From e4f868ea1625b06baf97d79bdb3de54e66af36b2 Mon Sep 17 00:00:00 2001 From: Craig Topper Date: Fri, 29 Jul 2016 06:06:04 +0000 Subject: [PATCH] [AVX512] Mark EVEX VMOVSSrm and VMOVSDrm as canFoldAsLoad and isReMaterializable. llvm-svn: 277120 --- llvm/lib/Target/X86/X86InstrAVX512.td | 1 + llvm/lib/Target/X86/X86InstrInfo.cpp | 2 + llvm/test/CodeGen/X86/fma-scalar-memfold.ll | 320 ++++++-------------- llvm/test/CodeGen/X86/fold-load-binops.ll | 112 ++----- 4 files changed, 131 insertions(+), 304 deletions(-) diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td index ebef4b194667..86ccc532514a 100644 --- a/llvm/lib/Target/X86/X86InstrAVX512.td +++ b/llvm/lib/Target/X86/X86InstrAVX512.td @@ -3050,6 +3050,7 @@ multiclass avx512_move_scalar, EVEX_4V, EVEX_K; + let canFoldAsLoad = 1, isReMaterializable = 1 in def rm : AVX512PI<0x10, MRMSrcMem, (outs _.FRC:$dst), (ins _.ScalarMemOp:$src), !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [(set _.FRC:$dst, (_.ScalarLdFrag addr:$src))], diff --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp index aca28c9c6c2b..44e4d5c9d72a 100644 --- a/llvm/lib/Target/X86/X86InstrInfo.cpp +++ b/llvm/lib/Target/X86/X86InstrInfo.cpp @@ -2617,6 +2617,8 @@ bool X86InstrInfo::isReallyTriviallyReMaterializable(const MachineInstr &MI, case X86::FsMOVAPSrm: case X86::FsMOVAPDrm: // AVX-512 + case X86::VMOVSSZrm: + case X86::VMOVSDZrm: case X86::VMOVAPDZ128rm: case X86::VMOVAPDZ256rm: case X86::VMOVAPDZrm: diff --git a/llvm/test/CodeGen/X86/fma-scalar-memfold.ll b/llvm/test/CodeGen/X86/fma-scalar-memfold.ll index c9e0721a1d35..23baeafe98dc 100644 --- a/llvm/test/CodeGen/X86/fma-scalar-memfold.ll +++ b/llvm/test/CodeGen/X86/fma-scalar-memfold.ll @@ -15,20 +15,12 @@ declare <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double>, <2 x double>, <2 x d declare <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double>, <2 x double>, <2 x double>) define void @fmadd_aab_ss(float* %a, float* %b) { -; AVX2-LABEL: fmadd_aab_ss: -; AVX2: # BB#0: -; AVX2-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX2-NEXT: vfmadd213ss (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovss %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fmadd_aab_ss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vfmadd213ss %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: vmovss %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fmadd_aab_ss: +; CHECK: # BB#0: +; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; CHECK-NEXT: vfmadd213ss (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovss %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load float, float* %a %av0 = insertelement <4 x float> undef, float %a.val, i32 0 %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1 @@ -49,20 +41,12 @@ define void @fmadd_aab_ss(float* %a, float* %b) { } define void @fmadd_aba_ss(float* %a, float* %b) { -; AVX2-LABEL: fmadd_aba_ss: -; AVX2: # BB#0: -; AVX2-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX2-NEXT: vfmadd132ss (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovss %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fmadd_aba_ss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vfmadd213ss %xmm0, %xmm1, %xmm0 -; AVX512-NEXT: vmovss %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fmadd_aba_ss: +; CHECK: # BB#0: +; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; CHECK-NEXT: vfmadd132ss (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovss %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load float, float* %a %av0 = insertelement <4 x float> undef, float %a.val, i32 0 %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1 @@ -83,20 +67,12 @@ define void @fmadd_aba_ss(float* %a, float* %b) { } define void @fmsub_aab_ss(float* %a, float* %b) { -; AVX2-LABEL: fmsub_aab_ss: -; AVX2: # BB#0: -; AVX2-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX2-NEXT: vfmsub213ss (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovss %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fmsub_aab_ss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vfmsub213ss %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: vmovss %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fmsub_aab_ss: +; CHECK: # BB#0: +; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; CHECK-NEXT: vfmsub213ss (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovss %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load float, float* %a %av0 = insertelement <4 x float> undef, float %a.val, i32 0 %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1 @@ -117,20 +93,12 @@ define void @fmsub_aab_ss(float* %a, float* %b) { } define void @fmsub_aba_ss(float* %a, float* %b) { -; AVX2-LABEL: fmsub_aba_ss: -; AVX2: # BB#0: -; AVX2-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX2-NEXT: vfmsub132ss (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovss %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fmsub_aba_ss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vfmsub213ss %xmm0, %xmm1, %xmm0 -; AVX512-NEXT: vmovss %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fmsub_aba_ss: +; CHECK: # BB#0: +; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; CHECK-NEXT: vfmsub132ss (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovss %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load float, float* %a %av0 = insertelement <4 x float> undef, float %a.val, i32 0 %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1 @@ -151,20 +119,12 @@ define void @fmsub_aba_ss(float* %a, float* %b) { } define void @fnmadd_aab_ss(float* %a, float* %b) { -; AVX2-LABEL: fnmadd_aab_ss: -; AVX2: # BB#0: -; AVX2-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX2-NEXT: vfnmadd213ss (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovss %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fnmadd_aab_ss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vfnmadd213ss %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: vmovss %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fnmadd_aab_ss: +; CHECK: # BB#0: +; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; CHECK-NEXT: vfnmadd213ss (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovss %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load float, float* %a %av0 = insertelement <4 x float> undef, float %a.val, i32 0 %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1 @@ -185,20 +145,12 @@ define void @fnmadd_aab_ss(float* %a, float* %b) { } define void @fnmadd_aba_ss(float* %a, float* %b) { -; AVX2-LABEL: fnmadd_aba_ss: -; AVX2: # BB#0: -; AVX2-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX2-NEXT: vfnmadd132ss (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovss %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fnmadd_aba_ss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vfnmadd213ss %xmm0, %xmm1, %xmm0 -; AVX512-NEXT: vmovss %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fnmadd_aba_ss: +; CHECK: # BB#0: +; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; CHECK-NEXT: vfnmadd132ss (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovss %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load float, float* %a %av0 = insertelement <4 x float> undef, float %a.val, i32 0 %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1 @@ -219,20 +171,12 @@ define void @fnmadd_aba_ss(float* %a, float* %b) { } define void @fnmsub_aab_ss(float* %a, float* %b) { -; AVX2-LABEL: fnmsub_aab_ss: -; AVX2: # BB#0: -; AVX2-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX2-NEXT: vfnmsub213ss (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovss %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fnmsub_aab_ss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vfnmsub213ss %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: vmovss %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fnmsub_aab_ss: +; CHECK: # BB#0: +; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; CHECK-NEXT: vfnmsub213ss (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovss %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load float, float* %a %av0 = insertelement <4 x float> undef, float %a.val, i32 0 %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1 @@ -253,20 +197,12 @@ define void @fnmsub_aab_ss(float* %a, float* %b) { } define void @fnmsub_aba_ss(float* %a, float* %b) { -; AVX2-LABEL: fnmsub_aba_ss: -; AVX2: # BB#0: -; AVX2-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX2-NEXT: vfnmsub132ss (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovss %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fnmsub_aba_ss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vfnmsub213ss %xmm0, %xmm1, %xmm0 -; AVX512-NEXT: vmovss %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fnmsub_aba_ss: +; CHECK: # BB#0: +; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; CHECK-NEXT: vfnmsub132ss (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovss %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load float, float* %a %av0 = insertelement <4 x float> undef, float %a.val, i32 0 %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1 @@ -287,20 +223,12 @@ define void @fnmsub_aba_ss(float* %a, float* %b) { } define void @fmadd_aab_sd(double* %a, double* %b) { -; AVX2-LABEL: fmadd_aab_sd: -; AVX2: # BB#0: -; AVX2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX2-NEXT: vfmadd213sd (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovlpd %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fmadd_aab_sd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vfmadd213sd %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: vmovlpd %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fmadd_aab_sd: +; CHECK: # BB#0: +; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero +; CHECK-NEXT: vfmadd213sd (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovlpd %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load double, double* %a %av0 = insertelement <2 x double> undef, double %a.val, i32 0 %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1 @@ -317,20 +245,12 @@ define void @fmadd_aab_sd(double* %a, double* %b) { } define void @fmadd_aba_sd(double* %a, double* %b) { -; AVX2-LABEL: fmadd_aba_sd: -; AVX2: # BB#0: -; AVX2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX2-NEXT: vfmadd132sd (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovlpd %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fmadd_aba_sd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vfmadd213sd %xmm0, %xmm1, %xmm0 -; AVX512-NEXT: vmovlpd %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fmadd_aba_sd: +; CHECK: # BB#0: +; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero +; CHECK-NEXT: vfmadd132sd (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovlpd %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load double, double* %a %av0 = insertelement <2 x double> undef, double %a.val, i32 0 %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1 @@ -347,20 +267,12 @@ define void @fmadd_aba_sd(double* %a, double* %b) { } define void @fmsub_aab_sd(double* %a, double* %b) { -; AVX2-LABEL: fmsub_aab_sd: -; AVX2: # BB#0: -; AVX2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX2-NEXT: vfmsub213sd (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovlpd %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fmsub_aab_sd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vfmsub213sd %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: vmovlpd %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fmsub_aab_sd: +; CHECK: # BB#0: +; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero +; CHECK-NEXT: vfmsub213sd (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovlpd %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load double, double* %a %av0 = insertelement <2 x double> undef, double %a.val, i32 0 %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1 @@ -377,20 +289,12 @@ define void @fmsub_aab_sd(double* %a, double* %b) { } define void @fmsub_aba_sd(double* %a, double* %b) { -; AVX2-LABEL: fmsub_aba_sd: -; AVX2: # BB#0: -; AVX2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX2-NEXT: vfmsub132sd (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovlpd %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fmsub_aba_sd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vfmsub213sd %xmm0, %xmm1, %xmm0 -; AVX512-NEXT: vmovlpd %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fmsub_aba_sd: +; CHECK: # BB#0: +; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero +; CHECK-NEXT: vfmsub132sd (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovlpd %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load double, double* %a %av0 = insertelement <2 x double> undef, double %a.val, i32 0 %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1 @@ -407,20 +311,12 @@ define void @fmsub_aba_sd(double* %a, double* %b) { } define void @fnmadd_aab_sd(double* %a, double* %b) { -; AVX2-LABEL: fnmadd_aab_sd: -; AVX2: # BB#0: -; AVX2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX2-NEXT: vfnmadd213sd (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovlpd %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fnmadd_aab_sd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vfnmadd213sd %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: vmovlpd %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fnmadd_aab_sd: +; CHECK: # BB#0: +; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero +; CHECK-NEXT: vfnmadd213sd (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovlpd %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load double, double* %a %av0 = insertelement <2 x double> undef, double %a.val, i32 0 %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1 @@ -437,20 +333,12 @@ define void @fnmadd_aab_sd(double* %a, double* %b) { } define void @fnmadd_aba_sd(double* %a, double* %b) { -; AVX2-LABEL: fnmadd_aba_sd: -; AVX2: # BB#0: -; AVX2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX2-NEXT: vfnmadd132sd (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovlpd %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fnmadd_aba_sd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vfnmadd213sd %xmm0, %xmm1, %xmm0 -; AVX512-NEXT: vmovlpd %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fnmadd_aba_sd: +; CHECK: # BB#0: +; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero +; CHECK-NEXT: vfnmadd132sd (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovlpd %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load double, double* %a %av0 = insertelement <2 x double> undef, double %a.val, i32 0 %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1 @@ -467,20 +355,12 @@ define void @fnmadd_aba_sd(double* %a, double* %b) { } define void @fnmsub_aab_sd(double* %a, double* %b) { -; AVX2-LABEL: fnmsub_aab_sd: -; AVX2: # BB#0: -; AVX2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX2-NEXT: vfnmsub213sd (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovlpd %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fnmsub_aab_sd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vfnmsub213sd %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: vmovlpd %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fnmsub_aab_sd: +; CHECK: # BB#0: +; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero +; CHECK-NEXT: vfnmsub213sd (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovlpd %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load double, double* %a %av0 = insertelement <2 x double> undef, double %a.val, i32 0 %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1 @@ -497,20 +377,12 @@ define void @fnmsub_aab_sd(double* %a, double* %b) { } define void @fnmsub_aba_sd(double* %a, double* %b) { -; AVX2-LABEL: fnmsub_aba_sd: -; AVX2: # BB#0: -; AVX2-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX2-NEXT: vfnmsub132sd (%rsi), %xmm0, %xmm0 -; AVX2-NEXT: vmovlpd %xmm0, (%rdi) -; AVX2-NEXT: retq -; -; AVX512-LABEL: fnmsub_aba_sd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vfnmsub213sd %xmm0, %xmm1, %xmm0 -; AVX512-NEXT: vmovlpd %xmm0, (%rdi) -; AVX512-NEXT: retq +; CHECK-LABEL: fnmsub_aba_sd: +; CHECK: # BB#0: +; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero +; CHECK-NEXT: vfnmsub132sd (%rsi), %xmm0, %xmm0 +; CHECK-NEXT: vmovlpd %xmm0, (%rdi) +; CHECK-NEXT: retq %a.val = load double, double* %a %av0 = insertelement <2 x double> undef, double %a.val, i32 0 %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1 diff --git a/llvm/test/CodeGen/X86/fold-load-binops.ll b/llvm/test/CodeGen/X86/fold-load-binops.ll index c6fc64b43874..4662a1521a38 100644 --- a/llvm/test/CodeGen/X86/fold-load-binops.ll +++ b/llvm/test/CodeGen/X86/fold-load-binops.ll @@ -13,16 +13,10 @@ define <4 x float> @addss(<4 x float> %va, float* %pb) { ; SSE-NEXT: addss (%rdi), %xmm0 ; SSE-NEXT: retq ; -; AVX1-LABEL: addss: -; AVX1: # BB#0: -; AVX1-NEXT: vaddss (%rdi), %xmm0, %xmm0 -; AVX1-NEXT: retq -; -; AVX512-LABEL: addss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vaddss %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: retq +; AVX-LABEL: addss: +; AVX: # BB#0: +; AVX-NEXT: vaddss (%rdi), %xmm0, %xmm0 +; AVX-NEXT: retq %a = extractelement <4 x float> %va, i32 0 %b = load float, float* %pb %r = fadd float %a, %b @@ -36,16 +30,10 @@ define <2 x double> @addsd(<2 x double> %va, double* %pb) { ; SSE-NEXT: addsd (%rdi), %xmm0 ; SSE-NEXT: retq ; -; AVX1-LABEL: addsd: -; AVX1: # BB#0: -; AVX1-NEXT: vaddsd (%rdi), %xmm0, %xmm0 -; AVX1-NEXT: retq -; -; AVX512-LABEL: addsd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vaddsd %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: retq +; AVX-LABEL: addsd: +; AVX: # BB#0: +; AVX-NEXT: vaddsd (%rdi), %xmm0, %xmm0 +; AVX-NEXT: retq %a = extractelement <2 x double> %va, i32 0 %b = load double, double* %pb %r = fadd double %a, %b @@ -59,16 +47,10 @@ define <4 x float> @subss(<4 x float> %va, float* %pb) { ; SSE-NEXT: subss (%rdi), %xmm0 ; SSE-NEXT: retq ; -; AVX1-LABEL: subss: -; AVX1: # BB#0: -; AVX1-NEXT: vsubss (%rdi), %xmm0, %xmm0 -; AVX1-NEXT: retq -; -; AVX512-LABEL: subss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vsubss %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: retq +; AVX-LABEL: subss: +; AVX: # BB#0: +; AVX-NEXT: vsubss (%rdi), %xmm0, %xmm0 +; AVX-NEXT: retq %a = extractelement <4 x float> %va, i32 0 %b = load float, float* %pb %r = fsub float %a, %b @@ -82,16 +64,10 @@ define <2 x double> @subsd(<2 x double> %va, double* %pb) { ; SSE-NEXT: subsd (%rdi), %xmm0 ; SSE-NEXT: retq ; -; AVX1-LABEL: subsd: -; AVX1: # BB#0: -; AVX1-NEXT: vsubsd (%rdi), %xmm0, %xmm0 -; AVX1-NEXT: retq -; -; AVX512-LABEL: subsd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vsubsd %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: retq +; AVX-LABEL: subsd: +; AVX: # BB#0: +; AVX-NEXT: vsubsd (%rdi), %xmm0, %xmm0 +; AVX-NEXT: retq %a = extractelement <2 x double> %va, i32 0 %b = load double, double* %pb %r = fsub double %a, %b @@ -105,16 +81,10 @@ define <4 x float> @mulss(<4 x float> %va, float* %pb) { ; SSE-NEXT: mulss (%rdi), %xmm0 ; SSE-NEXT: retq ; -; AVX1-LABEL: mulss: -; AVX1: # BB#0: -; AVX1-NEXT: vmulss (%rdi), %xmm0, %xmm0 -; AVX1-NEXT: retq -; -; AVX512-LABEL: mulss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vmulss %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: retq +; AVX-LABEL: mulss: +; AVX: # BB#0: +; AVX-NEXT: vmulss (%rdi), %xmm0, %xmm0 +; AVX-NEXT: retq %a = extractelement <4 x float> %va, i32 0 %b = load float, float* %pb %r = fmul float %a, %b @@ -128,16 +98,10 @@ define <2 x double> @mulsd(<2 x double> %va, double* %pb) { ; SSE-NEXT: mulsd (%rdi), %xmm0 ; SSE-NEXT: retq ; -; AVX1-LABEL: mulsd: -; AVX1: # BB#0: -; AVX1-NEXT: vmulsd (%rdi), %xmm0, %xmm0 -; AVX1-NEXT: retq -; -; AVX512-LABEL: mulsd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vmulsd %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: retq +; AVX-LABEL: mulsd: +; AVX: # BB#0: +; AVX-NEXT: vmulsd (%rdi), %xmm0, %xmm0 +; AVX-NEXT: retq %a = extractelement <2 x double> %va, i32 0 %b = load double, double* %pb %r = fmul double %a, %b @@ -151,16 +115,10 @@ define <4 x float> @divss(<4 x float> %va, float* %pb) { ; SSE-NEXT: divss (%rdi), %xmm0 ; SSE-NEXT: retq ; -; AVX1-LABEL: divss: -; AVX1: # BB#0: -; AVX1-NEXT: vdivss (%rdi), %xmm0, %xmm0 -; AVX1-NEXT: retq -; -; AVX512-LABEL: divss: -; AVX512: # BB#0: -; AVX512-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; AVX512-NEXT: vdivss %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: retq +; AVX-LABEL: divss: +; AVX: # BB#0: +; AVX-NEXT: vdivss (%rdi), %xmm0, %xmm0 +; AVX-NEXT: retq %a = extractelement <4 x float> %va, i32 0 %b = load float, float* %pb %r = fdiv float %a, %b @@ -174,16 +132,10 @@ define <2 x double> @divsd(<2 x double> %va, double* %pb) { ; SSE-NEXT: divsd (%rdi), %xmm0 ; SSE-NEXT: retq ; -; AVX1-LABEL: divsd: -; AVX1: # BB#0: -; AVX1-NEXT: vdivsd (%rdi), %xmm0, %xmm0 -; AVX1-NEXT: retq -; -; AVX512-LABEL: divsd: -; AVX512: # BB#0: -; AVX512-NEXT: vmovsd {{.*#+}} xmm1 = mem[0],zero -; AVX512-NEXT: vdivsd %xmm1, %xmm0, %xmm0 -; AVX512-NEXT: retq +; AVX-LABEL: divsd: +; AVX: # BB#0: +; AVX-NEXT: vdivsd (%rdi), %xmm0, %xmm0 +; AVX-NEXT: retq %a = extractelement <2 x double> %va, i32 0 %b = load double, double* %pb %r = fdiv double %a, %b