From e3f8e44b00ecb95818bc68c693b6637460112b2a Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Mon, 8 Jan 2024 12:35:48 +0000
Subject: [X86] vector-lzcnt-256.ll / vector-tzcnt-256.ll - replace X32 checks
 with X86. NFC.

We try to use X32 for gnux32 triples only.
---
 llvm/test/CodeGen/X86/vector-lzcnt-256.ll | 394 +++++++++++++++---------------
 llvm/test/CodeGen/X86/vector-tzcnt-256.ll | 302 +++++++++++------------
 2 files changed, 348 insertions(+), 348 deletions(-)

diff --git a/llvm/test/CodeGen/X86/vector-lzcnt-256.ll b/llvm/test/CodeGen/X86/vector-lzcnt-256.ll
index 3c53d21..fe6836c 100644
--- a/llvm/test/CodeGen/X86/vector-lzcnt-256.ll
+++ b/llvm/test/CodeGen/X86/vector-lzcnt-256.ll
@@ -7,7 +7,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+avx512cd | FileCheck %s --check-prefixes=X64,AVX512,AVX512CD
 ;
 ; Just one 32-bit run to make sure we do reasonable things for i64 lzcnt.
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X32-AVX
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X86-AVX
 
 define <4 x i64> @testv4i64(<4 x i64> %in) nounwind {
 ; AVX1-LABEL: testv4i64:
@@ -162,34 +162,34 @@ define <4 x i64> @testv4i64(<4 x i64> %in) nounwind {
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512CD-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv4i64:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
-; X32-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
-; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
-; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
-; X32-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
-; X32-AVX-NEXT:    vpsrld $16, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpcmpeqd %ymm4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpsrlq $32, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpsrlq $32, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv4i64:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X86-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X86-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
+; X86-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
+; X86-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
+; X86-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
+; X86-AVX-NEXT:    vpsrld $16, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpcmpeqd %ymm4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpsrlq $32, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpsrlq $32, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    retl
 
   %out = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %in, i1 0)
   ret <4 x i64> %out
@@ -348,34 +348,34 @@ define <4 x i64> @testv4i64u(<4 x i64> %in) nounwind {
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512CD-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv4i64u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
-; X32-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
-; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
-; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
-; X32-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
-; X32-AVX-NEXT:    vpsrld $16, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpcmpeqd %ymm4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpsrlq $32, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpsrlq $32, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv4i64u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X86-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X86-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
+; X86-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
+; X86-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
+; X86-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm2
+; X86-AVX-NEXT:    vpsrld $16, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddd %ymm2, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpcmpeqd %ymm4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpsrlq $32, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpsrlq $32, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddq %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    retl
 
   %out = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %in, i1 -1)
   ret <4 x i64> %out
@@ -509,29 +509,29 @@ define <8 x i32> @testv8i32(<8 x i32> %in) nounwind {
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512CD-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv8i32:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
-; X32-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
-; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
-; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
-; X32-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpsrld $16, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv8i32:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X86-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X86-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
+; X86-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
+; X86-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
+; X86-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpsrld $16, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    retl
 
   %out = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %in, i1 0)
   ret <8 x i32> %out
@@ -665,29 +665,29 @@ define <8 x i32> @testv8i32u(<8 x i32> %in) nounwind {
 ; AVX512CD-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512CD-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv8i32u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
-; X32-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
-; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
-; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
-; X32-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpsrld $16, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv8i32u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X86-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X86-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
+; X86-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
+; X86-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm2
+; X86-AVX-NEXT:    vpsrlw $8, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpand %ymm2, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddw %ymm2, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpcmpeqw %ymm4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpsrld $16, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpsrld $16, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    retl
 
   %out = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %in, i1 -1)
   ret <8 x i32> %out
@@ -792,24 +792,24 @@ define <16 x i16> @testv16i16(<16 x i16> %in) nounwind {
 ; AVX512-NEXT:    vpsubw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX512-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv16i16:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
-; X32-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
-; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
-; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv16i16:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X86-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X86-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
+; X86-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
+; X86-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %in, i1 0)
   ret <16 x i16> %out
 }
@@ -913,24 +913,24 @@ define <16 x i16> @testv16i16u(<16 x i16> %in) nounwind {
 ; AVX512-NEXT:    vpsubw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX512-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv16i16u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
-; X32-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
-; X32-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
-; X32-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
-; X32-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv16i16u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X86-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm3
+; X86-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm3, %ymm3
+; X86-AVX-NEXT:    vpxor %xmm4, %xmm4, %xmm4
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm3, %ymm5
+; X86-AVX-NEXT:    vpand %ymm5, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpshufb %ymm3, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm1, %ymm2, %ymm1
+; X86-AVX-NEXT:    vpcmpeqb %ymm4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpsrlw $8, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddw %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %in, i1 -1)
   ret <16 x i16> %out
 }
@@ -1014,19 +1014,19 @@ define <32 x i8> @testv32i8(<32 x i8> %in) nounwind {
 ; AVX512-NEXT:    vpsubb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX512-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv32i8:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; X32-AVX-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
-; X32-AVX-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv32i8:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X86-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; X86-AVX-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; X86-AVX-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %in, i1 0)
   ret <32 x i8> %out
 }
@@ -1110,19 +1110,19 @@ define <32 x i8> @testv32i8u(<32 x i8> %in) nounwind {
 ; AVX512-NEXT:    vpsubb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
 ; AVX512-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv32i8u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
-; X32-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; X32-AVX-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
-; X32-AVX-NEXT:    vpand %ymm3, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv32i8u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm1 = [4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0,4,3,2,2,1,1,1,1,0,0,0,0,0,0,0,0]
+; X86-AVX-NEXT:    # ymm1 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; X86-AVX-NEXT:    vpcmpeqb %ymm3, %ymm0, %ymm3
+; X86-AVX-NEXT:    vpand %ymm3, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm0, %ymm2, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %in, i1 -1)
   ret <32 x i8> %out
 }
@@ -1133,10 +1133,10 @@ define <4 x i64> @foldv4i64() nounwind {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm0 = [55,0,64,56]
 ; X64-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv4i64:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [55,0,0,0,64,0,56,0]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv4i64:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [55,0,0,0,64,0,56,0]
+; X86-AVX-NEXT:    retl
   %out = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> <i64 256, i64 -1, i64 0, i64 255>, i1 0)
   ret <4 x i64> %out
 }
@@ -1147,10 +1147,10 @@ define <4 x i64> @foldv4i64u() nounwind {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm0 = [55,0,64,56]
 ; X64-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv4i64u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [55,0,0,0,64,0,56,0]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv4i64u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [55,0,0,0,64,0,56,0]
+; X86-AVX-NEXT:    retl
   %out = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> <i64 256, i64 -1, i64 0, i64 255>, i1 -1)
   ret <4 x i64> %out
 }
@@ -1161,10 +1161,10 @@ define <8 x i32> @foldv8i32() nounwind {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
 ; X64-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv8i32:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv8i32:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
+; X86-AVX-NEXT:    retl
   %out = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> <i32 256, i32 -1, i32 0, i32 255, i32 -65536, i32 7, i32 24, i32 88>, i1 0)
   ret <8 x i32> %out
 }
@@ -1175,10 +1175,10 @@ define <8 x i32> @foldv8i32u() nounwind {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
 ; X64-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv8i32u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv8i32u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [23,0,32,24,0,29,27,25]
+; X86-AVX-NEXT:    retl
   %out = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> <i32 256, i32 -1, i32 0, i32 255, i32 -65536, i32 7, i32 24, i32 88>, i1 -1)
   ret <8 x i32> %out
 }
@@ -1189,10 +1189,10 @@ define <16 x i16> @foldv16i16() nounwind {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
 ; X64-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv16i16:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv16i16:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
+; X86-AVX-NEXT:    retl
   %out = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> <i16 256, i16 -1, i16 0, i16 255, i16 -65536, i16 7, i16 24, i16 88, i16 -2, i16 254, i16 1, i16 2, i16 4, i16 8, i16 16, i16 32>, i1 0)
   ret <16 x i16> %out
 }
@@ -1203,10 +1203,10 @@ define <16 x i16> @foldv16i16u() nounwind {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
 ; X64-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv16i16u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv16i16u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [7,0,16,8,16,13,11,9,0,8,15,14,13,12,11,10]
+; X86-AVX-NEXT:    retl
   %out = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> <i16 256, i16 -1, i16 0, i16 255, i16 -65536, i16 7, i16 24, i16 88, i16 -2, i16 254, i16 1, i16 2, i16 4, i16 8, i16 16, i16 32>, i1 -1)
   ret <16 x i16> %out
 }
@@ -1217,10 +1217,10 @@ define <32 x i8> @foldv32i8() nounwind {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
 ; X64-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv32i8:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv32i8:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
+; X86-AVX-NEXT:    retl
   %out = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> <i8 256, i8 -1, i8 0, i8 255, i8 -65536, i8 7, i8 24, i8 88, i8 -2, i8 254, i8 1, i8 2, i8 4, i8 8, i8 16, i8 32, i8 64, i8 128, i8 256, i8 -256, i8 -128, i8 -64, i8 -32, i8 -16, i8 -8, i8 -4, i8 -2, i8 -1, i8 3, i8 5, i8 7, i8 127>, i1 0)
   ret <32 x i8> %out
 }
@@ -1231,10 +1231,10 @@ define <32 x i8> @foldv32i8u() nounwind {
 ; X64-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
 ; X64-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv32i8u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv32i8u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,8,0,8,5,3,1,0,0,7,6,5,4,3,2,1,0,8,8,0,0,0,0,0,0,0,0,6,5,5,1]
+; X86-AVX-NEXT:    retl
   %out = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> <i8 256, i8 -1, i8 0, i8 255, i8 -65536, i8 7, i8 24, i8 88, i8 -2, i8 254, i8 1, i8 2, i8 4, i8 8, i8 16, i8 32, i8 64, i8 128, i8 256, i8 -256, i8 -128, i8 -64, i8 -32, i8 -16, i8 -8, i8 -4, i8 -2, i8 -1, i8 3, i8 5, i8 7, i8 127>, i1 -1)
   ret <32 x i8> %out
 }
diff --git a/llvm/test/CodeGen/X86/vector-tzcnt-256.ll b/llvm/test/CodeGen/X86/vector-tzcnt-256.ll
index cf3803a..5bcdf0e 100644
--- a/llvm/test/CodeGen/X86/vector-tzcnt-256.ll
+++ b/llvm/test/CodeGen/X86/vector-tzcnt-256.ll
@@ -9,7 +9,7 @@
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bitalg,+avx512vl | FileCheck %s --check-prefixes=ALL,BITALG
 ;
 ; Just one 32-bit run to make sure we do reasonable things for i64 tzcnt.
-; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=ALL,X32-AVX
+; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=ALL,X86-AVX
 
 define <4 x i64> @testv4i64(<4 x i64> %in) nounwind {
 ; AVX1-LABEL: testv4i64:
@@ -115,23 +115,23 @@ define <4 x i64> @testv4i64(<4 x i64> %in) nounwind {
 ; BITALG-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv4i64:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddq %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv4i64:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddq %ymm1, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X86-AVX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %in, i1 0)
   ret <4 x i64> %out
 }
@@ -240,23 +240,23 @@ define <4 x i64> @testv4i64u(<4 x i64> %in) nounwind {
 ; BITALG-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv4i64u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddq %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv4i64u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddq %ymm1, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X86-AVX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %in, i1 -1)
   ret <4 x i64> %out
 }
@@ -385,27 +385,27 @@ define <8 x i32> @testv8i32(<8 x i32> %in) nounwind {
 ; BITALG-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv8i32:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddd %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
-; X32-AVX-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
-; X32-AVX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv8i32:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddd %ymm1, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X86-AVX-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; X86-AVX-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
+; X86-AVX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %in, i1 0)
   ret <8 x i32> %out
 }
@@ -534,27 +534,27 @@ define <8 x i32> @testv8i32u(<8 x i32> %in) nounwind {
 ; BITALG-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv8i32u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddd %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; X32-AVX-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
-; X32-AVX-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
-; X32-AVX-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
-; X32-AVX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv8i32u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddd %ymm1, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpxor %xmm1, %xmm1, %xmm1
+; X86-AVX-NEXT:    vpunpckhdq {{.*#+}} ymm2 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
+; X86-AVX-NEXT:    vpsadbw %ymm1, %ymm2, %ymm2
+; X86-AVX-NEXT:    vpunpckldq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
+; X86-AVX-NEXT:    vpsadbw %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpackuswb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %in, i1 -1)
   ret <8 x i32> %out
 }
@@ -685,24 +685,24 @@ define <16 x i16> @testv16i16(<16 x i16> %in) nounwind {
 ; BITALG-NEXT:    vpopcntw %ymm0, %ymm0
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv16i16:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddw %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpsllw $8, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv16i16:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddw %ymm1, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpsllw $8, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %in, i1 0)
   ret <16 x i16> %out
 }
@@ -833,24 +833,24 @@ define <16 x i16> @testv16i16u(<16 x i16> %in) nounwind {
 ; BITALG-NEXT:    vpopcntw %ymm0, %ymm0
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv16i16u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddw %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpsllw $8, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
-; X32-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv16i16u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddw %ymm1, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpsllw $8, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm0, %ymm1, %ymm0
+; X86-AVX-NEXT:    vpsrlw $8, %ymm0, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %in, i1 -1)
   ret <16 x i16> %out
 }
@@ -978,21 +978,21 @@ define <32 x i8> @testv32i8(<32 x i8> %in) nounwind {
 ; BITALG-NEXT:    vpopcntb %ymm0, %ymm0
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv32i8:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv32i8:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm1, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %in, i1 0)
   ret <32 x i8> %out
 }
@@ -1120,21 +1120,21 @@ define <32 x i8> @testv32i8u(<32 x i8> %in) nounwind {
 ; BITALG-NEXT:    vpopcntb %ymm0, %ymm0
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: testv32i8u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
-; X32-AVX-NEXT:    vpaddb %ymm1, %ymm0, %ymm1
-; X32-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
-; X32-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
-; X32-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
-; X32-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
-; X32-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; X32-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
-; X32-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: testv32i8u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vpcmpeqd %ymm1, %ymm1, %ymm1
+; X86-AVX-NEXT:    vpaddb %ymm1, %ymm0, %ymm1
+; X86-AVX-NEXT:    vpandn %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpbroadcastb {{.*#+}} ymm1 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm2
+; X86-AVX-NEXT:    vbroadcasti128 {{.*#+}} ymm3 = [0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4,0,1,1,2,1,2,2,3,1,2,2,3,2,3,3,4]
+; X86-AVX-NEXT:    # ymm3 = mem[0,1,0,1]
+; X86-AVX-NEXT:    vpshufb %ymm2, %ymm3, %ymm2
+; X86-AVX-NEXT:    vpsrlw $4, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
+; X86-AVX-NEXT:    vpshufb %ymm0, %ymm3, %ymm0
+; X86-AVX-NEXT:    vpaddb %ymm2, %ymm0, %ymm0
+; X86-AVX-NEXT:    retl
   %out = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %in, i1 -1)
   ret <32 x i8> %out
 }
@@ -1155,10 +1155,10 @@ define <4 x i64> @foldv4i64() nounwind {
 ; BITALG-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,64,0]
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv4i64:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,0,0,64,0,0,0]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv4i64:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,0,0,64,0,0,0]
+; X86-AVX-NEXT:    retl
   %out = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> <i64 256, i64 -1, i64 0, i64 255>, i1 0)
   ret <4 x i64> %out
 }
@@ -1179,10 +1179,10 @@ define <4 x i64> @foldv4i64u() nounwind {
 ; BITALG-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,64,0]
 ; BITALG-NEXT:    retq
 ;
-; X32-AVX-LABEL: foldv4i64u:
-; X32-AVX:       # %bb.0:
-; X32-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,0,0,64,0,0,0]
-; X32-AVX-NEXT:    retl
+; X86-AVX-LABEL: foldv4i64u:
+; X86-AVX:       # %bb.0:
+; X86-AVX-NEXT:    vmovaps {{.*#+}} ymm0 = [8,0,0,0,64,0,0,0]
+; X86-AVX-NEXT:    retl
   %out = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> <i64 256, i64 -1, i64 0, i64 255>, i1 -1)
   ret <4 x i64> %out
 }
-- 
cgit v1.1