61 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -mtriple=x86_64-linux-gnu -mcpu=x86-64-v2 -passes=slp-vectorizer -S | FileCheck %s3 4; FIXME: Ensure llvm.powi.* intrinsics are vectorized.5 6define <2 x double> @PR53887_v2f64(<2 x double> noundef %x) {7; CHECK-LABEL: @PR53887_v2f64(8; CHECK-NEXT: entry:9; CHECK-NEXT: [[VECEXT:%.*]] = extractelement <2 x double> [[X:%.*]], i64 010; CHECK-NEXT: [[TMP2:%.*]] = tail call fast double @llvm.powi.f64.i32(double [[VECEXT]], i32 6)11; CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x double> zeroinitializer, double [[TMP2]], i64 012; CHECK-NEXT: [[VECEXT1:%.*]] = extractelement <2 x double> [[X]], i64 113; CHECK-NEXT: [[TMP1:%.*]] = tail call fast double @llvm.powi.f64.i32(double [[VECEXT1]], i32 6)14; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x double> [[VECINIT]], double [[TMP1]], i64 115; CHECK-NEXT: ret <2 x double> [[TMP0]]16;17entry:18 %vecext = extractelement <2 x double> %x, i64 019 %0 = tail call fast double @llvm.powi.f64.i32(double %vecext, i32 6)20 %vecinit = insertelement <2 x double> zeroinitializer, double %0, i64 021 %vecext1 = extractelement <2 x double> %x, i64 122 %1 = tail call fast double @llvm.powi.f64.i32(double %vecext1, i32 6)23 %vecinit3 = insertelement <2 x double> %vecinit, double %1, i64 124 ret <2 x double> %vecinit325}26 27define <4 x double> @PR53887_v4f64(<4 x double> noundef %x) {28; CHECK-LABEL: @PR53887_v4f64(29; CHECK-NEXT: entry:30; CHECK-NEXT: [[VECEXT:%.*]] = extractelement <4 x double> [[X:%.*]], i64 031; CHECK-NEXT: [[TMP4:%.*]] = tail call fast double @llvm.powi.f64.i32(double [[VECEXT]], i32 6)32; CHECK-NEXT: [[VECINIT:%.*]] = insertelement <4 x double> zeroinitializer, double [[TMP4]], i64 033; CHECK-NEXT: [[VECEXT1:%.*]] = extractelement <4 x double> [[X]], i64 134; CHECK-NEXT: [[TMP1:%.*]] = tail call fast double @llvm.powi.f64.i32(double [[VECEXT1]], i32 6)35; CHECK-NEXT: [[VECINIT3:%.*]] = insertelement <4 x double> [[VECINIT]], double [[TMP1]], i64 136; CHECK-NEXT: [[VECEXT4:%.*]] = extractelement <4 x double> [[X]], i64 237; CHECK-NEXT: [[TMP2:%.*]] = tail call fast double @llvm.powi.f64.i32(double [[VECEXT4]], i32 6)38; CHECK-NEXT: [[VECINIT6:%.*]] = insertelement <4 x double> [[VECINIT3]], double [[TMP2]], i64 239; CHECK-NEXT: [[VECEXT7:%.*]] = extractelement <4 x double> [[X]], i64 340; CHECK-NEXT: [[TMP3:%.*]] = tail call fast double @llvm.powi.f64.i32(double [[VECEXT7]], i32 6)41; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x double> [[VECINIT6]], double [[TMP3]], i64 342; CHECK-NEXT: ret <4 x double> [[TMP0]]43;44entry:45 %vecext = extractelement <4 x double> %x, i64 046 %0 = tail call fast double @llvm.powi.f64.i32(double %vecext, i32 6) #247 %vecinit = insertelement <4 x double> zeroinitializer, double %0, i64 048 %vecext1 = extractelement <4 x double> %x, i64 149 %1 = tail call fast double @llvm.powi.f64.i32(double %vecext1, i32 6) #250 %vecinit3 = insertelement <4 x double> %vecinit, double %1, i64 151 %vecext4 = extractelement <4 x double> %x, i64 252 %2 = tail call fast double @llvm.powi.f64.i32(double %vecext4, i32 6) #253 %vecinit6 = insertelement <4 x double> %vecinit3, double %2, i64 254 %vecext7 = extractelement <4 x double> %x, i64 355 %3 = tail call fast double @llvm.powi.f64.i32(double %vecext7, i32 6) #256 %vecinit9 = insertelement <4 x double> %vecinit6, double %3, i64 357 ret <4 x double> %vecinit958}59 60declare double @llvm.powi.f64.i32(double, i32)61