; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx1010 -S | FileCheck -check-prefixes=ALL,GFX9-10 %s
; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx900 -S | FileCheck -check-prefixes=ALL,GFX9-10 %s
; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -mtriple=amdgcn-unknown-amdhsa -mcpu=fiji -S | FileCheck -check-prefixes=ALL,VI %s
; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx1010 -cost-kind=code-size -S | FileCheck -check-prefixes=ALL-SIZE,GFX9-10-SIZE %s
; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx900 -cost-kind=code-size -S | FileCheck -check-prefixes=ALL-SIZE,GFX9-10-SIZE %s
; RUN: opt < %s -passes="print<cost-model>" 2>&1 -disable-output -mtriple=amdgcn-unknown-amdhsa -mcpu=fiji -cost-kind=code-size -S | FileCheck -check-prefixes=ALL-SIZE,VI-SIZE %s
; END.

define amdgpu_kernel void @shufflevector_i16(<2 x i16> %vec1, <2 x i16> %vec2) {
; GFX9-10-LABEL: 'shufflevector_i16'
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf00 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> zeroinitializer
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf10 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf11 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf02 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf22 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 3>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf33 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 3>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf13 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 3>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 3>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf32 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf000 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> zeroinitializer
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf001 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf010 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf011 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf100 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf101 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf110 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf111 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf002 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf020 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf022 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf200 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf202 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf220 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf222 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf112 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf121 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf122 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf211 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf212 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf221 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf00_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> zeroinitializer
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf10_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf11_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf02_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf22_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 3>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf33_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 3>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf13_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 3>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 3>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf32_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf000_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> zeroinitializer
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf001_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf010_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf011_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf100_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf101_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf110_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf111_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf002_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf020_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf022_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf200_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf202_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf220_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf222_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf112_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf121_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf122_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf211_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf212_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf221_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
;
; VI-LABEL: 'shufflevector_i16'
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf00 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> zeroinitializer
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf10 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf11 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf02 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf22 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 3>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf33 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 3>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf13 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 3>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 3>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf32 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf000 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> zeroinitializer
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf001 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf010 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf011 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf100 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf101 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf110 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf111 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf002 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf020 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf022 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf200 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf202 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf220 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf222 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf112 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf121 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf122 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf211 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf212 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf221 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf00_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> zeroinitializer
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf10_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf11_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf02_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf22_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 3>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf33_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 3>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf13_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 3>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 3>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf32_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf000_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> zeroinitializer
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf001_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf010_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf011_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf100_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf101_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf110_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf111_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf002_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf020_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf022_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf200_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf202_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf220_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf222_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf112_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf121_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf122_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf211_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf212_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf221_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
;
; GFX9-10-SIZE-LABEL: 'shufflevector_i16'
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf00 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> zeroinitializer
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf10 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf11 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf02 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf22 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 3>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf33 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 3>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf13 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 3>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 3>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf32 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf000 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> zeroinitializer
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf001 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf010 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf011 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf100 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf101 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf110 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf111 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf002 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf020 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf022 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf200 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf202 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf220 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf222 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf112 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf121 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf122 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf211 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf212 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf221 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf00_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> zeroinitializer
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf10_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf11_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf02_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf22_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 3>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf33_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 3>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf13_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 3>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 3>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf32_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf000_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> zeroinitializer
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf001_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf010_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf011_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf100_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf101_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf110_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf111_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf002_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf020_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf022_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf200_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf202_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf220_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf222_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf112_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf121_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf122_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf211_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf212_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf221_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; VI-SIZE-LABEL: 'shufflevector_i16'
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf00 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> zeroinitializer
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf10 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf11 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf02 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf22 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 3>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf33 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 3>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf13 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 3>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 3>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf32 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf000 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> zeroinitializer
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf001 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf010 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf011 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf100 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf101 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf110 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf111 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf002 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf020 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf022 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf200 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf202 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf220 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf222 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf112 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf121 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf122 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf211 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf212 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf221 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf00_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> zeroinitializer
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf10_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf11_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf02_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf22_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 3>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf33_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 3>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf13_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 3>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 3>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf32_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf000_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> zeroinitializer
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf001_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf010_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf011_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf100_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf101_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf110_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %shuf111_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf002_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf020_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf022_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf200_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf202_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf220_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf222_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf112_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf121_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf122_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf211_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf212_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf221_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
;
  %shuf00 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> zeroinitializer
  %shuf01 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 1>
  %shuf10 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 0>
  %shuf11 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 1>
  %shuf02 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 2>
  %shuf20 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 0>
  %shuf22 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 2>
  %shuf03 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 0, i32 3>
  %shuf30 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 0>
  %shuf33 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 3>
  %shuf12 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 2>
  %shuf21 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 1>
  %shuf13 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 1, i32 3>
  %shuf31 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 1>
  %shuf23 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 2, i32 3>
  %shuf32 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <2 x i32> <i32 3, i32 2>
  %shuf000 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 0>
  %shuf001 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
  %shuf010 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
  %shuf011 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
  %shuf100 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
  %shuf101 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
  %shuf110 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
  %shuf111 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
  %shuf002 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
  %shuf020 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
  %shuf022 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
  %shuf200 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
  %shuf202 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
  %shuf220 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
  %shuf222 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
  %shuf112 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
  %shuf121 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
  %shuf122 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
  %shuf211 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
  %shuf212 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
  %shuf221 = shufflevector <2 x i16> %vec1, <2 x i16> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
  %shuf00_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> zeroinitializer
  %shuf01_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 1>
  %shuf10_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 0>
  %shuf11_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 1>
  %shuf02_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 2>
  %shuf20_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 0>
  %shuf22_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 2>
  %shuf03_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 0, i32 3>
  %shuf30_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 0>
  %shuf33_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 3>
  %shuf12_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 2>
  %shuf21_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 1>
  %shuf13_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 1, i32 3>
  %shuf31_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 1>
  %shuf23_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 2, i32 3>
  %shuf32_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <2 x i32> <i32 3, i32 2>
  %shuf000_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 0>
  %shuf001_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
  %shuf010_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
  %shuf011_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
  %shuf100_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
  %shuf101_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
  %shuf110_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
  %shuf111_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
  %shuf002_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
  %shuf020_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
  %shuf022_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
  %shuf200_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
  %shuf202_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
  %shuf220_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
  %shuf222_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
  %shuf112_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
  %shuf121_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
  %shuf122_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
  %shuf211_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
  %shuf212_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
  %shuf221_2 = shufflevector <2 x i16> %vec1, <2 x i16> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
  ret void
}

; Should not assert
define amdgpu_kernel void @shufflevector_i8(<2 x i8> %vec1, <2 x i8> %vec2) {
; ALL-LABEL: 'shufflevector_i8'
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf00 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> zeroinitializer
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf10 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf11 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf02 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf22 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 0, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf33 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf13 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf32 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf000 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> zeroinitializer
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf001 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf010 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf011 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf100 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf101 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf110 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf111 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf002 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf020 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf022 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf200 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf202 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf220 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf222 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf112 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf121 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf122 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf211 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf212 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf221 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf00_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> zeroinitializer
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf10_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf11_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf02_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf22_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 0, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf33_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf13_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf32_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf000_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> zeroinitializer
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf001_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf010_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf011_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf100_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf101_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf110_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf111_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf002_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf020_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf022_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf200_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf202_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf220_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf222_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf112_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf121_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf122_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf211_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf212_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf221_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
;
; ALL-SIZE-LABEL: 'shufflevector_i8'
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf00 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> zeroinitializer
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf10 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf11 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf02 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf22 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 0, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf33 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf13 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf32 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf000 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> zeroinitializer
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf001 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf010 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf011 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf100 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf101 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf110 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf111 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf002 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf020 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf022 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf200 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf202 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf220 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf222 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf112 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf121 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf122 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf211 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf212 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf221 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf00_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> zeroinitializer
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf10_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf11_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf02_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %shuf20_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf22_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 0, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf30_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf33_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf12_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %shuf13_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf31_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf32_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf000_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> zeroinitializer
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf001_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf010_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf011_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf100_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf101_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf110_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf111_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf002_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf020_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf022_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf200_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf202_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf220_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf222_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf112_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf121_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf122_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf211_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf212_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %shuf221_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
;
  %shuf00 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> zeroinitializer
  %shuf01 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 0, i32 1>
  %shuf10 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 0>
  %shuf11 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 1>
  %shuf02 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 0, i32 2>
  %shuf20 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 0>
  %shuf22 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 2>
  %shuf03 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 0, i32 3>
  %shuf30 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 0>
  %shuf33 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 3>
  %shuf12 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 2>
  %shuf21 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 1>
  %shuf13 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 1, i32 3>
  %shuf31 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 1>
  %shuf23 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 2, i32 3>
  %shuf32 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <2 x i32> <i32 3, i32 2>
  %shuf000 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 0, i32 0>
  %shuf001 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
  %shuf010 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
  %shuf011 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
  %shuf100 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
  %shuf101 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
  %shuf110 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
  %shuf111 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
  %shuf002 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
  %shuf020 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
  %shuf022 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
  %shuf200 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
  %shuf202 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
  %shuf220 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
  %shuf222 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
  %shuf112 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
  %shuf121 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
  %shuf122 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
  %shuf211 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
  %shuf212 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
  %shuf221 = shufflevector <2 x i8> %vec1, <2 x i8> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
  %shuf00_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> zeroinitializer
  %shuf01_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 0, i32 1>
  %shuf10_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 0>
  %shuf11_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 1>
  %shuf02_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 0, i32 2>
  %shuf20_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 0>
  %shuf22_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 2>
  %shuf03_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 0, i32 3>
  %shuf30_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 0>
  %shuf33_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 3>
  %shuf12_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 2>
  %shuf21_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 1>
  %shuf13_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 1, i32 3>
  %shuf31_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 1>
  %shuf23_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 2, i32 3>
  %shuf32_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <2 x i32> <i32 3, i32 2>
  %shuf000_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 0, i32 0>
  %shuf001_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
  %shuf010_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
  %shuf011_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
  %shuf100_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
  %shuf101_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
  %shuf110_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
  %shuf111_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
  %shuf002_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
  %shuf020_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
  %shuf022_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
  %shuf200_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
  %shuf202_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
  %shuf220_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
  %shuf222_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
  %shuf112_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
  %shuf121_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
  %shuf122_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
  %shuf211_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
  %shuf212_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
  %shuf221_2 = shufflevector <2 x i8> %vec1, <2 x i8> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
  ret void
}

define amdgpu_kernel void @shufflevector_i32(<2 x i32> %vec1, <2 x i32> %vec2) {
; ALL-LABEL: 'shufflevector_i32'
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf00 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> zeroinitializer
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf10 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf11 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf02 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf20 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf22 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 0, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf30 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf33 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf12 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf13 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf31 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf32 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf000 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> zeroinitializer
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf001 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf010 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf011 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf100 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf101 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf110 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf111 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf002 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf020 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf022 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf200 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf202 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf220 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf222 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf112 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf121 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf122 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf211 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf212 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf221 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf00_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> zeroinitializer
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf10_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf11_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf02_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf20_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf22_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 0, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf30_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf33_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf12_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf13_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf31_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf32_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf000_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> zeroinitializer
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf001_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf010_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf011_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf100_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf101_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf110_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf111_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf002_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf020_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf022_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf200_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf202_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf220_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf222_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf112_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf121_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf122_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf211_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf212_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf221_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
; ALL-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
;
; ALL-SIZE-LABEL: 'shufflevector_i32'
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf00 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> zeroinitializer
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf10 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf11 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf02 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf20 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf22 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 0, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf30 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf33 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf12 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf13 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf31 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf32 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf000 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> zeroinitializer
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf001 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf010 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf011 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf100 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf101 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf110 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf111 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf002 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf020 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf022 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf200 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf202 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf220 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf222 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf112 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf121 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf122 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf211 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf212 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf221 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf00_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> zeroinitializer
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf01_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf10_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf11_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf02_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf20_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf22_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf03_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 0, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf30_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf33_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf12_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf21_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf13_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf31_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf23_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf32_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf000_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> zeroinitializer
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf001_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf010_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf011_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf100_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf101_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf110_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf111_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf002_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf020_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf022_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf200_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf202_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf220_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf222_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf112_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf121_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf122_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf211_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf212_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %shuf221_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
;
  %shuf00 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> zeroinitializer
  %shuf01 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 0, i32 1>
  %shuf10 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 0>
  %shuf11 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 1>
  %shuf02 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 0, i32 2>
  %shuf20 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 0>
  %shuf22 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 2>
  %shuf03 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 0, i32 3>
  %shuf30 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 0>
  %shuf33 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 3>
  %shuf12 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 2>
  %shuf21 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 1>
  %shuf13 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 1, i32 3>
  %shuf31 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 1>
  %shuf23 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 2, i32 3>
  %shuf32 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <2 x i32> <i32 3, i32 2>
  %shuf000 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 0, i32 0>
  %shuf001 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 0, i32 1>
  %shuf010 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 1, i32 0>
  %shuf011 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 1, i32 1>
  %shuf100 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 0, i32 0>
  %shuf101 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 0, i32 1>
  %shuf110 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 1, i32 0>
  %shuf111 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 1, i32 1>
  %shuf002 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 0, i32 2>
  %shuf020 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 2, i32 0>
  %shuf022 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 0, i32 2, i32 2>
  %shuf200 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 0, i32 0>
  %shuf202 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 0, i32 2>
  %shuf220 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 2, i32 0>
  %shuf222 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 2, i32 2>
  %shuf112 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 1, i32 2>
  %shuf121 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 2, i32 1>
  %shuf122 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 1, i32 2, i32 2>
  %shuf211 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 1, i32 1>
  %shuf212 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 1, i32 2>
  %shuf221 = shufflevector <2 x i32> %vec1, <2 x i32> %vec1, <3 x i32> <i32 2, i32 2, i32 1>
  %shuf00_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> zeroinitializer
  %shuf01_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 0, i32 1>
  %shuf10_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 0>
  %shuf11_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 1>
  %shuf02_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 0, i32 2>
  %shuf20_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 0>
  %shuf22_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 2>
  %shuf03_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 0, i32 3>
  %shuf30_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 0>
  %shuf33_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 3>
  %shuf12_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 2>
  %shuf21_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 1>
  %shuf13_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 1, i32 3>
  %shuf31_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 1>
  %shuf23_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 2, i32 3>
  %shuf32_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <2 x i32> <i32 3, i32 2>
  %shuf000_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 0, i32 0>
  %shuf001_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 0, i32 1>
  %shuf010_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 1, i32 0>
  %shuf011_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 1, i32 1>
  %shuf100_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 0, i32 0>
  %shuf101_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 0, i32 1>
  %shuf110_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 1, i32 0>
  %shuf111_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 1, i32 1>
  %shuf002_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 0, i32 2>
  %shuf020_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 2, i32 0>
  %shuf022_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 0, i32 2, i32 2>
  %shuf200_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 0, i32 0>
  %shuf202_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 0, i32 2>
  %shuf220_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 2, i32 0>
  %shuf222_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 2, i32 2>
  %shuf112_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 1, i32 2>
  %shuf121_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 2, i32 1>
  %shuf122_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 1, i32 2, i32 2>
  %shuf211_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 1, i32 1>
  %shuf212_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 1, i32 2>
  %shuf221_2 = shufflevector <2 x i32> %vec1, <2 x i32> %vec2, <3 x i32> <i32 2, i32 2, i32 1>
  ret void
}

; Other shuffle cases
define void @shuffle(<2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i8> %i8v4, <4 x i8> %i8v4_2, <6 x i8> %i8v6, <6 x i8> %i8v6_2, <8 x i8> %i8v8, <8 x i8> %i8v8_2, <16 x i8> %i8v16, <16 x i8> %i8v16_2, <2 x i16> %i16v2, <2 x i16> %i16v2_2, <4 x i16> %i16v4, <4 x i16> %i16v4_2, <8 x i16> %i16v8, <8 x i16> %i16v8_2, <2 x i32> %i32v2, <2 x i32> %i32v2_2, <4 x i32> %i32v4, <4 x i32> %i32v4_2, <2 x float> %floatv2, <2 x float> %floatv2_2, <4 x float> %floatv4, <4 x float> %floatv4_2,<2 x i64> %i64v2, <2 x i64> %i64v2_2,<2 x double> %doublev2, <2 x double> %doublev2_2) {
; GFX9-10-LABEL: 'shuffle'
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_2_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_4 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_4_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8_4 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8_4_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_8 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_8 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_8_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v6i8_8 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v6i8_8_2 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_8 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_8_2 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_16 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_16_2 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16_2, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i16_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i16_2_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16_4 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16_4_2 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_8 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_8_2 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i32_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i32_2_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_4 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_4_2 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f32_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f32_2_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_4 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_4_2 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i64_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i64_2_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f64_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f64_2_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
;
; VI-LABEL: 'shuffle'
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_2_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_4 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_4_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8_4 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8_4_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_8 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_8 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_8_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v6i8_8 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v6i8_8_2 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_8 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_8_2 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_16 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_16_2 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16_2, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i16_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i16_2_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2_2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16_4 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16_4_2 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_8 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_8_2 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i32_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i32_2_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2_2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_4 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_4_2 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f32_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f32_2_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2_2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_4 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_4_2 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i64_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i64_2_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2_2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f64_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f64_2_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2_2, <2 x i32> <i32 1, i32 0>
; VI-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
;
; GFX9-10-SIZE-LABEL: 'shuffle'
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_2_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_4 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_4_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8_4 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8_4_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_8 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_8 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_8_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v6i8_8 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v6i8_8_2 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_8 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_8_2 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_16 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_16_2 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16_2, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i16_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i16_2_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16_4 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16_4_2 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_8 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_8_2 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i32_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i32_2_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_4 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_4_2 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f32_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f32_2_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_4 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_4_2 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i64_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i64_2_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f64_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f64_2_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2_2, <2 x i32> <i32 1, i32 0>
; GFX9-10-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; VI-SIZE-LABEL: 'shuffle'
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_2_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_4 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i8_4_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8_4 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i8_4_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_8 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v2i8_8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_8 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i8_8_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v6i8_8 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v6i8_8_2 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_8 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i8_8_2 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_16 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i8_16_2 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16_2, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i16_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v2i16_2_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2_2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16_4 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v4i16_4_2 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_8 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v8i16_8_2 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i32_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i32_2_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2_2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_4 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_4_2 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f32_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f32_2_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2_2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_4 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_4_2 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i64_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2i64_2_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2_2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f64_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v2f64_2_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2_2, <2 x i32> <i32 1, i32 0>
; VI-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
;
  %v2i8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <2 x i32> <i32 1, i32 0>
  %v2i8_2_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <2 x i32> <i32 1, i32 0>
  %v2i8_4 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v2i8_4_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v4i8_4 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v4i8_4_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v2i8_8 = shufflevector <2 x i8>  %i8v2, <2 x i8>  %i8v2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v2i8_8_2 = shufflevector <2 x i8>  %i8v2, <2 x i8>  %i8v2_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v4i8_8 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v4i8_8_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v6i8_8 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v6i8_8_2 = shufflevector <6 x i8> %i8v6, <6 x i8> %i8v6_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v8i8_8 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v8i8_8_2 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v16i8_16 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v16i8_16_2 = shufflevector <16 x i8> %i8v16, <16 x i8> %i8v16_2, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v2i16_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2, <2 x i32> <i32 1, i32 0>
  %v2i16_2_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2_2, <2 x i32> <i32 1, i32 0>
  %v4i16_4 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v4i16_4_2 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v8i16_8 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v8i16_8_2 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8_2, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
  %v2i32_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2, <2 x i32> <i32 1, i32 0>
  %v2i32_2_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2_2, <2 x i32> <i32 1, i32 0>
  %v4i32_4 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v4i32_4_2 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v2f32_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2, <2 x i32> <i32 1, i32 0>
  %v2f32_2_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2_2, <2 x i32> <i32 1, i32 0>
  %v4f32_4 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v4f32_4_2 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4_2, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %v2i64_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2, <2 x i32> <i32 1, i32 0>
  %v2i64_2_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2_2, <2 x i32> <i32 1, i32 0>
  %v2f64_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2, <2 x i32> <i32 1, i32 0>
  %v2f64_2_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2_2, <2 x i32> <i32 1, i32 0>
  ret void
}

define void @concat(<2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i8> %i8v4, <4 x i8> %i8v4_2, <8 x i8> %i8v8, <8 x i8> %i8v8_2, <2 x half> %halfv2, <2 x half> %halfv2_2, <4 x half> %halfv4, <4 x half> %halfv4_2, <8 x half> %halfv8, <8 x half> %halfv8_2, <2 x i16> %i16v2, <2 x i16> %i16v2_2, <4 x i16> %i16v4, <4 x i16> %i16v4_2, <8 x i16> %i16v8, <8 x i16> %i16v8_2, <2 x i32> %i32v2, <2 x i32> %i32v2_2, <4 x i32> %i32v4, <4 x i32> %i32v4_2, <2 x float> %floatv2, <2 x float> %floatv2_2, <4 x float> %floatv4, <4 x float> %floatv4_2,<2 x i64> %i64v2, <2 x i64> %i64v2_2,<2 x double> %doublev2, <2 x double> %doublev2_2) {
; ALL-LABEL: 'concat'
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i8 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i8 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i8 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i16 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i16 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i16 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i32 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i64 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f16 = shufflevector <2 x half> %halfv2, <2 x half> %halfv2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8f16 = shufflevector <4 x half> %halfv4, <4 x half> %halfv4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16f16 = shufflevector <8 x half> %halfv8, <8 x half> %halfv8, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8f32 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f64 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i8_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i8_2 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8_2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i16_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i16_2 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i16_2 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8_2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i32_2 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i64_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f16_2 = shufflevector <2 x half> %halfv2, <2 x half> %halfv2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8f16_2 = shufflevector <4 x half> %halfv4, <4 x half> %halfv4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16f16_2 = shufflevector <8 x half> %halfv8, <8 x half> %halfv8_2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8f32_2 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f64_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: ret void
;
; ALL-SIZE-LABEL: 'concat'
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i8 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i8 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i8 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i16 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i16 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i16 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i32 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i64 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f16 = shufflevector <2 x half> %halfv2, <2 x half> %halfv2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8f16 = shufflevector <4 x half> %halfv4, <4 x half> %halfv4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16f16 = shufflevector <8 x half> %halfv8, <8 x half> %halfv8, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8f32 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f64 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i8_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i8_2 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8_2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i16_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i16_2 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i16_2 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8_2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i32_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8i32_2 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4i64_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f16_2 = shufflevector <2 x half> %halfv2, <2 x half> %halfv2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8f16_2 = shufflevector <4 x half> %halfv4, <4 x half> %halfv4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16f16_2 = shufflevector <8 x half> %halfv8, <8 x half> %halfv8_2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f32_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v8f32_2 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v4f64_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; ALL-SIZE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
;
  %v4i8 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8i8 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v16i8 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  %v4i16 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8i16 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v16i16 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  %v4i32 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8i32 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v4i64 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v4f16 = shufflevector <2 x half> %halfv2, <2 x half> %halfv2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8f16 = shufflevector <4 x half> %halfv4, <4 x half> %halfv4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v16f16 = shufflevector <8 x half> %halfv8, <8 x half> %halfv8, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  %v4f32 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8f32 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v4f64 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v4i8_2 = shufflevector <2 x i8> %i8v2, <2 x i8> %i8v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8i8_2 = shufflevector <4 x i8> %i8v4, <4 x i8> %i8v4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v16i8_2 = shufflevector <8 x i8> %i8v8, <8 x i8> %i8v8_2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  %v4i16_2 = shufflevector <2 x i16> %i16v2, <2 x i16> %i16v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8i16_2 = shufflevector <4 x i16> %i16v4, <4 x i16> %i16v4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v16i16_2 = shufflevector <8 x i16> %i16v8, <8 x i16> %i16v8_2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  %v4i32_2 = shufflevector <2 x i32> %i32v2, <2 x i32> %i32v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8i32_2 = shufflevector <4 x i32> %i32v4, <4 x i32> %i32v4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v4i64_2 = shufflevector <2 x i64> %i64v2, <2 x i64> %i64v2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v4f16_2 = shufflevector <2 x half> %halfv2, <2 x half> %halfv2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8f16_2 = shufflevector <4 x half> %halfv4, <4 x half> %halfv4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v16f16_2 = shufflevector <8 x half> %halfv8, <8 x half> %halfv8_2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
  %v4f32_2 = shufflevector <2 x float> %floatv2, <2 x float> %floatv2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %v8f32_2 = shufflevector <4 x float> %floatv4, <4 x float> %floatv4_2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  %v4f64_2 = shufflevector <2 x double> %doublev2, <2 x double> %doublev2_2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  ret void
}