1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fma | FileCheck %s --check-prefix=X32 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fma | FileCheck %s --check-prefix=X64 4 5; This test checks combinations of FNEG and FMA intrinsics 6 7define <8 x float> @test1(<8 x float> %a, <8 x float> %b, <8 x float> %c) { 8; X32-LABEL: test1: 9; X32: # %bb.0: # %entry 10; X32-NEXT: vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 11; X32-NEXT: retl 12; 13; X64-LABEL: test1: 14; X64: # %bb.0: # %entry 15; X64-NEXT: vfmsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) - ymm2 16; X64-NEXT: retq 17entry: 18 %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c 19 %0 = tail call <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %sub.i) #2 20 ret <8 x float> %0 21} 22 23declare <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>) 24 25define <4 x float> @test2(<4 x float> %a, <4 x float> %b, <4 x float> %c) { 26; X32-LABEL: test2: 27; X32: # %bb.0: # %entry 28; X32-NEXT: vfnmsub213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 29; X32-NEXT: retl 30; 31; X64-LABEL: test2: 32; X64: # %bb.0: # %entry 33; X64-NEXT: vfnmsub213ps {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 34; X64-NEXT: retq 35entry: 36 %0 = tail call <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2 37 %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %0 38 ret <4 x float> %sub.i 39} 40 41declare <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) 42 43define <4 x float> @test3(<4 x float> %a, <4 x float> %b, <4 x float> %c) { 44; X32-LABEL: test3: 45; X32: # %bb.0: # %entry 46; X32-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 47; X32-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0,-0,-0,-0] 48; X32-NEXT: vxorps %xmm1, %xmm0, %xmm0 49; X32-NEXT: retl 50; 51; X64-LABEL: test3: 52; X64: # %bb.0: # %entry 53; X64-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2 54; X64-NEXT: vbroadcastss {{.*#+}} xmm1 = [-0,-0,-0,-0] 55; X64-NEXT: vxorps %xmm1, %xmm0, %xmm0 56; X64-NEXT: retq 57entry: 58 %0 = tail call <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2 59 %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %0 60 ret <4 x float> %sub.i 61} 62 63declare <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) 64 65define <8 x float> @test4(<8 x float> %a, <8 x float> %b, <8 x float> %c) { 66; X32-LABEL: test4: 67; X32: # %bb.0: # %entry 68; X32-NEXT: vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 69; X32-NEXT: retl 70; 71; X64-LABEL: test4: 72; X64: # %bb.0: # %entry 73; X64-NEXT: vfnmadd213ps {{.*#+}} ymm0 = -(ymm1 * ymm0) + ymm2 74; X64-NEXT: retq 75entry: 76 %0 = tail call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2 77 %sub.i = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %0 78 ret <8 x float> %sub.i 79} 80 81define <8 x float> @test5(<8 x float> %a, <8 x float> %b, <8 x float> %c) { 82; X32-LABEL: test5: 83; X32: # %bb.0: # %entry 84; X32-NEXT: vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 85; X32-NEXT: retl 86; 87; X64-LABEL: test5: 88; X64: # %bb.0: # %entry 89; X64-NEXT: vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 90; X64-NEXT: retq 91entry: 92 %sub.c = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c 93 %0 = tail call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %sub.c) #2 94 ret <8 x float> %0 95} 96 97declare <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float>, <8 x float>, <8 x float>) 98 99 100define <2 x double> @test6(<2 x double> %a, <2 x double> %b, <2 x double> %c) { 101; X32-LABEL: test6: 102; X32: # %bb.0: # %entry 103; X32-NEXT: vfnmsub213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 104; X32-NEXT: retl 105; 106; X64-LABEL: test6: 107; X64: # %bb.0: # %entry 108; X64-NEXT: vfnmsub213pd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2 109; X64-NEXT: retq 110entry: 111 %0 = tail call <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) #2 112 %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %0 113 ret <2 x double> %sub.i 114} 115 116declare <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) 117 118define <8 x float> @test7(float %a, <8 x float> %b, <8 x float> %c) { 119; X32-LABEL: test7: 120; X32: # %bb.0: # %entry 121; X32-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero 122; X32-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero 123; X32-NEXT: vsubps %ymm2, %ymm3, %ymm2 124; X32-NEXT: vbroadcastss %xmm2, %ymm2 125; X32-NEXT: vfmadd213ps {{.*#+}} ymm0 = (ymm2 * ymm0) + ymm1 126; X32-NEXT: retl 127; 128; X64-LABEL: test7: 129; X64: # %bb.0: # %entry 130; X64-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0 131; X64-NEXT: vmovss {{.*#+}} xmm3 = mem[0],zero,zero,zero 132; X64-NEXT: vsubps %ymm0, %ymm3, %ymm0 133; X64-NEXT: vbroadcastss %xmm0, %ymm0 134; X64-NEXT: vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 135; X64-NEXT: retq 136entry: 137 %0 = insertelement <8 x float> undef, float %a, i32 0 138 %1 = fsub <8 x float> <float -0.000000e+00, float undef, float undef, float undef, float undef, float undef, float undef, float undef>, %0 139 %2 = shufflevector <8 x float> %1, <8 x float> undef, <8 x i32> zeroinitializer 140 %3 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %2, <8 x float> %b, <8 x float> %c) 141 ret <8 x float> %3 142 143} 144 145define <8 x float> @test8(float %a, <8 x float> %b, <8 x float> %c) { 146; X32-LABEL: test8: 147; X32: # %bb.0: # %entry 148; X32-NEXT: vmovss {{.*#+}} xmm2 = mem[0],zero,zero,zero 149; X32-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] 150; X32-NEXT: vxorps %xmm3, %xmm2, %xmm2 151; X32-NEXT: vbroadcastss %xmm2, %ymm2 152; X32-NEXT: vfmadd213ps {{.*#+}} ymm0 = (ymm2 * ymm0) + ymm1 153; X32-NEXT: retl 154; 155; X64-LABEL: test8: 156; X64: # %bb.0: # %entry 157; X64-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0,-0,-0,-0] 158; X64-NEXT: vxorps %xmm3, %xmm0, %xmm0 159; X64-NEXT: vbroadcastss %xmm0, %ymm0 160; X64-NEXT: vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2 161; X64-NEXT: retq 162entry: 163 %0 = fsub float -0.0, %a 164 %1 = insertelement <8 x float> undef, float %0, i32 0 165 %2 = shufflevector <8 x float> %1, <8 x float> undef, <8 x i32> zeroinitializer 166 %3 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %2, <8 x float> %b, <8 x float> %c) 167 ret <8 x float> %3 168} 169 170declare <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) 171