From 1b1457fd624d6ada42b2cb5ec1d925b76da75e37 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Sat, 5 Jan 2019 21:40:07 +0000
Subject: [PATCH] [X86] Allow combinevxi1Bitcast to use pmovmskb on avx512
 targets if the input is a truncate from v16i8/v32i8.

This is especially helpful on targets without avx512bw since we don't have a good way to convert from v16i8/v32i8 to v16i1/v32i1 for the truncate anyway. If we're just going to convert it to a GPR we might as well use pmovmskb to accomplish both.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@350480 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp        |  10 +-
 test/CodeGen/X86/avx512-ext.ll            |  33 +----
 test/CodeGen/X86/avx512-schedule.ll       |   6 +-
 test/CodeGen/X86/bitcast-and-setcc-128.ll |   5 +-
 test/CodeGen/X86/bitcast-and-setcc-256.ll |  10 +-
 test/CodeGen/X86/bitcast-setcc-128.ll     |   5 +-
 test/CodeGen/X86/bitcast-setcc-256.ll     |  10 +-
 test/CodeGen/X86/broadcastm-lowering.ll   |  12 +-
 test/CodeGen/X86/movmsk-cmp.ll            | 170 +++++-----------------
 9 files changed, 61 insertions(+), 200 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 4056b4982b0..67f2929dae7 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -32737,9 +32737,17 @@ static SDValue combineBitcastvxi1(SelectionDAG &DAG, SDValue BitCast,
   if (!VT.isScalarInteger() || !VecVT.isSimple())
     return SDValue();
 
+  // If the input is a truncate from v16i8 or v32i8 go ahead and use a
+  // movmskb even with avx512. This will be better than truncating to vXi1 and
+  // using a kmov. This can especially help KNL if the input is a v16i8/v32i8
+  // vpcmpeqb/vpcmpgtb.
+  bool IsTruncated = N0.getOpcode() == ISD::TRUNCATE && N0.hasOneUse() &&
+                     (N0.getOperand(0).getValueType() == MVT::v16i8 ||
+                      N0.getOperand(0).getValueType() == MVT::v32i8);
+
   // With AVX512 vxi1 types are legal and we prefer using k-regs.
   // MOVMSK is supported in SSE2 or later.
-  if (Subtarget.hasAVX512() || !Subtarget.hasSSE2())
+  if (!Subtarget.hasSSE2() || (Subtarget.hasAVX512() && !IsTruncated))
     return SDValue();
 
   // There are MOVMSK flavors for types v16i8, v32i8, v4f32, v8f32, v4f64 and
diff --git a/test/CodeGen/X86/avx512-ext.ll b/test/CodeGen/X86/avx512-ext.ll
index 2381180af76..072e3c8bdae 100644
--- a/test/CodeGen/X86/avx512-ext.ll
+++ b/test/CodeGen/X86/avx512-ext.ll
@@ -1644,33 +1644,12 @@ define   <8 x i64> @zext_8i1_to_8xi64(i8 %b) {
 }
 
 define i16 @trunc_16i8_to_16i1(<16 x i8> %a) {
-; KNL-LABEL: trunc_16i8_to_16i1:
-; KNL:       # %bb.0:
-; KNL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; KNL-NEXT:    vpslld $31, %zmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    # kill: def $ax killed $ax killed $eax
-; KNL-NEXT:    vzeroupper
-; KNL-NEXT:    retq
-;
-; SKX-LABEL: trunc_16i8_to_16i1:
-; SKX:       # %bb.0:
-; SKX-NEXT:    vpsllw $7, %xmm0, %xmm0
-; SKX-NEXT:    vpmovb2m %xmm0, %k0
-; SKX-NEXT:    kmovd %k0, %eax
-; SKX-NEXT:    # kill: def $ax killed $ax killed $eax
-; SKX-NEXT:    retq
-;
-; AVX512DQNOBW-LABEL: trunc_16i8_to_16i1:
-; AVX512DQNOBW:       # %bb.0:
-; AVX512DQNOBW-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
-; AVX512DQNOBW-NEXT:    vpslld $31, %zmm0, %zmm0
-; AVX512DQNOBW-NEXT:    vpmovd2m %zmm0, %k0
-; AVX512DQNOBW-NEXT:    kmovw %k0, %eax
-; AVX512DQNOBW-NEXT:    # kill: def $ax killed $ax killed $eax
-; AVX512DQNOBW-NEXT:    vzeroupper
-; AVX512DQNOBW-NEXT:    retq
+; ALL-LABEL: trunc_16i8_to_16i1:
+; ALL:       # %bb.0:
+; ALL-NEXT:    vpsllw $7, %xmm0, %xmm0
+; ALL-NEXT:    vpmovmskb %xmm0, %eax
+; ALL-NEXT:    # kill: def $ax killed $ax killed $eax
+; ALL-NEXT:    retq
   %mask_b = trunc <16 x i8>%a to <16 x i1>
   %mask = bitcast <16 x i1> %mask_b to i16
   ret i16 %mask
diff --git a/test/CodeGen/X86/avx512-schedule.ll b/test/CodeGen/X86/avx512-schedule.ll
index 5c44d8679b9..2c9b6e13481 100755
--- a/test/CodeGen/X86/avx512-schedule.ll
+++ b/test/CodeGen/X86/avx512-schedule.ll
@@ -4285,16 +4285,14 @@ define i16 @trunc_16i8_to_16i1(<16 x i8> %a) {
 ; GENERIC-LABEL: trunc_16i8_to_16i1:
 ; GENERIC:       # %bb.0:
 ; GENERIC-NEXT:    vpsllw $7, %xmm0, %xmm0 # sched: [1:1.00]
-; GENERIC-NEXT:    vpmovb2m %xmm0, %k0 # sched: [1:0.33]
-; GENERIC-NEXT:    kmovd %k0, %eax # sched: [1:0.33]
+; GENERIC-NEXT:    vpmovmskb %xmm0, %eax # sched: [2:1.00]
 ; GENERIC-NEXT:    # kill: def $ax killed $ax killed $eax
 ; GENERIC-NEXT:    retq # sched: [1:1.00]
 ;
 ; SKX-LABEL: trunc_16i8_to_16i1:
 ; SKX:       # %bb.0:
 ; SKX-NEXT:    vpsllw $7, %xmm0, %xmm0 # sched: [1:0.50]
-; SKX-NEXT:    vpmovb2m %xmm0, %k0 # sched: [1:1.00]
-; SKX-NEXT:    kmovd %k0, %eax # sched: [3:1.00]
+; SKX-NEXT:    vpmovmskb %xmm0, %eax # sched: [2:1.00]
 ; SKX-NEXT:    # kill: def $ax killed $ax killed $eax
 ; SKX-NEXT:    retq # sched: [7:1.00]
   %mask_b = trunc <16 x i8>%a to <16 x i1>
diff --git a/test/CodeGen/X86/bitcast-and-setcc-128.ll b/test/CodeGen/X86/bitcast-and-setcc-128.ll
index 0a2e154f5b9..289ddcb194b 100644
--- a/test/CodeGen/X86/bitcast-and-setcc-128.ll
+++ b/test/CodeGen/X86/bitcast-and-setcc-128.ll
@@ -159,11 +159,8 @@ define i16 @v16i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> %c, <16 x i8> %d) {
 ; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
 ; AVX512F-NEXT:    vpcmpgtb %xmm3, %xmm2, %xmm1
 ; AVX512F-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX512F-NEXT:    # kill: def $ax killed $ax killed $eax
-; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: v16i8:
diff --git a/test/CodeGen/X86/bitcast-and-setcc-256.ll b/test/CodeGen/X86/bitcast-and-setcc-256.ll
index 0f3b8c94540..426cabe5f4c 100644
--- a/test/CodeGen/X86/bitcast-and-setcc-256.ll
+++ b/test/CodeGen/X86/bitcast-and-setcc-256.ll
@@ -399,15 +399,7 @@ define i32 @v32i8(<32 x i8> %a, <32 x i8> %b, <32 x i8> %c, <32 x i8> %d) {
 ; AVX512F-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
 ; AVX512F-NEXT:    vpcmpgtb %ymm3, %ymm2, %ymm1
 ; AVX512F-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm1
-; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, %ecx
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, %eax
-; AVX512F-NEXT:    shll $16, %eax
-; AVX512F-NEXT:    orl %ecx, %eax
+; AVX512F-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
diff --git a/test/CodeGen/X86/bitcast-setcc-128.ll b/test/CodeGen/X86/bitcast-setcc-128.ll
index f803901c0e7..fb585974e5b 100644
--- a/test/CodeGen/X86/bitcast-setcc-128.ll
+++ b/test/CodeGen/X86/bitcast-setcc-128.ll
@@ -128,11 +128,8 @@ define i16 @v16i8(<16 x i8> %a, <16 x i8> %b) {
 ; AVX512F-LABEL: v16i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpcmpgtb %xmm1, %xmm0, %xmm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, %eax
+; AVX512F-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX512F-NEXT:    # kill: def $ax killed $ax killed $eax
-; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512BW-LABEL: v16i8:
diff --git a/test/CodeGen/X86/bitcast-setcc-256.ll b/test/CodeGen/X86/bitcast-setcc-256.ll
index d349ae3bc40..b0af971366c 100644
--- a/test/CodeGen/X86/bitcast-setcc-256.ll
+++ b/test/CodeGen/X86/bitcast-setcc-256.ll
@@ -184,15 +184,7 @@ define i32 @v32i8(<32 x i8> %a, <32 x i8> %b) {
 ; AVX512F-LABEL: v32i8:
 ; AVX512F:       # %bb.0:
 ; AVX512F-NEXT:    vpcmpgtb %ymm1, %ymm0, %ymm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm1
-; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, %ecx
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, %eax
-; AVX512F-NEXT:    shll $16, %eax
-; AVX512F-NEXT:    orl %ecx, %eax
+; AVX512F-NEXT:    vpmovmskb %ymm0, %eax
 ; AVX512F-NEXT:    vzeroupper
 ; AVX512F-NEXT:    retq
 ;
diff --git a/test/CodeGen/X86/broadcastm-lowering.ll b/test/CodeGen/X86/broadcastm-lowering.ll
index d25a4da9843..986d313cb8d 100644
--- a/test/CodeGen/X86/broadcastm-lowering.ll
+++ b/test/CodeGen/X86/broadcastm-lowering.ll
@@ -43,15 +43,9 @@ define <4 x i32> @test_mm_epi32(<16 x i8> %a, <16 x i8> %b) {
 ; AVX512CD-LABEL: test_mm_epi32:
 ; AVX512CD:       # %bb.0: # %entry
 ; AVX512CD-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; AVX512CD-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512CD-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512CD-NEXT:    kmovw %k0, %eax
-; AVX512CD-NEXT:    vpxor %xmm0, %xmm0, %xmm0
-; AVX512CD-NEXT:    vpinsrw $0, %eax, %xmm0, %xmm0
-; AVX512CD-NEXT:    vpinsrw $2, %eax, %xmm0, %xmm0
-; AVX512CD-NEXT:    vpinsrw $4, %eax, %xmm0, %xmm0
-; AVX512CD-NEXT:    vpinsrw $6, %eax, %xmm0, %xmm0
-; AVX512CD-NEXT:    vzeroupper
+; AVX512CD-NEXT:    vpmovmskb %xmm0, %eax
+; AVX512CD-NEXT:    vmovd %eax, %xmm0
+; AVX512CD-NEXT:    vpbroadcastd %xmm0, %xmm0
 ; AVX512CD-NEXT:    retq
 ;
 ; AVX512VLCDBW-LABEL: test_mm_epi32:
diff --git a/test/CodeGen/X86/movmsk-cmp.ll b/test/CodeGen/X86/movmsk-cmp.ll
index 718ade02435..93d86b0e11d 100644
--- a/test/CodeGen/X86/movmsk-cmp.ll
+++ b/test/CodeGen/X86/movmsk-cmp.ll
@@ -22,13 +22,9 @@ define i1 @allones_v16i8_sign(<16 x i8> %arg) {
 ;
 ; KNL-LABEL: allones_v16i8_sign:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; KNL-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kortestw %k0, %k0
-; KNL-NEXT:    setb %al
-; KNL-NEXT:    vzeroupper
+; KNL-NEXT:    vpmovmskb %xmm0, %eax
+; KNL-NEXT:    cmpw $-1, %ax
+; KNL-NEXT:    sete %al
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: allones_v16i8_sign:
@@ -60,13 +56,9 @@ define i1 @allzeros_v16i8_sign(<16 x i8> %arg) {
 ;
 ; KNL-LABEL: allzeros_v16i8_sign:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; KNL-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kortestw %k0, %k0
+; KNL-NEXT:    vpmovmskb %xmm0, %eax
+; KNL-NEXT:    testw %ax, %ax
 ; KNL-NEXT:    sete %al
-; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: allzeros_v16i8_sign:
@@ -117,18 +109,8 @@ define i1 @allones_v32i8_sign(<32 x i8> %arg) {
 ;
 ; KNL-LABEL: allones_v32i8_sign:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; KNL-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, %ecx
-; KNL-NEXT:    shll $16, %ecx
-; KNL-NEXT:    orl %eax, %ecx
-; KNL-NEXT:    cmpl $-1, %ecx
+; KNL-NEXT:    vpmovmskb %ymm0, %eax
+; KNL-NEXT:    cmpl $-1, %eax
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -180,17 +162,8 @@ define i1 @allzeros_v32i8_sign(<32 x i8> %arg) {
 ;
 ; KNL-LABEL: allzeros_v32i8_sign:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; KNL-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, %ecx
-; KNL-NEXT:    shll $16, %ecx
-; KNL-NEXT:    orl %eax, %ecx
+; KNL-NEXT:    vpmovmskb %ymm0, %eax
+; KNL-NEXT:    testl %eax, %eax
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -1438,14 +1411,10 @@ define i1 @allones_v16i8_and1(<16 x i8> %arg) {
 ;
 ; KNL-LABEL: allones_v16i8_and1:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; KNL-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; KNL-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kortestw %k0, %k0
-; KNL-NEXT:    setb %al
-; KNL-NEXT:    vzeroupper
+; KNL-NEXT:    vpsllw $7, %xmm0, %xmm0
+; KNL-NEXT:    vpmovmskb %xmm0, %eax
+; KNL-NEXT:    cmpw $-1, %ax
+; KNL-NEXT:    sete %al
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: allones_v16i8_and1:
@@ -1480,14 +1449,10 @@ define i1 @allzeros_v16i8_and1(<16 x i8> %arg) {
 ;
 ; KNL-LABEL: allzeros_v16i8_and1:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} xmm1 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; KNL-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; KNL-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kortestw %k0, %k0
+; KNL-NEXT:    vpsllw $7, %xmm0, %xmm0
+; KNL-NEXT:    vpmovmskb %xmm0, %eax
+; KNL-NEXT:    testw %ax, %ax
 ; KNL-NEXT:    sete %al
-; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: allzeros_v16i8_and1:
@@ -1546,19 +1511,9 @@ define i1 @allones_v32i8_and1(<32 x i8> %arg) {
 ;
 ; KNL-LABEL: allones_v32i8_and1:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; KNL-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, %ecx
-; KNL-NEXT:    shll $16, %ecx
-; KNL-NEXT:    orl %eax, %ecx
-; KNL-NEXT:    cmpl $-1, %ecx
+; KNL-NEXT:    vpsllw $7, %ymm0, %ymm0
+; KNL-NEXT:    vpmovmskb %ymm0, %eax
+; KNL-NEXT:    cmpl $-1, %eax
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -1618,18 +1573,9 @@ define i1 @allzeros_v32i8_and1(<32 x i8> %arg) {
 ;
 ; KNL-LABEL: allzeros_v32i8_and1:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm1 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
-; KNL-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, %ecx
-; KNL-NEXT:    shll $16, %ecx
-; KNL-NEXT:    orl %eax, %ecx
+; KNL-NEXT:    vpsllw $7, %ymm0, %ymm0
+; KNL-NEXT:    vpmovmskb %ymm0, %eax
+; KNL-NEXT:    testl %eax, %eax
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -3102,14 +3048,10 @@ define i1 @allones_v16i8_and4(<16 x i8> %arg) {
 ;
 ; KNL-LABEL: allones_v16i8_and4:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4]
-; KNL-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; KNL-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kortestw %k0, %k0
-; KNL-NEXT:    setb %al
-; KNL-NEXT:    vzeroupper
+; KNL-NEXT:    vpsllw $5, %xmm0, %xmm0
+; KNL-NEXT:    vpmovmskb %xmm0, %eax
+; KNL-NEXT:    cmpw $-1, %ax
+; KNL-NEXT:    sete %al
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: allones_v16i8_and4:
@@ -3144,14 +3086,10 @@ define i1 @allzeros_v16i8_and4(<16 x i8> %arg) {
 ;
 ; KNL-LABEL: allzeros_v16i8_and4:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} xmm1 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4]
-; KNL-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; KNL-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kortestw %k0, %k0
+; KNL-NEXT:    vpsllw $5, %xmm0, %xmm0
+; KNL-NEXT:    vpmovmskb %xmm0, %eax
+; KNL-NEXT:    testw %ax, %ax
 ; KNL-NEXT:    sete %al
-; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: allzeros_v16i8_and4:
@@ -3210,19 +3148,9 @@ define i1 @allones_v32i8_and4(<32 x i8> %arg) {
 ;
 ; KNL-LABEL: allones_v32i8_and4:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4]
-; KNL-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, %ecx
-; KNL-NEXT:    shll $16, %ecx
-; KNL-NEXT:    orl %eax, %ecx
-; KNL-NEXT:    cmpl $-1, %ecx
+; KNL-NEXT:    vpsllw $5, %ymm0, %ymm0
+; KNL-NEXT:    vpmovmskb %ymm0, %eax
+; KNL-NEXT:    cmpl $-1, %eax
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -3282,18 +3210,9 @@ define i1 @allzeros_v32i8_and4(<32 x i8> %arg) {
 ;
 ; KNL-LABEL: allzeros_v32i8_and4:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm1 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4]
-; KNL-NEXT:    vpand %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpeqb %ymm1, %ymm0, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, %ecx
-; KNL-NEXT:    shll $16, %ecx
-; KNL-NEXT:    orl %eax, %ecx
+; KNL-NEXT:    vpsllw $5, %ymm0, %ymm0
+; KNL-NEXT:    vpmovmskb %ymm0, %eax
+; KNL-NEXT:    testl %eax, %eax
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -4926,12 +4845,7 @@ define i32 @movmskb(<16 x i8> %x) {
 ;
 ; KNL-LABEL: movmskb:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; KNL-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    vzeroupper
+; KNL-NEXT:    vpmovmskb %xmm0, %eax
 ; KNL-NEXT:    retq
 ;
 ; SKX-LABEL: movmskb:
@@ -4975,17 +4889,7 @@ define i32 @movmskb256(<32 x i8> %x) {
 ;
 ; KNL-LABEL: movmskb256:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; KNL-NEXT:    vpcmpgtb %ymm0, %ymm1, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm1
-; KNL-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; KNL-NEXT:    kmovw %k0, %ecx
-; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm0
-; KNL-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; KNL-NEXT:    kmovw %k0, %eax
-; KNL-NEXT:    shll $16, %eax
-; KNL-NEXT:    orl %ecx, %eax
+; KNL-NEXT:    vpmovmskb %ymm0, %eax
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
 ;
-- 
2.50.1