From 81eb63ad7f328a8c2f0293d85aecde18b6edf16a Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Fri, 25 Jul 2025 15:49:47 +0100
Subject: [PATCH] [CostModel][X86] Complicate the cross lane single/two source
 shuffle masks

Try to ensure shuffle masks don't simplify too much to easier shuffle kinds when splitting
---
 .../CostModel/X86/shuffle-single-src-fp16.ll  |  16 +-
 .../CostModel/X86/shuffle-single-src.ll       | 228 +++++++++--------
 .../Analysis/CostModel/X86/shuffle-two-src.ll | 232 ++++++++----------
 3 files changed, 237 insertions(+), 239 deletions(-)

diff --git a/llvm/test/Analysis/CostModel/X86/shuffle-single-src-fp16.ll b/llvm/test/Analysis/CostModel/X86/shuffle-single-src-fp16.ll
index 007aa59bc497..21d576677de0 100644
--- a/llvm/test/Analysis/CostModel/X86/shuffle-single-src-fp16.ll
+++ b/llvm/test/Analysis/CostModel/X86/shuffle-single-src-fp16.ll
@@ -3,15 +3,15 @@
 
 define void @test_vXf16(<8 x half> %src128, <16 x half> %src256, <32 x half> %src512, <64 x half> %src1024) {
 ; CHECK-LABEL: 'test_vXf16'
-; CHECK-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:  Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x half> %src1024, <64 x half> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; CHECK-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; CHECK-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:  Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x half> %src1024, <64 x half> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; CHECK-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
-  %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V1024 = shufflevector <64 x half> %src1024, <64 x half> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+  %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+  %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V1024 = shufflevector <64 x half> %src1024, <64 x half> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
   ret void
 }
diff --git a/llvm/test/Analysis/CostModel/X86/shuffle-single-src.ll b/llvm/test/Analysis/CostModel/X86/shuffle-single-src.ll
index 3727d7b64d7a..a7732920dd24 100644
--- a/llvm/test/Analysis/CostModel/X86/shuffle-single-src.ll
+++ b/llvm/test/Analysis/CostModel/X86/shuffle-single-src.ll
@@ -21,42 +21,42 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
 ; SSE-LABEL: 'test_vXf64'
 ; SSE-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>
 ; SSE-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; SSE-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>
 ; SSE-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXf64'
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>
 ; XOP-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 5 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 11 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXf64'
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>
 ; AVX1-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 5 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 11 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXf64'
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 4 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 10 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512-LABEL: 'test_vXf64'
 ; AVX512-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:2 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:2 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> <i32 1, i32 1>
   %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-  %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
+  %V1024 = shufflevector <16 x double> %src1024, <16 x double> undef, <16 x i32> <i32 15, i32 0, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 13, i32 3, i32 11, i32 1, i32 0>
   ret void
 }
 
@@ -64,79 +64,93 @@ define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512)
 ; SSE-LABEL: 'test_vXi64'
 ; SSE-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>
 ; SSE-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; SSE-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
 ; SSE-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXi64'
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>
 ; XOP-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 5 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXi64'
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>
 ; AVX1-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 5 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXi64'
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512-LABEL: 'test_vXi64'
 ; AVX512-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> <i32 1, i32 1>
   %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-  %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 5, i32 1, i32 0>
   ret void
 }
 
 define void @test_vXf32(<2 x float> %src64, <4 x float> %src128, <8 x float> %src256, <16 x float> %src512) {
-; SSE-LABEL: 'test_vXf32'
-; SSE-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>
-; SSE-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; SSE-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
+; SSE2-LABEL: 'test_vXf32'
+; SSE2-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>
+; SSE2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of 11 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
+;
+; SSSE3-LABEL: 'test_vXf32'
+; SSSE3-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>
+; SSSE3-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 11 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; SSSE3-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
+;
+; SSE42-LABEL: 'test_vXf32'
+; SSE42-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>
+; SSE42-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; SSE42-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXf32'
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 12 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXf32'
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 12 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXf32'
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 9 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512-LABEL: 'test_vXf32'
 ; AVX512-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V64  = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> <i32 1, i32 1>
   %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
   %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
   ret void
 }
 
@@ -144,48 +158,48 @@ define void @test_vXi32(<2 x i32> %src64, <4 x i32> %src128, <8 x i32> %src256,
 ; SSE-LABEL: 'test_vXi32'
 ; SSE-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>
 ; SSE-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; SSE-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 2, i32 1, i32 0>
-; SSE-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 13, i32 10, i32 9, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE-NEXT:  Cost Model: Found costs of 10 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 11 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; SSE-NEXT:  Cost Model: Found costs of 18 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; SSE-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXi32'
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 13, i32 10, i32 9, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 12 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 12 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; XOP-NEXT:  Cost Model: Found costs of 26 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXi32'
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 13, i32 10, i32 9, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 14 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 12 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; AVX1-NEXT:  Cost Model: Found costs of 26 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXi32'
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 13, i32 10, i32 9, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 9 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; AVX2-NEXT:  Cost Model: Found costs of 20 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512-LABEL: 'test_vXi32'
 ; AVX512-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 13, i32 10, i32 9, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:2 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:2 CodeSize:2 Lat:6 SizeLat:2 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V64  = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> <i32 1, i32 1>
   %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 0>
-  %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 2, i32 1, i32 0>
-  %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 13, i32 10, i32 9, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> <i32 7, i32 6, i32 5, i32 5, i32 3, i32 0, i32 1, i32 0>
+  %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+  %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
   ret void
 }
 
@@ -193,90 +207,90 @@ define void @test_vXi16(<2 x i16> %src32, <4 x i16> %src64, <8 x i16> %src128, <
 ; SSE2-LABEL: 'test_vXi16'
 ; SSE2-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
 ; SSE2-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:5 Lat:5 SizeLat:5 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:6 CodeSize:8 Lat:8 SizeLat:8 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:18 CodeSize:24 Lat:24 SizeLat:24 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:24 CodeSize:26 Lat:26 SizeLat:26 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:5 Lat:5 SizeLat:5 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:18 CodeSize:24 Lat:24 SizeLat:24 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:42 CodeSize:56 Lat:56 SizeLat:56 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:72 CodeSize:94 Lat:94 SizeLat:94 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; SSE2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSSE3-LABEL: 'test_vXi16'
 ; SSSE3-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
 ; SSSE3-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-; SSSE3-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; SSSE3-NEXT:  Cost Model: Found costs of 21 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 39 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; SSSE3-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSE42-LABEL: 'test_vXi16'
 ; SSE42-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
 ; SSE42-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-; SSE42-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; SSE42-NEXT:  Cost Model: Found costs of 21 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 35 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; SSE42-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXi16'
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 16 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; XOP-NEXT:  Cost Model: Found costs of 18 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 63 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXi16'
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 8 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 16 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 20 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 15 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; AVX1-NEXT:  Cost Model: Found costs of 30 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 99 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXi16'
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 10 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; AVX2-NEXT:  Cost Model: Found costs of 14 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 44 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512F-LABEL: 'test_vXi16'
 ; AVX512F-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
 ; AVX512F-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-; AVX512F-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 14 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 21 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; AVX512F-NEXT:  Cost Model: Found costs of 14 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 84 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX512F-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512BW-LABEL: 'test_vXi16'
 ; AVX512BW-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
 ; AVX512BW-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-; AVX512BW-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX512BW-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512VBMI-LABEL: 'test_vXi16'
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V32  = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> <i32 1, i32 1>
   %V64  = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> <i32 2, i32 2, i32 1, i32 1>
-  %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 13, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 31, i32 30, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> <i32 7, i32 6, i32 6, i32 4, i32 3, i32 0, i32 1, i32 0>
+  %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> <i32 4, i32 15, i32 13, i32 13, i32 11, i32 10, i32 8, i32 8, i32 3, i32 6, i32 5, i32 14, i32 7, i32 2, i32 1, i32 16>
+  %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
   ret void
 }
 
@@ -286,8 +300,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; SSE2-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
 ; SSE2-NEXT:  Cost Model: Found costs of 5 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
 ; SSE2-NEXT:  Cost Model: Found costs of RThru:8 CodeSize:10 Lat:10 SizeLat:10 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:17 CodeSize:19 Lat:19 SizeLat:19 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:35 CodeSize:37 Lat:37 SizeLat:37 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:22 CodeSize:26 Lat:26 SizeLat:26 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:77 CodeSize:91 Lat:91 SizeLat:91 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; SSE2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSSE3-LABEL: 'test_vXi8'
@@ -295,8 +309,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; SSSE3-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
 ; SSSE3-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
 ; SSSE3-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 6 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 21 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; SSSE3-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSE42-LABEL: 'test_vXi8'
@@ -304,8 +318,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; SSE42-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
 ; SSE42-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
 ; SSE42-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 6 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 21 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; SSE42-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXi8'
@@ -313,8 +327,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 18 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXi8'
@@ -322,8 +336,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 8 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 12 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 8 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 30 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXi8'
@@ -331,8 +345,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 14 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512F-LABEL: 'test_vXi8'
@@ -340,8 +354,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX512F-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
 ; AVX512F-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
 ; AVX512F-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 14 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 14 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX512F-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512BW-LABEL: 'test_vXi8'
@@ -349,8 +363,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX512BW-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
 ; AVX512BW-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
 ; AVX512BW-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX512BW-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512VBMI-LABEL: 'test_vXi8'
@@ -358,16 +372,16 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 1 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 1 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 1 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 1 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V16  = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> <i32 1, i32 1>
   %V32  = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 1>
   %V64  = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> <i32 7, i32 7, i32 5, i32 5, i32 3, i32 3, i32 1, i32 1>
   %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 63, i32 62, i32 61, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> <i32 8, i32 30, i32 20, i32 28, i32 27, i32 15, i32 13, i32 24, i32 11, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 26, i32 15, i32 14, i32 25, i32 12, i32 11, i32 23, i32 9, i32 31, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> <i32 0, i32 62, i32 61, i32 60, i32 7, i32 58, i32 57, i32 55, i32 56, i32 54, i32 53, i32 52, i32 1, i32 50, i32 47, i32 48, i32 49, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 11, i32 37, i32 36, i32 33, i32 34, i32 35, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 23, i32 24, i32 20, i32 35, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 42, i32 12, i32 34, i32 10, i32 8, i32 9, i32 24, i32 6, i32 5, i32 4, i32 29, i32 2, i32 1, i32 0>
   ret void
 }
 
diff --git a/llvm/test/Analysis/CostModel/X86/shuffle-two-src.ll b/llvm/test/Analysis/CostModel/X86/shuffle-two-src.ll
index 56afc013e6e0..30163163b0db 100644
--- a/llvm/test/Analysis/CostModel/X86/shuffle-two-src.ll
+++ b/llvm/test/Analysis/CostModel/X86/shuffle-two-src.ll
@@ -22,41 +22,41 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
 ; SSE-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> %src128_1, <2 x i32> <i32 3, i32 0>
 ; SSE-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x double> %src256, <4 x double> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; SSE-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x double> %src512, <8 x double> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; SSE-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; SSE-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXf64'
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> %src128_1, <2 x i32> <i32 3, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <4 x double> %src256, <4 x double> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; XOP-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <8 x double> %src512, <8 x double> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; XOP-NEXT:  Cost Model: Found costs of 12 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 16 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXf64'
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> %src128_1, <2 x i32> <i32 3, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <4 x double> %src256, <4 x double> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; AVX1-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <8 x double> %src512, <8 x double> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; AVX1-NEXT:  Cost Model: Found costs of 12 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 16 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXf64'
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> %src128_1, <2 x i32> <i32 3, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <4 x double> %src256, <4 x double> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; AVX2-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <8 x double> %src512, <8 x double> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; AVX2-NEXT:  Cost Model: Found costs of 11 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 16 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512-LABEL: 'test_vXf64'
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <2 x double> %src128, <2 x double> %src128_1, <2 x i32> <i32 3, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <4 x double> %src256, <4 x double> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <8 x double> %src512, <8 x double> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:3 Lat:9 SizeLat:3 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:4 Lat:12 SizeLat:4 for: %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V128 = shufflevector <2 x double> %src128, <2 x double> %src128_1, <2 x i32> <i32 3, i32 0>
   %V256 = shufflevector <4 x double> %src256, <4 x double> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
   %V512 = shufflevector <8 x double> %src512, <8 x double> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-  %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V1024 = shufflevector <16 x double> %src1024, <16 x double> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
   ret void
 }
 
@@ -65,41 +65,41 @@ define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512,
 ; SSE-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> %src128_1, <2 x i32> <i32 3, i32 0>
 ; SSE-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; SSE-NEXT:  Cost Model: Found costs of 4 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; SSE-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; SSE-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXi64'
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> %src128_1, <2 x i32> <i32 3, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; XOP-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; XOP-NEXT:  Cost Model: Found costs of 12 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 16 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXi64'
 ; AVX1-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> %src128_1, <2 x i32> <i32 3, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; AVX1-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; AVX1-NEXT:  Cost Model: Found costs of 12 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 16 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXi64'
 ; AVX2-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> %src128_1, <2 x i32> <i32 3, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; AVX2-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; AVX2-NEXT:  Cost Model: Found costs of 11 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 16 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512-LABEL: 'test_vXi64'
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <2 x i64> %src128, <2 x i64> %src128_1, <2 x i32> <i32 3, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <4 x i64> %src256, <4 x i64> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <8 x i64> %src512, <8 x i64> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:3 Lat:9 SizeLat:3 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:4 Lat:12 SizeLat:4 for: %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V128 = shufflevector <2 x i64> %src128, <2 x i64> %src128_1, <2 x i32> <i32 3, i32 0>
   %V256 = shufflevector <4 x i64> %src256, <4 x i64> %src256_1, <4 x i32> <i32 3, i32 3, i32 7, i32 6>
   %V512 = shufflevector <8 x i64> %src512, <8 x i64> %src512_1, <8 x i32> <i32 7, i32 6, i32 12, i32 4, i32 3, i32 2, i32 1, i32 15>
-  %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 12, i32 13, i32 10, i32 18, i32 8, i32 8, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V1024 = shufflevector <16 x i64> %src1024, <16 x i64> %src1024_1, <16 x i32> <i32 30, i32 14, i32 13, i32 28, i32 13, i32 10, i32 18, i32 8, i32 8, i32 24, i32 31, i32 4, i32 9, i32 2, i32 8, i32 0>
   ret void
 }
 
@@ -107,129 +107,113 @@ define void @test_vXf32(<2 x float> %src64, <4 x float> %src128, <8 x float> %sr
 ; SSE2-LABEL: 'test_vXf32'
 ; SSE2-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x float> %src64, <2 x float> %src64_1, <2 x i32> <i32 3, i32 0>
 ; SSE2-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x float> %src128, <4 x float> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; SSE2-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of 14 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of 22 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; SSE2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSSE3-LABEL: 'test_vXf32'
 ; SSSE3-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x float> %src64, <2 x float> %src64_1, <2 x i32> <i32 3, i32 0>
 ; SSSE3-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x float> %src128, <4 x float> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; SSSE3-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 14 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 22 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; SSSE3-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSE42-LABEL: 'test_vXf32'
 ; SSE42-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x float> %src64, <2 x float> %src64_1, <2 x i32> <i32 3, i32 0>
 ; SSE42-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x float> %src128, <4 x float> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; SSE42-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 7 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 13 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 9 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 20 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; SSE42-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXf32'
 ; XOP-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x float> %src64, <2 x float> %src64_1, <2 x i32> <i32 3, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x float> %src128, <4 x float> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; XOP-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 17 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 27 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXf32'
 ; AVX1-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x float> %src64, <2 x float> %src64_1, <2 x i32> <i32 3, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x float> %src128, <4 x float> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; AVX1-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 17 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 27 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXf32'
 ; AVX2-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x float> %src64, <2 x float> %src64_1, <2 x i32> <i32 3, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x float> %src128, <4 x float> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 13 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 21 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512-LABEL: 'test_vXf32'
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V64 = shufflevector <2 x float> %src64, <2 x float> %src64_1, <2 x i32> <i32 3, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <4 x float> %src128, <4 x float> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:3 Lat:9 SizeLat:3 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:4 Lat:12 SizeLat:4 for: %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V64 = shufflevector <2 x float> %src64, <2 x float> %src64_1, <2 x i32> <i32 3, i32 0>
   %V128 = shufflevector <4 x float> %src128, <4 x float> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-  %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-  %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V256 = shufflevector <8 x float> %src256, <8 x float> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+  %V512 = shufflevector <16 x float> %src512, <16 x float> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+  %V1024 = shufflevector <32 x float> %src1024, <32 x float> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
   ret void
 }
 
 define void @test_vXi32(<2 x i32> %src64, <4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512, <32 x i32> %src1024, <2 x i32> %src64_1, <4 x i32> %src128_1, <8 x i32> %src256_1, <16 x i32> %src512_1, <32 x i32> %src1024_1) {
-; SSE2-LABEL: 'test_vXi32'
-; SSE2-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> %src64_1, <2 x i32> <i32 3, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; SSE2-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of 14 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
-;
-; SSSE3-LABEL: 'test_vXi32'
-; SSSE3-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> %src64_1, <2 x i32> <i32 3, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; SSSE3-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 14 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
-;
-; SSE42-LABEL: 'test_vXi32'
-; SSE42-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> %src64_1, <2 x i32> <i32 3, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; SSE42-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 13 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
+; SSE-LABEL: 'test_vXi32'
+; SSE-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> %src64_1, <2 x i32> <i32 3, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
+; SSE-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of 23 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
+; SSE-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXi32'
 ; XOP-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> %src64_1, <2 x i32> <i32 3, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; XOP-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 17 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 27 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXi32'
 ; AVX1-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> %src64_1, <2 x i32> <i32 3, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; AVX1-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 17 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 27 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXi32'
 ; AVX2-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> %src64_1, <2 x i32> <i32 3, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 13 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 8 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 21 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512-LABEL: 'test_vXi32'
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V64 = shufflevector <2 x i32> %src64, <2 x i32> %src64_1, <2 x i32> <i32 3, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V128 = shufflevector <4 x i32> %src128, <4 x i32> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512-NEXT:  Cost Model: Found costs of RThru:3 CodeSize:3 Lat:9 SizeLat:3 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:1 CodeSize:1 Lat:3 SizeLat:1 for: %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX512-NEXT:  Cost Model: Found costs of RThru:4 CodeSize:4 Lat:12 SizeLat:4 for: %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
 ; AVX512-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V64 = shufflevector <2 x i32> %src64, <2 x i32> %src64_1, <2 x i32> <i32 3, i32 0>
   %V128 = shufflevector <4 x i32> %src128, <4 x i32> %src128_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
-  %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
-  %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 11, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 12, i32 11, i32 11, i32 9, i32 45, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V256 = shufflevector <8 x i32> %src256, <8 x i32> %src256_1, <8 x i32> <i32 11, i32 6, i32 8, i32 4, i32 3, i32 2, i32 12, i32 0>
+  %V512 = shufflevector <16 x i32> %src512, <16 x i32> %src512_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+  %V1024 = shufflevector <32 x i32> %src1024, <32 x i32> %src1024_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 20, i32 53, i32 22, i32 20, i32 19, i32 18, i32 40, i32 16, i32 15, i32 48, i32 13, i32 40, i32 11, i32 11, i32 9, i32 45, i32 33, i32 11, i32 5, i32 4, i32 35, i32 2, i32 33, i32 0>
   ret void
 }
 
@@ -238,89 +222,89 @@ define void @test_vXi16(<2 x i16> %src32, <4 x i16> %src64, <8 x i16> %src128, <
 ; SSE2-NEXT:  Cost Model: Found costs of 2 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
 ; SSE2-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; SSE2-NEXT:  Cost Model: Found costs of RThru:6 CodeSize:8 Lat:8 SizeLat:8 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:9 CodeSize:11 Lat:11 SizeLat:11 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:21 CodeSize:27 Lat:27 SizeLat:27 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:30 CodeSize:36 Lat:36 SizeLat:36 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:18 CodeSize:24 Lat:24 SizeLat:24 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:39 CodeSize:51 Lat:51 SizeLat:51 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:69 CodeSize:91 Lat:91 SizeLat:91 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
 ; SSE2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSSE3-LABEL: 'test_vXi16'
 ; SSSE3-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
 ; SSSE3-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; SSSE3-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 12 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 21 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 37 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
 ; SSSE3-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSE42-LABEL: 'test_vXi16'
 ; SSE42-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
 ; SSE42-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; SSE42-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 10 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 12 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 19 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 33 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
 ; SSE42-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXi16'
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 18 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 26 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 9 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 30 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 54 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXi16'
 ; AVX1-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX1-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 15 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 30 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 42 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 15 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 42 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 90 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXi16'
 ; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 14 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 20 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 17 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 42 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512F-LABEL: 'test_vXi16'
 ; AVX512F-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
 ; AVX512F-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX512F-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 42 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 84 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 42 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 168 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
 ; AVX512F-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512BW-LABEL: 'test_vXi16'
 ; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
 ; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
 ; AVX512BW-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512VBMI-LABEL: 'test_vXi16'
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 4 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 8 for: %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V32 = shufflevector <2 x i16> %src32, <2 x i16> %src32_1, <2 x i32> <i32 3, i32 0>
   %V64 = shufflevector <4 x i16> %src64, <4 x i16> %src64_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
   %V128 = shufflevector <8 x i16> %src128, <8 x i16> %src128_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
-  %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 14, i32 13, i32 20, i32 21, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 38, i32 11, i32 11, i32 9, i32 8, i32 7, i32 11, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 63, i32 62, i32 71, i32 60, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 66, i32 2, i32 1, i32 0>
+  %V256 = shufflevector <16 x i16> %src256, <16 x i16> %src256_1, <16 x i32> <i32 15, i32 17, i32 13, i32 20, i32 29, i32 10, i32 8, i32 8, i32 7, i32 22, i32 5, i32 4, i32 22, i32 2, i32 1, i32 0>
+  %V512 = shufflevector <32 x i16> %src512, <32 x i16> %src512_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+  %V1024 = shufflevector <64 x i16> %src1024, <64 x i16> %src1024_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 41, i32 40, i32 39, i32 39, i32 42, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 0, i32 1, i32 5, i32 6, i32 2, i32 66, i32 2, i32 6, i32 7>
   ret void
 }
 
@@ -330,8 +314,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; SSE2-NEXT:  Cost Model: Found costs of 4 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; SSE2-NEXT:  Cost Model: Found costs of 7 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
 ; SSE2-NEXT:  Cost Model: Found costs of RThru:11 CodeSize:13 Lat:13 SizeLat:13 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:19 CodeSize:23 Lat:23 SizeLat:23 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE2-NEXT:  Cost Model: Found costs of RThru:37 CodeSize:41 Lat:41 SizeLat:41 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:36 CodeSize:42 Lat:42 SizeLat:42 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; SSE2-NEXT:  Cost Model: Found costs of RThru:66 CodeSize:78 Lat:78 SizeLat:78 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
 ; SSE2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSSE3-LABEL: 'test_vXi8'
@@ -339,8 +323,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; SSSE3-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; SSSE3-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
 ; SSSE3-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSSE3-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 12 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; SSSE3-NEXT:  Cost Model: Found costs of 18 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
 ; SSSE3-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; SSE42-LABEL: 'test_vXi8'
@@ -348,8 +332,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; SSE42-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; SSE42-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
 ; SSE42-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 4 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; SSE42-NEXT:  Cost Model: Found costs of 6 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 10 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; SSE42-NEXT:  Cost Model: Found costs of 18 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
 ; SSE42-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; XOP-LABEL: 'test_vXi8'
@@ -357,8 +341,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of 1 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 9 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; XOP-NEXT:  Cost Model: Found costs of 13 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 9 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; XOP-NEXT:  Cost Model: Found costs of 36 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
 ; XOP-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX1-LABEL: 'test_vXi8'
@@ -366,8 +350,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX1-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX1-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 15 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX1-NEXT:  Cost Model: Found costs of 23 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 15 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX1-NEXT:  Cost Model: Found costs of 54 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
 ; AVX1-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX2-LABEL: 'test_vXi8'
@@ -375,8 +359,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX2-NEXT:  Cost Model: Found costs of 11 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX2-NEXT:  Cost Model: Found costs of 23 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
 ; AVX2-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512F-LABEL: 'test_vXi8'
@@ -384,8 +368,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX512F-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX512F-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
 ; AVX512F-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512F-NEXT:  Cost Model: Found costs of 42 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX512F-NEXT:  Cost Model: Found costs of 42 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
 ; AVX512F-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512BW-LABEL: 'test_vXi8'
@@ -393,8 +377,8 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX512BW-NEXT:  Cost Model: Found costs of 3 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX512BW-NEXT:  Cost Model: Found costs of 3 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
 ; AVX512BW-NEXT:  Cost Model: Found costs of 3 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512BW-NEXT:  Cost Model: Found costs of 19 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 7 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX512BW-NEXT:  Cost Model: Found costs of 19 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
 ; AVX512BW-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
 ; AVX512VBMI-LABEL: 'test_vXi8'
@@ -402,15 +386,15 @@ define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+; AVX512VBMI-NEXT:  Cost Model: Found costs of 2 for: %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
 ; AVX512VBMI-NEXT:  Cost Model: Found costs of RThru:0 CodeSize:1 Lat:1 SizeLat:1 for: ret void
 ;
   %V16 = shufflevector <2 x i8> %src16, <2 x i8> %src16_1, <2 x i32> <i32 3, i32 0>
   %V32 = shufflevector <4 x i8> %src32, <4 x i8> %src32_1, <4 x i32> <i32 3, i32 6, i32 1, i32 5>
   %V64 = shufflevector <8 x i8> %src64, <8 x i8> %src64_1, <8 x i32> <i32 7, i32 6, i32 6, i32 8, i32 9, i32 2, i32 1, i32 0>
   %V128 = shufflevector <16 x i8> %src128, <16 x i8> %src128_1, <16 x i32> <i32 29, i32 14, i32 28, i32 12, i32 11, i32 10, i32 11, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 30, i32 45, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 8, i32 8, i32 7, i32 6, i32 8, i32 4, i32 3, i32 2, i32 1, i32 0>
-  %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 63, i32 100, i32 61, i32 96, i32 59, i32 58, i32 57, i32 56, i32 55, i32 54, i32 53, i32 52, i32 51, i32 50, i32 49, i32 48, i32 47, i32 46, i32 45, i32 44, i32 43, i32 42, i32 41, i32 40, i32 39, i32 38, i32 37, i32 36, i32 35, i32 34, i32 33, i32 32, i32 31, i32 30, i32 29, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 20, i32 21, i32 20, i32 19, i32 18, i32 17, i32 16, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  %V256 = shufflevector <32 x i8> %src256, <32 x i8> %src256_1, <32 x i32> <i32 31, i32 33, i32 20, i32 27, i32 28, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 7, i32 19, i32 18, i32 17, i32 16, i32 15, i32 48, i32 13, i32 40, i32 55, i32 11, i32 9, i32 45, i32 4, i32 11, i32 4, i32 5, i32 35, i32 2, i32 33, i32 0>
+  %V512 = shufflevector <64 x i8> %src512, <64 x i8> %src512_1, <64 x i32> <i32 124, i32 62, i32 71, i32 127, i32 58, i32 60, i32 55, i32 56, i32 57, i32 54, i32 127, i32 52, i32 50, i32 51, i32 49, i32 48, i32 47, i32 45, i32 44, i32 44, i32 100, i32 42, i32 41, i32 39, i32 39, i32 40, i32 37, i32 38, i32 35, i32 34, i32 32, i32 33, i32 31, i32 27, i32 30, i32 28, i32 29, i32 26, i32 25, i32 24, i32 23, i32 99, i32 21, i32 20, i32 19, i32 18, i32 17, i32 72, i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 5, i32 6, i32 4, i32 66, i32 2, i32 1, i32 0>
   ret void
 }