From 4628374ad1b65f9a43cc47928c134a68a3cbc9cb Mon Sep 17 00:00:00 2001 From: Simon Pilgrim Date: Sat, 8 Feb 2020 14:01:36 +0000 Subject: [PATCH] Add missing encoding comments from fma scalar folded intrinsics tests --- test/CodeGen/X86/fma-scalar-memfold.ll | 372 ++++++++++++++++++------- 1 file changed, 274 insertions(+), 98 deletions(-) diff --git a/test/CodeGen/X86/fma-scalar-memfold.ll b/test/CodeGen/X86/fma-scalar-memfold.ll index 0cdf2707664..06ef8293adf 100644 --- a/test/CodeGen/X86/fma-scalar-memfold.ll +++ b/test/CodeGen/X86/fma-scalar-memfold.ll @@ -1,6 +1,6 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc < %s -disable-peephole -mcpu=core-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=AVX2 -; RUN: llc < %s -disable-peephole -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 +; RUN: llc < %s -disable-peephole -mcpu=core-avx2 -show-mc-encoding | FileCheck %s --check-prefixes=CHECK,AVX2 +; RUN: llc < %s -disable-peephole -mcpu=skx -show-mc-encoding | FileCheck %s --check-prefixes=CHECK,AVX512 target triple = "x86_64-unknown-unknown" @@ -15,12 +15,23 @@ 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) { -; 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 +; AVX2-LABEL: fmadd_aab_ss: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovss (%rdi), %xmm0 # encoding: [0xc5,0xfa,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX2-NEXT: vfmadd213ss (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0xa9,0x06] +; AVX2-NEXT: # xmm0 = (xmm0 * xmm0) + mem +; AVX2-NEXT: vmovss %xmm0, (%rdi) # encoding: [0xc5,0xfa,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fmadd_aab_ss: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovss (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX512-NEXT: vfmadd213ss (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xa9,0x06] +; AVX512-NEXT: # xmm0 = (xmm0 * xmm0) + mem +; AVX512-NEXT: vmovss %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -41,12 +52,23 @@ define void @fmadd_aab_ss(float* %a, float* %b) { } define void @fmadd_aba_ss(float* %a, float* %b) { -; CHECK-LABEL: fmadd_aba_ss: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; CHECK-NEXT: vfmadd231ss {{.*#+}} xmm0 = (xmm0 * mem) + xmm0 -; CHECK-NEXT: vmovss %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fmadd_aba_ss: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovss (%rdi), %xmm0 # encoding: [0xc5,0xfa,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX2-NEXT: vfmadd231ss (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0xb9,0x06] +; AVX2-NEXT: # xmm0 = (xmm0 * mem) + xmm0 +; AVX2-NEXT: vmovss %xmm0, (%rdi) # encoding: [0xc5,0xfa,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fmadd_aba_ss: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovss (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX512-NEXT: vfmadd231ss (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xb9,0x06] +; AVX512-NEXT: # xmm0 = (xmm0 * mem) + xmm0 +; AVX512-NEXT: vmovss %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -67,12 +89,23 @@ define void @fmadd_aba_ss(float* %a, float* %b) { } define void @fmsub_aab_ss(float* %a, float* %b) { -; 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 +; AVX2-LABEL: fmsub_aab_ss: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovss (%rdi), %xmm0 # encoding: [0xc5,0xfa,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX2-NEXT: vfmsub213ss (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0xab,0x06] +; AVX2-NEXT: # xmm0 = (xmm0 * xmm0) - mem +; AVX2-NEXT: vmovss %xmm0, (%rdi) # encoding: [0xc5,0xfa,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fmsub_aab_ss: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovss (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX512-NEXT: vfmsub213ss (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xab,0x06] +; AVX512-NEXT: # xmm0 = (xmm0 * xmm0) - mem +; AVX512-NEXT: vmovss %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -93,12 +126,23 @@ define void @fmsub_aab_ss(float* %a, float* %b) { } define void @fmsub_aba_ss(float* %a, float* %b) { -; CHECK-LABEL: fmsub_aba_ss: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; CHECK-NEXT: vfmsub231ss {{.*#+}} xmm0 = (xmm0 * mem) - xmm0 -; CHECK-NEXT: vmovss %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fmsub_aba_ss: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovss (%rdi), %xmm0 # encoding: [0xc5,0xfa,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX2-NEXT: vfmsub231ss (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0xbb,0x06] +; AVX2-NEXT: # xmm0 = (xmm0 * mem) - xmm0 +; AVX2-NEXT: vmovss %xmm0, (%rdi) # encoding: [0xc5,0xfa,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fmsub_aba_ss: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovss (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX512-NEXT: vfmsub231ss (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xbb,0x06] +; AVX512-NEXT: # xmm0 = (xmm0 * mem) - xmm0 +; AVX512-NEXT: vmovss %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -119,12 +163,23 @@ define void @fmsub_aba_ss(float* %a, float* %b) { } define void @fnmadd_aab_ss(float* %a, float* %b) { -; 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 +; AVX2-LABEL: fnmadd_aab_ss: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovss (%rdi), %xmm0 # encoding: [0xc5,0xfa,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX2-NEXT: vfnmadd213ss (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0xad,0x06] +; AVX2-NEXT: # xmm0 = -(xmm0 * xmm0) + mem +; AVX2-NEXT: vmovss %xmm0, (%rdi) # encoding: [0xc5,0xfa,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fnmadd_aab_ss: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovss (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX512-NEXT: vfnmadd213ss (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xad,0x06] +; AVX512-NEXT: # xmm0 = -(xmm0 * xmm0) + mem +; AVX512-NEXT: vmovss %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -145,12 +200,23 @@ define void @fnmadd_aab_ss(float* %a, float* %b) { } define void @fnmadd_aba_ss(float* %a, float* %b) { -; CHECK-LABEL: fnmadd_aba_ss: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; CHECK-NEXT: vfnmadd231ss {{.*#+}} xmm0 = -(xmm0 * mem) + xmm0 -; CHECK-NEXT: vmovss %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fnmadd_aba_ss: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovss (%rdi), %xmm0 # encoding: [0xc5,0xfa,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX2-NEXT: vfnmadd231ss (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0xbd,0x06] +; AVX2-NEXT: # xmm0 = -(xmm0 * mem) + xmm0 +; AVX2-NEXT: vmovss %xmm0, (%rdi) # encoding: [0xc5,0xfa,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fnmadd_aba_ss: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovss (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX512-NEXT: vfnmadd231ss (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xbd,0x06] +; AVX512-NEXT: # xmm0 = -(xmm0 * mem) + xmm0 +; AVX512-NEXT: vmovss %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -171,12 +237,23 @@ define void @fnmadd_aba_ss(float* %a, float* %b) { } define void @fnmsub_aab_ss(float* %a, float* %b) { -; 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 +; AVX2-LABEL: fnmsub_aab_ss: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovss (%rdi), %xmm0 # encoding: [0xc5,0xfa,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX2-NEXT: vfnmsub213ss (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0xaf,0x06] +; AVX2-NEXT: # xmm0 = -(xmm0 * xmm0) - mem +; AVX2-NEXT: vmovss %xmm0, (%rdi) # encoding: [0xc5,0xfa,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fnmsub_aab_ss: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovss (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX512-NEXT: vfnmsub213ss (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xaf,0x06] +; AVX512-NEXT: # xmm0 = -(xmm0 * xmm0) - mem +; AVX512-NEXT: vmovss %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -197,12 +274,23 @@ define void @fnmsub_aab_ss(float* %a, float* %b) { } define void @fnmsub_aba_ss(float* %a, float* %b) { -; CHECK-LABEL: fnmsub_aba_ss: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; CHECK-NEXT: vfnmsub231ss {{.*#+}} xmm0 = -(xmm0 * mem) - xmm0 -; CHECK-NEXT: vmovss %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fnmsub_aba_ss: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovss (%rdi), %xmm0 # encoding: [0xc5,0xfa,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX2-NEXT: vfnmsub231ss (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0xbf,0x06] +; AVX2-NEXT: # xmm0 = -(xmm0 * mem) - xmm0 +; AVX2-NEXT: vmovss %xmm0, (%rdi) # encoding: [0xc5,0xfa,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fnmsub_aba_ss: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovss (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero,zero,zero +; AVX512-NEXT: vfnmsub231ss (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0xbf,0x06] +; AVX512-NEXT: # xmm0 = -(xmm0 * mem) - xmm0 +; AVX512-NEXT: vmovss %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfa,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -223,12 +311,23 @@ define void @fnmsub_aba_ss(float* %a, float* %b) { } define void @fmadd_aab_sd(double* %a, double* %b) { -; CHECK-LABEL: fmadd_aab_sd: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; CHECK-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm0 * xmm0) + mem -; CHECK-NEXT: vmovsd %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fmadd_aab_sd: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovsd (%rdi), %xmm0 # encoding: [0xc5,0xfb,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero +; AVX2-NEXT: vfmadd213sd (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0xf9,0xa9,0x06] +; AVX2-NEXT: # xmm0 = (xmm0 * xmm0) + mem +; AVX2-NEXT: vmovsd %xmm0, (%rdi) # encoding: [0xc5,0xfb,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fmadd_aab_sd: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovsd (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero +; AVX512-NEXT: vfmadd213sd (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xa9,0x06] +; AVX512-NEXT: # xmm0 = (xmm0 * xmm0) + mem +; AVX512-NEXT: vmovsd %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -245,12 +344,23 @@ define void @fmadd_aab_sd(double* %a, double* %b) { } define void @fmadd_aba_sd(double* %a, double* %b) { -; CHECK-LABEL: fmadd_aba_sd: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; CHECK-NEXT: vfmadd231sd {{.*#+}} xmm0 = (xmm0 * mem) + xmm0 -; CHECK-NEXT: vmovsd %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fmadd_aba_sd: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovsd (%rdi), %xmm0 # encoding: [0xc5,0xfb,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero +; AVX2-NEXT: vfmadd231sd (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0xf9,0xb9,0x06] +; AVX2-NEXT: # xmm0 = (xmm0 * mem) + xmm0 +; AVX2-NEXT: vmovsd %xmm0, (%rdi) # encoding: [0xc5,0xfb,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fmadd_aba_sd: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovsd (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero +; AVX512-NEXT: vfmadd231sd (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xb9,0x06] +; AVX512-NEXT: # xmm0 = (xmm0 * mem) + xmm0 +; AVX512-NEXT: vmovsd %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -267,12 +377,23 @@ define void @fmadd_aba_sd(double* %a, double* %b) { } define void @fmsub_aab_sd(double* %a, double* %b) { -; CHECK-LABEL: fmsub_aab_sd: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; CHECK-NEXT: vfmsub213sd {{.*#+}} xmm0 = (xmm0 * xmm0) - mem -; CHECK-NEXT: vmovsd %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fmsub_aab_sd: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovsd (%rdi), %xmm0 # encoding: [0xc5,0xfb,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero +; AVX2-NEXT: vfmsub213sd (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0xf9,0xab,0x06] +; AVX2-NEXT: # xmm0 = (xmm0 * xmm0) - mem +; AVX2-NEXT: vmovsd %xmm0, (%rdi) # encoding: [0xc5,0xfb,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fmsub_aab_sd: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovsd (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero +; AVX512-NEXT: vfmsub213sd (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xab,0x06] +; AVX512-NEXT: # xmm0 = (xmm0 * xmm0) - mem +; AVX512-NEXT: vmovsd %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -289,12 +410,23 @@ define void @fmsub_aab_sd(double* %a, double* %b) { } define void @fmsub_aba_sd(double* %a, double* %b) { -; CHECK-LABEL: fmsub_aba_sd: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; CHECK-NEXT: vfmsub231sd {{.*#+}} xmm0 = (xmm0 * mem) - xmm0 -; CHECK-NEXT: vmovsd %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fmsub_aba_sd: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovsd (%rdi), %xmm0 # encoding: [0xc5,0xfb,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero +; AVX2-NEXT: vfmsub231sd (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0xf9,0xbb,0x06] +; AVX2-NEXT: # xmm0 = (xmm0 * mem) - xmm0 +; AVX2-NEXT: vmovsd %xmm0, (%rdi) # encoding: [0xc5,0xfb,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fmsub_aba_sd: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovsd (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero +; AVX512-NEXT: vfmsub231sd (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xbb,0x06] +; AVX512-NEXT: # xmm0 = (xmm0 * mem) - xmm0 +; AVX512-NEXT: vmovsd %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -311,12 +443,23 @@ define void @fmsub_aba_sd(double* %a, double* %b) { } define void @fnmadd_aab_sd(double* %a, double* %b) { -; CHECK-LABEL: fnmadd_aab_sd: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; CHECK-NEXT: vfnmadd213sd {{.*#+}} xmm0 = -(xmm0 * xmm0) + mem -; CHECK-NEXT: vmovsd %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fnmadd_aab_sd: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovsd (%rdi), %xmm0 # encoding: [0xc5,0xfb,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero +; AVX2-NEXT: vfnmadd213sd (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0xf9,0xad,0x06] +; AVX2-NEXT: # xmm0 = -(xmm0 * xmm0) + mem +; AVX2-NEXT: vmovsd %xmm0, (%rdi) # encoding: [0xc5,0xfb,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fnmadd_aab_sd: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovsd (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero +; AVX512-NEXT: vfnmadd213sd (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xad,0x06] +; AVX512-NEXT: # xmm0 = -(xmm0 * xmm0) + mem +; AVX512-NEXT: vmovsd %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -333,12 +476,23 @@ define void @fnmadd_aab_sd(double* %a, double* %b) { } define void @fnmadd_aba_sd(double* %a, double* %b) { -; CHECK-LABEL: fnmadd_aba_sd: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; CHECK-NEXT: vfnmadd231sd {{.*#+}} xmm0 = -(xmm0 * mem) + xmm0 -; CHECK-NEXT: vmovsd %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fnmadd_aba_sd: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovsd (%rdi), %xmm0 # encoding: [0xc5,0xfb,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero +; AVX2-NEXT: vfnmadd231sd (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0xf9,0xbd,0x06] +; AVX2-NEXT: # xmm0 = -(xmm0 * mem) + xmm0 +; AVX2-NEXT: vmovsd %xmm0, (%rdi) # encoding: [0xc5,0xfb,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fnmadd_aba_sd: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovsd (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero +; AVX512-NEXT: vfnmadd231sd (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xbd,0x06] +; AVX512-NEXT: # xmm0 = -(xmm0 * mem) + xmm0 +; AVX512-NEXT: vmovsd %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -355,12 +509,23 @@ define void @fnmadd_aba_sd(double* %a, double* %b) { } define void @fnmsub_aab_sd(double* %a, double* %b) { -; CHECK-LABEL: fnmsub_aab_sd: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; CHECK-NEXT: vfnmsub213sd {{.*#+}} xmm0 = -(xmm0 * xmm0) - mem -; CHECK-NEXT: vmovsd %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fnmsub_aab_sd: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovsd (%rdi), %xmm0 # encoding: [0xc5,0xfb,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero +; AVX2-NEXT: vfnmsub213sd (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0xf9,0xaf,0x06] +; AVX2-NEXT: # xmm0 = -(xmm0 * xmm0) - mem +; AVX2-NEXT: vmovsd %xmm0, (%rdi) # encoding: [0xc5,0xfb,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fnmsub_aab_sd: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovsd (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero +; AVX512-NEXT: vfnmsub213sd (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xaf,0x06] +; AVX512-NEXT: # xmm0 = -(xmm0 * xmm0) - mem +; AVX512-NEXT: vmovsd %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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 @@ -377,12 +542,23 @@ define void @fnmsub_aab_sd(double* %a, double* %b) { } define void @fnmsub_aba_sd(double* %a, double* %b) { -; CHECK-LABEL: fnmsub_aba_sd: -; CHECK: # %bb.0: -; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero -; CHECK-NEXT: vfnmsub231sd {{.*#+}} xmm0 = -(xmm0 * mem) - xmm0 -; CHECK-NEXT: vmovsd %xmm0, (%rdi) -; CHECK-NEXT: retq +; AVX2-LABEL: fnmsub_aba_sd: +; AVX2: # %bb.0: +; AVX2-NEXT: vmovsd (%rdi), %xmm0 # encoding: [0xc5,0xfb,0x10,0x07] +; AVX2-NEXT: # xmm0 = mem[0],zero +; AVX2-NEXT: vfnmsub231sd (%rsi), %xmm0, %xmm0 # encoding: [0xc4,0xe2,0xf9,0xbf,0x06] +; AVX2-NEXT: # xmm0 = -(xmm0 * mem) - xmm0 +; AVX2-NEXT: vmovsd %xmm0, (%rdi) # encoding: [0xc5,0xfb,0x11,0x07] +; AVX2-NEXT: retq # encoding: [0xc3] +; +; AVX512-LABEL: fnmsub_aba_sd: +; AVX512: # %bb.0: +; AVX512-NEXT: vmovsd (%rdi), %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x10,0x07] +; AVX512-NEXT: # xmm0 = mem[0],zero +; AVX512-NEXT: vfnmsub231sd (%rsi), %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf9,0xbf,0x06] +; AVX512-NEXT: # xmm0 = -(xmm0 * mem) - xmm0 +; AVX512-NEXT: vmovsd %xmm0, (%rdi) # EVEX TO VEX Compression encoding: [0xc5,0xfb,0x11,0x07] +; AVX512-NEXT: retq # encoding: [0xc3] %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