31 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 62; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-apple-macosx15.0.0 -mcpu=skylake-avx512 -S < %s | FileCheck %s3 4define void @test(ptr %output) {5; CHECK-LABEL: define void @test(6; CHECK-SAME: ptr [[OUTPUT:%.*]]) #[[ATTR0:[0-9]+]] {7; CHECK-NEXT: [[ENTRY:.*:]]8; CHECK-NEXT: [[ARRAYIDX_2_I:%.*]] = getelementptr i8, ptr [[OUTPUT]], i64 89; CHECK-NEXT: [[TMP0:%.*]] = call <4 x i32> @llvm.masked.load.v4i32.p0(ptr align 4 [[OUTPUT]], <4 x i1> <i1 true, i1 false, i1 false, i1 true>, <4 x i32> poison)10; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x i32> [[TMP0]], <4 x i32> poison, <2 x i32> <i32 0, i32 3>11; CHECK-NEXT: [[TMP2:%.*]] = xor <2 x i32> <i32 -1, i32 0>, [[TMP1]]12; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <2 x i32> [[TMP1]], <2 x i32> <i32 0, i32 poison>, <2 x i32> <i32 2, i32 1>13; CHECK-NEXT: [[TMP4:%.*]] = call <2 x i32> @llvm.fshl.v2i32(<2 x i32> [[TMP2]], <2 x i32> [[TMP3]], <2 x i32> splat (i32 2))14; CHECK-NEXT: store <2 x i32> [[TMP4]], ptr [[ARRAYIDX_2_I]], align 415; CHECK-NEXT: ret void16;17entry:18 %arrayidx.2.i = getelementptr i8, ptr %output, i64 819 %0 = load i32, ptr %output, align 420 %arrayidx.3.i = getelementptr i8, ptr %output, i64 1221 %1 = load i32, ptr %arrayidx.3.i, align 422 %xor7 = xor i32 -1, %023 %or.i = tail call i32 @llvm.fshl.i32(i32 %xor7, i32 0, i32 2)24 %or.i11 = tail call i32 @llvm.fshl.i32(i32 %1, i32 %1, i32 2)25 store i32 %or.i, ptr %arrayidx.2.i, align 426 store i32 %or.i11, ptr %arrayidx.3.i, align 427 ret void28}29 30declare i32 @llvm.fshl.i32(i32, i32, i32)31