32 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: %if x86-registered-target %{ opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown-linux -slp-threshold=-1 | FileCheck %s %}3; RUN: %if aarch64-registered-target %{ opt < %s -passes=slp-vectorizer -S -mtriple=aarch64-unknown-linux-gnu -slp-threshold=-1 | FileCheck %s %}4 5define i32 @diamond_broadcast(ptr noalias nocapture %B, ptr noalias nocapture %A) {6; CHECK-LABEL: @diamond_broadcast(7; CHECK-NEXT: entry:8; CHECK-NEXT: [[LD:%.*]] = load i32, ptr [[A:%.*]], align 49; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x i32> poison, i32 [[LD]], i32 010; CHECK-NEXT: [[SHUFFLE:%.*]] = shufflevector <4 x i32> [[TMP0]], <4 x i32> poison, <4 x i32> zeroinitializer11; CHECK-NEXT: [[TMP1:%.*]] = mul <4 x i32> [[SHUFFLE]], [[SHUFFLE]]12; CHECK-NEXT: store <4 x i32> [[TMP1]], ptr [[B:%.*]], align 413; CHECK-NEXT: ret i32 014;15entry:16 %ld = load i32, ptr %A, align 417 %mul = mul i32 %ld, %ld18 store i32 %mul, ptr %B, align 419 %mul8 = mul i32 %ld, %ld20 %arrayidx9 = getelementptr inbounds i32, ptr %B, i64 121 store i32 %mul8, ptr %arrayidx9, align 422 %mul14 = mul i32 %ld, %ld23 %arrayidx15 = getelementptr inbounds i32, ptr %B, i64 224 store i32 %mul14, ptr %arrayidx15, align 425 %mul20 = mul i32 %ld, %ld26 %arrayidx21 = getelementptr inbounds i32, ptr %B, i64 327 store i32 %mul20, ptr %arrayidx21, align 428 ret i32 029}30 31 32