarTy.isSimple())
return false;
switch (ScalarTy.getSimpleVT().SimpleTy) {
case MVT::iPTR:
return Subtarget.is64Bit() ? Subtarget.hasVInstructionsI64() : true;
case MVT::i8:
case MVT::i16:
case MVT::i32:
return true;
case MVT::i64:
return Subtarget.hasVInstructionsI64();
case MVT::f16:
return Subtarget.hasVInstructionsF16Minimal();
case MVT::bf16:
return Subtarget.hasVInstructionsBF16Minimal();
case MVT::f32:
return Subtarget.hasVInstructionsF32();
case MVT::f64:
return Subtarget.hasVInstructionsF64();
default:
return false;
}
}
unsigned RISCVTargetLowering::combineRepeatedFPDivisors() const {
return NumRepeatedDivisors;
}
static SDValue getVLOperand(SDValue Op) {
assert((Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN ||
Op.getOpcode() == ISD::INTRINSIC_W_CHAIN) &&
"Unexpected opcode");
bool HasChain = Op.getOpcode() == ISD::INTRINSIC_W_CHAIN;
unsigned IntNo = Op.getConstantOperandVal(HasChain ? 1 : 0);
const RISCVVIntrinsicsTable::RISCVVIntrinsicInfo *II =
RISCVVIntrinsicsTable::getRISCVVIntrinsicInfo(IntNo);
if (!II)
return SDValue();
return Op.getOperand(II->VLOperand + 1 + HasChain);
}
static bool useRVVForFixedLengthVectorVT(MVT VT,
const RISCVSubtarget &Subtarget) {
assert(VT.isFixedLengthVector() && "Expected a fixed length vector type!");
if (!Subtarget.useRVVForFixedLengthVectors())
return false;
// We only support a set of vector types with a consistent maximum fixed size
// across all supported vector element types to avoid legalization issues.
// Therefore -- since the largest is v1024i8/v512i16/etc -- the largest
// fixed-length vector type we support is 1024 bytes.
if (VT.getVectorNumElements() > 1024 || VT.getFixedSizeInBits() > 1024 * 8)
return false;
unsigned MinVLen = Subtarget.getRealMinVLen();
MVT EltVT = VT.getVectorElementType();
// Don't use RVV for vectors we cannot scalarize if required.
switch (EltVT.SimpleTy) {
// i1 is supported but has different rules.
default:
return false;
case MVT::i1:
// Masks can only use a single register.
if (VT.getVectorNumElements() > MinVLen)
return false;
MinVLen /= 8;
break;
case MVT::i8:
case MVT::i16:
case MVT::i32:
break;
case MVT::i64:
if (!Subtarget.hasVInstructionsI64())
return false;
break;
case MVT::f16:
if (!Subtarget.hasVInstructionsF16Minimal())
return false;
break;
case MVT::bf16:
if (!Subtarget.hasVInstructionsBF16Minimal())
return false;
break;
case MVT::f32:
if (!Subtarget.hasVInstructionsF32())
return false;
break;
case MVT::f64:
if (!Subtarget.hasVInstructionsF64())
return false;
break;
}
// Reject elements larger than ELEN.
if (EltVT.getSizeInBits() > Subtarget.getELen())
return false;
unsigned LMul = divideCeil(VT.getSizeInBits(), MinVLen);
// Don't use RVV for types that don't fit.
if (LMul > Subtarget.getMaxLMULForFixedLengthVectors())
return false;
// TODO: Perhaps an artificial restriction, but worth having whilst getting
// the base fixed length RVV support in place.
if (!VT.isPow2VectorType())
return false;
return true;
}
bool RISCVTargetLowering::useRVVForFixedLengthVectorVT(MVT VT) const {
return ::useRVVForFixedLengthVectorVT(VT, Subtarget);
}
// Return the largest legal scalable vector type that matches VT's element type.
static MVT getContainerForFixedLengthVector(const TargetLowering &TLI, MVT VT,
const RISCVSubtarget &Subtarget) {
// This may be called before legal types are setup.
assert(((VT.isFixedLengthVector() && TLI.isTypeLegal(VT)) ||
useRVVForFixedLengthVectorVT(VT, Subtarget)) &&
"Expected legal fixed length vector!");
unsigned MinVLen = Subtarget.getRealMinVLen();
unsigned MaxELen = Subtarget.getELen();
MVT EltVT = VT.getVectorElementType();
switch (EltVT.SimpleTy) {
default:
llvm_unreachable("unexpected element type for RVV container");
case MVT::i1:
case MVT::i8:
case MVT::i16:
case MVT::i32:
case MVT::i64:
case MVT::bf16:
case MVT::f16:
case MVT::f32:
case MVT::f64: {
// We prefer to use LMUL=1 for VLEN sized types. Use fractional lmuls for
// narrower types. The smallest fractional LMUL we support is 8/ELEN. Within
// each fractional LMUL we support SEW between 8 and LMUL*ELEN.
unsigned NumElts =
(VT.getVectorNumElements() * RISCV::RVVBitsPerBlock) / MinVLen;
NumElts = std::max(NumElts, RISCV::RVVBitsPerBlock / MaxELen);
assert(isPowerOf2_32(NumElts) && "Expected power of 2 NumElts");
return MVT::getScalableVectorVT(EltVT, NumElts);
}
}
}
static MVT getContainerForFixedLengthVector(SelectionDAG &DAG, MVT VT,
const RISCVSubtarget &Subtarget) {
return getContainerForFixedLengthVector(DAG.getTargetLoweringInfo(), VT,
Subtarget);
}
MVT RISCVTargetLowering::getContainerForFixedLengthVector(MVT VT) const {
return ::getContainerForFixedLengthVector(*this, VT, getSubtarget());
}
// Grow V to consume an entire RVV register.
static SDValue convertToScalableVector(EVT VT, SDValue V, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(VT.isScalableVector() &&
"Expected to convert into a scalable vector!");
assert(V.getValueType().isFixedLengthVector() &&
"Expected a fixed length vector operand!");
SDLoc DL(V);
return DAG.getInsertSubvector(DL, DAG.getUNDEF(VT), V, 0);
}
// Shrink V so it's just big enough to maintain a VT's worth of data.
static SDValue convertFromScalableVector(EVT VT, SDValue V, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(VT.isFixedLengthVector() &&
"Expected to convert into a fixed length vector!");
assert(V.getValueType().isScalableVector() &&
"Expected a scalable vector operand!");
SDLoc DL(V);
return DAG.getExtractSubvector(DL, VT, V, 0);
}
/// Return the type of the mask type suitable for masking the provided
/// vector type. This is simply an i1 element type vector of the same
/// (possibly scalable) length.
static MVT getMaskTypeFor(MVT VecVT) {
assert(VecVT.isVector());
ElementCount EC = VecVT.getVectorElementCount();
return MVT::getVectorVT(MVT::i1, EC);
}
/// Creates an all ones mask suitable for masking a vector of type VecTy with
/// vector length VL. .
static SDValue getAllOnesMask(MVT VecVT, SDValue VL, const SDLoc &DL,
SelectionDAG &DAG) {
MVT MaskVT = getMaskTypeFor(VecVT);
return DAG.getNode(RISCVISD::VMSET_VL, DL, MaskVT, VL);
}
static std::pair<SDValue, SDValue>
getDefaultScalableVLOps(MVT VecVT, const SDLoc &DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(VecVT.isScalableVector() && "Expecting a scalable vector");
SDValue VL = DAG.getRegister(RISCV::X0, Subtarget.getXLenVT());
SDValue Mask = getAllOnesMask(VecVT, VL, DL, DAG);
return {Mask, VL};
}
static std::pair<SDValue, SDValue>
getDefaultVLOps(uint64_t NumElts, MVT ContainerVT, const SDLoc &DL,
SelectionDAG &DAG, const RISCVSubtarget &Subtarget) {
assert(ContainerVT.isScalableVector() && "Expecting scalable container type");
SDValue VL = DAG.getConstant(NumElts, DL, Subtarget.getXLenVT());
SDValue Mask = getAllOnesMask(ContainerVT, VL, DL, DAG);
return {Mask, VL};
}
// Gets the two common "VL" operands: an all-ones mask and the vector length.
// VecVT is a vector type, either fixed-length or scalable, and ContainerVT is
// the vector type that the fixed-length vector is contained in. Otherwise if
// VecVT is scalable, then ContainerVT should be the same as VecVT.
static std::pair<SDValue, SDValue>
getDefaultVLOps(MVT VecVT, MVT ContainerVT, const SDLoc &DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (VecVT.isFixedLengthVector())
return getDefaultVLOps(VecVT.getVectorNumElements(), ContainerVT, DL, DAG,
Subtarget);
assert(ContainerVT.isScalableVector() && "Expecting scalable container type");
return getDefaultScalableVLOps(ContainerVT, DL, DAG, Subtarget);
}
SDValue RISCVTargetLowering::computeVLMax(MVT VecVT, const SDLoc &DL,
SelectionDAG &DAG) const {
assert(VecVT.isScalableVector() && "Expected scalable vector");
return DAG.getElementCount(DL, Subtarget.getXLenVT(),
VecVT.getVectorElementCount());
}
std::pair<unsigned, unsigned>
RISCVTargetLowering::computeVLMAXBounds(MVT VecVT,
const RISCVSubtarget &Subtarget) {
assert(VecVT.isScalableVector() && "Expected scalable vector");
unsigned EltSize = VecVT.getScalarSizeInBits();
unsigned MinSize = VecVT.getSizeInBits().getKnownMinValue();
unsigned VectorBitsMax = Subtarget.getRealMaxVLen();
unsigned MaxVLMAX =
RISCVTargetLowering::computeVLMAX(VectorBitsMax, EltSize, MinSize);
unsigned VectorBitsMin = Subtarget.getRealMinVLen();
unsigned MinVLMAX =
RISCVTargetLowering::computeVLMAX(VectorBitsMin, EltSize, MinSize);
return std::make_pair(MinVLMAX, MaxVLMAX);
}
// The state of RVV BUILD_VECTOR and VECTOR_SHUFFLE lowering is that very few
// of either is (currently) supported. This can get us into an infinite loop
// where we try to lower a BUILD_VECTOR as a VECTOR_SHUFFLE as a BUILD_VECTOR
// as a ..., etc.
// Until either (or both) of these can reliably lower any node, reporting that
// we don't want to expand BUILD_VECTORs via VECTOR_SHUFFLEs at least breaks
// the infinite loop. Note that this lowers BUILD_VECTOR through the stack,
// which is not desirable.
bool RISCVTargetLowering::shouldExpandBuildVectorWithShuffles(
EVT VT, unsigned DefinedValues) const {
return false;
}
InstructionCost RISCVTargetLowering::getLMULCost(MVT VT) const {
// TODO: Here assume reciprocal throughput is 1 for LMUL_1, it is
// implementation-defined.
if (!VT.isVector())
return InstructionCost::getInvalid();
unsigned DLenFactor = Subtarget.getDLenFactor();
unsigned Cost;
if (VT.isScalableVector()) {
unsigned LMul;
bool Fractional;
std::tie(LMul, Fractional) =
RISCVVType::decodeVLMUL(RISCVTargetLowering::getLMUL(VT));
if (Fractional)
Cost = LMul <= DLenFactor ? (DLenFactor / LMul) : 1;
else
Cost = (LMul * DLenFactor);
} else {
Cost = divideCeil(VT.getSizeInBits(), Subtarget.getRealMinVLen() / DLenFactor);
}
return Cost;
}
/// Return the cost of a vrgather.vv instruction for the type VT. vrgather.vv
/// may be quadratic in the number of vreg implied by LMUL, and is assumed to
/// be by default. VRGatherCostModel reflects available options. Note that
/// operand (index and possibly mask) are handled separately.
InstructionCost RISCVTargetLowering::getVRGatherVVCost(MVT VT) const {
auto LMULCost = getLMULCost(VT);
bool Log2CostModel =
Subtarget.getVRGatherCostModel() == llvm::RISCVSubtarget::NLog2N;
if (Log2CostModel && LMULCost.isValid()) {
unsigned Log = Log2_64(LMULCost.getValue());
if (Log > 0)
return LMULCost * Log;
}
return LMULCost * LMULCost;
}
/// Return the cost of a vrgather.vi (or vx) instruction for the type VT.
/// vrgather.vi/vx may be linear in the number of vregs implied by LMUL,
/// or may track the vrgather.vv cost. It is implementation-dependent.
InstructionCost RISCVTargetLowering::getVRGatherVICost(MVT VT) const {
return getLMULCost(VT);
}
/// Return the cost of a vslidedown.vx or vslideup.vx instruction
/// for the type VT. (This does not cover the vslide1up or vslide1down
/// variants.) Slides may be linear in the number of vregs implied by LMUL,
/// or may track the vrgather.vv cost. It is implementation-dependent.
InstructionCost RISCVTargetLowering::getVSlideVXCost(MVT VT) const {
return getLMULCost(VT);
}
/// Return the cost of a vslidedown.vi or vslideup.vi instruction
/// for the type VT. (This does not cover the vslide1up or vslide1down
/// variants.) Slides may be linear in the number of vregs implied by LMUL,
/// or may track the vrgather.vv cost. It is implementation-dependent.
InstructionCost RISCVTargetLowering::getVSlideVICost(MVT VT) const {
return getLMULCost(VT);
}
static SDValue lowerINT_TO_FP(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
// f16 conversions are promoted to f32 when Zfh/Zhinx are not supported.
// bf16 conversions are always promoted to f32.
if ((Op.getValueType() == MVT::f16 && !Subtarget.hasStdExtZfhOrZhinx()) ||
Op.getValueType() == MVT::bf16) {
bool IsStrict = Op->isStrictFPOpcode();
SDLoc DL(Op);
if (IsStrict) {
SDValue Val = DAG.getNode(Op.getOpcode(), DL, {MVT::f32, MVT::Other},
{Op.getOperand(0), Op.getOperand(1)});
return DAG.getNode(ISD::STRICT_FP_ROUND, DL,
{Op.getValueType(), MVT::Other},
{Val.getValue(1), Val.getValue(0),
DAG.getIntPtrConstant(0, DL, /*isTarget=*/true)});
}
return DAG.getNode(
ISD::FP_ROUND, DL, Op.getValueType(),
DAG.getNode(Op.getOpcode(), DL, MVT::f32, Op.getOperand(0)),
DAG.getIntPtrConstant(0, DL, /*isTarget=*/true));
}
// Other operations are legal.
return Op;
}
static SDValue lowerFP_TO_INT_SAT(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
// RISC-V FP-to-int conversions saturate to the destination register size, but
// don't produce 0 for nan. We can use a conversion instruction and fix the
// nan case with a compare and a select.
SDValue Src = Op.getOperand(0);
MVT DstVT = Op.getSimpleValueType();
EVT SatVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
bool IsSigned = Op.getOpcode() == ISD::FP_TO_SINT_SAT;
if (!DstVT.isVector()) {
// For bf16 or for f16 in absence of Zfh, promote to f32, then saturate
// the result.
if ((Src.getValueType() == MVT::f16 && !Subtarget.hasStdExtZfhOrZhinx()) ||
Src.getValueType() == MVT::bf16) {
Src = DAG.getNode(ISD::FP_EXTEND, SDLoc(Op), MVT::f32, Src);
}
unsigned Opc;
if (SatVT == DstVT)
Opc = IsSigned ? RISCVISD::FCVT_X : RISCVISD::FCVT_XU;
else if (DstVT == MVT::i64 && SatVT == MVT::i32)
Opc = IsSigned ? RISCVISD::FCVT_W_RV64 : RISCVISD::FCVT_WU_RV64;
else
return SDValue();
// FIXME: Support other SatVTs by clamping before or after the conversion.
SDLoc DL(Op);
SDValue FpToInt = DAG.getNode(
Opc, DL, DstVT, Src,
DAG.getTargetConstant(RISCVFPRndMode::RTZ, DL, Subtarget.getXLenVT()));
if (Opc == RISCVISD::FCVT_WU_RV64)
FpToInt = DAG.getZeroExtendInReg(FpToInt, DL, MVT::i32);
SDValue ZeroInt = DAG.getConstant(0, DL, DstVT);
return DAG.getSelectCC(DL, Src, Src, ZeroInt, FpToInt,
ISD::CondCode::SETUO);
}
// Vectors.
MVT DstEltVT = DstVT.getVectorElementType();
MVT SrcVT = Src.getSimpleValueType();
MVT SrcEltVT = SrcVT.getVectorElementType();
unsigned SrcEltSize = SrcEltVT.getSizeInBits();
unsigned DstEltSize = DstEltVT.getSizeInBits();
// Only handle saturating to the destination type.
if (SatVT != DstEltVT)
return SDValue();
MVT DstContainerVT = DstVT;
MVT SrcContainerVT = SrcVT;
if (DstVT.isFixedLengthVector()) {
DstContainerVT = getContainerForFixedLengthVector(DAG, DstVT, Subtarget);
SrcContainerVT = getContainerForFixedLengthVector(DAG, SrcVT, Subtarget);
assert(DstContainerVT.getVectorElementCount() ==
SrcContainerVT.getVectorElementCount() &&
"Expected same element count");
Src = convertToScalableVector(SrcContainerVT, Src, DAG, Subtarget);
}
SDLoc DL(Op);
auto [Mask, VL] = getDefaultVLOps(DstVT, DstContainerVT, DL, DAG, Subtarget);
SDValue IsNan = DAG.getNode(RISCVISD::SETCC_VL, DL, Mask.getValueType(),
{Src, Src, DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(Mask.getValueType()), Mask, VL});
// Need to widen by more than 1 step, promote the FP type, then do a widening
// convert.
if (DstEltSize > (2 * SrcEltSize)) {
assert(SrcContainerVT.getVectorElementType() == MVT::f16 && "Unexpected VT!");
MVT InterVT = SrcContainerVT.changeVectorElementType(MVT::f32);
Src = DAG.getNode(RISCVISD::FP_EXTEND_VL, DL, InterVT, Src, Mask, VL);
}
MVT CvtContainerVT = DstContainerVT;
MVT CvtEltVT = DstEltVT;
if (SrcEltSize > (2 * DstEltSize)) {
CvtEltVT = MVT::getIntegerVT(SrcEltVT.getSizeInBits() / 2);
CvtContainerVT = CvtContainerVT.changeVectorElementType(CvtEltVT);
}
unsigned RVVOpc =
IsSigned ? RISCVISD::VFCVT_RTZ_X_F_VL : RISCVISD::VFCVT_RTZ_XU_F_VL;
SDValue Res = DAG.getNode(RVVOpc, DL, CvtContainerVT, Src, Mask, VL);
while (CvtContainerVT != DstContainerVT) {
CvtEltVT = MVT::getIntegerVT(CvtEltVT.getSizeInBits() / 2);
CvtContainerVT = CvtContainerVT.changeVectorElementType(CvtEltVT);
// Rounding mode here is arbitrary since we aren't shifting out any bits.
unsigned ClipOpc = IsSigned ? RISCVISD::TRUNCATE_VECTOR_VL_SSAT
: RISCVISD::TRUNCATE_VECTOR_VL_USAT;
Res = DAG.getNode(ClipOpc, DL, CvtContainerVT, Res, Mask, VL);
}
SDValue SplatZero = DAG.getNode(
RISCVISD::VMV_V_X_VL, DL, DstContainerVT, DAG.getUNDEF(DstContainerVT),
DAG.getConstant(0, DL, Subtarget.getXLenVT()), VL);
Res = DAG.getNode(RISCVISD::VMERGE_VL, DL, DstContainerVT, IsNan, SplatZero,
Res, DAG.getUNDEF(DstContainerVT), VL);
if (DstVT.isFixedLengthVector())
Res = convertFromScalableVector(DstVT, Res, DAG, Subtarget);
return Res;
}
static SDValue lowerFP_TO_INT(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
bool IsStrict = Op->isStrictFPOpcode();
SDValue SrcVal = Op.getOperand(IsStrict ? 1 : 0);
// f16 conversions are promoted to f32 when Zfh/Zhinx is not enabled.
// bf16 conversions are always promoted to f32.
if ((SrcVal.getValueType() == MVT::f16 && !Subtarget.hasStdExtZfhOrZhinx()) ||
SrcVal.getValueType() == MVT::bf16) {
SDLoc DL(Op);
if (IsStrict) {
SDValue Ext =
DAG.getNode(ISD::STRICT_FP_EXTEND, DL, {MVT::f32, MVT::Other},
{Op.getOperand(0), SrcVal});
return DAG.getNode(Op.getOpcode(), DL, {Op.getValueType(), MVT::Other},
{Ext.getValue(1), Ext.getValue(0)});
}
return DAG.getNode(Op.getOpcode(), DL, Op.getValueType(),
DAG.getNode(ISD::FP_EXTEND, DL, MVT::f32, SrcVal));
}
// Other operations are legal.
return Op;
}
static RISCVFPRndMode::RoundingMode matchRoundingOp(unsigned Opc) {
switch (Opc) {
case ISD::FROUNDEVEN:
case ISD::STRICT_FROUNDEVEN:
case ISD::VP_FROUNDEVEN:
return RISCVFPRndMode::RNE;
case ISD::FTRUNC:
case ISD::STRICT_FTRUNC:
case ISD::VP_FROUNDTOZERO:
return RISCVFPRndMode::RTZ;
case ISD::FFLOOR:
case ISD::STRICT_FFLOOR:
case ISD::VP_FFLOOR:
return RISCVFPRndMode::RDN;
case ISD::FCEIL:
case ISD::STRICT_FCEIL:
case ISD::VP_FCEIL:
return RISCVFPRndMode::RUP;
case ISD::FROUND:
case ISD::LROUND:
case ISD::LLROUND:
case ISD::STRICT_FROUND:
case ISD::STRICT_LROUND:
case ISD::STRICT_LLROUND:
case ISD::VP_FROUND:
return RISCVFPRndMode::RMM;
case ISD::FRINT:
case ISD::LRINT:
case ISD::LLRINT:
case ISD::STRICT_FRINT:
case ISD::STRICT_LRINT:
case ISD::STRICT_LLRINT:
case ISD::VP_FRINT:
case ISD::VP_LRINT:
case ISD::VP_LLRINT:
return RISCVFPRndMode::DYN;
}
return RISCVFPRndMode::Invalid;
}
// Expand vector FTRUNC, FCEIL, FFLOOR, FROUND, VP_FCEIL, VP_FFLOOR, VP_FROUND
// VP_FROUNDEVEN, VP_FROUNDTOZERO, VP_FRINT and VP_FNEARBYINT by converting to
// the integer domain and back. Taking care to avoid converting values that are
// nan or already correct.
static SDValue
lowerVectorFTRUNC_FCEIL_FFLOOR_FROUND(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
MVT VT = Op.getSimpleValueType();
assert(VT.isVector() && "Unexpected type");
SDLoc DL(Op);
SDValue Src = Op.getOperand(0);
// Freeze the source since we are increasing the number of uses.
Src = DAG.getFreeze(Src);
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
Src = convertToScalableVector(ContainerVT, Src, DAG, Subtarget);
}
SDValue Mask, VL;
if (Op->isVPOpcode()) {
Mask = Op.getOperand(1);
if (VT.isFixedLengthVector())
Mask = convertToScalableVector(getMaskTypeFor(ContainerVT), Mask, DAG,
Subtarget);
VL = Op.getOperand(2);
} else {
std::tie(Mask, VL) = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
}
// We do the conversion on the absolute value and fix the sign at the end.
SDValue Abs = DAG.getNode(RISCVISD::FABS_VL, DL, ContainerVT, Src, Mask, VL);
// Determine the largest integer that can be represented exactly. This and
// values larger than it don't have any fractional bits so don't need to
// be converted.
const fltSemantics &FltSem = ContainerVT.getFltSemantics();
unsigned Precision = APFloat::semanticsPrecision(FltSem);
APFloat MaxVal = APFloat(FltSem);
MaxVal.convertFromAPInt(APInt::getOneBitSet(Precision, Precision - 1),
/*IsSigned*/ false, APFloat::rmNearestTiesToEven);
SDValue MaxValNode =
DAG.getConstantFP(MaxVal, DL, ContainerVT.getVectorElementType());
SDValue MaxValSplat = DAG.getNode(RISCVISD::VFMV_V_F_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), MaxValNode, VL);
// If abs(Src) was larger than MaxVal or nan, keep it.
MVT SetccVT = MVT::getVectorVT(MVT::i1, ContainerVT.getVectorElementCount());
Mask =
DAG.getNode(RISCVISD::SETCC_VL, DL, SetccVT,
{Abs, MaxValSplat, DAG.getCondCode(ISD::SETOLT),
Mask, Mask, VL});
// Truncate to integer and convert back to FP.
MVT IntVT = ContainerVT.changeVectorElementTypeToInteger();
MVT XLenVT = Subtarget.getXLenVT();
SDValue Truncated;
switch (Op.getOpcode()) {
default:
llvm_unreachable("Unexpected opcode");
case ISD::FRINT:
case ISD::VP_FRINT:
case ISD::FCEIL:
case ISD::VP_FCEIL:
case ISD::FFLOOR:
case ISD::VP_FFLOOR:
case ISD::FROUND:
case ISD::FROUNDEVEN:
case ISD::VP_FROUND:
case ISD::VP_FROUNDEVEN:
case ISD::VP_FROUNDTOZERO: {
RISCVFPRndMode::RoundingMode FRM = matchRoundingOp(Op.getOpcode());
assert(FRM != RISCVFPRndMode::Invalid);
Truncated = DAG.getNode(RISCVISD::VFCVT_RM_X_F_VL, DL, IntVT, Src, Mask,
DAG.getTargetConstant(FRM, DL, XLenVT), VL);
break;
}
case ISD::FTRUNC:
Truncated = DAG.getNode(RISCVISD::VFCVT_RTZ_X_F_VL, DL, IntVT, Src,
Mask, VL);
break;
case ISD::FNEARBYINT:
case ISD::VP_FNEARBYINT:
Truncated = DAG.getNode(RISCVISD::VFROUND_NOEXCEPT_VL, DL, ContainerVT, Src,
Mask, VL);
break;
}
// VFROUND_NOEXCEPT_VL includes SINT_TO_FP_VL.
if (Truncated.getOpcode() != RISCVISD::VFROUND_NOEXCEPT_VL)
Truncated = DAG.getNode(RISCVISD::SINT_TO_FP_VL, DL, ContainerVT, Truncated,
Mask, VL);
// Restore the original sign so that -0.0 is preserved.
Truncated = DAG.getNode(RISCVISD::FCOPYSIGN_VL, DL, ContainerVT, Truncated,
Src, Src, Mask, VL);
if (!VT.isFixedLengthVector())
return Truncated;
return convertFromScalableVector(VT, Truncated, DAG, Subtarget);
}
// Expand vector STRICT_FTRUNC, STRICT_FCEIL, STRICT_FFLOOR, STRICT_FROUND
// STRICT_FROUNDEVEN and STRICT_FNEARBYINT by converting sNan of the source to
// qNan and converting the new source to integer and back to FP.
static SDValue
lowerVectorStrictFTRUNC_FCEIL_FFLOOR_FROUND(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
SDValue Chain = Op.getOperand(0);
SDValue Src = Op.getOperand(1);
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
Src = convertToScalableVector(ContainerVT, Src, DAG, Subtarget);
}
auto [Mask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
// Freeze the source since we are increasing the number of uses.
Src = DAG.getFreeze(Src);
// Convert sNan to qNan by executing x + x for all unordered element x in Src.
MVT MaskVT = Mask.getSimpleValueType();
SDValue Unorder = DAG.getNode(RISCVISD::STRICT_FSETCC_VL, DL,
DAG.getVTList(MaskVT, MVT::Other),
{Chain, Src, Src, DAG.getCondCode(ISD::SETUNE),
DAG.getUNDEF(MaskVT), Mask, VL});
Chain = Unorder.getValue(1);
Src = DAG.getNode(RISCVISD::STRICT_FADD_VL, DL,
DAG.getVTList(ContainerVT, MVT::Other),
{Chain, Src, Src, Src, Unorder, VL});
Chain = Src.getValue(1);
// We do the conversion on the absolute value and fix the sign at the end.
SDValue Abs = DAG.getNode(RISCVISD::FABS_VL, DL, ContainerVT, Src, Mask, VL);
// Determine the largest integer that can be represented exactly. This and
// values larger than it don't have any fractional bits so don't need to
// be converted.
const fltSemantics &FltSem = ContainerVT.getFltSemantics();
unsigned Precision = APFloat::semanticsPrecision(FltSem);
APFloat MaxVal = APFloat(FltSem);
MaxVal.convertFromAPInt(APInt::getOneBitSet(Precision, Precision - 1),
/*IsSigned*/ false, APFloat::rmNearestTiesToEven);
SDValue MaxValNode =
DAG.getConstantFP(MaxVal, DL, ContainerVT.getVectorElementType());
SDValue MaxValSplat = DAG.getNode(RISCVISD::VFMV_V_F_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), MaxValNode, VL);
// If abs(Src) was larger than MaxVal or nan, keep it.
Mask = DAG.getNode(
RISCVISD::SETCC_VL, DL, MaskVT,
{Abs, MaxValSplat, DAG.getCondCode(ISD::SETOLT), Mask, Mask, VL});
// Truncate to integer and convert back to FP.
MVT IntVT = ContainerVT.changeVectorElementTypeToInteger();
MVT XLenVT = Subtarget.getXLenVT();
SDValue Truncated;
switch (Op.getOpcode()) {
default:
llvm_unreachable("Unexpected opcode");
case ISD::STRICT_FCEIL:
case ISD::STRICT_FFLOOR:
case ISD::STRICT_FROUND:
case ISD::STRICT_FROUNDEVEN: {
RISCVFPRndMode::RoundingMode FRM = matchRoundingOp(Op.getOpcode());
assert(FRM != RISCVFPRndMode::Invalid);
Truncated = DAG.getNode(
RISCVISD::STRICT_VFCVT_RM_X_F_VL, DL, DAG.getVTList(IntVT, MVT::Other),
{Chain, Src, Mask, DAG.getTargetConstant(FRM, DL, XLenVT), VL});
break;
}
case ISD::STRICT_FTRUNC:
Truncated =
DAG.getNode(RISCVISD::STRICT_VFCVT_RTZ_X_F_VL, DL,
DAG.getVTList(IntVT, MVT::Other), Chain, Src, Mask, VL);
break;
case ISD::STRICT_FNEARBYINT:
Truncated = DAG.getNode(RISCVISD::STRICT_VFROUND_NOEXCEPT_VL, DL,
DAG.getVTList(ContainerVT, MVT::Other), Chain, Src,
Mask, VL);
break;
}
Chain = Truncated.getValue(1);
// VFROUND_NOEXCEPT_VL includes SINT_TO_FP_VL.
if (Op.getOpcode() != ISD::STRICT_FNEARBYINT) {
Truncated = DAG.getNode(RISCVISD::STRICT_SINT_TO_FP_VL, DL,
DAG.getVTList(ContainerVT, MVT::Other), Chain,
Truncated, Mask, VL);
Chain = Truncated.getValue(1);
}
// Restore the original sign so that -0.0 is preserved.
Truncated = DAG.getNode(RISCVISD::FCOPYSIGN_VL, DL, ContainerVT, Truncated,
Src, Src, Mask, VL);
if (VT.isFixedLengthVector())
Truncated = convertFromScalableVector(VT, Truncated, DAG, Subtarget);
return DAG.getMergeValues({Truncated, Chain}, DL);
}
static SDValue
lowerFTRUNC_FCEIL_FFLOOR_FROUND(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
MVT VT = Op.getSimpleValueType();
if (VT.isVector())
return lowerVectorFTRUNC_FCEIL_FFLOOR_FROUND(Op, DAG, Subtarget);
if (DAG.shouldOptForSize())
return SDValue();
SDLoc DL(Op);
SDValue Src = Op.getOperand(0);
// Create an integer the size of the mantissa with the MSB set. This and all
// values larger than it don't have any fractional bits so don't need to be
// converted.
const fltSemantics &FltSem = VT.getFltSemantics();
unsigned Precision = APFloat::semanticsPrecision(FltSem);
APFloat MaxVal = APFloat(FltSem);
MaxVal.convertFromAPInt(APInt::getOneBitSet(Precision, Precision - 1),
/*IsSigned*/ false, APFloat::rmNearestTiesToEven);
SDValue MaxValNode = DAG.getConstantFP(MaxVal, DL, VT);
RISCVFPRndMode::RoundingMode FRM = matchRoundingOp(Op.getOpcode());
return DAG.getNode(RISCVISD::FROUND, DL, VT, Src, MaxValNode,
DAG.getTargetConstant(FRM, DL, Subtarget.getXLenVT()));
}
// Expand vector [L]LRINT and [L]LROUND by converting to the integer domain.
static SDValue lowerVectorXRINT_XROUND(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDLoc DL(Op);
MVT DstVT = Op.getSimpleValueType();
SDValue Src = Op.getOperand(0);
MVT SrcVT = Src.getSimpleValueType();
assert(SrcVT.isVector() && DstVT.isVector() &&
!(SrcVT.isFixedLengthVector() ^ DstVT.isFixedLengthVector()) &&
"Unexpected type");
MVT DstContainerVT = DstVT;
MVT SrcContainerVT = SrcVT;
if (DstVT.isFixedLengthVector()) {
DstContainerVT = getContainerForFixedLengthVector(DAG, DstVT, Subtarget);
SrcContainerVT = getContainerForFixedLengthVector(DAG, SrcVT, Subtarget);
Src = convertToScalableVector(SrcContainerVT, Src, DAG, Subtarget);
}
auto [Mask, VL] = getDefaultVLOps(SrcVT, SrcContainerVT, DL, DAG, Subtarget);
// [b]f16 -> f32
MVT SrcElemType = SrcVT.getVectorElementType();
if (SrcElemType == MVT::f16 || SrcElemType == MVT::bf16) {
MVT F32VT = SrcContainerVT.changeVectorElementType(MVT::f32);
Src = DAG.getNode(RISCVISD::FP_EXTEND_VL, DL, F32VT, Src, Mask, VL);
}
SDValue Res =
DAG.getNode(RISCVISD::VFCVT_RM_X_F_VL, DL, DstContainerVT, Src, Mask,
DAG.getTargetConstant(matchRoundingOp(Op.getOpcode()), DL,
Subtarget.getXLenVT()),
VL);
if (!DstVT.isFixedLengthVector())
return Res;
return convertFromScalableVector(DstVT, Res, DAG, Subtarget);
}
static SDValue
getVSlidedown(SelectionDAG &DAG, const RISCVSubtarget &Subtarget,
const SDLoc &DL, EVT VT, SDValue Passthru, SDValue Op,
SDValue Offset, SDValue Mask, SDValue VL,
unsigned Policy = RISCVVType::TAIL_UNDISTURBED_MASK_UNDISTURBED) {
if (Passthru.isUndef())
Policy = RISCVVType::TAIL_AGNOSTIC | RISCVVType::MASK_AGNOSTIC;
SDValue PolicyOp = DAG.getTargetConstant(Policy, DL, Subtarget.getXLenVT());
SDValue Ops[] = {Passthru, Op, Offset, Mask, VL, PolicyOp};
return DAG.getNode(RISCVISD::VSLIDEDOWN_VL, DL, VT, Ops);
}
static SDValue
getVSlideup(SelectionDAG &DAG, const RISCVSubtarget &Subtarget, const SDLoc &DL,
EVT VT, SDValue Passthru, SDValue Op, SDValue Offset, SDValue Mask,
SDValue VL,
unsigned Policy = RISCVVType::TAIL_UNDISTURBED_MASK_UNDISTURBED) {
if (Passthru.isUndef())
Policy = RISCVVType::TAIL_AGNOSTIC | RISCVVType::MASK_AGNOSTIC;
SDValue PolicyOp = DAG.getTargetConstant(Policy, DL, Subtarget.getXLenVT());
SDValue Ops[] = {Passthru, Op, Offset, Mask, VL, PolicyOp};
return DAG.getNode(RISCVISD::VSLIDEUP_VL, DL, VT, Ops);
}
struct VIDSequence {
int64_t StepNumerator;
unsigned StepDenominator;
int64_t Addend;
};
static std::optional<APInt> getExactInteger(const APFloat &APF,
uint32_t BitWidth) {
// We will use a SINT_TO_FP to materialize this constant so we should use a
// signed APSInt here.
APSInt ValInt(BitWidth, /*IsUnsigned*/ false);
// We use an arbitrary rounding mode here. If a floating-point is an exact
// integer (e.g., 1.0), the rounding mode does not affect the output value. If
// the rounding mode changes the output value, then it is not an exact
// integer.
RoundingMode ArbitraryRM = RoundingMode::TowardZero;
bool IsExact;
// If it is out of signed integer range, it will return an invalid operation.
// If it is not an exact integer, IsExact is false.
if ((APF.convertToInteger(ValInt, ArbitraryRM, &IsExact) ==
APFloatBase::opInvalidOp) ||
!IsExact)
return std::nullopt;
return ValInt.extractBits(BitWidth, 0);
}
// Try to match an arithmetic-sequence BUILD_VECTOR [X,X+S,X+2*S,...,X+(N-1)*S]
// to the (non-zero) step S and start value X. This can be then lowered as the
// RVV sequence (VID * S) + X, for example.
// The step S is represented as an integer numerator divided by a positive
// denominator. Note that the implementation currently only identifies
// sequences in which either the numerator is +/- 1 or the denominator is 1. It
// cannot detect 2/3, for example.
// Note that this method will also match potentially unappealing index
// sequences, like <i32 0, i32 50939494>, however it is left to the caller to
// determine whether this is worth generating code for.
//
// EltSizeInBits is the size of the type that the sequence will be calculated
// in, i.e. SEW for build_vectors or XLEN for address calculations.
static std::optional<VIDSequence> isSimpleVIDSequence(SDValue Op,
unsigned EltSizeInBits) {
assert(Op.getOpcode() == ISD::BUILD_VECTOR && "Unexpected BUILD_VECTOR");
if (!cast<BuildVectorSDNode>(Op)->isConstant())
return std::nullopt;
bool IsInteger = Op.getValueType().isInteger();
std::optional<unsigned> SeqStepDenom;
std::optional<APInt> SeqStepNum;
std::optional<APInt> SeqAddend;
std::optional<std::pair<APInt, unsigned>> PrevElt;
assert(EltSizeInBits >= Op.getValueType().getScalarSizeInBits());
// First extract the ops into a list of constant integer values. This may not
// be possible for floats if they're not all representable as integers.
SmallVector<std::optional<APInt>> Elts(Op.getNumOperands());
const unsigned OpSize = Op.getScalarValueSizeInBits();
for (auto [Idx, Elt] : enumerate(Op->op_values())) {
if (Elt.isUndef()) {
Elts[Idx] = std::nullopt;
continue;
}
if (IsInteger) {
Elts[Idx] = Elt->getAsAPIntVal().trunc(OpSize).zext(EltSizeInBits);
} else {
auto ExactInteger =
getExactInteger(cast<ConstantFPSDNode>(Elt)->getValueAPF(), OpSize);
if (!ExactInteger)
return std::nullopt;
Elts[Idx] = *ExactInteger;
}
}
for (auto [Idx, Elt] : enumerate(Elts)) {
// Assume undef elements match the sequence; we just have to be careful
// when interpolating across them.
if (!Elt)
continue;
if (PrevElt) {
// Calculate the step since the last non-undef element, and ensure
// it's consistent across the entire sequence.
unsigned IdxDiff = Idx - PrevElt->second;
APInt ValDiff = *Elt - PrevElt->first;
// A zero-value value difference means that we're somewhere in the middle
// of a fractional step, e.g. <0,0,0*,0,1,1,1,1>. Wait until we notice a
// step change before evaluating the sequence.
if (ValDiff == 0)
continue;
int64_t Remainder = ValDiff.srem(IdxDiff);
// Normalize the step if it's greater than 1.
if (Remainder != ValDiff.getSExtValue()) {
// The difference must cleanly divide the element span.
if (Remainder != 0)
return std::nullopt;
ValDiff = ValDiff.sdiv(IdxDiff);
IdxDiff = 1;
}
if (!SeqStepNum)
SeqStepNum = ValDiff;
else if (ValDiff != SeqStepNum)
return std::nullopt;
if (!SeqStepDenom)
SeqStepDenom = IdxDiff;
else if (IdxDiff != *SeqStepDenom)
return std::nullopt;
}
// Record this non-undef element for later.
if (!PrevElt || PrevElt->first != *Elt)
PrevElt = std::make_pair(*Elt, Idx);
}
// We need to have logged a step for this to count as a legal index sequence.
if (!SeqStepNum || !SeqStepDenom)
return std::nullopt;
// Loop back through the sequence and validate elements we might have skipped
// while waiting for a valid step. While doing this, log any sequence addend.
for (auto [Idx, Elt] : enumerate(Elts)) {
if (!Elt)
continue;
APInt ExpectedVal =
(APInt(EltSizeInBits, Idx, /*isSigned=*/false, /*implicitTrunc=*/true) *
*SeqStepNum)
.sdiv(*SeqStepDenom);
APInt Addend = *Elt - ExpectedVal;
if (!SeqAddend)
SeqAddend = Addend;
else if (Addend != SeqAddend)
return std::nullopt;
}
assert(SeqAddend && "Must have an addend if we have a step");
return VIDSequence{SeqStepNum->getSExtValue(), *SeqStepDenom,
SeqAddend->getSExtValue()};
}
// Match a splatted value (SPLAT_VECTOR/BUILD_VECTOR) of an EXTRACT_VECTOR_ELT
// and lower it as a VRGATHER_VX_VL from the source vector.
static SDValue matchSplatAsGather(SDValue SplatVal, MVT VT, const SDLoc &DL,
SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (SplatVal.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
return SDValue();
SDValue Src = SplatVal.getOperand(0);
// Don't perform this optimization for i1 vectors, or if the element types are
// different
// FIXME: Support i1 vectors, maybe by promoting to i8?
MVT EltTy = VT.getVectorElementType();
if (EltTy == MVT::i1 ||
!DAG.getTargetLoweringInfo().isTypeLegal(Src.getValueType()))
return SDValue();
MVT SrcVT = Src.getSimpleValueType();
if (EltTy != SrcVT.getVectorElementType())
return SDValue();
SDValue Idx = SplatVal.getOperand(1);
// The index must be a legal type.
if (Idx.getValueType() != Subtarget.getXLenVT())
return SDValue();
// Check that we know Idx lies within VT
if (!TypeSize::isKnownLE(SrcVT.getSizeInBits(), VT.getSizeInBits())) {
auto *CIdx = dyn_cast<ConstantSDNode>(Idx);
if (!CIdx || CIdx->getZExtValue() >= VT.getVectorMinNumElements())
return SDValue();
}
// Convert fixed length vectors to scalable
MVT ContainerVT = VT;
if (VT.isFixedLengthVector())
ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
MVT SrcContainerVT = SrcVT;
if (SrcVT.isFixedLengthVector()) {
SrcContainerVT = getContainerForFixedLengthVector(DAG, SrcVT, Subtarget);
Src = convertToScalableVector(SrcContainerVT, Src, DAG, Subtarget);
}
// Put Vec in a VT sized vector
if (SrcContainerVT.getVectorMinNumElements() <
ContainerVT.getVectorMinNumElements())
Src = DAG.getInsertSubvector(DL, DAG.getUNDEF(ContainerVT), Src, 0);
else
Src = DAG.getExtractSubvector(DL, ContainerVT, Src, 0);
// We checked that Idx fits inside VT earlier
auto [Mask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
SDValue Gather = DAG.getNode(RISCVISD::VRGATHER_VX_VL, DL, ContainerVT, Src,
Idx, DAG.getUNDEF(ContainerVT), Mask, VL);
if (VT.isFixedLengthVector())
Gather = convertFromScalableVector(VT, Gather, DAG, Subtarget);
return Gather;
}
static SDValue lowerBuildVectorViaVID(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
MVT VT = Op.getSimpleValueType();
assert(VT.isFixedLengthVector() && "Unexpected vector!");
MVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
SDLoc DL(Op);
auto [Mask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
if (auto SimpleVID = isSimpleVIDSequence(Op, Op.getScalarValueSizeInBits())) {
int64_t StepNumerator = SimpleVID->StepNumerator;
unsigned StepDenominator = SimpleVID->StepDenominator;
int64_t Addend = SimpleVID->Addend;
assert(StepNumerator != 0 && "Invalid step");
bool Negate = false;
int64_t SplatStepVal = StepNumerator;
unsigned StepOpcode = ISD::MUL;
// Exclude INT64_MIN to avoid passing it to std::abs. We won't optimize it
// anyway as the shift of 63 won't fit in uimm5.
if (StepNumerator != 1 && StepNumerator != INT64_MIN &&
isPowerOf2_64(std::abs(StepNumerator))) {
Negate = StepNumerator < 0;
StepOpcode = ISD::SHL;
SplatStepVal = Log2_64(std::abs(StepNumerator));
}
// Only emit VIDs with suitably-small steps. We use imm5 as a threshold
// since it's the immediate value many RVV instructions accept. There is
// no vmul.vi instruction so ensure multiply constant can fit in a
// single addi instruction. For the addend, we allow up to 32 bits..
if (((StepOpcode == ISD::MUL && isInt<12>(SplatStepVal)) ||
(StepOpcode == ISD::SHL && isUInt<5>(SplatStepVal))) &&
isPowerOf2_32(StepDenominator) &&
(SplatStepVal >= 0 || StepDenominator == 1) && isInt<32>(Addend)) {
MVT VIDVT =
VT.isFloatingPoint() ? VT.changeVectorElementTypeToInteger() : VT;
MVT VIDContainerVT =
getContainerForFixedLengthVector(DAG, VIDVT, Subtarget);
SDValue VID = DAG.getNode(RISCVISD::VID_VL, DL, VIDContainerVT, Mask, VL);
// Convert right out of the scalable type so we can use standard ISD
// nodes for the rest of the computation. If we used scalable types with
// these, we'd lose the fixed-length vector info and generate worse
// vsetvli code.
VID = convertFromScalableVector(VIDVT, VID, DAG, Subtarget);
if ((StepOpcode == ISD::MUL && SplatStepVal != 1) ||
(StepOpcode == ISD::SHL && SplatStepVal != 0)) {
SDValue SplatStep = DAG.getSignedConstant(SplatStepVal, DL, VIDVT);
VID = DAG.getNode(StepOpcode, DL, VIDVT, VID, SplatStep);
}
if (StepDenominator != 1) {
SDValue SplatStep =
DAG.getConstant(Log2_64(StepDenominator), DL, VIDVT);
VID = DAG.getNode(ISD::SRL, DL, VIDVT, VID, SplatStep);
}
if (Addend != 0 || Negate) {
SDValue SplatAddend = DAG.getSignedConstant(Addend, DL, VIDVT);
VID = DAG.getNode(Negate ? ISD::SUB : ISD::ADD, DL, VIDVT, SplatAddend,
VID);
}
if (VT.isFloatingPoint()) {
// TODO: Use vfwcvt to reduce register pressure.
VID = DAG.getNode(ISD::SINT_TO_FP, DL, VT, VID);
}
return VID;
}
}
return SDValue();
}
/// Try and optimize BUILD_VECTORs with "dominant values" - these are values
/// which constitute a large proportion of the elements. In such cases we can
/// splat a vector with the dominant element and make up the shortfall with
/// INSERT_VECTOR_ELTs. Returns SDValue if not profitable.
/// Note that this includes vectors of 2 elements by association. The
/// upper-most element is the "dominant" one, allowing us to use a splat to
/// "insert" the upper element, and an insert of the lower element at position
/// 0, which improves codegen.
static SDValue lowerBuildVectorViaDominantValues(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
MVT VT = Op.getSimpleValueType();
assert(VT.isFixedLengthVector() && "Unexpected vector!");
MVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
SDLoc DL(Op);
auto [Mask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
MVT XLenVT = Subtarget.getXLenVT();
unsigned NumElts = Op.getNumOperands();
SDValue DominantValue;
unsigned MostCommonCount = 0;
DenseMap<SDValue, unsigned> ValueCounts;
unsigned NumUndefElts =
count_if(Op->op_values(), [](const SDValue &V) { return V.isUndef(); });
// Track the number of scalar loads we know we'd be inserting, estimated as
// any non-zero floating-point constant. Other kinds of element are either
// already in registers or are materialized on demand. The threshold at which
// a vector load is more desirable than several scalar materializion and
// vector-insertion instructions is not known.
unsigned NumScalarLoads = 0;
for (SDValue V : Op->op_values()) {
if (V.isUndef())
continue;
unsigned &Count = ValueCounts[V];
if (0 == Count)
if (auto *CFP = dyn_cast<ConstantFPSDNode>(V))
NumScalarLoads += !CFP->isExactlyValue(+0.0);
// Is this value dominant? In case of a tie, prefer the highest element as
// it's cheaper to insert near the beginning of a vector than it is at the
// end.
if (++Count >= MostCommonCount) {
DominantValue = V;
MostCommonCount = Count;
}
}
assert(DominantValue && "Not expecting an all-undef BUILD_VECTOR");
unsigned NumDefElts = NumElts - NumUndefElts;
unsigned DominantValueCountThreshold = NumDefElts <= 2 ? 0 : NumDefElts - 2;
// Don't perform this optimization when optimizing for size, since
// materializing elements and inserting them tends to cause code bloat.
if (!DAG.shouldOptForSize() && NumScalarLoads < NumElts &&
(NumElts != 2 || ISD::isBuildVectorOfConstantSDNodes(Op.getNode())) &&
((MostCommonCount > DominantValueCountThreshold) ||
(ValueCounts.size() <= Log2_32(NumDefElts)))) {
// Start by splatting the most common element.
SDValue Vec = DAG.getSplatBuildVector(VT, DL, DominantValue);
DenseSet<SDValue> Processed{DominantValue};
// We can handle an insert into the last element (of a splat) via
// v(f)slide1down. This is slightly better than the vslideup insert
// lowering as it avoids the need for a vector group temporary. It
// is also better than using vmerge.vx as it avoids the need to
// materialize the mask in a vector register.
if (SDValue LastOp = Op->getOperand(Op->getNumOperands() - 1);
!LastOp.isUndef() && ValueCounts[LastOp] == 1 &&
LastOp != DominantValue) {
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
auto OpCode =
VT.isFloatingPoint() ? RISCVISD::VFSLIDE1DOWN_VL : RISCVISD::VSLIDE1DOWN_VL;
if (!VT.isFloatingPoint())
LastOp = DAG.getNode(ISD::ANY_EXTEND, DL, XLenVT, LastOp);
Vec = DAG.getNode(OpCode, DL, ContainerVT, DAG.getUNDEF(ContainerVT), Vec,
LastOp, Mask, VL);
Vec = convertFromScalableVector(VT, Vec, DAG, Subtarget);
Processed.insert(LastOp);
}
MVT SelMaskTy = VT.changeVectorElementType(MVT::i1);
for (const auto &OpIdx : enumerate(Op->ops())) {
const SDValue &V = OpIdx.value();
if (V.isUndef() || !Processed.insert(V).second)
continue;
if (ValueCounts[V] == 1) {
Vec = DAG.getInsertVectorElt(DL, Vec, V, OpIdx.index());
} else {
// Blend in all instances of this value using a VSELECT, using a
// mask where each bit signals whether that element is the one
// we're after.
SmallVector<SDValue> Ops;
transform(Op->op_values(), std::back_inserter(Ops), [&](SDValue V1) {
return DAG.getConstant(V == V1, DL, XLenVT);
});
Vec = DAG.getNode(ISD::VSELECT, DL, VT,
DAG.getBuildVector(SelMaskTy, DL, Ops),
DAG.getSplatBuildVector(VT, DL, V), Vec);
}
}
return Vec;
}
return SDValue();
}
static SDValue lowerBuildVectorOfConstants(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
MVT VT = Op.getSimpleValueType();
assert(VT.isFixedLengthVector() && "Unexpected vector!");
MVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
SDLoc DL(Op);
auto [Mask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
MVT XLenVT = Subtarget.getXLenVT();
unsigned NumElts = Op.getNumOperands();
if (VT.getVectorElementType() == MVT::i1) {
if (ISD::isBuildVectorAllZeros(Op.getNode())) {
SDValue VMClr = DAG.getNode(RISCVISD::VMCLR_VL, DL, ContainerVT, VL);
return convertFromScalableVector(VT, VMClr, DAG, Subtarget);
}
if (ISD::isBuildVectorAllOnes(Op.getNode())) {
SDValue VMSet = DAG.getNode(RISCVISD::VMSET_VL, DL, ContainerVT, VL);
return convertFromScalableVector(VT, VMSet, DAG, Subtarget);
}
// Lower constant mask BUILD_VECTORs via an integer vector type, in
// scalar integer chunks whose bit-width depends on the number of mask
// bits and XLEN.
// First, determine the most appropriate scalar integer type to use. This
// is at most XLenVT, but may be shrunk to a smaller vector element type
// according to the size of the final vector - use i8 chunks rather than
// XLenVT if we're producing a v8i1. This results in more consistent
// codegen across RV32 and RV64.
unsigned NumViaIntegerBits = std::clamp(NumElts, 8u, Subtarget.getXLen());
NumViaIntegerBits = std::min(NumViaIntegerBits, Subtarget.getELen());
// If we have to use more than one INSERT_VECTOR_ELT then this
// optimization is likely to increase code size; avoid performing it in
// such a case. We can use a load from a constant pool in this case.
if (DAG.shouldOptForSize() && NumElts > NumViaIntegerBits)
return SDValue();
// Now we can create our integer vector type. Note that it may be larger
// than the resulting mask type: v4i1 would use v1i8 as its integer type.
unsigned IntegerViaVecElts = divideCeil(NumElts, NumViaIntegerBits);
MVT IntegerViaVecVT =
MVT::getVectorVT(MVT::getIntegerVT(NumViaIntegerBits),
IntegerViaVecElts);
uint64_t Bits = 0;
unsigned BitPos = 0, IntegerEltIdx = 0;
SmallVector<SDValue, 8> Elts(IntegerViaVecElts);
for (unsigned I = 0; I < NumElts;) {
SDValue V = Op.getOperand(I);
bool BitValue = !V.isUndef() && V->getAsZExtVal();
Bits |= ((uint64_t)BitValue << BitPos);
++BitPos;
++I;
// Once we accumulate enough bits to fill our scalar type or process the
// last element, insert into our vector and clear our accumulated data.
if (I % NumViaIntegerBits == 0 || I == NumElts) {
if (NumViaIntegerBits <= 32)
Bits = SignExtend64<32>(Bits);
SDValue Elt = DAG.getSignedConstant(Bits, DL, XLenVT);
Elts[IntegerEltIdx] = Elt;
Bits = 0;
BitPos = 0;
IntegerEltIdx++;
}
}
SDValue Vec = DAG.getBuildVector(IntegerViaVecVT, DL, Elts);
if (NumElts < NumViaIntegerBits) {
// If we're producing a smaller vector than our minimum legal integer
// type, bitcast to the equivalent (known-legal) mask type, and extract
// our final mask.
assert(IntegerViaVecVT == MVT::v1i8 && "Unexpected mask vector type");
Vec = DAG.getBitcast(MVT::v8i1, Vec);
Vec = DAG.getExtractSubvector(DL, VT, Vec, 0);
} else {
// Else we must have produced an integer type with the same size as the
// mask type; bitcast for the final result.
assert(VT.getSizeInBits() == IntegerViaVecVT.getSizeInBits());
Vec = DAG.getBitcast(VT, Vec);
}
return Vec;
}
if (SDValue Splat = cast<BuildVectorSDNode>(Op)->getSplatValue()) {
unsigned Opc = VT.isFloatingPoint() ? RISCVISD::VFMV_V_F_VL
: RISCVISD::VMV_V_X_VL;
if (!VT.isFloatingPoint())
Splat = DAG.getNode(ISD::ANY_EXTEND, DL, XLenVT, Splat);
Splat =
DAG.getNode(Opc, DL, ContainerVT, DAG.getUNDEF(ContainerVT), Splat, VL);
return convertFromScalableVector(VT, Splat, DAG, Subtarget);
}
// Try and match index sequences, which we can lower to the vid instruction
// with optional modifications. An all-undef vector is matched by
// getSplatValue, above.
if (SDValue Res = lowerBuildVectorViaVID(Op, DAG, Subtarget))
return Res;
// For very small build_vectors, use a single scalar insert of a constant.
// TODO: Base this on constant rematerialization cost, not size.
const unsigned EltBitSize = VT.getScalarSizeInBits();
if (VT.getSizeInBits() <= 32 &&
ISD::isBuildVectorOfConstantSDNodes(Op.getNode())) {
MVT ViaIntVT = MVT::getIntegerVT(VT.getSizeInBits());
assert((ViaIntVT == MVT::i16 || ViaIntVT == MVT::i32) &&
"Unexpected sequence type");
// If we can use the original VL with the modified element type, this
// means we only have a VTYPE toggle, not a VL toggle. TODO: Should this
// be moved into InsertVSETVLI?
unsigned ViaVecLen =
(Subtarget.getRealMinVLen() >= VT.getSizeInBits() * NumElts) ? NumElts : 1;
MVT ViaVecVT = MVT::getVectorVT(ViaIntVT, ViaVecLen);
uint64_t EltMask = maskTrailingOnes<uint64_t>(EltBitSize);
uint64_t SplatValue = 0;
// Construct the amalgamated value at this larger vector type.
for (const auto &OpIdx : enumerate(Op->op_values())) {
const auto &SeqV = OpIdx.value();
if (!SeqV.isUndef())
SplatValue |=
((SeqV->getAsZExtVal() & EltMask) << (OpIdx.index() * EltBitSize));
}
// On RV64, sign-extend from 32 to 64 bits where possible in order to
// achieve better constant materializion.
// On RV32, we need to sign-extend to use getSignedConstant.
if (ViaIntVT == MVT::i32)
SplatValue = SignExtend64<32>(SplatValue);
SDValue Vec = DAG.getInsertVectorElt(
DL, DAG.getUNDEF(ViaVecVT),
DAG.getSignedConstant(SplatValue, DL, XLenVT), 0);
if (ViaVecLen != 1)
Vec = DAG.getExtractSubvector(DL, MVT::getVectorVT(ViaIntVT, 1), Vec, 0);
return DAG.getBitcast(VT, Vec);
}
// Attempt to detect "hidden" splats, which only reveal themselves as splats
// when re-interpreted as a vector with a larger element type. For example,
// v4i16 = build_vector i16 0, i16 1, i16 0, i16 1
// could be instead splat as
// v2i32 = build_vector i32 0x00010000, i32 0x00010000
// TODO: This optimization could also work on non-constant splats, but it
// would require bit-manipulation instructions to construct the splat value.
SmallVector<SDValue> Sequence;
const auto *BV = cast<BuildVectorSDNode>(Op);
if (VT.isInteger() && EltBitSize < Subtarget.getELen() &&
ISD::isBuildVectorOfConstantSDNodes(Op.getNode()) &&
BV->getRepeatedSequence(Sequence) &&
(Sequence.size() * EltBitSize) <= Subtarget.getELen()) {
unsigned SeqLen = Sequence.size();
MVT ViaIntVT = MVT::getIntegerVT(EltBitSize * SeqLen);
assert((ViaIntVT == MVT::i16 || ViaIntVT == MVT::i32 ||
ViaIntVT == MVT::i64) &&
"Unexpected sequence type");
// If we can use the original VL with the modified element type, this
// means we only have a VTYPE toggle, not a VL toggle. TODO: Should this
// be moved into InsertVSETVLI?
const unsigned RequiredVL = NumElts / SeqLen;
const unsigned ViaVecLen =
(Subtarget.getRealMinVLen() >= ViaIntVT.getSizeInBits() * NumElts) ?
NumElts : RequiredVL;
MVT ViaVecVT = MVT::getVectorVT(ViaIntVT, ViaVecLen);
unsigned EltIdx = 0;
uint64_t EltMask = maskTrailingOnes<uint64_t>(EltBitSize);
uint64_t SplatValue = 0;
// Construct the amalgamated value which can be splatted as this larger
// vector type.
for (const auto &SeqV : Sequence) {
if (!SeqV.isUndef())
SplatValue |=
((SeqV->getAsZExtVal() & EltMask) << (EltIdx * EltBitSize));
EltIdx++;
}
// On RV64, sign-extend from 32 to 64 bits where possible in order to
// achieve better constant materializion.
// On RV32, we need to sign-extend to use getSignedConstant.
if (ViaIntVT == MVT::i32)
SplatValue = SignExtend64<32>(SplatValue);
// Since we can't introduce illegal i64 types at this stage, we can only
// perform an i64 splat on RV32 if it is its own sign-extended value. That
// way we can use RVV instructions to splat.
assert((ViaIntVT.bitsLE(XLenVT) ||
(!Subtarget.is64Bit() && ViaIntVT == MVT::i64)) &&
"Unexpected bitcast sequence");
if (ViaIntVT.bitsLE(XLenVT) || isInt<32>(SplatValue)) {
SDValue ViaVL =
DAG.getConstant(ViaVecVT.getVectorNumElements(), DL, XLenVT);
MVT ViaContainerVT =
getContainerForFixedLengthVector(DAG, ViaVecVT, Subtarget);
SDValue Splat =
DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ViaContainerVT,
DAG.getUNDEF(ViaContainerVT),
DAG.getSignedConstant(SplatValue, DL, XLenVT), ViaVL);
Splat = convertFromScalableVector(ViaVecVT, Splat, DAG, Subtarget);
if (ViaVecLen != RequiredVL)
Splat = DAG.getExtractSubvector(
DL, MVT::getVectorVT(ViaIntVT, RequiredVL), Splat, 0);
return DAG.getBitcast(VT, Splat);
}
}
// If the number of signbits allows, see if we can lower as a <N x i8>.
// Our main goal here is to reduce LMUL (and thus work) required to
// build the constant, but we will also narrow if the resulting
// narrow vector is known to materialize cheaply.
// TODO: We really should be costing the smaller vector. There are
// profitable cases this misses.
if (EltBitSize > 8 && VT.isInteger() &&
(NumElts <= 4 || VT.getSizeInBits() > Subtarget.getRealMinVLen()) &&
DAG.ComputeMaxSignificantBits(Op) <= 8) {
SDValue Source = DAG.getBuildVector(VT.changeVectorElementType(MVT::i8),
DL, Op->ops());
Source = convertToScalableVector(ContainerVT.changeVectorElementType(MVT::i8),
Source, DAG, Subtarget);
SDValue Res = DAG.getNode(RISCVISD::VSEXT_VL, DL, ContainerVT, Source, Mask, VL);
return convertFromScalableVector(VT, Res, DAG, Subtarget);
}
if (SDValue Res = lowerBuildVectorViaDominantValues(Op, DAG, Subtarget))
return Res;
// For constant vectors, use generic constant pool lowering. Otherwise,
// we'd have to materialize constants in GPRs just to move them into the
// vector.
return SDValue();
}
static unsigned getPACKOpcode(unsigned DestBW,
const RISCVSubtarget &Subtarget) {
switch (DestBW) {
default:
llvm_unreachable("Unsupported pack size");
case 16:
return RISCV::PACKH;
case 32:
return Subtarget.is64Bit() ? RISCV::PACKW : RISCV::PACK;
case 64:
assert(Subtarget.is64Bit());
return RISCV::PACK;
}
}
/// Double the element size of the build vector to reduce the number
/// of vslide1down in the build vector chain. In the worst case, this
/// trades three scalar operations for 1 vector operation. Scalar
/// operations are generally lower latency, and for out-of-order cores
/// we also benefit from additional parallelism.
static SDValue lowerBuildVectorViaPacking(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
assert(VT.isFixedLengthVector() && "Unexpected vector!");
MVT ElemVT = VT.getVectorElementType();
if (!ElemVT.isInteger())
return SDValue();
// TODO: Relax these architectural restrictions, possibly with costing
// of the actual instructions required.
if (!Subtarget.hasStdExtZbb() || !Subtarget.hasStdExtZba())
return SDValue();
unsigned NumElts = VT.getVectorNumElements();
unsigned ElemSizeInBits = ElemVT.getSizeInBits();
if (ElemSizeInBits >= std::min(Subtarget.getELen(), Subtarget.getXLen()) ||
NumElts % 2 != 0)
return SDValue();
// Produce [B,A] packed into a type twice as wide. Note that all
// scalars are XLenVT, possibly masked (see below).
MVT XLenVT = Subtarget.getXLenVT();
SDValue Mask = DAG.getConstant(
APInt::getLowBitsSet(XLenVT.getSizeInBits(), ElemSizeInBits), DL, XLenVT);
auto pack = [&](SDValue A, SDValue B) {
// Bias the scheduling of the inserted operations to near the
// definition of the element - this tends to reduce register
// pressure overall.
SDLoc ElemDL(B);
if (Subtarget.hasStdExtZbkb())
// Note that we're relying on the high bits of the result being
// don't care. For PACKW, the result is *sign* extended.
return SDValue(
DAG.getMachineNode(getPACKOpcode(ElemSizeInBits * 2, Subtarget),
ElemDL, XLenVT, A, B),
0);
A = DAG.getNode(ISD::AND, SDLoc(A), XLenVT, A, Mask);
B = DAG.getNode(ISD::AND, SDLoc(B), XLenVT, B, Mask);
SDValue ShtAmt = DAG.getConstant(ElemSizeInBits, ElemDL, XLenVT);
return DAG.getNode(ISD::OR, ElemDL, XLenVT, A,
DAG.getNode(ISD::SHL, ElemDL, XLenVT, B, ShtAmt),
SDNodeFlags::Disjoint);
};
SmallVector<SDValue> NewOperands;
NewOperands.reserve(NumElts / 2);
for (unsigned i = 0; i < VT.getVectorNumElements(); i += 2)
NewOperands.push_back(pack(Op.getOperand(i), Op.getOperand(i + 1)));
assert(NumElts == NewOperands.size() * 2);
MVT WideVT = MVT::getIntegerVT(ElemSizeInBits * 2);
MVT WideVecVT = MVT::getVectorVT(WideVT, NumElts / 2);
return DAG.getNode(ISD::BITCAST, DL, VT,
DAG.getBuildVector(WideVecVT, DL, NewOperands));
}
static SDValue lowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
MVT VT = Op.getSimpleValueType();
assert(VT.isFixedLengthVector() && "Unexpected vector!");
MVT EltVT = VT.getVectorElementType();
MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(Op);
// Proper support for f16 requires Zvfh. bf16 always requires special
// handling. We need to cast the scalar to integer and create an integer
// build_vector.
if ((EltVT == MVT::f16 && !Subtarget.hasStdExtZvfh()) || EltVT == MVT::bf16) {
MVT IVT = VT.changeVectorElementType(MVT::i16);
SmallVector<SDValue, 16> NewOps(Op.getNumOperands());
for (const auto &[I, U] : enumerate(Op->ops())) {
SDValue Elem = U.get();
if ((EltVT == MVT::bf16 && Subtarget.hasStdExtZfbfmin()) ||
(EltVT == MVT::f16 && Subtarget.hasStdExtZfhmin())) {
// Called by LegalizeDAG, we need to use XLenVT operations since we
// can't create illegal types.
if (auto *C = dyn_cast<ConstantFPSDNode>(Elem)) {
// Manually constant fold so the integer build_vector can be lowered
// better. Waiting for DAGCombine will be too late.
APInt V =
C->getValueAPF().bitcastToAPInt().sext(XLenVT.getSizeInBits());
NewOps[I] = DAG.getConstant(V, DL, XLenVT);
} else {
NewOps[I] = DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, XLenVT, Elem);
}
} else {
// Called by scalar type legalizer, we can use i16.
NewOps[I] = DAG.getBitcast(MVT::i16, Op.getOperand(I));
}
}
SDValue Res = DAG.getNode(ISD::BUILD_VECTOR, DL, IVT, NewOps);
return DAG.getBitcast(VT, Res);
}
if (ISD::isBuildVectorOfConstantSDNodes(Op.getNode()) ||
ISD::isBuildVectorOfConstantFPSDNodes(Op.getNode()))
return lowerBuildVectorOfConstants(Op, DAG, Subtarget);
MVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
auto [Mask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
if (VT.getVectorElementType() == MVT::i1) {
// A BUILD_VECTOR can be lowered as a SETCC. For each fixed-length mask
// vector type, we have a legal equivalently-sized i8 type, so we can use
// that.
MVT WideVecVT = VT.changeVectorElementType(MVT::i8);
SDValue VecZero = DAG.getConstant(0, DL, WideVecVT);
SDValue WideVec;
if (SDValue Splat = cast<BuildVectorSDNode>(Op)->getSplatValue()) {
// For a splat, perform a scalar truncate before creating the wider
// vector.
Splat = DAG.getNode(ISD::AND, DL, Splat.getValueType(), Splat,
DAG.getConstant(1, DL, Splat.getValueType()));
WideVec = DAG.getSplatBuildVector(WideVecVT, DL, Splat);
} else {
SmallVector<SDValue, 8> Ops(Op->op_values());
WideVec = DAG.getBuildVector(WideVecVT, DL, Ops);
SDValue VecOne = DAG.getConstant(1, DL, WideVecVT);
WideVec = DAG.getNode(ISD::AND, DL, WideVecVT, WideVec, VecOne);
}
return DAG.getSetCC(DL, VT, WideVec, VecZero, ISD::SETNE);
}
if (SDValue Splat = cast<BuildVectorSDNode>(Op)->getSplatValue()) {
if (auto Gather = matchSplatAsGather(Splat, VT, DL, DAG, Subtarget))
return Gather;
// Prefer vmv.s.x/vfmv.s.f if legal to reduce work and register
// pressure at high LMUL.
if (all_of(Op->ops().drop_front(),
[](const SDUse &U) { return U.get().isUndef(); })) {
unsigned Opc =
VT.isFloatingPoint() ? RISCVISD::VFMV_S_F_VL : RISCVISD::VMV_S_X_VL;
if (!VT.isFloatingPoint())
Splat = DAG.getNode(ISD::ANY_EXTEND, DL, XLenVT, Splat);
Splat = DAG.getNode(Opc, DL, ContainerVT, DAG.getUNDEF(ContainerVT),
Splat, VL);
return convertFromScalableVector(VT, Splat, DAG, Subtarget);
}
unsigned Opc =
VT.isFloatingPoint() ? RISCVISD::VFMV_V_F_VL : RISCVISD::VMV_V_X_VL;
if (!VT.isFloatingPoint())
Splat = DAG.getNode(ISD::ANY_EXTEND, DL, XLenVT, Splat);
Splat =
DAG.getNode(Opc, DL, ContainerVT, DAG.getUNDEF(ContainerVT), Splat, VL);
return convertFromScalableVector(VT, Splat, DAG, Subtarget);
}
if (SDValue Res = lowerBuildVectorViaDominantValues(Op, DAG, Subtarget))
return Res;
// If we're compiling for an exact VLEN value, we can split our work per
// register in the register group.
if (const auto VLen = Subtarget.getRealVLen();
VLen && VT.getSizeInBits().getKnownMinValue() > *VLen) {
MVT ElemVT = VT.getVectorElementType();
unsigned ElemsPerVReg = *VLen / ElemVT.getFixedSizeInBits();
EVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
MVT OneRegVT = MVT::getVectorVT(ElemVT, ElemsPerVReg);
MVT M1VT = getContainerForFixedLengthVector(DAG, OneRegVT, Subtarget);
assert(M1VT == RISCVTargetLowering::getM1VT(M1VT));
// The following semantically builds up a fixed length concat_vector
// of the component build_vectors. We eagerly lower to scalable and
// insert_subvector here to avoid DAG combining it back to a large
// build_vector.
SmallVector<SDValue> BuildVectorOps(Op->ops());
unsigned NumOpElts = M1VT.getVectorMinNumElements();
SDValue Vec = DAG.getUNDEF(ContainerVT);
for (unsigned i = 0; i < VT.getVectorNumElements(); i += ElemsPerVReg) {
auto OneVRegOfOps = ArrayRef(BuildVectorOps).slice(i, ElemsPerVReg);
SDValue SubBV =
DAG.getNode(ISD::BUILD_VECTOR, DL, OneRegVT, OneVRegOfOps);
SubBV = convertToScalableVector(M1VT, SubBV, DAG, Subtarget);
unsigned InsertIdx = (i / ElemsPerVReg) * NumOpElts;
Vec = DAG.getInsertSubvector(DL, Vec, SubBV, InsertIdx);
}
return convertFromScalableVector(VT, Vec, DAG, Subtarget);
}
// If we're about to resort to vslide1down (or stack usage), pack our
// elements into the widest scalar type we can. This will force a VL/VTYPE
// toggle, but reduces the critical path, the number of vslide1down ops
// required, and possibly enables scalar folds of the values.
if (SDValue Res = lowerBuildVectorViaPacking(Op, DAG, Subtarget))
return Res;
// For m1 vectors, if we have non-undef values in both halves of our vector,
// split the vector into low and high halves, build them separately, then
// use a vselect to combine them. For long vectors, this cuts the critical
// path of the vslide1down sequence in half, and gives us an opportunity
// to special case each half independently. Note that we don't change the
// length of the sub-vectors here, so if both fallback to the generic
// vslide1down path, we should be able to fold the vselect into the final
// vslidedown (for the undef tail) for the first half w/ masking.
unsigned NumElts = VT.getVectorNumElements();
unsigned NumUndefElts =
count_if(Op->op_values(), [](const SDValue &V) { return V.isUndef(); });
unsigned NumDefElts = NumElts - NumUndefElts;
if (NumDefElts >= 8 && NumDefElts > NumElts / 2 &&
ContainerVT.bitsLE(RISCVTargetLowering::getM1VT(ContainerVT))) {
SmallVector<SDValue> SubVecAOps, SubVecBOps;
SmallVector<SDValue> MaskVals;
SDValue UndefElem = DAG.getUNDEF(Op->getOperand(0)->getValueType(0));
SubVecAOps.reserve(NumElts);
SubVecBOps.reserve(NumElts);
for (const auto &[Idx, U] : enumerate(Op->ops())) {
SDValue Elem = U.get();
if (Idx < NumElts / 2) {
SubVecAOps.push_back(Elem);
SubVecBOps.push_back(UndefElem);
} else {
SubVecAOps.push_back(UndefElem);
SubVecBOps.push_back(Elem);
}
bool SelectMaskVal = (Idx < NumElts / 2);
MaskVals.push_back(DAG.getConstant(SelectMaskVal, DL, XLenVT));
}
assert(SubVecAOps.size() == NumElts && SubVecBOps.size() == NumElts &&
MaskVals.size() == NumElts);
SDValue SubVecA = DAG.getBuildVector(VT, DL, SubVecAOps);
SDValue SubVecB = DAG.getBuildVector(VT, DL, SubVecBOps);
MVT MaskVT = MVT::getVectorVT(MVT::i1, NumElts);
SDValue SelectMask = DAG.getBuildVector(MaskVT, DL, MaskVals);
return DAG.getNode(ISD::VSELECT, DL, VT, SelectMask, SubVecA, SubVecB);
}
// Cap the cost at a value linear to the number of elements in the vector.
// The default lowering is to use the stack. The vector store + scalar loads
// is linear in VL. However, at high lmuls vslide1down and vslidedown end up
// being (at least) linear in LMUL. As a result, using the vslidedown
// lowering for every element ends up being VL*LMUL..
// TODO: Should we be directly costing the stack alternative? Doing so might
// give us a more accurate upper bound.
InstructionCost LinearBudget = VT.getVectorNumElements() * 2;
// TODO: unify with TTI getSlideCost.
InstructionCost PerSlideCost = 1;
switch (RISCVTargetLowering::getLMUL(ContainerVT)) {
default: break;
case RISCVVType::LMUL_2:
PerSlideCost = 2;
break;
case RISCVVType::LMUL_4:
PerSlideCost = 4;
break;
case RISCVVType::LMUL_8:
PerSlideCost = 8;
break;
}
// TODO: Should we be using the build instseq then cost + evaluate scheme
// we use for integer constants here?
unsigned UndefCount = 0;
for (const SDValue &V : Op->ops()) {
if (V.isUndef()) {
UndefCount++;
continue;
}
if (UndefCount) {
LinearBudget -= PerSlideCost;
UndefCount = 0;
}
LinearBudget -= PerSlideCost;
}
if (UndefCount) {
LinearBudget -= PerSlideCost;
}
if (LinearBudget < 0)
return SDValue();
assert((!VT.isFloatingPoint() ||
VT.getVectorElementType().getSizeInBits() <= Subtarget.getFLen()) &&
"Illegal type which will result in reserved encoding");
const unsigned Policy = RISCVVType::TAIL_AGNOSTIC | RISCVVType::MASK_AGNOSTIC;
// General case: splat the first operand and slide other operands down one
// by one to form a vector. Alternatively, if every operand is an
// extraction from element 0 of a vector, we use that vector from the last
// extraction as the start value and slide up instead of slide down. Such that
// (1) we can avoid the initial splat (2) we can turn those vslide1up into
// vslideup of 1 later and eliminate the vector to scalar movement, which is
// something we cannot do with vslide1down/vslidedown.
// Of course, using vslide1up/vslideup might increase the register pressure,
// and that's why we conservatively limit to cases where every operand is an
// extraction from the first element.
SmallVector<SDValue> Operands(Op->op_begin(), Op->op_end());
SDValue EVec;
bool SlideUp = false;
auto getVSlide = [&](EVT ContainerVT, SDValue Passthru, SDValue Vec,
SDValue Offset, SDValue Mask, SDValue VL) -> SDValue {
if (SlideUp)
return getVSlideup(DAG, Subtarget, DL, ContainerVT, Passthru, Vec, Offset,
Mask, VL, Policy);
return getVSlidedown(DAG, Subtarget, DL, ContainerVT, Passthru, Vec, Offset,
Mask, VL, Policy);
};
// The reason we don't use all_of here is because we're also capturing EVec
// from the last non-undef operand. If the std::execution_policy of the
// underlying std::all_of is anything but std::sequenced_policy we might
// capture the wrong EVec.
for (SDValue V : Operands) {
using namespace SDPatternMatch;
SlideUp = V.isUndef() || sd_match(V, m_ExtractElt(m_Value(EVec), m_Zero()));
if (!SlideUp)
break;
}
if (SlideUp) {
MVT EVecContainerVT = EVec.getSimpleValueType();
// Make sure the original vector has scalable vector type.
if (EVecContainerVT.isFixedLengthVector()) {
EVecContainerVT =
getContainerForFixedLengthVector(DAG, EVecContainerVT, Subtarget);
EVec = convertToScalableVector(EVecContainerVT, EVec, DAG, Subtarget);
}
// Adapt EVec's type into ContainerVT.
if (EVecContainerVT.getVectorMinNumElements() <
ContainerVT.getVectorMinNumElements())
EVec = DAG.getInsertSubvector(DL, DAG.getUNDEF(ContainerVT), EVec, 0);
else
EVec = DAG.getExtractSubvector(DL, ContainerVT, EVec, 0);
// Reverse the elements as we're going to slide up from the last element.
std::reverse(Operands.begin(), Operands.end());
}
SDValue Vec;
UndefCount = 0;
for (SDValue V : Operands) {
if (V.isUndef()) {
UndefCount++;
continue;
}
// Start our sequence with either a TA splat or extract source in the
// hopes that hardware is able to recognize there's no dependency on the
// prior value of our temporary register.
if (!Vec) {
if (SlideUp) {
Vec = EVec;
} else {
Vec = DAG.getSplatVector(VT, DL, V);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
UndefCount = 0;
continue;
}
if (UndefCount) {
const SDValue Offset = DAG.getConstant(UndefCount, DL, Subtarget.getXLenVT());
Vec = getVSlide(ContainerVT, DAG.getUNDEF(ContainerVT), Vec, Offset, Mask,
VL);
UndefCount = 0;
}
unsigned Opcode;
if (VT.isFloatingPoint())
Opcode = SlideUp ? RISCVISD::VFSLIDE1UP_VL : RISCVISD::VFSLIDE1DOWN_VL;
else
Opcode = SlideUp ? RISCVISD::VSLIDE1UP_VL : RISCVISD::VSLIDE1DOWN_VL;
if (!VT.isFloatingPoint())
V = DAG.getNode(ISD::ANY_EXTEND, DL, Subtarget.getXLenVT(), V);
Vec = DAG.getNode(Opcode, DL, ContainerVT, DAG.getUNDEF(ContainerVT), Vec,
V, Mask, VL);
}
if (UndefCount) {
const SDValue Offset = DAG.getConstant(UndefCount, DL, Subtarget.getXLenVT());
Vec = getVSlide(ContainerVT, DAG.getUNDEF(ContainerVT), Vec, Offset, Mask,
VL);
}
return convertFromScalableVector(VT, Vec, DAG, Subtarget);
}
static SDValue splatPartsI64WithVL(const SDLoc &DL, MVT VT, SDValue Passthru,
SDValue Lo, SDValue Hi, SDValue VL,
SelectionDAG &DAG) {
if (!Passthru)
Passthru = DAG.getUNDEF(VT);
if (isa<ConstantSDNode>(Lo) && isa<ConstantSDNode>(Hi)) {
int32_t LoC = cast<ConstantSDNode>(Lo)->getSExtValue();
int32_t HiC = cast<ConstantSDNode>(Hi)->getSExtValue();
// If Hi constant is all the same sign bit as Lo, lower this as a custom
// node in order to try and match RVV vector/scalar instructions.
if ((LoC >> 31) == HiC)
return DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, Passthru, Lo, VL);
// Use vmv.v.x with EEW=32. Use either a vsetivli or vsetvli to change
// VL. This can temporarily increase VL if VL less than VLMAX.
if (LoC == HiC) {
SDValue NewVL;
if (isa<ConstantSDNode>(VL) && isUInt<4>(VL->getAsZExtVal()))
NewVL = DAG.getNode(ISD::ADD, DL, VL.getValueType(), VL, VL);
else
NewVL = DAG.getRegister(RISCV::X0, MVT::i32);
MVT InterVT =
MVT::getVectorVT(MVT::i32, VT.getVectorElementCount() * 2);
auto InterVec = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, InterVT,
DAG.getUNDEF(InterVT), Lo, NewVL);
return DAG.getNode(ISD::BITCAST, DL, VT, InterVec);
}
}
// Detect cases where Hi is (SRA Lo, 31) which means Hi is Lo sign extended.
if (Hi.getOpcode() == ISD::SRA && Hi.getOperand(0) == Lo &&
isa<ConstantSDNode>(Hi.getOperand(1)) &&
Hi.getConstantOperandVal(1) == 31)
return DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, Passthru, Lo, VL);
// If the hi bits of the splat are undefined, then it's fine to just splat Lo
// even if it might be sign extended.
if (Hi.isUndef())
return DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, Passthru, Lo, VL);
// Fall back to a stack store and stride x0 vector load.
return DAG.getNode(RISCVISD::SPLAT_VECTOR_SPLIT_I64_VL, DL, VT, Passthru, Lo,
Hi, VL);
}
// Called by type legalization to handle splat of i64 on RV32.
// FIXME: We can optimize this when the type has sign or zero bits in one
// of the halves.
static SDValue splatSplitI64WithVL(const SDLoc &DL, MVT VT, SDValue Passthru,
SDValue Scalar, SDValue VL,
SelectionDAG &DAG) {
assert(Scalar.getValueType() == MVT::i64 && "Unexpected VT!");
SDValue Lo, Hi;
std::tie(Lo, Hi) = DAG.SplitScalar(Scalar, DL, MVT::i32, MVT::i32);
return splatPartsI64WithVL(DL, VT, Passthru, Lo, Hi, VL, DAG);
}
// This function lowers a splat of a scalar operand Splat with the vector
// length VL. It ensures the final sequence is type legal, which is useful when
// lowering a splat after type legalization.
static SDValue lowerScalarSplat(SDValue Passthru, SDValue Scalar, SDValue VL,
MVT VT, const SDLoc &DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
bool HasPassthru = Passthru && !Passthru.isUndef();
if (!HasPassthru && !Passthru)
Passthru = DAG.getUNDEF(VT);
MVT EltVT = VT.getVectorElementType();
MVT XLenVT = Subtarget.getXLenVT();
if (VT.isFloatingPoint()) {
if ((EltVT == MVT::f16 && !Subtarget.hasStdExtZvfh()) ||
EltVT == MVT::bf16) {
if ((EltVT == MVT::bf16 && Subtarget.hasStdExtZfbfmin()) ||
(EltVT == MVT::f16 && Subtarget.hasStdExtZfhmin()))
Scalar = DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, XLenVT, Scalar);
else
Scalar = DAG.getNode(ISD::BITCAST, DL, MVT::i16, Scalar);
MVT IVT = VT.changeVectorElementType(MVT::i16);
Passthru = DAG.getNode(ISD::BITCAST, DL, IVT, Passthru);
SDValue Splat =
lowerScalarSplat(Passthru, Scalar, VL, IVT, DL, DAG, Subtarget);
return DAG.getNode(ISD::BITCAST, DL, VT, Splat);
}
return DAG.getNode(RISCVISD::VFMV_V_F_VL, DL, VT, Passthru, Scalar, VL);
}
// Simplest case is that the operand needs to be promoted to XLenVT.
if (Scalar.getValueType().bitsLE(XLenVT)) {
// If the operand is a constant, sign extend to increase our chances
// of being able to use a .vi instruction. ANY_EXTEND would become a
// a zero extend and the simm5 check in isel would fail.
// FIXME: Should we ignore the upper bits in isel instead?
unsigned ExtOpc =
isa<ConstantSDNode>(Scalar) ? ISD::SIGN_EXTEND : ISD::ANY_EXTEND;
Scalar = DAG.getNode(ExtOpc, DL, XLenVT, Scalar);
return DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, Passthru, Scalar, VL);
}
assert(XLenVT == MVT::i32 && Scalar.getValueType() == MVT::i64 &&
"Unexpected scalar for splat lowering!");
if (isOneConstant(VL) && isNullConstant(Scalar))
return DAG.getNode(RISCVISD::VMV_S_X_VL, DL, VT, Passthru,
DAG.getConstant(0, DL, XLenVT), VL);
// Otherwise use the more complicated splatting algorithm.
return splatSplitI64WithVL(DL, VT, Passthru, Scalar, VL, DAG);
}
// This function lowers an insert of a scalar operand Scalar into lane
// 0 of the vector regardless of the value of VL. The contents of the
// remaining lanes of the result vector are unspecified. VL is assumed
// to be non-zero.
static SDValue lowerScalarInsert(SDValue Scalar, SDValue VL, MVT VT,
const SDLoc &DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(VT.isScalableVector() && "Expect VT is scalable vector type.");
const MVT XLenVT = Subtarget.getXLenVT();
SDValue Passthru = DAG.getUNDEF(VT);
if (Scalar.getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
isNullConstant(Scalar.getOperand(1))) {
SDValue ExtractedVal = Scalar.getOperand(0);
// The element types must be the same.
if (ExtractedVal.getValueType().getVectorElementType() ==
VT.getVectorElementType()) {
MVT ExtractedVT = ExtractedVal.getSimpleValueType();
MVT ExtractedContainerVT = ExtractedVT;
if (ExtractedContainerVT.isFixedLengthVector()) {
ExtractedContainerVT = getContainerForFixedLengthVector(
DAG, ExtractedContainerVT, Subtarget);
ExtractedVal = convertToScalableVector(ExtractedContainerVT,
ExtractedVal, DAG, Subtarget);
}
if (ExtractedContainerVT.bitsLE(VT))
return DAG.getInsertSubvector(DL, Passthru, ExtractedVal, 0);
return DAG.getExtractSubvector(DL, VT, ExtractedVal, 0);
}
}
if (VT.isFloatingPoint())
return DAG.getNode(RISCVISD::VFMV_S_F_VL, DL, VT, DAG.getUNDEF(VT), Scalar,
VL);
// Avoid the tricky legalization cases by falling back to using the
// splat code which already handles it gracefully.
if (!Scalar.getValueType().bitsLE(XLenVT))
return lowerScalarSplat(DAG.getUNDEF(VT), Scalar,
DAG.getConstant(1, DL, XLenVT),
VT, DL, DAG, Subtarget);
// If the operand is a constant, sign extend to increase our chances
// of being able to use a .vi instruction. ANY_EXTEND would become a
// a zero extend and the simm5 check in isel would fail.
// FIXME: Should we ignore the upper bits in isel instead?
unsigned ExtOpc =
isa<ConstantSDNode>(Scalar) ? ISD::SIGN_EXTEND : ISD::ANY_EXTEND;
Scalar = DAG.getNode(ExtOpc, DL, XLenVT, Scalar);
return DAG.getNode(RISCVISD::VMV_S_X_VL, DL, VT, DAG.getUNDEF(VT), Scalar,
VL);
}
/// If concat_vector(V1,V2) could be folded away to some existing
/// vector source, return it. Note that the source may be larger
/// than the requested concat_vector (i.e. a extract_subvector
/// might be required.)
static SDValue foldConcatVector(SDValue V1, SDValue V2) {
EVT VT = V1.getValueType();
assert(VT == V2.getValueType() && "argument types must match");
// Both input must be extracts.
if (V1.getOpcode() != ISD::EXTRACT_SUBVECTOR ||
V2.getOpcode() != ISD::EXTRACT_SUBVECTOR)
return SDValue();
// Extracting from the same source.
SDValue Src = V1.getOperand(0);
if (Src != V2.getOperand(0) ||
VT.isScalableVector() != Src.getValueType().isScalableVector())
return SDValue();
// The extracts must extract the two halves of the source.
if (V1.getConstantOperandVal(1) != 0 ||
V2.getConstantOperandVal(1) != VT.getVectorMinNumElements())
return SDValue();
return Src;
}
// Can this shuffle be performed on exactly one (possibly larger) input?
static SDValue getSingleShuffleSrc(MVT VT, SDValue V1, SDValue V2) {
if (V2.isUndef())
return V1;
unsigned NumElts = VT.getVectorNumElements();
// Src needs to have twice the number of elements.
// TODO: Update shuffle lowering to add the extract subvector
if (SDValue Src = foldConcatVector(V1, V2);
Src && Src.getValueType().getVectorNumElements() == (NumElts * 2))
return Src;
return SDValue();
}
/// Is this shuffle interleaving contiguous elements from one vector into the
/// even elements and contiguous elements from another vector into the odd
/// elements. \p EvenSrc will contain the element that should be in the first
/// even element. \p OddSrc will contain the element that should be in the first
/// odd element. These can be the first element in a source or the element half
/// way through the source.
static bool isInterleaveShuffle(ArrayRef<int> Mask, MVT VT, int &EvenSrc,
int &OddSrc, const RISCVSubtarget &Subtarget) {
// We need to be able to widen elements to the next larger integer type or
// use the zip2a instruction at e64.
if (VT.getScalarSizeInBits() >= Subtarget.getELen() &&
!Subtarget.hasVendorXRivosVizip())
return false;
int Size = Mask.size();
int NumElts = VT.getVectorNumElements();
assert(Size == (int)NumElts && "Unexpected mask size");
SmallVector<unsigned, 2> StartIndexes;
if (!ShuffleVectorInst::isInterleaveMask(Mask, 2, Size * 2, StartIndexes))
return false;
EvenSrc = StartIndexes[0];
OddSrc = StartIndexes[1];
// One source should be low half of first vector.
if (EvenSrc != 0 && OddSrc != 0)
return false;
// Subvectors will be subtracted from either at the start of the two input
// vectors, or at the start and middle of the first vector if it's an unary
// interleave.
// In both cases, HalfNumElts will be extracted.
// We need to ensure that the extract indices are 0 or HalfNumElts otherwise
// we'll create an illegal extract_subvector.
// FIXME: We could support other values using a slidedown first.
int HalfNumElts = NumElts / 2;
return ((EvenSrc % HalfNumElts) == 0) && ((OddSrc % HalfNumElts) == 0);
}
/// Is this mask representing a masked combination of two slides?
static bool isMaskedSlidePair(ArrayRef<int> Mask,
std::array<std::pair<int, int>, 2> &SrcInfo) {
if (!llvm::isMaskedSlidePair(Mask, Mask.size(), SrcInfo))
return false;
// Avoid matching vselect idioms
if (SrcInfo[0].second == 0 && SrcInfo[1].second == 0)
return false;
// Prefer vslideup as the second instruction, and identity
// only as the initial instruction.
if ((SrcInfo[0].second > 0 && SrcInfo[1].second < 0) ||
SrcInfo[1].second == 0)
std::swap(SrcInfo[0], SrcInfo[1]);
assert(SrcInfo[0].first != -1 && "Must find one slide");
return true;
}
// Exactly matches the semantics of a previously existing custom matcher
// to allow migration to new matcher without changing output.
static bool isElementRotate(const std::array<std::pair<int, int>, 2> &SrcInfo,
unsigned NumElts) {
if (SrcInfo[1].first == -1)
return true;
return SrcInfo[0].second < 0 && SrcInfo[1].second > 0 &&
SrcInfo[1].second - SrcInfo[0].second == (int)NumElts;
}
static bool isAlternating(const std::array<std::pair<int, int>, 2> &SrcInfo,
ArrayRef<int> Mask, unsigned Factor,
bool RequiredPolarity) {
int NumElts = Mask.size();
for (const auto &[Idx, M] : enumerate(Mask)) {
if (M < 0)
continue;
int Src = M >= NumElts;
int Diff = (int)Idx - (M % NumElts);
bool C = Src == SrcInfo[1].first && Diff == SrcInfo[1].second;
assert(C != (Src == SrcInfo[0].first && Diff == SrcInfo[0].second) &&
"Must match exactly one of the two slides");
if (RequiredPolarity != (C == (Idx / Factor) % 2))
return false;
}
return true;
}
/// Given a shuffle which can be represented as a pair of two slides,
/// see if it is a zipeven idiom. Zipeven is:
/// vs2: a0 a1 a2 a3
/// vs1: b0 b1 b2 b3
/// vd: a0 b0 a2 b2
static bool isZipEven(const std::array<std::pair<int, int>, 2> &SrcInfo,
ArrayRef<int> Mask, unsigned &Factor) {
Factor = SrcInfo[1].second;
return SrcInfo[0].second == 0 && isPowerOf2_32(Factor) &&
Mask.size() % Factor == 0 &&
isAlternating(SrcInfo, Mask, Factor, true);
}
/// Given a shuffle which can be represented as a pair of two slides,
/// see if it is a zipodd idiom. Zipodd is:
/// vs2: a0 a1 a2 a3
/// vs1: b0 b1 b2 b3
/// vd: a1 b1 a3 b3
/// Note that the operand order is swapped due to the way we canonicalize
/// the slides, so SrCInfo[0] is vs1, and SrcInfo[1] is vs2.
static bool isZipOdd(const std::array<std::pair<int, int>, 2> &SrcInfo,
ArrayRef<int> Mask, unsigned &Factor) {
Factor = -SrcInfo[1].second;
return SrcInfo[0].second == 0 && isPowerOf2_32(Factor) &&
Mask.size() % Factor == 0 &&
isAlternating(SrcInfo, Mask, Factor, false);
}
// Lower a deinterleave shuffle to SRL and TRUNC. Factor must be
// 2, 4, 8 and the integer type Factor-times larger than VT's
// element type must be a legal element type.
// [a, p, b, q, c, r, d, s] -> [a, b, c, d] (Factor=2, Index=0)
// -> [p, q, r, s] (Factor=2, Index=1)
static SDValue getDeinterleaveShiftAndTrunc(const SDLoc &DL, MVT VT,
SDValue Src, unsigned Factor,
unsigned Index, SelectionDAG &DAG) {
unsigned EltBits = VT.getScalarSizeInBits();
ElementCount SrcEC = Src.getValueType().getVectorElementCount();
MVT WideSrcVT = MVT::getVectorVT(MVT::getIntegerVT(EltBits * Factor),
SrcEC.divideCoefficientBy(Factor));
MVT ResVT = MVT::getVectorVT(MVT::getIntegerVT(EltBits),
SrcEC.divideCoefficientBy(Factor));
Src = DAG.getBitcast(WideSrcVT, Src);
unsigned Shift = Index * EltBits;
SDValue Res = DAG.getNode(ISD::SRL, DL, WideSrcVT, Src,
DAG.getConstant(Shift, DL, WideSrcVT));
Res = DAG.getNode(ISD::TRUNCATE, DL, ResVT, Res);
MVT CastVT = ResVT.changeVectorElementType(VT.getVectorElementType());
Res = DAG.getBitcast(CastVT, Res);
return DAG.getInsertSubvector(DL, DAG.getUNDEF(VT), Res, 0);
}
/// Match a single source shuffle which is an identity except that some
/// particular element is repeated. This can be lowered as a masked
/// vrgather.vi/vx. Note that the two source form of this is handled
/// by the recursive splitting logic and doesn't need special handling.
static SDValue lowerVECTOR_SHUFFLEAsVRGatherVX(ShuffleVectorSDNode *SVN,
const RISCVSubtarget &Subtarget,
SelectionDAG &DAG) {
SDLoc DL(SVN);
MVT VT = SVN->getSimpleValueType(0);
SDValue V1 = SVN->getOperand(0);
assert(SVN->getOperand(1).isUndef());
ArrayRef<int> Mask = SVN->getMask();
const unsigned NumElts = VT.getVectorNumElements();
MVT XLenVT = Subtarget.getXLenVT();
std::optional<int> SplatIdx;
for (auto [I, M] : enumerate(Mask)) {
if (M == -1 || I == (unsigned)M)
continue;
if (SplatIdx && *SplatIdx != M)
return SDValue();
SplatIdx = M;
}
if (!SplatIdx)
return SDValue();
SmallVector<SDValue> MaskVals;
for (int MaskIndex : Mask) {
bool SelectMaskVal = MaskIndex == *SplatIdx;
MaskVals.push_back(DAG.getConstant(SelectMaskVal, DL, XLenVT));
}
assert(MaskVals.size() == NumElts && "Unexpected select-like shuffle");
MVT MaskVT = MVT::getVectorVT(MVT::i1, NumElts);
SDValue SelectMask = DAG.getBuildVector(MaskVT, DL, MaskVals);
SDValue Splat = DAG.getVectorShuffle(VT, DL, V1, DAG.getUNDEF(VT),
SmallVector<int>(NumElts, *SplatIdx));
return DAG.getNode(ISD::VSELECT, DL, VT, SelectMask, Splat, V1);
}
// Lower the following shuffle to vslidedown.
// a)
// t49: v8i8 = extract_subvector t13, Constant:i64<0>
// t109: v8i8 = extract_subvector t13, Constant:i64<8>
// t108: v8i8 = vector_shuffle<1,2,3,4,5,6,7,8> t49, t106
// b)
// t69: v16i16 = extract_subvector t68, Constant:i64<0>
// t23: v8i16 = extract_subvector t69, Constant:i64<0>
// t29: v4i16 = extract_subvector t23, Constant:i64<4>
// t26: v8i16 = extract_subvector t69, Constant:i64<8>
// t30: v4i16 = extract_subvector t26, Constant:i64<0>
// t54: v4i16 = vector_shuffle<1,2,3,4> t29, t30
static SDValue lowerVECTOR_SHUFFLEAsVSlidedown(const SDLoc &DL, MVT VT,
SDValue V1, SDValue V2,
ArrayRef<int> Mask,
const RISCVSubtarget &Subtarget,
SelectionDAG &DAG) {
auto findNonEXTRACT_SUBVECTORParent =
[](SDValue Parent) -> std::pair<SDValue, uint64_t> {
uint64_t Offset = 0;
while (Parent.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
// EXTRACT_SUBVECTOR can be used to extract a fixed-width vector from
// a scalable vector. But we don't want to match the case.
Parent.getOperand(0).getSimpleValueType().isFixedLengthVector()) {
Offset += Parent.getConstantOperandVal(1);
Parent = Parent.getOperand(0);
}
return std::make_pair(Parent, Offset);
};
auto [V1Src, V1IndexOffset] = findNonEXTRACT_SUBVECTORParent(V1);
auto [V2Src, V2IndexOffset] = findNonEXTRACT_SUBVECTORParent(V2);
// Extracting from the same source.
SDValue Src = V1Src;
if (Src != V2Src)
return SDValue();
// Rebuild mask because Src may be from multiple EXTRACT_SUBVECTORs.
SmallVector<int, 16> NewMask(Mask);
for (size_t i = 0; i != NewMask.size(); ++i) {
if (NewMask[i] == -1)
continue;
if (static_cast<size_t>(NewMask[i]) < NewMask.size()) {
NewMask[i] = NewMask[i] + V1IndexOffset;
} else {
// Minus NewMask.size() is needed. Otherwise, the b case would be
// <5,6,7,12> instead of <5,6,7,8>.
NewMask[i] = NewMask[i] - NewMask.size() + V2IndexOffset;
}
}
// First index must be known and non-zero. It will be used as the slidedown
// amount.
if (NewMask[0] <= 0)
return SDValue();
// NewMask is also continuous.
for (unsigned i = 1; i != NewMask.size(); ++i)
if (NewMask[i - 1] + 1 != NewMask[i])
return SDValue();
MVT XLenVT = Subtarget.getXLenVT();
MVT SrcVT = Src.getSimpleValueType();
MVT ContainerVT = getContainerForFixedLengthVector(DAG, SrcVT, Subtarget);
auto [TrueMask, VL] = getDefaultVLOps(SrcVT, ContainerVT, DL, DAG, Subtarget);
SDValue Slidedown =
getVSlidedown(DAG, Subtarget, DL, ContainerVT, DAG.getUNDEF(ContainerVT),
convertToScalableVector(ContainerVT, Src, DAG, Subtarget),
DAG.getConstant(NewMask[0], DL, XLenVT), TrueMask, VL);
return DAG.getExtractSubvector(
DL, VT, convertFromScalableVector(SrcVT, Slidedown, DAG, Subtarget), 0);
}
// Because vslideup leaves the destination elements at the start intact, we can
// use it to perform shuffles that insert subvectors:
//
// vector_shuffle v8:v8i8, v9:v8i8, <0, 1, 2, 3, 8, 9, 10, 11>
// ->
// vsetvli zero, 8, e8, mf2, ta, ma
// vslideup.vi v8, v9, 4
//
// vector_shuffle v8:v8i8, v9:v8i8 <0, 1, 8, 9, 10, 5, 6, 7>
// ->
// vsetvli zero, 5, e8, mf2, tu, ma
// vslideup.v1 v8, v9, 2
static SDValue lowerVECTOR_SHUFFLEAsVSlideup(const SDLoc &DL, MVT VT,
SDValue V1, SDValue V2,
ArrayRef<int> Mask,
const RISCVSubtarget &Subtarget,
SelectionDAG &DAG) {
unsigned NumElts = VT.getVectorNumElements();
int NumSubElts, Index;
if (!ShuffleVectorInst::isInsertSubvectorMask(Mask, NumElts, NumSubElts,
Index))
return SDValue();
bool OpsSwapped = Mask[Index] < (int)NumElts;
SDValue InPlace = OpsSwapped ? V2 : V1;
SDValue ToInsert = OpsSwapped ? V1 : V2;
MVT XLenVT = Subtarget.getXLenVT();
MVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
auto TrueMask = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget).first;
// We slide up by the index that the subvector is being inserted at, and set
// VL to the index + the number of elements being inserted.
unsigned Policy =
RISCVVType::TAIL_UNDISTURBED_MASK_UNDISTURBED | RISCVVType::MASK_AGNOSTIC;
// If the we're adding a suffix to the in place vector, i.e. inserting right
// up to the very end of it, then we don't actually care about the tail.
if (NumSubElts + Index >= (int)NumElts)
Policy |= RISCVVType::TAIL_AGNOSTIC;
InPlace = convertToScalableVector(ContainerVT, InPlace, DAG, Subtarget);
ToInsert = convertToScalableVector(ContainerVT, ToInsert, DAG, Subtarget);
SDValue VL = DAG.getConstant(NumSubElts + Index, DL, XLenVT);
SDValue Res;
// If we're inserting into the lowest elements, use a tail undisturbed
// vmv.v.v.
if (Index == 0)
Res = DAG.getNode(RISCVISD::VMV_V_V_VL, DL, ContainerVT, InPlace, ToInsert,
VL);
else
Res = getVSlideup(DAG, Subtarget, DL, ContainerVT, InPlace, ToInsert,
DAG.getConstant(Index, DL, XLenVT), TrueMask, VL, Policy);
return convertFromScalableVector(VT, Res, DAG, Subtarget);
}
/// Match v(f)slide1up/down idioms. These operations involve sliding
/// N-1 elements to make room for an inserted scalar at one end.
static SDValue lowerVECTOR_SHUFFLEAsVSlide1(const SDLoc &DL, MVT VT,
SDValue V1, SDValue V2,
ArrayRef<int> Mask,
const RISCVSubtarget &Subtarget,
SelectionDAG &DAG) {
bool OpsSwapped = false;
if (!isa<BuildVectorSDNode>(V1)) {
if (!isa<BuildVectorSDNode>(V2))
return SDValue();
std::swap(V1, V2);
OpsSwapped = true;
}
SDValue Splat = cast<BuildVectorSDNode>(V1)->getSplatValue();
if (!Splat)
return SDValue();
// Return true if the mask could describe a slide of Mask.size() - 1
// elements from concat_vector(V1, V2)[Base:] to [Offset:].
auto isSlideMask = [](ArrayRef<int> Mask, unsigned Base, int Offset) {
const unsigned S = (Offset > 0) ? 0 : -Offset;
const unsigned E = Mask.size() - ((Offset > 0) ? Offset : 0);
for (unsigned i = S; i != E; ++i)
if (Mask[i] >= 0 && (unsigned)Mask[i] != Base + i + Offset)
return false;
return true;
};
const unsigned NumElts = VT.getVectorNumElements();
bool IsVSlidedown = isSlideMask(Mask, OpsSwapped ? 0 : NumElts, 1);
if (!IsVSlidedown && !isSlideMask(Mask, OpsSwapped ? 0 : NumElts, -1))
return SDValue();
const int InsertIdx = Mask[IsVSlidedown ? (NumElts - 1) : 0];
// Inserted lane must come from splat, undef scalar is legal but not profitable.
if (InsertIdx < 0 || InsertIdx / NumElts != (unsigned)OpsSwapped)
return SDValue();
MVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
auto [TrueMask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
// zvfhmin and zvfbfmin don't have vfslide1{down,up}.vf so use fmv.x.h +
// vslide1{down,up}.vx instead.
if (VT.getVectorElementType() == MVT::bf16 ||
(VT.getVectorElementType() == MVT::f16 &&
!Subtarget.hasVInstructionsF16())) {
MVT IntVT = ContainerVT.changeVectorElementTypeToInteger();
Splat =
DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, Subtarget.getXLenVT(), Splat);
V2 = DAG.getBitcast(
IntVT, convertToScalableVector(ContainerVT, V2, DAG, Subtarget));
SDValue Vec = DAG.getNode(
IsVSlidedown ? RISCVISD::VSLIDE1DOWN_VL : RISCVISD::VSLIDE1UP_VL, DL,
IntVT, DAG.getUNDEF(IntVT), V2, Splat, TrueMask, VL);
Vec = DAG.getBitcast(ContainerVT, Vec);
return convertFromScalableVector(VT, Vec, DAG, Subtarget);
}
auto OpCode = IsVSlidedown ?
(VT.isFloatingPoint() ? RISCVISD::VFSLIDE1DOWN_VL : RISCVISD::VSLIDE1DOWN_VL) :
(VT.isFloatingPoint() ? RISCVISD::VFSLIDE1UP_VL : RISCVISD::VSLIDE1UP_VL);
if (!VT.isFloatingPoint())
Splat = DAG.getNode(ISD::ANY_EXTEND, DL, Subtarget.getXLenVT(), Splat);
auto Vec = DAG.getNode(OpCode, DL, ContainerVT,
DAG.getUNDEF(ContainerVT),
convertToScalableVector(ContainerVT, V2, DAG, Subtarget),
Splat, TrueMask, VL);
return convertFromScalableVector(VT, Vec, DAG, Subtarget);
}
/// Match a mask which "spreads" the leading elements of a vector evenly
/// across the result. Factor is the spread amount, and Index is the
/// offset applied. (on success, Index < Factor) This is the inverse
/// of a deinterleave with the same Factor and Index. This is analogous
/// to an interleave, except that all but one lane is undef.
bool RISCVTargetLowering::isSpreadMask(ArrayRef<int> Mask, unsigned Factor,
unsigned &Index) {
SmallVector<bool> LaneIsUndef(Factor, true);
for (unsigned i = 0; i < Mask.size(); i++)
LaneIsUndef[i % Factor] &= (Mask[i] == -1);
bool Found = false;
for (unsigned i = 0; i < Factor; i++) {
if (LaneIsUndef[i])
continue;
if (Found)
return false;
Index = i;
Found = true;
}
if (!Found)
return false;
for (unsigned i = 0; i < Mask.size() / Factor; i++) {
unsigned j = i * Factor + Index;
if (Mask[j] != -1 && (unsigned)Mask[j] != i)
return false;
}
return true;
}
static SDValue lowerVZIP(unsigned Opc, SDValue Op0, SDValue Op1,
const SDLoc &DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(RISCVISD::RI_VZIPEVEN_VL == Opc || RISCVISD::RI_VZIPODD_VL == Opc ||
RISCVISD::RI_VZIP2A_VL == Opc || RISCVISD::RI_VZIP2B_VL == Opc ||
RISCVISD::RI_VUNZIP2A_VL == Opc || RISCVISD::RI_VUNZIP2B_VL == Opc);
assert(Op0.getSimpleValueType() == Op1.getSimpleValueType());
MVT VT = Op0.getSimpleValueType();
MVT IntVT = VT.changeVectorElementTypeToInteger();
Op0 = DAG.getBitcast(IntVT, Op0);
Op1 = DAG.getBitcast(IntVT, Op1);
MVT ContainerVT = IntVT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(DAG, IntVT, Subtarget);
Op0 = convertToScalableVector(ContainerVT, Op0, DAG, Subtarget);
Op1 = convertToScalableVector(ContainerVT, Op1, DAG, Subtarget);
}
MVT InnerVT = ContainerVT;
auto [Mask, VL] = getDefaultVLOps(IntVT, InnerVT, DL, DAG, Subtarget);
if (Op1.isUndef() &&
ContainerVT.bitsGT(RISCVTargetLowering::getM1VT(ContainerVT)) &&
(RISCVISD::RI_VUNZIP2A_VL == Opc || RISCVISD::RI_VUNZIP2B_VL == Opc)) {
InnerVT = ContainerVT.getHalfNumVectorElementsVT();
VL = DAG.getConstant(VT.getVectorNumElements() / 2, DL,
Subtarget.getXLenVT());
Mask = getAllOnesMask(InnerVT, VL, DL, DAG);
unsigned HighIdx = InnerVT.getVectorElementCount().getKnownMinValue();
Op1 = DAG.getExtractSubvector(DL, InnerVT, Op0, HighIdx);
Op0 = DAG.getExtractSubvector(DL, InnerVT, Op0, 0);
}
SDValue Passthru = DAG.getUNDEF(InnerVT);
SDValue Res = DAG.getNode(Opc, DL, InnerVT, Op0, Op1, Passthru, Mask, VL);
if (InnerVT.bitsLT(ContainerVT))
Res = DAG.getInsertSubvector(DL, DAG.getUNDEF(ContainerVT), Res, 0);
if (IntVT.isFixedLengthVector())
Res = convertFromScalableVector(IntVT, Res, DAG, Subtarget);
Res = DAG.getBitcast(VT, Res);
return Res;
}
// Given a vector a, b, c, d return a vector Factor times longer
// with Factor-1 undef's between elements. Ex:
// a, undef, b, undef, c, undef, d, undef (Factor=2, Index=0)
// undef, a, undef, b, undef, c, undef, d (Factor=2, Index=1)
static SDValue getWideningSpread(SDValue V, unsigned Factor, unsigned Index,
const SDLoc &DL, SelectionDAG &DAG) {
MVT VT = V.getSimpleValueType();
unsigned EltBits = VT.getScalarSizeInBits();
ElementCount EC = VT.getVectorElementCount();
V = DAG.getBitcast(VT.changeTypeToInteger(), V);
MVT WideVT = MVT::getVectorVT(MVT::getIntegerVT(EltBits * Factor), EC);
SDValue Result = DAG.getNode(ISD::ZERO_EXTEND, DL, WideVT, V);
// TODO: On rv32, the constant becomes a splat_vector_parts which does not
// allow the SHL to fold away if Index is 0.
if (Index != 0)
Result = DAG.getNode(ISD::SHL, DL, WideVT, Result,
DAG.getConstant(EltBits * Index, DL, WideVT));
// Make sure to use original element type
MVT ResultVT = MVT::getVectorVT(VT.getVectorElementType(),
EC.multiplyCoefficientBy(Factor));
return DAG.getBitcast(ResultVT, Result);
}
// Given two input vectors of <[vscale x ]n x ty>, use vwaddu.vv and vwmaccu.vx
// to create an interleaved vector of <[vscale x] n*2 x ty>.
// This requires that the size of ty is less than the subtarget's maximum ELEN.
static SDValue getWideningInterleave(SDValue EvenV, SDValue OddV,
const SDLoc &DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
// FIXME: Not only does this optimize the code, it fixes some correctness
// issues because MIR does not have freeze.
if (EvenV.isUndef())
return getWideningSpread(OddV, 2, 1, DL, DAG);
if (OddV.isUndef())
return getWideningSpread(EvenV, 2, 0, DL, DAG);
MVT VecVT = EvenV.getSimpleValueType();
MVT VecContainerVT = VecVT; // <vscale x n x ty>
// Convert fixed vectors to scalable if needed
if (VecContainerVT.isFixedLengthVector()) {
VecContainerVT = getContainerForFixedLengthVector(DAG, VecVT, Subtarget);
EvenV = convertToScalableVector(VecContainerVT, EvenV, DAG, Subtarget);
OddV = convertToScalableVector(VecContainerVT, OddV, DAG, Subtarget);
}
assert(VecVT.getScalarSizeInBits() < Subtarget.getELen());
// We're working with a vector of the same size as the resulting
// interleaved vector, but with half the number of elements and
// twice the SEW (Hence the restriction on not using the maximum
// ELEN)
MVT WideVT =
MVT::getVectorVT(MVT::getIntegerVT(VecVT.getScalarSizeInBits() * 2),
VecVT.getVectorElementCount());
MVT WideContainerVT = WideVT; // <vscale x n x ty*2>
if (WideContainerVT.isFixedLengthVector())
WideContainerVT = getContainerForFixedLengthVector(DAG, WideVT, Subtarget);
// Bitcast the input vectors to integers in case they are FP
VecContainerVT = VecContainerVT.changeTypeToInteger();
EvenV = DAG.getBitcast(VecContainerVT, EvenV);
OddV = DAG.getBitcast(VecContainerVT, OddV);
auto [Mask, VL] = getDefaultVLOps(VecVT, VecContainerVT, DL, DAG, Subtarget);
SDValue Passthru = DAG.getUNDEF(WideContainerVT);
SDValue Interleaved;
if (Subtarget.hasStdExtZvbb()) {
// Interleaved = (OddV << VecVT.getScalarSizeInBits()) + EvenV.
SDValue OffsetVec =
DAG.getConstant(VecVT.getScalarSizeInBits(), DL, VecContainerVT);
Interleaved = DAG.getNode(RISCVISD::VWSLL_VL, DL, WideContainerVT, OddV,
OffsetVec, Passthru, Mask, VL);
Interleaved = DAG.getNode(RISCVISD::VWADDU_W_VL, DL, WideContainerVT,
Interleaved, EvenV, Passthru, Mask, VL);
} else {
// FIXME: We should freeze the odd vector here. We already handled the case
// of provably undef/poison above.
// Widen EvenV and OddV with 0s and add one copy of OddV to EvenV with
// vwaddu.vv
Interleaved = DAG.getNode(RISCVISD::VWADDU_VL, DL, WideContainerVT, EvenV,
OddV, Passthru, Mask, VL);
// Then get OddV * by 2^(VecVT.getScalarSizeInBits() - 1)
SDValue AllOnesVec = DAG.getSplatVector(
VecContainerVT, DL, DAG.getAllOnesConstant(DL, Subtarget.getXLenVT()));
SDValue OddsMul = DAG.getNode(RISCVISD::VWMULU_VL, DL, WideContainerVT,
OddV, AllOnesVec, Passthru, Mask, VL);
// Add the two together so we get
// (OddV * 0xff...ff) + (OddV + EvenV)
// = (OddV * 0x100...00) + EvenV
// = (OddV << VecVT.getScalarSizeInBits()) + EvenV
// Note the ADD_VL and VLMULU_VL should get selected as vwmaccu.vx
Interleaved = DAG.getNode(RISCVISD::ADD_VL, DL, WideContainerVT,
Interleaved, OddsMul, Passthru, Mask, VL);
}
// Bitcast from <vscale x n * ty*2> to <vscale x 2*n x ty>
MVT ResultContainerVT = MVT::getVectorVT(
VecVT.getVectorElementType(), // Make sure to use original type
VecContainerVT.getVectorElementCount().multiplyCoefficientBy(2));
Interleaved = DAG.getBitcast(ResultContainerVT, Interleaved);
// Convert back to a fixed vector if needed
MVT ResultVT =
MVT::getVectorVT(VecVT.getVectorElementType(),
VecVT.getVectorElementCount().multiplyCoefficientBy(2));
if (ResultVT.isFixedLengthVector())
Interleaved =
convertFromScalableVector(ResultVT, Interleaved, DAG, Subtarget);
return Interleaved;
}
// If we have a vector of bits that we want to reverse, we can use a vbrev on a
// larger element type, e.g. v32i1 can be reversed with a v1i32 bitreverse.
static SDValue lowerBitreverseShuffle(ShuffleVectorSDNode *SVN,
SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDLoc DL(SVN);
MVT VT = SVN->getSimpleValueType(0);
SDValue V = SVN->getOperand(0);
unsigned NumElts = VT.getVectorNumElements();
assert(VT.getVectorElementType() == MVT::i1);
if (!ShuffleVectorInst::isReverseMask(SVN->getMask(),
SVN->getMask().size()) ||
!SVN->getOperand(1).isUndef())
return SDValue();
unsigned ViaEltSize = std::max((uint64_t)8, PowerOf2Ceil(NumElts));
EVT ViaVT = EVT::getVectorVT(
*DAG.getContext(), EVT::getIntegerVT(*DAG.getContext(), ViaEltSize), 1);
EVT ViaBitVT =
EVT::getVectorVT(*DAG.getContext(), MVT::i1, ViaVT.getScalarSizeInBits());
// If we don't have zvbb or the larger element type > ELEN, the operation will
// be illegal.
if (!Subtarget.getTargetLowering()->isOperationLegalOrCustom(ISD::BITREVERSE,
ViaVT) ||
!Subtarget.getTargetLowering()->isTypeLegal(ViaBitVT))
return SDValue();
// If the bit vector doesn't fit exactly into the larger element type, we need
// to insert it into the larger vector and then shift up the reversed bits
// afterwards to get rid of the gap introduced.
if (ViaEltSize > NumElts)
V = DAG.getInsertSubvector(DL, DAG.getUNDEF(ViaBitVT), V, 0);
SDValue Res =
DAG.getNode(ISD::BITREVERSE, DL, ViaVT, DAG.getBitcast(ViaVT, V));
// Shift up the reversed bits if the vector didn't exactly fit into the larger
// element type.
if (ViaEltSize > NumElts)
Res = DAG.getNode(ISD::SRL, DL, ViaVT, Res,
DAG.getConstant(ViaEltSize - NumElts, DL, ViaVT));
Res = DAG.getBitcast(ViaBitVT, Res);
if (ViaEltSize > NumElts)
Res = DAG.getExtractSubvector(DL, VT, Res, 0);
return Res;
}
static bool isLegalBitRotate(ArrayRef<int> Mask, EVT VT,
const RISCVSubtarget &Subtarget,
MVT &RotateVT, unsigned &RotateAmt) {
unsigned NumElts = VT.getVectorNumElements();
unsigned EltSizeInBits = VT.getScalarSizeInBits();
unsigned NumSubElts;
if (!ShuffleVectorInst::isBitRotateMask(Mask, EltSizeInBits, 2,
NumElts, NumSubElts, RotateAmt))
return false;
RotateVT = MVT::getVectorVT(MVT::getIntegerVT(EltSizeInBits * NumSubElts),
NumElts / NumSubElts);
// We might have a RotateVT that isn't legal, e.g. v4i64 on zve32x.
return Subtarget.getTargetLowering()->isTypeLegal(RotateVT);
}
// Given a shuffle mask like <3, 0, 1, 2, 7, 4, 5, 6> for v8i8, we can
// reinterpret it as a v2i32 and rotate it right by 8 instead. We can lower this
// as a vror.vi if we have Zvkb, or otherwise as a vsll, vsrl and vor.
static SDValue lowerVECTOR_SHUFFLEAsRotate(ShuffleVectorSDNode *SVN,
SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDLoc DL(SVN);
EVT VT = SVN->getValueType(0);
unsigned RotateAmt;
MVT RotateVT;
if (!isLegalBitRotate(SVN->getMask(), VT, Subtarget, RotateVT, RotateAmt))
return SDValue();
SDValue Op = DAG.getBitcast(RotateVT, SVN->getOperand(0));
SDValue Rotate;
// A rotate of an i16 by 8 bits either direction is equivalent to a byteswap,
// so canonicalize to vrev8.
if (RotateVT.getScalarType() == MVT::i16 && RotateAmt == 8)
Rotate = DAG.getNode(ISD::BSWAP, DL, RotateVT, Op);
else
Rotate = DAG.getNode(ISD::ROTL, DL, RotateVT, Op,
DAG.getConstant(RotateAmt, DL, RotateVT));
return DAG.getBitcast(VT, Rotate);
}
// If compiling with an exactly known VLEN, see if we can split a
// shuffle on m2 or larger into a small number of m1 sized shuffles
// which write each destination registers exactly once.
static SDValue lowerShuffleViaVRegSplitting(ShuffleVectorSDNode *SVN,
SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDLoc DL(SVN);
MVT VT = SVN->getSimpleValueType(0);
SDValue V1 = SVN->getOperand(0);
SDValue V2 = SVN->getOperand(1);
ArrayRef<int> Mask = SVN->getMask();
// If we don't know exact data layout, not much we can do. If this
// is already m1 or smaller, no point in splitting further.
const auto VLen = Subtarget.getRealVLen();
if (!VLen || VT.getSizeInBits().getFixedValue() <= *VLen)
return SDValue();
// Avoid picking up bitrotate patterns which we have a linear-in-lmul
// expansion for.
unsigned RotateAmt;
MVT RotateVT;
if (isLegalBitRotate(Mask, VT, Subtarget, RotateVT, RotateAmt))
return SDValue();
MVT ElemVT = VT.getVectorElementType();
unsigned ElemsPerVReg = *VLen / ElemVT.getFixedSizeInBits();
EVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
MVT OneRegVT = MVT::getVectorVT(ElemVT, ElemsPerVReg);
MVT M1VT = getContainerForFixedLengthVector(DAG, OneRegVT, Subtarget);
assert(M1VT == RISCVTargetLowering::getM1VT(M1VT));
unsigned NumOpElts = M1VT.getVectorMinNumElements();
unsigned NumElts = ContainerVT.getVectorMinNumElements();
unsigned NumOfSrcRegs = NumElts / NumOpElts;
unsigned NumOfDestRegs = NumElts / NumOpElts;
// The following semantically builds up a fixed length concat_vector
// of the component shuffle_vectors. We eagerly lower to scalable here
// to avoid DAG combining it back to a large shuffle_vector again.
V1 = convertToScalableVector(ContainerVT, V1, DAG, Subtarget);
V2 = convertToScalableVector(ContainerVT, V2, DAG, Subtarget);
SmallVector<SmallVector<std::tuple<unsigned, unsigned, SmallVector<int>>>>
Operands;
processShuffleMasks(
Mask, NumOfSrcRegs, NumOfDestRegs, NumOfDestRegs,
[&]() { Operands.emplace_back(); },
[&](ArrayRef<int> SrcSubMask, unsigned SrcVecIdx, unsigned DstVecIdx) {
Operands.emplace_back().emplace_back(SrcVecIdx, UINT_MAX,
SmallVector<int>(SrcSubMask));
},
[&](ArrayRef<int> SrcSubMask, unsigned Idx1, unsigned Idx2, bool NewReg) {
if (NewReg)
Operands.emplace_back();
Operands.back().emplace_back(Idx1, Idx2, SmallVector<int>(SrcSubMask));
});
assert(Operands.size() == NumOfDestRegs && "Whole vector must be processed");
// Note: check that we do not emit too many shuffles here to prevent code
// size explosion.
// TODO: investigate, if it can be improved by extra analysis of the masks to
// check if the code is more profitable.
unsigned NumShuffles = std::accumulate(
Operands.begin(), Operands.end(), 0u,
[&](unsigned N,
ArrayRef<std::tuple<unsigned, unsigned, SmallVector<int>>> Data) {
if (Data.empty())
return N;
N += Data.size();
for (const auto &P : Data) {
unsigned Idx2 = std::get<1>(P);
ArrayRef<int> Mask = std::get<2>(P);
if (Idx2 != UINT_MAX)
++N;
else if (ShuffleVectorInst::isIdentityMask(Mask, Mask.size()))
--N;
}
return N;
});
if ((NumOfDestRegs > 2 && NumShuffles > NumOfDestRegs) ||
(NumOfDestRegs <= 2 && NumShuffles >= 4))
return SDValue();
auto ExtractValue = [&, &DAG = DAG](SDValue SrcVec, unsigned ExtractIdx) {
SDValue SubVec = DAG.getExtractSubvector(DL, M1VT, SrcVec, ExtractIdx);
SubVec = convertFromScalableVector(OneRegVT, SubVec, DAG, Subtarget);
return SubVec;
};
auto PerformShuffle = [&, &DAG = DAG](SDValue SubVec1, SDValue SubVec2,
ArrayRef<int> Mask) {
SDValue SubVec = DAG.getVectorShuffle(OneRegVT, DL, SubVec1, SubVec2, Mask);
return SubVec;
};
SDValue Vec = DAG.getUNDEF(ContainerVT);
for (auto [I, Data] : enumerate(Operands)) {
if (Data.empty())
continue;
SmallDenseMap<unsigned, SDValue, 4> Values;
for (unsigned I : seq<unsigned>(Data.size())) {
const auto &[Idx1, Idx2, _] = Data[I];
// If the shuffle contains permutation of odd number of elements,
// Idx1 might be used already in the first iteration.
//
// Idx1 = shuffle Idx1, Idx2
// Idx1 = shuffle Idx1, Idx3
SDValue &V = Values.try_emplace(Idx1).first->getSecond();
if (!V)
V = ExtractValue(Idx1 >= NumOfSrcRegs ? V2 : V1,
(Idx1 % NumOfSrcRegs) * NumOpElts);
if (Idx2 != UINT_MAX) {
SDValue &V = Values.try_emplace(Idx2).first->getSecond();
if (!V)
V = ExtractValue(Idx2 >= NumOfSrcRegs ? V2 : V1,
(Idx2 % NumOfSrcRegs) * NumOpElts);
}
}
SDValue V;
for (const auto &[Idx1, Idx2, Mask] : Data) {
SDValue V1 = Values.at(Idx1);
SDValue V2 = Idx2 == UINT_MAX ? V1 : Values.at(Idx2);
V = PerformShuffle(V1, V2, Mask);
Values[Idx1] = V;
}
unsigned InsertIdx = I * NumOpElts;
V = convertToScalableVector(M1VT, V, DAG, Subtarget);
Vec = DAG.getInsertSubvector(DL, Vec, V, InsertIdx);
}
return convertFromScalableVector(VT, Vec, DAG, Subtarget);
}
// Matches a subset of compress masks with a contiguous prefix of output
// elements. This could be extended to allow gaps by deciding which
// source elements to spuriously demand.
static bool isCompressMask(ArrayRef<int> Mask) {
int Last = -1;
bool SawUndef = false;
for (const auto &[Idx, M] : enumerate(Mask)) {
if (M == -1) {
SawUndef = true;
continue;
}
if (SawUndef)
return false;
if (Idx > (unsigned)M)
return false;
if (M <= Last)
return false;
Last = M;
}
return true;
}
/// Given a shuffle where the indices are disjoint between the two sources,
/// e.g.:
///
/// t2:v4i8 = vector_shuffle t0:v4i8, t1:v4i8, <2, 7, 1, 4>
///
/// Merge the two sources into one and do a single source shuffle:
///
/// t2:v4i8 = vselect t1:v4i8, t0:v4i8, <0, 1, 0, 1>
/// t3:v4i8 = vector_shuffle t2:v4i8, undef, <2, 3, 1, 0>
///
/// A vselect will either be merged into a masked instruction or be lowered as a
/// vmerge.vvm, which is cheaper than a vrgather.vv.
static SDValue lowerDisjointIndicesShuffle(ShuffleVectorSDNode *SVN,
SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
MVT VT = SVN->getSimpleValueType(0);
MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(SVN);
const ArrayRef<int> Mask = SVN->getMask();
// Work out which source each lane will come from.
SmallVector<int, 16> Srcs(Mask.size(), -1);
for (int Idx : Mask) {
if (Idx == -1)
continue;
unsigned SrcIdx = Idx % Mask.size();
int Src = (uint32_t)Idx < Mask.size() ? 0 : 1;
if (Srcs[SrcIdx] == -1)
// Mark this source as using this lane.
Srcs[SrcIdx] = Src;
else if (Srcs[SrcIdx] != Src)
// The other source is using this lane: not disjoint.
return SDValue();
}
SmallVector<SDValue> SelectMaskVals;
for (int Lane : Srcs) {
if (Lane == -1)
SelectMaskVals.push_back(DAG.getUNDEF(XLenVT));
else
SelectMaskVals.push_back(DAG.getConstant(Lane ? 0 : 1, DL, XLenVT));
}
MVT MaskVT = VT.changeVectorElementType(MVT::i1);
SDValue SelectMask = DAG.getBuildVector(MaskVT, DL, SelectMaskVals);
SDValue Select = DAG.getNode(ISD::VSELECT, DL, VT, SelectMask,
SVN->getOperand(0), SVN->getOperand(1));
// Move all indices relative to the first source.
SmallVector<int> NewMask(Mask.size());
for (unsigned I = 0; I < Mask.size(); I++) {
if (Mask[I] == -1)
NewMask[I] = -1;
else
NewMask[I] = Mask[I] % Mask.size();
}
return DAG.getVectorShuffle(VT, DL, Select, DAG.getUNDEF(VT), NewMask);
}
/// Is this mask local (i.e. elements only move within their local span), and
/// repeating (that is, the same rearrangement is being done within each span)?
static bool isLocalRepeatingShuffle(ArrayRef<int> Mask, int Span) {
// Require a prefix from the original mask until the consumer code
// is adjusted to rewrite the mask instead of just taking a prefix.
for (auto [I, M] : enumerate(Mask)) {
if (M == -1)
continue;
if ((M / Span) != (int)(I / Span))
return false;
int SpanIdx = I % Span;
int Expected = M % Span;
if (Mask[SpanIdx] != Expected)
return false;
}
return true;
}
/// Is this mask only using elements from the first span of the input?
static bool isLowSourceShuffle(ArrayRef<int> Mask, int Span) {
return all_of(Mask, [&](const auto &Idx) { return Idx == -1 || Idx < Span; });
}
/// Return true for a mask which performs an arbitrary shuffle within the first
/// span, and then repeats that same result across all remaining spans. Note
/// that this doesn't check if all the inputs come from a single span!
static bool isSpanSplatShuffle(ArrayRef<int> Mask, int Span) {
// Require a prefix from the original mask until the consumer code
// is adjusted to rewrite the mask instead of just taking a prefix.
for (auto [I, M] : enumerate(Mask)) {
if (M == -1)
continue;
int SpanIdx = I % Span;
if (Mask[SpanIdx] != M)
return false;
}
return true;
}
/// Try to widen element type to get a new mask value for a better permutation
/// sequence. This doesn't try to inspect the widened mask for profitability;
/// we speculate the widened form is equal or better. This has the effect of
/// reducing mask constant sizes - allowing cheaper materialization sequences
/// - and index sequence sizes - reducing register pressure and materialization
/// cost, at the cost of (possibly) an extra VTYPE toggle.
static SDValue tryWidenMaskForShuffle(SDValue Op, SelectionDAG &DAG) {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
MVT ScalarVT = VT.getVectorElementType();
unsigned ElementSize = ScalarVT.getFixedSizeInBits();
SDValue V0 = Op.getOperand(0);
SDValue V1 = Op.getOperand(1);
ArrayRef<int> Mask = cast<ShuffleVectorSDNode>(Op)->getMask();
// Avoid wasted work leading to isTypeLegal check failing below
if (ElementSize > 32)
return SDValue();
SmallVector<int, 8> NewMask;
if (!widenShuffleMaskElts(Mask, NewMask))
return SDValue();
MVT NewEltVT = VT.isFloatingPoint() ? MVT::getFloatingPointVT(ElementSize * 2)
: MVT::getIntegerVT(ElementSize * 2);
MVT NewVT = MVT::getVectorVT(NewEltVT, VT.getVectorNumElements() / 2);
if (!DAG.getTargetLoweringInfo().isTypeLegal(NewVT))
return SDValue();
V0 = DAG.getBitcast(NewVT, V0);
V1 = DAG.getBitcast(NewVT, V1);
return DAG.getBitcast(VT, DAG.getVectorShuffle(NewVT, DL, V0, V1, NewMask));
}
static SDValue lowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
MVT VT = Op.getSimpleValueType();
unsigned NumElts = VT.getVectorNumElements();
ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(Op.getNode());
if (VT.getVectorElementType() == MVT::i1) {
// Lower to a vror.vi of a larger element type if possible before we promote
// i1s to i8s.
if (SDValue V = lowerVECTOR_SHUFFLEAsRotate(SVN, DAG, Subtarget))
return V;
if (SDValue V = lowerBitreverseShuffle(SVN, DAG, Subtarget))
return V;
// Promote i1 shuffle to i8 shuffle.
MVT WidenVT = MVT::getVectorVT(MVT::i8, VT.getVectorElementCount());
V1 = DAG.getNode(ISD::ZERO_EXTEND, DL, WidenVT, V1);
V2 = V2.isUndef() ? DAG.getUNDEF(WidenVT)
: DAG.getNode(ISD::ZERO_EXTEND, DL, WidenVT, V2);
SDValue Shuffled = DAG.getVectorShuffle(WidenVT, DL, V1, V2, SVN->getMask());
return DAG.getSetCC(DL, VT, Shuffled, DAG.getConstant(0, DL, WidenVT),
ISD::SETNE);
}
MVT ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
// Store the return value in a single variable instead of structured bindings
// so that we can pass it to GetSlide below, which cannot capture structured
// bindings until C++20.
auto TrueMaskVL = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
auto [TrueMask, VL] = TrueMaskVL;
if (SVN->isSplat()) {
const int Lane = SVN->getSplatIndex();
if (Lane >= 0) {
MVT SVT = VT.getVectorElementType();
// Turn splatted vector load into a strided load with an X0 stride.
SDValue V = V1;
// Peek through CONCAT_VECTORS as VectorCombine can concat a vector
// with undef.
// FIXME: Peek through INSERT_SUBVECTOR, EXTRACT_SUBVECTOR, bitcasts?
int Offset = Lane;
if (V.getOpcode() == ISD::CONCAT_VECTORS) {
int OpElements =
V.getOperand(0).getSimpleValueType().getVectorNumElements();
V = V.getOperand(Offset / OpElements);
Offset %= OpElements;
}
// We need to ensure the load isn't atomic or volatile.
if (ISD::isNormalLoad(V.getNode()) && cast<LoadSDNode>(V)->isSimple()) {
auto *Ld = cast<LoadSDNode>(V);
Offset *= SVT.getStoreSize();
SDValue NewAddr = DAG.getMemBasePlusOffset(
Ld->getBasePtr(), TypeSize::getFixed(Offset), DL);
// If this is SEW=64 on RV32, use a strided load with a stride of x0.
if (SVT.isInteger() && SVT.bitsGT(XLenVT)) {
SDVTList VTs = DAG.getVTList({ContainerVT, MVT::Other});
SDValue IntID =
DAG.getTargetConstant(Intrinsic::riscv_vlse, DL, XLenVT);
SDValue Ops[] = {Ld->getChain(),
IntID,
DAG.getUNDEF(ContainerVT),
NewAddr,
DAG.getRegister(RISCV::X0, XLenVT),
VL};
SDValue NewLoad = DAG.getMemIntrinsicNode(
ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops, SVT,
DAG.getMachineFunction().getMachineMemOperand(
Ld->getMemOperand(), Offset, SVT.getStoreSize()));
DAG.makeEquivalentMemoryOrdering(Ld, NewLoad);
return convertFromScalableVector(VT, NewLoad, DAG, Subtarget);
}
MVT SplatVT = ContainerVT;
// f16 with zvfhmin and bf16 need to use an integer scalar load.
if (SVT == MVT::bf16 ||
(SVT == MVT::f16 && !Subtarget.hasStdExtZfh())) {
SVT = MVT::i16;
SplatVT = ContainerVT.changeVectorElementType(SVT);
}
// Otherwise use a scalar load and splat. This will give the best
// opportunity to fold a splat into the operation. ISel can turn it into
// the x0 strided load if we aren't able to fold away the select.
if (SVT.isFloatingPoint())
V = DAG.getLoad(SVT, DL, Ld->getChain(), NewAddr,
Ld->getPointerInfo().getWithOffset(Offset),
Ld->getBaseAlign(), Ld->getMemOperand()->getFlags());
else
V = DAG.getExtLoad(ISD::EXTLOAD, DL, XLenVT, Ld->getChain(), NewAddr,
Ld->getPointerInfo().getWithOffset(Offset), SVT,
Ld->getBaseAlign(),
Ld->getMemOperand()->getFlags());
DAG.makeEquivalentMemoryOrdering(Ld, V);
unsigned Opc = SplatVT.isFloatingPoint() ? RISCVISD::VFMV_V_F_VL
: RISCVISD::VMV_V_X_VL;
SDValue Splat =
DAG.getNode(Opc, DL, SplatVT, DAG.getUNDEF(ContainerVT), V, VL);
Splat = DAG.getBitcast(ContainerVT, Splat);
return convertFromScalableVector(VT, Splat, DAG, Subtarget);
}
V1 = convertToScalableVector(ContainerVT, V1, DAG, Subtarget);
assert(Lane < (int)NumElts && "Unexpected lane!");
SDValue Gather = DAG.getNode(RISCVISD::VRGATHER_VX_VL, DL, ContainerVT,
V1, DAG.getConstant(Lane, DL, XLenVT),
DAG.getUNDEF(ContainerVT), TrueMask, VL);
return convertFromScalableVector(VT, Gather, DAG, Subtarget);
}
}
// For exact VLEN m2 or greater, try to split to m1 operations if we
// can split cleanly.
if (SDValue V = lowerShuffleViaVRegSplitting(SVN, DAG, Subtarget))
return V;
ArrayRef<int> Mask = SVN->getMask();
if (SDValue V =
lowerVECTOR_SHUFFLEAsVSlide1(DL, VT, V1, V2, Mask, Subtarget, DAG))
return V;
if (SDValue V =
lowerVECTOR_SHUFFLEAsVSlidedown(DL, VT, V1, V2, Mask, Subtarget, DAG))
return V;
// A bitrotate will be one instruction on Zvkb, so try to lower to it first if
// available.
if (Subtarget.hasStdExtZvkb())
if (SDValue V = lowerVECTOR_SHUFFLEAsRotate(SVN, DAG, Subtarget))
return V;
if (ShuffleVectorInst::isReverseMask(Mask, NumElts) && V2.isUndef() &&
NumElts != 2)
return DAG.getNode(ISD::VECTOR_REVERSE, DL, VT, V1);
// If this is a deinterleave(2,4,8) and we can widen the vector, then we can
// use shift and truncate to perform the shuffle.
// TODO: For Factor=6, we can perform the first step of the deinterleave via
// shift-and-trunc reducing total cost for everything except an mf8 result.
// TODO: For Factor=4,8, we can do the same when the ratio isn't high enough
// to do the entire operation.
if (VT.getScalarSizeInBits() < Subtarget.getELen()) {
const unsigned MaxFactor = Subtarget.getELen() / VT.getScalarSizeInBits();
assert(MaxFactor == 2 || MaxFactor == 4 || MaxFactor == 8);
for (unsigned Factor = 2; Factor <= MaxFactor; Factor <<= 1) {
unsigned Index = 0;
if (ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, Factor, Index) &&
1 < count_if(Mask, [](int Idx) { return Idx != -1; })) {
if (SDValue Src = getSingleShuffleSrc(VT, V1, V2))
return getDeinterleaveShiftAndTrunc(DL, VT, Src, Factor, Index, DAG);
if (1 < count_if(Mask,
[&Mask](int Idx) { return Idx < (int)Mask.size(); }) &&
1 < count_if(Mask, [&Mask](int Idx) {
return Idx >= (int)Mask.size();
})) {
// Narrow each source and concatenate them.
// FIXME: For small LMUL it is better to concatenate first.
MVT EltVT = VT.getVectorElementType();
auto EltCnt = VT.getVectorElementCount();
MVT SubVT =
MVT::getVectorVT(EltVT, EltCnt.divideCoefficientBy(Factor));
SDValue Lo =
getDeinterleaveShiftAndTrunc(DL, SubVT, V1, Factor, Index, DAG);
SDValue Hi =
getDeinterleaveShiftAndTrunc(DL, SubVT, V2, Factor, Index, DAG);
SDValue Concat =
DAG.getNode(ISD::CONCAT_VECTORS, DL,
SubVT.getDoubleNumVectorElementsVT(), Lo, Hi);
if (Factor == 2)
return Concat;
SDValue Vec = DAG.getUNDEF(VT);
return DAG.getInsertSubvector(DL, Vec, Concat, 0);
}
}
}
}
// If this is a deinterleave(2), try using vunzip{a,b}. This mostly catches
// e64 which can't match above.
unsigned Index = 0;
if (Subtarget.hasVendorXRivosVizip() &&
ShuffleVectorInst::isDeInterleaveMaskOfFactor(Mask, 2, Index) &&
1 < count_if(Mask, [](int Idx) { return Idx != -1; })) {
unsigned Opc =
Index == 0 ? RISCVISD::RI_VUNZIP2A_VL : RISCVISD::RI_VUNZIP2B_VL;
if (V2.isUndef())
return lowerVZIP(Opc, V1, V2, DL, DAG, Subtarget);
if (auto VLEN = Subtarget.getRealVLen();
VLEN && VT.getSizeInBits().getKnownMinValue() % *VLEN == 0)
return lowerVZIP(Opc, V1, V2, DL, DAG, Subtarget);
if (SDValue Src = foldConcatVector(V1, V2)) {
EVT NewVT = VT.getDoubleNumVectorElementsVT();
Src = DAG.getExtractSubvector(DL, NewVT, Src, 0);
SDValue Res =
lowerVZIP(Opc, Src, DAG.getUNDEF(NewVT), DL, DAG, Subtarget);
return DAG.getExtractSubvector(DL, VT, Res, 0);
}
// Deinterleave each source and concatenate them, or concat first, then
// deinterleave.
if (1 < count_if(Mask,
[&Mask](int Idx) { return Idx < (int)Mask.size(); }) &&
1 < count_if(Mask,
[&Mask](int Idx) { return Idx >= (int)Mask.size(); })) {
const unsigned EltSize = VT.getScalarSizeInBits();
const unsigned MinVLMAX = Subtarget.getRealMinVLen() / EltSize;
if (NumElts < MinVLMAX) {
MVT ConcatVT = VT.getDoubleNumVectorElementsVT();
SDValue Concat = DAG.getNode(ISD::CONCAT_VECTORS, DL, ConcatVT, V1, V2);
SDValue Res =
lowerVZIP(Opc, Concat, DAG.getUNDEF(ConcatVT), DL, DAG, Subtarget);
return DAG.getExtractSubvector(DL, VT, Res, 0);
}
SDValue Lo = lowerVZIP(Opc, V1, DAG.getUNDEF(VT), DL, DAG, Subtarget);
SDValue Hi = lowerVZIP(Opc, V2, DAG.getUNDEF(VT), DL, DAG, Subtarget);
MVT SubVT = VT.getHalfNumVectorElementsVT();
return DAG.getNode(ISD::CONCAT_VECTORS, DL, VT,
DAG.getExtractSubvector(DL, SubVT, Lo, 0),
DAG.getExtractSubvector(DL, SubVT, Hi, 0));
}
}
if (SDValue V =
lowerVECTOR_SHUFFLEAsVSlideup(DL, VT, V1, V2, Mask, Subtarget, DAG))
return V;
// Detect an interleave shuffle and lower to
// (vmaccu.vx (vwaddu.vx lohalf(V1), lohalf(V2)), lohalf(V2), (2^eltbits - 1))
int EvenSrc, OddSrc;
if (isInterleaveShuffle(Mask, VT, EvenSrc, OddSrc, Subtarget) &&
!(NumElts == 2 &&
ShuffleVectorInst::isSingleSourceMask(Mask, Mask.size()))) {
// Extract the halves of the vectors.
MVT HalfVT = VT.getHalfNumVectorElementsVT();
// Recognize if one half is actually undef; the matching above will
// otherwise reuse the even stream for the undef one. This improves
// spread(2) shuffles.
bool LaneIsUndef[2] = { true, true};
for (const auto &[Idx, M] : enumerate(Mask))
LaneIsUndef[Idx % 2] &= (M == -1);
int Size = Mask.size();
SDValue EvenV, OddV;
if (LaneIsUndef[0]) {
EvenV = DAG.getUNDEF(HalfVT);
} else {
assert(EvenSrc >= 0 && "Undef source?");
EvenV = (EvenSrc / Size) == 0 ? V1 : V2;
EvenV = DAG.getExtractSubvector(DL, HalfVT, EvenV, EvenSrc % Size);
}
if (LaneIsUndef[1]) {
OddV = DAG.getUNDEF(HalfVT);
} else {
assert(OddSrc >= 0 && "Undef source?");
OddV = (OddSrc / Size) == 0 ? V1 : V2;
OddV = DAG.getExtractSubvector(DL, HalfVT, OddV, OddSrc % Size);
}
// Prefer vzip2a if available.
// TODO: Extend to matching zip2b if EvenSrc and OddSrc allow.
if (Subtarget.hasVendorXRivosVizip()) {
EvenV = DAG.getInsertSubvector(DL, DAG.getUNDEF(VT), EvenV, 0);
OddV = DAG.getInsertSubvector(DL, DAG.getUNDEF(VT), OddV, 0);
return lowerVZIP(RISCVISD::RI_VZIP2A_VL, EvenV, OddV, DL, DAG, Subtarget);
}
return getWideningInterleave(EvenV, OddV, DL, DAG, Subtarget);
}
// Recognize a pattern which can handled via a pair of vslideup/vslidedown
// instructions (in any combination) with masking on the second instruction.
// Also handles masked slides into an identity source, and single slides
// without masking. Avoid matching bit rotates (which are not also element
// rotates) as slide pairs. This is a performance heuristic, not a
// functional check.
std::array<std::pair<int, int>, 2> SrcInfo;
unsigned RotateAmt;
MVT RotateVT;
if (::isMaskedSlidePair(Mask, SrcInfo) &&
(isElementRotate(SrcInfo, NumElts) ||
!isLegalBitRotate(Mask, VT, Subtarget, RotateVT, RotateAmt))) {
SDValue Sources[2];
auto GetSourceFor = [&](const std::pair<int, int> &Info) {
int SrcIdx = Info.first;
assert(SrcIdx == 0 || SrcIdx == 1);
SDValue &Src = Sources[SrcIdx];
if (!Src) {
SDValue SrcV = SrcIdx == 0 ? V1 : V2;
Src = convertToScalableVector(ContainerVT, SrcV, DAG, Subtarget);
}
return Src;
};
auto GetSlide = [&](const std::pair<int, int> &Src, SDValue Mask,
SDValue Passthru) {
auto [TrueMask, VL] = TrueMaskVL;
SDValue SrcV = GetSourceFor(Src);
int SlideAmt = Src.second;
if (SlideAmt == 0) {
// Should never be second operation
assert(Mask == TrueMask);
return SrcV;
}
if (SlideAmt < 0)
return getVSlidedown(DAG, Subtarget, DL, ContainerVT, Passthru, SrcV,
DAG.getConstant(-SlideAmt, DL, XLenVT), Mask, VL,
RISCVVType::TAIL_AGNOSTIC);
return getVSlideup(DAG, Subtarget, DL, ContainerVT, Passthru, SrcV,
DAG.getConstant(SlideAmt, DL, XLenVT), Mask, VL,
RISCVVType::TAIL_AGNOSTIC);
};
if (SrcInfo[1].first == -1) {
SDValue Res = DAG.getUNDEF(ContainerVT);
Res = GetSlide(SrcInfo[0], TrueMask, Res);
return convertFromScalableVector(VT, Res, DAG, Subtarget);
}
if (Subtarget.hasVendorXRivosVizip()) {
bool TryWiden = false;
unsigned Factor;
if (isZipEven(SrcInfo, Mask, Factor)) {
if (Factor == 1) {
SDValue Src1 = SrcInfo[0].first == 0 ? V1 : V2;
SDValue Src2 = SrcInfo[1].first == 0 ? V1 : V2;
return lowerVZIP(RISCVISD::RI_VZIPEVEN_VL, Src1, Src2, DL, DAG,
Subtarget);
}
TryWiden = true;
}
if (isZipOdd(SrcInfo, Mask, Factor)) {
if (Factor == 1) {
SDValue Src1 = SrcInfo[1].first == 0 ? V1 : V2;
SDValue Src2 = SrcInfo[0].first == 0 ? V1 : V2;
return lowerVZIP(RISCVISD::RI_VZIPODD_VL, Src1, Src2, DL, DAG,
Subtarget);
}
TryWiden = true;
}
// If we found a widening oppurtunity which would let us form a
// zipeven or zipodd, use the generic code to widen the shuffle
// and recurse through this logic.
if (TryWiden)
if (SDValue V = tryWidenMaskForShuffle(Op, DAG))
return V;
}
// Build the mask. Note that vslideup unconditionally preserves elements
// below the slide amount in the destination, and thus those elements are
// undefined in the mask. If the mask ends up all true (or undef), it
// will be folded away by general logic.
SmallVector<SDValue> MaskVals;
for (const auto &[Idx, M] : enumerate(Mask)) {
if (M < 0 ||
(SrcInfo[1].second > 0 && Idx < (unsigned)SrcInfo[1].second)) {
MaskVals.push_back(DAG.getUNDEF(XLenVT));
continue;
}
int Src = M >= (int)NumElts;
int Diff = (int)Idx - (M % NumElts);
bool C = Src == SrcInfo[1].first && Diff == SrcInfo[1].second;
assert(C ^ (Src == SrcInfo[0].first && Diff == SrcInfo[0].second) &&
"Must match exactly one of the two slides");
MaskVals.push_back(DAG.getConstant(C, DL, XLenVT));
}
assert(MaskVals.size() == NumElts && "Unexpected select-like shuffle");
MVT MaskVT = MVT::getVectorVT(MVT::i1, NumElts);
SDValue SelectMask = convertToScalableVector(
ContainerVT.changeVectorElementType(MVT::i1),
DAG.getBuildVector(MaskVT, DL, MaskVals), DAG, Subtarget);
SDValue Res = DAG.getUNDEF(ContainerVT);
Res = GetSlide(SrcInfo[0], TrueMask, Res);
Res = GetSlide(SrcInfo[1], SelectMask, Res);
return convertFromScalableVector(VT, Res, DAG, Subtarget);
}
// Handle any remaining single source shuffles
assert(!V1.isUndef() && "Unexpected shuffle canonicalization");
if (V2.isUndef()) {
// We might be able to express the shuffle as a bitrotate. But even if we
// don't have Zvkb and have to expand, the expanded sequence of approx. 2
// shifts and a vor will have a higher throughput than a vrgather.
if (SDValue V = lowerVECTOR_SHUFFLEAsRotate(SVN, DAG, Subtarget))
return V;
if (SDValue V = lowerVECTOR_SHUFFLEAsVRGatherVX(SVN, Subtarget, DAG))
return V;
// Match a spread(4,8) which can be done via extend and shift. Spread(2)
// is fully covered in interleave(2) above, so it is ignored here.
if (VT.getScalarSizeInBits() < Subtarget.getELen()) {
unsigned MaxFactor = Subtarget.getELen() / VT.getScalarSizeInBits();
assert(MaxFactor == 2 || MaxFactor == 4 || MaxFactor == 8);
for (unsigned Factor = 4; Factor <= MaxFactor; Factor <<= 1) {
unsigned Index;
if (RISCVTargetLowering::isSpreadMask(Mask, Factor, Index)) {
MVT NarrowVT =
MVT::getVectorVT(VT.getVectorElementType(), NumElts / Factor);
SDValue Src = DAG.getExtractSubvector(DL, NarrowVT, V1, 0);
return getWideningSpread(Src, Factor, Index, DL, DAG);
}
}
}
// If only a prefix of the source elements influence a prefix of the
// destination elements, try to see if we can reduce the required LMUL
unsigned MinVLen = Subtarget.getRealMinVLen();
unsigned MinVLMAX = MinVLen / VT.getScalarSizeInBits();
if (NumElts > MinVLMAX) {
unsigned MaxIdx = 0;
for (auto [I, M] : enumerate(Mask)) {
if (M == -1)
continue;
MaxIdx = std::max(std::max((unsigned)I, (unsigned)M), MaxIdx);
}
unsigned NewNumElts =
std::max((uint64_t)MinVLMAX, PowerOf2Ceil(MaxIdx + 1));
if (NewNumElts != NumElts) {
MVT NewVT = MVT::getVectorVT(VT.getVectorElementType(), NewNumElts);
V1 = DAG.getExtractSubvector(DL, NewVT, V1, 0);
SDValue Res = DAG.getVectorShuffle(NewVT, DL, V1, DAG.getUNDEF(NewVT),
Mask.take_front(NewNumElts));
return DAG.getInsertSubvector(DL, DAG.getUNDEF(VT), Res, 0);
}
}
// Before hitting generic lowering fallbacks, try to widen the mask
// to a wider SEW.
if (SDValue V = tryWidenMaskForShuffle(Op, DAG))
return V;
// Can we generate a vcompress instead of a vrgather? These scale better
// at high LMUL, at the cost of not being able to fold a following select
// into them. The mask constants are also smaller than the index vector
// constants, and thus easier to materialize.
if (isCompressMask(Mask)) {
SmallVector<SDValue> MaskVals(NumElts,
DAG.getConstant(false, DL, XLenVT));
for (auto Idx : Mask) {
if (Idx == -1)
break;
assert(Idx >= 0 && (unsigned)Idx < NumElts);
MaskVals[Idx] = DAG.getConstant(true, DL, XLenVT);
}
MVT MaskVT = MVT::getVectorVT(MVT::i1, NumElts);
SDValue CompressMask = DAG.getBuildVector(MaskVT, DL, MaskVals);
return DAG.getNode(ISD::VECTOR_COMPRESS, DL, VT, V1, CompressMask,
DAG.getUNDEF(VT));
}
if (VT.getScalarSizeInBits() == 8 &&
any_of(Mask, [&](const auto &Idx) { return Idx > 255; })) {
// On such a vector we're unable to use i8 as the index type.
// FIXME: We could promote the index to i16 and use vrgatherei16, but that
// may involve vector splitting if we're already at LMUL=8, or our
// user-supplied maximum fixed-length LMUL.
return SDValue();
}
// Base case for the two operand recursion below - handle the worst case
// single source shuffle.
unsigned GatherVVOpc = RISCVISD::VRGATHER_VV_VL;
MVT IndexVT = VT.changeTypeToInteger();
// Since we can't introduce illegal index types at this stage, use i16 and
// vrgatherei16 if the corresponding index type for plain vrgather is greater
// than XLenVT.
if (IndexVT.getScalarType().bitsGT(XLenVT)) {
GatherVVOpc = RISCVISD::VRGATHEREI16_VV_VL;
IndexVT = IndexVT.changeVectorElementType(MVT::i16);
}
// If the mask allows, we can do all the index computation in 16 bits. This
// requires less work and less register pressure at high LMUL, and creates
// smaller constants which may be cheaper to materialize.
if (IndexVT.getScalarType().bitsGT(MVT::i16) && isUInt<16>(NumElts - 1) &&
(IndexVT.getSizeInBits() / Subtarget.getRealMinVLen()) > 1) {
GatherVVOpc = RISCVISD::VRGATHEREI16_VV_VL;
IndexVT = IndexVT.changeVectorElementType(MVT::i16);
}
MVT IndexContainerVT =
ContainerVT.changeVectorElementType(IndexVT.getScalarType());
V1 = convertToScalableVector(ContainerVT, V1, DAG, Subtarget);
SmallVector<SDValue> GatherIndicesLHS;
for (int MaskIndex : Mask) {
bool IsLHSIndex = MaskIndex < (int)NumElts && MaskIndex >= 0;
GatherIndicesLHS.push_back(IsLHSIndex
? DAG.getConstant(MaskIndex, DL, XLenVT)
: DAG.getUNDEF(XLenVT));
}
SDValue LHSIndices = DAG.getBuildVector(IndexVT, DL, GatherIndicesLHS);
LHSIndices =
convertToScalableVector(IndexContainerVT, LHSIndices, DAG, Subtarget);
// At m1 and less, there's no point trying any of the high LMUL splitting
// techniques. TODO: Should we reconsider this for DLEN < VLEN?
if (NumElts <= MinVLMAX) {
SDValue Gather = DAG.getNode(GatherVVOpc, DL, ContainerVT, V1, LHSIndices,
DAG.getUNDEF(ContainerVT), TrueMask, VL);
return convertFromScalableVector(VT, Gather, DAG, Subtarget);
}
const MVT M1VT = RISCVTargetLowering::getM1VT(ContainerVT);
EVT SubIndexVT = M1VT.changeVectorElementType(IndexVT.getScalarType());
auto [InnerTrueMask, InnerVL] =
getDefaultScalableVLOps(M1VT, DL, DAG, Subtarget);
int N =
ContainerVT.getVectorMinNumElements() / M1VT.getVectorMinNumElements();
assert(isPowerOf2_32(N) && N <= 8);
// If we have a locally repeating mask, then we can reuse the first
// register in the index register group for all registers within the
// source register group. TODO: This generalizes to m2, and m4.
if (isLocalRepeatingShuffle(Mask, MinVLMAX)) {
SDValue SubIndex = DAG.getExtractSubvector(DL, SubIndexVT, LHSIndices, 0);
SDValue Gather = DAG.getUNDEF(ContainerVT);
for (int i = 0; i < N; i++) {
unsigned SubIdx = M1VT.getVectorMinNumElements() * i;
SDValue SubV1 = DAG.getExtractSubvector(DL, M1VT, V1, SubIdx);
SDValue SubVec =
DAG.getNode(GatherVVOpc, DL, M1VT, SubV1, SubIndex,
DAG.getUNDEF(M1VT), InnerTrueMask, InnerVL);
Gather = DAG.getInsertSubvector(DL, Gather, SubVec, SubIdx);
}
return convertFromScalableVector(VT, Gather, DAG, Subtarget);
}
// If we have a shuffle which only uses the first register in our source
// register group, and repeats the same index across all spans, we can
// use a single vrgather (and possibly some register moves).
// TODO: This can be generalized for m2 or m4, or for any shuffle for
// which we can do a linear number of shuffles to form an m1 which
// contains all the output elements.
if (isLowSourceShuffle(Mask, MinVLMAX) &&
isSpanSplatShuffle(Mask, MinVLMAX)) {
SDValue SubV1 = DAG.getExtractSubvector(DL, M1VT, V1, 0);
SDValue SubIndex = DAG.getExtractSubvector(DL, SubIndexVT, LHSIndices, 0);
SDValue SubVec = DAG.getNode(GatherVVOpc, DL, M1VT, SubV1, SubIndex,
DAG.getUNDEF(M1VT), InnerTrueMask, InnerVL);
SDValue Gather = DAG.getUNDEF(ContainerVT);
for (int i = 0; i < N; i++)
Gather = DAG.getInsertSubvector(DL, Gather, SubVec,
M1VT.getVectorMinNumElements() * i);
return convertFromScalableVector(VT, Gather, DAG, Subtarget);
}
// If we have a shuffle which only uses the first register in our
// source register group, we can do a linear number of m1 vrgathers
// reusing the same source register (but with different indices)
// TODO: This can be generalized for m2 or m4, or for any shuffle
// for which we can do a vslidedown followed by this expansion.
if (isLowSourceShuffle(Mask, MinVLMAX)) {
SDValue SlideAmt =
DAG.getElementCount(DL, XLenVT, M1VT.getVectorElementCount());
SDValue SubV1 = DAG.getExtractSubvector(DL, M1VT, V1, 0);
SDValue Gather = DAG.getUNDEF(ContainerVT);
for (int i = 0; i < N; i++) {
if (i != 0)
LHSIndices = getVSlidedown(DAG, Subtarget, DL, IndexContainerVT,
DAG.getUNDEF(IndexContainerVT), LHSIndices,
SlideAmt, TrueMask, VL);
SDValue SubIndex =
DAG.getExtractSubvector(DL, SubIndexVT, LHSIndices, 0);
SDValue SubVec =
DAG.getNode(GatherVVOpc, DL, M1VT, SubV1, SubIndex,
DAG.getUNDEF(M1VT), InnerTrueMask, InnerVL);
Gather = DAG.getInsertSubvector(DL, Gather, SubVec,
M1VT.getVectorMinNumElements() * i);
}
return convertFromScalableVector(VT, Gather, DAG, Subtarget);
}
// Fallback to generic vrgather if we can't find anything better.
// On many machines, this will be O(LMUL^2)
SDValue Gather = DAG.getNode(GatherVVOpc, DL, ContainerVT, V1, LHSIndices,
DAG.getUNDEF(ContainerVT), TrueMask, VL);
return convertFromScalableVector(VT, Gather, DAG, Subtarget);
}
// As a backup, shuffles can be lowered via a vrgather instruction, possibly
// merged with a second vrgather.
SmallVector<int> ShuffleMaskLHS, ShuffleMaskRHS;
// Now construct the mask that will be used by the blended vrgather operation.
// Construct the appropriate indices into each vector.
for (int MaskIndex : Mask) {
bool IsLHSOrUndefIndex = MaskIndex < (int)NumElts;
ShuffleMaskLHS.push_back(IsLHSOrUndefIndex && MaskIndex >= 0
? MaskIndex : -1);
ShuffleMaskRHS.push_back(IsLHSOrUndefIndex ? -1 : (MaskIndex - NumElts));
}
// If the mask indices are disjoint between the two sources, we can lower it
// as a vselect + a single source vrgather.vv. Don't do this if we think the
// operands may end up being lowered to something cheaper than a vrgather.vv.
if (!DAG.isSplatValue(V2) && !DAG.isSplatValue(V1) &&
!ShuffleVectorSDNode::isSplatMask(ShuffleMaskLHS) &&
!ShuffleVectorSDNode::isSplatMask(ShuffleMaskRHS) &&
!ShuffleVectorInst::isIdentityMask(ShuffleMaskLHS, NumElts) &&
!ShuffleVectorInst::isIdentityMask(ShuffleMaskRHS, NumElts))
if (SDValue V = lowerDisjointIndicesShuffle(SVN, DAG, Subtarget))
return V;
// Before hitting generic lowering fallbacks, try to widen the mask
// to a wider SEW.
if (SDValue V = tryWidenMaskForShuffle(Op, DAG))
return V;
// Try to pick a profitable operand order.
bool SwapOps = DAG.isSplatValue(V2) && !DAG.isSplatValue(V1);
SwapOps = SwapOps ^ ShuffleVectorInst::isIdentityMask(ShuffleMaskRHS, NumElts);
// Recursively invoke lowering for each operand if we had two
// independent single source shuffles, and then combine the result via a
// vselect. Note that the vselect will likely be folded back into the
// second permute (vrgather, or other) by the post-isel combine.
V1 = DAG.getVectorShuffle(VT, DL, V1, DAG.getUNDEF(VT), ShuffleMaskLHS);
V2 = DAG.getVectorShuffle(VT, DL, V2, DAG.getUNDEF(VT), ShuffleMaskRHS);
SmallVector<SDValue> MaskVals;
for (int MaskIndex : Mask) {
bool SelectMaskVal = (MaskIndex < (int)NumElts) ^ !SwapOps;
MaskVals.push_back(DAG.getConstant(SelectMaskVal, DL, XLenVT));
}
assert(MaskVals.size() == NumElts && "Unexpected select-like shuffle");
MVT MaskVT = MVT::getVectorVT(MVT::i1, NumElts);
SDValue SelectMask = DAG.getBuildVector(MaskVT, DL, MaskVals);
if (SwapOps)
return DAG.getNode(ISD::VSELECT, DL, VT, SelectMask, V1, V2);
return DAG.getNode(ISD::VSELECT, DL, VT, SelectMask, V2, V1);
}
bool RISCVTargetLowering::isShuffleMaskLegal(ArrayRef<int> M, EVT VT) const {
// Only support legal VTs for other shuffles for now.
if (!isTypeLegal(VT))
return false;
// Support splats for any type. These should type legalize well.
if (ShuffleVectorSDNode::isSplatMask(M))
return true;
const unsigned NumElts = M.size();
MVT SVT = VT.getSimpleVT();
// Not for i1 vectors.
if (SVT.getScalarType() == MVT::i1)
return false;
std::array<std::pair<int, int>, 2> SrcInfo;
int Dummy1, Dummy2;
return ShuffleVectorInst::isReverseMask(M, NumElts) ||
(::isMaskedSlidePair(M, SrcInfo) &&
isElementRotate(SrcInfo, NumElts)) ||
isInterleaveShuffle(M, SVT, Dummy1, Dummy2, Subtarget);
}
// Lower CTLZ_ZERO_UNDEF or CTTZ_ZERO_UNDEF by converting to FP and extracting
// the exponent.
SDValue
RISCVTargetLowering::lowerCTLZ_CTTZ_ZERO_UNDEF(SDValue Op,
SelectionDAG &DAG) const {
MVT VT = Op.getSimpleValueType();
unsigned EltSize = VT.getScalarSizeInBits();
SDValue Src = Op.getOperand(0);
SDLoc DL(Op);
MVT ContainerVT = VT;
SDValue Mask, VL;
if (Op->isVPOpcode()) {
Mask = Op.getOperand(1);
if (VT.isFixedLengthVector())
Mask = convertToScalableVector(getMaskTypeFor(ContainerVT), Mask, DAG,
Subtarget);
VL = Op.getOperand(2);
}
// We choose FP type that can represent the value if possible. Otherwise, we
// use rounding to zero conversion for correct exponent of the result.
// TODO: Use f16 for i8 when possible?
MVT FloatEltVT = (EltSize >= 32) ? MVT::f64 : MVT::f32;
if (!isTypeLegal(MVT::getVectorVT(FloatEltVT, VT.getVectorElementCount())))
FloatEltVT = MVT::f32;
MVT FloatVT = MVT::getVectorVT(FloatEltVT, VT.getVectorElementCount());
// Legal types should have been checked in the RISCVTargetLowering
// constructor.
// TODO: Splitting may make sense in some cases.
assert(DAG.getTargetLoweringInfo().isTypeLegal(FloatVT) &&
"Expected legal float type!");
// For CTTZ_ZERO_UNDEF, we need to extract the lowest set bit using X & -X.
// The trailing zero count is equal to log2 of this single bit value.
if (Op.getOpcode() == ISD::CTTZ_ZERO_UNDEF) {
SDValue Neg = DAG.getNegative(Src, DL, VT);
Src = DAG.getNode(ISD::AND, DL, VT, Src, Neg);
} else if (Op.getOpcode() == ISD::VP_CTTZ_ZERO_UNDEF) {
SDValue Neg = DAG.getNode(ISD::VP_SUB, DL, VT, DAG.getConstant(0, DL, VT),
Src, Mask, VL);
Src = DAG.getNode(ISD::VP_AND, DL, VT, Src, Neg, Mask, VL);
}
// We have a legal FP type, convert to it.
SDValue FloatVal;
if (FloatVT.bitsGT(VT)) {
if (Op->isVPOpcode())
FloatVal = DAG.getNode(ISD::VP_UINT_TO_FP, DL, FloatVT, Src, Mask, VL);
else
FloatVal = DAG.getNode(ISD::UINT_TO_FP, DL, FloatVT, Src);
} else {
// Use RTZ to avoid rounding influencing exponent of FloatVal.
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
Src = convertToScalableVector(ContainerVT, Src, DAG, Subtarget);
}
if (!Op->isVPOpcode())
std::tie(Mask, VL) = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
SDValue RTZRM =
DAG.getTargetConstant(RISCVFPRndMode::RTZ, DL, Subtarget.getXLenVT());
MVT ContainerFloatVT =
MVT::getVectorVT(FloatEltVT, ContainerVT.getVectorElementCount());
FloatVal = DAG.getNode(RISCVISD::VFCVT_RM_F_XU_VL, DL, ContainerFloatVT,
Src, Mask, RTZRM, VL);
if (VT.isFixedLengthVector())
FloatVal = convertFromScalableVector(FloatVT, FloatVal, DAG, Subtarget);
}
// Bitcast to integer and shift the exponent to the LSB.
EVT IntVT = FloatVT.changeVectorElementTypeToInteger();
SDValue Bitcast = DAG.getBitcast(IntVT, FloatVal);
unsigned ShiftAmt = FloatEltVT == MVT::f64 ? 52 : 23;
SDValue Exp;
// Restore back to original type. Truncation after SRL is to generate vnsrl.
if (Op->isVPOpcode()) {
Exp = DAG.getNode(ISD::VP_SRL, DL, IntVT, Bitcast,
DAG.getConstant(ShiftAmt, DL, IntVT), Mask, VL);
Exp = DAG.getVPZExtOrTrunc(DL, VT, Exp, Mask, VL);
} else {
Exp = DAG.getNode(ISD::SRL, DL, IntVT, Bitcast,
DAG.getConstant(ShiftAmt, DL, IntVT));
if (IntVT.bitsLT(VT))
Exp = DAG.getNode(ISD::ZERO_EXTEND, DL, VT, Exp);
else if (IntVT.bitsGT(VT))
Exp = DAG.getNode(ISD::TRUNCATE, DL, VT, Exp);
}
// The exponent contains log2 of the value in biased form.
unsigned ExponentBias = FloatEltVT == MVT::f64 ? 1023 : 127;
// For trailing zeros, we just need to subtract the bias.
if (Op.getOpcode() == ISD::CTTZ_ZERO_UNDEF)
return DAG.getNode(ISD::SUB, DL, VT, Exp,
DAG.getConstant(ExponentBias, DL, VT));
if (Op.getOpcode() == ISD::VP_CTTZ_ZERO_UNDEF)
return DAG.getNode(ISD::VP_SUB, DL, VT, Exp,
DAG.getConstant(ExponentBias, DL, VT), Mask, VL);
// For leading zeros, we need to remove the bias and convert from log2 to
// leading zeros. We can do this by subtracting from (Bias + (EltSize - 1)).
unsigned Adjust = ExponentBias + (EltSize - 1);
SDValue Res;
if (Op->isVPOpcode())
Res = DAG.getNode(ISD::VP_SUB, DL, VT, DAG.getConstant(Adjust, DL, VT), Exp,
Mask, VL);
else
Res = DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(Adjust, DL, VT), Exp);
// The above result with zero input equals to Adjust which is greater than
// EltSize. Hence, we can do min(Res, EltSize) for CTLZ.
if (Op.getOpcode() == ISD::CTLZ)
Res = DAG.getNode(ISD::UMIN, DL, VT, Res, DAG.getConstant(EltSize, DL, VT));
else if (Op.getOpcode() == ISD::VP_CTLZ)
Res = DAG.getNode(ISD::VP_UMIN, DL, VT, Res,
DAG.getConstant(EltSize, DL, VT), Mask, VL);
return Res;
}
SDValue RISCVTargetLowering::lowerVPCttzElements(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
SDValue Source = Op->getOperand(0);
MVT SrcVT = Source.getSimpleValueType();
SDValue Mask = Op->getOperand(1);
SDValue EVL = Op->getOperand(2);
if (SrcVT.isFixedLengthVector()) {
MVT ContainerVT = getContainerForFixedLengthVector(SrcVT);
Source = convertToScalableVector(ContainerVT, Source, DAG, Subtarget);
Mask = convertToScalableVector(getMaskTypeFor(ContainerVT), Mask, DAG,
Subtarget);
SrcVT = ContainerVT;
}
// Convert to boolean vector.
if (SrcVT.getScalarType() != MVT::i1) {
SDValue AllZero = DAG.getConstant(0, DL, SrcVT);
SrcVT = MVT::getVectorVT(MVT::i1, SrcVT.getVectorElementCount());
Source = DAG.getNode(RISCVISD::SETCC_VL, DL, SrcVT,
{Source, AllZero, DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(SrcVT), Mask, EVL});
}
SDValue Res = DAG.getNode(RISCVISD::VFIRST_VL, DL, XLenVT, Source, Mask, EVL);
if (Op->getOpcode() == ISD::VP_CTTZ_ELTS_ZERO_UNDEF)
// In this case, we can interpret poison as -1, so nothing to do further.
return Res;
// Convert -1 to VL.
SDValue SetCC =
DAG.getSetCC(DL, XLenVT, Res, DAG.getConstant(0, DL, XLenVT), ISD::SETLT);
Res = DAG.getSelect(DL, XLenVT, SetCC, EVL, Res);
return DAG.getNode(ISD::TRUNCATE, DL, Op.getValueType(), Res);
}
// While RVV has alignment restrictions, we should always be able to load as a
// legal equivalently-sized byte-typed vector instead. This method is
// responsible for re-expressing a ISD::LOAD via a correctly-aligned type. If
// the load is already correctly-aligned, it returns SDValue().
SDValue RISCVTargetLowering::expandUnalignedRVVLoad(SDValue Op,
SelectionDAG &DAG) const {
auto *Load = cast<LoadSDNode>(Op);
assert(Load && Load->getMemoryVT().isVector() && "Expected vector load");
if (allowsMemoryAccessForAlignment(*DAG.getContext(), DAG.getDataLayout(),
Load->getMemoryVT(),
*Load->getMemOperand()))
return SDValue();
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
unsigned EltSizeBits = VT.getScalarSizeInBits();
assert((EltSizeBits == 16 || EltSizeBits == 32 || EltSizeBits == 64) &&
"Unexpected unaligned RVV load type");
MVT NewVT =
MVT::getVectorVT(MVT::i8, VT.getVectorElementCount() * (EltSizeBits / 8));
assert(NewVT.isValid() &&
"Expecting equally-sized RVV vector types to be legal");
SDValue L = DAG.getLoad(NewVT, DL, Load->getChain(), Load->getBasePtr(),
Load->getPointerInfo(), Load->getBaseAlign(),
Load->getMemOperand()->getFlags());
return DAG.getMergeValues({DAG.getBitcast(VT, L), L.getValue(1)}, DL);
}
// While RVV has alignment restrictions, we should always be able to store as a
// legal equivalently-sized byte-typed vector instead. This method is
// responsible for re-expressing a ISD::STORE via a correctly-aligned type. It
// returns SDValue() if the store is already correctly aligned.
SDValue RISCVTargetLowering::expandUnalignedRVVStore(SDValue Op,
SelectionDAG &DAG) const {
auto *Store = cast<StoreSDNode>(Op);
assert(Store && Store->getValue().getValueType().isVector() &&
"Expected vector store");
if (allowsMemoryAccessForAlignment(*DAG.getContext(), DAG.getDataLayout(),
Store->getMemoryVT(),
*Store->getMemOperand()))
return SDValue();
SDLoc DL(Op);
SDValue StoredVal = Store->getValue();
MVT VT = StoredVal.getSimpleValueType();
unsigned EltSizeBits = VT.getScalarSizeInBits();
assert((EltSizeBits == 16 || EltSizeBits == 32 || EltSizeBits == 64) &&
"Unexpected unaligned RVV store type");
MVT NewVT =
MVT::getVectorVT(MVT::i8, VT.getVectorElementCount() * (EltSizeBits / 8));
assert(NewVT.isValid() &&
"Expecting equally-sized RVV vector types to be legal");
StoredVal = DAG.getBitcast(NewVT, StoredVal);
return DAG.getStore(Store->getChain(), DL, StoredVal, Store->getBasePtr(),
Store->getPointerInfo(), Store->getBaseAlign(),
Store->getMemOperand()->getFlags());
}
static SDValue lowerConstant(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(Op.getValueType() == MVT::i64 && "Unexpected VT");
int64_t Imm = cast<ConstantSDNode>(Op)->getSExtValue();
// All simm32 constants should be handled by isel.
// NOTE: The getMaxBuildIntsCost call below should return a value >= 2 making
// this check redundant, but small immediates are common so this check
// should have better compile time.
if (isInt<32>(Imm))
return Op;
// We only need to cost the immediate, if constant pool lowering is enabled.
if (!Subtarget.useConstantPoolForLargeInts())
return Op;
RISCVMatInt::InstSeq Seq = RISCVMatInt::generateInstSeq(Imm, Subtarget);
if (Seq.size() <= Subtarget.getMaxBuildIntsCost())
return Op;
// Optimizations below are disabled for opt size. If we're optimizing for
// size, use a constant pool.
if (DAG.shouldOptForSize())
return SDValue();
// Special case. See if we can build the constant as (ADD (SLLI X, C), X) do
// that if it will avoid a constant pool.
// It will require an extra temporary register though.
// If we have Zba we can use (ADD_UW X, (SLLI X, 32)) to handle cases where
// low and high 32 bits are the same and bit 31 and 63 are set.
unsigned ShiftAmt, AddOpc;
RISCVMatInt::InstSeq SeqLo =
RISCVMatInt::generateTwoRegInstSeq(Imm, Subtarget, ShiftAmt, AddOpc);
if (!SeqLo.empty() && (SeqLo.size() + 2) <= Subtarget.getMaxBuildIntsCost())
return Op;
return SDValue();
}
SDValue RISCVTargetLowering::lowerConstantFP(SDValue Op,
SelectionDAG &DAG) const {
MVT VT = Op.getSimpleValueType();
const APFloat &Imm = cast<ConstantFPSDNode>(Op)->getValueAPF();
// Can this constant be selected by a Zfa FLI instruction?
bool Negate = false;
int Index = getLegalZfaFPImm(Imm, VT);
// If the constant is negative, try negating.
if (Index < 0 && Imm.isNegative()) {
Index = getLegalZfaFPImm(-Imm, VT);
Negate = true;
}
// If we couldn't find a FLI lowering, fall back to generic code.
if (Index < 0)
return SDValue();
// Emit an FLI+FNEG. We use a custom node to hide from constant folding.
SDLoc DL(Op);
SDValue Const =
DAG.getNode(RISCVISD::FLI, DL, VT,
DAG.getTargetConstant(Index, DL, Subtarget.getXLenVT()));
if (!Negate)
return Const;
return DAG.getNode(ISD::FNEG, DL, VT, Const);
}
static SDValue LowerPREFETCH(SDValue Op, const RISCVSubtarget &Subtarget,
SelectionDAG &DAG) {
unsigned IsData = Op.getConstantOperandVal(4);
// mips-p8700 we support data prefetch for now.
if (Subtarget.hasVendorXMIPSCBOP() && !IsData)
return Op.getOperand(0);
return Op;
}
static SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDLoc dl(Op);
AtomicOrdering FenceOrdering =
static_cast<AtomicOrdering>(Op.getConstantOperandVal(1));
SyncScope::ID FenceSSID =
static_cast<SyncScope::ID>(Op.getConstantOperandVal(2));
if (Subtarget.hasStdExtZtso()) {
// The only fence that needs an instruction is a sequentially-consistent
// cross-thread fence.
if (FenceOrdering == AtomicOrdering::SequentiallyConsistent &&
FenceSSID == SyncScope::System)
return Op;
// MEMBARRIER is a compiler barrier; it codegens to a no-op.
return DAG.getNode(ISD::MEMBARRIER, dl, MVT::Other, Op.getOperand(0));
}
// singlethread fences only synchronize with signal handlers on the same
// thread and thus only need to preserve instruction order, not actually
// enforce memory ordering.
if (FenceSSID == SyncScope::SingleThread)
// MEMBARRIER is a compiler barrier; it codegens to a no-op.
return DAG.getNode(ISD::MEMBARRIER, dl, MVT::Other, Op.getOperand(0));
return Op;
}
SDValue RISCVTargetLowering::LowerIS_FPCLASS(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
unsigned Check = Op.getConstantOperandVal(1);
unsigned TDCMask = 0;
if (Check & fcSNan)
TDCMask |= RISCV::FPMASK_Signaling_NaN;
if (Check & fcQNan)
TDCMask |= RISCV::FPMASK_Quiet_NaN;
if (Check & fcPosInf)
TDCMask |= RISCV::FPMASK_Positive_Infinity;
if (Check & fcNegInf)
TDCMask |= RISCV::FPMASK_Negative_Infinity;
if (Check & fcPosNormal)
TDCMask |= RISCV::FPMASK_Positive_Normal;
if (Check & fcNegNormal)
TDCMask |= RISCV::FPMASK_Negative_Normal;
if (Check & fcPosSubnormal)
TDCMask |= RISCV::FPMASK_Positive_Subnormal;
if (Check & fcNegSubnormal)
TDCMask |= RISCV::FPMASK_Negative_Subnormal;
if (Check & fcPosZero)
TDCMask |= RISCV::FPMASK_Positive_Zero;
if (Check & fcNegZero)
TDCMask |= RISCV::FPMASK_Negative_Zero;
bool IsOneBitMask = isPowerOf2_32(TDCMask);
SDValue TDCMaskV = DAG.getConstant(TDCMask, DL, XLenVT);
if (VT.isVector()) {
SDValue Op0 = Op.getOperand(0);
MVT VT0 = Op.getOperand(0).getSimpleValueType();
if (VT.isScalableVector()) {
MVT DstVT = VT0.changeVectorElementTypeToInteger();
auto [Mask, VL] = getDefaultScalableVLOps(VT0, DL, DAG, Subtarget);
if (Op.getOpcode() == ISD::VP_IS_FPCLASS) {
Mask = Op.getOperand(2);
VL = Op.getOperand(3);
}
SDValue FPCLASS = DAG.getNode(RISCVISD::FCLASS_VL, DL, DstVT, Op0, Mask,
VL, Op->getFlags());
if (IsOneBitMask)
return DAG.getSetCC(DL, VT, FPCLASS,
DAG.getConstant(TDCMask, DL, DstVT),
ISD::CondCode::SETEQ);
SDValue AND = DAG.getNode(ISD::AND, DL, DstVT, FPCLASS,
DAG.getConstant(TDCMask, DL, DstVT));
return DAG.getSetCC(DL, VT, AND, DAG.getConstant(0, DL, DstVT),
ISD::SETNE);
}
MVT ContainerVT0 = getContainerForFixedLengthVector(VT0);
MVT ContainerVT = getContainerForFixedLengthVector(VT);
MVT ContainerDstVT = ContainerVT0.changeVectorElementTypeToInteger();
auto [Mask, VL] = getDefaultVLOps(VT0, ContainerVT0, DL, DAG, Subtarget);
if (Op.getOpcode() == ISD::VP_IS_FPCLASS) {
Mask = Op.getOperand(2);
MVT MaskContainerVT =
getContainerForFixedLengthVector(Mask.getSimpleValueType());
Mask = convertToScalableVector(MaskContainerVT, Mask, DAG, Subtarget);
VL = Op.getOperand(3);
}
Op0 = convertToScalableVector(ContainerVT0, Op0, DAG, Subtarget);
SDValue FPCLASS = DAG.getNode(RISCVISD::FCLASS_VL, DL, ContainerDstVT, Op0,
Mask, VL, Op->getFlags());
TDCMaskV = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerDstVT,
DAG.getUNDEF(ContainerDstVT), TDCMaskV, VL);
if (IsOneBitMask) {
SDValue VMSEQ =
DAG.getNode(RISCVISD::SETCC_VL, DL, ContainerVT,
{FPCLASS, TDCMaskV, DAG.getCondCode(ISD::SETEQ),
DAG.getUNDEF(ContainerVT), Mask, VL});
return convertFromScalableVector(VT, VMSEQ, DAG, Subtarget);
}
SDValue AND = DAG.getNode(RISCVISD::AND_VL, DL, ContainerDstVT, FPCLASS,
TDCMaskV, DAG.getUNDEF(ContainerDstVT), Mask, VL);
SDValue SplatZero = DAG.getConstant(0, DL, XLenVT);
SplatZero = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerDstVT,
DAG.getUNDEF(ContainerDstVT), SplatZero, VL);
SDValue VMSNE = DAG.getNode(RISCVISD::SETCC_VL, DL, ContainerVT,
{AND, SplatZero, DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(ContainerVT), Mask, VL});
return convertFromScalableVector(VT, VMSNE, DAG, Subtarget);
}
SDValue FCLASS = DAG.getNode(RISCVISD::FCLASS, DL, XLenVT, Op.getOperand(0));
SDValue AND = DAG.getNode(ISD::AND, DL, XLenVT, FCLASS, TDCMaskV);
SDValue Res = DAG.getSetCC(DL, XLenVT, AND, DAG.getConstant(0, DL, XLenVT),
ISD::CondCode::SETNE);
return DAG.getNode(ISD::TRUNCATE, DL, VT, Res);
}
// Lower fmaximum and fminimum. Unlike our fmax and fmin instructions, these
// operations propagate nans.
static SDValue lowerFMAXIMUM_FMINIMUM(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
SDValue X = Op.getOperand(0);
SDValue Y = Op.getOperand(1);
if (!VT.isVector()) {
MVT XLenVT = Subtarget.getXLenVT();
// If X is a nan, replace Y with X. If Y is a nan, replace X with Y. This
// ensures that when one input is a nan, the other will also be a nan
// allowing the nan to propagate. If both inputs are nan, this will swap the
// inputs which is harmless.
SDValue NewY = Y;
if (!Op->getFlags().hasNoNaNs() && !DAG.isKnownNeverNaN(X)) {
SDValue XIsNonNan = DAG.getSetCC(DL, XLenVT, X, X, ISD::SETOEQ);
NewY = DAG.getSelect(DL, VT, XIsNonNan, Y, X);
}
SDValue NewX = X;
if (!Op->getFlags().hasNoNaNs() && !DAG.isKnownNeverNaN(Y)) {
SDValue YIsNonNan = DAG.getSetCC(DL, XLenVT, Y, Y, ISD::SETOEQ);
NewX = DAG.getSelect(DL, VT, YIsNonNan, X, Y);
}
unsigned Opc =
Op.getOpcode() == ISD::FMAXIMUM ? RISCVISD::FMAX : RISCVISD::FMIN;
return DAG.getNode(Opc, DL, VT, NewX, NewY);
}
// Check no NaNs before converting to fixed vector scalable.
bool XIsNeverNan = Op->getFlags().hasNoNaNs() || DAG.isKnownNeverNaN(X);
bool YIsNeverNan = Op->getFlags().hasNoNaNs() || DAG.isKnownNeverNaN(Y);
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(DAG, VT, Subtarget);
X = convertToScalableVector(ContainerVT, X, DAG, Subtarget);
Y = convertToScalableVector(ContainerVT, Y, DAG, Subtarget);
}
SDValue Mask, VL;
if (Op->isVPOpcode()) {
Mask = Op.getOperand(2);
if (VT.isFixedLengthVector())
Mask = convertToScalableVector(getMaskTypeFor(ContainerVT), Mask, DAG,
Subtarget);
VL = Op.getOperand(3);
} else {
std::tie(Mask, VL) = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
}
SDValue NewY = Y;
if (!XIsNeverNan) {
SDValue XIsNonNan = DAG.getNode(RISCVISD::SETCC_VL, DL, Mask.getValueType(),
{X, X, DAG.getCondCode(ISD::SETOEQ),
DAG.getUNDEF(ContainerVT), Mask, VL});
NewY = DAG.getNode(RISCVISD::VMERGE_VL, DL, ContainerVT, XIsNonNan, Y, X,
DAG.getUNDEF(ContainerVT), VL);
}
SDValue NewX = X;
if (!YIsNeverNan) {
SDValue YIsNonNan = DAG.getNode(RISCVISD::SETCC_VL, DL, Mask.getValueType(),
{Y, Y, DAG.getCondCode(ISD::SETOEQ),
DAG.getUNDEF(ContainerVT), Mask, VL});
NewX = DAG.getNode(RISCVISD::VMERGE_VL, DL, ContainerVT, YIsNonNan, X, Y,
DAG.getUNDEF(ContainerVT), VL);
}
unsigned Opc =
Op.getOpcode() == ISD::FMAXIMUM || Op->getOpcode() == ISD::VP_FMAXIMUM
? RISCVISD::VFMAX_VL
: RISCVISD::VFMIN_VL;
SDValue Res = DAG.getNode(Opc, DL, ContainerVT, NewX, NewY,
DAG.getUNDEF(ContainerVT), Mask, VL);
if (VT.isFixedLengthVector())
Res = convertFromScalableVector(VT, Res, DAG, Subtarget);
return Res;
}
static SDValue lowerFABSorFNEG(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
bool IsFABS = Op.getOpcode() == ISD::FABS;
assert((IsFABS || Op.getOpcode() == ISD::FNEG) &&
"Wrong opcode for lowering FABS or FNEG.");
MVT XLenVT = Subtarget.getXLenVT();
MVT VT = Op.getSimpleValueType();
assert((VT == MVT::f16 || VT == MVT::bf16) && "Unexpected type");
SDLoc DL(Op);
SDValue Fmv =
DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, XLenVT, Op.getOperand(0));
APInt Mask = IsFABS ? APInt::getSignedMaxValue(16) : APInt::getSignMask(16);
Mask = Mask.sext(Subtarget.getXLen());
unsigned LogicOpc = IsFABS ? ISD::AND : ISD::XOR;
SDValue Logic =
DAG.getNode(LogicOpc, DL, XLenVT, Fmv, DAG.getConstant(Mask, DL, XLenVT));
return DAG.getNode(RISCVISD::FMV_H_X, DL, VT, Logic);
}
static SDValue lowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(Op.getOpcode() == ISD::FCOPYSIGN && "Unexpected opcode");
MVT XLenVT = Subtarget.getXLenVT();
MVT VT = Op.getSimpleValueType();
assert((VT == MVT::f16 || VT == MVT::bf16) && "Unexpected type");
SDValue Mag = Op.getOperand(0);
SDValue Sign = Op.getOperand(1);
SDLoc DL(Op);
// Get sign bit into an integer value.
unsigned SignSize = Sign.getValueSizeInBits();
SDValue SignAsInt = [&]() {
if (SignSize == Subtarget.getXLen())
return DAG.getNode(ISD::BITCAST, DL, XLenVT, Sign);
switch (SignSize) {
case 16:
return DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, XLenVT, Sign);
case 32:
return DAG.getNode(RISCVISD::FMV_X_ANYEXTW_RV64, DL, XLenVT, Sign);
case 64: {
assert(XLenVT == MVT::i32 && "Unexpected type");
// Copy the upper word to integer.
SignSize = 32;
return DAG.getNode(RISCVISD::SplitF64, DL, {MVT::i32, MVT::i32}, Sign)
.getValue(1);
}
default:
llvm_unreachable("Unexpected sign size");
}
}();
// Get the signbit at the right position for MagAsInt.
if (int ShiftAmount = (int)SignSize - (int)Mag.getValueSizeInBits())
SignAsInt = DAG.getNode(ShiftAmount > 0 ? ISD::SRL : ISD::SHL, DL, XLenVT,
SignAsInt,
DAG.getConstant(std::abs(ShiftAmount), DL, XLenVT));
// Mask the sign bit and any bits above it. The extra bits will be dropped
// when we convert back to FP.
SDValue SignMask = DAG.getConstant(
APInt::getSignMask(16).sext(Subtarget.getXLen()), DL, XLenVT);
SDValue SignBit = DAG.getNode(ISD::AND, DL, XLenVT, SignAsInt, SignMask);
// Transform Mag value to integer, and clear the sign bit.
SDValue MagAsInt = DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, XLenVT, Mag);
SDValue ClearSignMask = DAG.getConstant(
APInt::getSignedMaxValue(16).sext(Subtarget.getXLen()), DL, XLenVT);
SDValue ClearedSign =
DAG.getNode(ISD::AND, DL, XLenVT, MagAsInt, ClearSignMask);
SDValue CopiedSign = DAG.getNode(ISD::OR, DL, XLenVT, ClearedSign, SignBit,
SDNodeFlags::Disjoint);
return DAG.getNode(RISCVISD::FMV_H_X, DL, VT, CopiedSign);
}
/// Get a RISC-V target specified VL op for a given SDNode.
static unsigned getRISCVVLOp(SDValue Op) {
#define OP_CASE(NODE) \
case ISD::NODE: \
return RISCVISD::NODE##_VL;
#define VP_CASE(NODE) \
case ISD::VP_##NODE: \
return RISCVISD::NODE##_VL;
// clang-format off
switch (Op.getOpcode()) {
default:
llvm_unreachable("don't have RISC-V specified VL op for this SDNode");
OP_CASE(ADD)
OP_CASE(SUB)
OP_CASE(MUL)
OP_CASE(MULHS)
OP_CASE(MULHU)
OP_CASE(SDIV)
OP_CASE(SREM)
OP_CASE(UDIV)
OP_CASE(UREM)
OP_CASE(SHL)
OP_CASE(SRA)
OP_CASE(SRL)
OP_CASE(ROTL)
OP_CASE(ROTR)
OP_CASE(BSWAP)
OP_CASE(CTTZ)
OP_CASE(CTLZ)
OP_CASE(CTPOP)
OP_CASE(BITREVERSE)
OP_CASE(SADDSAT)
OP_CASE(UADDSAT)
OP_CASE(SSUBSAT)
OP_CASE(USUBSAT)
OP_CASE(AVGFLOORS)
OP_CASE(AVGFLOORU)
OP_CASE(AVGCEILS)
OP_CASE(AVGCEILU)
OP_CASE(FADD)
OP_CASE(FSUB)
OP_CASE(FMUL)
OP_CASE(FDIV)
OP_CASE(FNEG)
OP_CASE(FABS)
OP_CASE(FCOPYSIGN)
OP_CASE(FSQRT)
OP_CASE(SMIN)
OP_CASE(SMAX)
OP_CASE(UMIN)
OP_CASE(UMAX)
OP_CASE(STRICT_FADD)
OP_CASE(STRICT_FSUB)
OP_CASE(STRICT_FMUL)
OP_CASE(STRICT_FDIV)
OP_CASE(STRICT_FSQRT)
VP_CASE(ADD) // VP_ADD
VP_CASE(SUB) // VP_SUB
VP_CASE(MUL) // VP_MUL
VP_CASE(SDIV) // VP_SDIV
VP_CASE(SREM) // VP_SREM
VP_CASE(UDIV) // VP_UDIV
VP_CASE(UREM) // VP_UREM
VP_CASE(SHL) // VP_SHL
VP_CASE(FADD) // VP_FADD
VP_CASE(FSUB) // VP_FSUB
VP_CASE(FMUL) // VP_FMUL
VP_CASE(FDIV) // VP_FDIV
VP_CASE(FNEG) // VP_FNEG
VP_CASE(FABS) // VP_FABS
VP_CASE(SMIN) // VP_SMIN
VP_CASE(SMAX) // VP_SMAX
VP_CASE(UMIN) // VP_UMIN
VP_CASE(UMAX) // VP_UMAX
VP_CASE(FCOPYSIGN) // VP_FCOPYSIGN
VP_CASE(SETCC) // VP_SETCC
VP_CASE(SINT_TO_FP) // VP_SINT_TO_FP
VP_CASE(UINT_TO_FP) // VP_UINT_TO_FP
VP_CASE(BITREVERSE) // VP_BITREVERSE
VP_CASE(SADDSAT) // VP_SADDSAT
VP_CASE(UADDSAT) // VP_UADDSAT
VP_CASE(SSUBSAT) // VP_SSUBSAT
VP_CASE(USUBSAT) // VP_USUBSAT
VP_CASE(BSWAP) // VP_BSWAP
VP_CASE(CTLZ) // VP_CTLZ
VP_CASE(CTTZ) // VP_CTTZ
VP_CASE(CTPOP) // VP_CTPOP
case ISD::CTLZ_ZERO_UNDEF:
case ISD::VP_CTLZ_ZERO_UNDEF:
return RISCVISD::CTLZ_VL;
case ISD::CTTZ_ZERO_UNDEF:
case ISD::VP_CTTZ_ZERO_UNDEF:
return RISCVISD::CTTZ_VL;
case ISD::FMA:
case ISD::VP_FMA:
return RISCVISD::VFMADD_VL;
case ISD::STRICT_FMA:
return RISCVISD::STRICT_VFMADD_VL;
case ISD::AND:
case ISD::VP_AND:
if (Op.getSimpleValueType().getVectorElementType() == MVT::i1)
return RISCVISD::VMAND_VL;
return RISCVISD::AND_VL;
case ISD::OR:
case ISD::VP_OR:
if (Op.getSimpleValueType().getVectorElementType() == MVT::i1)
return RISCVISD::VMOR_VL;
return RISCVISD::OR_VL;
case ISD::XOR:
case ISD::VP_XOR:
if (Op.getSimpleValueType().getVectorElementType() == MVT::i1)
return RISCVISD::VMXOR_VL;
return RISCVISD::XOR_VL;
case ISD::ANY_EXTEND:
case ISD::ZERO_EXTEND:
return RISCVISD::VZEXT_VL;
case ISD::SIGN_EXTEND:
return RISCVISD::VSEXT_VL;
case ISD::SETCC:
return RISCVISD::SETCC_VL;
case ISD::VSELECT:
return RISCVISD::VMERGE_VL;
case ISD::VP_SELECT:
case ISD::VP_MERGE:
return RISCVISD::VMERGE_VL;
case ISD::VP_SRA:
return RISCVISD::SRA_VL;
case ISD::VP_SRL:
return RISCVISD::SRL_VL;
case ISD::VP_SQRT:
return RISCVISD::FSQRT_VL;
case ISD::VP_SIGN_EXTEND:
return RISCVISD::VSEXT_VL;
case ISD::VP_ZERO_EXTEND:
return RISCVISD::VZEXT_VL;
case ISD::VP_FP_TO_SINT:
return RISCVISD::VFCVT_RTZ_X_F_VL;
case ISD::VP_FP_TO_UINT:
return RISCVISD::VFCVT_RTZ_XU_F_VL;
case ISD::FMINNUM:
case ISD::FMINIMUMNUM:
case ISD::VP_FMINNUM:
return RISCVISD::VFMIN_VL;
case ISD::FMAXNUM:
case ISD::FMAXIMUMNUM:
case ISD::VP_FMAXNUM:
return RISCVISD::VFMAX_VL;
case ISD::LRINT:
case ISD::VP_LRINT:
case ISD::LLRINT:
case ISD::VP_LLRINT:
return RISCVISD::VFCVT_RM_X_F_VL;
}
// clang-format on
#undef OP_CASE
#undef VP_CASE
}
static bool isPromotedOpNeedingSplit(SDValue Op,
const RISCVSubtarget &Subtarget) {
return (Op.getValueType() == MVT::nxv32f16 &&
(Subtarget.hasVInstructionsF16Minimal() &&
!Subtarget.hasVInstructionsF16())) ||
Op.getValueType() == MVT::nxv32bf16;
}
static SDValue SplitVectorOp(SDValue Op, SelectionDAG &DAG) {
auto [LoVT, HiVT] = DAG.GetSplitDestVTs(Op.getValueType());
SDLoc DL(Op);
SmallVector<SDValue, 4> LoOperands(Op.getNumOperands());
SmallVector<SDValue, 4> HiOperands(Op.getNumOperands());
for (unsigned j = 0; j != Op.getNumOperands(); ++j) {
if (!Op.getOperand(j).getValueType().isVector()) {
LoOperands[j] = Op.getOperand(j);
HiOperands[j] = Op.getOperand(j);
continue;
}
std::tie(LoOperands[j], HiOperands[j]) =
DAG.SplitVector(Op.getOperand(j), DL);
}
SDValue LoRes =
DAG.getNode(Op.getOpcode(), DL, LoVT, LoOperands, Op->getFlags());
SDValue HiRes =
DAG.getNode(Op.getOpcode(), DL, HiVT, HiOperands, Op->getFlags());
return DAG.getNode(ISD::CONCAT_VECTORS, DL, Op.getValueType(), LoRes, HiRes);
}
static SDValue SplitVPOp(SDValue Op, SelectionDAG &DAG) {
assert(ISD::isVPOpcode(Op.getOpcode()) && "Not a VP op");
auto [LoVT, HiVT] = DAG.GetSplitDestVTs(Op.getValueType());
SDLoc DL(Op);
SmallVector<SDValue, 4> LoOperands(Op.getNumOperands());
SmallVector<SDValue, 4> HiOperands(Op.getNumOperands());
for (unsigned j = 0; j != Op.getNumOperands(); ++j) {
if (ISD::getVPExplicitVectorLengthIdx(Op.getOpcode()) == j) {
std::tie(LoOperands[j], HiOperands[j]) =
DAG.SplitEVL(Op.getOperand(j), Op.getValueType(), DL);
continue;
}
if (!Op.getOperand(j).getValueType().isVector()) {
LoOperands[j] = Op.getOperand(j);
HiOperands[j] = Op.getOperand(j);
continue;
}
std::tie(LoOperands[j], HiOperands[j]) =
DAG.SplitVector(Op.getOperand(j), DL);
}
SDValue LoRes =
DAG.getNode(Op.getOpcode(), DL, LoVT, LoOperands, Op->getFlags());
SDValue HiRes =
DAG.getNode(Op.getOpcode(), DL, HiVT, HiOperands, Op->getFlags());
return DAG.getNode(ISD::CONCAT_VECTORS, DL, Op.getValueType(), LoRes, HiRes);
}
static SDValue SplitVectorReductionOp(SDValue Op, SelectionDAG &DAG) {
SDLoc DL(Op);
auto [Lo, Hi] = DAG.SplitVector(Op.getOperand(1), DL);
auto [MaskLo, MaskHi] = DAG.SplitVector(Op.getOperand(2), DL);
auto [EVLLo, EVLHi] =
DAG.SplitEVL(Op.getOperand(3), Op.getOperand(1).getValueType(), DL);
SDValue ResLo =
DAG.getNode(Op.getOpcode(), DL, Op.getValueType(),
{Op.getOperand(0), Lo, MaskLo, EVLLo}, Op->getFlags());
return DAG.getNode(Op.getOpcode(), DL, Op.getValueType(),
{ResLo, Hi, MaskHi, EVLHi}, Op->getFlags());
}
static SDValue SplitStrictFPVectorOp(SDValue Op, SelectionDAG &DAG) {
assert(Op->isStrictFPOpcode());
auto [LoVT, HiVT] = DAG.GetSplitDestVTs(Op->getValueType(0));
SDVTList LoVTs = DAG.getVTList(LoVT, Op->getValueType(1));
SDVTList HiVTs = DAG.getVTList(HiVT, Op->getValueType(1));
SDLoc DL(Op);
SmallVector<SDValue, 4> LoOperands(Op.getNumOperands());
SmallVector<SDValue, 4> HiOperands(Op.getNumOperands());
for (unsigned j = 0; j != Op.getNumOperands(); ++j) {
if (!Op.getOperand(j).getValueType().isVector()) {
LoOperands[j] = Op.getOperand(j);
HiOperands[j] = Op.getOperand(j);
continue;
}
std::tie(LoOperands[j], HiOperands[j]) =
DAG.SplitVector(Op.getOperand(j), DL);
}
SDValue LoRes =
DAG.getNode(Op.getOpcode(), DL, LoVTs, LoOperands, Op->getFlags());
HiOperands[0] = LoRes.getValue(1);
SDValue HiRes =
DAG.getNode(Op.getOpcode(), DL, HiVTs, HiOperands, Op->getFlags());
SDValue V = DAG.getNode(ISD::CONCAT_VECTORS, DL, Op->getValueType(0),
LoRes.getValue(0), HiRes.getValue(0));
return DAG.getMergeValues({V, HiRes.getValue(1)}, DL);
}
SDValue
RISCVTargetLowering::lowerXAndesBfHCvtBFloat16Load(SDValue Op,
SelectionDAG &DAG) const {
assert(Subtarget.hasVendorXAndesBFHCvt() && !Subtarget.hasStdExtZfh() &&
"Unexpected bfloat16 load lowering");
SDLoc DL(Op);
LoadSDNode *LD = cast<LoadSDNode>(Op.getNode());
EVT MemVT = LD->getMemoryVT();
SDValue Load = DAG.getExtLoad(
ISD::ZEXTLOAD, DL, Subtarget.getXLenVT(), LD->getChain(),
LD->getBasePtr(),
EVT::getIntegerVT(*DAG.getContext(), MemVT.getSizeInBits()),
LD->getMemOperand());
// Using mask to make bf16 nan-boxing valid when we don't have flh
// instruction. -65536 would be treat as a small number and thus it can be
// directly used lui to get the constant.
SDValue mask = DAG.getSignedConstant(-65536, DL, Subtarget.getXLenVT());
SDValue OrSixteenOne =
DAG.getNode(ISD::OR, DL, Load.getValueType(), {Load, mask});
SDValue ConvertedResult =
DAG.getNode(RISCVISD::NDS_FMV_BF16_X, DL, MVT::bf16, OrSixteenOne);
return DAG.getMergeValues({ConvertedResult, Load.getValue(1)}, DL);
}
SDValue
RISCVTargetLowering::lowerXAndesBfHCvtBFloat16Store(SDValue Op,
SelectionDAG &DAG) const {
assert(Subtarget.hasVendorXAndesBFHCvt() && !Subtarget.hasStdExtZfh() &&
"Unexpected bfloat16 store lowering");
StoreSDNode *ST = cast<StoreSDNode>(Op.getNode());
SDLoc DL(Op);
SDValue FMV = DAG.getNode(RISCVISD::NDS_FMV_X_ANYEXTBF16, DL,
Subtarget.getXLenVT(), ST->getValue());
return DAG.getTruncStore(
ST->getChain(), DL, FMV, ST->getBasePtr(),
EVT::getIntegerVT(*DAG.getContext(), ST->getMemoryVT().getSizeInBits()),
ST->getMemOperand());
}
SDValue RISCVTargetLowering::LowerOperation(SDValue Op,
SelectionDAG &DAG) const {
switch (Op.getOpcode()) {
default:
reportFatalInternalError(
"Unimplemented RISCVTargetLowering::LowerOperation Case");
case ISD::PREFETCH:
return LowerPREFETCH(Op, Subtarget, DAG);
case ISD::ATOMIC_FENCE:
return LowerATOMIC_FENCE(Op, DAG, Subtarget);
case ISD::GlobalAddress:
return lowerGlobalAddress(Op, DAG);
case ISD::BlockAddress:
return lowerBlockAddress(Op, DAG);
case ISD::ConstantPool:
return lowerConstantPool(Op, DAG);
case ISD::JumpTable:
return lowerJumpTable(Op, DAG);
case ISD::GlobalTLSAddress:
return lowerGlobalTLSAddress(Op, DAG);
case ISD::Constant:
return lowerConstant(Op, DAG, Subtarget);
case ISD::ConstantFP:
return lowerConstantFP(Op, DAG);
case ISD::SELECT:
return lowerSELECT(Op, DAG);
case ISD::BRCOND:
return lowerBRCOND(Op, DAG);
case ISD::VASTART:
return lowerVASTART(Op, DAG);
case ISD::FRAMEADDR:
return lowerFRAMEADDR(Op, DAG);
case ISD::RETURNADDR:
return lowerRETURNADDR(Op, DAG);
case ISD::SHL_PARTS:
return lowerShiftLeftParts(Op, DAG);
case ISD::SRA_PARTS:
return lowerShiftRightParts(Op, DAG, true);
case ISD::SRL_PARTS:
return lowerShiftRightParts(Op, DAG, false);
case ISD::ROTL:
case ISD::ROTR:
if (Op.getValueType().isFixedLengthVector()) {
assert(Subtarget.hasStdExtZvkb());
return lowerToScalableOp(Op, DAG);
}
assert(Subtarget.hasVendorXTHeadBb() &&
!(Subtarget.hasStdExtZbb() || Subtarget.hasStdExtZbkb()) &&
"Unexpected custom legalization");
// XTHeadBb only supports rotate by constant.
if (!isa<ConstantSDNode>(Op.getOperand(1)))
return SDValue();
return Op;
case ISD::BITCAST: {
SDLoc DL(Op);
EVT VT = Op.getValueType();
SDValue Op0 = Op.getOperand(0);
EVT Op0VT = Op0.getValueType();
MVT XLenVT = Subtarget.getXLenVT();
if (Op0VT == MVT::i16 &&
((VT == MVT::f16 && Subtarget.hasStdExtZfhminOrZhinxmin()) ||
(VT == MVT::bf16 && Subtarget.hasStdExtZfbfmin()))) {
SDValue NewOp0 = DAG.getNode(ISD::ANY_EXTEND, DL, XLenVT, Op0);
return DAG.getNode(RISCVISD::FMV_H_X, DL, VT, NewOp0);
}
if (VT == MVT::f32 && Op0VT == MVT::i32 && Subtarget.is64Bit() &&
Subtarget.hasStdExtFOrZfinx()) {
SDValue NewOp0 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op0);
return DAG.getNode(RISCVISD::FMV_W_X_RV64, DL, MVT::f32, NewOp0);
}
if (VT == MVT::f64 && Op0VT == MVT::i64 && !Subtarget.is64Bit() &&
Subtarget.hasStdExtDOrZdinx()) {
SDValue Lo, Hi;
std::tie(Lo, Hi) = DAG.SplitScalar(Op0, DL, MVT::i32, MVT::i32);
return DAG.getNode(RISCVISD::BuildPairF64, DL, MVT::f64, Lo, Hi);
}
// Consider other scalar<->scalar casts as legal if the types are legal.
// Otherwise expand them.
if (!VT.isVector() && !Op0VT.isVector()) {
if (isTypeLegal(VT) && isTypeLegal(Op0VT))
return Op;
return SDValue();
}
assert(!VT.isScalableVector() && !Op0VT.isScalableVector() &&
"Unexpected types");
if (VT.isFixedLengthVector()) {
// We can handle fixed length vector bitcasts with a simple replacement
// in isel.
if (Op0VT.isFixedLengthVector())
return Op;
// When bitcasting from scalar to fixed-length vector, insert the scalar
// into a one-element vector of the result type, and perform a vector
// bitcast.
if (!Op0VT.isVector()) {
EVT BVT = EVT::getVectorVT(*DAG.getContext(), Op0VT, 1);
if (!isTypeLegal(BVT))
return SDValue();
return DAG.getBitcast(
VT, DAG.getInsertVectorElt(DL, DAG.getUNDEF(BVT), Op0, 0));
}
return SDValue();
}
// Custom-legalize bitcasts from fixed-length vector types to scalar types
// thus: bitcast the vector to a one-element vector type whose element type
// is the same as the result type, and extract the first element.
if (!VT.isVector() && Op0VT.isFixedLengthVector()) {
EVT BVT = EVT::getVectorVT(*DAG.getContext(), VT, 1);
if (!isTypeLegal(BVT))
return SDValue();
SDValue BVec = DAG.getBitcast(BVT, Op0);
return DAG.getExtractVectorElt(DL, VT, BVec, 0);
}
return SDValue();
}
case ISD::INTRINSIC_WO_CHAIN:
return LowerINTRINSIC_WO_CHAIN(Op, DAG);
case ISD::INTRINSIC_W_CHAIN:
return LowerINTRINSIC_W_CHAIN(Op, DAG);
case ISD::INTRINSIC_VOID:
return LowerINTRINSIC_VOID(Op, DAG);
case ISD::IS_FPCLASS:
return LowerIS_FPCLASS(Op, DAG);
case ISD::BITREVERSE: {
MVT VT = Op.getSimpleValueType();
if (VT.isFixedLengthVector()) {
assert(Subtarget.hasStdExtZvbb());
return lowerToScalableOp(Op, DAG);
}
SDLoc DL(Op);
assert(Subtarget.hasStdExtZbkb() && "Unexpected custom legalization");
assert(Op.getOpcode() == ISD::BITREVERSE && "Unexpected opcode");
// Expand bitreverse to a bswap(rev8) followed by brev8.
SDValue BSwap = DAG.getNode(ISD::BSWAP, DL, VT, Op.getOperand(0));
return DAG.getNode(RISCVISD::BREV8, DL, VT, BSwap);
}
case ISD::TRUNCATE:
case ISD::TRUNCATE_SSAT_S:
case ISD::TRUNCATE_USAT_U:
// Only custom-lower vector truncates
if (!Op.getSimpleValueType().isVector())
return Op;
return lowerVectorTruncLike(Op, DAG);
case ISD::ANY_EXTEND:
case ISD::ZERO_EXTEND:
if (Op.getOperand(0).getValueType().isVector() &&
Op.getOperand(0).getValueType().getVectorElementType() == MVT::i1)
return lowerVectorMaskExt(Op, DAG, /*ExtVal*/ 1);
if (Op.getValueType().isScalableVector())
return Op;
return lowerToScalableOp(Op, DAG);
case ISD::SIGN_EXTEND:
if (Op.getOperand(0).getValueType().isVector() &&
Op.getOperand(0).getValueType().getVectorElementType() == MVT::i1)
return lowerVectorMaskExt(Op, DAG, /*ExtVal*/ -1);
if (Op.getValueType().isScalableVector())
return Op;
return lowerToScalableOp(Op, DAG);
case ISD::SPLAT_VECTOR_PARTS:
return lowerSPLAT_VECTOR_PARTS(Op, DAG);
case ISD::INSERT_VECTOR_ELT:
return lowerINSERT_VECTOR_ELT(Op, DAG);
case ISD::EXTRACT_VECTOR_ELT:
return lowerEXTRACT_VECTOR_ELT(Op, DAG);
case ISD::SCALAR_TO_VECTOR: {
MVT VT = Op.getSimpleValueType();
SDLoc DL(Op);
SDValue Scalar = Op.getOperand(0);
if (VT.getVectorElementType() == MVT::i1) {
MVT WideVT = VT.changeVectorElementType(MVT::i8);
SDValue V = DAG.getNode(ISD::SCALAR_TO_VECTOR, DL, WideVT, Scalar);
return DAG.getNode(ISD::TRUNCATE, DL, VT, V);
}
MVT ContainerVT = VT;
if (VT.isFixedLengthVector())
ContainerVT = getContainerForFixedLengthVector(VT);
SDValue VL = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget).second;
SDValue V;
if (VT.isFloatingPoint()) {
V = DAG.getNode(RISCVISD::VFMV_S_F_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), Scalar, VL);
} else {
Scalar = DAG.getNode(ISD::ANY_EXTEND, DL, Subtarget.getXLenVT(), Scalar);
V = DAG.getNode(RISCVISD::VMV_S_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), Scalar, VL);
}
if (VT.isFixedLengthVector())
V = convertFromScalableVector(VT, V, DAG, Subtarget);
return V;
}
case ISD::VSCALE: {
MVT XLenVT = Subtarget.getXLenVT();
MVT VT = Op.getSimpleValueType();
SDLoc DL(Op);
SDValue Res = DAG.getNode(RISCVISD::READ_VLENB, DL, XLenVT);
// We define our scalable vector types for lmul=1 to use a 64 bit known
// minimum size. e.g. <vscale x 2 x i32>. VLENB is in bytes so we calculate
// vscale as VLENB / 8.
static_assert(RISCV::RVVBitsPerBlock == 64, "Unexpected bits per block!");
if (Subtarget.getRealMinVLen() < RISCV::RVVBitsPerBlock)
reportFatalInternalError("Support for VLEN==32 is incomplete.");
// We assume VLENB is a multiple of 8. We manually choose the best shift
// here because SimplifyDemandedBits isn't always able to simplify it.
uint64_t Val = Op.getConstantOperandVal(0);
if (isPowerOf2_64(Val)) {
uint64_t Log2 = Log2_64(Val);
if (Log2 < 3) {
SDNodeFlags Flags;
Flags.setExact(true);
Res = DAG.getNode(ISD::SRL, DL, XLenVT, Res,
DAG.getConstant(3 - Log2, DL, XLenVT), Flags);
} else if (Log2 > 3) {
Res = DAG.getNode(ISD::SHL, DL, XLenVT, Res,
DAG.getConstant(Log2 - 3, DL, XLenVT));
}
} else if ((Val % 8) == 0) {
// If the multiplier is a multiple of 8, scale it down to avoid needing
// to shift the VLENB value.
Res = DAG.getNode(ISD::MUL, DL, XLenVT, Res,
DAG.getConstant(Val / 8, DL, XLenVT));
} else {
SDNodeFlags Flags;
Flags.setExact(true);
SDValue VScale = DAG.getNode(ISD::SRL, DL, XLenVT, Res,
DAG.getConstant(3, DL, XLenVT), Flags);
Res = DAG.getNode(ISD::MUL, DL, XLenVT, VScale,
DAG.getConstant(Val, DL, XLenVT));
}
return DAG.getNode(ISD::TRUNCATE, DL, VT, Res);
}
case ISD::FPOWI: {
// Custom promote f16 powi with illegal i32 integer type on RV64. Once
// promoted this will be legalized into a libcall by LegalizeIntegerTypes.
if (Op.getValueType() == MVT::f16 && Subtarget.is64Bit() &&
Op.getOperand(1).getValueType() == MVT::i32) {
SDLoc DL(Op);
SDValue Op0 = DAG.getNode(ISD::FP_EXTEND, DL, MVT::f32, Op.getOperand(0));
SDValue Powi =
DAG.getNode(ISD::FPOWI, DL, MVT::f32, Op0, Op.getOperand(1));
return DAG.getNode(ISD::FP_ROUND, DL, MVT::f16, Powi,
DAG.getIntPtrConstant(0, DL, /*isTarget=*/true));
}
return SDValue();
}
case ISD::FMAXIMUM:
case ISD::FMINIMUM:
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitVectorOp(Op, DAG);
return lowerFMAXIMUM_FMINIMUM(Op, DAG, Subtarget);
case ISD::FP_EXTEND:
case ISD::FP_ROUND:
return lowerVectorFPExtendOrRoundLike(Op, DAG);
case ISD::STRICT_FP_ROUND:
case ISD::STRICT_FP_EXTEND:
return lowerStrictFPExtendOrRoundLike(Op, DAG);
case ISD::SINT_TO_FP:
case ISD::UINT_TO_FP:
if (Op.getValueType().isVector() &&
((Op.getValueType().getScalarType() == MVT::f16 &&
(Subtarget.hasVInstructionsF16Minimal() &&
!Subtarget.hasVInstructionsF16())) ||
Op.getValueType().getScalarType() == MVT::bf16)) {
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitVectorOp(Op, DAG);
// int -> f32
SDLoc DL(Op);
MVT NVT =
MVT::getVectorVT(MVT::f32, Op.getValueType().getVectorElementCount());
SDValue NC = DAG.getNode(Op.getOpcode(), DL, NVT, Op->ops());
// f32 -> [b]f16
return DAG.getNode(ISD::FP_ROUND, DL, Op.getValueType(), NC,
DAG.getIntPtrConstant(0, DL, /*isTarget=*/true));
}
[[fallthrough]];
case ISD::FP_TO_SINT:
case ISD::FP_TO_UINT:
if (SDValue Op1 = Op.getOperand(0);
Op1.getValueType().isVector() &&
((Op1.getValueType().getScalarType() == MVT::f16 &&
(Subtarget.hasVInstructionsF16Minimal() &&
!Subtarget.hasVInstructionsF16())) ||
Op1.getValueType().getScalarType() == MVT::bf16)) {
if (isPromotedOpNeedingSplit(Op1, Subtarget))
return SplitVectorOp(Op, DAG);
// [b]f16 -> f32
SDLoc DL(Op);
MVT NVT = MVT::getVectorVT(MVT::f32,
Op1.getValueType().getVectorElementCount());
SDValue WidenVec = DAG.getNode(ISD::FP_EXTEND, DL, NVT, Op1);
// f32 -> int
return DAG.getNode(Op.getOpcode(), DL, Op.getValueType(), WidenVec);
}
[[fallthrough]];
case ISD::STRICT_FP_TO_SINT:
case ISD::STRICT_FP_TO_UINT:
case ISD::STRICT_SINT_TO_FP:
case ISD::STRICT_UINT_TO_FP: {
// RVV can only do fp<->int conversions to types half/double the size as
// the source. We custom-lower any conversions that do two hops into
// sequences.
MVT VT = Op.getSimpleValueType();
if (VT.isScalarInteger())
return lowerFP_TO_INT(Op, DAG, Subtarget);
bool IsStrict = Op->isStrictFPOpcode();
SDValue Src = Op.getOperand(0 + IsStrict);
MVT SrcVT = Src.getSimpleValueType();
if (SrcVT.isScalarInteger())
return lowerINT_TO_FP(Op, DAG, Subtarget);
if (!VT.isVector())
return Op;
SDLoc DL(Op);
MVT EltVT = VT.getVectorElementType();
MVT SrcEltVT = SrcVT.getVectorElementType();
unsigned EltSize = EltVT.getSizeInBits();
unsigned SrcEltSize = SrcEltVT.getSizeInBits();
assert(isPowerOf2_32(EltSize) && isPowerOf2_32(SrcEltSize) &&
"Unexpected vector element types");
bool IsInt2FP = SrcEltVT.isInteger();
// Widening conversions
if (EltSize > (2 * SrcEltSize)) {
if (IsInt2FP) {
// Do a regular integer sign/zero extension then convert to float.
MVT IVecVT = MVT::getVectorVT(MVT::getIntegerVT(EltSize / 2),
VT.getVectorElementCount());
unsigned ExtOpcode = (Op.getOpcode() == ISD::UINT_TO_FP ||
Op.getOpcode() == ISD::STRICT_UINT_TO_FP)
? ISD::ZERO_EXTEND
: ISD::SIGN_EXTEND;
SDValue Ext = DAG.getNode(ExtOpcode, DL, IVecVT, Src);
if (IsStrict)
return DAG.getNode(Op.getOpcode(), DL, Op->getVTList(),
Op.getOperand(0), Ext);
return DAG.getNode(Op.getOpcode(), DL, VT, Ext);
}
// FP2Int
assert(SrcEltVT == MVT::f16 && "Unexpected FP_TO_[US]INT lowering");
// Do one doubling fp_extend then complete the operation by converting
// to int.
MVT InterimFVT = MVT::getVectorVT(MVT::f32, VT.getVectorElementCount());
if (IsStrict) {
auto [FExt, Chain] =
DAG.getStrictFPExtendOrRound(Src, Op.getOperand(0), DL, InterimFVT);
return DAG.getNode(Op.getOpcode(), DL, Op->getVTList(), Chain, FExt);
}
SDValue FExt = DAG.getFPExtendOrRound(Src, DL, InterimFVT);
return DAG.getNode(Op.getOpcode(), DL, VT, FExt);
}
// Narrowing conversions
if (SrcEltSize > (2 * EltSize)) {
if (IsInt2FP) {
// One narrowing int_to_fp, then an fp_round.
assert(EltVT == MVT::f16 && "Unexpected [US]_TO_FP lowering");
MVT InterimFVT = MVT::getVectorVT(MVT::f32, VT.getVectorElementCount());
if (IsStrict) {
SDValue Int2FP = DAG.getNode(Op.getOpcode(), DL,
DAG.getVTList(InterimFVT, MVT::Other),
Op.getOperand(0), Src);
SDValue Chain = Int2FP.getValue(1);
return DAG.getStrictFPExtendOrRound(Int2FP, Chain, DL, VT).first;
}
SDValue Int2FP = DAG.getNode(Op.getOpcode(), DL, InterimFVT, Src);
return DAG.getFPExtendOrRound(Int2FP, DL, VT);
}
// FP2Int
// One narrowing fp_to_int, then truncate the integer. If the float isn't
// representable by the integer, the result is poison.
MVT IVecVT = MVT::getVectorVT(MVT::getIntegerVT(SrcEltSize / 2),
VT.getVectorElementCount());
if (IsStrict) {
SDValue FP2Int =
DAG.getNode(Op.getOpcode(), DL, DAG.getVTList(IVecVT, MVT::Other),
Op.getOperand(0), Src);
SDValue Res = DAG.getNode(ISD::TRUNCATE, DL, VT, FP2Int);
return DAG.getMergeValues({Res, FP2Int.getValue(1)}, DL);
}
SDValue FP2Int = DAG.getNode(Op.getOpcode(), DL, IVecVT, Src);
return DAG.getNode(ISD::TRUNCATE, DL, VT, FP2Int);
}
// Scalable vectors can exit here. Patterns will handle equally-sized
// conversions halving/doubling ones.
if (!VT.isFixedLengthVector())
return Op;
// For fixed-length vectors we lower to a custom "VL" node.
unsigned RVVOpc = 0;
switch (Op.getOpcode()) {
default:
llvm_unreachable("Impossible opcode");
case ISD::FP_TO_SINT:
RVVOpc = RISCVISD::VFCVT_RTZ_X_F_VL;
break;
case ISD::FP_TO_UINT:
RVVOpc = RISCVISD::VFCVT_RTZ_XU_F_VL;
break;
case ISD::SINT_TO_FP:
RVVOpc = RISCVISD::SINT_TO_FP_VL;
break;
case ISD::UINT_TO_FP:
RVVOpc = RISCVISD::UINT_TO_FP_VL;
break;
case ISD::STRICT_FP_TO_SINT:
RVVOpc = RISCVISD::STRICT_VFCVT_RTZ_X_F_VL;
break;
case ISD::STRICT_FP_TO_UINT:
RVVOpc = RISCVISD::STRICT_VFCVT_RTZ_XU_F_VL;
break;
case ISD::STRICT_SINT_TO_FP:
RVVOpc = RISCVISD::STRICT_SINT_TO_FP_VL;
break;
case ISD::STRICT_UINT_TO_FP:
RVVOpc = RISCVISD::STRICT_UINT_TO_FP_VL;
break;
}
MVT ContainerVT = getContainerForFixedLengthVector(VT);
MVT SrcContainerVT = getContainerForFixedLengthVector(SrcVT);
assert(ContainerVT.getVectorElementCount() == SrcContainerVT.getVectorElementCount() &&
"Expected same element count");
auto [Mask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
Src = convertToScalableVector(SrcContainerVT, Src, DAG, Subtarget);
if (IsStrict) {
Src = DAG.getNode(RVVOpc, DL, DAG.getVTList(ContainerVT, MVT::Other),
Op.getOperand(0), Src, Mask, VL);
SDValue SubVec = convertFromScalableVector(VT, Src, DAG, Subtarget);
return DAG.getMergeValues({SubVec, Src.getValue(1)}, DL);
}
Src = DAG.getNode(RVVOpc, DL, ContainerVT, Src, Mask, VL);
return convertFromScalableVector(VT, Src, DAG, Subtarget);
}
case ISD::FP_TO_SINT_SAT:
case ISD::FP_TO_UINT_SAT:
return lowerFP_TO_INT_SAT(Op, DAG, Subtarget);
case ISD::FP_TO_BF16: {
// Custom lower to ensure the libcall return is passed in an FPR on hard
// float ABIs.
assert(!Subtarget.isSoftFPABI() && "Unexpected custom legalization");
SDLoc DL(Op);
MakeLibCallOptions CallOptions;
RTLIB::Libcall LC =
RTLIB::getFPROUND(Op.getOperand(0).getValueType(), MVT::bf16);
SDValue Res =
makeLibCall(DAG, LC, MVT::f32, Op.getOperand(0), CallOptions, DL).first;
if (Subtarget.is64Bit())
return DAG.getNode(RISCVISD::FMV_X_ANYEXTW_RV64, DL, MVT::i64, Res);
return DAG.getBitcast(MVT::i32, Res);
}
case ISD::BF16_TO_FP: {
assert(Subtarget.hasStdExtFOrZfinx() && "Unexpected custom legalization");
MVT VT = Op.getSimpleValueType();
SDLoc DL(Op);
Op = DAG.getNode(
ISD::SHL, DL, Op.getOperand(0).getValueType(), Op.getOperand(0),
DAG.getShiftAmountConstant(16, Op.getOperand(0).getValueType(), DL));
SDValue Res = Subtarget.is64Bit()
? DAG.getNode(RISCVISD::FMV_W_X_RV64, DL, MVT::f32, Op)
: DAG.getBitcast(MVT::f32, Op);
// fp_extend if the target VT is bigger than f32.
if (VT != MVT::f32)
return DAG.getNode(ISD::FP_EXTEND, DL, VT, Res);
return Res;
}
case ISD::STRICT_FP_TO_FP16:
case ISD::FP_TO_FP16: {
// Custom lower to ensure the libcall return is passed in an FPR on hard
// float ABIs.
assert(Subtarget.hasStdExtFOrZfinx() && "Unexpected custom legalisation");
SDLoc DL(Op);
MakeLibCallOptions CallOptions;
bool IsStrict = Op->isStrictFPOpcode();
SDValue Op0 = IsStrict ? Op.getOperand(1) : Op.getOperand(0);
SDValue Chain = IsStrict ? Op.getOperand(0) : SDValue();
RTLIB::Libcall LC = RTLIB::getFPROUND(Op0.getValueType(), MVT::f16);
SDValue Res;
std::tie(Res, Chain) =
makeLibCall(DAG, LC, MVT::f32, Op0, CallOptions, DL, Chain);
if (Subtarget.is64Bit())
return DAG.getNode(RISCVISD::FMV_X_ANYEXTW_RV64, DL, MVT::i64, Res);
SDValue Result = DAG.getBitcast(MVT::i32, IsStrict ? Res.getValue(0) : Res);
if (IsStrict)
return DAG.getMergeValues({Result, Chain}, DL);
return Result;
}
case ISD::STRICT_FP16_TO_FP:
case ISD::FP16_TO_FP: {
// Custom lower to ensure the libcall argument is passed in an FPR on hard
// float ABIs.
assert(Subtarget.hasStdExtFOrZfinx() && "Unexpected custom legalisation");
SDLoc DL(Op);
MakeLibCallOptions CallOptions;
bool IsStrict = Op->isStrictFPOpcode();
SDValue Op0 = IsStrict ? Op.getOperand(1) : Op.getOperand(0);
SDValue Chain = IsStrict ? Op.getOperand(0) : SDValue();
SDValue Arg = Subtarget.is64Bit()
? DAG.getNode(RISCVISD::FMV_W_X_RV64, DL, MVT::f32, Op0)
: DAG.getBitcast(MVT::f32, Op0);
SDValue Res;
std::tie(Res, Chain) = makeLibCall(DAG, RTLIB::FPEXT_F16_F32, MVT::f32, Arg,
CallOptions, DL, Chain);
if (IsStrict)
return DAG.getMergeValues({Res, Chain}, DL);
return Res;
}
case ISD::FTRUNC:
case ISD::FCEIL:
case ISD::FFLOOR:
case ISD::FNEARBYINT:
case ISD::FRINT:
case ISD::FROUND:
case ISD::FROUNDEVEN:
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitVectorOp(Op, DAG);
return lowerFTRUNC_FCEIL_FFLOOR_FROUND(Op, DAG, Subtarget);
case ISD::LRINT:
case ISD::LLRINT:
case ISD::LROUND:
case ISD::LLROUND: {
if (Op.getValueType().isVector())
return lowerVectorXRINT_XROUND(Op, DAG, Subtarget);
assert(Op.getOperand(0).getValueType() == MVT::f16 &&
"Unexpected custom legalisation");
SDLoc DL(Op);
SDValue Ext = DAG.getNode(ISD::FP_EXTEND, DL, MVT::f32, Op.getOperand(0));
return DAG.getNode(Op.getOpcode(), DL, Op.getValueType(), Ext);
}
case ISD::STRICT_LRINT:
case ISD::STRICT_LLRINT:
case ISD::STRICT_LROUND:
case ISD::STRICT_LLROUND: {
assert(Op.getOperand(1).getValueType() == MVT::f16 &&
"Unexpected custom legalisation");
SDLoc DL(Op);
SDValue Ext = DAG.getNode(ISD::STRICT_FP_EXTEND, DL, {MVT::f32, MVT::Other},
{Op.getOperand(0), Op.getOperand(1)});
return DAG.getNode(Op.getOpcode(), DL, {Op.getValueType(), MVT::Other},
{Ext.getValue(1), Ext.getValue(0)});
}
case ISD::VECREDUCE_ADD:
case ISD::VECREDUCE_UMAX:
case ISD::VECREDUCE_SMAX:
case ISD::VECREDUCE_UMIN:
case ISD::VECREDUCE_SMIN:
return lowerVECREDUCE(Op, DAG);
case ISD::VECREDUCE_AND:
case ISD::VECREDUCE_OR:
case ISD::VECREDUCE_XOR:
if (Op.getOperand(0).getValueType().getVectorElementType() == MVT::i1)
return lowerVectorMaskVecReduction(Op, DAG, /*IsVP*/ false);
return lowerVECREDUCE(Op, DAG);
case ISD::VECREDUCE_FADD:
case ISD::VECREDUCE_SEQ_FADD:
case ISD::VECREDUCE_FMIN:
case ISD::VECREDUCE_FMAX:
case ISD::VECREDUCE_FMAXIMUM:
case ISD::VECREDUCE_FMINIMUM:
return lowerFPVECREDUCE(Op, DAG);
case ISD::VP_REDUCE_ADD:
case ISD::VP_REDUCE_UMAX:
case ISD::VP_REDUCE_SMAX:
case ISD::VP_REDUCE_UMIN:
case ISD::VP_REDUCE_SMIN:
case ISD::VP_REDUCE_FADD:
case ISD::VP_REDUCE_SEQ_FADD:
case ISD::VP_REDUCE_FMIN:
case ISD::VP_REDUCE_FMAX:
case ISD::VP_REDUCE_FMINIMUM:
case ISD::VP_REDUCE_FMAXIMUM:
if (isPromotedOpNeedingSplit(Op.getOperand(1), Subtarget))
return SplitVectorReductionOp(Op, DAG);
return lowerVPREDUCE(Op, DAG);
case ISD::VP_REDUCE_AND:
case ISD::VP_REDUCE_OR:
case ISD::VP_REDUCE_XOR:
if (Op.getOperand(1).getValueType().getVectorElementType() == MVT::i1)
return lowerVectorMaskVecReduction(Op, DAG, /*IsVP*/ true);
return lowerVPREDUCE(Op, DAG);
case ISD::VP_CTTZ_ELTS:
case ISD::VP_CTTZ_ELTS_ZERO_UNDEF:
return lowerVPCttzElements(Op, DAG);
case ISD::UNDEF: {
MVT ContainerVT = getContainerForFixedLengthVector(Op.getSimpleValueType());
return convertFromScalableVector(Op.getSimpleValueType(),
DAG.getUNDEF(ContainerVT), DAG, Subtarget);
}
case ISD::INSERT_SUBVECTOR:
return lowerINSERT_SUBVECTOR(Op, DAG);
case ISD::EXTRACT_SUBVECTOR:
return lowerEXTRACT_SUBVECTOR(Op, DAG);
case ISD::VECTOR_DEINTERLEAVE:
return lowerVECTOR_DEINTERLEAVE(Op, DAG);
case ISD::VECTOR_INTERLEAVE:
return lowerVECTOR_INTERLEAVE(Op, DAG);
case ISD::STEP_VECTOR:
return lowerSTEP_VECTOR(Op, DAG);
case ISD::VECTOR_REVERSE:
return lowerVECTOR_REVERSE(Op, DAG);
case ISD::VECTOR_SPLICE:
return lowerVECTOR_SPLICE(Op, DAG);
case ISD::BUILD_VECTOR: {
MVT VT = Op.getSimpleValueType();
MVT EltVT = VT.getVectorElementType();
if (!Subtarget.is64Bit() && EltVT == MVT::i64)
return lowerBuildVectorViaVID(Op, DAG, Subtarget);
return lowerBUILD_VECTOR(Op, DAG, Subtarget);
}
case ISD::SPLAT_VECTOR: {
MVT VT = Op.getSimpleValueType();
MVT EltVT = VT.getVectorElementType();
if ((EltVT == MVT::f16 && !Subtarget.hasStdExtZvfh()) ||
EltVT == MVT::bf16) {
SDLoc DL(Op);
SDValue Elt;
if ((EltVT == MVT::bf16 && Subtarget.hasStdExtZfbfmin()) ||
(EltVT == MVT::f16 && Subtarget.hasStdExtZfhmin()))
Elt = DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, Subtarget.getXLenVT(),
Op.getOperand(0));
else
Elt = DAG.getNode(ISD::BITCAST, DL, MVT::i16, Op.getOperand(0));
MVT IVT = VT.changeVectorElementType(MVT::i16);
return DAG.getNode(ISD::BITCAST, DL, VT,
DAG.getNode(ISD::SPLAT_VECTOR, DL, IVT, Elt));
}
if (EltVT == MVT::i1)
return lowerVectorMaskSplat(Op, DAG);
return SDValue();
}
case ISD::VECTOR_SHUFFLE:
return lowerVECTOR_SHUFFLE(Op, DAG, Subtarget);
case ISD::CONCAT_VECTORS: {
// Split CONCAT_VECTORS into a series of INSERT_SUBVECTOR nodes. This is
// better than going through the stack, as the default expansion does.
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector())
ContainerVT = ::getContainerForFixedLengthVector(DAG, VT, Subtarget);
// Recursively split concat_vectors with more than 2 operands:
//
// concat_vector op1, op2, op3, op4
// ->
// concat_vector (concat_vector op1, op2), (concat_vector op3, op4)
//
// This reduces the length of the chain of vslideups and allows us to
// perform the vslideups at a smaller LMUL, limited to MF2.
if (Op.getNumOperands() > 2 &&
ContainerVT.bitsGE(RISCVTargetLowering::getM1VT(ContainerVT))) {
MVT HalfVT = VT.getHalfNumVectorElementsVT();
assert(isPowerOf2_32(Op.getNumOperands()));
size_t HalfNumOps = Op.getNumOperands() / 2;
SDValue Lo = DAG.getNode(ISD::CONCAT_VECTORS, DL, HalfVT,
Op->ops().take_front(HalfNumOps));
SDValue Hi = DAG.getNode(ISD::CONCAT_VECTORS, DL, HalfVT,
Op->ops().drop_front(HalfNumOps));
return DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, Lo, Hi);
}
unsigned NumOpElts =
Op.getOperand(0).getSimpleValueType().getVectorMinNumElements();
SDValue Vec = DAG.getUNDEF(VT);
for (const auto &OpIdx : enumerate(Op->ops())) {
SDValue SubVec = OpIdx.value();
// Don't insert undef subvectors.
if (SubVec.isUndef())
continue;
Vec = DAG.getInsertSubvector(DL, Vec, SubVec, OpIdx.index() * NumOpElts);
}
return Vec;
}
case ISD::LOAD: {
auto *Load = cast<LoadSDNode>(Op);
EVT VT = Load->getValueType(0);
if (VT == MVT::f64) {
assert(Subtarget.hasStdExtZdinx() && !Subtarget.hasStdExtZilsd() &&
!Subtarget.is64Bit() && "Unexpected custom legalisation");
// Replace a double precision load with two i32 loads and a BuildPairF64.
SDLoc DL(Op);
SDValue BasePtr = Load->getBasePtr();
SDValue Chain = Load->getChain();
SDValue Lo =
DAG.getLoad(MVT::i32, DL, Chain, BasePtr, Load->getPointerInfo(),
Load->getBaseAlign(), Load->getMemOperand()->getFlags());
BasePtr = DAG.getObjectPtrOffset(DL, BasePtr, TypeSize::getFixed(4));
SDValue Hi = DAG.getLoad(
MVT::i32, DL, Chain, BasePtr, Load->getPointerInfo().getWithOffset(4),
Load->getBaseAlign(), Load->getMemOperand()->getFlags());
Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo.getValue(1),
Hi.getValue(1));
SDValue Pair = DAG.getNode(RISCVISD::BuildPairF64, DL, MVT::f64, Lo, Hi);
return DAG.getMergeValues({Pair, Chain}, DL);
}
if (VT == MVT::bf16)
return lowerXAndesBfHCvtBFloat16Load(Op, DAG);
// Handle normal vector tuple load.
if (VT.isRISCVVectorTuple()) {
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
unsigned NF = VT.getRISCVVectorTupleNumFields();
unsigned Sz = VT.getSizeInBits().getKnownMinValue();
unsigned NumElts = Sz / (NF * 8);
int Log2LMUL = Log2_64(NumElts) - 3;
auto Flag = SDNodeFlags();
Flag.setNoUnsignedWrap(true);
SDValue Ret = DAG.getUNDEF(VT);
SDValue BasePtr = Load->getBasePtr();
SDValue VROffset = DAG.getNode(RISCVISD::READ_VLENB, DL, XLenVT);
VROffset =
DAG.getNode(ISD::SHL, DL, XLenVT, VROffset,
DAG.getConstant(std::max(Log2LMUL, 0), DL, XLenVT));
SmallVector<SDValue, 8> OutChains;
// Load NF vector registers and combine them to a vector tuple.
for (unsigned i = 0; i < NF; ++i) {
SDValue LoadVal = DAG.getLoad(
MVT::getScalableVectorVT(MVT::i8, NumElts), DL, Load->getChain(),
BasePtr, MachinePointerInfo(Load->getAddressSpace()), Align(8));
OutChains.push_back(LoadVal.getValue(1));
Ret = DAG.getNode(RISCVISD::TUPLE_INSERT, DL, VT, Ret, LoadVal,
DAG.getTargetConstant(i, DL, MVT::i32));
BasePtr = DAG.getNode(ISD::ADD, DL, XLenVT, BasePtr, VROffset, Flag);
}
return DAG.getMergeValues(
{Ret, DAG.getNode(ISD::TokenFactor, DL, MVT::Other, OutChains)}, DL);
}
if (auto V = expandUnalignedRVVLoad(Op, DAG))
return V;
if (Op.getValueType().isFixedLengthVector())
return lowerFixedLengthVectorLoadToRVV(Op, DAG);
return Op;
}
case ISD::STORE: {
auto *Store = cast<StoreSDNode>(Op);
SDValue StoredVal = Store->getValue();
EVT VT = StoredVal.getValueType();
if (VT == MVT::f64) {
assert(Subtarget.hasStdExtZdinx() && !Subtarget.hasStdExtZilsd() &&
!Subtarget.is64Bit() && "Unexpected custom legalisation");
// Replace a double precision store with a SplitF64 and i32 stores.
SDValue DL(Op);
SDValue BasePtr = Store->getBasePtr();
SDValue Chain = Store->getChain();
SDValue Split = DAG.getNode(RISCVISD::SplitF64, DL,
DAG.getVTList(MVT::i32, MVT::i32), StoredVal);
SDValue Lo = DAG.getStore(Chain, DL, Split.getValue(0), BasePtr,
Store->getPointerInfo(), Store->getBaseAlign(),
Store->getMemOperand()->getFlags());
BasePtr = DAG.getObjectPtrOffset(DL, BasePtr, TypeSize::getFixed(4));
SDValue Hi = DAG.getStore(Chain, DL, Split.getValue(1), BasePtr,
Store->getPointerInfo().getWithOffset(4),
Store->getBaseAlign(),
Store->getMemOperand()->getFlags());
return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Lo, Hi);
}
if (VT == MVT::i64) {
assert(Subtarget.hasStdExtZilsd() && !Subtarget.is64Bit() &&
"Unexpected custom legalisation");
if (Store->isTruncatingStore())
return SDValue();
if (!Subtarget.enableUnalignedScalarMem() && Store->getAlign() < 8)
return SDValue();
SDLoc DL(Op);
SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, StoredVal,
DAG.getTargetConstant(0, DL, MVT::i32));
SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, StoredVal,
DAG.getTargetConstant(1, DL, MVT::i32));
return DAG.getMemIntrinsicNode(
RISCVISD::SD_RV32, DL, DAG.getVTList(MVT::Other),
{Store->getChain(), Lo, Hi, Store->getBasePtr()}, MVT::i64,
Store->getMemOperand());
}
if (VT == MVT::bf16)
return lowerXAndesBfHCvtBFloat16Store(Op, DAG);
// Handle normal vector tuple store.
if (VT.isRISCVVectorTuple()) {
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
unsigned NF = VT.getRISCVVectorTupleNumFields();
unsigned Sz = VT.getSizeInBits().getKnownMinValue();
unsigned NumElts = Sz / (NF * 8);
int Log2LMUL = Log2_64(NumElts) - 3;
auto Flag = SDNodeFlags();
Flag.setNoUnsignedWrap(true);
SDValue Ret;
SDValue Chain = Store->getChain();
SDValue BasePtr = Store->getBasePtr();
SDValue VROffset = DAG.getNode(RISCVISD::READ_VLENB, DL, XLenVT);
VROffset =
DAG.getNode(ISD::SHL, DL, XLenVT, VROffset,
DAG.getConstant(std::max(Log2LMUL, 0), DL, XLenVT));
// Extract subregisters in a vector tuple and store them individually.
for (unsigned i = 0; i < NF; ++i) {
auto Extract =
DAG.getNode(RISCVISD::TUPLE_EXTRACT, DL,
MVT::getScalableVectorVT(MVT::i8, NumElts), StoredVal,
DAG.getTargetConstant(i, DL, MVT::i32));
Ret = DAG.getStore(Chain, DL, Extract, BasePtr,
MachinePointerInfo(Store->getAddressSpace()),
Store->getBaseAlign(),
Store->getMemOperand()->getFlags());
Chain = Ret.getValue(0);
BasePtr = DAG.getNode(ISD::ADD, DL, XLenVT, BasePtr, VROffset, Flag);
}
return Ret;
}
if (auto V = expandUnalignedRVVStore(Op, DAG))
return V;
if (Op.getOperand(1).getValueType().isFixedLengthVector())
return lowerFixedLengthVectorStoreToRVV(Op, DAG);
return Op;
}
case ISD::MLOAD:
case ISD::VP_LOAD:
return lowerMaskedLoad(Op, DAG);
case ISD::VP_LOAD_FF:
return lowerLoadFF(Op, DAG);
case ISD::MSTORE:
case ISD::VP_STORE:
return lowerMaskedStore(Op, DAG);
case ISD::VECTOR_COMPRESS:
return lowerVectorCompress(Op, DAG);
case ISD::SELECT_CC: {
// This occurs because we custom legalize SETGT and SETUGT for setcc. That
// causes LegalizeDAG to think we need to custom legalize select_cc. Expand
// into separate SETCC+SELECT just like LegalizeDAG.
SDValue Tmp1 = Op.getOperand(0);
SDValue Tmp2 = Op.getOperand(1);
SDValue True = Op.getOperand(2);
SDValue False = Op.getOperand(3);
EVT VT = Op.getValueType();
SDValue CC = Op.getOperand(4);
EVT CmpVT = Tmp1.getValueType();
EVT CCVT =
getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), CmpVT);
SDLoc DL(Op);
SDValue Cond =
DAG.getNode(ISD::SETCC, DL, CCVT, Tmp1, Tmp2, CC, Op->getFlags());
return DAG.getSelect(DL, VT, Cond, True, False);
}
case ISD::SETCC: {
MVT OpVT = Op.getOperand(0).getSimpleValueType();
if (OpVT.isScalarInteger()) {
MVT VT = Op.getSimpleValueType();
SDValue LHS = Op.getOperand(0);
SDValue RHS = Op.getOperand(1);
ISD::CondCode CCVal = cast<CondCodeSDNode>(Op.getOperand(2))->get();
assert((CCVal == ISD::SETGT || CCVal == ISD::SETUGT) &&
"Unexpected CondCode");
SDLoc DL(Op);
// If the RHS is a constant in the range [-2049, 0) or (0, 2046], we can
// convert this to the equivalent of (set(u)ge X, C+1) by using
// (xori (slti(u) X, C+1), 1). This avoids materializing a small constant
// in a register.
if (isa<ConstantSDNode>(RHS)) {
int64_t Imm = cast<ConstantSDNode>(RHS)->getSExtValue();
if (Imm != 0 && isInt<12>((uint64_t)Imm + 1)) {
// If this is an unsigned compare and the constant is -1, incrementing
// the constant would change behavior. The result should be false.
if (CCVal == ISD::SETUGT && Imm == -1)
return DAG.getConstant(0, DL, VT);
// Using getSetCCSwappedOperands will convert SET(U)GT->SET(U)LT.
CCVal = ISD::getSetCCSwappedOperands(CCVal);
SDValue SetCC = DAG.getSetCC(
DL, VT, LHS, DAG.getSignedConstant(Imm + 1, DL, OpVT), CCVal);
return DAG.getLogicalNOT(DL, SetCC, VT);
}
// Lower (setugt X, 2047) as (setne (srl X, 11), 0).
if (CCVal == ISD::SETUGT && Imm == 2047) {
SDValue Shift = DAG.getNode(ISD::SRL, DL, OpVT, LHS,
DAG.getShiftAmountConstant(11, OpVT, DL));
return DAG.getSetCC(DL, VT, Shift, DAG.getConstant(0, DL, OpVT),
ISD::SETNE);
}
}
// Not a constant we could handle, swap the operands and condition code to
// SETLT/SETULT.
CCVal = ISD::getSetCCSwappedOperands(CCVal);
return DAG.getSetCC(DL, VT, RHS, LHS, CCVal);
}
if (isPromotedOpNeedingSplit(Op.getOperand(0), Subtarget))
return SplitVectorOp(Op, DAG);
return lowerToScalableOp(Op, DAG);
}
case ISD::ADD:
case ISD::SUB:
case ISD::MUL:
case ISD::MULHS:
case ISD::MULHU:
case ISD::AND:
case ISD::OR:
case ISD::XOR:
case ISD::SDIV:
case ISD::SREM:
case ISD::UDIV:
case ISD::UREM:
case ISD::BSWAP:
case ISD::CTPOP:
case ISD::VSELECT:
return lowerToScalableOp(Op, DAG);
case ISD::SHL:
case ISD::SRA:
case ISD::SRL:
if (Op.getSimpleValueType().isFixedLengthVector())
return lowerToScalableOp(Op, DAG);
// This can be called for an i32 shift amount that needs to be promoted.
assert(Op.getOperand(1).getValueType() == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
return SDValue();
case ISD::FABS:
case ISD::FNEG:
if (Op.getValueType() == MVT::f16 || Op.getValueType() == MVT::bf16)
return lowerFABSorFNEG(Op, DAG, Subtarget);
[[fallthrough]];
case ISD::FADD:
case ISD::FSUB:
case ISD::FMUL:
case ISD::FDIV:
case ISD::FSQRT:
case ISD::FMA:
case ISD::FMINNUM:
case ISD::FMAXNUM:
case ISD::FMINIMUMNUM:
case ISD::FMAXIMUMNUM:
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitVectorOp(Op, DAG);
[[fallthrough]];
case ISD::AVGFLOORS:
case ISD::AVGFLOORU:
case ISD::AVGCEILS:
case ISD::AVGCEILU:
case ISD::SMIN:
case ISD::SMAX:
case ISD::UMIN:
case ISD::UMAX:
case ISD::UADDSAT:
case ISD::USUBSAT:
case ISD::SADDSAT:
case ISD::SSUBSAT:
return lowerToScalableOp(Op, DAG);
case ISD::ABDS:
case ISD::ABDU: {
SDLoc dl(Op);
EVT VT = Op->getValueType(0);
SDValue LHS = DAG.getFreeze(Op->getOperand(0));
SDValue RHS = DAG.getFreeze(Op->getOperand(1));
bool IsSigned = Op->getOpcode() == ISD::ABDS;
// abds(lhs, rhs) -> sub(smax(lhs,rhs), smin(lhs,rhs))
// abdu(lhs, rhs) -> sub(umax(lhs,rhs), umin(lhs,rhs))
unsigned MaxOpc = IsSigned ? ISD::SMAX : ISD::UMAX;
unsigned MinOpc = IsSigned ? ISD::SMIN : ISD::UMIN;
SDValue Max = DAG.getNode(MaxOpc, dl, VT, LHS, RHS);
SDValue Min = DAG.getNode(MinOpc, dl, VT, LHS, RHS);
return DAG.getNode(ISD::SUB, dl, VT, Max, Min);
}
case ISD::ABS:
case ISD::VP_ABS:
return lowerABS(Op, DAG);
case ISD::CTLZ:
case ISD::CTLZ_ZERO_UNDEF:
case ISD::CTTZ:
case ISD::CTTZ_ZERO_UNDEF:
if (Subtarget.hasStdExtZvbb())
return lowerToScalableOp(Op, DAG);
assert(Op.getOpcode() != ISD::CTTZ);
return lowerCTLZ_CTTZ_ZERO_UNDEF(Op, DAG);
case ISD::FCOPYSIGN:
if (Op.getValueType() == MVT::f16 || Op.getValueType() == MVT::bf16)
return lowerFCOPYSIGN(Op, DAG, Subtarget);
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitVectorOp(Op, DAG);
return lowerToScalableOp(Op, DAG);
case ISD::STRICT_FADD:
case ISD::STRICT_FSUB:
case ISD::STRICT_FMUL:
case ISD::STRICT_FDIV:
case ISD::STRICT_FSQRT:
case ISD::STRICT_FMA:
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitStrictFPVectorOp(Op, DAG);
return lowerToScalableOp(Op, DAG);
case ISD::STRICT_FSETCC:
case ISD::STRICT_FSETCCS:
return lowerVectorStrictFSetcc(Op, DAG);
case ISD::STRICT_FCEIL:
case ISD::STRICT_FRINT:
case ISD::STRICT_FFLOOR:
case ISD::STRICT_FTRUNC:
case ISD::STRICT_FNEARBYINT:
case ISD::STRICT_FROUND:
case ISD::STRICT_FROUNDEVEN:
return lowerVectorStrictFTRUNC_FCEIL_FFLOOR_FROUND(Op, DAG, Subtarget);
case ISD::MGATHER:
case ISD::VP_GATHER:
return lowerMaskedGather(Op, DAG);
case ISD::MSCATTER:
case ISD::VP_SCATTER:
return lowerMaskedScatter(Op, DAG);
case ISD::GET_ROUNDING:
return lowerGET_ROUNDING(Op, DAG);
case ISD::SET_ROUNDING:
return lowerSET_ROUNDING(Op, DAG);
case ISD::GET_FPENV:
return lowerGET_FPENV(Op, DAG);
case ISD::SET_FPENV:
return lowerSET_FPENV(Op, DAG);
case ISD::RESET_FPENV:
return lowerRESET_FPENV(Op, DAG);
case ISD::GET_FPMODE:
return lowerGET_FPMODE(Op, DAG);
case ISD::SET_FPMODE:
return lowerSET_FPMODE(Op, DAG);
case ISD::RESET_FPMODE:
return lowerRESET_FPMODE(Op, DAG);
case ISD::EH_DWARF_CFA:
return lowerEH_DWARF_CFA(Op, DAG);
case ISD::VP_MERGE:
if (Op.getSimpleValueType().getVectorElementType() == MVT::i1)
return lowerVPMergeMask(Op, DAG);
[[fallthrough]];
case ISD::VP_SELECT:
case ISD::VP_ADD:
case ISD::VP_SUB:
case ISD::VP_MUL:
case ISD::VP_SDIV:
case ISD::VP_UDIV:
case ISD::VP_SREM:
case ISD::VP_UREM:
case ISD::VP_UADDSAT:
case ISD::VP_USUBSAT:
case ISD::VP_SADDSAT:
case ISD::VP_SSUBSAT:
case ISD::VP_LRINT:
case ISD::VP_LLRINT:
return lowerVPOp(Op, DAG);
case ISD::VP_AND:
case ISD::VP_OR:
case ISD::VP_XOR:
return lowerLogicVPOp(Op, DAG);
case ISD::VP_FADD:
case ISD::VP_FSUB:
case ISD::VP_FMUL:
case ISD::VP_FDIV:
case ISD::VP_FNEG:
case ISD::VP_FABS:
case ISD::VP_SQRT:
case ISD::VP_FMA:
case ISD::VP_FMINNUM:
case ISD::VP_FMAXNUM:
case ISD::VP_FCOPYSIGN:
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitVPOp(Op, DAG);
[[fallthrough]];
case ISD::VP_SRA:
case ISD::VP_SRL:
case ISD::VP_SHL:
return lowerVPOp(Op, DAG);
case ISD::VP_IS_FPCLASS:
return LowerIS_FPCLASS(Op, DAG);
case ISD::VP_SIGN_EXTEND:
case ISD::VP_ZERO_EXTEND:
if (Op.getOperand(0).getSimpleValueType().getVectorElementType() == MVT::i1)
return lowerVPExtMaskOp(Op, DAG);
return lowerVPOp(Op, DAG);
case ISD::VP_TRUNCATE:
return lowerVectorTruncLike(Op, DAG);
case ISD::VP_FP_EXTEND:
case ISD::VP_FP_ROUND:
return lowerVectorFPExtendOrRoundLike(Op, DAG);
case ISD::VP_SINT_TO_FP:
case ISD::VP_UINT_TO_FP:
if (Op.getValueType().isVector() &&
((Op.getValueType().getScalarType() == MVT::f16 &&
(Subtarget.hasVInstructionsF16Minimal() &&
!Subtarget.hasVInstructionsF16())) ||
Op.getValueType().getScalarType() == MVT::bf16)) {
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitVectorOp(Op, DAG);
// int -> f32
SDLoc DL(Op);
MVT NVT =
MVT::getVectorVT(MVT::f32, Op.getValueType().getVectorElementCount());
auto NC = DAG.getNode(Op.getOpcode(), DL, NVT, Op->ops());
// f32 -> [b]f16
return DAG.getNode(ISD::FP_ROUND, DL, Op.getValueType(), NC,
DAG.getIntPtrConstant(0, DL, /*isTarget=*/true));
}
[[fallthrough]];
case ISD::VP_FP_TO_SINT:
case ISD::VP_FP_TO_UINT:
if (SDValue Op1 = Op.getOperand(0);
Op1.getValueType().isVector() &&
((Op1.getValueType().getScalarType() == MVT::f16 &&
(Subtarget.hasVInstructionsF16Minimal() &&
!Subtarget.hasVInstructionsF16())) ||
Op1.getValueType().getScalarType() == MVT::bf16)) {
if (isPromotedOpNeedingSplit(Op1, Subtarget))
return SplitVectorOp(Op, DAG);
// [b]f16 -> f32
SDLoc DL(Op);
MVT NVT = MVT::getVectorVT(MVT::f32,
Op1.getValueType().getVectorElementCount());
SDValue WidenVec = DAG.getNode(ISD::FP_EXTEND, DL, NVT, Op1);
// f32 -> int
return DAG.getNode(Op.getOpcode(), DL, Op.getValueType(),
{WidenVec, Op.getOperand(1), Op.getOperand(2)});
}
return lowerVPFPIntConvOp(Op, DAG);
case ISD::VP_SETCC:
if (isPromotedOpNeedingSplit(Op.getOperand(0), Subtarget))
return SplitVPOp(Op, DAG);
if (Op.getOperand(0).getSimpleValueType().getVectorElementType() == MVT::i1)
return lowerVPSetCCMaskOp(Op, DAG);
[[fallthrough]];
case ISD::VP_SMIN:
case ISD::VP_SMAX:
case ISD::VP_UMIN:
case ISD::VP_UMAX:
case ISD::VP_BITREVERSE:
case ISD::VP_BSWAP:
return lowerVPOp(Op, DAG);
case ISD::VP_CTLZ:
case ISD::VP_CTLZ_ZERO_UNDEF:
if (Subtarget.hasStdExtZvbb())
return lowerVPOp(Op, DAG);
return lowerCTLZ_CTTZ_ZERO_UNDEF(Op, DAG);
case ISD::VP_CTTZ:
case ISD::VP_CTTZ_ZERO_UNDEF:
if (Subtarget.hasStdExtZvbb())
return lowerVPOp(Op, DAG);
return lowerCTLZ_CTTZ_ZERO_UNDEF(Op, DAG);
case ISD::VP_CTPOP:
return lowerVPOp(Op, DAG);
case ISD::EXPERIMENTAL_VP_STRIDED_LOAD:
return lowerVPStridedLoad(Op, DAG);
case ISD::EXPERIMENTAL_VP_STRIDED_STORE:
return lowerVPStridedStore(Op, DAG);
case ISD::VP_FCEIL:
case ISD::VP_FFLOOR:
case ISD::VP_FRINT:
case ISD::VP_FNEARBYINT:
case ISD::VP_FROUND:
case ISD::VP_FROUNDEVEN:
case ISD::VP_FROUNDTOZERO:
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitVPOp(Op, DAG);
return lowerVectorFTRUNC_FCEIL_FFLOOR_FROUND(Op, DAG, Subtarget);
case ISD::VP_FMAXIMUM:
case ISD::VP_FMINIMUM:
if (isPromotedOpNeedingSplit(Op, Subtarget))
return SplitVPOp(Op, DAG);
return lowerFMAXIMUM_FMINIMUM(Op, DAG, Subtarget);
case ISD::EXPERIMENTAL_VP_SPLICE:
return lowerVPSpliceExperimental(Op, DAG);
case ISD::EXPERIMENTAL_VP_REVERSE:
return lowerVPReverseExperimental(Op, DAG);
case ISD::EXPERIMENTAL_VP_SPLAT:
return lowerVPSplatExperimental(Op, DAG);
case ISD::CLEAR_CACHE: {
assert(getTargetMachine().getTargetTriple().isOSLinux() &&
"llvm.clear_cache only needs custom lower on Linux targets");
SDLoc DL(Op);
SDValue Flags = DAG.getConstant(0, DL, Subtarget.getXLenVT());
return emitFlushICache(DAG, Op.getOperand(0), Op.getOperand(1),
Op.getOperand(2), Flags, DL);
}
case ISD::DYNAMIC_STACKALLOC:
return lowerDYNAMIC_STACKALLOC(Op, DAG);
case ISD::INIT_TRAMPOLINE:
return lowerINIT_TRAMPOLINE(Op, DAG);
case ISD::ADJUST_TRAMPOLINE:
return lowerADJUST_TRAMPOLINE(Op, DAG);
case ISD::PARTIAL_REDUCE_UMLA:
case ISD::PARTIAL_REDUCE_SMLA:
case ISD::PARTIAL_REDUCE_SUMLA:
return lowerPARTIAL_REDUCE_MLA(Op, DAG);
}
}
SDValue RISCVTargetLowering::emitFlushICache(SelectionDAG &DAG, SDValue InChain,
SDValue Start, SDValue End,
SDValue Flags, SDLoc DL) const {
MakeLibCallOptions CallOptions;
std::pair<SDValue, SDValue> CallResult =
makeLibCall(DAG, RTLIB::RISCV_FLUSH_ICACHE, MVT::isVoid,
{Start, End, Flags}, CallOptions, DL, InChain);
// This function returns void so only the out chain matters.
return CallResult.second;
}
SDValue RISCVTargetLowering::lowerINIT_TRAMPOLINE(SDValue Op,
SelectionDAG &DAG) const {
if (!Subtarget.is64Bit())
llvm::reportFatalUsageError("Trampolines only implemented for RV64");
// Create an MCCodeEmitter to encode instructions.
TargetLoweringObjectFile *TLO = getTargetMachine().getObjFileLowering();
assert(TLO);
MCContext &MCCtx = TLO->getContext();
std::unique_ptr<MCCodeEmitter> CodeEmitter(
createRISCVMCCodeEmitter(*getTargetMachine().getMCInstrInfo(), MCCtx));
SDValue Root = Op.getOperand(0);
SDValue Trmp = Op.getOperand(1); // trampoline
SDLoc dl(Op);
const Value *TrmpAddr = cast<SrcValueSDNode>(Op.getOperand(4))->getValue();
// We store in the trampoline buffer the following instructions and data.
// Offset:
// 0: auipc t2, 0
// 4: ld t0, 24(t2)
// 8: ld t2, 16(t2)
// 12: jalr t0
// 16: <StaticChainOffset>
// 24: <FunctionAddressOffset>
// 32:
// Offset with branch control flow protection enabled:
// 0: lpad <imm20>
// 4: auipc t3, 0
// 8: ld t2, 28(t3)
// 12: ld t3, 20(t3)
// 16: jalr t2
// 20: <StaticChainOffset>
// 28: <FunctionAddressOffset>
// 36:
const bool HasCFBranch =
Subtarget.hasStdExtZicfilp() &&
DAG.getMachineFunction().getFunction().getParent()->getModuleFlag(
"cf-protection-branch");
const unsigned StaticChainIdx = HasCFBranch ? 5 : 4;
const unsigned StaticChainOffset = StaticChainIdx * 4;
const unsigned FunctionAddressOffset = StaticChainOffset + 8;
const MCSubtargetInfo *STI = getTargetMachine().getMCSubtargetInfo();
assert(STI);
auto GetEncoding = [&](const MCInst &MC) {
SmallVector<char, 4> CB;
SmallVector<MCFixup> Fixups;
CodeEmitter->encodeInstruction(MC, CB, Fixups, *STI);
uint32_t Encoding = support::endian::read32le(CB.data());
return Encoding;
};
SmallVector<SDValue> OutChains;
SmallVector<uint32_t> Encodings;
if (!HasCFBranch) {
Encodings.append(
{// auipc t2, 0
// Loads the current PC into t2.
GetEncoding(MCInstBuilder(RISCV::AUIPC).addReg(RISCV::X7).addImm(0)),
// ld t0, 24(t2)
// Loads the function address into t0. Note that we are using offsets
// pc-relative to the first instruction of the trampoline.
GetEncoding(MCInstBuilder(RISCV::LD)
.addReg(RISCV::X5)
.addReg(RISCV::X7)
.addImm(FunctionAddressOffset)),
// ld t2, 16(t2)
// Load the value of the static chain.
GetEncoding(MCInstBuilder(RISCV::LD)
.addReg(RISCV::X7)
.addReg(RISCV::X7)
.addImm(StaticChainOffset)),
// jalr t0
// Jump to the function.
GetEncoding(MCInstBuilder(RISCV::JALR)
.addReg(RISCV::X0)
.addReg(RISCV::X5)
.addImm(0))});
} else {
Encodings.append(
{// auipc x0, <imm20> (lpad <imm20>)
// Landing pad.
GetEncoding(MCInstBuilder(RISCV::AUIPC).addReg(RISCV::X0).addImm(0)),
// auipc t3, 0
// Loads the current PC into t3.
GetEncoding(MCInstBuilder(RISCV::AUIPC).addReg(RISCV::X28).addImm(0)),
// ld t2, (FunctionAddressOffset - 4)(t3)
// Loads the function address into t2. Note that we are using offsets
// pc-relative to the SECOND instruction of the trampoline.
GetEncoding(MCInstBuilder(RISCV::LD)
.addReg(RISCV::X7)
.addReg(RISCV::X28)
.addImm(FunctionAddressOffset - 4)),
// ld t3, (StaticChainOffset - 4)(t3)
// Load the value of the static chain.
GetEncoding(MCInstBuilder(RISCV::LD)
.addReg(RISCV::X28)
.addReg(RISCV::X28)
.addImm(StaticChainOffset - 4)),
// jalr t2
// Software-guarded jump to the function.
GetEncoding(MCInstBuilder(RISCV::JALR)
.addReg(RISCV::X0)
.addReg(RISCV::X7)
.addImm(0))});
}
// Store encoded instructions.
for (auto [Idx, Encoding] : llvm::enumerate(Encodings)) {
SDValue Addr = Idx > 0 ? DAG.getNode(ISD::ADD, dl, MVT::i64, Trmp,
DAG.getConstant(Idx * 4, dl, MVT::i64))
: Trmp;
OutChains.push_back(DAG.getTruncStore(
Root, dl, DAG.getConstant(Encoding, dl, MVT::i64), Addr,
MachinePointerInfo(TrmpAddr, Idx * 4), MVT::i32));
}
// Now store the variable part of the trampoline.
SDValue FunctionAddress = Op.getOperand(2);
SDValue StaticChain = Op.getOperand(3);
// Store the given static chain and function pointer in the trampoline buffer.
struct OffsetValuePair {
const unsigned Offset;
const SDValue Value;
SDValue Addr = SDValue(); // Used to cache the address.
} OffsetValues[] = {
{StaticChainOffset, StaticChain},
{FunctionAddressOffset, FunctionAddress},
};
for (auto &OffsetValue : OffsetValues) {
SDValue Addr =
DAG.getNode(ISD::ADD, dl, MVT::i64, Trmp,
DAG.getConstant(OffsetValue.Offset, dl, MVT::i64));
OffsetValue.Addr = Addr;
OutChains.push_back(
DAG.getStore(Root, dl, OffsetValue.Value, Addr,
MachinePointerInfo(TrmpAddr, OffsetValue.Offset)));
}
assert(OutChains.size() == StaticChainIdx + 2 &&
"Size of OutChains mismatch");
SDValue StoreToken = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains);
// The end of instructions of trampoline is the same as the static chain
// address that we computed earlier.
SDValue EndOfTrmp = OffsetValues[0].Addr;
// Call clear cache on the trampoline instructions.
SDValue Chain = DAG.getNode(ISD::CLEAR_CACHE, dl, MVT::Other, StoreToken,
Trmp, EndOfTrmp);
return Chain;
}
SDValue RISCVTargetLowering::lowerADJUST_TRAMPOLINE(SDValue Op,
SelectionDAG &DAG) const {
if (!Subtarget.is64Bit())
llvm::reportFatalUsageError("Trampolines only implemented for RV64");
return Op.getOperand(0);
}
SDValue RISCVTargetLowering::lowerPARTIAL_REDUCE_MLA(SDValue Op,
SelectionDAG &DAG) const {
// Currently, only the vqdot and vqdotu case (from zvqdotq) should be legal.
// TODO: There are many other sub-cases we could potentially lower, are
// any of them worthwhile? Ex: via vredsum, vwredsum, vwwmaccu, etc..
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
SDValue Accum = Op.getOperand(0);
assert(Accum.getSimpleValueType() == VT &&
VT.getVectorElementType() == MVT::i32);
SDValue A = Op.getOperand(1);
SDValue B = Op.getOperand(2);
MVT ArgVT = A.getSimpleValueType();
assert(ArgVT == B.getSimpleValueType() &&
ArgVT.getVectorElementType() == MVT::i8);
(void)ArgVT;
// The zvqdotq pseudos are defined with sources and destination both
// being i32. This cast is needed for correctness to avoid incorrect
// .vx matching of i8 splats.
A = DAG.getBitcast(VT, A);
B = DAG.getBitcast(VT, B);
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
Accum = convertToScalableVector(ContainerVT, Accum, DAG, Subtarget);
A = convertToScalableVector(ContainerVT, A, DAG, Subtarget);
B = convertToScalableVector(ContainerVT, B, DAG, Subtarget);
}
unsigned Opc;
switch (Op.getOpcode()) {
case ISD::PARTIAL_REDUCE_SMLA:
Opc = RISCVISD::VQDOT_VL;
break;
case ISD::PARTIAL_REDUCE_UMLA:
Opc = RISCVISD::VQDOTU_VL;
break;
case ISD::PARTIAL_REDUCE_SUMLA:
Opc = RISCVISD::VQDOTSU_VL;
break;
default:
llvm_unreachable("Unexpected opcode");
}
auto [Mask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
SDValue Res = DAG.getNode(Opc, DL, ContainerVT, {A, B, Accum, Mask, VL});
if (VT.isFixedLengthVector())
Res = convertFromScalableVector(VT, Res, DAG, Subtarget);
return Res;
}
static SDValue getTargetNode(GlobalAddressSDNode *N, const SDLoc &DL, EVT Ty,
SelectionDAG &DAG, unsigned Flags) {
return DAG.getTargetGlobalAddress(N->getGlobal(), DL, Ty, 0, Flags);
}
static SDValue getTargetNode(BlockAddressSDNode *N, const SDLoc &DL, EVT Ty,
SelectionDAG &DAG, unsigned Flags) {
return DAG.getTargetBlockAddress(N->getBlockAddress(), Ty, N->getOffset(),
Flags);
}
static SDValue getTargetNode(ConstantPoolSDNode *N, const SDLoc &DL, EVT Ty,
SelectionDAG &DAG, unsigned Flags) {
return DAG.getTargetConstantPool(N->getConstVal(), Ty, N->getAlign(),
N->getOffset(), Flags);
}
static SDValue getTargetNode(JumpTableSDNode *N, const SDLoc &DL, EVT Ty,
SelectionDAG &DAG, unsigned Flags) {
return DAG.getTargetJumpTable(N->getIndex(), Ty, Flags);
}
static SDValue getLargeGlobalAddress(GlobalAddressSDNode *N, const SDLoc &DL,
EVT Ty, SelectionDAG &DAG) {
RISCVConstantPoolValue *CPV = RISCVConstantPoolValue::Create(N->getGlobal());
SDValue CPAddr = DAG.getTargetConstantPool(CPV, Ty, Align(8));
SDValue LC = DAG.getNode(RISCVISD::LLA, DL, Ty, CPAddr);
return DAG.getLoad(
Ty, DL, DAG.getEntryNode(), LC,
MachinePointerInfo::getConstantPool(DAG.getMachineFunction()));
}
static SDValue getLargeExternalSymbol(ExternalSymbolSDNode *N, const SDLoc &DL,
EVT Ty, SelectionDAG &DAG) {
RISCVConstantPoolValue *CPV =
RISCVConstantPoolValue::Create(*DAG.getContext(), N->getSymbol());
SDValue CPAddr = DAG.getTargetConstantPool(CPV, Ty, Align(8));
SDValue LC = DAG.getNode(RISCVISD::LLA, DL, Ty, CPAddr);
return DAG.getLoad(
Ty, DL, DAG.getEntryNode(), LC,
MachinePointerInfo::getConstantPool(DAG.getMachineFunction()));
}
template <class NodeTy>
SDValue RISCVTargetLowering::getAddr(NodeTy *N, SelectionDAG &DAG,
bool IsLocal, bool IsExternWeak) const {
SDLoc DL(N);
EVT Ty = getPointerTy(DAG.getDataLayout());
// When HWASAN is used and tagging of global variables is enabled
// they should be accessed via the GOT, since the tagged address of a global
// is incompatible with existing code models. This also applies to non-pic
// mode.
if (isPositionIndependent() || Subtarget.allowTaggedGlobals()) {
SDValue Addr = getTargetNode(N, DL, Ty, DAG, 0);
if (IsLocal && !Subtarget.allowTaggedGlobals())
// Use PC-relative addressing to access the symbol. This generates the
// pattern (PseudoLLA sym), which expands to (addi (auipc %pcrel_hi(sym))
// %pcrel_lo(auipc)).
return DAG.getNode(RISCVISD::LLA, DL, Ty, Addr);
// Use PC-relative addressing to access the GOT for this symbol, then load
// the address from the GOT. This generates the pattern (PseudoLGA sym),
// which expands to (ld (addi (auipc %got_pcrel_hi(sym)) %pcrel_lo(auipc))).
SDValue Load =
SDValue(DAG.getMachineNode(RISCV::PseudoLGA, DL, Ty, Addr), 0);
MachineFunction &MF = DAG.getMachineFunction();
MachineMemOperand *MemOp = MF.getMachineMemOperand(
MachinePointerInfo::getGOT(MF),
MachineMemOperand::MOLoad | MachineMemOperand::MODereferenceable |
MachineMemOperand::MOInvariant,
LLT(Ty.getSimpleVT()), Align(Ty.getFixedSizeInBits() / 8));
DAG.setNodeMemRefs(cast<MachineSDNode>(Load.getNode()), {MemOp});
return Load;
}
switch (getTargetMachine().getCodeModel()) {
default:
reportFatalUsageError("Unsupported code model for lowering");
case CodeModel::Small: {
// Generate a sequence for accessing addresses within the first 2 GiB of
// address space.
if (Subtarget.hasVendorXqcili()) {
// Use QC.E.LI to generate the address, as this is easier to relax than
// LUI/ADDI.
SDValue Addr = getTargetNode(N, DL, Ty, DAG, 0);
return DAG.getNode(RISCVISD::QC_E_LI, DL, Ty, Addr);
}
// This generates the pattern (addi (lui %hi(sym)) %lo(sym)).
SDValue AddrHi = getTargetNode(N, DL, Ty, DAG, RISCVII::MO_HI);
SDValue AddrLo = getTargetNode(N, DL, Ty, DAG, RISCVII::MO_LO);
SDValue MNHi = DAG.getNode(RISCVISD::HI, DL, Ty, AddrHi);
return DAG.getNode(RISCVISD::ADD_LO, DL, Ty, MNHi, AddrLo);
}
case CodeModel::Medium: {
SDValue Addr = getTargetNode(N, DL, Ty, DAG, 0);
if (IsExternWeak) {
// An extern weak symbol may be undefined, i.e. have value 0, which may
// not be within 2GiB of PC, so use GOT-indirect addressing to access the
// symbol. This generates the pattern (PseudoLGA sym), which expands to
// (ld (addi (auipc %got_pcrel_hi(sym)) %pcrel_lo(auipc))).
SDValue Load =
SDValue(DAG.getMachineNode(RISCV::PseudoLGA, DL, Ty, Addr), 0);
MachineFunction &MF = DAG.getMachineFunction();
MachineMemOperand *MemOp = MF.getMachineMemOperand(
MachinePointerInfo::getGOT(MF),
MachineMemOperand::MOLoad | MachineMemOperand::MODereferenceable |
MachineMemOperand::MOInvariant,
LLT(Ty.getSimpleVT()), Align(Ty.getFixedSizeInBits() / 8));
DAG.setNodeMemRefs(cast<MachineSDNode>(Load.getNode()), {MemOp});
return Load;
}
// Generate a sequence for accessing addresses within any 2GiB range within
// the address space. This generates the pattern (PseudoLLA sym), which
// expands to (addi (auipc %pcrel_hi(sym)) %pcrel_lo(auipc)).
return DAG.getNode(RISCVISD::LLA, DL, Ty, Addr);
}
case CodeModel::Large: {
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(N))
return getLargeGlobalAddress(G, DL, Ty, DAG);
// Using pc-relative mode for other node type.
SDValue Addr = getTargetNode(N, DL, Ty, DAG, 0);
return DAG.getNode(RISCVISD::LLA, DL, Ty, Addr);
}
}
}
SDValue RISCVTargetLowering::lowerGlobalAddress(SDValue Op,
SelectionDAG &DAG) const {
GlobalAddressSDNode *N = cast<GlobalAddressSDNode>(Op);
assert(N->getOffset() == 0 && "unexpected offset in global node");
const GlobalValue *GV = N->getGlobal();
return getAddr(N, DAG, GV->isDSOLocal(), GV->hasExternalWeakLinkage());
}
SDValue RISCVTargetLowering::lowerBlockAddress(SDValue Op,
SelectionDAG &DAG) const {
BlockAddressSDNode *N = cast<BlockAddressSDNode>(Op);
return getAddr(N, DAG);
}
SDValue RISCVTargetLowering::lowerConstantPool(SDValue Op,
SelectionDAG &DAG) const {
ConstantPoolSDNode *N = cast<ConstantPoolSDNode>(Op);
return getAddr(N, DAG);
}
SDValue RISCVTargetLowering::lowerJumpTable(SDValue Op,
SelectionDAG &DAG) const {
JumpTableSDNode *N = cast<JumpTableSDNode>(Op);
return getAddr(N, DAG);
}
SDValue RISCVTargetLowering::getStaticTLSAddr(GlobalAddressSDNode *N,
SelectionDAG &DAG,
bool UseGOT) const {
SDLoc DL(N);
EVT Ty = getPointerTy(DAG.getDataLayout());
const GlobalValue *GV = N->getGlobal();
MVT XLenVT = Subtarget.getXLenVT();
if (UseGOT) {
// Use PC-relative addressing to access the GOT for this TLS symbol, then
// load the address from the GOT and add the thread pointer. This generates
// the pattern (PseudoLA_TLS_IE sym), which expands to
// (ld (auipc %tls_ie_pcrel_hi(sym)) %pcrel_lo(auipc)).
SDValue Addr = DAG.getTargetGlobalAddress(GV, DL, Ty, 0, 0);
SDValue Load =
SDValue(DAG.getMachineNode(RISCV::PseudoLA_TLS_IE, DL, Ty, Addr), 0);
MachineFunction &MF = DAG.getMachineFunction();
MachineMemOperand *MemOp = MF.getMachineMemOperand(
MachinePointerInfo::getGOT(MF),
MachineMemOperand::MOLoad | MachineMemOperand::MODereferenceable |
MachineMemOperand::MOInvariant,
LLT(Ty.getSimpleVT()), Align(Ty.getFixedSizeInBits() / 8));
DAG.setNodeMemRefs(cast<MachineSDNode>(Load.getNode()), {MemOp});
// Add the thread pointer.
SDValue TPReg = DAG.getRegister(RISCV::X4, XLenVT);
return DAG.getNode(ISD::ADD, DL, Ty, Load, TPReg);
}
// Generate a sequence for accessing the address relative to the thread
// pointer, with the appropriate adjustment for the thread pointer offset.
// This generates the pattern
// (add (add_tprel (lui %tprel_hi(sym)) tp %tprel_add(sym)) %tprel_lo(sym))
SDValue AddrHi =
DAG.getTargetGlobalAddress(GV, DL, Ty, 0, RISCVII::MO_TPREL_HI);
SDValue AddrAdd =
DAG.getTargetGlobalAddress(GV, DL, Ty, 0, RISCVII::MO_TPREL_ADD);
SDValue AddrLo =
DAG.getTargetGlobalAddress(GV, DL, Ty, 0, RISCVII::MO_TPREL_LO);
SDValue MNHi = DAG.getNode(RISCVISD::HI, DL, Ty, AddrHi);
SDValue TPReg = DAG.getRegister(RISCV::X4, XLenVT);
SDValue MNAdd =
DAG.getNode(RISCVISD::ADD_TPREL, DL, Ty, MNHi, TPReg, AddrAdd);
return DAG.getNode(RISCVISD::ADD_LO, DL, Ty, MNAdd, AddrLo);
}
SDValue RISCVTargetLowering::getDynamicTLSAddr(GlobalAddressSDNode *N,
SelectionDAG &DAG) const {
SDLoc DL(N);
EVT Ty = getPointerTy(DAG.getDataLayout());
IntegerType *CallTy = Type::getIntNTy(*DAG.getContext(), Ty.getSizeInBits());
const GlobalValue *GV = N->getGlobal();
// Use a PC-relative addressing mode to access the global dynamic GOT address.
// This generates the pattern (PseudoLA_TLS_GD sym), which expands to
// (addi (auipc %tls_gd_pcrel_hi(sym)) %pcrel_lo(auipc)).
SDValue Addr = DAG.getTargetGlobalAddress(GV, DL, Ty, 0, 0);
SDValue Load =
SDValue(DAG.getMachineNode(RISCV::PseudoLA_TLS_GD, DL, Ty, Addr), 0);
// Prepare argument list to generate call.
ArgListTy Args;
Args.emplace_back(Load, CallTy);
// Setup call to __tls_get_addr.
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(DL)
.setChain(DAG.getEntryNode())
.setLibCallee(CallingConv::C, CallTy,
DAG.getExternalSymbol("__tls_get_addr", Ty),
std::move(Args));
return LowerCallTo(CLI).first;
}
SDValue RISCVTargetLowering::getTLSDescAddr(GlobalAddressSDNode *N,
SelectionDAG &DAG) const {
SDLoc DL(N);
EVT Ty = getPointerTy(DAG.getDataLayout());
const GlobalValue *GV = N->getGlobal();
// Use a PC-relative addressing mode to access the global dynamic GOT address.
// This generates the pattern (PseudoLA_TLSDESC sym), which expands to
//
// auipc tX, %tlsdesc_hi(symbol) // R_RISCV_TLSDESC_HI20(symbol)
// lw tY, tX, %tlsdesc_load_lo(label) // R_RISCV_TLSDESC_LOAD_LO12(label)
// addi a0, tX, %tlsdesc_add_lo(label) // R_RISCV_TLSDESC_ADD_LO12(label)
// jalr t0, tY // R_RISCV_TLSDESC_CALL(label)
SDValue Addr = DAG.getTargetGlobalAddress(GV, DL, Ty, 0, 0);
return SDValue(DAG.getMachineNode(RISCV::PseudoLA_TLSDESC, DL, Ty, Addr), 0);
}
SDValue RISCVTargetLowering::lowerGlobalTLSAddress(SDValue Op,
SelectionDAG &DAG) const {
GlobalAddressSDNode *N = cast<GlobalAddressSDNode>(Op);
assert(N->getOffset() == 0 && "unexpected offset in global node");
if (DAG.getTarget().useEmulatedTLS())
return LowerToTLSEmulatedModel(N, DAG);
TLSModel::Model Model = getTargetMachine().getTLSModel(N->getGlobal());
if (DAG.getMachineFunction().getFunction().getCallingConv() ==
CallingConv::GHC)
reportFatalUsageError("In GHC calling convention TLS is not supported");
SDValue Addr;
switch (Model) {
case TLSModel::LocalExec:
Addr = getStaticTLSAddr(N, DAG, /*UseGOT=*/false);
break;
case TLSModel::InitialExec:
Addr = getStaticTLSAddr(N, DAG, /*UseGOT=*/true);
break;
case TLSModel::LocalDynamic:
case TLSModel::GeneralDynamic:
Addr = DAG.getTarget().useTLSDESC() ? getTLSDescAddr(N, DAG)
: getDynamicTLSAddr(N, DAG);
break;
}
return Addr;
}
// Return true if Val is equal to (setcc LHS, RHS, CC).
// Return false if Val is the inverse of (setcc LHS, RHS, CC).
// Otherwise, return std::nullopt.
static std::optional<bool> matchSetCC(SDValue LHS, SDValue RHS,
ISD::CondCode CC, SDValue Val) {
assert(Val->getOpcode() == ISD::SETCC);
SDValue LHS2 = Val.getOperand(0);
SDValue RHS2 = Val.getOperand(1);
ISD::CondCode CC2 = cast<CondCodeSDNode>(Val.getOperand(2))->get();
if (LHS == LHS2 && RHS == RHS2) {
if (CC == CC2)
return true;
if (CC == ISD::getSetCCInverse(CC2, LHS2.getValueType()))
return false;
} else if (LHS == RHS2 && RHS == LHS2) {
CC2 = ISD::getSetCCSwappedOperands(CC2);
if (CC == CC2)
return true;
if (CC == ISD::getSetCCInverse(CC2, LHS2.getValueType()))
return false;
}
return std::nullopt;
}
static bool isSimm12Constant(SDValue V) {
return isa<ConstantSDNode>(V) && V->getAsAPIntVal().isSignedIntN(12);
}
static SDValue lowerSelectToBinOp(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDValue CondV = N->getOperand(0);
SDValue TrueV = N->getOperand(1);
SDValue FalseV = N->getOperand(2);
MVT VT = N->getSimpleValueType(0);
SDLoc DL(N);
if (!Subtarget.hasConditionalMoveFusion()) {
// (select c, -1, y) -> -c | y
if (isAllOnesConstant(TrueV)) {
SDValue Neg = DAG.getNegative(CondV, DL, VT);
return DAG.getNode(ISD::OR, DL, VT, Neg, DAG.getFreeze(FalseV));
}
// (select c, y, -1) -> (c-1) | y
if (isAllOnesConstant(FalseV)) {
SDValue Neg = DAG.getNode(ISD::ADD, DL, VT, CondV,
DAG.getAllOnesConstant(DL, VT));
return DAG.getNode(ISD::OR, DL, VT, Neg, DAG.getFreeze(TrueV));
}
const bool HasCZero = VT.isScalarInteger() && Subtarget.hasCZEROLike();
// (select c, 0, y) -> (c-1) & y
if (isNullConstant(TrueV) && (!HasCZero || isSimm12Constant(FalseV))) {
SDValue Neg =
DAG.getNode(ISD::ADD, DL, VT, CondV, DAG.getAllOnesConstant(DL, VT));
return DAG.getNode(ISD::AND, DL, VT, Neg, DAG.getFreeze(FalseV));
}
// (select c, y, 0) -> -c & y
if (isNullConstant(FalseV) && (!HasCZero || isSimm12Constant(TrueV))) {
SDValue Neg = DAG.getNegative(CondV, DL, VT);
return DAG.getNode(ISD::AND, DL, VT, Neg, DAG.getFreeze(TrueV));
}
}
// select c, ~x, x --> xor -c, x
if (isa<ConstantSDNode>(TrueV) && isa<ConstantSDNode>(FalseV)) {
const APInt &TrueVal = TrueV->getAsAPIntVal();
const APInt &FalseVal = FalseV->getAsAPIntVal();
if (~TrueVal == FalseVal) {
SDValue Neg = DAG.getNegative(CondV, DL, VT);
return DAG.getNode(ISD::XOR, DL, VT, Neg, FalseV);
}
}
// Try to fold (select (setcc lhs, rhs, cc), truev, falsev) into bitwise ops
// when both truev and falsev are also setcc.
if (CondV.getOpcode() == ISD::SETCC && TrueV.getOpcode() == ISD::SETCC &&
FalseV.getOpcode() == ISD::SETCC) {
SDValue LHS = CondV.getOperand(0);
SDValue RHS = CondV.getOperand(1);
ISD::CondCode CC = cast<CondCodeSDNode>(CondV.getOperand(2))->get();
// (select x, x, y) -> x | y
// (select !x, x, y) -> x & y
if (std::optional<bool> MatchResult = matchSetCC(LHS, RHS, CC, TrueV)) {
return DAG.getNode(*MatchResult ? ISD::OR : ISD::AND, DL, VT, TrueV,
DAG.getFreeze(FalseV));
}
// (select x, y, x) -> x & y
// (select !x, y, x) -> x | y
if (std::optional<bool> MatchResult = matchSetCC(LHS, RHS, CC, FalseV)) {
return DAG.getNode(*MatchResult ? ISD::AND : ISD::OR, DL, VT,
DAG.getFreeze(TrueV), FalseV);
}
}
return SDValue();
}
// Transform `binOp (select cond, x, c0), c1` where `c0` and `c1` are constants
// into `select cond, binOp(x, c1), binOp(c0, c1)` if profitable.
// For now we only consider transformation profitable if `binOp(c0, c1)` ends up
// being `0` or `-1`. In such cases we can replace `select` with `and`.
// TODO: Should we also do this if `binOp(c0, c1)` is cheaper to materialize
// than `c0`?
static SDValue
foldBinOpIntoSelectIfProfitable(SDNode *BO, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (Subtarget.hasShortForwardBranchOpt())
return SDValue();
unsigned SelOpNo = 0;
SDValue Sel = BO->getOperand(0);
if (Sel.getOpcode() != ISD::SELECT || !Sel.hasOneUse()) {
SelOpNo = 1;
Sel = BO->getOperand(1);
}
if (Sel.getOpcode() != ISD::SELECT || !Sel.hasOneUse())
return SDValue();
unsigned ConstSelOpNo = 1;
unsigned OtherSelOpNo = 2;
if (!isa<ConstantSDNode>(Sel->getOperand(ConstSelOpNo))) {
ConstSelOpNo = 2;
OtherSelOpNo = 1;
}
SDValue ConstSelOp = Sel->getOperand(ConstSelOpNo);
ConstantSDNode *ConstSelOpNode = dyn_cast<ConstantSDNode>(ConstSelOp);
if (!ConstSelOpNode || ConstSelOpNode->isOpaque())
return SDValue();
SDValue ConstBinOp = BO->getOperand(SelOpNo ^ 1);
ConstantSDNode *ConstBinOpNode = dyn_cast<ConstantSDNode>(ConstBinOp);
if (!ConstBinOpNode || ConstBinOpNode->isOpaque())
return SDValue();
SDLoc DL(Sel);
EVT VT = BO->getValueType(0);
SDValue NewConstOps[2] = {ConstSelOp, ConstBinOp};
if (SelOpNo == 1)
std::swap(NewConstOps[0], NewConstOps[1]);
SDValue NewConstOp =
DAG.FoldConstantArithmetic(BO->getOpcode(), DL, VT, NewConstOps);
if (!NewConstOp)
return SDValue();
const APInt &NewConstAPInt = NewConstOp->getAsAPIntVal();
if (!NewConstAPInt.isZero() && !NewConstAPInt.isAllOnes())
return SDValue();
SDValue OtherSelOp = Sel->getOperand(OtherSelOpNo);
SDValue NewNonConstOps[2] = {OtherSelOp, ConstBinOp};
if (SelOpNo == 1)
std::swap(NewNonConstOps[0], NewNonConstOps[1]);
SDValue NewNonConstOp = DAG.getNode(BO->getOpcode(), DL, VT, NewNonConstOps);
SDValue NewT = (ConstSelOpNo == 1) ? NewConstOp : NewNonConstOp;
SDValue NewF = (ConstSelOpNo == 1) ? NewNonConstOp : NewConstOp;
return DAG.getSelect(DL, VT, Sel.getOperand(0), NewT, NewF);
}
SDValue RISCVTargetLowering::lowerSELECT(SDValue Op, SelectionDAG &DAG) const {
SDValue CondV = Op.getOperand(0);
SDValue TrueV = Op.getOperand(1);
SDValue FalseV = Op.getOperand(2);
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
// Lower vector SELECTs to VSELECTs by splatting the condition.
if (VT.isVector()) {
MVT SplatCondVT = VT.changeVectorElementType(MVT::i1);
SDValue CondSplat = DAG.getSplat(SplatCondVT, DL, CondV);
return DAG.getNode(ISD::VSELECT, DL, VT, CondSplat, TrueV, FalseV);
}
// Try some other optimizations before falling back to generic lowering.
if (SDValue V = lowerSelectToBinOp(Op.getNode(), DAG, Subtarget))
return V;
// When Zicond or XVentanaCondOps is present, emit CZERO_EQZ and CZERO_NEZ
// nodes to implement the SELECT. Performing the lowering here allows for
// greater control over when CZERO_{EQZ/NEZ} are used vs another branchless
// sequence or RISCVISD::SELECT_CC node (branch-based select).
if (Subtarget.hasCZEROLike() && VT.isScalarInteger()) {
// (select c, t, 0) -> (czero_eqz t, c)
if (isNullConstant(FalseV))
return DAG.getNode(RISCVISD::CZERO_EQZ, DL, VT, TrueV, CondV);
// (select c, 0, f) -> (czero_nez f, c)
if (isNullConstant(TrueV))
return DAG.getNode(RISCVISD::CZERO_NEZ, DL, VT, FalseV, CondV);
// Check to see if a given operation is a 'NOT', if so return the negated
// operand
auto getNotOperand = [](const SDValue &Op) -> std::optional<const SDValue> {
using namespace llvm::SDPatternMatch;
SDValue Xor;
if (sd_match(Op, m_OneUse(m_Not(m_Value(Xor))))) {
return Xor;
}
return std::nullopt;
};
// (select c, (and f, x), f) -> (or (and f, x), (czero_nez f, c))
// (select c, (and f, ~x), f) -> (andn f, (czero_eqz x, c))
if (TrueV.getOpcode() == ISD::AND &&
(TrueV.getOperand(0) == FalseV || TrueV.getOperand(1) == FalseV)) {
auto NotOperand = (TrueV.getOperand(0) == FalseV)
? getNotOperand(TrueV.getOperand(1))
: getNotOperand(TrueV.getOperand(0));
if (NotOperand) {
SDValue CMOV =
DAG.getNode(RISCVISD::CZERO_EQZ, DL, VT, *NotOperand, CondV);
SDValue NOT = DAG.getNOT(DL, CMOV, VT);
return DAG.getNode(ISD::AND, DL, VT, FalseV, NOT);
}
return DAG.getNode(
ISD::OR, DL, VT, TrueV,
DAG.getNode(RISCVISD::CZERO_NEZ, DL, VT, FalseV, CondV));
}
// (select c, t, (and t, x)) -> (or (czero_eqz t, c), (and t, x))
// (select c, t, (and t, ~x)) -> (andn t, (czero_nez x, c))
if (FalseV.getOpcode() == ISD::AND &&
(FalseV.getOperand(0) == TrueV || FalseV.getOperand(1) == TrueV)) {
auto NotOperand = (FalseV.getOperand(0) == TrueV)
? getNotOperand(FalseV.getOperand(1))
: getNotOperand(FalseV.getOperand(0));
if (NotOperand) {
SDValue CMOV =
DAG.getNode(RISCVISD::CZERO_NEZ, DL, VT, *NotOperand, CondV);
SDValue NOT = DAG.getNOT(DL, CMOV, VT);
return DAG.getNode(ISD::AND, DL, VT, TrueV, NOT);
}
return DAG.getNode(
ISD::OR, DL, VT, FalseV,
DAG.getNode(RISCVISD::CZERO_EQZ, DL, VT, TrueV, CondV));
}
// (select c, c1, c2) -> (add (czero_nez c2 - c1, c), c1)
// (select c, c1, c2) -> (add (czero_eqz c1 - c2, c), c2)
if (isa<ConstantSDNode>(TrueV) && isa<ConstantSDNode>(FalseV)) {
const APInt &TrueVal = TrueV->getAsAPIntVal();
const APInt &FalseVal = FalseV->getAsAPIntVal();
// Prefer these over Zicond to avoid materializing an immediate:
// (select (x < 0), y, z) -> x >> (XLEN - 1) & (y - z) + z
// (select (x > -1), z, y) -> x >> (XLEN - 1) & (y - z) + z
if (CondV.getOpcode() == ISD::SETCC &&
CondV.getOperand(0).getValueType() == VT && CondV.hasOneUse()) {
ISD::CondCode CCVal = cast<CondCodeSDNode>(CondV.getOperand(2))->get();
if ((CCVal == ISD::SETLT && isNullConstant(CondV.getOperand(1))) ||
(CCVal == ISD::SETGT && isAllOnesConstant(CondV.getOperand(1)))) {
int64_t TrueImm = TrueVal.getSExtValue();
int64_t FalseImm = FalseVal.getSExtValue();
if (CCVal == ISD::SETGT)
std::swap(TrueImm, FalseImm);
if (isInt<12>(TrueImm) && isInt<12>(FalseImm) &&
isInt<12>(TrueImm - FalseImm)) {
SDValue SRA =
DAG.getNode(ISD::SRA, DL, VT, CondV.getOperand(0),
DAG.getConstant(Subtarget.getXLen() - 1, DL, VT));
SDValue AND =
DAG.getNode(ISD::AND, DL, VT, SRA,
DAG.getSignedConstant(TrueImm - FalseImm, DL, VT));
return DAG.getNode(ISD::ADD, DL, VT, AND,
DAG.getSignedConstant(FalseImm, DL, VT));
}
}
}
// Use SHL/ADDI (and possible XORI) to avoid having to materialize
// a constant in register
if ((TrueVal - FalseVal).isPowerOf2() && FalseVal.isSignedIntN(12)) {
SDValue Log2 = DAG.getConstant((TrueVal - FalseVal).logBase2(), DL, VT);
SDValue BitDiff = DAG.getNode(ISD::SHL, DL, VT, CondV, Log2);
return DAG.getNode(ISD::ADD, DL, VT, FalseV, BitDiff);
}
if ((FalseVal - TrueVal).isPowerOf2() && TrueVal.isSignedIntN(12)) {
SDValue Log2 = DAG.getConstant((FalseVal - TrueVal).logBase2(), DL, VT);
CondV = DAG.getLogicalNOT(DL, CondV, CondV->getValueType(0));
SDValue BitDiff = DAG.getNode(ISD::SHL, DL, VT, CondV, Log2);
return DAG.getNode(ISD::ADD, DL, VT, TrueV, BitDiff);
}
auto getCost = [&](const APInt &Delta, const APInt &Addend) {
const int DeltaCost = RISCVMatInt::getIntMatCost(
Delta, Subtarget.getXLen(), Subtarget, /*CompressionCost=*/true);
// Does the addend fold into an ADDI
if (Addend.isSignedIntN(12))
return DeltaCost;
const int AddendCost = RISCVMatInt::getIntMatCost(
Addend, Subtarget.getXLen(), Subtarget, /*CompressionCost=*/true);
return AddendCost + DeltaCost;
};
bool IsCZERO_NEZ = getCost(FalseVal - TrueVal, TrueVal) <=
getCost(TrueVal - FalseVal, FalseVal);
SDValue LHSVal = DAG.getConstant(
IsCZERO_NEZ ? FalseVal - TrueVal : TrueVal - FalseVal, DL, VT);
SDValue CMOV =
DAG.getNode(IsCZERO_NEZ ? RISCVISD::CZERO_NEZ : RISCVISD::CZERO_EQZ,
DL, VT, LHSVal, CondV);
return DAG.getNode(ISD::ADD, DL, VT, CMOV, IsCZERO_NEZ ? TrueV : FalseV);
}
// (select c, c1, t) -> (add (czero_nez t - c1, c), c1)
// (select c, t, c1) -> (add (czero_eqz t - c1, c), c1)
if (isa<ConstantSDNode>(TrueV) != isa<ConstantSDNode>(FalseV)) {
bool IsCZERO_NEZ = isa<ConstantSDNode>(TrueV);
SDValue ConstVal = IsCZERO_NEZ ? TrueV : FalseV;
SDValue RegV = IsCZERO_NEZ ? FalseV : TrueV;
int64_t RawConstVal = cast<ConstantSDNode>(ConstVal)->getSExtValue();
// Fall back to XORI if Const == -0x800
if (RawConstVal == -0x800) {
SDValue XorOp = DAG.getNode(ISD::XOR, DL, VT, RegV, ConstVal);
SDValue CMOV =
DAG.getNode(IsCZERO_NEZ ? RISCVISD::CZERO_NEZ : RISCVISD::CZERO_EQZ,
DL, VT, XorOp, CondV);
return DAG.getNode(ISD::XOR, DL, VT, CMOV, ConstVal);
}
// Efficient only if the constant and its negation fit into `ADDI`
// Prefer Add/Sub over Xor since can be compressed for small immediates
if (isInt<12>(RawConstVal)) {
SDValue SubOp = DAG.getNode(ISD::SUB, DL, VT, RegV, ConstVal);
SDValue CMOV =
DAG.getNode(IsCZERO_NEZ ? RISCVISD::CZERO_NEZ : RISCVISD::CZERO_EQZ,
DL, VT, SubOp, CondV);
return DAG.getNode(ISD::ADD, DL, VT, CMOV, ConstVal);
}
}
// (select c, t, f) -> (or (czero_eqz t, c), (czero_nez f, c))
// Unless we have the short forward branch optimization.
if (!Subtarget.hasConditionalMoveFusion())
return DAG.getNode(
ISD::OR, DL, VT,
DAG.getNode(RISCVISD::CZERO_EQZ, DL, VT, TrueV, CondV),
DAG.getNode(RISCVISD::CZERO_NEZ, DL, VT, FalseV, CondV),
SDNodeFlags::Disjoint);
}
if (Op.hasOneUse()) {
unsigned UseOpc = Op->user_begin()->getOpcode();
if (isBinOp(UseOpc) && DAG.isSafeToSpeculativelyExecute(UseOpc)) {
SDNode *BinOp = *Op->user_begin();
if (SDValue NewSel = foldBinOpIntoSelectIfProfitable(*Op->user_begin(),
DAG, Subtarget)) {
DAG.ReplaceAllUsesWith(BinOp, &NewSel);
// Opcode check is necessary because foldBinOpIntoSelectIfProfitable
// may return a constant node and cause crash in lowerSELECT.
if (NewSel.getOpcode() == ISD::SELECT)
return lowerSELECT(NewSel, DAG);
return NewSel;
}
}
}
// (select cc, 1.0, 0.0) -> (sint_to_fp (zext cc))
// (select cc, 0.0, 1.0) -> (sint_to_fp (zext (xor cc, 1)))
const ConstantFPSDNode *FPTV = dyn_cast<ConstantFPSDNode>(TrueV);
const ConstantFPSDNode *FPFV = dyn_cast<ConstantFPSDNode>(FalseV);
if (FPTV && FPFV) {
if (FPTV->isExactlyValue(1.0) && FPFV->isExactlyValue(0.0))
return DAG.getNode(ISD::SINT_TO_FP, DL, VT, CondV);
if (FPTV->isExactlyValue(0.0) && FPFV->isExactlyValue(1.0)) {
SDValue XOR = DAG.getNode(ISD::XOR, DL, XLenVT, CondV,
DAG.getConstant(1, DL, XLenVT));
return DAG.getNode(ISD::SINT_TO_FP, DL, VT, XOR);
}
}
// If the condition is not an integer SETCC which operates on XLenVT, we need
// to emit a RISCVISD::SELECT_CC comparing the condition to zero. i.e.:
// (select condv, truev, falsev)
// -> (riscvisd::select_cc condv, zero, setne, truev, falsev)
if (CondV.getOpcode() != ISD::SETCC ||
CondV.getOperand(0).getSimpleValueType() != XLenVT) {
SDValue Zero = DAG.getConstant(0, DL, XLenVT);
SDValue SetNE = DAG.getCondCode(ISD::SETNE);
SDValue Ops[] = {CondV, Zero, SetNE, TrueV, FalseV};
return DAG.getNode(RISCVISD::SELECT_CC, DL, VT, Ops);
}
// If the CondV is the output of a SETCC node which operates on XLenVT inputs,
// then merge the SETCC node into the lowered RISCVISD::SELECT_CC to take
// advantage of the integer compare+branch instructions. i.e.:
// (select (setcc lhs, rhs, cc), truev, falsev)
// -> (riscvisd::select_cc lhs, rhs, cc, truev, falsev)
SDValue LHS = CondV.getOperand(0);
SDValue RHS = CondV.getOperand(1);
ISD::CondCode CCVal = cast<CondCodeSDNode>(CondV.getOperand(2))->get();
// Special case for a select of 2 constants that have a difference of 1.
// Normally this is done by DAGCombine, but if the select is introduced by
// type legalization or op legalization, we miss it. Restricting to SETLT
// case for now because that is what signed saturating add/sub need.
// FIXME: We don't need the condition to be SETLT or even a SETCC,
// but we would probably want to swap the true/false values if the condition
// is SETGE/SETLE to avoid an XORI.
if (isa<ConstantSDNode>(TrueV) && isa<ConstantSDNode>(FalseV) &&
CCVal == ISD::SETLT) {
const APInt &TrueVal = TrueV->getAsAPIntVal();
const APInt &FalseVal = FalseV->getAsAPIntVal();
if (TrueVal - 1 == FalseVal)
return DAG.getNode(ISD::ADD, DL, VT, CondV, FalseV);
if (TrueVal + 1 == FalseVal)
return DAG.getNode(ISD::SUB, DL, VT, FalseV, CondV);
}
translateSetCCForBranch(DL, LHS, RHS, CCVal, DAG, Subtarget);
// 1 < x ? x : 1 -> 0 < x ? x : 1
if (isOneConstant(LHS) && (CCVal == ISD::SETLT || CCVal == ISD::SETULT) &&
RHS == TrueV && LHS == FalseV) {
LHS = DAG.getConstant(0, DL, VT);
// 0 <u x is the same as x != 0.
if (CCVal == ISD::SETULT) {
std::swap(LHS, RHS);
CCVal = ISD::SETNE;
}
}
// x <s -1 ? x : -1 -> x <s 0 ? x : -1
if (isAllOnesConstant(RHS) && CCVal == ISD::SETLT && LHS == TrueV &&
RHS == FalseV) {
RHS = DAG.getConstant(0, DL, VT);
}
SDValue TargetCC = DAG.getCondCode(CCVal);
if (isa<ConstantSDNode>(TrueV) && !isa<ConstantSDNode>(FalseV)) {
// (select (setcc lhs, rhs, CC), constant, falsev)
// -> (select (setcc lhs, rhs, InverseCC), falsev, constant)
std::swap(TrueV, FalseV);
TargetCC = DAG.getCondCode(ISD::getSetCCInverse(CCVal, LHS.getValueType()));
}
SDValue Ops[] = {LHS, RHS, TargetCC, TrueV, FalseV};
return DAG.getNode(RISCVISD::SELECT_CC, DL, VT, Ops);
}
SDValue RISCVTargetLowering::lowerBRCOND(SDValue Op, SelectionDAG &DAG) const {
SDValue CondV = Op.getOperand(1);
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
if (CondV.getOpcode() == ISD::SETCC &&
CondV.getOperand(0).getValueType() == XLenVT) {
SDValue LHS = CondV.getOperand(0);
SDValue RHS = CondV.getOperand(1);
ISD::CondCode CCVal = cast<CondCodeSDNode>(CondV.getOperand(2))->get();
translateSetCCForBranch(DL, LHS, RHS, CCVal, DAG, Subtarget);
SDValue TargetCC = DAG.getCondCode(CCVal);
return DAG.getNode(RISCVISD::BR_CC, DL, Op.getValueType(), Op.getOperand(0),
LHS, RHS, TargetCC, Op.getOperand(2));
}
return DAG.getNode(RISCVISD::BR_CC, DL, Op.getValueType(), Op.getOperand(0),
CondV, DAG.getConstant(0, DL, XLenVT),
DAG.getCondCode(ISD::SETNE), Op.getOperand(2));
}
SDValue RISCVTargetLowering::lowerVASTART(SDValue Op, SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
RISCVMachineFunctionInfo *FuncInfo = MF.getInfo<RISCVMachineFunctionInfo>();
SDLoc DL(Op);
SDValue FI = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(),
getPointerTy(MF.getDataLayout()));
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
return DAG.getStore(Op.getOperand(0), DL, FI, Op.getOperand(1),
MachinePointerInfo(SV));
}
SDValue RISCVTargetLowering::lowerFRAMEADDR(SDValue Op,
SelectionDAG &DAG) const {
const RISCVRegisterInfo &RI = *Subtarget.getRegisterInfo();
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
MFI.setFrameAddressIsTaken(true);
Register FrameReg = RI.getFrameRegister(MF);
int XLenInBytes = Subtarget.getXLen() / 8;
EVT VT = Op.getValueType();
SDLoc DL(Op);
SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), DL, FrameReg, VT);
unsigned Depth = Op.getConstantOperandVal(0);
while (Depth--) {
int Offset = -(XLenInBytes * 2);
SDValue Ptr = DAG.getNode(
ISD::ADD, DL, VT, FrameAddr,
DAG.getSignedConstant(Offset, DL, getPointerTy(DAG.getDataLayout())));
FrameAddr =
DAG.getLoad(VT, DL, DAG.getEntryNode(), Ptr, MachinePointerInfo());
}
return FrameAddr;
}
SDValue RISCVTargetLowering::lowerRETURNADDR(SDValue Op,
SelectionDAG &DAG) const {
const RISCVRegisterInfo &RI = *Subtarget.getRegisterInfo();
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
MFI.setReturnAddressIsTaken(true);
MVT XLenVT = Subtarget.getXLenVT();
int XLenInBytes = Subtarget.getXLen() / 8;
EVT VT = Op.getValueType();
SDLoc DL(Op);
unsigned Depth = Op.getConstantOperandVal(0);
if (Depth) {
int Off = -XLenInBytes;
SDValue FrameAddr = lowerFRAMEADDR(Op, DAG);
SDValue Offset = DAG.getSignedConstant(Off, DL, VT);
return DAG.getLoad(VT, DL, DAG.getEntryNode(),
DAG.getNode(ISD::ADD, DL, VT, FrameAddr, Offset),
MachinePointerInfo());
}
// Return the value of the return address register, marking it an implicit
// live-in.
Register Reg = MF.addLiveIn(RI.getRARegister(), getRegClassFor(XLenVT));
return DAG.getCopyFromReg(DAG.getEntryNode(), DL, Reg, XLenVT);
}
SDValue RISCVTargetLowering::lowerShiftLeftParts(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
SDValue Lo = Op.getOperand(0);
SDValue Hi = Op.getOperand(1);
SDValue Shamt = Op.getOperand(2);
EVT VT = Lo.getValueType();
// if Shamt-XLEN < 0: // Shamt < XLEN
// Lo = Lo << Shamt
// Hi = (Hi << Shamt) | ((Lo >>u 1) >>u (XLEN-1 - Shamt))
// else:
// Lo = 0
// Hi = Lo << (Shamt-XLEN)
SDValue Zero = DAG.getConstant(0, DL, VT);
SDValue One = DAG.getConstant(1, DL, VT);
SDValue MinusXLen = DAG.getSignedConstant(-(int)Subtarget.getXLen(), DL, VT);
SDValue XLenMinus1 = DAG.getConstant(Subtarget.getXLen() - 1, DL, VT);
SDValue ShamtMinusXLen = DAG.getNode(ISD::ADD, DL, VT, Shamt, MinusXLen);
SDValue XLenMinus1Shamt = DAG.getNode(ISD::SUB, DL, VT, XLenMinus1, Shamt);
SDValue LoTrue = DAG.getNode(ISD::SHL, DL, VT, Lo, Shamt);
SDValue ShiftRight1Lo = DAG.getNode(ISD::SRL, DL, VT, Lo, One);
SDValue ShiftRightLo =
DAG.getNode(ISD::SRL, DL, VT, ShiftRight1Lo, XLenMinus1Shamt);
SDValue ShiftLeftHi = DAG.getNode(ISD::SHL, DL, VT, Hi, Shamt);
SDValue HiTrue = DAG.getNode(ISD::OR, DL, VT, ShiftLeftHi, ShiftRightLo);
SDValue HiFalse = DAG.getNode(ISD::SHL, DL, VT, Lo, ShamtMinusXLen);
SDValue CC = DAG.getSetCC(DL, VT, ShamtMinusXLen, Zero, ISD::SETLT);
Lo = DAG.getNode(ISD::SELECT, DL, VT, CC, LoTrue, Zero);
Hi = DAG.getNode(ISD::SELECT, DL, VT, CC, HiTrue, HiFalse);
SDValue Parts[2] = {Lo, Hi};
return DAG.getMergeValues(Parts, DL);
}
SDValue RISCVTargetLowering::lowerShiftRightParts(SDValue Op, SelectionDAG &DAG,
bool IsSRA) const {
SDLoc DL(Op);
SDValue Lo = Op.getOperand(0);
SDValue Hi = Op.getOperand(1);
SDValue Shamt = Op.getOperand(2);
EVT VT = Lo.getValueType();
// SRA expansion:
// if Shamt-XLEN < 0: // Shamt < XLEN
// Lo = (Lo >>u Shamt) | ((Hi << 1) << (XLEN-1 - ShAmt))
// Hi = Hi >>s Shamt
// else:
// Lo = Hi >>s (Shamt-XLEN);
// Hi = Hi >>s (XLEN-1)
//
// SRL expansion:
// if Shamt-XLEN < 0: // Shamt < XLEN
// Lo = (Lo >>u Shamt) | ((Hi << 1) << (XLEN-1 - ShAmt))
// Hi = Hi >>u Shamt
// else:
// Lo = Hi >>u (Shamt-XLEN);
// Hi = 0;
unsigned ShiftRightOp = IsSRA ? ISD::SRA : ISD::SRL;
SDValue Zero = DAG.getConstant(0, DL, VT);
SDValue One = DAG.getConstant(1, DL, VT);
SDValue MinusXLen = DAG.getSignedConstant(-(int)Subtarget.getXLen(), DL, VT);
SDValue XLenMinus1 = DAG.getConstant(Subtarget.getXLen() - 1, DL, VT);
SDValue ShamtMinusXLen = DAG.getNode(ISD::ADD, DL, VT, Shamt, MinusXLen);
SDValue XLenMinus1Shamt = DAG.getNode(ISD::SUB, DL, VT, XLenMinus1, Shamt);
SDValue ShiftRightLo = DAG.getNode(ISD::SRL, DL, VT, Lo, Shamt);
SDValue ShiftLeftHi1 = DAG.getNode(ISD::SHL, DL, VT, Hi, One);
SDValue ShiftLeftHi =
DAG.getNode(ISD::SHL, DL, VT, ShiftLeftHi1, XLenMinus1Shamt);
SDValue LoTrue = DAG.getNode(ISD::OR, DL, VT, ShiftRightLo, ShiftLeftHi);
SDValue HiTrue = DAG.getNode(ShiftRightOp, DL, VT, Hi, Shamt);
SDValue LoFalse = DAG.getNode(ShiftRightOp, DL, VT, Hi, ShamtMinusXLen);
SDValue HiFalse =
IsSRA ? DAG.getNode(ISD::SRA, DL, VT, Hi, XLenMinus1) : Zero;
SDValue CC = DAG.getSetCC(DL, VT, ShamtMinusXLen, Zero, ISD::SETLT);
Lo = DAG.getNode(ISD::SELECT, DL, VT, CC, LoTrue, LoFalse);
Hi = DAG.getNode(ISD::SELECT, DL, VT, CC, HiTrue, HiFalse);
SDValue Parts[2] = {Lo, Hi};
return DAG.getMergeValues(Parts, DL);
}
// Lower splats of i1 types to SETCC. For each mask vector type, we have a
// legal equivalently-sized i8 type, so we can use that as a go-between.
SDValue RISCVTargetLowering::lowerVectorMaskSplat(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
SDValue SplatVal = Op.getOperand(0);
// All-zeros or all-ones splats are handled specially.
if (ISD::isConstantSplatVectorAllOnes(Op.getNode())) {
SDValue VL = getDefaultScalableVLOps(VT, DL, DAG, Subtarget).second;
return DAG.getNode(RISCVISD::VMSET_VL, DL, VT, VL);
}
if (ISD::isConstantSplatVectorAllZeros(Op.getNode())) {
SDValue VL = getDefaultScalableVLOps(VT, DL, DAG, Subtarget).second;
return DAG.getNode(RISCVISD::VMCLR_VL, DL, VT, VL);
}
MVT InterVT = VT.changeVectorElementType(MVT::i8);
SplatVal = DAG.getNode(ISD::AND, DL, SplatVal.getValueType(), SplatVal,
DAG.getConstant(1, DL, SplatVal.getValueType()));
SDValue LHS = DAG.getSplatVector(InterVT, DL, SplatVal);
SDValue Zero = DAG.getConstant(0, DL, InterVT);
return DAG.getSetCC(DL, VT, LHS, Zero, ISD::SETNE);
}
// Custom-lower a SPLAT_VECTOR_PARTS where XLEN<SEW, as the SEW element type is
// illegal (currently only vXi64 RV32).
// FIXME: We could also catch non-constant sign-extended i32 values and lower
// them to VMV_V_X_VL.
SDValue RISCVTargetLowering::lowerSPLAT_VECTOR_PARTS(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VecVT = Op.getSimpleValueType();
assert(!Subtarget.is64Bit() && VecVT.getVectorElementType() == MVT::i64 &&
"Unexpected SPLAT_VECTOR_PARTS lowering");
assert(Op.getNumOperands() == 2 && "Unexpected number of operands!");
SDValue Lo = Op.getOperand(0);
SDValue Hi = Op.getOperand(1);
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector())
ContainerVT = getContainerForFixedLengthVector(VecVT);
auto VL = getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget).second;
SDValue Res =
splatPartsI64WithVL(DL, ContainerVT, SDValue(), Lo, Hi, VL, DAG);
if (VecVT.isFixedLengthVector())
Res = convertFromScalableVector(VecVT, Res, DAG, Subtarget);
return Res;
}
// Custom-lower extensions from mask vectors by using a vselect either with 1
// for zero/any-extension or -1 for sign-extension:
// (vXiN = (s|z)ext vXi1:vmask) -> (vXiN = vselect vmask, (-1 or 1), 0)
// Note that any-extension is lowered identically to zero-extension.
SDValue RISCVTargetLowering::lowerVectorMaskExt(SDValue Op, SelectionDAG &DAG,
int64_t ExtTrueVal) const {
SDLoc DL(Op);
MVT VecVT = Op.getSimpleValueType();
SDValue Src = Op.getOperand(0);
// Only custom-lower extensions from mask types
assert(Src.getValueType().isVector() &&
Src.getValueType().getVectorElementType() == MVT::i1);
if (VecVT.isScalableVector()) {
SDValue SplatZero = DAG.getConstant(0, DL, VecVT);
SDValue SplatTrueVal = DAG.getSignedConstant(ExtTrueVal, DL, VecVT);
if (Src.getOpcode() == ISD::XOR &&
ISD::isConstantSplatVectorAllOnes(Src.getOperand(1).getNode()))
return DAG.getNode(ISD::VSELECT, DL, VecVT, Src.getOperand(0), SplatZero,
SplatTrueVal);
return DAG.getNode(ISD::VSELECT, DL, VecVT, Src, SplatTrueVal, SplatZero);
}
MVT ContainerVT = getContainerForFixedLengthVector(VecVT);
MVT I1ContainerVT =
MVT::getVectorVT(MVT::i1, ContainerVT.getVectorElementCount());
SDValue CC = convertToScalableVector(I1ContainerVT, Src, DAG, Subtarget);
SDValue VL = getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget).second;
MVT XLenVT = Subtarget.getXLenVT();
SDValue SplatZero = DAG.getConstant(0, DL, XLenVT);
SDValue SplatTrueVal = DAG.getSignedConstant(ExtTrueVal, DL, XLenVT);
if (Src.getOpcode() == ISD::EXTRACT_SUBVECTOR) {
SDValue Xor = Src.getOperand(0);
if (Xor.getOpcode() == RISCVISD::VMXOR_VL) {
SDValue ScalableOnes = Xor.getOperand(1);
if (ScalableOnes.getOpcode() == ISD::INSERT_SUBVECTOR &&
ScalableOnes.getOperand(0).isUndef() &&
ISD::isConstantSplatVectorAllOnes(
ScalableOnes.getOperand(1).getNode())) {
CC = Xor.getOperand(0);
std::swap(SplatZero, SplatTrueVal);
}
}
}
SplatZero = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), SplatZero, VL);
SplatTrueVal = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), SplatTrueVal, VL);
SDValue Select =
DAG.getNode(RISCVISD::VMERGE_VL, DL, ContainerVT, CC, SplatTrueVal,
SplatZero, DAG.getUNDEF(ContainerVT), VL);
return convertFromScalableVector(VecVT, Select, DAG, Subtarget);
}
// Custom-lower truncations from vectors to mask vectors by using a mask and a
// setcc operation:
// (vXi1 = trunc vXiN vec) -> (vXi1 = setcc (and vec, 1), 0, ne)
SDValue RISCVTargetLowering::lowerVectorMaskTruncLike(SDValue Op,
SelectionDAG &DAG) const {
bool IsVPTrunc = Op.getOpcode() == ISD::VP_TRUNCATE;
SDLoc DL(Op);
EVT MaskVT = Op.getValueType();
// Only expect to custom-lower truncations to mask types
assert(MaskVT.isVector() && MaskVT.getVectorElementType() == MVT::i1 &&
"Unexpected type for vector mask lowering");
SDValue Src = Op.getOperand(0);
MVT VecVT = Src.getSimpleValueType();
SDValue Mask, VL;
if (IsVPTrunc) {
Mask = Op.getOperand(1);
VL = Op.getOperand(2);
}
// If this is a fixed vector, we need to convert it to a scalable vector.
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Src = convertToScalableVector(ContainerVT, Src, DAG, Subtarget);
if (IsVPTrunc) {
MVT MaskContainerVT =
getContainerForFixedLengthVector(Mask.getSimpleValueType());
Mask = convertToScalableVector(MaskContainerVT, Mask, DAG, Subtarget);
}
}
if (!IsVPTrunc) {
std::tie(Mask, VL) =
getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget);
}
SDValue SplatOne = DAG.getConstant(1, DL, Subtarget.getXLenVT());
SDValue SplatZero = DAG.getConstant(0, DL, Subtarget.getXLenVT());
SplatOne = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), SplatOne, VL);
SplatZero = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), SplatZero, VL);
MVT MaskContainerVT = ContainerVT.changeVectorElementType(MVT::i1);
SDValue Trunc = DAG.getNode(RISCVISD::AND_VL, DL, ContainerVT, Src, SplatOne,
DAG.getUNDEF(ContainerVT), Mask, VL);
Trunc = DAG.getNode(RISCVISD::SETCC_VL, DL, MaskContainerVT,
{Trunc, SplatZero, DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(MaskContainerVT), Mask, VL});
if (MaskVT.isFixedLengthVector())
Trunc = convertFromScalableVector(MaskVT, Trunc, DAG, Subtarget);
return Trunc;
}
SDValue RISCVTargetLowering::lowerVectorTruncLike(SDValue Op,
SelectionDAG &DAG) const {
unsigned Opc = Op.getOpcode();
bool IsVPTrunc = Opc == ISD::VP_TRUNCATE;
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
// Only custom-lower vector truncates
assert(VT.isVector() && "Unexpected type for vector truncate lowering");
// Truncates to mask types are handled differently
if (VT.getVectorElementType() == MVT::i1)
return lowerVectorMaskTruncLike(Op, DAG);
// RVV only has truncates which operate from SEW*2->SEW, so lower arbitrary
// truncates as a series of "RISCVISD::TRUNCATE_VECTOR_VL" nodes which
// truncate by one power of two at a time.
MVT DstEltVT = VT.getVectorElementType();
SDValue Src = Op.getOperand(0);
MVT SrcVT = Src.getSimpleValueType();
MVT SrcEltVT = SrcVT.getVectorElementType();
assert(DstEltVT.bitsLT(SrcEltVT) && isPowerOf2_64(DstEltVT.getSizeInBits()) &&
isPowerOf2_64(SrcEltVT.getSizeInBits()) &&
"Unexpected vector truncate lowering");
MVT ContainerVT = SrcVT;
SDValue Mask, VL;
if (IsVPTrunc) {
Mask = Op.getOperand(1);
VL = Op.getOperand(2);
}
if (SrcVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(SrcVT);
Src = convertToScalableVector(ContainerVT, Src, DAG, Subtarget);
if (IsVPTrunc) {
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
}
SDValue Result = Src;
if (!IsVPTrunc) {
std::tie(Mask, VL) =
getDefaultVLOps(SrcVT, ContainerVT, DL, DAG, Subtarget);
}
unsigned NewOpc;
if (Opc == ISD::TRUNCATE_SSAT_S)
NewOpc = RISCVISD::TRUNCATE_VECTOR_VL_SSAT;
else if (Opc == ISD::TRUNCATE_USAT_U)
NewOpc = RISCVISD::TRUNCATE_VECTOR_VL_USAT;
else
NewOpc = RISCVISD::TRUNCATE_VECTOR_VL;
do {
SrcEltVT = MVT::getIntegerVT(SrcEltVT.getSizeInBits() / 2);
MVT ResultVT = ContainerVT.changeVectorElementType(SrcEltVT);
Result = DAG.getNode(NewOpc, DL, ResultVT, Result, Mask, VL);
} while (SrcEltVT != DstEltVT);
if (SrcVT.isFixedLengthVector())
Result = convertFromScalableVector(VT, Result, DAG, Subtarget);
return Result;
}
SDValue
RISCVTargetLowering::lowerStrictFPExtendOrRoundLike(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
SDValue Chain = Op.getOperand(0);
SDValue Src = Op.getOperand(1);
MVT VT = Op.getSimpleValueType();
MVT SrcVT = Src.getSimpleValueType();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
MVT SrcContainerVT = getContainerForFixedLengthVector(SrcVT);
ContainerVT =
SrcContainerVT.changeVectorElementType(VT.getVectorElementType());
Src = convertToScalableVector(SrcContainerVT, Src, DAG, Subtarget);
}
auto [Mask, VL] = getDefaultVLOps(SrcVT, ContainerVT, DL, DAG, Subtarget);
// RVV can only widen/truncate fp to types double/half the size as the source.
if ((VT.getVectorElementType() == MVT::f64 &&
(SrcVT.getVectorElementType() == MVT::f16 ||
SrcVT.getVectorElementType() == MVT::bf16)) ||
((VT.getVectorElementType() == MVT::f16 ||
VT.getVectorElementType() == MVT::bf16) &&
SrcVT.getVectorElementType() == MVT::f64)) {
// For double rounding, the intermediate rounding should be round-to-odd.
unsigned InterConvOpc = Op.getOpcode() == ISD::STRICT_FP_EXTEND
? RISCVISD::STRICT_FP_EXTEND_VL
: RISCVISD::STRICT_VFNCVT_ROD_VL;
MVT InterVT = ContainerVT.changeVectorElementType(MVT::f32);
Src = DAG.getNode(InterConvOpc, DL, DAG.getVTList(InterVT, MVT::Other),
Chain, Src, Mask, VL);
Chain = Src.getValue(1);
}
unsigned ConvOpc = Op.getOpcode() == ISD::STRICT_FP_EXTEND
? RISCVISD::STRICT_FP_EXTEND_VL
: RISCVISD::STRICT_FP_ROUND_VL;
SDValue Res = DAG.getNode(ConvOpc, DL, DAG.getVTList(ContainerVT, MVT::Other),
Chain, Src, Mask, VL);
if (VT.isFixedLengthVector()) {
// StrictFP operations have two result values. Their lowered result should
// have same result count.
SDValue SubVec = convertFromScalableVector(VT, Res, DAG, Subtarget);
Res = DAG.getMergeValues({SubVec, Res.getValue(1)}, DL);
}
return Res;
}
SDValue
RISCVTargetLowering::lowerVectorFPExtendOrRoundLike(SDValue Op,
SelectionDAG &DAG) const {
bool IsVP =
Op.getOpcode() == ISD::VP_FP_ROUND || Op.getOpcode() == ISD::VP_FP_EXTEND;
bool IsExtend =
Op.getOpcode() == ISD::VP_FP_EXTEND || Op.getOpcode() == ISD::FP_EXTEND;
// RVV can only do truncate fp to types half the size as the source. We
// custom-lower f64->f16 rounds via RVV's round-to-odd float
// conversion instruction.
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
assert(VT.isVector() && "Unexpected type for vector truncate lowering");
SDValue Src = Op.getOperand(0);
MVT SrcVT = Src.getSimpleValueType();
bool IsDirectExtend =
IsExtend && (VT.getVectorElementType() != MVT::f64 ||
(SrcVT.getVectorElementType() != MVT::f16 &&
SrcVT.getVectorElementType() != MVT::bf16));
bool IsDirectTrunc = !IsExtend && ((VT.getVectorElementType() != MVT::f16 &&
VT.getVectorElementType() != MVT::bf16) ||
SrcVT.getVectorElementType() != MVT::f64);
bool IsDirectConv = IsDirectExtend || IsDirectTrunc;
// We have regular SD node patterns for direct non-VL extends.
if (VT.isScalableVector() && IsDirectConv && !IsVP)
return Op;
// Prepare any fixed-length vector operands.
MVT ContainerVT = VT;
SDValue Mask, VL;
if (IsVP) {
Mask = Op.getOperand(1);
VL = Op.getOperand(2);
}
if (VT.isFixedLengthVector()) {
MVT SrcContainerVT = getContainerForFixedLengthVector(SrcVT);
ContainerVT =
SrcContainerVT.changeVectorElementType(VT.getVectorElementType());
Src = convertToScalableVector(SrcContainerVT, Src, DAG, Subtarget);
if (IsVP) {
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
}
if (!IsVP)
std::tie(Mask, VL) =
getDefaultVLOps(SrcVT, ContainerVT, DL, DAG, Subtarget);
unsigned ConvOpc = IsExtend ? RISCVISD::FP_EXTEND_VL : RISCVISD::FP_ROUND_VL;
if (IsDirectConv) {
Src = DAG.getNode(ConvOpc, DL, ContainerVT, Src, Mask, VL);
if (VT.isFixedLengthVector())
Src = convertFromScalableVector(VT, Src, DAG, Subtarget);
return Src;
}
unsigned InterConvOpc =
IsExtend ? RISCVISD::FP_EXTEND_VL : RISCVISD::VFNCVT_ROD_VL;
MVT InterVT = ContainerVT.changeVectorElementType(MVT::f32);
SDValue IntermediateConv =
DAG.getNode(InterConvOpc, DL, InterVT, Src, Mask, VL);
SDValue Result =
DAG.getNode(ConvOpc, DL, ContainerVT, IntermediateConv, Mask, VL);
if (VT.isFixedLengthVector())
return convertFromScalableVector(VT, Result, DAG, Subtarget);
return Result;
}
// Given a scalable vector type and an index into it, returns the type for the
// smallest subvector that the index fits in. This can be used to reduce LMUL
// for operations like vslidedown.
//
// E.g. With Zvl128b, index 3 in a nxv4i32 fits within the first nxv2i32.
static std::optional<MVT>
getSmallestVTForIndex(MVT VecVT, unsigned MaxIdx, SDLoc DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(VecVT.isScalableVector());
const unsigned EltSize = VecVT.getScalarSizeInBits();
const unsigned VectorBitsMin = Subtarget.getRealMinVLen();
const unsigned MinVLMAX = VectorBitsMin / EltSize;
MVT SmallerVT;
if (MaxIdx < MinVLMAX)
SmallerVT = RISCVTargetLowering::getM1VT(VecVT);
else if (MaxIdx < MinVLMAX * 2)
SmallerVT =
RISCVTargetLowering::getM1VT(VecVT).getDoubleNumVectorElementsVT();
else if (MaxIdx < MinVLMAX * 4)
SmallerVT = RISCVTargetLowering::getM1VT(VecVT)
.getDoubleNumVectorElementsVT()
.getDoubleNumVectorElementsVT();
if (!SmallerVT.isValid() || !VecVT.bitsGT(SmallerVT))
return std::nullopt;
return SmallerVT;
}
static bool isValidVisniInsertExtractIndex(SDValue Idx) {
auto *IdxC = dyn_cast<ConstantSDNode>(Idx);
if (!IdxC || isNullConstant(Idx))
return false;
return isUInt<5>(IdxC->getZExtValue());
}
// Custom-legalize INSERT_VECTOR_ELT so that the value is inserted into the
// first position of a vector, and that vector is slid up to the insert index.
// By limiting the active vector length to index+1 and merging with the
// original vector (with an undisturbed tail policy for elements >= VL), we
// achieve the desired result of leaving all elements untouched except the one
// at VL-1, which is replaced with the desired value.
SDValue RISCVTargetLowering::lowerINSERT_VECTOR_ELT(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VecVT = Op.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
SDValue Vec = Op.getOperand(0);
SDValue Val = Op.getOperand(1);
MVT ValVT = Val.getSimpleValueType();
SDValue Idx = Op.getOperand(2);
if (VecVT.getVectorElementType() == MVT::i1) {
// FIXME: For now we just promote to an i8 vector and insert into that,
// but this is probably not optimal.
MVT WideVT = MVT::getVectorVT(MVT::i8, VecVT.getVectorElementCount());
Vec = DAG.getNode(ISD::ZERO_EXTEND, DL, WideVT, Vec);
Vec = DAG.getNode(ISD::INSERT_VECTOR_ELT, DL, WideVT, Vec, Val, Idx);
return DAG.getNode(ISD::TRUNCATE, DL, VecVT, Vec);
}
if ((ValVT == MVT::f16 && !Subtarget.hasVInstructionsF16()) ||
ValVT == MVT::bf16) {
// If we don't have vfmv.s.f for f16/bf16, use fmv.x.h first.
MVT IntVT = VecVT.changeTypeToInteger();
SDValue IntInsert = DAG.getNode(
ISD::INSERT_VECTOR_ELT, DL, IntVT, DAG.getBitcast(IntVT, Vec),
DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, XLenVT, Val), Idx);
return DAG.getBitcast(VecVT, IntInsert);
}
MVT ContainerVT = VecVT;
// If the operand is a fixed-length vector, convert to a scalable one.
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
// If we know the index we're going to insert at, we can shrink Vec so that
// we're performing the scalar inserts and slideup on a smaller LMUL.
SDValue OrigVec = Vec;
std::optional<unsigned> AlignedIdx;
if (auto *IdxC = dyn_cast<ConstantSDNode>(Idx)) {
const unsigned OrigIdx = IdxC->getZExtValue();
// Do we know an upper bound on LMUL?
if (auto ShrunkVT = getSmallestVTForIndex(ContainerVT, OrigIdx,
DL, DAG, Subtarget)) {
ContainerVT = *ShrunkVT;
AlignedIdx = 0;
}
// If we're compiling for an exact VLEN value, we can always perform
// the insert in m1 as we can determine the register corresponding to
// the index in the register group.
const MVT M1VT = RISCVTargetLowering::getM1VT(ContainerVT);
if (auto VLEN = Subtarget.getRealVLen(); VLEN && ContainerVT.bitsGT(M1VT)) {
EVT ElemVT = VecVT.getVectorElementType();
unsigned ElemsPerVReg = *VLEN / ElemVT.getFixedSizeInBits();
unsigned RemIdx = OrigIdx % ElemsPerVReg;
unsigned SubRegIdx = OrigIdx / ElemsPerVReg;
AlignedIdx = SubRegIdx * M1VT.getVectorElementCount().getKnownMinValue();
Idx = DAG.getVectorIdxConstant(RemIdx, DL);
ContainerVT = M1VT;
}
if (AlignedIdx)
Vec = DAG.getExtractSubvector(DL, ContainerVT, Vec, *AlignedIdx);
}
bool IsLegalInsert = Subtarget.is64Bit() || Val.getValueType() != MVT::i64;
// Even i64-element vectors on RV32 can be lowered without scalar
// legalization if the most-significant 32 bits of the value are not affected
// by the sign-extension of the lower 32 bits.
// TODO: We could also catch sign extensions of a 32-bit value.
if (!IsLegalInsert && isa<ConstantSDNode>(Val)) {
const auto *CVal = cast<ConstantSDNode>(Val);
if (isInt<32>(CVal->getSExtValue())) {
IsLegalInsert = true;
Val = DAG.getSignedConstant(CVal->getSExtValue(), DL, MVT::i32);
}
}
auto [Mask, VL] = getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget);
SDValue ValInVec;
if (IsLegalInsert) {
unsigned Opc =
VecVT.isFloatingPoint() ? RISCVISD::VFMV_S_F_VL : RISCVISD::VMV_S_X_VL;
if (isNullConstant(Idx)) {
if (!VecVT.isFloatingPoint())
Val = DAG.getNode(ISD::ANY_EXTEND, DL, XLenVT, Val);
Vec = DAG.getNode(Opc, DL, ContainerVT, Vec, Val, VL);
if (AlignedIdx)
Vec = DAG.getInsertSubvector(DL, OrigVec, Vec, *AlignedIdx);
if (!VecVT.isFixedLengthVector())
return Vec;
return convertFromScalableVector(VecVT, Vec, DAG, Subtarget);
}
// Use ri.vinsert.v.x if available.
if (Subtarget.hasVendorXRivosVisni() && VecVT.isInteger() &&
isValidVisniInsertExtractIndex(Idx)) {
// Tail policy applies to elements past VLMAX (by assumption Idx < VLMAX)
SDValue PolicyOp =
DAG.getTargetConstant(RISCVVType::TAIL_AGNOSTIC, DL, XLenVT);
Vec = DAG.getNode(RISCVISD::RI_VINSERT_VL, DL, ContainerVT, Vec, Val, Idx,
VL, PolicyOp);
if (AlignedIdx)
Vec = DAG.getInsertSubvector(DL, OrigVec, Vec, *AlignedIdx);
if (!VecVT.isFixedLengthVector())
return Vec;
return convertFromScalableVector(VecVT, Vec, DAG, Subtarget);
}
ValInVec = lowerScalarInsert(Val, VL, ContainerVT, DL, DAG, Subtarget);
} else {
// On RV32, i64-element vectors must be specially handled to place the
// value at element 0, by using two vslide1down instructions in sequence on
// the i32 split lo/hi value. Use an equivalently-sized i32 vector for
// this.
SDValue ValLo, ValHi;
std::tie(ValLo, ValHi) = DAG.SplitScalar(Val, DL, MVT::i32, MVT::i32);
MVT I32ContainerVT =
MVT::getVectorVT(MVT::i32, ContainerVT.getVectorElementCount() * 2);
SDValue I32Mask =
getDefaultScalableVLOps(I32ContainerVT, DL, DAG, Subtarget).first;
// Limit the active VL to two.
SDValue InsertI64VL = DAG.getConstant(2, DL, XLenVT);
// If the Idx is 0 we can insert directly into the vector.
if (isNullConstant(Idx)) {
// First slide in the lo value, then the hi in above it. We use slide1down
// to avoid the register group overlap constraint of vslide1up.
ValInVec = DAG.getNode(RISCVISD::VSLIDE1DOWN_VL, DL, I32ContainerVT,
Vec, Vec, ValLo, I32Mask, InsertI64VL);
// If the source vector is undef don't pass along the tail elements from
// the previous slide1down.
SDValue Tail = Vec.isUndef() ? Vec : ValInVec;
ValInVec = DAG.getNode(RISCVISD::VSLIDE1DOWN_VL, DL, I32ContainerVT,
Tail, ValInVec, ValHi, I32Mask, InsertI64VL);
// Bitcast back to the right container type.
ValInVec = DAG.getBitcast(ContainerVT, ValInVec);
if (AlignedIdx)
ValInVec = DAG.getInsertSubvector(DL, OrigVec, ValInVec, *AlignedIdx);
if (!VecVT.isFixedLengthVector())
return ValInVec;
return convertFromScalableVector(VecVT, ValInVec, DAG, Subtarget);
}
// First slide in the lo value, then the hi in above it. We use slide1down
// to avoid the register group overlap constraint of vslide1up.
ValInVec = DAG.getNode(RISCVISD::VSLIDE1DOWN_VL, DL, I32ContainerVT,
DAG.getUNDEF(I32ContainerVT),
DAG.getUNDEF(I32ContainerVT), ValLo,
I32Mask, InsertI64VL);
ValInVec = DAG.getNode(RISCVISD::VSLIDE1DOWN_VL, DL, I32ContainerVT,
DAG.getUNDEF(I32ContainerVT), ValInVec, ValHi,
I32Mask, InsertI64VL);
// Bitcast back to the right container type.
ValInVec = DAG.getBitcast(ContainerVT, ValInVec);
}
// Now that the value is in a vector, slide it into position.
SDValue InsertVL =
DAG.getNode(ISD::ADD, DL, XLenVT, Idx, DAG.getConstant(1, DL, XLenVT));
// Use tail agnostic policy if Idx is the last index of Vec.
unsigned Policy = RISCVVType::TAIL_UNDISTURBED_MASK_UNDISTURBED;
if (VecVT.isFixedLengthVector() && isa<ConstantSDNode>(Idx) &&
Idx->getAsZExtVal() + 1 == VecVT.getVectorNumElements())
Policy = RISCVVType::TAIL_AGNOSTIC;
SDValue Slideup = getVSlideup(DAG, Subtarget, DL, ContainerVT, Vec, ValInVec,
Idx, Mask, InsertVL, Policy);
if (AlignedIdx)
Slideup = DAG.getInsertSubvector(DL, OrigVec, Slideup, *AlignedIdx);
if (!VecVT.isFixedLengthVector())
return Slideup;
return convertFromScalableVector(VecVT, Slideup, DAG, Subtarget);
}
// Custom-lower EXTRACT_VECTOR_ELT operations to slide the vector down, then
// extract the first element: (extractelt (slidedown vec, idx), 0). For integer
// types this is done using VMV_X_S to allow us to glean information about the
// sign bits of the result.
SDValue RISCVTargetLowering::lowerEXTRACT_VECTOR_ELT(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
SDValue Idx = Op.getOperand(1);
SDValue Vec = Op.getOperand(0);
EVT EltVT = Op.getValueType();
MVT VecVT = Vec.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
if (VecVT.getVectorElementType() == MVT::i1) {
// Use vfirst.m to extract the first bit.
if (isNullConstant(Idx)) {
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
auto [Mask, VL] = getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget);
SDValue Vfirst =
DAG.getNode(RISCVISD::VFIRST_VL, DL, XLenVT, Vec, Mask, VL);
SDValue Res = DAG.getSetCC(DL, XLenVT, Vfirst,
DAG.getConstant(0, DL, XLenVT), ISD::SETEQ);
return DAG.getNode(ISD::TRUNCATE, DL, EltVT, Res);
}
if (VecVT.isFixedLengthVector()) {
unsigned NumElts = VecVT.getVectorNumElements();
if (NumElts >= 8) {
MVT WideEltVT;
unsigned WidenVecLen;
SDValue ExtractElementIdx;
SDValue ExtractBitIdx;
unsigned MaxEEW = Subtarget.getELen();
MVT LargestEltVT = MVT::getIntegerVT(
std::min(MaxEEW, unsigned(XLenVT.getSizeInBits())));
if (NumElts <= LargestEltVT.getSizeInBits()) {
assert(isPowerOf2_32(NumElts) &&
"the number of elements should be power of 2");
WideEltVT = MVT::getIntegerVT(NumElts);
WidenVecLen = 1;
ExtractElementIdx = DAG.getConstant(0, DL, XLenVT);
ExtractBitIdx = Idx;
} else {
WideEltVT = LargestEltVT;
WidenVecLen = NumElts / WideEltVT.getSizeInBits();
// extract element index = index / element width
ExtractElementIdx = DAG.getNode(
ISD::SRL, DL, XLenVT, Idx,
DAG.getConstant(Log2_64(WideEltVT.getSizeInBits()), DL, XLenVT));
// mask bit index = index % element width
ExtractBitIdx = DAG.getNode(
ISD::AND, DL, XLenVT, Idx,
DAG.getConstant(WideEltVT.getSizeInBits() - 1, DL, XLenVT));
}
MVT WideVT = MVT::getVectorVT(WideEltVT, WidenVecLen);
Vec = DAG.getNode(ISD::BITCAST, DL, WideVT, Vec);
SDValue ExtractElt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, XLenVT,
Vec, ExtractElementIdx);
// Extract the bit from GPR.
SDValue ShiftRight =
DAG.getNode(ISD::SRL, DL, XLenVT, ExtractElt, ExtractBitIdx);
SDValue Res = DAG.getNode(ISD::AND, DL, XLenVT, ShiftRight,
DAG.getConstant(1, DL, XLenVT));
return DAG.getNode(ISD::TRUNCATE, DL, EltVT, Res);
}
}
// Otherwise, promote to an i8 vector and extract from that.
MVT WideVT = MVT::getVectorVT(MVT::i8, VecVT.getVectorElementCount());
Vec = DAG.getNode(ISD::ZERO_EXTEND, DL, WideVT, Vec);
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, EltVT, Vec, Idx);
}
if ((EltVT == MVT::f16 && !Subtarget.hasVInstructionsF16()) ||
EltVT == MVT::bf16) {
// If we don't have vfmv.f.s for f16/bf16, extract to a gpr then use fmv.h.x
MVT IntVT = VecVT.changeTypeToInteger();
SDValue IntVec = DAG.getBitcast(IntVT, Vec);
SDValue IntExtract =
DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, XLenVT, IntVec, Idx);
return DAG.getNode(RISCVISD::FMV_H_X, DL, EltVT, IntExtract);
}
// If this is a fixed vector, we need to convert it to a scalable vector.
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
// If we're compiling for an exact VLEN value and we have a known
// constant index, we can always perform the extract in m1 (or
// smaller) as we can determine the register corresponding to
// the index in the register group.
const auto VLen = Subtarget.getRealVLen();
if (auto *IdxC = dyn_cast<ConstantSDNode>(Idx);
IdxC && VLen && VecVT.getSizeInBits().getKnownMinValue() > *VLen) {
MVT M1VT = RISCVTargetLowering::getM1VT(ContainerVT);
unsigned OrigIdx = IdxC->getZExtValue();
EVT ElemVT = VecVT.getVectorElementType();
unsigned ElemsPerVReg = *VLen / ElemVT.getFixedSizeInBits();
unsigned RemIdx = OrigIdx % ElemsPerVReg;
unsigned SubRegIdx = OrigIdx / ElemsPerVReg;
unsigned ExtractIdx =
SubRegIdx * M1VT.getVectorElementCount().getKnownMinValue();
Vec = DAG.getExtractSubvector(DL, M1VT, Vec, ExtractIdx);
Idx = DAG.getVectorIdxConstant(RemIdx, DL);
ContainerVT = M1VT;
}
// Reduce the LMUL of our slidedown and vmv.x.s to the smallest LMUL which
// contains our index.
std::optional<uint64_t> MaxIdx;
if (VecVT.isFixedLengthVector())
MaxIdx = VecVT.getVectorNumElements() - 1;
if (auto *IdxC = dyn_cast<ConstantSDNode>(Idx))
MaxIdx = IdxC->getZExtValue();
if (MaxIdx) {
if (auto SmallerVT =
getSmallestVTForIndex(ContainerVT, *MaxIdx, DL, DAG, Subtarget)) {
ContainerVT = *SmallerVT;
Vec = DAG.getExtractSubvector(DL, ContainerVT, Vec, 0);
}
}
// Use ri.vextract.x.v if available.
// TODO: Avoid index 0 and just use the vmv.x.s
if (Subtarget.hasVendorXRivosVisni() && EltVT.isInteger() &&
isValidVisniInsertExtractIndex(Idx)) {
SDValue Elt = DAG.getNode(RISCVISD::RI_VEXTRACT, DL, XLenVT, Vec, Idx);
return DAG.getNode(ISD::TRUNCATE, DL, EltVT, Elt);
}
// If after narrowing, the required slide is still greater than LMUL2,
// fallback to generic expansion and go through the stack. This is done
// for a subtle reason: extracting *all* elements out of a vector is
// widely expected to be linear in vector size, but because vslidedown
// is linear in LMUL, performing N extracts using vslidedown becomes
// O(n^2) / (VLEN/ETYPE) work. On the surface, going through the stack
// seems to have the same problem (the store is linear in LMUL), but the
// generic expansion *memoizes* the store, and thus for many extracts of
// the same vector we end up with one store and a bunch of loads.
// TODO: We don't have the same code for insert_vector_elt because we
// have BUILD_VECTOR and handle the degenerate case there. Should we
// consider adding an inverse BUILD_VECTOR node?
MVT LMUL2VT =
RISCVTargetLowering::getM1VT(ContainerVT).getDoubleNumVectorElementsVT();
if (ContainerVT.bitsGT(LMUL2VT) && VecVT.isFixedLengthVector())
return SDValue();
// If the index is 0, the vector is already in the right position.
if (!isNullConstant(Idx)) {
// Use a VL of 1 to avoid processing more elements than we need.
auto [Mask, VL] = getDefaultVLOps(1, ContainerVT, DL, DAG, Subtarget);
Vec = getVSlidedown(DAG, Subtarget, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), Vec, Idx, Mask, VL);
}
if (!EltVT.isInteger()) {
// Floating-point extracts are handled in TableGen.
return DAG.getExtractVectorElt(DL, EltVT, Vec, 0);
}
SDValue Elt0 = DAG.getNode(RISCVISD::VMV_X_S, DL, XLenVT, Vec);
return DAG.getNode(ISD::TRUNCATE, DL, EltVT, Elt0);
}
// Some RVV intrinsics may claim that they want an integer operand to be
// promoted or expanded.
static SDValue lowerVectorIntrinsicScalars(SDValue Op, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert((Op.getOpcode() == ISD::INTRINSIC_VOID ||
Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN ||
Op.getOpcode() == ISD::INTRINSIC_W_CHAIN) &&
"Unexpected opcode");
if (!Subtarget.hasVInstructions())
return SDValue();
bool HasChain = Op.getOpcode() == ISD::INTRINSIC_VOID ||
Op.getOpcode() == ISD::INTRINSIC_W_CHAIN;
unsigned IntNo = Op.getConstantOperandVal(HasChain ? 1 : 0);
SDLoc DL(Op);
const RISCVVIntrinsicsTable::RISCVVIntrinsicInfo *II =
RISCVVIntrinsicsTable::getRISCVVIntrinsicInfo(IntNo);
if (!II || !II->hasScalarOperand())
return SDValue();
unsigned SplatOp = II->ScalarOperand + 1 + HasChain;
assert(SplatOp < Op.getNumOperands());
SmallVector<SDValue, 8> Operands(Op->ops());
SDValue &ScalarOp = Operands[SplatOp];
MVT OpVT = ScalarOp.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
// If this isn't a scalar, or its type is XLenVT we're done.
if (!OpVT.isScalarInteger() || OpVT == XLenVT)
return SDValue();
// Simplest case is that the operand needs to be promoted to XLenVT.
if (OpVT.bitsLT(XLenVT)) {
// If the operand is a constant, sign extend to increase our chances
// of being able to use a .vi instruction. ANY_EXTEND would become a
// a zero extend and the simm5 check in isel would fail.
// FIXME: Should we ignore the upper bits in isel instead?
unsigned ExtOpc =
isa<ConstantSDNode>(ScalarOp) ? ISD::SIGN_EXTEND : ISD::ANY_EXTEND;
ScalarOp = DAG.getNode(ExtOpc, DL, XLenVT, ScalarOp);
return DAG.getNode(Op->getOpcode(), DL, Op->getVTList(), Operands);
}
// Use the previous operand to get the vXi64 VT. The result might be a mask
// VT for compares. Using the previous operand assumes that the previous
// operand will never have a smaller element size than a scalar operand and
// that a widening operation never uses SEW=64.
// NOTE: If this fails the below assert, we can probably just find the
// element count from any operand or result and use it to construct the VT.
assert(II->ScalarOperand > 0 && "Unexpected splat operand!");
MVT VT = Op.getOperand(SplatOp - 1).getSimpleValueType();
// The more complex case is when the scalar is larger than XLenVT.
assert(XLenVT == MVT::i32 && OpVT == MVT::i64 &&
VT.getVectorElementType() == MVT::i64 && "Unexpected VTs!");
// If this is a sign-extended 32-bit value, we can truncate it and rely on the
// instruction to sign-extend since SEW>XLEN.
if (DAG.ComputeNumSignBits(ScalarOp) > 32) {
ScalarOp = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, ScalarOp);
return DAG.getNode(Op->getOpcode(), DL, Op->getVTList(), Operands);
}
switch (IntNo) {
case Intrinsic::riscv_vslide1up:
case Intrinsic::riscv_vslide1down:
case Intrinsic::riscv_vslide1up_mask:
case Intrinsic::riscv_vslide1down_mask: {
// We need to special case these when the scalar is larger than XLen.
unsigned NumOps = Op.getNumOperands();
bool IsMasked = NumOps == 7;
// Convert the vector source to the equivalent nxvXi32 vector.
MVT I32VT = MVT::getVectorVT(MVT::i32, VT.getVectorElementCount() * 2);
SDValue Vec = DAG.getBitcast(I32VT, Operands[2]);
SDValue ScalarLo, ScalarHi;
std::tie(ScalarLo, ScalarHi) =
DAG.SplitScalar(ScalarOp, DL, MVT::i32, MVT::i32);
// Double the VL since we halved SEW.
SDValue AVL = getVLOperand(Op);
SDValue I32VL;
// Optimize for constant AVL
if (isa<ConstantSDNode>(AVL)) {
const auto [MinVLMAX, MaxVLMAX] =
RISCVTargetLowering::computeVLMAXBounds(VT, Subtarget);
uint64_t AVLInt = AVL->getAsZExtVal();
if (AVLInt <= MinVLMAX) {
I32VL = DAG.getConstant(2 * AVLInt, DL, XLenVT);
} else if (AVLInt >= 2 * MaxVLMAX) {
// Just set vl to VLMAX in this situation
I32VL = DAG.getRegister(RISCV::X0, XLenVT);
} else {
// For AVL between (MinVLMAX, 2 * MaxVLMAX), the actual working vl
// is related to the hardware implementation.
// So let the following code handle
}
}
if (!I32VL) {
RISCVVType::VLMUL Lmul = RISCVTargetLowering::getLMUL(VT);
SDValue LMUL = DAG.getConstant(Lmul, DL, XLenVT);
unsigned Sew = RISCVVType::encodeSEW(VT.getScalarSizeInBits());
SDValue SEW = DAG.getConstant(Sew, DL, XLenVT);
SDValue SETVL =
DAG.getTargetConstant(Intrinsic::riscv_vsetvli, DL, MVT::i32);
// Using vsetvli instruction to get actually used length which related to
// the hardware implementation
SDValue VL = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, XLenVT, SETVL, AVL,
SEW, LMUL);
I32VL =
DAG.getNode(ISD::SHL, DL, XLenVT, VL, DAG.getConstant(1, DL, XLenVT));
}
SDValue I32Mask = getAllOnesMask(I32VT, I32VL, DL, DAG);
// Shift the two scalar parts in using SEW=32 slide1up/slide1down
// instructions.
SDValue Passthru;
if (IsMasked)
Passthru = DAG.getUNDEF(I32VT);
else
Passthru = DAG.getBitcast(I32VT, Operands[1]);
if (IntNo == Intrinsic::riscv_vslide1up ||
IntNo == Intrinsic::riscv_vslide1up_mask) {
Vec = DAG.getNode(RISCVISD::VSLIDE1UP_VL, DL, I32VT, Passthru, Vec,
ScalarHi, I32Mask, I32VL);
Vec = DAG.getNode(RISCVISD::VSLIDE1UP_VL, DL, I32VT, Passthru, Vec,
ScalarLo, I32Mask, I32VL);
} else {
Vec = DAG.getNode(RISCVISD::VSLIDE1DOWN_VL, DL, I32VT, Passthru, Vec,
ScalarLo, I32Mask, I32VL);
Vec = DAG.getNode(RISCVISD::VSLIDE1DOWN_VL, DL, I32VT, Passthru, Vec,
ScalarHi, I32Mask, I32VL);
}
// Convert back to nxvXi64.
Vec = DAG.getBitcast(VT, Vec);
if (!IsMasked)
return Vec;
// Apply mask after the operation.
SDValue Mask = Operands[NumOps - 3];
SDValue MaskedOff = Operands[1];
// Assume Policy operand is the last operand.
uint64_t Policy = Operands[NumOps - 1]->getAsZExtVal();
// We don't need to select maskedoff if it's undef.
if (MaskedOff.isUndef())
return Vec;
// TAMU
if (Policy == RISCVVType::TAIL_AGNOSTIC)
return DAG.getNode(RISCVISD::VMERGE_VL, DL, VT, Mask, Vec, MaskedOff,
DAG.getUNDEF(VT), AVL);
// TUMA or TUMU: Currently we always emit tumu policy regardless of tuma.
// It's fine because vmerge does not care mask policy.
return DAG.getNode(RISCVISD::VMERGE_VL, DL, VT, Mask, Vec, MaskedOff,
MaskedOff, AVL);
}
}
// We need to convert the scalar to a splat vector.
SDValue VL = getVLOperand(Op);
assert(VL.getValueType() == XLenVT);
ScalarOp = splatSplitI64WithVL(DL, VT, SDValue(), ScalarOp, VL, DAG);
return DAG.getNode(Op->getOpcode(), DL, Op->getVTList(), Operands);
}
// Lower the llvm.get.vector.length intrinsic to vsetvli. We only support
// scalable vector llvm.get.vector.length for now.
//
// We need to convert from a scalable VF to a vsetvli with VLMax equal to
// (vscale * VF). The vscale and VF are independent of element width. We use
// SEW=8 for the vsetvli because it is the only element width that supports all
// fractional LMULs. The LMUL is chosen so that with SEW=8 the VLMax is
// (vscale * VF). Where vscale is defined as VLEN/RVVBitsPerBlock. The
// InsertVSETVLI pass can fix up the vtype of the vsetvli if a different
// SEW and LMUL are better for the surrounding vector instructions.
static SDValue lowerGetVectorLength(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
MVT XLenVT = Subtarget.getXLenVT();
// The smallest LMUL is only valid for the smallest element width.
const unsigned ElementWidth = 8;
// Determine the VF that corresponds to LMUL 1 for ElementWidth.
unsigned LMul1VF = RISCV::RVVBitsPerBlock / ElementWidth;
// We don't support VF==1 with ELEN==32.
[[maybe_unused]] unsigned MinVF =
RISCV::RVVBitsPerBlock / Subtarget.getELen();
[[maybe_unused]] unsigned VF = N->getConstantOperandVal(2);
assert(VF >= MinVF && VF <= (LMul1VF * 8) && isPowerOf2_32(VF) &&
"Unexpected VF");
bool Fractional = VF < LMul1VF;
unsigned LMulVal = Fractional ? LMul1VF / VF : VF / LMul1VF;
unsigned VLMUL = (unsigned)RISCVVType::encodeLMUL(LMulVal, Fractional);
unsigned VSEW = RISCVVType::encodeSEW(ElementWidth);
SDLoc DL(N);
SDValue LMul = DAG.getTargetConstant(VLMUL, DL, XLenVT);
SDValue Sew = DAG.getTargetConstant(VSEW, DL, XLenVT);
SDValue AVL = DAG.getNode(ISD::ZERO_EXTEND, DL, XLenVT, N->getOperand(1));
SDValue ID = DAG.getTargetConstant(Intrinsic::riscv_vsetvli, DL, XLenVT);
SDValue Res =
DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, XLenVT, ID, AVL, Sew, LMul);
return DAG.getNode(ISD::TRUNCATE, DL, N->getValueType(0), Res);
}
static SDValue lowerCttzElts(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDValue Op0 = N->getOperand(1);
MVT OpVT = Op0.getSimpleValueType();
MVT ContainerVT = OpVT;
if (OpVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(DAG, OpVT, Subtarget);
Op0 = convertToScalableVector(ContainerVT, Op0, DAG, Subtarget);
}
MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(N);
auto [Mask, VL] = getDefaultVLOps(OpVT, ContainerVT, DL, DAG, Subtarget);
SDValue Res = DAG.getNode(RISCVISD::VFIRST_VL, DL, XLenVT, Op0, Mask, VL);
if (isOneConstant(N->getOperand(2)))
return Res;
// Convert -1 to VL.
SDValue Setcc =
DAG.getSetCC(DL, XLenVT, Res, DAG.getConstant(0, DL, XLenVT), ISD::SETLT);
VL = DAG.getElementCount(DL, XLenVT, OpVT.getVectorElementCount());
return DAG.getSelect(DL, XLenVT, Setcc, VL, Res);
}
static inline void promoteVCIXScalar(const SDValue &Op,
SmallVectorImpl<SDValue> &Operands,
SelectionDAG &DAG) {
const RISCVSubtarget &Subtarget =
DAG.getMachineFunction().getSubtarget<RISCVSubtarget>();
bool HasChain = Op.getOpcode() == ISD::INTRINSIC_VOID ||
Op.getOpcode() == ISD::INTRINSIC_W_CHAIN;
unsigned IntNo = Op.getConstantOperandVal(HasChain ? 1 : 0);
SDLoc DL(Op);
const RISCVVIntrinsicsTable::RISCVVIntrinsicInfo *II =
RISCVVIntrinsicsTable::getRISCVVIntrinsicInfo(IntNo);
if (!II || !II->hasScalarOperand())
return;
unsigned SplatOp = II->ScalarOperand + 1;
assert(SplatOp < Op.getNumOperands());
SDValue &ScalarOp = Operands[SplatOp];
MVT OpVT = ScalarOp.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
// The code below is partially copied from lowerVectorIntrinsicScalars.
// If this isn't a scalar, or its type is XLenVT we're done.
if (!OpVT.isScalarInteger() || OpVT == XLenVT)
return;
// Manually emit promote operation for scalar operation.
if (OpVT.bitsLT(XLenVT)) {
unsigned ExtOpc =
isa<ConstantSDNode>(ScalarOp) ? ISD::SIGN_EXTEND : ISD::ANY_EXTEND;
ScalarOp = DAG.getNode(ExtOpc, DL, XLenVT, ScalarOp);
}
}
static void processVCIXOperands(SDValue &OrigOp,
SmallVectorImpl<SDValue> &Operands,
SelectionDAG &DAG) {
promoteVCIXScalar(OrigOp, Operands, DAG);
const RISCVSubtarget &Subtarget =
DAG.getMachineFunction().getSubtarget<RISCVSubtarget>();
for (SDValue &V : Operands) {
EVT ValType = V.getValueType();
if (ValType.isVector() && ValType.isFloatingPoint()) {
MVT InterimIVT =
MVT::getVectorVT(MVT::getIntegerVT(ValType.getScalarSizeInBits()),
ValType.getVectorElementCount());
V = DAG.getBitcast(InterimIVT, V);
}
if (ValType.isFixedLengthVector()) {
MVT OpContainerVT = getContainerForFixedLengthVector(
DAG, V.getSimpleValueType(), Subtarget);
V = convertToScalableVector(OpContainerVT, V, DAG, Subtarget);
}
}
}
// LMUL * VLEN should be greater than or equal to EGS * SEW
static inline bool isValidEGW(int EGS, EVT VT,
const RISCVSubtarget &Subtarget) {
return (Subtarget.getRealMinVLen() *
VT.getSizeInBits().getKnownMinValue()) / RISCV::RVVBitsPerBlock >=
EGS * VT.getScalarSizeInBits();
}
SDValue RISCVTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
SelectionDAG &DAG) const {
unsigned IntNo = Op.getConstantOperandVal(0);
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
switch (IntNo) {
default:
break; // Don't custom lower most intrinsics.
case Intrinsic::riscv_tuple_insert: {
SDValue Vec = Op.getOperand(1);
SDValue SubVec = Op.getOperand(2);
SDValue Index = Op.getOperand(3);
return DAG.getNode(RISCVISD::TUPLE_INSERT, DL, Op.getValueType(), Vec,
SubVec, Index);
}
case Intrinsic::riscv_tuple_extract: {
SDValue Vec = Op.getOperand(1);
SDValue Index = Op.getOperand(2);
return DAG.getNode(RISCVISD::TUPLE_EXTRACT, DL, Op.getValueType(), Vec,
Index);
}
case Intrinsic::thread_pointer: {
EVT PtrVT = getPointerTy(DAG.getDataLayout());
return DAG.getRegister(RISCV::X4, PtrVT);
}
case Intrinsic::riscv_orc_b:
case Intrinsic::riscv_brev8:
case Intrinsic::riscv_sha256sig0:
case Intrinsic::riscv_sha256sig1:
case Intrinsic::riscv_sha256sum0:
case Intrinsic::riscv_sha256sum1:
case Intrinsic::riscv_sm3p0:
case Intrinsic::riscv_sm3p1: {
unsigned Opc;
switch (IntNo) {
case Intrinsic::riscv_orc_b: Opc = RISCVISD::ORC_B; break;
case Intrinsic::riscv_brev8: Opc = RISCVISD::BREV8; break;
case Intrinsic::riscv_sha256sig0: Opc = RISCVISD::SHA256SIG0; break;
case Intrinsic::riscv_sha256sig1: Opc = RISCVISD::SHA256SIG1; break;
case Intrinsic::riscv_sha256sum0: Opc = RISCVISD::SHA256SUM0; break;
case Intrinsic::riscv_sha256sum1: Opc = RISCVISD::SHA256SUM1; break;
case Intrinsic::riscv_sm3p0: Opc = RISCVISD::SM3P0; break;
case Intrinsic::riscv_sm3p1: Opc = RISCVISD::SM3P1; break;
}
return DAG.getNode(Opc, DL, XLenVT, Op.getOperand(1));
}
case Intrinsic::riscv_sm4ks:
case Intrinsic::riscv_sm4ed: {
unsigned Opc =
IntNo == Intrinsic::riscv_sm4ks ? RISCVISD::SM4KS : RISCVISD::SM4ED;
return DAG.getNode(Opc, DL, XLenVT, Op.getOperand(1), Op.getOperand(2),
Op.getOperand(3));
}
case Intrinsic::riscv_zip:
case Intrinsic::riscv_unzip: {
unsigned Opc =
IntNo == Intrinsic::riscv_zip ? RISCVISD::ZIP : RISCVISD::UNZIP;
return DAG.getNode(Opc, DL, XLenVT, Op.getOperand(1));
}
case Intrinsic::riscv_mopr:
return DAG.getNode(RISCVISD::MOP_R, DL, XLenVT, Op.getOperand(1),
Op.getOperand(2));
case Intrinsic::riscv_moprr: {
return DAG.getNode(RISCVISD::MOP_RR, DL, XLenVT, Op.getOperand(1),
Op.getOperand(2), Op.getOperand(3));
}
case Intrinsic::riscv_clmul:
return DAG.getNode(RISCVISD::CLMUL, DL, XLenVT, Op.getOperand(1),
Op.getOperand(2));
case Intrinsic::riscv_clmulh:
case Intrinsic::riscv_clmulr: {
unsigned Opc =
IntNo == Intrinsic::riscv_clmulh ? RISCVISD::CLMULH : RISCVISD::CLMULR;
return DAG.getNode(Opc, DL, XLenVT, Op.getOperand(1), Op.getOperand(2));
}
case Intrinsic::experimental_get_vector_length:
return lowerGetVectorLength(Op.getNode(), DAG, Subtarget);
case Intrinsic::experimental_cttz_elts:
return lowerCttzElts(Op.getNode(), DAG, Subtarget);
case Intrinsic::riscv_vmv_x_s: {
SDValue Res = DAG.getNode(RISCVISD::VMV_X_S, DL, XLenVT, Op.getOperand(1));
return DAG.getNode(ISD::TRUNCATE, DL, Op.getValueType(), Res);
}
case Intrinsic::riscv_vfmv_f_s:
return DAG.getExtractVectorElt(DL, Op.getValueType(), Op.getOperand(1), 0);
case Intrinsic::riscv_vmv_v_x:
return lowerScalarSplat(Op.getOperand(1), Op.getOperand(2),
Op.getOperand(3), Op.getSimpleValueType(), DL, DAG,
Subtarget);
case Intrinsic::riscv_vfmv_v_f:
return DAG.getNode(RISCVISD::VFMV_V_F_VL, DL, Op.getValueType(),
Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
case Intrinsic::riscv_vmv_s_x: {
SDValue Scalar = Op.getOperand(2);
if (Scalar.getValueType().bitsLE(XLenVT)) {
Scalar = DAG.getNode(ISD::ANY_EXTEND, DL, XLenVT, Scalar);
return DAG.getNode(RISCVISD::VMV_S_X_VL, DL, Op.getValueType(),
Op.getOperand(1), Scalar, Op.getOperand(3));
}
assert(Scalar.getValueType() == MVT::i64 && "Unexpected scalar VT!");
// This is an i64 value that lives in two scalar registers. We have to
// insert this in a convoluted way. First we build vXi64 splat containing
// the two values that we assemble using some bit math. Next we'll use
// vid.v and vmseq to build a mask with bit 0 set. Then we'll use that mask
// to merge element 0 from our splat into the source vector.
// FIXME: This is probably not the best way to do this, but it is
// consistent with INSERT_VECTOR_ELT lowering so it is a good starting
// point.
// sw lo, (a0)
// sw hi, 4(a0)
// vlse vX, (a0)
//
// vid.v vVid
// vmseq.vx mMask, vVid, 0
// vmerge.vvm vDest, vSrc, vVal, mMask
MVT VT = Op.getSimpleValueType();
SDValue Vec = Op.getOperand(1);
SDValue VL = getVLOperand(Op);
SDValue SplattedVal = splatSplitI64WithVL(DL, VT, SDValue(), Scalar, VL, DAG);
if (Op.getOperand(1).isUndef())
return SplattedVal;
SDValue SplattedIdx =
DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, DAG.getUNDEF(VT),
DAG.getConstant(0, DL, MVT::i32), VL);
MVT MaskVT = getMaskTypeFor(VT);
SDValue Mask = getAllOnesMask(VT, VL, DL, DAG);
SDValue VID = DAG.getNode(RISCVISD::VID_VL, DL, VT, Mask, VL);
SDValue SelectCond =
DAG.getNode(RISCVISD::SETCC_VL, DL, MaskVT,
{VID, SplattedIdx, DAG.getCondCode(ISD::SETEQ),
DAG.getUNDEF(MaskVT), Mask, VL});
return DAG.getNode(RISCVISD::VMERGE_VL, DL, VT, SelectCond, SplattedVal,
Vec, DAG.getUNDEF(VT), VL);
}
case Intrinsic::riscv_vfmv_s_f:
return DAG.getNode(RISCVISD::VFMV_S_F_VL, DL, Op.getSimpleValueType(),
Op.getOperand(1), Op.getOperand(2), Op.getOperand(3));
// EGS * EEW >= 128 bits
case Intrinsic::riscv_vaesdf_vv:
case Intrinsic::riscv_vaesdf_vs:
case Intrinsic::riscv_vaesdm_vv:
case Intrinsic::riscv_vaesdm_vs:
case Intrinsic::riscv_vaesef_vv:
case Intrinsic::riscv_vaesef_vs:
case Intrinsic::riscv_vaesem_vv:
case Intrinsic::riscv_vaesem_vs:
case Intrinsic::riscv_vaeskf1:
case Intrinsic::riscv_vaeskf2:
case Intrinsic::riscv_vaesz_vs:
case Intrinsic::riscv_vsm4k:
case Intrinsic::riscv_vsm4r_vv:
case Intrinsic::riscv_vsm4r_vs: {
if (!isValidEGW(4, Op.getSimpleValueType(), Subtarget) ||
!isValidEGW(4, Op->getOperand(1).getSimpleValueType(), Subtarget) ||
!isValidEGW(4, Op->getOperand(2).getSimpleValueType(), Subtarget))
reportFatalUsageError("EGW should be greater than or equal to 4 * SEW.");
return Op;
}
// EGS * EEW >= 256 bits
case Intrinsic::riscv_vsm3c:
case Intrinsic::riscv_vsm3me: {
if (!isValidEGW(8, Op.getSimpleValueType(), Subtarget) ||
!isValidEGW(8, Op->getOperand(1).getSimpleValueType(), Subtarget))
reportFatalUsageError("EGW should be greater than or equal to 8 * SEW.");
return Op;
}
// zvknha(SEW=32)/zvknhb(SEW=[32|64])
case Intrinsic::riscv_vsha2ch:
case Intrinsic::riscv_vsha2cl:
case Intrinsic::riscv_vsha2ms: {
if (Op->getSimpleValueType(0).getScalarSizeInBits() == 64 &&
!Subtarget.hasStdExtZvknhb())
reportFatalUsageError("SEW=64 needs Zvknhb to be enabled.");
if (!isValidEGW(4, Op.getSimpleValueType(), Subtarget) ||
!isValidEGW(4, Op->getOperand(1).getSimpleValueType(), Subtarget) ||
!isValidEGW(4, Op->getOperand(2).getSimpleValueType(), Subtarget))
reportFatalUsageError("EGW should be greater than or equal to 4 * SEW.");
return Op;
}
case Intrinsic::riscv_sf_vc_v_x:
case Intrinsic::riscv_sf_vc_v_i:
case Intrinsic::riscv_sf_vc_v_xv:
case Intrinsic::riscv_sf_vc_v_iv:
case Intrinsic::riscv_sf_vc_v_vv:
case Intrinsic::riscv_sf_vc_v_fv:
case Intrinsic::riscv_sf_vc_v_xvv:
case Intrinsic::riscv_sf_vc_v_ivv:
case Intrinsic::riscv_sf_vc_v_vvv:
case Intrinsic::riscv_sf_vc_v_fvv:
case Intrinsic::riscv_sf_vc_v_xvw:
case Intrinsic::riscv_sf_vc_v_ivw:
case Intrinsic::riscv_sf_vc_v_vvw:
case Intrinsic::riscv_sf_vc_v_fvw: {
MVT VT = Op.getSimpleValueType();
SmallVector<SDValue> Operands{Op->op_values()};
processVCIXOperands(Op, Operands, DAG);
MVT RetVT = VT;
if (VT.isFixedLengthVector())
RetVT = getContainerForFixedLengthVector(VT);
else if (VT.isFloatingPoint())
RetVT = MVT::getVectorVT(MVT::getIntegerVT(VT.getScalarSizeInBits()),
VT.getVectorElementCount());
SDValue NewNode = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, RetVT, Operands);
if (VT.isFixedLengthVector())
NewNode = convertFromScalableVector(VT, NewNode, DAG, Subtarget);
else if (VT.isFloatingPoint())
NewNode = DAG.getBitcast(VT, NewNode);
if (Op == NewNode)
break;
return NewNode;
}
}
return lowerVectorIntrinsicScalars(Op, DAG, Subtarget);
}
static inline SDValue getVCIXISDNodeWCHAIN(SDValue &Op, SelectionDAG &DAG,
unsigned Type) {
SDLoc DL(Op);
SmallVector<SDValue> Operands{Op->op_values()};
Operands.erase(Operands.begin() + 1);
const RISCVSubtarget &Subtarget =
DAG.getMachineFunction().getSubtarget<RISCVSubtarget>();
MVT VT = Op.getSimpleValueType();
MVT RetVT = VT;
MVT FloatVT = VT;
if (VT.isFloatingPoint()) {
RetVT = MVT::getVectorVT(MVT::getIntegerVT(VT.getScalarSizeInBits()),
VT.getVectorElementCount());
FloatVT = RetVT;
}
if (VT.isFixedLengthVector())
RetVT = getContainerForFixedLengthVector(DAG.getTargetLoweringInfo(), RetVT,
Subtarget);
processVCIXOperands(Op, Operands, DAG);
SDVTList VTs = DAG.getVTList({RetVT, MVT::Other});
SDValue NewNode = DAG.getNode(Type, DL, VTs, Operands);
SDValue Chain = NewNode.getValue(1);
if (VT.isFixedLengthVector())
NewNode = convertFromScalableVector(FloatVT, NewNode, DAG, Subtarget);
if (VT.isFloatingPoint())
NewNode = DAG.getBitcast(VT, NewNode);
NewNode = DAG.getMergeValues({NewNode, Chain}, DL);
return NewNode;
}
static inline SDValue getVCIXISDNodeVOID(SDValue &Op, SelectionDAG &DAG,
unsigned Type) {
SmallVector<SDValue> Operands{Op->op_values()};
Operands.erase(Operands.begin() + 1);
processVCIXOperands(Op, Operands, DAG);
return DAG.getNode(Type, SDLoc(Op), Op.getValueType(), Operands);
}
static SDValue
lowerFixedVectorSegLoadIntrinsics(unsigned IntNo, SDValue Op,
const RISCVSubtarget &Subtarget,
SelectionDAG &DAG) {
bool IsStrided;
switch (IntNo) {
case Intrinsic::riscv_seg2_load_mask:
case Intrinsic::riscv_seg3_load_mask:
case Intrinsic::riscv_seg4_load_mask:
case Intrinsic::riscv_seg5_load_mask:
case Intrinsic::riscv_seg6_load_mask:
case Intrinsic::riscv_seg7_load_mask:
case Intrinsic::riscv_seg8_load_mask:
IsStrided = false;
break;
case Intrinsic::riscv_sseg2_load_mask:
case Intrinsic::riscv_sseg3_load_mask:
case Intrinsic::riscv_sseg4_load_mask:
case Intrinsic::riscv_sseg5_load_mask:
case Intrinsic::riscv_sseg6_load_mask:
case Intrinsic::riscv_sseg7_load_mask:
case Intrinsic::riscv_sseg8_load_mask:
IsStrided = true;
break;
default:
llvm_unreachable("unexpected intrinsic ID");
};
static const Intrinsic::ID VlsegInts[7] = {
Intrinsic::riscv_vlseg2_mask, Intrinsic::riscv_vlseg3_mask,
Intrinsic::riscv_vlseg4_mask, Intrinsic::riscv_vlseg5_mask,
Intrinsic::riscv_vlseg6_mask, Intrinsic::riscv_vlseg7_mask,
Intrinsic::riscv_vlseg8_mask};
static const Intrinsic::ID VlssegInts[7] = {
Intrinsic::riscv_vlsseg2_mask, Intrinsic::riscv_vlsseg3_mask,
Intrinsic::riscv_vlsseg4_mask, Intrinsic::riscv_vlsseg5_mask,
Intrinsic::riscv_vlsseg6_mask, Intrinsic::riscv_vlsseg7_mask,
Intrinsic::riscv_vlsseg8_mask};
SDLoc DL(Op);
unsigned NF = Op->getNumValues() - 1;
assert(NF >= 2 && NF <= 8 && "Unexpected seg number");
MVT XLenVT = Subtarget.getXLenVT();
MVT VT = Op->getSimpleValueType(0);
MVT ContainerVT = ::getContainerForFixedLengthVector(DAG, VT, Subtarget);
unsigned Sz = NF * ContainerVT.getVectorMinNumElements() *
ContainerVT.getScalarSizeInBits();
EVT VecTupTy = MVT::getRISCVVectorTupleVT(Sz, NF);
// Operands: (chain, int_id, pointer, mask, vl) or
// (chain, int_id, pointer, offset, mask, vl)
SDValue VL = Op.getOperand(Op.getNumOperands() - 1);
SDValue Mask = Op.getOperand(Op.getNumOperands() - 2);
MVT MaskVT = Mask.getSimpleValueType();
MVT MaskContainerVT =
::getContainerForFixedLengthVector(DAG, MaskVT, Subtarget);
Mask = convertToScalableVector(MaskContainerVT, Mask, DAG, Subtarget);
SDValue IntID = DAG.getTargetConstant(
IsStrided ? VlssegInts[NF - 2] : VlsegInts[NF - 2], DL, XLenVT);
auto *Load = cast<MemIntrinsicSDNode>(Op);
SDVTList VTs = DAG.getVTList({VecTupTy, MVT::Other});
SmallVector<SDValue, 9> Ops = {
Load->getChain(),
IntID,
DAG.getUNDEF(VecTupTy),
Op.getOperand(2),
Mask,
VL,
DAG.getTargetConstant(
RISCVVType::TAIL_AGNOSTIC | RISCVVType::MASK_AGNOSTIC, DL, XLenVT),
DAG.getTargetConstant(Log2_64(VT.getScalarSizeInBits()), DL, XLenVT)};
// Insert the stride operand.
if (IsStrided)
Ops.insert(std::next(Ops.begin(), 4), Op.getOperand(3));
SDValue Result =
DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops,
Load->getMemoryVT(), Load->getMemOperand());
SmallVector<SDValue, 9> Results;
for (unsigned int RetIdx = 0; RetIdx < NF; RetIdx++) {
SDValue SubVec = DAG.getNode(RISCVISD::TUPLE_EXTRACT, DL, ContainerVT,
Result.getValue(0),
DAG.getTargetConstant(RetIdx, DL, MVT::i32));
Results.push_back(convertFromScalableVector(VT, SubVec, DAG, Subtarget));
}
Results.push_back(Result.getValue(1));
return DAG.getMergeValues(Results, DL);
}
SDValue RISCVTargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op,
SelectionDAG &DAG) const {
unsigned IntNo = Op.getConstantOperandVal(1);
switch (IntNo) {
default:
break;
case Intrinsic::riscv_seg2_load_mask:
case Intrinsic::riscv_seg3_load_mask:
case Intrinsic::riscv_seg4_load_mask:
case Intrinsic::riscv_seg5_load_mask:
case Intrinsic::riscv_seg6_load_mask:
case Intrinsic::riscv_seg7_load_mask:
case Intrinsic::riscv_seg8_load_mask:
case Intrinsic::riscv_sseg2_load_mask:
case Intrinsic::riscv_sseg3_load_mask:
case Intrinsic::riscv_sseg4_load_mask:
case Intrinsic::riscv_sseg5_load_mask:
case Intrinsic::riscv_sseg6_load_mask:
case Intrinsic::riscv_sseg7_load_mask:
case Intrinsic::riscv_sseg8_load_mask:
return lowerFixedVectorSegLoadIntrinsics(IntNo, Op, Subtarget, DAG);
case Intrinsic::riscv_sf_vc_v_x_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_X_SE);
case Intrinsic::riscv_sf_vc_v_i_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_I_SE);
case Intrinsic::riscv_sf_vc_v_xv_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_XV_SE);
case Intrinsic::riscv_sf_vc_v_iv_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_IV_SE);
case Intrinsic::riscv_sf_vc_v_vv_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_VV_SE);
case Intrinsic::riscv_sf_vc_v_fv_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_FV_SE);
case Intrinsic::riscv_sf_vc_v_xvv_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_XVV_SE);
case Intrinsic::riscv_sf_vc_v_ivv_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_IVV_SE);
case Intrinsic::riscv_sf_vc_v_vvv_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_VVV_SE);
case Intrinsic::riscv_sf_vc_v_fvv_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_FVV_SE);
case Intrinsic::riscv_sf_vc_v_xvw_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_XVW_SE);
case Intrinsic::riscv_sf_vc_v_ivw_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_IVW_SE);
case Intrinsic::riscv_sf_vc_v_vvw_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_VVW_SE);
case Intrinsic::riscv_sf_vc_v_fvw_se:
return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_FVW_SE);
}
return lowerVectorIntrinsicScalars(Op, DAG, Subtarget);
}
static SDValue
lowerFixedVectorSegStoreIntrinsics(unsigned IntNo, SDValue Op,
const RISCVSubtarget &Subtarget,
SelectionDAG &DAG) {
bool IsStrided;
switch (IntNo) {
case Intrinsic::riscv_seg2_store_mask:
case Intrinsic::riscv_seg3_store_mask:
case Intrinsic::riscv_seg4_store_mask:
case Intrinsic::riscv_seg5_store_mask:
case Intrinsic::riscv_seg6_store_mask:
case Intrinsic::riscv_seg7_store_mask:
case Intrinsic::riscv_seg8_store_mask:
IsStrided = false;
break;
case Intrinsic::riscv_sseg2_store_mask:
case Intrinsic::riscv_sseg3_store_mask:
case Intrinsic::riscv_sseg4_store_mask:
case Intrinsic::riscv_sseg5_store_mask:
case Intrinsic::riscv_sseg6_store_mask:
case Intrinsic::riscv_sseg7_store_mask:
case Intrinsic::riscv_sseg8_store_mask:
IsStrided = true;
break;
default:
llvm_unreachable("unexpected intrinsic ID");
}
SDLoc DL(Op);
static const Intrinsic::ID VssegInts[] = {
Intrinsic::riscv_vsseg2_mask, Intrinsic::riscv_vsseg3_mask,
Intrinsic::riscv_vsseg4_mask, Intrinsic::riscv_vsseg5_mask,
Intrinsic::riscv_vsseg6_mask, Intrinsic::riscv_vsseg7_mask,
Intrinsic::riscv_vsseg8_mask};
static const Intrinsic::ID VsssegInts[] = {
Intrinsic::riscv_vssseg2_mask, Intrinsic::riscv_vssseg3_mask,
Intrinsic::riscv_vssseg4_mask, Intrinsic::riscv_vssseg5_mask,
Intrinsic::riscv_vssseg6_mask, Intrinsic::riscv_vssseg7_mask,
Intrinsic::riscv_vssseg8_mask};
// Operands: (chain, int_id, vec*, ptr, mask, vl) or
// (chain, int_id, vec*, ptr, stride, mask, vl)
unsigned NF = Op->getNumOperands() - (IsStrided ? 6 : 5);
assert(NF >= 2 && NF <= 8 && "Unexpected seg number");
MVT XLenVT = Subtarget.getXLenVT();
MVT VT = Op->getOperand(2).getSimpleValueType();
MVT ContainerVT = ::getContainerForFixedLengthVector(DAG, VT, Subtarget);
unsigned Sz = NF * ContainerVT.getVectorMinNumElements() *
ContainerVT.getScalarSizeInBits();
EVT VecTupTy = MVT::getRISCVVectorTupleVT(Sz, NF);
SDValue VL = Op.getOperand(Op.getNumOperands() - 1);
SDValue Mask = Op.getOperand(Op.getNumOperands() - 2);
MVT MaskVT = Mask.getSimpleValueType();
MVT MaskContainerVT =
::getContainerForFixedLengthVector(DAG, MaskVT, Subtarget);
Mask = convertToScalableVector(MaskContainerVT, Mask, DAG, Subtarget);
SDValue IntID = DAG.getTargetConstant(
IsStrided ? VsssegInts[NF - 2] : VssegInts[NF - 2], DL, XLenVT);
SDValue Ptr = Op->getOperand(NF + 2);
auto *FixedIntrinsic = cast<MemIntrinsicSDNode>(Op);
SDValue StoredVal = DAG.getUNDEF(VecTupTy);
for (unsigned i = 0; i < NF; i++)
StoredVal = DAG.getNode(
RISCVISD::TUPLE_INSERT, DL, VecTupTy, StoredVal,
convertToScalableVector(ContainerVT, FixedIntrinsic->getOperand(2 + i),
DAG, Subtarget),
DAG.getTargetConstant(i, DL, MVT::i32));
SmallVector<SDValue, 10> Ops = {
FixedIntrinsic->getChain(),
IntID,
StoredVal,
Ptr,
Mask,
VL,
DAG.getTargetConstant(Log2_64(VT.getScalarSizeInBits()), DL, XLenVT)};
// Insert the stride operand.
if (IsStrided)
Ops.insert(std::next(Ops.begin(), 4),
Op.getOperand(Op.getNumOperands() - 3));
return DAG.getMemIntrinsicNode(
ISD::INTRINSIC_VOID, DL, DAG.getVTList(MVT::Other), Ops,
FixedIntrinsic->getMemoryVT(), FixedIntrinsic->getMemOperand());
}
SDValue RISCVTargetLowering::LowerINTRINSIC_VOID(SDValue Op,
SelectionDAG &DAG) const {
unsigned IntNo = Op.getConstantOperandVal(1);
switch (IntNo) {
default:
break;
case Intrinsic::riscv_seg2_store_mask:
case Intrinsic::riscv_seg3_store_mask:
case Intrinsic::riscv_seg4_store_mask:
case Intrinsic::riscv_seg5_store_mask:
case Intrinsic::riscv_seg6_store_mask:
case Intrinsic::riscv_seg7_store_mask:
case Intrinsic::riscv_seg8_store_mask:
case Intrinsic::riscv_sseg2_store_mask:
case Intrinsic::riscv_sseg3_store_mask:
case Intrinsic::riscv_sseg4_store_mask:
case Intrinsic::riscv_sseg5_store_mask:
case Intrinsic::riscv_sseg6_store_mask:
case Intrinsic::riscv_sseg7_store_mask:
case Intrinsic::riscv_sseg8_store_mask:
return lowerFixedVectorSegStoreIntrinsics(IntNo, Op, Subtarget, DAG);
case Intrinsic::riscv_sf_vc_xv_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_XV_SE);
case Intrinsic::riscv_sf_vc_iv_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_IV_SE);
case Intrinsic::riscv_sf_vc_vv_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_VV_SE);
case Intrinsic::riscv_sf_vc_fv_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_FV_SE);
case Intrinsic::riscv_sf_vc_xvv_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_XVV_SE);
case Intrinsic::riscv_sf_vc_ivv_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_IVV_SE);
case Intrinsic::riscv_sf_vc_vvv_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_VVV_SE);
case Intrinsic::riscv_sf_vc_fvv_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_FVV_SE);
case Intrinsic::riscv_sf_vc_xvw_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_XVW_SE);
case Intrinsic::riscv_sf_vc_ivw_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_IVW_SE);
case Intrinsic::riscv_sf_vc_vvw_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_VVW_SE);
case Intrinsic::riscv_sf_vc_fvw_se:
return getVCIXISDNodeVOID(Op, DAG, RISCVISD::SF_VC_FVW_SE);
}
return lowerVectorIntrinsicScalars(Op, DAG, Subtarget);
}
static unsigned getRVVReductionOp(unsigned ISDOpcode) {
switch (ISDOpcode) {
default:
llvm_unreachable("Unhandled reduction");
case ISD::VP_REDUCE_ADD:
case ISD::VECREDUCE_ADD:
return RISCVISD::VECREDUCE_ADD_VL;
case ISD::VP_REDUCE_UMAX:
case ISD::VECREDUCE_UMAX:
return RISCVISD::VECREDUCE_UMAX_VL;
case ISD::VP_REDUCE_SMAX:
case ISD::VECREDUCE_SMAX:
return RISCVISD::VECREDUCE_SMAX_VL;
case ISD::VP_REDUCE_UMIN:
case ISD::VECREDUCE_UMIN:
return RISCVISD::VECREDUCE_UMIN_VL;
case ISD::VP_REDUCE_SMIN:
case ISD::VECREDUCE_SMIN:
return RISCVISD::VECREDUCE_SMIN_VL;
case ISD::VP_REDUCE_AND:
case ISD::VECREDUCE_AND:
return RISCVISD::VECREDUCE_AND_VL;
case ISD::VP_REDUCE_OR:
case ISD::VECREDUCE_OR:
return RISCVISD::VECREDUCE_OR_VL;
case ISD::VP_REDUCE_XOR:
case ISD::VECREDUCE_XOR:
return RISCVISD::VECREDUCE_XOR_VL;
case ISD::VP_REDUCE_FADD:
return RISCVISD::VECREDUCE_FADD_VL;
case ISD::VP_REDUCE_SEQ_FADD:
return RISCVISD::VECREDUCE_SEQ_FADD_VL;
case ISD::VP_REDUCE_FMAX:
case ISD::VP_REDUCE_FMAXIMUM:
return RISCVISD::VECREDUCE_FMAX_VL;
case ISD::VP_REDUCE_FMIN:
case ISD::VP_REDUCE_FMINIMUM:
return RISCVISD::VECREDUCE_FMIN_VL;
}
}
SDValue RISCVTargetLowering::lowerVectorMaskVecReduction(SDValue Op,
SelectionDAG &DAG,
bool IsVP) const {
SDLoc DL(Op);
SDValue Vec = Op.getOperand(IsVP ? 1 : 0);
MVT VecVT = Vec.getSimpleValueType();
assert((Op.getOpcode() == ISD::VECREDUCE_AND ||
Op.getOpcode() == ISD::VECREDUCE_OR ||
Op.getOpcode() == ISD::VECREDUCE_XOR ||
Op.getOpcode() == ISD::VP_REDUCE_AND ||
Op.getOpcode() == ISD::VP_REDUCE_OR ||
Op.getOpcode() == ISD::VP_REDUCE_XOR) &&
"Unexpected reduction lowering");
MVT XLenVT = Subtarget.getXLenVT();
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
SDValue Mask, VL;
if (IsVP) {
Mask = Op.getOperand(2);
VL = Op.getOperand(3);
} else {
std::tie(Mask, VL) =
getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget);
}
ISD::CondCode CC;
switch (Op.getOpcode()) {
default:
llvm_unreachable("Unhandled reduction");
case ISD::VECREDUCE_AND:
case ISD::VP_REDUCE_AND: {
// vcpop ~x == 0
SDValue TrueMask = DAG.getNode(RISCVISD::VMSET_VL, DL, ContainerVT, VL);
if (IsVP || VecVT.isFixedLengthVector())
Vec = DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Vec, TrueMask, VL);
else
Vec = DAG.getNode(ISD::XOR, DL, ContainerVT, Vec, TrueMask);
Vec = DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, Vec, Mask, VL);
CC = ISD::SETEQ;
break;
}
case ISD::VECREDUCE_OR:
case ISD::VP_REDUCE_OR:
// vcpop x != 0
Vec = DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, Vec, Mask, VL);
CC = ISD::SETNE;
break;
case ISD::VECREDUCE_XOR:
case ISD::VP_REDUCE_XOR: {
// ((vcpop x) & 1) != 0
SDValue One = DAG.getConstant(1, DL, XLenVT);
Vec = DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, Vec, Mask, VL);
Vec = DAG.getNode(ISD::AND, DL, XLenVT, Vec, One);
CC = ISD::SETNE;
break;
}
}
SDValue Zero = DAG.getConstant(0, DL, XLenVT);
SDValue SetCC = DAG.getSetCC(DL, XLenVT, Vec, Zero, CC);
SetCC = DAG.getNode(ISD::TRUNCATE, DL, Op.getValueType(), SetCC);
if (!IsVP)
return SetCC;
// Now include the start value in the operation.
// Note that we must return the start value when no elements are operated
// upon. The vcpop instructions we've emitted in each case above will return
// 0 for an inactive vector, and so we've already received the neutral value:
// AND gives us (0 == 0) -> 1 and OR/XOR give us (0 != 0) -> 0. Therefore we
// can simply include the start value.
unsigned BaseOpc = ISD::getVecReduceBaseOpcode(Op.getOpcode());
return DAG.getNode(BaseOpc, DL, Op.getValueType(), SetCC, Op.getOperand(0));
}
static bool isNonZeroAVL(SDValue AVL) {
auto *RegisterAVL = dyn_cast<RegisterSDNode>(AVL);
auto *ImmAVL = dyn_cast<ConstantSDNode>(AVL);
return (RegisterAVL && RegisterAVL->getReg() == RISCV::X0) ||
(ImmAVL && ImmAVL->getZExtValue() >= 1);
}
/// Helper to lower a reduction sequence of the form:
/// scalar = reduce_op vec, scalar_start
static SDValue lowerReductionSeq(unsigned RVVOpcode, MVT ResVT,
SDValue StartValue, SDValue Vec, SDValue Mask,
SDValue VL, const SDLoc &DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
const MVT VecVT = Vec.getSimpleValueType();
const MVT M1VT = RISCVTargetLowering::getM1VT(VecVT);
const MVT XLenVT = Subtarget.getXLenVT();
const bool NonZeroAVL = isNonZeroAVL(VL);
// The reduction needs an LMUL1 input; do the splat at either LMUL1
// or the original VT if fractional.
auto InnerVT = VecVT.bitsLE(M1VT) ? VecVT : M1VT;
// We reuse the VL of the reduction to reduce vsetvli toggles if we can
// prove it is non-zero. For the AVL=0 case, we need the scalar to
// be the result of the reduction operation.
auto InnerVL = NonZeroAVL ? VL : DAG.getConstant(1, DL, XLenVT);
SDValue InitialValue =
lowerScalarInsert(StartValue, InnerVL, InnerVT, DL, DAG, Subtarget);
if (M1VT != InnerVT)
InitialValue =
DAG.getInsertSubvector(DL, DAG.getUNDEF(M1VT), InitialValue, 0);
SDValue PassThru = NonZeroAVL ? DAG.getUNDEF(M1VT) : InitialValue;
SDValue Policy = DAG.getTargetConstant(RISCVVType::TAIL_AGNOSTIC, DL, XLenVT);
SDValue Ops[] = {PassThru, Vec, InitialValue, Mask, VL, Policy};
SDValue Reduction = DAG.getNode(RVVOpcode, DL, M1VT, Ops);
return DAG.getExtractVectorElt(DL, ResVT, Reduction, 0);
}
SDValue RISCVTargetLowering::lowerVECREDUCE(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
SDValue Vec = Op.getOperand(0);
EVT VecEVT = Vec.getValueType();
unsigned BaseOpc = ISD::getVecReduceBaseOpcode(Op.getOpcode());
// Due to ordering in legalize types we may have a vector type that needs to
// be split. Do that manually so we can get down to a legal type.
while (getTypeAction(*DAG.getContext(), VecEVT) ==
TargetLowering::TypeSplitVector) {
auto [Lo, Hi] = DAG.SplitVector(Vec, DL);
VecEVT = Lo.getValueType();
Vec = DAG.getNode(BaseOpc, DL, VecEVT, Lo, Hi);
}
// TODO: The type may need to be widened rather than split. Or widened before
// it can be split.
if (!isTypeLegal(VecEVT))
return SDValue();
MVT VecVT = VecEVT.getSimpleVT();
MVT VecEltVT = VecVT.getVectorElementType();
unsigned RVVOpcode = getRVVReductionOp(Op.getOpcode());
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
auto [Mask, VL] = getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget);
SDValue StartV = DAG.getNeutralElement(BaseOpc, DL, VecEltVT, SDNodeFlags());
switch (BaseOpc) {
case ISD::AND:
case ISD::OR:
case ISD::UMAX:
case ISD::UMIN:
case ISD::SMAX:
case ISD::SMIN:
StartV = DAG.getExtractVectorElt(DL, VecEltVT, Vec, 0);
}
return lowerReductionSeq(RVVOpcode, Op.getSimpleValueType(), StartV, Vec,
Mask, VL, DL, DAG, Subtarget);
}
// Given a reduction op, this function returns the matching reduction opcode,
// the vector SDValue and the scalar SDValue required to lower this to a
// RISCVISD node.
static std::tuple<unsigned, SDValue, SDValue>
getRVVFPReductionOpAndOperands(SDValue Op, SelectionDAG &DAG, EVT EltVT,
const RISCVSubtarget &Subtarget) {
SDLoc DL(Op);
auto Flags = Op->getFlags();
unsigned Opcode = Op.getOpcode();
switch (Opcode) {
default:
llvm_unreachable("Unhandled reduction");
case ISD::VECREDUCE_FADD: {
// Use positive zero if we can. It is cheaper to materialize.
SDValue Zero =
DAG.getConstantFP(Flags.hasNoSignedZeros() ? 0.0 : -0.0, DL, EltVT);
return std::make_tuple(RISCVISD::VECREDUCE_FADD_VL, Op.getOperand(0), Zero);
}
case ISD::VECREDUCE_SEQ_FADD:
return std::make_tuple(RISCVISD::VECREDUCE_SEQ_FADD_VL, Op.getOperand(1),
Op.getOperand(0));
case ISD::VECREDUCE_FMINIMUM:
case ISD::VECREDUCE_FMAXIMUM:
case ISD::VECREDUCE_FMIN:
case ISD::VECREDUCE_FMAX: {
SDValue Front = DAG.getExtractVectorElt(DL, EltVT, Op.getOperand(0), 0);
unsigned RVVOpc =
(Opcode == ISD::VECREDUCE_FMIN || Opcode == ISD::VECREDUCE_FMINIMUM)
? RISCVISD::VECREDUCE_FMIN_VL
: RISCVISD::VECREDUCE_FMAX_VL;
return std::make_tuple(RVVOpc, Op.getOperand(0), Front);
}
}
}
SDValue RISCVTargetLowering::lowerFPVECREDUCE(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VecEltVT = Op.getSimpleValueType();
unsigned RVVOpcode;
SDValue VectorVal, ScalarVal;
std::tie(RVVOpcode, VectorVal, ScalarVal) =
getRVVFPReductionOpAndOperands(Op, DAG, VecEltVT, Subtarget);
MVT VecVT = VectorVal.getSimpleValueType();
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
VectorVal = convertToScalableVector(ContainerVT, VectorVal, DAG, Subtarget);
}
MVT ResVT = Op.getSimpleValueType();
auto [Mask, VL] = getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget);
SDValue Res = lowerReductionSeq(RVVOpcode, ResVT, ScalarVal, VectorVal, Mask,
VL, DL, DAG, Subtarget);
if (Op.getOpcode() != ISD::VECREDUCE_FMINIMUM &&
Op.getOpcode() != ISD::VECREDUCE_FMAXIMUM)
return Res;
if (Op->getFlags().hasNoNaNs())
return Res;
// Force output to NaN if any element is Nan.
SDValue IsNan =
DAG.getNode(RISCVISD::SETCC_VL, DL, Mask.getValueType(),
{VectorVal, VectorVal, DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(Mask.getValueType()), Mask, VL});
MVT XLenVT = Subtarget.getXLenVT();
SDValue CPop = DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, IsNan, Mask, VL);
SDValue NoNaNs = DAG.getSetCC(DL, XLenVT, CPop,
DAG.getConstant(0, DL, XLenVT), ISD::SETEQ);
return DAG.getSelect(
DL, ResVT, NoNaNs, Res,
DAG.getConstantFP(APFloat::getNaN(ResVT.getFltSemantics()), DL, ResVT));
}
SDValue RISCVTargetLowering::lowerVPREDUCE(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
unsigned Opc = Op.getOpcode();
SDValue Start = Op.getOperand(0);
SDValue Vec = Op.getOperand(1);
EVT VecEVT = Vec.getValueType();
MVT XLenVT = Subtarget.getXLenVT();
// TODO: The type may need to be widened rather than split. Or widened before
// it can be split.
if (!isTypeLegal(VecEVT))
return SDValue();
MVT VecVT = VecEVT.getSimpleVT();
unsigned RVVOpcode = getRVVReductionOp(Opc);
if (VecVT.isFixedLengthVector()) {
auto ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
SDValue VL = Op.getOperand(3);
SDValue Mask = Op.getOperand(2);
SDValue Res =
lowerReductionSeq(RVVOpcode, Op.getSimpleValueType(), Op.getOperand(0),
Vec, Mask, VL, DL, DAG, Subtarget);
if ((Opc != ISD::VP_REDUCE_FMINIMUM && Opc != ISD::VP_REDUCE_FMAXIMUM) ||
Op->getFlags().hasNoNaNs())
return Res;
// Propagate NaNs.
MVT PredVT = getMaskTypeFor(Vec.getSimpleValueType());
// Check if any of the elements in Vec is NaN.
SDValue IsNaN = DAG.getNode(
RISCVISD::SETCC_VL, DL, PredVT,
{Vec, Vec, DAG.getCondCode(ISD::SETNE), DAG.getUNDEF(PredVT), Mask, VL});
SDValue VCPop = DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, IsNaN, Mask, VL);
// Check if the start value is NaN.
SDValue StartIsNaN = DAG.getSetCC(DL, XLenVT, Start, Start, ISD::SETUO);
VCPop = DAG.getNode(ISD::OR, DL, XLenVT, VCPop, StartIsNaN);
SDValue NoNaNs = DAG.getSetCC(DL, XLenVT, VCPop,
DAG.getConstant(0, DL, XLenVT), ISD::SETEQ);
MVT ResVT = Res.getSimpleValueType();
return DAG.getSelect(
DL, ResVT, NoNaNs, Res,
DAG.getConstantFP(APFloat::getNaN(ResVT.getFltSemantics()), DL, ResVT));
}
SDValue RISCVTargetLowering::lowerINSERT_SUBVECTOR(SDValue Op,
SelectionDAG &DAG) const {
SDValue Vec = Op.getOperand(0);
SDValue SubVec = Op.getOperand(1);
MVT VecVT = Vec.getSimpleValueType();
MVT SubVecVT = SubVec.getSimpleValueType();
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
unsigned OrigIdx = Op.getConstantOperandVal(2);
const RISCVRegisterInfo *TRI = Subtarget.getRegisterInfo();
if (OrigIdx == 0 && Vec.isUndef())
return Op;
// We don't have the ability to slide mask vectors up indexed by their i1
// elements; the smallest we can do is i8. Often we are able to bitcast to
// equivalent i8 vectors. Note that when inserting a fixed-length vector
// into a scalable one, we might not necessarily have enough scalable
// elements to safely divide by 8: nxv1i1 = insert nxv1i1, v4i1 is valid.
if (SubVecVT.getVectorElementType() == MVT::i1) {
if (VecVT.getVectorMinNumElements() >= 8 &&
SubVecVT.getVectorMinNumElements() >= 8) {
assert(OrigIdx % 8 == 0 && "Invalid index");
assert(VecVT.getVectorMinNumElements() % 8 == 0 &&
SubVecVT.getVectorMinNumElements() % 8 == 0 &&
"Unexpected mask vector lowering");
OrigIdx /= 8;
SubVecVT =
MVT::getVectorVT(MVT::i8, SubVecVT.getVectorMinNumElements() / 8,
SubVecVT.isScalableVector());
VecVT = MVT::getVectorVT(MVT::i8, VecVT.getVectorMinNumElements() / 8,
VecVT.isScalableVector());
Vec = DAG.getBitcast(VecVT, Vec);
SubVec = DAG.getBitcast(SubVecVT, SubVec);
} else {
// We can't slide this mask vector up indexed by its i1 elements.
// This poses a problem when we wish to insert a scalable vector which
// can't be re-expressed as a larger type. Just choose the slow path and
// extend to a larger type, then truncate back down.
MVT ExtVecVT = VecVT.changeVectorElementType(MVT::i8);
MVT ExtSubVecVT = SubVecVT.changeVectorElementType(MVT::i8);
Vec = DAG.getNode(ISD::ZERO_EXTEND, DL, ExtVecVT, Vec);
SubVec = DAG.getNode(ISD::ZERO_EXTEND, DL, ExtSubVecVT, SubVec);
Vec = DAG.getNode(ISD::INSERT_SUBVECTOR, DL, ExtVecVT, Vec, SubVec,
Op.getOperand(2));
SDValue SplatZero = DAG.getConstant(0, DL, ExtVecVT);
return DAG.getSetCC(DL, VecVT, Vec, SplatZero, ISD::SETNE);
}
}
// If the subvector vector is a fixed-length type and we don't know VLEN
// exactly, we cannot use subregister manipulation to simplify the codegen; we
// don't know which register of a LMUL group contains the specific subvector
// as we only know the minimum register size. Therefore we must slide the
// vector group up the full amount.
const auto VLen = Subtarget.getRealVLen();
if (SubVecVT.isFixedLengthVector() && !VLen) {
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
SubVec = DAG.getInsertSubvector(DL, DAG.getUNDEF(ContainerVT), SubVec, 0);
SDValue Mask =
getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget).first;
// Set the vector length to only the number of elements we care about. Note
// that for slideup this includes the offset.
unsigned EndIndex = OrigIdx + SubVecVT.getVectorNumElements();
SDValue VL = DAG.getConstant(EndIndex, DL, XLenVT);
// Use tail agnostic policy if we're inserting over Vec's tail.
unsigned Policy = RISCVVType::TAIL_UNDISTURBED_MASK_UNDISTURBED;
if (VecVT.isFixedLengthVector() && EndIndex == VecVT.getVectorNumElements())
Policy = RISCVVType::TAIL_AGNOSTIC;
// If we're inserting into the lowest elements, use a tail undisturbed
// vmv.v.v.
if (OrigIdx == 0) {
SubVec =
DAG.getNode(RISCVISD::VMV_V_V_VL, DL, ContainerVT, Vec, SubVec, VL);
} else {
SDValue SlideupAmt = DAG.getConstant(OrigIdx, DL, XLenVT);
SubVec = getVSlideup(DAG, Subtarget, DL, ContainerVT, Vec, SubVec,
SlideupAmt, Mask, VL, Policy);
}
if (VecVT.isFixedLengthVector())
SubVec = convertFromScalableVector(VecVT, SubVec, DAG, Subtarget);
return DAG.getBitcast(Op.getValueType(), SubVec);
}
MVT ContainerVecVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVecVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVecVT, Vec, DAG, Subtarget);
}
MVT ContainerSubVecVT = SubVecVT;
if (SubVecVT.isFixedLengthVector()) {
ContainerSubVecVT = getContainerForFixedLengthVector(SubVecVT);
SubVec = convertToScalableVector(ContainerSubVecVT, SubVec, DAG, Subtarget);
}
unsigned SubRegIdx;
ElementCount RemIdx;
// insert_subvector scales the index by vscale if the subvector is scalable,
// and decomposeSubvectorInsertExtractToSubRegs takes this into account. So if
// we have a fixed length subvector, we need to adjust the index by 1/vscale.
if (SubVecVT.isFixedLengthVector()) {
assert(VLen);
unsigned Vscale = *VLen / RISCV::RVVBitsPerBlock;
auto Decompose =
RISCVTargetLowering::decomposeSubvectorInsertExtractToSubRegs(
ContainerVecVT, ContainerSubVecVT, OrigIdx / Vscale, TRI);
SubRegIdx = Decompose.first;
RemIdx = ElementCount::getFixed((Decompose.second * Vscale) +
(OrigIdx % Vscale));
} else {
auto Decompose =
RISCVTargetLowering::decomposeSubvectorInsertExtractToSubRegs(
ContainerVecVT, ContainerSubVecVT, OrigIdx, TRI);
SubRegIdx = Decompose.first;
RemIdx = ElementCount::getScalable(Decompose.second);
}
TypeSize VecRegSize = TypeSize::getScalable(RISCV::RVVBitsPerBlock);
assert(isPowerOf2_64(
Subtarget.expandVScale(SubVecVT.getSizeInBits()).getKnownMinValue()));
bool ExactlyVecRegSized =
Subtarget.expandVScale(SubVecVT.getSizeInBits())
.isKnownMultipleOf(Subtarget.expandVScale(VecRegSize));
// 1. If the Idx has been completely eliminated and this subvector's size is
// a vector register or a multiple thereof, or the surrounding elements are
// undef, then this is a subvector insert which naturally aligns to a vector
// register. These can easily be handled using subregister manipulation.
// 2. If the subvector isn't an exact multiple of a valid register group size,
// then the insertion must preserve the undisturbed elements of the register.
// We do this by lowering to an EXTRACT_SUBVECTOR grabbing the nearest LMUL=1
// vector type (which resolves to a subregister copy), performing a VSLIDEUP
// to place the subvector within the vector register, and an INSERT_SUBVECTOR
// of that LMUL=1 type back into the larger vector (resolving to another
// subregister operation). See below for how our VSLIDEUP works. We go via a
// LMUL=1 type to avoid allocating a large register group to hold our
// subvector.
if (RemIdx.isZero() && (ExactlyVecRegSized || Vec.isUndef())) {
if (SubVecVT.isFixedLengthVector()) {
// We may get NoSubRegister if inserting at index 0 and the subvec
// container is the same as the vector, e.g. vec=v4i32,subvec=v4i32,idx=0
if (SubRegIdx == RISCV::NoSubRegister) {
assert(OrigIdx == 0);
return Op;
}
// Use a insert_subvector that will resolve to an insert subreg.
assert(VLen);
unsigned Vscale = *VLen / RISCV::RVVBitsPerBlock;
SDValue Insert =
DAG.getInsertSubvector(DL, Vec, SubVec, OrigIdx / Vscale);
if (VecVT.isFixedLengthVector())
Insert = convertFromScalableVector(VecVT, Insert, DAG, Subtarget);
return Insert;
}
return Op;
}
// VSLIDEUP works by leaving elements 0<i<OFFSET undisturbed, elements
// OFFSET<=i<VL set to the "subvector" and vl<=i<VLMAX set to the tail policy
// (in our case undisturbed). This means we can set up a subvector insertion
// where OFFSET is the insertion offset, and the VL is the OFFSET plus the
// size of the subvector.
MVT InterSubVT = ContainerVecVT;
SDValue AlignedExtract = Vec;
unsigned AlignedIdx = OrigIdx - RemIdx.getKnownMinValue();
if (SubVecVT.isFixedLengthVector()) {
assert(VLen);
AlignedIdx /= *VLen / RISCV::RVVBitsPerBlock;
}
if (ContainerVecVT.bitsGT(RISCVTargetLowering::getM1VT(ContainerVecVT))) {
InterSubVT = RISCVTargetLowering::getM1VT(ContainerVecVT);
// Extract a subvector equal to the nearest full vector register type. This
// should resolve to a EXTRACT_SUBREG instruction.
AlignedExtract = DAG.getExtractSubvector(DL, InterSubVT, Vec, AlignedIdx);
}
SubVec = DAG.getInsertSubvector(DL, DAG.getUNDEF(InterSubVT), SubVec, 0);
auto [Mask, VL] = getDefaultVLOps(VecVT, ContainerVecVT, DL, DAG, Subtarget);
ElementCount EndIndex = RemIdx + SubVecVT.getVectorElementCount();
VL = DAG.getElementCount(DL, XLenVT, SubVecVT.getVectorElementCount());
// Use tail agnostic policy if we're inserting over InterSubVT's tail.
unsigned Policy = RISCVVType::TAIL_UNDISTURBED_MASK_UNDISTURBED;
if (Subtarget.expandVScale(EndIndex) ==
Subtarget.expandVScale(InterSubVT.getVectorElementCount()))
Policy = RISCVVType::TAIL_AGNOSTIC;
// If we're inserting into the lowest elements, use a tail undisturbed
// vmv.v.v.
if (RemIdx.isZero()) {
SubVec = DAG.getNode(RISCVISD::VMV_V_V_VL, DL, InterSubVT, AlignedExtract,
SubVec, VL);
} else {
SDValue SlideupAmt = DAG.getElementCount(DL, XLenVT, RemIdx);
// Construct the vector length corresponding to RemIdx + length(SubVecVT).
VL = DAG.getNode(ISD::ADD, DL, XLenVT, SlideupAmt, VL);
SubVec = getVSlideup(DAG, Subtarget, DL, InterSubVT, AlignedExtract, SubVec,
SlideupAmt, Mask, VL, Policy);
}
// If required, insert this subvector back into the correct vector register.
// This should resolve to an INSERT_SUBREG instruction.
if (ContainerVecVT.bitsGT(InterSubVT))
SubVec = DAG.getInsertSubvector(DL, Vec, SubVec, AlignedIdx);
if (VecVT.isFixedLengthVector())
SubVec = convertFromScalableVector(VecVT, SubVec, DAG, Subtarget);
// We might have bitcast from a mask type: cast back to the original type if
// required.
return DAG.getBitcast(Op.getSimpleValueType(), SubVec);
}
SDValue RISCVTargetLowering::lowerEXTRACT_SUBVECTOR(SDValue Op,
SelectionDAG &DAG) const {
SDValue Vec = Op.getOperand(0);
MVT SubVecVT = Op.getSimpleValueType();
MVT VecVT = Vec.getSimpleValueType();
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
unsigned OrigIdx = Op.getConstantOperandVal(1);
const RISCVRegisterInfo *TRI = Subtarget.getRegisterInfo();
// With an index of 0 this is a cast-like subvector, which can be performed
// with subregister operations.
if (OrigIdx == 0)
return Op;
// We don't have the ability to slide mask vectors down indexed by their i1
// elements; the smallest we can do is i8. Often we are able to bitcast to
// equivalent i8 vectors. Note that when extracting a fixed-length vector
// from a scalable one, we might not necessarily have enough scalable
// elements to safely divide by 8: v8i1 = extract nxv1i1 is valid.
if (SubVecVT.getVectorElementType() == MVT::i1) {
if (VecVT.getVectorMinNumElements() >= 8 &&
SubVecVT.getVectorMinNumElements() >= 8) {
assert(OrigIdx % 8 == 0 && "Invalid index");
assert(VecVT.getVectorMinNumElements() % 8 == 0 &&
SubVecVT.getVectorMinNumElements() % 8 == 0 &&
"Unexpected mask vector lowering");
OrigIdx /= 8;
SubVecVT =
MVT::getVectorVT(MVT::i8, SubVecVT.getVectorMinNumElements() / 8,
SubVecVT.isScalableVector());
VecVT = MVT::getVectorVT(MVT::i8, VecVT.getVectorMinNumElements() / 8,
VecVT.isScalableVector());
Vec = DAG.getBitcast(VecVT, Vec);
} else {
// We can't slide this mask vector down, indexed by its i1 elements.
// This poses a problem when we wish to extract a scalable vector which
// can't be re-expressed as a larger type. Just choose the slow path and
// extend to a larger type, then truncate back down.
// TODO: We could probably improve this when extracting certain fixed
// from fixed, where we can extract as i8 and shift the correct element
// right to reach the desired subvector?
MVT ExtVecVT = VecVT.changeVectorElementType(MVT::i8);
MVT ExtSubVecVT = SubVecVT.changeVectorElementType(MVT::i8);
Vec = DAG.getNode(ISD::ZERO_EXTEND, DL, ExtVecVT, Vec);
Vec = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, ExtSubVecVT, Vec,
Op.getOperand(1));
SDValue SplatZero = DAG.getConstant(0, DL, ExtSubVecVT);
return DAG.getSetCC(DL, SubVecVT, Vec, SplatZero, ISD::SETNE);
}
}
const auto VLen = Subtarget.getRealVLen();
// If the subvector vector is a fixed-length type and we don't know VLEN
// exactly, we cannot use subregister manipulation to simplify the codegen; we
// don't know which register of a LMUL group contains the specific subvector
// as we only know the minimum register size. Therefore we must slide the
// vector group down the full amount.
if (SubVecVT.isFixedLengthVector() && !VLen) {
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
// Shrink down Vec so we're performing the slidedown on a smaller LMUL.
unsigned LastIdx = OrigIdx + SubVecVT.getVectorNumElements() - 1;
if (auto ShrunkVT =
getSmallestVTForIndex(ContainerVT, LastIdx, DL, DAG, Subtarget)) {
ContainerVT = *ShrunkVT;
Vec = DAG.getExtractSubvector(DL, ContainerVT, Vec, 0);
}
SDValue Mask =
getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget).first;
// Set the vector length to only the number of elements we care about. This
// avoids sliding down elements we're going to discard straight away.
SDValue VL = DAG.getConstant(SubVecVT.getVectorNumElements(), DL, XLenVT);
SDValue SlidedownAmt = DAG.getConstant(OrigIdx, DL, XLenVT);
SDValue Slidedown =
getVSlidedown(DAG, Subtarget, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), Vec, SlidedownAmt, Mask, VL);
// Now we can use a cast-like subvector extract to get the result.
Slidedown = DAG.getExtractSubvector(DL, SubVecVT, Slidedown, 0);
return DAG.getBitcast(Op.getValueType(), Slidedown);
}
if (VecVT.isFixedLengthVector()) {
VecVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(VecVT, Vec, DAG, Subtarget);
}
MVT ContainerSubVecVT = SubVecVT;
if (SubVecVT.isFixedLengthVector())
ContainerSubVecVT = getContainerForFixedLengthVector(SubVecVT);
unsigned SubRegIdx;
ElementCount RemIdx;
// extract_subvector scales the index by vscale if the subvector is scalable,
// and decomposeSubvectorInsertExtractToSubRegs takes this into account. So if
// we have a fixed length subvector, we need to adjust the index by 1/vscale.
if (SubVecVT.isFixedLengthVector()) {
assert(VLen);
unsigned Vscale = *VLen / RISCV::RVVBitsPerBlock;
auto Decompose =
RISCVTargetLowering::decomposeSubvectorInsertExtractToSubRegs(
VecVT, ContainerSubVecVT, OrigIdx / Vscale, TRI);
SubRegIdx = Decompose.first;
RemIdx = ElementCount::getFixed((Decompose.second * Vscale) +
(OrigIdx % Vscale));
} else {
auto Decompose =
RISCVTargetLowering::decomposeSubvectorInsertExtractToSubRegs(
VecVT, ContainerSubVecVT, OrigIdx, TRI);
SubRegIdx = Decompose.first;
RemIdx = ElementCount::getScalable(Decompose.second);
}
// If the Idx has been completely eliminated then this is a subvector extract
// which naturally aligns to a vector register. These can easily be handled
// using subregister manipulation. We use an extract_subvector that will
// resolve to an extract subreg.
if (RemIdx.isZero()) {
if (SubVecVT.isFixedLengthVector()) {
assert(VLen);
unsigned Vscale = *VLen / RISCV::RVVBitsPerBlock;
Vec =
DAG.getExtractSubvector(DL, ContainerSubVecVT, Vec, OrigIdx / Vscale);
return convertFromScalableVector(SubVecVT, Vec, DAG, Subtarget);
}
return Op;
}
// Else SubVecVT is M1 or smaller and may need to be slid down: if SubVecVT
// was > M1 then the index would need to be a multiple of VLMAX, and so would
// divide exactly.
assert(RISCVVType::decodeVLMUL(getLMUL(ContainerSubVecVT)).second ||
getLMUL(ContainerSubVecVT) == RISCVVType::LMUL_1);
// If the vector type is an LMUL-group type, extract a subvector equal to the
// nearest full vector register type.
MVT InterSubVT = VecVT;
if (VecVT.bitsGT(RISCVTargetLowering::getM1VT(VecVT))) {
// If VecVT has an LMUL > 1, then SubVecVT should have a smaller LMUL, and
// we should have successfully decomposed the extract into a subregister.
// We use an extract_subvector that will resolve to a subreg extract.
assert(SubRegIdx != RISCV::NoSubRegister);
(void)SubRegIdx;
unsigned Idx = OrigIdx - RemIdx.getKnownMinValue();
if (SubVecVT.isFixedLengthVector()) {
assert(VLen);
Idx /= *VLen / RISCV::RVVBitsPerBlock;
}
InterSubVT = RISCVTargetLowering::getM1VT(VecVT);
Vec = DAG.getExtractSubvector(DL, InterSubVT, Vec, Idx);
}
// Slide this vector register down by the desired number of elements in order
// to place the desired subvector starting at element 0.
SDValue SlidedownAmt = DAG.getElementCount(DL, XLenVT, RemIdx);
auto [Mask, VL] = getDefaultScalableVLOps(InterSubVT, DL, DAG, Subtarget);
if (SubVecVT.isFixedLengthVector())
VL = DAG.getConstant(SubVecVT.getVectorNumElements(), DL, XLenVT);
SDValue Slidedown =
getVSlidedown(DAG, Subtarget, DL, InterSubVT, DAG.getUNDEF(InterSubVT),
Vec, SlidedownAmt, Mask, VL);
// Now the vector is in the right position, extract our final subvector. This
// should resolve to a COPY.
Slidedown = DAG.getExtractSubvector(DL, SubVecVT, Slidedown, 0);
// We might have bitcast from a mask type: cast back to the original type if
// required.
return DAG.getBitcast(Op.getSimpleValueType(), Slidedown);
}
// Widen a vector's operands to i8, then truncate its results back to the
// original type, typically i1. All operand and result types must be the same.
static SDValue widenVectorOpsToi8(SDValue N, const SDLoc &DL,
SelectionDAG &DAG) {
MVT VT = N.getSimpleValueType();
MVT WideVT = VT.changeVectorElementType(MVT::i8);
SmallVector<SDValue, 4> WideOps;
for (SDValue Op : N->ops()) {
assert(Op.getSimpleValueType() == VT &&
"Operands and result must be same type");
WideOps.push_back(DAG.getNode(ISD::ZERO_EXTEND, DL, WideVT, Op));
}
unsigned NumVals = N->getNumValues();
SDVTList VTs = DAG.getVTList(SmallVector<EVT, 4>(
NumVals, N.getValueType().changeVectorElementType(MVT::i8)));
SDValue WideN = DAG.getNode(N.getOpcode(), DL, VTs, WideOps);
SmallVector<SDValue, 4> TruncVals;
for (unsigned I = 0; I < NumVals; I++) {
TruncVals.push_back(
DAG.getSetCC(DL, N->getSimpleValueType(I), WideN.getValue(I),
DAG.getConstant(0, DL, WideVT), ISD::SETNE));
}
if (TruncVals.size() > 1)
return DAG.getMergeValues(TruncVals, DL);
return TruncVals.front();
}
SDValue RISCVTargetLowering::lowerVECTOR_DEINTERLEAVE(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VecVT = Op.getSimpleValueType();
const unsigned Factor = Op->getNumValues();
assert(Factor <= 8);
// 1 bit element vectors need to be widened to e8
if (VecVT.getVectorElementType() == MVT::i1)
return widenVectorOpsToi8(Op, DL, DAG);
// Convert to scalable vectors first.
if (VecVT.isFixedLengthVector()) {
MVT ContainerVT = getContainerForFixedLengthVector(VecVT);
SmallVector<SDValue, 8> Ops(Factor);
for (unsigned i = 0U; i < Factor; ++i)
Ops[i] = convertToScalableVector(ContainerVT, Op.getOperand(i), DAG,
Subtarget);
SmallVector<EVT, 8> VTs(Factor, ContainerVT);
SDValue NewDeinterleave =
DAG.getNode(ISD::VECTOR_DEINTERLEAVE, DL, VTs, Ops);
SmallVector<SDValue, 8> Res(Factor);
for (unsigned i = 0U; i < Factor; ++i)
Res[i] = convertFromScalableVector(VecVT, NewDeinterleave.getValue(i),
DAG, Subtarget);
return DAG.getMergeValues(Res, DL);
}
// If concatenating would exceed LMUL=8, we need to split.
if ((VecVT.getSizeInBits().getKnownMinValue() * Factor) >
(8 * RISCV::RVVBitsPerBlock)) {
SmallVector<SDValue, 8> Ops(Factor * 2);
for (unsigned i = 0; i != Factor; ++i) {
auto [OpLo, OpHi] = DAG.SplitVectorOperand(Op.getNode(), i);
Ops[i * 2] = OpLo;
Ops[i * 2 + 1] = OpHi;
}
SmallVector<EVT, 8> VTs(Factor, Ops[0].getValueType());
SDValue Lo = DAG.getNode(ISD::VECTOR_DEINTERLEAVE, DL, VTs,
ArrayRef(Ops).slice(0, Factor));
SDValue Hi = DAG.getNode(ISD::VECTOR_DEINTERLEAVE, DL, VTs,
ArrayRef(Ops).slice(Factor, Factor));
SmallVector<SDValue, 8> Res(Factor);
for (unsigned i = 0; i != Factor; ++i)
Res[i] = DAG.getNode(ISD::CONCAT_VECTORS, DL, VecVT, Lo.getValue(i),
Hi.getValue(i));
return DAG.getMergeValues(Res, DL);
}
if (Subtarget.hasVendorXRivosVizip() && Factor == 2) {
MVT VT = Op->getSimpleValueType(0);
SDValue V1 = Op->getOperand(0);
SDValue V2 = Op->getOperand(1);
// For fractional LMUL, check if we can use a higher LMUL
// instruction to avoid a vslidedown.
if (SDValue Src = foldConcatVector(V1, V2);
Src && RISCVTargetLowering::getM1VT(VT).bitsGT(VT)) {
EVT NewVT = VT.getDoubleNumVectorElementsVT();
Src = DAG.getExtractSubvector(DL, NewVT, Src, 0);
// Freeze the source so we can increase its use count.
Src = DAG.getFreeze(Src);
SDValue Even = lowerVZIP(RISCVISD::RI_VUNZIP2A_VL, Src,
DAG.getUNDEF(NewVT), DL, DAG, Subtarget);
SDValue Odd = lowerVZIP(RISCVISD::RI_VUNZIP2B_VL, Src,
DAG.getUNDEF(NewVT), DL, DAG, Subtarget);
Even = DAG.getExtractSubvector(DL, VT, Even, 0);
Odd = DAG.getExtractSubvector(DL, VT, Odd, 0);
return DAG.getMergeValues({Even, Odd}, DL);
}
// Freeze the sources so we can increase their use count.
V1 = DAG.getFreeze(V1);
V2 = DAG.getFreeze(V2);
SDValue Even =
lowerVZIP(RISCVISD::RI_VUNZIP2A_VL, V1, V2, DL, DAG, Subtarget);
SDValue Odd =
lowerVZIP(RISCVISD::RI_VUNZIP2B_VL, V1, V2, DL, DAG, Subtarget);
return DAG.getMergeValues({Even, Odd}, DL);
}
SmallVector<SDValue, 8> Ops(Op->op_values());
// Concatenate the vectors as one vector to deinterleave
MVT ConcatVT =
MVT::getVectorVT(VecVT.getVectorElementType(),
VecVT.getVectorElementCount().multiplyCoefficientBy(
PowerOf2Ceil(Factor)));
if (Ops.size() < PowerOf2Ceil(Factor))
Ops.append(PowerOf2Ceil(Factor) - Factor, DAG.getUNDEF(VecVT));
SDValue Concat = DAG.getNode(ISD::CONCAT_VECTORS, DL, ConcatVT, Ops);
if (Factor == 2) {
// We can deinterleave through vnsrl.wi if the element type is smaller than
// ELEN
if (VecVT.getScalarSizeInBits() < Subtarget.getELen()) {
SDValue Even = getDeinterleaveShiftAndTrunc(DL, VecVT, Concat, 2, 0, DAG);
SDValue Odd = getDeinterleaveShiftAndTrunc(DL, VecVT, Concat, 2, 1, DAG);
return DAG.getMergeValues({Even, Odd}, DL);
}
// For the indices, use the vmv.v.x of an i8 constant to fill the largest
// possibly mask vector, then extract the required subvector. Doing this
// (instead of a vid, vmsne sequence) reduces LMUL, and allows the mask
// creation to be rematerialized during register allocation to reduce
// register pressure if needed.
MVT MaskVT = ConcatVT.changeVectorElementType(MVT::i1);
SDValue EvenSplat = DAG.getConstant(0b01010101, DL, MVT::nxv8i8);
EvenSplat = DAG.getBitcast(MVT::nxv64i1, EvenSplat);
SDValue EvenMask = DAG.getExtractSubvector(DL, MaskVT, EvenSplat, 0);
SDValue OddSplat = DAG.getConstant(0b10101010, DL, MVT::nxv8i8);
OddSplat = DAG.getBitcast(MVT::nxv64i1, OddSplat);
SDValue OddMask = DAG.getExtractSubvector(DL, MaskVT, OddSplat, 0);
// vcompress the even and odd elements into two separate vectors
SDValue EvenWide = DAG.getNode(ISD::VECTOR_COMPRESS, DL, ConcatVT, Concat,
EvenMask, DAG.getUNDEF(ConcatVT));
SDValue OddWide = DAG.getNode(ISD::VECTOR_COMPRESS, DL, ConcatVT, Concat,
OddMask, DAG.getUNDEF(ConcatVT));
// Extract the result half of the gather for even and odd
SDValue Even = DAG.getExtractSubvector(DL, VecVT, EvenWide, 0);
SDValue Odd = DAG.getExtractSubvector(DL, VecVT, OddWide, 0);
return DAG.getMergeValues({Even, Odd}, DL);
}
// Store with unit-stride store and load it back with segmented load.
MVT XLenVT = Subtarget.getXLenVT();
auto [Mask, VL] = getDefaultScalableVLOps(VecVT, DL, DAG, Subtarget);
SDValue Passthru = DAG.getUNDEF(ConcatVT);
// Allocate a stack slot.
Align Alignment = DAG.getReducedAlign(VecVT, /*UseABI=*/false);
SDValue StackPtr =
DAG.CreateStackTemporary(ConcatVT.getStoreSize(), Alignment);
auto &MF = DAG.getMachineFunction();
auto FrameIndex = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
auto PtrInfo = MachinePointerInfo::getFixedStack(MF, FrameIndex);
SDValue StoreOps[] = {DAG.getEntryNode(),
DAG.getTargetConstant(Intrinsic::riscv_vse, DL, XLenVT),
Concat, StackPtr, VL};
SDValue Chain = DAG.getMemIntrinsicNode(
ISD::INTRINSIC_VOID, DL, DAG.getVTList(MVT::Other), StoreOps,
ConcatVT.getVectorElementType(), PtrInfo, Alignment,
MachineMemOperand::MOStore, LocationSize::beforeOrAfterPointer());
static const Intrinsic::ID VlsegIntrinsicsIds[] = {
Intrinsic::riscv_vlseg2_mask, Intrinsic::riscv_vlseg3_mask,
Intrinsic::riscv_vlseg4_mask, Intrinsic::riscv_vlseg5_mask,
Intrinsic::riscv_vlseg6_mask, Intrinsic::riscv_vlseg7_mask,
Intrinsic::riscv_vlseg8_mask};
SDValue LoadOps[] = {
Chain,
DAG.getTargetConstant(VlsegIntrinsicsIds[Factor - 2], DL, XLenVT),
Passthru,
StackPtr,
Mask,
VL,
DAG.getTargetConstant(
RISCVVType::TAIL_AGNOSTIC | RISCVVType::MASK_AGNOSTIC, DL, XLenVT),
DAG.getTargetConstant(Log2_64(VecVT.getScalarSizeInBits()), DL, XLenVT)};
unsigned Sz =
Factor * VecVT.getVectorMinNumElements() * VecVT.getScalarSizeInBits();
EVT VecTupTy = MVT::getRISCVVectorTupleVT(Sz, Factor);
SDValue Load = DAG.getMemIntrinsicNode(
ISD::INTRINSIC_W_CHAIN, DL, DAG.getVTList({VecTupTy, MVT::Other}),
LoadOps, ConcatVT.getVectorElementType(), PtrInfo, Alignment,
MachineMemOperand::MOLoad, LocationSize::beforeOrAfterPointer());
SmallVector<SDValue, 8> Res(Factor);
for (unsigned i = 0U; i < Factor; ++i)
Res[i] = DAG.getNode(RISCVISD::TUPLE_EXTRACT, DL, VecVT, Load,
DAG.getTargetConstant(i, DL, MVT::i32));
return DAG.getMergeValues(Res, DL);
}
SDValue RISCVTargetLowering::lowerVECTOR_INTERLEAVE(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VecVT = Op.getSimpleValueType();
const unsigned Factor = Op.getNumOperands();
assert(Factor <= 8);
// i1 vectors need to be widened to i8
if (VecVT.getVectorElementType() == MVT::i1)
return widenVectorOpsToi8(Op, DL, DAG);
// Convert to scalable vectors first.
if (VecVT.isFixedLengthVector()) {
MVT ContainerVT = getContainerForFixedLengthVector(VecVT);
SmallVector<SDValue, 8> Ops(Factor);
for (unsigned i = 0U; i < Factor; ++i)
Ops[i] = convertToScalableVector(ContainerVT, Op.getOperand(i), DAG,
Subtarget);
SmallVector<EVT, 8> VTs(Factor, ContainerVT);
SDValue NewInterleave = DAG.getNode(ISD::VECTOR_INTERLEAVE, DL, VTs, Ops);
SmallVector<SDValue, 8> Res(Factor);
for (unsigned i = 0U; i < Factor; ++i)
Res[i] = convertFromScalableVector(VecVT, NewInterleave.getValue(i), DAG,
Subtarget);
return DAG.getMergeValues(Res, DL);
}
MVT XLenVT = Subtarget.getXLenVT();
auto [Mask, VL] = getDefaultScalableVLOps(VecVT, DL, DAG, Subtarget);
// If the VT is larger than LMUL=8, we need to split and reassemble.
if ((VecVT.getSizeInBits().getKnownMinValue() * Factor) >
(8 * RISCV::RVVBitsPerBlock)) {
SmallVector<SDValue, 8> Ops(Factor * 2);
for (unsigned i = 0; i != Factor; ++i) {
auto [OpLo, OpHi] = DAG.SplitVectorOperand(Op.getNode(), i);
Ops[i] = OpLo;
Ops[i + Factor] = OpHi;
}
SmallVector<EVT, 8> VTs(Factor, Ops[0].getValueType());
SDValue Res[] = {DAG.getNode(ISD::VECTOR_INTERLEAVE, DL, VTs,
ArrayRef(Ops).take_front(Factor)),
DAG.getNode(ISD::VECTOR_INTERLEAVE, DL, VTs,
ArrayRef(Ops).drop_front(Factor))};
SmallVector<SDValue, 8> Concats(Factor);
for (unsigned i = 0; i != Factor; ++i) {
unsigned IdxLo = 2 * i;
unsigned IdxHi = 2 * i + 1;
Concats[i] = DAG.getNode(ISD::CONCAT_VECTORS, DL, VecVT,
Res[IdxLo / Factor].getValue(IdxLo % Factor),
Res[IdxHi / Factor].getValue(IdxHi % Factor));
}
return DAG.getMergeValues(Concats, DL);
}
SDValue Interleaved;
// Spill to the stack using a segment store for simplicity.
if (Factor != 2) {
EVT MemVT =
EVT::getVectorVT(*DAG.getContext(), VecVT.getVectorElementType(),
VecVT.getVectorElementCount() * Factor);
// Allocate a stack slot.
Align Alignment = DAG.getReducedAlign(VecVT, /*UseABI=*/false);
SDValue StackPtr =
DAG.CreateStackTemporary(MemVT.getStoreSize(), Alignment);
EVT PtrVT = StackPtr.getValueType();
auto &MF = DAG.getMachineFunction();
auto FrameIndex = cast<FrameIndexSDNode>(StackPtr.getNode())->getIndex();
auto PtrInfo = MachinePointerInfo::getFixedStack(MF, FrameIndex);
static const Intrinsic::ID IntrIds[] = {
Intrinsic::riscv_vsseg2_mask, Intrinsic::riscv_vsseg3_mask,
Intrinsic::riscv_vsseg4_mask, Intrinsic::riscv_vsseg5_mask,
Intrinsic::riscv_vsseg6_mask, Intrinsic::riscv_vsseg7_mask,
Intrinsic::riscv_vsseg8_mask,
};
unsigned Sz =
Factor * VecVT.getVectorMinNumElements() * VecVT.getScalarSizeInBits();
EVT VecTupTy = MVT::getRISCVVectorTupleVT(Sz, Factor);
SDValue StoredVal = DAG.getUNDEF(VecTupTy);
for (unsigned i = 0; i < Factor; i++)
StoredVal =
DAG.getNode(RISCVISD::TUPLE_INSERT, DL, VecTupTy, StoredVal,
Op.getOperand(i), DAG.getTargetConstant(i, DL, MVT::i32));
SDValue Ops[] = {DAG.getEntryNode(),
DAG.getTargetConstant(IntrIds[Factor - 2], DL, XLenVT),
StoredVal,
StackPtr,
Mask,
VL,
DAG.getTargetConstant(Log2_64(VecVT.getScalarSizeInBits()),
DL, XLenVT)};
SDValue Chain = DAG.getMemIntrinsicNode(
ISD::INTRINSIC_VOID, DL, DAG.getVTList(MVT::Other), Ops,
VecVT.getVectorElementType(), PtrInfo, Alignment,
MachineMemOperand::MOStore, LocationSize::beforeOrAfterPointer());
SmallVector<SDValue, 8> Loads(Factor);
SDValue Increment =
DAG.getVScale(DL, PtrVT,
APInt(PtrVT.getFixedSizeInBits(),
VecVT.getStoreSize().getKnownMinValue()));
for (unsigned i = 0; i != Factor; ++i) {
if (i != 0)
StackPtr = DAG.getNode(ISD::ADD, DL, PtrVT, StackPtr, Increment);
Loads[i] = DAG.getLoad(VecVT, DL, Chain, StackPtr, PtrInfo);
}
return DAG.getMergeValues(Loads, DL);
}
// Use ri.vzip2{a,b} if available
// TODO: Figure out the best lowering for the spread variants
if (Subtarget.hasVendorXRivosVizip() && !Op.getOperand(0).isUndef() &&
!Op.getOperand(1).isUndef()) {
// Freeze the sources so we can increase their use count.
SDValue V1 = DAG.getFreeze(Op->getOperand(0));
SDValue V2 = DAG.getFreeze(Op->getOperand(1));
SDValue Lo = lowerVZIP(RISCVISD::RI_VZIP2A_VL, V1, V2, DL, DAG, Subtarget);
SDValue Hi = lowerVZIP(RISCVISD::RI_VZIP2B_VL, V1, V2, DL, DAG, Subtarget);
return DAG.getMergeValues({Lo, Hi}, DL);
}
// If the element type is smaller than ELEN, then we can interleave with
// vwaddu.vv and vwmaccu.vx
if (VecVT.getScalarSizeInBits() < Subtarget.getELen()) {
Interleaved = getWideningInterleave(Op.getOperand(0), Op.getOperand(1), DL,
DAG, Subtarget);
} else {
// Otherwise, fallback to using vrgathere16.vv
MVT ConcatVT =
MVT::getVectorVT(VecVT.getVectorElementType(),
VecVT.getVectorElementCount().multiplyCoefficientBy(2));
SDValue Concat = DAG.getNode(ISD::CONCAT_VECTORS, DL, ConcatVT,
Op.getOperand(0), Op.getOperand(1));
MVT IdxVT = ConcatVT.changeVectorElementType(MVT::i16);
// 0 1 2 3 4 5 6 7 ...
SDValue StepVec = DAG.getStepVector(DL, IdxVT);
// 1 1 1 1 1 1 1 1 ...
SDValue Ones = DAG.getSplatVector(IdxVT, DL, DAG.getConstant(1, DL, XLenVT));
// 1 0 1 0 1 0 1 0 ...
SDValue OddMask = DAG.getNode(ISD::AND, DL, IdxVT, StepVec, Ones);
OddMask = DAG.getSetCC(
DL, IdxVT.changeVectorElementType(MVT::i1), OddMask,
DAG.getSplatVector(IdxVT, DL, DAG.getConstant(0, DL, XLenVT)),
ISD::CondCode::SETNE);
SDValue VLMax = DAG.getSplatVector(IdxVT, DL, computeVLMax(VecVT, DL, DAG));
// Build up the index vector for interleaving the concatenated vector
// 0 0 1 1 2 2 3 3 ...
SDValue Idx = DAG.getNode(ISD::SRL, DL, IdxVT, StepVec, Ones);
// 0 n 1 n+1 2 n+2 3 n+3 ...
Idx =
DAG.getNode(RISCVISD::ADD_VL, DL, IdxVT, Idx, VLMax, Idx, OddMask, VL);
// Then perform the interleave
// v[0] v[n] v[1] v[n+1] v[2] v[n+2] v[3] v[n+3] ...
SDValue TrueMask = getAllOnesMask(IdxVT, VL, DL, DAG);
Interleaved = DAG.getNode(RISCVISD::VRGATHEREI16_VV_VL, DL, ConcatVT,
Concat, Idx, DAG.getUNDEF(ConcatVT), TrueMask, VL);
}
// Extract the two halves from the interleaved result
SDValue Lo = DAG.getExtractSubvector(DL, VecVT, Interleaved, 0);
SDValue Hi = DAG.getExtractSubvector(DL, VecVT, Interleaved,
VecVT.getVectorMinNumElements());
return DAG.getMergeValues({Lo, Hi}, DL);
}
// Lower step_vector to the vid instruction. Any non-identity step value must
// be accounted for my manual expansion.
SDValue RISCVTargetLowering::lowerSTEP_VECTOR(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
assert(VT.isScalableVector() && "Expected scalable vector");
MVT XLenVT = Subtarget.getXLenVT();
auto [Mask, VL] = getDefaultScalableVLOps(VT, DL, DAG, Subtarget);
SDValue StepVec = DAG.getNode(RISCVISD::VID_VL, DL, VT, Mask, VL);
uint64_t StepValImm = Op.getConstantOperandVal(0);
if (StepValImm != 1) {
if (isPowerOf2_64(StepValImm)) {
SDValue StepVal =
DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, DAG.getUNDEF(VT),
DAG.getConstant(Log2_64(StepValImm), DL, XLenVT), VL);
StepVec = DAG.getNode(ISD::SHL, DL, VT, StepVec, StepVal);
} else {
SDValue StepVal = lowerScalarSplat(
SDValue(), DAG.getConstant(StepValImm, DL, VT.getVectorElementType()),
VL, VT, DL, DAG, Subtarget);
StepVec = DAG.getNode(ISD::MUL, DL, VT, StepVec, StepVal);
}
}
return StepVec;
}
// Implement vector_reverse using vrgather.vv with indices determined by
// subtracting the id of each element from (VLMAX-1). This will convert
// the indices like so:
// (0, 1,..., VLMAX-2, VLMAX-1) -> (VLMAX-1, VLMAX-2,..., 1, 0).
// TODO: This code assumes VLMAX <= 65536 for LMUL=8 SEW=16.
SDValue RISCVTargetLowering::lowerVECTOR_REVERSE(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VecVT = Op.getSimpleValueType();
if (VecVT.getVectorElementType() == MVT::i1) {
MVT WidenVT = MVT::getVectorVT(MVT::i8, VecVT.getVectorElementCount());
SDValue Op1 = DAG.getNode(ISD::ZERO_EXTEND, DL, WidenVT, Op.getOperand(0));
SDValue Op2 = DAG.getNode(ISD::VECTOR_REVERSE, DL, WidenVT, Op1);
return DAG.getSetCC(DL, VecVT, Op2,
DAG.getConstant(0, DL, Op2.getValueType()), ISD::SETNE);
}
MVT ContainerVT = VecVT;
SDValue Vec = Op.getOperand(0);
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
MVT XLenVT = Subtarget.getXLenVT();
auto [Mask, VL] = getDefaultVLOps(VecVT, ContainerVT, DL, DAG, Subtarget);
// On some uarchs vrgather.vv will read from every input register for each
// output register, regardless of the indices. However to reverse a vector
// each output register only needs to read from one register. So decompose it
// into LMUL * M1 vrgather.vvs, so we get O(LMUL) performance instead of
// O(LMUL^2).
//
// vsetvli a1, zero, e64, m4, ta, ma
// vrgatherei16.vv v12, v8, v16
// ->
// vsetvli a1, zero, e64, m1, ta, ma
// vrgather.vv v15, v8, v16
// vrgather.vv v14, v9, v16
// vrgather.vv v13, v10, v16
// vrgather.vv v12, v11, v16
if (ContainerVT.bitsGT(RISCVTargetLowering::getM1VT(ContainerVT)) &&
ContainerVT.getVectorElementCount().isKnownMultipleOf(2)) {
auto [Lo, Hi] = DAG.SplitVector(Vec, DL);
Lo = DAG.getNode(ISD::VECTOR_REVERSE, DL, Lo.getSimpleValueType(), Lo);
Hi = DAG.getNode(ISD::VECTOR_REVERSE, DL, Hi.getSimpleValueType(), Hi);
SDValue Concat = DAG.getNode(ISD::CONCAT_VECTORS, DL, ContainerVT, Hi, Lo);
// Fixed length vectors might not fit exactly into their container, and so
// leave a gap in the front of the vector after being reversed. Slide this
// away.
//
// x x x x 3 2 1 0 <- v4i16 @ vlen=128
// 0 1 2 3 x x x x <- reverse
// x x x x 0 1 2 3 <- vslidedown.vx
if (VecVT.isFixedLengthVector()) {
SDValue Offset = DAG.getNode(
ISD::SUB, DL, XLenVT,
DAG.getElementCount(DL, XLenVT, ContainerVT.getVectorElementCount()),
DAG.getElementCount(DL, XLenVT, VecVT.getVectorElementCount()));
Concat =
getVSlidedown(DAG, Subtarget, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), Concat, Offset, Mask, VL);
Concat = convertFromScalableVector(VecVT, Concat, DAG, Subtarget);
}
return Concat;
}
unsigned EltSize = ContainerVT.getScalarSizeInBits();
unsigned MinSize = ContainerVT.getSizeInBits().getKnownMinValue();
unsigned VectorBitsMax = Subtarget.getRealMaxVLen();
unsigned MaxVLMAX =
VecVT.isFixedLengthVector()
? VecVT.getVectorNumElements()
: RISCVTargetLowering::computeVLMAX(VectorBitsMax, EltSize, MinSize);
unsigned GatherOpc = RISCVISD::VRGATHER_VV_VL;
MVT IntVT = ContainerVT.changeVectorElementTypeToInteger();
// If this is SEW=8 and VLMAX is potentially more than 256, we need
// to use vrgatherei16.vv.
if (MaxVLMAX > 256 && EltSize == 8) {
// If this is LMUL=8, we have to split before can use vrgatherei16.vv.
// Reverse each half, then reassemble them in reverse order.
// NOTE: It's also possible that after splitting that VLMAX no longer
// requires vrgatherei16.vv.
if (MinSize == (8 * RISCV::RVVBitsPerBlock)) {
auto [Lo, Hi] = DAG.SplitVectorOperand(Op.getNode(), 0);
auto [LoVT, HiVT] = DAG.GetSplitDestVTs(VecVT);
Lo = DAG.getNode(ISD::VECTOR_REVERSE, DL, LoVT, Lo);
Hi = DAG.getNode(ISD::VECTOR_REVERSE, DL, HiVT, Hi);
// Reassemble the low and high pieces reversed.
// FIXME: This is a CONCAT_VECTORS.
SDValue Res = DAG.getInsertSubvector(DL, DAG.getUNDEF(VecVT), Hi, 0);
return DAG.getInsertSubvector(DL, Res, Lo,
LoVT.getVectorMinNumElements());
}
// Just promote the int type to i16 which will double the LMUL.
IntVT = MVT::getVectorVT(MVT::i16, ContainerVT.getVectorElementCount());
GatherOpc = RISCVISD::VRGATHEREI16_VV_VL;
}
// At LMUL > 1, do the index computation in 16 bits to reduce register
// pressure.
if (IntVT.getScalarType().bitsGT(MVT::i16) &&
IntVT.bitsGT(RISCVTargetLowering::getM1VT(IntVT))) {
assert(isUInt<16>(MaxVLMAX - 1)); // Largest VLMAX is 65536 @ zvl65536b
GatherOpc = RISCVISD::VRGATHEREI16_VV_VL;
IntVT = IntVT.changeVectorElementType(MVT::i16);
}
// Calculate VLMAX-1 for the desired SEW.
SDValue VLMinus1 = DAG.getNode(
ISD::SUB, DL, XLenVT,
DAG.getElementCount(DL, XLenVT, VecVT.getVectorElementCount()),
DAG.getConstant(1, DL, XLenVT));
// Splat VLMAX-1 taking care to handle SEW==64 on RV32.
bool IsRV32E64 =
!Subtarget.is64Bit() && IntVT.getVectorElementType() == MVT::i64;
SDValue SplatVL;
if (!IsRV32E64)
SplatVL = DAG.getSplatVector(IntVT, DL, VLMinus1);
else
SplatVL = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, IntVT, DAG.getUNDEF(IntVT),
VLMinus1, DAG.getRegister(RISCV::X0, XLenVT));
SDValue VID = DAG.getNode(RISCVISD::VID_VL, DL, IntVT, Mask, VL);
SDValue Indices = DAG.getNode(RISCVISD::SUB_VL, DL, IntVT, SplatVL, VID,
DAG.getUNDEF(IntVT), Mask, VL);
SDValue Gather = DAG.getNode(GatherOpc, DL, ContainerVT, Vec, Indices,
DAG.getUNDEF(ContainerVT), Mask, VL);
if (VecVT.isFixedLengthVector())
Gather = convertFromScalableVector(VecVT, Gather, DAG, Subtarget);
return Gather;
}
SDValue RISCVTargetLowering::lowerVECTOR_SPLICE(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
MVT XLenVT = Subtarget.getXLenVT();
MVT VecVT = Op.getSimpleValueType();
SDValue VLMax = computeVLMax(VecVT, DL, DAG);
int64_t ImmValue = cast<ConstantSDNode>(Op.getOperand(2))->getSExtValue();
SDValue DownOffset, UpOffset;
if (ImmValue >= 0) {
// The operand is a TargetConstant, we need to rebuild it as a regular
// constant.
DownOffset = DAG.getConstant(ImmValue, DL, XLenVT);
UpOffset = DAG.getNode(ISD::SUB, DL, XLenVT, VLMax, DownOffset);
} else {
// The operand is a TargetConstant, we need to rebuild it as a regular
// constant rather than negating the original operand.
UpOffset = DAG.getConstant(-ImmValue, DL, XLenVT);
DownOffset = DAG.getNode(ISD::SUB, DL, XLenVT, VLMax, UpOffset);
}
SDValue TrueMask = getAllOnesMask(VecVT, VLMax, DL, DAG);
SDValue SlideDown = getVSlidedown(
DAG, Subtarget, DL, VecVT, DAG.getUNDEF(VecVT), V1, DownOffset, TrueMask,
Subtarget.hasVLDependentLatency() ? UpOffset
: DAG.getRegister(RISCV::X0, XLenVT));
return getVSlideup(DAG, Subtarget, DL, VecVT, SlideDown, V2, UpOffset,
TrueMask, DAG.getRegister(RISCV::X0, XLenVT),
RISCVVType::TAIL_AGNOSTIC);
}
SDValue
RISCVTargetLowering::lowerFixedLengthVectorLoadToRVV(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
auto *Load = cast<LoadSDNode>(Op);
assert(allowsMemoryAccessForAlignment(*DAG.getContext(), DAG.getDataLayout(),
Load->getMemoryVT(),
*Load->getMemOperand()) &&
"Expecting a correctly-aligned load");
MVT VT = Op.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
MVT ContainerVT = getContainerForFixedLengthVector(VT);
// If we know the exact VLEN and our fixed length vector completely fills
// the container, use a whole register load instead.
const auto [MinVLMAX, MaxVLMAX] =
RISCVTargetLowering::computeVLMAXBounds(ContainerVT, Subtarget);
if (MinVLMAX == MaxVLMAX && MinVLMAX == VT.getVectorNumElements() &&
RISCVTargetLowering::getM1VT(ContainerVT).bitsLE(ContainerVT)) {
MachineMemOperand *MMO = Load->getMemOperand();
SDValue NewLoad =
DAG.getLoad(ContainerVT, DL, Load->getChain(), Load->getBasePtr(),
MMO->getPointerInfo(), MMO->getBaseAlign(), MMO->getFlags(),
MMO->getAAInfo(), MMO->getRanges());
SDValue Result = convertFromScalableVector(VT, NewLoad, DAG, Subtarget);
return DAG.getMergeValues({Result, NewLoad.getValue(1)}, DL);
}
SDValue VL = DAG.getConstant(VT.getVectorNumElements(), DL, XLenVT);
bool IsMaskOp = VT.getVectorElementType() == MVT::i1;
SDValue IntID = DAG.getTargetConstant(
IsMaskOp ? Intrinsic::riscv_vlm : Intrinsic::riscv_vle, DL, XLenVT);
SmallVector<SDValue, 4> Ops{Load->getChain(), IntID};
if (!IsMaskOp)
Ops.push_back(DAG.getUNDEF(ContainerVT));
Ops.push_back(Load->getBasePtr());
Ops.push_back(VL);
SDVTList VTs = DAG.getVTList({ContainerVT, MVT::Other});
SDValue NewLoad =
DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops,
Load->getMemoryVT(), Load->getMemOperand());
SDValue Result = convertFromScalableVector(VT, NewLoad, DAG, Subtarget);
return DAG.getMergeValues({Result, NewLoad.getValue(1)}, DL);
}
SDValue
RISCVTargetLowering::lowerFixedLengthVectorStoreToRVV(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
auto *Store = cast<StoreSDNode>(Op);
assert(allowsMemoryAccessForAlignment(*DAG.getContext(), DAG.getDataLayout(),
Store->getMemoryVT(),
*Store->getMemOperand()) &&
"Expecting a correctly-aligned store");
SDValue StoreVal = Store->getValue();
MVT VT = StoreVal.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
// If the size less than a byte, we need to pad with zeros to make a byte.
if (VT.getVectorElementType() == MVT::i1 && VT.getVectorNumElements() < 8) {
VT = MVT::v8i1;
StoreVal =
DAG.getInsertSubvector(DL, DAG.getConstant(0, DL, VT), StoreVal, 0);
}
MVT ContainerVT = getContainerForFixedLengthVector(VT);
SDValue NewValue =
convertToScalableVector(ContainerVT, StoreVal, DAG, Subtarget);
// If we know the exact VLEN and our fixed length vector completely fills
// the container, use a whole register store instead.
const auto [MinVLMAX, MaxVLMAX] =
RISCVTargetLowering::computeVLMAXBounds(ContainerVT, Subtarget);
if (MinVLMAX == MaxVLMAX && MinVLMAX == VT.getVectorNumElements() &&
RISCVTargetLowering::getM1VT(ContainerVT).bitsLE(ContainerVT)) {
MachineMemOperand *MMO = Store->getMemOperand();
return DAG.getStore(Store->getChain(), DL, NewValue, Store->getBasePtr(),
MMO->getPointerInfo(), MMO->getBaseAlign(),
MMO->getFlags(), MMO->getAAInfo());
}
SDValue VL = DAG.getConstant(VT.getVectorNumElements(), DL, XLenVT);
bool IsMaskOp = VT.getVectorElementType() == MVT::i1;
SDValue IntID = DAG.getTargetConstant(
IsMaskOp ? Intrinsic::riscv_vsm : Intrinsic::riscv_vse, DL, XLenVT);
return DAG.getMemIntrinsicNode(
ISD::INTRINSIC_VOID, DL, DAG.getVTList(MVT::Other),
{Store->getChain(), IntID, NewValue, Store->getBasePtr(), VL},
Store->getMemoryVT(), Store->getMemOperand());
}
SDValue RISCVTargetLowering::lowerMaskedLoad(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
const auto *MemSD = cast<MemSDNode>(Op);
EVT MemVT = MemSD->getMemoryVT();
MachineMemOperand *MMO = MemSD->getMemOperand();
SDValue Chain = MemSD->getChain();
SDValue BasePtr = MemSD->getBasePtr();
SDValue Mask, PassThru, VL;
bool IsExpandingLoad = false;
if (const auto *VPLoad = dyn_cast<VPLoadSDNode>(Op)) {
Mask = VPLoad->getMask();
PassThru = DAG.getUNDEF(VT);
VL = VPLoad->getVectorLength();
} else {
const auto *MLoad = cast<MaskedLoadSDNode>(Op);
Mask = MLoad->getMask();
PassThru = MLoad->getPassThru();
IsExpandingLoad = MLoad->isExpandingLoad();
}
bool IsUnmasked = ISD::isConstantSplatVectorAllOnes(Mask.getNode());
MVT XLenVT = Subtarget.getXLenVT();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
PassThru = convertToScalableVector(ContainerVT, PassThru, DAG, Subtarget);
if (!IsUnmasked) {
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
}
if (!VL)
VL = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget).second;
SDValue ExpandingVL;
if (!IsUnmasked && IsExpandingLoad) {
ExpandingVL = VL;
VL =
DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, Mask,
getAllOnesMask(Mask.getSimpleValueType(), VL, DL, DAG), VL);
}
unsigned IntID = IsUnmasked || IsExpandingLoad ? Intrinsic::riscv_vle
: Intrinsic::riscv_vle_mask;
SmallVector<SDValue, 8> Ops{Chain, DAG.getTargetConstant(IntID, DL, XLenVT)};
if (IntID == Intrinsic::riscv_vle)
Ops.push_back(DAG.getUNDEF(ContainerVT));
else
Ops.push_back(PassThru);
Ops.push_back(BasePtr);
if (IntID == Intrinsic::riscv_vle_mask)
Ops.push_back(Mask);
Ops.push_back(VL);
if (IntID == Intrinsic::riscv_vle_mask)
Ops.push_back(DAG.getTargetConstant(RISCVVType::TAIL_AGNOSTIC, DL, XLenVT));
SDVTList VTs = DAG.getVTList({ContainerVT, MVT::Other});
SDValue Result =
DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops, MemVT, MMO);
Chain = Result.getValue(1);
if (ExpandingVL) {
MVT IndexVT = ContainerVT;
if (ContainerVT.isFloatingPoint())
IndexVT = ContainerVT.changeVectorElementTypeToInteger();
MVT IndexEltVT = IndexVT.getVectorElementType();
bool UseVRGATHEREI16 = false;
// If index vector is an i8 vector and the element count exceeds 256, we
// should change the element type of index vector to i16 to avoid
// overflow.
if (IndexEltVT == MVT::i8 && VT.getVectorNumElements() > 256) {
// FIXME: We need to do vector splitting manually for LMUL=8 cases.
assert(getLMUL(IndexVT) != RISCVVType::LMUL_8);
IndexVT = IndexVT.changeVectorElementType(MVT::i16);
UseVRGATHEREI16 = true;
}
SDValue Iota =
DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, IndexVT,
DAG.getConstant(Intrinsic::riscv_viota, DL, XLenVT),
DAG.getUNDEF(IndexVT), Mask, ExpandingVL);
Result =
DAG.getNode(UseVRGATHEREI16 ? RISCVISD::VRGATHEREI16_VV_VL
: RISCVISD::VRGATHER_VV_VL,
DL, ContainerVT, Result, Iota, PassThru, Mask, ExpandingVL);
}
if (VT.isFixedLengthVector())
Result = convertFromScalableVector(VT, Result, DAG, Subtarget);
return DAG.getMergeValues({Result, Chain}, DL);
}
SDValue RISCVTargetLowering::lowerLoadFF(SDValue Op, SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op->getSimpleValueType(0);
const auto *VPLoadFF = cast<VPLoadFFSDNode>(Op);
EVT MemVT = VPLoadFF->getMemoryVT();
MachineMemOperand *MMO = VPLoadFF->getMemOperand();
SDValue Chain = VPLoadFF->getChain();
SDValue BasePtr = VPLoadFF->getBasePtr();
SDValue Mask = VPLoadFF->getMask();
SDValue VL = VPLoadFF->getVectorLength();
MVT XLenVT = Subtarget.getXLenVT();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
unsigned IntID = Intrinsic::riscv_vleff_mask;
SDValue Ops[] = {
Chain,
DAG.getTargetConstant(IntID, DL, XLenVT),
DAG.getUNDEF(ContainerVT),
BasePtr,
Mask,
VL,
DAG.getTargetConstant(RISCVVType::TAIL_AGNOSTIC, DL, XLenVT)};
SDVTList VTs = DAG.getVTList({ContainerVT, Op->getValueType(1), MVT::Other});
SDValue Result =
DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops, MemVT, MMO);
SDValue OutVL = Result.getValue(1);
Chain = Result.getValue(2);
if (VT.isFixedLengthVector())
Result = convertFromScalableVector(VT, Result, DAG, Subtarget);
return DAG.getMergeValues({Result, OutVL, Chain}, DL);
}
SDValue RISCVTargetLowering::lowerMaskedStore(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
const auto *MemSD = cast<MemSDNode>(Op);
EVT MemVT = MemSD->getMemoryVT();
MachineMemOperand *MMO = MemSD->getMemOperand();
SDValue Chain = MemSD->getChain();
SDValue BasePtr = MemSD->getBasePtr();
SDValue Val, Mask, VL;
bool IsCompressingStore = false;
if (const auto *VPStore = dyn_cast<VPStoreSDNode>(Op)) {
Val = VPStore->getValue();
Mask = VPStore->getMask();
VL = VPStore->getVectorLength();
} else {
const auto *MStore = cast<MaskedStoreSDNode>(Op);
Val = MStore->getValue();
Mask = MStore->getMask();
IsCompressingStore = MStore->isCompressingStore();
}
bool IsUnmasked =
ISD::isConstantSplatVectorAllOnes(Mask.getNode()) || IsCompressingStore;
MVT VT = Val.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
Val = convertToScalableVector(ContainerVT, Val, DAG, Subtarget);
if (!IsUnmasked || IsCompressingStore) {
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
}
if (!VL)
VL = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget).second;
if (IsCompressingStore) {
Val = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, ContainerVT,
DAG.getConstant(Intrinsic::riscv_vcompress, DL, XLenVT),
DAG.getUNDEF(ContainerVT), Val, Mask, VL);
VL =
DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, Mask,
getAllOnesMask(Mask.getSimpleValueType(), VL, DL, DAG), VL);
}
unsigned IntID =
IsUnmasked ? Intrinsic::riscv_vse : Intrinsic::riscv_vse_mask;
SmallVector<SDValue, 8> Ops{Chain, DAG.getTargetConstant(IntID, DL, XLenVT)};
Ops.push_back(Val);
Ops.push_back(BasePtr);
if (!IsUnmasked)
Ops.push_back(Mask);
Ops.push_back(VL);
return DAG.getMemIntrinsicNode(ISD::INTRINSIC_VOID, DL,
DAG.getVTList(MVT::Other), Ops, MemVT, MMO);
}
SDValue RISCVTargetLowering::lowerVectorCompress(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
SDValue Val = Op.getOperand(0);
SDValue Mask = Op.getOperand(1);
SDValue Passthru = Op.getOperand(2);
MVT VT = Val.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
MVT MaskVT = getMaskTypeFor(ContainerVT);
Val = convertToScalableVector(ContainerVT, Val, DAG, Subtarget);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
Passthru = convertToScalableVector(ContainerVT, Passthru, DAG, Subtarget);
}
SDValue VL = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget).second;
SDValue Res =
DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, ContainerVT,
DAG.getConstant(Intrinsic::riscv_vcompress, DL, XLenVT),
Passthru, Val, Mask, VL);
if (VT.isFixedLengthVector())
Res = convertFromScalableVector(VT, Res, DAG, Subtarget);
return Res;
}
SDValue RISCVTargetLowering::lowerVectorStrictFSetcc(SDValue Op,
SelectionDAG &DAG) const {
unsigned Opc = Op.getOpcode();
SDLoc DL(Op);
SDValue Chain = Op.getOperand(0);
SDValue Op1 = Op.getOperand(1);
SDValue Op2 = Op.getOperand(2);
SDValue CC = Op.getOperand(3);
ISD::CondCode CCVal = cast<CondCodeSDNode>(CC)->get();
MVT VT = Op.getSimpleValueType();
MVT InVT = Op1.getSimpleValueType();
// RVV VMFEQ/VMFNE ignores qNan, so we expand strict_fsetccs with OEQ/UNE
// condition code.
if (Opc == ISD::STRICT_FSETCCS) {
// Expand strict_fsetccs(x, oeq) to
// (and strict_fsetccs(x, y, oge), strict_fsetccs(x, y, ole))
SDVTList VTList = Op->getVTList();
if (CCVal == ISD::SETEQ || CCVal == ISD::SETOEQ) {
SDValue OLECCVal = DAG.getCondCode(ISD::SETOLE);
SDValue Tmp1 = DAG.getNode(ISD::STRICT_FSETCCS, DL, VTList, Chain, Op1,
Op2, OLECCVal);
SDValue Tmp2 = DAG.getNode(ISD::STRICT_FSETCCS, DL, VTList, Chain, Op2,
Op1, OLECCVal);
SDValue OutChain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other,
Tmp1.getValue(1), Tmp2.getValue(1));
// Tmp1 and Tmp2 might be the same node.
if (Tmp1 != Tmp2)
Tmp1 = DAG.getNode(ISD::AND, DL, VT, Tmp1, Tmp2);
return DAG.getMergeValues({Tmp1, OutChain}, DL);
}
// Expand (strict_fsetccs x, y, une) to (not (strict_fsetccs x, y, oeq))
if (CCVal == ISD::SETNE || CCVal == ISD::SETUNE) {
SDValue OEQCCVal = DAG.getCondCode(ISD::SETOEQ);
SDValue OEQ = DAG.getNode(ISD::STRICT_FSETCCS, DL, VTList, Chain, Op1,
Op2, OEQCCVal);
SDValue Res = DAG.getNOT(DL, OEQ, VT);
return DAG.getMergeValues({Res, OEQ.getValue(1)}, DL);
}
}
MVT ContainerInVT = InVT;
if (InVT.isFixedLengthVector()) {
ContainerInVT = getContainerForFixedLengthVector(InVT);
Op1 = convertToScalableVector(ContainerInVT, Op1, DAG, Subtarget);
Op2 = convertToScalableVector(ContainerInVT, Op2, DAG, Subtarget);
}
MVT MaskVT = getMaskTypeFor(ContainerInVT);
auto [Mask, VL] = getDefaultVLOps(InVT, ContainerInVT, DL, DAG, Subtarget);
SDValue Res;
if (Opc == ISD::STRICT_FSETCC &&
(CCVal == ISD::SETLT || CCVal == ISD::SETOLT || CCVal == ISD::SETLE ||
CCVal == ISD::SETOLE)) {
// VMFLT/VMFLE/VMFGT/VMFGE raise exception for qNan. Generate a mask to only
// active when both input elements are ordered.
SDValue True = getAllOnesMask(ContainerInVT, VL, DL, DAG);
SDValue OrderMask1 = DAG.getNode(
RISCVISD::STRICT_FSETCC_VL, DL, DAG.getVTList(MaskVT, MVT::Other),
{Chain, Op1, Op1, DAG.getCondCode(ISD::SETOEQ), DAG.getUNDEF(MaskVT),
True, VL});
SDValue OrderMask2 = DAG.getNode(
RISCVISD::STRICT_FSETCC_VL, DL, DAG.getVTList(MaskVT, MVT::Other),
{Chain, Op2, Op2, DAG.getCondCode(ISD::SETOEQ), DAG.getUNDEF(MaskVT),
True, VL});
Mask =
DAG.getNode(RISCVISD::VMAND_VL, DL, MaskVT, OrderMask1, OrderMask2, VL);
// Use Mask as the passthru operand to let the result be 0 if either of the
// inputs is unordered.
Res = DAG.getNode(RISCVISD::STRICT_FSETCCS_VL, DL,
DAG.getVTList(MaskVT, MVT::Other),
{Chain, Op1, Op2, CC, Mask, Mask, VL});
} else {
unsigned RVVOpc = Opc == ISD::STRICT_FSETCC ? RISCVISD::STRICT_FSETCC_VL
: RISCVISD::STRICT_FSETCCS_VL;
Res = DAG.getNode(RVVOpc, DL, DAG.getVTList(MaskVT, MVT::Other),
{Chain, Op1, Op2, CC, DAG.getUNDEF(MaskVT), Mask, VL});
}
if (VT.isFixedLengthVector()) {
SDValue SubVec = convertFromScalableVector(VT, Res, DAG, Subtarget);
return DAG.getMergeValues({SubVec, Res.getValue(1)}, DL);
}
return Res;
}
// Lower vector ABS to smax(X, sub(0, X)).
SDValue RISCVTargetLowering::lowerABS(SDValue Op, SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
SDValue X = Op.getOperand(0);
assert((Op.getOpcode() == ISD::VP_ABS || VT.isFixedLengthVector()) &&
"Unexpected type for ISD::ABS");
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
X = convertToScalableVector(ContainerVT, X, DAG, Subtarget);
}
SDValue Mask, VL;
if (Op->getOpcode() == ISD::VP_ABS) {
Mask = Op->getOperand(1);
if (VT.isFixedLengthVector())
Mask = convertToScalableVector(getMaskTypeFor(ContainerVT), Mask, DAG,
Subtarget);
VL = Op->getOperand(2);
} else
std::tie(Mask, VL) = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
SDValue SplatZero = DAG.getNode(
RISCVISD::VMV_V_X_VL, DL, ContainerVT, DAG.getUNDEF(ContainerVT),
DAG.getConstant(0, DL, Subtarget.getXLenVT()), VL);
SDValue NegX = DAG.getNode(RISCVISD::SUB_VL, DL, ContainerVT, SplatZero, X,
DAG.getUNDEF(ContainerVT), Mask, VL);
SDValue Max = DAG.getNode(RISCVISD::SMAX_VL, DL, ContainerVT, X, NegX,
DAG.getUNDEF(ContainerVT), Mask, VL);
if (VT.isFixedLengthVector())
Max = convertFromScalableVector(VT, Max, DAG, Subtarget);
return Max;
}
SDValue RISCVTargetLowering::lowerToScalableOp(SDValue Op,
SelectionDAG &DAG) const {
const auto &TSInfo =
static_cast<const RISCVSelectionDAGInfo &>(DAG.getSelectionDAGInfo());
unsigned NewOpc = getRISCVVLOp(Op);
bool HasPassthruOp = TSInfo.hasPassthruOp(NewOpc);
bool HasMask = TSInfo.hasMaskOp(NewOpc);
MVT VT = Op.getSimpleValueType();
MVT ContainerVT = getContainerForFixedLengthVector(VT);
// Create list of operands by converting existing ones to scalable types.
SmallVector<SDValue, 6> Ops;
for (const SDValue &V : Op->op_values()) {
assert(!isa<VTSDNode>(V) && "Unexpected VTSDNode node!");
// Pass through non-vector operands.
if (!V.getValueType().isVector()) {
Ops.push_back(V);
continue;
}
// "cast" fixed length vector to a scalable vector.
assert(useRVVForFixedLengthVectorVT(V.getSimpleValueType()) &&
"Only fixed length vectors are supported!");
MVT VContainerVT = ContainerVT.changeVectorElementType(
V.getSimpleValueType().getVectorElementType());
Ops.push_back(convertToScalableVector(VContainerVT, V, DAG, Subtarget));
}
SDLoc DL(Op);
auto [Mask, VL] = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget);
if (HasPassthruOp)
Ops.push_back(DAG.getUNDEF(ContainerVT));
if (HasMask)
Ops.push_back(Mask);
Ops.push_back(VL);
// StrictFP operations have two result values. Their lowered result should
// have same result count.
if (Op->isStrictFPOpcode()) {
SDValue ScalableRes =
DAG.getNode(NewOpc, DL, DAG.getVTList(ContainerVT, MVT::Other), Ops,
Op->getFlags());
SDValue SubVec = convertFromScalableVector(VT, ScalableRes, DAG, Subtarget);
return DAG.getMergeValues({SubVec, ScalableRes.getValue(1)}, DL);
}
SDValue ScalableRes =
DAG.getNode(NewOpc, DL, ContainerVT, Ops, Op->getFlags());
return convertFromScalableVector(VT, ScalableRes, DAG, Subtarget);
}
// Lower a VP_* ISD node to the corresponding RISCVISD::*_VL node:
// * Operands of each node are assumed to be in the same order.
// * The EVL operand is promoted from i32 to i64 on RV64.
// * Fixed-length vectors are converted to their scalable-vector container
// types.
SDValue RISCVTargetLowering::lowerVPOp(SDValue Op, SelectionDAG &DAG) const {
const auto &TSInfo =
static_cast<const RISCVSelectionDAGInfo &>(DAG.getSelectionDAGInfo());
unsigned RISCVISDOpc = getRISCVVLOp(Op);
bool HasPassthruOp = TSInfo.hasPassthruOp(RISCVISDOpc);
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
SmallVector<SDValue, 4> Ops;
MVT ContainerVT = VT;
if (VT.isFixedLengthVector())
ContainerVT = getContainerForFixedLengthVector(VT);
for (const auto &OpIdx : enumerate(Op->ops())) {
SDValue V = OpIdx.value();
assert(!isa<VTSDNode>(V) && "Unexpected VTSDNode node!");
// Add dummy passthru value before the mask. Or if there isn't a mask,
// before EVL.
if (HasPassthruOp) {
auto MaskIdx = ISD::getVPMaskIdx(Op.getOpcode());
if (MaskIdx) {
if (*MaskIdx == OpIdx.index())
Ops.push_back(DAG.getUNDEF(ContainerVT));
} else if (ISD::getVPExplicitVectorLengthIdx(Op.getOpcode()) ==
OpIdx.index()) {
if (Op.getOpcode() == ISD::VP_MERGE) {
// For VP_MERGE, copy the false operand instead of an undef value.
Ops.push_back(Ops.back());
} else {
assert(Op.getOpcode() == ISD::VP_SELECT);
// For VP_SELECT, add an undef value.
Ops.push_back(DAG.getUNDEF(ContainerVT));
}
}
}
// VFCVT_RM_X_F_VL requires a rounding mode to be injected before the VL.
if (RISCVISDOpc == RISCVISD::VFCVT_RM_X_F_VL &&
ISD::getVPExplicitVectorLengthIdx(Op.getOpcode()) == OpIdx.index())
Ops.push_back(DAG.getTargetConstant(RISCVFPRndMode::DYN, DL,
Subtarget.getXLenVT()));
// Pass through operands which aren't fixed-length vectors.
if (!V.getValueType().isFixedLengthVector()) {
Ops.push_back(V);
continue;
}
// "cast" fixed length vector to a scalable vector.
MVT OpVT = V.getSimpleValueType();
MVT ContainerVT = getContainerForFixedLengthVector(OpVT);
assert(useRVVForFixedLengthVectorVT(OpVT) &&
"Only fixed length vectors are supported!");
Ops.push_back(convertToScalableVector(ContainerVT, V, DAG, Subtarget));
}
if (!VT.isFixedLengthVector())
return DAG.getNode(RISCVISDOpc, DL, VT, Ops, Op->getFlags());
SDValue VPOp = DAG.getNode(RISCVISDOpc, DL, ContainerVT, Ops, Op->getFlags());
return convertFromScalableVector(VT, VPOp, DAG, Subtarget);
}
SDValue RISCVTargetLowering::lowerVPExtMaskOp(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
SDValue Src = Op.getOperand(0);
// NOTE: Mask is dropped.
SDValue VL = Op.getOperand(2);
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
MVT SrcVT = MVT::getVectorVT(MVT::i1, ContainerVT.getVectorElementCount());
Src = convertToScalableVector(SrcVT, Src, DAG, Subtarget);
}
MVT XLenVT = Subtarget.getXLenVT();
SDValue Zero = DAG.getConstant(0, DL, XLenVT);
SDValue ZeroSplat = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), Zero, VL);
SDValue SplatValue = DAG.getSignedConstant(
Op.getOpcode() == ISD::VP_ZERO_EXTEND ? 1 : -1, DL, XLenVT);
SDValue Splat = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), SplatValue, VL);
SDValue Result = DAG.getNode(RISCVISD::VMERGE_VL, DL, ContainerVT, Src, Splat,
ZeroSplat, DAG.getUNDEF(ContainerVT), VL);
if (!VT.isFixedLengthVector())
return Result;
return convertFromScalableVector(VT, Result, DAG, Subtarget);
}
SDValue RISCVTargetLowering::lowerVPSetCCMaskOp(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
SDValue Op1 = Op.getOperand(0);
SDValue Op2 = Op.getOperand(1);
ISD::CondCode Condition = cast<CondCodeSDNode>(Op.getOperand(2))->get();
// NOTE: Mask is dropped.
SDValue VL = Op.getOperand(4);
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
Op1 = convertToScalableVector(ContainerVT, Op1, DAG, Subtarget);
Op2 = convertToScalableVector(ContainerVT, Op2, DAG, Subtarget);
}
SDValue Result;
SDValue AllOneMask = DAG.getNode(RISCVISD::VMSET_VL, DL, ContainerVT, VL);
switch (Condition) {
default:
break;
// X != Y --> (X^Y)
case ISD::SETNE:
Result = DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Op1, Op2, VL);
break;
// X == Y --> ~(X^Y)
case ISD::SETEQ: {
SDValue Temp =
DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Op1, Op2, VL);
Result =
DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Temp, AllOneMask, VL);
break;
}
// X >s Y --> X == 0 & Y == 1 --> ~X & Y
// X <u Y --> X == 0 & Y == 1 --> ~X & Y
case ISD::SETGT:
case ISD::SETULT: {
SDValue Temp =
DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Op1, AllOneMask, VL);
Result = DAG.getNode(RISCVISD::VMAND_VL, DL, ContainerVT, Temp, Op2, VL);
break;
}
// X <s Y --> X == 1 & Y == 0 --> ~Y & X
// X >u Y --> X == 1 & Y == 0 --> ~Y & X
case ISD::SETLT:
case ISD::SETUGT: {
SDValue Temp =
DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Op2, AllOneMask, VL);
Result = DAG.getNode(RISCVISD::VMAND_VL, DL, ContainerVT, Op1, Temp, VL);
break;
}
// X >=s Y --> X == 0 | Y == 1 --> ~X | Y
// X <=u Y --> X == 0 | Y == 1 --> ~X | Y
case ISD::SETGE:
case ISD::SETULE: {
SDValue Temp =
DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Op1, AllOneMask, VL);
Result = DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Temp, Op2, VL);
break;
}
// X <=s Y --> X == 1 | Y == 0 --> ~Y | X
// X >=u Y --> X == 1 | Y == 0 --> ~Y | X
case ISD::SETLE:
case ISD::SETUGE: {
SDValue Temp =
DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Op2, AllOneMask, VL);
Result = DAG.getNode(RISCVISD::VMXOR_VL, DL, ContainerVT, Temp, Op1, VL);
break;
}
}
if (!VT.isFixedLengthVector())
return Result;
return convertFromScalableVector(VT, Result, DAG, Subtarget);
}
// Lower Floating-Point/Integer Type-Convert VP SDNodes
SDValue RISCVTargetLowering::lowerVPFPIntConvOp(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
SDValue Src = Op.getOperand(0);
SDValue Mask = Op.getOperand(1);
SDValue VL = Op.getOperand(2);
unsigned RISCVISDOpc = getRISCVVLOp(Op);
MVT DstVT = Op.getSimpleValueType();
MVT SrcVT = Src.getSimpleValueType();
if (DstVT.isFixedLengthVector()) {
DstVT = getContainerForFixedLengthVector(DstVT);
SrcVT = getContainerForFixedLengthVector(SrcVT);
Src = convertToScalableVector(SrcVT, Src, DAG, Subtarget);
MVT MaskVT = getMaskTypeFor(DstVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
unsigned DstEltSize = DstVT.getScalarSizeInBits();
unsigned SrcEltSize = SrcVT.getScalarSizeInBits();
SDValue Result;
if (DstEltSize >= SrcEltSize) { // Single-width and widening conversion.
if (SrcVT.isInteger()) {
assert(DstVT.isFloatingPoint() && "Wrong input/output vector types");
unsigned RISCVISDExtOpc = RISCVISDOpc == RISCVISD::SINT_TO_FP_VL
? RISCVISD::VSEXT_VL
: RISCVISD::VZEXT_VL;
// Do we need to do any pre-widening before converting?
if (SrcEltSize == 1) {
MVT IntVT = DstVT.changeVectorElementTypeToInteger();
MVT XLenVT = Subtarget.getXLenVT();
SDValue Zero = DAG.getConstant(0, DL, XLenVT);
SDValue ZeroSplat = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, IntVT,
DAG.getUNDEF(IntVT), Zero, VL);
SDValue One = DAG.getSignedConstant(
RISCVISDExtOpc == RISCVISD::VZEXT_VL ? 1 : -1, DL, XLenVT);
SDValue OneSplat = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, IntVT,
DAG.getUNDEF(IntVT), One, VL);
Src = DAG.getNode(RISCVISD::VMERGE_VL, DL, IntVT, Src, OneSplat,
ZeroSplat, DAG.getUNDEF(IntVT), VL);
} else if (DstEltSize > (2 * SrcEltSize)) {
// Widen before converting.
MVT IntVT = MVT::getVectorVT(MVT::getIntegerVT(DstEltSize / 2),
DstVT.getVectorElementCount());
Src = DAG.getNode(RISCVISDExtOpc, DL, IntVT, Src, Mask, VL);
}
Result = DAG.getNode(RISCVISDOpc, DL, DstVT, Src, Mask, VL);
} else {
assert(SrcVT.isFloatingPoint() && DstVT.isInteger() &&
"Wrong input/output vector types");
// Convert f16 to f32 then convert f32 to i64.
if (DstEltSize > (2 * SrcEltSize)) {
assert(SrcVT.getVectorElementType() == MVT::f16 && "Unexpected type!");
MVT InterimFVT =
MVT::getVectorVT(MVT::f32, DstVT.getVectorElementCount());
Src =
DAG.getNode(RISCVISD::FP_EXTEND_VL, DL, InterimFVT, Src, Mask, VL);
}
Result = DAG.getNode(RISCVISDOpc, DL, DstVT, Src, Mask, VL);
}
} else { // Narrowing + Conversion
if (SrcVT.isInteger()) {
assert(DstVT.isFloatingPoint() && "Wrong input/output vector types");
// First do a narrowing convert to an FP type half the size, then round
// the FP type to a small FP type if needed.
MVT InterimFVT = DstVT;
if (SrcEltSize > (2 * DstEltSize)) {
assert(SrcEltSize == (4 * DstEltSize) && "Unexpected types!");
assert(DstVT.getVectorElementType() == MVT::f16 && "Unexpected type!");
InterimFVT = MVT::getVectorVT(MVT::f32, DstVT.getVectorElementCount());
}
Result = DAG.getNode(RISCVISDOpc, DL, InterimFVT, Src, Mask, VL);
if (InterimFVT != DstVT) {
Src = Result;
Result = DAG.getNode(RISCVISD::FP_ROUND_VL, DL, DstVT, Src, Mask, VL);
}
} else {
assert(SrcVT.isFloatingPoint() && DstVT.isInteger() &&
"Wrong input/output vector types");
// First do a narrowing conversion to an integer half the size, then
// truncate if needed.
if (DstEltSize == 1) {
// First convert to the same size integer, then convert to mask using
// setcc.
assert(SrcEltSize >= 16 && "Unexpected FP type!");
MVT InterimIVT = MVT::getVectorVT(MVT::getIntegerVT(SrcEltSize),
DstVT.getVectorElementCount());
Result = DAG.getNode(RISCVISDOpc, DL, InterimIVT, Src, Mask, VL);
// Compare the integer result to 0. The integer should be 0 or 1/-1,
// otherwise the conversion was undefined.
MVT XLenVT = Subtarget.getXLenVT();
SDValue SplatZero = DAG.getConstant(0, DL, XLenVT);
SplatZero = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, InterimIVT,
DAG.getUNDEF(InterimIVT), SplatZero, VL);
Result = DAG.getNode(RISCVISD::SETCC_VL, DL, DstVT,
{Result, SplatZero, DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(DstVT), Mask, VL});
} else {
MVT InterimIVT = MVT::getVectorVT(MVT::getIntegerVT(SrcEltSize / 2),
DstVT.getVectorElementCount());
Result = DAG.getNode(RISCVISDOpc, DL, InterimIVT, Src, Mask, VL);
while (InterimIVT != DstVT) {
SrcEltSize /= 2;
Src = Result;
InterimIVT = MVT::getVectorVT(MVT::getIntegerVT(SrcEltSize / 2),
DstVT.getVectorElementCount());
Result = DAG.getNode(RISCVISD::TRUNCATE_VECTOR_VL, DL, InterimIVT,
Src, Mask, VL);
}
}
}
}
MVT VT = Op.getSimpleValueType();
if (!VT.isFixedLengthVector())
return Result;
return convertFromScalableVector(VT, Result, DAG, Subtarget);
}
SDValue RISCVTargetLowering::lowerVPMergeMask(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
SDValue Mask = Op.getOperand(0);
SDValue TrueVal = Op.getOperand(1);
SDValue FalseVal = Op.getOperand(2);
SDValue VL = Op.getOperand(3);
// Use default legalization if a vector of EVL type would be legal.
EVT EVLVecVT = EVT::getVectorVT(*DAG.getContext(), VL.getValueType(),
VT.getVectorElementCount());
if (isTypeLegal(EVLVecVT))
return SDValue();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
Mask = convertToScalableVector(ContainerVT, Mask, DAG, Subtarget);
TrueVal = convertToScalableVector(ContainerVT, TrueVal, DAG, Subtarget);
FalseVal = convertToScalableVector(ContainerVT, FalseVal, DAG, Subtarget);
}
// Promote to a vector of i8.
MVT PromotedVT = ContainerVT.changeVectorElementType(MVT::i8);
// Promote TrueVal and FalseVal using VLMax.
// FIXME: Is there a better way to do this?
SDValue VLMax = DAG.getRegister(RISCV::X0, XLenVT);
SDValue SplatOne = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, PromotedVT,
DAG.getUNDEF(PromotedVT),
DAG.getConstant(1, DL, XLenVT), VLMax);
SDValue SplatZero = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, PromotedVT,
DAG.getUNDEF(PromotedVT),
DAG.getConstant(0, DL, XLenVT), VLMax);
TrueVal = DAG.getNode(RISCVISD::VMERGE_VL, DL, PromotedVT, TrueVal, SplatOne,
SplatZero, DAG.getUNDEF(PromotedVT), VL);
// Any element past VL uses FalseVal, so use VLMax
FalseVal = DAG.getNode(RISCVISD::VMERGE_VL, DL, PromotedVT, FalseVal,
SplatOne, SplatZero, DAG.getUNDEF(PromotedVT), VLMax);
// VP_MERGE the two promoted values.
SDValue VPMerge = DAG.getNode(RISCVISD::VMERGE_VL, DL, PromotedVT, Mask,
TrueVal, FalseVal, FalseVal, VL);
// Convert back to mask.
SDValue TrueMask = DAG.getNode(RISCVISD::VMSET_VL, DL, ContainerVT, VL);
SDValue Result = DAG.getNode(
RISCVISD::SETCC_VL, DL, ContainerVT,
{VPMerge, DAG.getConstant(0, DL, PromotedVT), DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(getMaskTypeFor(ContainerVT)), TrueMask, VLMax});
if (VT.isFixedLengthVector())
Result = convertFromScalableVector(VT, Result, DAG, Subtarget);
return Result;
}
SDValue
RISCVTargetLowering::lowerVPSpliceExperimental(SDValue Op,
SelectionDAG &DAG) const {
using namespace SDPatternMatch;
SDLoc DL(Op);
SDValue Op1 = Op.getOperand(0);
SDValue Op2 = Op.getOperand(1);
SDValue Offset = Op.getOperand(2);
SDValue Mask = Op.getOperand(3);
SDValue EVL1 = Op.getOperand(4);
SDValue EVL2 = Op.getOperand(5);
const MVT XLenVT = Subtarget.getXLenVT();
MVT VT = Op.getSimpleValueType();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
Op1 = convertToScalableVector(ContainerVT, Op1, DAG, Subtarget);
Op2 = convertToScalableVector(ContainerVT, Op2, DAG, Subtarget);
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
bool IsMaskVector = VT.getVectorElementType() == MVT::i1;
if (IsMaskVector) {
ContainerVT = ContainerVT.changeVectorElementType(MVT::i8);
// Expand input operands
SDValue SplatOneOp1 = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT),
DAG.getConstant(1, DL, XLenVT), EVL1);
SDValue SplatZeroOp1 = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT),
DAG.getConstant(0, DL, XLenVT), EVL1);
Op1 = DAG.getNode(RISCVISD::VMERGE_VL, DL, ContainerVT, Op1, SplatOneOp1,
SplatZeroOp1, DAG.getUNDEF(ContainerVT), EVL1);
SDValue SplatOneOp2 = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT),
DAG.getConstant(1, DL, XLenVT), EVL2);
SDValue SplatZeroOp2 = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT),
DAG.getConstant(0, DL, XLenVT), EVL2);
Op2 = DAG.getNode(RISCVISD::VMERGE_VL, DL, ContainerVT, Op2, SplatOneOp2,
SplatZeroOp2, DAG.getUNDEF(ContainerVT), EVL2);
}
auto getVectorFirstEle = [](SDValue Vec) {
SDValue FirstEle;
if (sd_match(Vec, m_InsertElt(m_Value(), m_Value(FirstEle), m_Zero())))
return FirstEle;
if (Vec.getOpcode() == ISD::SPLAT_VECTOR ||
Vec.getOpcode() == ISD::BUILD_VECTOR)
return Vec.getOperand(0);
return SDValue();
};
if (!IsMaskVector && isNullConstant(Offset) && isOneConstant(EVL1))
if (auto FirstEle = getVectorFirstEle(Op->getOperand(0))) {
MVT EltVT = ContainerVT.getVectorElementType();
SDValue Result;
if ((EltVT == MVT::f16 && !Subtarget.hasVInstructionsF16()) ||
EltVT == MVT::bf16) {
EltVT = EltVT.changeTypeToInteger();
ContainerVT = ContainerVT.changeVectorElementType(EltVT);
Op2 = DAG.getBitcast(ContainerVT, Op2);
FirstEle =
DAG.getAnyExtOrTrunc(DAG.getBitcast(EltVT, FirstEle), DL, XLenVT);
}
Result = DAG.getNode(EltVT.isFloatingPoint() ? RISCVISD::VFSLIDE1UP_VL
: RISCVISD::VSLIDE1UP_VL,
DL, ContainerVT, DAG.getUNDEF(ContainerVT), Op2,
FirstEle, Mask, EVL2);
Result = DAG.getBitcast(
ContainerVT.changeVectorElementType(VT.getVectorElementType()),
Result);
return VT.isFixedLengthVector()
? convertFromScalableVector(VT, Result, DAG, Subtarget)
: Result;
}
int64_t ImmValue = cast<ConstantSDNode>(Offset)->getSExtValue();
SDValue DownOffset, UpOffset;
if (ImmValue >= 0) {
// The operand is a TargetConstant, we need to rebuild it as a regular
// constant.
DownOffset = DAG.getConstant(ImmValue, DL, XLenVT);
UpOffset = DAG.getNode(ISD::SUB, DL, XLenVT, EVL1, DownOffset);
} else {
// The operand is a TargetConstant, we need to rebuild it as a regular
// constant rather than negating the original operand.
UpOffset = DAG.getConstant(-ImmValue, DL, XLenVT);
DownOffset = DAG.getNode(ISD::SUB, DL, XLenVT, EVL1, UpOffset);
}
if (ImmValue != 0)
Op1 = getVSlidedown(DAG, Subtarget, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), Op1, DownOffset, Mask,
Subtarget.hasVLDependentLatency() ? UpOffset : EVL2);
SDValue Result = getVSlideup(DAG, Subtarget, DL, ContainerVT, Op1, Op2,
UpOffset, Mask, EVL2, RISCVVType::TAIL_AGNOSTIC);
if (IsMaskVector) {
// Truncate Result back to a mask vector (Result has same EVL as Op2)
Result = DAG.getNode(
RISCVISD::SETCC_VL, DL, ContainerVT.changeVectorElementType(MVT::i1),
{Result, DAG.getConstant(0, DL, ContainerVT),
DAG.getCondCode(ISD::SETNE), DAG.getUNDEF(getMaskTypeFor(ContainerVT)),
Mask, EVL2});
}
if (!VT.isFixedLengthVector())
return Result;
return convertFromScalableVector(VT, Result, DAG, Subtarget);
}
SDValue RISCVTargetLowering::lowerVPSplatExperimental(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
SDValue Val = Op.getOperand(0);
SDValue Mask = Op.getOperand(1);
SDValue VL = Op.getOperand(2);
MVT VT = Op.getSimpleValueType();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
SDValue Result;
if (VT.getScalarType() == MVT::i1) {
if (auto *C = dyn_cast<ConstantSDNode>(Val)) {
Result =
DAG.getNode(C->isZero() ? RISCVISD::VMCLR_VL : RISCVISD::VMSET_VL, DL,
ContainerVT, VL);
} else {
MVT WidenVT = ContainerVT.changeVectorElementType(MVT::i8);
SDValue LHS =
DAG.getNode(RISCVISD::VMV_V_X_VL, DL, WidenVT, DAG.getUNDEF(WidenVT),
DAG.getZExtOrTrunc(Val, DL, Subtarget.getXLenVT()), VL);
SDValue RHS = DAG.getConstant(0, DL, WidenVT);
Result = DAG.getNode(RISCVISD::SETCC_VL, DL, ContainerVT,
{LHS, RHS, DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(ContainerVT), Mask, VL});
}
} else {
Result =
lowerScalarSplat(SDValue(), Val, VL, ContainerVT, DL, DAG, Subtarget);
}
if (!VT.isFixedLengthVector())
return Result;
return convertFromScalableVector(VT, Result, DAG, Subtarget);
}
SDValue
RISCVTargetLowering::lowerVPReverseExperimental(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
SDValue Op1 = Op.getOperand(0);
SDValue Mask = Op.getOperand(1);
SDValue EVL = Op.getOperand(2);
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
Op1 = convertToScalableVector(ContainerVT, Op1, DAG, Subtarget);
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
MVT GatherVT = ContainerVT;
MVT IndicesVT = ContainerVT.changeVectorElementTypeToInteger();
// Check if we are working with mask vectors
bool IsMaskVector = ContainerVT.getVectorElementType() == MVT::i1;
if (IsMaskVector) {
GatherVT = IndicesVT = ContainerVT.changeVectorElementType(MVT::i8);
// Expand input operand
SDValue SplatOne = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, IndicesVT,
DAG.getUNDEF(IndicesVT),
DAG.getConstant(1, DL, XLenVT), EVL);
SDValue SplatZero = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, IndicesVT,
DAG.getUNDEF(IndicesVT),
DAG.getConstant(0, DL, XLenVT), EVL);
Op1 = DAG.getNode(RISCVISD::VMERGE_VL, DL, IndicesVT, Op1, SplatOne,
SplatZero, DAG.getUNDEF(IndicesVT), EVL);
}
unsigned EltSize = GatherVT.getScalarSizeInBits();
unsigned MinSize = GatherVT.getSizeInBits().getKnownMinValue();
unsigned VectorBitsMax = Subtarget.getRealMaxVLen();
unsigned MaxVLMAX =
RISCVTargetLowering::computeVLMAX(VectorBitsMax, EltSize, MinSize);
unsigned GatherOpc = RISCVISD::VRGATHER_VV_VL;
// If this is SEW=8 and VLMAX is unknown or more than 256, we need
// to use vrgatherei16.vv.
// TODO: It's also possible to use vrgatherei16.vv for other types to
// decrease register width for the index calculation.
// NOTE: This code assumes VLMAX <= 65536 for LMUL=8 SEW=16.
if (MaxVLMAX > 256 && EltSize == 8) {
// If this is LMUL=8, we have to split before using vrgatherei16.vv.
// Split the vector in half and reverse each half using a full register
// reverse.
// Swap the halves and concatenate them.
// Slide the concatenated result by (VLMax - VL).
if (MinSize == (8 * RISCV::RVVBitsPerBlock)) {
auto [LoVT, HiVT] = DAG.GetSplitDestVTs(GatherVT);
auto [Lo, Hi] = DAG.SplitVector(Op1, DL);
SDValue LoRev = DAG.getNode(ISD::VECTOR_REVERSE, DL, LoVT, Lo);
SDValue HiRev = DAG.getNode(ISD::VECTOR_REVERSE, DL, HiVT, Hi);
// Reassemble the low and high pieces reversed.
// NOTE: this Result is unmasked (because we do not need masks for
// shuffles). If in the future this has to change, we can use a SELECT_VL
// between Result and UNDEF using the mask originally passed to VP_REVERSE
SDValue Result =
DAG.getNode(ISD::CONCAT_VECTORS, DL, GatherVT, HiRev, LoRev);
// Slide off any elements from past EVL that were reversed into the low
// elements.
unsigned MinElts = GatherVT.getVectorMinNumElements();
SDValue VLMax =
DAG.getVScale(DL, XLenVT, APInt(XLenVT.getSizeInBits(), MinElts));
SDValue Diff = DAG.getNode(ISD::SUB, DL, XLenVT, VLMax, EVL);
Result = getVSlidedown(DAG, Subtarget, DL, GatherVT,
DAG.getUNDEF(GatherVT), Result, Diff, Mask, EVL);
if (IsMaskVector) {
// Truncate Result back to a mask vector
Result =
DAG.getNode(RISCVISD::SETCC_VL, DL, ContainerVT,
{Result, DAG.getConstant(0, DL, GatherVT),
DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(getMaskTypeFor(ContainerVT)), Mask, EVL});
}
if (!VT.isFixedLengthVector())
return Result;
return convertFromScalableVector(VT, Result, DAG, Subtarget);
}
// Just promote the int type to i16 which will double the LMUL.
IndicesVT = MVT::getVectorVT(MVT::i16, IndicesVT.getVectorElementCount());
GatherOpc = RISCVISD::VRGATHEREI16_VV_VL;
}
SDValue VID = DAG.getNode(RISCVISD::VID_VL, DL, IndicesVT, Mask, EVL);
SDValue VecLen =
DAG.getNode(ISD::SUB, DL, XLenVT, EVL, DAG.getConstant(1, DL, XLenVT));
SDValue VecLenSplat = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, IndicesVT,
DAG.getUNDEF(IndicesVT), VecLen, EVL);
SDValue VRSUB = DAG.getNode(RISCVISD::SUB_VL, DL, IndicesVT, VecLenSplat, VID,
DAG.getUNDEF(IndicesVT), Mask, EVL);
SDValue Result = DAG.getNode(GatherOpc, DL, GatherVT, Op1, VRSUB,
DAG.getUNDEF(GatherVT), Mask, EVL);
if (IsMaskVector) {
// Truncate Result back to a mask vector
Result = DAG.getNode(
RISCVISD::SETCC_VL, DL, ContainerVT,
{Result, DAG.getConstant(0, DL, GatherVT), DAG.getCondCode(ISD::SETNE),
DAG.getUNDEF(getMaskTypeFor(ContainerVT)), Mask, EVL});
}
if (!VT.isFixedLengthVector())
return Result;
return convertFromScalableVector(VT, Result, DAG, Subtarget);
}
SDValue RISCVTargetLowering::lowerLogicVPOp(SDValue Op,
SelectionDAG &DAG) const {
MVT VT = Op.getSimpleValueType();
if (VT.getVectorElementType() != MVT::i1)
return lowerVPOp(Op, DAG);
// It is safe to drop mask parameter as masked-off elements are undef.
SDValue Op1 = Op->getOperand(0);
SDValue Op2 = Op->getOperand(1);
SDValue VL = Op->getOperand(3);
MVT ContainerVT = VT;
const bool IsFixed = VT.isFixedLengthVector();
if (IsFixed) {
ContainerVT = getContainerForFixedLengthVector(VT);
Op1 = convertToScalableVector(ContainerVT, Op1, DAG, Subtarget);
Op2 = convertToScalableVector(ContainerVT, Op2, DAG, Subtarget);
}
SDLoc DL(Op);
SDValue Val = DAG.getNode(getRISCVVLOp(Op), DL, ContainerVT, Op1, Op2, VL);
if (!IsFixed)
return Val;
return convertFromScalableVector(VT, Val, DAG, Subtarget);
}
SDValue RISCVTargetLowering::lowerVPStridedLoad(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
MVT VT = Op.getSimpleValueType();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector())
ContainerVT = getContainerForFixedLengthVector(VT);
SDVTList VTs = DAG.getVTList({ContainerVT, MVT::Other});
auto *VPNode = cast<VPStridedLoadSDNode>(Op);
// Check if the mask is known to be all ones
SDValue Mask = VPNode->getMask();
bool IsUnmasked = ISD::isConstantSplatVectorAllOnes(Mask.getNode());
SDValue IntID = DAG.getTargetConstant(IsUnmasked ? Intrinsic::riscv_vlse
: Intrinsic::riscv_vlse_mask,
DL, XLenVT);
SmallVector<SDValue, 8> Ops{VPNode->getChain(), IntID,
DAG.getUNDEF(ContainerVT), VPNode->getBasePtr(),
VPNode->getStride()};
if (!IsUnmasked) {
if (VT.isFixedLengthVector()) {
MVT MaskVT = ContainerVT.changeVectorElementType(MVT::i1);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
Ops.push_back(Mask);
}
Ops.push_back(VPNode->getVectorLength());
if (!IsUnmasked) {
SDValue Policy =
DAG.getTargetConstant(RISCVVType::TAIL_AGNOSTIC, DL, XLenVT);
Ops.push_back(Policy);
}
SDValue Result =
DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops,
VPNode->getMemoryVT(), VPNode->getMemOperand());
SDValue Chain = Result.getValue(1);
if (VT.isFixedLengthVector())
Result = convertFromScalableVector(VT, Result, DAG, Subtarget);
return DAG.getMergeValues({Result, Chain}, DL);
}
SDValue RISCVTargetLowering::lowerVPStridedStore(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT XLenVT = Subtarget.getXLenVT();
auto *VPNode = cast<VPStridedStoreSDNode>(Op);
SDValue StoreVal = VPNode->getValue();
MVT VT = StoreVal.getSimpleValueType();
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
StoreVal = convertToScalableVector(ContainerVT, StoreVal, DAG, Subtarget);
}
// Check if the mask is known to be all ones
SDValue Mask = VPNode->getMask();
bool IsUnmasked = ISD::isConstantSplatVectorAllOnes(Mask.getNode());
SDValue IntID = DAG.getTargetConstant(IsUnmasked ? Intrinsic::riscv_vsse
: Intrinsic::riscv_vsse_mask,
DL, XLenVT);
SmallVector<SDValue, 8> Ops{VPNode->getChain(), IntID, StoreVal,
VPNode->getBasePtr(), VPNode->getStride()};
if (!IsUnmasked) {
if (VT.isFixedLengthVector()) {
MVT MaskVT = ContainerVT.changeVectorElementType(MVT::i1);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
Ops.push_back(Mask);
}
Ops.push_back(VPNode->getVectorLength());
return DAG.getMemIntrinsicNode(ISD::INTRINSIC_VOID, DL, VPNode->getVTList(),
Ops, VPNode->getMemoryVT(),
VPNode->getMemOperand());
}
// Custom lower MGATHER/VP_GATHER to a legalized form for RVV. It will then be
// matched to a RVV indexed load. The RVV indexed load instructions only
// support the "unsigned unscaled" addressing mode; indices are implicitly
// zero-extended or truncated to XLEN and are treated as byte offsets. Any
// signed or scaled indexing is extended to the XLEN value type and scaled
// accordingly.
SDValue RISCVTargetLowering::lowerMaskedGather(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
MVT VT = Op.getSimpleValueType();
const auto *MemSD = cast<MemSDNode>(Op.getNode());
EVT MemVT = MemSD->getMemoryVT();
MachineMemOperand *MMO = MemSD->getMemOperand();
SDValue Chain = MemSD->getChain();
SDValue BasePtr = MemSD->getBasePtr();
[[maybe_unused]] ISD::LoadExtType LoadExtType;
SDValue Index, Mask, PassThru, VL;
if (auto *VPGN = dyn_cast<VPGatherSDNode>(Op.getNode())) {
Index = VPGN->getIndex();
Mask = VPGN->getMask();
PassThru = DAG.getUNDEF(VT);
VL = VPGN->getVectorLength();
// VP doesn't support extending loads.
LoadExtType = ISD::NON_EXTLOAD;
} else {
// Else it must be a MGATHER.
auto *MGN = cast<MaskedGatherSDNode>(Op.getNode());
Index = MGN->getIndex();
Mask = MGN->getMask();
PassThru = MGN->getPassThru();
LoadExtType = MGN->getExtensionType();
}
MVT IndexVT = Index.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
assert(VT.getVectorElementCount() == IndexVT.getVectorElementCount() &&
"Unexpected VTs!");
assert(BasePtr.getSimpleValueType() == XLenVT && "Unexpected pointer type");
// Targets have to explicitly opt-in for extending vector loads.
assert(LoadExtType == ISD::NON_EXTLOAD &&
"Unexpected extending MGATHER/VP_GATHER");
// If the mask is known to be all ones, optimize to an unmasked intrinsic;
// the selection of the masked intrinsics doesn't do this for us.
bool IsUnmasked = ISD::isConstantSplatVectorAllOnes(Mask.getNode());
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
IndexVT = MVT::getVectorVT(IndexVT.getVectorElementType(),
ContainerVT.getVectorElementCount());
Index = convertToScalableVector(IndexVT, Index, DAG, Subtarget);
if (!IsUnmasked) {
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
PassThru = convertToScalableVector(ContainerVT, PassThru, DAG, Subtarget);
}
}
if (!VL)
VL = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget).second;
if (XLenVT == MVT::i32 && IndexVT.getVectorElementType().bitsGT(XLenVT)) {
IndexVT = IndexVT.changeVectorElementType(XLenVT);
Index = DAG.getNode(ISD::TRUNCATE, DL, IndexVT, Index);
}
unsigned IntID =
IsUnmasked ? Intrinsic::riscv_vluxei : Intrinsic::riscv_vluxei_mask;
SmallVector<SDValue, 8> Ops{Chain, DAG.getTargetConstant(IntID, DL, XLenVT)};
if (IsUnmasked)
Ops.push_back(DAG.getUNDEF(ContainerVT));
else
Ops.push_back(PassThru);
Ops.push_back(BasePtr);
Ops.push_back(Index);
if (!IsUnmasked)
Ops.push_back(Mask);
Ops.push_back(VL);
if (!IsUnmasked)
Ops.push_back(DAG.getTargetConstant(RISCVVType::TAIL_AGNOSTIC, DL, XLenVT));
SDVTList VTs = DAG.getVTList({ContainerVT, MVT::Other});
SDValue Result =
DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops, MemVT, MMO);
Chain = Result.getValue(1);
if (VT.isFixedLengthVector())
Result = convertFromScalableVector(VT, Result, DAG, Subtarget);
return DAG.getMergeValues({Result, Chain}, DL);
}
// Custom lower MSCATTER/VP_SCATTER to a legalized form for RVV. It will then be
// matched to a RVV indexed store. The RVV indexed store instructions only
// support the "unsigned unscaled" addressing mode; indices are implicitly
// zero-extended or truncated to XLEN and are treated as byte offsets. Any
// signed or scaled indexing is extended to the XLEN value type and scaled
// accordingly.
SDValue RISCVTargetLowering::lowerMaskedScatter(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
const auto *MemSD = cast<MemSDNode>(Op.getNode());
EVT MemVT = MemSD->getMemoryVT();
MachineMemOperand *MMO = MemSD->getMemOperand();
SDValue Chain = MemSD->getChain();
SDValue BasePtr = MemSD->getBasePtr();
[[maybe_unused]] bool IsTruncatingStore = false;
SDValue Index, Mask, Val, VL;
if (auto *VPSN = dyn_cast<VPScatterSDNode>(Op.getNode())) {
Index = VPSN->getIndex();
Mask = VPSN->getMask();
Val = VPSN->getValue();
VL = VPSN->getVectorLength();
// VP doesn't support truncating stores.
IsTruncatingStore = false;
} else {
// Else it must be a MSCATTER.
auto *MSN = cast<MaskedScatterSDNode>(Op.getNode());
Index = MSN->getIndex();
Mask = MSN->getMask();
Val = MSN->getValue();
IsTruncatingStore = MSN->isTruncatingStore();
}
MVT VT = Val.getSimpleValueType();
MVT IndexVT = Index.getSimpleValueType();
MVT XLenVT = Subtarget.getXLenVT();
assert(VT.getVectorElementCount() == IndexVT.getVectorElementCount() &&
"Unexpected VTs!");
assert(BasePtr.getSimpleValueType() == XLenVT && "Unexpected pointer type");
// Targets have to explicitly opt-in for extending vector loads and
// truncating vector stores.
assert(!IsTruncatingStore && "Unexpected truncating MSCATTER/VP_SCATTER");
// If the mask is known to be all ones, optimize to an unmasked intrinsic;
// the selection of the masked intrinsics doesn't do this for us.
bool IsUnmasked = ISD::isConstantSplatVectorAllOnes(Mask.getNode());
MVT ContainerVT = VT;
if (VT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VT);
IndexVT = MVT::getVectorVT(IndexVT.getVectorElementType(),
ContainerVT.getVectorElementCount());
Index = convertToScalableVector(IndexVT, Index, DAG, Subtarget);
Val = convertToScalableVector(ContainerVT, Val, DAG, Subtarget);
if (!IsUnmasked) {
MVT MaskVT = getMaskTypeFor(ContainerVT);
Mask = convertToScalableVector(MaskVT, Mask, DAG, Subtarget);
}
}
if (!VL)
VL = getDefaultVLOps(VT, ContainerVT, DL, DAG, Subtarget).second;
if (XLenVT == MVT::i32 && IndexVT.getVectorElementType().bitsGT(XLenVT)) {
IndexVT = IndexVT.changeVectorElementType(XLenVT);
Index = DAG.getNode(ISD::TRUNCATE, DL, IndexVT, Index);
}
unsigned IntID =
IsUnmasked ? Intrinsic::riscv_vsoxei : Intrinsic::riscv_vsoxei_mask;
SmallVector<SDValue, 8> Ops{Chain, DAG.getTargetConstant(IntID, DL, XLenVT)};
Ops.push_back(Val);
Ops.push_back(BasePtr);
Ops.push_back(Index);
if (!IsUnmasked)
Ops.push_back(Mask);
Ops.push_back(VL);
return DAG.getMemIntrinsicNode(ISD::INTRINSIC_VOID, DL,
DAG.getVTList(MVT::Other), Ops, MemVT, MMO);
}
SDValue RISCVTargetLowering::lowerGET_ROUNDING(SDValue Op,
SelectionDAG &DAG) const {
const MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(Op);
SDValue Chain = Op->getOperand(0);
SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::frm, DL, XLenVT);
SDVTList VTs = DAG.getVTList(XLenVT, MVT::Other);
SDValue RM = DAG.getNode(RISCVISD::READ_CSR, DL, VTs, Chain, SysRegNo);
// Encoding used for rounding mode in RISC-V differs from that used in
// FLT_ROUNDS. To convert it the RISC-V rounding mode is used as an index in a
// table, which consists of a sequence of 4-bit fields, each representing
// corresponding FLT_ROUNDS mode.
static const int Table =
(int(RoundingMode::NearestTiesToEven) << 4 * RISCVFPRndMode::RNE) |
(int(RoundingMode::TowardZero) << 4 * RISCVFPRndMode::RTZ) |
(int(RoundingMode::TowardNegative) << 4 * RISCVFPRndMode::RDN) |
(int(RoundingMode::TowardPositive) << 4 * RISCVFPRndMode::RUP) |
(int(RoundingMode::NearestTiesToAway) << 4 * RISCVFPRndMode::RMM);
SDValue Shift =
DAG.getNode(ISD::SHL, DL, XLenVT, RM, DAG.getConstant(2, DL, XLenVT));
SDValue Shifted = DAG.getNode(ISD::SRL, DL, XLenVT,
DAG.getConstant(Table, DL, XLenVT), Shift);
SDValue Masked = DAG.getNode(ISD::AND, DL, XLenVT, Shifted,
DAG.getConstant(7, DL, XLenVT));
return DAG.getMergeValues({Masked, Chain}, DL);
}
SDValue RISCVTargetLowering::lowerSET_ROUNDING(SDValue Op,
SelectionDAG &DAG) const {
const MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(Op);
SDValue Chain = Op->getOperand(0);
SDValue RMValue = Op->getOperand(1);
SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::frm, DL, XLenVT);
// Encoding used for rounding mode in RISC-V differs from that used in
// FLT_ROUNDS. To convert it the C rounding mode is used as an index in
// a table, which consists of a sequence of 4-bit fields, each representing
// corresponding RISC-V mode.
static const unsigned Table =
(RISCVFPRndMode::RNE << 4 * int(RoundingMode::NearestTiesToEven)) |
(RISCVFPRndMode::RTZ << 4 * int(RoundingMode::TowardZero)) |
(RISCVFPRndMode::RDN << 4 * int(RoundingMode::TowardNegative)) |
(RISCVFPRndMode::RUP << 4 * int(RoundingMode::TowardPositive)) |
(RISCVFPRndMode::RMM << 4 * int(RoundingMode::NearestTiesToAway));
RMValue = DAG.getNode(ISD::ZERO_EXTEND, DL, XLenVT, RMValue);
SDValue Shift = DAG.getNode(ISD::SHL, DL, XLenVT, RMValue,
DAG.getConstant(2, DL, XLenVT));
SDValue Shifted = DAG.getNode(ISD::SRL, DL, XLenVT,
DAG.getConstant(Table, DL, XLenVT), Shift);
RMValue = DAG.getNode(ISD::AND, DL, XLenVT, Shifted,
DAG.getConstant(0x7, DL, XLenVT));
return DAG.getNode(RISCVISD::WRITE_CSR, DL, MVT::Other, Chain, SysRegNo,
RMValue);
}
SDValue RISCVTargetLowering::lowerGET_FPENV(SDValue Op,
SelectionDAG &DAG) const {
const MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(Op);
SDValue Chain = Op->getOperand(0);
SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::fcsr, DL, XLenVT);
SDVTList VTs = DAG.getVTList(XLenVT, MVT::Other);
return DAG.getNode(RISCVISD::READ_CSR, DL, VTs, Chain, SysRegNo);
}
SDValue RISCVTargetLowering::lowerSET_FPENV(SDValue Op,
SelectionDAG &DAG) const {
const MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(Op);
SDValue Chain = Op->getOperand(0);
SDValue EnvValue = Op->getOperand(1);
SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::fcsr, DL, XLenVT);
EnvValue = DAG.getNode(ISD::ZERO_EXTEND, DL, XLenVT, EnvValue);
return DAG.getNode(RISCVISD::WRITE_CSR, DL, MVT::Other, Chain, SysRegNo,
EnvValue);
}
SDValue RISCVTargetLowering::lowerRESET_FPENV(SDValue Op,
SelectionDAG &DAG) const {
const MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(Op);
SDValue Chain = Op->getOperand(0);
SDValue EnvValue = DAG.getRegister(RISCV::X0, XLenVT);
SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::fcsr, DL, XLenVT);
return DAG.getNode(RISCVISD::WRITE_CSR, DL, MVT::Other, Chain, SysRegNo,
EnvValue);
}
const uint64_t ModeMask64 = ~RISCVExceptFlags::ALL;
const uint32_t ModeMask32 = ~RISCVExceptFlags::ALL;
SDValue RISCVTargetLowering::lowerGET_FPMODE(SDValue Op,
SelectionDAG &DAG) const {
const MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(Op);
SDValue Chain = Op->getOperand(0);
SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::fcsr, DL, XLenVT);
SDVTList VTs = DAG.getVTList(XLenVT, MVT::Other);
SDValue Result = DAG.getNode(RISCVISD::READ_CSR, DL, VTs, Chain, SysRegNo);
Chain = Result.getValue(1);
return DAG.getMergeValues({Result, Chain}, DL);
}
SDValue RISCVTargetLowering::lowerSET_FPMODE(SDValue Op,
SelectionDAG &DAG) const {
const MVT XLenVT = Subtarget.getXLenVT();
const uint64_t ModeMaskValue = Subtarget.is64Bit() ? ModeMask64 : ModeMask32;
SDLoc DL(Op);
SDValue Chain = Op->getOperand(0);
SDValue EnvValue = Op->getOperand(1);
SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::fcsr, DL, XLenVT);
SDValue ModeMask = DAG.getConstant(ModeMaskValue, DL, XLenVT);
EnvValue = DAG.getNode(ISD::ZERO_EXTEND, DL, XLenVT, EnvValue);
EnvValue = DAG.getNode(ISD::AND, DL, XLenVT, EnvValue, ModeMask);
Chain = DAG.getNode(RISCVISD::CLEAR_CSR, DL, MVT::Other, Chain, SysRegNo,
ModeMask);
return DAG.getNode(RISCVISD::SET_CSR, DL, MVT::Other, Chain, SysRegNo,
EnvValue);
}
SDValue RISCVTargetLowering::lowerRESET_FPMODE(SDValue Op,
SelectionDAG &DAG) const {
const MVT XLenVT = Subtarget.getXLenVT();
const uint64_t ModeMaskValue = Subtarget.is64Bit() ? ModeMask64 : ModeMask32;
SDLoc DL(Op);
SDValue Chain = Op->getOperand(0);
SDValue SysRegNo = DAG.getTargetConstant(RISCVSysReg::fcsr, DL, XLenVT);
SDValue ModeMask = DAG.getConstant(ModeMaskValue, DL, XLenVT);
return DAG.getNode(RISCVISD::CLEAR_CSR, DL, MVT::Other, Chain, SysRegNo,
ModeMask);
}
SDValue RISCVTargetLowering::lowerEH_DWARF_CFA(SDValue Op,
SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
bool isRISCV64 = Subtarget.is64Bit();
EVT PtrVT = getPointerTy(DAG.getDataLayout());
int FI = MF.getFrameInfo().CreateFixedObject(isRISCV64 ? 8 : 4, 0, false);
return DAG.getFrameIndex(FI, PtrVT);
}
// Returns the opcode of the target-specific SDNode that implements the 32-bit
// form of the given Opcode.
static unsigned getRISCVWOpcode(unsigned Opcode) {
switch (Opcode) {
default:
llvm_unreachable("Unexpected opcode");
case ISD::SHL:
return RISCVISD::SLLW;
case ISD::SRA:
return RISCVISD::SRAW;
case ISD::SRL:
return RISCVISD::SRLW;
case ISD::SDIV:
return RISCVISD::DIVW;
case ISD::UDIV:
return RISCVISD::DIVUW;
case ISD::UREM:
return RISCVISD::REMUW;
case ISD::ROTL:
return RISCVISD::ROLW;
case ISD::ROTR:
return RISCVISD::RORW;
}
}
// Converts the given i8/i16/i32 operation to a target-specific SelectionDAG
// node. Because i8/i16/i32 isn't a legal type for RV64, these operations would
// otherwise be promoted to i64, making it difficult to select the
// SLLW/DIVUW/.../*W later one because the fact the operation was originally of
// type i8/i16/i32 is lost.
static SDValue customLegalizeToWOp(SDNode *N, SelectionDAG &DAG,
unsigned ExtOpc = ISD::ANY_EXTEND) {
SDLoc DL(N);
unsigned WOpcode = getRISCVWOpcode(N->getOpcode());
SDValue NewOp0 = DAG.getNode(ExtOpc, DL, MVT::i64, N->getOperand(0));
SDValue NewOp1 = DAG.getNode(ExtOpc, DL, MVT::i64, N->getOperand(1));
SDValue NewRes = DAG.getNode(WOpcode, DL, MVT::i64, NewOp0, NewOp1);
// ReplaceNodeResults requires we maintain the same type for the return value.
return DAG.getNode(ISD::TRUNCATE, DL, N->getValueType(0), NewRes);
}
// Converts the given 32-bit operation to a i64 operation with signed extension
// semantic to reduce the signed extension instructions.
static SDValue customLegalizeToWOpWithSExt(SDNode *N, SelectionDAG &DAG) {
SDLoc DL(N);
SDValue NewOp0 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(0));
SDValue NewOp1 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue NewWOp = DAG.getNode(N->getOpcode(), DL, MVT::i64, NewOp0, NewOp1);
SDValue NewRes = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i64, NewWOp,
DAG.getValueType(MVT::i32));
return DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, NewRes);
}
void RISCVTargetLowering::ReplaceNodeResults(SDNode *N,
SmallVectorImpl<SDValue> &Results,
SelectionDAG &DAG) const {
SDLoc DL(N);
switch (N->getOpcode()) {
default:
llvm_unreachable("Don't know how to custom type legalize this operation!");
case ISD::STRICT_FP_TO_SINT:
case ISD::STRICT_FP_TO_UINT:
case ISD::FP_TO_SINT:
case ISD::FP_TO_UINT: {
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
bool IsStrict = N->isStrictFPOpcode();
bool IsSigned = N->getOpcode() == ISD::FP_TO_SINT ||
N->getOpcode() == ISD::STRICT_FP_TO_SINT;
SDValue Op0 = IsStrict ? N->getOperand(1) : N->getOperand(0);
if (getTypeAction(*DAG.getContext(), Op0.getValueType()) !=
TargetLowering::TypeSoftenFloat) {
if (!isTypeLegal(Op0.getValueType()))
return;
if (IsStrict) {
SDValue Chain = N->getOperand(0);
// In absence of Zfh, promote f16 to f32, then convert.
if (Op0.getValueType() == MVT::f16 &&
!Subtarget.hasStdExtZfhOrZhinx()) {
Op0 = DAG.getNode(ISD::STRICT_FP_EXTEND, DL, {MVT::f32, MVT::Other},
{Chain, Op0});
Chain = Op0.getValue(1);
}
unsigned Opc = IsSigned ? RISCVISD::STRICT_FCVT_W_RV64
: RISCVISD::STRICT_FCVT_WU_RV64;
SDVTList VTs = DAG.getVTList(MVT::i64, MVT::Other);
SDValue Res = DAG.getNode(
Opc, DL, VTs, Chain, Op0,
DAG.getTargetConstant(RISCVFPRndMode::RTZ, DL, MVT::i64));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
Results.push_back(Res.getValue(1));
return;
}
// For bf16, or f16 in absence of Zfh, promote [b]f16 to f32 and then
// convert.
if ((Op0.getValueType() == MVT::f16 &&
!Subtarget.hasStdExtZfhOrZhinx()) ||
Op0.getValueType() == MVT::bf16)
Op0 = DAG.getNode(ISD::FP_EXTEND, DL, MVT::f32, Op0);
unsigned Opc = IsSigned ? RISCVISD::FCVT_W_RV64 : RISCVISD::FCVT_WU_RV64;
SDValue Res =
DAG.getNode(Opc, DL, MVT::i64, Op0,
DAG.getTargetConstant(RISCVFPRndMode::RTZ, DL, MVT::i64));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
// If the FP type needs to be softened, emit a library call using the 'si'
// version. If we left it to default legalization we'd end up with 'di'. If
// the FP type doesn't need to be softened just let generic type
// legalization promote the result type.
RTLIB::Libcall LC;
if (IsSigned)
LC = RTLIB::getFPTOSINT(Op0.getValueType(), N->getValueType(0));
else
LC = RTLIB::getFPTOUINT(Op0.getValueType(), N->getValueType(0));
MakeLibCallOptions CallOptions;
EVT OpVT = Op0.getValueType();
CallOptions.setTypeListBeforeSoften(OpVT, N->getValueType(0));
SDValue Chain = IsStrict ? N->getOperand(0) : SDValue();
SDValue Result;
std::tie(Result, Chain) =
makeLibCall(DAG, LC, N->getValueType(0), Op0, CallOptions, DL, Chain);
Results.push_back(Result);
if (IsStrict)
Results.push_back(Chain);
break;
}
case ISD::LROUND: {
SDValue Op0 = N->getOperand(0);
EVT Op0VT = Op0.getValueType();
if (getTypeAction(*DAG.getContext(), Op0.getValueType()) !=
TargetLowering::TypeSoftenFloat) {
if (!isTypeLegal(Op0VT))
return;
// In absence of Zfh, promote f16 to f32, then convert.
if (Op0.getValueType() == MVT::f16 && !Subtarget.hasStdExtZfhOrZhinx())
Op0 = DAG.getNode(ISD::FP_EXTEND, DL, MVT::f32, Op0);
SDValue Res =
DAG.getNode(RISCVISD::FCVT_W_RV64, DL, MVT::i64, Op0,
DAG.getTargetConstant(RISCVFPRndMode::RMM, DL, MVT::i64));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
// If the FP type needs to be softened, emit a library call to lround. We'll
// need to truncate the result. We assume any value that doesn't fit in i32
// is allowed to return an unspecified value.
RTLIB::Libcall LC =
Op0.getValueType() == MVT::f64 ? RTLIB::LROUND_F64 : RTLIB::LROUND_F32;
MakeLibCallOptions CallOptions;
EVT OpVT = Op0.getValueType();
CallOptions.setTypeListBeforeSoften(OpVT, MVT::i64);
SDValue Result = makeLibCall(DAG, LC, MVT::i64, Op0, CallOptions, DL).first;
Result = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Result);
Results.push_back(Result);
break;
}
case ISD::READCYCLECOUNTER:
case ISD::READSTEADYCOUNTER: {
assert(!Subtarget.is64Bit() && "READCYCLECOUNTER/READSTEADYCOUNTER only "
"has custom type legalization on riscv32");
SDValue LoCounter, HiCounter;
MVT XLenVT = Subtarget.getXLenVT();
if (N->getOpcode() == ISD::READCYCLECOUNTER) {
LoCounter = DAG.getTargetConstant(RISCVSysReg::cycle, DL, XLenVT);
HiCounter = DAG.getTargetConstant(RISCVSysReg::cycleh, DL, XLenVT);
} else {
LoCounter = DAG.getTargetConstant(RISCVSysReg::time, DL, XLenVT);
HiCounter = DAG.getTargetConstant(RISCVSysReg::timeh, DL, XLenVT);
}
SDVTList VTs = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other);
SDValue RCW = DAG.getNode(RISCVISD::READ_COUNTER_WIDE, DL, VTs,
N->getOperand(0), LoCounter, HiCounter);
Results.push_back(
DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, RCW, RCW.getValue(1)));
Results.push_back(RCW.getValue(2));
break;
}
case ISD::LOAD: {
if (!ISD::isNON_EXTLoad(N))
return;
// Use a SEXTLOAD instead of the default EXTLOAD. Similar to the
// sext_inreg we emit for ADD/SUB/MUL/SLLI.
LoadSDNode *Ld = cast<LoadSDNode>(N);
if (N->getValueType(0) == MVT::i64) {
assert(Subtarget.hasStdExtZilsd() && !Subtarget.is64Bit() &&
"Unexpected custom legalisation");
if (!Subtarget.enableUnalignedScalarMem() && Ld->getAlign() < 8)
return;
SDLoc DL(N);
SDValue Result = DAG.getMemIntrinsicNode(
RISCVISD::LD_RV32, DL,
DAG.getVTList({MVT::i32, MVT::i32, MVT::Other}),
{Ld->getChain(), Ld->getBasePtr()}, MVT::i64, Ld->getMemOperand());
SDValue Lo = Result.getValue(0);
SDValue Hi = Result.getValue(1);
SDValue Pair = DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Lo, Hi);
Results.append({Pair, Result.getValue(2)});
return;
}
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
SDLoc dl(N);
SDValue Res = DAG.getExtLoad(ISD::SEXTLOAD, dl, MVT::i64, Ld->getChain(),
Ld->getBasePtr(), Ld->getMemoryVT(),
Ld->getMemOperand());
Results.push_back(DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Res));
Results.push_back(Res.getValue(1));
return;
}
case ISD::MUL: {
unsigned Size = N->getSimpleValueType(0).getSizeInBits();
unsigned XLen = Subtarget.getXLen();
// This multiply needs to be expanded, try to use MULHSU+MUL if possible.
if (Size > XLen) {
assert(Size == (XLen * 2) && "Unexpected custom legalisation");
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
APInt HighMask = APInt::getHighBitsSet(Size, XLen);
bool LHSIsU = DAG.MaskedValueIsZero(LHS, HighMask);
bool RHSIsU = DAG.MaskedValueIsZero(RHS, HighMask);
// We need exactly one side to be unsigned.
if (LHSIsU == RHSIsU)
return;
auto MakeMULPair = [&](SDValue S, SDValue U) {
MVT XLenVT = Subtarget.getXLenVT();
S = DAG.getNode(ISD::TRUNCATE, DL, XLenVT, S);
U = DAG.getNode(ISD::TRUNCATE, DL, XLenVT, U);
SDValue Lo = DAG.getNode(ISD::MUL, DL, XLenVT, S, U);
SDValue Hi = DAG.getNode(RISCVISD::MULHSU, DL, XLenVT, S, U);
return DAG.getNode(ISD::BUILD_PAIR, DL, N->getValueType(0), Lo, Hi);
};
bool LHSIsS = DAG.ComputeNumSignBits(LHS) > XLen;
bool RHSIsS = DAG.ComputeNumSignBits(RHS) > XLen;
// The other operand should be signed, but still prefer MULH when
// possible.
if (RHSIsU && LHSIsS && !RHSIsS)
Results.push_back(MakeMULPair(LHS, RHS));
else if (LHSIsU && RHSIsS && !LHSIsS)
Results.push_back(MakeMULPair(RHS, LHS));
return;
}
[[fallthrough]];
}
case ISD::ADD:
case ISD::SUB:
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
Results.push_back(customLegalizeToWOpWithSExt(N, DAG));
break;
case ISD::SHL:
case ISD::SRA:
case ISD::SRL:
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
if (N->getOperand(1).getOpcode() != ISD::Constant) {
// If we can use a BSET instruction, allow default promotion to apply.
if (N->getOpcode() == ISD::SHL && Subtarget.hasStdExtZbs() &&
isOneConstant(N->getOperand(0)))
break;
Results.push_back(customLegalizeToWOp(N, DAG));
break;
}
// Custom legalize ISD::SHL by placing a SIGN_EXTEND_INREG after. This is
// similar to customLegalizeToWOpWithSExt, but we must zero_extend the
// shift amount.
if (N->getOpcode() == ISD::SHL) {
SDLoc DL(N);
SDValue NewOp0 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(0));
SDValue NewOp1 =
DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue NewWOp = DAG.getNode(ISD::SHL, DL, MVT::i64, NewOp0, NewOp1);
SDValue NewRes = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i64, NewWOp,
DAG.getValueType(MVT::i32));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, NewRes));
}
break;
case ISD::ROTL:
case ISD::ROTR:
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
assert((Subtarget.hasStdExtZbb() || Subtarget.hasStdExtZbkb() ||
Subtarget.hasVendorXTHeadBb()) &&
"Unexpected custom legalization");
if (!isa<ConstantSDNode>(N->getOperand(1)) &&
!(Subtarget.hasStdExtZbb() || Subtarget.hasStdExtZbkb()))
return;
Results.push_back(customLegalizeToWOp(N, DAG));
break;
case ISD::CTTZ:
case ISD::CTTZ_ZERO_UNDEF:
case ISD::CTLZ:
case ISD::CTLZ_ZERO_UNDEF: {
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
SDValue NewOp0 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(0));
bool IsCTZ =
N->getOpcode() == ISD::CTTZ || N->getOpcode() == ISD::CTTZ_ZERO_UNDEF;
unsigned Opc = IsCTZ ? RISCVISD::CTZW : RISCVISD::CLZW;
SDValue Res = DAG.getNode(Opc, DL, MVT::i64, NewOp0);
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
case ISD::SDIV:
case ISD::UDIV:
case ISD::UREM: {
MVT VT = N->getSimpleValueType(0);
assert((VT == MVT::i8 || VT == MVT::i16 || VT == MVT::i32) &&
Subtarget.is64Bit() && Subtarget.hasStdExtM() &&
"Unexpected custom legalisation");
// Don't promote division/remainder by constant since we should expand those
// to multiply by magic constant.
AttributeList Attr = DAG.getMachineFunction().getFunction().getAttributes();
if (N->getOperand(1).getOpcode() == ISD::Constant &&
!isIntDivCheap(N->getValueType(0), Attr))
return;
// If the input is i32, use ANY_EXTEND since the W instructions don't read
// the upper 32 bits. For other types we need to sign or zero extend
// based on the opcode.
unsigned ExtOpc = ISD::ANY_EXTEND;
if (VT != MVT::i32)
ExtOpc = N->getOpcode() == ISD::SDIV ? ISD::SIGN_EXTEND
: ISD::ZERO_EXTEND;
Results.push_back(customLegalizeToWOp(N, DAG, ExtOpc));
break;
}
case ISD::SADDO: {
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
// If the RHS is a constant, we can simplify ConditionRHS below. Otherwise
// use the default legalization.
if (!isa<ConstantSDNode>(N->getOperand(1)))
return;
SDValue LHS = DAG.getNode(ISD::SIGN_EXTEND, DL, MVT::i64, N->getOperand(0));
SDValue RHS = DAG.getNode(ISD::SIGN_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue Res = DAG.getNode(ISD::ADD, DL, MVT::i64, LHS, RHS);
Res = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i64, Res,
DAG.getValueType(MVT::i32));
SDValue Zero = DAG.getConstant(0, DL, MVT::i64);
// For an addition, the result should be less than one of the operands (LHS)
// if and only if the other operand (RHS) is negative, otherwise there will
// be overflow.
// For a subtraction, the result should be less than one of the operands
// (LHS) if and only if the other operand (RHS) is (non-zero) positive,
// otherwise there will be overflow.
EVT OType = N->getValueType(1);
SDValue ResultLowerThanLHS = DAG.getSetCC(DL, OType, Res, LHS, ISD::SETLT);
SDValue ConditionRHS = DAG.getSetCC(DL, OType, RHS, Zero, ISD::SETLT);
SDValue Overflow =
DAG.getNode(ISD::XOR, DL, OType, ConditionRHS, ResultLowerThanLHS);
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
Results.push_back(Overflow);
return;
}
case ISD::UADDO:
case ISD::USUBO: {
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
bool IsAdd = N->getOpcode() == ISD::UADDO;
// Create an ADDW or SUBW.
SDValue LHS = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(0));
SDValue RHS = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue Res =
DAG.getNode(IsAdd ? ISD::ADD : ISD::SUB, DL, MVT::i64, LHS, RHS);
Res = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i64, Res,
DAG.getValueType(MVT::i32));
SDValue Overflow;
if (IsAdd && isOneConstant(RHS)) {
// Special case uaddo X, 1 overflowed if the addition result is 0.
// The general case (X + C) < C is not necessarily beneficial. Although we
// reduce the live range of X, we may introduce the materialization of
// constant C, especially when the setcc result is used by branch. We have
// no compare with constant and branch instructions.
Overflow = DAG.getSetCC(DL, N->getValueType(1), Res,
DAG.getConstant(0, DL, MVT::i64), ISD::SETEQ);
} else if (IsAdd && isAllOnesConstant(RHS)) {
// Special case uaddo X, -1 overflowed if X != 0.
Overflow = DAG.getSetCC(DL, N->getValueType(1), N->getOperand(0),
DAG.getConstant(0, DL, MVT::i32), ISD::SETNE);
} else {
// Sign extend the LHS and perform an unsigned compare with the ADDW
// result. Since the inputs are sign extended from i32, this is equivalent
// to comparing the lower 32 bits.
LHS = DAG.getNode(ISD::SIGN_EXTEND, DL, MVT::i64, N->getOperand(0));
Overflow = DAG.getSetCC(DL, N->getValueType(1), Res, LHS,
IsAdd ? ISD::SETULT : ISD::SETUGT);
}
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
Results.push_back(Overflow);
return;
}
case ISD::UADDSAT:
case ISD::USUBSAT: {
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
!Subtarget.hasStdExtZbb() && "Unexpected custom legalisation");
// Without Zbb, expand to UADDO/USUBO+select which will trigger our custom
// promotion for UADDO/USUBO.
Results.push_back(expandAddSubSat(N, DAG));
return;
}
case ISD::SADDSAT:
case ISD::SSUBSAT: {
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
Results.push_back(expandAddSubSat(N, DAG));
return;
}
case ISD::ABS: {
assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
"Unexpected custom legalisation");
if (Subtarget.hasStdExtZbb()) {
// Emit a special ABSW node that will be expanded to NEGW+MAX at isel.
// This allows us to remember that the result is sign extended. Expanding
// to NEGW+MAX here requires a Freeze which breaks ComputeNumSignBits.
SDValue Src = DAG.getNode(ISD::SIGN_EXTEND, DL, MVT::i64,
N->getOperand(0));
SDValue Abs = DAG.getNode(RISCVISD::ABSW, DL, MVT::i64, Src);
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Abs));
return;
}
// Expand abs to Y = (sraiw X, 31); subw(xor(X, Y), Y)
SDValue Src = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(0));
// Freeze the source so we can increase it's use count.
Src = DAG.getFreeze(Src);
// Copy sign bit to all bits using the sraiw pattern.
SDValue SignFill = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i64, Src,
DAG.getValueType(MVT::i32));
SignFill = DAG.getNode(ISD::SRA, DL, MVT::i64, SignFill,
DAG.getConstant(31, DL, MVT::i64));
SDValue NewRes = DAG.getNode(ISD::XOR, DL, MVT::i64, Src, SignFill);
NewRes = DAG.getNode(ISD::SUB, DL, MVT::i64, NewRes, SignFill);
// NOTE: The result is only required to be anyextended, but sext is
// consistent with type legalization of sub.
NewRes = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i64, NewRes,
DAG.getValueType(MVT::i32));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, NewRes));
return;
}
case ISD::BITCAST: {
EVT VT = N->getValueType(0);
assert(VT.isInteger() && !VT.isVector() && "Unexpected VT!");
SDValue Op0 = N->getOperand(0);
EVT Op0VT = Op0.getValueType();
MVT XLenVT = Subtarget.getXLenVT();
if (VT == MVT::i16 &&
((Op0VT == MVT::f16 && Subtarget.hasStdExtZfhminOrZhinxmin()) ||
(Op0VT == MVT::bf16 && Subtarget.hasStdExtZfbfmin()))) {
SDValue FPConv = DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, XLenVT, Op0);
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i16, FPConv));
} else if (VT == MVT::i32 && Op0VT == MVT::f32 && Subtarget.is64Bit() &&
Subtarget.hasStdExtFOrZfinx()) {
SDValue FPConv =
DAG.getNode(RISCVISD::FMV_X_ANYEXTW_RV64, DL, MVT::i64, Op0);
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, FPConv));
} else if (VT == MVT::i64 && Op0VT == MVT::f64 && !Subtarget.is64Bit() &&
Subtarget.hasStdExtDOrZdinx()) {
SDValue NewReg = DAG.getNode(RISCVISD::SplitF64, DL,
DAG.getVTList(MVT::i32, MVT::i32), Op0);
SDValue RetReg = DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64,
NewReg.getValue(0), NewReg.getValue(1));
Results.push_back(RetReg);
} else if (!VT.isVector() && Op0VT.isFixedLengthVector() &&
isTypeLegal(Op0VT)) {
// Custom-legalize bitcasts from fixed-length vector types to illegal
// scalar types in order to improve codegen. Bitcast the vector to a
// one-element vector type whose element type is the same as the result
// type, and extract the first element.
EVT BVT = EVT::getVectorVT(*DAG.getContext(), VT, 1);
if (isTypeLegal(BVT)) {
SDValue BVec = DAG.getBitcast(BVT, Op0);
Results.push_back(DAG.getExtractVectorElt(DL, VT, BVec, 0));
}
}
break;
}
case ISD::BITREVERSE: {
assert(N->getValueType(0) == MVT::i8 && Subtarget.hasStdExtZbkb() &&
"Unexpected custom legalisation");
MVT XLenVT = Subtarget.getXLenVT();
SDValue NewOp = DAG.getNode(ISD::ANY_EXTEND, DL, XLenVT, N->getOperand(0));
SDValue NewRes = DAG.getNode(RISCVISD::BREV8, DL, XLenVT, NewOp);
// ReplaceNodeResults requires we maintain the same type for the return
// value.
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i8, NewRes));
break;
}
case RISCVISD::BREV8:
case RISCVISD::ORC_B: {
MVT VT = N->getSimpleValueType(0);
MVT XLenVT = Subtarget.getXLenVT();
assert((VT == MVT::i16 || (VT == MVT::i32 && Subtarget.is64Bit())) &&
"Unexpected custom legalisation");
assert(((N->getOpcode() == RISCVISD::BREV8 && Subtarget.hasStdExtZbkb()) ||
(N->getOpcode() == RISCVISD::ORC_B && Subtarget.hasStdExtZbb())) &&
"Unexpected extension");
SDValue NewOp = DAG.getNode(ISD::ANY_EXTEND, DL, XLenVT, N->getOperand(0));
SDValue NewRes = DAG.getNode(N->getOpcode(), DL, XLenVT, NewOp);
// ReplaceNodeResults requires we maintain the same type for the return
// value.
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, VT, NewRes));
break;
}
case ISD::EXTRACT_VECTOR_ELT: {
// Custom-legalize an EXTRACT_VECTOR_ELT where XLEN<SEW, as the SEW element
// type is illegal (currently only vXi64 RV32).
// With vmv.x.s, when SEW > XLEN, only the least-significant XLEN bits are
// transferred to the destination register. We issue two of these from the
// upper- and lower- halves of the SEW-bit vector element, slid down to the
// first element.
SDValue Vec = N->getOperand(0);
SDValue Idx = N->getOperand(1);
// The vector type hasn't been legalized yet so we can't issue target
// specific nodes if it needs legalization.
// FIXME: We would manually legalize if it's important.
if (!isTypeLegal(Vec.getValueType()))
return;
MVT VecVT = Vec.getSimpleValueType();
assert(!Subtarget.is64Bit() && N->getValueType(0) == MVT::i64 &&
VecVT.getVectorElementType() == MVT::i64 &&
"Unexpected EXTRACT_VECTOR_ELT legalization");
// If this is a fixed vector, we need to convert it to a scalable vector.
MVT ContainerVT = VecVT;
if (VecVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(VecVT);
Vec = convertToScalableVector(ContainerVT, Vec, DAG, Subtarget);
}
MVT XLenVT = Subtarget.getXLenVT();
// Use a VL of 1 to avoid processing more elements than we need.
auto [Mask, VL] = getDefaultVLOps(1, ContainerVT, DL, DAG, Subtarget);
// Unless the index is known to be 0, we must slide the vector down to get
// the desired element into index 0.
if (!isNullConstant(Idx)) {
Vec = getVSlidedown(DAG, Subtarget, DL, ContainerVT,
DAG.getUNDEF(ContainerVT), Vec, Idx, Mask, VL);
}
// Extract the lower XLEN bits of the correct vector element.
SDValue EltLo = DAG.getNode(RISCVISD::VMV_X_S, DL, XLenVT, Vec);
// To extract the upper XLEN bits of the vector element, shift the first
// element right by 32 bits and re-extract the lower XLEN bits.
SDValue ThirtyTwoV = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, ContainerVT,
DAG.getUNDEF(ContainerVT),
DAG.getConstant(32, DL, XLenVT), VL);
SDValue LShr32 =
DAG.getNode(RISCVISD::SRL_VL, DL, ContainerVT, Vec, ThirtyTwoV,
DAG.getUNDEF(ContainerVT), Mask, VL);
SDValue EltHi = DAG.getNode(RISCVISD::VMV_X_S, DL, XLenVT, LShr32);
Results.push_back(DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, EltLo, EltHi));
break;
}
case ISD::INTRINSIC_WO_CHAIN: {
unsigned IntNo = N->getConstantOperandVal(0);
switch (IntNo) {
default:
llvm_unreachable(
"Don't know how to custom type legalize this intrinsic!");
case Intrinsic::experimental_get_vector_length: {
SDValue Res = lowerGetVectorLength(N, DAG, Subtarget);
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
case Intrinsic::experimental_cttz_elts: {
SDValue Res = lowerCttzElts(N, DAG, Subtarget);
Results.push_back(
DAG.getNode(ISD::TRUNCATE, DL, N->getValueType(0), Res));
return;
}
case Intrinsic::riscv_orc_b:
case Intrinsic::riscv_brev8:
case Intrinsic::riscv_sha256sig0:
case Intrinsic::riscv_sha256sig1:
case Intrinsic::riscv_sha256sum0:
case Intrinsic::riscv_sha256sum1:
case Intrinsic::riscv_sm3p0:
case Intrinsic::riscv_sm3p1: {
if (!Subtarget.is64Bit() || N->getValueType(0) != MVT::i32)
return;
unsigned Opc;
switch (IntNo) {
case Intrinsic::riscv_orc_b: Opc = RISCVISD::ORC_B; break;
case Intrinsic::riscv_brev8: Opc = RISCVISD::BREV8; break;
case Intrinsic::riscv_sha256sig0: Opc = RISCVISD::SHA256SIG0; break;
case Intrinsic::riscv_sha256sig1: Opc = RISCVISD::SHA256SIG1; break;
case Intrinsic::riscv_sha256sum0: Opc = RISCVISD::SHA256SUM0; break;
case Intrinsic::riscv_sha256sum1: Opc = RISCVISD::SHA256SUM1; break;
case Intrinsic::riscv_sm3p0: Opc = RISCVISD::SM3P0; break;
case Intrinsic::riscv_sm3p1: Opc = RISCVISD::SM3P1; break;
}
SDValue NewOp =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue Res = DAG.getNode(Opc, DL, MVT::i64, NewOp);
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
case Intrinsic::riscv_sm4ks:
case Intrinsic::riscv_sm4ed: {
unsigned Opc =
IntNo == Intrinsic::riscv_sm4ks ? RISCVISD::SM4KS : RISCVISD::SM4ED;
SDValue NewOp0 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue NewOp1 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(2));
SDValue Res =
DAG.getNode(Opc, DL, MVT::i64, NewOp0, NewOp1, N->getOperand(3));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
case Intrinsic::riscv_mopr: {
if (!Subtarget.is64Bit() || N->getValueType(0) != MVT::i32)
return;
SDValue NewOp =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue Res = DAG.getNode(
RISCVISD::MOP_R, DL, MVT::i64, NewOp,
DAG.getTargetConstant(N->getConstantOperandVal(2), DL, MVT::i64));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
case Intrinsic::riscv_moprr: {
if (!Subtarget.is64Bit() || N->getValueType(0) != MVT::i32)
return;
SDValue NewOp0 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue NewOp1 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(2));
SDValue Res = DAG.getNode(
RISCVISD::MOP_RR, DL, MVT::i64, NewOp0, NewOp1,
DAG.getTargetConstant(N->getConstantOperandVal(3), DL, MVT::i64));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
case Intrinsic::riscv_clmul: {
if (!Subtarget.is64Bit() || N->getValueType(0) != MVT::i32)
return;
SDValue NewOp0 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue NewOp1 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(2));
SDValue Res = DAG.getNode(RISCVISD::CLMUL, DL, MVT::i64, NewOp0, NewOp1);
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
case Intrinsic::riscv_clmulh:
case Intrinsic::riscv_clmulr: {
if (!Subtarget.is64Bit() || N->getValueType(0) != MVT::i32)
return;
// Extend inputs to XLen, and shift by 32. This will add 64 trailing zeros
// to the full 128-bit clmul result of multiplying two xlen values.
// Perform clmulr or clmulh on the shifted values. Finally, extract the
// upper 32 bits.
//
// The alternative is to mask the inputs to 32 bits and use clmul, but
// that requires two shifts to mask each input without zext.w.
// FIXME: If the inputs are known zero extended or could be freely
// zero extended, the mask form would be better.
SDValue NewOp0 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
SDValue NewOp1 =
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(2));
NewOp0 = DAG.getNode(ISD::SHL, DL, MVT::i64, NewOp0,
DAG.getConstant(32, DL, MVT::i64));
NewOp1 = DAG.getNode(ISD::SHL, DL, MVT::i64, NewOp1,
DAG.getConstant(32, DL, MVT::i64));
unsigned Opc = IntNo == Intrinsic::riscv_clmulh ? RISCVISD::CLMULH
: RISCVISD::CLMULR;
SDValue Res = DAG.getNode(Opc, DL, MVT::i64, NewOp0, NewOp1);
Res = DAG.getNode(ISD::SRL, DL, MVT::i64, Res,
DAG.getConstant(32, DL, MVT::i64));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Res));
return;
}
case Intrinsic::riscv_vmv_x_s: {
EVT VT = N->getValueType(0);
MVT XLenVT = Subtarget.getXLenVT();
if (VT.bitsLT(XLenVT)) {
// Simple case just extract using vmv.x.s and truncate.
SDValue Extract = DAG.getNode(RISCVISD::VMV_X_S, DL,
Subtarget.getXLenVT(), N->getOperand(1));
Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, VT, Extract));
return;
}
assert(VT == MVT::i64 && !Subtarget.is64Bit() &&
"Unexpected custom legalization");
// We need to do the move in two steps.
SDValue Vec = N->getOperand(1);
MVT VecVT = Vec.getSimpleValueType();
// First extract the lower XLEN bits of the element.
SDValue EltLo = DAG.getNode(RISCVISD::VMV_X_S, DL, XLenVT, Vec);
// To extract the upper XLEN bits of the vector element, shift the first
// element right by 32 bits and re-extract the lower XLEN bits.
auto [Mask, VL] = getDefaultVLOps(1, VecVT, DL, DAG, Subtarget);
SDValue ThirtyTwoV =
DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VecVT, DAG.getUNDEF(VecVT),
DAG.getConstant(32, DL, XLenVT), VL);
SDValue LShr32 = DAG.getNode(RISCVISD::SRL_VL, DL, VecVT, Vec, ThirtyTwoV,
DAG.getUNDEF(VecVT), Mask, VL);
SDValue EltHi = DAG.getNode(RISCVISD::VMV_X_S, DL, XLenVT, LShr32);
Results.push_back(
DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, EltLo, EltHi));
break;
}
}
break;
}
case ISD::VECREDUCE_ADD:
case ISD::VECREDUCE_AND:
case ISD::VECREDUCE_OR:
case ISD::VECREDUCE_XOR:
case ISD::VECREDUCE_SMAX:
case ISD::VECREDUCE_UMAX:
case ISD::VECREDUCE_SMIN:
case ISD::VECREDUCE_UMIN:
if (SDValue V = lowerVECREDUCE(SDValue(N, 0), DAG))
Results.push_back(V);
break;
case ISD::VP_REDUCE_ADD:
case ISD::VP_REDUCE_AND:
case ISD::VP_REDUCE_OR:
case ISD::VP_REDUCE_XOR:
case ISD::VP_REDUCE_SMAX:
case ISD::VP_REDUCE_UMAX:
case ISD::VP_REDUCE_SMIN:
case ISD::VP_REDUCE_UMIN:
if (SDValue V = lowerVPREDUCE(SDValue(N, 0), DAG))
Results.push_back(V);
break;
case ISD::GET_ROUNDING: {
SDVTList VTs = DAG.getVTList(Subtarget.getXLenVT(), MVT::Other);
SDValue Res = DAG.getNode(ISD::GET_ROUNDING, DL, VTs, N->getOperand(0));
Results.push_back(Res.getValue(0));
Results.push_back(Res.getValue(1));
break;
}
}
}
/// Given a binary operator, return the *associative* generic ISD::VECREDUCE_OP
/// which corresponds to it.
static unsigned getVecReduceOpcode(unsigned Opc) {
switch (Opc) {
default:
llvm_unreachable("Unhandled binary to transform reduction");
case ISD::ADD:
return ISD::VECREDUCE_ADD;
case ISD::UMAX:
return ISD::VECREDUCE_UMAX;
case ISD::SMAX:
return ISD::VECREDUCE_SMAX;
case ISD::UMIN:
return ISD::VECREDUCE_UMIN;
case ISD::SMIN:
return ISD::VECREDUCE_SMIN;
case ISD::AND:
return ISD::VECREDUCE_AND;
case ISD::OR:
return ISD::VECREDUCE_OR;
case ISD::XOR:
return ISD::VECREDUCE_XOR;
case ISD::FADD:
// Note: This is the associative form of the generic reduction opcode.
return ISD::VECREDUCE_FADD;
}
}
/// Perform two related transforms whose purpose is to incrementally recognize
/// an explode_vector followed by scalar reduction as a vector reduction node.
/// This exists to recover from a deficiency in SLP which can't handle
/// forests with multiple roots sharing common nodes. In some cases, one
/// of the trees will be vectorized, and the other will remain (unprofitably)
/// scalarized.
static SDValue
combineBinOpOfExtractToReduceTree(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
// This transforms need to run before all integer types have been legalized
// to i64 (so that the vector element type matches the add type), and while
// it's safe to introduce odd sized vector types.
if (DAG.NewNodesMustHaveLegalTypes)
return SDValue();
// Without V, this transform isn't useful. We could form the (illegal)
// operations and let them be scalarized again, but there's really no point.
if (!Subtarget.hasVInstructions())
return SDValue();
const SDLoc DL(N);
const EVT VT = N->getValueType(0);
const unsigned Opc = N->getOpcode();
// For FADD, we only handle the case with reassociation allowed. We
// could handle strict reduction order, but at the moment, there's no
// known reason to, and the complexity isn't worth it.
// TODO: Handle fminnum and fmaxnum here
if (!VT.isInteger() &&
(Opc != ISD::FADD || !N->getFlags().hasAllowReassociation()))
return SDValue();
const unsigned ReduceOpc = getVecReduceOpcode(Opc);
assert(Opc == ISD::getVecReduceBaseOpcode(ReduceOpc) &&
"Inconsistent mappings");
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
if (!LHS.hasOneUse() || !RHS.hasOneUse())
return SDValue();
if (RHS.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
std::swap(LHS, RHS);
if (RHS.getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
!isa<ConstantSDNode>(RHS.getOperand(1)))
return SDValue();
uint64_t RHSIdx = cast<ConstantSDNode>(RHS.getOperand(1))->getLimitedValue();
SDValue SrcVec = RHS.getOperand(0);
EVT SrcVecVT = SrcVec.getValueType();
assert(SrcVecVT.getVectorElementType() == VT);
if (SrcVecVT.isScalableVector())
return SDValue();
if (SrcVecVT.getScalarSizeInBits() > Subtarget.getELen())
return SDValue();
// match binop (extract_vector_elt V, 0), (extract_vector_elt V, 1) to
// reduce_op (extract_subvector [2 x VT] from V). This will form the
// root of our reduction tree. TODO: We could extend this to any two
// adjacent aligned constant indices if desired.
if (LHS.getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
LHS.getOperand(0) == SrcVec && isa<ConstantSDNode>(LHS.getOperand(1))) {
uint64_t LHSIdx =
cast<ConstantSDNode>(LHS.getOperand(1))->getLimitedValue();
if (0 == std::min(LHSIdx, RHSIdx) && 1 == std::max(LHSIdx, RHSIdx)) {
EVT ReduceVT = EVT::getVectorVT(*DAG.getContext(), VT, 2);
SDValue Vec = DAG.getExtractSubvector(DL, ReduceVT, SrcVec, 0);
return DAG.getNode(ReduceOpc, DL, VT, Vec, N->getFlags());
}
}
// Match (binop (reduce (extract_subvector V, 0),
// (extract_vector_elt V, sizeof(SubVec))))
// into a reduction of one more element from the original vector V.
if (LHS.getOpcode() != ReduceOpc)
return SDValue();
SDValue ReduceVec = LHS.getOperand(0);
if (ReduceVec.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
ReduceVec.hasOneUse() && ReduceVec.getOperand(0) == RHS.getOperand(0) &&
isNullConstant(ReduceVec.getOperand(1)) &&
ReduceVec.getValueType().getVectorNumElements() == RHSIdx) {
// For illegal types (e.g. 3xi32), most will be combined again into a
// wider (hopefully legal) type. If this is a terminal state, we are
// relying on type legalization here to produce something reasonable
// and this lowering quality could probably be improved. (TODO)
EVT ReduceVT = EVT::getVectorVT(*DAG.getContext(), VT, RHSIdx + 1);
SDValue Vec = DAG.getExtractSubvector(DL, ReduceVT, SrcVec, 0);
return DAG.getNode(ReduceOpc, DL, VT, Vec,
ReduceVec->getFlags() & N->getFlags());
}
return SDValue();
}
// Try to fold (<bop> x, (reduction.<bop> vec, start))
static SDValue combineBinOpToReduce(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
auto BinOpToRVVReduce = [](unsigned Opc) {
switch (Opc) {
default:
llvm_unreachable("Unhandled binary to transform reduction");
case ISD::ADD:
return RISCVISD::VECREDUCE_ADD_VL;
case ISD::UMAX:
return RISCVISD::VECREDUCE_UMAX_VL;
case ISD::SMAX:
return RISCVISD::VECREDUCE_SMAX_VL;
case ISD::UMIN:
return RISCVISD::VECREDUCE_UMIN_VL;
case ISD::SMIN:
return RISCVISD::VECREDUCE_SMIN_VL;
case ISD::AND:
return RISCVISD::VECREDUCE_AND_VL;
case ISD::OR:
return RISCVISD::VECREDUCE_OR_VL;
case ISD::XOR:
return RISCVISD::VECREDUCE_XOR_VL;
case ISD::FADD:
return RISCVISD::VECREDUCE_FADD_VL;
case ISD::FMAXNUM:
return RISCVISD::VECREDUCE_FMAX_VL;
case ISD::FMINNUM:
return RISCVISD::VECREDUCE_FMIN_VL;
}
};
auto IsReduction = [&BinOpToRVVReduce](SDValue V, unsigned Opc) {
return V.getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
isNullConstant(V.getOperand(1)) &&
V.getOperand(0).getOpcode() == BinOpToRVVReduce(Opc);
};
unsigned Opc = N->getOpcode();
unsigned ReduceIdx;
if (IsReduction(N->getOperand(0), Opc))
ReduceIdx = 0;
else if (IsReduction(N->getOperand(1), Opc))
ReduceIdx = 1;
else
return SDValue();
// Skip if FADD disallows reassociation but the combiner needs.
if (Opc == ISD::FADD && !N->getFlags().hasAllowReassociation())
return SDValue();
SDValue Extract = N->getOperand(ReduceIdx);
SDValue Reduce = Extract.getOperand(0);
if (!Extract.hasOneUse() || !Reduce.hasOneUse())
return SDValue();
SDValue ScalarV = Reduce.getOperand(2);
EVT ScalarVT = ScalarV.getValueType();
if (ScalarV.getOpcode() == ISD::INSERT_SUBVECTOR &&
ScalarV.getOperand(0)->isUndef() &&
isNullConstant(ScalarV.getOperand(2)))
ScalarV = ScalarV.getOperand(1);
// Make sure that ScalarV is a splat with VL=1.
if (ScalarV.getOpcode() != RISCVISD::VFMV_S_F_VL &&
ScalarV.getOpcode() != RISCVISD::VMV_S_X_VL &&
ScalarV.getOpcode() != RISCVISD::VMV_V_X_VL)
return SDValue();
if (!isNonZeroAVL(ScalarV.getOperand(2)))
return SDValue();
// Check the scalar of ScalarV is neutral element
// TODO: Deal with value other than neutral element.
if (!isNeutralConstant(N->getOpcode(), N->getFlags(), ScalarV.getOperand(1),
0))
return SDValue();
// If the AVL is zero, operand 0 will be returned. So it's not safe to fold.
// FIXME: We might be able to improve this if operand 0 is undef.
if (!isNonZeroAVL(Reduce.getOperand(5)))
return SDValue();
SDValue NewStart = N->getOperand(1 - ReduceIdx);
SDLoc DL(N);
SDValue NewScalarV =
lowerScalarInsert(NewStart, ScalarV.getOperand(2),
ScalarV.getSimpleValueType(), DL, DAG, Subtarget);
// If we looked through an INSERT_SUBVECTOR we need to restore it.
if (ScalarVT != ScalarV.getValueType())
NewScalarV =
DAG.getInsertSubvector(DL, DAG.getUNDEF(ScalarVT), NewScalarV, 0);
SDValue Ops[] = {Reduce.getOperand(0), Reduce.getOperand(1),
NewScalarV, Reduce.getOperand(3),
Reduce.getOperand(4), Reduce.getOperand(5)};
SDValue NewReduce =
DAG.getNode(Reduce.getOpcode(), DL, Reduce.getValueType(), Ops);
return DAG.getNode(Extract.getOpcode(), DL, Extract.getValueType(), NewReduce,
Extract.getOperand(1));
}
// Optimize (add (shl x, c0), (shl y, c1)) ->
// (SLLI (SH*ADD x, y), c0), if c1-c0 equals to [1|2|3].
// or
// (SLLI (QC.SHLADD x, y, c1 - c0), c0), if 4 <= (c1-c0) <=31.
static SDValue transformAddShlImm(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
const bool HasStdExtZba = Subtarget.hasStdExtZba();
const bool HasVendorXAndesPerf = Subtarget.hasVendorXAndesPerf();
const bool HasVendorXqciac = Subtarget.hasVendorXqciac();
// Perform this optimization only in the zba/xandesperf/xqciac extension.
if (!HasStdExtZba && !HasVendorXAndesPerf && !HasVendorXqciac)
return SDValue();
// Skip for vector types and larger types.
EVT VT = N->getValueType(0);
if (VT.isVector() || VT.getSizeInBits() > Subtarget.getXLen())
return SDValue();
// The two operand nodes must be SHL and have no other use.
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (N0->getOpcode() != ISD::SHL || N1->getOpcode() != ISD::SHL ||
!N0->hasOneUse() || !N1->hasOneUse())
return SDValue();
// Check c0 and c1.
auto *N0C = dyn_cast<ConstantSDNode>(N0->getOperand(1));
auto *N1C = dyn_cast<ConstantSDNode>(N1->getOperand(1));
if (!N0C || !N1C)
return SDValue();
int64_t C0 = N0C->getSExtValue();
int64_t C1 = N1C->getSExtValue();
if (C0 <= 0 || C1 <= 0)
return SDValue();
int64_t Diff = std::abs(C0 - C1);
bool IsShXaddDiff = Diff == 1 || Diff == 2 || Diff == 3;
bool HasShXadd = HasStdExtZba || HasVendorXAndesPerf;
// Skip if SH1ADD/SH2ADD/SH3ADD are not applicable.
if ((!IsShXaddDiff && HasShXadd && !HasVendorXqciac) ||
(IsShXaddDiff && !HasShXadd && HasVendorXqciac))
return SDValue();
// Skip if QC_SHLADD is not applicable.
if (Diff == 0 || Diff > 31)
return SDValue();
// Build nodes.
SDLoc DL(N);
int64_t Bits = std::min(C0, C1);
SDValue NS = (C0 < C1) ? N0->getOperand(0) : N1->getOperand(0);
SDValue NL = (C0 > C1) ? N0->getOperand(0) : N1->getOperand(0);
SDValue SHADD = DAG.getNode(RISCVISD::SHL_ADD, DL, VT, NL,
DAG.getConstant(Diff, DL, VT), NS);
return DAG.getNode(ISD::SHL, DL, VT, SHADD, DAG.getConstant(Bits, DL, VT));
}
// Check if this SDValue is an add immediate that is fed by a shift of 1, 2,
// or 3.
static SDValue combineShlAddIAddImpl(SDNode *N, SDValue AddI, SDValue Other,
SelectionDAG &DAG) {
using namespace llvm::SDPatternMatch;
// Looking for a reg-reg add and not an addi.
if (isa<ConstantSDNode>(N->getOperand(1)))
return SDValue();
// Based on testing it seems that performance degrades if the ADDI has
// more than 2 uses.
if (AddI->use_size() > 2)
return SDValue();
APInt AddVal;
SDValue SHLVal;
if (!sd_match(AddI, m_Add(m_Value(SHLVal), m_ConstInt(AddVal))))
return SDValue();
APInt VShift;
if (!sd_match(SHLVal, m_OneUse(m_Shl(m_Value(), m_ConstInt(VShift)))))
return SDValue();
if (VShift.slt(1) || VShift.sgt(3))
return SDValue();
SDLoc DL(N);
EVT VT = N->getValueType(0);
// The shift must be positive but the add can be signed.
uint64_t ShlConst = VShift.getZExtValue();
int64_t AddConst = AddVal.getSExtValue();
SDValue SHADD = DAG.getNode(RISCVISD::SHL_ADD, DL, VT, SHLVal->getOperand(0),
DAG.getConstant(ShlConst, DL, VT), Other);
return DAG.getNode(ISD::ADD, DL, VT, SHADD,
DAG.getSignedConstant(AddConst, DL, VT));
}
// Optimize (add (add (shl x, c0), c1), y) ->
// (ADDI (SH*ADD y, x), c1), if c0 equals to [1|2|3].
static SDValue combineShlAddIAdd(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
// Perform this optimization only in the zba extension.
if (!ReassocShlAddiAdd || !Subtarget.hasStdExtZba())
return SDValue();
// Skip for vector types and larger types.
EVT VT = N->getValueType(0);
if (VT != Subtarget.getXLenVT())
return SDValue();
SDValue AddI = N->getOperand(0);
SDValue Other = N->getOperand(1);
if (SDValue V = combineShlAddIAddImpl(N, AddI, Other, DAG))
return V;
if (SDValue V = combineShlAddIAddImpl(N, Other, AddI, DAG))
return V;
return SDValue();
}
// Combine a constant select operand into its use:
//
// (and (select cond, -1, c), x)
// -> (select cond, x, (and x, c)) [AllOnes=1]
// (or (select cond, 0, c), x)
// -> (select cond, x, (or x, c)) [AllOnes=0]
// (xor (select cond, 0, c), x)
// -> (select cond, x, (xor x, c)) [AllOnes=0]
// (add (select cond, 0, c), x)
// -> (select cond, x, (add x, c)) [AllOnes=0]
// (sub x, (select cond, 0, c))
// -> (select cond, x, (sub x, c)) [AllOnes=0]
static SDValue combineSelectAndUse(SDNode *N, SDValue Slct, SDValue OtherOp,
SelectionDAG &DAG, bool AllOnes,
const RISCVSubtarget &Subtarget) {
EVT VT = N->getValueType(0);
// Skip vectors.
if (VT.isVector())
return SDValue();
if (!Subtarget.hasConditionalMoveFusion()) {
// (select cond, x, (and x, c)) has custom lowering with Zicond.
if (!Subtarget.hasCZEROLike() || N->getOpcode() != ISD::AND)
return SDValue();
// Maybe harmful when condition code has multiple use.
if (Slct.getOpcode() == ISD::SELECT && !Slct.getOperand(0).hasOneUse())
return SDValue();
// Maybe harmful when VT is wider than XLen.
if (VT.getSizeInBits() > Subtarget.getXLen())
return SDValue();
}
if ((Slct.getOpcode() != ISD::SELECT &&
Slct.getOpcode() != RISCVISD::SELECT_CC) ||
!Slct.hasOneUse())
return SDValue();
auto isZeroOrAllOnes = [](SDValue N, bool AllOnes) {
return AllOnes ? isAllOnesConstant(N) : isNullConstant(N);
};
bool SwapSelectOps;
unsigned OpOffset = Slct.getOpcode() == RISCVISD::SELECT_CC ? 2 : 0;
SDValue TrueVal = Slct.getOperand(1 + OpOffset);
SDValue FalseVal = Slct.getOperand(2 + OpOffset);
SDValue NonConstantVal;
if (isZeroOrAllOnes(TrueVal, AllOnes)) {
SwapSelectOps = false;
NonConstantVal = FalseVal;
} else if (isZeroOrAllOnes(FalseVal, AllOnes)) {
SwapSelectOps = true;
NonConstantVal = TrueVal;
} else
return SDValue();
// Slct is now know to be the desired identity constant when CC is true.
TrueVal = OtherOp;
FalseVal = DAG.getNode(N->getOpcode(), SDLoc(N), VT, OtherOp, NonConstantVal);
// Unless SwapSelectOps says the condition should be false.
if (SwapSelectOps)
std::swap(TrueVal, FalseVal);
if (Slct.getOpcode() == RISCVISD::SELECT_CC)
return DAG.getNode(RISCVISD::SELECT_CC, SDLoc(N), VT,
{Slct.getOperand(0), Slct.getOperand(1),
Slct.getOperand(2), TrueVal, FalseVal});
return DAG.getNode(ISD::SELECT, SDLoc(N), VT,
{Slct.getOperand(0), TrueVal, FalseVal});
}
// Attempt combineSelectAndUse on each operand of a commutative operator N.
static SDValue combineSelectAndUseCommutative(SDNode *N, SelectionDAG &DAG,
bool AllOnes,
const RISCVSubtarget &Subtarget) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (SDValue Result = combineSelectAndUse(N, N0, N1, DAG, AllOnes, Subtarget))
return Result;
if (SDValue Result = combineSelectAndUse(N, N1, N0, DAG, AllOnes, Subtarget))
return Result;
return SDValue();
}
// Transform (add (mul x, c0), c1) ->
// (add (mul (add x, c1/c0), c0), c1%c0).
// if c1/c0 and c1%c0 are simm12, while c1 is not. A special corner case
// that should be excluded is when c0*(c1/c0) is simm12, which will lead
// to an infinite loop in DAGCombine if transformed.
// Or transform (add (mul x, c0), c1) ->
// (add (mul (add x, c1/c0+1), c0), c1%c0-c0),
// if c1/c0+1 and c1%c0-c0 are simm12, while c1 is not. A special corner
// case that should be excluded is when c0*(c1/c0+1) is simm12, which will
// lead to an infinite loop in DAGCombine if transformed.
// Or transform (add (mul x, c0), c1) ->
// (add (mul (add x, c1/c0-1), c0), c1%c0+c0),
// if c1/c0-1 and c1%c0+c0 are simm12, while c1 is not. A special corner
// case that should be excluded is when c0*(c1/c0-1) is simm12, which will
// lead to an infinite loop in DAGCombine if transformed.
// Or transform (add (mul x, c0), c1) ->
// (mul (add x, c1/c0), c0).
// if c1%c0 is zero, and c1/c0 is simm12 while c1 is not.
static SDValue transformAddImmMulImm(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
// Skip for vector types and larger types.
EVT VT = N->getValueType(0);
if (VT.isVector() || VT.getSizeInBits() > Subtarget.getXLen())
return SDValue();
// The first operand node must be a MUL and has no other use.
SDValue N0 = N->getOperand(0);
if (!N0->hasOneUse() || N0->getOpcode() != ISD::MUL)
return SDValue();
// Check if c0 and c1 match above conditions.
auto *N0C = dyn_cast<ConstantSDNode>(N0->getOperand(1));
auto *N1C = dyn_cast<ConstantSDNode>(N->getOperand(1));
if (!N0C || !N1C)
return SDValue();
// If N0C has multiple uses it's possible one of the cases in
// DAGCombiner::isMulAddWithConstProfitable will be true, which would result
// in an infinite loop.
if (!N0C->hasOneUse())
return SDValue();
int64_t C0 = N0C->getSExtValue();
int64_t C1 = N1C->getSExtValue();
int64_t CA, CB;
if (C0 == -1 || C0 == 0 || C0 == 1 || isInt<12>(C1))
return SDValue();
// Search for proper CA (non-zero) and CB that both are simm12.
if ((C1 / C0) != 0 && isInt<12>(C1 / C0) && isInt<12>(C1 % C0) &&
!isInt<12>(C0 * (C1 / C0))) {
CA = C1 / C0;
CB = C1 % C0;
} else if ((C1 / C0 + 1) != 0 && isInt<12>(C1 / C0 + 1) &&
isInt<12>(C1 % C0 - C0) && !isInt<12>(C0 * (C1 / C0 + 1))) {
CA = C1 / C0 + 1;
CB = C1 % C0 - C0;
} else if ((C1 / C0 - 1) != 0 && isInt<12>(C1 / C0 - 1) &&
isInt<12>(C1 % C0 + C0) && !isInt<12>(C0 * (C1 / C0 - 1))) {
CA = C1 / C0 - 1;
CB = C1 % C0 + C0;
} else
return SDValue();
// Build new nodes (add (mul (add x, c1/c0), c0), c1%c0).
SDLoc DL(N);
SDValue New0 = DAG.getNode(ISD::ADD, DL, VT, N0->getOperand(0),
DAG.getSignedConstant(CA, DL, VT));
SDValue New1 =
DAG.getNode(ISD::MUL, DL, VT, New0, DAG.getSignedConstant(C0, DL, VT));
return DAG.getNode(ISD::ADD, DL, VT, New1, DAG.getSignedConstant(CB, DL, VT));
}
// add (zext, zext) -> zext (add (zext, zext))
// sub (zext, zext) -> sext (sub (zext, zext))
// mul (zext, zext) -> zext (mul (zext, zext))
// sdiv (zext, zext) -> zext (sdiv (zext, zext))
// udiv (zext, zext) -> zext (udiv (zext, zext))
// srem (zext, zext) -> zext (srem (zext, zext))
// urem (zext, zext) -> zext (urem (zext, zext))
//
// where the sum of the extend widths match, and the the range of the bin op
// fits inside the width of the narrower bin op. (For profitability on rvv, we
// use a power of two for both inner and outer extend.)
static SDValue combineBinOpOfZExt(SDNode *N, SelectionDAG &DAG) {
EVT VT = N->getValueType(0);
if (!VT.isVector() || !DAG.getTargetLoweringInfo().isTypeLegal(VT))
return SDValue();
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (N0.getOpcode() != ISD::ZERO_EXTEND || N1.getOpcode() != ISD::ZERO_EXTEND)
return SDValue();
if (!N0.hasOneUse() || !N1.hasOneUse())
return SDValue();
SDValue Src0 = N0.getOperand(0);
SDValue Src1 = N1.getOperand(0);
EVT SrcVT = Src0.getValueType();
if (!DAG.getTargetLoweringInfo().isTypeLegal(SrcVT) ||
SrcVT != Src1.getValueType() || SrcVT.getScalarSizeInBits() < 8 ||
SrcVT.getScalarSizeInBits() >= VT.getScalarSizeInBits() / 2)
return SDValue();
LLVMContext &C = *DAG.getContext();
EVT ElemVT = VT.getVectorElementType().getHalfSizedIntegerVT(C);
EVT NarrowVT = EVT::getVectorVT(C, ElemVT, VT.getVectorElementCount());
Src0 = DAG.getNode(ISD::ZERO_EXTEND, SDLoc(Src0), NarrowVT, Src0);
Src1 = DAG.getNode(ISD::ZERO_EXTEND, SDLoc(Src1), NarrowVT, Src1);
// Src0 and Src1 are zero extended, so they're always positive if signed.
//
// sub can produce a negative from two positive operands, so it needs sign
// extended. Other nodes produce a positive from two positive operands, so
// zero extend instead.
unsigned OuterExtend =
N->getOpcode() == ISD::SUB ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
return DAG.getNode(
OuterExtend, SDLoc(N), VT,
DAG.getNode(N->getOpcode(), SDLoc(N), NarrowVT, Src0, Src1));
}
// Try to turn (add (xor bool, 1) -1) into (neg bool).
static SDValue combineAddOfBooleanXor(SDNode *N, SelectionDAG &DAG) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N->getValueType(0);
SDLoc DL(N);
// RHS should be -1.
if (!isAllOnesConstant(N1))
return SDValue();
// Look for (xor X, 1).
if (N0.getOpcode() != ISD::XOR || !isOneConstant(N0.getOperand(1)))
return SDValue();
// First xor input should be 0 or 1.
APInt Mask = APInt::getBitsSetFrom(VT.getSizeInBits(), 1);
if (!DAG.MaskedValueIsZero(N0.getOperand(0), Mask))
return SDValue();
// Emit a negate of the setcc.
return DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, DL, VT),
N0.getOperand(0));
}
static SDValue performADDCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
SelectionDAG &DAG = DCI.DAG;
if (SDValue V = combineAddOfBooleanXor(N, DAG))
return V;
if (SDValue V = transformAddImmMulImm(N, DAG, Subtarget))
return V;
if (!DCI.isBeforeLegalize() && !DCI.isCalledByLegalizer()) {
if (SDValue V = transformAddShlImm(N, DAG, Subtarget))
return V;
if (SDValue V = combineShlAddIAdd(N, DAG, Subtarget))
return V;
}
if (SDValue V = combineBinOpToReduce(N, DAG, Subtarget))
return V;
if (SDValue V = combineBinOpOfExtractToReduceTree(N, DAG, Subtarget))
return V;
if (SDValue V = combineBinOpOfZExt(N, DAG))
return V;
// fold (add (select lhs, rhs, cc, 0, y), x) ->
// (select lhs, rhs, cc, x, (add x, y))
return combineSelectAndUseCommutative(N, DAG, /*AllOnes*/ false, Subtarget);
}
// Try to turn a sub boolean RHS and constant LHS into an addi.
static SDValue combineSubOfBoolean(SDNode *N, SelectionDAG &DAG) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N->getValueType(0);
SDLoc DL(N);
// Require a constant LHS.
auto *N0C = dyn_cast<ConstantSDNode>(N0);
if (!N0C)
return SDValue();
// All our optimizations involve subtracting 1 from the immediate and forming
// an ADDI. Make sure the new immediate is valid for an ADDI.
APInt ImmValMinus1 = N0C->getAPIntValue() - 1;
if (!ImmValMinus1.isSignedIntN(12))
return SDValue();
SDValue NewLHS;
if (N1.getOpcode() == ISD::SETCC && N1.hasOneUse()) {
// (sub constant, (setcc x, y, eq/neq)) ->
// (add (setcc x, y, neq/eq), constant - 1)
ISD::CondCode CCVal = cast<CondCodeSDNode>(N1.getOperand(2))->get();
EVT SetCCOpVT = N1.getOperand(0).getValueType();
if (!isIntEqualitySetCC(CCVal) || !SetCCOpVT.isInteger())
return SDValue();
CCVal = ISD::getSetCCInverse(CCVal, SetCCOpVT);
NewLHS =
DAG.getSetCC(SDLoc(N1), VT, N1.getOperand(0), N1.getOperand(1), CCVal);
} else if (N1.getOpcode() == ISD::XOR && isOneConstant(N1.getOperand(1)) &&
N1.getOperand(0).getOpcode() == ISD::SETCC) {
// (sub C, (xor (setcc), 1)) -> (add (setcc), C-1).
// Since setcc returns a bool the xor is equivalent to 1-setcc.
NewLHS = N1.getOperand(0);
} else
return SDValue();
SDValue NewRHS = DAG.getConstant(ImmValMinus1, DL, VT);
return DAG.getNode(ISD::ADD, DL, VT, NewLHS, NewRHS);
}
// Looks for (sub (shl X, 8-Y), (shr X, Y)) where the Y-th bit in each byte is
// potentially set. It is fine for Y to be 0, meaning that (sub (shl X, 8), X)
// is also valid. Replace with (orc.b X). For example, 0b0000_1000_0000_1000 is
// valid with Y=3, while 0b0000_1000_0000_0100 is not.
static SDValue combineSubShiftToOrcB(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (!Subtarget.hasStdExtZbb())
return SDValue();
EVT VT = N->getValueType(0);
if (VT != Subtarget.getXLenVT() && VT != MVT::i32 && VT != MVT::i16)
return SDValue();
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (N0->getOpcode() != ISD::SHL)
return SDValue();
auto *ShAmtCLeft = dyn_cast<ConstantSDNode>(N0.getOperand(1));
if (!ShAmtCLeft)
return SDValue();
unsigned ShiftedAmount = 8 - ShAmtCLeft->getZExtValue();
if (ShiftedAmount >= 8)
return SDValue();
SDValue LeftShiftOperand = N0->getOperand(0);
SDValue RightShiftOperand = N1;
if (ShiftedAmount != 0) { // Right operand must be a right shift.
if (N1->getOpcode() != ISD::SRL)
return SDValue();
auto *ShAmtCRight = dyn_cast<ConstantSDNode>(N1.getOperand(1));
if (!ShAmtCRight || ShAmtCRight->getZExtValue() != ShiftedAmount)
return SDValue();
RightShiftOperand = N1.getOperand(0);
}
// At least one shift should have a single use.
if (!N0.hasOneUse() && (ShiftedAmount == 0 || !N1.hasOneUse()))
return SDValue();
if (LeftShiftOperand != RightShiftOperand)
return SDValue();
APInt Mask = APInt::getSplat(VT.getSizeInBits(), APInt(8, 0x1));
Mask <<= ShiftedAmount;
// Check that X has indeed the right shape (only the Y-th bit can be set in
// every byte).
if (!DAG.MaskedValueIsZero(LeftShiftOperand, ~Mask))
return SDValue();
return DAG.getNode(RISCVISD::ORC_B, SDLoc(N), VT, LeftShiftOperand);
}
static SDValue performSUBCombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (SDValue V = combineSubOfBoolean(N, DAG))
return V;
EVT VT = N->getValueType(0);
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
// fold (sub 0, (setcc x, 0, setlt)) -> (sra x, xlen - 1)
if (isNullConstant(N0) && N1.getOpcode() == ISD::SETCC && N1.hasOneUse() &&
isNullConstant(N1.getOperand(1))) {
ISD::CondCode CCVal = cast<CondCodeSDNode>(N1.getOperand(2))->get();
if (CCVal == ISD::SETLT) {
SDLoc DL(N);
unsigned ShAmt = N0.getValueSizeInBits() - 1;
return DAG.getNode(ISD::SRA, DL, VT, N1.getOperand(0),
DAG.getConstant(ShAmt, DL, VT));
}
}
if (SDValue V = combineBinOpOfZExt(N, DAG))
return V;
if (SDValue V = combineSubShiftToOrcB(N, DAG, Subtarget))
return V;
// fold (sub x, (select lhs, rhs, cc, 0, y)) ->
// (select lhs, rhs, cc, x, (sub x, y))
return combineSelectAndUse(N, N1, N0, DAG, /*AllOnes*/ false, Subtarget);
}
// Apply DeMorgan's law to (and/or (xor X, 1), (xor Y, 1)) if X and Y are 0/1.
// Legalizing setcc can introduce xors like this. Doing this transform reduces
// the number of xors and may allow the xor to fold into a branch condition.
static SDValue combineDeMorganOfBoolean(SDNode *N, SelectionDAG &DAG) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
bool IsAnd = N->getOpcode() == ISD::AND;
if (N0.getOpcode() != ISD::XOR || N1.getOpcode() != ISD::XOR)
return SDValue();
if (!N0.hasOneUse() || !N1.hasOneUse())
return SDValue();
SDValue N01 = N0.getOperand(1);
SDValue N11 = N1.getOperand(1);
// For AND, SimplifyDemandedBits may have turned one of the (xor X, 1) into
// (xor X, -1) based on the upper bits of the other operand being 0. If the
// operation is And, allow one of the Xors to use -1.
if (isOneConstant(N01)) {
if (!isOneConstant(N11) && !(IsAnd && isAllOnesConstant(N11)))
return SDValue();
} else if (isOneConstant(N11)) {
// N01 and N11 being 1 was already handled. Handle N11==1 and N01==-1.
if (!(IsAnd && isAllOnesConstant(N01)))
return SDValue();
} else
return SDValue();
EVT VT = N->getValueType(0);
SDValue N00 = N0.getOperand(0);
SDValue N10 = N1.getOperand(0);
// The LHS of the xors needs to be 0/1.
APInt Mask = APInt::getBitsSetFrom(VT.getSizeInBits(), 1);
if (!DAG.MaskedValueIsZero(N00, Mask) || !DAG.MaskedValueIsZero(N10, Mask))
return SDValue();
// Invert the opcode and insert a new xor.
SDLoc DL(N);
unsigned Opc = IsAnd ? ISD::OR : ISD::AND;
SDValue Logic = DAG.getNode(Opc, DL, VT, N00, N10);
return DAG.getNode(ISD::XOR, DL, VT, Logic, DAG.getConstant(1, DL, VT));
}
// Fold (vXi8 (trunc (vselect (setltu, X, 256), X, (sext (setgt X, 0))))) to
// (vXi8 (trunc (smin (smax X, 0), 255))). This represents saturating a signed
// value to an unsigned value. This will be lowered to vmax and series of
// vnclipu instructions later. This can be extended to other truncated types
// other than i8 by replacing 256 and 255 with the equivalent constants for the
// type.
static SDValue combineTruncSelectToSMaxUSat(SDNode *N, SelectionDAG &DAG) {
EVT VT = N->getValueType(0);
SDValue N0 = N->getOperand(0);
EVT SrcVT = N0.getValueType();
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (!VT.isVector() || !TLI.isTypeLegal(VT) || !TLI.isTypeLegal(SrcVT))
return SDValue();
if (N0.getOpcode() != ISD::VSELECT || !N0.hasOneUse())
return SDValue();
SDValue Cond = N0.getOperand(0);
SDValue True = N0.getOperand(1);
SDValue False = N0.getOperand(2);
if (Cond.getOpcode() != ISD::SETCC)
return SDValue();
// FIXME: Support the version of this pattern with the select operands
// swapped.
ISD::CondCode CCVal = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
if (CCVal != ISD::SETULT)
return SDValue();
SDValue CondLHS = Cond.getOperand(0);
SDValue CondRHS = Cond.getOperand(1);
if (CondLHS != True)
return SDValue();
unsigned ScalarBits = VT.getScalarSizeInBits();
// FIXME: Support other constants.
ConstantSDNode *CondRHSC = isConstOrConstSplat(CondRHS);
if (!CondRHSC || CondRHSC->getAPIntValue() != (1ULL << ScalarBits))
return SDValue();
if (False.getOpcode() != ISD::SIGN_EXTEND)
return SDValue();
False = False.getOperand(0);
if (False.getOpcode() != ISD::SETCC || False.getOperand(0) != True)
return SDValue();
ConstantSDNode *FalseRHSC = isConstOrConstSplat(False.getOperand(1));
if (!FalseRHSC || !FalseRHSC->isZero())
return SDValue();
ISD::CondCode CCVal2 = cast<CondCodeSDNode>(False.getOperand(2))->get();
if (CCVal2 != ISD::SETGT)
return SDValue();
// Emit the signed to unsigned saturation pattern.
SDLoc DL(N);
SDValue Max =
DAG.getNode(ISD::SMAX, DL, SrcVT, True, DAG.getConstant(0, DL, SrcVT));
SDValue Min =
DAG.getNode(ISD::SMIN, DL, SrcVT, Max,
DAG.getConstant((1ULL << ScalarBits) - 1, DL, SrcVT));
return DAG.getNode(ISD::TRUNCATE, DL, VT, Min);
}
static SDValue performTRUNCATECombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
// Pre-promote (i1 (truncate (srl X, Y))) on RV64 with Zbs without zero
// extending X. This is safe since we only need the LSB after the shift and
// shift amounts larger than 31 would produce poison. If we wait until
// type legalization, we'll create RISCVISD::SRLW and we can't recover it
// to use a BEXT instruction.
if (Subtarget.is64Bit() && Subtarget.hasStdExtZbs() && VT == MVT::i1 &&
N0.getValueType() == MVT::i32 && N0.getOpcode() == ISD::SRL &&
!isa<ConstantSDNode>(N0.getOperand(1)) && N0.hasOneUse()) {
SDLoc DL(N0);
SDValue Op0 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N0.getOperand(0));
SDValue Op1 = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, N0.getOperand(1));
SDValue Srl = DAG.getNode(ISD::SRL, DL, MVT::i64, Op0, Op1);
return DAG.getNode(ISD::TRUNCATE, SDLoc(N), VT, Srl);
}
return combineTruncSelectToSMaxUSat(N, DAG);
}
// InstCombinerImpl::transformZExtICmp will narrow a zext of an icmp with a
// truncation. But RVV doesn't have truncation instructions for more than twice
// the bitwidth.
//
// E.g. trunc <vscale x 1 x i64> %x to <vscale x 1 x i8> will generate:
//
// vsetvli a0, zero, e32, m2, ta, ma
// vnsrl.wi v12, v8, 0
// vsetvli zero, zero, e16, m1, ta, ma
// vnsrl.wi v8, v12, 0
// vsetvli zero, zero, e8, mf2, ta, ma
// vnsrl.wi v8, v8, 0
//
// So reverse the combine so we generate an vmseq/vmsne again:
//
// and (lshr (trunc X), ShAmt), 1
// -->
// zext (icmp ne (and X, (1 << ShAmt)), 0)
//
// and (lshr (not (trunc X)), ShAmt), 1
// -->
// zext (icmp eq (and X, (1 << ShAmt)), 0)
static SDValue reverseZExtICmpCombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
using namespace SDPatternMatch;
SDLoc DL(N);
if (!Subtarget.hasVInstructions())
return SDValue();
EVT VT = N->getValueType(0);
if (!VT.isVector())
return SDValue();
APInt ShAmt;
SDValue Inner;
if (!sd_match(N, m_And(m_OneUse(m_Srl(m_Value(Inner), m_ConstInt(ShAmt))),
m_One())))
return SDValue();
SDValue X;
bool IsNot;
if (sd_match(Inner, m_Not(m_Trunc(m_Value(X)))))
IsNot = true;
else if (sd_match(Inner, m_Trunc(m_Value(X))))
IsNot = false;
else
return SDValue();
EVT WideVT = X.getValueType();
if (VT.getScalarSizeInBits() >= WideVT.getScalarSizeInBits() / 2)
return SDValue();
SDValue Res =
DAG.getNode(ISD::AND, DL, WideVT, X,
DAG.getConstant(1ULL << ShAmt.getZExtValue(), DL, WideVT));
Res = DAG.getSetCC(DL,
EVT::getVectorVT(*DAG.getContext(), MVT::i1,
WideVT.getVectorElementCount()),
Res, DAG.getConstant(0, DL, WideVT),
IsNot ? ISD::SETEQ : ISD::SETNE);
return DAG.getNode(ISD::ZERO_EXTEND, DL, VT, Res);
}
static SDValue reduceANDOfAtomicLoad(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
SelectionDAG &DAG = DCI.DAG;
if (N->getOpcode() != ISD::AND)
return SDValue();
SDValue N0 = N->getOperand(0);
if (N0.getOpcode() != ISD::ATOMIC_LOAD)
return SDValue();
if (!N0.hasOneUse())
return SDValue();
AtomicSDNode *ALoad = cast<AtomicSDNode>(N0.getNode());
if (isStrongerThanMonotonic(ALoad->getSuccessOrdering()))
return SDValue();
EVT LoadedVT = ALoad->getMemoryVT();
ConstantSDNode *MaskConst = dyn_cast<ConstantSDNode>(N->getOperand(1));
if (!MaskConst)
return SDValue();
uint64_t Mask = MaskConst->getZExtValue();
uint64_t ExpectedMask = maskTrailingOnes<uint64_t>(LoadedVT.getSizeInBits());
if (Mask != ExpectedMask)
return SDValue();
SDValue ZextLoad = DAG.getAtomicLoad(
ISD::ZEXTLOAD, SDLoc(N), ALoad->getMemoryVT(), N->getValueType(0),
ALoad->getChain(), ALoad->getBasePtr(), ALoad->getMemOperand());
DCI.CombineTo(N, ZextLoad);
DAG.ReplaceAllUsesOfValueWith(SDValue(N0.getNode(), 1), ZextLoad.getValue(1));
DCI.recursivelyDeleteUnusedNodes(N0.getNode());
return SDValue(N, 0);
}
// Combines two comparison operation and logic operation to one selection
// operation(min, max) and logic operation. Returns new constructed Node if
// conditions for optimization are satisfied.
static SDValue performANDCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
SelectionDAG &DAG = DCI.DAG;
SDValue N0 = N->getOperand(0);
// Pre-promote (i32 (and (srl X, Y), 1)) on RV64 with Zbs without zero
// extending X. This is safe since we only need the LSB after the shift and
// shift amounts larger than 31 would produce poison. If we wait until
// type legalization, we'll create RISCVISD::SRLW and we can't recover it
// to use a BEXT instruction.
if (Subtarget.is64Bit() && Subtarget.hasStdExtZbs() &&
N->getValueType(0) == MVT::i32 && isOneConstant(N->getOperand(1)) &&
N0.getOpcode() == ISD::SRL && !isa<ConstantSDNode>(N0.getOperand(1)) &&
N0.hasOneUse()) {
SDLoc DL(N);
SDValue Op0 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N0.getOperand(0));
SDValue Op1 = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, N0.getOperand(1));
SDValue Srl = DAG.getNode(ISD::SRL, DL, MVT::i64, Op0, Op1);
SDValue And = DAG.getNode(ISD::AND, DL, MVT::i64, Srl,
DAG.getConstant(1, DL, MVT::i64));
return DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, And);
}
if (SDValue V = reverseZExtICmpCombine(N, DAG, Subtarget))
return V;
if (SDValue V = combineBinOpToReduce(N, DAG, Subtarget))
return V;
if (SDValue V = combineBinOpOfExtractToReduceTree(N, DAG, Subtarget))
return V;
if (SDValue V = reduceANDOfAtomicLoad(N, DCI))
return V;
if (DCI.isAfterLegalizeDAG())
if (SDValue V = combineDeMorganOfBoolean(N, DAG))
return V;
// fold (and (select lhs, rhs, cc, -1, y), x) ->
// (select lhs, rhs, cc, x, (and x, y))
return combineSelectAndUseCommutative(N, DAG, /*AllOnes*/ true, Subtarget);
}
// Try to pull an xor with 1 through a select idiom that uses czero_eqz/nez.
// FIXME: Generalize to other binary operators with same operand.
static SDValue combineOrOfCZERO(SDNode *N, SDValue N0, SDValue N1,
SelectionDAG &DAG) {
assert(N->getOpcode() == ISD::OR && "Unexpected opcode");
if (N0.getOpcode() != RISCVISD::CZERO_EQZ ||
N1.getOpcode() != RISCVISD::CZERO_NEZ ||
!N0.hasOneUse() || !N1.hasOneUse())
return SDValue();
// Should have the same condition.
SDValue Cond = N0.getOperand(1);
if (Cond != N1.getOperand(1))
return SDValue();
SDValue TrueV = N0.getOperand(0);
SDValue FalseV = N1.getOperand(0);
if (TrueV.getOpcode() != ISD::XOR || FalseV.getOpcode() != ISD::XOR ||
TrueV.getOperand(1) != FalseV.getOperand(1) ||
!isOneConstant(TrueV.getOperand(1)) ||
!TrueV.hasOneUse() || !FalseV.hasOneUse())
return SDValue();
EVT VT = N->getValueType(0);
SDLoc DL(N);
SDValue NewN0 = DAG.getNode(RISCVISD::CZERO_EQZ, DL, VT, TrueV.getOperand(0),
Cond);
SDValue NewN1 =
DAG.getNode(RISCVISD::CZERO_NEZ, DL, VT, FalseV.getOperand(0), Cond);
SDValue NewOr =
DAG.getNode(ISD::OR, DL, VT, NewN0, NewN1, SDNodeFlags::Disjoint);
return DAG.getNode(ISD::XOR, DL, VT, NewOr, TrueV.getOperand(1));
}
// (xor X, (xor (and X, C2), Y))
// ->(qc_insb X, (sra Y, ShAmt), Width, ShAmt)
// where C2 is a shifted mask with width = Width and shift = ShAmt
// qc_insb might become qc.insb or qc.insbi depending on the operands.
static SDValue combineXorToBitfieldInsert(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (!Subtarget.hasVendorXqcibm())
return SDValue();
using namespace SDPatternMatch;
SDValue Base, Inserted;
APInt CMask;
if (!sd_match(N, m_Xor(m_Value(Base),
m_OneUse(m_Xor(m_OneUse(m_And(m_Deferred(Base),
m_ConstInt(CMask))),
m_Value(Inserted))))))
return SDValue();
if (N->getValueType(0) != MVT::i32)
return SDValue();
unsigned Width, ShAmt;
if (!CMask.isShiftedMask(ShAmt, Width))
return SDValue();
// Check if all zero bits in CMask are also zero in Inserted
if (!DAG.MaskedValueIsZero(Inserted, ~CMask))
return SDValue();
SDLoc DL(N);
// `Inserted` needs to be right shifted before it is put into the
// instruction.
Inserted = DAG.getNode(ISD::SRA, DL, MVT::i32, Inserted,
DAG.getShiftAmountConstant(ShAmt, MVT::i32, DL));
SDValue Ops[] = {Base, Inserted, DAG.getConstant(Width, DL, MVT::i32),
DAG.getConstant(ShAmt, DL, MVT::i32)};
return DAG.getNode(RISCVISD::QC_INSB, DL, MVT::i32, Ops);
}
static SDValue performORCombine(SDNode *N, TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
SelectionDAG &DAG = DCI.DAG;
if (SDValue V = combineBinOpToReduce(N, DAG, Subtarget))
return V;
if (SDValue V = combineBinOpOfExtractToReduceTree(N, DAG, Subtarget))
return V;
if (DCI.isAfterLegalizeDAG())
if (SDValue V = combineDeMorganOfBoolean(N, DAG))
return V;
// Look for Or of CZERO_EQZ/NEZ with same condition which is the select idiom.
// We may be able to pull a common operation out of the true and false value.
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (SDValue V = combineOrOfCZERO(N, N0, N1, DAG))
return V;
if (SDValue V = combineOrOfCZERO(N, N1, N0, DAG))
return V;
// fold (or (select cond, 0, y), x) ->
// (select cond, x, (or x, y))
return combineSelectAndUseCommutative(N, DAG, /*AllOnes*/ false, Subtarget);
}
static SDValue performXORCombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
// Pre-promote (i32 (xor (shl -1, X), ~0)) on RV64 with Zbs so we can use
// (ADDI (BSET X0, X), -1). If we wait until/ type legalization, we'll create
// RISCVISD:::SLLW and we can't recover it to use a BSET instruction.
if (Subtarget.is64Bit() && Subtarget.hasStdExtZbs() &&
N->getValueType(0) == MVT::i32 && isAllOnesConstant(N1) &&
N0.getOpcode() == ISD::SHL && isAllOnesConstant(N0.getOperand(0)) &&
!isa<ConstantSDNode>(N0.getOperand(1)) && N0.hasOneUse()) {
SDLoc DL(N);
SDValue Op0 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N0.getOperand(0));
SDValue Op1 = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, N0.getOperand(1));
SDValue Shl = DAG.getNode(ISD::SHL, DL, MVT::i64, Op0, Op1);
SDValue Not = DAG.getNOT(DL, Shl, MVT::i64);
return DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Not);
}
// fold (xor (sllw 1, x), -1) -> (rolw ~1, x)
// NOTE: Assumes ROL being legal means ROLW is legal.
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (N0.getOpcode() == RISCVISD::SLLW &&
isAllOnesConstant(N1) && isOneConstant(N0.getOperand(0)) &&
TLI.isOperationLegal(ISD::ROTL, MVT::i64)) {
SDLoc DL(N);
return DAG.getNode(RISCVISD::ROLW, DL, MVT::i64,
DAG.getConstant(~1, DL, MVT::i64), N0.getOperand(1));
}
// Fold (xor (setcc constant, y, setlt), 1) -> (setcc y, constant + 1, setlt)
if (N0.getOpcode() == ISD::SETCC && isOneConstant(N1) && N0.hasOneUse()) {
auto *ConstN00 = dyn_cast<ConstantSDNode>(N0.getOperand(0));
ISD::CondCode CC = cast<CondCodeSDNode>(N0.getOperand(2))->get();
if (ConstN00 && CC == ISD::SETLT) {
EVT VT = N0.getValueType();
SDLoc DL(N0);
const APInt &Imm = ConstN00->getAPIntValue();
if ((Imm + 1).isSignedIntN(12))
return DAG.getSetCC(DL, VT, N0.getOperand(1),
DAG.getConstant(Imm + 1, DL, VT), CC);
}
}
if (SDValue V = combineXorToBitfieldInsert(N, DAG, Subtarget))
return V;
if (SDValue V = combineBinOpToReduce(N, DAG, Subtarget))
return V;
if (SDValue V = combineBinOpOfExtractToReduceTree(N, DAG, Subtarget))
return V;
// fold (xor (select cond, 0, y), x) ->
// (select cond, x, (xor x, y))
return combineSelectAndUseCommutative(N, DAG, /*AllOnes*/ false, Subtarget);
}
// Try to expand a multiply to a sequence of shifts and add/subs,
// for a machine without native mul instruction.
static SDValue expandMulToNAFSequence(SDNode *N, SelectionDAG &DAG,
uint64_t MulAmt) {
SDLoc DL(N);
EVT VT = N->getValueType(0);
const uint64_t BitWidth = VT.getFixedSizeInBits();
SDValue Result = DAG.getConstant(0, DL, N->getValueType(0));
SDValue N0 = N->getOperand(0);
// Find the Non-adjacent form of the multiplier.
for (uint64_t E = MulAmt, I = 0; E && I < BitWidth; ++I, E >>= 1) {
if (E & 1) {
bool IsAdd = (E & 3) == 1;
E -= IsAdd ? 1 : -1;
SDValue ShiftVal = DAG.getNode(ISD::SHL, DL, VT, N0,
DAG.getShiftAmountConstant(I, VT, DL));
ISD::NodeType AddSubOp = IsAdd ? ISD::ADD : ISD::SUB;
Result = DAG.getNode(AddSubOp, DL, VT, Result, ShiftVal);
}
}
return Result;
}
// X * (2^N +/- 2^M) -> (add/sub (shl X, C1), (shl X, C2))
static SDValue expandMulToAddOrSubOfShl(SDNode *N, SelectionDAG &DAG,
uint64_t MulAmt) {
uint64_t MulAmtLowBit = MulAmt & (-MulAmt);
ISD::NodeType Op;
uint64_t ShiftAmt1;
if (isPowerOf2_64(MulAmt + MulAmtLowBit)) {
Op = ISD::SUB;
ShiftAmt1 = MulAmt + MulAmtLowBit;
} else if (isPowerOf2_64(MulAmt - MulAmtLowBit)) {
Op = ISD::ADD;
ShiftAmt1 = MulAmt - MulAmtLowBit;
} else {
return SDValue();
}
EVT VT = N->getValueType(0);
SDLoc DL(N);
SDValue Shift1 = DAG.getNode(ISD::SHL, DL, VT, N->getOperand(0),
DAG.getConstant(Log2_64(ShiftAmt1), DL, VT));
SDValue Shift2 = DAG.getNode(ISD::SHL, DL, VT, N->getOperand(0),
DAG.getConstant(Log2_64(MulAmtLowBit), DL, VT));
return DAG.getNode(Op, DL, VT, Shift1, Shift2);
}
// Try to expand a scalar multiply to a faster sequence.
static SDValue expandMul(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
EVT VT = N->getValueType(0);
// LI + MUL is usually smaller than the alternative sequence.
if (DAG.getMachineFunction().getFunction().hasMinSize())
return SDValue();
if (VT != Subtarget.getXLenVT())
return SDValue();
bool ShouldExpandMul =
(!DCI.isBeforeLegalize() && !DCI.isCalledByLegalizer()) ||
!Subtarget.hasStdExtZmmul();
if (!ShouldExpandMul)
return SDValue();
ConstantSDNode *CNode = dyn_cast<ConstantSDNode>(N->getOperand(1));
if (!CNode)
return SDValue();
uint64_t MulAmt = CNode->getZExtValue();
// Don't do this if the Xqciac extension is enabled and the MulAmt in simm12.
if (Subtarget.hasVendorXqciac() && isInt<12>(CNode->getSExtValue()))
return SDValue();
const bool HasShlAdd = Subtarget.hasStdExtZba() ||
Subtarget.hasVendorXTHeadBa() ||
Subtarget.hasVendorXAndesPerf();
// WARNING: The code below is knowingly incorrect with regards to undef semantics.
// We're adding additional uses of X here, and in principle, we should be freezing
// X before doing so. However, adding freeze here causes real regressions, and no
// other target properly freezes X in these cases either.
SDValue X = N->getOperand(0);
if (HasShlAdd) {
for (uint64_t Divisor : {3, 5, 9}) {
if (MulAmt % Divisor != 0)
continue;
uint64_t MulAmt2 = MulAmt / Divisor;
// 3/5/9 * 2^N -> shl (shXadd X, X), N
if (isPowerOf2_64(MulAmt2)) {
SDLoc DL(N);
SDValue X = N->getOperand(0);
// Put the shift first if we can fold a zext into the
// shift forming a slli.uw.
if (X.getOpcode() == ISD::AND && isa<ConstantSDNode>(X.getOperand(1)) &&
X.getConstantOperandVal(1) == UINT64_C(0xffffffff)) {
SDValue Shl = DAG.getNode(ISD::SHL, DL, VT, X,
DAG.getConstant(Log2_64(MulAmt2), DL, VT));
return DAG.getNode(RISCVISD::SHL_ADD, DL, VT, Shl,
DAG.getConstant(Log2_64(Divisor - 1), DL, VT),
Shl);
}
// Otherwise, put rhe shl second so that it can fold with following
// instructions (e.g. sext or add).
SDValue Mul359 =
DAG.getNode(RISCVISD::SHL_ADD, DL, VT, X,
DAG.getConstant(Log2_64(Divisor - 1), DL, VT), X);
return DAG.getNode(ISD::SHL, DL, VT, Mul359,
DAG.getConstant(Log2_64(MulAmt2), DL, VT));
}
// 3/5/9 * 3/5/9 -> shXadd (shYadd X, X), (shYadd X, X)
if (MulAmt2 == 3 || MulAmt2 == 5 || MulAmt2 == 9) {
SDLoc DL(N);
SDValue Mul359 =
DAG.getNode(RISCVISD::SHL_ADD, DL, VT, X,
DAG.getConstant(Log2_64(Divisor - 1), DL, VT), X);
return DAG.getNode(RISCVISD::SHL_ADD, DL, VT, Mul359,
DAG.getConstant(Log2_64(MulAmt2 - 1), DL, VT),
Mul359);
}
}
// If this is a power 2 + 2/4/8, we can use a shift followed by a single
// shXadd. First check if this a sum of two power of 2s because that's
// easy. Then count how many zeros are up to the first bit.
if (isPowerOf2_64(MulAmt & (MulAmt - 1))) {
unsigned ScaleShift = llvm::countr_zero(MulAmt);
if (ScaleShift >= 1 && ScaleShift < 4) {
unsigned ShiftAmt = Log2_64((MulAmt & (MulAmt - 1)));
SDLoc DL(N);
SDValue Shift1 =
DAG.getNode(ISD::SHL, DL, VT, X, DAG.getConstant(ShiftAmt, DL, VT));
return DAG.getNode(RISCVISD::SHL_ADD, DL, VT, X,
DAG.getConstant(ScaleShift, DL, VT), Shift1);
}
}
// 2^(1,2,3) * 3,5,9 + 1 -> (shXadd (shYadd x, x), x)
// This is the two instruction form, there are also three instruction
// variants we could implement. e.g.
// (2^(1,2,3) * 3,5,9 + 1) << C2
// 2^(C1>3) * 3,5,9 +/- 1
for (uint64_t Divisor : {3, 5, 9}) {
uint64_t C = MulAmt - 1;
if (C <= Divisor)
continue;
unsigned TZ = llvm::countr_zero(C);
if ((C >> TZ) == Divisor && (TZ == 1 || TZ == 2 || TZ == 3)) {
SDLoc DL(N);
SDValue Mul359 =
DAG.getNode(RISCVISD::SHL_ADD, DL, VT, X,
DAG.getConstant(Log2_64(Divisor - 1), DL, VT), X);
return DAG.getNode(RISCVISD::SHL_ADD, DL, VT, Mul359,
DAG.getConstant(TZ, DL, VT), X);
}
}
// 2^n + 2/4/8 + 1 -> (add (shl X, C1), (shXadd X, X))
if (MulAmt > 2 && isPowerOf2_64((MulAmt - 1) & (MulAmt - 2))) {
unsigned ScaleShift = llvm::countr_zero(MulAmt - 1);
if (ScaleShift >= 1 && ScaleShift < 4) {
unsigned ShiftAmt = Log2_64(((MulAmt - 1) & (MulAmt - 2)));
SDLoc DL(N);
SDValue Shift1 =
DAG.getNode(ISD::SHL, DL, VT, X, DAG.getConstant(ShiftAmt, DL, VT));
return DAG.getNode(ISD::ADD, DL, VT, Shift1,
DAG.getNode(RISCVISD::SHL_ADD, DL, VT, X,
DAG.getConstant(ScaleShift, DL, VT), X));
}
}
// 2^N - 3/5/9 --> (sub (shl X, C1), (shXadd X, x))
for (uint64_t Offset : {3, 5, 9}) {
if (isPowerOf2_64(MulAmt + Offset)) {
unsigned ShAmt = Log2_64(MulAmt + Offset);
if (ShAmt >= VT.getSizeInBits())
continue;
SDLoc DL(N);
SDValue Shift1 =
DAG.getNode(ISD::SHL, DL, VT, X, DAG.getConstant(ShAmt, DL, VT));
SDValue Mul359 =
DAG.getNode(RISCVISD::SHL_ADD, DL, VT, X,
DAG.getConstant(Log2_64(Offset - 1), DL, VT), X);
return DAG.getNode(ISD::SUB, DL, VT, Shift1, Mul359);
}
}
for (uint64_t Divisor : {3, 5, 9}) {
if (MulAmt % Divisor != 0)
continue;
uint64_t MulAmt2 = MulAmt / Divisor;
// 3/5/9 * 3/5/9 * 2^N - In particular, this covers multiples
// of 25 which happen to be quite common.
for (uint64_t Divisor2 : {3, 5, 9}) {
if (MulAmt2 % Divisor2 != 0)
continue;
uint64_t MulAmt3 = MulAmt2 / Divisor2;
if (isPowerOf2_64(MulAmt3)) {
SDLoc DL(N);
SDValue Mul359A =
DAG.getNode(RISCVISD::SHL_ADD, DL, VT, X,
DAG.getConstant(Log2_64(Divisor - 1), DL, VT), X);
SDValue Mul359B = DAG.getNode(
RISCVISD::SHL_ADD, DL, VT, Mul359A,
DAG.getConstant(Log2_64(Divisor2 - 1), DL, VT), Mul359A);
return DAG.getNode(ISD::SHL, DL, VT, Mul359B,
DAG.getConstant(Log2_64(MulAmt3), DL, VT));
}
}
}
}
if (SDValue V = expandMulToAddOrSubOfShl(N, DAG, MulAmt))
return V;
if (!Subtarget.hasStdExtZmmul())
return expandMulToNAFSequence(N, DAG, MulAmt);
return SDValue();
}
// Combine vXi32 (mul (and (lshr X, 15), 0x10001), 0xffff) ->
// (bitcast (sra (v2Xi16 (bitcast X)), 15))
// Same for other equivalent types with other equivalent constants.
static SDValue combineVectorMulToSraBitcast(SDNode *N, SelectionDAG &DAG) {
EVT VT = N->getValueType(0);
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
// Do this for legal vectors unless they are i1 or i8 vectors.
if (!VT.isVector() || !TLI.isTypeLegal(VT) || VT.getScalarSizeInBits() < 16)
return SDValue();
if (N->getOperand(0).getOpcode() != ISD::AND ||
N->getOperand(0).getOperand(0).getOpcode() != ISD::SRL)
return SDValue();
SDValue And = N->getOperand(0);
SDValue Srl = And.getOperand(0);
APInt V1, V2, V3;
if (!ISD::isConstantSplatVector(N->getOperand(1).getNode(), V1) ||
!ISD::isConstantSplatVector(And.getOperand(1).getNode(), V2) ||
!ISD::isConstantSplatVector(Srl.getOperand(1).getNode(), V3))
return SDValue();
unsigned HalfSize = VT.getScalarSizeInBits() / 2;
if (!V1.isMask(HalfSize) || V2 != (1ULL | 1ULL << HalfSize) ||
V3 != (HalfSize - 1))
return SDValue();
EVT HalfVT = EVT::getVectorVT(*DAG.getContext(),
EVT::getIntegerVT(*DAG.getContext(), HalfSize),
VT.getVectorElementCount() * 2);
SDLoc DL(N);
SDValue Cast = DAG.getNode(ISD::BITCAST, DL, HalfVT, Srl.getOperand(0));
SDValue Sra = DAG.getNode(ISD::SRA, DL, HalfVT, Cast,
DAG.getConstant(HalfSize - 1, DL, HalfVT));
return DAG.getNode(ISD::BITCAST, DL, VT, Sra);
}
static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
EVT VT = N->getValueType(0);
if (!VT.isVector())
return expandMul(N, DAG, DCI, Subtarget);
SDLoc DL(N);
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
SDValue MulOper;
unsigned AddSubOpc;
// vmadd: (mul (add x, 1), y) -> (add (mul x, y), y)
// (mul x, add (y, 1)) -> (add x, (mul x, y))
// vnmsub: (mul (sub 1, x), y) -> (sub y, (mul x, y))
// (mul x, (sub 1, y)) -> (sub x, (mul x, y))
auto IsAddSubWith1 = [&](SDValue V) -> bool {
AddSubOpc = V->getOpcode();
if ((AddSubOpc == ISD::ADD || AddSubOpc == ISD::SUB) && V->hasOneUse()) {
SDValue Opnd = V->getOperand(1);
MulOper = V->getOperand(0);
if (AddSubOpc == ISD::SUB)
std::swap(Opnd, MulOper);
if (isOneOrOneSplat(Opnd))
return true;
}
return false;
};
if (IsAddSubWith1(N0)) {
SDValue MulVal = DAG.getNode(ISD::MUL, DL, VT, N1, MulOper);
return DAG.getNode(AddSubOpc, DL, VT, N1, MulVal);
}
if (IsAddSubWith1(N1)) {
SDValue MulVal = DAG.getNode(ISD::MUL, DL, VT, N0, MulOper);
return DAG.getNode(AddSubOpc, DL, VT, N0, MulVal);
}
if (SDValue V = combineBinOpOfZExt(N, DAG))
return V;
if (SDValue V = combineVectorMulToSraBitcast(N, DAG))
return V;
return SDValue();
}
/// According to the property that indexed load/store instructions zero-extend
/// their indices, try to narrow the type of index operand.
static bool narrowIndex(SDValue &N, ISD::MemIndexType IndexType, SelectionDAG &DAG) {
if (isIndexTypeSigned(IndexType))
return false;
if (!N->hasOneUse())
return false;
EVT VT = N.getValueType();
SDLoc DL(N);
// In general, what we're doing here is seeing if we can sink a truncate to
// a smaller element type into the expression tree building our index.
// TODO: We can generalize this and handle a bunch more cases if useful.
// Narrow a buildvector to the narrowest element type. This requires less
// work and less register pressure at high LMUL, and creates smaller constants
// which may be cheaper to materialize.
if (ISD::isBuildVectorOfConstantSDNodes(N.getNode())) {
KnownBits Known = DAG.computeKnownBits(N);
unsigned ActiveBits = std::max(8u, Known.countMaxActiveBits());
LLVMContext &C = *DAG.getContext();
EVT ResultVT = EVT::getIntegerVT(C, ActiveBits).getRoundIntegerType(C);
if (ResultVT.bitsLT(VT.getVectorElementType())) {
N = DAG.getNode(ISD::TRUNCATE, DL,
VT.changeVectorElementType(ResultVT), N);
return true;
}
}
// Handle the pattern (shl (zext x to ty), C) and bits(x) + C < bits(ty).
if (N.getOpcode() != ISD::SHL)
return false;
SDValue N0 = N.getOperand(0);
if (N0.getOpcode() != ISD::ZERO_EXTEND &&
N0.getOpcode() != RISCVISD::VZEXT_VL)
return false;
if (!N0->hasOneUse())
return false;
APInt ShAmt;
SDValue N1 = N.getOperand(1);
if (!ISD::isConstantSplatVector(N1.getNode(), ShAmt))
return false;
SDValue Src = N0.getOperand(0);
EVT SrcVT = Src.getValueType();
unsigned SrcElen = SrcVT.getScalarSizeInBits();
unsigned ShAmtV = ShAmt.getZExtValue();
unsigned NewElen = PowerOf2Ceil(SrcElen + ShAmtV);
NewElen = std::max(NewElen, 8U);
// Skip if NewElen is not narrower than the original extended type.
if (NewElen >= N0.getValueType().getScalarSizeInBits())
return false;
EVT NewEltVT = EVT::getIntegerVT(*DAG.getContext(), NewElen);
EVT NewVT = SrcVT.changeVectorElementType(NewEltVT);
SDValue NewExt = DAG.getNode(N0->getOpcode(), DL, NewVT, N0->ops());
SDValue NewShAmtVec = DAG.getConstant(ShAmtV, DL, NewVT);
N = DAG.getNode(ISD::SHL, DL, NewVT, NewExt, NewShAmtVec);
return true;
}
/// Try to map an integer comparison with size > XLEN to vector instructions
/// before type legalization splits it up into chunks.
static SDValue
combineVectorSizedSetCCEquality(EVT VT, SDValue X, SDValue Y, ISD::CondCode CC,
const SDLoc &DL, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(ISD::isIntEqualitySetCC(CC) && "Bad comparison predicate");
if (!Subtarget.hasVInstructions())
return SDValue();
MVT XLenVT = Subtarget.getXLenVT();
EVT OpVT = X.getValueType();
// We're looking for an oversized integer equality comparison.
if (!OpVT.isScalarInteger())
return SDValue();
unsigned OpSize = OpVT.getSizeInBits();
// The size should be larger than XLen and smaller than the maximum vector
// size.
if (OpSize <= Subtarget.getXLen() ||
OpSize > Subtarget.getRealMinVLen() *
Subtarget.getMaxLMULForFixedLengthVectors())
return SDValue();
// Don't perform this combine if constructing the vector will be expensive.
auto IsVectorBitCastCheap = [](SDValue X) {
X = peekThroughBitcasts(X);
return isa<ConstantSDNode>(X) || X.getValueType().isVector() ||
X.getOpcode() == ISD::LOAD;
};
if (!IsVectorBitCastCheap(X) || !IsVectorBitCastCheap(Y))
return SDValue();
if (DAG.getMachineFunction().getFunction().hasFnAttribute(
Attribute::NoImplicitFloat))
return SDValue();
// Bail out for non-byte-sized types.
if (!OpVT.isByteSized())
return SDValue();
unsigned VecSize = OpSize / 8;
EVT VecVT = EVT::getVectorVT(*DAG.getContext(), MVT::i8, VecSize);
EVT CmpVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1, VecSize);
SDValue VecX = DAG.getBitcast(VecVT, X);
SDValue VecY = DAG.getBitcast(VecVT, Y);
SDValue Mask = DAG.getAllOnesConstant(DL, CmpVT);
SDValue VL = DAG.getConstant(VecSize, DL, XLenVT);
SDValue Cmp = DAG.getNode(ISD::VP_SETCC, DL, CmpVT, VecX, VecY,
DAG.getCondCode(ISD::SETNE), Mask, VL);
return DAG.getSetCC(DL, VT,
DAG.getNode(ISD::VP_REDUCE_OR, DL, XLenVT,
DAG.getConstant(0, DL, XLenVT), Cmp, Mask,
VL),
DAG.getConstant(0, DL, XLenVT), CC);
}
static SDValue performSETCCCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N->getValueType(0);
EVT OpVT = N0.getValueType();
ISD::CondCode Cond = cast<CondCodeSDNode>(N->getOperand(2))->get();
// Looking for an equality compare.
if (!isIntEqualitySetCC(Cond))
return SDValue();
if (SDValue V =
combineVectorSizedSetCCEquality(VT, N0, N1, Cond, dl, DAG, Subtarget))
return V;
if (DCI.isAfterLegalizeDAG() && isa<ConstantSDNode>(N1) &&
N0.getOpcode() == ISD::AND && N0.hasOneUse() &&
isa<ConstantSDNode>(N0.getOperand(1))) {
const APInt &AndRHSC = N0.getConstantOperandAPInt(1);
// (X & -(1 << C)) == 0 -> (X >> C) == 0 if the AND constant can't use ANDI.
if (isNullConstant(N1) && !isInt<12>(AndRHSC.getSExtValue()) &&
AndRHSC.isNegatedPowerOf2()) {
unsigned ShiftBits = AndRHSC.countr_zero();
SDValue Shift = DAG.getNode(ISD::SRL, dl, OpVT, N0.getOperand(0),
DAG.getConstant(ShiftBits, dl, OpVT));
return DAG.getSetCC(dl, VT, Shift, N1, Cond);
}
// Similar to above but handling the lower 32 bits by using sraiw. Allow
// comparing with constants other than 0 if the constant can be folded into
// addi or xori after shifting.
uint64_t N1Int = cast<ConstantSDNode>(N1)->getZExtValue();
uint64_t AndRHSInt = AndRHSC.getZExtValue();
if (OpVT == MVT::i64 && AndRHSInt <= 0xffffffff &&
isPowerOf2_32(-uint32_t(AndRHSInt)) && (N1Int & AndRHSInt) == N1Int) {
unsigned ShiftBits = llvm::countr_zero(AndRHSInt);
int64_t NewC = SignExtend64<32>(N1Int) >> ShiftBits;
if (NewC >= -2048 && NewC <= 2048) {
SDValue SExt =
DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, OpVT, N0.getOperand(0),
DAG.getValueType(MVT::i32));
SDValue Shift = DAG.getNode(ISD::SRA, dl, OpVT, SExt,
DAG.getConstant(ShiftBits, dl, OpVT));
return DAG.getSetCC(dl, VT, Shift,
DAG.getSignedConstant(NewC, dl, OpVT), Cond);
}
}
}
// Replace (seteq (i64 (and X, 0xffffffff)), C1) with
// (seteq (i64 (sext_inreg (X, i32)), C1')) where C1' is C1 sign extended from
// bit 31. Same for setne. C1' may be cheaper to materialize and the
// sext_inreg can become a sext.w instead of a shift pair.
if (OpVT != MVT::i64 || !Subtarget.is64Bit())
return SDValue();
// RHS needs to be a constant.
auto *N1C = dyn_cast<ConstantSDNode>(N1);
if (!N1C)
return SDValue();
// LHS needs to be (and X, 0xffffffff).
if (N0.getOpcode() != ISD::AND || !N0.hasOneUse() ||
!isa<ConstantSDNode>(N0.getOperand(1)) ||
N0.getConstantOperandVal(1) != UINT64_C(0xffffffff))
return SDValue();
// Don't do this if the sign bit is provably zero, it will be turned back into
// an AND.
APInt SignMask = APInt::getOneBitSet(64, 31);
if (DAG.MaskedValueIsZero(N0.getOperand(0), SignMask))
return SDValue();
const APInt &C1 = N1C->getAPIntValue();
// If the constant is larger than 2^32 - 1 it is impossible for both sides
// to be equal.
if (C1.getActiveBits() > 32)
return DAG.getBoolConstant(Cond == ISD::SETNE, dl, VT, OpVT);
SDValue SExtOp = DAG.getNode(ISD::SIGN_EXTEND_INREG, N, OpVT,
N0.getOperand(0), DAG.getValueType(MVT::i32));
return DAG.getSetCC(dl, VT, SExtOp, DAG.getConstant(C1.trunc(32).sext(64),
dl, OpVT), Cond);
}
static SDValue
performSIGN_EXTEND_INREGCombine(SDNode *N, TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
SelectionDAG &DAG = DCI.DAG;
SDValue Src = N->getOperand(0);
EVT VT = N->getValueType(0);
EVT SrcVT = cast<VTSDNode>(N->getOperand(1))->getVT();
unsigned Opc = Src.getOpcode();
SDLoc DL(N);
// Fold (sext_inreg (fmv_x_anyexth X), i16) -> (fmv_x_signexth X)
// Don't do this with Zhinx. We need to explicitly sign extend the GPR.
if (Opc == RISCVISD::FMV_X_ANYEXTH && SrcVT.bitsGE(MVT::i16) &&
Subtarget.hasStdExtZfhmin())
return DAG.getNode(RISCVISD::FMV_X_SIGNEXTH, DL, VT, Src.getOperand(0));
// Fold (sext_inreg (shl X, Y), i32) -> (sllw X, Y) iff Y u< 32
if (Opc == ISD::SHL && Subtarget.is64Bit() && SrcVT == MVT::i32 &&
VT == MVT::i64 && !isa<ConstantSDNode>(Src.getOperand(1)) &&
DAG.computeKnownBits(Src.getOperand(1)).countMaxActiveBits() <= 5)
return DAG.getNode(RISCVISD::SLLW, DL, VT, Src.getOperand(0),
Src.getOperand(1));
// Fold (sext_inreg (setcc), i1) -> (sub 0, (setcc))
if (Opc == ISD::SETCC && SrcVT == MVT::i1 && DCI.isAfterLegalizeDAG())
return DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, DL, VT), Src);
// Fold (sext_inreg (xor (setcc), -1), i1) -> (add (setcc), -1)
if (Opc == ISD::XOR && SrcVT == MVT::i1 &&
isAllOnesConstant(Src.getOperand(1)) &&
Src.getOperand(0).getOpcode() == ISD::SETCC && DCI.isAfterLegalizeDAG())
return DAG.getNode(ISD::ADD, DL, VT, Src.getOperand(0),
DAG.getAllOnesConstant(DL, VT));
return SDValue();
}
namespace {
// Forward declaration of the structure holding the necessary information to
// apply a combine.
struct CombineResult;
enum ExtKind : uint8_t {
ZExt = 1 << 0,
SExt = 1 << 1,
FPExt = 1 << 2,
BF16Ext = 1 << 3
};
/// Helper class for folding sign/zero extensions.
/// In particular, this class is used for the following combines:
/// add | add_vl | or disjoint -> vwadd(u) | vwadd(u)_w
/// sub | sub_vl -> vwsub(u) | vwsub(u)_w
/// mul | mul_vl -> vwmul(u) | vwmul_su
/// shl | shl_vl -> vwsll
/// fadd -> vfwadd | vfwadd_w
/// fsub -> vfwsub | vfwsub_w
/// fmul -> vfwmul
/// An object of this class represents an operand of the operation we want to
/// combine.
/// E.g., when trying to combine `mul_vl a, b`, we will have one instance of
/// NodeExtensionHelper for `a` and one for `b`.
///
/// This class abstracts away how the extension is materialized and
/// how its number of users affect the combines.
///
/// In particular:
/// - VWADD_W is conceptually == add(op0, sext(op1))
/// - VWADDU_W == add(op0, zext(op1))
/// - VWSUB_W == sub(op0, sext(op1))
/// - VWSUBU_W == sub(op0, zext(op1))
/// - VFWADD_W == fadd(op0, fpext(op1))
/// - VFWSUB_W == fsub(op0, fpext(op1))
/// And VMV_V_X_VL, depending on the value, is conceptually equivalent to
/// zext|sext(smaller_value).
struct NodeExtensionHelper {
/// Records if this operand is like being zero extended.
bool SupportsZExt;
/// Records if this operand is like being sign extended.
/// Note: SupportsZExt and SupportsSExt are not mutually exclusive. For
/// instance, a splat constant (e.g., 3), would support being both sign and
/// zero extended.
bool SupportsSExt;
/// Records if this operand is like being floating point extended.
bool SupportsFPExt;
/// Records if this operand is extended from bf16.
bool SupportsBF16Ext;
/// This boolean captures whether we care if this operand would still be
/// around after the folding happens.
bool EnforceOneUse;
/// Original value that this NodeExtensionHelper represents.
SDValue OrigOperand;
/// Get the value feeding the extension or the value itself.
/// E.g., for zext(a), this would return a.
SDValue getSource() const {
switch (OrigOperand.getOpcode()) {
case ISD::ZERO_EXTEND:
case ISD::SIGN_EXTEND:
case RISCVISD::VSEXT_VL:
case RISCVISD::VZEXT_VL:
case RISCVISD::FP_EXTEND_VL:
return OrigOperand.getOperand(0);
default:
return OrigOperand;
}
}
/// Check if this instance represents a splat.
bool isSplat() const {
return OrigOperand.getOpcode() == RISCVISD::VMV_V_X_VL ||
OrigOperand.getOpcode() == ISD::SPLAT_VECTOR;
}
/// Get the extended opcode.
unsigned getExtOpc(ExtKind SupportsExt) const {
switch (SupportsExt) {
case ExtKind::SExt:
return RISCVISD::VSEXT_VL;
case ExtKind::ZExt:
return RISCVISD::VZEXT_VL;
case ExtKind::FPExt:
case ExtKind::BF16Ext:
return RISCVISD::FP_EXTEND_VL;
}
llvm_unreachable("Unknown ExtKind enum");
}
/// Get or create a value that can feed \p Root with the given extension \p
/// SupportsExt. If \p SExt is std::nullopt, this returns the source of this
/// operand. \see ::getSource().
SDValue getOrCreateExtendedOp(SDNode *Root, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget,
std::optional<ExtKind> SupportsExt) const {
if (!SupportsExt.has_value())
return OrigOperand;
MVT NarrowVT = getNarrowType(Root, *SupportsExt);
SDValue Source = getSource();
assert(Subtarget.getTargetLowering()->isTypeLegal(Source.getValueType()));
if (Source.getValueType() == NarrowVT)
return Source;
unsigned ExtOpc = getExtOpc(*SupportsExt);
// If we need an extension, we should be changing the type.
SDLoc DL(OrigOperand);
auto [Mask, VL] = getMaskAndVL(Root, DAG, Subtarget);
switch (OrigOperand.getOpcode()) {
case ISD::ZERO_EXTEND:
case ISD::SIGN_EXTEND:
case RISCVISD::VSEXT_VL:
case RISCVISD::VZEXT_VL:
case RISCVISD::FP_EXTEND_VL:
return DAG.getNode(ExtOpc, DL, NarrowVT, Source, Mask, VL);
case ISD::SPLAT_VECTOR:
return DAG.getSplat(NarrowVT, DL, Source.getOperand(0));
case RISCVISD::VMV_V_X_VL:
return DAG.getNode(RISCVISD::VMV_V_X_VL, DL, NarrowVT,
DAG.getUNDEF(NarrowVT), Source.getOperand(1), VL);
case RISCVISD::VFMV_V_F_VL:
Source = Source.getOperand(1);
assert(Source.getOpcode() == ISD::FP_EXTEND && "Unexpected source");
Source = Source.getOperand(0);
assert(Source.getValueType() == NarrowVT.getVectorElementType());
return DAG.getNode(RISCVISD::VFMV_V_F_VL, DL, NarrowVT,
DAG.getUNDEF(NarrowVT), Source, VL);
default:
// Other opcodes can only come from the original LHS of VW(ADD|SUB)_W_VL
// and that operand should already have the right NarrowVT so no
// extension should be required at this point.
llvm_unreachable("Unsupported opcode");
}
}
/// Helper function to get the narrow type for \p Root.
/// The narrow type is the type of \p Root where we divided the size of each
/// element by 2. E.g., if Root's type <2xi16> -> narrow type <2xi8>.
/// \pre Both the narrow type and the original type should be legal.
static MVT getNarrowType(const SDNode *Root, ExtKind SupportsExt) {
MVT VT = Root->getSimpleValueType(0);
// Determine the narrow size.
unsigned NarrowSize = VT.getScalarSizeInBits() / 2;
MVT EltVT = SupportsExt == ExtKind::BF16Ext ? MVT::bf16
: SupportsExt == ExtKind::FPExt
? MVT::getFloatingPointVT(NarrowSize)
: MVT::getIntegerVT(NarrowSize);
assert((int)NarrowSize >= (SupportsExt == ExtKind::FPExt ? 16 : 8) &&
"Trying to extend something we can't represent");
MVT NarrowVT = MVT::getVectorVT(EltVT, VT.getVectorElementCount());
return NarrowVT;
}
/// Get the opcode to materialize:
/// Opcode(sext(a), sext(b)) -> newOpcode(a, b)
static unsigned getSExtOpcode(unsigned Opcode) {
switch (Opcode) {
case ISD::ADD:
case RISCVISD::ADD_VL:
case RISCVISD::VWADD_W_VL:
case RISCVISD::VWADDU_W_VL:
case ISD::OR:
case RISCVISD::OR_VL:
return RISCVISD::VWADD_VL;
case ISD::SUB:
case RISCVISD::SUB_VL:
case RISCVISD::VWSUB_W_VL:
case RISCVISD::VWSUBU_W_VL:
return RISCVISD::VWSUB_VL;
case ISD::MUL:
case RISCVISD::MUL_VL:
return RISCVISD::VWMUL_VL;
default:
llvm_unreachable("Unexpected opcode");
}
}
/// Get the opcode to materialize:
/// Opcode(zext(a), zext(b)) -> newOpcode(a, b)
static unsigned getZExtOpcode(unsigned Opcode) {
switch (Opcode) {
case ISD::ADD:
case RISCVISD::ADD_VL:
case RISCVISD::VWADD_W_VL:
case RISCVISD::VWADDU_W_VL:
case ISD::OR:
case RISCVISD::OR_VL:
return RISCVISD::VWADDU_VL;
case ISD::SUB:
case RISCVISD::SUB_VL:
case RISCVISD::VWSUB_W_VL:
case RISCVISD::VWSUBU_W_VL:
return RISCVISD::VWSUBU_VL;
case ISD::MUL:
case RISCVISD::MUL_VL:
return RISCVISD::VWMULU_VL;
case ISD::SHL:
case RISCVISD::SHL_VL:
return RISCVISD::VWSLL_VL;
default:
llvm_unreachable("Unexpected opcode");
}
}
/// Get the opcode to materialize:
/// Opcode(fpext(a), fpext(b)) -> newOpcode(a, b)
static unsigned getFPExtOpcode(unsigned Opcode) {
switch (Opcode) {
case RISCVISD::FADD_VL:
case RISCVISD::VFWADD_W_VL:
return RISCVISD::VFWADD_VL;
case RISCVISD::FSUB_VL:
case RISCVISD::VFWSUB_W_VL:
return RISCVISD::VFWSUB_VL;
case RISCVISD::FMUL_VL:
return RISCVISD::VFWMUL_VL;
case RISCVISD::VFMADD_VL:
return RISCVISD::VFWMADD_VL;
case RISCVISD::VFMSUB_VL:
return RISCVISD::VFWMSUB_VL;
case RISCVISD::VFNMADD_VL:
return RISCVISD::VFWNMADD_VL;
case RISCVISD::VFNMSUB_VL:
return RISCVISD::VFWNMSUB_VL;
default:
llvm_unreachable("Unexpected opcode");
}
}
/// Get the opcode to materialize \p Opcode(sext(a), zext(b)) ->
/// newOpcode(a, b).
static unsigned getSUOpcode(unsigned Opcode) {
assert((Opcode == RISCVISD::MUL_VL || Opcode == ISD::MUL) &&
"SU is only supported for MUL");
return RISCVISD::VWMULSU_VL;
}
/// Get the opcode to materialize
/// \p Opcode(a, s|z|fpext(b)) -> newOpcode(a, b).
static unsigned getWOpcode(unsigned Opcode, ExtKind SupportsExt) {
switch (Opcode) {
case ISD::ADD:
case RISCVISD::ADD_VL:
case ISD::OR:
case RISCVISD::OR_VL:
return SupportsExt == ExtKind::SExt ? RISCVISD::VWADD_W_VL
: RISCVISD::VWADDU_W_VL;
case ISD::SUB:
case RISCVISD::SUB_VL:
return SupportsExt == ExtKind::SExt ? RISCVISD::VWSUB_W_VL
: RISCVISD::VWSUBU_W_VL;
case RISCVISD::FADD_VL:
return RISCVISD::VFWADD_W_VL;
case RISCVISD::FSUB_VL:
return RISCVISD::VFWSUB_W_VL;
default:
llvm_unreachable("Unexpected opcode");
}
}
using CombineToTry = std::function<std::optional<CombineResult>(
SDNode * /*Root*/, const NodeExtensionHelper & /*LHS*/,
const NodeExtensionHelper & /*RHS*/, SelectionDAG &,
const RISCVSubtarget &)>;
/// Check if this node needs to be fully folded or extended for all users.
bool needToPromoteOtherUsers() const { return EnforceOneUse; }
void fillUpExtensionSupportForSplat(SDNode *Root, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
unsigned Opc = OrigOperand.getOpcode();
MVT VT = OrigOperand.getSimpleValueType();
assert((Opc == ISD::SPLAT_VECTOR || Opc == RISCVISD::VMV_V_X_VL) &&
"Unexpected Opcode");
// The pasthru must be undef for tail agnostic.
if (Opc == RISCVISD::VMV_V_X_VL && !OrigOperand.getOperand(0).isUndef())
return;
// Get the scalar value.
SDValue Op = Opc == ISD::SPLAT_VECTOR ? OrigOperand.getOperand(0)
: OrigOperand.getOperand(1);
// See if we have enough sign bits or zero bits in the scalar to use a
// widening opcode by splatting to smaller element size.
unsigned EltBits = VT.getScalarSizeInBits();
unsigned ScalarBits = Op.getValueSizeInBits();
// If we're not getting all bits from the element, we need special handling.
if (ScalarBits < EltBits) {
// This should only occur on RV32.
assert(Opc == RISCVISD::VMV_V_X_VL && EltBits == 64 && ScalarBits == 32 &&
!Subtarget.is64Bit() && "Unexpected splat");
// vmv.v.x sign extends narrow inputs.
SupportsSExt = true;
// If the input is positive, then sign extend is also zero extend.
if (DAG.SignBitIsZero(Op))
SupportsZExt = true;
EnforceOneUse = false;
return;
}
unsigned NarrowSize = EltBits / 2;
// If the narrow type cannot be expressed with a legal VMV,
// this is not a valid candidate.
if (NarrowSize < 8)
return;
if (DAG.ComputeMaxSignificantBits(Op) <= NarrowSize)
SupportsSExt = true;
if (DAG.MaskedValueIsZero(Op,
APInt::getBitsSetFrom(ScalarBits, NarrowSize)))
SupportsZExt = true;
EnforceOneUse = false;
}
bool isSupportedFPExtend(MVT NarrowEltVT, const RISCVSubtarget &Subtarget) {
return (NarrowEltVT == MVT::f32 ||
(NarrowEltVT == MVT::f16 && Subtarget.hasVInstructionsF16()));
}
bool isSupportedBF16Extend(MVT NarrowEltVT, const RISCVSubtarget &Subtarget) {
return NarrowEltVT == MVT::bf16 && Subtarget.hasStdExtZvfbfwma();
}
/// Helper method to set the various fields of this struct based on the
/// type of \p Root.
void fillUpExtensionSupport(SDNode *Root, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SupportsZExt = false;
SupportsSExt = false;
SupportsFPExt = false;
SupportsBF16Ext = false;
EnforceOneUse = true;
unsigned Opc = OrigOperand.getOpcode();
// For the nodes we handle below, we end up using their inputs directly: see
// getSource(). However since they either don't have a passthru or we check
// that their passthru is undef, we can safely ignore their mask and VL.
switch (Opc) {
case ISD::ZERO_EXTEND:
case ISD::SIGN_EXTEND: {
MVT VT = OrigOperand.getSimpleValueType();
if (!VT.isVector())
break;
SDValue NarrowElt = OrigOperand.getOperand(0);
MVT NarrowVT = NarrowElt.getSimpleValueType();
// i1 types are legal but we can't select V{S,Z}EXT_VLs with them.
if (NarrowVT.getVectorElementType() == MVT::i1)
break;
SupportsZExt = Opc == ISD::ZERO_EXTEND;
SupportsSExt = Opc == ISD::SIGN_EXTEND;
break;
}
case RISCVISD::VZEXT_VL:
SupportsZExt = true;
break;
case RISCVISD::VSEXT_VL:
SupportsSExt = true;
break;
case RISCVISD::FP_EXTEND_VL: {
MVT NarrowEltVT =
OrigOperand.getOperand(0).getSimpleValueType().getVectorElementType();
if (isSupportedFPExtend(NarrowEltVT, Subtarget))
SupportsFPExt = true;
if (isSupportedBF16Extend(NarrowEltVT, Subtarget))
SupportsBF16Ext = true;
break;
}
case ISD::SPLAT_VECTOR:
case RISCVISD::VMV_V_X_VL:
fillUpExtensionSupportForSplat(Root, DAG, Subtarget);
break;
case RISCVISD::VFMV_V_F_VL: {
MVT VT = OrigOperand.getSimpleValueType();
if (!OrigOperand.getOperand(0).isUndef())
break;
SDValue Op = OrigOperand.getOperand(1);
if (Op.getOpcode() != ISD::FP_EXTEND)
break;
unsigned NarrowSize = VT.getScalarSizeInBits() / 2;
unsigned ScalarBits = Op.getOperand(0).getValueSizeInBits();
if (NarrowSize != ScalarBits)
break;
if (isSupportedFPExtend(Op.getOperand(0).getSimpleValueType(), Subtarget))
SupportsFPExt = true;
if (isSupportedBF16Extend(Op.getOperand(0).getSimpleValueType(),
Subtarget))
SupportsBF16Ext = true;
break;
}
default:
break;
}
}
/// Check if \p Root supports any extension folding combines.
static bool isSupportedRoot(const SDNode *Root,
const RISCVSubtarget &Subtarget) {
switch (Root->getOpcode()) {
case ISD::ADD:
case ISD::SUB:
case ISD::MUL: {
return Root->getValueType(0).isScalableVector();
}
case ISD::OR: {
return Root->getValueType(0).isScalableVector() &&
Root->getFlags().hasDisjoint();
}
// Vector Widening Integer Add/Sub/Mul Instructions
case RISCVISD::ADD_VL:
case RISCVISD::MUL_VL:
case RISCVISD::VWADD_W_VL:
case RISCVISD::VWADDU_W_VL:
case RISCVISD::SUB_VL:
case RISCVISD::VWSUB_W_VL:
case RISCVISD::VWSUBU_W_VL:
// Vector Widening Floating-Point Add/Sub/Mul Instructions
case RISCVISD::FADD_VL:
case RISCVISD::FSUB_VL:
case RISCVISD::FMUL_VL:
case RISCVISD::VFWADD_W_VL:
case RISCVISD::VFWSUB_W_VL:
return true;
case RISCVISD::OR_VL:
return Root->getFlags().hasDisjoint();
case ISD::SHL:
return Root->getValueType(0).isScalableVector() &&
Subtarget.hasStdExtZvbb();
case RISCVISD::SHL_VL:
return Subtarget.hasStdExtZvbb();
case RISCVISD::VFMADD_VL:
case RISCVISD::VFNMSUB_VL:
case RISCVISD::VFNMADD_VL:
case RISCVISD::VFMSUB_VL:
return true;
default:
return false;
}
}
/// Build a NodeExtensionHelper for \p Root.getOperand(\p OperandIdx).
NodeExtensionHelper(SDNode *Root, unsigned OperandIdx, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(isSupportedRoot(Root, Subtarget) &&
"Trying to build an helper with an "
"unsupported root");
assert(OperandIdx < 2 && "Requesting something else than LHS or RHS");
assert(DAG.getTargetLoweringInfo().isTypeLegal(Root->getValueType(0)));
OrigOperand = Root->getOperand(OperandIdx);
unsigned Opc = Root->getOpcode();
switch (Opc) {
// We consider
// VW<ADD|SUB>_W(LHS, RHS) -> <ADD|SUB>(LHS, SEXT(RHS))
// VW<ADD|SUB>U_W(LHS, RHS) -> <ADD|SUB>(LHS, ZEXT(RHS))
// VFW<ADD|SUB>_W(LHS, RHS) -> F<ADD|SUB>(LHS, FPEXT(RHS))
case RISCVISD::VWADD_W_VL:
case RISCVISD::VWADDU_W_VL:
case RISCVISD::VWSUB_W_VL:
case RISCVISD::VWSUBU_W_VL:
case RISCVISD::VFWADD_W_VL:
case RISCVISD::VFWSUB_W_VL:
if (OperandIdx == 1) {
SupportsZExt =
Opc == RISCVISD::VWADDU_W_VL || Opc == RISCVISD::VWSUBU_W_VL;
SupportsSExt =
Opc == RISCVISD::VWADD_W_VL || Opc == RISCVISD::VWSUB_W_VL;
SupportsFPExt =
Opc == RISCVISD::VFWADD_W_VL || Opc == RISCVISD::VFWSUB_W_VL;
// There's no existing extension here, so we don't have to worry about
// making sure it gets removed.
EnforceOneUse = false;
break;
}
[[fallthrough]];
default:
fillUpExtensionSupport(Root, DAG, Subtarget);
break;
}
}
/// Helper function to get the Mask and VL from \p Root.
static std::pair<SDValue, SDValue>
getMaskAndVL(const SDNode *Root, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(isSupportedRoot(Root, Subtarget) && "Unexpected root");
switch (Root->getOpcode()) {
case ISD::ADD:
case ISD::SUB:
case ISD::MUL:
case ISD::OR:
case ISD::SHL: {
SDLoc DL(Root);
MVT VT = Root->getSimpleValueType(0);
return getDefaultScalableVLOps(VT, DL, DAG, Subtarget);
}
default:
return std::make_pair(Root->getOperand(3), Root->getOperand(4));
}
}
/// Helper function to check if \p N is commutative with respect to the
/// foldings that are supported by this class.
static bool isCommutative(const SDNode *N) {
switch (N->getOpcode()) {
case ISD::ADD:
case ISD::MUL:
case ISD::OR:
case RISCVISD::ADD_VL:
case RISCVISD::MUL_VL:
case RISCVISD::OR_VL:
case RISCVISD::VWADD_W_VL:
case RISCVISD::VWADDU_W_VL:
case RISCVISD::FADD_VL:
case RISCVISD::FMUL_VL:
case RISCVISD::VFWADD_W_VL:
case RISCVISD::VFMADD_VL:
case RISCVISD::VFNMSUB_VL:
case RISCVISD::VFNMADD_VL:
case RISCVISD::VFMSUB_VL:
return true;
case ISD::SUB:
case RISCVISD::SUB_VL:
case RISCVISD::VWSUB_W_VL:
case RISCVISD::VWSUBU_W_VL:
case RISCVISD::FSUB_VL:
case RISCVISD::VFWSUB_W_VL:
case ISD::SHL:
case RISCVISD::SHL_VL:
return false;
default:
llvm_unreachable("Unexpected opcode");
}
}
/// Get a list of combine to try for folding extensions in \p Root.
/// Note that each returned CombineToTry function doesn't actually modify
/// anything. Instead they produce an optional CombineResult that if not None,
/// need to be materialized for the combine to be applied.
/// \see CombineResult::materialize.
/// If the related CombineToTry function returns std::nullopt, that means the
/// combine didn't match.
static SmallVector<CombineToTry> getSupportedFoldings(const SDNode *Root);
};
/// Helper structure that holds all the necessary information to materialize a
/// combine that does some extension folding.
struct CombineResult {
/// Opcode to be generated when materializing the combine.
unsigned TargetOpcode;
// No value means no extension is needed.
std::optional<ExtKind> LHSExt;
std::optional<ExtKind> RHSExt;
/// Root of the combine.
SDNode *Root;
/// LHS of the TargetOpcode.
NodeExtensionHelper LHS;
/// RHS of the TargetOpcode.
NodeExtensionHelper RHS;
CombineResult(unsigned TargetOpcode, SDNode *Root,
const NodeExtensionHelper &LHS, std::optional<ExtKind> LHSExt,
const NodeExtensionHelper &RHS, std::optional<ExtKind> RHSExt)
: TargetOpcode(TargetOpcode), LHSExt(LHSExt), RHSExt(RHSExt), Root(Root),
LHS(LHS), RHS(RHS) {}
/// Return a value that uses TargetOpcode and that can be used to replace
/// Root.
/// The actual replacement is *not* done in that method.
SDValue materialize(SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) const {
SDValue Mask, VL, Passthru;
std::tie(Mask, VL) =
NodeExtensionHelper::getMaskAndVL(Root, DAG, Subtarget);
switch (Root->getOpcode()) {
default:
Passthru = Root->getOperand(2);
break;
case ISD::ADD:
case ISD::SUB:
case ISD::MUL:
case ISD::OR:
case ISD::SHL:
Passthru = DAG.getUNDEF(Root->getValueType(0));
break;
}
return DAG.getNode(TargetOpcode, SDLoc(Root), Root->getValueType(0),
LHS.getOrCreateExtendedOp(Root, DAG, Subtarget, LHSExt),
RHS.getOrCreateExtendedOp(Root, DAG, Subtarget, RHSExt),
Passthru, Mask, VL);
}
};
/// Check if \p Root follows a pattern Root(ext(LHS), ext(RHS))
/// where `ext` is the same for both LHS and RHS (i.e., both are sext or both
/// are zext) and LHS and RHS can be folded into Root.
/// AllowExtMask define which form `ext` can take in this pattern.
///
/// \note If the pattern can match with both zext and sext, the returned
/// CombineResult will feature the zext result.
///
/// \returns std::nullopt if the pattern doesn't match or a CombineResult that
/// can be used to apply the pattern.
static std::optional<CombineResult>
canFoldToVWWithSameExtensionImpl(SDNode *Root, const NodeExtensionHelper &LHS,
const NodeExtensionHelper &RHS,
uint8_t AllowExtMask, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if ((AllowExtMask & ExtKind::ZExt) && LHS.SupportsZExt && RHS.SupportsZExt)
return CombineResult(NodeExtensionHelper::getZExtOpcode(Root->getOpcode()),
Root, LHS, /*LHSExt=*/{ExtKind::ZExt}, RHS,
/*RHSExt=*/{ExtKind::ZExt});
if ((AllowExtMask & ExtKind::SExt) && LHS.SupportsSExt && RHS.SupportsSExt)
return CombineResult(NodeExtensionHelper::getSExtOpcode(Root->getOpcode()),
Root, LHS, /*LHSExt=*/{ExtKind::SExt}, RHS,
/*RHSExt=*/{ExtKind::SExt});
if ((AllowExtMask & ExtKind::FPExt) && LHS.SupportsFPExt && RHS.SupportsFPExt)
return CombineResult(NodeExtensionHelper::getFPExtOpcode(Root->getOpcode()),
Root, LHS, /*LHSExt=*/{ExtKind::FPExt}, RHS,
/*RHSExt=*/{ExtKind::FPExt});
if ((AllowExtMask & ExtKind::BF16Ext) && LHS.SupportsBF16Ext &&
RHS.SupportsBF16Ext)
return CombineResult(NodeExtensionHelper::getFPExtOpcode(Root->getOpcode()),
Root, LHS, /*LHSExt=*/{ExtKind::BF16Ext}, RHS,
/*RHSExt=*/{ExtKind::BF16Ext});
return std::nullopt;
}
/// Check if \p Root follows a pattern Root(ext(LHS), ext(RHS))
/// where `ext` is the same for both LHS and RHS (i.e., both are sext or both
/// are zext) and LHS and RHS can be folded into Root.
///
/// \returns std::nullopt if the pattern doesn't match or a CombineResult that
/// can be used to apply the pattern.
static std::optional<CombineResult>
canFoldToVWWithSameExtension(SDNode *Root, const NodeExtensionHelper &LHS,
const NodeExtensionHelper &RHS, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
return canFoldToVWWithSameExtensionImpl(
Root, LHS, RHS, ExtKind::ZExt | ExtKind::SExt | ExtKind::FPExt, DAG,
Subtarget);
}
/// Check if \p Root follows a pattern Root(LHS, ext(RHS))
///
/// \returns std::nullopt if the pattern doesn't match or a CombineResult that
/// can be used to apply the pattern.
static std::optional<CombineResult>
canFoldToVW_W(SDNode *Root, const NodeExtensionHelper &LHS,
const NodeExtensionHelper &RHS, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (RHS.SupportsFPExt)
return CombineResult(
NodeExtensionHelper::getWOpcode(Root->getOpcode(), ExtKind::FPExt),
Root, LHS, /*LHSExt=*/std::nullopt, RHS, /*RHSExt=*/{ExtKind::FPExt});
// FIXME: Is it useful to form a vwadd.wx or vwsub.wx if it removes a scalar
// sext/zext?
// Control this behavior behind an option (AllowSplatInVW_W) for testing
// purposes.
if (RHS.SupportsZExt && (!RHS.isSplat() || AllowSplatInVW_W))
return CombineResult(
NodeExtensionHelper::getWOpcode(Root->getOpcode(), ExtKind::ZExt), Root,
LHS, /*LHSExt=*/std::nullopt, RHS, /*RHSExt=*/{ExtKind::ZExt});
if (RHS.SupportsSExt && (!RHS.isSplat() || AllowSplatInVW_W))
return CombineResult(
NodeExtensionHelper::getWOpcode(Root->getOpcode(), ExtKind::SExt), Root,
LHS, /*LHSExt=*/std::nullopt, RHS, /*RHSExt=*/{ExtKind::SExt});
return std::nullopt;
}
/// Check if \p Root follows a pattern Root(sext(LHS), sext(RHS))
///
/// \returns std::nullopt if the pattern doesn't match or a CombineResult that
/// can be used to apply the pattern.
static std::optional<CombineResult>
canFoldToVWWithSEXT(SDNode *Root, const NodeExtensionHelper &LHS,
const NodeExtensionHelper &RHS, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
return canFoldToVWWithSameExtensionImpl(Root, LHS, RHS, ExtKind::SExt, DAG,
Subtarget);
}
/// Check if \p Root follows a pattern Root(zext(LHS), zext(RHS))
///
/// \returns std::nullopt if the pattern doesn't match or a CombineResult that
/// can be used to apply the pattern.
static std::optional<CombineResult>
canFoldToVWWithZEXT(SDNode *Root, const NodeExtensionHelper &LHS,
const NodeExtensionHelper &RHS, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
return canFoldToVWWithSameExtensionImpl(Root, LHS, RHS, ExtKind::ZExt, DAG,
Subtarget);
}
/// Check if \p Root follows a pattern Root(fpext(LHS), fpext(RHS))
///
/// \returns std::nullopt if the pattern doesn't match or a CombineResult that
/// can be used to apply the pattern.
static std::optional<CombineResult>
canFoldToVWWithFPEXT(SDNode *Root, const NodeExtensionHelper &LHS,
const NodeExtensionHelper &RHS, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
return canFoldToVWWithSameExtensionImpl(Root, LHS, RHS, ExtKind::FPExt, DAG,
Subtarget);
}
/// Check if \p Root follows a pattern Root(bf16ext(LHS), bf16ext(RHS))
///
/// \returns std::nullopt if the pattern doesn't match or a CombineResult that
/// can be used to apply the pattern.
static std::optional<CombineResult>
canFoldToVWWithBF16EXT(SDNode *Root, const NodeExtensionHelper &LHS,
const NodeExtensionHelper &RHS, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
return canFoldToVWWithSameExtensionImpl(Root, LHS, RHS, ExtKind::BF16Ext, DAG,
Subtarget);
}
/// Check if \p Root follows a pattern Root(sext(LHS), zext(RHS))
///
/// \returns std::nullopt if the pattern doesn't match or a CombineResult that
/// can be used to apply the pattern.
static std::optional<CombineResult>
canFoldToVW_SU(SDNode *Root, const NodeExtensionHelper &LHS,
const NodeExtensionHelper &RHS, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (!LHS.SupportsSExt || !RHS.SupportsZExt)
return std::nullopt;
return CombineResult(NodeExtensionHelper::getSUOpcode(Root->getOpcode()),
Root, LHS, /*LHSExt=*/{ExtKind::SExt}, RHS,
/*RHSExt=*/{ExtKind::ZExt});
}
SmallVector<NodeExtensionHelper::CombineToTry>
NodeExtensionHelper::getSupportedFoldings(const SDNode *Root) {
SmallVector<CombineToTry> Strategies;
switch (Root->getOpcode()) {
case ISD::ADD:
case ISD::SUB:
case ISD::OR:
case RISCVISD::ADD_VL:
case RISCVISD::SUB_VL:
case RISCVISD::OR_VL:
case RISCVISD::FADD_VL:
case RISCVISD::FSUB_VL:
// add|sub|fadd|fsub-> vwadd(u)|vwsub(u)|vfwadd|vfwsub
Strategies.push_back(canFoldToVWWithSameExtension);
// add|sub|fadd|fsub -> vwadd(u)_w|vwsub(u)_w}|vfwadd_w|vfwsub_w
Strategies.push_back(canFoldToVW_W);
break;
case RISCVISD::FMUL_VL:
case RISCVISD::VFMADD_VL:
case RISCVISD::VFMSUB_VL:
case RISCVISD::VFNMADD_VL:
case RISCVISD::VFNMSUB_VL:
Strategies.push_back(canFoldToVWWithSameExtension);
if (Root->getOpcode() == RISCVISD::VFMADD_VL)
Strategies.push_back(canFoldToVWWithBF16EXT);
break;
case ISD::MUL:
case RISCVISD::MUL_VL:
// mul -> vwmul(u)
Strategies.push_back(canFoldToVWWithSameExtension);
// mul -> vwmulsu
Strategies.push_back(canFoldToVW_SU);
break;
case ISD::SHL:
case RISCVISD::SHL_VL:
// shl -> vwsll
Strategies.push_back(canFoldToVWWithZEXT);
break;
case RISCVISD::VWADD_W_VL:
case RISCVISD::VWSUB_W_VL:
// vwadd_w|vwsub_w -> vwadd|vwsub
Strategies.push_back(canFoldToVWWithSEXT);
break;
case RISCVISD::VWADDU_W_VL:
case RISCVISD::VWSUBU_W_VL:
// vwaddu_w|vwsubu_w -> vwaddu|vwsubu
Strategies.push_back(canFoldToVWWithZEXT);
break;
case RISCVISD::VFWADD_W_VL:
case RISCVISD::VFWSUB_W_VL:
// vfwadd_w|vfwsub_w -> vfwadd|vfwsub
Strategies.push_back(canFoldToVWWithFPEXT);
break;
default:
llvm_unreachable("Unexpected opcode");
}
return Strategies;
}
} // End anonymous namespace.
static SDValue simplifyOp_VL(SDNode *N) {
// TODO: Extend this to other binops using generic identity logic
assert(N->getOpcode() == RISCVISD::ADD_VL);
SDValue A = N->getOperand(0);
SDValue B = N->getOperand(1);
SDValue Passthru = N->getOperand(2);
if (!Passthru.isUndef())
// TODO:This could be a vmerge instead
return SDValue();
;
if (ISD::isConstantSplatVectorAllZeros(B.getNode()))
return A;
// Peek through fixed to scalable
if (B.getOpcode() == ISD::INSERT_SUBVECTOR && B.getOperand(0).isUndef() &&
ISD::isConstantSplatVectorAllZeros(B.getOperand(1).getNode()))
return A;
return SDValue();
}
/// Combine a binary or FMA operation to its equivalent VW or VW_W form.
/// The supported combines are:
/// add | add_vl | or disjoint | or_vl disjoint -> vwadd(u) | vwadd(u)_w
/// sub | sub_vl -> vwsub(u) | vwsub(u)_w
/// mul | mul_vl -> vwmul(u) | vwmul_su
/// shl | shl_vl -> vwsll
/// fadd_vl -> vfwadd | vfwadd_w
/// fsub_vl -> vfwsub | vfwsub_w
/// fmul_vl -> vfwmul
/// vwadd_w(u) -> vwadd(u)
/// vwsub_w(u) -> vwsub(u)
/// vfwadd_w -> vfwadd
/// vfwsub_w -> vfwsub
static SDValue combineOp_VLToVWOp_VL(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
SelectionDAG &DAG = DCI.DAG;
if (DCI.isBeforeLegalize())
return SDValue();
if (!NodeExtensionHelper::isSupportedRoot(N, Subtarget))
return SDValue();
SmallVector<SDNode *> Worklist;
SmallPtrSet<SDNode *, 8> Inserted;
Worklist.push_back(N);
Inserted.insert(N);
SmallVector<CombineResult> CombinesToApply;
while (!Worklist.empty()) {
SDNode *Root = Worklist.pop_back_val();
NodeExtensionHelper LHS(Root, 0, DAG, Subtarget);
NodeExtensionHelper RHS(Root, 1, DAG, Subtarget);
auto AppendUsersIfNeeded = [&Worklist, &Subtarget,
&Inserted](const NodeExtensionHelper &Op) {
if (Op.needToPromoteOtherUsers()) {
for (SDUse &Use : Op.OrigOperand->uses()) {
SDNode *TheUser = Use.getUser();
if (!NodeExtensionHelper::isSupportedRoot(TheUser, Subtarget))
return false;
// We only support the first 2 operands of FMA.
if (Use.getOperandNo() >= 2)
return false;
if (Inserted.insert(TheUser).second)
Worklist.push_back(TheUser);
}
}
return true;
};
// Control the compile time by limiting the number of node we look at in
// total.
if (Inserted.size() > ExtensionMaxWebSize)
return SDValue();
SmallVector<NodeExtensionHelper::CombineToTry> FoldingStrategies =
NodeExtensionHelper::getSupportedFoldings(Root);
assert(!FoldingStrategies.empty() && "Nothing to be folded");
bool Matched = false;
for (int Attempt = 0;
(Attempt != 1 + NodeExtensionHelper::isCommutative(Root)) && !Matched;
++Attempt) {
for (NodeExtensionHelper::CombineToTry FoldingStrategy :
FoldingStrategies) {
std::optional<CombineResult> Res =
FoldingStrategy(Root, LHS, RHS, DAG, Subtarget);
if (Res) {
Matched = true;
CombinesToApply.push_back(*Res);
// All the inputs that are extended need to be folded, otherwise
// we would be leaving the old input (since it is may still be used),
// and the new one.
if (Res->LHSExt.has_value())
if (!AppendUsersIfNeeded(LHS))
return SDValue();
if (Res->RHSExt.has_value())
if (!AppendUsersIfNeeded(RHS))
return SDValue();
break;
}
}
std::swap(LHS, RHS);
}
// Right now we do an all or nothing approach.
if (!Matched)
return SDValue();
}
// Store the value for the replacement of the input node separately.
SDValue InputRootReplacement;
// We do the RAUW after we materialize all the combines, because some replaced
// nodes may be feeding some of the yet-to-be-replaced nodes. Put differently,
// some of these nodes may appear in the NodeExtensionHelpers of some of the
// yet-to-be-visited CombinesToApply roots.
SmallVector<std::pair<SDValue, SDValue>> ValuesToReplace;
ValuesToReplace.reserve(CombinesToApply.size());
for (CombineResult Res : CombinesToApply) {
SDValue NewValue = Res.materialize(DAG, Subtarget);
if (!InputRootReplacement) {
assert(Res.Root == N &&
"First element is expected to be the current node");
InputRootReplacement = NewValue;
} else {
ValuesToReplace.emplace_back(SDValue(Res.Root, 0), NewValue);
}
}
for (std::pair<SDValue, SDValue> OldNewValues : ValuesToReplace) {
DAG.ReplaceAllUsesOfValueWith(OldNewValues.first, OldNewValues.second);
DCI.AddToWorklist(OldNewValues.second.getNode());
}
return InputRootReplacement;
}
// Fold (vwadd(u).wv y, (vmerge cond, x, 0)) -> vwadd(u).wv y, x, y, cond
// (vwsub(u).wv y, (vmerge cond, x, 0)) -> vwsub(u).wv y, x, y, cond
// y will be the Passthru and cond will be the Mask.
static SDValue combineVWADDSUBWSelect(SDNode *N, SelectionDAG &DAG) {
unsigned Opc = N->getOpcode();
assert(Opc == RISCVISD::VWADD_W_VL || Opc == RISCVISD::VWADDU_W_VL ||
Opc == RISCVISD::VWSUB_W_VL || Opc == RISCVISD::VWSUBU_W_VL);
SDValue Y = N->getOperand(0);
SDValue MergeOp = N->getOperand(1);
unsigned MergeOpc = MergeOp.getOpcode();
if (MergeOpc != RISCVISD::VMERGE_VL && MergeOpc != ISD::VSELECT)
return SDValue();
SDValue X = MergeOp->getOperand(1);
if (!MergeOp.hasOneUse())
return SDValue();
// Passthru should be undef
SDValue Passthru = N->getOperand(2);
if (!Passthru.isUndef())
return SDValue();
// Mask should be all ones
SDValue Mask = N->getOperand(3);
if (Mask.getOpcode() != RISCVISD::VMSET_VL)
return SDValue();
// False value of MergeOp should be all zeros
SDValue Z = MergeOp->getOperand(2);
if (Z.getOpcode() == ISD::INSERT_SUBVECTOR &&
(isNullOrNullSplat(Z.getOperand(0)) || Z.getOperand(0).isUndef()))
Z = Z.getOperand(1);
if (!ISD::isConstantSplatVectorAllZeros(Z.getNode()))
return SDValue();
return DAG.getNode(Opc, SDLoc(N), N->getValueType(0),
{Y, X, Y, MergeOp->getOperand(0), N->getOperand(4)},
N->getFlags());
}
static SDValue performVWADDSUBW_VLCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
[[maybe_unused]] unsigned Opc = N->getOpcode();
assert(Opc == RISCVISD::VWADD_W_VL || Opc == RISCVISD::VWADDU_W_VL ||
Opc == RISCVISD::VWSUB_W_VL || Opc == RISCVISD::VWSUBU_W_VL);
if (SDValue V = combineOp_VLToVWOp_VL(N, DCI, Subtarget))
return V;
return combineVWADDSUBWSelect(N, DCI.DAG);
}
// Helper function for performMemPairCombine.
// Try to combine the memory loads/stores LSNode1 and LSNode2
// into a single memory pair operation.
static SDValue tryMemPairCombine(SelectionDAG &DAG, LSBaseSDNode *LSNode1,
LSBaseSDNode *LSNode2, SDValue BasePtr,
uint64_t Imm) {
SmallPtrSet<const SDNode *, 32> Visited;
SmallVector<const SDNode *, 8> Worklist = {LSNode1, LSNode2};
if (SDNode::hasPredecessorHelper(LSNode1, Visited, Worklist) ||
SDNode::hasPredecessorHelper(LSNode2, Visited, Worklist))
return SDValue();
MachineFunction &MF = DAG.getMachineFunction();
const RISCVSubtarget &Subtarget = MF.getSubtarget<RISCVSubtarget>();
// The new operation has twice the width.
MVT XLenVT = Subtarget.getXLenVT();
EVT MemVT = LSNode1->getMemoryVT();
EVT NewMemVT = (MemVT == MVT::i32) ? MVT::i64 : MVT::i128;
MachineMemOperand *MMO = LSNode1->getMemOperand();
MachineMemOperand *NewMMO = MF.getMachineMemOperand(
MMO, MMO->getPointerInfo(), MemVT == MVT::i32 ? 8 : 16);
if (LSNode1->getOpcode() == ISD::LOAD) {
auto Ext = cast<LoadSDNode>(LSNode1)->getExtensionType();
unsigned Opcode;
if (MemVT == MVT::i32)
Opcode = (Ext == ISD::ZEXTLOAD) ? RISCVISD::TH_LWUD : RISCVISD::TH_LWD;
else
Opcode = RISCVISD::TH_LDD;
SDValue Res = DAG.getMemIntrinsicNode(
Opcode, SDLoc(LSNode1), DAG.getVTList({XLenVT, XLenVT, MVT::Other}),
{LSNode1->getChain(), BasePtr,
DAG.getConstant(Imm, SDLoc(LSNode1), XLenVT)},
NewMemVT, NewMMO);
SDValue Node1 =
DAG.getMergeValues({Res.getValue(0), Res.getValue(2)}, SDLoc(LSNode1));
SDValue Node2 =
DAG.getMergeValues({Res.getValue(1), Res.getValue(2)}, SDLoc(LSNode2));
DAG.ReplaceAllUsesWith(LSNode2, Node2.getNode());
return Node1;
} else {
unsigned Opcode = (MemVT == MVT::i32) ? RISCVISD::TH_SWD : RISCVISD::TH_SDD;
SDValue Res = DAG.getMemIntrinsicNode(
Opcode, SDLoc(LSNode1), DAG.getVTList(MVT::Other),
{LSNode1->getChain(), LSNode1->getOperand(1), LSNode2->getOperand(1),
BasePtr, DAG.getConstant(Imm, SDLoc(LSNode1), XLenVT)},
NewMemVT, NewMMO);
DAG.ReplaceAllUsesWith(LSNode2, Res.getNode());
return Res;
}
}
// Try to combine two adjacent loads/stores to a single pair instruction from
// the XTHeadMemPair vendor extension.
static SDValue performMemPairCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI) {
SelectionDAG &DAG = DCI.DAG;
MachineFunction &MF = DAG.getMachineFunction();
const RISCVSubtarget &Subtarget = MF.getSubtarget<RISCVSubtarget>();
// Target does not support load/store pair.
if (!Subtarget.hasVendorXTHeadMemPair())
return SDValue();
LSBaseSDNode *LSNode1 = cast<LSBaseSDNode>(N);
EVT MemVT = LSNode1->getMemoryVT();
unsigned OpNum = LSNode1->getOpcode() == ISD::LOAD ? 1 : 2;
// No volatile, indexed or atomic loads/stores.
if (!LSNode1->isSimple() || LSNode1->isIndexed())
return SDValue();
// Function to get a base + constant representation from a memory value.
auto ExtractBaseAndOffset = [](SDValue Ptr) -> std::pair<SDValue, uint64_t> {
if (Ptr->getOpcode() == ISD::ADD)
if (auto *C1 = dyn_cast<ConstantSDNode>(Ptr->getOperand(1)))
return {Ptr->getOperand(0), C1->getZExtValue()};
return {Ptr, 0};
};
auto [Base1, Offset1] = ExtractBaseAndOffset(LSNode1->getOperand(OpNum));
SDValue Chain = N->getOperand(0);
for (SDUse &Use : Chain->uses()) {
if (Use.getUser() != N && Use.getResNo() == 0 &&
Use.getUser()->getOpcode() == N->getOpcode()) {
LSBaseSDNode *LSNode2 = cast<LSBaseSDNode>(Use.getUser());
// No volatile, indexed or atomic loads/stores.
if (!LSNode2->isSimple() || LSNode2->isIndexed())
continue;
// Check if LSNode1 and LSNode2 have the same type and extension.
if (LSNode1->getOpcode() == ISD::LOAD)
if (cast<LoadSDNode>(LSNode2)->getExtensionType() !=
cast<LoadSDNode>(LSNode1)->getExtensionType())
continue;
if (LSNode1->getMemoryVT() != LSNode2->getMemoryVT())
continue;
auto [Base2, Offset2] = ExtractBaseAndOffset(LSNode2->getOperand(OpNum));
// Check if the base pointer is the same for both instruction.
if (Base1 != Base2)
continue;
// Check if the offsets match the XTHeadMemPair encoding constraints.
bool Valid = false;
if (MemVT == MVT::i32) {
// Check for adjacent i32 values and a 2-bit index.
if ((Offset1 + 4 == Offset2) && isShiftedUInt<2, 3>(Offset1))
Valid = true;
} else if (MemVT == MVT::i64) {
// Check for adjacent i64 values and a 2-bit index.
if ((Offset1 + 8 == Offset2) && isShiftedUInt<2, 4>(Offset1))
Valid = true;
}
if (!Valid)
continue;
// Try to combine.
if (SDValue Res =
tryMemPairCombine(DAG, LSNode1, LSNode2, Base1, Offset1))
return Res;
}
}
return SDValue();
}
// Fold
// (fp_to_int (froundeven X)) -> fcvt X, rne
// (fp_to_int (ftrunc X)) -> fcvt X, rtz
// (fp_to_int (ffloor X)) -> fcvt X, rdn
// (fp_to_int (fceil X)) -> fcvt X, rup
// (fp_to_int (fround X)) -> fcvt X, rmm
// (fp_to_int (frint X)) -> fcvt X
static SDValue performFP_TO_INTCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
SelectionDAG &DAG = DCI.DAG;
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
MVT XLenVT = Subtarget.getXLenVT();
SDValue Src = N->getOperand(0);
// Don't do this for strict-fp Src.
if (Src->isStrictFPOpcode())
return SDValue();
// Ensure the FP type is legal.
if (!TLI.isTypeLegal(Src.getValueType()))
return SDValue();
// Don't do this for f16 with Zfhmin and not Zfh.
if (Src.getValueType() == MVT::f16 && !Subtarget.hasStdExtZfh())
return SDValue();
RISCVFPRndMode::RoundingMode FRM = matchRoundingOp(Src.getOpcode());
// If the result is invalid, we didn't find a foldable instruction.
if (FRM == RISCVFPRndMode::Invalid)
return SDValue();
SDLoc DL(N);
bool IsSigned = N->getOpcode() == ISD::FP_TO_SINT;
EVT VT = N->getValueType(0);
if (VT.isVector() && TLI.isTypeLegal(VT)) {
MVT SrcVT = Src.getSimpleValueType();
MVT SrcContainerVT = SrcVT;
MVT ContainerVT = VT.getSimpleVT();
SDValue XVal = Src.getOperand(0);
// For widening and narrowing conversions we just combine it into a
// VFCVT_..._VL node, as there are no specific VFWCVT/VFNCVT VL nodes. They
// end up getting lowered to their appropriate pseudo instructions based on
// their operand types
if (VT.getScalarSizeInBits() > SrcVT.getScalarSizeInBits() * 2 ||
VT.getScalarSizeInBits() * 2 < SrcVT.getScalarSizeInBits())
return SDValue();
// Make fixed-length vectors scalable first
if (SrcVT.isFixedLengthVector()) {
SrcContainerVT = getContainerForFixedLengthVector(DAG, SrcVT, Subtarget);
XVal = convertToScalableVector(SrcContainerVT, XVal, DAG, Subtarget);
ContainerVT =
getContainerForFixedLengthVector(DAG, ContainerVT, Subtarget);
}
auto [Mask, VL] =
getDefaultVLOps(SrcVT, SrcContainerVT, DL, DAG, Subtarget);
SDValue FpToInt;
if (FRM == RISCVFPRndMode::RTZ) {
// Use the dedicated trunc static rounding mode if we're truncating so we
// don't need to generate calls to fsrmi/fsrm
unsigned Opc =
IsSigned ? RISCVISD::VFCVT_RTZ_X_F_VL : RISCVISD::VFCVT_RTZ_XU_F_VL;
FpToInt = DAG.getNode(Opc, DL, ContainerVT, XVal, Mask, VL);
} else {
unsigned Opc =
IsSigned ? RISCVISD::VFCVT_RM_X_F_VL : RISCVISD::VFCVT_RM_XU_F_VL;
FpToInt = DAG.getNode(Opc, DL, ContainerVT, XVal, Mask,
DAG.getTargetConstant(FRM, DL, XLenVT), VL);
}
// If converted from fixed-length to scalable, convert back
if (VT.isFixedLengthVector())
FpToInt = convertFromScalableVector(VT, FpToInt, DAG, Subtarget);
return FpToInt;
}
// Only handle XLen or i32 types. Other types narrower than XLen will
// eventually be legalized to XLenVT.
if (VT != MVT::i32 && VT != XLenVT)
return SDValue();
unsigned Opc;
if (VT == XLenVT)
Opc = IsSigned ? RISCVISD::FCVT_X : RISCVISD::FCVT_XU;
else
Opc = IsSigned ? RISCVISD::FCVT_W_RV64 : RISCVISD::FCVT_WU_RV64;
SDValue FpToInt = DAG.getNode(Opc, DL, XLenVT, Src.getOperand(0),
DAG.getTargetConstant(FRM, DL, XLenVT));
return DAG.getNode(ISD::TRUNCATE, DL, VT, FpToInt);
}
// Fold
// (fp_to_int_sat (froundeven X)) -> (select X == nan, 0, (fcvt X, rne))
// (fp_to_int_sat (ftrunc X)) -> (select X == nan, 0, (fcvt X, rtz))
// (fp_to_int_sat (ffloor X)) -> (select X == nan, 0, (fcvt X, rdn))
// (fp_to_int_sat (fceil X)) -> (select X == nan, 0, (fcvt X, rup))
// (fp_to_int_sat (fround X)) -> (select X == nan, 0, (fcvt X, rmm))
// (fp_to_int_sat (frint X)) -> (select X == nan, 0, (fcvt X, dyn))
static SDValue performFP_TO_INT_SATCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
SelectionDAG &DAG = DCI.DAG;
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
MVT XLenVT = Subtarget.getXLenVT();
// Only handle XLen types. Other types narrower than XLen will eventually be
// legalized to XLenVT.
EVT DstVT = N->getValueType(0);
if (DstVT != XLenVT)
return SDValue();
SDValue Src = N->getOperand(0);
// Don't do this for strict-fp Src.
if (Src->isStrictFPOpcode())
return SDValue();
// Ensure the FP type is also legal.
if (!TLI.isTypeLegal(Src.getValueType()))
return SDValue();
// Don't do this for f16 with Zfhmin and not Zfh.
if (Src.getValueType() == MVT::f16 && !Subtarget.hasStdExtZfh())
return SDValue();
EVT SatVT = cast<VTSDNode>(N->getOperand(1))->getVT();
RISCVFPRndMode::RoundingMode FRM = matchRoundingOp(Src.getOpcode());
if (FRM == RISCVFPRndMode::Invalid)
return SDValue();
bool IsSigned = N->getOpcode() == ISD::FP_TO_SINT_SAT;
unsigned Opc;
if (SatVT == DstVT)
Opc = IsSigned ? RISCVISD::FCVT_X : RISCVISD::FCVT_XU;
else if (DstVT == MVT::i64 && SatVT == MVT::i32)
Opc = IsSigned ? RISCVISD::FCVT_W_RV64 : RISCVISD::FCVT_WU_RV64;
else
return SDValue();
// FIXME: Support other SatVTs by clamping before or after the conversion.
Src = Src.getOperand(0);
SDLoc DL(N);
SDValue FpToInt = DAG.getNode(Opc, DL, XLenVT, Src,
DAG.getTargetConstant(FRM, DL, XLenVT));
// fcvt.wu.* sign extends bit 31 on RV64. FP_TO_UINT_SAT expects to zero
// extend.
if (Opc == RISCVISD::FCVT_WU_RV64)
FpToInt = DAG.getZeroExtendInReg(FpToInt, DL, MVT::i32);
// RISC-V FP-to-int conversions saturate to the destination register size, but
// don't produce 0 for nan.
SDValue ZeroInt = DAG.getConstant(0, DL, DstVT);
return DAG.getSelectCC(DL, Src, Src, ZeroInt, FpToInt, ISD::CondCode::SETUO);
}
// Combine (bitreverse (bswap X)) to the BREV8 GREVI encoding if the type is
// smaller than XLenVT.
static SDValue performBITREVERSECombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(Subtarget.hasStdExtZbkb() && "Unexpected extension");
SDValue Src = N->getOperand(0);
if (Src.getOpcode() != ISD::BSWAP)
return SDValue();
EVT VT = N->getValueType(0);
if (!VT.isScalarInteger() || VT.getSizeInBits() >= Subtarget.getXLen() ||
!llvm::has_single_bit<uint32_t>(VT.getSizeInBits()))
return SDValue();
SDLoc DL(N);
return DAG.getNode(RISCVISD::BREV8, DL, VT, Src.getOperand(0));
}
static SDValue performVP_REVERSECombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
// Fold:
// vp.reverse(vp.load(ADDR, MASK)) -> vp.strided.load(ADDR, -1, MASK)
// Check if its first operand is a vp.load.
auto *VPLoad = dyn_cast<VPLoadSDNode>(N->getOperand(0));
if (!VPLoad)
return SDValue();
EVT LoadVT = VPLoad->getValueType(0);
// We do not have a strided_load version for masks, and the evl of vp.reverse
// and vp.load should always be the same.
if (!LoadVT.getVectorElementType().isByteSized() ||
N->getOperand(2) != VPLoad->getVectorLength() ||
!N->getOperand(0).hasOneUse())
return SDValue();
// Check if the mask of outer vp.reverse are all 1's.
if (!isOneOrOneSplat(N->getOperand(1)))
return SDValue();
SDValue LoadMask = VPLoad->getMask();
// If Mask is all ones, then load is unmasked and can be reversed.
if (!isOneOrOneSplat(LoadMask)) {
// If the mask is not all ones, we can reverse the load if the mask was also
// reversed by an unmasked vp.reverse with the same EVL.
if (LoadMask.getOpcode() != ISD::EXPERIMENTAL_VP_REVERSE ||
!isOneOrOneSplat(LoadMask.getOperand(1)) ||
LoadMask.getOperand(2) != VPLoad->getVectorLength())
return SDValue();
LoadMask = LoadMask.getOperand(0);
}
// Base = LoadAddr + (NumElem - 1) * ElemWidthByte
SDLoc DL(N);
MVT XLenVT = Subtarget.getXLenVT();
SDValue NumElem = VPLoad->getVectorLength();
uint64_t ElemWidthByte = VPLoad->getValueType(0).getScalarSizeInBits() / 8;
SDValue Temp1 = DAG.getNode(ISD::SUB, DL, XLenVT, NumElem,
DAG.getConstant(1, DL, XLenVT));
SDValue Temp2 = DAG.getNode(ISD::MUL, DL, XLenVT, Temp1,
DAG.getConstant(ElemWidthByte, DL, XLenVT));
SDValue Base = DAG.getNode(ISD::ADD, DL, XLenVT, VPLoad->getBasePtr(), Temp2);
SDValue Stride = DAG.getSignedConstant(-ElemWidthByte, DL, XLenVT);
MachineFunction &MF = DAG.getMachineFunction();
MachinePointerInfo PtrInfo(VPLoad->getAddressSpace());
MachineMemOperand *MMO = MF.getMachineMemOperand(
PtrInfo, VPLoad->getMemOperand()->getFlags(),
LocationSize::beforeOrAfterPointer(), VPLoad->getAlign());
SDValue Ret = DAG.getStridedLoadVP(
LoadVT, DL, VPLoad->getChain(), Base, Stride, LoadMask,
VPLoad->getVectorLength(), MMO, VPLoad->isExpandingLoad());
DAG.ReplaceAllUsesOfValueWith(SDValue(VPLoad, 1), Ret.getValue(1));
return Ret;
}
static SDValue performVP_STORECombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
// Fold:
// vp.store(vp.reverse(VAL), ADDR, MASK) -> vp.strided.store(VAL, NEW_ADDR,
// -1, MASK)
auto *VPStore = cast<VPStoreSDNode>(N);
if (VPStore->getValue().getOpcode() != ISD::EXPERIMENTAL_VP_REVERSE)
return SDValue();
SDValue VPReverse = VPStore->getValue();
EVT ReverseVT = VPReverse->getValueType(0);
// We do not have a strided_store version for masks, and the evl of vp.reverse
// and vp.store should always be the same.
if (!ReverseVT.getVectorElementType().isByteSized() ||
VPStore->getVectorLength() != VPReverse.getOperand(2) ||
!VPReverse.hasOneUse())
return SDValue();
SDValue StoreMask = VPStore->getMask();
// If Mask is all ones, then load is unmasked and can be reversed.
if (!isOneOrOneSplat(StoreMask)) {
// If the mask is not all ones, we can reverse the store if the mask was
// also reversed by an unmasked vp.reverse with the same EVL.
if (StoreMask.getOpcode() != ISD::EXPERIMENTAL_VP_REVERSE ||
!isOneOrOneSplat(StoreMask.getOperand(1)) ||
StoreMask.getOperand(2) != VPStore->getVectorLength())
return SDValue();
StoreMask = StoreMask.getOperand(0);
}
// Base = StoreAddr + (NumElem - 1) * ElemWidthByte
SDLoc DL(N);
MVT XLenVT = Subtarget.getXLenVT();
SDValue NumElem = VPStore->getVectorLength();
uint64_t ElemWidthByte = VPReverse.getValueType().getScalarSizeInBits() / 8;
SDValue Temp1 = DAG.getNode(ISD::SUB, DL, XLenVT, NumElem,
DAG.getConstant(1, DL, XLenVT));
SDValue Temp2 = DAG.getNode(ISD::MUL, DL, XLenVT, Temp1,
DAG.getConstant(ElemWidthByte, DL, XLenVT));
SDValue Base =
DAG.getNode(ISD::ADD, DL, XLenVT, VPStore->getBasePtr(), Temp2);
SDValue Stride = DAG.getSignedConstant(-ElemWidthByte, DL, XLenVT);
MachineFunction &MF = DAG.getMachineFunction();
MachinePointerInfo PtrInfo(VPStore->getAddressSpace());
MachineMemOperand *MMO = MF.getMachineMemOperand(
PtrInfo, VPStore->getMemOperand()->getFlags(),
LocationSize::beforeOrAfterPointer(), VPStore->getAlign());
return DAG.getStridedStoreVP(
VPStore->getChain(), DL, VPReverse.getOperand(0), Base,
VPStore->getOffset(), Stride, StoreMask, VPStore->getVectorLength(),
VPStore->getMemoryVT(), MMO, VPStore->getAddressingMode(),
VPStore->isTruncatingStore(), VPStore->isCompressingStore());
}
// Peephole avgceil pattern.
// %1 = zext <N x i8> %a to <N x i32>
// %2 = zext <N x i8> %b to <N x i32>
// %3 = add nuw nsw <N x i32> %1, splat (i32 1)
// %4 = add nuw nsw <N x i32> %3, %2
// %5 = lshr <N x i32> %4, splat (i32 1)
// %6 = trunc <N x i32> %5 to <N x i8>
static SDValue performVP_TRUNCATECombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
EVT VT = N->getValueType(0);
// Ignore fixed vectors.
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (!VT.isScalableVector() || !TLI.isTypeLegal(VT))
return SDValue();
SDValue In = N->getOperand(0);
SDValue Mask = N->getOperand(1);
SDValue VL = N->getOperand(2);
// Input should be a vp_srl with same mask and VL.
if (In.getOpcode() != ISD::VP_SRL || In.getOperand(2) != Mask ||
In.getOperand(3) != VL)
return SDValue();
// Shift amount should be 1.
if (!isOneOrOneSplat(In.getOperand(1)))
return SDValue();
// Shifted value should be a vp_add with same mask and VL.
SDValue LHS = In.getOperand(0);
if (LHS.getOpcode() != ISD::VP_ADD || LHS.getOperand(2) != Mask ||
LHS.getOperand(3) != VL)
return SDValue();
SDValue Operands[3];
// Matches another VP_ADD with same VL and Mask.
auto FindAdd = [&](SDValue V, SDValue Other) {
if (V.getOpcode() != ISD::VP_ADD || V.getOperand(2) != Mask ||
V.getOperand(3) != VL)
return false;
Operands[0] = Other;
Operands[1] = V.getOperand(1);
Operands[2] = V.getOperand(0);
return true;
};
// We need to find another VP_ADD in one of the operands.
SDValue LHS0 = LHS.getOperand(0);
SDValue LHS1 = LHS.getOperand(1);
if (!FindAdd(LHS0, LHS1) && !FindAdd(LHS1, LHS0))
return SDValue();
// Now we have three operands of two additions. Check that one of them is a
// constant vector with ones.
auto I = llvm::find_if(Operands,
[](const SDValue &Op) { return isOneOrOneSplat(Op); });
if (I == std::end(Operands))
return SDValue();
// We found a vector with ones, move if it to the end of the Operands array.
std::swap(*I, Operands[2]);
// Make sure the other 2 operands can be promoted from the result type.
for (SDValue Op : drop_end(Operands)) {
if (Op.getOpcode() != ISD::VP_ZERO_EXTEND || Op.getOperand(1) != Mask ||
Op.getOperand(2) != VL)
return SDValue();
// Input must be the same size or smaller than our result.
if (Op.getOperand(0).getScalarValueSizeInBits() > VT.getScalarSizeInBits())
return SDValue();
}
// Pattern is detected.
// Rebuild the zero extends in case the inputs are smaller than our result.
SDValue NewOp0 = DAG.getNode(ISD::VP_ZERO_EXTEND, SDLoc(Operands[0]), VT,
Operands[0].getOperand(0), Mask, VL);
SDValue NewOp1 = DAG.getNode(ISD::VP_ZERO_EXTEND, SDLoc(Operands[1]), VT,
Operands[1].getOperand(0), Mask, VL);
// Build a AVGCEILU_VL which will be selected as a VAADDU with RNU rounding
// mode.
SDLoc DL(N);
return DAG.getNode(RISCVISD::AVGCEILU_VL, DL, VT,
{NewOp0, NewOp1, DAG.getUNDEF(VT), Mask, VL});
}
// Convert from one FMA opcode to another based on whether we are negating the
// multiply result and/or the accumulator.
// NOTE: Only supports RVV operations with VL.
static unsigned negateFMAOpcode(unsigned Opcode, bool NegMul, bool NegAcc) {
// Negating the multiply result changes ADD<->SUB and toggles 'N'.
if (NegMul) {
// clang-format off
switch (Opcode) {
default: llvm_unreachable("Unexpected opcode");
case RISCVISD::VFMADD_VL: Opcode = RISCVISD::VFNMSUB_VL; break;
case RISCVISD::VFNMSUB_VL: Opcode = RISCVISD::VFMADD_VL; break;
case RISCVISD::VFNMADD_VL: Opcode = RISCVISD::VFMSUB_VL; break;
case RISCVISD::VFMSUB_VL: Opcode = RISCVISD::VFNMADD_VL; break;
case RISCVISD::STRICT_VFMADD_VL: Opcode = RISCVISD::STRICT_VFNMSUB_VL; break;
case RISCVISD::STRICT_VFNMSUB_VL: Opcode = RISCVISD::STRICT_VFMADD_VL; break;
case RISCVISD::STRICT_VFNMADD_VL: Opcode = RISCVISD::STRICT_VFMSUB_VL; break;
case RISCVISD::STRICT_VFMSUB_VL: Opcode = RISCVISD::STRICT_VFNMADD_VL; break;
}
// clang-format on
}
// Negating the accumulator changes ADD<->SUB.
if (NegAcc) {
// clang-format off
switch (Opcode) {
default: llvm_unreachable("Unexpected opcode");
case RISCVISD::VFMADD_VL: Opcode = RISCVISD::VFMSUB_VL; break;
case RISCVISD::VFMSUB_VL: Opcode = RISCVISD::VFMADD_VL; break;
case RISCVISD::VFNMADD_VL: Opcode = RISCVISD::VFNMSUB_VL; break;
case RISCVISD::VFNMSUB_VL: Opcode = RISCVISD::VFNMADD_VL; break;
case RISCVISD::STRICT_VFMADD_VL: Opcode = RISCVISD::STRICT_VFMSUB_VL; break;
case RISCVISD::STRICT_VFMSUB_VL: Opcode = RISCVISD::STRICT_VFMADD_VL; break;
case RISCVISD::STRICT_VFNMADD_VL: Opcode = RISCVISD::STRICT_VFNMSUB_VL; break;
case RISCVISD::STRICT_VFNMSUB_VL: Opcode = RISCVISD::STRICT_VFNMADD_VL; break;
}
// clang-format on
}
return Opcode;
}
static SDValue combineVFMADD_VLWithVFNEG_VL(SDNode *N, SelectionDAG &DAG) {
// Fold FNEG_VL into FMA opcodes.
// The first operand of strict-fp is chain.
bool IsStrict =
DAG.getSelectionDAGInfo().isTargetStrictFPOpcode(N->getOpcode());
unsigned Offset = IsStrict ? 1 : 0;
SDValue A = N->getOperand(0 + Offset);
SDValue B = N->getOperand(1 + Offset);
SDValue C = N->getOperand(2 + Offset);
SDValue Mask = N->getOperand(3 + Offset);
SDValue VL = N->getOperand(4 + Offset);
auto invertIfNegative = [&Mask, &VL](SDValue &V) {
if (V.getOpcode() == RISCVISD::FNEG_VL && V.getOperand(1) == Mask &&
V.getOperand(2) == VL) {
// Return the negated input.
V = V.getOperand(0);
return true;
}
return false;
};
bool NegA = invertIfNegative(A);
bool NegB = invertIfNegative(B);
bool NegC = invertIfNegative(C);
// If no operands are negated, we're done.
if (!NegA && !NegB && !NegC)
return SDValue();
unsigned NewOpcode = negateFMAOpcode(N->getOpcode(), NegA != NegB, NegC);
if (IsStrict)
return DAG.getNode(NewOpcode, SDLoc(N), N->getVTList(),
{N->getOperand(0), A, B, C, Mask, VL});
return DAG.getNode(NewOpcode, SDLoc(N), N->getValueType(0), A, B, C, Mask,
VL);
}
static SDValue performVFMADD_VLCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
SelectionDAG &DAG = DCI.DAG;
if (SDValue V = combineVFMADD_VLWithVFNEG_VL(N, DAG))
return V;
// FIXME: Ignore strict opcodes for now.
if (DAG.getSelectionDAGInfo().isTargetStrictFPOpcode(N->getOpcode()))
return SDValue();
return combineOp_VLToVWOp_VL(N, DCI, Subtarget);
}
static SDValue performSRACombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(N->getOpcode() == ISD::SRA && "Unexpected opcode");
EVT VT = N->getValueType(0);
if (VT != Subtarget.getXLenVT())
return SDValue();
if (!isa<ConstantSDNode>(N->getOperand(1)))
return SDValue();
uint64_t ShAmt = N->getConstantOperandVal(1);
SDValue N0 = N->getOperand(0);
// Combine (sra (sext_inreg (shl X, C1), iX), C2) ->
// (sra (shl X, C1+(XLen-iX)), C2+(XLen-iX)) so it gets selected as SLLI+SRAI.
if (N0.getOpcode() == ISD::SIGN_EXTEND_INREG && N0.hasOneUse()) {
unsigned ExtSize =
cast<VTSDNode>(N0.getOperand(1))->getVT().getSizeInBits();
if (ShAmt < ExtSize && N0.getOperand(0).getOpcode() == ISD::SHL &&
N0.getOperand(0).hasOneUse() &&
isa<ConstantSDNode>(N0.getOperand(0).getOperand(1))) {
uint64_t LShAmt = N0.getOperand(0).getConstantOperandVal(1);
if (LShAmt < ExtSize) {
unsigned Size = VT.getSizeInBits();
SDLoc ShlDL(N0.getOperand(0));
SDValue Shl =
DAG.getNode(ISD::SHL, ShlDL, VT, N0.getOperand(0).getOperand(0),
DAG.getConstant(LShAmt + (Size - ExtSize), ShlDL, VT));
SDLoc DL(N);
return DAG.getNode(ISD::SRA, DL, VT, Shl,
DAG.getConstant(ShAmt + (Size - ExtSize), DL, VT));
}
}
}
if (ShAmt > 32 || VT != MVT::i64)
return SDValue();
// Combine (sra (shl X, 32), 32 - C) -> (shl (sext_inreg X, i32), C)
// FIXME: Should this be a generic combine? There's a similar combine on X86.
//
// Also try these folds where an add or sub is in the middle.
// (sra (add (shl X, 32), C1), 32 - C) -> (shl (sext_inreg (add X, C1), C)
// (sra (sub C1, (shl X, 32)), 32 - C) -> (shl (sext_inreg (sub C1, X), C)
SDValue Shl;
ConstantSDNode *AddC = nullptr;
// We might have an ADD or SUB between the SRA and SHL.
bool IsAdd = N0.getOpcode() == ISD::ADD;
if ((IsAdd || N0.getOpcode() == ISD::SUB)) {
// Other operand needs to be a constant we can modify.
AddC = dyn_cast<ConstantSDNode>(N0.getOperand(IsAdd ? 1 : 0));
if (!AddC)
return SDValue();
// AddC needs to have at least 32 trailing zeros.
if (llvm::countr_zero(AddC->getZExtValue()) < 32)
return SDValue();
// All users should be a shift by constant less than or equal to 32. This
// ensures we'll do this optimization for each of them to produce an
// add/sub+sext_inreg they can all share.
for (SDNode *U : N0->users()) {
if (U->getOpcode() != ISD::SRA ||
!isa<ConstantSDNode>(U->getOperand(1)) ||
U->getConstantOperandVal(1) > 32)
return SDValue();
}
Shl = N0.getOperand(IsAdd ? 0 : 1);
} else {
// Not an ADD or SUB.
Shl = N0;
}
// Look for a shift left by 32.
if (Shl.getOpcode() != ISD::SHL || !isa<ConstantSDNode>(Shl.getOperand(1)) ||
Shl.getConstantOperandVal(1) != 32)
return SDValue();
// We if we didn't look through an add/sub, then the shl should have one use.
// If we did look through an add/sub, the sext_inreg we create is free so
// we're only creating 2 new instructions. It's enough to only remove the
// original sra+add/sub.
if (!AddC && !Shl.hasOneUse())
return SDValue();
SDLoc DL(N);
SDValue In = Shl.getOperand(0);
// If we looked through an ADD or SUB, we need to rebuild it with the shifted
// constant.
if (AddC) {
SDValue ShiftedAddC =
DAG.getConstant(AddC->getZExtValue() >> 32, DL, MVT::i64);
if (IsAdd)
In = DAG.getNode(ISD::ADD, DL, MVT::i64, In, ShiftedAddC);
else
In = DAG.getNode(ISD::SUB, DL, MVT::i64, ShiftedAddC, In);
}
SDValue SExt = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i64, In,
DAG.getValueType(MVT::i32));
if (ShAmt == 32)
return SExt;
return DAG.getNode(
ISD::SHL, DL, MVT::i64, SExt,
DAG.getConstant(32 - ShAmt, DL, MVT::i64));
}
// Invert (and/or (set cc X, Y), (xor Z, 1)) to (or/and (set !cc X, Y)), Z) if
// the result is used as the condition of a br_cc or select_cc we can invert,
// inverting the setcc is free, and Z is 0/1. Caller will invert the
// br_cc/select_cc.
static SDValue tryDemorganOfBooleanCondition(SDValue Cond, SelectionDAG &DAG) {
bool IsAnd = Cond.getOpcode() == ISD::AND;
if (!IsAnd && Cond.getOpcode() != ISD::OR)
return SDValue();
if (!Cond.hasOneUse())
return SDValue();
SDValue Setcc = Cond.getOperand(0);
SDValue Xor = Cond.getOperand(1);
// Canonicalize setcc to LHS.
if (Setcc.getOpcode() != ISD::SETCC)
std::swap(Setcc, Xor);
// LHS should be a setcc and RHS should be an xor.
if (Setcc.getOpcode() != ISD::SETCC || !Setcc.hasOneUse() ||
Xor.getOpcode() != ISD::XOR || !Xor.hasOneUse())
return SDValue();
// If the condition is an And, SimplifyDemandedBits may have changed
// (xor Z, 1) to (not Z).
SDValue Xor1 = Xor.getOperand(1);
if (!isOneConstant(Xor1) && !(IsAnd && isAllOnesConstant(Xor1)))
return SDValue();
EVT VT = Cond.getValueType();
SDValue Xor0 = Xor.getOperand(0);
// The LHS of the xor needs to be 0/1.
APInt Mask = APInt::getBitsSetFrom(VT.getSizeInBits(), 1);
if (!DAG.MaskedValueIsZero(Xor0, Mask))
return SDValue();
// We can only invert integer setccs.
EVT SetCCOpVT = Setcc.getOperand(0).getValueType();
if (!SetCCOpVT.isScalarInteger())
return SDValue();
ISD::CondCode CCVal = cast<CondCodeSDNode>(Setcc.getOperand(2))->get();
if (ISD::isIntEqualitySetCC(CCVal)) {
CCVal = ISD::getSetCCInverse(CCVal, SetCCOpVT);
Setcc = DAG.getSetCC(SDLoc(Setcc), VT, Setcc.getOperand(0),
Setcc.getOperand(1), CCVal);
} else if (CCVal == ISD::SETLT && isNullConstant(Setcc.getOperand(0))) {
// Invert (setlt 0, X) by converting to (setlt X, 1).
Setcc = DAG.getSetCC(SDLoc(Setcc), VT, Setcc.getOperand(1),
DAG.getConstant(1, SDLoc(Setcc), VT), CCVal);
} else if (CCVal == ISD::SETLT && isOneConstant(Setcc.getOperand(1))) {
// (setlt X, 1) by converting to (setlt 0, X).
Setcc = DAG.getSetCC(SDLoc(Setcc), VT,
DAG.getConstant(0, SDLoc(Setcc), VT),
Setcc.getOperand(0), CCVal);
} else
return SDValue();
unsigned Opc = IsAnd ? ISD::OR : ISD::AND;
return DAG.getNode(Opc, SDLoc(Cond), VT, Setcc, Xor.getOperand(0));
}
// Perform common combines for BR_CC and SELECT_CC conditions.
static bool combine_CC(SDValue &LHS, SDValue &RHS, SDValue &CC, const SDLoc &DL,
SelectionDAG &DAG, const RISCVSubtarget &Subtarget) {
ISD::CondCode CCVal = cast<CondCodeSDNode>(CC)->get();
// As far as arithmetic right shift always saves the sign,
// shift can be omitted.
// Fold setlt (sra X, N), 0 -> setlt X, 0 and
// setge (sra X, N), 0 -> setge X, 0
if (isNullConstant(RHS) && (CCVal == ISD::SETGE || CCVal == ISD::SETLT) &&
LHS.getOpcode() == ISD::SRA) {
LHS = LHS.getOperand(0);
return true;
}
if (!ISD::isIntEqualitySetCC(CCVal))
return false;
// Fold ((setlt X, Y), 0, ne) -> (X, Y, lt)
// Sometimes the setcc is introduced after br_cc/select_cc has been formed.
if (LHS.getOpcode() == ISD::SETCC && isNullConstant(RHS) &&
LHS.getOperand(0).getValueType() == Subtarget.getXLenVT()) {
// If we're looking for eq 0 instead of ne 0, we need to invert the
// condition.
bool Invert = CCVal == ISD::SETEQ;
CCVal = cast<CondCodeSDNode>(LHS.getOperand(2))->get();
if (Invert)
CCVal = ISD::getSetCCInverse(CCVal, LHS.getValueType());
RHS = LHS.getOperand(1);
LHS = LHS.getOperand(0);
translateSetCCForBranch(DL, LHS, RHS, CCVal, DAG, Subtarget);
CC = DAG.getCondCode(CCVal);
return true;
}
// If XOR is reused and has an immediate that will fit in XORI,
// do not fold.
auto isXorImmediate = [](const SDValue &Op) -> bool {
if (const auto *XorCnst = dyn_cast<ConstantSDNode>(Op))
return isInt<12>(XorCnst->getSExtValue());
return false;
};
// Fold (X(i1) ^ 1) == 0 -> X != 0
auto singleBitOp = [&DAG](const SDValue &VarOp,
const SDValue &ConstOp) -> bool {
if (const auto *XorCnst = dyn_cast<ConstantSDNode>(ConstOp)) {
const APInt Mask = APInt::getBitsSetFrom(VarOp.getValueSizeInBits(), 1);
return (XorCnst->getSExtValue() == 1) &&
DAG.MaskedValueIsZero(VarOp, Mask);
}
return false;
};
auto onlyUsedBySelectOrBR = [](const SDValue &Op) -> bool {
for (const SDNode *UserNode : Op->users()) {
const unsigned Opcode = UserNode->getOpcode();
if (Opcode != RISCVISD::SELECT_CC && Opcode != RISCVISD::BR_CC)
return false;
}
return true;
};
auto isFoldableXorEq = [isXorImmediate, singleBitOp, onlyUsedBySelectOrBR](
const SDValue &LHS, const SDValue &RHS) -> bool {
return LHS.getOpcode() == ISD::XOR && isNullConstant(RHS) &&
(!isXorImmediate(LHS.getOperand(1)) ||
singleBitOp(LHS.getOperand(0), LHS.getOperand(1)) ||
onlyUsedBySelectOrBR(LHS));
};
// Fold ((xor X, Y), 0, eq/ne) -> (X, Y, eq/ne)
if (isFoldableXorEq(LHS, RHS)) {
RHS = LHS.getOperand(1);
LHS = LHS.getOperand(0);
return true;
}
// Fold ((sext (xor X, C)), 0, eq/ne) -> ((sext(X), C, eq/ne)
if (LHS.getOpcode() == ISD::SIGN_EXTEND_INREG) {
const SDValue LHS0 = LHS.getOperand(0);
if (isFoldableXorEq(LHS0, RHS) && isa<ConstantSDNode>(LHS0.getOperand(1))) {
// SEXT(XOR(X, Y)) -> XOR(SEXT(X), SEXT(Y)))
RHS = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, LHS.getValueType(),
LHS0.getOperand(1), LHS.getOperand(1));
LHS = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, LHS.getValueType(),
LHS0.getOperand(0), LHS.getOperand(1));
return true;
}
}
// Fold ((srl (and X, 1<<C), C), 0, eq/ne) -> ((shl X, XLen-1-C), 0, ge/lt)
if (isNullConstant(RHS) && LHS.getOpcode() == ISD::SRL && LHS.hasOneUse() &&
LHS.getOperand(1).getOpcode() == ISD::Constant) {
SDValue LHS0 = LHS.getOperand(0);
if (LHS0.getOpcode() == ISD::AND &&
LHS0.getOperand(1).getOpcode() == ISD::Constant) {
uint64_t Mask = LHS0.getConstantOperandVal(1);
uint64_t ShAmt = LHS.getConstantOperandVal(1);
if (isPowerOf2_64(Mask) && Log2_64(Mask) == ShAmt) {
// XAndesPerf supports branch on test bit.
if (Subtarget.hasVendorXAndesPerf()) {
LHS =
DAG.getNode(ISD::AND, DL, LHS.getValueType(), LHS0.getOperand(0),
DAG.getConstant(Mask, DL, LHS.getValueType()));
return true;
}
CCVal = CCVal == ISD::SETEQ ? ISD::SETGE : ISD::SETLT;
CC = DAG.getCondCode(CCVal);
ShAmt = LHS.getValueSizeInBits() - 1 - ShAmt;
LHS = LHS0.getOperand(0);
if (ShAmt != 0)
LHS =
DAG.getNode(ISD::SHL, DL, LHS.getValueType(), LHS0.getOperand(0),
DAG.getConstant(ShAmt, DL, LHS.getValueType()));
return true;
}
}
}
// (X, 1, setne) -> // (X, 0, seteq) if we can prove X is 0/1.
// This can occur when legalizing some floating point comparisons.
APInt Mask = APInt::getBitsSetFrom(LHS.getValueSizeInBits(), 1);
if (isOneConstant(RHS) && DAG.MaskedValueIsZero(LHS, Mask)) {
CCVal = ISD::getSetCCInverse(CCVal, LHS.getValueType());
CC = DAG.getCondCode(CCVal);
RHS = DAG.getConstant(0, DL, LHS.getValueType());
return true;
}
if (isNullConstant(RHS)) {
if (SDValue NewCond = tryDemorganOfBooleanCondition(LHS, DAG)) {
CCVal = ISD::getSetCCInverse(CCVal, LHS.getValueType());
CC = DAG.getCondCode(CCVal);
LHS = NewCond;
return true;
}
}
return false;
}
// Fold
// (select C, (add Y, X), Y) -> (add Y, (select C, X, 0)).
// (select C, (sub Y, X), Y) -> (sub Y, (select C, X, 0)).
// (select C, (or Y, X), Y) -> (or Y, (select C, X, 0)).
// (select C, (xor Y, X), Y) -> (xor Y, (select C, X, 0)).
// (select C, (rotl Y, X), Y) -> (rotl Y, (select C, X, 0)).
// (select C, (rotr Y, X), Y) -> (rotr Y, (select C, X, 0)).
static SDValue tryFoldSelectIntoOp(SDNode *N, SelectionDAG &DAG,
SDValue TrueVal, SDValue FalseVal,
bool Swapped) {
bool Commutative = true;
unsigned Opc = TrueVal.getOpcode();
switch (Opc) {
default:
return SDValue();
case ISD::SHL:
case ISD::SRA:
case ISD::SRL:
case ISD::SUB:
case ISD::ROTL:
case ISD::ROTR:
Commutative = false;
break;
case ISD::ADD:
case ISD::OR:
case ISD::XOR:
break;
}
if (!TrueVal.hasOneUse())
return SDValue();
unsigned OpToFold;
if (FalseVal == TrueVal.getOperand(0))
OpToFold = 0;
else if (Commutative && FalseVal == TrueVal.getOperand(1))
OpToFold = 1;
else
return SDValue();
EVT VT = N->getValueType(0);
SDLoc DL(N);
SDValue OtherOp = TrueVal.getOperand(1 - OpToFold);
EVT OtherOpVT = OtherOp.getValueType();
SDValue IdentityOperand =
DAG.getNeutralElement(Opc, DL, OtherOpVT, N->getFlags());
if (!Commutative)
IdentityOperand = DAG.getConstant(0, DL, OtherOpVT);
assert(IdentityOperand && "No identity operand!");
if (Swapped)
std::swap(OtherOp, IdentityOperand);
SDValue NewSel =
DAG.getSelect(DL, OtherOpVT, N->getOperand(0), OtherOp, IdentityOperand);
return DAG.getNode(TrueVal.getOpcode(), DL, VT, FalseVal, NewSel);
}
// This tries to get rid of `select` and `icmp` that are being used to handle
// `Targets` that do not support `cttz(0)`/`ctlz(0)`.
static SDValue foldSelectOfCTTZOrCTLZ(SDNode *N, SelectionDAG &DAG) {
SDValue Cond = N->getOperand(0);
// This represents either CTTZ or CTLZ instruction.
SDValue CountZeroes;
SDValue ValOnZero;
if (Cond.getOpcode() != ISD::SETCC)
return SDValue();
if (!isNullConstant(Cond->getOperand(1)))
return SDValue();
ISD::CondCode CCVal = cast<CondCodeSDNode>(Cond->getOperand(2))->get();
if (CCVal == ISD::CondCode::SETEQ) {
CountZeroes = N->getOperand(2);
ValOnZero = N->getOperand(1);
} else if (CCVal == ISD::CondCode::SETNE) {
CountZeroes = N->getOperand(1);
ValOnZero = N->getOperand(2);
} else {
return SDValue();
}
if (CountZeroes.getOpcode() == ISD::TRUNCATE ||
CountZeroes.getOpcode() == ISD::ZERO_EXTEND)
CountZeroes = CountZeroes.getOperand(0);
if (CountZeroes.getOpcode() != ISD::CTTZ &&
CountZeroes.getOpcode() != ISD::CTTZ_ZERO_UNDEF &&
CountZeroes.getOpcode() != ISD::CTLZ &&
CountZeroes.getOpcode() != ISD::CTLZ_ZERO_UNDEF)
return SDValue();
if (!isNullConstant(ValOnZero))
return SDValue();
SDValue CountZeroesArgument = CountZeroes->getOperand(0);
if (Cond->getOperand(0) != CountZeroesArgument)
return SDValue();
unsigned BitWidth = CountZeroes.getValueSizeInBits();
if (!isPowerOf2_32(BitWidth))
return SDValue();
if (CountZeroes.getOpcode() == ISD::CTTZ_ZERO_UNDEF) {
CountZeroes = DAG.getNode(ISD::CTTZ, SDLoc(CountZeroes),
CountZeroes.getValueType(), CountZeroesArgument);
} else if (CountZeroes.getOpcode() == ISD::CTLZ_ZERO_UNDEF) {
CountZeroes = DAG.getNode(ISD::CTLZ, SDLoc(CountZeroes),
CountZeroes.getValueType(), CountZeroesArgument);
}
SDValue BitWidthMinusOne =
DAG.getConstant(BitWidth - 1, SDLoc(N), CountZeroes.getValueType());
auto AndNode = DAG.getNode(ISD::AND, SDLoc(N), CountZeroes.getValueType(),
CountZeroes, BitWidthMinusOne);
return DAG.getZExtOrTrunc(AndNode, SDLoc(N), N->getValueType(0));
}
static SDValue useInversedSetcc(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
SDValue Cond = N->getOperand(0);
SDValue True = N->getOperand(1);
SDValue False = N->getOperand(2);
SDLoc DL(N);
EVT VT = N->getValueType(0);
EVT CondVT = Cond.getValueType();
if (Cond.getOpcode() != ISD::SETCC || !Cond.hasOneUse())
return SDValue();
// Replace (setcc eq (and x, C)) with (setcc ne (and x, C))) to generate
// BEXTI, where C is power of 2.
if (Subtarget.hasStdExtZbs() && VT.isScalarInteger() &&
(Subtarget.hasCZEROLike() || Subtarget.hasVendorXTHeadCondMov())) {
SDValue LHS = Cond.getOperand(0);
SDValue RHS = Cond.getOperand(1);
ISD::CondCode CC = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
if (CC == ISD::SETEQ && LHS.getOpcode() == ISD::AND &&
isa<ConstantSDNode>(LHS.getOperand(1)) && isNullConstant(RHS)) {
const APInt &MaskVal = LHS.getConstantOperandAPInt(1);
if (MaskVal.isPowerOf2() && !MaskVal.isSignedIntN(12))
return DAG.getSelect(DL, VT,
DAG.getSetCC(DL, CondVT, LHS, RHS, ISD::SETNE),
False, True);
}
}
return SDValue();
}
static bool matchSelectAddSub(SDValue TrueVal, SDValue FalseVal, bool &SwapCC) {
if (!TrueVal.hasOneUse() || !FalseVal.hasOneUse())
return false;
SwapCC = false;
if (TrueVal.getOpcode() == ISD::SUB && FalseVal.getOpcode() == ISD::ADD) {
std::swap(TrueVal, FalseVal);
SwapCC = true;
}
if (TrueVal.getOpcode() != ISD::ADD || FalseVal.getOpcode() != ISD::SUB)
return false;
SDValue A = FalseVal.getOperand(0);
SDValue B = FalseVal.getOperand(1);
// Add is commutative, so check both orders
return ((TrueVal.getOperand(0) == A && TrueVal.getOperand(1) == B) ||
(TrueVal.getOperand(1) == A && TrueVal.getOperand(0) == B));
}
/// Convert vselect CC, (add a, b), (sub a, b) to add a, (vselect CC, -b, b).
/// This allows us match a vadd.vv fed by a masked vrsub, which reduces
/// register pressure over the add followed by masked vsub sequence.
static SDValue performVSELECTCombine(SDNode *N, SelectionDAG &DAG) {
SDLoc DL(N);
EVT VT = N->getValueType(0);
SDValue CC = N->getOperand(0);
SDValue TrueVal = N->getOperand(1);
SDValue FalseVal = N->getOperand(2);
bool SwapCC;
if (!matchSelectAddSub(TrueVal, FalseVal, SwapCC))
return SDValue();
SDValue Sub = SwapCC ? TrueVal : FalseVal;
SDValue A = Sub.getOperand(0);
SDValue B = Sub.getOperand(1);
// Arrange the select such that we can match a masked
// vrsub.vi to perform the conditional negate
SDValue NegB = DAG.getNegative(B, DL, VT);
if (!SwapCC)
CC = DAG.getLogicalNOT(DL, CC, CC->getValueType(0));
SDValue NewB = DAG.getNode(ISD::VSELECT, DL, VT, CC, NegB, B);
return DAG.getNode(ISD::ADD, DL, VT, A, NewB);
}
static SDValue performSELECTCombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (SDValue Folded = foldSelectOfCTTZOrCTLZ(N, DAG))
return Folded;
if (SDValue V = useInversedSetcc(N, DAG, Subtarget))
return V;
if (Subtarget.hasConditionalMoveFusion())
return SDValue();
SDValue TrueVal = N->getOperand(1);
SDValue FalseVal = N->getOperand(2);
if (SDValue V = tryFoldSelectIntoOp(N, DAG, TrueVal, FalseVal, /*Swapped*/false))
return V;
return tryFoldSelectIntoOp(N, DAG, FalseVal, TrueVal, /*Swapped*/true);
}
/// If we have a build_vector where each lane is binop X, C, where C
/// is a constant (but not necessarily the same constant on all lanes),
/// form binop (build_vector x1, x2, ...), (build_vector c1, c2, c3, ..).
/// We assume that materializing a constant build vector will be no more
/// expensive that performing O(n) binops.
static SDValue performBUILD_VECTORCombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget,
const RISCVTargetLowering &TLI) {
SDLoc DL(N);
EVT VT = N->getValueType(0);
assert(!VT.isScalableVector() && "unexpected build vector");
if (VT.getVectorNumElements() == 1)
return SDValue();
const unsigned Opcode = N->op_begin()->getNode()->getOpcode();
if (!TLI.isBinOp(Opcode))
return SDValue();
if (!TLI.isOperationLegalOrCustom(Opcode, VT) || !TLI.isTypeLegal(VT))
return SDValue();
// This BUILD_VECTOR involves an implicit truncation, and sinking
// truncates through binops is non-trivial.
if (N->op_begin()->getValueType() != VT.getVectorElementType())
return SDValue();
SmallVector<SDValue> LHSOps;
SmallVector<SDValue> RHSOps;
for (SDValue Op : N->ops()) {
if (Op.isUndef()) {
// We can't form a divide or remainder from undef.
if (!DAG.isSafeToSpeculativelyExecute(Opcode))
return SDValue();
LHSOps.push_back(Op);
RHSOps.push_back(Op);
continue;
}
// TODO: We can handle operations which have an neutral rhs value
// (e.g. x + 0, a * 1 or a << 0), but we then have to keep track
// of profit in a more explicit manner.
if (Op.getOpcode() != Opcode || !Op.hasOneUse())
return SDValue();
LHSOps.push_back(Op.getOperand(0));
if (!isa<ConstantSDNode>(Op.getOperand(1)) &&
!isa<ConstantFPSDNode>(Op.getOperand(1)))
return SDValue();
// FIXME: Return failure if the RHS type doesn't match the LHS. Shifts may
// have different LHS and RHS types.
if (Op.getOperand(0).getValueType() != Op.getOperand(1).getValueType())
return SDValue();
RHSOps.push_back(Op.getOperand(1));
}
return DAG.getNode(Opcode, DL, VT, DAG.getBuildVector(VT, DL, LHSOps),
DAG.getBuildVector(VT, DL, RHSOps));
}
static MVT getQDOTXResultType(MVT OpVT) {
ElementCount OpEC = OpVT.getVectorElementCount();
assert(OpEC.isKnownMultipleOf(4) && OpVT.getVectorElementType() == MVT::i8);
return MVT::getVectorVT(MVT::i32, OpEC.divideCoefficientBy(4));
}
/// Given fixed length vectors A and B with equal element types, but possibly
/// different number of elements, return A + B where either A or B is zero
/// padded to the larger number of elements.
static SDValue getZeroPaddedAdd(const SDLoc &DL, SDValue A, SDValue B,
SelectionDAG &DAG) {
// NOTE: Manually doing the extract/add/insert scheme produces
// significantly better codegen than the naive pad with zeros
// and add scheme.
EVT AVT = A.getValueType();
EVT BVT = B.getValueType();
assert(AVT.getVectorElementType() == BVT.getVectorElementType());
if (AVT.getVectorMinNumElements() > BVT.getVectorMinNumElements()) {
std::swap(A, B);
std::swap(AVT, BVT);
}
SDValue BPart = DAG.getExtractSubvector(DL, AVT, B, 0);
SDValue Res = DAG.getNode(ISD::ADD, DL, AVT, A, BPart);
return DAG.getInsertSubvector(DL, B, Res, 0);
}
static SDValue foldReduceOperandViaVQDOT(SDValue InVec, const SDLoc &DL,
SelectionDAG &DAG,
const RISCVSubtarget &Subtarget,
const RISCVTargetLowering &TLI) {
// Note: We intentionally do not check the legality of the reduction type.
// We want to handle the m4/m8 *src* types, and thus need to let illegal
// intermediate types flow through here.
if (InVec.getValueType().getVectorElementType() != MVT::i32 ||
!InVec.getValueType().getVectorElementCount().isKnownMultipleOf(4))
return SDValue();
// Recurse through adds (since generic dag canonicalizes to that
// form). TODO: Handle disjoint or here.
if (InVec->getOpcode() == ISD::ADD) {
SDValue A = InVec.getOperand(0);
SDValue B = InVec.getOperand(1);
SDValue AOpt = foldReduceOperandViaVQDOT(A, DL, DAG, Subtarget, TLI);
SDValue BOpt = foldReduceOperandViaVQDOT(B, DL, DAG, Subtarget, TLI);
if (AOpt || BOpt) {
if (AOpt)
A = AOpt;
if (BOpt)
B = BOpt;
// From here, we're doing A + B with mixed types, implicitly zero
// padded to the wider type. Note that we *don't* need the result
// type to be the original VT, and in fact prefer narrower ones
// if possible.
return getZeroPaddedAdd(DL, A, B, DAG);
}
}
// zext a <--> partial_reduce_umla 0, a, 1
// sext a <--> partial_reduce_smla 0, a, 1
if (InVec.getOpcode() == ISD::ZERO_EXTEND ||
InVec.getOpcode() == ISD::SIGN_EXTEND) {
SDValue A = InVec.getOperand(0);
EVT OpVT = A.getValueType();
if (OpVT.getVectorElementType() != MVT::i8 || !TLI.isTypeLegal(OpVT))
return SDValue();
MVT ResVT = getQDOTXResultType(A.getSimpleValueType());
SDValue B = DAG.getConstant(0x1, DL, OpVT);
bool IsSigned = InVec.getOpcode() == ISD::SIGN_EXTEND;
unsigned Opc =
IsSigned ? ISD::PARTIAL_REDUCE_SMLA : ISD::PARTIAL_REDUCE_UMLA;
return DAG.getNode(Opc, DL, ResVT, {DAG.getConstant(0, DL, ResVT), A, B});
}
// mul (sext a, sext b) -> partial_reduce_smla 0, a, b
// mul (zext a, zext b) -> partial_reduce_umla 0, a, b
// mul (sext a, zext b) -> partial_reduce_ssmla 0, a, b
// mul (zext a, sext b) -> partial_reduce_smla 0, b, a (swapped)
if (InVec.getOpcode() != ISD::MUL)
return SDValue();
SDValue A = InVec.getOperand(0);
SDValue B = InVec.getOperand(1);
if (!ISD::isExtOpcode(A.getOpcode()))
return SDValue();
EVT OpVT = A.getOperand(0).getValueType();
if (OpVT.getVectorElementType() != MVT::i8 ||
OpVT != B.getOperand(0).getValueType() ||
!TLI.isTypeLegal(A.getValueType()))
return SDValue();
unsigned Opc;
if (A.getOpcode() == ISD::SIGN_EXTEND && B.getOpcode() == ISD::SIGN_EXTEND)
Opc = ISD::PARTIAL_REDUCE_SMLA;
else if (A.getOpcode() == ISD::ZERO_EXTEND &&
B.getOpcode() == ISD::ZERO_EXTEND)
Opc = ISD::PARTIAL_REDUCE_UMLA;
else if (A.getOpcode() == ISD::SIGN_EXTEND &&
B.getOpcode() == ISD::ZERO_EXTEND)
Opc = ISD::PARTIAL_REDUCE_SUMLA;
else if (A.getOpcode() == ISD::ZERO_EXTEND &&
B.getOpcode() == ISD::SIGN_EXTEND) {
Opc = ISD::PARTIAL_REDUCE_SUMLA;
std::swap(A, B);
} else
return SDValue();
MVT ResVT = getQDOTXResultType(OpVT.getSimpleVT());
return DAG.getNode(
Opc, DL, ResVT,
{DAG.getConstant(0, DL, ResVT), A.getOperand(0), B.getOperand(0)});
}
static SDValue performVECREDUCECombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget,
const RISCVTargetLowering &TLI) {
if (!Subtarget.hasStdExtZvqdotq())
return SDValue();
SDLoc DL(N);
EVT VT = N->getValueType(0);
SDValue InVec = N->getOperand(0);
if (SDValue V = foldReduceOperandViaVQDOT(InVec, DL, DAG, Subtarget, TLI))
return DAG.getNode(ISD::VECREDUCE_ADD, DL, VT, V);
return SDValue();
}
static SDValue performINSERT_VECTOR_ELTCombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget,
const RISCVTargetLowering &TLI) {
SDValue InVec = N->getOperand(0);
SDValue InVal = N->getOperand(1);
SDValue EltNo = N->getOperand(2);
SDLoc DL(N);
EVT VT = InVec.getValueType();
if (VT.isScalableVector())
return SDValue();
if (!InVec.hasOneUse())
return SDValue();
// Given insert_vector_elt (binop a, VecC), (same_binop b, C2), Elt
// move the insert_vector_elts into the arms of the binop. Note that
// the new RHS must be a constant.
const unsigned InVecOpcode = InVec->getOpcode();
if (InVecOpcode == InVal->getOpcode() && TLI.isBinOp(InVecOpcode) &&
InVal.hasOneUse()) {
SDValue InVecLHS = InVec->getOperand(0);
SDValue InVecRHS = InVec->getOperand(1);
SDValue InValLHS = InVal->getOperand(0);
SDValue InValRHS = InVal->getOperand(1);
if (!ISD::isBuildVectorOfConstantSDNodes(InVecRHS.getNode()))
return SDValue();
if (!isa<ConstantSDNode>(InValRHS) && !isa<ConstantFPSDNode>(InValRHS))
return SDValue();
// FIXME: Return failure if the RHS type doesn't match the LHS. Shifts may
// have different LHS and RHS types.
if (InVec.getOperand(0).getValueType() != InVec.getOperand(1).getValueType())
return SDValue();
SDValue LHS = DAG.getNode(ISD::INSERT_VECTOR_ELT, DL, VT,
InVecLHS, InValLHS, EltNo);
SDValue RHS = DAG.getNode(ISD::INSERT_VECTOR_ELT, DL, VT,
InVecRHS, InValRHS, EltNo);
return DAG.getNode(InVecOpcode, DL, VT, LHS, RHS);
}
// Given insert_vector_elt (concat_vectors ...), InVal, Elt
// move the insert_vector_elt to the source operand of the concat_vector.
if (InVec.getOpcode() != ISD::CONCAT_VECTORS)
return SDValue();
auto *IndexC = dyn_cast<ConstantSDNode>(EltNo);
if (!IndexC)
return SDValue();
unsigned Elt = IndexC->getZExtValue();
EVT ConcatVT = InVec.getOperand(0).getValueType();
if (ConcatVT.getVectorElementType() != InVal.getValueType())
return SDValue();
unsigned ConcatNumElts = ConcatVT.getVectorNumElements();
unsigned NewIdx = Elt % ConcatNumElts;
unsigned ConcatOpIdx = Elt / ConcatNumElts;
SDValue ConcatOp = InVec.getOperand(ConcatOpIdx);
ConcatOp = DAG.getInsertVectorElt(DL, ConcatOp, InVal, NewIdx);
SmallVector<SDValue> ConcatOps(InVec->ops());
ConcatOps[ConcatOpIdx] = ConcatOp;
return DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, ConcatOps);
}
// If we're concatenating a series of vector loads like
// concat_vectors (load v4i8, p+0), (load v4i8, p+n), (load v4i8, p+n*2) ...
// Then we can turn this into a strided load by widening the vector elements
// vlse32 p, stride=n
static SDValue performCONCAT_VECTORSCombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget,
const RISCVTargetLowering &TLI) {
SDLoc DL(N);
EVT VT = N->getValueType(0);
// Only perform this combine on legal MVTs.
if (!TLI.isTypeLegal(VT))
return SDValue();
// TODO: Potentially extend this to scalable vectors
if (VT.isScalableVector())
return SDValue();
auto *BaseLd = dyn_cast<LoadSDNode>(N->getOperand(0));
if (!BaseLd || !BaseLd->isSimple() || !ISD::isNormalLoad(BaseLd) ||
!SDValue(BaseLd, 0).hasOneUse())
return SDValue();
EVT BaseLdVT = BaseLd->getValueType(0);
// Go through the loads and check that they're strided
SmallVector<LoadSDNode *> Lds;
Lds.push_back(BaseLd);
Align Align = BaseLd->getAlign();
for (SDValue Op : N->ops().drop_front()) {
auto *Ld = dyn_cast<LoadSDNode>(Op);
if (!Ld || !Ld->isSimple() || !Op.hasOneUse() ||
Ld->getChain() != BaseLd->getChain() || !ISD::isNormalLoad(Ld) ||
Ld->getValueType(0) != BaseLdVT)
return SDValue();
Lds.push_back(Ld);
// The common alignment is the most restrictive (smallest) of all the loads
Align = std::min(Align, Ld->getAlign());
}
using PtrDiff = std::pair<std::variant<int64_t, SDValue>, bool>;
auto GetPtrDiff = [&DAG](LoadSDNode *Ld1,
LoadSDNode *Ld2) -> std::optional<PtrDiff> {
// If the load ptrs can be decomposed into a common (Base + Index) with a
// common constant stride, then return the constant stride.
BaseIndexOffset BIO1 = BaseIndexOffset::match(Ld1, DAG);
BaseIndexOffset BIO2 = BaseIndexOffset::match(Ld2, DAG);
if (BIO1.equalBaseIndex(BIO2, DAG))
return {{BIO2.getOffset() - BIO1.getOffset(), false}};
// Otherwise try to match (add LastPtr, Stride) or (add NextPtr, Stride)
SDValue P1 = Ld1->getBasePtr();
SDValue P2 = Ld2->getBasePtr();
if (P2.getOpcode() == ISD::ADD && P2.getOperand(0) == P1)
return {{P2.getOperand(1), false}};
if (P1.getOpcode() == ISD::ADD && P1.getOperand(0) == P2)
return {{P1.getOperand(1), true}};
return std::nullopt;
};
// Get the distance between the first and second loads
auto BaseDiff = GetPtrDiff(Lds[0], Lds[1]);
if (!BaseDiff)
return SDValue();
// Check all the loads are the same distance apart
for (auto *It = Lds.begin() + 1; It != Lds.end() - 1; It++)
if (GetPtrDiff(*It, *std::next(It)) != BaseDiff)
return SDValue();
// TODO: At this point, we've successfully matched a generalized gather
// load. Maybe we should emit that, and then move the specialized
// matchers above and below into a DAG combine?
// Get the widened scalar type, e.g. v4i8 -> i64
unsigned WideScalarBitWidth =
BaseLdVT.getScalarSizeInBits() * BaseLdVT.getVectorNumElements();
MVT WideScalarVT = MVT::getIntegerVT(WideScalarBitWidth);
// Get the vector type for the strided load, e.g. 4 x v4i8 -> v4i64
MVT WideVecVT = MVT::getVectorVT(WideScalarVT, N->getNumOperands());
if (!TLI.isTypeLegal(WideVecVT))
return SDValue();
// Check that the operation is legal
if (!TLI.isLegalStridedLoadStore(WideVecVT, Align))
return SDValue();
auto [StrideVariant, MustNegateStride] = *BaseDiff;
SDValue Stride =
std::holds_alternative<SDValue>(StrideVariant)
? std::get<SDValue>(StrideVariant)
: DAG.getSignedConstant(std::get<int64_t>(StrideVariant), DL,
Lds[0]->getOffset().getValueType());
if (MustNegateStride)
Stride = DAG.getNegative(Stride, DL, Stride.getValueType());
SDValue AllOneMask =
DAG.getSplat(WideVecVT.changeVectorElementType(MVT::i1), DL,
DAG.getConstant(1, DL, MVT::i1));
uint64_t MemSize;
if (auto *ConstStride = dyn_cast<ConstantSDNode>(Stride);
ConstStride && ConstStride->getSExtValue() >= 0)
// total size = (elsize * n) + (stride - elsize) * (n-1)
// = elsize + stride * (n-1)
MemSize = WideScalarVT.getSizeInBits() +
ConstStride->getSExtValue() * (N->getNumOperands() - 1);
else
// If Stride isn't constant, then we can't know how much it will load
MemSize = MemoryLocation::UnknownSize;
MachineMemOperand *MMO = DAG.getMachineFunction().getMachineMemOperand(
BaseLd->getPointerInfo(), BaseLd->getMemOperand()->getFlags(), MemSize,
Align);
SDValue StridedLoad = DAG.getStridedLoadVP(
WideVecVT, DL, BaseLd->getChain(), BaseLd->getBasePtr(), Stride,
AllOneMask,
DAG.getConstant(N->getNumOperands(), DL, Subtarget.getXLenVT()), MMO);
for (SDValue Ld : N->ops())
DAG.makeEquivalentMemoryOrdering(cast<LoadSDNode>(Ld), StridedLoad);
return DAG.getBitcast(VT.getSimpleVT(), StridedLoad);
}
static SDValue performVECTOR_SHUFFLECombine(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget,
const RISCVTargetLowering &TLI) {
SDLoc DL(N);
EVT VT = N->getValueType(0);
const unsigned ElementSize = VT.getScalarSizeInBits();
const unsigned NumElts = VT.getVectorNumElements();
SDValue V1 = N->getOperand(0);
SDValue V2 = N->getOperand(1);
ArrayRef<int> Mask = cast<ShuffleVectorSDNode>(N)->getMask();
MVT XLenVT = Subtarget.getXLenVT();
// Recognized a disguised select of add/sub.
bool SwapCC;
if (ShuffleVectorInst::isSelectMask(Mask, NumElts) &&
matchSelectAddSub(V1, V2, SwapCC)) {
SDValue Sub = SwapCC ? V1 : V2;
SDValue A = Sub.getOperand(0);
SDValue B = Sub.getOperand(1);
SmallVector<SDValue> MaskVals;
for (int MaskIndex : Mask) {
bool SelectMaskVal = (MaskIndex < (int)NumElts);
MaskVals.push_back(DAG.getConstant(SelectMaskVal, DL, XLenVT));
}
assert(MaskVals.size() == NumElts && "Unexpected select-like shuffle");
EVT MaskVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1, NumElts);
SDValue CC = DAG.getBuildVector(MaskVT, DL, MaskVals);
// Arrange the select such that we can match a masked
// vrsub.vi to perform the conditional negate
SDValue NegB = DAG.getNegative(B, DL, VT);
if (!SwapCC)
CC = DAG.getLogicalNOT(DL, CC, CC->getValueType(0));
SDValue NewB = DAG.getNode(ISD::VSELECT, DL, VT, CC, NegB, B);
return DAG.getNode(ISD::ADD, DL, VT, A, NewB);
}
// Custom legalize <N x i128> or <N x i256> to <M x ELEN>. This runs
// during the combine phase before type legalization, and relies on
// DAGCombine not undoing the transform if isShuffleMaskLegal returns false
// for the source mask.
if (TLI.isTypeLegal(VT) || ElementSize <= Subtarget.getELen() ||
!isPowerOf2_64(ElementSize) || VT.getVectorNumElements() % 2 != 0 ||
VT.isFloatingPoint() || TLI.isShuffleMaskLegal(Mask, VT))
return SDValue();
SmallVector<int, 8> NewMask;
narrowShuffleMaskElts(2, Mask, NewMask);
LLVMContext &C = *DAG.getContext();
EVT NewEltVT = EVT::getIntegerVT(C, ElementSize / 2);
EVT NewVT = EVT::getVectorVT(C, NewEltVT, VT.getVectorNumElements() * 2);
SDValue Res = DAG.getVectorShuffle(NewVT, DL, DAG.getBitcast(NewVT, V1),
DAG.getBitcast(NewVT, V2), NewMask);
return DAG.getBitcast(VT, Res);
}
static SDValue combineToVWMACC(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(N->getOpcode() == RISCVISD::ADD_VL || N->getOpcode() == ISD::ADD);
if (N->getValueType(0).isFixedLengthVector())
return SDValue();
SDValue Addend = N->getOperand(0);
SDValue MulOp = N->getOperand(1);
if (N->getOpcode() == RISCVISD::ADD_VL) {
SDValue AddPassthruOp = N->getOperand(2);
if (!AddPassthruOp.isUndef())
return SDValue();
}
auto IsVWMulOpc = [](unsigned Opc) {
switch (Opc) {
case RISCVISD::VWMUL_VL:
case RISCVISD::VWMULU_VL:
case RISCVISD::VWMULSU_VL:
return true;
default:
return false;
}
};
if (!IsVWMulOpc(MulOp.getOpcode()))
std::swap(Addend, MulOp);
if (!IsVWMulOpc(MulOp.getOpcode()))
return SDValue();
SDValue MulPassthruOp = MulOp.getOperand(2);
if (!MulPassthruOp.isUndef())
return SDValue();
auto [AddMask, AddVL] = [](SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (N->getOpcode() == ISD::ADD) {
SDLoc DL(N);
return getDefaultScalableVLOps(N->getSimpleValueType(0), DL, DAG,
Subtarget);
}
return std::make_pair(N->getOperand(3), N->getOperand(4));
}(N, DAG, Subtarget);
SDValue MulMask = MulOp.getOperand(3);
SDValue MulVL = MulOp.getOperand(4);
if (AddMask != MulMask || AddVL != MulVL)
return SDValue();
const auto &TSInfo =
static_cast<const RISCVSelectionDAGInfo &>(DAG.getSelectionDAGInfo());
unsigned Opc = TSInfo.getMAccOpcode(MulOp.getOpcode());
SDLoc DL(N);
EVT VT = N->getValueType(0);
SDValue Ops[] = {MulOp.getOperand(0), MulOp.getOperand(1), Addend, AddMask,
AddVL};
return DAG.getNode(Opc, DL, VT, Ops);
}
static SDValue combineVqdotAccum(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(N->getOpcode() == RISCVISD::ADD_VL || N->getOpcode() == ISD::ADD);
if (!N->getValueType(0).isVector())
return SDValue();
SDValue Addend = N->getOperand(0);
SDValue DotOp = N->getOperand(1);
if (N->getOpcode() == RISCVISD::ADD_VL) {
SDValue AddPassthruOp = N->getOperand(2);
if (!AddPassthruOp.isUndef())
return SDValue();
}
auto IsVqdotqOpc = [](unsigned Opc) {
switch (Opc) {
case RISCVISD::VQDOT_VL:
case RISCVISD::VQDOTU_VL:
case RISCVISD::VQDOTSU_VL:
return true;
default:
return false;
}
};
if (!IsVqdotqOpc(DotOp.getOpcode()))
std::swap(Addend, DotOp);
if (!IsVqdotqOpc(DotOp.getOpcode()))
return SDValue();
auto [AddMask, AddVL] = [](SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
if (N->getOpcode() == ISD::ADD) {
SDLoc DL(N);
return getDefaultScalableVLOps(N->getSimpleValueType(0), DL, DAG,
Subtarget);
}
return std::make_pair(N->getOperand(3), N->getOperand(4));
}(N, DAG, Subtarget);
SDValue MulVL = DotOp.getOperand(4);
if (AddVL != MulVL)
return SDValue();
if (AddMask.getOpcode() != RISCVISD::VMSET_VL ||
AddMask.getOperand(0) != MulVL)
return SDValue();
SDValue AccumOp = DotOp.getOperand(2);
SDLoc DL(N);
EVT VT = N->getValueType(0);
Addend = DAG.getNode(RISCVISD::ADD_VL, DL, VT, Addend, AccumOp,
DAG.getUNDEF(VT), AddMask, AddVL);
SDValue Ops[] = {DotOp.getOperand(0), DotOp.getOperand(1), Addend,
DotOp.getOperand(3), DotOp->getOperand(4)};
return DAG.getNode(DotOp->getOpcode(), DL, VT, Ops);
}
static bool
legalizeScatterGatherIndexType(SDLoc DL, SDValue &Index,
ISD::MemIndexType &IndexType,
RISCVTargetLowering::DAGCombinerInfo &DCI) {
if (!DCI.isBeforeLegalize())
return false;
SelectionDAG &DAG = DCI.DAG;
const MVT XLenVT =
DAG.getMachineFunction().getSubtarget<RISCVSubtarget>().getXLenVT();
const EVT IndexVT = Index.getValueType();
// RISC-V indexed loads only support the "unsigned unscaled" addressing
// mode, so anything else must be manually legalized.
if (!isIndexTypeSigned(IndexType))
return false;
if (IndexVT.getVectorElementType().bitsLT(XLenVT)) {
// Any index legalization should first promote to XLenVT, so we don't lose
// bits when scaling. This may create an illegal index type so we let
// LLVM's legalization take care of the splitting.
// FIXME: LLVM can't split VP_GATHER or VP_SCATTER yet.
Index = DAG.getNode(ISD::SIGN_EXTEND, DL,
IndexVT.changeVectorElementType(XLenVT), Index);
}
IndexType = ISD::UNSIGNED_SCALED;
return true;
}
/// Match the index vector of a scatter or gather node as the shuffle mask
/// which performs the rearrangement if possible. Will only match if
/// all lanes are touched, and thus replacing the scatter or gather with
/// a unit strided access and shuffle is legal.
static bool matchIndexAsShuffle(EVT VT, SDValue Index, SDValue Mask,
SmallVector<int> &ShuffleMask) {
if (!ISD::isConstantSplatVectorAllOnes(Mask.getNode()))
return false;
if (!ISD::isBuildVectorOfConstantSDNodes(Index.getNode()))
return false;
const unsigned ElementSize = VT.getScalarStoreSize();
const unsigned NumElems = VT.getVectorNumElements();
// Create the shuffle mask and check all bits active
assert(ShuffleMask.empty());
BitVector ActiveLanes(NumElems);
for (unsigned i = 0; i < Index->getNumOperands(); i++) {
// TODO: We've found an active bit of UB, and could be
// more aggressive here if desired.
if (Index->getOperand(i)->isUndef())
return false;
uint64_t C = Index->getConstantOperandVal(i);
if (C % ElementSize != 0)
return false;
C = C / ElementSize;
if (C >= NumElems)
return false;
ShuffleMask.push_back(C);
ActiveLanes.set(C);
}
return ActiveLanes.all();
}
/// Match the index of a gather or scatter operation as an operation
/// with twice the element width and half the number of elements. This is
/// generally profitable (if legal) because these operations are linear
/// in VL, so even if we cause some extract VTYPE/VL toggles, we still
/// come out ahead.
static bool matchIndexAsWiderOp(EVT VT, SDValue Index, SDValue Mask,
Align BaseAlign, const RISCVSubtarget &ST) {
if (!ISD::isConstantSplatVectorAllOnes(Mask.getNode()))
return false;
if (!ISD::isBuildVectorOfConstantSDNodes(Index.getNode()))
return false;
// Attempt a doubling. If we can use a element type 4x or 8x in
// size, this will happen via multiply iterations of the transform.
const unsigned NumElems = VT.getVectorNumElements();
if (NumElems % 2 != 0)
return false;
const unsigned ElementSize = VT.getScalarStoreSize();
const unsigned WiderElementSize = ElementSize * 2;
if (WiderElementSize > ST.getELen()/8)
return false;
if (!ST.enableUnalignedVectorMem() && BaseAlign < WiderElementSize)
return false;
for (unsigned i = 0; i < Index->getNumOperands(); i++) {
// TODO: We've found an active bit of UB, and could be
// more aggressive here if desired.
if (Index->getOperand(i)->isUndef())
return false;
// TODO: This offset check is too strict if we support fully
// misaligned memory operations.
uint64_t C = Index->getConstantOperandVal(i);
if (i % 2 == 0) {
if (C % WiderElementSize != 0)
return false;
continue;
}
uint64_t Last = Index->getConstantOperandVal(i-1);
if (C != Last + ElementSize)
return false;
}
return true;
}
// trunc (sra sext (X), zext (Y)) -> sra (X, smin (Y, scalarsize(Y) - 1))
// This would be benefit for the cases where X and Y are both the same value
// type of low precision vectors. Since the truncate would be lowered into
// n-levels TRUNCATE_VECTOR_VL to satisfy RVV's SEW*2->SEW truncate
// restriction, such pattern would be expanded into a series of "vsetvli"
// and "vnsrl" instructions later to reach this point.
static SDValue combineTruncOfSraSext(SDNode *N, SelectionDAG &DAG) {
SDValue Mask = N->getOperand(1);
SDValue VL = N->getOperand(2);
bool IsVLMAX = isAllOnesConstant(VL) ||
(isa<RegisterSDNode>(VL) &&
cast<RegisterSDNode>(VL)->getReg() == RISCV::X0);
if (!IsVLMAX || Mask.getOpcode() != RISCVISD::VMSET_VL ||
Mask.getOperand(0) != VL)
return SDValue();
auto IsTruncNode = [&](SDValue V) {
return V.getOpcode() == RISCVISD::TRUNCATE_VECTOR_VL &&
V.getOperand(1) == Mask && V.getOperand(2) == VL;
};
SDValue Op = N->getOperand(0);
// We need to first find the inner level of TRUNCATE_VECTOR_VL node
// to distinguish such pattern.
while (IsTruncNode(Op)) {
if (!Op.hasOneUse())
return SDValue();
Op = Op.getOperand(0);
}
if (Op.getOpcode() != ISD::SRA || !Op.hasOneUse())
return SDValue();
SDValue N0 = Op.getOperand(0);
SDValue N1 = Op.getOperand(1);
if (N0.getOpcode() != ISD::SIGN_EXTEND || !N0.hasOneUse() ||
N1.getOpcode() != ISD::ZERO_EXTEND || !N1.hasOneUse())
return SDValue();
SDValue N00 = N0.getOperand(0);
SDValue N10 = N1.getOperand(0);
if (!N00.getValueType().isVector() ||
N00.getValueType() != N10.getValueType() ||
N->getValueType(0) != N10.getValueType())
return SDValue();
unsigned MaxShAmt = N10.getValueType().getScalarSizeInBits() - 1;
SDValue SMin =
DAG.getNode(ISD::SMIN, SDLoc(N1), N->getValueType(0), N10,
DAG.getConstant(MaxShAmt, SDLoc(N1), N->getValueType(0)));
return DAG.getNode(ISD::SRA, SDLoc(N), N->getValueType(0), N00, SMin);
}
// Combine (truncate_vector_vl (umin X, C)) -> (vnclipu_vl X) if C is the
// maximum value for the truncated type.
// Combine (truncate_vector_vl (smin (smax X, C2), C1)) -> (vnclip_vl X) if C1
// is the signed maximum value for the truncated type and C2 is the signed
// minimum value.
static SDValue combineTruncToVnclip(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
assert(N->getOpcode() == RISCVISD::TRUNCATE_VECTOR_VL);
MVT VT = N->getSimpleValueType(0);
SDValue Mask = N->getOperand(1);
SDValue VL = N->getOperand(2);
auto MatchMinMax = [&VL, &Mask](SDValue V, unsigned Opc, unsigned OpcVL,
APInt &SplatVal) {
if (V.getOpcode() != Opc &&
!(V.getOpcode() == OpcVL && V.getOperand(2).isUndef() &&
V.getOperand(3) == Mask && V.getOperand(4) == VL))
return SDValue();
SDValue Op = V.getOperand(1);
// Peek through conversion between fixed and scalable vectors.
if (Op.getOpcode() == ISD::INSERT_SUBVECTOR && Op.getOperand(0).isUndef() &&
isNullConstant(Op.getOperand(2)) &&
Op.getOperand(1).getValueType().isFixedLengthVector() &&
Op.getOperand(1).getOpcode() == ISD::EXTRACT_SUBVECTOR &&
Op.getOperand(1).getOperand(0).getValueType() == Op.getValueType() &&
isNullConstant(Op.getOperand(1).getOperand(1)))
Op = Op.getOperand(1).getOperand(0);
if (ISD::isConstantSplatVector(Op.getNode(), SplatVal))
return V.getOperand(0);
if (Op.getOpcode() == RISCVISD::VMV_V_X_VL && Op.getOperand(0).isUndef() &&
Op.getOperand(2) == VL) {
if (auto *Op1 = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
SplatVal =
Op1->getAPIntValue().sextOrTrunc(Op.getScalarValueSizeInBits());
return V.getOperand(0);
}
}
return SDValue();
};
SDLoc DL(N);
auto DetectUSatPattern = [&](SDValue V) {
APInt LoC, HiC;
// Simple case, V is a UMIN.
if (SDValue UMinOp = MatchMinMax(V, ISD::UMIN, RISCVISD::UMIN_VL, HiC))
if (HiC.isMask(VT.getScalarSizeInBits()))
return UMinOp;
// If we have an SMAX that removes negative numbers first, then we can match
// SMIN instead of UMIN.
if (SDValue SMinOp = MatchMinMax(V, ISD::SMIN, RISCVISD::SMIN_VL, HiC))
if (SDValue SMaxOp =
MatchMinMax(SMinOp, ISD::SMAX, RISCVISD::SMAX_VL, LoC))
if (LoC.isNonNegative() && HiC.isMask(VT.getScalarSizeInBits()))
return SMinOp;
// If we have an SMIN before an SMAX and the SMAX constant is less than or
// equal to the SMIN constant, we can use vnclipu if we insert a new SMAX
// first.
if (SDValue SMaxOp = MatchMinMax(V, ISD::SMAX, RISCVISD::SMAX_VL, LoC))
if (SDValue SMinOp =
MatchMinMax(SMaxOp, ISD::SMIN, RISCVISD::SMIN_VL, HiC))
if (LoC.isNonNegative() && HiC.isMask(VT.getScalarSizeInBits()) &&
HiC.uge(LoC))
return DAG.getNode(RISCVISD::SMAX_VL, DL, V.getValueType(), SMinOp,
V.getOperand(1), DAG.getUNDEF(V.getValueType()),
Mask, VL);
return SDValue();
};
auto DetectSSatPattern = [&](SDValue V) {
unsigned NumDstBits = VT.getScalarSizeInBits();
unsigned NumSrcBits = V.getScalarValueSizeInBits();
APInt SignedMax = APInt::getSignedMaxValue(NumDstBits).sext(NumSrcBits);
APInt SignedMin = APInt::getSignedMinValue(NumDstBits).sext(NumSrcBits);
APInt HiC, LoC;
if (SDValue SMinOp = MatchMinMax(V, ISD::SMIN, RISCVISD::SMIN_VL, HiC))
if (SDValue SMaxOp =
MatchMinMax(SMinOp, ISD::SMAX, RISCVISD::SMAX_VL, LoC))
if (HiC == SignedMax && LoC == SignedMin)
return SMaxOp;
if (SDValue SMaxOp = MatchMinMax(V, ISD::SMAX, RISCVISD::SMAX_VL, LoC))
if (SDValue SMinOp =
MatchMinMax(SMaxOp, ISD::SMIN, RISCVISD::SMIN_VL, HiC))
if (HiC == SignedMax && LoC == SignedMin)
return SMinOp;
return SDValue();
};
SDValue Src = N->getOperand(0);
// Look through multiple layers of truncates.
while (Src.getOpcode() == RISCVISD::TRUNCATE_VECTOR_VL &&
Src.getOperand(1) == Mask && Src.getOperand(2) == VL &&
Src.hasOneUse())
Src = Src.getOperand(0);
SDValue Val;
unsigned ClipOpc;
if ((Val = DetectUSatPattern(Src)))
ClipOpc = RISCVISD::TRUNCATE_VECTOR_VL_USAT;
else if ((Val = DetectSSatPattern(Src)))
ClipOpc = RISCVISD::TRUNCATE_VECTOR_VL_SSAT;
else
return SDValue();
MVT ValVT = Val.getSimpleValueType();
do {
MVT ValEltVT = MVT::getIntegerVT(ValVT.getScalarSizeInBits() / 2);
ValVT = ValVT.changeVectorElementType(ValEltVT);
Val = DAG.getNode(ClipOpc, DL, ValVT, Val, Mask, VL);
} while (ValVT != VT);
return Val;
}
// Convert
// (iX ctpop (bitcast (vXi1 A)))
// ->
// (zext (vcpop.m (nxvYi1 (insert_subvec (vXi1 A)))))
// and
// (iN reduce.add (zext (vXi1 A to vXiN))
// ->
// (zext (vcpop.m (nxvYi1 (insert_subvec (vXi1 A)))))
// FIXME: It's complicated to match all the variations of this after type
// legalization so we only handle the pre-type legalization pattern, but that
// requires the fixed vector type to be legal.
static SDValue combineToVCPOP(SDNode *N, SelectionDAG &DAG,
const RISCVSubtarget &Subtarget) {
unsigned Opc = N->getOpcode();
assert((Opc == ISD::CTPOP || Opc == ISD::VECREDUCE_ADD) &&
"Unexpected opcode");
EVT VT = N->getValueType(0);
if (!VT.isScalarInteger())
return SDValue();
SDValue Src = N->getOperand(0);
if (Opc == ISD::CTPOP) {
// Peek through zero_extend. It doesn't change the count.
if (Src.getOpcode() == ISD::ZERO_EXTEND)
Src = Src.getOperand(0);
if (Src.getOpcode() != ISD::BITCAST)
return SDValue();
Src = Src.getOperand(0);
} else if (Opc == ISD::VECREDUCE_ADD) {
if (Src.getOpcode() != ISD::ZERO_EXTEND)
return SDValue();
Src = Src.getOperand(0);
}
EVT SrcEVT = Src.getValueType();
if (!SrcEVT.isSimple())
return SDValue();
MVT SrcMVT = SrcEVT.getSimpleVT();
// Make sure the input is an i1 vector.
if (!SrcMVT.isVector() || SrcMVT.getVectorElementType() != MVT::i1)
return SDValue();
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
if (!TLI.isTypeLegal(SrcMVT))
return SDValue();
// Check that destination type is large enough to hold result without
// overflow.
if (Opc == ISD::VECREDUCE_ADD) {
unsigned EltSize = SrcMVT.getScalarSizeInBits();
unsigned MinSize = SrcMVT.getSizeInBits().getKnownMinValue();
unsigned VectorBitsMax = Subtarget.getRealMaxVLen();
unsigned MaxVLMAX = SrcMVT.isFixedLengthVector()
? SrcMVT.getVectorNumElements()
: RISCVTargetLowering::computeVLMAX(
VectorBitsMax, EltSize, MinSize);
if (VT.getFixedSizeInBits() < Log2_32(MaxVLMAX) + 1)
return SDValue();
}
MVT ContainerVT = SrcMVT;
if (SrcMVT.isFixedLengthVector()) {
ContainerVT = getContainerForFixedLengthVector(DAG, SrcMVT, Subtarget);
Src = convertToScalableVector(ContainerVT, Src, DAG, Subtarget);
}
SDLoc DL(N);
auto [Mask, VL] = getDefaultVLOps(SrcMVT, ContainerVT, DL, DAG, Subtarget);
MVT XLenVT = Subtarget.getXLenVT();
SDValue Pop = DAG.getNode(RISCVISD::VCPOP_VL, DL, XLenVT, Src, Mask, VL);
return DAG.getZExtOrTrunc(Pop, DL, VT);
}
static SDValue performSHLCombine(SDNode *N,
TargetLowering::DAGCombinerInfo &DCI,
const RISCVSubtarget &Subtarget) {
// (shl (zext x), y) -> (vwsll x, y)
if (SDValue V = combineOp_VLToVWOp_VL(N, DCI, Subtarget))
return V;
// (shl (sext x), C) -> (vwmulsu x, 1u << C)
// (shl (zext x), C) -> (vwmulu x, 1u << C)
if (!DCI.isAfterLegalizeDAG())
return SDValue();
SDValue LHS = N->getOperand(0);
if (!LHS.hasOneUse())
return SDValue();
unsigned Opcode;
switch (LHS.getOpcode()) {
case ISD::SIGN_EXTEND:
case RISCVISD::VSEXT_VL:
Opcode = RISCVISD::VWMULSU_VL;
break;
case ISD::ZERO_EXTEND:
case RISCVISD::VZEXT_VL:
Opcode = RISCVISD::VWMULU_VL;
break;
default:
return SDValue();
}
SDValue RHS = N->getOperand(1);
APInt ShAmt;
uint64_t ShAmtInt;
if (ISD::isConstantSplatVector(RHS.getNode(), ShAmt))
ShAmtInt = ShAmt.getZExtValue();
else if (RHS.getOpcode() == RISCVISD::VMV_V_X_VL &&
RHS.getOperand(1).getOpcode() == ISD::Constant)
ShAmtInt = RHS.getConstantOperandVal(1);
else
return SDValue();
// Better foldings:
// (shl (sext x), 1) -> (vwadd x, x)
// (shl (zext x), 1) -> (vwaddu x, x)
if (ShAmtInt <= 1)
return SDValue();
SDValue NarrowOp = LHS.getOperand(0);
MVT NarrowVT = NarrowOp.getSimpleValueType();
uint64_t NarrowBits = NarrowVT.getScalarSizeInBits();
if (ShAmtInt >= NarrowBits)
return SDValue();
MVT VT = N->getSimpleValueType(0);
if (NarrowBits * 2 != VT.getScalarSizeInBits())
return SDValue();
SelectionDAG &DAG = DCI.DAG;
SDLoc DL(N);
SDValue Passthru, Mask, VL;
switch (N->getOpcode()) {
case ISD::SHL:
Passthru = DAG.getUNDEF(VT);
std::tie(Mask, VL) = getDefaultScalableVLOps(VT, DL, DAG, Subtarget);
break;
case RISCVISD::SHL_VL:
Passthru = N->getOperand(2);
Mask = N->getOperand(3);
VL = N->getOperand(4);
break;
default:
llvm_unreachable("Expected SHL");
}
return DAG.getNode(Opcode, DL, VT, NarrowOp,
DAG.getConstant(1ULL << ShAmtInt, SDLoc(RHS), NarrowVT),
Passthru, Mask, VL);
}
SDValue RISCVTargetLowering::PerformDAGCombine(SDNode *N,
DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
const MVT XLenVT = Subtarget.getXLenVT();
SDLoc DL(N);
// Helper to call SimplifyDemandedBits on an operand of N where only some low
// bits are demanded. N will be added to the Worklist if it was not deleted.
// Caller should return SDValue(N, 0) if this returns true.
auto SimplifyDemandedLowBitsHelper = [&](unsigned OpNo, unsigned LowBits) {
SDValue Op = N->getOperand(OpNo);
APInt Mask = APInt::getLowBitsSet(Op.getValueSizeInBits(), LowBits);
if (!SimplifyDemandedBits(Op, Mask, DCI))
return false;
if (N->getOpcode() != ISD::DELETED_NODE)
DCI.AddToWorklist(N);
return true;
};
switch (N->getOpcode()) {
default:
break;
case RISCVISD::SplitF64: {
SDValue Op0 = N->getOperand(0);
// If the input to SplitF64 is just BuildPairF64 then the operation is
// redundant. Instead, use BuildPairF64's operands directly.
if (Op0->getOpcode() == RISCVISD::BuildPairF64)
return DCI.CombineTo(N, Op0.getOperand(0), Op0.getOperand(1));
if (Op0->isUndef()) {
SDValue Lo = DAG.getUNDEF(MVT::i32);
SDValue Hi = DAG.getUNDEF(MVT::i32);
return DCI.CombineTo(N, Lo, Hi);
}
// It's cheaper to materialise two 32-bit integers than to load a double
// from the constant pool and transfer it to integer registers through the
// stack.
if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(Op0)) {
APInt V = C->getValueAPF().bitcastToAPInt();
SDValue Lo = DAG.getConstant(V.trunc(32), DL, MVT::i32);
SDValue Hi = DAG.getConstant(V.lshr(32).trunc(32), DL, MVT::i32);
return DCI.CombineTo(N, Lo, Hi);
}
// This is a target-specific version of a DAGCombine performed in
// DAGCombiner::visitBITCAST. It performs the equivalent of:
// fold (bitconvert (fneg x)) -> (xor (bitconvert x), signbit)
// fold (bitconvert (fabs x)) -> (and (bitconvert x), (not signbit))
if (!(Op0.getOpcode() == ISD::FNEG || Op0.getOpcode() == ISD::FABS) ||
!Op0.getNode()->hasOneUse() || Subtarget.hasStdExtZdinx())
break;
SDValue NewSplitF64 =
DAG.getNode(RISCVISD::SplitF64, DL, DAG.getVTList(MVT::i32, MVT::i32),
Op0.getOperand(0));
SDValue Lo = NewSplitF64.getValue(0);
SDValue Hi = NewSplitF64.getValue(1);
APInt SignBit = APInt::getSignMask(32);
if (Op0.getOpcode() == ISD::FNEG) {
SDValue NewHi = DAG.getNode(ISD::XOR, DL, MVT::i32, Hi,
DAG.getConstant(SignBit, DL, MVT::i32));
return DCI.CombineTo(N, Lo, NewHi);
}
assert(Op0.getOpcode() == ISD::FABS);
SDValue NewHi = DAG.getNode(ISD::AND, DL, MVT::i32, Hi,
DAG.getConstant(~SignBit, DL, MVT::i32));
return DCI.CombineTo(N, Lo, NewHi);
}
case RISCVISD::SLLW:
case RISCVISD::SRAW:
case RISCVISD::SRLW:
case RISCVISD::RORW:
case RISCVISD::ROLW: {
// Only the lower 32 bits of LHS and lower 5 bits of RHS are read.
if (SimplifyDemandedLowBitsHelper(0, 32) ||
SimplifyDemandedLowBitsHelper(1, 5))
return SDValue(N, 0);
break;
}
case RISCVISD::CLZW:
case RISCVISD::CTZW: {
// Only the lower 32 bits of the first operand are read
if (SimplifyDemandedLowBitsHelper(0, 32))
return SDValue(N, 0);
break;
}
case RISCVISD::FMV_W_X_RV64: {
// If the input to FMV_W_X_RV64 is just FMV_X_ANYEXTW_RV64 the the
// conversion is unnecessary and can be replaced with the
// FMV_X_ANYEXTW_RV64 operand.
SDValue Op0 = N->getOperand(0);
if (Op0.getOpcode() == RISCVISD::FMV_X_ANYEXTW_RV64)
return Op0.getOperand(0);
break;
}
case RISCVISD::FMV_X_ANYEXTH:
case RISCVISD::FMV_X_ANYEXTW_RV64: {
SDLoc DL(N);
SDValue Op0 = N->getOperand(0);
MVT VT = N->getSimpleValueType(0);
// Constant fold.
if (auto *CFP = dyn_cast<ConstantFPSDNode>(Op0)) {
APInt Val = CFP->getValueAPF().bitcastToAPInt().sext(VT.getSizeInBits());
return DAG.getConstant(Val, DL, VT);
}
// If the input to FMV_X_ANYEXTW_RV64 is just FMV_W_X_RV64 then the
// conversion is unnecessary and can be replaced with the FMV_W_X_RV64
// operand. Similar for FMV_X_ANYEXTH and FMV_H_X.
if ((N->getOpcode() == RISCVISD::FMV_X_ANYEXTW_RV64 &&
Op0->getOpcode() == RISCVISD::FMV_W_X_RV64) ||
(N->getOpcode() == RISCVISD::FMV_X_ANYEXTH &&
Op0->getOpcode() == RISCVISD::FMV_H_X)) {
assert(Op0.getOperand(0).getValueType() == VT &&
"Unexpected value type!");
return Op0.getOperand(0);
}
if (ISD::isNormalLoad(Op0.getNode()) && Op0.hasOneUse() &&
cast<LoadSDNode>(Op0)->isSimple()) {
MVT IVT = MVT::getIntegerVT(Op0.getValueSizeInBits());
auto *LN0 = cast<LoadSDNode>(Op0);
SDValue Load =
DAG.getExtLoad(ISD::EXTLOAD, SDLoc(N), VT, LN0->getChain(),
LN0->getBasePtr(), IVT, LN0->getMemOperand());
DAG.ReplaceAllUsesOfValueWith(Op0.getValue(1), Load.getValue(1));
return Load;
}
// This is a target-specific version of a DAGCombine performed in
// DAGCombiner::visitBITCAST. It performs the equivalent of:
// fold (bitconvert (fneg x)) -> (xor (bitconvert x), signbit)
// fold (bitconvert (fabs x)) -> (and (bitconvert x), (not signbit))
if (!(Op0.getOpcode() == ISD::FNEG || Op0.getOpcode() == ISD::FABS) ||
!Op0.getNode()->hasOneUse())
break;
SDValue NewFMV = DAG.getNode(N->getOpcode(), DL, VT, Op0.getOperand(0));
unsigned FPBits = N->getOpcode() == RISCVISD::FMV_X_ANYEXTW_RV64 ? 32 : 16;
APInt SignBit = APInt::getSignMask(FPBits).sext(VT.getSizeInBits());
if (Op0.getOpcode() == ISD::FNEG)
return DAG.getNode(ISD::XOR, DL, VT, NewFMV,
DAG.getConstant(SignBit, DL, VT));
assert(Op0.getOpcode() == ISD::FABS);
return DAG.getNode(ISD::AND, DL, VT, NewFMV,
DAG.getConstant(~SignBit, DL, VT));
}
case ISD::ABS: {
EVT VT = N->getValueType(0);
SDValue N0 = N->getOperand(0);
// abs (sext) -> zext (abs)
// abs (zext) -> zext (handled elsewhere)
if (VT.isVector() && N0.hasOneUse() && N0.getOpcode() == ISD::SIGN_EXTEND) {
SDValue Src = N0.getOperand(0);
SDLoc DL(N);
return DAG.getNode(ISD::ZERO_EXTEND, DL, VT,
DAG.getNode(ISD::ABS, DL, Src.getValueType(), Src));
}
break;
}
case ISD::ADD: {
if (SDValue V = combineOp_VLToVWOp_VL(N, DCI, Subtarget))
return V;
if (SDValue V = combineToVWMACC(N, DAG, Subtarget))
return V;
if (SDValue V = combineVqdotAccum(N, DAG, Subtarget))
return V;
return performADDCombine(N, DCI, Subtarget);
}
case ISD::SUB: {
if (SDValue V = combineOp_VLToVWOp_VL(N, DCI, Subtarget))
return V;
return performSUBCombine(N, DAG, Subtarget);
}
case ISD::AND:
return performANDCombine(N, DCI, Subtarget);
case ISD::OR: {
if (SDValue V = combineOp_VLToVWOp_VL(N, DCI, Subtarget))
return V;
return performORCombine(N, DCI, Subtarget);
}
case ISD::XOR:
return performXORCombine(N, DAG, Subtarget);
case ISD::MUL:
if (SDValue V = combineOp_VLToVWOp_VL(N, DCI, Subtarget))
return V;
return performMULCombine(N, DAG, DCI, Subtarget);
case ISD::SDIV:
case ISD::UDIV:
case ISD::SREM:
case ISD::UREM:
if (SDValue V = combineBinOpOfZExt(N, DAG))
return V;
break;
case ISD::FMUL: {
using namespace SDPatternMatch;
SDLoc DL(N);
EVT VT = N->getValueType(0);
SDValue X, Y;
// InstCombine canonicalizes fneg (fmul x, y) -> fmul x, (fneg y), see
// hoistFNegAboveFMulFDiv.
// Undo this and sink the fneg so we match more fmsub/fnmadd patterns.
if (sd_match(N, m_FMul(m_Value(X), m_OneUse(m_FNeg(m_Value(Y))))))
return DAG.getNode(ISD::FNEG, DL, VT,
DAG.getNode(ISD::FMUL, DL, VT, X, Y));
// fmul X, (copysign 1.0, Y) -> fsgnjx X, Y
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
if (N0->getOpcode() != ISD::FCOPYSIGN)
std::swap(N0, N1);
if (N0->getOpcode() != ISD::FCOPYSIGN)
return SDValue();
ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(N0->getOperand(0));
if (!C || !C->getValueAPF().isExactlyValue(+1.0))
return SDValue();
if (VT.isVector() || !isOperationLegal(ISD::FCOPYSIGN, VT))
return SDValue();
SDValue Sign = N0->getOperand(1);
if (Sign.getValueType() != VT)
return SDValue();
return DAG.getNode(RISCVISD::FSGNJX, DL, VT, N1, N0->getOperand(1));
}
case ISD::FADD:
case ISD::UMAX:
case ISD::UMIN:
case ISD::SMAX:
case ISD::SMIN:
case ISD::FMAXNUM:
case ISD::FMINNUM: {
if (SDValue V = combineBinOpToReduce(N, DAG, Subtarget))
return V;
if (SDValue V = combineBinOpOfExtractToReduceTree(N, DAG, Subtarget))
return V;
return SDValue();
}
case ISD::SETCC:
return performSETCCCombine(N, DCI, Subtarget);
case ISD::SIGN_EXTEND_INREG:
return performSIGN_EXTEND_INREGCombine(N, DCI, Subtarget);
case ISD::ZERO_EXTEND:
// Fold (zero_extend (fp_to_uint X)) to prevent forming fcvt+zexti32 during
// type legalization. This is safe because fp_to_uint produces poison if
// it overflows.
if (N->getValueType(0) == MVT::i64 && Subtarget.is64Bit()) {
SDValue Src = N->getOperand(0);
if (Src.getOpcode() == ISD::FP_TO_UINT &&
isTypeLegal(Src.getOperand(0).getValueType()))
return DAG.getNode(ISD::FP_TO_UINT, SDLoc(N), MVT::i64,
Src.getOperand(0));
if (Src.getOpcode() == ISD::STRICT_FP_TO_UINT && Src.hasOneUse() &&
isTypeLegal(Src.getOperand(1).getValueType())) {
SDVTList VTs = DAG.getVTList(MVT::i64, MVT::Other);
SDValue Res = DAG.getNode(ISD::STRICT_FP_TO_UINT, SDLoc(N), VTs,
Src.getOperand(0), Src.getOperand(1));
DCI.CombineTo(N, Res);
DAG.ReplaceAllUsesOfValueWith(Src.getValue(1), Res.getValue(1));
DCI.recursivelyDeleteUnusedNodes(Src.getNode());
return SDValue(N, 0); // Return N so it doesn't get rechecked.
}
}
return SDValue();
case RISCVISD::TRUNCATE_VECTOR_VL:
if (SDValue V = combineTruncOfSraSext(N, DAG))
return V;
return combineTruncToVnclip(N, DAG, Subtarget);
case ISD::VP_TRUNCATE:
return performVP_TRUNCATECombine(N, DAG, Subtarget);
case ISD::TRUNCATE:
return performTRUNCATECombine(N, DAG, Subtarget);
case ISD::SELECT:
return performSELECTCombine(N, DAG, Subtarget);
case ISD::VSELECT:
return performVSELECTCombine(N, DAG);
case RISCVISD::CZERO_EQZ:
case RISCVISD::CZERO_NEZ: {
SDValue Val = N->getOperand(0);
SDValue Cond = N->getOperand(1);
unsigned Opc = N->getOpcode();
// czero_eqz x, x -> x
if (Opc == RISCVISD::CZERO_EQZ && Val == Cond)
return Val;
unsigned InvOpc =
Opc == RISCVISD::CZERO_EQZ ? RISCVISD::CZERO_NEZ : RISCVISD::CZERO_EQZ;
// czero_eqz X, (xor Y, 1) -> czero_nez X, Y if Y is 0 or 1.
// czero_nez X, (xor Y, 1) -> czero_eqz X, Y if Y is 0 or 1.
if (Cond.getOpcode() == ISD::XOR && isOneConstant(Cond.getOperand(1))) {
SDValue NewCond = Cond.getOperand(0);
APInt Mask = APInt::getBitsSetFrom(NewCond.getValueSizeInBits(), 1);
if (DAG.MaskedValueIsZero(NewCond, Mask))
return DAG.getNode(InvOpc, SDLoc(N), N->getValueType(0), Val, NewCond);
}
// czero_eqz x, (setcc y, 0, ne) -> czero_eqz x, y
// czero_nez x, (setcc y, 0, ne) -> czero_nez x, y
// czero_eqz x, (setcc y, 0, eq) -> czero_nez x, y
// czero_nez x, (setcc y, 0, eq) -> czero_eqz x, y
if (Cond.getOpcode() == ISD::SETCC && isNullConstant(Cond.getOperand(1))) {
ISD::CondCode CCVal = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
if (ISD::isIntEqualitySetCC(CCVal))
return DAG.getNode(CCVal == ISD::SETNE ? Opc : InvOpc, SDLoc(N),
N->getValueType(0), Val, Cond.getOperand(0));
}
return SDValue();
}
case RISCVISD::SELECT_CC: {
// Transform
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
SDValue CC = N->getOperand(2);
ISD::CondCode CCVal = cast<CondCodeSDNode>(CC)->get();
SDValue TrueV = N->getOperand(3);
SDValue FalseV = N->getOperand(4);
SDLoc DL(N);
EVT VT = N->getValueType(0);
// If the True and False values are the same, we don't need a select_cc.
if (TrueV == FalseV)
return TrueV;
// (select (x < 0), y, z) -> x >> (XLEN - 1) & (y - z) + z
// (select (x >= 0), y, z) -> x >> (XLEN - 1) & (z - y) + y
if (!Subtarget.hasShortForwardBranchOpt() && isa<ConstantSDNode>(TrueV) &&
isa<ConstantSDNode>(FalseV) && isNullConstant(RHS) &&
(CCVal == ISD::CondCode::SETLT || CCVal == ISD::CondCode::SETGE)) {
if (CCVal == ISD::CondCode::SETGE)
std::swap(TrueV, FalseV);
int64_t TrueSImm = cast<ConstantSDNode>(TrueV)->getSExtValue();
int64_t FalseSImm = cast<ConstantSDNode>(FalseV)->getSExtValue();
// Only handle simm12, if it is not in this range, it can be considered as
// register.
if (isInt<12>(TrueSImm) && isInt<12>(FalseSImm) &&
isInt<12>(TrueSImm - FalseSImm)) {
SDValue SRA =
DAG.getNode(ISD::SRA, DL, VT, LHS,
DAG.getConstant(Subtarget.getXLen() - 1, DL, VT));
SDValue AND =
DAG.getNode(ISD::AND, DL, VT, SRA,
DAG.getSignedConstant(TrueSImm - FalseSImm, DL, VT));
return DAG.getNode(ISD::ADD, DL, VT, AND, FalseV);
}
if (CCVal == ISD::CondCode::SETGE)
std::swap(TrueV, FalseV);
}
if (combine_CC(LHS, RHS, CC, DL, DAG, Subtarget))
return DAG.getNode(RISCVISD::SELECT_CC, DL, N->getValueType(0),
{LHS, RHS, CC, TrueV, FalseV});
if (!Subtarget.hasConditionalMoveFusion()) {
// (select c, -1, y) -> -c | y
if (isAllOnesConstant(TrueV)) {
SDValue C = DAG.getSetCC(DL, VT, LHS, RHS, CCVal);
SDValue Neg = DAG.getNegative(C, DL, VT);
return DAG.getNode(ISD::OR, DL, VT, Neg, FalseV);
}
// (select c, y, -1) -> -!c | y
if (isAllOnesConstant(FalseV)) {
SDValue C =
DAG.getSetCC(DL, VT, LHS, RHS, ISD::getSetCCInverse(CCVal, VT));
SDValue Neg = DAG.getNegative(C, DL, VT);
return DAG.getNode(ISD::OR, DL, VT, Neg, TrueV);
}
// (select c, 0, y) -> -!c & y
if (isNullConstant(TrueV)) {
SDValue C =
DAG.getSetCC(DL, VT, LHS, RHS, ISD::getSetCCInverse(CCVal, VT));
SDValue Neg = DAG.getNegative(C, DL, VT);
return DAG.getNode(ISD::AND, DL, VT, Neg, FalseV);
}
// (select c, y, 0) -> -c & y
if (isNullConstant(FalseV)) {
SDValue C = DAG.getSetCC(DL, VT, LHS, RHS, CCVal);
SDValue Neg = DAG.getNegative(C, DL, VT);
return DAG.getNode(ISD::AND, DL, VT, Neg, TrueV);
}
// (riscvisd::select_cc x, 0, ne, x, 1) -> (add x, (setcc x, 0, eq))
// (riscvisd::select_cc x, 0, eq, 1, x) -> (add x, (setcc x, 0, eq))
if (((isOneConstant(FalseV) && LHS == TrueV &&
CCVal == ISD::CondCode::SETNE) ||
(isOneConstant(TrueV) && LHS == FalseV &&
CCVal == ISD::CondCode::SETEQ)) &&
isNullConstant(RHS)) {
// freeze it to be safe.
LHS = DAG.getFreeze(LHS);
SDValue C = DAG.getSetCC(DL, VT, LHS, RHS, ISD::CondCode::SETEQ);
return DAG.getNode(ISD::ADD, DL, VT, LHS, C);
}
}
// If both true/false are an xor with 1, pull through the select.
// This can occur after op legalization if both operands are setccs that
// require an xor to invert.
// FIXME: Generalize to other binary ops with identical operand?
if (TrueV.getOpcode() == ISD::XOR && FalseV.getOpcode() == ISD::XOR &&
TrueV.getOperand(1) == FalseV.getOperand(1) &&
isOneConstant(TrueV.getOperand(1)) &&
TrueV.hasOneUse() && FalseV.hasOneUse()) {
SDValue NewSel = DAG.getNode(RISCVISD::SELECT_CC, DL, VT, LHS, RHS, CC,
TrueV.getOperand(0), FalseV.getOperand(0));
return DAG.getNode(ISD::XOR, DL, VT, NewSel, TrueV.getOperand(1));
}
return SDValue();
}
case RISCVISD::BR_CC: {
SDValue LHS = N->getOperand(1);
SDValue RHS = N->getOperand(2);
SDValue CC = N->getOperand(3);
SDLoc DL(N);
if (combine_CC(LHS, RHS, CC, DL, DAG, Subtarget))
return DAG.getNode(RISCVISD::BR_CC, DL, N->getValueType(0),
N->getOperand(0), LHS, RHS, CC, N->getOperand(4));
return SDValue();
}
case ISD::BITREVERSE:
return performBITREVERSECombine(N, DAG, Subtarget);
case ISD::FP_TO_SINT:
case ISD::FP_TO_UINT:
return performFP_TO_INTCombine(N, DCI, Subtarget);
case ISD::FP_TO_SINT_SAT:
case ISD::FP_TO_UINT_SAT:
return performFP_TO_INT_SATCombine(N, DCI, Subtarget);
case ISD::FCOPYSIGN: {
EVT VT = N->getValueType(0);
if (!VT.isVector())
break;
// There is a form of VFSGNJ which injects the negated sign of its second
// operand. Try and bubble any FNEG up after the extend/round to produce
// this optimized pattern. Avoid modifying cases where FP_ROUND and
// TRUNC=1.
SDValue In2 = N->getOperand(1);
// Avoid cases where the extend/round has multiple uses, as duplicating
// those is typically more expensive than removing a fneg.
if (!In2.hasOneUse())
break;
if (In2.getOpcode() != ISD::FP_EXTEND &&
(In2.getOpcode() != ISD::FP_ROUND || In2.getConstantOperandVal(1) != 0))
break;
In2 = In2.getOperand(0);
if (In2.getOpcode() != ISD::FNEG)
break;
SDLoc DL(N);
SDValue NewFPExtRound = DAG.getFPExtendOrRound(In2.getOperand(0), DL, VT);
return DAG.getNode(ISD::FCOPYSIGN, DL, VT, N->getOperand(0),
DAG.getNode(ISD::FNEG, DL, VT, NewFPExtRound));
}
case ISD::MGATHER: {
const auto *MGN = cast<MaskedGatherSDNode>(N);
const EVT VT = N->getValueType(0);
SDValue Index = MGN->getIndex();
SDValue ScaleOp = MGN->getScale();
ISD::MemIndexType IndexType = MGN->getIndexType();
assert(!MGN->isIndexScaled() &&
"Scaled gather/scatter should not be formed");
SDLoc DL(N);
if (legalizeScatterGatherIndexType(DL, Index, IndexType, DCI))
return DAG.getMaskedGather(
N->getVTList(), MGN->getMemoryVT(), DL,
{MGN->getChain(), MGN->getPassThru(), MGN->getMask(),
MGN->getBasePtr(), Index, ScaleOp},
MGN->getMemOperand(), IndexType, MGN->getExtensionType());
if (narrowIndex(Index, IndexType, DAG))
return DAG.getMaskedGather(
N->getVTList(), MGN->getMemoryVT(), DL,
{MGN->getChain(), MGN->getPassThru(), MGN->getMask(),
MGN->getBasePtr(), Index, ScaleOp},
MGN->getMemOperand(), IndexType, MGN->getExtensionType());
if (Index.getOpcode() == ISD::BUILD_VECTOR &&
MGN->getExtensionType() == ISD::NON_EXTLOAD && isTypeLegal(VT)) {
// The sequence will be XLenVT, not the type of Index. Tell
// isSimpleVIDSequence this so we avoid overflow.
if (std::optional<VIDSequence> SimpleVID =
isSimpleVIDSequence(Index, Subtarget.getXLen());
SimpleVID && SimpleVID->StepDenominator == 1) {
const int64_t StepNumerator = SimpleVID->StepNumerator;
const int64_t Addend = SimpleVID->Addend;
// Note: We don't need to check alignment here since (by assumption
// from the existence of the gather), our offsets must be sufficiently
// aligned.
const EVT PtrVT = getPointerTy(DAG.getDataLayout());
assert(MGN->getBasePtr()->getValueType(0) == PtrVT);
assert(IndexType == ISD::UNSIGNED_SCALED);
SDValue BasePtr = DAG.getNode(ISD::ADD, DL, PtrVT, MGN->getBasePtr(),
DAG.getSignedConstant(Addend, DL, PtrVT));
SDValue EVL = DAG.getElementCount(DL, Subtarget.getXLenVT(),
VT.getVectorElementCount());
SDValue StridedLoad = DAG.getStridedLoadVP(
VT, DL, MGN->getChain(), BasePtr,
DAG.getSignedConstant(StepNumerator, DL, XLenVT), MGN->getMask(),
EVL, MGN->getMemOperand());
SDValue Select = DAG.getSelect(DL, VT, MGN->getMask(), StridedLoad,
MGN->getPassThru());
return DAG.getMergeValues({Select, SDValue(StridedLoad.getNode(), 1)},
DL);
}
}
SmallVector<int> ShuffleMask;
if (MGN->getExtensionType() == ISD::NON_EXTLOAD &&
matchIndexAsShuffle(VT, Index, MGN->getMask(), ShuffleMask)) {
SDValue Load = DAG.getMaskedLoad(VT, DL, MGN->getChain(),
MGN->getBasePtr(), DAG.getUNDEF(XLenVT),
MGN->getMask(), DAG.getUNDEF(VT),
MGN->getMemoryVT(), MGN->getMemOperand(),
ISD::UNINDEXED, ISD::NON_EXTLOAD);
SDValue Shuffle =
DAG.getVectorShuffle(VT, DL, Load, DAG.getUNDEF(VT), ShuffleMask);
return DAG.getMergeValues({Shuffle, Load.getValue(1)}, DL);
}
if (MGN->getExtensionType() == ISD::NON_EXTLOAD &&
matchIndexAsWiderOp(VT, Index, MGN->getMask(),
MGN->getMemOperand()->getBaseAlign(), Subtarget)) {
SmallVector<SDValue> NewIndices;
for (unsigned i = 0; i < Index->getNumOperands(); i += 2)
NewIndices.push_back(Index.getOperand(i));
EVT IndexVT = Index.getValueType()
.getHalfNumVectorElementsVT(*DAG.getContext());
Index = DAG.getBuildVector(IndexVT, DL, NewIndices);
unsigned ElementSize = VT.getScalarStoreSize();
EVT WideScalarVT = MVT::getIntegerVT(ElementSize * 8 * 2);
auto EltCnt = VT.getVectorElementCount();
assert(EltCnt.isKnownEven() && "Splitting vector, but not in half!");
EVT WideVT = EVT::getVectorVT(*DAG.getContext(), WideScalarVT,
EltCnt.divideCoefficientBy(2));
SDValue Passthru = DAG.getBitcast(WideVT, MGN->getPassThru());
EVT MaskVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
EltCnt.divideCoefficientBy(2));
SDValue Mask = DAG.getSplat(MaskVT, DL, DAG.getConstant(1, DL, MVT::i1));
SDValue Gather =
DAG.getMaskedGather(DAG.getVTList(WideVT, MVT::Other), WideVT, DL,
{MGN->getChain(), Passthru, Mask, MGN->getBasePtr(),
Index, ScaleOp},
MGN->getMemOperand(), IndexType, ISD::NON_EXTLOAD);
SDValue Result = DAG.getBitcast(VT, Gather.getValue(0));
return DAG.getMergeValues({Result, Gather.getValue(1)}, DL);
}
break;
}
case ISD::MSCATTER:{
const auto *MSN = cast<MaskedScatterSDNode>(N);
SDValue Index = MSN->getIndex();
SDValue ScaleOp = MSN->getScale();
ISD::MemIndexType IndexType = MSN->getIndexType();
assert(!MSN->isIndexScaled() &&
"Scaled gather/scatter should not be formed");
SDLoc DL(N);
if (legalizeScatterGatherIndexType(DL, Index, IndexType, DCI))
return DAG.getMaskedScatter(
N->getVTList(), MSN->getMemoryVT(), DL,
{MSN->getChain(), MSN->getValue(), MSN->getMask(), MSN->getBasePtr(),
Index, ScaleOp},
MSN->getMemOperand(), IndexType, MSN->isTruncatingStore());
if (narrowIndex(Index, IndexType, DAG))
return DAG.getMaskedScatter(
N->getVTList(), MSN->getMemoryVT(), DL,
{MSN->getChain(), MSN->getValue(), MSN->getMask(), MSN->getBasePtr(),
Index, ScaleOp},
MSN->getMemOperand(), IndexType, MSN->isTruncatingStore());
EVT VT = MSN->getValue()->getValueType(0);
SmallVector<int> ShuffleMask;
if (!MSN->isTruncatingStore() &&
matchIndexAsShuffle(VT, Index, MSN->getMask(), ShuffleMask)) {
SDValue Shuffle = DAG.getVectorShuffle(VT, DL, MSN->getValue(),
DAG.getUNDEF(VT), ShuffleMask);
return DAG.getMaskedStore(MSN->getChain(), DL, Shuffle, MSN->getBasePtr(),
DAG.getUNDEF(XLenVT), MSN->getMask(),
MSN->getMemoryVT(), MSN->getMemOperand(),
ISD::UNINDEXED, false);
}
break;
}
case ISD::VP_GATHER: {
const auto *VPGN = cast<VPGatherSDNode>(N);
SDValue Index = VPGN->getIndex();
SDValue ScaleOp = VPGN->getScale();
ISD::MemIndexType IndexType = VPGN->getIndexType();
assert(!VPGN->isIndexScaled() &&
"Scaled gather/scatter should not be formed");
SDLoc DL(N);
if (legalizeScatterGatherIndexType(DL, Index, IndexType, DCI))
return DAG.getGatherVP(N->getVTList(), VPGN->getMemoryVT(), DL,
{VPGN->getChain(), VPGN->getBasePtr(), Index,
ScaleOp, VPGN->getMask(),
VPGN->getVectorLength()},
VPGN->getMemOperand(), IndexType);
if (narrowIndex(Index, IndexType, DAG))
return DAG.getGatherVP(N->getVTList(), VPGN->getMemoryVT(), DL,
{VPGN->getChain(), VPGN->getBasePtr(), Index,
ScaleOp, VPGN->getMask(),
VPGN->getVectorLength()},
VPGN->getMemOperand(), IndexType);
break;
}
case ISD::VP_SCATTER: {
const auto *VPSN = cast<VPScatterSDNode>(N);
SDValue Index = VPSN->getIndex();
SDValue ScaleOp = VPSN->getScale();
ISD::MemIndexType IndexType = VPSN->getIndexType();
assert(!VPSN->isIndexScaled() &&
"Scaled gather/scatter should not be formed");
SDLoc DL(N);
if (legalizeScatterGatherIndexType(DL, Index, IndexType, DCI))
return DAG.getScatterVP(N->getVTList(), VPSN->getMemoryVT(), DL,
{VPSN->getChain(), VPSN->getValue(),
VPSN->getBasePtr(), Index, ScaleOp,
VPSN->getMask(), VPSN->getVectorLength()},
VPSN->getMemOperand(), IndexType);
if (narrowIndex(Index, IndexType, DAG))
return DAG.getScatterVP(N->getVTList(), VPSN->getMemoryVT(), DL,
{VPSN->getChain(), VPSN->getValue(),
VPSN->getBasePtr(), Index, ScaleOp,
VPSN->getMask(), VPSN->getVectorLength()},
VPSN->getMemOperand(), IndexType);
break;
}
case RISCVISD::SHL_VL:
if (SDValue V = performSHLCombine(N, DCI, Subtarget))
return V;
[[fallthrough]];
case RISCVISD::SRA_VL:
case RISCVISD::SRL_VL: {
SDValue ShAmt = N->getOperand(1);
if (ShAmt.getOpcode() == RISCVISD::SPLAT_VECTOR_SPLIT_I64_VL) {
// We don't need the upper 32 bits of a 64-bit element for a shift amount.
SDLoc DL(N);
SDValue VL = N->getOperand(4);
EVT VT = N->getValueType(0);
ShAmt = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, DAG.getUNDEF(VT),
ShAmt.getOperand(1), VL);
return DAG.getNode(N->getOpcode(), DL, VT, N->getOperand(0), ShAmt,
N->getOperand(2), N->getOperand(3), N->getOperand(4));
}
break;
}
case ISD::SRA:
if (SDValue V = performSRACombine(N, DAG, Subtarget))
return V;
[[fallthrough]];
case ISD::SRL:
case ISD::SHL: {
if (N->getOpcode() == ISD::SHL) {
if (SDValue V = performSHLCombine(N, DCI, Subtarget))
return V;
}
SDValue ShAmt = N->getOperand(1);
if (ShAmt.getOpcode() == RISCVISD::SPLAT_VECTOR_SPLIT_I64_VL) {
// We don't need the upper 32 bits of a 64-bit element for a shift amount.
SDLoc DL(N);
EVT VT = N->getValueType(0);
ShAmt = DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, DAG.getUNDEF(VT),
ShAmt.getOperand(1),
DAG.getRegister(RISCV::X0, Subtarget.getXLenVT()));
return DAG.getNode(N->getOpcode(), DL, VT, N->getOperand(0), ShAmt);
}
break;
}
case RISCVISD::ADD_VL:
if (SDValue V = simplifyOp_VL(N))
return V;
if (SDValue V = combineOp_VLToVWOp_VL(N, DCI, Subtarget))
return V;
if (SDValue V = combineVqdotAccum(N, DAG, Subtarget))
return V;
return combineToVWMACC(N, DAG, Subtarget);
case RISCVISD::VWADD_W_VL:
case RISCVISD::VWADDU_W_VL:
case RISCVISD::VWSUB_W_VL:
case RISCVISD::VWSUBU_W_VL:
return performVWADDSUBW_VLCombine(N, DCI, Subtarget);
case RISCVISD::OR_VL:
case RISCVISD::SUB_VL:
case RISCVISD::MUL_VL:
return combineOp_VLToVWOp_VL(N, DCI, Subtarget);
case RISCVISD::VFMADD_VL:
case RISCVISD::VFNMADD_VL:
case RISCVISD::VFMSUB_VL:
case RISCVISD::VFNMSUB_VL:
case RISCVISD::STRICT_VFMADD_VL:
case RISCVISD::STRICT_VFNMADD_VL:
case RISCVISD::STRICT_VFMSUB_VL:
case RISCVISD::STRICT_VFNMSUB_VL:
return performVFMADD_VLCombine(N, DCI, Subtarget);
case RISCVISD::FADD_VL:
case RISCVISD::FSUB_VL:
case RISCVISD::FMUL_VL:
case RISCVISD::VFWADD_W_VL:
case RISCVISD::VFWSUB_W_VL:
return combineOp_VLToVWOp_VL(N, DCI, Subtarget);
case ISD::LOAD:
case ISD::STORE: {
if (DCI.isAfterLegalizeDAG())
if (SDValue V = performMemPairCombine(N, DCI))
return V;
if (N->getOpcode() != ISD::STORE)
break;
auto *Store = cast<StoreSDNode>(N);
SDValue Chain = Store->getChain();
EVT MemVT = Store->getMemoryVT();
SDValue Val = Store->getValue();
SDLoc DL(N);
bool IsScalarizable =
MemVT.isFixedLengthVector() && ISD::isNormalStore(Store) &&
Store->isSimple() &&
MemVT.getVectorElementType().bitsLE(Subtarget.getXLenVT()) &&
isPowerOf2_64(MemVT.getSizeInBits()) &&
MemVT.getSizeInBits() <= Subtarget.getXLen();
// If sufficiently aligned we can scalarize stores of constant vectors of
// any power-of-two size up to XLen bits, provided that they aren't too
// expensive to materialize.
// vsetivli zero, 2, e8, m1, ta, ma
// vmv.v.i v8, 4
// vse64.v v8, (a0)
// ->
// li a1, 1028
// sh a1, 0(a0)
if (DCI.isBeforeLegalize() && IsScalarizable &&
ISD::isBuildVectorOfConstantSDNodes(Val.getNode())) {
// Get the constant vector bits
APInt NewC(Val.getValueSizeInBits(), 0);
uint64_t EltSize = Val.getScalarValueSizeInBits();
for (unsigned i = 0; i < Val.getNumOperands(); i++) {
if (Val.getOperand(i).isUndef())
continue;
NewC.insertBits(Val.getConstantOperandAPInt(i).trunc(EltSize),
i * EltSize);
}
MVT NewVT = MVT::getIntegerVT(MemVT.getSizeInBits());
if (RISCVMatInt::getIntMatCost(NewC, Subtarget.getXLen(), Subtarget,
true) <= 2 &&
allowsMemoryAccessForAlignment(*DAG.getContext(), DAG.getDataLayout(),
NewVT, *Store->getMemOperand())) {
SDValue NewV = DAG.getConstant(NewC, DL, NewVT);
return DAG.getStore(Chain, DL, NewV, Store->getBasePtr(),
Store->getPointerInfo(), Store->getBaseAlign(),
Store->getMemOperand()->getFlags());
}
}
// Similarly, if sufficiently aligned we can scalarize vector copies, e.g.
// vsetivli zero, 2, e16, m1, ta, ma
// vle16.v v8, (a0)
// vse16.v v8, (a1)
if (auto *L = dyn_cast<LoadSDNode>(Val);
L && DCI.isBeforeLegalize() && IsScalarizable && L->isSimple() &&
L->hasNUsesOfValue(1, 0) && L->hasNUsesOfValue(1, 1) &&
Store->getChain() == SDValue(L, 1) && ISD::isNormalLoad(L) &&
L->getMemoryVT() == MemVT) {
MVT NewVT = MVT::getIntegerVT(MemVT.getSizeInBits());
if (allowsMemoryAccessForAlignment(*DAG.getContext(), DAG.getDataLayout(),
NewVT, *Store->getMemOperand()) &&
allowsMemoryAccessForAlignment(*DAG.getContext(), DAG.getDataLayout(),
NewVT, *L->getMemOperand())) {
SDValue NewL = DAG.getLoad(NewVT, DL, L->getChain(), L->getBasePtr(),
L->getPointerInfo(), L->getBaseAlign(),
L->getMemOperand()->getFlags());
return DAG.getStore(Chain, DL, NewL, Store->getBasePtr(),
Store->getPointerInfo(), Store->getBaseAlign(),
Store->getMemOperand()->getFlags());
}
}
// Combine store of vmv.x.s/vfmv.f.s to vse with VL of 1.
// vfmv.f.s is represented as extract element from 0. Match it late to avoid
// any illegal types.
if ((Val.getOpcode() == RISCVISD::VMV_X_S ||
(DCI.isAfterLegalizeDAG() &&
Val.getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
isNullConstant(Val.getOperand(1)))) &&
Val.hasOneUse()) {
SDValue Src = Val.getOperand(0);
MVT VecVT = Src.getSimpleValueType();
// VecVT should be scalable and memory VT should match the element type.
if (!Store->isIndexed() && VecVT.isScalableVector() &&
MemVT == VecVT.getVectorElementType()) {
SDLoc DL(N);
MVT MaskVT = getMaskTypeFor(VecVT);
return DAG.getStoreVP(
Store->getChain(), DL, Src, Store->getBasePtr(), Store->getOffset(),
DAG.getConstant(1, DL, MaskVT),
DAG.getConstant(1, DL, Subtarget.getXLenVT()), MemVT,
Store->getMemOperand(), Store->getAddressingMode(),
Store->isTruncatingStore(), /*IsCompress*/ false);
}
}
break;
}
case ISD::SPLAT_VECTOR: {
EVT VT = N->getValueType(0);
// Only perform this combine on legal MVT types.
if (!isTypeLegal(VT))
break;
if (auto Gather = matchSplatAsGather(N->getOperand(0), VT.getSimpleVT(), N,
DAG, Subtarget))
return Gather;
break;
}
case ISD::BUILD_VECTOR:
if (SDValue V = performBUILD_VECTORCombine(N, DAG, Subtarget, *this))
return V;
break;
case ISD::CONCAT_VECTORS:
if (SDValue V = performCONCAT_VECTORSCombine(N, DAG, Subtarget, *this))
return V;
break;
case ISD::VECTOR_SHUFFLE:
if (SDValue V = performVECTOR_SHUFFLECombine(N, DAG, Subtarget, *this))
return V;
break;
case ISD::INSERT_VECTOR_ELT:
if (SDValue V = performINSERT_VECTOR_ELTCombine(N, DAG, Subtarget, *this))
return V;
break;
case RISCVISD::VFMV_V_F_VL: {
const MVT VT = N->getSimpleValueType(0);
SDValue Passthru = N->getOperand(0);
SDValue Scalar = N->getOperand(1);
SDValue VL = N->getOperand(2);
// If VL is 1, we can use vfmv.s.f.
if (isOneConstant(VL))
return DAG.getNode(RISCVISD::VFMV_S_F_VL, DL, VT, Passthru, Scalar, VL);
break;
}
case RISCVISD::VMV_V_X_VL: {
const MVT VT = N->getSimpleValueType(0);
SDValue Passthru = N->getOperand(0);
SDValue Scalar = N->getOperand(1);
SDValue VL = N->getOperand(2);
// Tail agnostic VMV.V.X only demands the vector element bitwidth from the
// scalar input.
unsigned ScalarSize = Scalar.getValueSizeInBits();
unsigned EltWidth = VT.getScalarSizeInBits();
if (ScalarSize > EltWidth && Passthru.isUndef())
if (SimplifyDemandedLowBitsHelper(1, EltWidth))
return SDValue(N, 0);
// If VL is 1 and the scalar value won't benefit from immediate, we can
// use vmv.s.x.
ConstantSDNode *Const = dyn_cast<ConstantSDNode>(Scalar);
if (isOneConstant(VL) &&
(!Const || Const->isZero() ||
!Const->getAPIntValue().sextOrTrunc(EltWidth).isSignedIntN(5)))
return DAG.getNode(RISCVISD::VMV_S_X_VL, DL, VT, Passthru, Scalar, VL);
break;
}
case RISCVISD::VFMV_S_F_VL: {
SDValue Src = N->getOperand(1);
// Try to remove vector->scalar->vector if the scalar->vector is inserting
// into an undef vector.
// TODO: Could use a vslide or vmv.v.v for non-undef.
if (N->getOperand(0).isUndef() &&
Src.getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
isNullConstant(Src.getOperand(1)) &&
Src.getOperand(0).getValueType().isScalableVector()) {
EVT VT = N->getValueType(0);
SDValue EVSrc = Src.getOperand(0);
EVT EVSrcVT = EVSrc.getValueType();
assert(EVSrcVT.getVectorElementType() == VT.getVectorElementType());
// Widths match, just return the original vector.
if (EVSrcVT == VT)
return EVSrc;
SDLoc DL(N);
// Width is narrower, using insert_subvector.
if (EVSrcVT.getVectorMinNumElements() < VT.getVectorMinNumElements()) {
return DAG.getNode(ISD::INSERT_SUBVECTOR, DL, VT, DAG.getUNDEF(VT),
EVSrc,
DAG.getConstant(0, DL, Subtarget.getXLenVT()));
}
// Width is wider, using extract_subvector.
return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, VT, EVSrc,
DAG.getConstant(0, DL, Subtarget.getXLenVT()));
}
[[fallthrough]];
}
case RISCVISD::VMV_S_X_VL: {
const MVT VT = N->getSimpleValueType(0);
SDValue Passthru = N->getOperand(0);
SDValue Scalar = N->getOperand(1);
SDValue VL = N->getOperand(2);
// The vmv.s.x instruction copies the scalar integer register to element 0
// of the destination vector register. If SEW < XLEN, the least-significant
// bits are copied and the upper XLEN-SEW bits are ignored.
unsigned ScalarSize = Scalar.getValueSizeInBits();
unsigned EltWidth = VT.getScalarSizeInBits();
if (ScalarSize > EltWidth && SimplifyDemandedLowBitsHelper(1, EltWidth))
return SDValue(N, 0);
if (Scalar.getOpcode() == RISCVISD::VMV_X_S && Passthru.isUndef() &&
Scalar.getOperand(0).getValueType() == N->getValueType(0))
return Scalar.getOperand(0);
// Use M1 or smaller to avoid over constraining register allocation
const MVT M1VT = RISCVTargetLowering::getM1VT(VT);
if (M1VT.bitsLT(VT)) {
SDValue M1Passthru = DAG.getExtractSubvector(DL, M1VT, Passthru, 0);
SDValue Result =
DAG.getNode(N->getOpcode(), DL, M1VT, M1Passthru, Scalar, VL);
Result = DAG.getInsertSubvector(DL, Passthru, Result, 0);
return Result;
}
// We use a vmv.v.i if possible. We limit this to LMUL1. LMUL2 or
// higher would involve overly constraining the register allocator for
// no purpose.
if (ConstantSDNode *Const = dyn_cast<ConstantSDNode>(Scalar);
Const && !Const->isZero() && isInt<5>(Const->getSExtValue()) &&
VT.bitsLE(RISCVTargetLowering::getM1VT(VT)) && Passthru.isUndef())
return DAG.getNode(RISCVISD::VMV_V_X_VL, DL, VT, Passthru, Scalar, VL);
break;
}
case RISCVISD::VMV_X_S: {
SDValue Vec = N->getOperand(0);
MVT VecVT = N->getOperand(0).getSimpleValueType();
const MVT M1VT = RISCVTargetLowering::getM1VT(VecVT);
if (M1VT.bitsLT(VecVT)) {
Vec = DAG.getExtractSubvector(DL, M1VT, Vec, 0);
return DAG.getNode(RISCVISD::VMV_X_S, DL, N->getSimpleValueType(0), Vec);
}
break;
}
case ISD::INTRINSIC_VOID:
case ISD::INTRINSIC_W_CHAIN:
case ISD::INTRINSIC_WO_CHAIN: {
unsigned IntOpNo = N->getOpcode() == ISD::INTRINSIC_WO_CHAIN ? 0 : 1;
unsigned IntNo = N->getConstantOperandVal(IntOpNo);
switch (IntNo) {
// By default we do not combine any intrinsic.
default:
return SDValue();
case Intrinsic::riscv_vcpop:
case Intrinsic::riscv_vcpop_mask:
case Intrinsic::riscv_vfirst:
case Intrinsic::riscv_vfirst_mask: {
SDValue VL = N->getOperand(2);
if (IntNo == Intrinsic::riscv_vcpop_mask ||
IntNo == Intrinsic::riscv_vfirst_mask)
VL = N->getOperand(3);
if (!isNullConstant(VL))
return SDValue();
// If VL is 0, vcpop -> li 0, vfirst -> li -1.
SDLoc DL(N);
EVT VT = N->getValueType(0);
if (IntNo == Intrinsic::riscv_vfirst ||
IntNo == Intrinsic::riscv_vfirst_mask)
return DAG.getAllOnesConstant(DL, VT);
return DAG.getConstant(0, DL, VT);
}
case Intrinsic::riscv_vsseg2_mask:
case Intrinsic::riscv_vsseg3_mask:
case Intrinsic::riscv_vsseg4_mask:
case Intrinsic::riscv_vsseg5_mask:
case Intrinsic::riscv_vsseg6_mask:
case Intrinsic::riscv_vsseg7_mask:
case Intrinsic::riscv_vsseg8_mask: {
SDValue Tuple = N->getOperand(2);
unsigned NF = Tuple.getValueType().getRISCVVectorTupleNumFields();
if (Subtarget.hasOptimizedSegmentLoadStore(NF) || !Tuple.hasOneUse() ||
Tuple.getOpcode() != RISCVISD::TUPLE_INSERT ||
!Tuple.getOperand(0).isUndef())
return SDValue();
SDValue Val = Tuple.getOperand(1);
unsigned Idx = Tuple.getConstantOperandVal(2);
unsigned SEW = Val.getValueType().getScalarSizeInBits();
assert(Log2_64(SEW) == N->getConstantOperandVal(6) &&
"Type mismatch without bitcast?");
unsigned Stride = SEW / 8 * NF;
unsigned Offset = SEW / 8 * Idx;
SDValue Ops[] = {
/*Chain=*/N->getOperand(0),
/*IntID=*/
DAG.getTargetConstant(Intrinsic::riscv_vsse_mask, DL, XLenVT),
/*StoredVal=*/Val,
/*Ptr=*/
DAG.getNode(ISD::ADD, DL, XLenVT, N->getOperand(3),
DAG.getConstant(Offset, DL, XLenVT)),
/*Stride=*/DAG.getConstant(Stride, DL, XLenVT),
/*Mask=*/N->getOperand(4),
/*VL=*/N->getOperand(5)};
auto *OldMemSD = cast<MemIntrinsicSDNode>(N);
// Match getTgtMemIntrinsic for non-unit stride case
EVT MemVT = OldMemSD->getMemoryVT().getScalarType();
MachineFunction &MF = DAG.getMachineFunction();
MachineMemOperand *MMO = MF.getMachineMemOperand(
OldMemSD->getMemOperand(), Offset, MemoryLocation::UnknownSize);
SDVTList VTs = DAG.getVTList(MVT::Other);
return DAG.getMemIntrinsicNode(ISD::INTRINSIC_VOID, DL, VTs, Ops, MemVT,
MMO);
}
}
}
case ISD::EXPERIMENTAL_VP_REVERSE:
return performVP_REVERSECombine(N, DAG, Subtarget);
case ISD::VP_STORE:
return performVP_STORECombine(N, DAG, Subtarget);
case ISD::BITCAST: {
assert(Subtarget.useRVVForFixedLengthVectors());
SDValue N0 = N->getOperand(0);
EVT VT = N->getValueType(0);
EVT SrcVT = N0.getValueType();
if (VT.isRISCVVectorTuple() && N0->getOpcode() == ISD::SPLAT_VECTOR) {
unsigned NF = VT.getRISCVVectorTupleNumFields();
unsigned NumScalElts = VT.getSizeInBits().getKnownMinValue() / (NF * 8);
SDValue EltVal = DAG.getConstant(0, DL, Subtarget.getXLenVT());
MVT ScalTy = MVT::getScalableVectorVT(MVT::getIntegerVT(8), NumScalElts);
SDValue Splat = DAG.getNode(ISD::SPLAT_VECTOR, DL, ScalTy, EltVal);
SDValue Result = DAG.getUNDEF(VT);
for (unsigned i = 0; i < NF; ++i)
Result = DAG.getNode(RISCVISD::TUPLE_INSERT, DL, VT, Result, Splat,
DAG.getTargetConstant(i, DL, MVT::i32));
return Result;
}
// If this is a bitcast between a MVT::v4i1/v2i1/v1i1 and an illegal integer
// type, widen both sides to avoid a trip through memory.
if ((SrcVT == MVT::v1i1 || SrcVT == MVT::v2i1 || SrcVT == MVT::v4i1) &&
VT.isScalarInteger()) {
unsigned NumConcats = 8 / SrcVT.getVectorNumElements();
SmallVector<SDValue, 4> Ops(NumConcats, DAG.getUNDEF(SrcVT));
Ops[0] = N0;
SDLoc DL(N);
N0 = DAG.getNode(ISD::CONCAT_VECTORS, DL, MVT::v8i1, Ops);
N0 = DAG.getBitcast(MVT::i8, N0);
return DAG.getNode(ISD::TRUNCATE, DL, VT, N0);
}
return SDValue();
}
case ISD::VECREDUCE_ADD:
if (SDValue V = performVECREDUCECombine(N, DAG, Subtarget, *this))
return V;
[[fallthrough]];
case ISD::CTPOP:
if (SDValue V = combineToVCPOP(N, DAG, Subtarget))
return V;
break;
case RISCVISD::VRGATHER_VX_VL: {
// Note this assumes that out of bounds indices produce poison
// and can thus be replaced without having to prove them inbounds..
EVT VT = N->getValueType(0);
SDValue Src = N->getOperand(0);
SDValue Idx = N->getOperand(1);
SDValue Passthru = N->getOperand(2);
SDValue VL = N->getOperand(4);
// Warning: Unlike most cases we strip an insert_subvector, this one
// does not require the first operand to be undef.
if (Src.getOpcode() == ISD::INSERT_SUBVECTOR &&
isNullConstant(Src.getOperand(2)))
Src = Src.getOperand(1);
switch (Src.getOpcode()) {
default:
break;
case RISCVISD::VMV_V_X_VL:
case RISCVISD::VFMV_V_F_VL:
// Drop a redundant vrgather_vx.
// TODO: Remove the type restriction if we find a motivating
// test case?
if (Passthru.isUndef() && VL == Src.getOperand(2) &&
Src.getValueType() == VT)
return Src;
break;
case RISCVISD::VMV_S_X_VL:
case RISCVISD::VFMV_S_F_VL:
// If this use only demands lane zero from the source vmv.s.x, and
// doesn't have a passthru, then this vrgather.vi/vx is equivalent to
// a vmv.v.x. Note that there can be other uses of the original
// vmv.s.x and thus we can't eliminate it. (vfmv.s.f is analogous)
if (isNullConstant(Idx) && Passthru.isUndef() &&
VL == Src.getOperand(2)) {
unsigned Opc =
VT.isFloatingPoint() ? RISCVISD::VFMV_V_F_VL : RISCVISD::VMV_V_X_VL;
return DAG.getNode(Opc, DL, VT, DAG.getUNDEF(VT), Src.getOperand(1),
VL);
}
break;
}
break;
}
case RISCVISD::TUPLE_EXTRACT: {
EVT VT = N->getValueType(0);
SDValue Tuple = N->getOperand(0);
unsigned Idx = N->getConstantOperandVal(1);
if (!Tuple.hasOneUse() || Tuple.getOpcode() != ISD::INTRINSIC_W_CHAIN)
break;
unsigned NF = 0;
switch (Tuple.getConstantOperandVal(1)) {
default:
break;
case Intrinsic::riscv_vlseg2_mask:
case Intrinsic::riscv_vlseg3_mask:
case Intrinsic::riscv_vlseg4_mask:
case Intrinsic::riscv_vlseg5_mask:
case Intrinsic::riscv_vlseg6_mask:
case Intrinsic::riscv_vlseg7_mask:
case Intrinsic::riscv_vlseg8_mask:
NF = Tuple.getValueType().getRISCVVectorTupleNumFields();
break;
}
if (!NF || Subtarget.hasOptimizedSegmentLoadStore(NF))
break;
unsigned SEW = VT.getScalarSizeInBits();
assert(Log2_64(SEW) == Tuple.getConstantOperandVal(7) &&
"Type mismatch without bitcast?");
unsigned Stride = SEW / 8 * NF;
unsigned Offset = SEW / 8 * Idx;
SDValue Ops[] = {
/*Chain=*/Tuple.getOperand(0),
/*IntID=*/DAG.getTargetConstant(Intrinsic::riscv_vlse_mask, DL, XLenVT),
/*Passthru=*/Tuple.getOperand(2),
/*Ptr=*/
DAG.getNode(ISD::ADD, DL, XLenVT, Tuple.getOperand(3),
DAG.getConstant(Offset, DL, XLenVT)),
/*Stride=*/DAG.getConstant(Stride, DL, XLenVT),
/*Mask=*/Tuple.getOperand(4),
/*VL=*/Tuple.getOperand(5),
/*Policy=*/Tuple.getOperand(6)};
auto *TupleMemSD = cast<MemIntrinsicSDNode>(Tuple);
// Match getTgtMemIntrinsic for non-unit stride case
EVT MemVT = TupleMemSD->getMemoryVT().getScalarType();
MachineFunction &MF = DAG.getMachineFunction();
MachineMemOperand *MMO = MF.getMachineMemOperand(
TupleMemSD->getMemOperand(), Offset, MemoryLocation::UnknownSize);
SDVTList VTs = DAG.getVTList({VT, MVT::Other});
SDValue Result = DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs,
Ops, MemVT, MMO);
DAG.ReplaceAllUsesOfValueWith(Tuple.getValue(1), Result.getValue(1));
return Result.getValue(0);
}
case RISCVISD::TUPLE_INSERT: {
// tuple_insert tuple, undef, idx -> tuple
if (N->getOperand(1).isUndef())
return N->getOperand(0);
break;
}
case RISCVISD::VSLIDE1UP_VL:
case RISCVISD::VFSLIDE1UP_VL: {
using namespace SDPatternMatch;
SDValue SrcVec;
SDLoc DL(N);
MVT VT = N->getSimpleValueType(0);
// If the scalar we're sliding in was extracted from the first element of a
// vector, we can use that vector as the passthru in a normal slideup of 1.
// This saves us an extract_element instruction (i.e. vfmv.f.s, vmv.x.s).
if (!N->getOperand(0).isUndef() ||
!sd_match(N->getOperand(2),
m_AnyOf(m_ExtractElt(m_Value(SrcVec), m_Zero()),
m_Node(RISCVISD::VMV_X_S, m_Value(SrcVec)))))
break;
MVT SrcVecVT = SrcVec.getSimpleValueType();
if (SrcVecVT.getVectorElementType() != VT.getVectorElementType())
break;
// Adapt the value type of source vector.
if (SrcVecVT.isFixedLengthVector()) {
SrcVecVT = getContainerForFixedLengthVector(SrcVecVT);
SrcVec = convertToScalableVector(SrcVecVT, SrcVec, DAG, Subtarget);
}
if (SrcVecVT.getVectorMinNumElements() < VT.getVectorMinNumElements())
SrcVec = DAG.getInsertSubvector(DL, DAG.getUNDEF(VT), SrcVec, 0);
else
SrcVec = DAG.getExtractSubvector(DL, VT, SrcVec, 0);
return getVSlideup(DAG, Subtarget, DL, VT, SrcVec, N->getOperand(1),
DAG.getConstant(1, DL, XLenVT), N->getOperand(3),
N->getOperand(4));
}
}
return SDValue();
}
bool RISCVTargetLowering::shouldTransformSignedTruncationCheck(
EVT XVT, unsigned KeptBits) const {
// For vectors, we don't have a preference..
if (XVT.isVector())
return false;
if (XVT != MVT::i32 && XVT != MVT::i64)
return false;
// We can use sext.w for RV64 or an srai 31 on RV32.
if (KeptBits == 32 || KeptBits == 64)
return true;
// With Zbb we can use sext.h/sext.b.
return Subtarget.hasStdExtZbb() &&
((KeptBits == 8 && XVT == MVT::i64 && !Subtarget.is64Bit()) ||
KeptBits == 16);
}
bool RISCVTargetLowering::isDesirableToCommuteWithShift(
const SDNode *N, CombineLevel Level) const {
assert((N->getOpcode() == ISD::SHL || N->getOpcode() == ISD::SRA ||
N->getOpcode() == ISD::SRL) &&
"Expected shift op");
// The following folds are only desirable if `(OP _, c1 << c2)` can be
// materialised in fewer instructions than `(OP _, c1)`:
//
// (shl (add x, c1), c2) -> (add (shl x, c2), c1 << c2)
// (shl (or x, c1), c2) -> (or (shl x, c2), c1 << c2)
SDValue N0 = N->getOperand(0);
EVT Ty = N0.getValueType();
// LD/ST will optimize constant Offset extraction, so when AddNode is used by
// LD/ST, it can still complete the folding optimization operation performed
// above.
auto isUsedByLdSt = [](const SDNode *X, const SDNode *User) {
for (SDNode *Use : X->users()) {
// This use is the one we're on right now. Skip it
if (Use == User || Use->getOpcode() == ISD::SELECT)
continue;
if (!isa<StoreSDNode>(Use) && !isa<LoadSDNode>(Use))
return false;
}
return true;
};
if (Ty.isScalarInteger() &&
(N0.getOpcode() == ISD::ADD || N0.getOpcode() == ISD::OR)) {
if (N0.getOpcode() == ISD::ADD && !N0->hasOneUse())
return isUsedByLdSt(N0.getNode(), N);
auto *C1 = dyn_cast<ConstantSDNode>(N0->getOperand(1));
auto *C2 = dyn_cast<ConstantSDNode>(N->getOperand(1));
bool IsShXAdd =
(Subtarget.hasStdExtZba() || Subtarget.hasVendorXAndesPerf()) && C2 &&
C2->getZExtValue() >= 1 && C2->getZExtValue() <= 3;
bool IsQCShlAdd = Subtarget.hasVendorXqciac() && C2 &&
C2->getZExtValue() >= 4 && C2->getZExtValue() <= 31;
// Bail if we might break a sh{1,2,3}add/qc.shladd pattern.
if ((IsShXAdd || IsQCShlAdd) && N->hasOneUse() &&
N->user_begin()->getOpcode() == ISD::ADD &&
!isUsedByLdSt(*N->user_begin(), nullptr) &&
!isa<ConstantSDNode>(N->user_begin()->getOperand(1)))
return false;
if (C1 && C2) {
const APInt &C1Int = C1->getAPIntValue();
APInt ShiftedC1Int = C1Int << C2->getAPIntValue();
// We can materialise `c1 << c2` into an add immediate, so it's "free",
// and the combine should happen, to potentially allow further combines
// later.
if (ShiftedC1Int.getSignificantBits() <= 64 &&
isLegalAddImmediate(ShiftedC1Int.getSExtValue()))
return true;
// We can materialise `c1` in an add immediate, so it's "free", and the
// combine should be prevented.
if (C1Int.getSignificantBits() <= 64 &&
isLegalAddImmediate(C1Int.getSExtValue()))
return false;
// Neither constant will fit into an immediate, so find materialisation
// costs.
int C1Cost =
RISCVMatInt::getIntMatCost(C1Int, Ty.getSizeInBits(), Subtarget,
/*CompressionCost*/ true);
int ShiftedC1Cost = RISCVMatInt::getIntMatCost(
ShiftedC1Int, Ty.getSizeInBits(), Subtarget,
/*CompressionCost*/ true);
// Materialising `c1` is cheaper than materialising `c1 << c2`, so the
// combine should be prevented.
if (C1Cost < ShiftedC1Cost)
return false;
}
}
if (!N0->hasOneUse())
return false;
if (N0->getOpcode() == ISD::SIGN_EXTEND &&
N0->getOperand(0)->getOpcode() == ISD::ADD &&
!N0->getOperand(0)->hasOneUse())
return isUsedByLdSt(N0->getOperand(0).getNode(), N0.getNode());
return true;
}
bool RISCVTargetLowering::targetShrinkDemandedConstant(
SDValue Op, const APInt &DemandedBits, const APInt &DemandedElts,
TargetLoweringOpt &TLO) const {
// Delay this optimization as late as possible.
if (!TLO.LegalOps)
return false;
EVT VT = Op.getValueType();
if (VT.isVector())
return false;
unsigned Opcode = Op.getOpcode();
if (Opcode != ISD::AND && Opcode != ISD::OR && Opcode != ISD::XOR)
return false;
ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1));
if (!C)
return false;
const APInt &Mask = C->getAPIntValue();
// Clear all non-demanded bits initially.
APInt ShrunkMask = Mask & DemandedBits;
// Try to make a smaller immediate by setting undemanded bits.
APInt ExpandedMask = Mask | ~DemandedBits;
auto IsLegalMask = [ShrunkMask, ExpandedMask](const APInt &Mask) -> bool {
return ShrunkMask.isSubsetOf(Mask) && Mask.isSubsetOf(ExpandedMask);
};
auto UseMask = [Mask, Op, &TLO](const APInt &NewMask) -> bool {
if (NewMask == Mask)
return true;
SDLoc DL(Op);
SDValue NewC = TLO.DAG.getConstant(NewMask, DL, Op.getValueType());
SDValue NewOp = TLO.DAG.getNode(Op.getOpcode(), DL, Op.getValueType(),
Op.getOperand(0), NewC);
return TLO.CombineTo(Op, NewOp);
};
// If the shrunk mask fits in sign extended 12 bits, let the target
// independent code apply it.
if (ShrunkMask.isSignedIntN(12))
return false;
// And has a few special cases for zext.
if (Opcode == ISD::AND) {
// Preserve (and X, 0xffff), if zext.h exists use zext.h,
// otherwise use SLLI + SRLI.
APInt NewMask = APInt(Mask.getBitWidth(), 0xffff);
if (IsLegalMask(NewMask))
return UseMask(NewMask);
// Try to preserve (and X, 0xffffffff), the (zext_inreg X, i32) pattern.
if (VT == MVT::i64) {
APInt NewMask = APInt(64, 0xffffffff);
if (IsLegalMask(NewMask))
return UseMask(NewMask);
}
}
// For the remaining optimizations, we need to be able to make a negative
// number through a combination of mask and undemanded bits.
if (!ExpandedMask.isNegative())
return false;
// What is the fewest number of bits we need to represent the negative number.
unsigned MinSignedBits = ExpandedMask.getSignificantBits();
// Try to make a 12 bit negative immediate. If that fails try to make a 32
// bit negative immediate unless the shrunk immediate already fits in 32 bits.
// If we can't create a simm12, we shouldn't change opaque constants.
APInt NewMask = ShrunkMask;
if (MinSignedBits <= 12)
NewMask.setBitsFrom(11);
else if (!C->isOpaque() && MinSignedBits <= 32 && !ShrunkMask.isSignedIntN(32))
NewMask.setBitsFrom(31);
else
return false;
// Check that our new mask is a subset of the demanded mask.
assert(IsLegalMask(NewMask));
return UseMask(NewMask);
}
static uint64_t computeGREVOrGORC(uint64_t x, unsigned ShAmt, bool IsGORC) {
static const uint64_t GREVMasks[] = {
0x5555555555555555ULL, 0x3333333333333333ULL, 0x0F0F0F0F0F0F0F0FULL,
0x00FF00FF00FF00FFULL, 0x0000FFFF0000FFFFULL, 0x00000000FFFFFFFFULL};
for (unsigned Stage = 0; Stage != 6; ++Stage) {
unsigned Shift = 1 << Stage;
if (ShAmt & Shift) {
uint64_t Mask = GREVMasks[Stage];
uint64_t Res = ((x & Mask) << Shift) | ((x >> Shift) & Mask);
if (IsGORC)
Res |= x;
x = Res;
}
}
return x;
}
void RISCVTargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
KnownBits &Known,
const APInt &DemandedElts,
const SelectionDAG &DAG,
unsigned Depth) const {
unsigned BitWidth = Known.getBitWidth();
unsigned Opc = Op.getOpcode();
assert((Opc >= ISD::BUILTIN_OP_END ||
Opc == ISD::INTRINSIC_WO_CHAIN ||
Opc == ISD::INTRINSIC_W_CHAIN ||
Opc == ISD::INTRINSIC_VOID) &&
"Should use MaskedValueIsZero if you don't know whether Op"
" is a target node!");
Known.resetAll();
switch (Opc) {
default: break;
case RISCVISD::SELECT_CC: {
Known = DAG.computeKnownBits(Op.getOperand(4), Depth + 1);
// If we don't know any bits, early out.
if (Known.isUnknown())
break;
KnownBits Known2 = DAG.computeKnownBits(Op.getOperand(3), Depth + 1);
// Only known if known in both the LHS and RHS.
Known = Known.intersectWith(Known2);
break;
}
case RISCVISD::VCPOP_VL: {
KnownBits Known2 = DAG.computeKnownBits(Op.getOperand(2), Depth + 1);
Known.Zero.setBitsFrom(Known2.countMaxActiveBits());
break;
}
case RISCVISD::CZERO_EQZ:
case RISCVISD::CZERO_NEZ:
Known = DAG.computeKnownBits(Op.getOperand(0), Depth + 1);
// Result is either all zero or operand 0. We can propagate zeros, but not
// ones.
Known.One.clearAllBits();
break;
case RISCVISD::REMUW: {
KnownBits Known2;
Known = DAG.computeKnownBits(Op.getOperand(0), DemandedElts, Depth + 1);
Known2 = DAG.computeKnownBits(Op.getOperand(1), DemandedElts, Depth + 1);
// We only care about the lower 32 bits.
Known = KnownBits::urem(Known.trunc(32), Known2.trunc(32));
// Restore the original width by sign extending.
Known = Known.sext(BitWidth);
break;
}
case RISCVISD::DIVUW: {
KnownBits Known2;
Known = DAG.computeKnownBits(Op.getOperand(0), DemandedElts, Depth + 1);
Known2 = DAG.computeKnownBits(Op.getOperand(1), DemandedElts, Depth + 1);
// We only care about the lower 32 bits.
Known = KnownBits::udiv(Known.trunc(32), Known2.trunc(32));
// Restore the original width by sign extending.
Known = Known.sext(BitWidth);
break;
}
case RISCVISD::SLLW: {
KnownBits Known2;
Known = DAG.computeKnownBits(Op.getOperand(0), DemandedElts, Depth + 1);
Known2 = DAG.computeKnownBits(Op.getOperand(1), DemandedElts, Depth + 1);
Known = KnownBits::shl(Known.trunc(32), Known2.trunc(5).zext(32));
// Restore the original width by sign extending.
Known = Known.sext(BitWidth);
break;
}
case RISCVISD::SRLW: {
KnownBits Known2;
Known = DAG.computeKnownBits(Op.getOperand(0), DemandedElts, Depth + 1);
Known2 = DAG.computeKnownBits(Op.getOperand(1), DemandedElts, Depth + 1);
Known = KnownBits::lshr(Known.trunc(32), Known2.trunc(5).zext(32));
// Restore the original width by sign extending.
Known = Known.sext(BitWidth);
break;
}
case RISCVISD::SRAW: {
KnownBits Known2;
Known = DAG.computeKnownBits(Op.getOperand(0), DemandedElts, Depth + 1);
Known2 = DAG.computeKnownBits(Op.getOperand(1), DemandedElts, Depth + 1);
Known = KnownBits::ashr(Known.trunc(32), Known2.trunc(5).zext(32));
// Restore the original width by sign extending.
Known = Known.sext(BitWidth);
break;
}
case RISCVISD::CTZW: {
KnownBits Known2 = DAG.computeKnownBits(Op.getOperand(0), Depth + 1);
unsigned PossibleTZ = Known2.trunc(32).countMaxTrailingZeros();
unsigned LowBits = llvm::bit_width(PossibleTZ);
Known.Zero.setBitsFrom(LowBits);
break;
}
case RISCVISD::CLZW: {
KnownBits Known2 = DAG.computeKnownBits(Op.getOperand(0), Depth + 1);
unsigned PossibleLZ = Known2.trunc(32).countMaxLeadingZeros();
unsigned LowBits = llvm::bit_width(PossibleLZ);
Known.Zero.setBitsFrom(LowBits);
break;
}
case RISCVISD::BREV8:
case RISCVISD::ORC_B: {
// FIXME: This is based on the non-ratified Zbp GREV and GORC where a
// control value of 7 is equivalent to brev8 and orc.b.
Known = DAG.computeKnownBits(Op.getOperand(0), Depth + 1);
bool IsGORC = Op.getOpcode() == RISCVISD::ORC_B;
// To compute zeros for ORC_B, we need to invert the value and invert it
// back after. This inverting is harmless for BREV8.
Known.Zero =
~computeGREVOrGORC(~Known.Zero.getZExtValue(), 7, IsGORC);
Known.One = computeGREVOrGORC(Known.One.getZExtValue(), 7, IsGORC);
break;
}
case RISCVISD::READ_VLENB: {
// We can use the minimum and maximum VLEN values to bound VLENB. We
// know VLEN must be a power of two.
const unsigned MinVLenB = Subtarget.getRealMinVLen() / 8;
const unsigned MaxVLenB = Subtarget.getRealMaxVLen() / 8;
assert(MinVLenB > 0 && "READ_VLENB without vector extension enabled?");
Known.Zero.setLowBits(Log2_32(MinVLenB));
Known.Zero.setBitsFrom(Log2_32(MaxVLenB)+1);
if (MaxVLenB == MinVLenB)
Known.One.setBit(Log2_32(MinVLenB));
break;
}
case RISCVISD::FCLASS: {
// fclass will only set one of the low 10 bits.
Known.Zero.setBitsFrom(10);
break;
}
case ISD::INTRINSIC_W_CHAIN:
case ISD::INTRINSIC_WO_CHAIN: {
unsigned IntNo =
Op.getConstantOperandVal(Opc == ISD::INTRINSIC_WO_CHAIN ? 0 : 1);
switch (IntNo) {
default:
// We can't do anything for most intrinsics.
break;
case Intrinsic::riscv_vsetvli:
case Intrinsic::riscv_vsetvlimax: {
bool HasAVL = IntNo == Intrinsic::riscv_vsetvli;
unsigned VSEW = Op.getConstantOperandVal(HasAVL + 1);
RISCVVType::VLMUL VLMUL =
static_cast<RISCVVType::VLMUL>(Op.getConstantOperandVal(HasAVL + 2));
unsigned SEW = RISCVVType::decodeVSEW(VSEW);
auto [LMul, Fractional] = RISCVVType::decodeVLMUL(VLMUL);
uint64_t MaxVL = Subtarget.getRealMaxVLen() / SEW;
MaxVL = (Fractional) ? MaxVL / LMul : MaxVL * LMul;
// Result of vsetvli must be not larger than AVL.
if (HasAVL && isa<ConstantSDNode>(Op.getOperand(1)))
MaxVL = std::min(MaxVL, Op.getConstantOperandVal(1));
unsigned KnownZeroFirstBit = Log2_32(MaxVL) + 1;
if (BitWidth > KnownZeroFirstBit)
Known.Zero.setBitsFrom(KnownZeroFirstBit);
break;
}
}
break;
}
}
}
unsigned RISCVTargetLowering::ComputeNumSignBitsForTargetNode(
SDValue Op, const APInt &DemandedElts, const SelectionDAG &DAG,
unsigned Depth) const {
switch (Op.getOpcode()) {
default:
break;
case RISCVISD::SELECT_CC: {
unsigned Tmp =
DAG.ComputeNumSignBits(Op.getOperand(3), DemandedElts, Depth + 1);
if (Tmp == 1) return 1; // Early out.
unsigned Tmp2 =
DAG.ComputeNumSignBits(Op.getOperand(4), DemandedElts, Depth + 1);
return std::min(Tmp, Tmp2);
}
case RISCVISD::CZERO_EQZ:
case RISCVISD::CZERO_NEZ:
// Output is either all zero or operand 0. We can propagate sign bit count
// from operand 0.
return DAG.ComputeNumSignBits(Op.getOperand(0), DemandedElts, Depth + 1);
case RISCVISD::ABSW: {
// We expand this at isel to negw+max. The result will have 33 sign bits
// if the input has at least 33 sign bits.
unsigned Tmp =
DAG.ComputeNumSignBits(Op.getOperand(0), DemandedElts, Depth + 1);
if (Tmp < 33) return 1;
return 33;
}
case RISCVISD::SRAW: {
unsigned Tmp =
DAG.ComputeNumSignBits(Op.getOperand(0), DemandedElts, Depth + 1);
// sraw produces at least 33 sign bits. If the input already has more than
// 33 sign bits sraw, will preserve them.
// TODO: A more precise answer could be calculated depending on known bits
// in the shift amount.
return std::max(Tmp, 33U);
}
case RISCVISD::SLLW:
case RISCVISD::SRLW:
case RISCVISD::DIVW:
case RISCVISD::DIVUW:
case RISCVISD::REMUW:
case RISCVISD::ROLW:
case RISCVISD::RORW:
case RISCVISD::FCVT_W_RV64:
case RISCVISD::FCVT_WU_RV64:
case RISCVISD::STRICT_FCVT_W_RV64:
case RISCVISD::STRICT_FCVT_WU_RV64:
// TODO: As the result is sign-extended, this is conservatively correct.
return 33;
case RISCVISD::VMV_X_S: {
// The number of sign bits of the scalar result is computed by obtaining the
// element type of the input vector operand, subtracting its width from the
// XLEN, and then adding one (sign bit within the element type). If the
// element type is wider than XLen, the least-significant XLEN bits are
// taken.
unsigned XLen = Subtarget.getXLen();
unsigned EltBits = Op.getOperand(0).getScalarValueSizeInBits();
if (EltBits <= XLen)
return XLen - EltBits + 1;
break;
}
case ISD::INTRINSIC_W_CHAIN: {
unsigned IntNo = Op.getConstantOperandVal(1);
switch (IntNo) {
default:
break;
case Intrinsic::riscv_masked_atomicrmw_xchg:
case Intrinsic::riscv_masked_atomicrmw_add:
case Intrinsic::riscv_masked_atomicrmw_sub:
case Intrinsic::riscv_masked_atomicrmw_nand:
case Intrinsic::riscv_masked_atomicrmw_max:
case Intrinsic::riscv_masked_atomicrmw_min:
case Intrinsic::riscv_masked_atomicrmw_umax:
case Intrinsic::riscv_masked_atomicrmw_umin:
case Intrinsic::riscv_masked_cmpxchg:
// riscv_masked_{atomicrmw_*,cmpxchg} intrinsics represent an emulated
// narrow atomic operation. These are implemented using atomic
// operations at the minimum supported atomicrmw/cmpxchg width whose
// result is then sign extended to XLEN. With +A, the minimum width is
// 32 for both 64 and 32.
assert(getMinCmpXchgSizeInBits() == 32);
assert(Subtarget.hasStdExtA());
return Op.getValueSizeInBits() - 31;
}
break;
}
}
return 1;
}
bool RISCVTargetLowering::SimplifyDemandedBitsForTargetNode(
SDValue Op, const APInt &OriginalDemandedBits,
const APInt &OriginalDemandedElts, KnownBits &Known, TargetLoweringOpt &TLO,
unsigned Depth) const {
unsigned BitWidth = OriginalDemandedBits.getBitWidth();
switch (Op.getOpcode()) {
case RISCVISD::BREV8:
case RISCVISD::ORC_B: {
KnownBits Known2;
bool IsGORC = Op.getOpcode() == RISCVISD::ORC_B;
// For BREV8, we need to do BREV8 on the demanded bits.
// For ORC_B, any bit in the output demandeds all bits from the same byte.
// So we need to do ORC_B on the demanded bits.
APInt DemandedBits =
APInt(BitWidth, computeGREVOrGORC(OriginalDemandedBits.getZExtValue(),
7, IsGORC));
if (SimplifyDemandedBits(Op.getOperand(0), DemandedBits,
OriginalDemandedElts, Known2, TLO, Depth + 1))
return true;
// To compute zeros for ORC_B, we need to invert the value and invert it
// back after. This inverting is harmless for BREV8.
Known.Zero = ~computeGREVOrGORC(~Known2.Zero.getZExtValue(), 7, IsGORC);
Known.One = computeGREVOrGORC(Known2.One.getZExtValue(), 7, IsGORC);
return false;
}
}
return TargetLowering::SimplifyDemandedBitsForTargetNode(
Op, OriginalDemandedBits, OriginalDemandedElts, Known, TLO, Depth);
}
bool RISCVTargetLowering::canCreateUndefOrPoisonForTargetNode(
SDValue Op, const APInt &DemandedElts, const SelectionDAG &DAG,
bool PoisonOnly, bool ConsiderFlags, unsigned Depth) const {
// TODO: Add more target nodes.
switch (Op.getOpcode()) {
case RISCVISD::SLLW:
case RISCVISD::SRAW:
case RISCVISD::SRLW:
case RISCVISD::RORW:
case RISCVISD::ROLW:
// Only the lower 5 bits of RHS are read, guaranteeing the rotate/shift
// amount is bounds.
return false;
case RISCVISD::SELECT_CC:
// Integer comparisons cannot create poison.
assert(Op.getOperand(0).getValueType().isInteger() &&
"RISCVISD::SELECT_CC only compares integers");
return false;
}
return TargetLowering::canCreateUndefOrPoisonForTargetNode(
Op, DemandedElts, DAG, PoisonOnly, ConsiderFlags, Depth);
}
const Constant *
RISCVTargetLowering::getTargetConstantFromLoad(LoadSDNode *Ld) const {
assert(Ld && "Unexpected null LoadSDNode");
if (!ISD::isNormalLoad(Ld))
return nullptr;
SDValue Ptr = Ld->getBasePtr();
// Only constant pools with no offset are supported.
auto GetSupportedConstantPool = [](SDValue Ptr) -> ConstantPoolSDNode * {
auto *CNode = dyn_cast<ConstantPoolSDNode>(Ptr);
if (!CNode || CNode->isMachineConstantPoolEntry() ||
CNode->getOffset() != 0)
return nullptr;
return CNode;
};
// Simple case, LLA.
if (Ptr.getOpcode() == RISCVISD::LLA) {
auto *CNode = GetSupportedConstantPool(Ptr.getOperand(0));
if (!CNode || CNode->getTargetFlags() != 0)
return nullptr;
return CNode->getConstVal();
}
// Look for a HI and ADD_LO pair.
if (Ptr.getOpcode() != RISCVISD::ADD_LO ||
Ptr.getOperand(0).getOpcode() != RISCVISD::HI)
return nullptr;
auto *CNodeLo = GetSupportedConstantPool(Ptr.getOperand(1));
auto *CNodeHi = GetSupportedConstantPool(Ptr.getOperand(0).getOperand(0));
if (!CNodeLo || CNodeLo->getTargetFlags() != RISCVII::MO_LO ||
!CNodeHi || CNodeHi->getTargetFlags() != RISCVII::MO_HI)
return nullptr;
if (CNodeLo->getConstVal() != CNodeHi->getConstVal())
return nullptr;
return CNodeLo->getConstVal();
}
static MachineBasicBlock *emitReadCounterWidePseudo(MachineInstr &MI,
MachineBasicBlock *BB) {
assert(MI.getOpcode() == RISCV::ReadCounterWide && "Unexpected instruction");
// To read a 64-bit counter CSR on a 32-bit target, we read the two halves.
// Should the count have wrapped while it was being read, we need to try
// again.
// For example:
// ```
// read:
// csrrs x3, counterh # load high word of counter
// csrrs x2, counter # load low word of counter
// csrrs x4, counterh # load high word of counter
// bne x3, x4, read # check if high word reads match, otherwise try again
// ```
MachineFunction &MF = *BB->getParent();
const BasicBlock *LLVMBB = BB->getBasicBlock();
MachineFunction::iterator It = ++BB->getIterator();
MachineBasicBlock *LoopMBB = MF.CreateMachineBasicBlock(LLVMBB);
MF.insert(It, LoopMBB);
MachineBasicBlock *DoneMBB = MF.CreateMachineBasicBlock(LLVMBB);
MF.insert(It, DoneMBB);
// Transfer the remainder of BB and its successor edges to DoneMBB.
DoneMBB->splice(DoneMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
DoneMBB->transferSuccessorsAndUpdatePHIs(BB);
BB->addSuccessor(LoopMBB);
MachineRegisterInfo &RegInfo = MF.getRegInfo();
Register ReadAgainReg = RegInfo.createVirtualRegister(&RISCV::GPRRegClass);
Register LoReg = MI.getOperand(0).getReg();
Register HiReg = MI.getOperand(1).getReg();
int64_t LoCounter = MI.getOperand(2).getImm();
int64_t HiCounter = MI.getOperand(3).getImm();
DebugLoc DL = MI.getDebugLoc();
const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
BuildMI(LoopMBB, DL, TII->get(RISCV::CSRRS), HiReg)
.addImm(HiCounter)
.addReg(RISCV::X0);
BuildMI(LoopMBB, DL, TII->get(RISCV::CSRRS), LoReg)
.addImm(LoCounter)
.addReg(RISCV::X0);
BuildMI(LoopMBB, DL, TII->get(RISCV::CSRRS), ReadAgainReg)
.addImm(HiCounter)
.addReg(RISCV::X0);
BuildMI(LoopMBB, DL, TII->get(RISCV::BNE))
.addReg(HiReg)
.addReg(ReadAgainReg)
.addMBB(LoopMBB);
LoopMBB->addSuccessor(LoopMBB);
LoopMBB->addSuccessor(DoneMBB);
MI.eraseFromParent();
return DoneMBB;
}
static MachineBasicBlock *emitSplitF64Pseudo(MachineInstr &MI,
MachineBasicBlock *BB,
const RISCVSubtarget &Subtarget) {
assert(MI.getOpcode() == RISCV::SplitF64Pseudo && "Unexpected instruction");
MachineFunction &MF = *BB->getParent();
DebugLoc DL = MI.getDebugLoc();
const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
const TargetRegisterInfo *RI = MF.getSubtarget().getRegisterInfo();
Register LoReg = MI.getOperand(0).getReg();
Register HiReg = MI.getOperand(1).getReg();
Register SrcReg = MI.getOperand(2).getReg();
const TargetRegisterClass *SrcRC = &RISCV::FPR64RegClass;
int FI = MF.getInfo<RISCVMachineFunctionInfo>()->getMoveF64FrameIndex(MF);
TII.storeRegToStackSlot(*BB, MI, SrcReg, MI.getOperand(2).isKill(), FI, SrcRC,
RI, Register());
MachinePointerInfo MPI = MachinePointerInfo::getFixedStack(MF, FI);
MachineMemOperand *MMOLo =
MF.getMachineMemOperand(MPI, MachineMemOperand::MOLoad, 4, Align(8));
MachineMemOperand *MMOHi = MF.getMachineMemOperand(
MPI.getWithOffset(4), MachineMemOperand::MOLoad, 4, Align(8));
BuildMI(*BB, MI, DL, TII.get(RISCV::LW), LoReg)
.addFrameIndex(FI)
.addImm(0)
.addMemOperand(MMOLo);
BuildMI(*BB, MI, DL, TII.get(RISCV::LW), HiReg)
.addFrameIndex(FI)
.addImm(4)
.addMemOperand(MMOHi);
MI.eraseFromParent(); // The pseudo instruction is gone now.
return BB;
}
static MachineBasicBlock *emitBuildPairF64Pseudo(MachineInstr &MI,
MachineBasicBlock *BB,
const RISCVSubtarget &Subtarget) {
assert(MI.getOpcode() == RISCV::BuildPairF64Pseudo &&
"Unexpected instruction");
MachineFunction &MF = *BB->getParent();
DebugLoc DL = MI.getDebugLoc();
const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
const TargetRegisterInfo *RI = MF.getSubtarget().getRegisterInfo();
Register DstReg = MI.getOperand(0).getReg();
Register LoReg = MI.getOperand(1).getReg();
Register HiReg = MI.getOperand(2).getReg();
const TargetRegisterClass *DstRC = &RISCV::FPR64RegClass;
int FI = MF.getInfo<RISCVMachineFunctionInfo>()->getMoveF64FrameIndex(MF);
MachinePointerInfo MPI = MachinePointerInfo::getFixedStack(MF, FI);
MachineMemOperand *MMOLo =
MF.getMachineMemOperand(MPI, MachineMemOperand::MOStore, 4, Align(8));
MachineMemOperand *MMOHi = MF.getMachineMemOperand(
MPI.getWithOffset(4), MachineMemOperand::MOStore, 4, Align(8));
BuildMI(*BB, MI, DL, TII.get(RISCV::SW))
.addReg(LoReg, getKillRegState(MI.getOperand(1).isKill()))
.addFrameIndex(FI)
.addImm(0)
.addMemOperand(MMOLo);
BuildMI(*BB, MI, DL, TII.get(RISCV::SW))
.addReg(HiReg, getKillRegState(MI.getOperand(2).isKill()))
.addFrameIndex(FI)
.addImm(4)
.addMemOperand(MMOHi);
TII.loadRegFromStackSlot(*BB, MI, DstReg, FI, DstRC, RI, Register());
MI.eraseFromParent(); // The pseudo instruction is gone now.
return BB;
}
static MachineBasicBlock *emitQuietFCMP(MachineInstr &MI, MachineBasicBlock *BB,
unsigned RelOpcode, unsigned EqOpcode,
const RISCVSubtarget &Subtarget) {
DebugLoc DL = MI.getDebugLoc();
Register DstReg = MI.getOperand(0).getReg();
Register Src1Reg = MI.getOperand(1).getReg();
Register Src2Reg = MI.getOperand(2).getReg();
MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
Register SavedFFlags = MRI.createVirtualRegister(&RISCV::GPRRegClass);
const TargetInstrInfo &TII = *BB->getParent()->getSubtarget().getInstrInfo();
// Save the current FFLAGS.
BuildMI(*BB, MI, DL, TII.get(RISCV::ReadFFLAGS), SavedFFlags);
auto MIB = BuildMI(*BB, MI, DL, TII.get(RelOpcode), DstReg)
.addReg(Src1Reg)
.addReg(Src2Reg);
if (MI.getFlag(MachineInstr::MIFlag::NoFPExcept))
MIB->setFlag(MachineInstr::MIFlag::NoFPExcept);
// Restore the FFLAGS.
BuildMI(*BB, MI, DL, TII.get(RISCV::WriteFFLAGS))
.addReg(SavedFFlags, RegState::Kill);
// Issue a dummy FEQ opcode to raise exception for signaling NaNs.
auto MIB2 = BuildMI(*BB, MI, DL, TII.get(EqOpcode), RISCV::X0)
.addReg(Src1Reg, getKillRegState(MI.getOperand(1).isKill()))
.addReg(Src2Reg, getKillRegState(MI.getOperand(2).isKill()));
if (MI.getFlag(MachineInstr::MIFlag::NoFPExcept))
MIB2->setFlag(MachineInstr::MIFlag::NoFPExcept);
// Erase the pseudoinstruction.
MI.eraseFromParent();
return BB;
}
static MachineBasicBlock *
EmitLoweredCascadedSelect(MachineInstr &First, MachineInstr &Second,
MachineBasicBlock *ThisMBB,
const RISCVSubtarget &Subtarget) {
// Select_FPRX_ (rs1, rs2, imm, rs4, (Select_FPRX_ rs1, rs2, imm, rs4, rs5)
// Without this, custom-inserter would have generated:
//
// A
// | \
// | B
// | /
// C
// | \
// | D
// | /
// E
//
// A: X = ...; Y = ...
// B: empty
// C: Z = PHI [X, A], [Y, B]
// D: empty
// E: PHI [X, C], [Z, D]
//
// If we lower both Select_FPRX_ in a single step, we can instead generate:
//
// A
// | \
// | C
// | /|
// |/ |
// | |
// | D
// | /
// E
//
// A: X = ...; Y = ...
// D: empty
// E: PHI [X, A], [X, C], [Y, D]
const RISCVInstrInfo &TII = *Subtarget.getInstrInfo();
const DebugLoc &DL = First.getDebugLoc();
const BasicBlock *LLVM_BB = ThisMBB->getBasicBlock();
MachineFunction *F = ThisMBB->getParent();
MachineBasicBlock *FirstMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *SecondMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *SinkMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineFunction::iterator It = ++ThisMBB->getIterator();
F->insert(It, FirstMBB);
F->insert(It, SecondMBB);
F->insert(It, SinkMBB);
// Transfer the remainder of ThisMBB and its successor edges to SinkMBB.
SinkMBB->splice(SinkMBB->begin(), ThisMBB,
std::next(MachineBasicBlock::iterator(First)),
ThisMBB->end());
SinkMBB->transferSuccessorsAndUpdatePHIs(ThisMBB);
// Fallthrough block for ThisMBB.
ThisMBB->addSuccessor(FirstMBB);
// Fallthrough block for FirstMBB.
FirstMBB->addSuccessor(SecondMBB);
ThisMBB->addSuccessor(SinkMBB);
FirstMBB->addSuccessor(SinkMBB);
// This is fallthrough.
SecondMBB->addSuccessor(SinkMBB);
auto FirstCC = static_cast<RISCVCC::CondCode>(First.getOperand(3).getImm());
Register FLHS = First.getOperand(1).getReg();
Register FRHS = First.getOperand(2).getReg();
// Insert appropriate branch.
BuildMI(FirstMBB, DL, TII.get(RISCVCC::getBrCond(FirstCC, First.getOpcode())))
.addReg(FLHS)
.addReg(FRHS)
.addMBB(SinkMBB);
Register SLHS = Second.getOperand(1).getReg();
Register SRHS = Second.getOperand(2).getReg();
Register Op1Reg4 = First.getOperand(4).getReg();
Register Op1Reg5 = First.getOperand(5).getReg();
auto SecondCC = static_cast<RISCVCC::CondCode>(Second.getOperand(3).getImm());
// Insert appropriate branch.
BuildMI(ThisMBB, DL,
TII.get(RISCVCC::getBrCond(SecondCC, Second.getOpcode())))
.addReg(SLHS)
.addReg(SRHS)
.addMBB(SinkMBB);
Register DestReg = Second.getOperand(0).getReg();
Register Op2Reg4 = Second.getOperand(4).getReg();
BuildMI(*SinkMBB, SinkMBB->begin(), DL, TII.get(RISCV::PHI), DestReg)
.addReg(Op2Reg4)
.addMBB(ThisMBB)
.addReg(Op1Reg4)
.addMBB(FirstMBB)
.addReg(Op1Reg5)
.addMBB(SecondMBB);
// Now remove the Select_FPRX_s.
First.eraseFromParent();
Second.eraseFromParent();
return SinkMBB;
}
static MachineBasicBlock *emitSelectPseudo(MachineInstr &MI,
MachineBasicBlock *BB,
const RISCVSubtarget &Subtarget) {
// To "insert" Select_* instructions, we actually have to insert the triangle
// control-flow pattern. The incoming instructions know the destination vreg
// to set, the condition code register to branch on, the true/false values to
// select between, and the condcode to use to select the appropriate branch.
//
// We produce the following control flow:
// HeadMBB
// | \
// | IfFalseMBB
// | /
// TailMBB
//
// When we find a sequence of selects we attempt to optimize their emission
// by sharing the control flow. Currently we only handle cases where we have
// multiple selects with the exact same condition (same LHS, RHS and CC).
// The selects may be interleaved with other instructions if the other
// instructions meet some requirements we deem safe:
// - They are not pseudo instructions.
// - They are debug instructions. Otherwise,
// - They do not have side-effects, do not access memory and their inputs do
// not depend on the results of the select pseudo-instructions.
// The TrueV/FalseV operands of the selects cannot depend on the result of
// previous selects in the sequence.
// These conditions could be further relaxed. See the X86 target for a
// related approach and more information.
//
// Select_FPRX_ (rs1, rs2, imm, rs4, (Select_FPRX_ rs1, rs2, imm, rs4, rs5))
// is checked here and handled by a separate function -
// EmitLoweredCascadedSelect.
auto Next = next_nodbg(MI.getIterator(), BB->instr_end());
if (MI.getOpcode() != RISCV::Select_GPR_Using_CC_GPR &&
MI.getOperand(1).isReg() && MI.getOperand(2).isReg() &&
Next != BB->end() && Next->getOpcode() == MI.getOpcode() &&
Next->getOperand(5).getReg() == MI.getOperand(0).getReg() &&
Next->getOperand(5).isKill())
return EmitLoweredCascadedSelect(MI, *Next, BB, Subtarget);
Register LHS = MI.getOperand(1).getReg();
Register RHS;
if (MI.getOperand(2).isReg())
RHS = MI.getOperand(2).getReg();
auto CC = static_cast<RISCVCC::CondCode>(MI.getOperand(3).getImm());
SmallVector<MachineInstr *, 4> SelectDebugValues;
SmallSet<Register, 4> SelectDests;
SelectDests.insert(MI.getOperand(0).getReg());
MachineInstr *LastSelectPseudo = &MI;
for (auto E = BB->end(), SequenceMBBI = MachineBasicBlock::iterator(MI);
SequenceMBBI != E; ++SequenceMBBI) {
if (SequenceMBBI->isDebugInstr())
continue;
if (RISCVInstrInfo::isSelectPseudo(*SequenceMBBI)) {
if (SequenceMBBI->getOperand(1).getReg() != LHS ||
!SequenceMBBI->getOperand(2).isReg() ||
SequenceMBBI->getOperand(2).getReg() != RHS ||
SequenceMBBI->getOperand(3).getImm() != CC ||
SelectDests.count(SequenceMBBI->getOperand(4).getReg()) ||
SelectDests.count(SequenceMBBI->getOperand(5).getReg()))
break;
LastSelectPseudo = &*SequenceMBBI;
SequenceMBBI->collectDebugValues(SelectDebugValues);
SelectDests.insert(SequenceMBBI->getOperand(0).getReg());
continue;
}
if (SequenceMBBI->hasUnmodeledSideEffects() ||
SequenceMBBI->mayLoadOrStore() ||
SequenceMBBI->usesCustomInsertionHook())
break;
if (llvm::any_of(SequenceMBBI->operands(), [&](MachineOperand &MO) {
return MO.isReg() && MO.isUse() && SelectDests.count(MO.getReg());
}))
break;
}
const RISCVInstrInfo &TII = *Subtarget.getInstrInfo();
const BasicBlock *LLVM_BB = BB->getBasicBlock();
DebugLoc DL = MI.getDebugLoc();
MachineFunction::iterator I = ++BB->getIterator();
MachineBasicBlock *HeadMBB = BB;
MachineFunction *F = BB->getParent();
MachineBasicBlock *TailMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *IfFalseMBB = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(I, IfFalseMBB);
F->insert(I, TailMBB);
// Set the call frame size on entry to the new basic blocks.
unsigned CallFrameSize = TII.getCallFrameSizeAt(*LastSelectPseudo);
IfFalseMBB->setCallFrameSize(CallFrameSize);
TailMBB->setCallFrameSize(CallFrameSize);
// Transfer debug instructions associated with the selects to TailMBB.
for (MachineInstr *DebugInstr : SelectDebugValues) {
TailMBB->push_back(DebugInstr->removeFromParent());
}
// Move all instructions after the sequence to TailMBB.
TailMBB->splice(TailMBB->end(), HeadMBB,
std::next(LastSelectPseudo->getIterator()), HeadMBB->end());
// Update machine-CFG edges by transferring all successors of the current
// block to the new block which will contain the Phi nodes for the selects.
TailMBB->transferSuccessorsAndUpdatePHIs(HeadMBB);
// Set the successors for HeadMBB.
HeadMBB->addSuccessor(IfFalseMBB);
HeadMBB->addSuccessor(TailMBB);
// Insert appropriate branch.
if (MI.getOperand(2).isImm())
BuildMI(HeadMBB, DL, TII.get(RISCVCC::getBrCond(CC, MI.getOpcode())))
.addReg(LHS)
.addImm(MI.getOperand(2).getImm())
.addMBB(TailMBB);
else
BuildMI(HeadMBB, DL, TII.get(RISCVCC::getBrCond(CC, MI.getOpcode())))
.addReg(LHS)
.addReg(RHS)
.addMBB(TailMBB);
// IfFalseMBB just falls through to TailMBB.
IfFalseMBB->addSuccessor(TailMBB);
// Create PHIs for all of the select pseudo-instructions.
auto SelectMBBI = MI.getIterator();
auto SelectEnd = std::next(LastSelectPseudo->getIterator());
auto InsertionPoint = TailMBB->begin();
while (SelectMBBI != SelectEnd) {
auto Next = std::next(SelectMBBI);
if (RISCVInstrInfo::isSelectPseudo(*SelectMBBI)) {
// %Result = phi [ %TrueValue, HeadMBB ], [ %FalseValue, IfFalseMBB ]
BuildMI(*TailMBB, InsertionPoint, SelectMBBI->getDebugLoc(),
TII.get(RISCV::PHI), SelectMBBI->getOperand(0).getReg())
.addReg(SelectMBBI->getOperand(4).getReg())
.addMBB(HeadMBB)
.addReg(SelectMBBI->getOperand(5).getReg())
.addMBB(IfFalseMBB);
SelectMBBI->eraseFromParent();
}
SelectMBBI = Next;
}
F->getProperties().resetNoPHIs();
return TailMBB;
}
// Helper to find Masked Pseudo instruction from MC instruction, LMUL and SEW.
static const RISCV::RISCVMaskedPseudoInfo *
lookupMaskedIntrinsic(uint16_t MCOpcode, RISCVVType::VLMUL LMul, unsigned SEW) {
const RISCVVInversePseudosTable::PseudoInfo *Inverse =
RISCVVInversePseudosTable::getBaseInfo(MCOpcode, LMul, SEW);
assert(Inverse && "Unexpected LMUL and SEW pair for instruction");
const RISCV::RISCVMaskedPseudoInfo *Masked =
RISCV::lookupMaskedIntrinsicByUnmasked(Inverse->Pseudo);
assert(Masked && "Could not find masked instruction for LMUL and SEW pair");
return Masked;
}
static MachineBasicBlock *emitVFROUND_NOEXCEPT_MASK(MachineInstr &MI,
MachineBasicBlock *BB,
unsigned CVTXOpc) {
DebugLoc DL = MI.getDebugLoc();
const TargetInstrInfo &TII = *BB->getParent()->getSubtarget().getInstrInfo();
MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
Register SavedFFLAGS = MRI.createVirtualRegister(&RISCV::GPRRegClass);
// Save the old value of FFLAGS.
BuildMI(*BB, MI, DL, TII.get(RISCV::ReadFFLAGS), SavedFFLAGS);
assert(MI.getNumOperands() == 7);
// Emit a VFCVT_X_F
const TargetRegisterInfo *TRI =
BB->getParent()->getSubtarget().getRegisterInfo();
const TargetRegisterClass *RC = MI.getRegClassConstraint(0, &TII, TRI);
Register Tmp = MRI.createVirtualRegister(RC);
BuildMI(*BB, MI, DL, TII.get(CVTXOpc), Tmp)
.add(MI.getOperand(1))
.add(MI.getOperand(2))
.add(MI.getOperand(3))
.add(MachineOperand::CreateImm(7)) // frm = DYN
.add(MI.getOperand(4))
.add(MI.getOperand(5))
.add(MI.getOperand(6))
.add(MachineOperand::CreateReg(RISCV::FRM,
/*IsDef*/ false,
/*IsImp*/ true));
// Emit a VFCVT_F_X
RISCVVType::VLMUL LMul = RISCVII::getLMul(MI.getDesc().TSFlags);
unsigned Log2SEW = MI.getOperand(RISCVII::getSEWOpNum(MI.getDesc())).getImm();
// There is no E8 variant for VFCVT_F_X.
assert(Log2SEW >= 4);
unsigned CVTFOpc =
lookupMaskedIntrinsic(RISCV::VFCVT_F_X_V, LMul, 1 << Log2SEW)
->MaskedPseudo;
BuildMI(*BB, MI, DL, TII.get(CVTFOpc))
.add(MI.getOperand(0))
.add(MI.getOperand(1))
.addReg(Tmp)
.add(MI.getOperand(3))
.add(MachineOperand::CreateImm(7)) // frm = DYN
.add(MI.getOperand(4))
.add(MI.getOperand(5))
.add(MI.getOperand(6))
.add(MachineOperand::CreateReg(RISCV::FRM,
/*IsDef*/ false,
/*IsImp*/ true));
// Restore FFLAGS.
BuildMI(*BB, MI, DL, TII.get(RISCV::WriteFFLAGS))
.addReg(SavedFFLAGS, RegState::Kill);
// Erase the pseudoinstruction.
MI.eraseFromParent();
return BB;
}
static MachineBasicBlock *emitFROUND(MachineInstr &MI, MachineBasicBlock *MBB,
const RISCVSubtarget &Subtarget) {
unsigned CmpOpc, F2IOpc, I2FOpc, FSGNJOpc, FSGNJXOpc;
const TargetRegisterClass *RC;
switch (MI.getOpcode()) {
default:
llvm_unreachable("Unexpected opcode");
case RISCV::PseudoFROUND_H:
CmpOpc = RISCV::FLT_H;
F2IOpc = RISCV::FCVT_W_H;
I2FOpc = RISCV::FCVT_H_W;
FSGNJOpc = RISCV::FSGNJ_H;
FSGNJXOpc = RISCV::FSGNJX_H;
RC = &RISCV::FPR16RegClass;
break;
case RISCV::PseudoFROUND_H_INX:
CmpOpc = RISCV::FLT_H_INX;
F2IOpc = RISCV::FCVT_W_H_INX;
I2FOpc = RISCV::FCVT_H_W_INX;
FSGNJOpc = RISCV::FSGNJ_H_INX;
FSGNJXOpc = RISCV::FSGNJX_H_INX;
RC = &RISCV::GPRF16RegClass;
break;
case RISCV::PseudoFROUND_S:
CmpOpc = RISCV::FLT_S;
F2IOpc = RISCV::FCVT_W_S;
I2FOpc = RISCV::FCVT_S_W;
FSGNJOpc = RISCV::FSGNJ_S;
FSGNJXOpc = RISCV::FSGNJX_S;
RC = &RISCV::FPR32RegClass;
break;
case RISCV::PseudoFROUND_S_INX:
CmpOpc = RISCV::FLT_S_INX;
F2IOpc = RISCV::FCVT_W_S_INX;
I2FOpc = RISCV::FCVT_S_W_INX;
FSGNJOpc = RISCV::FSGNJ_S_INX;
FSGNJXOpc = RISCV::FSGNJX_S_INX;
RC = &RISCV::GPRF32RegClass;
break;
case RISCV::PseudoFROUND_D:
assert(Subtarget.is64Bit() && "Expected 64-bit GPR.");
CmpOpc = RISCV::FLT_D;
F2IOpc = RISCV::FCVT_L_D;
I2FOpc = RISCV::FCVT_D_L;
FSGNJOpc = RISCV::FSGNJ_D;
FSGNJXOpc = RISCV::FSGNJX_D;
RC = &RISCV::FPR64RegClass;
break;
case RISCV::PseudoFROUND_D_INX:
assert(Subtarget.is64Bit() && "Expected 64-bit GPR.");
CmpOpc = RISCV::FLT_D_INX;
F2IOpc = RISCV::FCVT_L_D_INX;
I2FOpc = RISCV::FCVT_D_L_INX;
FSGNJOpc = RISCV::FSGNJ_D_INX;
FSGNJXOpc = RISCV::FSGNJX_D_INX;
RC = &RISCV::GPRRegClass;
break;
}
const BasicBlock *BB = MBB->getBasicBlock();
DebugLoc DL = MI.getDebugLoc();
MachineFunction::iterator I = ++MBB->getIterator();
MachineFunction *F = MBB->getParent();
MachineBasicBlock *CvtMBB = F->CreateMachineBasicBlock(BB);
MachineBasicBlock *DoneMBB = F->CreateMachineBasicBlock(BB);
F->insert(I, CvtMBB);
F->insert(I, DoneMBB);
// Move all instructions after the sequence to DoneMBB.
DoneMBB->splice(DoneMBB->end(), MBB, MachineBasicBlock::iterator(MI),
MBB->end());
// Update machine-CFG edges by transferring all successors of the current
// block to the new block which will contain the Phi nodes for the selects.
DoneMBB->transferSuccessorsAndUpdatePHIs(MBB);
// Set the successors for MBB.
MBB->addSuccessor(CvtMBB);
MBB->addSuccessor(DoneMBB);
Register DstReg = MI.getOperand(0).getReg();
Register SrcReg = MI.getOperand(1).getReg();
Register MaxReg = MI.getOperand(2).getReg();
int64_t FRM = MI.getOperand(3).getImm();
const RISCVInstrInfo &TII = *Subtarget.getInstrInfo();
MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
Register FabsReg = MRI.createVirtualRegister(RC);
BuildMI(MBB, DL, TII.get(FSGNJXOpc), FabsReg).addReg(SrcReg).addReg(SrcReg);
// Compare the FP value to the max value.
Register CmpReg = MRI.createVirtualRegister(&RISCV::GPRRegClass);
auto MIB =
BuildMI(MBB, DL, TII.get(CmpOpc), CmpReg).addReg(FabsReg).addReg(MaxReg);
if (MI.getFlag(MachineInstr::MIFlag::NoFPExcept))
MIB->setFlag(MachineInstr::MIFlag::NoFPExcept);
// Insert branch.
BuildMI(MBB, DL, TII.get(RISCV::BEQ))
.addReg(CmpReg)
.addReg(RISCV::X0)
.addMBB(DoneMBB);
CvtMBB->addSuccessor(DoneMBB);
// Convert to integer.
Register F2IReg = MRI.createVirtualRegister(&RISCV::GPRRegClass);
MIB = BuildMI(CvtMBB, DL, TII.get(F2IOpc), F2IReg).addReg(SrcReg).addImm(FRM);
if (MI.getFlag(MachineInstr::MIFlag::NoFPExcept))
MIB->setFlag(MachineInstr::MIFlag::NoFPExcept);
// Convert back to FP.
Register I2FReg = MRI.createVirtualRegister(RC);
MIB = BuildMI(CvtMBB, DL, TII.get(I2FOpc), I2FReg).addReg(F2IReg).addImm(FRM);
if (MI.getFlag(MachineInstr::MIFlag::NoFPExcept))
MIB->setFlag(MachineInstr::MIFlag::NoFPExcept);
// Restore the sign bit.
Register CvtReg = MRI.createVirtualRegister(RC);
BuildMI(CvtMBB, DL, TII.get(FSGNJOpc), CvtReg).addReg(I2FReg).addReg(SrcReg);
// Merge the results.
BuildMI(*DoneMBB, DoneMBB->begin(), DL, TII.get(RISCV::PHI), DstReg)
.addReg(SrcReg)
.addMBB(MBB)
.addReg(CvtReg)
.addMBB(CvtMBB);
MI.eraseFromParent();
return DoneMBB;
}
MachineBasicBlock *
RISCVTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
MachineBasicBlock *BB) const {
switch (MI.getOpcode()) {
default:
llvm_unreachable("Unexpected instr type to insert");
case RISCV::ReadCounterWide:
assert(!Subtarget.is64Bit() &&
"ReadCounterWide is only to be used on riscv32");
return emitReadCounterWidePseudo(MI, BB);
case RISCV::Select_GPR_Using_CC_GPR:
case RISCV::Select_GPR_Using_CC_SImm5_CV:
case RISCV::Select_GPRNoX0_Using_CC_SImm5NonZero_QC:
case RISCV::Select_GPRNoX0_Using_CC_UImm5NonZero_QC:
case RISCV::Select_GPRNoX0_Using_CC_SImm16NonZero_QC:
case RISCV::Select_GPRNoX0_Using_CC_UImm16NonZero_QC:
case RISCV::Select_GPR_Using_CC_UImmLog2XLen_NDS:
case RISCV::Select_GPR_Using_CC_UImm7_NDS:
case RISCV::Select_FPR16_Using_CC_GPR:
case RISCV::Select_FPR16INX_Using_CC_GPR:
case RISCV::Select_FPR32_Using_CC_GPR:
case RISCV::Select_FPR32INX_Using_CC_GPR:
case RISCV::Select_FPR64_Using_CC_GPR:
case RISCV::Select_FPR64INX_Using_CC_GPR:
case RISCV::Select_FPR64IN32X_Using_CC_GPR:
return emitSelectPseudo(MI, BB, Subtarget);
case RISCV::BuildPairF64Pseudo:
return emitBuildPairF64Pseudo(MI, BB, Subtarget);
case RISCV::SplitF64Pseudo:
return emitSplitF64Pseudo(MI, BB, Subtarget);
case RISCV::PseudoQuietFLE_H:
return emitQuietFCMP(MI, BB, RISCV::FLE_H, RISCV::FEQ_H, Subtarget);
case RISCV::PseudoQuietFLE_H_INX:
return emitQuietFCMP(MI, BB, RISCV::FLE_H_INX, RISCV::FEQ_H_INX, Subtarget);
case RISCV::PseudoQuietFLT_H:
return emitQuietFCMP(MI, BB, RISCV::FLT_H, RISCV::FEQ_H, Subtarget);
case RISCV::PseudoQuietFLT_H_INX:
return emitQuietFCMP(MI, BB, RISCV::FLT_H_INX, RISCV::FEQ_H_INX, Subtarget);
case RISCV::PseudoQuietFLE_S:
return emitQuietFCMP(MI, BB, RISCV::FLE_S, RISCV::FEQ_S, Subtarget);
case RISCV::PseudoQuietFLE_S_INX:
return emitQuietFCMP(MI, BB, RISCV::FLE_S_INX, RISCV::FEQ_S_INX, Subtarget);
case RISCV::PseudoQuietFLT_S:
return emitQuietFCMP(MI, BB, RISCV::FLT_S, RISCV::FEQ_S, Subtarget);
case RISCV::PseudoQuietFLT_S_INX:
return emitQuietFCMP(MI, BB, RISCV::FLT_S_INX, RISCV::FEQ_S_INX, Subtarget);
case RISCV::PseudoQuietFLE_D:
return emitQuietFCMP(MI, BB, RISCV::FLE_D, RISCV::FEQ_D, Subtarget);
case RISCV::PseudoQuietFLE_D_INX:
return emitQuietFCMP(MI, BB, RISCV::FLE_D_INX, RISCV::FEQ_D_INX, Subtarget);
case RISCV::PseudoQuietFLE_D_IN32X:
return emitQuietFCMP(MI, BB, RISCV::FLE_D_IN32X, RISCV::FEQ_D_IN32X,
Subtarget);
case RISCV::PseudoQuietFLT_D:
return emitQuietFCMP(MI, BB, RISCV::FLT_D, RISCV::FEQ_D, Subtarget);
case RISCV::PseudoQuietFLT_D_INX:
return emitQuietFCMP(MI, BB, RISCV::FLT_D_INX, RISCV::FEQ_D_INX, Subtarget);
case RISCV::PseudoQuietFLT_D_IN32X:
return emitQuietFCMP(MI, BB, RISCV::FLT_D_IN32X, RISCV::FEQ_D_IN32X,
Subtarget);
case RISCV::PseudoVFROUND_NOEXCEPT_V_M1_MASK:
return emitVFROUND_NOEXCEPT_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_M1_MASK);
case RISCV::PseudoVFROUND_NOEXCEPT_V_M2_MASK:
return emitVFROUND_NOEXCEPT_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_M2_MASK);
case RISCV::PseudoVFROUND_NOEXCEPT_V_M4_MASK:
return emitVFROUND_NOEXCEPT_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_M4_MASK);
case RISCV::PseudoVFROUND_NOEXCEPT_V_M8_MASK:
return emitVFROUND_NOEXCEPT_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_M8_MASK);
case RISCV::PseudoVFROUND_NOEXCEPT_V_MF2_MASK:
return emitVFROUND_NOEXCEPT_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_MF2_MASK);
case RISCV::PseudoVFROUND_NOEXCEPT_V_MF4_MASK:
return emitVFROUND_NOEXCEPT_MASK(MI, BB, RISCV::PseudoVFCVT_X_F_V_MF4_MASK);
case RISCV::PseudoFROUND_H:
case RISCV::PseudoFROUND_H_INX:
case RISCV::PseudoFROUND_S:
case RISCV::PseudoFROUND_S_INX:
case RISCV::PseudoFROUND_D:
case RISCV::PseudoFROUND_D_INX:
case RISCV::PseudoFROUND_D_IN32X:
return emitFROUND(MI, BB, Subtarget);
case RISCV::PROBED_STACKALLOC_DYN:
return emitDynamicProbedAlloc(MI, BB);
case TargetOpcode::STATEPOINT:
// STATEPOINT is a pseudo instruction which has no implicit defs/uses
// while jal call instruction (where statepoint will be lowered at the end)
// has implicit def. This def is early-clobber as it will be set at
// the moment of the call and earlier than any use is read.
// Add this implicit dead def here as a workaround.
MI.addOperand(*MI.getMF(),
MachineOperand::CreateReg(
RISCV::X1, /*isDef*/ true,
/*isImp*/ true, /*isKill*/ false, /*isDead*/ true,
/*isUndef*/ false, /*isEarlyClobber*/ true));
[[fallthrough]];
case TargetOpcode::STACKMAP:
case TargetOpcode::PATCHPOINT:
if (!Subtarget.is64Bit())
reportFatalUsageError("STACKMAP, PATCHPOINT and STATEPOINT are only "
"supported on 64-bit targets");
return emitPatchPoint(MI, BB);
}
}
void RISCVTargetLowering::AdjustInstrPostInstrSelection(MachineInstr &MI,
SDNode *Node) const {
// If instruction defines FRM operand, conservatively set it as non-dead to
// express data dependency with FRM users and prevent incorrect instruction
// reordering.
if (auto *FRMDef = MI.findRegisterDefOperand(RISCV::FRM, /*TRI=*/nullptr)) {
FRMDef->setIsDead(false);
return;
}
// Add FRM dependency to any instructions with dynamic rounding mode.
int Idx = RISCV::getNamedOperandIdx(MI.getOpcode(), RISCV::OpName::frm);
if (Idx < 0) {
// Vector pseudos have FRM index indicated by TSFlags.
Idx = RISCVII::getFRMOpNum(MI.getDesc());
if (Idx < 0)
return;
}
if (MI.getOperand(Idx).getImm() != RISCVFPRndMode::DYN)
return;
// If the instruction already reads FRM, don't add another read.
if (MI.readsRegister(RISCV::FRM, /*TRI=*/nullptr))
return;
MI.addOperand(
MachineOperand::CreateReg(RISCV::FRM, /*isDef*/ false, /*isImp*/ true));
}
void RISCVTargetLowering::analyzeInputArgs(
MachineFunction &MF, CCState &CCInfo,
const SmallVectorImpl<ISD::InputArg> &Ins, bool IsRet,
RISCVCCAssignFn Fn) const {
for (const auto &[Idx, In] : enumerate(Ins)) {
MVT ArgVT = In.VT;
ISD::ArgFlagsTy ArgFlags = In.Flags;
if (Fn(Idx, ArgVT, ArgVT, CCValAssign::Full, ArgFlags, CCInfo, IsRet,
In.OrigTy)) {
LLVM_DEBUG(dbgs() << "InputArg #" << Idx << " has unhandled type "
<< ArgVT << '\n');
llvm_unreachable(nullptr);
}
}
}
void RISCVTargetLowering::analyzeOutputArgs(
MachineFunction &MF, CCState &CCInfo,
const SmallVectorImpl<ISD::OutputArg> &Outs, bool IsRet,
CallLoweringInfo *CLI, RISCVCCAssignFn Fn) const {
for (const auto &[Idx, Out] : enumerate(Outs)) {
MVT ArgVT = Out.VT;
ISD::ArgFlagsTy ArgFlags = Out.Flags;
if (Fn(Idx, ArgVT, ArgVT, CCValAssign::Full, ArgFlags, CCInfo, IsRet,
Out.OrigTy)) {
LLVM_DEBUG(dbgs() << "OutputArg #" << Idx << " has unhandled type "
<< ArgVT << "\n");
llvm_unreachable(nullptr);
}
}
}
// Convert Val to a ValVT. Should not be called for CCValAssign::Indirect
// values.
static SDValue convertLocVTToValVT(SelectionDAG &DAG, SDValue Val,
const CCValAssign &VA, const SDLoc &DL,
const RISCVSubtarget &Subtarget) {
if (VA.needsCustom()) {
if (VA.getLocVT().isInteger() &&
(VA.getValVT() == MVT::f16 || VA.getValVT() == MVT::bf16))
return DAG.getNode(RISCVISD::FMV_H_X, DL, VA.getValVT(), Val);
if (VA.getLocVT() == MVT::i64 && VA.getValVT() == MVT::f32)
return DAG.getNode(RISCVISD::FMV_W_X_RV64, DL, MVT::f32, Val);
if (VA.getValVT().isFixedLengthVector() && VA.getLocVT().isScalableVector())
return convertFromScalableVector(VA.getValVT(), Val, DAG, Subtarget);
llvm_unreachable("Unexpected Custom handling.");
}
switch (VA.getLocInfo()) {
default:
llvm_unreachable("Unexpected CCValAssign::LocInfo");
case CCValAssign::Full:
break;
case CCValAssign::BCvt:
Val = DAG.getNode(ISD::BITCAST, DL, VA.getValVT(), Val);
break;
}
return Val;
}
// The caller is responsible for loading the full value if the argument is
// passed with CCValAssign::Indirect.
static SDValue unpackFromRegLoc(SelectionDAG &DAG, SDValue Chain,
const CCValAssign &VA, const SDLoc &DL,
const ISD::InputArg &In,
const RISCVTargetLowering &TLI) {
MachineFunction &MF = DAG.getMachineFunction();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
EVT LocVT = VA.getLocVT();
SDValue Val;
const TargetRegisterClass *RC = TLI.getRegClassFor(LocVT.getSimpleVT());
Register VReg = RegInfo.createVirtualRegister(RC);
RegInfo.addLiveIn(VA.getLocReg(), VReg);
Val = DAG.getCopyFromReg(Chain, DL, VReg, LocVT);
// If input is sign extended from 32 bits, note it for the SExtWRemoval pass.
if (In.isOrigArg()) {
Argument *OrigArg = MF.getFunction().getArg(In.getOrigArgIndex());
if (OrigArg->getType()->isIntegerTy()) {
unsigned BitWidth = OrigArg->getType()->getIntegerBitWidth();
// An input zero extended from i31 can also be considered sign extended.
if ((BitWidth <= 32 && In.Flags.isSExt()) ||
(BitWidth < 32 && In.Flags.isZExt())) {
RISCVMachineFunctionInfo *RVFI = MF.getInfo<RISCVMachineFunctionInfo>();
RVFI->addSExt32Register(VReg);
}
}
}
if (VA.getLocInfo() == CCValAssign::Indirect)
return Val;
return convertLocVTToValVT(DAG, Val, VA, DL, TLI.getSubtarget());
}
static SDValue convertValVTToLocVT(SelectionDAG &DAG, SDValue Val,
const CCValAssign &VA, const SDLoc &DL,
const RISCVSubtarget &Subtarget) {
EVT LocVT = VA.getLocVT();
if (VA.needsCustom()) {
if (LocVT.isInteger() &&
(VA.getValVT() == MVT::f16 || VA.getValVT() == MVT::bf16))
return DAG.getNode(RISCVISD::FMV_X_ANYEXTH, DL, LocVT, Val);
if (LocVT == MVT::i64 && VA.getValVT() == MVT::f32)
return DAG.getNode(RISCVISD::FMV_X_ANYEXTW_RV64, DL, MVT::i64, Val);
if (VA.getValVT().isFixedLengthVector() && LocVT.isScalableVector())
return convertToScalableVector(LocVT, Val, DAG, Subtarget);
llvm_unreachable("Unexpected Custom handling.");
}
switch (VA.getLocInfo()) {
default:
llvm_unreachable("Unexpected CCValAssign::LocInfo");
case CCValAssign::Full:
break;
case CCValAssign::BCvt:
Val = DAG.getNode(ISD::BITCAST, DL, LocVT, Val);
break;
}
return Val;
}
// The caller is responsible for loading the full value if the argument is
// passed with CCValAssign::Indirect.
static SDValue unpackFromMemLoc(SelectionDAG &DAG, SDValue Chain,
const CCValAssign &VA, const SDLoc &DL) {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
EVT LocVT = VA.getLocVT();
EVT ValVT = VA.getValVT();
EVT PtrVT = MVT::getIntegerVT(DAG.getDataLayout().getPointerSizeInBits(0));
if (VA.getLocInfo() == CCValAssign::Indirect) {
// When the value is a scalable vector, we save the pointer which points to
// the scalable vector value in the stack. The ValVT will be the pointer
// type, instead of the scalable vector type.
ValVT = LocVT;
}
int FI = MFI.CreateFixedObject(ValVT.getStoreSize(), VA.getLocMemOffset(),
/*IsImmutable=*/true);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
SDValue Val;
ISD::LoadExtType ExtType = ISD::NON_EXTLOAD;
switch (VA.getLocInfo()) {
default:
llvm_unreachable("Unexpected CCValAssign::LocInfo");
case CCValAssign::Full:
case CCValAssign::Indirect:
case CCValAssign::BCvt:
break;
}
Val = DAG.getExtLoad(
ExtType, DL, LocVT, Chain, FIN,
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI), ValVT);
return Val;
}
static SDValue unpackF64OnRV32DSoftABI(SelectionDAG &DAG, SDValue Chain,
const CCValAssign &VA,
const CCValAssign &HiVA,
const SDLoc &DL) {
assert(VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64 &&
"Unexpected VA");
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
assert(VA.isRegLoc() && "Expected register VA assignment");
Register LoVReg = RegInfo.createVirtualRegister(&RISCV::GPRRegClass);
RegInfo.addLiveIn(VA.getLocReg(), LoVReg);
SDValue Lo = DAG.getCopyFromReg(Chain, DL, LoVReg, MVT::i32);
SDValue Hi;
if (HiVA.isMemLoc()) {
// Second half of f64 is passed on the stack.
int FI = MFI.CreateFixedObject(4, HiVA.getLocMemOffset(),
/*IsImmutable=*/true);
SDValue FIN = DAG.getFrameIndex(FI, MVT::i32);
Hi = DAG.getLoad(MVT::i32, DL, Chain, FIN,
MachinePointerInfo::getFixedStack(MF, FI));
} else {
// Second half of f64 is passed in another GPR.
Register HiVReg = RegInfo.createVirtualRegister(&RISCV::GPRRegClass);
RegInfo.addLiveIn(HiVA.getLocReg(), HiVReg);
Hi = DAG.getCopyFromReg(Chain, DL, HiVReg, MVT::i32);
}
return DAG.getNode(RISCVISD::BuildPairF64, DL, MVT::f64, Lo, Hi);
}
// Transform physical registers into virtual registers.
SDValue RISCVTargetLowering::LowerFormalArguments(
SDValue Chain, CallingConv::ID CallConv, bool IsVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &DL,
SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals) const {
MachineFunction &MF = DAG.getMachineFunction();
switch (CallConv) {
default:
reportFatalUsageError("Unsupported calling convention");
case CallingConv::C:
case CallingConv::Fast:
case CallingConv::SPIR_KERNEL:
case CallingConv::PreserveMost:
case CallingConv::GRAAL:
case CallingConv::RISCV_VectorCall:
#define CC_VLS_CASE(ABI_VLEN) case CallingConv::RISCV_VLSCall_##ABI_VLEN:
CC_VLS_CASE(32)
CC_VLS_CASE(64)
CC_VLS_CASE(128)
CC_VLS_CASE(256)
CC_VLS_CASE(512)
CC_VLS_CASE(1024)
CC_VLS_CASE(2048)
CC_VLS_CASE(4096)
CC_VLS_CASE(8192)
CC_VLS_CASE(16384)
CC_VLS_CASE(32768)
CC_VLS_CASE(65536)
#undef CC_VLS_CASE
break;
case CallingConv::GHC:
if (Subtarget.hasStdExtE())
reportFatalUsageError("GHC calling convention is not supported on RVE!");
if (!Subtarget.hasStdExtFOrZfinx() || !Subtarget.hasStdExtDOrZdinx())
reportFatalUsageError("GHC calling convention requires the (Zfinx/F) and "
"(Zdinx/D) instruction set extensions");
}
const Function &Func = MF.getFunction();
if (Func.hasFnAttribute("interrupt")) {
if (!Func.arg_empty())
reportFatalUsageError(
"Functions with the interrupt attribute cannot have arguments!");
StringRef Kind =
MF.getFunction().getFnAttribute("interrupt").getValueAsString();
constexpr StringLiteral SupportedInterruptKinds[] = {
"machine",
"supervisor",
"rnmi",
"qci-nest",
"qci-nonest",
"SiFive-CLIC-preemptible",
"SiFive-CLIC-stack-swap",
"SiFive-CLIC-preemptible-stack-swap",
};
if (!llvm::is_contained(SupportedInterruptKinds, Kind))
reportFatalUsageError(
"Function interrupt attribute argument not supported!");
if (Kind.starts_with("qci-") && !Subtarget.hasVendorXqciint())
reportFatalUsageError(
"'qci-*' interrupt kinds require Xqciint extension");
if (Kind.starts_with("SiFive-CLIC-") && !Subtarget.hasVendorXSfmclic())
reportFatalUsageError(
"'SiFive-CLIC-*' interrupt kinds require XSfmclic extension");
if (Kind == "rnmi" && !Subtarget.hasStdExtSmrnmi())
reportFatalUsageError("'rnmi' interrupt kind requires Srnmi extension");
const TargetFrameLowering *TFI = Subtarget.getFrameLowering();
if (Kind.starts_with("SiFive-CLIC-preemptible") && TFI->hasFP(MF))
reportFatalUsageError("'SiFive-CLIC-preemptible' interrupt kinds cannot "
"have a frame pointer");
}
EVT PtrVT = getPointerTy(DAG.getDataLayout());
MVT XLenVT = Subtarget.getXLenVT();
unsigned XLenInBytes = Subtarget.getXLen() / 8;
// Used with vargs to accumulate store chains.
std::vector<SDValue> OutChains;
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext());
if (CallConv == CallingConv::GHC)
CCInfo.AnalyzeFormalArguments(Ins, CC_RISCV_GHC);
else
analyzeInputArgs(MF, CCInfo, Ins, /*IsRet=*/false,
CallConv == CallingConv::Fast ? CC_RISCV_FastCC
: CC_RISCV);
for (unsigned i = 0, e = ArgLocs.size(), InsIdx = 0; i != e; ++i, ++InsIdx) {
CCValAssign &VA = ArgLocs[i];
SDValue ArgValue;
// Passing f64 on RV32D with a soft float ABI must be handled as a special
// case.
if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64) {
assert(VA.needsCustom());
ArgValue = unpackF64OnRV32DSoftABI(DAG, Chain, VA, ArgLocs[++i], DL);
} else if (VA.isRegLoc())
ArgValue = unpackFromRegLoc(DAG, Chain, VA, DL, Ins[InsIdx], *this);
else
ArgValue = unpackFromMemLoc(DAG, Chain, VA, DL);
if (VA.getLocInfo() == CCValAssign::Indirect) {
// If the original argument was split and passed by reference (e.g. i128
// on RV32), we need to load all parts of it here (using the same
// address). Vectors may be partly split to registers and partly to the
// stack, in which case the base address is partly offset and subsequent
// stores are relative to that.
InVals.push_back(DAG.getLoad(VA.getValVT(), DL, Chain, ArgValue,
MachinePointerInfo()));
unsigned ArgIndex = Ins[InsIdx].OrigArgIndex;
unsigned ArgPartOffset = Ins[InsIdx].PartOffset;
assert(VA.getValVT().isVector() || ArgPartOffset == 0);
while (i + 1 != e && Ins[InsIdx + 1].OrigArgIndex == ArgIndex) {
CCValAssign &PartVA = ArgLocs[i + 1];
unsigned PartOffset = Ins[InsIdx + 1].PartOffset - ArgPartOffset;
SDValue Offset = DAG.getIntPtrConstant(PartOffset, DL);
if (PartVA.getValVT().isScalableVector())
Offset = DAG.getNode(ISD::VSCALE, DL, XLenVT, Offset);
SDValue Address = DAG.getNode(ISD::ADD, DL, PtrVT, ArgValue, Offset);
InVals.push_back(DAG.getLoad(PartVA.getValVT(), DL, Chain, Address,
MachinePointerInfo()));
++i;
++InsIdx;
}
continue;
}
InVals.push_back(ArgValue);
}
if (any_of(ArgLocs,
[](CCValAssign &VA) { return VA.getLocVT().isScalableVector(); }))
MF.getInfo<RISCVMachineFunctionInfo>()->setIsVectorCall();
if (IsVarArg) {
ArrayRef<MCPhysReg> ArgRegs = RISCV::getArgGPRs(Subtarget.getTargetABI());
unsigned Idx = CCInfo.getFirstUnallocated(ArgRegs);
const TargetRegisterClass *RC = &RISCV::GPRRegClass;
MachineFrameInfo &MFI = MF.getFrameInfo();
MachineRegisterInfo &RegInfo = MF.getRegInfo();
RISCVMachineFunctionInfo *RVFI = MF.getInfo<RISCVMachineFunctionInfo>();
// Size of the vararg save area. For now, the varargs save area is either
// zero or large enough to hold a0-a7.
int VarArgsSaveSize = XLenInBytes * (ArgRegs.size() - Idx);
int FI;
// If all registers are allocated, then all varargs must be passed on the
// stack and we don't need to save any argregs.
if (VarArgsSaveSize == 0) {
int VaArgOffset = CCInfo.getStackSize();
FI = MFI.CreateFixedObject(XLenInBytes, VaArgOffset, true);
} else {
int VaArgOffset = -VarArgsSaveSize;
FI = MFI.CreateFixedObject(VarArgsSaveSize, VaArgOffset, true);
// If saving an odd number of registers then create an extra stack slot to
// ensure that the frame pointer is 2*XLEN-aligned, which in turn ensures
// offsets to even-numbered registered remain 2*XLEN-aligned.
if (Idx % 2) {
MFI.CreateFixedObject(
XLenInBytes, VaArgOffset - static_cast<int>(XLenInBytes), true);
VarArgsSaveSize += XLenInBytes;
}
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
// Copy the integer registers that may have been used for passing varargs
// to the vararg save area.
for (unsigned I = Idx; I < ArgRegs.size(); ++I) {
const Register Reg = RegInfo.createVirtualRegister(RC);
RegInfo.addLiveIn(ArgRegs[I], Reg);
SDValue ArgValue = DAG.getCopyFromReg(Chain, DL, Reg, XLenVT);
SDValue Store = DAG.getStore(
Chain, DL, ArgValue, FIN,
MachinePointerInfo::getFixedStack(MF, FI, (I - Idx) * XLenInBytes));
OutChains.push_back(Store);
FIN =
DAG.getMemBasePlusOffset(FIN, TypeSize::getFixed(XLenInBytes), DL);
}
}
// Record the frame index of the first variable argument
// which is a value necessary to VASTART.
RVFI->setVarArgsFrameIndex(FI);
RVFI->setVarArgsSaveSize(VarArgsSaveSize);
}
// All stores are grouped in one node to allow the matching between
// the size of Ins and InVals. This only happens for vararg functions.
if (!OutChains.empty()) {
OutChains.push_back(Chain);
Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, OutChains);
}
return Chain;
}
/// isEligibleForTailCallOptimization - Check whether the call is eligible
/// for tail call optimization.
/// Note: This is modelled after ARM's IsEligibleForTailCallOptimization.
bool RISCVTargetLowering::isEligibleForTailCallOptimization(
CCState &CCInfo, CallLoweringInfo &CLI, MachineFunction &MF,
const SmallVector<CCValAssign, 16> &ArgLocs) const {
auto CalleeCC = CLI.CallConv;
auto &Outs = CLI.Outs;
auto &Caller = MF.getFunction();
auto CallerCC = Caller.getCallingConv();
// Exception-handling functions need a special set of instructions to
// indicate a return to the hardware. Tail-calling another function would
// probably break this.
// TODO: The "interrupt" attribute isn't currently defined by RISC-V. This
// should be expanded as new function attributes are introduced.
if (Caller.hasFnAttribute("interrupt"))
return false;
// Do not tail call opt if the stack is used to pass parameters.
if (CCInfo.getStackSize() != 0)
return false;
// Do not tail call opt if any parameters need to be passed indirectly.
// Since long doubles (fp128) and i128 are larger than 2*XLEN, they are
// passed indirectly. So the address of the value will be passed in a
// register, or if not available, then the address is put on the stack. In
// order to pass indirectly, space on the stack often needs to be allocated
// in order to store the value. In this case the CCInfo.getNextStackOffset()
// != 0 check is not enough and we need to check if any CCValAssign ArgsLocs
// are passed CCValAssign::Indirect.
for (auto &VA : ArgLocs)
if (VA.getLocInfo() == CCValAssign::Indirect)
return false;
// Do not tail call opt if either caller or callee uses struct return
// semantics.
auto IsCallerStructRet = Caller.hasStructRetAttr();
auto IsCalleeStructRet = Outs.empty() ? false : Outs[0].Flags.isSRet();
if (IsCallerStructRet || IsCalleeStructRet)
return false;
// The callee has to preserve all registers the caller needs to preserve.
const RISCVRegisterInfo *TRI = Subtarget.getRegisterInfo();
const uint32_t *CallerPreserved = TRI->getCallPreservedMask(MF, CallerCC);
if (CalleeCC != CallerCC) {
const uint32_t *CalleePreserved = TRI->getCallPreservedMask(MF, CalleeCC);
if (!TRI->regmaskSubsetEqual(CallerPreserved, CalleePreserved))
return false;
}
// Byval parameters hand the function a pointer directly into the stack area
// we want to reuse during a tail call. Working around this *is* possible
// but less efficient and uglier in LowerCall.
for (auto &Arg : Outs)
if (Arg.Flags.isByVal())
return false;
return true;
}
static Align getPrefTypeAlign(EVT VT, SelectionDAG &DAG) {
return DAG.getDataLayout().getPrefTypeAlign(
VT.getTypeForEVT(*DAG.getContext()));
}
// Lower a call to a callseq_start + CALL + callseq_end chain, and add input
// and output parameter nodes.
SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
SelectionDAG &DAG = CLI.DAG;
SDLoc &DL = CLI.DL;
SmallVectorImpl<ISD::OutputArg> &Outs = CLI.Outs;
SmallVectorImpl<SDValue> &OutVals = CLI.OutVals;
SmallVectorImpl<ISD::InputArg> &Ins = CLI.Ins;
SDValue Chain = CLI.Chain;
SDValue Callee = CLI.Callee;
bool &IsTailCall = CLI.IsTailCall;
CallingConv::ID CallConv = CLI.CallConv;
bool IsVarArg = CLI.IsVarArg;
EVT PtrVT = getPointerTy(DAG.getDataLayout());
MVT XLenVT = Subtarget.getXLenVT();
const CallBase *CB = CLI.CB;
MachineFunction &MF = DAG.getMachineFunction();
MachineFunction::CallSiteInfo CSInfo;
// Set type id for call site info.
if (MF.getTarget().Options.EmitCallGraphSection && CB && CB->isIndirectCall())
CSInfo = MachineFunction::CallSiteInfo(*CB);
// Analyze the operands of the call, assigning locations to each operand.
SmallVector<CCValAssign, 16> ArgLocs;
CCState ArgCCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext());
if (CallConv == CallingConv::GHC) {
if (Subtarget.hasStdExtE())
reportFatalUsageError("GHC calling convention is not supported on RVE!");
ArgCCInfo.AnalyzeCallOperands(Outs, CC_RISCV_GHC);
} else
analyzeOutputArgs(MF, ArgCCInfo, Outs, /*IsRet=*/false, &CLI,
CallConv == CallingConv::Fast ? CC_RISCV_FastCC
: CC_RISCV);
// Check if it's really possible to do a tail call.
if (IsTailCall)
IsTailCall = isEligibleForTailCallOptimization(ArgCCInfo, CLI, MF, ArgLocs);
if (IsTailCall)
++NumTailCalls;
else if (CLI.CB && CLI.CB->isMustTailCall())
reportFatalInternalError("failed to perform tail call elimination on a "
"call site marked musttail");
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = ArgCCInfo.getStackSize();
// Create local copies for byval args
SmallVector<SDValue, 8> ByValArgs;
for (unsigned i = 0, e = Outs.size(); i != e; ++i) {
ISD::ArgFlagsTy Flags = Outs[i].Flags;
if (!Flags.isByVal())
continue;
SDValue Arg = OutVals[i];
unsigned Size = Flags.getByValSize();
Align Alignment = Flags.getNonZeroByValAlign();
int FI =
MF.getFrameInfo().CreateStackObject(Size, Alignment, /*isSS=*/false);
SDValue FIPtr = DAG.getFrameIndex(FI, getPointerTy(DAG.getDataLayout()));
SDValue SizeNode = DAG.getConstant(Size, DL, XLenVT);
Chain = DAG.getMemcpy(Chain, DL, FIPtr, Arg, SizeNode, Alignment,
/*IsVolatile=*/false,
/*AlwaysInline=*/false, /*CI*/ nullptr, IsTailCall,
MachinePointerInfo(), MachinePointerInfo());
ByValArgs.push_back(FIPtr);
}
if (!IsTailCall)
Chain = DAG.getCALLSEQ_START(Chain, NumBytes, 0, CLI.DL);
// Copy argument values to their designated locations.
SmallVector<std::pair<Register, SDValue>, 8> RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
SDValue StackPtr;
for (unsigned i = 0, j = 0, e = ArgLocs.size(), OutIdx = 0; i != e;
++i, ++OutIdx) {
CCValAssign &VA = ArgLocs[i];
SDValue ArgValue = OutVals[OutIdx];
ISD::ArgFlagsTy Flags = Outs[OutIdx].Flags;
// Handle passing f64 on RV32D with a soft float ABI as a special case.
if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64) {
assert(VA.isRegLoc() && "Expected register VA assignment");
assert(VA.needsCustom());
SDValue SplitF64 = DAG.getNode(
RISCVISD::SplitF64, DL, DAG.getVTList(MVT::i32, MVT::i32), ArgValue);
SDValue Lo = SplitF64.getValue(0);
SDValue Hi = SplitF64.getValue(1);
Register RegLo = VA.getLocReg();
RegsToPass.push_back(std::make_pair(RegLo, Lo));
// Get the CCValAssign for the Hi part.
CCValAssign &HiVA = ArgLocs[++i];
if (HiVA.isMemLoc()) {
// Second half of f64 is passed on the stack.
if (!StackPtr.getNode())
StackPtr = DAG.getCopyFromReg(Chain, DL, RISCV::X2, PtrVT);
SDValue Address =
DAG.getNode(ISD::ADD, DL, PtrVT, StackPtr,
DAG.getIntPtrConstant(HiVA.getLocMemOffset(), DL));
// Emit the store.
MemOpChains.push_back(DAG.getStore(
Chain, DL, Hi, Address,
MachinePointerInfo::getStack(MF, HiVA.getLocMemOffset())));
} else {
// Second half of f64 is passed in another GPR.
Register RegHigh = HiVA.getLocReg();
RegsToPass.push_back(std::make_pair(RegHigh, Hi));
}
continue;
}
// Promote the value if needed.
// For now, only handle fully promoted and indirect arguments.
if (VA.getLocInfo() == CCValAssign::Indirect) {
// Store the argument in a stack slot and pass its address.
Align StackAlign =
std::max(getPrefTypeAlign(Outs[OutIdx].ArgVT, DAG),
getPrefTypeAlign(ArgValue.getValueType(), DAG));
TypeSize StoredSize = ArgValue.getValueType().getStoreSize();
// If the original argument was split (e.g. i128), we need
// to store the required parts of it here (and pass just one address).
// Vectors may be partly split to registers and partly to the stack, in
// which case the base address is partly offset and subsequent stores are
// relative to that.
unsigned ArgIndex = Outs[OutIdx].OrigArgIndex;
unsigned ArgPartOffset = Outs[OutIdx].PartOffset;
assert(VA.getValVT().isVector() || ArgPartOffset == 0);
// Calculate the total size to store. We don't have access to what we're
// actually storing other than performing the loop and collecting the
// info.
SmallVector<std::pair<SDValue, SDValue>> Parts;
while (i + 1 != e && Outs[OutIdx + 1].OrigArgIndex == ArgIndex) {
SDValue PartValue = OutVals[OutIdx + 1];
unsigned PartOffset = Outs[OutIdx + 1].PartOffset - ArgPartOffset;
SDValue Offset = DAG.getIntPtrConstant(PartOffset, DL);
EVT PartVT = PartValue.getValueType();
if (PartVT.isScalableVector())
Offset = DAG.getNode(ISD::VSCALE, DL, XLenVT, Offset);
StoredSize += PartVT.getStoreSize();
StackAlign = std::max(StackAlign, getPrefTypeAlign(PartVT, DAG));
Parts.push_back(std::make_pair(PartValue, Offset));
++i;
++OutIdx;
}
SDValue SpillSlot = DAG.CreateStackTemporary(StoredSize, StackAlign);
int FI = cast<FrameIndexSDNode>(SpillSlot)->getIndex();
MemOpChains.push_back(
DAG.getStore(Chain, DL, ArgValue, SpillSlot,
MachinePointerInfo::getFixedStack(MF, FI)));
for (const auto &Part : Parts) {
SDValue PartValue = Part.first;
SDValue PartOffset = Part.second;
SDValue Address =
DAG.getNode(ISD::ADD, DL, PtrVT, SpillSlot, PartOffset);
MemOpChains.push_back(
DAG.getStore(Chain, DL, PartValue, Address,
MachinePointerInfo::getFixedStack(MF, FI)));
}
ArgValue = SpillSlot;
} else {
ArgValue = convertValVTToLocVT(DAG, ArgValue, VA, DL, Subtarget);
}
// Use local copy if it is a byval arg.
if (Flags.isByVal())
ArgValue = ByValArgs[j++];
if (VA.isRegLoc()) {
// Queue up the argument copies and emit them at the end.
RegsToPass.push_back(std::make_pair(VA.getLocReg(), ArgValue));
} else {
assert(VA.isMemLoc() && "Argument not register or memory");
assert(!IsTailCall && "Tail call not allowed if stack is used "
"for passing parameters");
// Work out the address of the stack slot.
if (!StackPtr.getNode())
StackPtr = DAG.getCopyFromReg(Chain, DL, RISCV::X2, PtrVT);
SDValue Address =
DAG.getNode(ISD::ADD, DL, PtrVT, StackPtr,
DAG.getIntPtrConstant(VA.getLocMemOffset(), DL));
// Emit the store.
MemOpChains.push_back(
DAG.getStore(Chain, DL, ArgValue, Address,
MachinePointerInfo::getStack(MF, VA.getLocMemOffset())));
}
}
// Join the stores, which are independent of one another.
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other, MemOpChains);
SDValue Glue;
// Build a sequence of copy-to-reg nodes, chained and glued together.
for (auto &Reg : RegsToPass) {
Chain = DAG.getCopyToReg(Chain, DL, Reg.first, Reg.second, Glue);
Glue = Chain.getValue(1);
}
// Validate that none of the argument registers have been marked as
// reserved, if so report an error. Do the same for the return address if this
// is not a tailcall.
validateCCReservedRegs(RegsToPass, MF);
if (!IsTailCall && MF.getSubtarget().isRegisterReservedByUser(RISCV::X1))
MF.getFunction().getContext().diagnose(DiagnosticInfoUnsupported{
MF.getFunction(),
"Return address register required, but has been reserved."});
// If the callee is a GlobalAddress/ExternalSymbol node, turn it into a
// TargetGlobalAddress/TargetExternalSymbol node so that legalize won't
// split it and then direct call can be matched by PseudoCALL.
bool CalleeIsLargeExternalSymbol = false;
if (getTargetMachine().getCodeModel() == CodeModel::Large) {
if (auto *S = dyn_cast<GlobalAddressSDNode>(Callee))
Callee = getLargeGlobalAddress(S, DL, PtrVT, DAG);
else if (auto *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
Callee = getLargeExternalSymbol(S, DL, PtrVT, DAG);
CalleeIsLargeExternalSymbol = true;
}
} else if (GlobalAddressSDNode *S = dyn_cast<GlobalAddressSDNode>(Callee)) {
const GlobalValue *GV = S->getGlobal();
Callee = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, RISCVII::MO_CALL);
} else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
Callee = DAG.getTargetExternalSymbol(S->getSymbol(), PtrVT, RISCVII::MO_CALL);
}
// The first call operand is the chain and the second is the target address.
SmallVector<SDValue, 8> Ops;
Ops.push_back(Chain);
Ops.push_back(Callee);
// Add argument registers to the end of the list so that they are
// known live into the call.
for (auto &Reg : RegsToPass)
Ops.push_back(DAG.getRegister(Reg.first, Reg.second.getValueType()));
// Add a register mask operand representing the call-preserved registers.
const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
const uint32_t *Mask = TRI->getCallPreservedMask(MF, CallConv);
assert(Mask && "Missing call preserved mask for calling convention");
Ops.push_back(DAG.getRegisterMask(Mask));
// Glue the call to the argument copies, if any.
if (Glue.getNode())
Ops.push_back(Glue);
assert((!CLI.CFIType || CLI.CB->isIndirectCall()) &&
"Unexpected CFI type for a direct call");
// Emit the call.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
// Use software guarded branch for large code model non-indirect calls
// Tail call to external symbol will have a null CLI.CB and we need another
// way to determine the callsite type
bool NeedSWGuarded = false;
if (getTargetMachine().getCodeModel() == CodeModel::Large &&
Subtarget.hasStdExtZicfilp() &&
((CLI.CB && !CLI.CB->isIndirectCall()) || CalleeIsLargeExternalSymbol))
NeedSWGuarded = true;
if (IsTailCall) {
MF.getFrameInfo().setHasTailCall();
unsigned CallOpc =
NeedSWGuarded ? RISCVISD::SW_GUARDED_TAIL : RISCVISD::TAIL;
SDValue Ret = DAG.getNode(CallOpc, DL, NodeTys, Ops);
if (CLI.CFIType)
Ret.getNode()->setCFIType(CLI.CFIType->getZExtValue());
DAG.addNoMergeSiteInfo(Ret.getNode(), CLI.NoMerge);
if (MF.getTarget().Options.EmitCallGraphSection && CB &&
CB->isIndirectCall())
DAG.addCallSiteInfo(Ret.getNode(), std::move(CSInfo));
return Ret;
}
unsigned CallOpc = NeedSWGuarded ? RISCVISD::SW_GUARDED_CALL : RISCVISD::CALL;
Chain = DAG.getNode(CallOpc, DL, NodeTys, Ops);
if (CLI.CFIType)
Chain.getNode()->setCFIType(CLI.CFIType->getZExtValue());
if (MF.getTarget().Options.EmitCallGraphSection && CB && CB->isIndirectCall())
DAG.addCallSiteInfo(Chain.getNode(), std::move(CSInfo));
DAG.addNoMergeSiteInfo(Chain.getNode(), CLI.NoMerge);
Glue = Chain.getValue(1);
// Mark the end of the call, which is glued to the call itself.
Chain = DAG.getCALLSEQ_END(Chain, NumBytes, 0, Glue, DL);
Glue = Chain.getValue(1);
// Assign locations to each value returned by this call.
SmallVector<CCValAssign, 16> RVLocs;
CCState RetCCInfo(CallConv, IsVarArg, MF, RVLocs, *DAG.getContext());
analyzeInputArgs(MF, RetCCInfo, Ins, /*IsRet=*/true, CC_RISCV);
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
auto &VA = RVLocs[i];
// Copy the value out
SDValue RetValue =
DAG.getCopyFromReg(Chain, DL, VA.getLocReg(), VA.getLocVT(), Glue);
// Glue the RetValue to the end of the call sequence
Chain = RetValue.getValue(1);
Glue = RetValue.getValue(2);
if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64) {
assert(VA.needsCustom());
SDValue RetValue2 = DAG.getCopyFromReg(Chain, DL, RVLocs[++i].getLocReg(),
MVT::i32, Glue);
Chain = RetValue2.getValue(1);
Glue = RetValue2.getValue(2);
RetValue = DAG.getNode(RISCVISD::BuildPairF64, DL, MVT::f64, RetValue,
RetValue2);
} else
RetValue = convertLocVTToValVT(DAG, RetValue, VA, DL, Subtarget);
InVals.push_back(RetValue);
}
return Chain;
}
bool RISCVTargetLowering::CanLowerReturn(
CallingConv::ID CallConv, MachineFunction &MF, bool IsVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs, LLVMContext &Context,
const Type *RetTy) const {
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, IsVarArg, MF, RVLocs, Context);
for (unsigned i = 0, e = Outs.size(); i != e; ++i) {
MVT VT = Outs[i].VT;
ISD::ArgFlagsTy ArgFlags = Outs[i].Flags;
if (CC_RISCV(i, VT, VT, CCValAssign::Full, ArgFlags, CCInfo,
/*IsRet=*/true, Outs[i].OrigTy))
return false;
}
return true;
}
SDValue
RISCVTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
bool IsVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SDLoc &DL, SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
const RISCVSubtarget &STI = MF.getSubtarget<RISCVSubtarget>();
// Stores the assignment of the return value to a location.
SmallVector<CCValAssign, 16> RVLocs;
// Info about the registers and stack slot.
CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), RVLocs,
*DAG.getContext());
analyzeOutputArgs(DAG.getMachineFunction(), CCInfo, Outs, /*IsRet=*/true,
nullptr, CC_RISCV);
if (CallConv == CallingConv::GHC && !RVLocs.empty())
reportFatalUsageError("GHC functions return void only");
SDValue Glue;
SmallVector<SDValue, 4> RetOps(1, Chain);
// Copy the result values into the output registers.
for (unsigned i = 0, e = RVLocs.size(), OutIdx = 0; i < e; ++i, ++OutIdx) {
SDValue Val = OutVals[OutIdx];
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
if (VA.getLocVT() == MVT::i32 && VA.getValVT() == MVT::f64) {
// Handle returning f64 on RV32D with a soft float ABI.
assert(VA.isRegLoc() && "Expected return via registers");
assert(VA.needsCustom());
SDValue SplitF64 = DAG.getNode(RISCVISD::SplitF64, DL,
DAG.getVTList(MVT::i32, MVT::i32), Val);
SDValue Lo = SplitF64.getValue(0);
SDValue Hi = SplitF64.getValue(1);
Register RegLo = VA.getLocReg();
Register RegHi = RVLocs[++i].getLocReg();
if (STI.isRegisterReservedByUser(RegLo) ||
STI.isRegisterReservedByUser(RegHi))
MF.getFunction().getContext().diagnose(DiagnosticInfoUnsupported{
MF.getFunction(),
"Return value register required, but has been reserved."});
Chain = DAG.getCopyToReg(Chain, DL, RegLo, Lo, Glue);
Glue = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(RegLo, MVT::i32));
Chain = DAG.getCopyToReg(Chain, DL, RegHi, Hi, Glue);
Glue = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(RegHi, MVT::i32));
} else {
// Handle a 'normal' return.
Val = convertValVTToLocVT(DAG, Val, VA, DL, Subtarget);
Chain = DAG.getCopyToReg(Chain, DL, VA.getLocReg(), Val, Glue);
if (STI.isRegisterReservedByUser(VA.getLocReg()))
MF.getFunction().getContext().diagnose(DiagnosticInfoUnsupported{
MF.getFunction(),
"Return value register required, but has been reserved."});
// Guarantee that all emitted copies are stuck together.
Glue = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
}
}
RetOps[0] = Chain; // Update chain.
// Add the glue node if we have it.
if (Glue.getNode()) {
RetOps.push_back(Glue);
}
if (any_of(RVLocs,
[](CCValAssign &VA) { return VA.getLocVT().isScalableVector(); }))
MF.getInfo<RISCVMachineFunctionInfo>()->setIsVectorCall();
unsigned RetOpc = RISCVISD::RET_GLUE;
// Interrupt service routines use different return instructions.
const Function &Func = DAG.getMachineFunction().getFunction();
if (Func.hasFnAttribute("interrupt")) {
if (!Func.getReturnType()->isVoidTy())
reportFatalUsageError(
"Functions with the interrupt attribute must have void return type!");
MachineFunction &MF = DAG.getMachineFunction();
StringRef Kind =
MF.getFunction().getFnAttribute("interrupt").getValueAsString();
if (Kind == "supervisor")
RetOpc = RISCVISD::SRET_GLUE;
else if (Kind == "rnmi") {
assert(STI.hasFeature(RISCV::FeatureStdExtSmrnmi) &&
"Need Smrnmi extension for rnmi");
RetOpc = RISCVISD::MNRET_GLUE;
} else if (Kind == "qci-nest" || Kind == "qci-nonest") {
assert(STI.hasFeature(RISCV::FeatureVendorXqciint) &&
"Need Xqciint for qci-(no)nest");
RetOpc = RISCVISD::QC_C_MILEAVERET_GLUE;
} else
RetOpc = RISCVISD::MRET_GLUE;
}
return DAG.getNode(RetOpc, DL, MVT::Other, RetOps);
}
void RISCVTargetLowering::validateCCReservedRegs(
const SmallVectorImpl<std::pair<llvm::Register, llvm::SDValue>> &Regs,
MachineFunction &MF) const {
const Function &F = MF.getFunction();
const RISCVSubtarget &STI = MF.getSubtarget<RISCVSubtarget>();
if (llvm::any_of(Regs, [&STI](auto Reg) {
return STI.isRegisterReservedByUser(Reg.first);
}))
F.getContext().diagnose(DiagnosticInfoUnsupported{
F, "Argument register required, but has been reserved."});
}
// Check if the result of the node is only used as a return value, as
// otherwise we can't perform a tail-call.
bool RISCVTargetLowering::isUsedByReturnOnly(SDNode *N, SDValue &Chain) const {
if (N->getNumValues() != 1)
return false;
if (!N->hasNUsesOfValue(1, 0))
return false;
SDNode *Copy = *N->user_begin();
if (Copy->getOpcode() == ISD::BITCAST) {
return isUsedByReturnOnly(Copy, Chain);
}
// TODO: Handle additional opcodes in order to support tail-calling libcalls
// with soft float ABIs.
if (Copy->getOpcode() != ISD::CopyToReg) {
return false;
}
// If the ISD::CopyToReg has a glue operand, we conservatively assume it
// isn't safe to perform a tail call.
if (Copy->getOperand(Copy->getNumOperands() - 1).getValueType() == MVT::Glue)
return false;
// The copy must be used by a RISCVISD::RET_GLUE, and nothing else.
bool HasRet = false;
for (SDNode *Node : Copy->users()) {
if (Node->getOpcode() != RISCVISD::RET_GLUE)
return false;
HasRet = true;
}
if (!HasRet)
return false;
Chain = Copy->getOperand(0);
return true;
}
bool RISCVTargetLowering::mayBeEmittedAsTailCall(const CallInst *CI) const {
return CI->isTailCall();
}
/// getConstraintType - Given a constraint letter, return the type of
/// constraint it is for this target.
RISCVTargetLowering::ConstraintType
RISCVTargetLowering::getConstraintType(StringRef Constraint) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
default:
break;
case 'f':
case 'R':
return C_RegisterClass;
case 'I':
case 'J':
case 'K':
return C_Immediate;
case 'A':
return C_Memory;
case 's':
case 'S': // A symbolic address
return C_Other;
}
} else {
if (Constraint == "vr" || Constraint == "vd" || Constraint == "vm")
return C_RegisterClass;
if (Constraint == "cr" || Constraint == "cR" || Constraint == "cf")
return C_RegisterClass;
}
return TargetLowering::getConstraintType(Constraint);
}
std::pair<unsigned, const TargetRegisterClass *>
RISCVTargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
StringRef Constraint,
MVT VT) const {
// First, see if this is a constraint that directly corresponds to a RISC-V
// register class.
if (Constraint.size() == 1) {
switch (Constraint[0]) {
case 'r':
// TODO: Support fixed vectors up to XLen for P extension?
if (VT.isVector())
break;
if (VT == MVT::f16 && Subtarget.hasStdExtZhinxmin())
return std::make_pair(0U, &RISCV::GPRF16NoX0RegClass);
if (VT == MVT::f32 && Subtarget.hasStdExtZfinx())
return std::make_pair(0U, &RISCV::GPRF32NoX0RegClass);
if (VT == MVT::f64 && Subtarget.hasStdExtZdinx() && !Subtarget.is64Bit())
return std::make_pair(0U, &RISCV::GPRPairNoX0RegClass);
return std::make_pair(0U, &RISCV::GPRNoX0RegClass);
case 'f':
if (VT == MVT::f16) {
if (Subtarget.hasStdExtZfhmin())
return std::make_pair(0U, &RISCV::FPR16RegClass);
if (Subtarget.hasStdExtZhinxmin())
return std::make_pair(0U, &RISCV::GPRF16NoX0RegClass);
} else if (VT == MVT::f32) {
if (Subtarget.hasStdExtF())
return std::make_pair(0U, &RISCV::FPR32RegClass);
if (Subtarget.hasStdExtZfinx())
return std::make_pair(0U, &RISCV::GPRF32NoX0RegClass);
} else if (VT == MVT::f64) {
if (Subtarget.hasStdExtD())
return std::make_pair(0U, &RISCV::FPR64RegClass);
if (Subtarget.hasStdExtZdinx() && !Subtarget.is64Bit())
return std::make_pair(0U, &RISCV::GPRPairNoX0RegClass);
if (Subtarget.hasStdExtZdinx() && Subtarget.is64Bit())
return std::make_pair(0U, &RISCV::GPRNoX0RegClass);
}
break;
case 'R':
if (((VT == MVT::i64 || VT == MVT::f64) && !Subtarget.is64Bit()) ||
(VT == MVT::i128 && Subtarget.is64Bit()))
return std::make_pair(0U, &RISCV::GPRPairNoX0RegClass);
break;
default:
break;
}
} else if (Constraint == "vr") {
for (const auto *RC :
{&RISCV::VRRegClass, &RISCV::VRM2RegClass, &RISCV::VRM4RegClass,
&RISCV::VRM8RegClass, &RISCV::VRN2M1RegClass, &RISCV::VRN3M1RegClass,
&RISCV::VRN4M1RegClass, &RISCV::VRN5M1RegClass,
&RISCV::VRN6M1RegClass, &RISCV::VRN7M1RegClass,
&RISCV::VRN8M1RegClass, &RISCV::VRN2M2RegClass,
&RISCV::VRN3M2RegClass, &RISCV::VRN4M2RegClass,
&RISCV::VRN2M4RegClass}) {
if (TRI->isTypeLegalForClass(*RC, VT.SimpleTy))
return std::make_pair(0U, RC);
if (VT.isFixedLengthVector() && useRVVForFixedLengthVectorVT(VT)) {
MVT ContainerVT = getContainerForFixedLengthVector(VT);
if (TRI->isTypeLegalForClass(*RC, ContainerVT))
return std::make_pair(0U, RC);
}
}
} else if (Constraint == "vd") {
for (const auto *RC :
{&RISCV::VRNoV0RegClass, &RISCV::VRM2NoV0RegClass,
&RISCV::VRM4NoV0RegClass, &RISCV::VRM8NoV0RegClass,
&RISCV::VRN2M1NoV0RegClass, &RISCV::VRN3M1NoV0RegClass,
&RISCV::VRN4M1NoV0RegClass, &RISCV::VRN5M1NoV0RegClass,
&RISCV::VRN6M1NoV0RegClass, &RISCV::VRN7M1NoV0RegClass,
&RISCV::VRN8M1NoV0RegClass, &RISCV::VRN2M2NoV0RegClass,
&RISCV::VRN3M2NoV0RegClass, &RISCV::VRN4M2NoV0RegClass,
&RISCV::VRN2M4NoV0RegClass}) {
if (TRI->isTypeLegalForClass(*RC, VT.SimpleTy))
return std::make_pair(0U, RC);
if (VT.isFixedLengthVector() && useRVVForFixedLengthVectorVT(VT)) {
MVT ContainerVT = getContainerForFixedLengthVector(VT);
if (TRI->isTypeLegalForClass(*RC, ContainerVT))
return std::make_pair(0U, RC);
}
}
} else if (Constraint == "vm") {
if (TRI->isTypeLegalForClass(RISCV::VMV0RegClass, VT.SimpleTy))
return std::make_pair(0U, &RISCV::VMV0RegClass);
if (VT.isFixedLengthVector() && useRVVForFixedLengthVectorVT(VT)) {
MVT ContainerVT = getContainerForFixedLengthVector(VT);
// VT here might be coerced to vector with i8 elements, so we need to
// check if this is a M1 register here instead of checking VMV0RegClass.
if (TRI->isTypeLegalForClass(RISCV::VRRegClass, ContainerVT))
return std::make_pair(0U, &RISCV::VMV0RegClass);
}
} else if (Constraint == "cr") {
if (VT == MVT::f16 && Subtarget.hasStdExtZhinxmin())
return std::make_pair(0U, &RISCV::GPRF16CRegClass);
if (VT == MVT::f32 && Subtarget.hasStdExtZfinx())
return std::make_pair(0U, &RISCV::GPRF32CRegClass);
if (VT == MVT::f64 && Subtarget.hasStdExtZdinx() && !Subtarget.is64Bit())
return std::make_pair(0U, &RISCV::GPRPairCRegClass);
if (!VT.isVector())
return std::make_pair(0U, &RISCV::GPRCRegClass);
} else if (Constraint == "cR") {
if (((VT == MVT::i64 || VT == MVT::f64) && !Subtarget.is64Bit()) ||
(VT == MVT::i128 && Subtarget.is64Bit()))
return std::make_pair(0U, &RISCV::GPRPairCRegClass);
} else if (Constraint == "cf") {
if (VT == MVT::f16) {
if (Subtarget.hasStdExtZfhmin())
return std::make_pair(0U, &RISCV::FPR16CRegClass);
if (Subtarget.hasStdExtZhinxmin())
return std::make_pair(0U, &RISCV::GPRF16CRegClass);
} else if (VT == MVT::f32) {
if (Subtarget.hasStdExtF())
return std::make_pair(0U, &RISCV::FPR32CRegClass);
if (Subtarget.hasStdExtZfinx())
return std::make_pair(0U, &RISCV::GPRF32CRegClass);
} else if (VT == MVT::f64) {
if (Subtarget.hasStdExtD())
return std::make_pair(0U, &RISCV::FPR64CRegClass);
if (Subtarget.hasStdExtZdinx() && !Subtarget.is64Bit())
return std::make_pair(0U, &RISCV::GPRPairCRegClass);
if (Subtarget.hasStdExtZdinx() && Subtarget.is64Bit())
return std::make_pair(0U, &RISCV::GPRCRegClass);
}
}
// Clang will correctly decode the usage of register name aliases into their
// official names. However, other frontends like `rustc` do not. This allows
// users of these frontends to use the ABI names for registers in LLVM-style
// register constraints.
unsigned XRegFromAlias = StringSwitch<unsigned>(Constraint.lower())
.Case("{zero}", RISCV::X0)
.Case("{ra}", RISCV::X1)
.Case("{sp}", RISCV::X2)
.Case("{gp}", RISCV::X3)
.Case("{tp}", RISCV::X4)
.Case("{t0}", RISCV::X5)
.Case("{t1}", RISCV::X6)
.Case("{t2}", RISCV::X7)
.Cases("{s0}", "{fp}", RISCV::X8)
.Case("{s1}", RISCV::X9)
.Case("{a0}", RISCV::X10)
.Case("{a1}", RISCV::X11)
.Case("{a2}", RISCV::X12)
.Case("{a3}", RISCV::X13)
.Case("{a4}", RISCV::X14)
.Case("{a5}", RISCV::X15)
.Case("{a6}", RISCV::X16)
.Case("{a7}", RISCV::X17)
.Case("{s2}", RISCV::X18)
.Case("{s3}", RISCV::X19)
.Case("{s4}", RISCV::X20)
.Case("{s5}", RISCV::X21)
.Case("{s6}", RISCV::X22)
.Case("{s7}", RISCV::X23)
.Case("{s8}", RISCV::X24)
.Case("{s9}", RISCV::X25)
.Case("{s10}", RISCV::X26)
.Case("{s11}", RISCV::X27)
.Case("{t3}", RISCV::X28)
.Case("{t4}", RISCV::X29)
.Case("{t5}", RISCV::X30)
.Case("{t6}", RISCV::X31)
.Default(RISCV::NoRegister);
if (XRegFromAlias != RISCV::NoRegister)
return std::make_pair(XRegFromAlias, &RISCV::GPRRegClass);
// Since TargetLowering::getRegForInlineAsmConstraint uses the name of the
// TableGen record rather than the AsmName to choose registers for InlineAsm
// constraints, plus we want to match those names to the widest floating point
// register type available, manually select floating point registers here.
//
// The second case is the ABI name of the register, so that frontends can also
// use the ABI names in register constraint lists.
if (Subtarget.hasStdExtF()) {
unsigned FReg = StringSwitch<unsigned>(Constraint.lower())
.Cases("{f0}", "{ft0}", RISCV::F0_F)
.Cases("{f1}", "{ft1}", RISCV::F1_F)
.Cases("{f2}", "{ft2}", RISCV::F2_F)
.Cases("{f3}", "{ft3}", RISCV::F3_F)
.Cases("{f4}", "{ft4}", RISCV::F4_F)
.Cases("{f5}", "{ft5}", RISCV::F5_F)
.Cases("{f6}", "{ft6}", RISCV::F6_F)
.Cases("{f7}", "{ft7}", RISCV::F7_F)
.Cases("{f8}", "{fs0}", RISCV::F8_F)
.Cases("{f9}", "{fs1}", RISCV::F9_F)
.Cases("{f10}", "{fa0}", RISCV::F10_F)
.Cases("{f11}", "{fa1}", RISCV::F11_F)
.Cases("{f12}", "{fa2}", RISCV::F12_F)
.Cases("{f13}", "{fa3}", RISCV::F13_F)
.Cases("{f14}", "{fa4}", RISCV::F14_F)
.Cases("{f15}", "{fa5}", RISCV::F15_F)
.Cases("{f16}", "{fa6}", RISCV::F16_F)
.Cases("{f17}", "{fa7}", RISCV::F17_F)
.Cases("{f18}", "{fs2}", RISCV::F18_F)
.Cases("{f19}", "{fs3}", RISCV::F19_F)
.Cases("{f20}", "{fs4}", RISCV::F20_F)
.Cases("{f21}", "{fs5}", RISCV::F21_F)
.Cases("{f22}", "{fs6}", RISCV::F22_F)
.Cases("{f23}", "{fs7}", RISCV::F23_F)
.Cases("{f24}", "{fs8}", RISCV::F24_F)
.Cases("{f25}", "{fs9}", RISCV::F25_F)
.Cases("{f26}", "{fs10}", RISCV::F26_F)
.Cases("{f27}", "{fs11}", RISCV::F27_F)
.Cases("{f28}", "{ft8}", RISCV::F28_F)
.Cases("{f29}", "{ft9}", RISCV::F29_F)
.Cases("{f30}", "{ft10}", RISCV::F30_F)
.Cases("{f31}", "{ft11}", RISCV::F31_F)
.Default(RISCV::NoRegister);
if (FReg != RISCV::NoRegister) {
assert(RISCV::F0_F <= FReg && FReg <= RISCV::F31_F && "Unknown fp-reg");
if (Subtarget.hasStdExtD() && (VT == MVT::f64 || VT == MVT::Other)) {
unsigned RegNo = FReg - RISCV::F0_F;
unsigned DReg = RISCV::F0_D + RegNo;
return std::make_pair(DReg, &RISCV::FPR64RegClass);
}
if (VT == MVT::f32 || VT == MVT::Other)
return std::make_pair(FReg, &RISCV::FPR32RegClass);
if (Subtarget.hasStdExtZfhmin() && VT == MVT::f16) {
unsigned RegNo = FReg - RISCV::F0_F;
unsigned HReg = RISCV::F0_H + RegNo;
return std::make_pair(HReg, &RISCV::FPR16RegClass);
}
}
}
if (Subtarget.hasVInstructions()) {
Register VReg = StringSwitch<Register>(Constraint.lower())
.Case("{v0}", RISCV::V0)
.Case("{v1}", RISCV::V1)
.Case("{v2}", RISCV::V2)
.Case("{v3}", RISCV::V3)
.Case("{v4}", RISCV::V4)
.Case("{v5}", RISCV::V5)
.Case("{v6}", RISCV::V6)
.Case("{v7}", RISCV::V7)
.Case("{v8}", RISCV::V8)
.Case("{v9}", RISCV::V9)
.Case("{v10}", RISCV::V10)
.Case("{v11}", RISCV::V11)
.Case("{v12}", RISCV::V12)
.Case("{v13}", RISCV::V13)
.Case("{v14}", RISCV::V14)
.Case("{v15}", RISCV::V15)
.Case("{v16}", RISCV::V16)
.Case("{v17}", RISCV::V17)
.Case("{v18}", RISCV::V18)
.Case("{v19}", RISCV::V19)
.Case("{v20}", RISCV::V20)
.Case("{v21}", RISCV::V21)
.Case("{v22}", RISCV::V22)
.Case("{v23}", RISCV::V23)
.Case("{v24}", RISCV::V24)
.Case("{v25}", RISCV::V25)
.Case("{v26}", RISCV::V26)
.Case("{v27}", RISCV::V27)
.Case("{v28}", RISCV::V28)
.Case("{v29}", RISCV::V29)
.Case("{v30}", RISCV::V30)
.Case("{v31}", RISCV::V31)
.Default(RISCV::NoRegister);
if (VReg != RISCV::NoRegister) {
if (TRI->isTypeLegalForClass(RISCV::VMRegClass, VT.SimpleTy))
return std::make_pair(VReg, &RISCV::VMRegClass);
if (TRI->isTypeLegalForClass(RISCV::VRRegClass, VT.SimpleTy))
return std::make_pair(VReg, &RISCV::VRRegClass);
for (const auto *RC :
{&RISCV::VRM2RegClass, &RISCV::VRM4RegClass, &RISCV::VRM8RegClass}) {
if (TRI->isTypeLegalForClass(*RC, VT.SimpleTy)) {
VReg = TRI->getMatchingSuperReg(VReg, RISCV::sub_vrm1_0, RC);
return std::make_pair(VReg, RC);
}
}
}
}
std::pair<Register, const TargetRegisterClass *> Res =
TargetLowering::getRegForInlineAsmConstraint(TRI, Constraint, VT);
// If we picked one of the Zfinx register classes, remap it to the GPR class.
// FIXME: When Zfinx is supported in CodeGen this will need to take the
// Subtarget into account.
if (Res.second == &RISCV::GPRF16RegClass ||
Res.second == &RISCV::GPRF32RegClass ||
Res.second == &RISCV::GPRPairRegClass)
return std::make_pair(Res.first, &RISCV::GPRRegClass);
return Res;
}
InlineAsm::ConstraintCode
RISCVTargetLowering::getInlineAsmMemConstraint(StringRef ConstraintCode) const {
// Currently only support length 1 constraints.
if (ConstraintCode.size() == 1) {
switch (ConstraintCode[0]) {
case 'A':
return InlineAsm::ConstraintCode::A;
default:
break;
}
}
return TargetLowering::getInlineAsmMemConstraint(ConstraintCode);
}
void RISCVTargetLowering::LowerAsmOperandForConstraint(
SDValue Op, StringRef Constraint, std::vector<SDValue> &Ops,
SelectionDAG &DAG) const {
// Currently only support length 1 constraints.
if (Constraint.size() == 1) {
switch (Constraint[0]) {
case 'I':
// Validate & create a 12-bit signed immediate operand.
if (auto *C = dyn_cast<ConstantSDNode>(Op)) {
uint64_t CVal = C->getSExtValue();
if (isInt<12>(CVal))
Ops.push_back(DAG.getSignedTargetConstant(CVal, SDLoc(Op),
Subtarget.getXLenVT()));
}
return;
case 'J':
// Validate & create an integer zero operand.
if (isNullConstant(Op))
Ops.push_back(
DAG.getTargetConstant(0, SDLoc(Op), Subtarget.getXLenVT()));
return;
case 'K':
// Validate & create a 5-bit unsigned immediate operand.
if (auto *C = dyn_cast<ConstantSDNode>(Op)) {
uint64_t CVal = C->getZExtValue();
if (isUInt<5>(CVal))
Ops.push_back(
DAG.getTargetConstant(CVal, SDLoc(Op), Subtarget.getXLenVT()));
}
return;
case 'S':
TargetLowering::LowerAsmOperandForConstraint(Op, "s", Ops, DAG);
return;
default:
break;
}
}
TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, Ops, DAG);
}
Instruction *RISCVTargetLowering::emitLeadingFence(IRBuilderBase &Builder,
Instruction *Inst,
AtomicOrdering Ord) const {
if (Subtarget.hasStdExtZtso()) {
if (isa<LoadInst>(Inst) && Ord == AtomicOrdering::SequentiallyConsistent)
return Builder.CreateFence(Ord);
return nullptr;
}
if (isa<LoadInst>(Inst) && Ord == AtomicOrdering::SequentiallyConsistent)
return Builder.CreateFence(Ord);
if (isa<StoreInst>(Inst) && isReleaseOrStronger(Ord))
return Builder.CreateFence(AtomicOrdering::Release);
return nullptr;
}
Instruction *RISCVTargetLowering::emitTrailingFence(IRBuilderBase &Builder,
Instruction *Inst,
AtomicOrdering Ord) const {
if (Subtarget.hasStdExtZtso()) {
if (isa<StoreInst>(Inst) && Ord == AtomicOrdering::SequentiallyConsistent)
return Builder.CreateFence(Ord);
return nullptr;
}
if (isa<LoadInst>(Inst) && isAcquireOrStronger(Ord))
return Builder.CreateFence(AtomicOrdering::Acquire);
if (Subtarget.enableTrailingSeqCstFence() && isa<StoreInst>(Inst) &&
Ord == AtomicOrdering::SequentiallyConsistent)
return Builder.CreateFence(AtomicOrdering::SequentiallyConsistent);
return nullptr;
}
TargetLowering::AtomicExpansionKind
RISCVTargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const {
// atomicrmw {fadd,fsub} must be expanded to use compare-exchange, as floating
// point operations can't be used in an lr/sc sequence without breaking the
// forward-progress guarantee.
if (AI->isFloatingPointOperation() ||
AI->getOperation() == AtomicRMWInst::UIncWrap ||
AI->getOperation() == AtomicRMWInst::UDecWrap ||
AI->getOperation() == AtomicRMWInst::USubCond ||
AI->getOperation() == AtomicRMWInst::USubSat)
return AtomicExpansionKind::CmpXChg;
// Don't expand forced atomics, we want to have __sync libcalls instead.
if (Subtarget.hasForcedAtomics())
return AtomicExpansionKind::None;
unsigned Size = AI->getType()->getPrimitiveSizeInBits();
if (AI->getOperation() == AtomicRMWInst::Nand) {
if (Subtarget.hasStdExtZacas() &&
(Size >= 32 || Subtarget.hasStdExtZabha()))
return AtomicExpansionKind::CmpXChg;
if (Size < 32)
return AtomicExpansionKind::MaskedIntrinsic;
}
if (Size < 32 && !Subtarget.hasStdExtZabha())
return AtomicExpansionKind::MaskedIntrinsic;
return AtomicExpansionKind::None;
}
static Intrinsic::ID
getIntrinsicForMaskedAtomicRMWBinOp(unsigned XLen, AtomicRMWInst::BinOp BinOp) {
switch (BinOp) {
default:
llvm_unreachable("Unexpected AtomicRMW BinOp");
case AtomicRMWInst::Xchg:
return Intrinsic::riscv_masked_atomicrmw_xchg;
case AtomicRMWInst::Add:
return Intrinsic::riscv_masked_atomicrmw_add;
case AtomicRMWInst::Sub:
return Intrinsic::riscv_masked_atomicrmw_sub;
case AtomicRMWInst::Nand:
return Intrinsic::riscv_masked_atomicrmw_nand;
case AtomicRMWInst::Max:
return Intrinsic::riscv_masked_atomicrmw_max;
case AtomicRMWInst::Min:
return Intrinsic::riscv_masked_atomicrmw_min;
case AtomicRMWInst::UMax:
return Intrinsic::riscv_masked_atomicrmw_umax;
case AtomicRMWInst::UMin:
return Intrinsic::riscv_masked_atomicrmw_umin;
}
}
Value *RISCVTargetLowering::emitMaskedAtomicRMWIntrinsic(
IRBuilderBase &Builder, AtomicRMWInst *AI, Value *AlignedAddr, Value *Incr,
Value *Mask, Value *ShiftAmt, AtomicOrdering Ord) const {
// In the case of an atomicrmw xchg with a constant 0/-1 operand, replace
// the atomic instruction with an AtomicRMWInst::And/Or with appropriate
// mask, as this produces better code than the LR/SC loop emitted by
// int_riscv_masked_atomicrmw_xchg.
if (AI->getOperation() == AtomicRMWInst::Xchg &&
isa<ConstantInt>(AI->getValOperand())) {
ConstantInt *CVal = cast<ConstantInt>(AI->getValOperand());
if (CVal->isZero())
return Builder.CreateAtomicRMW(AtomicRMWInst::And, AlignedAddr,
Builder.CreateNot(Mask, "Inv_Mask"),
AI->getAlign(), Ord);
if (CVal->isMinusOne())
return Builder.CreateAtomicRMW(AtomicRMWInst::Or, AlignedAddr, Mask,
AI->getAlign(), Ord);
}
unsigned XLen = Subtarget.getXLen();
Value *Ordering =
Builder.getIntN(XLen, static_cast<uint64_t>(AI->getOrdering()));
Type *Tys[] = {Builder.getIntNTy(XLen), AlignedAddr->getType()};
Function *LrwOpScwLoop = Intrinsic::getOrInsertDeclaration(
AI->getModule(),
getIntrinsicForMaskedAtomicRMWBinOp(XLen, AI->getOperation()), Tys);
if (XLen == 64) {
Incr = Builder.CreateSExt(Incr, Builder.getInt64Ty());
Mask = Builder.CreateSExt(Mask, Builder.getInt64Ty());
ShiftAmt = Builder.CreateSExt(ShiftAmt, Builder.getInt64Ty());
}
Value *Result;
// Must pass the shift amount needed to sign extend the loaded value prior
// to performing a signed comparison for min/max. ShiftAmt is the number of
// bits to shift the value into position. Pass XLen-ShiftAmt-ValWidth, which
// is the number of bits to left+right shift the value in order to
// sign-extend.
if (AI->getOperation() == AtomicRMWInst::Min ||
AI->getOperation() == AtomicRMWInst::Max) {
const DataLayout &DL = AI->getDataLayout();
unsigned ValWidth =
DL.getTypeStoreSizeInBits(AI->getValOperand()->getType());
Value *SextShamt =
Builder.CreateSub(Builder.getIntN(XLen, XLen - ValWidth), ShiftAmt);
Result = Builder.CreateCall(LrwOpScwLoop,
{AlignedAddr, Incr, Mask, SextShamt, Ordering});
} else {
Result =
Builder.CreateCall(LrwOpScwLoop, {AlignedAddr, Incr, Mask, Ordering});
}
if (XLen == 64)
Result = Builder.CreateTrunc(Result, Builder.getInt32Ty());
return Result;
}
TargetLowering::AtomicExpansionKind
RISCVTargetLowering::shouldExpandAtomicCmpXchgInIR(
AtomicCmpXchgInst *CI) const {
// Don't expand forced atomics, we want to have __sync libcalls instead.
if (Subtarget.hasForcedAtomics())
return AtomicExpansionKind::None;
unsigned Size = CI->getCompareOperand()->getType()->getPrimitiveSizeInBits();
if (!(Subtarget.hasStdExtZabha() && Subtarget.hasStdExtZacas()) &&
(Size == 8 || Size == 16))
return AtomicExpansionKind::MaskedIntrinsic;
return AtomicExpansionKind::None;
}
Value *RISCVTargetLowering::emitMaskedAtomicCmpXchgIntrinsic(
IRBuilderBase &Builder, AtomicCmpXchgInst *CI, Value *AlignedAddr,
Value *CmpVal, Value *NewVal, Value *Mask, AtomicOrdering Ord) const {
unsigned XLen = Subtarget.getXLen();
Value *Ordering = Builder.getIntN(XLen, static_cast<uint64_t>(Ord));
Intrinsic::ID CmpXchgIntrID = Intrinsic::riscv_masked_cmpxchg;
if (XLen == 64) {
CmpVal = Builder.CreateSExt(CmpVal, Builder.getInt64Ty());
NewVal = Builder.CreateSExt(NewVal, Builder.getInt64Ty());
Mask = Builder.CreateSExt(Mask, Builder.getInt64Ty());
}
Type *Tys[] = {Builder.getIntNTy(XLen), AlignedAddr->getType()};
Value *Result = Builder.CreateIntrinsic(
CmpXchgIntrID, Tys, {AlignedAddr, CmpVal, NewVal, Mask, Ordering});
if (XLen == 64)
Result = Builder.CreateTrunc(Result, Builder.getInt32Ty());
return Result;
}
bool RISCVTargetLowering::shouldRemoveExtendFromGSIndex(SDValue Extend,
EVT DataVT) const {
// We have indexed loads for all supported EEW types. Indices are always
// zero extended.
return Extend.getOpcode() == ISD::ZERO_EXTEND &&
isTypeLegal(Extend.getValueType()) &&
isTypeLegal(Extend.getOperand(0).getValueType()) &&
Extend.getOperand(0).getValueType().getVectorElementType() != MVT::i1;
}
bool RISCVTargetLowering::shouldConvertFpToSat(unsigned Op, EVT FPVT,
EVT VT) const {
if (!isOperationLegalOrCustom(Op, VT) || !FPVT.isSimple())
return false;
switch (FPVT.getSimpleVT().SimpleTy) {
case MVT::f16:
return Subtarget.hasStdExtZfhmin();
case MVT::f32:
return Subtarget.hasStdExtF();
case MVT::f64:
return Subtarget.hasStdExtD();
default:
return false;
}
}
unsigned RISCVTargetLowering::getJumpTableEncoding() const {
// If we are using the small code model, we can reduce size of jump table
// entry to 4 bytes.
if (Subtarget.is64Bit() && !isPositionIndependent() &&
getTargetMachine().getCodeModel() == CodeModel::Small) {
return MachineJumpTableInfo::EK_Custom32;
}
return TargetLowering::getJumpTableEncoding();
}
const MCExpr *RISCVTargetLowering::LowerCustomJumpTableEntry(
const MachineJumpTableInfo *MJTI, const MachineBasicBlock *MBB,
unsigned uid, MCContext &Ctx) const {
assert(Subtarget.is64Bit() && !isPositionIndependent() &&
getTargetMachine().getCodeModel() == CodeModel::Small);
return MCSymbolRefExpr::create(MBB->getSymbol(), Ctx);
}
bool RISCVTargetLowering::isVScaleKnownToBeAPowerOfTwo() const {
// We define vscale to be VLEN/RVVBitsPerBlock. VLEN is always a power
// of two >= 64, and RVVBitsPerBlock is 64. Thus, vscale must be
// a power of two as well.
// FIXME: This doesn't work for zve32, but that's already broken
// elsewhere for the same reason.
assert(Subtarget.getRealMinVLen() >= 64 && "zve32* unsupported");
static_assert(RISCV::RVVBitsPerBlock == 64,
"RVVBitsPerBlock changed, audit needed");
return true;
}
bool RISCVTargetLowering::getIndexedAddressParts(SDNode *Op, SDValue &Base,
SDValue &Offset,
ISD::MemIndexedMode &AM,
SelectionDAG &DAG) const {
// Target does not support indexed loads.
if (!Subtarget.hasVendorXTHeadMemIdx())
return false;
if (Op->getOpcode() != ISD::ADD && Op->getOpcode() != ISD::SUB)
return false;
Base = Op->getOperand(0);
if (ConstantSDNode *RHS = dyn_cast<ConstantSDNode>(Op->getOperand(1))) {
int64_t RHSC = RHS->getSExtValue();
if (Op->getOpcode() == ISD::SUB)
RHSC = -(uint64_t)RHSC;
// The constants that can be encoded in the THeadMemIdx instructions
// are of the form (sign_extend(imm5) << imm2).
bool isLegalIndexedOffset = false;
for (unsigned i = 0; i < 4; i++)
if (isInt<5>(RHSC >> i) && ((RHSC % (1LL << i)) == 0)) {
isLegalIndexedOffset = true;
break;
}
if (!isLegalIndexedOffset)
return false;
Offset = Op->getOperand(1);
return true;
}
return false;
}
bool RISCVTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
SDValue &Offset,
ISD::MemIndexedMode &AM,
SelectionDAG &DAG) const {
EVT VT;
SDValue Ptr;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
VT = LD->getMemoryVT();
Ptr = LD->getBasePtr();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
VT = ST->getMemoryVT();
Ptr = ST->getBasePtr();
} else
return false;
if (!getIndexedAddressParts(Ptr.getNode(), Base, Offset, AM, DAG))
return false;
AM = ISD::PRE_INC;
return true;
}
bool RISCVTargetLowering::getPostIndexedAddressParts(SDNode *N, SDNode *Op,
SDValue &Base,
SDValue &Offset,
ISD::MemIndexedMode &AM,
SelectionDAG &DAG) const {
if (Subtarget.hasVendorXCVmem() && !Subtarget.is64Bit()) {
if (Op->getOpcode() != ISD::ADD)
return false;
if (LSBaseSDNode *LS = dyn_cast<LSBaseSDNode>(N))
Base = LS->getBasePtr();
else
return false;
if (Base == Op->getOperand(0))
Offset = Op->getOperand(1);
else if (Base == Op->getOperand(1))
Offset = Op->getOperand(0);
else
return false;
AM = ISD::POST_INC;
return true;
}
EVT VT;
SDValue Ptr;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
VT = LD->getMemoryVT();
Ptr = LD->getBasePtr();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
VT = ST->getMemoryVT();
Ptr = ST->getBasePtr();
} else
return false;
if (!getIndexedAddressParts(Op, Base, Offset, AM, DAG))
return false;
// Post-indexing updates the base, so it's not a valid transform
// if that's not the same as the load's pointer.
if (Ptr != Base)
return false;
AM = ISD::POST_INC;
return true;
}
bool RISCVTargetLowering::isFMAFasterThanFMulAndFAdd(const MachineFunction &MF,
EVT VT) const {
EVT SVT = VT.getScalarType();
if (!SVT.isSimple())
return false;
switch (SVT.getSimpleVT().SimpleTy) {
case MVT::f16:
return VT.isVector() ? Subtarget.hasVInstructionsF16()
: Subtarget.hasStdExtZfhOrZhinx();
case MVT::f32:
return Subtarget.hasStdExtFOrZfinx();
case MVT::f64:
return Subtarget.hasStdExtDOrZdinx();
default:
break;
}
return false;
}
ISD::NodeType RISCVTargetLowering::getExtendForAtomicCmpSwapArg() const {
// Zacas will use amocas.w which does not require extension.
return Subtarget.hasStdExtZacas() ? ISD::ANY_EXTEND : ISD::SIGN_EXTEND;
}
Register RISCVTargetLowering::getExceptionPointerRegister(
const Constant *PersonalityFn) const {
return RISCV::X10;
}
Register RISCVTargetLowering::getExceptionSelectorRegister(
const Constant *PersonalityFn) const {
return RISCV::X11;
}
bool RISCVTargetLowering::shouldExtendTypeInLibCall(EVT Type) const {
// Return false to suppress the unnecessary extensions if the LibCall
// arguments or return value is a float narrower than XLEN on a soft FP ABI.
if (Subtarget.isSoftFPABI() && (Type.isFloatingPoint() && !Type.isVector() &&
Type.getSizeInBits() < Subtarget.getXLen()))
return false;
return true;
}
bool RISCVTargetLowering::shouldSignExtendTypeInLibCall(Type *Ty,
bool IsSigned) const {
if (Subtarget.is64Bit() && Ty->isIntegerTy(32))
return true;
return IsSigned;
}
bool RISCVTargetLowering::decomposeMulByConstant(LLVMContext &Context, EVT VT,
SDValue C) const {
// Check integral scalar types.
if (!VT.isScalarInteger())
return false;
// Omit the optimization if the sub target has the M extension and the data
// size exceeds XLen.
const bool HasZmmul = Subtarget.hasStdExtZmmul();
if (HasZmmul && VT.getSizeInBits() > Subtarget.getXLen())
return false;
auto *ConstNode = cast<ConstantSDNode>(C);
const APInt &Imm = ConstNode->getAPIntValue();
// Don't do this if the Xqciac extension is enabled and the Imm in simm12.
if (Subtarget.hasVendorXqciac() && Imm.isSignedIntN(12))
return false;
// Break the MUL to a SLLI and an ADD/SUB.
if ((Imm + 1).isPowerOf2() || (Imm - 1).isPowerOf2() ||
(1 - Imm).isPowerOf2() || (-1 - Imm).isPowerOf2())
return true;
// Optimize the MUL to (SH*ADD x, (SLLI x, bits)) if Imm is not simm12.
if (Subtarget.hasStdExtZba() && !Imm.isSignedIntN(12) &&
((Imm - 2).isPowerOf2() || (Imm - 4).isPowerOf2() ||
(Imm - 8).isPowerOf2()))
return true;
// Break the MUL to two SLLI instructions and an ADD/SUB, if Imm needs
// a pair of LUI/ADDI.
if (!Imm.isSignedIntN(12) && Imm.countr_zero() < 12 &&
ConstNode->hasOneUse()) {
APInt ImmS = Imm.ashr(Imm.countr_zero());
if ((ImmS + 1).isPowerOf2() || (ImmS - 1).isPowerOf2() ||
(1 - ImmS).isPowerOf2())
return true;
}
return false;
}
bool RISCVTargetLowering::isMulAddWithConstProfitable(SDValue AddNode,
SDValue ConstNode) const {
// Let the DAGCombiner decide for vectors.
EVT VT = AddNode.getValueType();
if (VT.isVector())
return true;
// Let the DAGCombiner decide for larger types.
if (VT.getScalarSizeInBits() > Subtarget.getXLen())
return true;
// It is worse if c1 is simm12 while c1*c2 is not.
ConstantSDNode *C1Node = cast<ConstantSDNode>(AddNode.getOperand(1));
ConstantSDNode *C2Node = cast<ConstantSDNode>(ConstNode);
const APInt &C1 = C1Node->getAPIntValue();
const APInt &C2 = C2Node->getAPIntValue();
if (C1.isSignedIntN(12) && !(C1 * C2).isSignedIntN(12))
return false;
// Default to true and let the DAGCombiner decide.
return true;
}
bool RISCVTargetLowering::allowsMisalignedMemoryAccesses(
EVT VT, unsigned AddrSpace, Align Alignment, MachineMemOperand::Flags Flags,
unsigned *Fast) const {
if (!VT.isVector()) {
if (Fast)
*Fast = Subtarget.enableUnalignedScalarMem();
return Subtarget.enableUnalignedScalarMem();
}
// All vector implementations must support element alignment
EVT ElemVT = VT.getVectorElementType();
if (Alignment >= ElemVT.getStoreSize()) {
if (Fast)
*Fast = 1;
return true;
}
// Note: We lower an unmasked unaligned vector access to an equally sized
// e8 element type access. Given this, we effectively support all unmasked
// misaligned accesses. TODO: Work through the codegen implications of
// allowing such accesses to be formed, and considered fast.
if (Fast)
*Fast = Subtarget.enableUnalignedVectorMem();
return Subtarget.enableUnalignedVectorMem();
}
EVT RISCVTargetLowering::getOptimalMemOpType(
LLVMContext &Context, const MemOp &Op,
const AttributeList &FuncAttributes) const {
if (!Subtarget.hasVInstructions())
return MVT::Other;
if (FuncAttributes.hasFnAttr(Attribute::NoImplicitFloat))
return MVT::Other;
// We use LMUL1 memory operations here for a non-obvious reason. Our caller
// has an expansion threshold, and we want the number of hardware memory
// operations to correspond roughly to that threshold. LMUL>1 operations
// are typically expanded linearly internally, and thus correspond to more
// than one actual memory operation. Note that store merging and load
// combining will typically form larger LMUL operations from the LMUL1
// operations emitted here, and that's okay because combining isn't
// introducing new memory operations; it's just merging existing ones.
// NOTE: We limit to 1024 bytes to avoid creating an invalid MVT.
const unsigned MinVLenInBytes =
std::min(Subtarget.getRealMinVLen() / 8, 1024U);
if (Op.size() < MinVLenInBytes)
// TODO: Figure out short memops. For the moment, do the default thing
// which ends up using scalar sequences.
return MVT::Other;
// If the minimum VLEN is less than RISCV::RVVBitsPerBlock we don't support
// fixed vectors.
if (MinVLenInBytes <= RISCV::RVVBytesPerBlock)
return MVT::Other;
// Prefer i8 for non-zero memset as it allows us to avoid materializing
// a large scalar constant and instead use vmv.v.x/i to do the
// broadcast. For everything else, prefer ELenVT to minimize VL and thus
// maximize the chance we can encode the size in the vsetvli.
MVT ELenVT = MVT::getIntegerVT(Subtarget.getELen());
MVT PreferredVT = (Op.isMemset() && !Op.isZeroMemset()) ? MVT::i8 : ELenVT;
// Do we have sufficient alignment for our preferred VT? If not, revert
// to largest size allowed by our alignment criteria.
if (PreferredVT != MVT::i8 && !Subtarget.enableUnalignedVectorMem()) {
Align RequiredAlign(PreferredVT.getStoreSize());
if (Op.isFixedDstAlign())
RequiredAlign = std::min(RequiredAlign, Op.getDstAlign());
if (Op.isMemcpy())
RequiredAlign = std::min(RequiredAlign, Op.getSrcAlign());
PreferredVT = MVT::getIntegerVT(RequiredAlign.value() * 8);
}
return MVT::getVectorVT(PreferredVT, MinVLenInBytes/PreferredVT.getStoreSize());
}
bool RISCVTargetLowering::splitValueIntoRegisterParts(
SelectionDAG &DAG, const SDLoc &DL, SDValue Val, SDValue *Parts,
unsigned NumParts, MVT PartVT, std::optional<CallingConv::ID> CC) const {
bool IsABIRegCopy = CC.has_value();
EVT ValueVT = Val.getValueType();
MVT PairVT = Subtarget.is64Bit() ? MVT::i128 : MVT::i64;
if ((ValueVT == PairVT ||
(!Subtarget.is64Bit() && Subtarget.hasStdExtZdinx() &&
ValueVT == MVT::f64)) &&
NumParts == 1 && PartVT == MVT::Untyped) {
// Pairs in Inline Assembly, f64 in Inline assembly on rv32_zdinx
MVT XLenVT = Subtarget.getXLenVT();
if (ValueVT == MVT::f64)
Val = DAG.getBitcast(MVT::i64, Val);
auto [Lo, Hi] = DAG.SplitScalar(Val, DL, XLenVT, XLenVT);
// Always creating an MVT::Untyped part, so always use
// RISCVISD::BuildGPRPair.
Parts[0] = DAG.getNode(RISCVISD::BuildGPRPair, DL, PartVT, Lo, Hi);
return true;
}
if (IsABIRegCopy && (ValueVT == MVT::f16 || ValueVT == MVT::bf16) &&
PartVT == MVT::f32) {
// Cast the [b]f16 to i16, extend to i32, pad with ones to make a float
// nan, and cast to f32.
Val = DAG.getNode(ISD::BITCAST, DL, MVT::i16, Val);
Val = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i32, Val);
Val = DAG.getNode(ISD::OR, DL, MVT::i32, Val,
DAG.getConstant(0xFFFF0000, DL, MVT::i32));
Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
Parts[0] = Val;
return true;
}
if (ValueVT.isRISCVVectorTuple() && PartVT.isRISCVVectorTuple()) {
#ifndef NDEBUG
unsigned ValNF = ValueVT.getRISCVVectorTupleNumFields();
[[maybe_unused]] unsigned ValLMUL =
divideCeil(ValueVT.getSizeInBits().getKnownMinValue(),
ValNF * RISCV::RVVBitsPerBlock);
unsigned PartNF = PartVT.getRISCVVectorTupleNumFields();
[[maybe_unused]] unsigned PartLMUL =
divideCeil(PartVT.getSizeInBits().getKnownMinValue(),
PartNF * RISCV::RVVBitsPerBlock);
assert(ValNF == PartNF && ValLMUL == PartLMUL &&
"RISC-V vector tuple type only accepts same register class type "
"TUPLE_INSERT");
#endif
Val = DAG.getNode(RISCVISD::TUPLE_INSERT, DL, PartVT, DAG.getUNDEF(PartVT),
Val, DAG.getTargetConstant(0, DL, MVT::i32));
Parts[0] = Val;
return true;
}
if ((ValueVT.isScalableVector() || ValueVT.isFixedLengthVector()) &&
PartVT.isScalableVector()) {
if (ValueVT.isFixedLengthVector()) {
ValueVT = getContainerForFixedLengthVector(ValueVT.getSimpleVT());
Val = convertToScalableVector(ValueVT, Val, DAG, Subtarget);
}
LLVMContext &Context = *DAG.getContext();
EVT ValueEltVT = ValueVT.getVectorElementType();
EVT PartEltVT = PartVT.getVectorElementType();
unsigned ValueVTBitSize = ValueVT.getSizeInBits().getKnownMinValue();
unsigned PartVTBitSize = PartVT.getSizeInBits().getKnownMinValue();
if (PartVTBitSize % ValueVTBitSize == 0) {
assert(PartVTBitSize >= ValueVTBitSize);
// If the element types are different, bitcast to the same element type of
// PartVT first.
// Give an example here, we want copy a <vscale x 1 x i8> value to
// <vscale x 4 x i16>.
// We need to convert <vscale x 1 x i8> to <vscale x 8 x i8> by insert
// subvector, then we can bitcast to <vscale x 4 x i16>.
if (ValueEltVT != PartEltVT) {
if (PartVTBitSize > ValueVTBitSize) {
unsigned Count = PartVTBitSize / ValueEltVT.getFixedSizeInBits();
assert(Count != 0 && "The number of element should not be zero.");
EVT SameEltTypeVT =
EVT::getVectorVT(Context, ValueEltVT, Count, /*IsScalable=*/true);
Val = DAG.getInsertSubvector(DL, DAG.getUNDEF(SameEltTypeVT), Val, 0);
}
Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
} else {
Val = DAG.getInsertSubvector(DL, DAG.getUNDEF(PartVT), Val, 0);
}
Parts[0] = Val;
return true;
}
}
return false;
}
SDValue RISCVTargetLowering::joinRegisterPartsIntoValue(
SelectionDAG &DAG, const SDLoc &DL, const SDValue *Parts, unsigned NumParts,
MVT PartVT, EVT ValueVT, std::optional<CallingConv::ID> CC) const {
bool IsABIRegCopy = CC.has_value();
MVT PairVT = Subtarget.is64Bit() ? MVT::i128 : MVT::i64;
if ((ValueVT == PairVT ||
(!Subtarget.is64Bit() && Subtarget.hasStdExtZdinx() &&
ValueVT == MVT::f64)) &&
NumParts == 1 && PartVT == MVT::Untyped) {
// Pairs in Inline Assembly, f64 in Inline assembly on rv32_zdinx
MVT XLenVT = Subtarget.getXLenVT();
SDValue Val = Parts[0];
// Always starting with an MVT::Untyped part, so always use
// RISCVISD::SplitGPRPair
Val = DAG.getNode(RISCVISD::SplitGPRPair, DL, DAG.getVTList(XLenVT, XLenVT),
Val);
Val = DAG.getNode(ISD::BUILD_PAIR, DL, PairVT, Val.getValue(0),
Val.getValue(1));
if (ValueVT == MVT::f64)
Val = DAG.getBitcast(ValueVT, Val);
return Val;
}
if (IsABIRegCopy && (ValueVT == MVT::f16 || ValueVT == MVT::bf16) &&
PartVT == MVT::f32) {
SDValue Val = Parts[0];
// Cast the f32 to i32, truncate to i16, and cast back to [b]f16.
Val = DAG.getNode(ISD::BITCAST, DL, MVT::i32, Val);
Val = DAG.getNode(ISD::TRUNCATE, DL, MVT::i16, Val);
Val = DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);
return Val;
}
if ((ValueVT.isScalableVector() || ValueVT.isFixedLengthVector()) &&
PartVT.isScalableVector()) {
LLVMContext &Context = *DAG.getContext();
SDValue Val = Parts[0];
EVT ValueEltVT = ValueVT.getVectorElementType();
EVT PartEltVT = PartVT.getVectorElementType();
unsigned ValueVTBitSize = ValueVT.getSizeInBits().getKnownMinValue();
if (ValueVT.isFixedLengthVector())
ValueVTBitSize = getContainerForFixedLengthVector(ValueVT.getSimpleVT())
.getSizeInBits()
.getKnownMinValue();
unsigned PartVTBitSize = PartVT.getSizeInBits().getKnownMinValue();
if (PartVTBitSize % ValueVTBitSize == 0) {
assert(PartVTBitSize >= ValueVTBitSize);
EVT SameEltTypeVT = ValueVT;
// If the element types are different, convert it to the same element type
// of PartVT.
// Give an example here, we want copy a <vscale x 1 x i8> value from
// <vscale x 4 x i16>.
// We need to convert <vscale x 4 x i16> to <vscale x 8 x i8> first,
// then we can extract <vscale x 1 x i8>.
if (ValueEltVT != PartEltVT) {
unsigned Count = PartVTBitSize / ValueEltVT.getFixedSizeInBits();
assert(Count != 0 && "The number of element should not be zero.");
SameEltTypeVT =
EVT::getVectorVT(Context, ValueEltVT, Count, /*IsScalable=*/true);
Val = DAG.getNode(ISD::BITCAST, DL, SameEltTypeVT, Val);
}
if (ValueVT.isFixedLengthVector())
Val = convertFromScalableVector(ValueVT, Val, DAG, Subtarget);
else
Val = DAG.getExtractSubvector(DL, ValueVT, Val, 0);
return Val;
}
}
return SDValue();
}
bool RISCVTargetLowering::isIntDivCheap(EVT VT, AttributeList Attr) const {
// When aggressively optimizing for code size, we prefer to use a div
// instruction, as it is usually smaller than the alternative sequence.
// TODO: Add vector division?
bool OptSize = Attr.hasFnAttr(Attribute::MinSize);
return OptSize && !VT.isVector();
}
bool RISCVTargetLowering::preferScalarizeSplat(SDNode *N) const {
// Scalarize zero_ext and sign_ext might stop match to widening instruction in
// some situation.
unsigned Opc = N->getOpcode();
if (Opc == ISD::ZERO_EXTEND || Opc == ISD::SIGN_EXTEND)
return false;
return true;
}
static Value *useTpOffset(IRBuilderBase &IRB, unsigned Offset) {
Module *M = IRB.GetInsertBlock()->getModule();
Function *ThreadPointerFunc = Intrinsic::getOrInsertDeclaration(
M, Intrinsic::thread_pointer, IRB.getPtrTy());
return IRB.CreateConstGEP1_32(IRB.getInt8Ty(),
IRB.CreateCall(ThreadPointerFunc), Offset);
}
Value *RISCVTargetLowering::getIRStackGuard(IRBuilderBase &IRB) const {
// Fuchsia provides a fixed TLS slot for the stack cookie.
// <zircon/tls.h> defines ZX_TLS_STACK_GUARD_OFFSET with this value.
if (Subtarget.isTargetFuchsia())
return useTpOffset(IRB, -0x10);
// Android provides a fixed TLS slot for the stack cookie. See the definition
// of TLS_SLOT_STACK_GUARD in
// https://android.googlesource.com/platform/bionic/+/main/libc/platform/bionic/tls_defines.h
if (Subtarget.isTargetAndroid())
return useTpOffset(IRB, -0x18);
Module *M = IRB.GetInsertBlock()->getModule();
if (M->getStackProtectorGuard() == "tls") {
// Users must specify the offset explicitly
int Offset = M->getStackProtectorGuardOffset();
return useTpOffset(IRB, Offset);
}
return TargetLowering::getIRStackGuard(IRB);
}
bool RISCVTargetLowering::isLegalStridedLoadStore(EVT DataType,
Align Alignment) const {
if (!Subtarget.hasVInstructions())
return false;
// Only support fixed vectors if we know the minimum vector size.
if (DataType.isFixedLengthVector() && !Subtarget.useRVVForFixedLengthVectors())
return false;
EVT ScalarType = DataType.getScalarType();
if (!isLegalElementTypeForRVV(ScalarType))
return false;
if (!Subtarget.enableUnalignedVectorMem() &&
Alignment < ScalarType.getStoreSize())
return false;
return true;
}
MachineInstr *
RISCVTargetLowering::EmitKCFICheck(MachineBasicBlock &MBB,
MachineBasicBlock::instr_iterator &MBBI,
const TargetInstrInfo *TII) const {
assert(MBBI->isCall() && MBBI->getCFIType() &&
"Invalid call instruction for a KCFI check");
assert(is_contained({RISCV::PseudoCALLIndirect, RISCV::PseudoTAILIndirect},
MBBI->getOpcode()));
MachineOperand &Target = MBBI->getOperand(0);
Target.setIsRenamable(false);
return BuildMI(MBB, MBBI, MBBI->getDebugLoc(), TII->get(RISCV::KCFI_CHECK))
.addReg(Target.getReg())
.addImm(MBBI->getCFIType())
.getInstr();
}
#define GET_REGISTER_MATCHER
#include "RISCVGenAsmMatcher.inc"
Register
RISCVTargetLowering::getRegisterByName(const char *RegName, LLT VT,
const MachineFunction &MF) const {
Register Reg = MatchRegisterAltName(RegName);
if (!Reg)
Reg = MatchRegisterName(RegName);
if (!Reg)
return Reg;
BitVector ReservedRegs = Subtarget.getRegisterInfo()->getReservedRegs(MF);
if (!ReservedRegs.test(Reg) && !Subtarget.isRegisterReservedByUser(Reg))
reportFatalUsageError(Twine("Trying to obtain non-reserved register \"" +
StringRef(RegName) + "\"."));
return Reg;
}
MachineMemOperand::Flags
RISCVTargetLowering::getTargetMMOFlags(const Instruction &I) const {
const MDNode *NontemporalInfo = I.getMetadata(LLVMContext::MD_nontemporal);
if (NontemporalInfo == nullptr)
return MachineMemOperand::MONone;
// 1 for default value work as __RISCV_NTLH_ALL
// 2 -> __RISCV_NTLH_INNERMOST_PRIVATE
// 3 -> __RISCV_NTLH_ALL_PRIVATE
// 4 -> __RISCV_NTLH_INNERMOST_SHARED
// 5 -> __RISCV_NTLH_ALL
int NontemporalLevel = 5;
const MDNode *RISCVNontemporalInfo =
I.getMetadata("riscv-nontemporal-domain");
if (RISCVNontemporalInfo != nullptr)
NontemporalLevel =
cast<ConstantInt>(
cast<ConstantAsMetadata>(RISCVNontemporalInfo->getOperand(0))
->getValue())
->getZExtValue();
assert((1 <= NontemporalLevel && NontemporalLevel <= 5) &&
"RISC-V target doesn't support this non-temporal domain.");
NontemporalLevel -= 2;
MachineMemOperand::Flags Flags = MachineMemOperand::MONone;
if (NontemporalLevel & 0b1)
Flags |= MONontemporalBit0;
if (NontemporalLevel & 0b10)
Flags |= MONontemporalBit1;
return Flags;
}
MachineMemOperand::Flags
RISCVTargetLowering::getTargetMMOFlags(const MemSDNode &Node) const {
MachineMemOperand::Flags NodeFlags = Node.getMemOperand()->getFlags();
MachineMemOperand::Flags TargetFlags = MachineMemOperand::MONone;
TargetFlags |= (NodeFlags & MONontemporalBit0);
TargetFlags |= (NodeFlags & MONontemporalBit1);
return TargetFlags;
}
bool RISCVTargetLowering::areTwoSDNodeTargetMMOFlagsMergeable(
const MemSDNode &NodeX, const MemSDNode &NodeY) const {
return getTargetMMOFlags(NodeX) == getTargetMMOFlags(NodeY);
}
bool RISCVTargetLowering::isCtpopFast(EVT VT) const {
if (VT.isScalableVector())
return isTypeLegal(VT) && Subtarget.hasStdExtZvbb();
if (VT.isFixedLengthVector() && Subtarget.hasStdExtZvbb())
return true;
return Subtarget.hasStdExtZbb() &&
(VT == MVT::i32 || VT == MVT::i64 || VT.isFixedLengthVector());
}
unsigned RISCVTargetLowering::getCustomCtpopCost(EVT VT,
ISD::CondCode Cond) const {
return isCtpopFast(VT) ? 0 : 1;
}
bool RISCVTargetLowering::shouldInsertFencesForAtomic(
const Instruction *I) const {
if (Subtarget.hasStdExtZalasr()) {
if (Subtarget.hasStdExtZtso()) {
// Zalasr + TSO means that atomic_load_acquire and atomic_store_release
// should be lowered to plain load/store. The easiest way to do this is
// to say we should insert fences for them, and the fence insertion code
// will just not insert any fences
auto *LI = dyn_cast<LoadInst>(I);
auto *SI = dyn_cast<StoreInst>(I);
if ((LI &&
(LI->getOrdering() == AtomicOrdering::SequentiallyConsistent)) ||
(SI &&
(SI->getOrdering() == AtomicOrdering::SequentiallyConsistent))) {
// Here, this is a load or store which is seq_cst, and needs a .aq or
// .rl therefore we shouldn't try to insert fences
return false;
}
// Here, we are a TSO inst that isn't a seq_cst load/store
return isa<LoadInst>(I) || isa<StoreInst>(I);
}
return false;
}
// Note that one specific case requires fence insertion for an
// AtomicCmpXchgInst but is handled via the RISCVZacasABIFix pass rather
// than this hook due to limitations in the interface here.
return isa<LoadInst>(I) || isa<StoreInst>(I);
}
bool RISCVTargetLowering::fallBackToDAGISel(const Instruction &Inst) const {
// GISel support is in progress or complete for these opcodes.
unsigned Op = Inst.getOpcode();
if (Op == Instruction::Add || Op == Instruction::Sub ||
Op == Instruction::And || Op == Instruction::Or ||
Op == Instruction::Xor || Op == Instruction::InsertElement ||
Op == Instruction::ShuffleVector || Op == Instruction::Load ||
Op == Instruction::Freeze || Op == Instruction::Store)
return false;
if (Inst.getType()->isScalableTy())
return true;
for (unsigned i = 0; i < Inst.getNumOperands(); ++i)
if (Inst.getOperand(i)->getType()->isScalableTy() &&
!isa<ReturnInst>(&Inst))
return true;
if (const AllocaInst *AI = dyn_cast<AllocaInst>(&Inst)) {
if (AI->getAllocatedType()->isScalableTy())
return true;
}
return false;
}
SDValue
RISCVTargetLowering::BuildSDIVPow2(SDNode *N, const APInt &Divisor,
SelectionDAG &DAG,
SmallVectorImpl<SDNode *> &Created) const {
AttributeList Attr = DAG.getMachineFunction().getFunction().getAttributes();
if (isIntDivCheap(N->getValueType(0), Attr))
return SDValue(N, 0); // Lower SDIV as SDIV
// Only perform this transform if short forward branch opt is supported.
if (!Subtarget.hasShortForwardBranchOpt())
return SDValue();
EVT VT = N->getValueType(0);
if (!(VT == MVT::i32 || (VT == MVT::i64 && Subtarget.is64Bit())))
return SDValue();
// Ensure 2**k-1 < 2048 so that we can just emit a single addi/addiw.
if (Divisor.sgt(2048) || Divisor.slt(-2048))
return SDValue();
return TargetLowering::buildSDIVPow2WithCMov(N, Divisor, DAG, Created);
}
bool RISCVTargetLowering::shouldFoldSelectWithSingleBitTest(
EVT VT, const APInt &AndMask) const {
if (Subtarget.hasCZEROLike())
return !Subtarget.hasStdExtZbs() && AndMask.ugt(1024);
return TargetLowering::shouldFoldSelectWithSingleBitTest(VT, AndMask);
}
unsigned RISCVTargetLowering::getMinimumJumpTableEntries() const {
return Subtarget.getMinimumJumpTableEntries();
}
SDValue RISCVTargetLowering::expandIndirectJTBranch(const SDLoc &dl,
SDValue Value, SDValue Addr,
int JTI,
SelectionDAG &DAG) const {
if (Subtarget.hasStdExtZicfilp()) {
// When Zicfilp enabled, we need to use software guarded branch for jump
// table branch.
SDValue Chain = Value;
// Jump table debug info is only needed if CodeView is enabled.
if (DAG.getTarget().getTargetTriple().isOSBinFormatCOFF())
Chain = DAG.getJumpTableDebugInfo(JTI, Chain, dl);
return DAG.getNode(RISCVISD::SW_GUARDED_BRIND, dl, MVT::Other, Chain, Addr);
}
return TargetLowering::expandIndirectJTBranch(dl, Value, Addr, JTI, DAG);
}
// If an output pattern produces multiple instructions tablegen may pick an
// arbitrary type from an instructions destination register class to use for the
// VT of that MachineSDNode. This VT may be used to look up the representative
// register class. If the type isn't legal, the default implementation will
// not find a register class.
//
// Some integer types smaller than XLen are listed in the GPR register class to
// support isel patterns for GISel, but are not legal in SelectionDAG. The
// arbitrary type tablegen picks may be one of these smaller types.
//
// f16 and bf16 are both valid for the FPR16 or GPRF16 register class. It's
// possible for tablegen to pick bf16 as the arbitrary type for an f16 pattern.
std::pair<const TargetRegisterClass *, uint8_t>
RISCVTargetLowering::findRepresentativeClass(const TargetRegisterInfo *TRI,
MVT VT) const {
switch (VT.SimpleTy) {
default:
break;
case MVT::i8:
case MVT::i16:
case MVT::i32:
return TargetLowering::findRepresentativeClass(TRI, Subtarget.getXLenVT());
case MVT::bf16:
case MVT::f16:
return TargetLowering::findRepresentativeClass(TRI, MVT::f32);
}
return TargetLowering::findRepresentativeClass(TRI, VT);
}
namespace llvm::RISCVVIntrinsicsTable {
#define GET_RISCVVIntrinsicsTable_IMPL
#include "RISCVGenSearchableTables.inc"
} // namespace llvm::RISCVVIntrinsicsTable
bool RISCVTargetLowering::hasInlineStackProbe(const MachineFunction &MF) const {
// If the function specifically requests inline stack probes, emit them.
if (MF.getFunction().hasFnAttribute("probe-stack"))
return MF.getFunction().getFnAttribute("probe-stack").getValueAsString() ==
"inline-asm";
return false;
}
unsigned RISCVTargetLowering::getStackProbeSize(const MachineFunction &MF,
Align StackAlign) const {
// The default stack probe size is 4096 if the function has no
// stack-probe-size attribute.
const Function &Fn = MF.getFunction();
unsigned StackProbeSize =
Fn.getFnAttributeAsParsedInteger("stack-probe-size", 4096);
// Round down to the stack alignment.
StackProbeSize = alignDown(StackProbeSize, StackAlign.value());
return StackProbeSize ? StackProbeSize : StackAlign.value();
}
SDValue RISCVTargetLowering::lowerDYNAMIC_STACKALLOC(SDValue Op,
SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
if (!hasInlineStackProbe(MF))
return SDValue();
MVT XLenVT = Subtarget.getXLenVT();
// Get the inputs.
SDValue Chain = Op.getOperand(0);
SDValue Size = Op.getOperand(1);
MaybeAlign Align =
cast<ConstantSDNode>(Op.getOperand(2))->getMaybeAlignValue();
SDLoc dl(Op);
EVT VT = Op.getValueType();
// Construct the new SP value in a GPR.
SDValue SP = DAG.getCopyFromReg(Chain, dl, RISCV::X2, XLenVT);
Chain = SP.getValue(1);
SP = DAG.getNode(ISD::SUB, dl, XLenVT, SP, Size);
if (Align)
SP = DAG.getNode(ISD::AND, dl, VT, SP.getValue(0),
DAG.getSignedConstant(-Align->value(), dl, VT));
// Set the real SP to the new value with a probing loop.
Chain = DAG.getNode(RISCVISD::PROBED_ALLOCA, dl, MVT::Other, Chain, SP);
return DAG.getMergeValues({SP, Chain}, dl);
}
MachineBasicBlock *
RISCVTargetLowering::emitDynamicProbedAlloc(MachineInstr &MI,
MachineBasicBlock *MBB) const {
MachineFunction &MF = *MBB->getParent();
MachineBasicBlock::iterator MBBI = MI.getIterator();
DebugLoc DL = MBB->findDebugLoc(MBBI);
Register TargetReg = MI.getOperand(0).getReg();
const RISCVInstrInfo *TII = Subtarget.getInstrInfo();
bool IsRV64 = Subtarget.is64Bit();
Align StackAlign = Subtarget.getFrameLowering()->getStackAlign();
const RISCVTargetLowering *TLI = Subtarget.getTargetLowering();
uint64_t ProbeSize = TLI->getStackProbeSize(MF, StackAlign);
MachineFunction::iterator MBBInsertPoint = std::next(MBB->getIterator());
MachineBasicBlock *LoopTestMBB =
MF.CreateMachineBasicBlock(MBB->getBasicBlock());
MF.insert(MBBInsertPoint, LoopTestMBB);
MachineBasicBlock *ExitMBB = MF.CreateMachineBasicBlock(MBB->getBasicBlock());
MF.insert(MBBInsertPoint, ExitMBB);
Register SPReg = RISCV::X2;
Register ScratchReg =
MF.getRegInfo().createVirtualRegister(&RISCV::GPRRegClass);
// ScratchReg = ProbeSize
TII->movImm(*MBB, MBBI, DL, ScratchReg, ProbeSize, MachineInstr::NoFlags);
// LoopTest:
// SUB SP, SP, ProbeSize
BuildMI(*LoopTestMBB, LoopTestMBB->end(), DL, TII->get(RISCV::SUB), SPReg)
.addReg(SPReg)
.addReg(ScratchReg);
// s[d|w] zero, 0(sp)
BuildMI(*LoopTestMBB, LoopTestMBB->end(), DL,
TII->get(IsRV64 ? RISCV::SD : RISCV::SW))
.addReg(RISCV::X0)
.addReg(SPReg)
.addImm(0);
// BLT TargetReg, SP, LoopTest
BuildMI(*LoopTestMBB, LoopTestMBB->end(), DL, TII->get(RISCV::BLT))
.addReg(TargetReg)
.addReg(SPReg)
.addMBB(LoopTestMBB);
// Adjust with: MV SP, TargetReg.
BuildMI(*ExitMBB, ExitMBB->end(), DL, TII->get(RISCV::ADDI), SPReg)
.addReg(TargetReg)
.addImm(0);
ExitMBB->splice(ExitMBB->end(), MBB, std::next(MBBI), MBB->end());
ExitMBB->transferSuccessorsAndUpdatePHIs(MBB);
LoopTestMBB->addSuccessor(ExitMBB);
LoopTestMBB->addSuccessor(LoopTestMBB);
MBB->addSuccessor(LoopTestMBB);
MI.eraseFromParent();
MF.getInfo<RISCVMachineFunctionInfo>()->setDynamicAllocation();
return ExitMBB->begin()->getParent();
}
ArrayRef<MCPhysReg> RISCVTargetLowering::getRoundingControlRegisters() const {
if (Subtarget.hasStdExtFOrZfinx()) {
static const MCPhysReg RCRegs[] = {RISCV::FRM, RISCV::FFLAGS};
return RCRegs;
}
return {};
}