133 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -O3 -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+sha < %s | FileCheck %s3 4target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"5target triple = "x86_64-unknown-unknown"6 7; Stack reload folding tests.8;9; By including a nop call with sideeffects we can force a partial register spill of the10; relevant registers and check that the reload is correctly folded into the instruction.11 12define <4 x i32> @stack_fold_sha1msg1(<4 x i32> %a0, <4 x i32> %a1) {13; CHECK-LABEL: stack_fold_sha1msg1:14; CHECK: # %bb.0:15; CHECK-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill16; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill17; CHECK-NEXT: #APP18; CHECK-NEXT: nop19; CHECK-NEXT: #NO_APP20; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload21; CHECK-NEXT: sha1msg1 {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload22; CHECK-NEXT: retq23 %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()24 %2 = tail call <4 x i32> @llvm.x86.sha1msg1(<4 x i32> %a0, <4 x i32> %a1)25 ret <4 x i32> %226}27declare <4 x i32> @llvm.x86.sha1msg1(<4 x i32>, <4 x i32>) nounwind readnone28 29define <4 x i32> @stack_fold_sha1msg2(<4 x i32> %a0, <4 x i32> %a1) {30; CHECK-LABEL: stack_fold_sha1msg2:31; CHECK: # %bb.0:32; CHECK-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill33; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill34; CHECK-NEXT: #APP35; CHECK-NEXT: nop36; CHECK-NEXT: #NO_APP37; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload38; CHECK-NEXT: sha1msg2 {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload39; CHECK-NEXT: retq40 %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()41 %2 = tail call <4 x i32> @llvm.x86.sha1msg2(<4 x i32> %a0, <4 x i32> %a1)42 ret <4 x i32> %243}44declare <4 x i32> @llvm.x86.sha1msg2(<4 x i32>, <4 x i32>) nounwind readnone45 46define <4 x i32> @stack_fold_sha1nexte(<4 x i32> %a0, <4 x i32> %a1) {47; CHECK-LABEL: stack_fold_sha1nexte:48; CHECK: # %bb.0:49; CHECK-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill50; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill51; CHECK-NEXT: #APP52; CHECK-NEXT: nop53; CHECK-NEXT: #NO_APP54; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload55; CHECK-NEXT: sha1nexte {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload56; CHECK-NEXT: retq57 %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()58 %2 = tail call <4 x i32> @llvm.x86.sha1nexte(<4 x i32> %a0, <4 x i32> %a1)59 ret <4 x i32> %260}61declare <4 x i32> @llvm.x86.sha1nexte(<4 x i32>, <4 x i32>) nounwind readnone62 63define <4 x i32> @stack_fold_sha1rnds4(<4 x i32> %a0, <4 x i32> %a1) {64; CHECK-LABEL: stack_fold_sha1rnds4:65; CHECK: # %bb.0:66; CHECK-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill67; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill68; CHECK-NEXT: #APP69; CHECK-NEXT: nop70; CHECK-NEXT: #NO_APP71; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload72; CHECK-NEXT: sha1rnds4 $3, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload73; CHECK-NEXT: retq74 %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()75 %2 = tail call <4 x i32> @llvm.x86.sha1rnds4(<4 x i32> %a0, <4 x i32> %a1, i8 3)76 ret <4 x i32> %277}78declare <4 x i32> @llvm.x86.sha1rnds4(<4 x i32>, <4 x i32>, i8) nounwind readnone79 80define <4 x i32> @stack_fold_sha256msg1(<4 x i32> %a0, <4 x i32> %a1) {81; CHECK-LABEL: stack_fold_sha256msg1:82; CHECK: # %bb.0:83; CHECK-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill84; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill85; CHECK-NEXT: #APP86; CHECK-NEXT: nop87; CHECK-NEXT: #NO_APP88; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload89; CHECK-NEXT: sha256msg1 {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload90; CHECK-NEXT: retq91 %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()92 %2 = tail call <4 x i32> @llvm.x86.sha256msg1(<4 x i32> %a0, <4 x i32> %a1)93 ret <4 x i32> %294}95declare <4 x i32> @llvm.x86.sha256msg1(<4 x i32>, <4 x i32>) nounwind readnone96 97define <4 x i32> @stack_fold_sha256msg2(<4 x i32> %a0, <4 x i32> %a1) {98; CHECK-LABEL: stack_fold_sha256msg2:99; CHECK: # %bb.0:100; CHECK-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill101; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill102; CHECK-NEXT: #APP103; CHECK-NEXT: nop104; CHECK-NEXT: #NO_APP105; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload106; CHECK-NEXT: sha256msg2 {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload107; CHECK-NEXT: retq108 %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()109 %2 = tail call <4 x i32> @llvm.x86.sha256msg2(<4 x i32> %a0, <4 x i32> %a1)110 ret <4 x i32> %2111}112declare <4 x i32> @llvm.x86.sha256msg2(<4 x i32>, <4 x i32>) nounwind readnone113 114define <4 x i32> @stack_fold_sha256rnds2(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {115; CHECK-LABEL: stack_fold_sha256rnds2:116; CHECK: # %bb.0:117; CHECK-NEXT: movaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill118; CHECK-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill119; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill120; CHECK-NEXT: #APP121; CHECK-NEXT: nop122; CHECK-NEXT: #NO_APP123; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload124; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload125; CHECK-NEXT: sha256rnds2 %xmm0, {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Folded Reload126; CHECK-NEXT: movaps %xmm1, %xmm0127; CHECK-NEXT: retq128 %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm1},~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()129 %2 = tail call <4 x i32> @llvm.x86.sha256rnds2(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2)130 ret <4 x i32> %2131}132declare <4 x i32> @llvm.x86.sha256rnds2(<4 x i32>, <4 x i32>, <4 x i32>) nounwind readnone133