124 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes='sroa<preserve-cfg>' -S | FileCheck %s --check-prefixes=CHECK,CHECK-PRESERVE-CFG3; RUN: opt < %s -passes='sroa<modify-cfg>' -S | FileCheck %s --check-prefixes=CHECK,CHECK-MODIFY-CFG4;5; Make sure the llvm.access.group meta-data is preserved6; when a load/store is replaced with another load/store by sroa7;8; class Complex {9; private:10; float real_;11; float imaginary_;12;13; public:14; Complex() : real_(0), imaginary_(0) { }15; Complex(float real, float imaginary) : real_(real), imaginary_(imaginary) { }16; Complex(const Complex &rhs) : real_(rhs.real()), imaginary_(rhs.imaginary()) { }17;18; inline float real() const { return real_; }19; inline float imaginary() const { return imaginary_; }20;21; Complex operator+(const Complex& rhs) const22; {23; return Complex(real_ + rhs.real_, imaginary_ + rhs.imaginary_);24; }25; };26;27; void test(Complex *out, long size)28; {29; #pragma clang loop vectorize(assume_safety)30; for (long offset = 0; offset < size; ++offset) {31; Complex t0 = out[offset];32; out[offset] = t0 + t0;33; }34; }35 36; ModuleID = '<stdin>'37source_filename = "mem-par-metadata-sroa1.cpp"38target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"39target triple = "x86_64-unknown-linux-gnu"40 41%class.Complex = type { float, float }42 43; Function Attrs: norecurse nounwind uwtable44define void @_Z4testP7Complexl(ptr nocapture %out, i64 %size) local_unnamed_addr {45; CHECK-LABEL: @_Z4testP7Complexl(46; CHECK-NEXT: entry:47; CHECK-NEXT: br label [[FOR_COND:%.*]]48; CHECK: for.cond:49; CHECK-NEXT: [[OFFSET_0:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_BODY:%.*]] ]50; CHECK-NEXT: [[CMP:%.*]] = icmp slt i64 [[OFFSET_0]], [[SIZE:%.*]]51; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]]52; CHECK: for.body:53; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [[CLASS_COMPLEX:%.*]], ptr [[OUT:%.*]], i64 [[OFFSET_0]]54; CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[ARRAYIDX]], align 4, !llvm.access.group [[ACC_GRP1:![0-9]+]]55; CHECK-NEXT: [[IMAGINARY__I_I:%.*]] = getelementptr inbounds [[CLASS_COMPLEX]], ptr [[ARRAYIDX]], i64 0, i32 156; CHECK-NEXT: [[TMP1:%.*]] = load float, ptr [[IMAGINARY__I_I]], align 4, !llvm.access.group [[ACC_GRP1]]57; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds [[CLASS_COMPLEX]], ptr [[OUT]], i64 [[OFFSET_0]]58; CHECK-NEXT: [[ADD_I:%.*]] = fadd float [[TMP0]], [[TMP0]]59; CHECK-NEXT: [[ADD4_I:%.*]] = fadd float [[TMP1]], [[TMP1]]60; CHECK-NEXT: [[TMP2:%.*]] = bitcast float [[ADD_I]] to i3261; CHECK-NEXT: [[TMP3:%.*]] = bitcast float [[ADD4_I]] to i3262; CHECK-NEXT: store i32 [[TMP2]], ptr [[ARRAYIDX1]], align 4, !llvm.access.group [[ACC_GRP1]]63; CHECK-NEXT: [[ARRAYIDX1_SROA_IDX:%.*]] = getelementptr inbounds i8, ptr [[ARRAYIDX1]], i64 464; CHECK-NEXT: store i32 [[TMP3]], ptr [[ARRAYIDX1_SROA_IDX]], align 4, !llvm.access.group [[ACC_GRP1]]65; CHECK-NEXT: [[INC]] = add nsw i64 [[OFFSET_0]], 166; CHECK-NEXT: br label [[FOR_COND]], !llvm.loop [[LOOP2:![0-9]+]]67; CHECK: for.end:68; CHECK-NEXT: ret void69;70entry:71 %t0 = alloca %class.Complex, align 472 %ref.tmp = alloca i64, align 873 br label %for.cond74 75for.cond: ; preds = %for.body, %entry76 %offset.0 = phi i64 [ 0, %entry ], [ %inc, %for.body ]77 %cmp = icmp slt i64 %offset.0, %size78 br i1 %cmp, label %for.body, label %for.end79 80for.body: ; preds = %for.cond81 %arrayidx = getelementptr inbounds %class.Complex, ptr %out, i64 %offset.082 %0 = load float, ptr %arrayidx, align 4, !llvm.access.group !1183 store float %0, ptr %t0, align 4, !llvm.access.group !1184 %imaginary_.i = getelementptr inbounds %class.Complex, ptr %t0, i64 0, i32 185 %imaginary_.i.i = getelementptr inbounds %class.Complex, ptr %arrayidx, i64 0, i32 186 %1 = load float, ptr %imaginary_.i.i, align 4, !llvm.access.group !1187 store float %1, ptr %imaginary_.i, align 4, !llvm.access.group !1188 %arrayidx1 = getelementptr inbounds %class.Complex, ptr %out, i64 %offset.089 %2 = load float, ptr %t0, align 4, !noalias !3, !llvm.access.group !1190 %3 = load float, ptr %t0, align 4, !noalias !3, !llvm.access.group !1191 %add.i = fadd float %2, %392 %imaginary_.i2 = getelementptr inbounds %class.Complex, ptr %t0, i64 0, i32 193 %4 = load float, ptr %imaginary_.i2, align 4, !noalias !3, !llvm.access.group !1194 %imaginary_3.i = getelementptr inbounds %class.Complex, ptr %t0, i64 0, i32 195 %5 = load float, ptr %imaginary_3.i, align 4, !noalias !3, !llvm.access.group !1196 %add4.i = fadd float %4, %597 store float %add.i, ptr %ref.tmp, align 4, !alias.scope !3, !llvm.access.group !1198 %imaginary_.i.i4 = getelementptr inbounds %class.Complex, ptr %ref.tmp, i64 0, i32 199 store float %add4.i, ptr %imaginary_.i.i4, align 4, !alias.scope !3, !llvm.access.group !11100 %6 = load i64, ptr %ref.tmp, align 8, !llvm.access.group !11101 store i64 %6, ptr %arrayidx1, align 4, !llvm.access.group !11102 %inc = add nsw i64 %offset.0, 1103 br label %for.cond, !llvm.loop !1104 105for.end: ; preds = %for.cond106 ret void107}108 109; Function Attrs: argmemonly nounwind110declare void @llvm.memcpy.p0.p0.i64(ptr nocapture writeonly, ptr nocapture readonly, i64, i1)111 112!llvm.ident = !{!0}113 114!0 = !{!"clang version 4.0.0 (cfe/trunk 277751)"}115!1 = distinct !{!1, !2, !{!"llvm.loop.parallel_accesses", !11}}116!2 = !{!"llvm.loop.vectorize.enable", i1 true}117!3 = !{!4}118!4 = distinct !{!4, !5, !"_ZNK7ComplexplERKS_: %agg.result"}119!5 = distinct !{!5, !"_ZNK7ComplexplERKS_"}120!11 = distinct !{}121;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:122; CHECK-MODIFY-CFG: {{.*}}123; CHECK-PRESERVE-CFG: {{.*}}124