98 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -S -mtriple=s390x-unknown-linux -mcpu=z13 -passes=loop-vectorize,dce,instcombine -force-vector-width=2 < %s | FileCheck %s3;4; Test that loop vectorizer does not generate vector addresses that must then5; always be extracted.6 7; Check that the addresses for a scalarized memory access is not extracted8; from a vector register.9define void @foo(ptr nocapture %A) {10; CHECK-LABEL: @foo(11; CHECK-NEXT: entry:12; CHECK-NEXT: br label [[VECTOR_PH:%.*]]13; CHECK: vector.ph:14; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]15; CHECK: vector.body:16; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]17; CHECK-NEXT: [[DOTIDX:%.*]] = shl nsw i64 [[INDEX]], 418; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[A:%.*]], i64 [[DOTIDX]]19; CHECK-NEXT: [[DOTIDX1:%.*]] = shl i64 [[INDEX]], 420; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[A]], i64 [[DOTIDX1]]21; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[TMP1]], i64 1622; CHECK-NEXT: store i32 4, ptr [[TMP2]], align 423; CHECK-NEXT: store i32 4, ptr [[TMP3]], align 424; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 225; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000026; CHECK-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]27; CHECK: middle.block:28; CHECK-NEXT: br label [[FOR_BODY:%.*]]29; CHECK: for.end:30; CHECK-NEXT: ret void31;32 33entry:34 br label %for.body35 36for.body:37 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]38 %0 = shl nsw i64 %indvars.iv, 239 %arrayidx = getelementptr inbounds i32, ptr %A, i64 %040 store i32 4, ptr %arrayidx, align 441 %indvars.iv.next = add i64 %indvars.iv, 142 %lftr.wideiv = trunc i64 %indvars.iv.next to i3243 %exitcond = icmp eq i32 %lftr.wideiv, 1000044 br i1 %exitcond, label %for.end, label %for.body45 46for.end:47 ret void48}49 50 51; Check that a load of address is scalarized.52define void @foo1(ptr nocapture noalias %A, ptr nocapture %PtrPtr) {53; CHECK-LABEL: @foo1(54; CHECK-NEXT: entry:55; CHECK-NEXT: br label [[VECTOR_PH:%.*]]56; CHECK: vector.ph:57; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]58; CHECK: vector.body:59; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]60; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds ptr, ptr [[PTRPTR:%.*]], i64 [[INDEX]]61; CHECK-NEXT: [[TMP11:%.*]] = getelementptr ptr, ptr [[PTRPTR]], i64 [[INDEX]]62; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP11]], i64 863; CHECK-NEXT: [[TMP3:%.*]] = load ptr, ptr [[TMP1]], align 864; CHECK-NEXT: [[TMP4:%.*]] = load ptr, ptr [[TMP2]], align 865; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[TMP3]], align 466; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr [[TMP4]], align 467; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x i32> poison, i32 [[TMP5]], i64 068; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i32> [[TMP7]], i32 [[TMP6]], i64 169; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]70; CHECK-NEXT: store <2 x i32> [[TMP8]], ptr [[TMP9]], align 471; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 272; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000073; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]74; CHECK: middle.block:75; CHECK-NEXT: br label [[FOR_BODY:%.*]]76; CHECK: for.end:77; CHECK-NEXT: ret void78;79 80entry:81 br label %for.body82 83for.body:84 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]85 %ptr = getelementptr inbounds ptr, ptr %PtrPtr, i64 %indvars.iv86 %el = load ptr, ptr %ptr87 %v = load i32, ptr %el88 %arrayidx = getelementptr inbounds i32, ptr %A, i64 %indvars.iv89 store i32 %v, ptr %arrayidx, align 490 %indvars.iv.next = add i64 %indvars.iv, 191 %lftr.wideiv = trunc i64 %indvars.iv.next to i3292 %exitcond = icmp eq i32 %lftr.wideiv, 1000093 br i1 %exitcond, label %for.end, label %for.body94 95for.end:96 ret void97}98