test/Transforms/SLPVectorizer/AArch64/nontemporal.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -S -basic-aa -slp-vectorizer -dce < %s | FileCheck %s
target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
target triple = "arm64-apple-ios5.0.0"

define void @foo(float* noalias %a, float* noalias %b, float* noalias %c) {
; CHECK-LABEL: @foo(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast float* [[B:%.*]] to <4 x float>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4, !nontemporal !0
; CHECK-NEXT:    [[TMP2:%.*]] = bitcast float* [[C:%.*]] to <4 x float>*
; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* [[TMP2]], align 4
; CHECK-NEXT:    [[TMP4:%.*]] = fadd <4 x float> [[TMP1]], [[TMP3]]
; CHECK-NEXT:    [[TMP5:%.*]] = bitcast float* [[A:%.*]] to <4 x float>*
; CHECK-NEXT:    store <4 x float> [[TMP4]], <4 x float>* [[TMP5]], align 4, !nontemporal !0
; CHECK-NEXT:    ret void
;
entry:
; Check that we don't lose !nontemporal hint when vectorizing loads.
  %b1 = load float, float* %b, align 4, !nontemporal !0
  %arrayidx.1 = getelementptr inbounds float, float* %b, i64 1
  %b2 = load float, float* %arrayidx.1, align 4, !nontemporal !0
  %arrayidx.2 = getelementptr inbounds float, float* %b, i64 2
  %b3 = load float, float* %arrayidx.2, align 4, !nontemporal !0
  %arrayidx.3 = getelementptr inbounds float, float* %b, i64 3
  %b4 = load float, float* %arrayidx.3, align 4, !nontemporal !0

; Check that we don't introduce !nontemporal hint when the original scalar loads didn't have it.
  %c1 = load float, float* %c, align 4
  %arrayidx2.1 = getelementptr inbounds float, float* %c, i64 1
  %c2 = load float, float* %arrayidx2.1, align 4
  %arrayidx2.2 = getelementptr inbounds float, float* %c, i64 2
  %c3 = load float, float* %arrayidx2.2, align 4
  %arrayidx2.3 = getelementptr inbounds float, float* %c, i64 3
  %c4 = load float, float* %arrayidx2.3, align 4

  %a1 = fadd float %b1, %c1
  %a2 = fadd float %b2, %c2
  %a3 = fadd float %b3, %c3
  %a4 = fadd float %b4, %c4

; Check that we don't lose !nontemporal hint when vectorizing stores.
  store float %a1, float* %a, align 4, !nontemporal !0
  %arrayidx3.1 = getelementptr inbounds float, float* %a, i64 1
  store float %a2, float* %arrayidx3.1, align 4, !nontemporal !0
  %arrayidx3.2 = getelementptr inbounds float, float* %a, i64 2
  store float %a3, float* %arrayidx3.2, align 4, !nontemporal !0
  %arrayidx3.3 = getelementptr inbounds float, float* %a, i64 3
  store float %a4, float* %arrayidx3.3, align 4, !nontemporal !0

  ret void
}

define void @foo2(float* noalias %a, float* noalias %b) {
; CHECK-LABEL: @foo2(
; CHECK-NEXT:  entry:
; CHECK-NEXT:    [[TMP0:%.*]] = bitcast float* [[B:%.*]] to <4 x float>*
; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
; CHECK-NEXT:    [[TMP2:%.*]] = bitcast float* [[A:%.*]] to <4 x float>*
; CHECK-NEXT:    store <4 x float> [[TMP1]], <4 x float>* [[TMP2]], align 4
; CHECK-NEXT:    ret void
;
entry:
; Check that we don't mark vector load with !nontemporal attribute if some of
; the original scalar loads don't have it.
  %b1 = load float, float* %b, align 4, !nontemporal !0
  %arrayidx.1 = getelementptr inbounds float, float* %b, i64 1
  %b2 = load float, float* %arrayidx.1, align 4
  %arrayidx.2 = getelementptr inbounds float, float* %b, i64 2
  %b3 = load float, float* %arrayidx.2, align 4
  %arrayidx.3 = getelementptr inbounds float, float* %b, i64 3
  %b4 = load float, float* %arrayidx.3, align 4, !nontemporal !0

; Check that we don't mark vector store with !nontemporal attribute if some of
; the original scalar stores don't have it.
  store float %b1, float* %a, align 4, !nontemporal !0
  %arrayidx3.1 = getelementptr inbounds float, float* %a, i64 1
  store float %b2, float* %arrayidx3.1, align 4
  %arrayidx3.2 = getelementptr inbounds float, float* %a, i64 2
  store float %b3, float* %arrayidx3.2, align 4
  %arrayidx3.3 = getelementptr inbounds float, float* %a, i64 3
  store float %b4, float* %arrayidx3.3, align 4, !nontemporal !0

  ret void
}

!0 = !{i32 1}