llvm/test/Transforms/SLPVectorizer/WebAssembly/simd-splat-shuffle-cost.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -mtriple=wasm32 -mattr=+simd128 -passes=slp-vectorizer -S %s | FileCheck %s --check-prefix=SIMD128
; RUN: opt -mtriple=wasm32 -passes=slp-vectorizer -S %s | FileCheck %s --check-prefix=NO-SIMD128

; Test that the SLP vectorizer correctly handles splat patterns when
; shuffle costs are properly modeled for WebAssembly SIMD128.

define void @splat_i8x16(i8 %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_i8x16(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    [[TMP0:%.*]] = insertelement <16 x i8> poison, i8 [[V:%.*]], i32 0
; SIMD128-NEXT:    [[TMP1:%.*]] = shufflevector <16 x i8> [[TMP0]], <16 x i8> poison, <16 x i32> zeroinitializer
; SIMD128-NEXT:    store <16 x i8> [[TMP1]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_i8x16(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store i8 [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 1
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX1]], align 1
; NO-SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 2
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX2]], align 1
; NO-SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 3
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX3]], align 1
; NO-SIMD128-NEXT:    [[IDX4:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX4]], align 1
; NO-SIMD128-NEXT:    [[IDX5:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 5
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX5]], align 1
; NO-SIMD128-NEXT:    [[IDX6:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 6
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX6]], align 1
; NO-SIMD128-NEXT:    [[IDX7:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 7
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX7]], align 1
; NO-SIMD128-NEXT:    [[IDX8:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX8]], align 1
; NO-SIMD128-NEXT:    [[IDX9:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 9
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX9]], align 1
; NO-SIMD128-NEXT:    [[IDX10:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 10
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX10]], align 1
; NO-SIMD128-NEXT:    [[IDX11:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 11
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX11]], align 1
; NO-SIMD128-NEXT:    [[IDX12:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 12
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX12]], align 1
; NO-SIMD128-NEXT:    [[IDX13:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 13
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX13]], align 1
; NO-SIMD128-NEXT:    [[IDX14:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 14
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX14]], align 1
; NO-SIMD128-NEXT:    [[IDX15:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 15
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX15]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store i8 %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 1
  store i8 %v, ptr %idx1, align 1
  %idx2 = getelementptr inbounds nuw i8, ptr %p, i32 2
  store i8 %v, ptr %idx2, align 1
  %idx3 = getelementptr inbounds nuw i8, ptr %p, i32 3
  store i8 %v, ptr %idx3, align 1
  %idx4 = getelementptr inbounds nuw i8, ptr %p, i32 4
  store i8 %v, ptr %idx4, align 1
  %idx5 = getelementptr inbounds nuw i8, ptr %p, i32 5
  store i8 %v, ptr %idx5, align 1
  %idx6 = getelementptr inbounds nuw i8, ptr %p, i32 6
  store i8 %v, ptr %idx6, align 1
  %idx7 = getelementptr inbounds nuw i8, ptr %p, i32 7
  store i8 %v, ptr %idx7, align 1
  %idx8 = getelementptr inbounds nuw i8, ptr %p, i32 8
  store i8 %v, ptr %idx8, align 1
  %idx9 = getelementptr inbounds nuw i8, ptr %p, i32 9
  store i8 %v, ptr %idx9, align 1
  %idx10 = getelementptr inbounds nuw i8, ptr %p, i32 10
  store i8 %v, ptr %idx10, align 1
  %idx11 = getelementptr inbounds nuw i8, ptr %p, i32 11
  store i8 %v, ptr %idx11, align 1
  %idx12 = getelementptr inbounds nuw i8, ptr %p, i32 12
  store i8 %v, ptr %idx12, align 1
  %idx13 = getelementptr inbounds nuw i8, ptr %p, i32 13
  store i8 %v, ptr %idx13, align 1
  %idx14 = getelementptr inbounds nuw i8, ptr %p, i32 14
  store i8 %v, ptr %idx14, align 1
  %idx15 = getelementptr inbounds nuw i8, ptr %p, i32 15
  store i8 %v, ptr %idx15, align 1
  ret void
}

define void @splat_i16x8(i16 %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_i16x8(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    [[TMP0:%.*]] = insertelement <8 x i16> poison, i16 [[V:%.*]], i32 0
; SIMD128-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i16> [[TMP0]], <8 x i16> poison, <8 x i32> zeroinitializer
; SIMD128-NEXT:    store <8 x i16> [[TMP1]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_i16x8(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store i16 [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 2
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX1]], align 1
; NO-SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX2]], align 1
; NO-SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 6
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX3]], align 1
; NO-SIMD128-NEXT:    [[IDX4:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX4]], align 1
; NO-SIMD128-NEXT:    [[IDX5:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 10
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX5]], align 1
; NO-SIMD128-NEXT:    [[IDX6:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 12
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX6]], align 1
; NO-SIMD128-NEXT:    [[IDX7:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 14
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX7]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store i16 %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 2
  store i16 %v, ptr %idx1, align 1
  %idx2 = getelementptr inbounds nuw i8, ptr %p, i32 4
  store i16 %v, ptr %idx2, align 1
  %idx3 = getelementptr inbounds nuw i8, ptr %p, i32 6
  store i16 %v, ptr %idx3, align 1
  %idx4 = getelementptr inbounds nuw i8, ptr %p, i32 8
  store i16 %v, ptr %idx4, align 1
  %idx5 = getelementptr inbounds nuw i8, ptr %p, i32 10
  store i16 %v, ptr %idx5, align 1
  %idx6 = getelementptr inbounds nuw i8, ptr %p, i32 12
  store i16 %v, ptr %idx6, align 1
  %idx7 = getelementptr inbounds nuw i8, ptr %p, i32 14
  store i16 %v, ptr %idx7, align 1
  ret void
}

define void @splat_i32x4(i32 %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_i32x4(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    store i32 [[V:%.*]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; SIMD128-NEXT:    store i32 [[V]], ptr [[IDX1]], align 1
; SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; SIMD128-NEXT:    store i32 [[V]], ptr [[IDX2]], align 1
; SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 12
; SIMD128-NEXT:    store i32 [[V]], ptr [[IDX3]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_i32x4(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store i32 [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX2]], align 1
; NO-SIMD128-NEXT:    [[IDX4:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX4]], align 1
; NO-SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 12
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX3]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store i32 %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 4
  store i32 %v, ptr %idx1, align 1
  %idx2 = getelementptr inbounds nuw i8, ptr %p, i32 8
  store i32 %v, ptr %idx2, align 1
  %idx3 = getelementptr inbounds nuw i8, ptr %p, i32 12
  store i32 %v, ptr %idx3, align 1
  ret void
}


define void @splat_i64x2(i64 %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_i64x2(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    store i64 [[V:%.*]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; SIMD128-NEXT:    store i64 [[V]], ptr [[IDX1]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_i64x2(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store i64 [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; NO-SIMD128-NEXT:    store i64 [[V]], ptr [[IDX1]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store i64 %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 8
  store i64 %v, ptr %idx1, align 1
  ret void
}

define void @splat_f32x4(float %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_f32x4(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    store float [[V:%.*]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; SIMD128-NEXT:    store float [[V]], ptr [[IDX1]], align 1
; SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; SIMD128-NEXT:    store float [[V]], ptr [[IDX2]], align 1
; SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 12
; SIMD128-NEXT:    store float [[V]], ptr [[IDX3]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_f32x4(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store float [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; NO-SIMD128-NEXT:    store float [[V]], ptr [[IDX1]], align 1
; NO-SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; NO-SIMD128-NEXT:    store float [[V]], ptr [[IDX2]], align 1
; NO-SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 12
; NO-SIMD128-NEXT:    store float [[V]], ptr [[IDX3]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store float %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 4
  store float %v, ptr %idx1, align 1
  %idx2 = getelementptr inbounds nuw i8, ptr %p, i32 8
  store float %v, ptr %idx2, align 1
  %idx3 = getelementptr inbounds nuw i8, ptr %p, i32 12
  store float %v, ptr %idx3, align 1
  ret void
}

define void @splat_f64x2(double %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_f64x2(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    store double [[V:%.*]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; SIMD128-NEXT:    store double [[V]], ptr [[IDX1]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_f64x2(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store double [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; NO-SIMD128-NEXT:    store double [[V]], ptr [[IDX1]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store double %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 8
  store double %v, ptr %idx1, align 1
  ret void
}

define void @splat_i32x8(i32 %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_i32x8(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    [[TMP0:%.*]] = insertelement <8 x i32> poison, i32 [[V:%.*]], i32 0
; SIMD128-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i32> [[TMP0]], <8 x i32> poison, <8 x i32> zeroinitializer
; SIMD128-NEXT:    store <8 x i32> [[TMP1]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_i32x8(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store i32 [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX1]], align 1
; NO-SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 8
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX2]], align 1
; NO-SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 12
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX3]], align 1
; NO-SIMD128-NEXT:    [[IDX4:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 16
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX4]], align 1
; NO-SIMD128-NEXT:    [[IDX5:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 20
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX5]], align 1
; NO-SIMD128-NEXT:    [[IDX6:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 24
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX6]], align 1
; NO-SIMD128-NEXT:    [[IDX7:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 28
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX7]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store i32 %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 4
  store i32 %v, ptr %idx1, align 1
  %idx2 = getelementptr inbounds nuw i8, ptr %p, i32 8
  store i32 %v, ptr %idx2, align 1
  %idx3 = getelementptr inbounds nuw i8, ptr %p, i32 12
  store i32 %v, ptr %idx3, align 1
  %idx4 = getelementptr inbounds nuw i8, ptr %p, i32 16
  store i32 %v, ptr %idx4, align 1
  %idx5 = getelementptr inbounds nuw i8, ptr %p, i32 20
  store i32 %v, ptr %idx5, align 1
  %idx6 = getelementptr inbounds nuw i8, ptr %p, i32 24
  store i32 %v, ptr %idx6, align 1
  %idx7 = getelementptr inbounds nuw i8, ptr %p, i32 28
  store i32 %v, ptr %idx7, align 1
  ret void
}

; Negative test sections, non-128 bit versions not vectorize.

define void @splat_i32x2(i32 %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_i32x2(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    store i32 [[V:%.*]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; SIMD128-NEXT:    store i32 [[V]], ptr [[IDX1]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_i32x2(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store i32 [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; NO-SIMD128-NEXT:    store i32 [[V]], ptr [[IDX1]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store i32 %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 4
  store i32 %v, ptr %idx1, align 1
  ret void
}

define void @splat_i16x4(i16 %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_i16x4(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    store i16 [[V:%.*]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 2
; SIMD128-NEXT:    store i16 [[V]], ptr [[IDX1]], align 1
; SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; SIMD128-NEXT:    store i16 [[V]], ptr [[IDX2]], align 1
; SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 6
; SIMD128-NEXT:    store i16 [[V]], ptr [[IDX3]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_i16x4(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store i16 [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 2
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX1]], align 1
; NO-SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX2]], align 1
; NO-SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 6
; NO-SIMD128-NEXT:    store i16 [[V]], ptr [[IDX3]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store i16 %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 2
  store i16 %v, ptr %idx1, align 1
  %idx2 = getelementptr inbounds nuw i8, ptr %p, i32 4
  store i16 %v, ptr %idx2, align 1
  %idx3 = getelementptr inbounds nuw i8, ptr %p, i32 6
  store i16 %v, ptr %idx3, align 1
  ret void
}

define void @splat_i8x8(i8 %v, ptr noalias %p) {
; SIMD128-LABEL: @splat_i8x8(
; SIMD128-NEXT:  entry:
; SIMD128-NEXT:    store i8 [[V:%.*]], ptr [[P:%.*]], align 1
; SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 1
; SIMD128-NEXT:    store i8 [[V]], ptr [[IDX1]], align 1
; SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 2
; SIMD128-NEXT:    store i8 [[V]], ptr [[IDX2]], align 1
; SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 3
; SIMD128-NEXT:    store i8 [[V]], ptr [[IDX3]], align 1
; SIMD128-NEXT:    [[IDX4:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; SIMD128-NEXT:    store i8 [[V]], ptr [[IDX4]], align 1
; SIMD128-NEXT:    [[IDX5:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 5
; SIMD128-NEXT:    store i8 [[V]], ptr [[IDX5]], align 1
; SIMD128-NEXT:    [[IDX6:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 6
; SIMD128-NEXT:    store i8 [[V]], ptr [[IDX6]], align 1
; SIMD128-NEXT:    [[IDX7:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 7
; SIMD128-NEXT:    store i8 [[V]], ptr [[IDX7]], align 1
; SIMD128-NEXT:    ret void
;
; NO-SIMD128-LABEL: @splat_i8x8(
; NO-SIMD128-NEXT:  entry:
; NO-SIMD128-NEXT:    store i8 [[V:%.*]], ptr [[P:%.*]], align 1
; NO-SIMD128-NEXT:    [[IDX1:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 1
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX1]], align 1
; NO-SIMD128-NEXT:    [[IDX2:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 2
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX2]], align 1
; NO-SIMD128-NEXT:    [[IDX3:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 3
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX3]], align 1
; NO-SIMD128-NEXT:    [[IDX4:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 4
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX4]], align 1
; NO-SIMD128-NEXT:    [[IDX5:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 5
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX5]], align 1
; NO-SIMD128-NEXT:    [[IDX6:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 6
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX6]], align 1
; NO-SIMD128-NEXT:    [[IDX7:%.*]] = getelementptr inbounds nuw i8, ptr [[P]], i32 7
; NO-SIMD128-NEXT:    store i8 [[V]], ptr [[IDX7]], align 1
; NO-SIMD128-NEXT:    ret void
;
entry:
  store i8 %v, ptr %p, align 1
  %idx1 = getelementptr inbounds nuw i8, ptr %p, i32 1
  store i8 %v, ptr %idx1, align 1
  %idx2 = getelementptr inbounds nuw i8, ptr %p, i32 2
  store i8 %v, ptr %idx2, align 1
  %idx3 = getelementptr inbounds nuw i8, ptr %p, i32 3
  store i8 %v, ptr %idx3, align 1
  %idx4 = getelementptr inbounds nuw i8, ptr %p, i32 4
  store i8 %v, ptr %idx4, align 1
  %idx5 = getelementptr inbounds nuw i8, ptr %p, i32 5
  store i8 %v, ptr %idx5, align 1
  %idx6 = getelementptr inbounds nuw i8, ptr %p, i32 6
  store i8 %v, ptr %idx6, align 1
  %idx7 = getelementptr inbounds nuw i8, ptr %p, i32 7
  store i8 %v, ptr %idx7, align 1
  ret void
}