129 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-apple-macosx10.8.0 -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 7; int foo(int * restrict B, int * restrict A, int n, int m) {8; B[0] = n * A[0] + m * A[0];9; B[1] = n * A[1] + m * A[1];10; B[2] = n * A[2] + m * A[2];11; B[3] = n * A[3] + m * A[3];12; return 0;13; }14 15define i32 @foo(ptr noalias nocapture %B, ptr noalias nocapture %A, i32 %n, i32 %m) #0 {16; CHECK-LABEL: @foo(17; CHECK-NEXT: entry:18; CHECK-NEXT: [[MUL238:%.*]] = add i32 [[M:%.*]], [[N:%.*]]19; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[A:%.*]], align 420; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> poison, i32 [[MUL238]], i32 021; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer22; CHECK-NEXT: [[TMP3:%.*]] = mul <4 x i32> [[TMP0]], [[TMP2]]23; CHECK-NEXT: store <4 x i32> [[TMP3]], ptr [[B:%.*]], align 424; CHECK-NEXT: ret i32 025;26entry:27 %0 = load i32, ptr %A, align 428 %mul238 = add i32 %m, %n29 %add = mul i32 %0, %mul23830 store i32 %add, ptr %B, align 431 %arrayidx4 = getelementptr inbounds i32, ptr %A, i64 132 %1 = load i32, ptr %arrayidx4, align 433 %add8 = mul i32 %1, %mul23834 %arrayidx9 = getelementptr inbounds i32, ptr %B, i64 135 store i32 %add8, ptr %arrayidx9, align 436 %arrayidx10 = getelementptr inbounds i32, ptr %A, i64 237 %2 = load i32, ptr %arrayidx10, align 438 %add14 = mul i32 %2, %mul23839 %arrayidx15 = getelementptr inbounds i32, ptr %B, i64 240 store i32 %add14, ptr %arrayidx15, align 441 %arrayidx16 = getelementptr inbounds i32, ptr %A, i64 342 %3 = load i32, ptr %arrayidx16, align 443 %add20 = mul i32 %3, %mul23844 %arrayidx21 = getelementptr inbounds i32, ptr %B, i64 345 store i32 %add20, ptr %arrayidx21, align 446 ret i32 047}48 49 50; int extr_user(int * restrict B, int * restrict A, int n, int m) {51; B[0] = n * A[0] + m * A[0];52; B[1] = n * A[1] + m * A[1];53; B[2] = n * A[2] + m * A[2];54; B[3] = n * A[3] + m * A[3];55; return A[0];56; }57 58define i32 @extr_user(ptr noalias nocapture %B, ptr noalias nocapture %A, i32 %n, i32 %m) {59; CHECK-LABEL: @extr_user(60; CHECK-NEXT: entry:61; CHECK-NEXT: [[MUL238:%.*]] = add i32 [[M:%.*]], [[N:%.*]]62; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[A:%.*]], align 463; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr [[A]], align 464; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> poison, i32 [[MUL238]], i32 065; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> poison, <4 x i32> zeroinitializer66; CHECK-NEXT: [[TMP4:%.*]] = mul <4 x i32> [[TMP0]], [[TMP3]]67; CHECK-NEXT: store <4 x i32> [[TMP4]], ptr [[B:%.*]], align 468; CHECK-NEXT: ret i32 [[TMP1]]69;70entry:71 %0 = load i32, ptr %A, align 472 %mul238 = add i32 %m, %n73 %add = mul i32 %0, %mul23874 store i32 %add, ptr %B, align 475 %arrayidx4 = getelementptr inbounds i32, ptr %A, i64 176 %1 = load i32, ptr %arrayidx4, align 477 %add8 = mul i32 %1, %mul23878 %arrayidx9 = getelementptr inbounds i32, ptr %B, i64 179 store i32 %add8, ptr %arrayidx9, align 480 %arrayidx10 = getelementptr inbounds i32, ptr %A, i64 281 %2 = load i32, ptr %arrayidx10, align 482 %add14 = mul i32 %2, %mul23883 %arrayidx15 = getelementptr inbounds i32, ptr %B, i64 284 store i32 %add14, ptr %arrayidx15, align 485 %arrayidx16 = getelementptr inbounds i32, ptr %A, i64 386 %3 = load i32, ptr %arrayidx16, align 487 %add20 = mul i32 %3, %mul23888 %arrayidx21 = getelementptr inbounds i32, ptr %B, i64 389 store i32 %add20, ptr %arrayidx21, align 490 ret i32 %0 ;<--------- This value has multiple users91}92 93; In this example we have an external user that is not the first element in the vector.94define i32 @extr_user1(ptr noalias nocapture %B, ptr noalias nocapture %A, i32 %n, i32 %m) {95; CHECK-LABEL: @extr_user1(96; CHECK-NEXT: entry:97; CHECK-NEXT: [[MUL238:%.*]] = add i32 [[M:%.*]], [[N:%.*]]98; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 199; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[A]], align 4100; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr [[ARRAYIDX4]], align 4101; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> poison, i32 [[MUL238]], i32 0102; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> poison, <4 x i32> zeroinitializer103; CHECK-NEXT: [[TMP4:%.*]] = mul <4 x i32> [[TMP0]], [[TMP3]]104; CHECK-NEXT: store <4 x i32> [[TMP4]], ptr [[B:%.*]], align 4105; CHECK-NEXT: ret i32 [[TMP1]]106;107entry:108 %0 = load i32, ptr %A, align 4109 %mul238 = add i32 %m, %n110 %add = mul i32 %0, %mul238111 store i32 %add, ptr %B, align 4112 %arrayidx4 = getelementptr inbounds i32, ptr %A, i64 1113 %1 = load i32, ptr %arrayidx4, align 4114 %add8 = mul i32 %1, %mul238115 %arrayidx9 = getelementptr inbounds i32, ptr %B, i64 1116 store i32 %add8, ptr %arrayidx9, align 4117 %arrayidx10 = getelementptr inbounds i32, ptr %A, i64 2118 %2 = load i32, ptr %arrayidx10, align 4119 %add14 = mul i32 %2, %mul238120 %arrayidx15 = getelementptr inbounds i32, ptr %B, i64 2121 store i32 %add14, ptr %arrayidx15, align 4122 %arrayidx16 = getelementptr inbounds i32, ptr %A, i64 3123 %3 = load i32, ptr %arrayidx16, align 4124 %add20 = mul i32 %3, %mul238125 %arrayidx21 = getelementptr inbounds i32, ptr %B, i64 3126 store i32 %add20, ptr %arrayidx21, align 4127 ret i32 %1 ;<--------- This value has multiple users128}129