[X86] Regenerate avx2-vector-shifts.ll check prefixes

Add common CHECK prefix

diff --git a/llvm/test/CodeGen/X86/avx2-vector-shifts.ll b/llvm/test/CodeGen/X86/avx2-vector-shifts.ll
index d3a304e..7348a8a 100644 (file)
--- a/llvm/test/CodeGen/X86/avx2-vector-shifts.ll
+++ b/llvm/test/CodeGen/X86/avx2-vector-shifts.ll
@@ -1,79 +1,56 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=X86,X86-SLOW
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X86,X86-FAST-ALL
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X86,X86-FAST-PERLANE
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=X64,X64-SLOW
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST-ALL
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=X64,X64-FAST-PERLANE
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,X86,X86-SLOW
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,X86,X86-FAST-ALL
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,X86,X86-FAST-PERLANE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,X64,X64-SLOW
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,X64,X64-FAST-ALL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck %s --check-prefixes=CHECK,X64,X64-FAST-PERLANE
 
 ; AVX2 Logical Shift Left
 
 define <16 x i16> @test_sllw_1(<16 x i16> %InVec) {
-; X86-LABEL: test_sllw_1:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_sllw_1:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    retq
+; CHECK-LABEL: test_sllw_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = shl <16 x i16> %InVec, <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>
   ret <16 x i16> %shl
 }
 
 define <16 x i16> @test_sllw_2(<16 x i16> %InVec) {
-; X86-LABEL: test_sllw_2:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpaddw %ymm0, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_sllw_2:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpaddw %ymm0, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_sllw_2:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpaddw %ymm0, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = shl <16 x i16> %InVec, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   ret <16 x i16> %shl
 }
 
 define <16 x i16> @test_sllw_3(<16 x i16> %InVec) {
-; X86-LABEL: test_sllw_3:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsllw $15, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_sllw_3:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsllw $15, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_sllw_3:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsllw $15, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = shl <16 x i16> %InVec, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
   ret <16 x i16> %shl
 }
 
 define <8 x i32> @test_slld_1(<8 x i32> %InVec) {
-; X86-LABEL: test_slld_1:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_slld_1:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    retq
+; CHECK-LABEL: test_slld_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = shl <8 x i32> %InVec, <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   ret <8 x i32> %shl
 }
 
 define <8 x i32> @test_slld_2(<8 x i32> %InVec) {
-; X86-LABEL: test_slld_2:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpaddd %ymm0, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_slld_2:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpaddd %ymm0, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_slld_2:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpaddd %ymm0, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = shl <8 x i32> %InVec, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
   ret <8 x i32> %shl
@@ -99,58 +76,39 @@ define <8 x i32> @test_vpslld_var(i32 %shift) {
 }
 
 define <8 x i32> @test_slld_3(<8 x i32> %InVec) {
-; X86-LABEL: test_slld_3:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpslld $31, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_slld_3:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpslld $31, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_slld_3:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpslld $31, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = shl <8 x i32> %InVec, <i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
   ret <8 x i32> %shl
 }
 
 define <4 x i64> @test_sllq_1(<4 x i64> %InVec) {
-; X86-LABEL: test_sllq_1:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_sllq_1:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    retq
+; CHECK-LABEL: test_sllq_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = shl <4 x i64> %InVec, <i64 0, i64 0, i64 0, i64 0>
   ret <4 x i64> %shl
 }
 
 define <4 x i64> @test_sllq_2(<4 x i64> %InVec) {
-; X86-LABEL: test_sllq_2:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpaddq %ymm0, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_sllq_2:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpaddq %ymm0, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_sllq_2:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpaddq %ymm0, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = shl <4 x i64> %InVec, <i64 1, i64 1, i64 1, i64 1>
   ret <4 x i64> %shl
 }
 
 define <4 x i64> @test_sllq_3(<4 x i64> %InVec) {
-; X86-LABEL: test_sllq_3:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsllq $63, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_sllq_3:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsllq $63, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_sllq_3:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsllq $63, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = shl <4 x i64> %InVec, <i64 63, i64 63, i64 63, i64 63>
   ret <4 x i64> %shl
@@ -159,86 +117,58 @@ entry:
 ; AVX2 Arithmetic Shift
 
 define <16 x i16> @test_sraw_1(<16 x i16> %InVec) {
-; X86-LABEL: test_sraw_1:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_sraw_1:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    retq
+; CHECK-LABEL: test_sraw_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = ashr <16 x i16> %InVec, <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>
   ret <16 x i16> %shl
 }
 
 define <16 x i16> @test_sraw_2(<16 x i16> %InVec) {
-; X86-LABEL: test_sraw_2:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsraw $1, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_sraw_2:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsraw $1, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_sraw_2:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsraw $1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = ashr <16 x i16> %InVec, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   ret <16 x i16> %shl
 }
 
 define <16 x i16> @test_sraw_3(<16 x i16> %InVec) {
-; X86-LABEL: test_sraw_3:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsraw $15, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_sraw_3:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsraw $15, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_sraw_3:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsraw $15, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = ashr <16 x i16> %InVec, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
   ret <16 x i16> %shl
 }
 
 define <8 x i32> @test_srad_1(<8 x i32> %InVec) {
-; X86-LABEL: test_srad_1:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srad_1:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srad_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = ashr <8 x i32> %InVec, <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   ret <8 x i32> %shl
 }
 
 define <8 x i32> @test_srad_2(<8 x i32> %InVec) {
-; X86-LABEL: test_srad_2:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsrad $1, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srad_2:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsrad $1, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srad_2:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsrad $1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = ashr <8 x i32> %InVec, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
   ret <8 x i32> %shl
 }
 
 define <8 x i32> @test_srad_3(<8 x i32> %InVec) {
-; X86-LABEL: test_srad_3:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsrad $31, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srad_3:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsrad $31, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srad_3:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsrad $31, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = ashr <8 x i32> %InVec, <i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
   ret <8 x i32> %shl
@@ -247,129 +177,87 @@ entry:
 ; SSE Logical Shift Right
 
 define <16 x i16> @test_srlw_1(<16 x i16> %InVec) {
-; X86-LABEL: test_srlw_1:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srlw_1:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srlw_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = lshr <16 x i16> %InVec, <i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0, i16 0>
   ret <16 x i16> %shl
 }
 
 define <16 x i16> @test_srlw_2(<16 x i16> %InVec) {
-; X86-LABEL: test_srlw_2:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsrlw $1, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srlw_2:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsrlw $1, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srlw_2:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsrlw $1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = lshr <16 x i16> %InVec, <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>
   ret <16 x i16> %shl
 }
 
 define <16 x i16> @test_srlw_3(<16 x i16> %InVec) {
-; X86-LABEL: test_srlw_3:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsrlw $15, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srlw_3:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsrlw $15, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srlw_3:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsrlw $15, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = lshr <16 x i16> %InVec, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
   ret <16 x i16> %shl
 }
 
 define <8 x i32> @test_srld_1(<8 x i32> %InVec) {
-; X86-LABEL: test_srld_1:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srld_1:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srld_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = lshr <8 x i32> %InVec, <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
   ret <8 x i32> %shl
 }
 
 define <8 x i32> @test_srld_2(<8 x i32> %InVec) {
-; X86-LABEL: test_srld_2:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsrld $1, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srld_2:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsrld $1, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srld_2:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsrld $1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = lshr <8 x i32> %InVec, <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
   ret <8 x i32> %shl
 }
 
 define <8 x i32> @test_srld_3(<8 x i32> %InVec) {
-; X86-LABEL: test_srld_3:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsrld $31, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srld_3:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsrld $31, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srld_3:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsrld $31, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = lshr <8 x i32> %InVec, <i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
   ret <8 x i32> %shl
 }
 
 define <4 x i64> @test_srlq_1(<4 x i64> %InVec) {
-; X86-LABEL: test_srlq_1:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srlq_1:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srlq_1:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = lshr <4 x i64> %InVec, <i64 0, i64 0, i64 0, i64 0>
   ret <4 x i64> %shl
 }
 
 define <4 x i64> @test_srlq_2(<4 x i64> %InVec) {
-; X86-LABEL: test_srlq_2:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsrlq $1, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srlq_2:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsrlq $1, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srlq_2:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsrlq $1, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = lshr <4 x i64> %InVec, <i64 1, i64 1, i64 1, i64 1>
   ret <4 x i64> %shl
 }
 
 define <4 x i64> @test_srlq_3(<4 x i64> %InVec) {
-; X86-LABEL: test_srlq_3:
-; X86:       # %bb.0: # %entry
-; X86-NEXT:    vpsrlq $63, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: test_srlq_3:
-; X64:       # %bb.0: # %entry
-; X64-NEXT:    vpsrlq $63, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: test_srlq_3:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vpsrlq $63, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
 entry:
   %shl = lshr <4 x i64> %InVec, <i64 63, i64 63, i64 63, i64 63>
   ret <4 x i64> %shl
@@ -446,59 +334,34 @@ define <4 x i32> @srl_trunc_and_v4i64(<4 x i32> %x, <4 x i64> %y) nounwind {
 ;
 
 define <8 x i16> @shl_8i16(<8 x i16> %r, <8 x i16> %a) nounwind {
-; X86-LABEL: shl_8i16:
-; X86:       # %bb.0:
-; X86-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; X86-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; X86-NEXT:    vpsllvd %ymm1, %ymm0, %ymm0
-; X86-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
-; X86-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; X86-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; X86-NEXT:    vzeroupper
-; X86-NEXT:    retl
-;
-; X64-LABEL: shl_8i16:
-; X64:       # %bb.0:
-; X64-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; X64-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; X64-NEXT:    vpsllvd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
-; X64-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; X64-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
-; X64-NEXT:    vzeroupper
-; X64-NEXT:    retq
+; CHECK-LABEL: shl_8i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; CHECK-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; CHECK-NEXT:    vpsllvd %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u]
+; CHECK-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
+; CHECK-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    ret{{[l|q]}}
   %shl = shl <8 x i16> %r, %a
   ret <8 x i16> %shl
 }
 
 define <16 x i16> @shl_16i16(<16 x i16> %r, <16 x i16> %a) nounwind {
-; X86-LABEL: shl_16i16:
-; X86:       # %bb.0:
-; X86-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X86-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
-; X86-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15]
-; X86-NEXT:    vpsllvd %ymm3, %ymm4, %ymm3
-; X86-NEXT:    vpsrld $16, %ymm3, %ymm3
-; X86-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
-; X86-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11]
-; X86-NEXT:    vpsllvd %ymm1, %ymm0, %ymm0
-; X86-NEXT:    vpsrld $16, %ymm0, %ymm0
-; X86-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: shl_16i16:
-; X64:       # %bb.0:
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X64-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
-; X64-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15]
-; X64-NEXT:    vpsllvd %ymm3, %ymm4, %ymm3
-; X64-NEXT:    vpsrld $16, %ymm3, %ymm3
-; X64-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
-; X64-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11]
-; X64-NEXT:    vpsllvd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpsrld $16, %ymm0, %ymm0
-; X64-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: shl_16i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; CHECK-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
+; CHECK-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15]
+; CHECK-NEXT:    vpsllvd %ymm3, %ymm4, %ymm3
+; CHECK-NEXT:    vpsrld $16, %ymm3, %ymm3
+; CHECK-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
+; CHECK-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11]
+; CHECK-NEXT:    vpsllvd %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vpsrld $16, %ymm0, %ymm0
+; CHECK-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
   %shl = shl <16 x i16> %r, %a
   ret <16 x i16> %shl
 }
@@ -538,173 +401,97 @@ define <32 x i8> @shl_32i8(<32 x i8> %r, <32 x i8> %a) nounwind {
 }
 
 define <8 x i16> @ashr_8i16(<8 x i16> %r, <8 x i16> %a) nounwind {
-; X86-LABEL: ashr_8i16:
-; X86:       # %bb.0:
-; X86-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; X86-NEXT:    vpmovsxwd %xmm0, %ymm0
-; X86-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
-; X86-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; X86-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; X86-NEXT:    vzeroupper
-; X86-NEXT:    retl
-;
-; X64-LABEL: ashr_8i16:
-; X64:       # %bb.0:
-; X64-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; X64-NEXT:    vpmovsxwd %xmm0, %ymm0
-; X64-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; X64-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vzeroupper
-; X64-NEXT:    retq
+; CHECK-LABEL: ashr_8i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; CHECK-NEXT:    vpmovsxwd %xmm0, %ymm0
+; CHECK-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; CHECK-NEXT:    vpackssdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    ret{{[l|q]}}
   %ashr = ashr <8 x i16> %r, %a
   ret <8 x i16> %ashr
 }
 
 define <16 x i16> @ashr_16i16(<16 x i16> %r, <16 x i16> %a) nounwind {
-; X86-LABEL: ashr_16i16:
-; X86:       # %bb.0:
-; X86-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X86-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
-; X86-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15]
-; X86-NEXT:    vpsravd %ymm3, %ymm4, %ymm3
-; X86-NEXT:    vpsrld $16, %ymm3, %ymm3
-; X86-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
-; X86-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11]
-; X86-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
-; X86-NEXT:    vpsrld $16, %ymm0, %ymm0
-; X86-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: ashr_16i16:
-; X64:       # %bb.0:
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X64-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
-; X64-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15]
-; X64-NEXT:    vpsravd %ymm3, %ymm4, %ymm3
-; X64-NEXT:    vpsrld $16, %ymm3, %ymm3
-; X64-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
-; X64-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11]
-; X64-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpsrld $16, %ymm0, %ymm0
-; X64-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: ashr_16i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; CHECK-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
+; CHECK-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15]
+; CHECK-NEXT:    vpsravd %ymm3, %ymm4, %ymm3
+; CHECK-NEXT:    vpsrld $16, %ymm3, %ymm3
+; CHECK-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
+; CHECK-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11]
+; CHECK-NEXT:    vpsravd %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vpsrld $16, %ymm0, %ymm0
+; CHECK-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
   %ashr = ashr <16 x i16> %r, %a
   ret <16 x i16> %ashr
 }
 
 define <32 x i8> @ashr_32i8(<32 x i8> %r, <32 x i8> %a) nounwind {
-; X86-LABEL: ashr_32i8:
-; X86:       # %bb.0:
-; X86-NEXT:    vpsllw $5, %ymm1, %ymm1
-; X86-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; X86-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; X86-NEXT:    vpsraw $4, %ymm3, %ymm4
-; X86-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; X86-NEXT:    vpsraw $2, %ymm3, %ymm4
-; X86-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; X86-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; X86-NEXT:    vpsraw $1, %ymm3, %ymm4
-; X86-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; X86-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm2
-; X86-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; X86-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; X86-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; X86-NEXT:    vpsraw $4, %ymm0, %ymm3
-; X86-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; X86-NEXT:    vpsraw $2, %ymm0, %ymm3
-; X86-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; X86-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; X86-NEXT:    vpsraw $1, %ymm0, %ymm3
-; X86-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; X86-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; X86-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; X86-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: ashr_32i8:
-; X64:       # %bb.0:
-; X64-NEXT:    vpsllw $5, %ymm1, %ymm1
-; X64-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; X64-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
-; X64-NEXT:    vpsraw $4, %ymm3, %ymm4
-; X64-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; X64-NEXT:    vpsraw $2, %ymm3, %ymm4
-; X64-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; X64-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
-; X64-NEXT:    vpsraw $1, %ymm3, %ymm4
-; X64-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
-; X64-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm2
-; X64-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; X64-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; X64-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
-; X64-NEXT:    vpsraw $4, %ymm0, %ymm3
-; X64-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; X64-NEXT:    vpsraw $2, %ymm0, %ymm3
-; X64-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; X64-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; X64-NEXT:    vpsraw $1, %ymm0, %ymm3
-; X64-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
-; X64-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
-; X64-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; X64-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: ashr_32i8:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpsllw $5, %ymm1, %ymm1
+; CHECK-NEXT:    vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; CHECK-NEXT:    vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
+; CHECK-NEXT:    vpsraw $4, %ymm3, %ymm4
+; CHECK-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
+; CHECK-NEXT:    vpsraw $2, %ymm3, %ymm4
+; CHECK-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
+; CHECK-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm3
+; CHECK-NEXT:    vpsraw $1, %ymm3, %ymm4
+; CHECK-NEXT:    vpaddw %ymm2, %ymm2, %ymm2
+; CHECK-NEXT:    vpblendvb %ymm2, %ymm4, %ymm3, %ymm2
+; CHECK-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; CHECK-NEXT:    vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
+; CHECK-NEXT:    vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
+; CHECK-NEXT:    vpsraw $4, %ymm0, %ymm3
+; CHECK-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
+; CHECK-NEXT:    vpsraw $2, %ymm0, %ymm3
+; CHECK-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
+; CHECK-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
+; CHECK-NEXT:    vpsraw $1, %ymm0, %ymm3
+; CHECK-NEXT:    vpaddw %ymm1, %ymm1, %ymm1
+; CHECK-NEXT:    vpblendvb %ymm1, %ymm3, %ymm0, %ymm0
+; CHECK-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; CHECK-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
   %ashr = ashr <32 x i8> %r, %a
   ret <32 x i8> %ashr
 }
 
 define <8 x i16> @lshr_8i16(<8 x i16> %r, <8 x i16> %a) nounwind {
-; X86-LABEL: lshr_8i16:
-; X86:       # %bb.0:
-; X86-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; X86-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; X86-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
-; X86-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; X86-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
-; X86-NEXT:    vzeroupper
-; X86-NEXT:    retl
-;
-; X64-LABEL: lshr_8i16:
-; X64:       # %bb.0:
-; X64-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
-; X64-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
-; X64-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; X64-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
-; X64-NEXT:    vzeroupper
-; X64-NEXT:    retq
+; CHECK-LABEL: lshr_8i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; CHECK-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; CHECK-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; CHECK-NEXT:    vpackusdw %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    vzeroupper
+; CHECK-NEXT:    ret{{[l|q]}}
   %lshr = lshr <8 x i16> %r, %a
   ret <8 x i16> %lshr
 }
 
 define <16 x i16> @lshr_16i16(<16 x i16> %r, <16 x i16> %a) nounwind {
-; X86-LABEL: lshr_16i16:
-; X86:       # %bb.0:
-; X86-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X86-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
-; X86-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15]
-; X86-NEXT:    vpsrlvd %ymm3, %ymm4, %ymm3
-; X86-NEXT:    vpsrld $16, %ymm3, %ymm3
-; X86-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
-; X86-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11]
-; X86-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
-; X86-NEXT:    vpsrld $16, %ymm0, %ymm0
-; X86-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
-; X86-NEXT:    retl
-;
-; X64-LABEL: lshr_16i16:
-; X64:       # %bb.0:
-; X64-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; X64-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
-; X64-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15]
-; X64-NEXT:    vpsrlvd %ymm3, %ymm4, %ymm3
-; X64-NEXT:    vpsrld $16, %ymm3, %ymm3
-; X64-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
-; X64-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11]
-; X64-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
-; X64-NEXT:    vpsrld $16, %ymm0, %ymm0
-; X64-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
-; X64-NEXT:    retq
+; CHECK-LABEL: lshr_16i16:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpxor %xmm2, %xmm2, %xmm2
+; CHECK-NEXT:    vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15]
+; CHECK-NEXT:    vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15]
+; CHECK-NEXT:    vpsrlvd %ymm3, %ymm4, %ymm3
+; CHECK-NEXT:    vpsrld $16, %ymm3, %ymm3
+; CHECK-NEXT:    vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11]
+; CHECK-NEXT:    vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11]
+; CHECK-NEXT:    vpsrlvd %ymm1, %ymm0, %ymm0
+; CHECK-NEXT:    vpsrld $16, %ymm0, %ymm0
+; CHECK-NEXT:    vpackusdw %ymm3, %ymm0, %ymm0
+; CHECK-NEXT:    ret{{[l|q]}}
   %lshr = lshr <16 x i16> %r, %a
   ret <16 x i16> %lshr
 }