blob: e1af02829c1da6fde2df5eccc613daad189c71a4 [file] [log] [blame]
//===-- UnrollLoopRuntime.cpp - Runtime Loop unrolling utilities ----------===//
//
// 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 implements some loop unrolling utilities for loops with run-time
// trip counts. See LoopUnroll.cpp for unrolling loops with compile-time
// trip counts.
//
// The functions in this file are used to generate extra code when the
// run-time trip count modulo the unroll factor is not 0. When this is the
// case, we need to generate code to execute these 'left over' iterations.
//
// The current strategy generates an if-then-else sequence prior to the
// unrolled loop to execute the 'left over' iterations before or after the
// unrolled loop.
//
//===----------------------------------------------------------------------===//
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/DomTreeUpdater.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/LoopIterator.h"
#include "llvm/Analysis/ScalarEvolution.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/MDBuilder.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/ProfDataUtils.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Cloning.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/LoopUtils.h"
#include "llvm/Transforms/Utils/ScalarEvolutionExpander.h"
#include "llvm/Transforms/Utils/UnrollLoop.h"
#include <algorithm>
using namespace llvm;
#define DEBUG_TYPE "loop-unroll"
STATISTIC(NumRuntimeUnrolled,
"Number of loops unrolled with run-time trip counts");
static cl::opt<bool> UnrollRuntimeMultiExit(
"unroll-runtime-multi-exit", cl::init(false), cl::Hidden,
cl::desc("Allow runtime unrolling for loops with multiple exits, when "
"epilog is generated"));
static cl::opt<bool> UnrollRuntimeOtherExitPredictable(
"unroll-runtime-other-exit-predictable", cl::init(false), cl::Hidden,
cl::desc("Assume the non latch exit block to be predictable"));
// Probability that the loop trip count is so small that after the prolog
// we do not enter the unrolled loop at all.
// It is unlikely that the loop trip count is smaller than the unroll factor;
// other than that, the choice of constant is not tuned yet.
static const uint32_t UnrolledLoopHeaderWeights[] = {1, 127};
// Probability that the loop trip count is so small that we skip the unrolled
// loop completely and immediately enter the epilogue loop.
// It is unlikely that the loop trip count is smaller than the unroll factor;
// other than that, the choice of constant is not tuned yet.
static const uint32_t EpilogHeaderWeights[] = {1, 127};
/// Connect the unrolling prolog code to the original loop.
/// The unrolling prolog code contains code to execute the
/// 'extra' iterations if the run-time trip count modulo the
/// unroll count is non-zero.
///
/// This function performs the following:
/// - Create PHI nodes at prolog end block to combine values
/// that exit the prolog code and jump around the prolog.
/// - Add a PHI operand to a PHI node at the loop exit block
/// for values that exit the prolog and go around the loop.
/// - Branch around the original loop if the trip count is less
/// than the unroll factor.
///
static void ConnectProlog(Loop *L, Value *BECount, unsigned Count,
BasicBlock *PrologExit,
BasicBlock *OriginalLoopLatchExit,
BasicBlock *PreHeader, BasicBlock *NewPreHeader,
ValueToValueMapTy &VMap, DominatorTree *DT,
LoopInfo *LI, bool PreserveLCSSA,
ScalarEvolution &SE) {
// Loop structure should be the following:
// Preheader
// PrologHeader
// ...
// PrologLatch
// PrologExit
// NewPreheader
// Header
// ...
// Latch
// LatchExit
BasicBlock *Latch = L->getLoopLatch();
assert(Latch && "Loop must have a latch");
BasicBlock *PrologLatch = cast<BasicBlock>(VMap[Latch]);
// Create a PHI node for each outgoing value from the original loop
// (which means it is an outgoing value from the prolog code too).
// The new PHI node is inserted in the prolog end basic block.
// The new PHI node value is added as an operand of a PHI node in either
// the loop header or the loop exit block.
for (BasicBlock *Succ : successors(Latch)) {
for (PHINode &PN : Succ->phis()) {
// Add a new PHI node to the prolog end block and add the
// appropriate incoming values.
// TODO: This code assumes that the PrologExit (or the LatchExit block for
// prolog loop) contains only one predecessor from the loop, i.e. the
// PrologLatch. When supporting multiple-exiting block loops, we can have
// two or more blocks that have the LatchExit as the target in the
// original loop.
PHINode *NewPN = PHINode::Create(PN.getType(), 2, PN.getName() + ".unr");
NewPN->insertBefore(PrologExit->getFirstNonPHIIt());
// Adding a value to the new PHI node from the original loop preheader.
// This is the value that skips all the prolog code.
if (L->contains(&PN)) {
// Succ is loop header.
NewPN->addIncoming(PN.getIncomingValueForBlock(NewPreHeader),
PreHeader);
} else {
// Succ is LatchExit.
NewPN->addIncoming(UndefValue::get(PN.getType()), PreHeader);
}
Value *V = PN.getIncomingValueForBlock(Latch);
if (Instruction *I = dyn_cast<Instruction>(V)) {
if (L->contains(I)) {
V = VMap.lookup(I);
}
}
// Adding a value to the new PHI node from the last prolog block
// that was created.
NewPN->addIncoming(V, PrologLatch);
// Update the existing PHI node operand with the value from the
// new PHI node. How this is done depends on if the existing
// PHI node is in the original loop block, or the exit block.
if (L->contains(&PN))
PN.setIncomingValueForBlock(NewPreHeader, NewPN);
else
PN.addIncoming(NewPN, PrologExit);
SE.forgetValue(&PN);
}
}
// Make sure that created prolog loop is in simplified form
SmallVector<BasicBlock *, 4> PrologExitPreds;
Loop *PrologLoop = LI->getLoopFor(PrologLatch);
if (PrologLoop) {
for (BasicBlock *PredBB : predecessors(PrologExit))
if (PrologLoop->contains(PredBB))
PrologExitPreds.push_back(PredBB);
SplitBlockPredecessors(PrologExit, PrologExitPreds, ".unr-lcssa", DT, LI,
nullptr, PreserveLCSSA);
}
// Create a branch around the original loop, which is taken if there are no
// iterations remaining to be executed after running the prologue.
Instruction *InsertPt = PrologExit->getTerminator();
IRBuilder<> B(InsertPt);
assert(Count != 0 && "nonsensical Count!");
// If BECount <u (Count - 1) then (BECount + 1) % Count == (BECount + 1)
// This means %xtraiter is (BECount + 1) and all of the iterations of this
// loop were executed by the prologue. Note that if BECount <u (Count - 1)
// then (BECount + 1) cannot unsigned-overflow.
Value *BrLoopExit =
B.CreateICmpULT(BECount, ConstantInt::get(BECount->getType(), Count - 1));
// Split the exit to maintain loop canonicalization guarantees
SmallVector<BasicBlock *, 4> Preds(predecessors(OriginalLoopLatchExit));
SplitBlockPredecessors(OriginalLoopLatchExit, Preds, ".unr-lcssa", DT, LI,
nullptr, PreserveLCSSA);
// Add the branch to the exit block (around the unrolled loop)
MDNode *BranchWeights = nullptr;
if (hasBranchWeightMD(*Latch->getTerminator())) {
// Assume loop is nearly always entered.
MDBuilder MDB(B.getContext());
BranchWeights = MDB.createBranchWeights(UnrolledLoopHeaderWeights);
}
B.CreateCondBr(BrLoopExit, OriginalLoopLatchExit, NewPreHeader,
BranchWeights);
InsertPt->eraseFromParent();
if (DT) {
auto *NewDom = DT->findNearestCommonDominator(OriginalLoopLatchExit,
PrologExit);
DT->changeImmediateDominator(OriginalLoopLatchExit, NewDom);
}
}
/// Connect the unrolling epilog code to the original loop.
/// The unrolling epilog code contains code to execute the
/// 'extra' iterations if the run-time trip count modulo the
/// unroll count is non-zero.
///
/// This function performs the following:
/// - Update PHI nodes at the unrolling loop exit and epilog loop exit
/// - Create PHI nodes at the unrolling loop exit to combine
/// values that exit the unrolling loop code and jump around it.
/// - Update PHI operands in the epilog loop by the new PHI nodes
/// - Branch around the epilog loop if extra iters (ModVal) is zero.
///
static void ConnectEpilog(Loop *L, Value *ModVal, BasicBlock *NewExit,
BasicBlock *Exit, BasicBlock *PreHeader,
BasicBlock *EpilogPreHeader, BasicBlock *NewPreHeader,
ValueToValueMapTy &VMap, DominatorTree *DT,
LoopInfo *LI, bool PreserveLCSSA, ScalarEvolution &SE,
unsigned Count) {
BasicBlock *Latch = L->getLoopLatch();
assert(Latch && "Loop must have a latch");
BasicBlock *EpilogLatch = cast<BasicBlock>(VMap[Latch]);
// Loop structure should be the following:
//
// PreHeader
// NewPreHeader
// Header
// ...
// Latch
// NewExit (PN)
// EpilogPreHeader
// EpilogHeader
// ...
// EpilogLatch
// Exit (EpilogPN)
// Update PHI nodes at NewExit and Exit.
for (PHINode &PN : NewExit->phis()) {
// PN should be used in another PHI located in Exit block as
// Exit was split by SplitBlockPredecessors into Exit and NewExit
// Basically it should look like:
// NewExit:
// PN = PHI [I, Latch]
// ...
// Exit:
// EpilogPN = PHI [PN, EpilogPreHeader], [X, Exit2], [Y, Exit2.epil]
//
// Exits from non-latch blocks point to the original exit block and the
// epilogue edges have already been added.
//
// There is EpilogPreHeader incoming block instead of NewExit as
// NewExit was spilt 1 more time to get EpilogPreHeader.
assert(PN.hasOneUse() && "The phi should have 1 use");
PHINode *EpilogPN = cast<PHINode>(PN.use_begin()->getUser());
assert(EpilogPN->getParent() == Exit && "EpilogPN should be in Exit block");
// Add incoming PreHeader from branch around the Loop
PN.addIncoming(UndefValue::get(PN.getType()), PreHeader);
SE.forgetValue(&PN);
Value *V = PN.getIncomingValueForBlock(Latch);
Instruction *I = dyn_cast<Instruction>(V);
if (I && L->contains(I))
// If value comes from an instruction in the loop add VMap value.
V = VMap.lookup(I);
// For the instruction out of the loop, constant or undefined value
// insert value itself.
EpilogPN->addIncoming(V, EpilogLatch);
assert(EpilogPN->getBasicBlockIndex(EpilogPreHeader) >= 0 &&
"EpilogPN should have EpilogPreHeader incoming block");
// Change EpilogPreHeader incoming block to NewExit.
EpilogPN->setIncomingBlock(EpilogPN->getBasicBlockIndex(EpilogPreHeader),
NewExit);
// Now PHIs should look like:
// NewExit:
// PN = PHI [I, Latch], [undef, PreHeader]
// ...
// Exit:
// EpilogPN = PHI [PN, NewExit], [VMap[I], EpilogLatch]
}
// Create PHI nodes at NewExit (from the unrolling loop Latch and PreHeader).
// Update corresponding PHI nodes in epilog loop.
for (BasicBlock *Succ : successors(Latch)) {
// Skip this as we already updated phis in exit blocks.
if (!L->contains(Succ))
continue;
for (PHINode &PN : Succ->phis()) {
// Add new PHI nodes to the loop exit block and update epilog
// PHIs with the new PHI values.
PHINode *NewPN = PHINode::Create(PN.getType(), 2, PN.getName() + ".unr");
NewPN->insertBefore(NewExit->getFirstNonPHIIt());
// Adding a value to the new PHI node from the unrolling loop preheader.
NewPN->addIncoming(PN.getIncomingValueForBlock(NewPreHeader), PreHeader);
// Adding a value to the new PHI node from the unrolling loop latch.
NewPN->addIncoming(PN.getIncomingValueForBlock(Latch), Latch);
// Update the existing PHI node operand with the value from the new PHI
// node. Corresponding instruction in epilog loop should be PHI.
PHINode *VPN = cast<PHINode>(VMap[&PN]);
VPN->setIncomingValueForBlock(EpilogPreHeader, NewPN);
}
}
Instruction *InsertPt = NewExit->getTerminator();
IRBuilder<> B(InsertPt);
Value *BrLoopExit = B.CreateIsNotNull(ModVal, "lcmp.mod");
assert(Exit && "Loop must have a single exit block only");
// Split the epilogue exit to maintain loop canonicalization guarantees
SmallVector<BasicBlock*, 4> Preds(predecessors(Exit));
SplitBlockPredecessors(Exit, Preds, ".epilog-lcssa", DT, LI, nullptr,
PreserveLCSSA);
// Add the branch to the exit block (around the unrolling loop)
MDNode *BranchWeights = nullptr;
if (hasBranchWeightMD(*Latch->getTerminator())) {
// Assume equal distribution in interval [0, Count).
MDBuilder MDB(B.getContext());
BranchWeights = MDB.createBranchWeights(1, Count - 1);
}
B.CreateCondBr(BrLoopExit, EpilogPreHeader, Exit, BranchWeights);
InsertPt->eraseFromParent();
if (DT) {
auto *NewDom = DT->findNearestCommonDominator(Exit, NewExit);
DT->changeImmediateDominator(Exit, NewDom);
}
// Split the main loop exit to maintain canonicalization guarantees.
SmallVector<BasicBlock*, 4> NewExitPreds{Latch};
SplitBlockPredecessors(NewExit, NewExitPreds, ".loopexit", DT, LI, nullptr,
PreserveLCSSA);
}
/// Create a clone of the blocks in a loop and connect them together. A new
/// loop will be created including all cloned blocks, and the iterator of the
/// new loop switched to count NewIter down to 0.
/// The cloned blocks should be inserted between InsertTop and InsertBot.
/// InsertTop should be new preheader, InsertBot new loop exit.
/// Returns the new cloned loop that is created.
static Loop *
CloneLoopBlocks(Loop *L, Value *NewIter, const bool UseEpilogRemainder,
const bool UnrollRemainder,
BasicBlock *InsertTop,
BasicBlock *InsertBot, BasicBlock *Preheader,
std::vector<BasicBlock *> &NewBlocks,
LoopBlocksDFS &LoopBlocks, ValueToValueMapTy &VMap,
DominatorTree *DT, LoopInfo *LI, unsigned Count) {
StringRef suffix = UseEpilogRemainder ? "epil" : "prol";
BasicBlock *Header = L->getHeader();
BasicBlock *Latch = L->getLoopLatch();
Function *F = Header->getParent();
LoopBlocksDFS::RPOIterator BlockBegin = LoopBlocks.beginRPO();
LoopBlocksDFS::RPOIterator BlockEnd = LoopBlocks.endRPO();
Loop *ParentLoop = L->getParentLoop();
NewLoopsMap NewLoops;
NewLoops[ParentLoop] = ParentLoop;
// For each block in the original loop, create a new copy,
// and update the value map with the newly created values.
for (LoopBlocksDFS::RPOIterator BB = BlockBegin; BB != BlockEnd; ++BB) {
BasicBlock *NewBB = CloneBasicBlock(*BB, VMap, "." + suffix, F);
NewBlocks.push_back(NewBB);
addClonedBlockToLoopInfo(*BB, NewBB, LI, NewLoops);
VMap[*BB] = NewBB;
if (Header == *BB) {
// For the first block, add a CFG connection to this newly
// created block.
InsertTop->getTerminator()->setSuccessor(0, NewBB);
}
if (DT) {
if (Header == *BB) {
// The header is dominated by the preheader.
DT->addNewBlock(NewBB, InsertTop);
} else {
// Copy information from original loop to unrolled loop.
BasicBlock *IDomBB = DT->getNode(*BB)->getIDom()->getBlock();
DT->addNewBlock(NewBB, cast<BasicBlock>(VMap[IDomBB]));
}
}
if (Latch == *BB) {
// For the last block, create a loop back to cloned head.
VMap.erase((*BB)->getTerminator());
// Use an incrementing IV. Pre-incr/post-incr is backedge/trip count.
// Subtle: NewIter can be 0 if we wrapped when computing the trip count,
// thus we must compare the post-increment (wrapping) value.
BasicBlock *FirstLoopBB = cast<BasicBlock>(VMap[Header]);
BranchInst *LatchBR = cast<BranchInst>(NewBB->getTerminator());
IRBuilder<> Builder(LatchBR);
PHINode *NewIdx =
PHINode::Create(NewIter->getType(), 2, suffix + ".iter");
NewIdx->insertBefore(FirstLoopBB->getFirstNonPHIIt());
auto *Zero = ConstantInt::get(NewIdx->getType(), 0);
auto *One = ConstantInt::get(NewIdx->getType(), 1);
Value *IdxNext =
Builder.CreateAdd(NewIdx, One, NewIdx->getName() + ".next");
Value *IdxCmp = Builder.CreateICmpNE(IdxNext, NewIter, NewIdx->getName() + ".cmp");
MDNode *BranchWeights = nullptr;
if (hasBranchWeightMD(*LatchBR)) {
uint32_t ExitWeight;
uint32_t BackEdgeWeight;
if (Count >= 3) {
// Note: We do not enter this loop for zero-remainders. The check
// is at the end of the loop. We assume equal distribution between
// possible remainders in [1, Count).
ExitWeight = 1;
BackEdgeWeight = (Count - 2) / 2;
} else {
// Unnecessary backedge, should never be taken. The conditional
// jump should be optimized away later.
ExitWeight = 1;
BackEdgeWeight = 0;
}
MDBuilder MDB(Builder.getContext());
BranchWeights = MDB.createBranchWeights(BackEdgeWeight, ExitWeight);
}
Builder.CreateCondBr(IdxCmp, FirstLoopBB, InsertBot, BranchWeights);
NewIdx->addIncoming(Zero, InsertTop);
NewIdx->addIncoming(IdxNext, NewBB);
LatchBR->eraseFromParent();
}
}
// Change the incoming values to the ones defined in the preheader or
// cloned loop.
for (BasicBlock::iterator I = Header->begin(); isa<PHINode>(I); ++I) {
PHINode *NewPHI = cast<PHINode>(VMap[&*I]);
unsigned idx = NewPHI->getBasicBlockIndex(Preheader);
NewPHI->setIncomingBlock(idx, InsertTop);
BasicBlock *NewLatch = cast<BasicBlock>(VMap[Latch]);
idx = NewPHI->getBasicBlockIndex(Latch);
Value *InVal = NewPHI->getIncomingValue(idx);
NewPHI->setIncomingBlock(idx, NewLatch);
if (Value *V = VMap.lookup(InVal))
NewPHI->setIncomingValue(idx, V);
}
Loop *NewLoop = NewLoops[L];
assert(NewLoop && "L should have been cloned");
MDNode *LoopID = NewLoop->getLoopID();
// Only add loop metadata if the loop is not going to be completely
// unrolled.
if (UnrollRemainder)
return NewLoop;
std::optional<MDNode *> NewLoopID = makeFollowupLoopID(
LoopID, {LLVMLoopUnrollFollowupAll, LLVMLoopUnrollFollowupRemainder});
if (NewLoopID) {
NewLoop->setLoopID(*NewLoopID);
// Do not setLoopAlreadyUnrolled if loop attributes have been defined
// explicitly.
return NewLoop;
}
// Add unroll disable metadata to disable future unrolling for this loop.
NewLoop->setLoopAlreadyUnrolled();
return NewLoop;
}
/// Returns true if we can profitably unroll the multi-exit loop L. Currently,
/// we return true only if UnrollRuntimeMultiExit is set to true.
static bool canProfitablyUnrollMultiExitLoop(
Loop *L, SmallVectorImpl<BasicBlock *> &OtherExits, BasicBlock *LatchExit,
bool UseEpilogRemainder) {
// Priority goes to UnrollRuntimeMultiExit if it's supplied.
if (UnrollRuntimeMultiExit.getNumOccurrences())
return UnrollRuntimeMultiExit;
// The main pain point with multi-exit loop unrolling is that once unrolled,
// we will not be able to merge all blocks into a straight line code.
// There are branches within the unrolled loop that go to the OtherExits.
// The second point is the increase in code size, but this is true
// irrespective of multiple exits.
// Note: Both the heuristics below are coarse grained. We are essentially
// enabling unrolling of loops that have a single side exit other than the
// normal LatchExit (i.e. exiting into a deoptimize block).
// The heuristics considered are:
// 1. low number of branches in the unrolled version.
// 2. high predictability of these extra branches.
// We avoid unrolling loops that have more than two exiting blocks. This
// limits the total number of branches in the unrolled loop to be atmost
// the unroll factor (since one of the exiting blocks is the latch block).
SmallVector<BasicBlock*, 4> ExitingBlocks;
L->getExitingBlocks(ExitingBlocks);
if (ExitingBlocks.size() > 2)
return false;
// Allow unrolling of loops with no non latch exit blocks.
if (OtherExits.size() == 0)
return true;
// The second heuristic is that L has one exit other than the latchexit and
// that exit is a deoptimize block. We know that deoptimize blocks are rarely
// taken, which also implies the branch leading to the deoptimize block is
// highly predictable. When UnrollRuntimeOtherExitPredictable is specified, we
// assume the other exit branch is predictable even if it has no deoptimize
// call.
return (OtherExits.size() == 1 &&
(UnrollRuntimeOtherExitPredictable ||
OtherExits[0]->getPostdominatingDeoptimizeCall()));
// TODO: These can be fine-tuned further to consider code size or deopt states
// that are captured by the deoptimize exit block.
// Also, we can extend this to support more cases, if we actually
// know of kinds of multiexit loops that would benefit from unrolling.
}
/// Calculate ModVal = (BECount + 1) % Count on the abstract integer domain
/// accounting for the possibility of unsigned overflow in the 2s complement
/// domain. Preconditions:
/// 1) TripCount = BECount + 1 (allowing overflow)
/// 2) Log2(Count) <= BitWidth(BECount)
static Value *CreateTripRemainder(IRBuilder<> &B, Value *BECount,
Value *TripCount, unsigned Count) {
// Note that TripCount is BECount + 1.
if (isPowerOf2_32(Count))
// If the expression is zero, then either:
// 1. There are no iterations to be run in the prolog/epilog loop.
// OR
// 2. The addition computing TripCount overflowed.
//
// If (2) is true, we know that TripCount really is (1 << BEWidth) and so
// the number of iterations that remain to be run in the original loop is a
// multiple Count == (1 << Log2(Count)) because Log2(Count) <= BEWidth (a
// precondition of this method).
return B.CreateAnd(TripCount, Count - 1, "xtraiter");
// As (BECount + 1) can potentially unsigned overflow we count
// (BECount % Count) + 1 which is overflow safe as BECount % Count < Count.
Constant *CountC = ConstantInt::get(BECount->getType(), Count);
Value *ModValTmp = B.CreateURem(BECount, CountC);
Value *ModValAdd = B.CreateAdd(ModValTmp,
ConstantInt::get(ModValTmp->getType(), 1));
// At that point (BECount % Count) + 1 could be equal to Count.
// To handle this case we need to take mod by Count one more time.
return B.CreateURem(ModValAdd, CountC, "xtraiter");
}
/// Insert code in the prolog/epilog code when unrolling a loop with a
/// run-time trip-count.
///
/// This method assumes that the loop unroll factor is total number
/// of loop bodies in the loop after unrolling. (Some folks refer
/// to the unroll factor as the number of *extra* copies added).
/// We assume also that the loop unroll factor is a power-of-two. So, after
/// unrolling the loop, the number of loop bodies executed is 2,
/// 4, 8, etc. Note - LLVM converts the if-then-sequence to a switch
/// instruction in SimplifyCFG.cpp. Then, the backend decides how code for
/// the switch instruction is generated.
///
/// ***Prolog case***
/// extraiters = tripcount % loopfactor
/// if (extraiters == 0) jump Loop:
/// else jump Prol:
/// Prol: LoopBody;
/// extraiters -= 1 // Omitted if unroll factor is 2.
/// if (extraiters != 0) jump Prol: // Omitted if unroll factor is 2.
/// if (tripcount < loopfactor) jump End:
/// Loop:
/// ...
/// End:
///
/// ***Epilog case***
/// extraiters = tripcount % loopfactor
/// if (tripcount < loopfactor) jump LoopExit:
/// unroll_iters = tripcount - extraiters
/// Loop: LoopBody; (executes unroll_iter times);
/// unroll_iter -= 1
/// if (unroll_iter != 0) jump Loop:
/// LoopExit:
/// if (extraiters == 0) jump EpilExit:
/// Epil: LoopBody; (executes extraiters times)
/// extraiters -= 1 // Omitted if unroll factor is 2.
/// if (extraiters != 0) jump Epil: // Omitted if unroll factor is 2.
/// EpilExit:
bool llvm::UnrollRuntimeLoopRemainder(
Loop *L, unsigned Count, bool AllowExpensiveTripCount,
bool UseEpilogRemainder, bool UnrollRemainder, bool ForgetAllSCEV,
LoopInfo *LI, ScalarEvolution *SE, DominatorTree *DT, AssumptionCache *AC,
const TargetTransformInfo *TTI, bool PreserveLCSSA, Loop **ResultLoop) {
LLVM_DEBUG(dbgs() << "Trying runtime unrolling on Loop: \n");
LLVM_DEBUG(L->dump());
LLVM_DEBUG(UseEpilogRemainder ? dbgs() << "Using epilog remainder.\n"
: dbgs() << "Using prolog remainder.\n");
// Make sure the loop is in canonical form.
if (!L->isLoopSimplifyForm()) {
LLVM_DEBUG(dbgs() << "Not in simplify form!\n");
return false;
}
// Guaranteed by LoopSimplifyForm.
BasicBlock *Latch = L->getLoopLatch();
BasicBlock *Header = L->getHeader();
BranchInst *LatchBR = cast<BranchInst>(Latch->getTerminator());
if (!LatchBR || LatchBR->isUnconditional()) {
// The loop-rotate pass can be helpful to avoid this in many cases.
LLVM_DEBUG(
dbgs()
<< "Loop latch not terminated by a conditional branch.\n");
return false;
}
unsigned ExitIndex = LatchBR->getSuccessor(0) == Header ? 1 : 0;
BasicBlock *LatchExit = LatchBR->getSuccessor(ExitIndex);
if (L->contains(LatchExit)) {
// Cloning the loop basic blocks (`CloneLoopBlocks`) requires that one of the
// targets of the Latch be an exit block out of the loop.
LLVM_DEBUG(
dbgs()
<< "One of the loop latch successors must be the exit block.\n");
return false;
}
// These are exit blocks other than the target of the latch exiting block.
SmallVector<BasicBlock *, 4> OtherExits;
L->getUniqueNonLatchExitBlocks(OtherExits);
// Support only single exit and exiting block unless multi-exit loop
// unrolling is enabled.
if (!L->getExitingBlock() || OtherExits.size()) {
// We rely on LCSSA form being preserved when the exit blocks are transformed.
// (Note that only an off-by-default mode of the old PM disables PreserveLCCA.)
if (!PreserveLCSSA)
return false;
if (!canProfitablyUnrollMultiExitLoop(L, OtherExits, LatchExit,
UseEpilogRemainder)) {
LLVM_DEBUG(
dbgs()
<< "Multiple exit/exiting blocks in loop and multi-exit unrolling not "
"enabled!\n");
return false;
}
}
// Use Scalar Evolution to compute the trip count. This allows more loops to
// be unrolled than relying on induction var simplification.
if (!SE)
return false;
// Only unroll loops with a computable trip count.
// We calculate the backedge count by using getExitCount on the Latch block,
// which is proven to be the only exiting block in this loop. This is same as
// calculating getBackedgeTakenCount on the loop (which computes SCEV for all
// exiting blocks).
const SCEV *BECountSC = SE->getExitCount(L, Latch);
if (isa<SCEVCouldNotCompute>(BECountSC)) {
LLVM_DEBUG(dbgs() << "Could not compute exit block SCEV\n");
return false;
}
unsigned BEWidth = cast<IntegerType>(BECountSC->getType())->getBitWidth();
// Add 1 since the backedge count doesn't include the first loop iteration.
// (Note that overflow can occur, this is handled explicitly below)
const SCEV *TripCountSC =
SE->getAddExpr(BECountSC, SE->getConstant(BECountSC->getType(), 1));
if (isa<SCEVCouldNotCompute>(TripCountSC)) {
LLVM_DEBUG(dbgs() << "Could not compute trip count SCEV.\n");
return false;
}
BasicBlock *PreHeader = L->getLoopPreheader();
BranchInst *PreHeaderBR = cast<BranchInst>(PreHeader->getTerminator());
const DataLayout &DL = Header->getModule()->getDataLayout();
SCEVExpander Expander(*SE, DL, "loop-unroll");
if (!AllowExpensiveTripCount &&
Expander.isHighCostExpansion(TripCountSC, L, SCEVCheapExpansionBudget,
TTI, PreHeaderBR)) {
LLVM_DEBUG(dbgs() << "High cost for expanding trip count scev!\n");
return false;
}
// This constraint lets us deal with an overflowing trip count easily; see the
// comment on ModVal below.
if (Log2_32(Count) > BEWidth) {
LLVM_DEBUG(
dbgs()
<< "Count failed constraint on overflow trip count calculation.\n");
return false;
}
// Loop structure is the following:
//
// PreHeader
// Header
// ...
// Latch
// LatchExit
BasicBlock *NewPreHeader;
BasicBlock *NewExit = nullptr;
BasicBlock *PrologExit = nullptr;
BasicBlock *EpilogPreHeader = nullptr;
BasicBlock *PrologPreHeader = nullptr;
if (UseEpilogRemainder) {
// If epilog remainder
// Split PreHeader to insert a branch around loop for unrolling.
NewPreHeader = SplitBlock(PreHeader, PreHeader->getTerminator(), DT, LI);
NewPreHeader->setName(PreHeader->getName() + ".new");
// Split LatchExit to create phi nodes from branch above.
NewExit = SplitBlockPredecessors(LatchExit, {Latch}, ".unr-lcssa", DT, LI,
nullptr, PreserveLCSSA);
// NewExit gets its DebugLoc from LatchExit, which is not part of the
// original Loop.
// Fix this by setting Loop's DebugLoc to NewExit.
auto *NewExitTerminator = NewExit->getTerminator();
NewExitTerminator->setDebugLoc(Header->getTerminator()->getDebugLoc());
// Split NewExit to insert epilog remainder loop.
EpilogPreHeader = SplitBlock(NewExit, NewExitTerminator, DT, LI);
EpilogPreHeader->setName(Header->getName() + ".epil.preheader");
// If the latch exits from multiple level of nested loops, then
// by assumption there must be another loop exit which branches to the
// outer loop and we must adjust the loop for the newly inserted blocks
// to account for the fact that our epilogue is still in the same outer
// loop. Note that this leaves loopinfo temporarily out of sync with the
// CFG until the actual epilogue loop is inserted.
if (auto *ParentL = L->getParentLoop())
if (LI->getLoopFor(LatchExit) != ParentL) {
LI->removeBlock(NewExit);
ParentL->addBasicBlockToLoop(NewExit, *LI);
LI->removeBlock(EpilogPreHeader);
ParentL->addBasicBlockToLoop(EpilogPreHeader, *LI);
}
} else {
// If prolog remainder
// Split the original preheader twice to insert prolog remainder loop
PrologPreHeader = SplitEdge(PreHeader, Header, DT, LI);
PrologPreHeader->setName(Header->getName() + ".prol.preheader");
PrologExit = SplitBlock(PrologPreHeader, PrologPreHeader->getTerminator(),
DT, LI);
PrologExit->setName(Header->getName() + ".prol.loopexit");
// Split PrologExit to get NewPreHeader.
NewPreHeader = SplitBlock(PrologExit, PrologExit->getTerminator(), DT, LI);
NewPreHeader->setName(PreHeader->getName() + ".new");
}
// Loop structure should be the following:
// Epilog Prolog
//
// PreHeader PreHeader
// *NewPreHeader *PrologPreHeader
// Header *PrologExit
// ... *NewPreHeader
// Latch Header
// *NewExit ...
// *EpilogPreHeader Latch
// LatchExit LatchExit
// Calculate conditions for branch around loop for unrolling
// in epilog case and around prolog remainder loop in prolog case.
// Compute the number of extra iterations required, which is:
// extra iterations = run-time trip count % loop unroll factor
PreHeaderBR = cast<BranchInst>(PreHeader->getTerminator());
IRBuilder<> B(PreHeaderBR);
Value *TripCount = Expander.expandCodeFor(TripCountSC, TripCountSC->getType(),
PreHeaderBR);
Value *BECount;
// If there are other exits before the latch, that may cause the latch exit
// branch to never be executed, and the latch exit count may be poison.
// In this case, freeze the TripCount and base BECount on the frozen
// TripCount. We will introduce two branches using these values, and it's
// important that they see a consistent value (which would not be guaranteed
// if were frozen independently.)
if ((!OtherExits.empty() || !SE->loopHasNoAbnormalExits(L)) &&
!isGuaranteedNotToBeUndefOrPoison(TripCount, AC, PreHeaderBR, DT)) {
TripCount = B.CreateFreeze(TripCount);
BECount =
B.CreateAdd(TripCount, Constant::getAllOnesValue(TripCount->getType()));
} else {
// If we don't need to freeze, use SCEVExpander for BECount as well, to
// allow slightly better value reuse.
BECount =
Expander.expandCodeFor(BECountSC, BECountSC->getType(), PreHeaderBR);
}
Value * const ModVal = CreateTripRemainder(B, BECount, TripCount, Count);
Value *BranchVal =
UseEpilogRemainder ? B.CreateICmpULT(BECount,
ConstantInt::get(BECount->getType(),
Count - 1)) :
B.CreateIsNotNull(ModVal, "lcmp.mod");
BasicBlock *RemainderLoop = UseEpilogRemainder ? NewExit : PrologPreHeader;
BasicBlock *UnrollingLoop = UseEpilogRemainder ? NewPreHeader : PrologExit;
// Branch to either remainder (extra iterations) loop or unrolling loop.
MDNode *BranchWeights = nullptr;
if (hasBranchWeightMD(*Latch->getTerminator())) {
// Assume loop is nearly always entered.
MDBuilder MDB(B.getContext());
BranchWeights = MDB.createBranchWeights(EpilogHeaderWeights);
}
B.CreateCondBr(BranchVal, RemainderLoop, UnrollingLoop, BranchWeights);
PreHeaderBR->eraseFromParent();
if (DT) {
if (UseEpilogRemainder)
DT->changeImmediateDominator(NewExit, PreHeader);
else
DT->changeImmediateDominator(PrologExit, PreHeader);
}
Function *F = Header->getParent();
// Get an ordered list of blocks in the loop to help with the ordering of the
// cloned blocks in the prolog/epilog code
LoopBlocksDFS LoopBlocks(L);
LoopBlocks.perform(LI);
//
// For each extra loop iteration, create a copy of the loop's basic blocks
// and generate a condition that branches to the copy depending on the
// number of 'left over' iterations.
//
std::vector<BasicBlock *> NewBlocks;
ValueToValueMapTy VMap;
// Clone all the basic blocks in the loop. If Count is 2, we don't clone
// the loop, otherwise we create a cloned loop to execute the extra
// iterations. This function adds the appropriate CFG connections.
BasicBlock *InsertBot = UseEpilogRemainder ? LatchExit : PrologExit;
BasicBlock *InsertTop = UseEpilogRemainder ? EpilogPreHeader : PrologPreHeader;
Loop *remainderLoop = CloneLoopBlocks(
L, ModVal, UseEpilogRemainder, UnrollRemainder, InsertTop, InsertBot,
NewPreHeader, NewBlocks, LoopBlocks, VMap, DT, LI, Count);
// Insert the cloned blocks into the function.
F->splice(InsertBot->getIterator(), F, NewBlocks[0]->getIterator(), F->end());
// Now the loop blocks are cloned and the other exiting blocks from the
// remainder are connected to the original Loop's exit blocks. The remaining
// work is to update the phi nodes in the original loop, and take in the
// values from the cloned region.
for (auto *BB : OtherExits) {
// Given we preserve LCSSA form, we know that the values used outside the
// loop will be used through these phi nodes at the exit blocks that are
// transformed below.
for (PHINode &PN : BB->phis()) {
unsigned oldNumOperands = PN.getNumIncomingValues();
// Add the incoming values from the remainder code to the end of the phi
// node.
for (unsigned i = 0; i < oldNumOperands; i++){
auto *PredBB =PN.getIncomingBlock(i);
if (PredBB == Latch)
// The latch exit is handled seperately, see connectX
continue;
if (!L->contains(PredBB))
// Even if we had dedicated exits, the code above inserted an
// extra branch which can reach the latch exit.
continue;
auto *V = PN.getIncomingValue(i);
if (Instruction *I = dyn_cast<Instruction>(V))
if (L->contains(I))
V = VMap.lookup(I);
PN.addIncoming(V, cast<BasicBlock>(VMap[PredBB]));
}
}
#if defined(EXPENSIVE_CHECKS) && !defined(NDEBUG)
for (BasicBlock *SuccBB : successors(BB)) {
assert(!(llvm::is_contained(OtherExits, SuccBB) || SuccBB == LatchExit) &&
"Breaks the definition of dedicated exits!");
}
#endif
}
// Update the immediate dominator of the exit blocks and blocks that are
// reachable from the exit blocks. This is needed because we now have paths
// from both the original loop and the remainder code reaching the exit
// blocks. While the IDom of these exit blocks were from the original loop,
// now the IDom is the preheader (which decides whether the original loop or
// remainder code should run).
if (DT && !L->getExitingBlock()) {
SmallVector<BasicBlock *, 16> ChildrenToUpdate;
// NB! We have to examine the dom children of all loop blocks, not just
// those which are the IDom of the exit blocks. This is because blocks
// reachable from the exit blocks can have their IDom as the nearest common
// dominator of the exit blocks.
for (auto *BB : L->blocks()) {
auto *DomNodeBB = DT->getNode(BB);
for (auto *DomChild : DomNodeBB->children()) {
auto *DomChildBB = DomChild->getBlock();
if (!L->contains(LI->getLoopFor(DomChildBB)))
ChildrenToUpdate.push_back(DomChildBB);
}
}
for (auto *BB : ChildrenToUpdate)
DT->changeImmediateDominator(BB, PreHeader);
}
// Loop structure should be the following:
// Epilog Prolog
//
// PreHeader PreHeader
// NewPreHeader PrologPreHeader
// Header PrologHeader
// ... ...
// Latch PrologLatch
// NewExit PrologExit
// EpilogPreHeader NewPreHeader
// EpilogHeader Header
// ... ...
// EpilogLatch Latch
// LatchExit LatchExit
// Rewrite the cloned instruction operands to use the values created when the
// clone is created.
for (BasicBlock *BB : NewBlocks) {
Module *M = BB->getModule();
for (Instruction &I : *BB) {
RemapInstruction(&I, VMap,
RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
RemapDbgRecordRange(M, I.getDbgRecordRange(), VMap,
RF_NoModuleLevelChanges | RF_IgnoreMissingLocals);
}
}
if (UseEpilogRemainder) {
// Connect the epilog code to the original loop and update the
// PHI functions.
ConnectEpilog(L, ModVal, NewExit, LatchExit, PreHeader, EpilogPreHeader,
NewPreHeader, VMap, DT, LI, PreserveLCSSA, *SE, Count);
// Update counter in loop for unrolling.
// Use an incrementing IV. Pre-incr/post-incr is backedge/trip count.
// Subtle: TestVal can be 0 if we wrapped when computing the trip count,
// thus we must compare the post-increment (wrapping) value.
IRBuilder<> B2(NewPreHeader->getTerminator());
Value *TestVal = B2.CreateSub(TripCount, ModVal, "unroll_iter");
BranchInst *LatchBR = cast<BranchInst>(Latch->getTerminator());
PHINode *NewIdx = PHINode::Create(TestVal->getType(), 2, "niter");
NewIdx->insertBefore(Header->getFirstNonPHIIt());
B2.SetInsertPoint(LatchBR);
auto *Zero = ConstantInt::get(NewIdx->getType(), 0);
auto *One = ConstantInt::get(NewIdx->getType(), 1);
Value *IdxNext = B2.CreateAdd(NewIdx, One, NewIdx->getName() + ".next");
auto Pred = LatchBR->getSuccessor(0) == Header ? ICmpInst::ICMP_NE : ICmpInst::ICMP_EQ;
Value *IdxCmp = B2.CreateICmp(Pred, IdxNext, TestVal, NewIdx->getName() + ".ncmp");
NewIdx->addIncoming(Zero, NewPreHeader);
NewIdx->addIncoming(IdxNext, Latch);
LatchBR->setCondition(IdxCmp);
} else {
// Connect the prolog code to the original loop and update the
// PHI functions.
ConnectProlog(L, BECount, Count, PrologExit, LatchExit, PreHeader,
NewPreHeader, VMap, DT, LI, PreserveLCSSA, *SE);
}
// If this loop is nested, then the loop unroller changes the code in the any
// of its parent loops, so the Scalar Evolution pass needs to be run again.
SE->forgetTopmostLoop(L);
// Verify that the Dom Tree and Loop Info are correct.
#if defined(EXPENSIVE_CHECKS) && !defined(NDEBUG)
if (DT) {
assert(DT->verify(DominatorTree::VerificationLevel::Full));
LI->verify(*DT);
}
#endif
// For unroll factor 2 remainder loop will have 1 iteration.
if (Count == 2 && DT && LI && SE) {
// TODO: This code could probably be pulled out into a helper function
// (e.g. breakLoopBackedgeAndSimplify) and reused in loop-deletion.
BasicBlock *RemainderLatch = remainderLoop->getLoopLatch();
assert(RemainderLatch);
SmallVector<BasicBlock*> RemainderBlocks(remainderLoop->getBlocks().begin(),
remainderLoop->getBlocks().end());
breakLoopBackedge(remainderLoop, *DT, *SE, *LI, nullptr);
remainderLoop = nullptr;
// Simplify loop values after breaking the backedge
const DataLayout &DL = L->getHeader()->getModule()->getDataLayout();
SmallVector<WeakTrackingVH, 16> DeadInsts;
for (BasicBlock *BB : RemainderBlocks) {
for (Instruction &Inst : llvm::make_early_inc_range(*BB)) {
if (Value *V = simplifyInstruction(&Inst, {DL, nullptr, DT, AC}))
if (LI->replacementPreservesLCSSAForm(&Inst, V))
Inst.replaceAllUsesWith(V);
if (isInstructionTriviallyDead(&Inst))
DeadInsts.emplace_back(&Inst);
}
// We can't do recursive deletion until we're done iterating, as we might
// have a phi which (potentially indirectly) uses instructions later in
// the block we're iterating through.
RecursivelyDeleteTriviallyDeadInstructions(DeadInsts);
}
// Merge latch into exit block.
auto *ExitBB = RemainderLatch->getSingleSuccessor();
assert(ExitBB && "required after breaking cond br backedge");
DomTreeUpdater DTU(DT, DomTreeUpdater::UpdateStrategy::Eager);
MergeBlockIntoPredecessor(ExitBB, &DTU, LI);
}
// Canonicalize to LoopSimplifyForm both original and remainder loops. We
// cannot rely on the LoopUnrollPass to do this because it only does
// canonicalization for parent/subloops and not the sibling loops.
if (OtherExits.size() > 0) {
// Generate dedicated exit blocks for the original loop, to preserve
// LoopSimplifyForm.
formDedicatedExitBlocks(L, DT, LI, nullptr, PreserveLCSSA);
// Generate dedicated exit blocks for the remainder loop if one exists, to
// preserve LoopSimplifyForm.
if (remainderLoop)
formDedicatedExitBlocks(remainderLoop, DT, LI, nullptr, PreserveLCSSA);
}
auto UnrollResult = LoopUnrollResult::Unmodified;
if (remainderLoop && UnrollRemainder) {
LLVM_DEBUG(dbgs() << "Unrolling remainder loop\n");
UnrollResult =
UnrollLoop(remainderLoop,
{/*Count*/ Count - 1, /*Force*/ false, /*Runtime*/ false,
/*AllowExpensiveTripCount*/ false,
/*UnrollRemainder*/ false, ForgetAllSCEV},
LI, SE, DT, AC, TTI, /*ORE*/ nullptr, PreserveLCSSA);
}
if (ResultLoop && UnrollResult != LoopUnrollResult::FullyUnrolled)
*ResultLoop = remainderLoop;
NumRuntimeUnrolled++;
return true;
}