431 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+fma,-fma4 | FileCheck %s --check-prefix=CHECK3 4define <4 x float> @test_mm_fmadd_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {5; CHECK-LABEL: test_mm_fmadd_ps:6; CHECK: # %bb.0: # %entry7; CHECK-NEXT: vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm28; CHECK-NEXT: retq9entry:10 %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #211 ret <4 x float> %012}13 14define <2 x double> @test_mm_fmadd_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {15; CHECK-LABEL: test_mm_fmadd_pd:16; CHECK: # %bb.0: # %entry17; CHECK-NEXT: vfmadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm218; CHECK-NEXT: retq19entry:20 %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #221 ret <2 x double> %022}23 24define <4 x float> @test_mm_fmadd_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {25; CHECK-LABEL: test_mm_fmadd_ss:26; CHECK: # %bb.0: # %entry27; CHECK-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm228; CHECK-NEXT: retq29entry:30 %0 = extractelement <4 x float> %a, i64 031 %1 = extractelement <4 x float> %b, i64 032 %2 = extractelement <4 x float> %c, i64 033 %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #234 %4 = insertelement <4 x float> %a, float %3, i64 035 ret <4 x float> %436}37 38define <2 x double> @test_mm_fmadd_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {39; CHECK-LABEL: test_mm_fmadd_sd:40; CHECK: # %bb.0: # %entry41; CHECK-NEXT: vfmadd213sd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm242; CHECK-NEXT: retq43entry:44 %0 = extractelement <2 x double> %a, i64 045 %1 = extractelement <2 x double> %b, i64 046 %2 = extractelement <2 x double> %c, i64 047 %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #248 %4 = insertelement <2 x double> %a, double %3, i64 049 ret <2 x double> %450}51 52define <4 x float> @test_mm_fmsub_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {53; CHECK-LABEL: test_mm_fmsub_ps:54; CHECK: # %bb.0: # %entry55; CHECK-NEXT: vxorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm256; CHECK-NEXT: vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm257; CHECK-NEXT: retq58entry:59 %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c60 %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %sub.i) #261 ret <4 x float> %062}63 64define <2 x double> @test_mm_fmsub_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {65; CHECK-LABEL: test_mm_fmsub_pd:66; CHECK: # %bb.0: # %entry67; CHECK-NEXT: vxorpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm268; CHECK-NEXT: vfmadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm269; CHECK-NEXT: retq70entry:71 %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c72 %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %sub.i) #273 ret <2 x double> %074}75 76define <4 x float> @test_mm_fmsub_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {77; CHECK-LABEL: test_mm_fmsub_ss:78; CHECK: # %bb.0: # %entry79; CHECK-NEXT: vfmsub213ss {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm280; CHECK-NEXT: retq81entry:82 %0 = extractelement <4 x float> %a, i64 083 %1 = extractelement <4 x float> %b, i64 084 %.rhs.i = extractelement <4 x float> %c, i64 085 %2 = fsub float -0.000000e+00, %.rhs.i86 %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #287 %4 = insertelement <4 x float> %a, float %3, i64 088 ret <4 x float> %489}90 91define <2 x double> @test_mm_fmsub_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {92; CHECK-LABEL: test_mm_fmsub_sd:93; CHECK: # %bb.0: # %entry94; CHECK-NEXT: vfmsub213sd {{.*#+}} xmm0 = (xmm1 * xmm0) - xmm295; CHECK-NEXT: retq96entry:97 %0 = extractelement <2 x double> %a, i64 098 %1 = extractelement <2 x double> %b, i64 099 %.rhs.i = extractelement <2 x double> %c, i64 0100 %2 = fsub double -0.000000e+00, %.rhs.i101 %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2102 %4 = insertelement <2 x double> %a, double %3, i64 0103 ret <2 x double> %4104}105 106define <4 x float> @test_mm_fnmadd_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {107; CHECK-LABEL: test_mm_fnmadd_ps:108; CHECK: # %bb.0: # %entry109; CHECK-NEXT: vxorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0110; CHECK-NEXT: vfmadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2111; CHECK-NEXT: retq112entry:113 %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a114 %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %b, <4 x float> %c) #2115 ret <4 x float> %0116}117 118define <2 x double> @test_mm_fnmadd_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {119; CHECK-LABEL: test_mm_fnmadd_pd:120; CHECK: # %bb.0: # %entry121; CHECK-NEXT: vxorpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0122; CHECK-NEXT: vfmadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm2123; CHECK-NEXT: retq124entry:125 %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a126 %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %b, <2 x double> %c) #2127 ret <2 x double> %0128}129 130define <4 x float> @test_mm_fnmadd_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {131; CHECK-LABEL: test_mm_fnmadd_ss:132; CHECK: # %bb.0: # %entry133; CHECK-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2134; CHECK-NEXT: retq135entry:136 %0 = extractelement <4 x float> %a, i64 0137 %.rhs.i = extractelement <4 x float> %b, i64 0138 %1 = fsub float -0.000000e+00, %.rhs.i139 %2 = extractelement <4 x float> %c, i64 0140 %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2141 %4 = insertelement <4 x float> %a, float %3, i64 0142 ret <4 x float> %4143}144 145define <2 x double> @test_mm_fnmadd_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {146; CHECK-LABEL: test_mm_fnmadd_sd:147; CHECK: # %bb.0: # %entry148; CHECK-NEXT: vfnmadd213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm2149; CHECK-NEXT: retq150entry:151 %0 = extractelement <2 x double> %a, i64 0152 %.rhs.i = extractelement <2 x double> %b, i64 0153 %1 = fsub double -0.000000e+00, %.rhs.i154 %2 = extractelement <2 x double> %c, i64 0155 %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2156 %4 = insertelement <2 x double> %a, double %3, i64 0157 ret <2 x double> %4158}159 160define <4 x float> @test_mm_fnmsub_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {161; CHECK-LABEL: test_mm_fnmsub_ps:162; CHECK: # %bb.0: # %entry163; CHECK-NEXT: vbroadcastss {{.*#+}} xmm3 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]164; CHECK-NEXT: vxorps %xmm3, %xmm0, %xmm4165; CHECK-NEXT: vxorps %xmm3, %xmm2, %xmm0166; CHECK-NEXT: vfmadd231ps {{.*#+}} xmm0 = (xmm1 * xmm4) + xmm0167; CHECK-NEXT: retq168entry:169 %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a170 %sub1.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c171 %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %sub.i, <4 x float> %b, <4 x float> %sub1.i) #2172 ret <4 x float> %0173}174 175define <2 x double> @test_mm_fnmsub_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {176; CHECK-LABEL: test_mm_fnmsub_pd:177; CHECK: # %bb.0: # %entry178; CHECK-NEXT: vmovddup {{.*#+}} xmm3 = [-0.0E+0,-0.0E+0]179; CHECK-NEXT: # xmm3 = mem[0,0]180; CHECK-NEXT: vxorpd %xmm3, %xmm0, %xmm4181; CHECK-NEXT: vxorpd %xmm3, %xmm2, %xmm0182; CHECK-NEXT: vfmadd231pd {{.*#+}} xmm0 = (xmm1 * xmm4) + xmm0183; CHECK-NEXT: retq184entry:185 %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a186 %sub1.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c187 %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %sub.i, <2 x double> %b, <2 x double> %sub1.i) #2188 ret <2 x double> %0189}190 191define <4 x float> @test_mm_fnmsub_ss(<4 x float> %a, <4 x float> %b, <4 x float> %c) {192; CHECK-LABEL: test_mm_fnmsub_ss:193; CHECK: # %bb.0: # %entry194; CHECK-NEXT: vfnmsub213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2195; CHECK-NEXT: retq196entry:197 %0 = extractelement <4 x float> %a, i64 0198 %.rhs.i = extractelement <4 x float> %b, i64 0199 %1 = fsub float -0.000000e+00, %.rhs.i200 %.rhs2.i = extractelement <4 x float> %c, i64 0201 %2 = fsub float -0.000000e+00, %.rhs2.i202 %3 = tail call float @llvm.fma.f32(float %0, float %1, float %2) #2203 %4 = insertelement <4 x float> %a, float %3, i64 0204 ret <4 x float> %4205}206 207define <2 x double> @test_mm_fnmsub_sd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {208; CHECK-LABEL: test_mm_fnmsub_sd:209; CHECK: # %bb.0: # %entry210; CHECK-NEXT: vfnmsub213sd {{.*#+}} xmm0 = -(xmm1 * xmm0) - xmm2211; CHECK-NEXT: retq212entry:213 %0 = extractelement <2 x double> %a, i64 0214 %.rhs.i = extractelement <2 x double> %b, i64 0215 %1 = fsub double -0.000000e+00, %.rhs.i216 %.rhs2.i = extractelement <2 x double> %c, i64 0217 %2 = fsub double -0.000000e+00, %.rhs2.i218 %3 = tail call double @llvm.fma.f64(double %0, double %1, double %2) #2219 %4 = insertelement <2 x double> %a, double %3, i64 0220 ret <2 x double> %4221}222 223define <4 x float> @test_mm_fmaddsub_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {224; CHECK-LABEL: test_mm_fmaddsub_ps:225; CHECK: # %bb.0: # %entry226; CHECK-NEXT: vfmaddsub213ps {{.*#+}} xmm0 = (xmm1 * xmm0) +/- xmm2227; CHECK-NEXT: retq228entry:229 %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2230 %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c231 %2 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %1) #2232 %3 = shufflevector <4 x float> %2, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>233 ret <4 x float> %3234}235 236define <2 x double> @test_mm_fmaddsub_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {237; CHECK-LABEL: test_mm_fmaddsub_pd:238; CHECK: # %bb.0: # %entry239; CHECK-NEXT: vfmaddsub213pd {{.*#+}} xmm0 = (xmm1 * xmm0) +/- xmm2240; CHECK-NEXT: retq241entry:242 %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #2243 %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c244 %2 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %1) #2245 %3 = shufflevector <2 x double> %2, <2 x double> %0, <2 x i32> <i32 0, i32 3>246 ret <2 x double> %3247}248 249define <4 x float> @test_mm_fmsubadd_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {250; CHECK-LABEL: test_mm_fmsubadd_ps:251; CHECK: # %bb.0: # %entry252; CHECK-NEXT: vfmsubadd213ps {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2253; CHECK-NEXT: retq254entry:255 %sub.i = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c256 %0 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %sub.i) #2257 %1 = tail call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c) #2258 %2 = shufflevector <4 x float> %1, <4 x float> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>259 ret <4 x float> %2260}261 262define <2 x double> @test_mm_fmsubadd_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {263; CHECK-LABEL: test_mm_fmsubadd_pd:264; CHECK: # %bb.0: # %entry265; CHECK-NEXT: vfmsubadd213pd {{.*#+}} xmm0 = (xmm1 * xmm0) -/+ xmm2266; CHECK-NEXT: retq267entry:268 %sub.i = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %c269 %0 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %sub.i) #2270 %1 = tail call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c) #2271 %2 = shufflevector <2 x double> %1, <2 x double> %0, <2 x i32> <i32 0, i32 3>272 ret <2 x double> %2273}274 275define <8 x float> @test_mm256_fmadd_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {276; CHECK-LABEL: test_mm256_fmadd_ps:277; CHECK: # %bb.0: # %entry278; CHECK-NEXT: vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2279; CHECK-NEXT: retq280entry:281 %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2282 ret <8 x float> %0283}284 285define <4 x double> @test_mm256_fmadd_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {286; CHECK-LABEL: test_mm256_fmadd_pd:287; CHECK: # %bb.0: # %entry288; CHECK-NEXT: vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2289; CHECK-NEXT: retq290entry:291 %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #2292 ret <4 x double> %0293}294 295define <8 x float> @test_mm256_fmsub_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {296; CHECK-LABEL: test_mm256_fmsub_ps:297; CHECK: # %bb.0: # %entry298; CHECK-NEXT: vxorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2299; CHECK-NEXT: vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2300; CHECK-NEXT: retq301entry:302 %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>, %c303 %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %sub.i) #2304 ret <8 x float> %0305}306 307define <4 x double> @test_mm256_fmsub_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {308; CHECK-LABEL: test_mm256_fmsub_pd:309; CHECK: # %bb.0: # %entry310; CHECK-NEXT: vxorpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2311; CHECK-NEXT: vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2312; CHECK-NEXT: retq313entry:314 %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c315 %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %sub.i) #2316 ret <4 x double> %0317}318 319define <8 x float> @test_mm256_fnmadd_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {320; CHECK-LABEL: test_mm256_fnmadd_ps:321; CHECK: # %bb.0: # %entry322; CHECK-NEXT: vxorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0323; CHECK-NEXT: vfmadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2324; CHECK-NEXT: retq325entry:326 %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>, %a327 %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %b, <8 x float> %c) #2328 ret <8 x float> %0329}330 331define <4 x double> @test_mm256_fnmadd_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {332; CHECK-LABEL: test_mm256_fnmadd_pd:333; CHECK: # %bb.0: # %entry334; CHECK-NEXT: vxorpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0335; CHECK-NEXT: vfmadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) + ymm2336; CHECK-NEXT: retq337entry:338 %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a339 %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %b, <4 x double> %c) #2340 ret <4 x double> %0341}342 343define <8 x float> @test_mm256_fnmsub_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {344; CHECK-LABEL: test_mm256_fnmsub_ps:345; CHECK: # %bb.0: # %entry346; CHECK-NEXT: vbroadcastss {{.*#+}} ymm3 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]347; CHECK-NEXT: vxorps %ymm3, %ymm0, %ymm4348; CHECK-NEXT: vxorps %ymm3, %ymm2, %ymm0349; CHECK-NEXT: vfmadd231ps {{.*#+}} ymm0 = (ymm1 * ymm4) + ymm0350; CHECK-NEXT: retq351entry:352 %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>, %a353 %sub1.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>, %c354 %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %sub.i, <8 x float> %b, <8 x float> %sub1.i) #2355 ret <8 x float> %0356}357 358define <4 x double> @test_mm256_fnmsub_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {359; CHECK-LABEL: test_mm256_fnmsub_pd:360; CHECK: # %bb.0: # %entry361; CHECK-NEXT: vbroadcastsd {{.*#+}} ymm3 = [-0.0E+0,-0.0E+0,-0.0E+0,-0.0E+0]362; CHECK-NEXT: vxorpd %ymm3, %ymm0, %ymm4363; CHECK-NEXT: vxorpd %ymm3, %ymm2, %ymm0364; CHECK-NEXT: vfmadd231pd {{.*#+}} ymm0 = (ymm1 * ymm4) + ymm0365; CHECK-NEXT: retq366entry:367 %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a368 %sub1.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c369 %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %sub.i, <4 x double> %b, <4 x double> %sub1.i) #2370 ret <4 x double> %0371}372 373define <8 x float> @test_mm256_fmaddsub_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {374; CHECK-LABEL: test_mm256_fmaddsub_ps:375; CHECK: # %bb.0: # %entry376; CHECK-NEXT: vfmaddsub213ps {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2377; CHECK-NEXT: retq378entry:379 %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2380 %1 = 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>, %c381 %2 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %1) #2382 %3 = shufflevector <8 x float> %2, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>383 ret <8 x float> %3384}385 386define <4 x double> @test_mm256_fmaddsub_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {387; CHECK-LABEL: test_mm256_fmaddsub_pd:388; CHECK: # %bb.0: # %entry389; CHECK-NEXT: vfmaddsub213pd {{.*#+}} ymm0 = (ymm1 * ymm0) +/- ymm2390; CHECK-NEXT: retq391entry:392 %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #2393 %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c394 %2 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %1) #2395 %3 = shufflevector <4 x double> %2, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>396 ret <4 x double> %3397}398 399define <8 x float> @test_mm256_fmsubadd_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {400; CHECK-LABEL: test_mm256_fmsubadd_ps:401; CHECK: # %bb.0: # %entry402; CHECK-NEXT: vfmsubadd213ps {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2403; CHECK-NEXT: retq404entry:405 %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>, %c406 %0 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %sub.i) #2407 %1 = tail call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c) #2408 %2 = shufflevector <8 x float> %1, <8 x float> %0, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>409 ret <8 x float> %2410}411 412define <4 x double> @test_mm256_fmsubadd_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {413; CHECK-LABEL: test_mm256_fmsubadd_pd:414; CHECK: # %bb.0: # %entry415; CHECK-NEXT: vfmsubadd213pd {{.*#+}} ymm0 = (ymm1 * ymm0) -/+ ymm2416; CHECK-NEXT: retq417entry:418 %sub.i = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %c419 %0 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %sub.i) #2420 %1 = tail call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c) #2421 %2 = shufflevector <4 x double> %1, <4 x double> %0, <4 x i32> <i32 0, i32 5, i32 2, i32 7>422 ret <4 x double> %2423}424 425declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>) #1426declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>) #1427declare float @llvm.fma.f32(float, float, float) #1428declare double @llvm.fma.f64(double, double, double) #1429declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>) #1430declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>) #1431