[X86] Use two pmovmskbs in combineBitcastvxi1 for (i64 (bitcast (v64i1 (truncate (v64i8)))) on KNL.

llvm-svn: 350481

3 files changed, 59 insertions, 170 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 67f2929dae7..f643482884e 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -32743,7 +32743,8 @@ static SDValue combineBitcastvxi1(SelectionDAG &DAG, SDValue BitCast,
   // vpcmpeqb/vpcmpgtb.
   bool IsTruncated = N0.getOpcode() == ISD::TRUNCATE && N0.hasOneUse() &&
                      (N0.getOperand(0).getValueType() == MVT::v16i8 ||
-                      N0.getOperand(0).getValueType() == MVT::v32i8);
+                      N0.getOperand(0).getValueType() == MVT::v32i8 ||
+                      N0.getOperand(0).getValueType() == MVT::v64i8);
 
   // With AVX512 vxi1 types are legal and we prefer using k-regs.
   // MOVMSK is supported in SSE2 or later.
@@ -32799,12 +32800,30 @@ static SDValue combineBitcastvxi1(SelectionDAG &DAG, SDValue BitCast,
   case MVT::v32i1:
     SExtVT = MVT::v32i8;
     break;
+  case MVT::v64i1:
+    // If we have AVX512F, but not AVX512BW and the input is truncated from
+    // v64i8 checked earlier. Then split the input and make two pmovmskbs.
+    if (Subtarget.hasAVX512() && !Subtarget.hasBWI()) {
+      SExtVT = MVT::v64i8;
+      break;
+    }
+    return SDValue();
   };
 
   SDLoc DL(BitCast);
   SDValue V = DAG.getNode(ISD::SIGN_EXTEND, DL, SExtVT, N0);
 
-  if (SExtVT == MVT::v16i8 || SExtVT == MVT::v32i8) {
+  if (SExtVT == MVT::v64i8) {
+    SDValue Lo, Hi;
+    std::tie(Lo, Hi) = DAG.SplitVector(V, DL);
+    Lo = DAG.getNode(X86ISD::MOVMSK, DL, MVT::i32, Lo);
+    Lo = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, Lo);
+    Hi = DAG.getNode(X86ISD::MOVMSK, DL, MVT::i32, Hi);
+    Hi = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Hi);
+    Hi = DAG.getNode(ISD::SHL, DL, MVT::i64, Hi,
+                     DAG.getConstant(32, DL, MVT::i8));
+    V = DAG.getNode(ISD::OR, DL, MVT::i64, Lo, Hi);
+  } else if (SExtVT == MVT::v16i8 || SExtVT == MVT::v32i8) {
     V = getPMOVMSKB(DL, V, DAG, Subtarget);
   } else {
     if (SExtVT == MVT::v8i16)
diff --git a/llvm/test/CodeGen/X86/bitcast-setcc-512.ll b/llvm/test/CodeGen/X86/bitcast-setcc-512.ll
index 1911aed2973..340c7ab3afd 100644
--- a/llvm/test/CodeGen/X86/bitcast-setcc-512.ll
+++ b/llvm/test/CodeGen/X86/bitcast-setcc-512.ll
@@ -256,26 +256,10 @@ define i64 @v64i8(<64 x i8> %a, <64 x i8> %b) {
 ;
 ; AVX512F-LABEL: v64i8:
 ; AVX512F:       # %bb.0:
+; AVX512F-NEXT:    vpcmpgtb %ymm3, %ymm1, %ymm1
 ; AVX512F-NEXT:    vpcmpgtb %ymm2, %ymm0, %ymm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm2
-; AVX512F-NEXT:    vptestmd %zmm2, %zmm2, %k0
-; AVX512F-NEXT:    kmovw %k0, %eax
-; AVX512F-NEXT:    vextracti128 $1, %ymm0, %xmm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm0
-; AVX512F-NEXT:    vptestmd %zmm0, %zmm0, %k0
-; AVX512F-NEXT:    kmovw %k0, %ecx
-; AVX512F-NEXT:    shll $16, %ecx
-; AVX512F-NEXT:    orl %eax, %ecx
-; AVX512F-NEXT:    vpcmpgtb %ymm3, %ymm1, %ymm0
-; AVX512F-NEXT:    vpmovsxbd %xmm0, %zmm1
-; AVX512F-NEXT:    vptestmd %zmm1, %zmm1, %k0
-; AVX512F-NEXT:    kmovw %k0, %edx
-; 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 %edx, %eax
+; AVX512F-NEXT:    vpmovmskb %ymm0, %ecx
+; AVX512F-NEXT:    vpmovmskb %ymm1, %eax
 ; AVX512F-NEXT:    shlq $32, %rax
 ; AVX512F-NEXT:    orq %rcx, %rax
 ; AVX512F-NEXT:    vzeroupper
diff --git a/llvm/test/CodeGen/X86/movmsk-cmp.ll b/llvm/test/CodeGen/X86/movmsk-cmp.ll
index 93d86b0e11d..bc16d8c710e 100644
--- a/llvm/test/CodeGen/X86/movmsk-cmp.ll
+++ b/llvm/test/CodeGen/X86/movmsk-cmp.ll
@@ -235,30 +235,11 @@ define i1 @allones_v64i8_sign(<64 x i8> %arg) {
 ;
 ; KNL-LABEL: allones_v64i8_sign:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; KNL-NEXT:    vpcmpgtb %ymm0, %ymm2, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm3
-; KNL-NEXT:    vptestmd %zmm3, %zmm3, %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:    vpcmpgtb %ymm1, %ymm2, %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, %edx
-; KNL-NEXT:    shll $16, %edx
-; KNL-NEXT:    orl %eax, %edx
-; KNL-NEXT:    shlq $32, %rdx
-; KNL-NEXT:    orq %rcx, %rdx
-; KNL-NEXT:    cmpq $-1, %rdx
+; KNL-NEXT:    vpmovmskb %ymm1, %eax
+; KNL-NEXT:    shlq $32, %rax
+; KNL-NEXT:    vpmovmskb %ymm0, %ecx
+; KNL-NEXT:    orq %rax, %rcx
+; KNL-NEXT:    cmpq $-1, %rcx
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -327,29 +308,10 @@ define i1 @allzeros_v64i8_sign(<64 x i8> %arg) {
 ;
 ; KNL-LABEL: allzeros_v64i8_sign:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; KNL-NEXT:    vpcmpgtb %ymm0, %ymm2, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm3
-; KNL-NEXT:    vptestmd %zmm3, %zmm3, %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:    vpcmpgtb %ymm1, %ymm2, %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, %edx
-; KNL-NEXT:    shll $16, %edx
-; KNL-NEXT:    orl %eax, %edx
-; KNL-NEXT:    shlq $32, %rdx
-; KNL-NEXT:    orq %rcx, %rdx
+; KNL-NEXT:    vpmovmskb %ymm1, %eax
+; KNL-NEXT:    shlq $32, %rax
+; KNL-NEXT:    vpmovmskb %ymm0, %ecx
+; KNL-NEXT:    orq %rax, %rcx
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -1662,32 +1624,13 @@ define i1 @allones_v64i8_and1(<64 x i8> %arg) {
 ;
 ; KNL-LABEL: allones_v64i8_and1:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [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 %ymm2, %ymm1, %ymm1
-; KNL-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm3
-; KNL-NEXT:    vptestmd %zmm3, %zmm3, %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:    vpcmpeqb %ymm2, %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, %edx
-; KNL-NEXT:    shll $16, %edx
-; KNL-NEXT:    orl %eax, %edx
-; KNL-NEXT:    shlq $32, %rdx
-; KNL-NEXT:    orq %rcx, %rdx
-; KNL-NEXT:    cmpq $-1, %rdx
+; KNL-NEXT:    vpsllw $7, %ymm0, %ymm0
+; KNL-NEXT:    vpsllw $7, %ymm1, %ymm1
+; KNL-NEXT:    vpmovmskb %ymm1, %eax
+; KNL-NEXT:    shlq $32, %rax
+; KNL-NEXT:    vpmovmskb %ymm0, %ecx
+; KNL-NEXT:    orq %rax, %rcx
+; KNL-NEXT:    cmpq $-1, %rcx
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -1771,31 +1714,12 @@ define i1 @allzeros_v64i8_and1(<64 x i8> %arg) {
 ;
 ; KNL-LABEL: allzeros_v64i8_and1:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [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 %ymm2, %ymm1, %ymm1
-; KNL-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm3
-; KNL-NEXT:    vptestmd %zmm3, %zmm3, %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:    vpcmpeqb %ymm2, %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, %edx
-; KNL-NEXT:    shll $16, %edx
-; KNL-NEXT:    orl %eax, %edx
-; KNL-NEXT:    shlq $32, %rdx
-; KNL-NEXT:    orq %rcx, %rdx
+; KNL-NEXT:    vpsllw $7, %ymm0, %ymm0
+; KNL-NEXT:    vpsllw $7, %ymm1, %ymm1
+; KNL-NEXT:    vpmovmskb %ymm1, %eax
+; KNL-NEXT:    shlq $32, %rax
+; KNL-NEXT:    vpmovmskb %ymm0, %ecx
+; KNL-NEXT:    orq %rax, %rcx
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -3299,32 +3223,13 @@ define i1 @allones_v64i8_and4(<64 x i8> %arg) {
 ;
 ; KNL-LABEL: allones_v64i8_and4:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [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 %ymm2, %ymm1, %ymm1
-; KNL-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm3
-; KNL-NEXT:    vptestmd %zmm3, %zmm3, %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:    vpcmpeqb %ymm2, %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, %edx
-; KNL-NEXT:    shll $16, %edx
-; KNL-NEXT:    orl %eax, %edx
-; KNL-NEXT:    shlq $32, %rdx
-; KNL-NEXT:    orq %rcx, %rdx
-; KNL-NEXT:    cmpq $-1, %rdx
+; KNL-NEXT:    vpsllw $5, %ymm0, %ymm0
+; KNL-NEXT:    vpsllw $5, %ymm1, %ymm1
+; KNL-NEXT:    vpmovmskb %ymm1, %eax
+; KNL-NEXT:    shlq $32, %rax
+; KNL-NEXT:    vpmovmskb %ymm0, %ecx
+; KNL-NEXT:    orq %rax, %rcx
+; KNL-NEXT:    cmpq $-1, %rcx
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq
@@ -3408,31 +3313,12 @@ define i1 @allzeros_v64i8_and4(<64 x i8> %arg) {
 ;
 ; KNL-LABEL: allzeros_v64i8_and4:
 ; KNL:       # %bb.0:
-; KNL-NEXT:    vmovdqa {{.*#+}} ymm2 = [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 %ymm2, %ymm1, %ymm1
-; KNL-NEXT:    vpand %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpcmpeqb %ymm2, %ymm0, %ymm0
-; KNL-NEXT:    vpmovsxbd %xmm0, %zmm3
-; KNL-NEXT:    vptestmd %zmm3, %zmm3, %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:    vpcmpeqb %ymm2, %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, %edx
-; KNL-NEXT:    shll $16, %edx
-; KNL-NEXT:    orl %eax, %edx
-; KNL-NEXT:    shlq $32, %rdx
-; KNL-NEXT:    orq %rcx, %rdx
+; KNL-NEXT:    vpsllw $5, %ymm0, %ymm0
+; KNL-NEXT:    vpsllw $5, %ymm1, %ymm1
+; KNL-NEXT:    vpmovmskb %ymm1, %eax
+; KNL-NEXT:    shlq $32, %rax
+; KNL-NEXT:    vpmovmskb %ymm0, %ecx
+; KNL-NEXT:    orq %rax, %rcx
 ; KNL-NEXT:    sete %al
 ; KNL-NEXT:    vzeroupper
 ; KNL-NEXT:    retq