36 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 22; RUN: %if x86-registered-target %{ opt -S -passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s %}3; RUN: %if aarch64-registered-target %{ opt -S -passes=slp-vectorizer -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s %}4 5define i32 @test() {6; CHECK-LABEL: define i32 @test() {7; CHECK-NEXT: entry:8; CHECK-NEXT: [[SQ:%.*]] = alloca [64 x i32], i32 0, align 169; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds [64 x i32], ptr [[SQ]], i64 0, i64 110; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[TMP0]], align 411; CHECK-NEXT: [[TMP2:%.*]] = mul <4 x i32> [[TMP1]], <i32 2, i32 3, i32 2, i32 1>12; CHECK-NEXT: [[TMP3:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP2]])13; CHECK-NEXT: ret i32 [[TMP3]]14;15entry:16 %sq = alloca [64 x i32], i32 0, align 1617 %0 = getelementptr inbounds [64 x i32], ptr %sq, i64 0, i64 118 %elt_1 = load i32, ptr %0, align 419 %1 = getelementptr [64 x i32], ptr %sq, i64 0, i64 220 %elt_2 = load i32, ptr %1, align 821 %2 = getelementptr [64 x i32], ptr %sq, i64 0, i64 322 %elt_3 = load i32, ptr %2, align 423 %3 = getelementptr [64 x i32], ptr %sq, i64 0, i64 424 %elt_4 = load i32, ptr %3, align 1625 26 %4 = add i32 %elt_2, %elt_327 %5 = add i32 %4, %elt_228 %6 = add i32 %5, %elt_129 %7 = add i32 %6, %elt_430 %8 = add i32 %7, %elt_331 %9 = add i32 %8, %elt_232 %10 = add i32 %9, %elt_133 34 ret i32 %1035}36