[X86] Add some non-AVX512VL command lines to the *vl-vec-test-testn.ll tests.

This will expose some regressions in the WIP and/or/xor promotion removal patch.

llvm-svn: 345297

2 files changed, 762 insertions, 357 deletions

diff --git a/llvm/test/CodeGen/X86/avx512bwvl-vec-test-testn.ll b/llvm/test/CodeGen/X86/avx512bwvl-vec-test-testn.ll
index 70cadc78c18..ecb76b3f9a7 100644
--- a/llvm/test/CodeGen/X86/avx512bwvl-vec-test-testn.ll
+++ b/llvm/test/CodeGen/X86/avx512bwvl-vec-test-testn.ll
@@ -1,14 +1,25 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefix=AVX512BWVL
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefix=AVX512BW
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i16 @TEST_mm_test_epi8_mask(<2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm_test_epi8_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestmb %xmm0, %xmm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm_test_epi8_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestmb %xmm0, %xmm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm_test_epi8_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512BW-NEXT:    vptestmb %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <16 x i8>
@@ -19,13 +30,24 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i16 @TEST_mm_mask_test_epi8_mask(i16 zeroext %__U, <2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm_mask_test_epi8_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestmb %xmm0, %xmm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    andl %edi, %eax
-; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm_mask_test_epi8_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestmb %xmm0, %xmm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    andl %edi, %eax
+; AVX512BWVL-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm_mask_test_epi8_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512BW-NEXT:    vptestmb %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    andl %edi, %eax
+; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <16 x i8>
@@ -38,12 +60,22 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_test_epi16_mask(<2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm_test_epi16_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestmw %xmm0, %xmm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    # kill: def $al killed $al killed $eax
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm_test_epi16_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestmw %xmm0, %xmm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm_test_epi16_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512BW-NEXT:    vptestmw %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <8 x i16>
@@ -54,13 +86,24 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_mask_test_epi16_mask(i8 zeroext %__U, <2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm_mask_test_epi16_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestmw %xmm0, %xmm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    andb %dil, %al
-; CHECK-NEXT:    # kill: def $al killed $al killed $eax
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm_mask_test_epi16_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestmw %xmm0, %xmm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    andb %dil, %al
+; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm_mask_test_epi16_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512BW-NEXT:    vptestmw %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    andb %dil, %al
+; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <8 x i16>
@@ -73,12 +116,22 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i16 @TEST_mm_testn_epi8_mask(<2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm_testn_epi8_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestnmb %xmm0, %xmm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm_testn_epi8_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestnmb %xmm0, %xmm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm_testn_epi8_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <16 x i8>
@@ -89,13 +142,24 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i16 @TEST_mm_mask_testn_epi8_mask(i16 zeroext %__U, <2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm_mask_testn_epi8_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestnmb %xmm0, %xmm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    andl %edi, %eax
-; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm_mask_testn_epi8_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestnmb %xmm0, %xmm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    andl %edi, %eax
+; AVX512BWVL-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm_mask_testn_epi8_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    andl %edi, %eax
+; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <16 x i8>
@@ -108,12 +172,22 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_testn_epi16_mask(<2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm_testn_epi16_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestnmw %xmm0, %xmm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    # kill: def $al killed $al killed $eax
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm_testn_epi16_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestnmw %xmm0, %xmm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm_testn_epi16_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512BW-NEXT:    vptestnmw %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <8 x i16>
@@ -124,13 +198,24 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_mask_testn_epi16_mask(i8 zeroext %__U, <2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm_mask_testn_epi16_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestnmw %xmm0, %xmm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    andb %dil, %al
-; CHECK-NEXT:    # kill: def $al killed $al killed $eax
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm_mask_testn_epi16_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestnmw %xmm0, %xmm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    andb %dil, %al
+; AVX512BWVL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm_mask_testn_epi16_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512BW-NEXT:    vptestnmw %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    andb %dil, %al
+; AVX512BW-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <8 x i16>
@@ -143,12 +228,21 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define i32 @TEST_mm256_test_epi8_mask(<4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm256_test_epi8_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestmb %ymm0, %ymm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm256_test_epi8_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestmb %ymm0, %ymm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    vzeroupper
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm256_test_epi8_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vptestmb %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <32 x i8>
@@ -159,13 +253,23 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define i32 @TEST_mm256_mask_test_epi8_mask(i32 %__U, <4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm256_mask_test_epi8_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestmb %ymm0, %ymm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    andl %edi, %eax
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm256_mask_test_epi8_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestmb %ymm0, %ymm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    andl %edi, %eax
+; AVX512BWVL-NEXT:    vzeroupper
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm256_mask_test_epi8_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vptestmb %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    andl %edi, %eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <32 x i8>
@@ -178,13 +282,23 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i16 @TEST_mm256_test_epi16_mask(<4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm256_test_epi16_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestmw %ymm0, %ymm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm256_test_epi16_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestmw %ymm0, %ymm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BWVL-NEXT:    vzeroupper
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm256_test_epi16_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vptestmw %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <16 x i16>
@@ -195,14 +309,25 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i16 @TEST_mm256_mask_test_epi16_mask(i16 zeroext %__U, <4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm256_mask_test_epi16_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestmw %ymm0, %ymm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    andl %edi, %eax
-; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm256_mask_test_epi16_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestmw %ymm0, %ymm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    andl %edi, %eax
+; AVX512BWVL-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BWVL-NEXT:    vzeroupper
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm256_mask_test_epi16_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vptestmw %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    andl %edi, %eax
+; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <16 x i16>
@@ -215,12 +340,21 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define i32 @TEST_mm256_testn_epi8_mask(<4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm256_testn_epi8_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestnmb %ymm0, %ymm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm256_testn_epi8_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestnmb %ymm0, %ymm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    vzeroupper
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm256_testn_epi8_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <32 x i8>
@@ -231,13 +365,23 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define i32 @TEST_mm256_mask_testn_epi8_mask(i32 %__U, <4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm256_mask_testn_epi8_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestnmb %ymm0, %ymm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    andl %edi, %eax
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm256_mask_testn_epi8_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestnmb %ymm0, %ymm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    andl %edi, %eax
+; AVX512BWVL-NEXT:    vzeroupper
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm256_mask_testn_epi8_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vptestnmb %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    andl %edi, %eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <32 x i8>
@@ -250,13 +394,23 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i16 @TEST_mm256_testn_epi16_mask(<4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm256_testn_epi16_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestnmw %ymm0, %ymm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm256_testn_epi16_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestnmw %ymm0, %ymm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BWVL-NEXT:    vzeroupper
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm256_testn_epi16_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vptestnmw %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <16 x i16>
@@ -267,14 +421,25 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i16 @TEST_mm256_mask_testn_epi16_mask(i16 zeroext %__U, <4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; CHECK-LABEL: TEST_mm256_mask_testn_epi16_mask:
-; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    vptestnmw %ymm0, %ymm1, %k0
-; CHECK-NEXT:    kmovd %k0, %eax
-; CHECK-NEXT:    andl %edi, %eax
-; CHECK-NEXT:    # kill: def $ax killed $ax killed $eax
-; CHECK-NEXT:    vzeroupper
-; CHECK-NEXT:    retq
+; AVX512BWVL-LABEL: TEST_mm256_mask_testn_epi16_mask:
+; AVX512BWVL:       # %bb.0: # %entry
+; AVX512BWVL-NEXT:    vptestnmw %ymm0, %ymm1, %k0
+; AVX512BWVL-NEXT:    kmovd %k0, %eax
+; AVX512BWVL-NEXT:    andl %edi, %eax
+; AVX512BWVL-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BWVL-NEXT:    vzeroupper
+; AVX512BWVL-NEXT:    retq
+;
+; AVX512BW-LABEL: TEST_mm256_mask_testn_epi16_mask:
+; AVX512BW:       # %bb.0: # %entry
+; AVX512BW-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512BW-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512BW-NEXT:    vptestnmw %zmm0, %zmm1, %k0
+; AVX512BW-NEXT:    kmovd %k0, %eax
+; AVX512BW-NEXT:    andl %edi, %eax
+; AVX512BW-NEXT:    # kill: def $ax killed $ax killed $eax
+; AVX512BW-NEXT:    vzeroupper
+; AVX512BW-NEXT:    retq
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <16 x i16>
diff --git a/llvm/test/CodeGen/X86/avx512vl-vec-test-testn.ll b/llvm/test/CodeGen/X86/avx512vl-vec-test-testn.ll
index c662226fde9..ae74be241d5 100644
--- a/llvm/test/CodeGen/X86/avx512vl-vec-test-testn.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-vec-test-testn.ll
@@ -1,22 +1,29 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=X86_64
-; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=I386
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=AVX512VL --check-prefix=AVX512VL-X64
+; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+avx512vl | FileCheck %s --check-prefix=AVX512VL --check-prefix=AVX512VL-X86
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512F --check-prefix=AVX512F-X64
+; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefix=AVX512F --check-prefix=AVX512F-X86
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_test_epi64_mask(<2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm_test_epi64_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestmq %xmm0, %xmm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm_test_epi64_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestmq %xmm0, %xmm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    retl
+; AVX512VL-LABEL: TEST_mm_test_epi64_mask:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vptestmq %xmm0, %xmm1, %k0
+; AVX512VL-NEXT:    kmovw %k0, %eax
+; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    ret{{[l|q]}}
+;
+; AVX512F-LABEL: TEST_mm_test_epi64_mask:
+; AVX512F:       # %bb.0: # %entry
+; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vptestmq %zmm0, %zmm1, %k0
+; AVX512F-NEXT:    kshiftlw $14, %k0, %k0
+; AVX512F-NEXT:    kshiftrw $14, %k0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    ret{{[l|q]}}
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = icmp ne <2 x i64> %and.i.i, zeroinitializer
@@ -27,19 +34,24 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_test_epi32_mask(<2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm_test_epi32_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestmd %xmm0, %xmm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm_test_epi32_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestmd %xmm0, %xmm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    retl
+; AVX512VL-LABEL: TEST_mm_test_epi32_mask:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vptestmd %xmm0, %xmm1, %k0
+; AVX512VL-NEXT:    kmovw %k0, %eax
+; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    ret{{[l|q]}}
+;
+; AVX512F-LABEL: TEST_mm_test_epi32_mask:
+; AVX512F:       # %bb.0: # %entry
+; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm1, %k0
+; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    ret{{[l|q]}}
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <4 x i32>
@@ -51,21 +63,25 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm256_test_epi64_mask(<4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm256_test_epi64_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestmq %ymm0, %ymm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    vzeroupper
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm256_test_epi64_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestmq %ymm0, %ymm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    vzeroupper
-; I386-NEXT:    retl
+; AVX512VL-LABEL: TEST_mm256_test_epi64_mask:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vptestmq %ymm0, %ymm1, %k0
+; AVX512VL-NEXT:    kmovw %k0, %eax
+; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    ret{{[l|q]}}
+;
+; AVX512F-LABEL: TEST_mm256_test_epi64_mask:
+; AVX512F:       # %bb.0: # %entry
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vptestmq %zmm0, %zmm1, %k0
+; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    ret{{[l|q]}}
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = icmp ne <4 x i64> %and.i.i, zeroinitializer
@@ -76,21 +92,23 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm256_test_epi32_mask(<4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm256_test_epi32_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestmd %ymm0, %ymm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    vzeroupper
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm256_test_epi32_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestmd %ymm0, %ymm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    vzeroupper
-; I386-NEXT:    retl
+; AVX512VL-LABEL: TEST_mm256_test_epi32_mask:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vptestmd %ymm0, %ymm1, %k0
+; AVX512VL-NEXT:    kmovw %k0, %eax
+; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    ret{{[l|q]}}
+;
+; AVX512F-LABEL: TEST_mm256_test_epi32_mask:
+; AVX512F:       # %bb.0: # %entry
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vptestmd %zmm0, %zmm1, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    ret{{[l|q]}}
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <8 x i32>
@@ -101,22 +119,49 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_mask_test_epi64_mask(i8 %__U, <2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm_mask_test_epi64_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    kmovw %edi, %k1
-; X86_64-NEXT:    vptestmq %xmm0, %xmm1, %k0 {%k1}
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm_mask_test_epi64_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; I386-NEXT:    kmovw %eax, %k1
-; I386-NEXT:    vptestmq %xmm0, %xmm1, %k0 {%k1}
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    retl
+; AVX512VL-X64-LABEL: TEST_mm_mask_test_epi64_mask:
+; AVX512VL-X64:       # %bb.0: # %entry
+; AVX512VL-X64-NEXT:    kmovw %edi, %k1
+; AVX512VL-X64-NEXT:    vptestmq %xmm0, %xmm1, %k0 {%k1}
+; AVX512VL-X64-NEXT:    kmovw %k0, %eax
+; AVX512VL-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X64-NEXT:    retq
+;
+; AVX512VL-X86-LABEL: TEST_mm_mask_test_epi64_mask:
+; AVX512VL-X86:       # %bb.0: # %entry
+; AVX512VL-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512VL-X86-NEXT:    kmovw %eax, %k1
+; AVX512VL-X86-NEXT:    vptestmq %xmm0, %xmm1, %k0 {%k1}
+; AVX512VL-X86-NEXT:    kmovw %k0, %eax
+; AVX512VL-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X86-NEXT:    retl
+;
+; AVX512F-X64-LABEL: TEST_mm_mask_test_epi64_mask:
+; AVX512F-X64:       # %bb.0: # %entry
+; AVX512F-X64-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-X64-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-X64-NEXT:    kmovw %edi, %k1
+; AVX512F-X64-NEXT:    vptestmq %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X64-NEXT:    kshiftlw $14, %k0, %k0
+; AVX512F-X64-NEXT:    kshiftrw $14, %k0, %k0
+; AVX512F-X64-NEXT:    kmovw %k0, %eax
+; AVX512F-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X64-NEXT:    vzeroupper
+; AVX512F-X64-NEXT:    retq
+;
+; AVX512F-X86-LABEL: TEST_mm_mask_test_epi64_mask:
+; AVX512F-X86:       # %bb.0: # %entry
+; AVX512F-X86-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512F-X86-NEXT:    kmovw %eax, %k1
+; AVX512F-X86-NEXT:    vptestmq %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X86-NEXT:    kshiftlw $14, %k0, %k0
+; AVX512F-X86-NEXT:    kshiftrw $14, %k0, %k0
+; AVX512F-X86-NEXT:    kmovw %k0, %eax
+; AVX512F-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X86-NEXT:    vzeroupper
+; AVX512F-X86-NEXT:    retl
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = icmp ne <2 x i64> %and.i.i, zeroinitializer
@@ -130,22 +175,49 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_mask_test_epi32_mask(i8 %__U, <2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm_mask_test_epi32_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    kmovw %edi, %k1
-; X86_64-NEXT:    vptestmd %xmm0, %xmm1, %k0 {%k1}
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm_mask_test_epi32_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; I386-NEXT:    kmovw %eax, %k1
-; I386-NEXT:    vptestmd %xmm0, %xmm1, %k0 {%k1}
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    retl
+; AVX512VL-X64-LABEL: TEST_mm_mask_test_epi32_mask:
+; AVX512VL-X64:       # %bb.0: # %entry
+; AVX512VL-X64-NEXT:    kmovw %edi, %k1
+; AVX512VL-X64-NEXT:    vptestmd %xmm0, %xmm1, %k0 {%k1}
+; AVX512VL-X64-NEXT:    kmovw %k0, %eax
+; AVX512VL-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X64-NEXT:    retq
+;
+; AVX512VL-X86-LABEL: TEST_mm_mask_test_epi32_mask:
+; AVX512VL-X86:       # %bb.0: # %entry
+; AVX512VL-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512VL-X86-NEXT:    kmovw %eax, %k1
+; AVX512VL-X86-NEXT:    vptestmd %xmm0, %xmm1, %k0 {%k1}
+; AVX512VL-X86-NEXT:    kmovw %k0, %eax
+; AVX512VL-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X86-NEXT:    retl
+;
+; AVX512F-X64-LABEL: TEST_mm_mask_test_epi32_mask:
+; AVX512F-X64:       # %bb.0: # %entry
+; AVX512F-X64-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-X64-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-X64-NEXT:    kmovw %edi, %k1
+; AVX512F-X64-NEXT:    vptestmd %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X64-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-X64-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-X64-NEXT:    kmovw %k0, %eax
+; AVX512F-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X64-NEXT:    vzeroupper
+; AVX512F-X64-NEXT:    retq
+;
+; AVX512F-X86-LABEL: TEST_mm_mask_test_epi32_mask:
+; AVX512F-X86:       # %bb.0: # %entry
+; AVX512F-X86-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512F-X86-NEXT:    kmovw %eax, %k1
+; AVX512F-X86-NEXT:    vptestmd %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X86-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-X86-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-X86-NEXT:    kmovw %k0, %eax
+; AVX512F-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X86-NEXT:    vzeroupper
+; AVX512F-X86-NEXT:    retl
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <4 x i32>
@@ -161,24 +233,51 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm256_mask_test_epi64_mask(i8 %__U, <4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm256_mask_test_epi64_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    kmovw %edi, %k1
-; X86_64-NEXT:    vptestmq %ymm0, %ymm1, %k0 {%k1}
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    vzeroupper
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm256_mask_test_epi64_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; I386-NEXT:    kmovw %eax, %k1
-; I386-NEXT:    vptestmq %ymm0, %ymm1, %k0 {%k1}
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    vzeroupper
-; I386-NEXT:    retl
+; AVX512VL-X64-LABEL: TEST_mm256_mask_test_epi64_mask:
+; AVX512VL-X64:       # %bb.0: # %entry
+; AVX512VL-X64-NEXT:    kmovw %edi, %k1
+; AVX512VL-X64-NEXT:    vptestmq %ymm0, %ymm1, %k0 {%k1}
+; AVX512VL-X64-NEXT:    kmovw %k0, %eax
+; AVX512VL-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X64-NEXT:    vzeroupper
+; AVX512VL-X64-NEXT:    retq
+;
+; AVX512VL-X86-LABEL: TEST_mm256_mask_test_epi64_mask:
+; AVX512VL-X86:       # %bb.0: # %entry
+; AVX512VL-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512VL-X86-NEXT:    kmovw %eax, %k1
+; AVX512VL-X86-NEXT:    vptestmq %ymm0, %ymm1, %k0 {%k1}
+; AVX512VL-X86-NEXT:    kmovw %k0, %eax
+; AVX512VL-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X86-NEXT:    vzeroupper
+; AVX512VL-X86-NEXT:    retl
+;
+; AVX512F-X64-LABEL: TEST_mm256_mask_test_epi64_mask:
+; AVX512F-X64:       # %bb.0: # %entry
+; AVX512F-X64-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-X64-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-X64-NEXT:    kmovw %edi, %k1
+; AVX512F-X64-NEXT:    vptestmq %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X64-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-X64-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-X64-NEXT:    kmovw %k0, %eax
+; AVX512F-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X64-NEXT:    vzeroupper
+; AVX512F-X64-NEXT:    retq
+;
+; AVX512F-X86-LABEL: TEST_mm256_mask_test_epi64_mask:
+; AVX512F-X86:       # %bb.0: # %entry
+; AVX512F-X86-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-X86-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512F-X86-NEXT:    kmovw %eax, %k1
+; AVX512F-X86-NEXT:    vptestmq %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X86-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-X86-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-X86-NEXT:    kmovw %k0, %eax
+; AVX512F-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X86-NEXT:    vzeroupper
+; AVX512F-X86-NEXT:    retl
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = icmp ne <4 x i64> %and.i.i, zeroinitializer
@@ -192,23 +291,45 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm256_mask_test_epi32_mask(i8 %__U, <4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm256_mask_test_epi32_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestmd %ymm0, %ymm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    andb %dil, %al
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    vzeroupper
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm256_mask_test_epi32_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestmd %ymm0, %ymm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    andb {{[0-9]+}}(%esp), %al
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    vzeroupper
-; I386-NEXT:    retl
+; AVX512VL-X64-LABEL: TEST_mm256_mask_test_epi32_mask:
+; AVX512VL-X64:       # %bb.0: # %entry
+; AVX512VL-X64-NEXT:    vptestmd %ymm0, %ymm1, %k0
+; AVX512VL-X64-NEXT:    kmovw %k0, %eax
+; AVX512VL-X64-NEXT:    andb %dil, %al
+; AVX512VL-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X64-NEXT:    vzeroupper
+; AVX512VL-X64-NEXT:    retq
+;
+; AVX512VL-X86-LABEL: TEST_mm256_mask_test_epi32_mask:
+; AVX512VL-X86:       # %bb.0: # %entry
+; AVX512VL-X86-NEXT:    vptestmd %ymm0, %ymm1, %k0
+; AVX512VL-X86-NEXT:    kmovw %k0, %eax
+; AVX512VL-X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; AVX512VL-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X86-NEXT:    vzeroupper
+; AVX512VL-X86-NEXT:    retl
+;
+; AVX512F-X64-LABEL: TEST_mm256_mask_test_epi32_mask:
+; AVX512F-X64:       # %bb.0: # %entry
+; AVX512F-X64-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-X64-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-X64-NEXT:    vptestmd %zmm0, %zmm1, %k0
+; AVX512F-X64-NEXT:    kmovw %k0, %eax
+; AVX512F-X64-NEXT:    andb %dil, %al
+; AVX512F-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X64-NEXT:    vzeroupper
+; AVX512F-X64-NEXT:    retq
+;
+; AVX512F-X86-LABEL: TEST_mm256_mask_test_epi32_mask:
+; AVX512F-X86:       # %bb.0: # %entry
+; AVX512F-X86-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-X86-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-X86-NEXT:    vptestmd %zmm0, %zmm1, %k0
+; AVX512F-X86-NEXT:    kmovw %k0, %eax
+; AVX512F-X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; AVX512F-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X86-NEXT:    vzeroupper
+; AVX512F-X86-NEXT:    retl
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <8 x i32>
@@ -221,19 +342,24 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_testn_epi64_mask(<2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm_testn_epi64_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestnmq %xmm0, %xmm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm_testn_epi64_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestnmq %xmm0, %xmm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    retl
+; AVX512VL-LABEL: TEST_mm_testn_epi64_mask:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vptestnmq %xmm0, %xmm1, %k0
+; AVX512VL-NEXT:    kmovw %k0, %eax
+; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    ret{{[l|q]}}
+;
+; AVX512F-LABEL: TEST_mm_testn_epi64_mask:
+; AVX512F:       # %bb.0: # %entry
+; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vptestnmq %zmm0, %zmm1, %k0
+; AVX512F-NEXT:    kshiftlw $14, %k0, %k0
+; AVX512F-NEXT:    kshiftrw $14, %k0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    ret{{[l|q]}}
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = icmp eq <2 x i64> %and.i.i, zeroinitializer
@@ -244,19 +370,24 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_testn_epi32_mask(<2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm_testn_epi32_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestnmd %xmm0, %xmm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm_testn_epi32_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestnmd %xmm0, %xmm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    retl
+; AVX512VL-LABEL: TEST_mm_testn_epi32_mask:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vptestnmd %xmm0, %xmm1, %k0
+; AVX512VL-NEXT:    kmovw %k0, %eax
+; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    ret{{[l|q]}}
+;
+; AVX512F-LABEL: TEST_mm_testn_epi32_mask:
+; AVX512F:       # %bb.0: # %entry
+; AVX512F-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vptestnmd %zmm0, %zmm1, %k0
+; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    ret{{[l|q]}}
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <4 x i32>
@@ -268,21 +399,25 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm256_testn_epi64_mask(<4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm256_testn_epi64_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestnmq %ymm0, %ymm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    vzeroupper
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm256_testn_epi64_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestnmq %ymm0, %ymm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    vzeroupper
-; I386-NEXT:    retl
+; AVX512VL-LABEL: TEST_mm256_testn_epi64_mask:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vptestnmq %ymm0, %ymm1, %k0
+; AVX512VL-NEXT:    kmovw %k0, %eax
+; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    ret{{[l|q]}}
+;
+; AVX512F-LABEL: TEST_mm256_testn_epi64_mask:
+; AVX512F:       # %bb.0: # %entry
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vptestnmq %zmm0, %zmm1, %k0
+; AVX512F-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    ret{{[l|q]}}
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = icmp eq <4 x i64> %and.i.i, zeroinitializer
@@ -293,21 +428,23 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm256_testn_epi32_mask(<4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm256_testn_epi32_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestnmd %ymm0, %ymm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    vzeroupper
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm256_testn_epi32_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestnmd %ymm0, %ymm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    vzeroupper
-; I386-NEXT:    retl
+; AVX512VL-LABEL: TEST_mm256_testn_epi32_mask:
+; AVX512VL:       # %bb.0: # %entry
+; AVX512VL-NEXT:    vptestnmd %ymm0, %ymm1, %k0
+; AVX512VL-NEXT:    kmovw %k0, %eax
+; AVX512VL-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-NEXT:    vzeroupper
+; AVX512VL-NEXT:    ret{{[l|q]}}
+;
+; AVX512F-LABEL: TEST_mm256_testn_epi32_mask:
+; AVX512F:       # %bb.0: # %entry
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vptestnmd %zmm0, %zmm1, %k0
+; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-NEXT:    vzeroupper
+; AVX512F-NEXT:    ret{{[l|q]}}
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <8 x i32>
@@ -318,22 +455,49 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_mask_testn_epi64_mask(i8 %__U, <2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm_mask_testn_epi64_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    kmovw %edi, %k1
-; X86_64-NEXT:    vptestnmq %xmm0, %xmm1, %k0 {%k1}
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm_mask_testn_epi64_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; I386-NEXT:    kmovw %eax, %k1
-; I386-NEXT:    vptestnmq %xmm0, %xmm1, %k0 {%k1}
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    retl
+; AVX512VL-X64-LABEL: TEST_mm_mask_testn_epi64_mask:
+; AVX512VL-X64:       # %bb.0: # %entry
+; AVX512VL-X64-NEXT:    kmovw %edi, %k1
+; AVX512VL-X64-NEXT:    vptestnmq %xmm0, %xmm1, %k0 {%k1}
+; AVX512VL-X64-NEXT:    kmovw %k0, %eax
+; AVX512VL-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X64-NEXT:    retq
+;
+; AVX512VL-X86-LABEL: TEST_mm_mask_testn_epi64_mask:
+; AVX512VL-X86:       # %bb.0: # %entry
+; AVX512VL-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512VL-X86-NEXT:    kmovw %eax, %k1
+; AVX512VL-X86-NEXT:    vptestnmq %xmm0, %xmm1, %k0 {%k1}
+; AVX512VL-X86-NEXT:    kmovw %k0, %eax
+; AVX512VL-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X86-NEXT:    retl
+;
+; AVX512F-X64-LABEL: TEST_mm_mask_testn_epi64_mask:
+; AVX512F-X64:       # %bb.0: # %entry
+; AVX512F-X64-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-X64-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-X64-NEXT:    kmovw %edi, %k1
+; AVX512F-X64-NEXT:    vptestnmq %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X64-NEXT:    kshiftlw $14, %k0, %k0
+; AVX512F-X64-NEXT:    kshiftrw $14, %k0, %k0
+; AVX512F-X64-NEXT:    kmovw %k0, %eax
+; AVX512F-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X64-NEXT:    vzeroupper
+; AVX512F-X64-NEXT:    retq
+;
+; AVX512F-X86-LABEL: TEST_mm_mask_testn_epi64_mask:
+; AVX512F-X86:       # %bb.0: # %entry
+; AVX512F-X86-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512F-X86-NEXT:    kmovw %eax, %k1
+; AVX512F-X86-NEXT:    vptestnmq %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X86-NEXT:    kshiftlw $14, %k0, %k0
+; AVX512F-X86-NEXT:    kshiftrw $14, %k0, %k0
+; AVX512F-X86-NEXT:    kmovw %k0, %eax
+; AVX512F-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X86-NEXT:    vzeroupper
+; AVX512F-X86-NEXT:    retl
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = icmp eq <2 x i64> %and.i.i, zeroinitializer
@@ -347,22 +511,49 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm_mask_testn_epi32_mask(i8 %__U, <2 x i64> %__A, <2 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm_mask_testn_epi32_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    kmovw %edi, %k1
-; X86_64-NEXT:    vptestnmd %xmm0, %xmm1, %k0 {%k1}
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm_mask_testn_epi32_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; I386-NEXT:    kmovw %eax, %k1
-; I386-NEXT:    vptestnmd %xmm0, %xmm1, %k0 {%k1}
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    retl
+; AVX512VL-X64-LABEL: TEST_mm_mask_testn_epi32_mask:
+; AVX512VL-X64:       # %bb.0: # %entry
+; AVX512VL-X64-NEXT:    kmovw %edi, %k1
+; AVX512VL-X64-NEXT:    vptestnmd %xmm0, %xmm1, %k0 {%k1}
+; AVX512VL-X64-NEXT:    kmovw %k0, %eax
+; AVX512VL-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X64-NEXT:    retq
+;
+; AVX512VL-X86-LABEL: TEST_mm_mask_testn_epi32_mask:
+; AVX512VL-X86:       # %bb.0: # %entry
+; AVX512VL-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512VL-X86-NEXT:    kmovw %eax, %k1
+; AVX512VL-X86-NEXT:    vptestnmd %xmm0, %xmm1, %k0 {%k1}
+; AVX512VL-X86-NEXT:    kmovw %k0, %eax
+; AVX512VL-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X86-NEXT:    retl
+;
+; AVX512F-X64-LABEL: TEST_mm_mask_testn_epi32_mask:
+; AVX512F-X64:       # %bb.0: # %entry
+; AVX512F-X64-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-X64-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-X64-NEXT:    kmovw %edi, %k1
+; AVX512F-X64-NEXT:    vptestnmd %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X64-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-X64-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-X64-NEXT:    kmovw %k0, %eax
+; AVX512F-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X64-NEXT:    vzeroupper
+; AVX512F-X64-NEXT:    retq
+;
+; AVX512F-X86-LABEL: TEST_mm_mask_testn_epi32_mask:
+; AVX512F-X86:       # %bb.0: # %entry
+; AVX512F-X86-NEXT:    # kill: def $xmm1 killed $xmm1 def $zmm1
+; AVX512F-X86-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512F-X86-NEXT:    kmovw %eax, %k1
+; AVX512F-X86-NEXT:    vptestnmd %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X86-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-X86-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-X86-NEXT:    kmovw %k0, %eax
+; AVX512F-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X86-NEXT:    vzeroupper
+; AVX512F-X86-NEXT:    retl
 entry:
   %and.i.i = and <2 x i64> %__B, %__A
   %0 = bitcast <2 x i64> %and.i.i to <4 x i32>
@@ -378,24 +569,51 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm256_mask_testn_epi64_mask(i8 %__U, <4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm256_mask_testn_epi64_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    kmovw %edi, %k1
-; X86_64-NEXT:    vptestnmq %ymm0, %ymm1, %k0 {%k1}
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    vzeroupper
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm256_mask_testn_epi64_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
-; I386-NEXT:    kmovw %eax, %k1
-; I386-NEXT:    vptestnmq %ymm0, %ymm1, %k0 {%k1}
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    vzeroupper
-; I386-NEXT:    retl
+; AVX512VL-X64-LABEL: TEST_mm256_mask_testn_epi64_mask:
+; AVX512VL-X64:       # %bb.0: # %entry
+; AVX512VL-X64-NEXT:    kmovw %edi, %k1
+; AVX512VL-X64-NEXT:    vptestnmq %ymm0, %ymm1, %k0 {%k1}
+; AVX512VL-X64-NEXT:    kmovw %k0, %eax
+; AVX512VL-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X64-NEXT:    vzeroupper
+; AVX512VL-X64-NEXT:    retq
+;
+; AVX512VL-X86-LABEL: TEST_mm256_mask_testn_epi64_mask:
+; AVX512VL-X86:       # %bb.0: # %entry
+; AVX512VL-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512VL-X86-NEXT:    kmovw %eax, %k1
+; AVX512VL-X86-NEXT:    vptestnmq %ymm0, %ymm1, %k0 {%k1}
+; AVX512VL-X86-NEXT:    kmovw %k0, %eax
+; AVX512VL-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X86-NEXT:    vzeroupper
+; AVX512VL-X86-NEXT:    retl
+;
+; AVX512F-X64-LABEL: TEST_mm256_mask_testn_epi64_mask:
+; AVX512F-X64:       # %bb.0: # %entry
+; AVX512F-X64-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-X64-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-X64-NEXT:    kmovw %edi, %k1
+; AVX512F-X64-NEXT:    vptestnmq %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X64-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-X64-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-X64-NEXT:    kmovw %k0, %eax
+; AVX512F-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X64-NEXT:    vzeroupper
+; AVX512F-X64-NEXT:    retq
+;
+; AVX512F-X86-LABEL: TEST_mm256_mask_testn_epi64_mask:
+; AVX512F-X86:       # %bb.0: # %entry
+; AVX512F-X86-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-X86-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
+; AVX512F-X86-NEXT:    kmovw %eax, %k1
+; AVX512F-X86-NEXT:    vptestnmq %zmm0, %zmm1, %k0 {%k1}
+; AVX512F-X86-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512F-X86-NEXT:    kshiftrw $12, %k0, %k0
+; AVX512F-X86-NEXT:    kmovw %k0, %eax
+; AVX512F-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X86-NEXT:    vzeroupper
+; AVX512F-X86-NEXT:    retl
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = icmp eq <4 x i64> %and.i.i, zeroinitializer
@@ -409,23 +627,45 @@ entry:
 
 ; Function Attrs: norecurse nounwind readnone
 define zeroext i8 @TEST_mm256_mask_testn_epi32_mask(i8 %__U, <4 x i64> %__A, <4 x i64> %__B) local_unnamed_addr #0 {
-; X86_64-LABEL: TEST_mm256_mask_testn_epi32_mask:
-; X86_64:       # %bb.0: # %entry
-; X86_64-NEXT:    vptestnmd %ymm0, %ymm1, %k0
-; X86_64-NEXT:    kmovw %k0, %eax
-; X86_64-NEXT:    andb %dil, %al
-; X86_64-NEXT:    # kill: def $al killed $al killed $eax
-; X86_64-NEXT:    vzeroupper
-; X86_64-NEXT:    retq
-;
-; I386-LABEL: TEST_mm256_mask_testn_epi32_mask:
-; I386:       # %bb.0: # %entry
-; I386-NEXT:    vptestnmd %ymm0, %ymm1, %k0
-; I386-NEXT:    kmovw %k0, %eax
-; I386-NEXT:    andb {{[0-9]+}}(%esp), %al
-; I386-NEXT:    # kill: def $al killed $al killed $eax
-; I386-NEXT:    vzeroupper
-; I386-NEXT:    retl
+; AVX512VL-X64-LABEL: TEST_mm256_mask_testn_epi32_mask:
+; AVX512VL-X64:       # %bb.0: # %entry
+; AVX512VL-X64-NEXT:    vptestnmd %ymm0, %ymm1, %k0
+; AVX512VL-X64-NEXT:    kmovw %k0, %eax
+; AVX512VL-X64-NEXT:    andb %dil, %al
+; AVX512VL-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X64-NEXT:    vzeroupper
+; AVX512VL-X64-NEXT:    retq
+;
+; AVX512VL-X86-LABEL: TEST_mm256_mask_testn_epi32_mask:
+; AVX512VL-X86:       # %bb.0: # %entry
+; AVX512VL-X86-NEXT:    vptestnmd %ymm0, %ymm1, %k0
+; AVX512VL-X86-NEXT:    kmovw %k0, %eax
+; AVX512VL-X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; AVX512VL-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512VL-X86-NEXT:    vzeroupper
+; AVX512VL-X86-NEXT:    retl
+;
+; AVX512F-X64-LABEL: TEST_mm256_mask_testn_epi32_mask:
+; AVX512F-X64:       # %bb.0: # %entry
+; AVX512F-X64-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-X64-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-X64-NEXT:    vptestnmd %zmm0, %zmm1, %k0
+; AVX512F-X64-NEXT:    kmovw %k0, %eax
+; AVX512F-X64-NEXT:    andb %dil, %al
+; AVX512F-X64-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X64-NEXT:    vzeroupper
+; AVX512F-X64-NEXT:    retq
+;
+; AVX512F-X86-LABEL: TEST_mm256_mask_testn_epi32_mask:
+; AVX512F-X86:       # %bb.0: # %entry
+; AVX512F-X86-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-X86-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-X86-NEXT:    vptestnmd %zmm0, %zmm1, %k0
+; AVX512F-X86-NEXT:    kmovw %k0, %eax
+; AVX512F-X86-NEXT:    andb {{[0-9]+}}(%esp), %al
+; AVX512F-X86-NEXT:    # kill: def $al killed $al killed $eax
+; AVX512F-X86-NEXT:    vzeroupper
+; AVX512F-X86-NEXT:    retl
 entry:
   %and.i.i = and <4 x i64> %__B, %__A
   %0 = bitcast <4 x i64> %and.i.i to <8 x i32>