80 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt < %s -passes=slp-vectorizer,dce -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; SLP vectorization example from http://cs.stanford.edu/people/eschkufz/research/asplos291-schkufza.pdf8define void @SAXPY(ptr noalias nocapture %x, ptr noalias nocapture %y, i32 %a, i64 %i) {9; CHECK-LABEL: @SAXPY(10; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[X:%.*]], i64 [[I:%.*]]11; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[Y:%.*]], i64 [[I]]12; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr [[TMP1]], align 413; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x i32> poison, i32 [[A:%.*]], i32 014; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[TMP5]], <4 x i32> poison, <4 x i32> zeroinitializer15; CHECK-NEXT: [[TMP6:%.*]] = mul nsw <4 x i32> [[TMP4]], [[SHUFFLE]]16; CHECK-NEXT: [[TMP8:%.*]] = load <4 x i32>, ptr [[TMP2]], align 417; CHECK-NEXT: [[TMP9:%.*]] = add nsw <4 x i32> [[TMP6]], [[TMP8]]18; CHECK-NEXT: store <4 x i32> [[TMP9]], ptr [[TMP1]], align 419; CHECK-NEXT: ret void20;21 %1 = getelementptr inbounds i32, ptr %x, i64 %i22 %2 = load i32, ptr %1, align 423 %3 = mul nsw i32 %2, %a24 %4 = getelementptr inbounds i32, ptr %y, i64 %i25 %5 = load i32, ptr %4, align 426 %6 = add nsw i32 %3, %527 store i32 %6, ptr %1, align 428 %7 = add i64 %i, 129 %8 = getelementptr inbounds i32, ptr %x, i64 %730 %9 = load i32, ptr %8, align 431 %10 = mul nsw i32 %9, %a32 %11 = getelementptr inbounds i32, ptr %y, i64 %733 %12 = load i32, ptr %11, align 434 %13 = add nsw i32 %10, %1235 store i32 %13, ptr %8, align 436 %14 = add i64 %i, 237 %15 = getelementptr inbounds i32, ptr %x, i64 %1438 %16 = load i32, ptr %15, align 439 %17 = mul nsw i32 %16, %a40 %18 = getelementptr inbounds i32, ptr %y, i64 %1441 %19 = load i32, ptr %18, align 442 %20 = add nsw i32 %17, %1943 store i32 %20, ptr %15, align 444 %21 = add i64 %i, 345 %22 = getelementptr inbounds i32, ptr %x, i64 %2146 %23 = load i32, ptr %22, align 447 %24 = mul nsw i32 %23, %a48 %25 = getelementptr inbounds i32, ptr %y, i64 %2149 %26 = load i32, ptr %25, align 450 %27 = add nsw i32 %24, %2651 store i32 %27, ptr %22, align 452 ret void53}54 55; Make sure we don't crash on this one.56define void @SAXPY_crash(ptr noalias nocapture %x, ptr noalias nocapture %y, i64 %i) {57; CHECK-LABEL: @SAXPY_crash(58; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[I:%.*]], 159; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i32, ptr [[X:%.*]], i64 [[TMP1]]60; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[Y:%.*]], i64 [[TMP1]]61; CHECK-NEXT: [[TMP5:%.*]] = load <2 x i32>, ptr [[TMP3]], align 462; CHECK-NEXT: [[TMP6:%.*]] = add nsw <2 x i32> undef, [[TMP5]]63; CHECK-NEXT: store <2 x i32> [[TMP6]], ptr [[TMP2]], align 464; CHECK-NEXT: ret void65;66 %1 = add i64 %i, 167 %2 = getelementptr inbounds i32, ptr %x, i64 %168 %3 = getelementptr inbounds i32, ptr %y, i64 %169 %4 = load i32, ptr %3, align 470 %5 = add nsw i32 undef, %471 store i32 %5, ptr %2, align 472 %6 = add i64 %i, 273 %7 = getelementptr inbounds i32, ptr %x, i64 %674 %8 = getelementptr inbounds i32, ptr %y, i64 %675 %9 = load i32, ptr %8, align 476 %10 = add nsw i32 undef, %977 store i32 %10, ptr %7, align 478 ret void79}80