81 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=slp-vectorizer -S -mcpu=corei7-avx | FileCheck %s3 4target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"5target triple = "x86_64-apple-macosx10.8.0"6 7define i32 @foo(ptr nocapture %A, i32 %n) {8; CHECK-LABEL: @foo(9; CHECK-NEXT: entry:10; CHECK-NEXT: [[CMP62:%.*]] = icmp sgt i32 [[N:%.*]], 011; CHECK-NEXT: br i1 [[CMP62]], label [[FOR_BODY:%.*]], label [[FOR_END:%.*]]12; CHECK: for.body:13; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ]14; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDVARS_IV]]15; CHECK-NEXT: [[TMP0:%.*]] = load <8 x i32>, ptr [[ARRAYIDX]], align 416; CHECK-NEXT: [[TMP1:%.*]] = insertelement <8 x i32> poison, i32 [[N]], i32 017; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <8 x i32> [[TMP1]], <8 x i32> poison, <8 x i32> zeroinitializer18; CHECK-NEXT: [[TMP3:%.*]] = add nsw <8 x i32> [[TMP0]], [[TMP2]]19; CHECK-NEXT: store <8 x i32> [[TMP3]], ptr [[ARRAYIDX]], align 420; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 821; CHECK-NEXT: [[TMP4:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i3222; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[TMP4]], [[N]]23; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END]]24; CHECK: for.end:25; CHECK-NEXT: ret i32 undef26;27entry:28 %cmp62 = icmp sgt i32 %n, 029 br i1 %cmp62, label %for.body, label %for.end30 31for.body:32 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]33 %arrayidx = getelementptr inbounds i32, ptr %A, i64 %indvars.iv34 %0 = load i32, ptr %arrayidx, align 435 %add1 = add nsw i32 %0, %n36 store i32 %add1, ptr %arrayidx, align 437 %1 = or disjoint i64 %indvars.iv, 138 %arrayidx4 = getelementptr inbounds i32, ptr %A, i64 %139 %2 = load i32, ptr %arrayidx4, align 440 %add5 = add nsw i32 %2, %n41 store i32 %add5, ptr %arrayidx4, align 442 %3 = or disjoint i64 %indvars.iv, 243 %arrayidx8 = getelementptr inbounds i32, ptr %A, i64 %344 %4 = load i32, ptr %arrayidx8, align 445 %add9 = add nsw i32 %4, %n46 store i32 %add9, ptr %arrayidx8, align 447 %5 = or disjoint i64 %indvars.iv, 348 %arrayidx12 = getelementptr inbounds i32, ptr %A, i64 %549 %6 = load i32, ptr %arrayidx12, align 450 %add13 = add nsw i32 %6, %n51 store i32 %add13, ptr %arrayidx12, align 452 %7 = or disjoint i64 %indvars.iv, 453 %arrayidx16 = getelementptr inbounds i32, ptr %A, i64 %754 %8 = load i32, ptr %arrayidx16, align 455 %add17 = add nsw i32 %8, %n56 store i32 %add17, ptr %arrayidx16, align 457 %9 = or disjoint i64 %indvars.iv, 558 %arrayidx20 = getelementptr inbounds i32, ptr %A, i64 %959 %10 = load i32, ptr %arrayidx20, align 460 %add21 = add nsw i32 %10, %n61 store i32 %add21, ptr %arrayidx20, align 462 %11 = or disjoint i64 %indvars.iv, 663 %arrayidx24 = getelementptr inbounds i32, ptr %A, i64 %1164 %12 = load i32, ptr %arrayidx24, align 465 %add25 = add nsw i32 %12, %n66 store i32 %add25, ptr %arrayidx24, align 467 %13 = or disjoint i64 %indvars.iv, 768 %arrayidx28 = getelementptr inbounds i32, ptr %A, i64 %1369 %14 = load i32, ptr %arrayidx28, align 470 %add29 = add nsw i32 %14, %n71 store i32 %add29, ptr %arrayidx28, align 472 %indvars.iv.next = add i64 %indvars.iv, 873 %15 = trunc i64 %indvars.iv.next to i3274 %cmp = icmp slt i32 %15, %n75 br i1 %cmp, label %for.body, label %for.end76 77for.end:78 ret i32 undef79}80 81