llvm/test/Analysis/CostModel/X86/load-broadcast.ll

; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 4
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 | FileCheck %s -check-prefixes=SSE,SSE2
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+ssse3 | FileCheck %s -check-prefixes=SSE,SSSE3
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s -check-prefixes=SSE,SSE42
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx | FileCheck %s -check-prefixes=AVX,AVX1
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 | FileCheck %s -check-prefixes=AVX,AVX2
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,AVX512
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX,AVX512
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f,+avx512bw,+avx512vbmi | FileCheck %s --check-prefixes=AVX,AVX512
;
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mcpu=slm | FileCheck %s --check-prefixes=SSE,SSE42
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mcpu=goldmont | FileCheck %s --check-prefixes=SSE,SSE42
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mcpu=btver2 | FileCheck %s --check-prefixes=AVX,AVX1

;
; vXf64
;

define <2 x double> @broadcast_load_v2f64_v2f64(ptr %src) {
; SSE2-LABEL: 'broadcast_load_v2f64_v2f64'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <2 x i32> zeroinitializer
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x double> %bcst
;
; SSSE3-LABEL: 'broadcast_load_v2f64_v2f64'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <2 x i32> zeroinitializer
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x double> %bcst
;
; SSE42-LABEL: 'broadcast_load_v2f64_v2f64'
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <2 x i32> zeroinitializer
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x double> %bcst
;
; AVX-LABEL: 'broadcast_load_v2f64_v2f64'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <2 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x double> %bcst
;
  %load = load <2 x double>, ptr %src
  %bcst = shufflevector <2 x double> %load, <2 x double> poison, <2 x i32> zeroinitializer
  ret <2 x double> %bcst
}

define <4 x double> @broadcast_load_v2f64_v4f64(ptr %src) {
; SSE2-LABEL: 'broadcast_load_v2f64_v4f64'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <4 x i32> zeroinitializer
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x double> %bcst
;
; SSSE3-LABEL: 'broadcast_load_v2f64_v4f64'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <4 x i32> zeroinitializer
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x double> %bcst
;
; SSE42-LABEL: 'broadcast_load_v2f64_v4f64'
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <4 x i32> zeroinitializer
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x double> %bcst
;
; AVX-LABEL: 'broadcast_load_v2f64_v4f64'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <4 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x double> %bcst
;
  %load = load <2 x double>, ptr %src
  %bcst = shufflevector <2 x double> %load, <2 x double> poison, <4 x i32> zeroinitializer
  ret <4 x double> %bcst
}

define <8 x double> @broadcast_load_v2f64_v8f64(ptr %src) {
; SSE2-LABEL: 'broadcast_load_v2f64_v8f64'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <8 x i32> zeroinitializer
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x double> %bcst
;
; SSSE3-LABEL: 'broadcast_load_v2f64_v8f64'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <8 x i32> zeroinitializer
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x double> %bcst
;
; SSE42-LABEL: 'broadcast_load_v2f64_v8f64'
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <8 x i32> zeroinitializer
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x double> %bcst
;
; AVX-LABEL: 'broadcast_load_v2f64_v8f64'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x double>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x double> %load, <2 x double> poison, <8 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x double> %bcst
;
  %load = load <2 x double>, ptr %src
  %bcst = shufflevector <2 x double> %load, <2 x double> poison, <8 x i32> zeroinitializer
  ret <8 x double> %bcst
}

define <2 x double> @broadcast_load_f64_v2f64(ptr %src) {
; SSE-LABEL: 'broadcast_load_f64_v2f64'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <2 x double> poison, double %load, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <2 x double> %insert, <2 x double> poison, <2 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x double> %bcst
;
; AVX-LABEL: 'broadcast_load_f64_v2f64'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <2 x double> poison, double %load, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <2 x double> %insert, <2 x double> poison, <2 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x double> %bcst
;
  %load = load double, ptr %src
  %insert = insertelement <2 x double> poison, double %load, i32 0
  %bcst = shufflevector <2 x double> %insert, <2 x double> poison, <2 x i32> zeroinitializer
  ret <2 x double> %bcst
}

define <4 x double> @broadcast_load_f64_v4f64(ptr %src) {
; SSE-LABEL: 'broadcast_load_f64_v4f64'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <4 x double> poison, double %load, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <4 x double> %insert, <4 x double> poison, <4 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x double> %bcst
;
; AVX1-LABEL: 'broadcast_load_f64_v4f64'
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <4 x double> poison, double %load, i32 0
; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bcst = shufflevector <4 x double> %insert, <4 x double> poison, <4 x i32> zeroinitializer
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x double> %bcst
;
; AVX2-LABEL: 'broadcast_load_f64_v4f64'
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <4 x double> poison, double %load, i32 0
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <4 x double> %insert, <4 x double> poison, <4 x i32> zeroinitializer
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x double> %bcst
;
; AVX512-LABEL: 'broadcast_load_f64_v4f64'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <4 x double> poison, double %load, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <4 x double> %insert, <4 x double> poison, <4 x i32> zeroinitializer
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x double> %bcst
;
  %load = load double, ptr %src
  %insert = insertelement <4 x double> poison, double %load, i32 0
  %bcst = shufflevector <4 x double> %insert, <4 x double> poison, <4 x i32> zeroinitializer
  ret <4 x double> %bcst
}

define <8 x double> @broadcast_load_f64_v8f64(ptr %src) {
; SSE-LABEL: 'broadcast_load_f64_v8f64'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <8 x double> poison, double %load, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x double> %insert, <8 x double> poison, <8 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x double> %bcst
;
; AVX1-LABEL: 'broadcast_load_f64_v8f64'
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <8 x double> poison, double %load, i32 0
; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bcst = shufflevector <8 x double> %insert, <8 x double> poison, <8 x i32> zeroinitializer
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x double> %bcst
;
; AVX2-LABEL: 'broadcast_load_f64_v8f64'
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <8 x double> poison, double %load, i32 0
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x double> %insert, <8 x double> poison, <8 x i32> zeroinitializer
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x double> %bcst
;
; AVX512-LABEL: 'broadcast_load_f64_v8f64'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load double, ptr %src, align 8
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %insert = insertelement <8 x double> poison, double %load, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x double> %insert, <8 x double> poison, <8 x i32> zeroinitializer
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x double> %bcst
;
  %load = load double, ptr %src
  %insert = insertelement <8 x double> poison, double %load, i32 0
  %bcst = shufflevector <8 x double> %insert, <8 x double> poison, <8 x i32> zeroinitializer
  ret <8 x double> %bcst
}

;
; vXf32
;

define <4 x float> @broadcast_load_v4f32_v4f32(ptr %src) {
; SSE-LABEL: 'broadcast_load_v4f32_v4f32'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x float>, ptr %src, align 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <4 x float> %load, <4 x float> poison, <4 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x float> %bcst
;
; AVX-LABEL: 'broadcast_load_v4f32_v4f32'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x float>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <4 x float> %load, <4 x float> poison, <4 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x float> %bcst
;
  %load = load <4 x float>, ptr %src
  %bcst = shufflevector <4 x float> %load, <4 x float> poison, <4 x i32> zeroinitializer
  ret <4 x float> %bcst
}

define <8 x float> @broadcast_load_v4f32_v8f32(ptr %src) {
; SSE-LABEL: 'broadcast_load_v4f32_v8f32'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x float>, ptr %src, align 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <4 x float> %load, <4 x float> poison, <8 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x float> %bcst
;
; AVX-LABEL: 'broadcast_load_v4f32_v8f32'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x float>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <4 x float> %load, <4 x float> poison, <8 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x float> %bcst
;
  %load = load <4 x float>, ptr %src
  %bcst = shufflevector <4 x float> %load, <4 x float> poison, <8 x i32> zeroinitializer
  ret <8 x float> %bcst
}

define <16 x float> @broadcast_load_v4f32_v16f32(ptr %src) {
; SSE-LABEL: 'broadcast_load_v4f32_v16f32'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x float>, ptr %src, align 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <4 x float> %load, <4 x float> poison, <16 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x float> %bcst
;
; AVX-LABEL: 'broadcast_load_v4f32_v16f32'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x float>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <4 x float> %load, <4 x float> poison, <16 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x float> %bcst
;
  %load = load <4 x float>, ptr %src
  %bcst = shufflevector <4 x float> %load, <4 x float> poison, <16 x i32> zeroinitializer
  ret <16 x float> %bcst
}

define <2 x i64> @broadcast_load_v2i64_v2i64(ptr %src) {
; SSE-LABEL: 'broadcast_load_v2i64_v2i64'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x i64>, ptr %src, align 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <2 x i64> %load, <2 x i64> poison, <2 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %bcst
;
; AVX-LABEL: 'broadcast_load_v2i64_v2i64'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x i64>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x i64> %load, <2 x i64> poison, <2 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %bcst
;
  %load = load <2 x i64>, ptr %src
  %bcst = shufflevector <2 x i64> %load, <2 x i64> poison, <2 x i32> zeroinitializer
  ret <2 x i64> %bcst
}

;
; vXi64
;

define <4 x i64> @broadcast_load_v2i64_v4i64(ptr %src) {
; SSE-LABEL: 'broadcast_load_v2i64_v4i64'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x i64>, ptr %src, align 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <2 x i64> %load, <2 x i64> poison, <4 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %bcst
;
; AVX-LABEL: 'broadcast_load_v2i64_v4i64'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x i64>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x i64> %load, <2 x i64> poison, <4 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %bcst
;
  %load = load <2 x i64>, ptr %src
  %bcst = shufflevector <2 x i64> %load, <2 x i64> poison, <4 x i32> zeroinitializer
  ret <4 x i64> %bcst
}

define <8 x i64> @broadcast_load_v2i64_v8i64(ptr %src) {
; SSE-LABEL: 'broadcast_load_v2i64_v8i64'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x i64>, ptr %src, align 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <2 x i64> %load, <2 x i64> poison, <8 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %bcst
;
; AVX-LABEL: 'broadcast_load_v2i64_v8i64'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <2 x i64>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <2 x i64> %load, <2 x i64> poison, <8 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %bcst
;
  %load = load <2 x i64>, ptr %src
  %bcst = shufflevector <2 x i64> %load, <2 x i64> poison, <8 x i32> zeroinitializer
  ret <8 x i64> %bcst
}

;
; vXi32
;

define <4 x i32> @broadcast_load_v4i32_v4i32(ptr %src) {
; SSE-LABEL: 'broadcast_load_v4i32_v4i32'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x i32>, ptr %src, align 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <4 x i32> %load, <4 x i32> poison, <4 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %bcst
;
; AVX-LABEL: 'broadcast_load_v4i32_v4i32'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x i32>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <4 x i32> %load, <4 x i32> poison, <4 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %bcst
;
  %load = load <4 x i32>, ptr %src
  %bcst = shufflevector <4 x i32> %load, <4 x i32> poison, <4 x i32> zeroinitializer
  ret <4 x i32> %bcst
}

define <8 x i32> @broadcast_load_v4i32_v8i32(ptr %src) {
; SSE-LABEL: 'broadcast_load_v4i32_v8i32'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x i32>, ptr %src, align 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <4 x i32> %load, <4 x i32> poison, <8 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %bcst
;
; AVX-LABEL: 'broadcast_load_v4i32_v8i32'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x i32>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <4 x i32> %load, <4 x i32> poison, <8 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %bcst
;
  %load = load <4 x i32>, ptr %src
  %bcst = shufflevector <4 x i32> %load, <4 x i32> poison, <8 x i32> zeroinitializer
  ret <8 x i32> %bcst
}

define <16 x i32> @broadcast_load_v4i32_v16i32(ptr %src) {
; SSE-LABEL: 'broadcast_load_v4i32_v16i32'
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x i32>, ptr %src, align 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <4 x i32> %load, <4 x i32> poison, <16 x i32> zeroinitializer
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %bcst
;
; AVX-LABEL: 'broadcast_load_v4i32_v16i32'
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <4 x i32>, ptr %src, align 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <4 x i32> %load, <4 x i32> poison, <16 x i32> zeroinitializer
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %bcst
;
  %load = load <4 x i32>, ptr %src
  %bcst = shufflevector <4 x i32> %load, <4 x i32> poison, <16 x i32> zeroinitializer
  ret <16 x i32> %bcst
}

;
; vXi16
;

define <8 x i16> @broadcast_load_v8i16_v8i16(ptr %src) {
; SSE2-LABEL: 'broadcast_load_v8i16_v8i16'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <8 x i32> zeroinitializer
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %bcst
;
; SSSE3-LABEL: 'broadcast_load_v8i16_v8i16'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <8 x i32> zeroinitializer
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %bcst
;
; SSE42-LABEL: 'broadcast_load_v8i16_v8i16'
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <8 x i32> zeroinitializer
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %bcst
;
; AVX1-LABEL: 'broadcast_load_v8i16_v8i16'
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <8 x i32> zeroinitializer
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %bcst
;
; AVX2-LABEL: 'broadcast_load_v8i16_v8i16'
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <8 x i32> zeroinitializer
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %bcst
;
; AVX512-LABEL: 'broadcast_load_v8i16_v8i16'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <8 x i32> zeroinitializer
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %bcst
;
  %load = load <8 x i16>, ptr %src
  %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <8 x i32> zeroinitializer
  ret <8 x i16> %bcst
}

define <16 x i16> @broadcast_load_v8i16_v16i16(ptr %src) {
; SSE2-LABEL: 'broadcast_load_v8i16_v16i16'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <16 x i32> zeroinitializer
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bcst
;
; SSSE3-LABEL: 'broadcast_load_v8i16_v16i16'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <16 x i32> zeroinitializer
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bcst
;
; SSE42-LABEL: 'broadcast_load_v8i16_v16i16'
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <16 x i32> zeroinitializer
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bcst
;
; AVX1-LABEL: 'broadcast_load_v8i16_v16i16'
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <16 x i32> zeroinitializer
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bcst
;
; AVX2-LABEL: 'broadcast_load_v8i16_v16i16'
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <16 x i32> zeroinitializer
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bcst
;
; AVX512-LABEL: 'broadcast_load_v8i16_v16i16'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <16 x i32> zeroinitializer
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %bcst
;
  %load = load <8 x i16>, ptr %src
  %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <16 x i32> zeroinitializer
  ret <16 x i16> %bcst
}

define <32 x i16> @broadcast_load_v8i16_v32i16(ptr %src) {
; SSE2-LABEL: 'broadcast_load_v8i16_v32i16'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <32 x i32> zeroinitializer
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bcst
;
; SSSE3-LABEL: 'broadcast_load_v8i16_v32i16'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <32 x i32> zeroinitializer
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bcst
;
; SSE42-LABEL: 'broadcast_load_v8i16_v32i16'
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <32 x i32> zeroinitializer
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bcst
;
; AVX1-LABEL: 'broadcast_load_v8i16_v32i16'
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <32 x i32> zeroinitializer
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bcst
;
; AVX2-LABEL: 'broadcast_load_v8i16_v32i16'
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <32 x i32> zeroinitializer
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bcst
;
; AVX512-LABEL: 'broadcast_load_v8i16_v32i16'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <8 x i16>, ptr %src, align 16
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <32 x i32> zeroinitializer
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %bcst
;
  %load = load <8 x i16>, ptr %src
  %bcst = shufflevector <8 x i16> %load, <8 x i16> poison, <32 x i32> zeroinitializer
  ret <32 x i16> %bcst
}

;
; vXi8
;

define <16 x i8> @broadcast_load_v16i8_v16i8(ptr %src) {
; SSE2-LABEL: 'broadcast_load_v16i8_v16i8'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <16 x i32> zeroinitializer
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %bcst
;
; SSSE3-LABEL: 'broadcast_load_v16i8_v16i8'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <16 x i32> zeroinitializer
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %bcst
;
; SSE42-LABEL: 'broadcast_load_v16i8_v16i8'
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <16 x i32> zeroinitializer
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %bcst
;
; AVX1-LABEL: 'broadcast_load_v16i8_v16i8'
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <16 x i32> zeroinitializer
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %bcst
;
; AVX2-LABEL: 'broadcast_load_v16i8_v16i8'
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <16 x i32> zeroinitializer
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %bcst
;
; AVX512-LABEL: 'broadcast_load_v16i8_v16i8'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <16 x i32> zeroinitializer
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %bcst
;
  %load = load <16 x i8>, ptr %src
  %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <16 x i32> zeroinitializer
  ret <16 x i8> %bcst
}

define <32 x i8> @broadcast_load_v16i8_v32i8(ptr %src) {
; SSE2-LABEL: 'broadcast_load_v16i8_v32i8'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <32 x i32> zeroinitializer
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %bcst
;
; SSSE3-LABEL: 'broadcast_load_v16i8_v32i8'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <32 x i32> zeroinitializer
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %bcst
;
; SSE42-LABEL: 'broadcast_load_v16i8_v32i8'
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <32 x i32> zeroinitializer
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %bcst
;
; AVX1-LABEL: 'broadcast_load_v16i8_v32i8'
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <32 x i32> zeroinitializer
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %bcst
;
; AVX2-LABEL: 'broadcast_load_v16i8_v32i8'
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <32 x i32> zeroinitializer
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %bcst
;
; AVX512-LABEL: 'broadcast_load_v16i8_v32i8'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <32 x i32> zeroinitializer
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %bcst
;
  %load = load <16 x i8>, ptr %src
  %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <32 x i32> zeroinitializer
  ret <32 x i8> %bcst
}

define <64 x i8> @broadcast_load_v16i8_v64i8(ptr %src) {
; SSE2-LABEL: 'broadcast_load_v16i8_v64i8'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <64 x i32> zeroinitializer
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %bcst
;
; SSSE3-LABEL: 'broadcast_load_v16i8_v64i8'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <64 x i32> zeroinitializer
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %bcst
;
; SSE42-LABEL: 'broadcast_load_v16i8_v64i8'
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <64 x i32> zeroinitializer
; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %bcst
;
; AVX1-LABEL: 'broadcast_load_v16i8_v64i8'
; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <64 x i32> zeroinitializer
; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %bcst
;
; AVX2-LABEL: 'broadcast_load_v16i8_v64i8'
; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <64 x i32> zeroinitializer
; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %bcst
;
; AVX512-LABEL: 'broadcast_load_v16i8_v64i8'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %load = load <16 x i8>, ptr %src, align 16
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <64 x i32> zeroinitializer
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %bcst
;
  %load = load <16 x i8>, ptr %src
  %bcst = shufflevector <16 x i8> %load, <16 x i8> poison, <64 x i32> zeroinitializer
  ret <64 x i8> %bcst
}