diff --git a/llvm/lib/Target/X86/X86Subtarget.h b/llvm/lib/Target/X86/X86Subtarget.h index 3f4cd7d0b1d7..c2c95658482d 100644 --- a/llvm/lib/Target/X86/X86Subtarget.h +++ b/llvm/lib/Target/X86/X86Subtarget.h @@ -435,9 +435,9 @@ public: bool hasPCLMUL() const { return HasPCLMUL; } // Prefer FMA4 to FMA - its better for commutation/memory folding and // has equal or better performance on all supported targets. - bool hasFMA() const { return HasFMA && !HasFMA4; } + bool hasFMA() const { return (HasFMA || hasAVX512()) && !HasFMA4; } bool hasFMA4() const { return HasFMA4; } - bool hasAnyFMA() const { return hasFMA() || hasFMA4() || hasAVX512(); } + bool hasAnyFMA() const { return hasFMA() || hasFMA4(); } bool hasXOP() const { return HasXOP; } bool hasTBM() const { return HasTBM; } bool hasMOVBE() const { return HasMOVBE; } diff --git a/llvm/test/CodeGen/X86/fma.ll b/llvm/test/CodeGen/X86/fma.ll index f5002c01846a..2c942347d54c 100644 --- a/llvm/test/CodeGen/X86/fma.ll +++ b/llvm/test/CodeGen/X86/fma.ll @@ -1,10 +1,10 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc < %s -mtriple=i386-apple-darwin10 -mattr=+sse2,+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=FMA32 -; RUN: llc < %s -mtriple=i386-apple-darwin10 -mattr=+sse2,-fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=FMACALL32 +; RUN: llc < %s -mtriple=i386-apple-darwin10 -mattr=+avx,+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=FMA32 +; RUN: llc < %s -mtriple=i386-apple-darwin10 -mattr=+avx,-fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=FMACALL32 ; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mattr=+fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=FMA64 ; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mattr=-fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=FMACALL64 -; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mattr=+avx512f,-fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=AVX51264NOFMA -; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mattr=+avx512f,fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=AVX51264 +; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mattr=+avx512f,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=AVX512 --check-prefix=AVX51264 +; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mattr=+avx512vl,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=AVX512VL ; RUN: llc < %s -mtriple=i386-apple-darwin10 -mcpu=bdver2 -mattr=-fma4 -show-mc-encoding | FileCheck %s --check-prefix=FMA32 ; RUN: llc < %s -mtriple=i386-apple-darwin10 -mcpu=bdver2 -mattr=-fma,-fma4 -show-mc-encoding | FileCheck %s --check-prefix=FMACALL32 @@ -39,15 +39,15 @@ define float @test_f32(float %a, float %b, float %c) #0 { ; FMACALL64-NEXT: ## encoding: [0xeb,A] ; FMACALL64-NEXT: ## fixup A - offset: 1, value: _fmaf-1, kind: FK_PCRel_1 ; -; AVX51264NOFMA-LABEL: test_f32: -; AVX51264NOFMA: ## BB#0: ## %entry -; AVX51264NOFMA-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa9,0xc2] -; AVX51264NOFMA-NEXT: retq ## encoding: [0xc3] +; AVX512-LABEL: test_f32: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa9,0xc2] +; AVX512-NEXT: retq ## encoding: [0xc3] ; -; AVX51264-LABEL: test_f32: -; AVX51264: ## BB#0: ## %entry -; AVX51264-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa9,0xc2] -; AVX51264-NEXT: retq ## encoding: [0xc3] +; AVX512VL-LABEL: test_f32: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa9,0xc2] +; AVX512VL-NEXT: retq ## encoding: [0xc3] entry: %call = call float @llvm.fma.f32(float %a, float %b, float %c) ret float %call @@ -84,15 +84,15 @@ define double @test_f64(double %a, double %b, double %c) #0 { ; FMACALL64-NEXT: ## encoding: [0xeb,A] ; FMACALL64-NEXT: ## fixup A - offset: 1, value: _fma-1, kind: FK_PCRel_1 ; -; AVX51264NOFMA-LABEL: test_f64: -; AVX51264NOFMA: ## BB#0: ## %entry -; AVX51264NOFMA-NEXT: vfmadd213sd %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa9,0xc2] -; AVX51264NOFMA-NEXT: retq ## encoding: [0xc3] +; AVX512-LABEL: test_f64: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: vfmadd213sd %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa9,0xc2] +; AVX512-NEXT: retq ## encoding: [0xc3] ; -; AVX51264-LABEL: test_f64: -; AVX51264: ## BB#0: ## %entry -; AVX51264-NEXT: vfmadd213sd %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa9,0xc2] -; AVX51264-NEXT: retq ## encoding: [0xc3] +; AVX512VL-LABEL: test_f64: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: vfmadd213sd %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa9,0xc2] +; AVX512VL-NEXT: retq ## encoding: [0xc3] entry: %call = call double @llvm.fma.f64(double %a, double %b, double %c) ret double %call @@ -155,33 +155,33 @@ define x86_fp80 @test_f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c) #0 { ; FMACALL64-NEXT: addq $56, %rsp ## encoding: [0x48,0x83,0xc4,0x38] ; FMACALL64-NEXT: retq ## encoding: [0xc3] ; -; AVX51264NOFMA-LABEL: test_f80: -; AVX51264NOFMA: ## BB#0: ## %entry -; AVX51264NOFMA-NEXT: subq $56, %rsp ## encoding: [0x48,0x83,0xec,0x38] -; AVX51264NOFMA-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x40] -; AVX51264NOFMA-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x50] -; AVX51264NOFMA-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x60] -; AVX51264NOFMA-NEXT: fstpt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x7c,0x24,0x20] -; AVX51264NOFMA-NEXT: fstpt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x7c,0x24,0x10] -; AVX51264NOFMA-NEXT: fstpt (%rsp) ## encoding: [0xdb,0x3c,0x24] -; AVX51264NOFMA-NEXT: callq _fmal ## encoding: [0xe8,A,A,A,A] -; AVX51264NOFMA-NEXT: ## fixup A - offset: 1, value: _fmal-4, kind: FK_PCRel_4 -; AVX51264NOFMA-NEXT: addq $56, %rsp ## encoding: [0x48,0x83,0xc4,0x38] -; AVX51264NOFMA-NEXT: retq ## encoding: [0xc3] +; AVX512-LABEL: test_f80: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: subq $56, %rsp ## encoding: [0x48,0x83,0xec,0x38] +; AVX512-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x40] +; AVX512-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x50] +; AVX512-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x60] +; AVX512-NEXT: fstpt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x7c,0x24,0x20] +; AVX512-NEXT: fstpt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x7c,0x24,0x10] +; AVX512-NEXT: fstpt (%rsp) ## encoding: [0xdb,0x3c,0x24] +; AVX512-NEXT: callq _fmal ## encoding: [0xe8,A,A,A,A] +; AVX512-NEXT: ## fixup A - offset: 1, value: _fmal-4, kind: FK_PCRel_4 +; AVX512-NEXT: addq $56, %rsp ## encoding: [0x48,0x83,0xc4,0x38] +; AVX512-NEXT: retq ## encoding: [0xc3] ; -; AVX51264-LABEL: test_f80: -; AVX51264: ## BB#0: ## %entry -; AVX51264-NEXT: subq $56, %rsp ## encoding: [0x48,0x83,0xec,0x38] -; AVX51264-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x40] -; AVX51264-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x50] -; AVX51264-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x60] -; AVX51264-NEXT: fstpt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x7c,0x24,0x20] -; AVX51264-NEXT: fstpt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x7c,0x24,0x10] -; AVX51264-NEXT: fstpt (%rsp) ## encoding: [0xdb,0x3c,0x24] -; AVX51264-NEXT: callq _fmal ## encoding: [0xe8,A,A,A,A] -; AVX51264-NEXT: ## fixup A - offset: 1, value: _fmal-4, kind: FK_PCRel_4 -; AVX51264-NEXT: addq $56, %rsp ## encoding: [0x48,0x83,0xc4,0x38] -; AVX51264-NEXT: retq ## encoding: [0xc3] +; AVX512VL-LABEL: test_f80: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: subq $56, %rsp ## encoding: [0x48,0x83,0xec,0x38] +; AVX512VL-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x40] +; AVX512VL-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x50] +; AVX512VL-NEXT: fldt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x6c,0x24,0x60] +; AVX512VL-NEXT: fstpt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x7c,0x24,0x20] +; AVX512VL-NEXT: fstpt {{[0-9]+}}(%rsp) ## encoding: [0xdb,0x7c,0x24,0x10] +; AVX512VL-NEXT: fstpt (%rsp) ## encoding: [0xdb,0x3c,0x24] +; AVX512VL-NEXT: callq _fmal ## encoding: [0xe8,A,A,A,A] +; AVX512VL-NEXT: ## fixup A - offset: 1, value: _fmal-4, kind: FK_PCRel_4 +; AVX512VL-NEXT: addq $56, %rsp ## encoding: [0x48,0x83,0xc4,0x38] +; AVX512VL-NEXT: retq ## encoding: [0xc3] entry: %call = call x86_fp80 @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c) ret x86_fp80 %call @@ -214,26 +214,200 @@ define float @test_f32_cst() #0 { ; FMACALL64-NEXT: ## xmm0 = mem[0],zero,zero,zero ; FMACALL64-NEXT: retq ## encoding: [0xc3] ; -; AVX51264NOFMA-LABEL: test_f32_cst: -; AVX51264NOFMA: ## BB#0: ## %entry -; AVX51264NOFMA-NEXT: vmovss {{.*}}(%rip), %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x05,A,A,A,A] -; AVX51264NOFMA-NEXT: ## fixup A - offset: 4, value: LCPI3_0-4, kind: reloc_riprel_4byte -; AVX51264NOFMA-NEXT: ## xmm0 = mem[0],zero,zero,zero -; AVX51264NOFMA-NEXT: retq ## encoding: [0xc3] +; AVX512-LABEL: test_f32_cst: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: vmovss {{.*}}(%rip), %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x05,A,A,A,A] +; AVX512-NEXT: ## fixup A - offset: 4, value: LCPI3_0-4, kind: reloc_riprel_4byte +; AVX512-NEXT: ## xmm0 = mem[0],zero,zero,zero +; AVX512-NEXT: retq ## encoding: [0xc3] ; -; AVX51264-LABEL: test_f32_cst: -; AVX51264: ## BB#0: ## %entry -; AVX51264-NEXT: vmovss {{.*}}(%rip), %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x05,A,A,A,A] -; AVX51264-NEXT: ## fixup A - offset: 4, value: LCPI3_0-4, kind: reloc_riprel_4byte -; AVX51264-NEXT: ## xmm0 = mem[0],zero,zero,zero -; AVX51264-NEXT: retq ## encoding: [0xc3] +; AVX512VL-LABEL: test_f32_cst: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: vmovss {{.*}}(%rip), %xmm0 ## EVEX TO VEX Compression encoding: [0xc5,0xfa,0x10,0x05,A,A,A,A] +; AVX512VL-NEXT: ## fixup A - offset: 4, value: LCPI3_0-4, kind: reloc_riprel_4byte +; AVX512VL-NEXT: ## xmm0 = mem[0],zero,zero,zero +; AVX512VL-NEXT: retq ## encoding: [0xc3] entry: %call = call float @llvm.fma.f32(float 3.0, float 3.0, float 3.0) ret float %call } +define <4 x float> @test_v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #0 { +; FMA32-LABEL: test_v4f32: +; FMA32: ## BB#0: ## %entry +; FMA32-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0xc2] +; FMA32-NEXT: retl ## encoding: [0xc3] +; +; FMA64-LABEL: test_v4f32: +; FMA64: ## BB#0: ## %entry +; FMA64-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0xc2] +; FMA64-NEXT: retq ## encoding: [0xc3] +; +; AVX512-LABEL: test_v4f32: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0xc2] +; AVX512-NEXT: retq ## encoding: [0xc3] +; +; AVX512VL-LABEL: test_v4f32: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: vfmadd213ps %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0xc2] +; AVX512VL-NEXT: retq ## encoding: [0xc3] +entry: + %call = call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) + ret <4 x float> %call +} + +define <8 x float> @test_v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #0 { +; FMA32-LABEL: test_v8f32: +; FMA32: ## BB#0: ## %entry +; FMA32-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0x75,0xa8,0xc2] +; FMA32-NEXT: retl ## encoding: [0xc3] +; +; FMA64-LABEL: test_v8f32: +; FMA64: ## BB#0: ## %entry +; FMA64-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0x75,0xa8,0xc2] +; FMA64-NEXT: retq ## encoding: [0xc3] +; +; AVX512-LABEL: test_v8f32: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0x75,0xa8,0xc2] +; AVX512-NEXT: retq ## encoding: [0xc3] +; +; AVX512VL-LABEL: test_v8f32: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: vfmadd213ps %ymm2, %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa8,0xc2] +; AVX512VL-NEXT: retq ## encoding: [0xc3] +entry: + %call = call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) + ret <8 x float> %call +} + +define <16 x float> @test_v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c) #0 { +; FMA32-LABEL: test_v16f32: +; FMA32: ## BB#0: ## %entry +; FMA32-NEXT: pushl %ebp ## encoding: [0x55] +; FMA32-NEXT: movl %esp, %ebp ## encoding: [0x89,0xe5] +; FMA32-NEXT: andl $-32, %esp ## encoding: [0x83,0xe4,0xe0] +; FMA32-NEXT: subl $32, %esp ## encoding: [0x83,0xec,0x20] +; FMA32-NEXT: vfmadd213ps 8(%ebp), %ymm2, %ymm0 ## encoding: [0xc4,0xe2,0x6d,0xa8,0x45,0x08] +; FMA32-NEXT: vfmadd213ps 40(%ebp), %ymm3, %ymm1 ## encoding: [0xc4,0xe2,0x65,0xa8,0x4d,0x28] +; FMA32-NEXT: movl %ebp, %esp ## encoding: [0x89,0xec] +; FMA32-NEXT: popl %ebp ## encoding: [0x5d] +; FMA32-NEXT: retl ## encoding: [0xc3] +; +; FMA64-LABEL: test_v16f32: +; FMA64: ## BB#0: ## %entry +; FMA64-NEXT: vfmadd213ps %ymm4, %ymm2, %ymm0 ## encoding: [0xc4,0xe2,0x6d,0xa8,0xc4] +; FMA64-NEXT: vfmadd213ps %ymm5, %ymm3, %ymm1 ## encoding: [0xc4,0xe2,0x65,0xa8,0xcd] +; FMA64-NEXT: retq ## encoding: [0xc3] +; +; AVX512-LABEL: test_v16f32: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: vfmadd213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xa8,0xc2] +; AVX512-NEXT: retq ## encoding: [0xc3] +; +; AVX512VL-LABEL: test_v16f32: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: vfmadd213ps %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0x75,0x48,0xa8,0xc2] +; AVX512VL-NEXT: retq ## encoding: [0xc3] +entry: + %call = call <16 x float> @llvm.fma.v16f32(<16 x float> %a, <16 x float> %b, <16 x float> %c) + ret <16 x float> %call +} + +define <2 x double> @test_v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #0 { +; FMA32-LABEL: test_v2f64: +; FMA32: ## BB#0: ## %entry +; FMA32-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0xf1,0xa8,0xc2] +; FMA32-NEXT: retl ## encoding: [0xc3] +; +; FMA64-LABEL: test_v2f64: +; FMA64: ## BB#0: ## %entry +; FMA64-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0xf1,0xa8,0xc2] +; FMA64-NEXT: retq ## encoding: [0xc3] +; +; AVX512-LABEL: test_v2f64: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0xf1,0xa8,0xc2] +; AVX512-NEXT: retq ## encoding: [0xc3] +; +; AVX512VL-LABEL: test_v2f64: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: vfmadd213pd %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa8,0xc2] +; AVX512VL-NEXT: retq ## encoding: [0xc3] +entry: + %call = call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) + ret <2 x double> %call +} + +define <4 x double> @test_v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #0 { +; FMA32-LABEL: test_v4f64: +; FMA32: ## BB#0: ## %entry +; FMA32-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0xf5,0xa8,0xc2] +; FMA32-NEXT: retl ## encoding: [0xc3] +; +; FMA64-LABEL: test_v4f64: +; FMA64: ## BB#0: ## %entry +; FMA64-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0xf5,0xa8,0xc2] +; FMA64-NEXT: retq ## encoding: [0xc3] +; +; AVX512-LABEL: test_v4f64: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0xf5,0xa8,0xc2] +; AVX512-NEXT: retq ## encoding: [0xc3] +; +; AVX512VL-LABEL: test_v4f64: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: vfmadd213pd %ymm2, %ymm1, %ymm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa8,0xc2] +; AVX512VL-NEXT: retq ## encoding: [0xc3] +entry: + %call = call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) + ret <4 x double> %call +} + +define <8 x double> @test_v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c) #0 { +; FMA32-LABEL: test_v8f64: +; FMA32: ## BB#0: ## %entry +; FMA32-NEXT: pushl %ebp ## encoding: [0x55] +; FMA32-NEXT: movl %esp, %ebp ## encoding: [0x89,0xe5] +; FMA32-NEXT: andl $-32, %esp ## encoding: [0x83,0xe4,0xe0] +; FMA32-NEXT: subl $32, %esp ## encoding: [0x83,0xec,0x20] +; FMA32-NEXT: vfmadd213pd 8(%ebp), %ymm2, %ymm0 ## encoding: [0xc4,0xe2,0xed,0xa8,0x45,0x08] +; FMA32-NEXT: vfmadd213pd 40(%ebp), %ymm3, %ymm1 ## encoding: [0xc4,0xe2,0xe5,0xa8,0x4d,0x28] +; FMA32-NEXT: movl %ebp, %esp ## encoding: [0x89,0xec] +; FMA32-NEXT: popl %ebp ## encoding: [0x5d] +; FMA32-NEXT: retl ## encoding: [0xc3] +; +; FMA64-LABEL: test_v8f64: +; FMA64: ## BB#0: ## %entry +; FMA64-NEXT: vfmadd213pd %ymm4, %ymm2, %ymm0 ## encoding: [0xc4,0xe2,0xed,0xa8,0xc4] +; FMA64-NEXT: vfmadd213pd %ymm5, %ymm3, %ymm1 ## encoding: [0xc4,0xe2,0xe5,0xa8,0xcd] +; FMA64-NEXT: retq ## encoding: [0xc3] +; +; AVX512-LABEL: test_v8f64: +; AVX512: ## BB#0: ## %entry +; AVX512-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xa8,0xc2] +; AVX512-NEXT: retq ## encoding: [0xc3] +; +; AVX512VL-LABEL: test_v8f64: +; AVX512VL: ## BB#0: ## %entry +; AVX512VL-NEXT: vfmadd213pd %zmm2, %zmm1, %zmm0 ## encoding: [0x62,0xf2,0xf5,0x48,0xa8,0xc2] +; AVX512VL-NEXT: retq ## encoding: [0xc3] +entry: + %call = call <8 x double> @llvm.fma.v8f64(<8 x double> %a, <8 x double> %b, <8 x double> %c) + ret <8 x double> %call +} + declare float @llvm.fma.f32(float, float, float) declare double @llvm.fma.f64(double, double, double) declare x86_fp80 @llvm.fma.f80(x86_fp80, x86_fp80, x86_fp80) +declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) +declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>) +declare <16 x float> @llvm.fma.v16f32(<16 x float>, <16 x float>, <16 x float>) + +declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) +declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) +declare <8 x double> @llvm.fma.v8f64(<8 x double>, <8 x double>, <8 x double>) + attributes #0 = { nounwind }