1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=skylake-avx512 -passes=slp-vectorizer -S | FileCheck %s 3; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=skylake-avx512 -passes=inject-tli-mappings,slp-vectorizer -vector-library=SVML -S | FileCheck %s --check-prefix=VECLIB 4; RUN: opt < %s -mtriple=x86_64-unknown-linux -mcpu=skylake-avx512 -passes=inject-tli-mappings,slp-vectorizer -vector-library=AMDLIBM -S | FileCheck %s --check-prefix=AMDLIBM 5 6@src = common global [8 x double] zeroinitializer, align 64 7@dst = common global [8 x double] zeroinitializer, align 64 8 9declare double @llvm.sqrt.f64(double) 10declare double @llvm.sin.f64(double) 11 12define void @test() { 13; CHECK-LABEL: @test( 14; CHECK-NEXT: [[TMP1:%.*]] = load <8 x double>, ptr @src, align 8 15; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 2, i32 6> 16; CHECK-NEXT: [[TMP3:%.*]] = call fast <2 x double> @llvm.sin.v2f64(<2 x double> [[TMP2]]) 17; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 3, i32 7> 18; CHECK-NEXT: [[TMP5:%.*]] = call fast <2 x double> @llvm.sin.v2f64(<2 x double> [[TMP4]]) 19; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 0, i32 4> 20; CHECK-NEXT: [[TMP7:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> [[TMP6]]) 21; CHECK-NEXT: [[TMP8:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 1, i32 5> 22; CHECK-NEXT: [[TMP9:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> [[TMP8]]) 23; CHECK-NEXT: [[TMP10:%.*]] = fadd fast <2 x double> [[TMP7]], [[TMP5]] 24; CHECK-NEXT: [[TMP11:%.*]] = fadd fast <2 x double> [[TMP3]], [[TMP9]] 25; CHECK-NEXT: [[TMP12:%.*]] = fadd fast <2 x double> [[TMP10]], [[TMP11]] 26; CHECK-NEXT: store <2 x double> [[TMP12]], ptr @dst, align 8 27; CHECK-NEXT: ret void 28; 29; VECLIB-LABEL: @test( 30; VECLIB-NEXT: [[TMP1:%.*]] = load <8 x double>, ptr @src, align 8 31; VECLIB-NEXT: [[TMP2:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 2, i32 6> 32; VECLIB-NEXT: [[TMP3:%.*]] = call fast <2 x double> @__svml_sin2(<2 x double> [[TMP2]]) 33; VECLIB-NEXT: [[TMP4:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 3, i32 7> 34; VECLIB-NEXT: [[TMP5:%.*]] = call fast <2 x double> @__svml_sin2(<2 x double> [[TMP4]]) 35; VECLIB-NEXT: [[TMP6:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 0, i32 4> 36; VECLIB-NEXT: [[TMP7:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> [[TMP6]]) 37; VECLIB-NEXT: [[TMP8:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 1, i32 5> 38; VECLIB-NEXT: [[TMP9:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> [[TMP8]]) 39; VECLIB-NEXT: [[TMP10:%.*]] = fadd fast <2 x double> [[TMP7]], [[TMP5]] 40; VECLIB-NEXT: [[TMP11:%.*]] = fadd fast <2 x double> [[TMP3]], [[TMP9]] 41; VECLIB-NEXT: [[TMP12:%.*]] = fadd fast <2 x double> [[TMP10]], [[TMP11]] 42; VECLIB-NEXT: store <2 x double> [[TMP12]], ptr @dst, align 8 43; VECLIB-NEXT: ret void 44; 45; AMDLIBM-LABEL: @test( 46; AMDLIBM-NEXT: [[TMP1:%.*]] = load <8 x double>, ptr @src, align 8 47; AMDLIBM-NEXT: [[TMP2:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 2, i32 6> 48; AMDLIBM-NEXT: [[TMP3:%.*]] = call fast <2 x double> @amd_vrd2_sin(<2 x double> [[TMP2]]) 49; AMDLIBM-NEXT: [[TMP4:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 3, i32 7> 50; AMDLIBM-NEXT: [[TMP5:%.*]] = call fast <2 x double> @amd_vrd2_sin(<2 x double> [[TMP4]]) 51; AMDLIBM-NEXT: [[TMP6:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 0, i32 4> 52; AMDLIBM-NEXT: [[TMP7:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> [[TMP6]]) 53; AMDLIBM-NEXT: [[TMP8:%.*]] = shufflevector <8 x double> [[TMP1]], <8 x double> poison, <2 x i32> <i32 1, i32 5> 54; AMDLIBM-NEXT: [[TMP9:%.*]] = call fast <2 x double> @llvm.sqrt.v2f64(<2 x double> [[TMP8]]) 55; AMDLIBM-NEXT: [[TMP10:%.*]] = fadd fast <2 x double> [[TMP7]], [[TMP5]] 56; AMDLIBM-NEXT: [[TMP11:%.*]] = fadd fast <2 x double> [[TMP3]], [[TMP9]] 57; AMDLIBM-NEXT: [[TMP12:%.*]] = fadd fast <2 x double> [[TMP10]], [[TMP11]] 58; AMDLIBM-NEXT: store <2 x double> [[TMP12]], ptr @dst, align 8 59; AMDLIBM-NEXT: ret void 60; 61 %a0 = load double, ptr @src, align 8 62 %a1 = load double, ptr getelementptr inbounds ([8 x double], ptr @src, i32 0, i64 1), align 8 63 %a2 = load double, ptr getelementptr inbounds ([8 x double], ptr @src, i32 0, i64 2), align 8 64 %a3 = load double, ptr getelementptr inbounds ([8 x double], ptr @src, i32 0, i64 3), align 8 65 %a4 = load double, ptr getelementptr inbounds ([8 x double], ptr @src, i32 0, i64 4), align 8 66 %a5 = load double, ptr getelementptr inbounds ([8 x double], ptr @src, i32 0, i64 5), align 8 67 %a6 = load double, ptr getelementptr inbounds ([8 x double], ptr @src, i32 0, i64 6), align 8 68 %a7 = load double, ptr getelementptr inbounds ([8 x double], ptr @src, i32 0, i64 7), align 8 69 %sin0 = call fast double @llvm.sin.f64(double %a2) 70 %sin1 = call fast double @llvm.sin.f64(double %a3) 71 %sqrt0 = call fast double @llvm.sqrt.f64(double %a0) 72 %sqrt1 = call fast double @llvm.sqrt.f64(double %a1) 73 %sin2 = call fast double @llvm.sin.f64(double %a6) 74 %sin3 = call fast double @llvm.sin.f64(double %a7) 75 %sqrt2 = call fast double @llvm.sqrt.f64(double %a4) 76 %sqrt3 = call fast double @llvm.sqrt.f64(double %a5) 77 %res1 = fadd fast double %sqrt0, %sin1 78 %res2 = fadd fast double %sin0, %sqrt1 79 %res00 = fadd fast double %res1, %res2 80 %res3 = fadd fast double %sqrt2, %sin3 81 %res4 = fadd fast double %sin2, %sqrt3 82 %res01 = fadd fast double %res3, %res4 83 store double %res00, ptr @dst, align 8 84 store double %res01, ptr getelementptr inbounds ([8 x double], ptr @dst, i32 0, i64 1), align 8 85 ret void 86} 87