40 lines · plain
1; RUN: opt -mtriple=x86_64-unknown-linux-gnu -passes=load-store-vectorizer -mcpu haswell -S -o - %s | FileCheck --check-prefix=CHECK-HSW %s2; RUN: opt -mtriple=x86_64-unknown-linux-gnu -passes=load-store-vectorizer -mcpu knl -S -o - %s | FileCheck --check-prefix=CHECK-KNL %s3; RUN: opt -mtriple=x86_64-unknown-linux-gnu -aa-pipeline=basic-aa -passes='function(load-store-vectorizer)' -mcpu haswell -S -o - %s | FileCheck --check-prefix=CHECK-HSW %s4; RUN: opt -mtriple=x86_64-unknown-linux-gnu -aa-pipeline=basic-aa -passes='function(load-store-vectorizer)' -mcpu knl -S -o - %s | FileCheck --check-prefix=CHECK-KNL %s5 6define <8 x double> @loadwidth_insert_extract(ptr %ptr) {7 %b = getelementptr <2 x double>, ptr %ptr, i32 18 %c = getelementptr <2 x double>, ptr %ptr, i32 29 %d = getelementptr <2 x double>, ptr %ptr, i32 310; CHECK-HSW: load <4 x double>11; CHECK-HSW: load <4 x double>12; CHECK-HSW-NOT: load13; CHECK-KNL: load <8 x double>14; CHECK-KNL-NOT: load15 %la = load <2 x double>, ptr %ptr16 %lb = load <2 x double>, ptr %b17 %lc = load <2 x double>, ptr %c18 %ld = load <2 x double>, ptr %d19 ; Scalarize everything - Explicitly not a shufflevector to test this code20 ; path in the LSV21 %v1 = extractelement <2 x double> %la, i32 022 %v2 = extractelement <2 x double> %la, i32 123 %v3 = extractelement <2 x double> %lb, i32 024 %v4 = extractelement <2 x double> %lb, i32 125 %v5 = extractelement <2 x double> %lc, i32 026 %v6 = extractelement <2 x double> %lc, i32 127 %v7 = extractelement <2 x double> %ld, i32 028 %v8 = extractelement <2 x double> %ld, i32 129 ; Make a vector again30 %i1 = insertelement <8 x double> undef, double %v1, i32 031 %i2 = insertelement <8 x double> %i1, double %v2, i32 132 %i3 = insertelement <8 x double> %i2, double %v3, i32 233 %i4 = insertelement <8 x double> %i3, double %v4, i32 334 %i5 = insertelement <8 x double> %i4, double %v5, i32 435 %i6 = insertelement <8 x double> %i5, double %v6, i32 536 %i7 = insertelement <8 x double> %i6, double %v7, i32 637 %i8 = insertelement <8 x double> %i7, double %v8, i32 738 ret <8 x double> %i839}40