llvm/test/CodeGen/AArch64/saturating-vec-smull.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=aarch64-none-elf < %s | FileCheck %s --check-prefixes=CHECK,CHECK-SD
; RUN: llc -mtriple=aarch64-none-elf -global-isel -global-isel-abort=2 2>&1 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-GI

; CHECK-GI:       warning: Instruction selection used fallback path for extend_to_illegal_type

define <2 x i16> @saturating_2xi16(<2 x i16> %a, <2 x i16> %b) {
; CHECK-SD-LABEL: saturating_2xi16:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    shl v0.2s, v0.2s, #16
; CHECK-SD-NEXT:    shl v1.2s, v1.2s, #16
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #16
; CHECK-SD-NEXT:    sshr v1.2s, v1.2s, #16
; CHECK-SD-NEXT:    sqdmulh v0.2s, v1.2s, v0.2s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: saturating_2xi16:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    shl v0.2s, v0.2s, #16
; CHECK-GI-NEXT:    shl v1.2s, v1.2s, #16
; CHECK-GI-NEXT:    sshr v0.2s, v0.2s, #16
; CHECK-GI-NEXT:    sshr v1.2s, v1.2s, #16
; CHECK-GI-NEXT:    mul v0.2s, v1.2s, v0.2s
; CHECK-GI-NEXT:    movi v1.2s, #127, msl #8
; CHECK-GI-NEXT:    sshr v0.2s, v0.2s, #15
; CHECK-GI-NEXT:    smin v0.2s, v0.2s, v1.2s
; CHECK-GI-NEXT:    ret
  %as = sext <2 x i16> %a to <2 x i32>
  %bs = sext <2 x i16> %b to <2 x i32>
  %m = mul nsw <2 x i32> %bs, %as
  %sh = ashr <2 x i32> %m, splat (i32 15)
  %ma = tail call <2 x i32> @llvm.smin.v2i32(<2 x i32> %sh, <2 x i32> splat (i32 32767))
  %t = trunc nsw <2 x i32> %ma to <2 x i16>
  ret <2 x i16> %t
}

define <4 x i16> @saturating_4xi16(<4 x i16> %a, <4 x i16> %b) {
; CHECK-LABEL: saturating_4xi16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    sqdmulh v0.4h, v1.4h, v0.4h
; CHECK-NEXT:    ret
  %as = sext <4 x i16> %a to <4 x i32>
  %bs = sext <4 x i16> %b to <4 x i32>
  %m = mul nsw <4 x i32> %bs, %as
  %sh = ashr <4 x i32> %m, splat (i32 15)
  %ma = tail call <4 x i32> @llvm.smin.v4i32(<4 x i32> %sh, <4 x i32> splat (i32 32767))
  %t = trunc nsw <4 x i32> %ma to <4 x i16>
  ret <4 x i16> %t
}

define <8 x i16> @saturating_8xi16(<8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: saturating_8xi16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    sqdmulh v0.8h, v1.8h, v0.8h
; CHECK-NEXT:    ret
  %as = sext <8 x i16> %a to <8 x i32>
  %bs = sext <8 x i16> %b to <8 x i32>
  %m = mul nsw <8 x i32> %bs, %as
  %sh = ashr <8 x i32> %m, splat (i32 15)
  %ma = tail call <8 x i32> @llvm.smin.v8i32(<8 x i32> %sh, <8 x i32> splat (i32 32767))
  %t = trunc nsw <8 x i32> %ma to <8 x i16>
  ret <8 x i16> %t
}

define <2 x i32> @saturating_2xi32(<2 x i32> %a, <2 x i32> %b) {
; CHECK-LABEL: saturating_2xi32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    sqdmulh v0.2s, v1.2s, v0.2s
; CHECK-NEXT:    ret
  %as = sext <2 x i32> %a to <2 x i64>
  %bs = sext <2 x i32> %b to <2 x i64>
  %m = mul nsw <2 x i64> %bs, %as
  %sh = ashr <2 x i64> %m, splat (i64 31)
  %ma = tail call <2 x i64> @llvm.smin.v2i64(<2 x i64> %sh, <2 x i64> splat (i64 2147483647))
  %t = trunc nsw <2 x i64> %ma to <2 x i32>
  ret <2 x i32> %t
}

define <4 x i32> @saturating_4xi32(<4 x i32> %a, <4 x i32> %b) {
; CHECK-LABEL: saturating_4xi32:
; CHECK:       // %bb.0:
; CHECK-NEXT:    sqdmulh v0.4s, v1.4s, v0.4s
; CHECK-NEXT:    ret
  %as = sext <4 x i32> %a to <4 x i64>
  %bs = sext <4 x i32> %b to <4 x i64>
  %m = mul nsw <4 x i64> %bs, %as
  %sh = ashr <4 x i64> %m, splat (i64 31)
  %ma = tail call <4 x i64> @llvm.smin.v4i64(<4 x i64> %sh, <4 x i64> splat (i64 2147483647))
  %t = trunc nsw <4 x i64> %ma to <4 x i32>
  ret <4 x i32> %t
}

define <8 x i32> @saturating_8xi32(<8 x i32> %a, <8 x i32> %b) {
; CHECK-SD-LABEL: saturating_8xi32:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    sqdmulh v1.4s, v3.4s, v1.4s
; CHECK-SD-NEXT:    sqdmulh v0.4s, v2.4s, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: saturating_8xi32:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    sqdmulh v0.4s, v2.4s, v0.4s
; CHECK-GI-NEXT:    sqdmulh v1.4s, v3.4s, v1.4s
; CHECK-GI-NEXT:    ret
  %as = sext <8 x i32> %a to <8 x i64>
  %bs = sext <8 x i32> %b to <8 x i64>
  %m = mul nsw <8 x i64> %bs, %as
  %sh = ashr <8 x i64> %m, splat (i64 31)
  %ma = tail call <8 x i64> @llvm.smin.v8i64(<8 x i64> %sh, <8 x i64> splat (i64 2147483647))
  %t = trunc nsw <8 x i64> %ma to <8 x i32>
  ret <8 x i32> %t
}

define <2 x i64> @saturating_2xi32_2xi64(<2 x i32> %a, <2 x i32> %b) {
; CHECK-SD-LABEL: saturating_2xi32_2xi64:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    sqdmulh v0.2s, v1.2s, v0.2s
; CHECK-SD-NEXT:    sshll v0.2d, v0.2s, #0
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: saturating_2xi32_2xi64:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    smull v0.2d, v1.2s, v0.2s
; CHECK-GI-NEXT:    adrp x8, .LCPI6_0
; CHECK-GI-NEXT:    ldr q1, [x8, :lo12:.LCPI6_0]
; CHECK-GI-NEXT:    sshr v0.2d, v0.2d, #31
; CHECK-GI-NEXT:    cmgt v2.2d, v1.2d, v0.2d
; CHECK-GI-NEXT:    bif v0.16b, v1.16b, v2.16b
; CHECK-GI-NEXT:    ret
  %as = sext <2 x i32> %a to <2 x i64>
  %bs = sext <2 x i32> %b to <2 x i64>
  %m = mul nsw <2 x i64> %bs, %as
  %sh = ashr <2 x i64> %m, splat (i64 31)
  %ma = tail call <2 x i64> @llvm.smin.v2i64(<2 x i64> %sh, <2 x i64> splat (i64 2147483647))
  ret <2 x i64> %ma
}

define <6 x i16> @saturating_6xi16(<6 x i16> %a, <6 x i16> %b) {
; CHECK-SD-LABEL: saturating_6xi16:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    smull2 v2.4s, v1.8h, v0.8h
; CHECK-SD-NEXT:    movi v3.4s, #127, msl #8
; CHECK-SD-NEXT:    sqdmulh v0.4h, v1.4h, v0.4h
; CHECK-SD-NEXT:    sshr v2.4s, v2.4s, #15
; CHECK-SD-NEXT:    smin v1.4s, v2.4s, v3.4s
; CHECK-SD-NEXT:    xtn2 v0.8h, v1.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: saturating_6xi16:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    smov w8, v1.h[0]
; CHECK-GI-NEXT:    smov w9, v0.h[0]
; CHECK-GI-NEXT:    smov w10, v1.h[1]
; CHECK-GI-NEXT:    smov w11, v0.h[1]
; CHECK-GI-NEXT:    fmov s2, w8
; CHECK-GI-NEXT:    fmov s3, w9
; CHECK-GI-NEXT:    smov w8, v1.h[2]
; CHECK-GI-NEXT:    smov w9, v0.h[2]
; CHECK-GI-NEXT:    mov v2.s[1], w10
; CHECK-GI-NEXT:    mov v3.s[1], w11
; CHECK-GI-NEXT:    smov w10, v1.h[3]
; CHECK-GI-NEXT:    sshll2 v1.4s, v1.8h, #0
; CHECK-GI-NEXT:    mov v2.s[2], w8
; CHECK-GI-NEXT:    smov w8, v0.h[3]
; CHECK-GI-NEXT:    mov v3.s[2], w9
; CHECK-GI-NEXT:    sshll2 v0.4s, v0.8h, #0
; CHECK-GI-NEXT:    mov v2.s[3], w10
; CHECK-GI-NEXT:    mov v3.s[3], w8
; CHECK-GI-NEXT:    mul v0.2s, v1.2s, v0.2s
; CHECK-GI-NEXT:    mul v2.4s, v2.4s, v3.4s
; CHECK-GI-NEXT:    sshr v0.2s, v0.2s, #15
; CHECK-GI-NEXT:    sqxtn v0.4h, v0.4s
; CHECK-GI-NEXT:    sqshrn v3.4h, v2.4s, #15
; CHECK-GI-NEXT:    mov v2.h[0], v3.h[0]
; CHECK-GI-NEXT:    mov v2.h[1], v3.h[1]
; CHECK-GI-NEXT:    mov v2.h[2], v3.h[2]
; CHECK-GI-NEXT:    mov v2.h[3], v3.h[3]
; CHECK-GI-NEXT:    mov v2.h[4], v0.h[0]
; CHECK-GI-NEXT:    mov v2.h[5], v0.h[1]
; CHECK-GI-NEXT:    mov v0.16b, v2.16b
; CHECK-GI-NEXT:    ret
  %as = sext <6 x i16> %a to <6 x i32>
  %bs = sext <6 x i16> %b to <6 x i32>
  %m = mul nsw <6 x i32> %bs, %as
  %sh = ashr <6 x i32> %m, splat (i32 15)
  %ma = tail call <6 x i32> @llvm.smin.v6i32(<6 x i32> %sh, <6 x i32> splat (i32 32767))
  %t = trunc nsw <6 x i32> %ma to <6 x i16>
  ret <6 x i16> %t
}

define <4 x i16> @unsupported_saturation_value_v4i16(<4 x i16> %a, <4 x i16> %b) {
; CHECK-LABEL: unsupported_saturation_value_v4i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    smull v0.4s, v1.4h, v0.4h
; CHECK-NEXT:    movi v1.4s, #42
; CHECK-NEXT:    sshr v0.4s, v0.4s, #15
; CHECK-NEXT:    smin v0.4s, v0.4s, v1.4s
; CHECK-NEXT:    xtn v0.4h, v0.4s
; CHECK-NEXT:    ret
  %as = sext <4 x i16> %a to <4 x i32>
  %bs = sext <4 x i16> %b to <4 x i32>
  %m = mul nsw <4 x i32> %bs, %as
  %sh = ashr <4 x i32> %m, splat (i32 15)
  %ma = tail call <4 x i32> @llvm.smin.v4i32(<4 x i32> %sh, <4 x i32> splat (i32 42))
  %t = trunc nsw <4 x i32> %ma to <4 x i16>
  ret <4 x i16> %t
}

define <4 x i16> @unsupported_shift_value_v4i16(<4 x i16> %a, <4 x i16> %b) {
; CHECK-LABEL: unsupported_shift_value_v4i16:
; CHECK:       // %bb.0:
; CHECK-NEXT:    smull v0.4s, v1.4h, v0.4h
; CHECK-NEXT:    movi v1.4s, #127, msl #8
; CHECK-NEXT:    sshr v0.4s, v0.4s, #3
; CHECK-NEXT:    smin v0.4s, v0.4s, v1.4s
; CHECK-NEXT:    xtn v0.4h, v0.4s
; CHECK-NEXT:    ret
  %as = sext <4 x i16> %a to <4 x i32>
  %bs = sext <4 x i16> %b to <4 x i32>
  %m = mul nsw <4 x i32> %bs, %as
  %sh = ashr <4 x i32> %m, splat (i32 3)
  %ma = tail call <4 x i32> @llvm.smin.v4i32(<4 x i32> %sh, <4 x i32> splat (i32 32767))
  %t = trunc <4 x i32> %ma to <4 x i16>
  ret <4 x i16> %t
}

define <2 x i16> @extend_to_illegal_type(<2 x i16> %a, <2 x i16> %b) {
; CHECK-LABEL: extend_to_illegal_type:
; CHECK:       // %bb.0:
; CHECK-NEXT:    shl v0.2s, v0.2s, #16
; CHECK-NEXT:    shl v1.2s, v1.2s, #16
; CHECK-NEXT:    sshr v0.2s, v0.2s, #16
; CHECK-NEXT:    sshr v1.2s, v1.2s, #16
; CHECK-NEXT:    sqdmulh v0.2s, v1.2s, v0.2s
; CHECK-NEXT:    ret
  %as = sext <2 x i16> %a to <2 x i48>
  %bs = sext <2 x i16> %b to <2 x i48>
  %m = mul nsw <2 x i48> %bs, %as
  %sh = ashr <2 x i48> %m, splat (i48 15)
  %ma = tail call <2 x i48> @llvm.smin.v2i48(<2 x i48> %sh, <2 x i48> splat (i48 32767))
  %t = trunc nsw <2 x i48> %ma to <2 x i16>
  ret <2 x i16> %t
}

define <2 x i11> @illegal_source(<2 x i11> %a, <2 x i11> %b) {
; CHECK-LABEL: illegal_source:
; CHECK:       // %bb.0:
; CHECK-NEXT:    shl v0.2s, v0.2s, #21
; CHECK-NEXT:    shl v1.2s, v1.2s, #21
; CHECK-NEXT:    sshr v0.2s, v0.2s, #21
; CHECK-NEXT:    sshr v1.2s, v1.2s, #21
; CHECK-NEXT:    mul v0.2s, v1.2s, v0.2s
; CHECK-NEXT:    movi v1.2s, #127, msl #8
; CHECK-NEXT:    sshr v0.2s, v0.2s, #15
; CHECK-NEXT:    smin v0.2s, v0.2s, v1.2s
; CHECK-NEXT:    ret
  %as = sext <2 x i11> %a to <2 x i32>
  %bs = sext <2 x i11> %b to <2 x i32>
  %m = mul <2 x i32> %bs, %as
  %sh = ashr <2 x i32> %m, splat (i32 15)
  %ma = tail call <2 x i32> @llvm.smin.v2i32(<2 x i32> %sh, <2 x i32> splat (i32 32767))
  %t = trunc <2 x i32> %ma to <2 x i11>
  ret <2 x i11> %t
}
define <1 x i16> @saturating_1xi16(<1 x i16> %a, <1 x i16> %b) {
; CHECK-SD-LABEL: saturating_1xi16:
; CHECK-SD:       // %bb.0:
; CHECK-SD-NEXT:    zip1 v0.4h, v0.4h, v0.4h
; CHECK-SD-NEXT:    zip1 v1.4h, v1.4h, v0.4h
; CHECK-SD-NEXT:    shl v0.2s, v0.2s, #16
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #16
; CHECK-SD-NEXT:    shl v1.2s, v1.2s, #16
; CHECK-SD-NEXT:    sshr v1.2s, v1.2s, #16
; CHECK-SD-NEXT:    mul v0.2s, v1.2s, v0.2s
; CHECK-SD-NEXT:    movi v1.2s, #127, msl #8
; CHECK-SD-NEXT:    sshr v0.2s, v0.2s, #15
; CHECK-SD-NEXT:    smin v0.2s, v0.2s, v1.2s
; CHECK-SD-NEXT:    xtn v0.4h, v0.4s
; CHECK-SD-NEXT:    ret
;
; CHECK-GI-LABEL: saturating_1xi16:
; CHECK-GI:       // %bb.0:
; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
; CHECK-GI-NEXT:    // kill: def $d1 killed $d1 def $q1
; CHECK-GI-NEXT:    smov w8, v0.h[0]
; CHECK-GI-NEXT:    smov w9, v1.h[0]
; CHECK-GI-NEXT:    mul w8, w9, w8
; CHECK-GI-NEXT:    mov w9, #32767 // =0x7fff
; CHECK-GI-NEXT:    asr w8, w8, #15
; CHECK-GI-NEXT:    cmp w8, w9
; CHECK-GI-NEXT:    csel w8, w8, w9, lt
; CHECK-GI-NEXT:    fmov s0, w8
; CHECK-GI-NEXT:    ret
  %as = sext <1 x i16> %a to <1 x i32>
  %bs = sext <1 x i16> %b to <1 x i32>
  %m = mul nsw <1 x i32> %bs, %as
  %sh = ashr <1 x i32> %m, splat (i32 15)
  %ma = tail call <1 x i32> @llvm.smin.v1i32(<1 x i32> %sh, <1 x i32> splat (i32 32767))
  %t = trunc nsw <1 x i32> %ma to <1 x i16>
  ret <1 x i16> %t
}