; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse3 | FileCheck %s --check-prefixes=SSE,SSE3
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE,SSSE3
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE,SSE4
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE4
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx | FileCheck %s --check-prefixes=AVX
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW
;
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mcpu=slm | FileCheck %s --check-prefixes=SSE,SLM
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mcpu=goldmont | FileCheck %s --check-prefixes=SSE,GLM
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -mcpu=btver2 | FileCheck %s --check-prefixes=AVX

define i32 @insert_double(i32 %arg, double %val, <2 x double> %src128, <4 x double> %src256, <8 x double> %src512) {
; SSE-LABEL: 'insert_double'
; SSE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f64_a = insertelement <2 x double> %src128, double %val, i32 %arg
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f64_0 = insertelement <2 x double> %src128, double %val, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f64_1 = insertelement <2 x double> %src128, double %val, i32 1
; SSE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4f64_a = insertelement <4 x double> %src256, double %val, i32 %arg
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f64_0 = insertelement <4 x double> %src256, double %val, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f64_3 = insertelement <4 x double> %src256, double %val, i32 3
; SSE-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8f64_a = insertelement <8 x double> %src512, double %val, i32 %arg
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f64_0 = insertelement <8 x double> %src512, double %val, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f64_3 = insertelement <8 x double> %src512, double %val, i32 3
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f64_4 = insertelement <8 x double> %src512, double %val, i32 4
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f64_7 = insertelement <8 x double> %src512, double %val, i32 7
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX-LABEL: 'insert_double'
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f64_a = insertelement <2 x double> %src128, double %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f64_0 = insertelement <2 x double> %src128, double %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f64_1 = insertelement <2 x double> %src128, double %val, i32 1
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f64_a = insertelement <4 x double> %src256, double %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f64_0 = insertelement <4 x double> %src256, double %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f64_3 = insertelement <4 x double> %src256, double %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f64_a = insertelement <8 x double> %src512, double %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f64_0 = insertelement <8 x double> %src512, double %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f64_3 = insertelement <8 x double> %src512, double %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f64_4 = insertelement <8 x double> %src512, double %val, i32 4
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f64_7 = insertelement <8 x double> %src512, double %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512-LABEL: 'insert_double'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f64_a = insertelement <2 x double> %src128, double %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f64_0 = insertelement <2 x double> %src128, double %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f64_1 = insertelement <2 x double> %src128, double %val, i32 1
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f64_a = insertelement <4 x double> %src256, double %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f64_0 = insertelement <4 x double> %src256, double %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f64_3 = insertelement <4 x double> %src256, double %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f64_a = insertelement <8 x double> %src512, double %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f64_0 = insertelement <8 x double> %src512, double %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f64_3 = insertelement <8 x double> %src512, double %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f64_4 = insertelement <8 x double> %src512, double %val, i32 4
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f64_7 = insertelement <8 x double> %src512, double %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
  %v2f64_a = insertelement <2 x double> %src128, double %val, i32 %arg
  %v2f64_0 = insertelement <2 x double> %src128, double %val, i32 0
  %v2f64_1 = insertelement <2 x double> %src128, double %val, i32 1

  %v4f64_a = insertelement <4 x double> %src256, double %val, i32 %arg
  %v4f64_0 = insertelement <4 x double> %src256, double %val, i32 0
  %v4f64_3 = insertelement <4 x double> %src256, double %val, i32 3

  %v8f64_a = insertelement <8 x double> %src512, double %val, i32 %arg
  %v8f64_0 = insertelement <8 x double> %src512, double %val, i32 0
  %v8f64_3 = insertelement <8 x double> %src512, double %val, i32 3
  %v8f64_4 = insertelement <8 x double> %src512, double %val, i32 4
  %v8f64_7 = insertelement <8 x double> %src512, double %val, i32 7

  ret i32 undef
}

define i32 @insert_float(i32 %arg, float %val, <2 x float> %src64, <4 x float> %src128, <8 x float> %src256, <16 x float> %src512) {
; SSE2-LABEL: 'insert_float'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32_a = insertelement <2 x float> %src64, float %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f32_0 = insertelement <2 x float> %src64, float %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f32_1 = insertelement <2 x float> %src64, float %val, i32 1
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32_a = insertelement <4 x float> %src128, float %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f32_0 = insertelement <4 x float> %src128, float %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f32_3 = insertelement <4 x float> %src128, float %val, i32 3
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32_a = insertelement <8 x float> %src256, float %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_0 = insertelement <8 x float> %src256, float %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_3 = insertelement <8 x float> %src256, float %val, i32 3
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_4 = insertelement <8 x float> %src256, float %val, i32 4
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_7 = insertelement <8 x float> %src256, float %val, i32 7
; SSE2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32_a = insertelement <16 x float> %src512, float %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_0 = insertelement <16 x float> %src512, float %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_3 = insertelement <16 x float> %src512, float %val, i32 3
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_8 = insertelement <16 x float> %src512, float %val, i32 8
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_15 = insertelement <16 x float> %src512, float %val, i32 15
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE3-LABEL: 'insert_float'
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32_a = insertelement <2 x float> %src64, float %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f32_0 = insertelement <2 x float> %src64, float %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f32_1 = insertelement <2 x float> %src64, float %val, i32 1
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32_a = insertelement <4 x float> %src128, float %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f32_0 = insertelement <4 x float> %src128, float %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f32_3 = insertelement <4 x float> %src128, float %val, i32 3
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32_a = insertelement <8 x float> %src256, float %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_0 = insertelement <8 x float> %src256, float %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_3 = insertelement <8 x float> %src256, float %val, i32 3
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_4 = insertelement <8 x float> %src256, float %val, i32 4
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_7 = insertelement <8 x float> %src256, float %val, i32 7
; SSE3-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32_a = insertelement <16 x float> %src512, float %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_0 = insertelement <16 x float> %src512, float %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_3 = insertelement <16 x float> %src512, float %val, i32 3
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_8 = insertelement <16 x float> %src512, float %val, i32 8
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_15 = insertelement <16 x float> %src512, float %val, i32 15
; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSSE3-LABEL: 'insert_float'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32_a = insertelement <2 x float> %src64, float %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f32_0 = insertelement <2 x float> %src64, float %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2f32_1 = insertelement <2 x float> %src64, float %val, i32 1
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32_a = insertelement <4 x float> %src128, float %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f32_0 = insertelement <4 x float> %src128, float %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4f32_3 = insertelement <4 x float> %src128, float %val, i32 3
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32_a = insertelement <8 x float> %src256, float %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_0 = insertelement <8 x float> %src256, float %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_3 = insertelement <8 x float> %src256, float %val, i32 3
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_4 = insertelement <8 x float> %src256, float %val, i32 4
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8f32_7 = insertelement <8 x float> %src256, float %val, i32 7
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32_a = insertelement <16 x float> %src512, float %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_0 = insertelement <16 x float> %src512, float %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_3 = insertelement <16 x float> %src512, float %val, i32 3
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_8 = insertelement <16 x float> %src512, float %val, i32 8
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v16f32_15 = insertelement <16 x float> %src512, float %val, i32 15
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE4-LABEL: 'insert_float'
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32_a = insertelement <2 x float> %src64, float %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_0 = insertelement <2 x float> %src64, float %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_1 = insertelement <2 x float> %src64, float %val, i32 1
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32_a = insertelement <4 x float> %src128, float %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_0 = insertelement <4 x float> %src128, float %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_3 = insertelement <4 x float> %src128, float %val, i32 3
; SSE4-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32_a = insertelement <8 x float> %src256, float %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_0 = insertelement <8 x float> %src256, float %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_3 = insertelement <8 x float> %src256, float %val, i32 3
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_4 = insertelement <8 x float> %src256, float %val, i32 4
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_7 = insertelement <8 x float> %src256, float %val, i32 7
; SSE4-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32_a = insertelement <16 x float> %src512, float %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_0 = insertelement <16 x float> %src512, float %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_3 = insertelement <16 x float> %src512, float %val, i32 3
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_8 = insertelement <16 x float> %src512, float %val, i32 8
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_15 = insertelement <16 x float> %src512, float %val, i32 15
; SSE4-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX-LABEL: 'insert_float'
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32_a = insertelement <2 x float> %src64, float %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_0 = insertelement <2 x float> %src64, float %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_1 = insertelement <2 x float> %src64, float %val, i32 1
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32_a = insertelement <4 x float> %src128, float %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_0 = insertelement <4 x float> %src128, float %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_3 = insertelement <4 x float> %src128, float %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f32_a = insertelement <8 x float> %src256, float %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_0 = insertelement <8 x float> %src256, float %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_3 = insertelement <8 x float> %src256, float %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f32_4 = insertelement <8 x float> %src256, float %val, i32 4
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f32_7 = insertelement <8 x float> %src256, float %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16f32_a = insertelement <16 x float> %src512, float %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_0 = insertelement <16 x float> %src512, float %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_3 = insertelement <16 x float> %src512, float %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_8 = insertelement <16 x float> %src512, float %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16f32_15 = insertelement <16 x float> %src512, float %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512-LABEL: 'insert_float'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32_a = insertelement <2 x float> %src64, float %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_0 = insertelement <2 x float> %src64, float %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_1 = insertelement <2 x float> %src64, float %val, i32 1
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32_a = insertelement <4 x float> %src128, float %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_0 = insertelement <4 x float> %src128, float %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_3 = insertelement <4 x float> %src128, float %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f32_a = insertelement <8 x float> %src256, float %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_0 = insertelement <8 x float> %src256, float %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_3 = insertelement <8 x float> %src256, float %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f32_4 = insertelement <8 x float> %src256, float %val, i32 4
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8f32_7 = insertelement <8 x float> %src256, float %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16f32_a = insertelement <16 x float> %src512, float %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_0 = insertelement <16 x float> %src512, float %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_3 = insertelement <16 x float> %src512, float %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16f32_8 = insertelement <16 x float> %src512, float %val, i32 8
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16f32_15 = insertelement <16 x float> %src512, float %val, i32 15
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SLM-LABEL: 'insert_float'
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32_a = insertelement <2 x float> %src64, float %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_0 = insertelement <2 x float> %src64, float %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_1 = insertelement <2 x float> %src64, float %val, i32 1
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32_a = insertelement <4 x float> %src128, float %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_0 = insertelement <4 x float> %src128, float %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_3 = insertelement <4 x float> %src128, float %val, i32 3
; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32_a = insertelement <8 x float> %src256, float %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_0 = insertelement <8 x float> %src256, float %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_3 = insertelement <8 x float> %src256, float %val, i32 3
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_4 = insertelement <8 x float> %src256, float %val, i32 4
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_7 = insertelement <8 x float> %src256, float %val, i32 7
; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32_a = insertelement <16 x float> %src512, float %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_0 = insertelement <16 x float> %src512, float %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_3 = insertelement <16 x float> %src512, float %val, i32 3
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_8 = insertelement <16 x float> %src512, float %val, i32 8
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_15 = insertelement <16 x float> %src512, float %val, i32 15
; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; GLM-LABEL: 'insert_float'
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2f32_a = insertelement <2 x float> %src64, float %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_0 = insertelement <2 x float> %src64, float %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2f32_1 = insertelement <2 x float> %src64, float %val, i32 1
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4f32_a = insertelement <4 x float> %src128, float %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_0 = insertelement <4 x float> %src128, float %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4f32_3 = insertelement <4 x float> %src128, float %val, i32 3
; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8f32_a = insertelement <8 x float> %src256, float %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_0 = insertelement <8 x float> %src256, float %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_3 = insertelement <8 x float> %src256, float %val, i32 3
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_4 = insertelement <8 x float> %src256, float %val, i32 4
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8f32_7 = insertelement <8 x float> %src256, float %val, i32 7
; GLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16f32_a = insertelement <16 x float> %src512, float %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_0 = insertelement <16 x float> %src512, float %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_3 = insertelement <16 x float> %src512, float %val, i32 3
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_8 = insertelement <16 x float> %src512, float %val, i32 8
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16f32_15 = insertelement <16 x float> %src512, float %val, i32 15
; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
  %v2f32_a = insertelement <2 x float> %src64, float %val, i32 %arg
  %v2f32_0 = insertelement <2 x float> %src64, float %val, i32 0
  %v2f32_1 = insertelement <2 x float> %src64, float %val, i32 1

  %v4f32_a = insertelement <4 x float> %src128, float %val, i32 %arg
  %v4f32_0 = insertelement <4 x float> %src128, float %val, i32 0
  %v4f32_3 = insertelement <4 x float> %src128, float %val, i32 3

  %v8f32_a = insertelement <8 x float> %src256, float %val, i32 %arg
  %v8f32_0 = insertelement <8 x float> %src256, float %val, i32 0
  %v8f32_3 = insertelement <8 x float> %src256, float %val, i32 3
  %v8f32_4 = insertelement <8 x float> %src256, float %val, i32 4
  %v8f32_7 = insertelement <8 x float> %src256, float %val, i32 7

  %v16f32_a  = insertelement <16 x float> %src512, float %val, i32 %arg
  %v16f32_0  = insertelement <16 x float> %src512, float %val, i32 0
  %v16f32_3  = insertelement <16 x float> %src512, float %val, i32 3
  %v16f32_8  = insertelement <16 x float> %src512, float %val, i32 8
  %v16f32_15 = insertelement <16 x float> %src512, float %val, i32 15

  ret i32 undef
}

define i32 @insert_i64(i32 %arg, i64 %val, <2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512) {
; SSE2-LABEL: 'insert_i64'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i64_a = insertelement <2 x i64> %src128, i64 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i64_0 = insertelement <2 x i64> %src128, i64 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i64_1 = insertelement <2 x i64> %src128, i64 %val, i32 1
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i64_a = insertelement <4 x i64> %src256, i64 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i64_0 = insertelement <4 x i64> %src256, i64 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i64_3 = insertelement <4 x i64> %src256, i64 %val, i32 3
; SSE2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8i64_a = insertelement <8 x i64> %src512, i64 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_0 = insertelement <8 x i64> %src512, i64 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_3 = insertelement <8 x i64> %src512, i64 %val, i32 3
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_4 = insertelement <8 x i64> %src512, i64 %val, i32 4
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_7 = insertelement <8 x i64> %src512, i64 %val, i32 7
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE3-LABEL: 'insert_i64'
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i64_a = insertelement <2 x i64> %src128, i64 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i64_0 = insertelement <2 x i64> %src128, i64 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i64_1 = insertelement <2 x i64> %src128, i64 %val, i32 1
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i64_a = insertelement <4 x i64> %src256, i64 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i64_0 = insertelement <4 x i64> %src256, i64 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i64_3 = insertelement <4 x i64> %src256, i64 %val, i32 3
; SSE3-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8i64_a = insertelement <8 x i64> %src512, i64 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_0 = insertelement <8 x i64> %src512, i64 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_3 = insertelement <8 x i64> %src512, i64 %val, i32 3
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_4 = insertelement <8 x i64> %src512, i64 %val, i32 4
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_7 = insertelement <8 x i64> %src512, i64 %val, i32 7
; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSSE3-LABEL: 'insert_i64'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i64_a = insertelement <2 x i64> %src128, i64 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i64_0 = insertelement <2 x i64> %src128, i64 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i64_1 = insertelement <2 x i64> %src128, i64 %val, i32 1
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i64_a = insertelement <4 x i64> %src256, i64 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i64_0 = insertelement <4 x i64> %src256, i64 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i64_3 = insertelement <4 x i64> %src256, i64 %val, i32 3
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8i64_a = insertelement <8 x i64> %src512, i64 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_0 = insertelement <8 x i64> %src512, i64 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_3 = insertelement <8 x i64> %src512, i64 %val, i32 3
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_4 = insertelement <8 x i64> %src512, i64 %val, i32 4
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i64_7 = insertelement <8 x i64> %src512, i64 %val, i32 7
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE4-LABEL: 'insert_i64'
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i64_a = insertelement <2 x i64> %src128, i64 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_0 = insertelement <2 x i64> %src128, i64 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_1 = insertelement <2 x i64> %src128, i64 %val, i32 1
; SSE4-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i64_a = insertelement <4 x i64> %src256, i64 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i64_0 = insertelement <4 x i64> %src256, i64 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i64_3 = insertelement <4 x i64> %src256, i64 %val, i32 3
; SSE4-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8i64_a = insertelement <8 x i64> %src512, i64 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_0 = insertelement <8 x i64> %src512, i64 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_3 = insertelement <8 x i64> %src512, i64 %val, i32 3
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_4 = insertelement <8 x i64> %src512, i64 %val, i32 4
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_7 = insertelement <8 x i64> %src512, i64 %val, i32 7
; SSE4-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX-LABEL: 'insert_i64'
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i64_a = insertelement <2 x i64> %src128, i64 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_0 = insertelement <2 x i64> %src128, i64 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_1 = insertelement <2 x i64> %src128, i64 %val, i32 1
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i64_a = insertelement <4 x i64> %src256, i64 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i64_0 = insertelement <4 x i64> %src256, i64 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i64_3 = insertelement <4 x i64> %src256, i64 %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i64_a = insertelement <8 x i64> %src512, i64 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_0 = insertelement <8 x i64> %src512, i64 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i64_3 = insertelement <8 x i64> %src512, i64 %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_4 = insertelement <8 x i64> %src512, i64 %val, i32 4
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i64_7 = insertelement <8 x i64> %src512, i64 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512-LABEL: 'insert_i64'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i64_a = insertelement <2 x i64> %src128, i64 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_0 = insertelement <2 x i64> %src128, i64 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_1 = insertelement <2 x i64> %src128, i64 %val, i32 1
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i64_a = insertelement <4 x i64> %src256, i64 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i64_0 = insertelement <4 x i64> %src256, i64 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i64_3 = insertelement <4 x i64> %src256, i64 %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i64_a = insertelement <8 x i64> %src512, i64 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_0 = insertelement <8 x i64> %src512, i64 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i64_3 = insertelement <8 x i64> %src512, i64 %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i64_4 = insertelement <8 x i64> %src512, i64 %val, i32 4
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i64_7 = insertelement <8 x i64> %src512, i64 %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SLM-LABEL: 'insert_i64'
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i64_a = insertelement <2 x i64> %src128, i64 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_0 = insertelement <2 x i64> %src128, i64 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_1 = insertelement <2 x i64> %src128, i64 %val, i32 1
; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i64_a = insertelement <4 x i64> %src256, i64 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i64_0 = insertelement <4 x i64> %src256, i64 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i64_3 = insertelement <4 x i64> %src256, i64 %val, i32 3
; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8i64_a = insertelement <8 x i64> %src512, i64 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_0 = insertelement <8 x i64> %src512, i64 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_3 = insertelement <8 x i64> %src512, i64 %val, i32 3
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_4 = insertelement <8 x i64> %src512, i64 %val, i32 4
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_7 = insertelement <8 x i64> %src512, i64 %val, i32 7
; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; GLM-LABEL: 'insert_i64'
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i64_a = insertelement <2 x i64> %src128, i64 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_0 = insertelement <2 x i64> %src128, i64 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i64_1 = insertelement <2 x i64> %src128, i64 %val, i32 1
; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i64_a = insertelement <4 x i64> %src256, i64 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i64_0 = insertelement <4 x i64> %src256, i64 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i64_3 = insertelement <4 x i64> %src256, i64 %val, i32 3
; GLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8i64_a = insertelement <8 x i64> %src512, i64 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_0 = insertelement <8 x i64> %src512, i64 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_3 = insertelement <8 x i64> %src512, i64 %val, i32 3
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_4 = insertelement <8 x i64> %src512, i64 %val, i32 4
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i64_7 = insertelement <8 x i64> %src512, i64 %val, i32 7
; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
  %v2i64_a = insertelement <2 x i64> %src128, i64 %val, i32 %arg
  %v2i64_0 = insertelement <2 x i64> %src128, i64 %val, i32 0
  %v2i64_1 = insertelement <2 x i64> %src128, i64 %val, i32 1

  %v4i64_a = insertelement <4 x i64> %src256, i64 %val, i32 %arg
  %v4i64_0 = insertelement <4 x i64> %src256, i64 %val, i32 0
  %v4i64_3 = insertelement <4 x i64> %src256, i64 %val, i32 3

  %v8i64_a = insertelement <8 x i64> %src512, i64 %val, i32 %arg
  %v8i64_0 = insertelement <8 x i64> %src512, i64 %val, i32 0
  %v8i64_3 = insertelement <8 x i64> %src512, i64 %val, i32 3
  %v8i64_4 = insertelement <8 x i64> %src512, i64 %val, i32 4
  %v8i64_7 = insertelement <8 x i64> %src512, i64 %val, i32 7

  ret i32 undef
}

define i32 @insert_i32(i32 %arg, i32 %val, <2 x i32> %src64, <4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512) {
; SSE2-LABEL: 'insert_i32'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_a = insertelement <2 x i32> %src64, i32 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_0 = insertelement <2 x i32> %src64, i32 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_1 = insertelement <2 x i32> %src64, i32 %val, i32 1
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_a = insertelement <4 x i32> %src128, i32 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_0 = insertelement <4 x i32> %src128, i32 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_3 = insertelement <4 x i32> %src128, i32 %val, i32 3
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i32_a = insertelement <8 x i32> %src256, i32 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_0 = insertelement <8 x i32> %src256, i32 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_3 = insertelement <8 x i32> %src256, i32 %val, i32 3
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_4 = insertelement <8 x i32> %src256, i32 %val, i32 4
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_7 = insertelement <8 x i32> %src256, i32 %val, i32 7
; SSE2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16i32_a = insertelement <16 x i32> %src512, i32 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_0 = insertelement <16 x i32> %src512, i32 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_3 = insertelement <16 x i32> %src512, i32 %val, i32 3
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_8 = insertelement <16 x i32> %src512, i32 %val, i32 8
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_15 = insertelement <16 x i32> %src512, i32 %val, i32 15
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE3-LABEL: 'insert_i32'
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_a = insertelement <2 x i32> %src64, i32 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_0 = insertelement <2 x i32> %src64, i32 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_1 = insertelement <2 x i32> %src64, i32 %val, i32 1
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_a = insertelement <4 x i32> %src128, i32 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_0 = insertelement <4 x i32> %src128, i32 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_3 = insertelement <4 x i32> %src128, i32 %val, i32 3
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i32_a = insertelement <8 x i32> %src256, i32 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_0 = insertelement <8 x i32> %src256, i32 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_3 = insertelement <8 x i32> %src256, i32 %val, i32 3
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_4 = insertelement <8 x i32> %src256, i32 %val, i32 4
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_7 = insertelement <8 x i32> %src256, i32 %val, i32 7
; SSE3-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16i32_a = insertelement <16 x i32> %src512, i32 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_0 = insertelement <16 x i32> %src512, i32 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_3 = insertelement <16 x i32> %src512, i32 %val, i32 3
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_8 = insertelement <16 x i32> %src512, i32 %val, i32 8
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_15 = insertelement <16 x i32> %src512, i32 %val, i32 15
; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSSE3-LABEL: 'insert_i32'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_a = insertelement <2 x i32> %src64, i32 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_0 = insertelement <2 x i32> %src64, i32 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_1 = insertelement <2 x i32> %src64, i32 %val, i32 1
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_a = insertelement <4 x i32> %src128, i32 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_0 = insertelement <4 x i32> %src128, i32 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_3 = insertelement <4 x i32> %src128, i32 %val, i32 3
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i32_a = insertelement <8 x i32> %src256, i32 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_0 = insertelement <8 x i32> %src256, i32 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_3 = insertelement <8 x i32> %src256, i32 %val, i32 3
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_4 = insertelement <8 x i32> %src256, i32 %val, i32 4
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_7 = insertelement <8 x i32> %src256, i32 %val, i32 7
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16i32_a = insertelement <16 x i32> %src512, i32 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_0 = insertelement <16 x i32> %src512, i32 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_3 = insertelement <16 x i32> %src512, i32 %val, i32 3
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_8 = insertelement <16 x i32> %src512, i32 %val, i32 8
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_15 = insertelement <16 x i32> %src512, i32 %val, i32 15
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE4-LABEL: 'insert_i32'
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_a = insertelement <2 x i32> %src64, i32 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_0 = insertelement <2 x i32> %src64, i32 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_1 = insertelement <2 x i32> %src64, i32 %val, i32 1
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_a = insertelement <4 x i32> %src128, i32 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_0 = insertelement <4 x i32> %src128, i32 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_3 = insertelement <4 x i32> %src128, i32 %val, i32 3
; SSE4-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i32_a = insertelement <8 x i32> %src256, i32 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_0 = insertelement <8 x i32> %src256, i32 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_3 = insertelement <8 x i32> %src256, i32 %val, i32 3
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_4 = insertelement <8 x i32> %src256, i32 %val, i32 4
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_7 = insertelement <8 x i32> %src256, i32 %val, i32 7
; SSE4-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16i32_a = insertelement <16 x i32> %src512, i32 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_0 = insertelement <16 x i32> %src512, i32 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_3 = insertelement <16 x i32> %src512, i32 %val, i32 3
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_8 = insertelement <16 x i32> %src512, i32 %val, i32 8
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_15 = insertelement <16 x i32> %src512, i32 %val, i32 15
; SSE4-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX-LABEL: 'insert_i32'
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_a = insertelement <2 x i32> %src64, i32 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_0 = insertelement <2 x i32> %src64, i32 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_1 = insertelement <2 x i32> %src64, i32 %val, i32 1
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_a = insertelement <4 x i32> %src128, i32 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_0 = insertelement <4 x i32> %src128, i32 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_3 = insertelement <4 x i32> %src128, i32 %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_a = insertelement <8 x i32> %src256, i32 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_0 = insertelement <8 x i32> %src256, i32 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_3 = insertelement <8 x i32> %src256, i32 %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_4 = insertelement <8 x i32> %src256, i32 %val, i32 4
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_7 = insertelement <8 x i32> %src256, i32 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i32_a = insertelement <16 x i32> %src512, i32 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_0 = insertelement <16 x i32> %src512, i32 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_3 = insertelement <16 x i32> %src512, i32 %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_8 = insertelement <16 x i32> %src512, i32 %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_15 = insertelement <16 x i32> %src512, i32 %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512-LABEL: 'insert_i32'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_a = insertelement <2 x i32> %src64, i32 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_0 = insertelement <2 x i32> %src64, i32 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_1 = insertelement <2 x i32> %src64, i32 %val, i32 1
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_a = insertelement <4 x i32> %src128, i32 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_0 = insertelement <4 x i32> %src128, i32 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_3 = insertelement <4 x i32> %src128, i32 %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_a = insertelement <8 x i32> %src256, i32 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_0 = insertelement <8 x i32> %src256, i32 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_3 = insertelement <8 x i32> %src256, i32 %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_4 = insertelement <8 x i32> %src256, i32 %val, i32 4
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i32_7 = insertelement <8 x i32> %src256, i32 %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_a = insertelement <16 x i32> %src512, i32 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_0 = insertelement <16 x i32> %src512, i32 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_3 = insertelement <16 x i32> %src512, i32 %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_8 = insertelement <16 x i32> %src512, i32 %val, i32 8
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i32_15 = insertelement <16 x i32> %src512, i32 %val, i32 15
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SLM-LABEL: 'insert_i32'
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_a = insertelement <2 x i32> %src64, i32 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_0 = insertelement <2 x i32> %src64, i32 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_1 = insertelement <2 x i32> %src64, i32 %val, i32 1
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_a = insertelement <4 x i32> %src128, i32 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_0 = insertelement <4 x i32> %src128, i32 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_3 = insertelement <4 x i32> %src128, i32 %val, i32 3
; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i32_a = insertelement <8 x i32> %src256, i32 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_0 = insertelement <8 x i32> %src256, i32 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_3 = insertelement <8 x i32> %src256, i32 %val, i32 3
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_4 = insertelement <8 x i32> %src256, i32 %val, i32 4
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_7 = insertelement <8 x i32> %src256, i32 %val, i32 7
; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16i32_a = insertelement <16 x i32> %src512, i32 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_0 = insertelement <16 x i32> %src512, i32 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_3 = insertelement <16 x i32> %src512, i32 %val, i32 3
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_8 = insertelement <16 x i32> %src512, i32 %val, i32 8
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_15 = insertelement <16 x i32> %src512, i32 %val, i32 15
; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; GLM-LABEL: 'insert_i32'
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i32_a = insertelement <2 x i32> %src64, i32 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_0 = insertelement <2 x i32> %src64, i32 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i32_1 = insertelement <2 x i32> %src64, i32 %val, i32 1
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i32_a = insertelement <4 x i32> %src128, i32 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_0 = insertelement <4 x i32> %src128, i32 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i32_3 = insertelement <4 x i32> %src128, i32 %val, i32 3
; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i32_a = insertelement <8 x i32> %src256, i32 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_0 = insertelement <8 x i32> %src256, i32 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_3 = insertelement <8 x i32> %src256, i32 %val, i32 3
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_4 = insertelement <8 x i32> %src256, i32 %val, i32 4
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_7 = insertelement <8 x i32> %src256, i32 %val, i32 7
; GLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v16i32_a = insertelement <16 x i32> %src512, i32 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_0 = insertelement <16 x i32> %src512, i32 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_3 = insertelement <16 x i32> %src512, i32 %val, i32 3
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_8 = insertelement <16 x i32> %src512, i32 %val, i32 8
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_15 = insertelement <16 x i32> %src512, i32 %val, i32 15
; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
  %v2i32_a = insertelement <2 x i32> %src64, i32 %val, i32 %arg
  %v2i32_0 = insertelement <2 x i32> %src64, i32 %val, i32 0
  %v2i32_1 = insertelement <2 x i32> %src64, i32 %val, i32 1

  %v4i32_a = insertelement <4 x i32> %src128, i32 %val, i32 %arg
  %v4i32_0 = insertelement <4 x i32> %src128, i32 %val, i32 0
  %v4i32_3 = insertelement <4 x i32> %src128, i32 %val, i32 3

  %v8i32_a = insertelement <8 x i32> %src256, i32 %val, i32 %arg
  %v8i32_0 = insertelement <8 x i32> %src256, i32 %val, i32 0
  %v8i32_3 = insertelement <8 x i32> %src256, i32 %val, i32 3
  %v8i32_4 = insertelement <8 x i32> %src256, i32 %val, i32 4
  %v8i32_7 = insertelement <8 x i32> %src256, i32 %val, i32 7

  %v16i32_a  = insertelement <16 x i32> %src512, i32 %val, i32 %arg
  %v16i32_0  = insertelement <16 x i32> %src512, i32 %val, i32 0
  %v16i32_3  = insertelement <16 x i32> %src512, i32 %val, i32 3
  %v16i32_8  = insertelement <16 x i32> %src512, i32 %val, i32 8
  %v16i32_15 = insertelement <16 x i32> %src512, i32 %val, i32 15

  ret i32 undef
}

define i32 @insert_i16(i32 %arg, i16 %val, <2 x i16> %src32, <4 x i16> %src64, <8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src512) {
; SSE-LABEL: 'insert_i16'
; SSE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i16_a = insertelement <2 x i16> %src32, i16 %val, i32 %arg
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i16_0 = insertelement <2 x i16> %src32, i16 %val, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i16_1 = insertelement <2 x i16> %src32, i16 %val, i32 1
; SSE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i16_a = insertelement <4 x i16> %src64, i16 %val, i32 %arg
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i16_0 = insertelement <4 x i16> %src64, i16 %val, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i16_3 = insertelement <4 x i16> %src64, i16 %val, i32 3
; SSE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i16_a = insertelement <8 x i16> %src128, i16 %val, i32 %arg
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_0 = insertelement <8 x i16> %src128, i16 %val, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_7 = insertelement <8 x i16> %src128, i16 %val, i32 7
; SSE-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i16_a = insertelement <16 x i16> %src256, i16 %val, i32 %arg
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_0 = insertelement <16 x i16> %src256, i16 %val, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_7 = insertelement <16 x i16> %src256, i16 %val, i32 7
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_8 = insertelement <16 x i16> %src256, i16 %val, i32 8
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_15 = insertelement <16 x i16> %src256, i16 %val, i32 15
; SSE-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v32i16_a = insertelement <32 x i16> %src512, i16 %val, i32 %arg
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_0 = insertelement <32 x i16> %src512, i16 %val, i32 0
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_7 = insertelement <32 x i16> %src512, i16 %val, i32 7
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_8 = insertelement <32 x i16> %src512, i16 %val, i32 8
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_15 = insertelement <32 x i16> %src512, i16 %val, i32 15
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_16 = insertelement <32 x i16> %src512, i16 %val, i32 16
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_24 = insertelement <32 x i16> %src512, i16 %val, i32 24
; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_31 = insertelement <32 x i16> %src512, i16 %val, i32 31
; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX-LABEL: 'insert_i16'
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i16_a = insertelement <2 x i16> %src32, i16 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i16_0 = insertelement <2 x i16> %src32, i16 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i16_1 = insertelement <2 x i16> %src32, i16 %val, i32 1
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i16_a = insertelement <4 x i16> %src64, i16 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i16_0 = insertelement <4 x i16> %src64, i16 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i16_3 = insertelement <4 x i16> %src64, i16 %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i16_a = insertelement <8 x i16> %src128, i16 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_0 = insertelement <8 x i16> %src128, i16 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_7 = insertelement <8 x i16> %src128, i16 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i16_a = insertelement <16 x i16> %src256, i16 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_0 = insertelement <16 x i16> %src256, i16 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_7 = insertelement <16 x i16> %src256, i16 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i16_8 = insertelement <16 x i16> %src256, i16 %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i16_15 = insertelement <16 x i16> %src256, i16 %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32i16_a = insertelement <32 x i16> %src512, i16 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_0 = insertelement <32 x i16> %src512, i16 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_7 = insertelement <32 x i16> %src512, i16 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_8 = insertelement <32 x i16> %src512, i16 %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_15 = insertelement <32 x i16> %src512, i16 %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_16 = insertelement <32 x i16> %src512, i16 %val, i32 16
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_24 = insertelement <32 x i16> %src512, i16 %val, i32 24
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_31 = insertelement <32 x i16> %src512, i16 %val, i32 31
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512-LABEL: 'insert_i16'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i16_a = insertelement <2 x i16> %src32, i16 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i16_0 = insertelement <2 x i16> %src32, i16 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i16_1 = insertelement <2 x i16> %src32, i16 %val, i32 1
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i16_a = insertelement <4 x i16> %src64, i16 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i16_0 = insertelement <4 x i16> %src64, i16 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i16_3 = insertelement <4 x i16> %src64, i16 %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i16_a = insertelement <8 x i16> %src128, i16 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_0 = insertelement <8 x i16> %src128, i16 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_7 = insertelement <8 x i16> %src128, i16 %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i16_a = insertelement <16 x i16> %src256, i16 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_0 = insertelement <16 x i16> %src256, i16 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_7 = insertelement <16 x i16> %src256, i16 %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i16_8 = insertelement <16 x i16> %src256, i16 %val, i32 8
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i16_15 = insertelement <16 x i16> %src256, i16 %val, i32 15
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_a = insertelement <32 x i16> %src512, i16 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_0 = insertelement <32 x i16> %src512, i16 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i16_7 = insertelement <32 x i16> %src512, i16 %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_8 = insertelement <32 x i16> %src512, i16 %val, i32 8
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_15 = insertelement <32 x i16> %src512, i16 %val, i32 15
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_16 = insertelement <32 x i16> %src512, i16 %val, i32 16
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_24 = insertelement <32 x i16> %src512, i16 %val, i32 24
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i16_31 = insertelement <32 x i16> %src512, i16 %val, i32 31
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
  %v2i16_a = insertelement <2 x i16> %src32, i16 %val, i32 %arg
  %v2i16_0 = insertelement <2 x i16> %src32, i16 %val, i32 0
  %v2i16_1 = insertelement <2 x i16> %src32, i16 %val, i32 1

  %v4i16_a = insertelement <4 x i16> %src64, i16 %val, i32 %arg
  %v4i16_0 = insertelement <4 x i16> %src64, i16 %val, i32 0
  %v4i16_3 = insertelement <4 x i16> %src64, i16 %val, i32 3

  %v8i16_a = insertelement <8 x i16> %src128, i16 %val, i32 %arg
  %v8i16_0 = insertelement <8 x i16> %src128, i16 %val, i32 0
  %v8i16_7 = insertelement <8 x i16> %src128, i16 %val, i32 7

  %v16i16_a  = insertelement <16 x i16> %src256, i16 %val, i32 %arg
  %v16i16_0  = insertelement <16 x i16> %src256, i16 %val, i32 0
  %v16i16_7  = insertelement <16 x i16> %src256, i16 %val, i32 7
  %v16i16_8  = insertelement <16 x i16> %src256, i16 %val, i32 8
  %v16i16_15 = insertelement <16 x i16> %src256, i16 %val, i32 15

  %v32i16_a  = insertelement <32 x i16> %src512, i16 %val, i32 %arg
  %v32i16_0  = insertelement <32 x i16> %src512, i16 %val, i32 0
  %v32i16_7  = insertelement <32 x i16> %src512, i16 %val, i32 7
  %v32i16_8  = insertelement <32 x i16> %src512, i16 %val, i32 8
  %v32i16_15 = insertelement <32 x i16> %src512, i16 %val, i32 15
  %v32i16_16 = insertelement <32 x i16> %src512, i16 %val, i32 16
  %v32i16_24 = insertelement <32 x i16> %src512, i16 %val, i32 24
  %v32i16_31 = insertelement <32 x i16> %src512, i16 %val, i32 31

  ret i32 undef
}

define i32 @insert_i8(i32 %arg, i8 %val, <2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512) {
; SSE2-LABEL: 'insert_i8'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i8_0 = insertelement <4 x i8> %src32, i8 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i8_3 = insertelement <4 x i8> %src32, i8 %val, i32 3
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i8_a = insertelement <8 x i8> %src64, i8 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i8_0 = insertelement <8 x i8> %src64, i8 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i8_7 = insertelement <8 x i8> %src64, i8 %val, i32 7
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i8_a = insertelement <16 x i8> %src128, i8 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i8_0 = insertelement <16 x i8> %src128, i8 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i8_8 = insertelement <16 x i8> %src128, i8 %val, i32 8
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i8_15 = insertelement <16 x i8> %src128, i8 %val, i32 15
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32i8_a = insertelement <32 x i8> %src256, i8 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_0 = insertelement <32 x i8> %src256, i8 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_7 = insertelement <32 x i8> %src256, i8 %val, i32 7
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_8 = insertelement <32 x i8> %src256, i8 %val, i32 8
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_15 = insertelement <32 x i8> %src256, i8 %val, i32 15
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_24 = insertelement <32 x i8> %src256, i8 %val, i32 24
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_31 = insertelement <32 x i8> %src256, i8 %val, i32 31
; SSE2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v64i8_a = insertelement <64 x i8> %src512, i8 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_0 = insertelement <64 x i8> %src512, i8 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_7 = insertelement <64 x i8> %src512, i8 %val, i32 7
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_8 = insertelement <64 x i8> %src512, i8 %val, i32 8
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_15 = insertelement <64 x i8> %src512, i8 %val, i32 15
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_24 = insertelement <64 x i8> %src512, i8 %val, i32 24
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_31 = insertelement <64 x i8> %src512, i8 %val, i32 31
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_32 = insertelement <64 x i8> %src512, i8 %val, i32 32
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_48 = insertelement <64 x i8> %src512, i8 %val, i32 48
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_63 = insertelement <64 x i8> %src512, i8 %val, i32 63
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE3-LABEL: 'insert_i8'
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i8_0 = insertelement <4 x i8> %src32, i8 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i8_3 = insertelement <4 x i8> %src32, i8 %val, i32 3
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i8_a = insertelement <8 x i8> %src64, i8 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i8_0 = insertelement <8 x i8> %src64, i8 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i8_7 = insertelement <8 x i8> %src64, i8 %val, i32 7
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i8_a = insertelement <16 x i8> %src128, i8 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i8_0 = insertelement <16 x i8> %src128, i8 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i8_8 = insertelement <16 x i8> %src128, i8 %val, i32 8
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i8_15 = insertelement <16 x i8> %src128, i8 %val, i32 15
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32i8_a = insertelement <32 x i8> %src256, i8 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_0 = insertelement <32 x i8> %src256, i8 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_7 = insertelement <32 x i8> %src256, i8 %val, i32 7
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_8 = insertelement <32 x i8> %src256, i8 %val, i32 8
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_15 = insertelement <32 x i8> %src256, i8 %val, i32 15
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_24 = insertelement <32 x i8> %src256, i8 %val, i32 24
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i8_31 = insertelement <32 x i8> %src256, i8 %val, i32 31
; SSE3-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v64i8_a = insertelement <64 x i8> %src512, i8 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_0 = insertelement <64 x i8> %src512, i8 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_7 = insertelement <64 x i8> %src512, i8 %val, i32 7
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_8 = insertelement <64 x i8> %src512, i8 %val, i32 8
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_15 = insertelement <64 x i8> %src512, i8 %val, i32 15
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_24 = insertelement <64 x i8> %src512, i8 %val, i32 24
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_31 = insertelement <64 x i8> %src512, i8 %val, i32 31
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_32 = insertelement <64 x i8> %src512, i8 %val, i32 32
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_48 = insertelement <64 x i8> %src512, i8 %val, i32 48
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i8_63 = insertelement <64 x i8> %src512, i8 %val, i32 63
; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSSE3-LABEL: 'insert_i8'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_0 = insertelement <4 x i8> %src32, i8 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_3 = insertelement <4 x i8> %src32, i8 %val, i32 3
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i8_a = insertelement <8 x i8> %src64, i8 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_0 = insertelement <8 x i8> %src64, i8 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_7 = insertelement <8 x i8> %src64, i8 %val, i32 7
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i8_a = insertelement <16 x i8> %src128, i8 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16i8_0 = insertelement <16 x i8> %src128, i8 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16i8_8 = insertelement <16 x i8> %src128, i8 %val, i32 8
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16i8_15 = insertelement <16 x i8> %src128, i8 %val, i32 15
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32i8_a = insertelement <32 x i8> %src256, i8 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i8_0 = insertelement <32 x i8> %src256, i8 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i8_7 = insertelement <32 x i8> %src256, i8 %val, i32 7
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i8_8 = insertelement <32 x i8> %src256, i8 %val, i32 8
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i8_15 = insertelement <32 x i8> %src256, i8 %val, i32 15
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i8_24 = insertelement <32 x i8> %src256, i8 %val, i32 24
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i8_31 = insertelement <32 x i8> %src256, i8 %val, i32 31
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v64i8_a = insertelement <64 x i8> %src512, i8 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i8_0 = insertelement <64 x i8> %src512, i8 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i8_7 = insertelement <64 x i8> %src512, i8 %val, i32 7
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i8_8 = insertelement <64 x i8> %src512, i8 %val, i32 8
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i8_15 = insertelement <64 x i8> %src512, i8 %val, i32 15
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i8_24 = insertelement <64 x i8> %src512, i8 %val, i32 24
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i8_31 = insertelement <64 x i8> %src512, i8 %val, i32 31
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i8_32 = insertelement <64 x i8> %src512, i8 %val, i32 32
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i8_48 = insertelement <64 x i8> %src512, i8 %val, i32 48
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i8_63 = insertelement <64 x i8> %src512, i8 %val, i32 63
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE4-LABEL: 'insert_i8'
; SSE4-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_0 = insertelement <4 x i8> %src32, i8 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_3 = insertelement <4 x i8> %src32, i8 %val, i32 3
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i8_a = insertelement <8 x i8> %src64, i8 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_0 = insertelement <8 x i8> %src64, i8 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_7 = insertelement <8 x i8> %src64, i8 %val, i32 7
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i8_a = insertelement <16 x i8> %src128, i8 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_0 = insertelement <16 x i8> %src128, i8 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_8 = insertelement <16 x i8> %src128, i8 %val, i32 8
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_15 = insertelement <16 x i8> %src128, i8 %val, i32 15
; SSE4-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32i8_a = insertelement <32 x i8> %src256, i8 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_0 = insertelement <32 x i8> %src256, i8 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_7 = insertelement <32 x i8> %src256, i8 %val, i32 7
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_8 = insertelement <32 x i8> %src256, i8 %val, i32 8
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_15 = insertelement <32 x i8> %src256, i8 %val, i32 15
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_24 = insertelement <32 x i8> %src256, i8 %val, i32 24
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_31 = insertelement <32 x i8> %src256, i8 %val, i32 31
; SSE4-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v64i8_a = insertelement <64 x i8> %src512, i8 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_0 = insertelement <64 x i8> %src512, i8 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_7 = insertelement <64 x i8> %src512, i8 %val, i32 7
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_8 = insertelement <64 x i8> %src512, i8 %val, i32 8
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_15 = insertelement <64 x i8> %src512, i8 %val, i32 15
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_24 = insertelement <64 x i8> %src512, i8 %val, i32 24
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_31 = insertelement <64 x i8> %src512, i8 %val, i32 31
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_32 = insertelement <64 x i8> %src512, i8 %val, i32 32
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_48 = insertelement <64 x i8> %src512, i8 %val, i32 48
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_63 = insertelement <64 x i8> %src512, i8 %val, i32 63
; SSE4-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX-LABEL: 'insert_i8'
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_0 = insertelement <4 x i8> %src32, i8 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_3 = insertelement <4 x i8> %src32, i8 %val, i32 3
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i8_a = insertelement <8 x i8> %src64, i8 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_0 = insertelement <8 x i8> %src64, i8 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_7 = insertelement <8 x i8> %src64, i8 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i8_a = insertelement <16 x i8> %src128, i8 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_0 = insertelement <16 x i8> %src128, i8 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_8 = insertelement <16 x i8> %src128, i8 %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_15 = insertelement <16 x i8> %src128, i8 %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i8_a = insertelement <32 x i8> %src256, i8 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_0 = insertelement <32 x i8> %src256, i8 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_7 = insertelement <32 x i8> %src256, i8 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_8 = insertelement <32 x i8> %src256, i8 %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_15 = insertelement <32 x i8> %src256, i8 %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i8_24 = insertelement <32 x i8> %src256, i8 %val, i32 24
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i8_31 = insertelement <32 x i8> %src256, i8 %val, i32 31
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v64i8_a = insertelement <64 x i8> %src512, i8 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_0 = insertelement <64 x i8> %src512, i8 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_7 = insertelement <64 x i8> %src512, i8 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_8 = insertelement <64 x i8> %src512, i8 %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_15 = insertelement <64 x i8> %src512, i8 %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_24 = insertelement <64 x i8> %src512, i8 %val, i32 24
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_31 = insertelement <64 x i8> %src512, i8 %val, i32 31
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_32 = insertelement <64 x i8> %src512, i8 %val, i32 32
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_48 = insertelement <64 x i8> %src512, i8 %val, i32 48
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_63 = insertelement <64 x i8> %src512, i8 %val, i32 63
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512-LABEL: 'insert_i8'
; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_0 = insertelement <4 x i8> %src32, i8 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_3 = insertelement <4 x i8> %src32, i8 %val, i32 3
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i8_a = insertelement <8 x i8> %src64, i8 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_0 = insertelement <8 x i8> %src64, i8 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_7 = insertelement <8 x i8> %src64, i8 %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i8_a = insertelement <16 x i8> %src128, i8 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_0 = insertelement <16 x i8> %src128, i8 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_8 = insertelement <16 x i8> %src128, i8 %val, i32 8
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_15 = insertelement <16 x i8> %src128, i8 %val, i32 15
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i8_a = insertelement <32 x i8> %src256, i8 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_0 = insertelement <32 x i8> %src256, i8 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_7 = insertelement <32 x i8> %src256, i8 %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_8 = insertelement <32 x i8> %src256, i8 %val, i32 8
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_15 = insertelement <32 x i8> %src256, i8 %val, i32 15
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i8_24 = insertelement <32 x i8> %src256, i8 %val, i32 24
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i8_31 = insertelement <32 x i8> %src256, i8 %val, i32 31
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_a = insertelement <64 x i8> %src512, i8 %val, i32 %arg
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_0 = insertelement <64 x i8> %src512, i8 %val, i32 0
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_7 = insertelement <64 x i8> %src512, i8 %val, i32 7
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_8 = insertelement <64 x i8> %src512, i8 %val, i32 8
; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_15 = insertelement <64 x i8> %src512, i8 %val, i32 15
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_24 = insertelement <64 x i8> %src512, i8 %val, i32 24
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_31 = insertelement <64 x i8> %src512, i8 %val, i32 31
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_32 = insertelement <64 x i8> %src512, i8 %val, i32 32
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_48 = insertelement <64 x i8> %src512, i8 %val, i32 48
; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i8_63 = insertelement <64 x i8> %src512, i8 %val, i32 63
; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SLM-LABEL: 'insert_i8'
; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_0 = insertelement <4 x i8> %src32, i8 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_3 = insertelement <4 x i8> %src32, i8 %val, i32 3
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i8_a = insertelement <8 x i8> %src64, i8 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_0 = insertelement <8 x i8> %src64, i8 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_7 = insertelement <8 x i8> %src64, i8 %val, i32 7
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i8_a = insertelement <16 x i8> %src128, i8 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_0 = insertelement <16 x i8> %src128, i8 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_8 = insertelement <16 x i8> %src128, i8 %val, i32 8
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_15 = insertelement <16 x i8> %src128, i8 %val, i32 15
; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32i8_a = insertelement <32 x i8> %src256, i8 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_0 = insertelement <32 x i8> %src256, i8 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_7 = insertelement <32 x i8> %src256, i8 %val, i32 7
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_8 = insertelement <32 x i8> %src256, i8 %val, i32 8
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_15 = insertelement <32 x i8> %src256, i8 %val, i32 15
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_24 = insertelement <32 x i8> %src256, i8 %val, i32 24
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_31 = insertelement <32 x i8> %src256, i8 %val, i32 31
; SLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v64i8_a = insertelement <64 x i8> %src512, i8 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_0 = insertelement <64 x i8> %src512, i8 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_7 = insertelement <64 x i8> %src512, i8 %val, i32 7
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_8 = insertelement <64 x i8> %src512, i8 %val, i32 8
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_15 = insertelement <64 x i8> %src512, i8 %val, i32 15
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_24 = insertelement <64 x i8> %src512, i8 %val, i32 24
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_31 = insertelement <64 x i8> %src512, i8 %val, i32 31
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_32 = insertelement <64 x i8> %src512, i8 %val, i32 32
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_48 = insertelement <64 x i8> %src512, i8 %val, i32 48
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_63 = insertelement <64 x i8> %src512, i8 %val, i32 63
; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; GLM-LABEL: 'insert_i8'
; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i8_a = insertelement <2 x i8> %src16, i8 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_0 = insertelement <2 x i8> %src16, i8 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i8_3 = insertelement <2 x i8> %src16, i8 %val, i32 1
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i8_a = insertelement <4 x i8> %src32, i8 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_0 = insertelement <4 x i8> %src32, i8 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i8_3 = insertelement <4 x i8> %src32, i8 %val, i32 3
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v8i8_a = insertelement <8 x i8> %src64, i8 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_0 = insertelement <8 x i8> %src64, i8 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i8_7 = insertelement <8 x i8> %src64, i8 %val, i32 7
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v16i8_a = insertelement <16 x i8> %src128, i8 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_0 = insertelement <16 x i8> %src128, i8 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_8 = insertelement <16 x i8> %src128, i8 %val, i32 8
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i8_15 = insertelement <16 x i8> %src128, i8 %val, i32 15
; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v32i8_a = insertelement <32 x i8> %src256, i8 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_0 = insertelement <32 x i8> %src256, i8 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_7 = insertelement <32 x i8> %src256, i8 %val, i32 7
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_8 = insertelement <32 x i8> %src256, i8 %val, i32 8
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_15 = insertelement <32 x i8> %src256, i8 %val, i32 15
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_24 = insertelement <32 x i8> %src256, i8 %val, i32 24
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i8_31 = insertelement <32 x i8> %src256, i8 %val, i32 31
; GLM-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v64i8_a = insertelement <64 x i8> %src512, i8 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_0 = insertelement <64 x i8> %src512, i8 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_7 = insertelement <64 x i8> %src512, i8 %val, i32 7
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_8 = insertelement <64 x i8> %src512, i8 %val, i32 8
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_15 = insertelement <64 x i8> %src512, i8 %val, i32 15
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_24 = insertelement <64 x i8> %src512, i8 %val, i32 24
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_31 = insertelement <64 x i8> %src512, i8 %val, i32 31
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_32 = insertelement <64 x i8> %src512, i8 %val, i32 32
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_48 = insertelement <64 x i8> %src512, i8 %val, i32 48
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i8_63 = insertelement <64 x i8> %src512, i8 %val, i32 63
; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
  %v2i8_a   = insertelement <2 x i8> %src16, i8 %val, i32 %arg
  %v2i8_0   = insertelement <2 x i8> %src16, i8 %val, i32 0
  %v2i8_3   = insertelement <2 x i8> %src16, i8 %val, i32 1

  %v4i8_a   = insertelement <4 x i8> %src32, i8 %val, i32 %arg
  %v4i8_0   = insertelement <4 x i8> %src32, i8 %val, i32 0
  %v4i8_3   = insertelement <4 x i8> %src32, i8 %val, i32 3

  %v8i8_a   = insertelement <8 x i8> %src64, i8 %val, i32 %arg
  %v8i8_0   = insertelement <8 x i8> %src64, i8 %val, i32 0
  %v8i8_7   = insertelement <8 x i8> %src64, i8 %val, i32 7

  %v16i8_a  = insertelement <16 x i8> %src128, i8 %val, i32 %arg
  %v16i8_0  = insertelement <16 x i8> %src128, i8 %val, i32 0
  %v16i8_8  = insertelement <16 x i8> %src128, i8 %val, i32 8
  %v16i8_15 = insertelement <16 x i8> %src128, i8 %val, i32 15

  %v32i8_a  = insertelement <32 x i8> %src256, i8 %val, i32 %arg
  %v32i8_0  = insertelement <32 x i8> %src256, i8 %val, i32 0
  %v32i8_7  = insertelement <32 x i8> %src256, i8 %val, i32 7
  %v32i8_8  = insertelement <32 x i8> %src256, i8 %val, i32 8
  %v32i8_15 = insertelement <32 x i8> %src256, i8 %val, i32 15
  %v32i8_24 = insertelement <32 x i8> %src256, i8 %val, i32 24
  %v32i8_31 = insertelement <32 x i8> %src256, i8 %val, i32 31

  %v64i8_a  = insertelement <64 x i8> %src512, i8 %val, i32 %arg
  %v64i8_0  = insertelement <64 x i8> %src512, i8 %val, i32 0
  %v64i8_7  = insertelement <64 x i8> %src512, i8 %val, i32 7
  %v64i8_8  = insertelement <64 x i8> %src512, i8 %val, i32 8
  %v64i8_15 = insertelement <64 x i8> %src512, i8 %val, i32 15
  %v64i8_24 = insertelement <64 x i8> %src512, i8 %val, i32 24
  %v64i8_31 = insertelement <64 x i8> %src512, i8 %val, i32 31
  %v64i8_32 = insertelement <64 x i8> %src512, i8 %val, i32 32
  %v64i8_48 = insertelement <64 x i8> %src512, i8 %val, i32 48
  %v64i8_63 = insertelement <64 x i8> %src512, i8 %val, i32 63

  ret i32 undef
}

define i32 @insert_i1(i32 %arg, i1 %val, <2 x i1> %src2, <4 x i1> %src4, <8 x i1> %src8, <16 x i1> %src16, <32 x i1> %src32, <64 x i1> %src64) {
; SSE2-LABEL: 'insert_i1'
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4
; SSE2-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31
; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
; SSE2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63
; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE3-LABEL: 'insert_i1'
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4
; SSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31
; SSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
; SSE3-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63
; SSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSSE3-LABEL: 'insert_i1'
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63
; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SSE4-LABEL: 'insert_i1'
; SSE4-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1
; SSE4-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2
; SSE4-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4
; SSE4-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31
; SSE4-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
; SSE4-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63
; SSE4-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX-LABEL: 'insert_i1'
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4
; AVX-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
; AVX-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63
; AVX-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'insert_i1'
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63
; AVX512F-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'insert_i1'
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63
; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; SLM-LABEL: 'insert_i1'
; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1
; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2
; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4
; SLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31
; SLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
; SLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63
; SLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; GLM-LABEL: 'insert_i1'
; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v2i1_a = insertelement <2 x i1> %src2, i1 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_0 = insertelement <2 x i1> %src2, i1 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v2i1_1 = insertelement <2 x i1> %src2, i1 %val, i32 1
; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v4i1_a = insertelement <4 x i1> %src4, i1 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_0 = insertelement <4 x i1> %src4, i1 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v4i1_2 = insertelement <4 x i1> %src4, i1 %val, i32 2
; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v8i1_a = insertelement <8 x i1> %src8, i1 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_0 = insertelement <8 x i1> %src8, i1 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i1_4 = insertelement <8 x i1> %src8, i1 %val, i32 4
; GLM-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %v16i1_a = insertelement <16 x i1> %src16, i1 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_0 = insertelement <16 x i1> %src16, i1 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_8 = insertelement <16 x i1> %src16, i1 %val, i32 8
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v32i1_a = insertelement <32 x i1> %src32, i1 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_0 = insertelement <32 x i1> %src32, i1 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_7 = insertelement <32 x i1> %src32, i1 %val, i32 7
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_8 = insertelement <32 x i1> %src32, i1 %val, i32 8
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31
; GLM-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %v64i1_a = insertelement <64 x i1> %src64, i1 %val, i32 %arg
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_0 = insertelement <64 x i1> %src64, i1 %val, i32 0
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_7 = insertelement <64 x i1> %src64, i1 %val, i32 7
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_8 = insertelement <64 x i1> %src64, i1 %val, i32 8
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
; GLM-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63
; GLM-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
  %v2i1_a  = insertelement <2 x i1> %src2, i1 %val, i32 %arg
  %v2i1_0  = insertelement <2 x i1> %src2, i1 %val, i32 0
  %v2i1_1  = insertelement <2 x i1> %src2, i1 %val, i32 1

  %v4i1_a  = insertelement <4 x i1> %src4, i1 %val, i32 %arg
  %v4i1_0  = insertelement <4 x i1> %src4, i1 %val, i32 0
  %v4i1_2  = insertelement <4 x i1> %src4, i1 %val, i32 2

  %v8i1_a  = insertelement <8 x i1> %src8, i1 %val, i32 %arg
  %v8i1_0  = insertelement <8 x i1> %src8, i1 %val, i32 0
  %v8i1_4  = insertelement <8 x i1> %src8, i1 %val, i32 4

  %v16i1_a  = insertelement <16 x i1> %src16, i1 %val, i32 %arg
  %v16i1_0  = insertelement <16 x i1> %src16, i1 %val, i32 0
  %v16i1_8  = insertelement <16 x i1> %src16, i1 %val, i32 8
  %v16i1_15 = insertelement <16 x i1> %src16, i1 %val, i32 15

  %v32i1_a  = insertelement <32 x i1> %src32, i1 %val, i32 %arg
  %v32i1_0  = insertelement <32 x i1> %src32, i1 %val, i32 0
  %v32i1_7  = insertelement <32 x i1> %src32, i1 %val, i32 7
  %v32i1_8  = insertelement <32 x i1> %src32, i1 %val, i32 8
  %v32i1_15 = insertelement <32 x i1> %src32, i1 %val, i32 15
  %v32i1_24 = insertelement <32 x i1> %src32, i1 %val, i32 24
  %v32i1_31 = insertelement <32 x i1> %src32, i1 %val, i32 31

  %v64i1_a  = insertelement <64 x i1> %src64, i1 %val, i32 %arg
  %v64i1_0  = insertelement <64 x i1> %src64, i1 %val, i32 0
  %v64i1_7  = insertelement <64 x i1> %src64, i1 %val, i32 7
  %v64i1_8  = insertelement <64 x i1> %src64, i1 %val, i32 8
  %v64i1_15 = insertelement <64 x i1> %src64, i1 %val, i32 15
  %v64i1_24 = insertelement <64 x i1> %src64, i1 %val, i32 24
  %v64i1_31 = insertelement <64 x i1> %src64, i1 %val, i32 31
  %v64i1_32 = insertelement <64 x i1> %src64, i1 %val, i32 32
  %v64i1_48 = insertelement <64 x i1> %src64, i1 %val, i32 48
  %v64i1_63 = insertelement <64 x i1> %src64, i1 %val, i32 63

  ret i32 undef
}