brintos

brintos / llvm-project-archived public Read only

0
0
Text · 25.6 KiB · c3334af Raw
344 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -mtriple=x86_64-unknown -passes=slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=SSE3; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=corei7-avx -passes=slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX2564; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=bdver1 -passes=slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX2565; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=core-avx2 -passes=slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX2566; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skylake-avx512 -mattr=-prefer-256-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX5127; RUN: opt < %s -mtriple=x86_64-unknown -mcpu=skylake-avx512 -mattr=+prefer-256-bit -passes=slp-vectorizer -S | FileCheck %s --check-prefix=CHECK --check-prefix=AVX --check-prefix=AVX2568 9target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"10 11@srcA64 = common global [8 x double] zeroinitializer, align 6412@srcB64 = common global [8 x double] zeroinitializer, align 6413@srcC64 = common global [8 x double] zeroinitializer, align 6414@srcA32 = common global [16 x float] zeroinitializer, align 6415@srcB32 = common global [16 x float] zeroinitializer, align 6416@srcC32 = common global [16 x float] zeroinitializer, align 6417@dst64 = common global [8 x double] zeroinitializer, align 6418@dst32 = common global [16 x float] zeroinitializer, align 6419 20declare float @llvm.copysign.f32(float, float)21declare double @llvm.copysign.f64(double, double)22 23;24; CHECK25;26 27define void @fcopysign_2f64() #0 {28; CHECK-LABEL: @fcopysign_2f64(29; CHECK-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr @srcA64, align 830; CHECK-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr @srcB64, align 831; CHECK-NEXT:    [[TMP3:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP1]], <2 x double> [[TMP2]])32; CHECK-NEXT:    store <2 x double> [[TMP3]], ptr @dst64, align 833; CHECK-NEXT:    ret void34;35  %a0 = load double, ptr @srcA64, align 836  %a1 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 1), align 837  %b0 = load double, ptr @srcB64, align 838  %b1 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 1), align 839  %fcopysign0 = call double @llvm.copysign.f64(double %a0, double %b0)40  %fcopysign1 = call double @llvm.copysign.f64(double %a1, double %b1)41  store double %fcopysign0, ptr @dst64, align 842  store double %fcopysign1, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 1), align 843  ret void44}45 46define void @fcopysign_4f64() #0 {47; SSE-LABEL: @fcopysign_4f64(48; SSE-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr @srcA64, align 849; SSE-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr @srcB64, align 850; SSE-NEXT:    [[TMP3:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP1]], <2 x double> [[TMP2]])51; SSE-NEXT:    store <2 x double> [[TMP3]], ptr @dst64, align 852; SSE-NEXT:    [[TMP4:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 2), align 853; SSE-NEXT:    [[TMP5:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 2), align 854; SSE-NEXT:    [[TMP6:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP4]], <2 x double> [[TMP5]])55; SSE-NEXT:    store <2 x double> [[TMP6]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 2), align 856; SSE-NEXT:    ret void57;58; AVX-LABEL: @fcopysign_4f64(59; AVX-NEXT:    [[TMP1:%.*]] = load <4 x double>, ptr @srcA64, align 860; AVX-NEXT:    [[TMP2:%.*]] = load <4 x double>, ptr @srcB64, align 861; AVX-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.copysign.v4f64(<4 x double> [[TMP1]], <4 x double> [[TMP2]])62; AVX-NEXT:    store <4 x double> [[TMP3]], ptr @dst64, align 863; AVX-NEXT:    ret void64;65  %a0 = load double, ptr @srcA64, align 866  %a1 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 1), align 867  %a2 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 2), align 868  %a3 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 3), align 869  %b0 = load double, ptr @srcB64, align 870  %b1 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 1), align 871  %b2 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 2), align 872  %b3 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 3), align 873  %fcopysign0 = call double @llvm.copysign.f64(double %a0, double %b0)74  %fcopysign1 = call double @llvm.copysign.f64(double %a1, double %b1)75  %fcopysign2 = call double @llvm.copysign.f64(double %a2, double %b2)76  %fcopysign3 = call double @llvm.copysign.f64(double %a3, double %b3)77  store double %fcopysign0, ptr @dst64, align 878  store double %fcopysign1, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 1), align 879  store double %fcopysign2, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 2), align 880  store double %fcopysign3, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 3), align 881  ret void82}83 84define void @fcopysign_8f64() #0 {85; SSE-LABEL: @fcopysign_8f64(86; SSE-NEXT:    [[TMP1:%.*]] = load <2 x double>, ptr @srcA64, align 487; SSE-NEXT:    [[TMP2:%.*]] = load <2 x double>, ptr @srcB64, align 488; SSE-NEXT:    [[TMP3:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP1]], <2 x double> [[TMP2]])89; SSE-NEXT:    store <2 x double> [[TMP3]], ptr @dst64, align 490; SSE-NEXT:    [[TMP4:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 2), align 491; SSE-NEXT:    [[TMP5:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 2), align 492; SSE-NEXT:    [[TMP6:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP4]], <2 x double> [[TMP5]])93; SSE-NEXT:    store <2 x double> [[TMP6]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 2), align 494; SSE-NEXT:    [[TMP7:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 4), align 495; SSE-NEXT:    [[TMP8:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 4), align 496; SSE-NEXT:    [[TMP9:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP7]], <2 x double> [[TMP8]])97; SSE-NEXT:    store <2 x double> [[TMP9]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 4), align 498; SSE-NEXT:    [[TMP10:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 6), align 499; SSE-NEXT:    [[TMP11:%.*]] = load <2 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 6), align 4100; SSE-NEXT:    [[TMP12:%.*]] = call <2 x double> @llvm.copysign.v2f64(<2 x double> [[TMP10]], <2 x double> [[TMP11]])101; SSE-NEXT:    store <2 x double> [[TMP12]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 6), align 4102; SSE-NEXT:    ret void103;104; AVX256-LABEL: @fcopysign_8f64(105; AVX256-NEXT:    [[TMP1:%.*]] = load <4 x double>, ptr @srcA64, align 4106; AVX256-NEXT:    [[TMP2:%.*]] = load <4 x double>, ptr @srcB64, align 4107; AVX256-NEXT:    [[TMP3:%.*]] = call <4 x double> @llvm.copysign.v4f64(<4 x double> [[TMP1]], <4 x double> [[TMP2]])108; AVX256-NEXT:    store <4 x double> [[TMP3]], ptr @dst64, align 4109; AVX256-NEXT:    [[TMP4:%.*]] = load <4 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 4), align 4110; AVX256-NEXT:    [[TMP5:%.*]] = load <4 x double>, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 4), align 4111; AVX256-NEXT:    [[TMP6:%.*]] = call <4 x double> @llvm.copysign.v4f64(<4 x double> [[TMP4]], <4 x double> [[TMP5]])112; AVX256-NEXT:    store <4 x double> [[TMP6]], ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 4), align 4113; AVX256-NEXT:    ret void114;115; AVX512-LABEL: @fcopysign_8f64(116; AVX512-NEXT:    [[TMP1:%.*]] = load <8 x double>, ptr @srcA64, align 4117; AVX512-NEXT:    [[TMP2:%.*]] = load <8 x double>, ptr @srcB64, align 4118; AVX512-NEXT:    [[TMP3:%.*]] = call <8 x double> @llvm.copysign.v8f64(<8 x double> [[TMP1]], <8 x double> [[TMP2]])119; AVX512-NEXT:    store <8 x double> [[TMP3]], ptr @dst64, align 4120; AVX512-NEXT:    ret void121;122  %a0 = load double, ptr @srcA64, align 4123  %a1 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 1), align 4124  %a2 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 2), align 4125  %a3 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 3), align 4126  %a4 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 4), align 4127  %a5 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 5), align 4128  %a6 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 6), align 4129  %a7 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcA64, i32 0, i64 7), align 4130  %b0 = load double, ptr @srcB64, align 4131  %b1 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 1), align 4132  %b2 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 2), align 4133  %b3 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 3), align 4134  %b4 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 4), align 4135  %b5 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 5), align 4136  %b6 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 6), align 4137  %b7 = load double, ptr getelementptr inbounds ([8 x double], ptr @srcB64, i32 0, i64 7), align 4138  %fcopysign0 = call double @llvm.copysign.f64(double %a0, double %b0)139  %fcopysign1 = call double @llvm.copysign.f64(double %a1, double %b1)140  %fcopysign2 = call double @llvm.copysign.f64(double %a2, double %b2)141  %fcopysign3 = call double @llvm.copysign.f64(double %a3, double %b3)142  %fcopysign4 = call double @llvm.copysign.f64(double %a4, double %b4)143  %fcopysign5 = call double @llvm.copysign.f64(double %a5, double %b5)144  %fcopysign6 = call double @llvm.copysign.f64(double %a6, double %b6)145  %fcopysign7 = call double @llvm.copysign.f64(double %a7, double %b7)146  store double %fcopysign0, ptr @dst64, align 4147  store double %fcopysign1, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 1), align 4148  store double %fcopysign2, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 2), align 4149  store double %fcopysign3, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 3), align 4150  store double %fcopysign4, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 4), align 4151  store double %fcopysign5, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 5), align 4152  store double %fcopysign6, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 6), align 4153  store double %fcopysign7, ptr getelementptr inbounds ([8 x double], ptr @dst64, i32 0, i64 7), align 4154  ret void155}156 157define void @fcopysign_4f32() #0 {158; CHECK-LABEL: @fcopysign_4f32(159; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, ptr @srcA32, align 4160; CHECK-NEXT:    [[TMP2:%.*]] = load <4 x float>, ptr @srcB32, align 4161; CHECK-NEXT:    [[TMP3:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP2]])162; CHECK-NEXT:    store <4 x float> [[TMP3]], ptr @dst32, align 4163; CHECK-NEXT:    ret void164;165  %a0 = load float, ptr @srcA32, align 4166  %a1 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 1), align 4167  %a2 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 2), align 4168  %a3 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 3), align 4169  %b0 = load float, ptr @srcB32, align 4170  %b1 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 1), align 4171  %b2 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 2), align 4172  %b3 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 3), align 4173  %fcopysign0 = call float @llvm.copysign.f32(float %a0, float %b0)174  %fcopysign1 = call float @llvm.copysign.f32(float %a1, float %b1)175  %fcopysign2 = call float @llvm.copysign.f32(float %a2, float %b2)176  %fcopysign3 = call float @llvm.copysign.f32(float %a3, float %b3)177  store float %fcopysign0, ptr @dst32, align 4178  store float %fcopysign1, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 1), align 4179  store float %fcopysign2, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 2), align 4180  store float %fcopysign3, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 3), align 4181  ret void182}183 184define void @fcopysign_8f32() #0 {185; SSE-LABEL: @fcopysign_8f32(186; SSE-NEXT:    [[TMP1:%.*]] = load <4 x float>, ptr @srcA32, align 4187; SSE-NEXT:    [[TMP2:%.*]] = load <4 x float>, ptr @srcB32, align 4188; SSE-NEXT:    [[TMP3:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP2]])189; SSE-NEXT:    store <4 x float> [[TMP3]], ptr @dst32, align 4190; SSE-NEXT:    [[TMP4:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 4), align 4191; SSE-NEXT:    [[TMP5:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 4), align 4192; SSE-NEXT:    [[TMP6:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP4]], <4 x float> [[TMP5]])193; SSE-NEXT:    store <4 x float> [[TMP6]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 4), align 4194; SSE-NEXT:    ret void195;196; AVX-LABEL: @fcopysign_8f32(197; AVX-NEXT:    [[TMP1:%.*]] = load <8 x float>, ptr @srcA32, align 4198; AVX-NEXT:    [[TMP2:%.*]] = load <8 x float>, ptr @srcB32, align 4199; AVX-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.copysign.v8f32(<8 x float> [[TMP1]], <8 x float> [[TMP2]])200; AVX-NEXT:    store <8 x float> [[TMP3]], ptr @dst32, align 4201; AVX-NEXT:    ret void202;203  %a0 = load float, ptr @srcA32, align 4204  %a1 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 1), align 4205  %a2 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 2), align 4206  %a3 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 3), align 4207  %a4 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 4), align 4208  %a5 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 5), align 4209  %a6 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 6), align 4210  %a7 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 7), align 4211  %b0 = load float, ptr @srcB32, align 4212  %b1 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 1), align 4213  %b2 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 2), align 4214  %b3 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 3), align 4215  %b4 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 4), align 4216  %b5 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 5), align 4217  %b6 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 6), align 4218  %b7 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 7), align 4219  %fcopysign0 = call float @llvm.copysign.f32(float %a0, float %b0)220  %fcopysign1 = call float @llvm.copysign.f32(float %a1, float %b1)221  %fcopysign2 = call float @llvm.copysign.f32(float %a2, float %b2)222  %fcopysign3 = call float @llvm.copysign.f32(float %a3, float %b3)223  %fcopysign4 = call float @llvm.copysign.f32(float %a4, float %b4)224  %fcopysign5 = call float @llvm.copysign.f32(float %a5, float %b5)225  %fcopysign6 = call float @llvm.copysign.f32(float %a6, float %b6)226  %fcopysign7 = call float @llvm.copysign.f32(float %a7, float %b7)227  store float %fcopysign0, ptr @dst32, align 4228  store float %fcopysign1, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 1), align 4229  store float %fcopysign2, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 2), align 4230  store float %fcopysign3, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 3), align 4231  store float %fcopysign4, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 4), align 4232  store float %fcopysign5, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 5), align 4233  store float %fcopysign6, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 6), align 4234  store float %fcopysign7, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 7), align 4235  ret void236}237 238define void @fcopysign_16f32() #0 {239; SSE-LABEL: @fcopysign_16f32(240; SSE-NEXT:    [[TMP1:%.*]] = load <4 x float>, ptr @srcA32, align 4241; SSE-NEXT:    [[TMP2:%.*]] = load <4 x float>, ptr @srcB32, align 4242; SSE-NEXT:    [[TMP3:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP1]], <4 x float> [[TMP2]])243; SSE-NEXT:    store <4 x float> [[TMP3]], ptr @dst32, align 4244; SSE-NEXT:    [[TMP4:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 4), align 4245; SSE-NEXT:    [[TMP5:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 4), align 4246; SSE-NEXT:    [[TMP6:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP4]], <4 x float> [[TMP5]])247; SSE-NEXT:    store <4 x float> [[TMP6]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 4), align 4248; SSE-NEXT:    [[TMP7:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 8), align 4249; SSE-NEXT:    [[TMP8:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 8), align 4250; SSE-NEXT:    [[TMP9:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP7]], <4 x float> [[TMP8]])251; SSE-NEXT:    store <4 x float> [[TMP9]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 8), align 4252; SSE-NEXT:    [[TMP10:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 12), align 4253; SSE-NEXT:    [[TMP11:%.*]] = load <4 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 12), align 4254; SSE-NEXT:    [[TMP12:%.*]] = call <4 x float> @llvm.copysign.v4f32(<4 x float> [[TMP10]], <4 x float> [[TMP11]])255; SSE-NEXT:    store <4 x float> [[TMP12]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 12), align 4256; SSE-NEXT:    ret void257;258; AVX256-LABEL: @fcopysign_16f32(259; AVX256-NEXT:    [[TMP1:%.*]] = load <8 x float>, ptr @srcA32, align 4260; AVX256-NEXT:    [[TMP2:%.*]] = load <8 x float>, ptr @srcB32, align 4261; AVX256-NEXT:    [[TMP3:%.*]] = call <8 x float> @llvm.copysign.v8f32(<8 x float> [[TMP1]], <8 x float> [[TMP2]])262; AVX256-NEXT:    store <8 x float> [[TMP3]], ptr @dst32, align 4263; AVX256-NEXT:    [[TMP4:%.*]] = load <8 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 8), align 4264; AVX256-NEXT:    [[TMP5:%.*]] = load <8 x float>, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 8), align 4265; AVX256-NEXT:    [[TMP6:%.*]] = call <8 x float> @llvm.copysign.v8f32(<8 x float> [[TMP4]], <8 x float> [[TMP5]])266; AVX256-NEXT:    store <8 x float> [[TMP6]], ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 8), align 4267; AVX256-NEXT:    ret void268;269; AVX512-LABEL: @fcopysign_16f32(270; AVX512-NEXT:    [[TMP1:%.*]] = load <16 x float>, ptr @srcA32, align 4271; AVX512-NEXT:    [[TMP2:%.*]] = load <16 x float>, ptr @srcB32, align 4272; AVX512-NEXT:    [[TMP3:%.*]] = call <16 x float> @llvm.copysign.v16f32(<16 x float> [[TMP1]], <16 x float> [[TMP2]])273; AVX512-NEXT:    store <16 x float> [[TMP3]], ptr @dst32, align 4274; AVX512-NEXT:    ret void275;276  %a0  = load float, ptr @srcA32, align 4277  %a1  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64  1), align 4278  %a2  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64  2), align 4279  %a3  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64  3), align 4280  %a4  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64  4), align 4281  %a5  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64  5), align 4282  %a6  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64  6), align 4283  %a7  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64  7), align 4284  %a8  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64  8), align 4285  %a9  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64  9), align 4286  %a10 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 10), align 4287  %a11 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 11), align 4288  %a12 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 12), align 4289  %a13 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 13), align 4290  %a14 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 14), align 4291  %a15 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcA32, i32 0, i64 15), align 4292  %b0  = load float, ptr @srcB32, align 4293  %b1  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64  1), align 4294  %b2  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64  2), align 4295  %b3  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64  3), align 4296  %b4  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64  4), align 4297  %b5  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64  5), align 4298  %b6  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64  6), align 4299  %b7  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64  7), align 4300  %b8  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64  8), align 4301  %b9  = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64  9), align 4302  %b10 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 10), align 4303  %b11 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 11), align 4304  %b12 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 12), align 4305  %b13 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 13), align 4306  %b14 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 14), align 4307  %b15 = load float, ptr getelementptr inbounds ([16 x float], ptr @srcB32, i32 0, i64 15), align 4308  %fcopysign0  = call float @llvm.copysign.f32(float %a0 , float %b0 )309  %fcopysign1  = call float @llvm.copysign.f32(float %a1 , float %b1 )310  %fcopysign2  = call float @llvm.copysign.f32(float %a2 , float %b2 )311  %fcopysign3  = call float @llvm.copysign.f32(float %a3 , float %b3 )312  %fcopysign4  = call float @llvm.copysign.f32(float %a4 , float %b4 )313  %fcopysign5  = call float @llvm.copysign.f32(float %a5 , float %b5 )314  %fcopysign6  = call float @llvm.copysign.f32(float %a6 , float %b6 )315  %fcopysign7  = call float @llvm.copysign.f32(float %a7 , float %b7 )316  %fcopysign8  = call float @llvm.copysign.f32(float %a8 , float %b8 )317  %fcopysign9  = call float @llvm.copysign.f32(float %a9 , float %b9 )318  %fcopysign10 = call float @llvm.copysign.f32(float %a10, float %b10)319  %fcopysign11 = call float @llvm.copysign.f32(float %a11, float %b11)320  %fcopysign12 = call float @llvm.copysign.f32(float %a12, float %b12)321  %fcopysign13 = call float @llvm.copysign.f32(float %a13, float %b13)322  %fcopysign14 = call float @llvm.copysign.f32(float %a14, float %b14)323  %fcopysign15 = call float @llvm.copysign.f32(float %a15, float %b15)324  store float %fcopysign0 , ptr @dst32, align 4325  store float %fcopysign1 , ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64  1), align 4326  store float %fcopysign2 , ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64  2), align 4327  store float %fcopysign3 , ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64  3), align 4328  store float %fcopysign4 , ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64  4), align 4329  store float %fcopysign5 , ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64  5), align 4330  store float %fcopysign6 , ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64  6), align 4331  store float %fcopysign7 , ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64  7), align 4332  store float %fcopysign8 , ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64  8), align 4333  store float %fcopysign9 , ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64  9), align 4334  store float %fcopysign10, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 10), align 4335  store float %fcopysign11, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 11), align 4336  store float %fcopysign12, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 12), align 4337  store float %fcopysign13, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 13), align 4338  store float %fcopysign14, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 14), align 4339  store float %fcopysign15, ptr getelementptr inbounds ([16 x float], ptr @dst32, i32 0, i64 15), align 4340  ret void341}342 343attributes #0 = { nounwind }344