75 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -interleaved-access -S %s | FileCheck %s3; RUN: opt -passes=interleaved-access -S %s | FileCheck %s4 5target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"6target triple = "x86_64-apple-macosx10.15.0"7 8; No interleaved load instruction is generated, but the shuffle is moved just9; after the load.10define <2 x double> @shuffle_binop_fol(ptr %ptr) {11; CHECK-LABEL: @shuffle_binop_fol(12; CHECK-NEXT: vector.body.preheader:13; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x double>, ptr [[PTR:%.*]], align 814; CHECK-NEXT: [[EXTRACTED1:%.*]] = shufflevector <4 x double> [[WIDE_LOAD]], <4 x double> poison, <2 x i32> <i32 0, i32 2>15; CHECK-NEXT: [[EXTRACTED2:%.*]] = shufflevector <4 x double> splat (double 1.000000e+00), <4 x double> poison, <2 x i32> <i32 0, i32 2>16; CHECK-NEXT: [[FADD3:%.*]] = fadd <2 x double> [[EXTRACTED1]], [[EXTRACTED2]]17; CHECK-NEXT: ret <2 x double> [[FADD3]]18;19vector.body.preheader:20 %wide.load = load <4 x double>, ptr %ptr, align 821 %fadd = fadd <4 x double> %wide.load, <double 1.0, double 1.0, double 1.0, double 1.0>22 %extracted = shufflevector <4 x double> %fadd, <4 x double> undef, <2 x i32> <i32 0, i32 2>23 ret <2 x double> %extracted24}25 26define <2 x double> @shuffle_binop_fol_oob(ptr %ptr) {27; CHECK-LABEL: @shuffle_binop_fol_oob(28; CHECK-NEXT: vector.body.preheader:29; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x double>, ptr [[PTR:%.*]], align 830; CHECK-NEXT: [[FADD:%.*]] = fadd <4 x double> [[WIDE_LOAD]], splat (double 1.000000e+00)31; CHECK-NEXT: [[EXTRACTED:%.*]] = shufflevector <4 x double> [[FADD]], <4 x double> undef, <2 x i32> <i32 0, i32 4>32; CHECK-NEXT: ret <2 x double> [[EXTRACTED]]33;34vector.body.preheader:35 %wide.load = load <4 x double>, ptr %ptr, align 836 %fadd = fadd <4 x double> %wide.load, <double 1.0, double 1.0, double 1.0, double 1.0>37 %extracted = shufflevector <4 x double> %fadd, <4 x double> undef, <2 x i32> <i32 0, i32 4>38 ret <2 x double> %extracted39}40 41; No interleaved load instruction is generated, but the extractelement42; instructions are updated to use the shuffle instead of the load.43define void @shuffle_extract(ptr %ptr, i1 %c) {44; CHECK-LABEL: @shuffle_extract(45; CHECK-NEXT: vector.body.preheader:46; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x double>, ptr [[PTR:%.*]], align 847; CHECK-NEXT: [[EXTRACTED:%.*]] = shufflevector <4 x double> [[WIDE_LOAD]], <4 x double> undef, <2 x i32> <i32 0, i32 2>48; CHECK-NEXT: br i1 [[C:%.*]], label [[IF_THEN:%.*]], label [[IF_MERGE:%.*]]49; CHECK: if.then:50; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[EXTRACTED]], i64 051; CHECK-NEXT: call void @use(double [[TMP0]])52; CHECK-NEXT: br label [[IF_MERGE]]53; CHECK: if.merge:54; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x double> [[EXTRACTED]], i64 155; CHECK-NEXT: call void @use(double [[TMP1]])56; CHECK-NEXT: ret void57;58vector.body.preheader:59 %wide.load = load <4 x double>, ptr %ptr, align 860 %extracted = shufflevector <4 x double> %wide.load, <4 x double> undef, <2 x i32> <i32 0, i32 2>61 br i1 %c, label %if.then, label %if.merge62 63if.then:64 %e0 = extractelement <4 x double> %wide.load, i32 065 call void @use(double %e0)66 br label %if.merge67 68if.merge:69 %e1 = extractelement <4 x double> %wide.load, i32 270 call void @use(double %e1)71 ret void72}73 74declare void @use(double)75