[X86] Add additional operations that masked instructions can combine with

[goldstein.w.n]

Summary

Added: OR, SMAX, SMIN, UMAX, UMIN, ABS, SHL, SRL, SRA, MUL

Intentionally not generically using TLI.isBinOp as that causes
regressions as there are many binops that cannot combine with masked
instructions.

[goldstein.w.n]

Rebase

[RKSimon]

llvm/lib/Target/X86/X86ISelLowering.cpp

19862

I can see test changes for some of the integer min/max opcodes but nothing else - could you add additional test coverage first?

[goldstein.w.n]

Rebase with tests

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[RKSimon]

Thanks - do you have ISD::SETCC coverage? That one I'm the most worried given it has different src/dst types

[goldstein.w.n]

In D143860#4131696, @RKSimon wrote:

Thanks - do you have ISD::SETCC coverage? That one I'm the most worried given it has different src/dst types

Dropped SETCC for now.

[goldstein.w.n]

Drop SETCC

[goldstein.w.n]

Drop SETCC (correct patch this time)

[RKSimon]

LGTM

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

[goldstein.w.n]

Rebase

llvm/lib/Target/X86/X86ISelLowering.cpp

auto HasMaskOperation = [&](SDValue V) {
     // it to all binary operation by checking TLI.isBinOp().
     switch (V->getOpcode()) {
     default:
       return false;
     case ISD::ADD:
     case ISD::SUB:
     case ISD::AND:
     case ISD::XOR:
+    case ISD::OR:
+    case ISD::SMAX:
+    case ISD::SMIN:
+    case ISD::UMAX:
+    case ISD::UMIN:
+    case ISD::ABS:
+    case ISD::SHL:
+    case ISD::SRL:
+    case ISD::SRA:
+    case ISD::MUL:
       break;
     }
     if (!V->hasOneUse())
       return false;
 
     return true;
   };
 

llvm/test/CodeGen/X86/combine-mask-with-shuffle.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512vpopcntdq,+avx512f,+avx512vl | FileCheck %s --check-prefixes=CHECK
 
 
 define <16 x i32> @combine_mask_with_or(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_or:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    movw $2570, %ax # imm = 0xA0A
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpblendmd %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpord %zmm2, %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0 = or <16 x i32> %shuf0_0, %shuf0_1
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 define <16 x i32> @combine_mask_with_mul(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_mul:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    movw $2570, %ax # imm = 0xA0A
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpblendmd %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpmulld %zmm2, %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0 = mul <16 x i32> %shuf0_0, %shuf0_1
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 define <16 x i32> @combine_mask_with_abs(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_abs:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    vpabsd %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0_tmp0 = tail call <16 x i32> @llvm.abs.v16i32(<16 x i32> %shuf0_0, i1 true)
   %op0_0_tmp1 = tail call <16 x i32> @llvm.abs.v16i32(<16 x i32> %shuf0_1, i1 true)
   %op0_0 = shufflevector <16 x i32> %op0_0_tmp0, <16 x i32> %op0_0_tmp0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 define <16 x i32> @combine_mask_with_umin(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_umin:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    movw $-21846, %ax # imm = 0xAAAA
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpblendmd %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpminud %zmm2, %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0 = tail call <16 x i32> @llvm.umin.v16i32(<16 x i32> %shuf0_0, <16 x i32> %shuf0_1)
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 define <16 x i32> @combine_mask_with_umax(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_umax:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    movw $-21846, %ax # imm = 0xAAAA
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpblendmd %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpmaxud %zmm2, %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0 = tail call <16 x i32> @llvm.umax.v16i32(<16 x i32> %shuf0_0, <16 x i32> %shuf0_1)
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 define <16 x i32> @combine_mask_with_smin(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_smin:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    movw $-21846, %ax # imm = 0xAAAA
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpblendmd %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpminsd %zmm2, %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0 = tail call <16 x i32> @llvm.smin.v16i32(<16 x i32> %shuf0_0, <16 x i32> %shuf0_1)
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 define <16 x i32> @combine_mask_with_smax(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_smax:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    movw $-21846, %ax # imm = 0xAAAA
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpblendmd %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpmaxsd %zmm2, %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0 = tail call <16 x i32> @llvm.smax.v16i32(<16 x i32> %shuf0_0, <16 x i32> %shuf0_1)
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 define <16 x i32> @combine_mask_with_shl(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_shl:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    movw $2570, %ax # imm = 0xA0A
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpblendmd %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpsllvd %zmm2, %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0 = shl <16 x i32> %shuf0_0, %shuf0_1
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 define <16 x i32> @combine_mask_with_ashr(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_ashr:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    movw $2570, %ax # imm = 0xA0A
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpblendmd %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpsravd %zmm2, %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0 = ashr <16 x i32> %shuf0_0, %shuf0_1
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 define <16 x i32> @combine_mask_with_lshr(<16 x i32> %v0) {
 ; CHECK-LABEL: combine_mask_with_lshr:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    movw $2570, %ax # imm = 0xA0A
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpblendmd %zmm1, %zmm0, %zmm2 {%k1}
 ; CHECK-NEXT:    vpsrlvd %zmm2, %zmm1, %zmm1
-; CHECK-NEXT:    vpopcntd %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
 ; CHECK-NEXT:    kmovw %eax, %k1
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpopcntd %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
 ; CHECK-NEXT:    retq
   %shuf0_0 = shufflevector <16 x i32> %v0, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %shuf0_1 = shufflevector <16 x i32> %v0, <16 x i32> %shuf0_0, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %op0_0 = lshr <16 x i32> %shuf0_0, %shuf0_1
   %op1_0 = tail call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %v0)
   %r = shufflevector <16 x i32> %op0_0, <16 x i32> %op1_0, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 20, i32 21, i32 22, i32 23, i32 8, i32 9, i32 10, i32 11, i32 28, i32 29, i32 30, i32 31>
   ret <16 x i32> %r
 }
 
 declare <16 x i32> @llvm.smin.v16i32(<16 x i32>, <16 x i32>)
 declare <16 x i32> @llvm.smax.v16i32(<16 x i32>, <16 x i32>)
 declare <16 x i32> @llvm.umin.v16i32(<16 x i32>, <16 x i32>)
 declare <16 x i32> @llvm.umax.v16i32(<16 x i32>, <16 x i32>)
 declare <16 x i32> @llvm.ctpop.v16i32(<16 x i32>)
 declare <16 x i32> @llvm.abs.v16i32(<16 x i32>, i1)

llvm/test/CodeGen/X86/vselect-avx512.ll

 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm1 = zmm0[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
 ; CHECK-NEXT:    vpminsd %zmm0, %zmm1, %zmm2
 ; CHECK-NEXT:    movw $-21846, %ax # imm = 0xAAAA
 ; CHECK-NEXT:    kmovw %eax, %k1
 ; CHECK-NEXT:    vpmaxsd %zmm0, %zmm1, %zmm2 {%k1}
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm2[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
 ; CHECK-NEXT:    vpminsd %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vpmaxsd %zmm2, %zmm0, %zmm0
-; CHECK-NEXT:    movb $-86, %al
-; CHECK-NEXT:    kmovw %eax, %k2
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k2}
-; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm1[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
-; CHECK-NEXT:    vpminsd %zmm1, %zmm0, %zmm2
-; CHECK-NEXT:    vpmaxsd %zmm1, %zmm0, %zmm2 {%k1}
-; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm2[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
+; CHECK-NEXT:    vshufps {{.*#+}} zmm2 = zmm1[0,1],zmm0[2,3],zmm1[4,5],zmm0[6,7],zmm1[8,9],zmm0[10,11],zmm1[12,13],zmm0[14,15]
+; CHECK-NEXT:    vshufps {{.*#+}} zmm0 = zmm1[1,0],zmm0[3,2],zmm1[5,4],zmm0[7,6],zmm1[9,8],zmm0[11,10],zmm1[13,12],zmm0[15,14]
+; CHECK-NEXT:    vpminsd %zmm2, %zmm0, %zmm1
+; CHECK-NEXT:    vpmaxsd %zmm2, %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm1[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
 ; CHECK-NEXT:    vpermq {{.*#+}} zmm0 = zmm0[2,3,0,1,6,7,4,5]
+; CHECK-NEXT:    vpminsd %zmm1, %zmm0, %zmm2
+; CHECK-NEXT:    movw $-3856, %ax # imm = 0xF0F0
+; CHECK-NEXT:    kmovw %eax, %k2
+; CHECK-NEXT:    vpmaxsd %zmm1, %zmm0, %zmm2 {%k2}
+; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm2[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
 ; CHECK-NEXT:    vpminsd %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vpmaxsd %zmm2, %zmm0, %zmm0
-; CHECK-NEXT:    movb $-52, %al
-; CHECK-NEXT:    kmovw %eax, %k3
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k3}
-; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm1[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
-; CHECK-NEXT:    vpminsd %zmm1, %zmm0, %zmm2
-; CHECK-NEXT:    vpmaxsd %zmm1, %zmm0, %zmm0
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm2 {%k2}
-; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm2[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
+; CHECK-NEXT:    vshufps {{.*#+}} zmm2 = zmm1[0,1],zmm0[2,3],zmm1[4,5],zmm0[6,7],zmm1[8,9],zmm0[10,11],zmm1[12,13],zmm0[14,15]
+; CHECK-NEXT:    vshufps {{.*#+}} zmm0 = zmm1[1,0],zmm0[3,2],zmm1[5,4],zmm0[7,6],zmm1[9,8],zmm0[11,10],zmm1[13,12],zmm0[15,14]
 ; CHECK-NEXT:    vpminsd %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vpmaxsd %zmm2, %zmm0, %zmm1 {%k1}
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm1[3,2,1,0,7,6,5,4,11,10,9,8,15,14,13,12]
 ; CHECK-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[4,5,6,7,0,1,2,3]
 ; CHECK-NEXT:    vpminsd %zmm1, %zmm0, %zmm2
 ; CHECK-NEXT:    vpmaxsd %zmm1, %zmm0, %zmm0
 ; CHECK-NEXT:    vshufi64x2 {{.*#+}} zmm1 = zmm2[2,3,0,1],zmm0[6,7,4,5]
 ; CHECK-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm2[0,1,2,3],zmm0[4,5,6,7]
 ; CHECK-NEXT:    vpminsd %zmm0, %zmm1, %zmm2
-; CHECK-NEXT:    vpmaxsd %zmm0, %zmm1, %zmm0
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm2 {%k3}
+; CHECK-NEXT:    vpmaxsd %zmm0, %zmm1, %zmm2 {%k2}
 ; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm2[2,3,0,1,6,7,4,5,10,11,8,9,14,15,12,13]
 ; CHECK-NEXT:    vpminsd %zmm2, %zmm0, %zmm1
 ; CHECK-NEXT:    vpmaxsd %zmm2, %zmm0, %zmm0
-; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm1 {%k2}
-; CHECK-NEXT:    vpshufd {{.*#+}} zmm0 = zmm1[1,0,3,2,5,4,7,6,9,8,11,10,13,12,15,14]
-; CHECK-NEXT:    vpminsd %zmm1, %zmm0, %zmm2
-; CHECK-NEXT:    vpmaxsd %zmm1, %zmm0, %zmm2 {%k1}
-; CHECK-NEXT:    vmovdqu64 %zmm2, (%rdi)
+; CHECK-NEXT:    vshufps {{.*#+}} zmm2 = zmm1[0,1],zmm0[2,3],zmm1[4,5],zmm0[6,7],zmm1[8,9],zmm0[10,11],zmm1[12,13],zmm0[14,15]
+; CHECK-NEXT:    vshufps {{.*#+}} zmm0 = zmm1[1,0],zmm0[3,2],zmm1[5,4],zmm0[7,6],zmm1[9,8],zmm0[11,10],zmm1[13,12],zmm0[15,14]
+; CHECK-NEXT:    vpminsd %zmm2, %zmm0, %zmm1
+; CHECK-NEXT:    vpmaxsd %zmm2, %zmm0, %zmm1 {%k1}
+; CHECK-NEXT:    vmovdqu64 %zmm1, (%rdi)
 ; CHECK-NEXT:    vzeroupper
 ; CHECK-NEXT:    retq
   %2 = load <16 x i32>, ptr %0, align 1
   %3 = shufflevector <16 x i32> %2, <16 x i32> poison, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 9, i32 8, i32 11, i32 10, i32 13, i32 12, i32 15, i32 14>
   %4 = tail call <16 x i32> @llvm.smin.v16i32(<16 x i32> %3, <16 x i32> %2) #2
   %5 = tail call <16 x i32> @llvm.smax.v16i32(<16 x i32> %3, <16 x i32> %2) #2
   %6 = shufflevector <16 x i32> %4, <16 x i32> %5, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
   %7 = shufflevector <16 x i32> %6, <16 x i32> poison, <16 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4, i32 11, i32 10, i32 9, i32 8, i32 15, i32 14, i32 13, i32 12>