This is an archive of the discontinued LLVM Phabricator instance.

Differential D42313

[X86] Don't reorder (srl (and X, C1), C2) if (and X, C1) can be matched as a movzx
ClosedPublic

Authored by craig.topper on Jan 19 2018, 12:47 PM.

Details

Reviewers
spatel
RKSimon
Commits
rGc92edd994ea6: [X86] Don't reorder (srl (and X, C1), C2) if (and X, C1) can be matched as a…
rL323175: [X86] Don't reorder (srl (and X, C1), C2) if (and X, C1) can be matched as a…
Summary

If we can match as a zero extend there's no need to flip the order to get an encoding benefit. As movzx is 3 bytes with independent source/dest registers. The shortest 'and' we could make is also 3 bytes unless we get lucky in the register allocator and its on AL/AX/EAX which have a 2 byte encoding.

This patch was more impressive before r322957 went in. It removed some of the same Ands that got deleted by that patch.

Diff Detail

Event Timeline

craig.topper created this revision.Jan 19 2018, 12:47 PM
Diffusion mentioned this in rL323030: [X86] Add test cases for failures to use movzx due to various issues with….Jan 19 2018, 11:53 PM
craig.topper updated this revision to Diff 130750.Jan 19 2018, 11:55 PM

Rebase after adding zext-demanded.ll test

spatel accepted this revision.Jan 22 2018, 6:57 AM

LGTM.

This revision is now accepted and ready to land.Jan 22 2018, 6:57 AM
Closed by commit rL323175: [X86] Don't reorder (srl (and X, C1), C2) if (and X, C1) can be matched as a… (authored by ctopper). · Explain WhyJan 22 2018, 9:47 PM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
lib/
Target/
X86/
8 lines
test/
CodeGen/
X86/
1470 lines
600 lines

Diff 130676

lib/Target/X86/X86ISelLowering.cpp

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 32,896 Lines • ▼ Show 20 Linesstatic SDValue combineShiftRightLogical(SDNode *N, SelectionDAG &DAG) {
auto *AndC = dyn_cast<ConstantSDNode>(N0.getOperand(1)); auto *AndC = dyn_cast<ConstantSDNode>(N0.getOperand(1));
if (!ShiftC || !AndC) if (!ShiftC || !AndC)
return SDValue(); return SDValue();
// If we can shrink the constant mask below 8-bits or 32-bits, then this // If we can shrink the constant mask below 8-bits or 32-bits, then this
// transform should reduce code size. It may also enable secondary transforms // transform should reduce code size. It may also enable secondary transforms
// from improved known-bits analysis or instruction selection. // from improved known-bits analysis or instruction selection.
APInt MaskVal = AndC->getAPIntValue(); APInt MaskVal = AndC->getAPIntValue();
// If this can be matched by a zero extend, don't optimize.
if (MaskVal.isMask()) {
unsigned TO = MaskVal.countTrailingOnes();
if (TO >= 8 && isPowerOf2_32(TO))
return SDValue();
}
APInt NewMaskVal = MaskVal.lshr(ShiftC->getAPIntValue()); APInt NewMaskVal = MaskVal.lshr(ShiftC->getAPIntValue());
unsigned OldMaskSize = MaskVal.getMinSignedBits(); unsigned OldMaskSize = MaskVal.getMinSignedBits();
unsigned NewMaskSize = NewMaskVal.getMinSignedBits(); unsigned NewMaskSize = NewMaskVal.getMinSignedBits();
if ((OldMaskSize > 8 && NewMaskSize <= 8) || if ((OldMaskSize > 8 && NewMaskSize <= 8) ||
(OldMaskSize > 32 && NewMaskSize <= 32)) { (OldMaskSize > 32 && NewMaskSize <= 32)) {
// srl (and X, AndC), ShiftC --> and (srl X, ShiftC), (AndC >> ShiftC) // srl (and X, AndC), ShiftC --> and (srl X, ShiftC), (AndC >> ShiftC)
SDLoc DL(N); SDLoc DL(N);
SDValue NewMask = DAG.getConstant(NewMaskVal, DL, VT); SDValue NewMask = DAG.getConstant(NewMaskVal, DL, VT);
▲ Show 20 Lines • Show All 6,037 Lines • Show Last 20 Lines

test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll

Show First 20 Lines • Show All 106 Lines • ▼ Show 20 Lines
; X32-LABEL: test_mm512_mask_set1_epi8: ; X32-LABEL: test_mm512_mask_set1_epi8:
; X32: # %bb.0: # %entry ; X32: # %bb.0: # %entry
; X32-NEXT: pushl %ebx ; X32-NEXT: pushl %ebx
; X32-NEXT: .cfi_def_cfa_offset 8 ; X32-NEXT: .cfi_def_cfa_offset 8
; X32-NEXT: pushl %esi ; X32-NEXT: pushl %esi
; X32-NEXT: .cfi_def_cfa_offset 12 ; X32-NEXT: .cfi_def_cfa_offset 12
; X32-NEXT: .cfi_offset %esi, -12 ; X32-NEXT: .cfi_offset %esi, -12
; X32-NEXT: .cfi_offset %ebx, -8 ; X32-NEXT: .cfi_offset %ebx, -8
; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-NEXT: kmovd %ecx, %k0 ; X32-NEXT: kmovd %eax, %k0
; X32-NEXT: kshiftrq $1, %k0, %k1 ; X32-NEXT: kshiftrq $1, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $2, %al ; X32-NEXT: andb $2, %cl
; X32-NEXT: shrb %al ; X32-NEXT: shrb %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $62, %k1, %k1 ; X32-NEXT: kshiftrq $62, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $2, %k0, %k1 ; X32-NEXT: kshiftrq $2, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %al ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %eax, %edx ; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $61, %k1, %k1 ; X32-NEXT: kshiftrq $61, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $3, %k0, %k1 ; X32-NEXT: kshiftrq $3, %k0, %k1
; X32-NEXT: shrb $3, %al ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $60, %k1, %k1 ; X32-NEXT: kshiftrq $60, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $4, %k0, %k1 ; X32-NEXT: kshiftrq $4, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $4, %al ; X32-NEXT: shrb $4, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $59, %k1, %k1 ; X32-NEXT: kshiftrq $59, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $5, %k0, %k1 ; X32-NEXT: kshiftrq $5, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $5, %al ; X32-NEXT: shrb $5, %cl
; X32-NEXT: andb $1, %al ; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $58, %k1, %k1 ; X32-NEXT: kshiftrq $58, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $6, %k0, %k1 ; X32-NEXT: kshiftrq $6, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $6, %al ; X32-NEXT: shrb $6, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $57, %k1, %k1 ; X32-NEXT: kshiftrq $57, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $7, %k0, %k1 ; X32-NEXT: kshiftrq $7, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $7, %al ; X32-NEXT: shrb $7, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $56, %k1, %k1 ; X32-NEXT: kshiftrq $56, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $8, %k0, %k1 ; X32-NEXT: kshiftrq $8, %k0, %k1
; X32-NEXT: movb %ch, %al ; X32-NEXT: movb %ah, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $55, %k1, %k1 ; X32-NEXT: kshiftrq $55, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $9, %k0, %k1 ; X32-NEXT: kshiftrq $9, %k0, %k1
; X32-NEXT: andb $2, %al ; X32-NEXT: andb $2, %cl
; X32-NEXT: shrb %al ; X32-NEXT: shrb %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $54, %k1, %k1 ; X32-NEXT: kshiftrq $54, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $10, %k0, %k1 ; X32-NEXT: kshiftrq $10, %k0, %k1
; X32-NEXT: movb %ch, %al ; X32-NEXT: movb %ah, %cl
; X32-NEXT: andb $15, %al ; X32-NEXT: movl %eax, %esi
; X32-NEXT: movl %eax, %edx ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: movzwl %si, %edx
; X32-NEXT: movl %edx, %ecx
; X32-NEXT: shrl $12, %ecx
; X32-NEXT: kmovd %ecx, %k3
; X32-NEXT: movl %esi, %ecx
; X32-NEXT: shrl $13, %ecx
; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %edx, %ecx
; X32-NEXT: shrl $14, %ecx
; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: shrl $16, %eax
; X32-NEXT: movl %eax, %ebx
; X32-NEXT: andb $2, %bl
; X32-NEXT: shrb %bl
; X32-NEXT: kmovd %ebx, %k6
; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %ebx
; X32-NEXT: shrb $2, %bl
; X32-NEXT: kmovd %ebx, %k7
; X32-NEXT: movl {{[0-9]+}}(%esp), %ebx
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $53, %k1, %k1 ; X32-NEXT: kshiftrq $53, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $11, %k0, %k1 ; X32-NEXT: kshiftrq $11, %k0, %k1
; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $12, %eax
; X32-NEXT: andl $15, %eax
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $13, %eax
; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %eax, %k3
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $14, %eax
; X32-NEXT: andl $3, %eax
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $15, %eax
; X32-NEXT: andl $1, %eax
; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrl $16, %edx
; X32-NEXT: movl %edx, %eax
; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %al
; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %edx, %ebx
; X32-NEXT: andb $15, %bl
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $52, %k1, %k1 ; X32-NEXT: kshiftrq $52, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $12, %k0, %k1 ; X32-NEXT: kshiftrq $12, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k3, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $51, %k1, %k1 ; X32-NEXT: kshiftrq $51, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $13, %k0, %k1 ; X32-NEXT: kshiftrq $13, %k0, %k1
; X32-NEXT: kxorq %k3, %k1, %k1 ; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $50, %k1, %k1 ; X32-NEXT: kshiftrq $50, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $14, %k0, %k1 ; X32-NEXT: kshiftrq $14, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1 ; X32-NEXT: kxorq %k5, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $49, %k1, %k1 ; X32-NEXT: kshiftrq $49, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $15, %k0, %k1 ; X32-NEXT: kshiftrq $15, %k0, %k1
; X32-NEXT: kxorq %k5, %k1, %k1 ; X32-NEXT: shrl $15, %edx
; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $48, %k1, %k1 ; X32-NEXT: kshiftrq $48, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $16, %k0, %k1 ; X32-NEXT: kshiftrq $16, %k0, %k1
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $47, %k1, %k1 ; X32-NEXT: kshiftrq $47, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $17, %k0, %k1 ; X32-NEXT: kshiftrq $17, %k0, %k1
; X32-NEXT: kxorq %k6, %k1, %k1 ; X32-NEXT: kxorq %k6, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $46, %k1, %k1 ; X32-NEXT: kshiftrq $46, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $18, %k0, %k1 ; X32-NEXT: kshiftrq $18, %k0, %k1
; X32-NEXT: kxorq %k7, %k1, %k1 ; X32-NEXT: kxorq %k7, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $45, %k1, %k1 ; X32-NEXT: kshiftrq $45, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $19, %k0, %k1 ; X32-NEXT: kshiftrq $19, %k0, %k1
; X32-NEXT: shrb $3, %bl ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $44, %k1, %k1 ; X32-NEXT: kshiftrq $44, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $20, %k0, %k1 ; X32-NEXT: kshiftrq $20, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $4, %bl ; X32-NEXT: shrb $4, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $43, %k1, %k1 ; X32-NEXT: kshiftrq $43, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $21, %k0, %k1 ; X32-NEXT: kshiftrq $21, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $5, %bl ; X32-NEXT: shrb $5, %cl
; X32-NEXT: andb $1, %bl ; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $42, %k1, %k1 ; X32-NEXT: kshiftrq $42, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $22, %k0, %k1 ; X32-NEXT: kshiftrq $22, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $6, %bl ; X32-NEXT: shrb $6, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $41, %k1, %k1 ; X32-NEXT: kshiftrq $41, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $23, %k0, %k1 ; X32-NEXT: kshiftrq $23, %k0, %k1
; X32-NEXT: shrb $7, %dl ; X32-NEXT: shrb $7, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $40, %k1, %k1 ; X32-NEXT: kshiftrq $40, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $24, %k0, %k1 ; X32-NEXT: kshiftrq $24, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %esi, %edx
; X32-NEXT: shrl $24, %edx ; X32-NEXT: shrl $24, %edx
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $39, %k1, %k1 ; X32-NEXT: kshiftrq $39, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $25, %k0, %k1 ; X32-NEXT: kshiftrq $25, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: andb $2, %bl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %bl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $38, %k1, %k1 ; X32-NEXT: kshiftrq $38, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $26, %k0, %k1 ; X32-NEXT: kshiftrq $26, %k0, %k1
; X32-NEXT: andb $15, %dl ; X32-NEXT: andb $15, %dl
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: shrb $2, %bl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $37, %k1, %k1 ; X32-NEXT: kshiftrq $37, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $27, %k0, %k1 ; X32-NEXT: kshiftrq $27, %k0, %k1
; X32-NEXT: shrb $3, %dl ; X32-NEXT: shrb $3, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $36, %k1, %k1 ; X32-NEXT: kshiftrq $36, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $28, %k0, %k1 ; X32-NEXT: kshiftrq $28, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %esi, %ecx
; X32-NEXT: shrl $28, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: shrl $28, %eax
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $35, %k1, %k1 ; X32-NEXT: kshiftrq $35, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $29, %k0, %k1 ; X32-NEXT: kshiftrq $29, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $29, %edx ; X32-NEXT: shrl $29, %eax
; X32-NEXT: andb $1, %dl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $34, %k1, %k1 ; X32-NEXT: kshiftrq $34, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $30, %k0, %k1 ; X32-NEXT: kshiftrq $30, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $30, %edx ; X32-NEXT: shrl $30, %eax
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $33, %k1, %k1 ; X32-NEXT: kshiftrq $33, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $31, %k0, %k1 ; X32-NEXT: kshiftrq $31, %k0, %k1
; X32-NEXT: shrl $31, %ecx ; X32-NEXT: shrl $31, %ecx
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $32, %k1, %k1 ; X32-NEXT: kshiftrq $32, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $32, %k0, %k1 ; X32-NEXT: kshiftrq $32, %k0, %k1
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ebx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $31, %k1, %k1 ; X32-NEXT: kshiftrq $31, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $33, %k0, %k1 ; X32-NEXT: kshiftrq $33, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: andb $2, %cl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %cl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $30, %k1, %k1 ; X32-NEXT: kshiftrq $30, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $34, %k0, %k1 ; X32-NEXT: kshiftrq $34, %k0, %k1
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: andb $15, %al
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %cl ; X32-NEXT: shrb $2, %cl
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $29, %k1, %k1 ; X32-NEXT: kshiftrq $29, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $35, %k0, %k1 ; X32-NEXT: kshiftrq $35, %k0, %k1
; X32-NEXT: shrb $3, %cl ; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $28, %k1, %k1 ; X32-NEXT: kshiftrq $28, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $36, %k0, %k1 ; X32-NEXT: kshiftrq $36, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $4, %cl ; X32-NEXT: shrb $4, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $27, %k1, %k1 ; X32-NEXT: kshiftrq $27, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $37, %k0, %k1 ; X32-NEXT: kshiftrq $37, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $5, %cl ; X32-NEXT: shrb $5, %al
; X32-NEXT: andb $1, %cl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $26, %k1, %k1 ; X32-NEXT: kshiftrq $26, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $38, %k0, %k1 ; X32-NEXT: kshiftrq $38, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $6, %cl ; X32-NEXT: shrb $6, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $25, %k1, %k1 ; X32-NEXT: kshiftrq $25, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $39, %k0, %k1 ; X32-NEXT: kshiftrq $39, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $7, %cl ; X32-NEXT: shrb $7, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $24, %k1, %k1 ; X32-NEXT: kshiftrq $24, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $40, %k0, %k1 ; X32-NEXT: kshiftrq $40, %k0, %k1
; X32-NEXT: movb %ah, %cl ; X32-NEXT: movb %bh, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: andb $2, %cl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %cl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movb %ah, %cl ; X32-NEXT: movb %bh, %al
; X32-NEXT: andb $15, %cl ; X32-NEXT: andb $15, %al
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k3
; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrl $13, %ecx
; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $2, %cl
; X32-NEXT: kmovd %ecx, %k3
; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $13, %eax
; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %ebx, %ecx
; X32-NEXT: shrl $16, %ecx ; X32-NEXT: shrl $16, %ecx
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: andb $2, %dl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %dl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %edx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %edx
; X32-NEXT: andb $15, %dl ; X32-NEXT: andb $15, %dl
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: shrb $2, %bl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %ebx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $23, %k1, %k1 ; X32-NEXT: kshiftrq $23, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $41, %k0, %k1 ; X32-NEXT: kshiftrq $41, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $22, %k1, %k1 ; X32-NEXT: kshiftrq $22, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $42, %k0, %k1 ; X32-NEXT: kshiftrq $42, %k0, %k1
; X32-NEXT: kxorq %k3, %k1, %k1 ; X32-NEXT: kxorq %k3, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: movzwl %bx, %eax
; X32-NEXT: kshiftrq $21, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $43, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: movl %eax, %esi ; X32-NEXT: movl %eax, %esi
; X32-NEXT: shrl $12, %esi ; X32-NEXT: shrl $12, %esi
; X32-NEXT: andl $15, %esi
; X32-NEXT: kmovd %esi, %k2 ; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: movl %eax, %esi ; X32-NEXT: movl %eax, %esi
; X32-NEXT: shrl $14, %esi ; X32-NEXT: shrl $14, %esi
; X32-NEXT: andl $3, %esi
; X32-NEXT: kmovd %esi, %k3 ; X32-NEXT: kmovd %esi, %k3
; X32-NEXT: movl %eax, %esi ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: shrl $15, %esi ; X32-NEXT: kshiftrq $21, %k1, %k1
; X32-NEXT: andl $1, %esi ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kmovd %esi, %k4 ; X32-NEXT: kshiftrq $43, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: shrl $15, %eax
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $20, %k1, %k1 ; X32-NEXT: kshiftrq $20, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $44, %k0, %k1 ; X32-NEXT: kshiftrq $44, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $19, %k1, %k1 ; X32-NEXT: kshiftrq $19, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
Show All 31 Lines
; X32-NEXT: kshiftrq $51, %k0, %k1 ; X32-NEXT: kshiftrq $51, %k0, %k1
; X32-NEXT: shrb $3, %dl ; X32-NEXT: shrb $3, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $12, %k1, %k1 ; X32-NEXT: kshiftrq $12, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k4 ; X32-NEXT: kxorq %k0, %k1, %k4
; X32-NEXT: kshiftrq $52, %k4, %k0 ; X32-NEXT: kshiftrq $52, %k4, %k0
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $4, %dl ; X32-NEXT: shrb $4, %al
; X32-NEXT: kmovd %edx, %k1 ; X32-NEXT: kmovd %eax, %k1
; X32-NEXT: kxorq %k1, %k0, %k5 ; X32-NEXT: kxorq %k1, %k0, %k5
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $5, %dl ; X32-NEXT: shrb $5, %al
; X32-NEXT: andb $1, %dl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %edx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $6, %dl ; X32-NEXT: shrb $6, %al
; X32-NEXT: kmovd %edx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: shrb $7, %cl ; X32-NEXT: shrb $7, %cl
; X32-NEXT: kmovd %ecx, %k0 ; X32-NEXT: kmovd %ecx, %k0
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %ecx
; X32-NEXT: shrl $24, %ecx ; X32-NEXT: shrl $24, %ecx
; X32-NEXT: kmovd %ecx, %k1 ; X32-NEXT: kmovd %ecx, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: andb $2, %dl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %dl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: andb $15, %cl ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %edx, %k3 ; X32-NEXT: kmovd %eax, %k3
; X32-NEXT: kshiftlq $63, %k5, %k5 ; X32-NEXT: kshiftlq $63, %k5, %k5
; X32-NEXT: kshiftrq $11, %k5, %k5 ; X32-NEXT: kshiftrq $11, %k5, %k5
; X32-NEXT: kxorq %k4, %k5, %k4 ; X32-NEXT: kxorq %k4, %k5, %k4
; X32-NEXT: kshiftrq $53, %k4, %k5 ; X32-NEXT: kshiftrq $53, %k4, %k5
; X32-NEXT: kxorq %k6, %k5, %k5 ; X32-NEXT: kxorq %k6, %k5, %k5
; X32-NEXT: kshiftlq $63, %k5, %k5 ; X32-NEXT: kshiftlq $63, %k5, %k5
; X32-NEXT: kshiftrq $10, %k5, %k5 ; X32-NEXT: kshiftrq $10, %k5, %k5
; X32-NEXT: kxorq %k4, %k5, %k5 ; X32-NEXT: kxorq %k4, %k5, %k5
; X32-NEXT: kshiftrq $54, %k5, %k4 ; X32-NEXT: kshiftrq $54, %k5, %k4
; X32-NEXT: kxorq %k7, %k4, %k6 ; X32-NEXT: kxorq %k7, %k4, %k6
; X32-NEXT: shrb $3, %cl ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k4 ; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $29, %ecx ; X32-NEXT: shrl $29, %eax
; X32-NEXT: andb $1, %cl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %ecx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: kshiftlq $63, %k6, %k6 ; X32-NEXT: kshiftlq $63, %k6, %k6
; X32-NEXT: kshiftrq $9, %k6, %k6 ; X32-NEXT: kshiftrq $9, %k6, %k6
; X32-NEXT: kxorq %k5, %k6, %k5 ; X32-NEXT: kxorq %k5, %k6, %k5
; X32-NEXT: kshiftrq $55, %k5, %k6 ; X32-NEXT: kshiftrq $55, %k5, %k6
; X32-NEXT: kxorq %k0, %k6, %k0 ; X32-NEXT: kxorq %k0, %k6, %k0
; X32-NEXT: kshiftlq $63, %k0, %k0 ; X32-NEXT: kshiftlq $63, %k0, %k0
; X32-NEXT: kshiftrq $8, %k0, %k0 ; X32-NEXT: kshiftrq $8, %k0, %k0
; X32-NEXT: kxorq %k5, %k0, %k0 ; X32-NEXT: kxorq %k5, %k0, %k0
; X32-NEXT: kshiftrq $56, %k0, %k5 ; X32-NEXT: kshiftrq $56, %k0, %k5
; X32-NEXT: kxorq %k1, %k5, %k1 ; X32-NEXT: kxorq %k1, %k5, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $28, %ecx ; X32-NEXT: shrl $28, %eax
; X32-NEXT: kmovd %ecx, %k5 ; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $30, %ecx ; X32-NEXT: shrl $30, %eax
; X32-NEXT: kmovd %ecx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $7, %k1, %k1 ; X32-NEXT: kshiftrq $7, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $57, %k0, %k1 ; X32-NEXT: kshiftrq $57, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $6, %k1, %k1 ; X32-NEXT: kshiftrq $6, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
Show All 14 Lines
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $61, %k0, %k1 ; X32-NEXT: kshiftrq $61, %k0, %k1
; X32-NEXT: kxorq %k7, %k1, %k1 ; X32-NEXT: kxorq %k7, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $2, %k1, %k1 ; X32-NEXT: kshiftrq $2, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $62, %k0, %k1 ; X32-NEXT: kshiftrq $62, %k0, %k1
; X32-NEXT: kxorq %k6, %k1, %k1 ; X32-NEXT: kxorq %k6, %k1, %k1
; X32-NEXT: shrl $31, %eax ; X32-NEXT: shrl $31, %ebx
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ebx, %k2
; X32-NEXT: movb {{[0-9]+}}(%esp), %al ; X32-NEXT: movb {{[0-9]+}}(%esp), %al
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $1, %k1, %k1 ; X32-NEXT: kshiftrq $1, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftlq $1, %k0, %k0 ; X32-NEXT: kshiftlq $1, %k0, %k0
; X32-NEXT: kshiftrq $1, %k0, %k0 ; X32-NEXT: kshiftrq $1, %k0, %k0
; X32-NEXT: kshiftlq $63, %k2, %k1 ; X32-NEXT: kshiftlq $63, %k2, %k1
; X32-NEXT: korq %k1, %k0, %k1 ; X32-NEXT: korq %k1, %k0, %k1
Show All 21 Lines
; X32-LABEL: test_mm512_maskz_set1_epi8: ; X32-LABEL: test_mm512_maskz_set1_epi8:
; X32: # %bb.0: # %entry ; X32: # %bb.0: # %entry
; X32-NEXT: pushl %ebx ; X32-NEXT: pushl %ebx
; X32-NEXT: .cfi_def_cfa_offset 8 ; X32-NEXT: .cfi_def_cfa_offset 8
; X32-NEXT: pushl %esi ; X32-NEXT: pushl %esi
; X32-NEXT: .cfi_def_cfa_offset 12 ; X32-NEXT: .cfi_def_cfa_offset 12
; X32-NEXT: .cfi_offset %esi, -12 ; X32-NEXT: .cfi_offset %esi, -12
; X32-NEXT: .cfi_offset %ebx, -8 ; X32-NEXT: .cfi_offset %ebx, -8
; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-NEXT: kmovd %ecx, %k0 ; X32-NEXT: kmovd %eax, %k0
; X32-NEXT: kshiftrq $1, %k0, %k1 ; X32-NEXT: kshiftrq $1, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $2, %al ; X32-NEXT: andb $2, %cl
; X32-NEXT: shrb %al ; X32-NEXT: shrb %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $62, %k1, %k1 ; X32-NEXT: kshiftrq $62, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $2, %k0, %k1 ; X32-NEXT: kshiftrq $2, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %al ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %eax, %edx ; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $61, %k1, %k1 ; X32-NEXT: kshiftrq $61, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $3, %k0, %k1 ; X32-NEXT: kshiftrq $3, %k0, %k1
; X32-NEXT: shrb $3, %al ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $60, %k1, %k1 ; X32-NEXT: kshiftrq $60, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $4, %k0, %k1 ; X32-NEXT: kshiftrq $4, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $4, %al ; X32-NEXT: shrb $4, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $59, %k1, %k1 ; X32-NEXT: kshiftrq $59, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $5, %k0, %k1 ; X32-NEXT: kshiftrq $5, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $5, %al ; X32-NEXT: shrb $5, %cl
; X32-NEXT: andb $1, %al ; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $58, %k1, %k1 ; X32-NEXT: kshiftrq $58, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $6, %k0, %k1 ; X32-NEXT: kshiftrq $6, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $6, %al ; X32-NEXT: shrb $6, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $57, %k1, %k1 ; X32-NEXT: kshiftrq $57, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $7, %k0, %k1 ; X32-NEXT: kshiftrq $7, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $7, %al ; X32-NEXT: shrb $7, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $56, %k1, %k1 ; X32-NEXT: kshiftrq $56, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $8, %k0, %k1 ; X32-NEXT: kshiftrq $8, %k0, %k1
; X32-NEXT: movb %ch, %al ; X32-NEXT: movb %ah, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $55, %k1, %k1 ; X32-NEXT: kshiftrq $55, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $9, %k0, %k1 ; X32-NEXT: kshiftrq $9, %k0, %k1
; X32-NEXT: andb $2, %al ; X32-NEXT: andb $2, %cl
; X32-NEXT: shrb %al ; X32-NEXT: shrb %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $54, %k1, %k1 ; X32-NEXT: kshiftrq $54, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $10, %k0, %k1 ; X32-NEXT: kshiftrq $10, %k0, %k1
; X32-NEXT: movb %ch, %al ; X32-NEXT: movb %ah, %cl
; X32-NEXT: andb $15, %al ; X32-NEXT: movl %eax, %esi
; X32-NEXT: movl %eax, %edx ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: movzwl %si, %edx
; X32-NEXT: movl %edx, %ecx
; X32-NEXT: shrl $12, %ecx
; X32-NEXT: kmovd %ecx, %k3
; X32-NEXT: movl %esi, %ecx
; X32-NEXT: shrl $13, %ecx
; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %edx, %ecx
; X32-NEXT: shrl $14, %ecx
; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: shrl $16, %eax
; X32-NEXT: movl %eax, %ebx
; X32-NEXT: andb $2, %bl
; X32-NEXT: shrb %bl
; X32-NEXT: kmovd %ebx, %k6
; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %ebx
; X32-NEXT: shrb $2, %bl
; X32-NEXT: kmovd %ebx, %k7
; X32-NEXT: movl {{[0-9]+}}(%esp), %ebx
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $53, %k1, %k1 ; X32-NEXT: kshiftrq $53, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $11, %k0, %k1 ; X32-NEXT: kshiftrq $11, %k0, %k1
; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $12, %eax
; X32-NEXT: andl $15, %eax
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $13, %eax
; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %eax, %k3
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $14, %eax
; X32-NEXT: andl $3, %eax
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $15, %eax
; X32-NEXT: andl $1, %eax
; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrl $16, %edx
; X32-NEXT: movl %edx, %eax
; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %al
; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %edx, %ebx
; X32-NEXT: andb $15, %bl
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $52, %k1, %k1 ; X32-NEXT: kshiftrq $52, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $12, %k0, %k1 ; X32-NEXT: kshiftrq $12, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k3, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $51, %k1, %k1 ; X32-NEXT: kshiftrq $51, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $13, %k0, %k1 ; X32-NEXT: kshiftrq $13, %k0, %k1
; X32-NEXT: kxorq %k3, %k1, %k1 ; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $50, %k1, %k1 ; X32-NEXT: kshiftrq $50, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $14, %k0, %k1 ; X32-NEXT: kshiftrq $14, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1 ; X32-NEXT: kxorq %k5, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $49, %k1, %k1 ; X32-NEXT: kshiftrq $49, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $15, %k0, %k1 ; X32-NEXT: kshiftrq $15, %k0, %k1
; X32-NEXT: kxorq %k5, %k1, %k1 ; X32-NEXT: shrl $15, %edx
; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $48, %k1, %k1 ; X32-NEXT: kshiftrq $48, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $16, %k0, %k1 ; X32-NEXT: kshiftrq $16, %k0, %k1
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $47, %k1, %k1 ; X32-NEXT: kshiftrq $47, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $17, %k0, %k1 ; X32-NEXT: kshiftrq $17, %k0, %k1
; X32-NEXT: kxorq %k6, %k1, %k1 ; X32-NEXT: kxorq %k6, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $46, %k1, %k1 ; X32-NEXT: kshiftrq $46, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $18, %k0, %k1 ; X32-NEXT: kshiftrq $18, %k0, %k1
; X32-NEXT: kxorq %k7, %k1, %k1 ; X32-NEXT: kxorq %k7, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $45, %k1, %k1 ; X32-NEXT: kshiftrq $45, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $19, %k0, %k1 ; X32-NEXT: kshiftrq $19, %k0, %k1
; X32-NEXT: shrb $3, %bl ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $44, %k1, %k1 ; X32-NEXT: kshiftrq $44, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $20, %k0, %k1 ; X32-NEXT: kshiftrq $20, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $4, %bl ; X32-NEXT: shrb $4, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $43, %k1, %k1 ; X32-NEXT: kshiftrq $43, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $21, %k0, %k1 ; X32-NEXT: kshiftrq $21, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $5, %bl ; X32-NEXT: shrb $5, %cl
; X32-NEXT: andb $1, %bl ; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $42, %k1, %k1 ; X32-NEXT: kshiftrq $42, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $22, %k0, %k1 ; X32-NEXT: kshiftrq $22, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $6, %bl ; X32-NEXT: shrb $6, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $41, %k1, %k1 ; X32-NEXT: kshiftrq $41, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $23, %k0, %k1 ; X32-NEXT: kshiftrq $23, %k0, %k1
; X32-NEXT: shrb $7, %dl ; X32-NEXT: shrb $7, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $40, %k1, %k1 ; X32-NEXT: kshiftrq $40, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $24, %k0, %k1 ; X32-NEXT: kshiftrq $24, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %esi, %edx
; X32-NEXT: shrl $24, %edx ; X32-NEXT: shrl $24, %edx
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $39, %k1, %k1 ; X32-NEXT: kshiftrq $39, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $25, %k0, %k1 ; X32-NEXT: kshiftrq $25, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: andb $2, %bl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %bl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $38, %k1, %k1 ; X32-NEXT: kshiftrq $38, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $26, %k0, %k1 ; X32-NEXT: kshiftrq $26, %k0, %k1
; X32-NEXT: andb $15, %dl ; X32-NEXT: andb $15, %dl
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: shrb $2, %bl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $37, %k1, %k1 ; X32-NEXT: kshiftrq $37, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $27, %k0, %k1 ; X32-NEXT: kshiftrq $27, %k0, %k1
; X32-NEXT: shrb $3, %dl ; X32-NEXT: shrb $3, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $36, %k1, %k1 ; X32-NEXT: kshiftrq $36, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $28, %k0, %k1 ; X32-NEXT: kshiftrq $28, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %esi, %ecx
; X32-NEXT: shrl $28, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: shrl $28, %eax
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $35, %k1, %k1 ; X32-NEXT: kshiftrq $35, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $29, %k0, %k1 ; X32-NEXT: kshiftrq $29, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $29, %edx ; X32-NEXT: shrl $29, %eax
; X32-NEXT: andb $1, %dl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $34, %k1, %k1 ; X32-NEXT: kshiftrq $34, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $30, %k0, %k1 ; X32-NEXT: kshiftrq $30, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $30, %edx ; X32-NEXT: shrl $30, %eax
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $33, %k1, %k1 ; X32-NEXT: kshiftrq $33, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $31, %k0, %k1 ; X32-NEXT: kshiftrq $31, %k0, %k1
; X32-NEXT: shrl $31, %ecx ; X32-NEXT: shrl $31, %ecx
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $32, %k1, %k1 ; X32-NEXT: kshiftrq $32, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $32, %k0, %k1 ; X32-NEXT: kshiftrq $32, %k0, %k1
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ebx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $31, %k1, %k1 ; X32-NEXT: kshiftrq $31, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $33, %k0, %k1 ; X32-NEXT: kshiftrq $33, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: andb $2, %cl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %cl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $30, %k1, %k1 ; X32-NEXT: kshiftrq $30, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $34, %k0, %k1 ; X32-NEXT: kshiftrq $34, %k0, %k1
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: andb $15, %al
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %cl ; X32-NEXT: shrb $2, %cl
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $29, %k1, %k1 ; X32-NEXT: kshiftrq $29, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $35, %k0, %k1 ; X32-NEXT: kshiftrq $35, %k0, %k1
; X32-NEXT: shrb $3, %cl ; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $28, %k1, %k1 ; X32-NEXT: kshiftrq $28, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $36, %k0, %k1 ; X32-NEXT: kshiftrq $36, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $4, %cl ; X32-NEXT: shrb $4, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $27, %k1, %k1 ; X32-NEXT: kshiftrq $27, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $37, %k0, %k1 ; X32-NEXT: kshiftrq $37, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $5, %cl ; X32-NEXT: shrb $5, %al
; X32-NEXT: andb $1, %cl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $26, %k1, %k1 ; X32-NEXT: kshiftrq $26, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $38, %k0, %k1 ; X32-NEXT: kshiftrq $38, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $6, %cl ; X32-NEXT: shrb $6, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $25, %k1, %k1 ; X32-NEXT: kshiftrq $25, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $39, %k0, %k1 ; X32-NEXT: kshiftrq $39, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $7, %cl ; X32-NEXT: shrb $7, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $24, %k1, %k1 ; X32-NEXT: kshiftrq $24, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $40, %k0, %k1 ; X32-NEXT: kshiftrq $40, %k0, %k1
; X32-NEXT: movb %ah, %cl ; X32-NEXT: movb %bh, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: andb $2, %cl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %cl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movb %ah, %cl ; X32-NEXT: movb %bh, %al
; X32-NEXT: andb $15, %cl ; X32-NEXT: andb $15, %al
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k3
; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrl $13, %ecx
; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $2, %cl
; X32-NEXT: kmovd %ecx, %k3
; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $13, %eax
; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %ebx, %ecx
; X32-NEXT: shrl $16, %ecx ; X32-NEXT: shrl $16, %ecx
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: andb $2, %dl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %dl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %edx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %edx
; X32-NEXT: andb $15, %dl ; X32-NEXT: andb $15, %dl
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: shrb $2, %bl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %ebx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $23, %k1, %k1 ; X32-NEXT: kshiftrq $23, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $41, %k0, %k1 ; X32-NEXT: kshiftrq $41, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $22, %k1, %k1 ; X32-NEXT: kshiftrq $22, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $42, %k0, %k1 ; X32-NEXT: kshiftrq $42, %k0, %k1
; X32-NEXT: kxorq %k3, %k1, %k1 ; X32-NEXT: kxorq %k3, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: movzwl %bx, %eax
; X32-NEXT: kshiftrq $21, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $43, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: movl %eax, %esi ; X32-NEXT: movl %eax, %esi
; X32-NEXT: shrl $12, %esi ; X32-NEXT: shrl $12, %esi
; X32-NEXT: andl $15, %esi
; X32-NEXT: kmovd %esi, %k2 ; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: movl %eax, %esi ; X32-NEXT: movl %eax, %esi
; X32-NEXT: shrl $14, %esi ; X32-NEXT: shrl $14, %esi
; X32-NEXT: andl $3, %esi
; X32-NEXT: kmovd %esi, %k3 ; X32-NEXT: kmovd %esi, %k3
; X32-NEXT: movl %eax, %esi ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: shrl $15, %esi ; X32-NEXT: kshiftrq $21, %k1, %k1
; X32-NEXT: andl $1, %esi ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kmovd %esi, %k4 ; X32-NEXT: kshiftrq $43, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: shrl $15, %eax
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $20, %k1, %k1 ; X32-NEXT: kshiftrq $20, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $44, %k0, %k1 ; X32-NEXT: kshiftrq $44, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $19, %k1, %k1 ; X32-NEXT: kshiftrq $19, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
Show All 31 Lines
; X32-NEXT: kshiftrq $51, %k0, %k1 ; X32-NEXT: kshiftrq $51, %k0, %k1
; X32-NEXT: shrb $3, %dl ; X32-NEXT: shrb $3, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $12, %k1, %k1 ; X32-NEXT: kshiftrq $12, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k4 ; X32-NEXT: kxorq %k0, %k1, %k4
; X32-NEXT: kshiftrq $52, %k4, %k0 ; X32-NEXT: kshiftrq $52, %k4, %k0
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $4, %dl ; X32-NEXT: shrb $4, %al
; X32-NEXT: kmovd %edx, %k1 ; X32-NEXT: kmovd %eax, %k1
; X32-NEXT: kxorq %k1, %k0, %k5 ; X32-NEXT: kxorq %k1, %k0, %k5
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $5, %dl ; X32-NEXT: shrb $5, %al
; X32-NEXT: andb $1, %dl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %edx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $6, %dl ; X32-NEXT: shrb $6, %al
; X32-NEXT: kmovd %edx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: shrb $7, %cl ; X32-NEXT: shrb $7, %cl
; X32-NEXT: kmovd %ecx, %k0 ; X32-NEXT: kmovd %ecx, %k0
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %ecx
; X32-NEXT: shrl $24, %ecx ; X32-NEXT: shrl $24, %ecx
; X32-NEXT: kmovd %ecx, %k1 ; X32-NEXT: kmovd %ecx, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: andb $2, %dl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %dl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: andb $15, %cl ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %edx, %k3 ; X32-NEXT: kmovd %eax, %k3
; X32-NEXT: kshiftlq $63, %k5, %k5 ; X32-NEXT: kshiftlq $63, %k5, %k5
; X32-NEXT: kshiftrq $11, %k5, %k5 ; X32-NEXT: kshiftrq $11, %k5, %k5
; X32-NEXT: kxorq %k4, %k5, %k4 ; X32-NEXT: kxorq %k4, %k5, %k4
; X32-NEXT: kshiftrq $53, %k4, %k5 ; X32-NEXT: kshiftrq $53, %k4, %k5
; X32-NEXT: kxorq %k6, %k5, %k5 ; X32-NEXT: kxorq %k6, %k5, %k5
; X32-NEXT: kshiftlq $63, %k5, %k5 ; X32-NEXT: kshiftlq $63, %k5, %k5
; X32-NEXT: kshiftrq $10, %k5, %k5 ; X32-NEXT: kshiftrq $10, %k5, %k5
; X32-NEXT: kxorq %k4, %k5, %k5 ; X32-NEXT: kxorq %k4, %k5, %k5
; X32-NEXT: kshiftrq $54, %k5, %k4 ; X32-NEXT: kshiftrq $54, %k5, %k4
; X32-NEXT: kxorq %k7, %k4, %k6 ; X32-NEXT: kxorq %k7, %k4, %k6
; X32-NEXT: shrb $3, %cl ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k4 ; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $29, %ecx ; X32-NEXT: shrl $29, %eax
; X32-NEXT: andb $1, %cl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %ecx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: kshiftlq $63, %k6, %k6 ; X32-NEXT: kshiftlq $63, %k6, %k6
; X32-NEXT: kshiftrq $9, %k6, %k6 ; X32-NEXT: kshiftrq $9, %k6, %k6
; X32-NEXT: kxorq %k5, %k6, %k5 ; X32-NEXT: kxorq %k5, %k6, %k5
; X32-NEXT: kshiftrq $55, %k5, %k6 ; X32-NEXT: kshiftrq $55, %k5, %k6
; X32-NEXT: kxorq %k0, %k6, %k0 ; X32-NEXT: kxorq %k0, %k6, %k0
; X32-NEXT: kshiftlq $63, %k0, %k0 ; X32-NEXT: kshiftlq $63, %k0, %k0
; X32-NEXT: kshiftrq $8, %k0, %k0 ; X32-NEXT: kshiftrq $8, %k0, %k0
; X32-NEXT: kxorq %k5, %k0, %k0 ; X32-NEXT: kxorq %k5, %k0, %k0
; X32-NEXT: kshiftrq $56, %k0, %k5 ; X32-NEXT: kshiftrq $56, %k0, %k5
; X32-NEXT: kxorq %k1, %k5, %k1 ; X32-NEXT: kxorq %k1, %k5, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $28, %ecx ; X32-NEXT: shrl $28, %eax
; X32-NEXT: kmovd %ecx, %k5 ; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $30, %ecx ; X32-NEXT: shrl $30, %eax
; X32-NEXT: kmovd %ecx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $7, %k1, %k1 ; X32-NEXT: kshiftrq $7, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $57, %k0, %k1 ; X32-NEXT: kshiftrq $57, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $6, %k1, %k1 ; X32-NEXT: kshiftrq $6, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
Show All 14 Lines
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $61, %k0, %k1 ; X32-NEXT: kshiftrq $61, %k0, %k1
; X32-NEXT: kxorq %k7, %k1, %k1 ; X32-NEXT: kxorq %k7, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $2, %k1, %k1 ; X32-NEXT: kshiftrq $2, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $62, %k0, %k1 ; X32-NEXT: kshiftrq $62, %k0, %k1
; X32-NEXT: kxorq %k6, %k1, %k1 ; X32-NEXT: kxorq %k6, %k1, %k1
; X32-NEXT: shrl $31, %eax ; X32-NEXT: shrl $31, %ebx
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ebx, %k2
; X32-NEXT: movb {{[0-9]+}}(%esp), %al ; X32-NEXT: movb {{[0-9]+}}(%esp), %al
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $1, %k1, %k1 ; X32-NEXT: kshiftrq $1, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftlq $1, %k0, %k0 ; X32-NEXT: kshiftlq $1, %k0, %k0
; X32-NEXT: kshiftrq $1, %k0, %k0 ; X32-NEXT: kshiftrq $1, %k0, %k0
; X32-NEXT: kshiftlq $63, %k2, %k1 ; X32-NEXT: kshiftlq $63, %k2, %k1
; X32-NEXT: korq %k1, %k0, %k1 ; X32-NEXT: korq %k1, %k0, %k1
▲ Show 20 Lines • Show All 502 Lines • ▼ Show 20 Lines
; X32-NEXT: movl %esp, %ebp ; X32-NEXT: movl %esp, %ebp
; X32-NEXT: .cfi_def_cfa_register %ebp ; X32-NEXT: .cfi_def_cfa_register %ebp
; X32-NEXT: pushl %ebx ; X32-NEXT: pushl %ebx
; X32-NEXT: pushl %esi ; X32-NEXT: pushl %esi
; X32-NEXT: andl $-8, %esp ; X32-NEXT: andl $-8, %esp
; X32-NEXT: subl $8, %esp ; X32-NEXT: subl $8, %esp
; X32-NEXT: .cfi_offset %esi, -16 ; X32-NEXT: .cfi_offset %esi, -16
; X32-NEXT: .cfi_offset %ebx, -12 ; X32-NEXT: .cfi_offset %ebx, -12
; X32-NEXT: movl 8(%ebp), %ecx ; X32-NEXT: movl 8(%ebp), %eax
; X32-NEXT: kmovd %ecx, %k0 ; X32-NEXT: kmovd %eax, %k0
; X32-NEXT: kshiftrq $1, %k0, %k1 ; X32-NEXT: kshiftrq $1, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $2, %al ; X32-NEXT: andb $2, %cl
; X32-NEXT: shrb %al ; X32-NEXT: shrb %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $62, %k1, %k1 ; X32-NEXT: kshiftrq $62, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $2, %k0, %k1 ; X32-NEXT: kshiftrq $2, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %al ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %eax, %edx ; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $61, %k1, %k1 ; X32-NEXT: kshiftrq $61, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $3, %k0, %k1 ; X32-NEXT: kshiftrq $3, %k0, %k1
; X32-NEXT: shrb $3, %al ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $60, %k1, %k1 ; X32-NEXT: kshiftrq $60, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $4, %k0, %k1 ; X32-NEXT: kshiftrq $4, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $4, %al ; X32-NEXT: shrb $4, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $59, %k1, %k1 ; X32-NEXT: kshiftrq $59, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $5, %k0, %k1 ; X32-NEXT: kshiftrq $5, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $5, %al ; X32-NEXT: shrb $5, %cl
; X32-NEXT: andb $1, %al ; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $58, %k1, %k1 ; X32-NEXT: kshiftrq $58, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $6, %k0, %k1 ; X32-NEXT: kshiftrq $6, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $6, %al ; X32-NEXT: shrb $6, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $57, %k1, %k1 ; X32-NEXT: kshiftrq $57, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $7, %k0, %k1 ; X32-NEXT: kshiftrq $7, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $7, %al ; X32-NEXT: shrb $7, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $56, %k1, %k1 ; X32-NEXT: kshiftrq $56, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $8, %k0, %k1 ; X32-NEXT: kshiftrq $8, %k0, %k1
; X32-NEXT: movb %ch, %al ; X32-NEXT: movb %ah, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %al
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movb %ch, %al
; X32-NEXT: andb $15, %al
; X32-NEXT: movl %eax, %edx
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k3
; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $13, %eax
; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrl $16, %edx
; X32-NEXT: movl %edx, %eax
; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %al
; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %edx, %ebx
; X32-NEXT: andb $15, %bl
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: movl 12(%ebp), %eax
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $55, %k1, %k1 ; X32-NEXT: kshiftrq $55, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $9, %k0, %k1 ; X32-NEXT: kshiftrq $9, %k0, %k1
; X32-NEXT: andb $2, %cl
; X32-NEXT: shrb %cl
; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $54, %k1, %k1 ; X32-NEXT: kshiftrq $54, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $10, %k0, %k1 ; X32-NEXT: kshiftrq $10, %k0, %k1
; X32-NEXT: kxorq %k3, %k1, %k1 ; X32-NEXT: movb %ah, %cl
; X32-NEXT: movl %eax, %esi
; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: movzwl %si, %edx
; X32-NEXT: movl %edx, %ecx
; X32-NEXT: shrl $12, %ecx
; X32-NEXT: kmovd %ecx, %k3
; X32-NEXT: movl %esi, %ecx
; X32-NEXT: shrl $13, %ecx
; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %edx, %ecx
; X32-NEXT: shrl $14, %ecx
; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: shrl $16, %eax
; X32-NEXT: movl %eax, %ebx
; X32-NEXT: andb $2, %bl
; X32-NEXT: shrb %bl
; X32-NEXT: kmovd %ebx, %k6
; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %ebx
; X32-NEXT: shrb $2, %bl
; X32-NEXT: kmovd %ebx, %k7
; X32-NEXT: movl 12(%ebp), %ebx
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $53, %k1, %k1 ; X32-NEXT: kshiftrq $53, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $11, %k0, %k1 ; X32-NEXT: kshiftrq $11, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $52, %k1, %k1 ; X32-NEXT: kshiftrq $52, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $12, %k0, %k1 ; X32-NEXT: kshiftrq $12, %k0, %k1
; X32-NEXT: movl %ecx, %esi ; X32-NEXT: kxorq %k3, %k1, %k1
; X32-NEXT: shrl $12, %esi
; X32-NEXT: andl $15, %esi
; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $51, %k1, %k1 ; X32-NEXT: kshiftrq $51, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $13, %k0, %k1 ; X32-NEXT: kshiftrq $13, %k0, %k1
; X32-NEXT: kxorq %k5, %k1, %k1 ; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $50, %k1, %k1 ; X32-NEXT: kshiftrq $50, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $14, %k0, %k1 ; X32-NEXT: kshiftrq $14, %k0, %k1
; X32-NEXT: movl %ecx, %esi ; X32-NEXT: kxorq %k5, %k1, %k1
; X32-NEXT: shrl $14, %esi
; X32-NEXT: andl $3, %esi
; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $49, %k1, %k1 ; X32-NEXT: kshiftrq $49, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $15, %k0, %k1 ; X32-NEXT: kshiftrq $15, %k0, %k1
; X32-NEXT: movl %ecx, %esi ; X32-NEXT: shrl $15, %edx
; X32-NEXT: shrl $15, %esi ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: andl $1, %esi
; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $48, %k1, %k1 ; X32-NEXT: kshiftrq $48, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $16, %k0, %k1 ; X32-NEXT: kshiftrq $16, %k0, %k1
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $47, %k1, %k1 ; X32-NEXT: kshiftrq $47, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $17, %k0, %k1 ; X32-NEXT: kshiftrq $17, %k0, %k1
; X32-NEXT: kxorq %k6, %k1, %k1 ; X32-NEXT: kxorq %k6, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $46, %k1, %k1 ; X32-NEXT: kshiftrq $46, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $18, %k0, %k1 ; X32-NEXT: kshiftrq $18, %k0, %k1
; X32-NEXT: kxorq %k7, %k1, %k1 ; X32-NEXT: kxorq %k7, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $45, %k1, %k1 ; X32-NEXT: kshiftrq $45, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $19, %k0, %k1 ; X32-NEXT: kshiftrq $19, %k0, %k1
; X32-NEXT: shrb $3, %bl ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $44, %k1, %k1 ; X32-NEXT: kshiftrq $44, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $20, %k0, %k1 ; X32-NEXT: kshiftrq $20, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $4, %bl ; X32-NEXT: shrb $4, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $43, %k1, %k1 ; X32-NEXT: kshiftrq $43, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $21, %k0, %k1 ; X32-NEXT: kshiftrq $21, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $5, %bl ; X32-NEXT: shrb $5, %cl
; X32-NEXT: andb $1, %bl ; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $42, %k1, %k1 ; X32-NEXT: kshiftrq $42, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $22, %k0, %k1 ; X32-NEXT: kshiftrq $22, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $6, %bl ; X32-NEXT: shrb $6, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $41, %k1, %k1 ; X32-NEXT: kshiftrq $41, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $23, %k0, %k1 ; X32-NEXT: kshiftrq $23, %k0, %k1
; X32-NEXT: shrb $7, %dl ; X32-NEXT: shrb $7, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $40, %k1, %k1 ; X32-NEXT: kshiftrq $40, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $24, %k0, %k1 ; X32-NEXT: kshiftrq $24, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %esi, %edx
; X32-NEXT: shrl $24, %edx ; X32-NEXT: shrl $24, %edx
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $39, %k1, %k1 ; X32-NEXT: kshiftrq $39, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $25, %k0, %k1 ; X32-NEXT: kshiftrq $25, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: andb $2, %bl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %bl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $38, %k1, %k1 ; X32-NEXT: kshiftrq $38, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $26, %k0, %k1 ; X32-NEXT: kshiftrq $26, %k0, %k1
; X32-NEXT: andb $15, %dl ; X32-NEXT: andb $15, %dl
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: shrb $2, %bl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $37, %k1, %k1 ; X32-NEXT: kshiftrq $37, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $27, %k0, %k1 ; X32-NEXT: kshiftrq $27, %k0, %k1
; X32-NEXT: shrb $3, %dl ; X32-NEXT: shrb $3, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $36, %k1, %k1 ; X32-NEXT: kshiftrq $36, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $28, %k0, %k1 ; X32-NEXT: kshiftrq $28, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %esi, %ecx
; X32-NEXT: shrl $28, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: shrl $28, %eax
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $35, %k1, %k1 ; X32-NEXT: kshiftrq $35, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $29, %k0, %k1 ; X32-NEXT: kshiftrq $29, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $29, %edx ; X32-NEXT: shrl $29, %eax
; X32-NEXT: andb $1, %dl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $34, %k1, %k1 ; X32-NEXT: kshiftrq $34, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $30, %k0, %k1 ; X32-NEXT: kshiftrq $30, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $30, %edx ; X32-NEXT: shrl $30, %eax
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $33, %k1, %k1 ; X32-NEXT: kshiftrq $33, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $31, %k0, %k1 ; X32-NEXT: kshiftrq $31, %k0, %k1
; X32-NEXT: shrl $31, %ecx ; X32-NEXT: shrl $31, %ecx
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $32, %k1, %k1 ; X32-NEXT: kshiftrq $32, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $32, %k0, %k1 ; X32-NEXT: kshiftrq $32, %k0, %k1
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ebx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $31, %k1, %k1 ; X32-NEXT: kshiftrq $31, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $33, %k0, %k1 ; X32-NEXT: kshiftrq $33, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: andb $2, %cl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %cl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $30, %k1, %k1 ; X32-NEXT: kshiftrq $30, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $34, %k0, %k1 ; X32-NEXT: kshiftrq $34, %k0, %k1
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: andb $15, %al
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %cl ; X32-NEXT: shrb $2, %cl
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $29, %k1, %k1 ; X32-NEXT: kshiftrq $29, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $35, %k0, %k1 ; X32-NEXT: kshiftrq $35, %k0, %k1
; X32-NEXT: shrb $3, %cl ; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $28, %k1, %k1 ; X32-NEXT: kshiftrq $28, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $36, %k0, %k1 ; X32-NEXT: kshiftrq $36, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $4, %cl ; X32-NEXT: shrb $4, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $27, %k1, %k1 ; X32-NEXT: kshiftrq $27, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $37, %k0, %k1 ; X32-NEXT: kshiftrq $37, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $5, %cl ; X32-NEXT: shrb $5, %al
; X32-NEXT: andb $1, %cl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $26, %k1, %k1 ; X32-NEXT: kshiftrq $26, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $38, %k0, %k1 ; X32-NEXT: kshiftrq $38, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $6, %cl ; X32-NEXT: shrb $6, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $25, %k1, %k1 ; X32-NEXT: kshiftrq $25, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $39, %k0, %k1 ; X32-NEXT: kshiftrq $39, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $7, %cl ; X32-NEXT: shrb $7, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $24, %k1, %k1 ; X32-NEXT: kshiftrq $24, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $40, %k0, %k1 ; X32-NEXT: kshiftrq $40, %k0, %k1
; X32-NEXT: movb %ah, %cl ; X32-NEXT: movb %bh, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: andb $2, %cl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %cl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movb %ah, %cl ; X32-NEXT: movb %bh, %al
; X32-NEXT: andb $15, %cl ; X32-NEXT: andb $15, %al
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k3
; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrl $13, %ecx
; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $2, %cl
; X32-NEXT: kmovd %ecx, %k3
; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $13, %eax
; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %ebx, %ecx
; X32-NEXT: shrl $16, %ecx ; X32-NEXT: shrl $16, %ecx
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: andb $2, %dl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %dl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %edx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %edx
; X32-NEXT: andb $15, %dl ; X32-NEXT: andb $15, %dl
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: shrb $2, %bl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %ebx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $23, %k1, %k1 ; X32-NEXT: kshiftrq $23, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $41, %k0, %k1 ; X32-NEXT: kshiftrq $41, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $22, %k1, %k1 ; X32-NEXT: kshiftrq $22, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $42, %k0, %k1 ; X32-NEXT: kshiftrq $42, %k0, %k1
; X32-NEXT: kxorq %k3, %k1, %k1 ; X32-NEXT: kxorq %k3, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $21, %k1, %k1 ; X32-NEXT: kshiftrq $21, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $43, %k0, %k1 ; X32-NEXT: kshiftrq $43, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1 ; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $20, %k1, %k1 ; X32-NEXT: kshiftrq $20, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $44, %k0, %k1 ; X32-NEXT: kshiftrq $44, %k0, %k1
; X32-NEXT: movzwl %bx, %eax
; X32-NEXT: movl %eax, %esi ; X32-NEXT: movl %eax, %esi
; X32-NEXT: shrl $12, %esi ; X32-NEXT: shrl $12, %esi
; X32-NEXT: andl $15, %esi
; X32-NEXT: kmovd %esi, %k2 ; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $19, %k1, %k1 ; X32-NEXT: kshiftrq $19, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $45, %k0, %k1 ; X32-NEXT: kshiftrq $45, %k0, %k1
; X32-NEXT: kxorq %k5, %k1, %k1 ; X32-NEXT: kxorq %k5, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $18, %k1, %k1 ; X32-NEXT: kshiftrq $18, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $46, %k0, %k1 ; X32-NEXT: kshiftrq $46, %k0, %k1
; X32-NEXT: movl %eax, %esi ; X32-NEXT: movl %eax, %esi
; X32-NEXT: shrl $14, %esi ; X32-NEXT: shrl $14, %esi
; X32-NEXT: andl $3, %esi
; X32-NEXT: kmovd %esi, %k2 ; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $17, %k1, %k1 ; X32-NEXT: kshiftrq $17, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $47, %k0, %k1 ; X32-NEXT: kshiftrq $47, %k0, %k1
; X32-NEXT: movl %eax, %esi ; X32-NEXT: shrl $15, %eax
; X32-NEXT: shrl $15, %esi ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: andl $1, %esi
; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $16, %k1, %k1 ; X32-NEXT: kshiftrq $16, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $48, %k0, %k1 ; X32-NEXT: kshiftrq $48, %k0, %k1
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
Show All 12 Lines
; X32-NEXT: kshiftrq $51, %k0, %k1 ; X32-NEXT: kshiftrq $51, %k0, %k1
; X32-NEXT: shrb $3, %dl ; X32-NEXT: shrb $3, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $12, %k1, %k1 ; X32-NEXT: kshiftrq $12, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $52, %k0, %k1 ; X32-NEXT: kshiftrq $52, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $4, %dl ; X32-NEXT: shrb $4, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $5, %dl ; X32-NEXT: shrb $5, %al
; X32-NEXT: andb $1, %dl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $6, %dl ; X32-NEXT: shrb $6, %al
; X32-NEXT: kmovd %edx, %k3 ; X32-NEXT: kmovd %eax, %k3
; X32-NEXT: shrb $7, %cl ; X32-NEXT: shrb $7, %cl
; X32-NEXT: kmovd %ecx, %k4 ; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %ecx
; X32-NEXT: shrl $24, %ecx ; X32-NEXT: shrl $24, %ecx
; X32-NEXT: kmovd %ecx, %k5 ; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: andb $2, %dl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %dl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %edx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: andb $15, %cl ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %edx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $11, %k1, %k1 ; X32-NEXT: kshiftrq $11, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $53, %k0, %k1 ; X32-NEXT: kshiftrq $53, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $10, %k1, %k1 ; X32-NEXT: kshiftrq $10, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
Show All 25 Lines
; X32-NEXT: kshiftrq $59, %k0, %k1 ; X32-NEXT: kshiftrq $59, %k0, %k1
; X32-NEXT: shrb $3, %cl ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $4, %k1, %k1 ; X32-NEXT: kshiftrq $4, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $60, %k0, %k1 ; X32-NEXT: kshiftrq $60, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $28, %ecx ; X32-NEXT: shrl $28, %eax
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $3, %k1, %k1 ; X32-NEXT: kshiftrq $3, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $61, %k0, %k1 ; X32-NEXT: kshiftrq $61, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $29, %ecx ; X32-NEXT: shrl $29, %eax
; X32-NEXT: andb $1, %cl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $2, %k1, %k1 ; X32-NEXT: kshiftrq $2, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $62, %k0, %k1 ; X32-NEXT: kshiftrq $62, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $30, %ecx ; X32-NEXT: shrl $30, %eax
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $1, %k1, %k1 ; X32-NEXT: kshiftrq $1, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftlq $1, %k0, %k0 ; X32-NEXT: kshiftlq $1, %k0, %k0
; X32-NEXT: kshiftrq $1, %k0, %k0 ; X32-NEXT: kshiftrq $1, %k0, %k0
; X32-NEXT: shrl $31, %eax ; X32-NEXT: shrl $31, %ebx
; X32-NEXT: kmovd %eax, %k1 ; X32-NEXT: kmovd %ebx, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: korq %k1, %k0, %k1 ; X32-NEXT: korq %k1, %k0, %k1
; X32-NEXT: vptestmb %zmm0, %zmm1, %k0 {%k1} ; X32-NEXT: vptestmb %zmm0, %zmm1, %k0 {%k1}
; X32-NEXT: kmovq %k0, (%esp) ; X32-NEXT: kmovq %k0, (%esp)
; X32-NEXT: movl (%esp), %eax ; X32-NEXT: movl (%esp), %eax
; X32-NEXT: movl {{[0-9]+}}(%esp), %edx ; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
; X32-NEXT: leal -8(%ebp), %esp ; X32-NEXT: leal -8(%ebp), %esp
; X32-NEXT: popl %esi ; X32-NEXT: popl %esi
▲ Show 20 Lines • Show All 109 Lines • ▼ Show 20 Lines
; X32-NEXT: movl %esp, %ebp ; X32-NEXT: movl %esp, %ebp
; X32-NEXT: .cfi_def_cfa_register %ebp ; X32-NEXT: .cfi_def_cfa_register %ebp
; X32-NEXT: pushl %ebx ; X32-NEXT: pushl %ebx
; X32-NEXT: pushl %esi ; X32-NEXT: pushl %esi
; X32-NEXT: andl $-8, %esp ; X32-NEXT: andl $-8, %esp
; X32-NEXT: subl $8, %esp ; X32-NEXT: subl $8, %esp
; X32-NEXT: .cfi_offset %esi, -16 ; X32-NEXT: .cfi_offset %esi, -16
; X32-NEXT: .cfi_offset %ebx, -12 ; X32-NEXT: .cfi_offset %ebx, -12
; X32-NEXT: movl 8(%ebp), %ecx ; X32-NEXT: movl 8(%ebp), %eax
; X32-NEXT: kmovd %ecx, %k0 ; X32-NEXT: kmovd %eax, %k0
; X32-NEXT: kshiftrq $1, %k0, %k1 ; X32-NEXT: kshiftrq $1, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $2, %al ; X32-NEXT: andb $2, %cl
; X32-NEXT: shrb %al ; X32-NEXT: shrb %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $62, %k1, %k1 ; X32-NEXT: kshiftrq $62, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $2, %k0, %k1 ; X32-NEXT: kshiftrq $2, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %al ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %eax, %edx ; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $61, %k1, %k1 ; X32-NEXT: kshiftrq $61, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $3, %k0, %k1 ; X32-NEXT: kshiftrq $3, %k0, %k1
; X32-NEXT: shrb $3, %al ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $60, %k1, %k1 ; X32-NEXT: kshiftrq $60, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $4, %k0, %k1 ; X32-NEXT: kshiftrq $4, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $4, %al ; X32-NEXT: shrb $4, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $59, %k1, %k1 ; X32-NEXT: kshiftrq $59, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $5, %k0, %k1 ; X32-NEXT: kshiftrq $5, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $5, %al ; X32-NEXT: shrb $5, %cl
; X32-NEXT: andb $1, %al ; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $58, %k1, %k1 ; X32-NEXT: kshiftrq $58, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $6, %k0, %k1 ; X32-NEXT: kshiftrq $6, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $6, %al ; X32-NEXT: shrb $6, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $57, %k1, %k1 ; X32-NEXT: kshiftrq $57, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $7, %k0, %k1 ; X32-NEXT: kshiftrq $7, %k0, %k1
; X32-NEXT: movl %ecx, %eax ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $7, %al ; X32-NEXT: shrb $7, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $56, %k1, %k1 ; X32-NEXT: kshiftrq $56, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $8, %k0, %k1 ; X32-NEXT: kshiftrq $8, %k0, %k1
; X32-NEXT: movb %ch, %al ; X32-NEXT: movb %ah, %cl
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %al
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movb %ch, %al
; X32-NEXT: andb $15, %al
; X32-NEXT: movl %eax, %edx
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k3
; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $13, %eax
; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrl $16, %edx
; X32-NEXT: movl %edx, %eax
; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %al
; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %edx, %ebx
; X32-NEXT: andb $15, %bl
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: movl 12(%ebp), %eax
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $55, %k1, %k1 ; X32-NEXT: kshiftrq $55, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $9, %k0, %k1 ; X32-NEXT: kshiftrq $9, %k0, %k1
; X32-NEXT: andb $2, %cl
; X32-NEXT: shrb %cl
; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $54, %k1, %k1 ; X32-NEXT: kshiftrq $54, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $10, %k0, %k1 ; X32-NEXT: kshiftrq $10, %k0, %k1
; X32-NEXT: kxorq %k3, %k1, %k1 ; X32-NEXT: movb %ah, %cl
; X32-NEXT: movl %eax, %esi
; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: movzwl %si, %edx
; X32-NEXT: movl %edx, %ecx
; X32-NEXT: shrl $12, %ecx
; X32-NEXT: kmovd %ecx, %k3
; X32-NEXT: movl %esi, %ecx
; X32-NEXT: shrl $13, %ecx
; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %edx, %ecx
; X32-NEXT: shrl $14, %ecx
; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: shrl $16, %eax
; X32-NEXT: movl %eax, %ebx
; X32-NEXT: andb $2, %bl
; X32-NEXT: shrb %bl
; X32-NEXT: kmovd %ebx, %k6
; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %ebx
; X32-NEXT: shrb $2, %bl
; X32-NEXT: kmovd %ebx, %k7
; X32-NEXT: movl 12(%ebp), %ebx
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $53, %k1, %k1 ; X32-NEXT: kshiftrq $53, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $11, %k0, %k1 ; X32-NEXT: kshiftrq $11, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $52, %k1, %k1 ; X32-NEXT: kshiftrq $52, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $12, %k0, %k1 ; X32-NEXT: kshiftrq $12, %k0, %k1
; X32-NEXT: movl %ecx, %esi ; X32-NEXT: kxorq %k3, %k1, %k1
; X32-NEXT: shrl $12, %esi
; X32-NEXT: andl $15, %esi
; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $51, %k1, %k1 ; X32-NEXT: kshiftrq $51, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $13, %k0, %k1 ; X32-NEXT: kshiftrq $13, %k0, %k1
; X32-NEXT: kxorq %k5, %k1, %k1 ; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $50, %k1, %k1 ; X32-NEXT: kshiftrq $50, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $14, %k0, %k1 ; X32-NEXT: kshiftrq $14, %k0, %k1
; X32-NEXT: movl %ecx, %esi ; X32-NEXT: kxorq %k5, %k1, %k1
; X32-NEXT: shrl $14, %esi
; X32-NEXT: andl $3, %esi
; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $49, %k1, %k1 ; X32-NEXT: kshiftrq $49, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $15, %k0, %k1 ; X32-NEXT: kshiftrq $15, %k0, %k1
; X32-NEXT: movl %ecx, %esi ; X32-NEXT: shrl $15, %edx
; X32-NEXT: shrl $15, %esi ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: andl $1, %esi
; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $48, %k1, %k1 ; X32-NEXT: kshiftrq $48, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $16, %k0, %k1 ; X32-NEXT: kshiftrq $16, %k0, %k1
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $47, %k1, %k1 ; X32-NEXT: kshiftrq $47, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $17, %k0, %k1 ; X32-NEXT: kshiftrq $17, %k0, %k1
; X32-NEXT: kxorq %k6, %k1, %k1 ; X32-NEXT: kxorq %k6, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $46, %k1, %k1 ; X32-NEXT: kshiftrq $46, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $18, %k0, %k1 ; X32-NEXT: kshiftrq $18, %k0, %k1
; X32-NEXT: kxorq %k7, %k1, %k1 ; X32-NEXT: kxorq %k7, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $45, %k1, %k1 ; X32-NEXT: kshiftrq $45, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $19, %k0, %k1 ; X32-NEXT: kshiftrq $19, %k0, %k1
; X32-NEXT: shrb $3, %bl ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $44, %k1, %k1 ; X32-NEXT: kshiftrq $44, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $20, %k0, %k1 ; X32-NEXT: kshiftrq $20, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $4, %bl ; X32-NEXT: shrb $4, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $43, %k1, %k1 ; X32-NEXT: kshiftrq $43, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $21, %k0, %k1 ; X32-NEXT: kshiftrq $21, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $5, %bl ; X32-NEXT: shrb $5, %cl
; X32-NEXT: andb $1, %bl ; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $42, %k1, %k1 ; X32-NEXT: kshiftrq $42, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $22, %k0, %k1 ; X32-NEXT: kshiftrq $22, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $6, %bl ; X32-NEXT: shrb $6, %cl
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $41, %k1, %k1 ; X32-NEXT: kshiftrq $41, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $23, %k0, %k1 ; X32-NEXT: kshiftrq $23, %k0, %k1
; X32-NEXT: shrb $7, %dl ; X32-NEXT: shrb $7, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $40, %k1, %k1 ; X32-NEXT: kshiftrq $40, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $24, %k0, %k1 ; X32-NEXT: kshiftrq $24, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %esi, %edx
; X32-NEXT: shrl $24, %edx ; X32-NEXT: shrl $24, %edx
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $39, %k1, %k1 ; X32-NEXT: kshiftrq $39, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $25, %k0, %k1 ; X32-NEXT: kshiftrq $25, %k0, %k1
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: andb $2, %bl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %bl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $38, %k1, %k1 ; X32-NEXT: kshiftrq $38, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $26, %k0, %k1 ; X32-NEXT: kshiftrq $26, %k0, %k1
; X32-NEXT: andb $15, %dl ; X32-NEXT: andb $15, %dl
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: shrb $2, %bl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %ebx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $37, %k1, %k1 ; X32-NEXT: kshiftrq $37, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $27, %k0, %k1 ; X32-NEXT: kshiftrq $27, %k0, %k1
; X32-NEXT: shrb $3, %dl ; X32-NEXT: shrb $3, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $36, %k1, %k1 ; X32-NEXT: kshiftrq $36, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $28, %k0, %k1 ; X32-NEXT: kshiftrq $28, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %esi, %ecx
; X32-NEXT: shrl $28, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: shrl $28, %eax
; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $35, %k1, %k1 ; X32-NEXT: kshiftrq $35, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $29, %k0, %k1 ; X32-NEXT: kshiftrq $29, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $29, %edx ; X32-NEXT: shrl $29, %eax
; X32-NEXT: andb $1, %dl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $34, %k1, %k1 ; X32-NEXT: kshiftrq $34, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $30, %k0, %k1 ; X32-NEXT: kshiftrq $30, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrl $30, %edx ; X32-NEXT: shrl $30, %eax
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $33, %k1, %k1 ; X32-NEXT: kshiftrq $33, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $31, %k0, %k1 ; X32-NEXT: kshiftrq $31, %k0, %k1
; X32-NEXT: shrl $31, %ecx ; X32-NEXT: shrl $31, %ecx
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $32, %k1, %k1 ; X32-NEXT: kshiftrq $32, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $32, %k0, %k1 ; X32-NEXT: kshiftrq $32, %k0, %k1
; X32-NEXT: kmovd %eax, %k2 ; X32-NEXT: kmovd %ebx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $31, %k1, %k1 ; X32-NEXT: kshiftrq $31, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $33, %k0, %k1 ; X32-NEXT: kshiftrq $33, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: andb $2, %cl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %cl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $30, %k1, %k1 ; X32-NEXT: kshiftrq $30, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $34, %k0, %k1 ; X32-NEXT: kshiftrq $34, %k0, %k1
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: andb $15, %al
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: andb $15, %cl ; X32-NEXT: shrb $2, %cl
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $29, %k1, %k1 ; X32-NEXT: kshiftrq $29, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $35, %k0, %k1 ; X32-NEXT: kshiftrq $35, %k0, %k1
; X32-NEXT: shrb $3, %cl ; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $28, %k1, %k1 ; X32-NEXT: kshiftrq $28, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $36, %k0, %k1 ; X32-NEXT: kshiftrq $36, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $4, %cl ; X32-NEXT: shrb $4, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $27, %k1, %k1 ; X32-NEXT: kshiftrq $27, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $37, %k0, %k1 ; X32-NEXT: kshiftrq $37, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $5, %cl ; X32-NEXT: shrb $5, %al
; X32-NEXT: andb $1, %cl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $26, %k1, %k1 ; X32-NEXT: kshiftrq $26, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $38, %k0, %k1 ; X32-NEXT: kshiftrq $38, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $6, %cl ; X32-NEXT: shrb $6, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $25, %k1, %k1 ; X32-NEXT: kshiftrq $25, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $39, %k0, %k1 ; X32-NEXT: kshiftrq $39, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrb $7, %cl ; X32-NEXT: shrb $7, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $24, %k1, %k1 ; X32-NEXT: kshiftrq $24, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $40, %k0, %k1 ; X32-NEXT: kshiftrq $40, %k0, %k1
; X32-NEXT: movb %ah, %cl ; X32-NEXT: movb %bh, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: andb $2, %cl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %cl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movb %ah, %cl ; X32-NEXT: movb %bh, %al
; X32-NEXT: andb $15, %cl ; X32-NEXT: andb $15, %al
; X32-NEXT: movl %ecx, %edx
; X32-NEXT: shrb $2, %dl
; X32-NEXT: kmovd %edx, %k3
; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrl $13, %ecx
; X32-NEXT: andb $1, %cl
; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %eax, %ecx
; X32-NEXT: shrb $2, %cl
; X32-NEXT: kmovd %ecx, %k3
; X32-NEXT: shrb $3, %al
; X32-NEXT: kmovd %eax, %k4
; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $13, %eax
; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %eax, %k5
; X32-NEXT: movl %ebx, %ecx
; X32-NEXT: shrl $16, %ecx ; X32-NEXT: shrl $16, %ecx
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: andb $2, %dl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %dl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %edx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %edx
; X32-NEXT: andb $15, %dl ; X32-NEXT: andb $15, %dl
; X32-NEXT: movl %edx, %ebx ; X32-NEXT: movl %edx, %eax
; X32-NEXT: shrb $2, %bl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %ebx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $23, %k1, %k1 ; X32-NEXT: kshiftrq $23, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $41, %k0, %k1 ; X32-NEXT: kshiftrq $41, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $22, %k1, %k1 ; X32-NEXT: kshiftrq $22, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $42, %k0, %k1 ; X32-NEXT: kshiftrq $42, %k0, %k1
; X32-NEXT: kxorq %k3, %k1, %k1 ; X32-NEXT: kxorq %k3, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $21, %k1, %k1 ; X32-NEXT: kshiftrq $21, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $43, %k0, %k1 ; X32-NEXT: kshiftrq $43, %k0, %k1
; X32-NEXT: kxorq %k4, %k1, %k1 ; X32-NEXT: kxorq %k4, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $20, %k1, %k1 ; X32-NEXT: kshiftrq $20, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $44, %k0, %k1 ; X32-NEXT: kshiftrq $44, %k0, %k1
; X32-NEXT: movzwl %bx, %eax
; X32-NEXT: movl %eax, %esi ; X32-NEXT: movl %eax, %esi
; X32-NEXT: shrl $12, %esi ; X32-NEXT: shrl $12, %esi
; X32-NEXT: andl $15, %esi
; X32-NEXT: kmovd %esi, %k2 ; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $19, %k1, %k1 ; X32-NEXT: kshiftrq $19, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $45, %k0, %k1 ; X32-NEXT: kshiftrq $45, %k0, %k1
; X32-NEXT: kxorq %k5, %k1, %k1 ; X32-NEXT: kxorq %k5, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $18, %k1, %k1 ; X32-NEXT: kshiftrq $18, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $46, %k0, %k1 ; X32-NEXT: kshiftrq $46, %k0, %k1
; X32-NEXT: movl %eax, %esi ; X32-NEXT: movl %eax, %esi
; X32-NEXT: shrl $14, %esi ; X32-NEXT: shrl $14, %esi
; X32-NEXT: andl $3, %esi
; X32-NEXT: kmovd %esi, %k2 ; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $17, %k1, %k1 ; X32-NEXT: kshiftrq $17, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $47, %k0, %k1 ; X32-NEXT: kshiftrq $47, %k0, %k1
; X32-NEXT: movl %eax, %esi ; X32-NEXT: shrl $15, %eax
; X32-NEXT: shrl $15, %esi ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: andl $1, %esi
; X32-NEXT: kmovd %esi, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $16, %k1, %k1 ; X32-NEXT: kshiftrq $16, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $48, %k0, %k1 ; X32-NEXT: kshiftrq $48, %k0, %k1
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
Show All 12 Lines
; X32-NEXT: kshiftrq $51, %k0, %k1 ; X32-NEXT: kshiftrq $51, %k0, %k1
; X32-NEXT: shrb $3, %dl ; X32-NEXT: shrb $3, %dl
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %edx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $12, %k1, %k1 ; X32-NEXT: kshiftrq $12, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $52, %k0, %k1 ; X32-NEXT: kshiftrq $52, %k0, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $4, %dl ; X32-NEXT: shrb $4, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $5, %dl ; X32-NEXT: shrb $5, %al
; X32-NEXT: andb $1, %dl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %edx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $6, %dl ; X32-NEXT: shrb $6, %al
; X32-NEXT: kmovd %edx, %k3 ; X32-NEXT: kmovd %eax, %k3
; X32-NEXT: shrb $7, %cl ; X32-NEXT: shrb $7, %cl
; X32-NEXT: kmovd %ecx, %k4 ; X32-NEXT: kmovd %ecx, %k4
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %ecx
; X32-NEXT: shrl $24, %ecx ; X32-NEXT: shrl $24, %ecx
; X32-NEXT: kmovd %ecx, %k5 ; X32-NEXT: kmovd %ecx, %k5
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: andb $2, %dl ; X32-NEXT: andb $2, %al
; X32-NEXT: shrb %dl ; X32-NEXT: shrb %al
; X32-NEXT: kmovd %edx, %k6 ; X32-NEXT: kmovd %eax, %k6
; X32-NEXT: andb $15, %cl ; X32-NEXT: andb $15, %cl
; X32-NEXT: movl %ecx, %edx ; X32-NEXT: movl %ecx, %eax
; X32-NEXT: shrb $2, %dl ; X32-NEXT: shrb $2, %al
; X32-NEXT: kmovd %edx, %k7 ; X32-NEXT: kmovd %eax, %k7
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $11, %k1, %k1 ; X32-NEXT: kshiftrq $11, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $53, %k0, %k1 ; X32-NEXT: kshiftrq $53, %k0, %k1
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $10, %k1, %k1 ; X32-NEXT: kshiftrq $10, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
Show All 25 Lines
; X32-NEXT: kshiftrq $59, %k0, %k1 ; X32-NEXT: kshiftrq $59, %k0, %k1
; X32-NEXT: shrb $3, %cl ; X32-NEXT: shrb $3, %cl
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %ecx, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $4, %k1, %k1 ; X32-NEXT: kshiftrq $4, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $60, %k0, %k1 ; X32-NEXT: kshiftrq $60, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $28, %ecx ; X32-NEXT: shrl $28, %eax
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $3, %k1, %k1 ; X32-NEXT: kshiftrq $3, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $61, %k0, %k1 ; X32-NEXT: kshiftrq $61, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $29, %ecx ; X32-NEXT: shrl $29, %eax
; X32-NEXT: andb $1, %cl ; X32-NEXT: andb $1, %al
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $2, %k1, %k1 ; X32-NEXT: kshiftrq $2, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftrq $62, %k0, %k1 ; X32-NEXT: kshiftrq $62, %k0, %k1
; X32-NEXT: movl %eax, %ecx ; X32-NEXT: movl %ebx, %eax
; X32-NEXT: shrl $30, %ecx ; X32-NEXT: shrl $30, %eax
; X32-NEXT: kmovd %ecx, %k2 ; X32-NEXT: kmovd %eax, %k2
; X32-NEXT: kxorq %k2, %k1, %k1 ; X32-NEXT: kxorq %k2, %k1, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: kshiftrq $1, %k1, %k1 ; X32-NEXT: kshiftrq $1, %k1, %k1
; X32-NEXT: kxorq %k0, %k1, %k0 ; X32-NEXT: kxorq %k0, %k1, %k0
; X32-NEXT: kshiftlq $1, %k0, %k0 ; X32-NEXT: kshiftlq $1, %k0, %k0
; X32-NEXT: kshiftrq $1, %k0, %k0 ; X32-NEXT: kshiftrq $1, %k0, %k0
; X32-NEXT: shrl $31, %eax ; X32-NEXT: shrl $31, %ebx
; X32-NEXT: kmovd %eax, %k1 ; X32-NEXT: kmovd %ebx, %k1
; X32-NEXT: kshiftlq $63, %k1, %k1 ; X32-NEXT: kshiftlq $63, %k1, %k1
; X32-NEXT: korq %k1, %k0, %k1 ; X32-NEXT: korq %k1, %k0, %k1
; X32-NEXT: vptestnmb %zmm0, %zmm1, %k0 {%k1} ; X32-NEXT: vptestnmb %zmm0, %zmm1, %k0 {%k1}
; X32-NEXT: kmovq %k0, (%esp) ; X32-NEXT: kmovq %k0, (%esp)
; X32-NEXT: movl (%esp), %eax ; X32-NEXT: movl (%esp), %eax
; X32-NEXT: movl {{[0-9]+}}(%esp), %edx ; X32-NEXT: movl {{[0-9]+}}(%esp), %edx
; X32-NEXT: leal -8(%ebp), %esp ; X32-NEXT: leal -8(%ebp), %esp
; X32-NEXT: popl %esi ; X32-NEXT: popl %esi
▲ Show 20 Lines • Show All 72 Lines • Show Last 20 Lines

test/CodeGen/X86/avx512bw-intrinsics-upgrade.ll

Show First 20 Lines • Show All 1,791 Lines • ▼ Show 20 Lines
; AVX512BW-NEXT: kmovq %k0, %rax ; AVX512BW-NEXT: kmovq %k0, %rax
; AVX512BW-NEXT: addq %rdx, %rax ; AVX512BW-NEXT: addq %rdx, %rax
; AVX512BW-NEXT: addq %rdi, %rax ; AVX512BW-NEXT: addq %rdi, %rax
; AVX512BW-NEXT: vzeroupper ; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq ; AVX512BW-NEXT: retq
; ;
; AVX512F-32-LABEL: test_mask_cmp_b_512: ; AVX512F-32-LABEL: test_mask_cmp_b_512:
; AVX512F-32: # %bb.0: ; AVX512F-32: # %bb.0:
; AVX512F-32-NEXT: pushl %ebx ; AVX512F-32-NEXT: pushl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 8 ; AVX512F-32-NEXT: .cfi_def_cfa_offset 8
; AVX512F-32-NEXT: pushl %esi ; AVX512F-32-NEXT: pushl %ebx
; AVX512F-32-NEXT: .cfi_def_cfa_offset 12 ; AVX512F-32-NEXT: .cfi_def_cfa_offset 12
; AVX512F-32-NEXT: pushl %edi
; AVX512F-32-NEXT: .cfi_def_cfa_offset 16
; AVX512F-32-NEXT: pushl %esi
; AVX512F-32-NEXT: .cfi_def_cfa_offset 20
; AVX512F-32-NEXT: subl $60, %esp ; AVX512F-32-NEXT: subl $60, %esp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 72 ; AVX512F-32-NEXT: .cfi_def_cfa_offset 80
; AVX512F-32-NEXT: .cfi_offset %esi, -12 ; AVX512F-32-NEXT: .cfi_offset %esi, -20
; AVX512F-32-NEXT: .cfi_offset %ebx, -8 ; AVX512F-32-NEXT: .cfi_offset %edi, -16
; AVX512F-32-NEXT: .cfi_offset %ebx, -12
; AVX512F-32-NEXT: .cfi_offset %ebp, -8
; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %ebx ; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %ebx
; AVX512F-32-NEXT: movl %ebx, %eax ; AVX512F-32-NEXT: movl %ebx, %eax
; AVX512F-32-NEXT: shrl $16, %eax ; AVX512F-32-NEXT: shrl $16, %eax
; AVX512F-32-NEXT: movl %ebx, %edx ; AVX512F-32-NEXT: movl %ebx, %edx
; AVX512F-32-NEXT: andb $15, %dl ; AVX512F-32-NEXT: andb $15, %dl
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: andb $2, %cl ; AVX512F-32-NEXT: andb $2, %cl
; AVX512F-32-NEXT: shrb %cl ; AVX512F-32-NEXT: shrb %cl
Show All 23 Lines
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $62, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $62, %k1, %k1
; AVX512F-32-NEXT: kxorq %k5, %k1, %k7 ; AVX512F-32-NEXT: kxorq %k5, %k1, %k7
; AVX512F-32-NEXT: kshiftrq $2, %k7, %k1 ; AVX512F-32-NEXT: kshiftrq $2, %k7, %k1
; AVX512F-32-NEXT: kxorq %k2, %k1, %k2 ; AVX512F-32-NEXT: kxorq %k2, %k1, %k2
; AVX512F-32-NEXT: kmovd %ecx, %k5 ; AVX512F-32-NEXT: kmovd %ecx, %k5
; AVX512F-32-NEXT: movb %bh, %cl ; AVX512F-32-NEXT: movb %bh, %cl
; AVX512F-32-NEXT: kmovd %ecx, %k1 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: movl %ebx, %esi ; AVX512F-32-NEXT: movl %ebx, %ebp
; AVX512F-32-NEXT: andb $2, %cl ; AVX512F-32-NEXT: andb $2, %cl
; AVX512F-32-NEXT: shrb %cl ; AVX512F-32-NEXT: shrb %cl
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $61, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $61, %k2, %k2
; AVX512F-32-NEXT: kxorq %k7, %k2, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k2, %k7
; AVX512F-32-NEXT: kshiftrq $3, %k7, %k2 ; AVX512F-32-NEXT: kshiftrq $3, %k7, %k2
; AVX512F-32-NEXT: kxorq %k0, %k2, %k0 ; AVX512F-32-NEXT: kxorq %k0, %k2, %k0
; AVX512F-32-NEXT: kmovd %ecx, %k2 ; AVX512F-32-NEXT: kmovd %ecx, %k2
Show All 9 Lines
; AVX512F-32-NEXT: andb $15, %dl ; AVX512F-32-NEXT: andb $15, %dl
; AVX512F-32-NEXT: shrb $3, %cl ; AVX512F-32-NEXT: shrb $3, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $59, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $59, %k7, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k0, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $5, %k7, %k0 ; AVX512F-32-NEXT: kshiftrq $5, %k7, %k0
; AVX512F-32-NEXT: kxorq %k4, %k0, %k4 ; AVX512F-32-NEXT: kxorq %k4, %k0, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k0 ; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %esi, %ecx ; AVX512F-32-NEXT: movl %ebp, %ecx
; AVX512F-32-NEXT: shrl $13, %ecx ; AVX512F-32-NEXT: shrl $13, %ecx
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $58, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $58, %k4, %k4
; AVX512F-32-NEXT: kxorq %k7, %k4, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k7
; AVX512F-32-NEXT: kshiftrq $6, %k7, %k4 ; AVX512F-32-NEXT: kshiftrq $6, %k7, %k4
; AVX512F-32-NEXT: kxorq %k6, %k4, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k4, %k6
; AVX512F-32-NEXT: kmovd %ecx, %k4 ; AVX512F-32-NEXT: kmovd %ecx, %k4
Show All 28 Lines
; AVX512F-32-NEXT: kshiftrq $54, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $54, %k2, %k2
; AVX512F-32-NEXT: kxorq %k7, %k2, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k2, %k7
; AVX512F-32-NEXT: kshiftrq $10, %k7, %k2 ; AVX512F-32-NEXT: kshiftrq $10, %k7, %k2
; AVX512F-32-NEXT: kxorq %k3, %k2, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k2, %k3
; AVX512F-32-NEXT: kmovd %ecx, %k2 ; AVX512F-32-NEXT: kmovd %ecx, %k2
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $5, %cl ; AVX512F-32-NEXT: shrb $5, %cl
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: movl %esi, %edx ; AVX512F-32-NEXT: movzwl %bp, %edx
; AVX512F-32-NEXT: movl %edx, %esi
; AVX512F-32-NEXT: movl %edx, %edi
; AVX512F-32-NEXT: shrl $12, %edx ; AVX512F-32-NEXT: shrl $12, %edx
; AVX512F-32-NEXT: andl $15, %edx
; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3 ; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $53, %k3, %k3 ; AVX512F-32-NEXT: kshiftrq $53, %k3, %k3
; AVX512F-32-NEXT: kxorq %k7, %k3, %k3 ; AVX512F-32-NEXT: kxorq %k7, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $11, %k3, %k7 ; AVX512F-32-NEXT: kshiftrq $11, %k3, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k0 ; AVX512F-32-NEXT: kxorq %k0, %k7, %k0
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $52, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $52, %k0, %k0
; AVX512F-32-NEXT: kxorq %k3, %k0, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k0, %k3
; AVX512F-32-NEXT: kshiftrq $12, %k3, %k0 ; AVX512F-32-NEXT: kshiftrq $12, %k3, %k0
; AVX512F-32-NEXT: kmovd %edx, %k7 ; AVX512F-32-NEXT: kmovd %edx, %k7
; AVX512F-32-NEXT: kxorq %k7, %k0, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k0, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k0 ; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %eax, %edx ; AVX512F-32-NEXT: movl %eax, %edx
; AVX512F-32-NEXT: shrb $6, %dl ; AVX512F-32-NEXT: shrb $6, %dl
; AVX512F-32-NEXT: movl %esi, %ecx ; AVX512F-32-NEXT: shrl $15, %esi
; AVX512F-32-NEXT: shrl $14, %ecx ; AVX512F-32-NEXT: shrl $14, %edi
; AVX512F-32-NEXT: andl $3, %ecx
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $51, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $51, %k7, %k7
; AVX512F-32-NEXT: kxorq %k3, %k7, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k7, %k3
; AVX512F-32-NEXT: kshiftrq $13, %k3, %k7 ; AVX512F-32-NEXT: kshiftrq $13, %k3, %k7
; AVX512F-32-NEXT: kxorq %k4, %k7, %k4 ; AVX512F-32-NEXT: kxorq %k4, %k7, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k7
; AVX512F-32-NEXT: movl %esi, %ecx
; AVX512F-32-NEXT: shrl $15, %ecx
; AVX512F-32-NEXT: andl $1, %ecx
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $50, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $50, %k4, %k4
; AVX512F-32-NEXT: kxorq %k3, %k4, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k4, %k3
; AVX512F-32-NEXT: kshiftrq $14, %k3, %k4 ; AVX512F-32-NEXT: kshiftrq $14, %k3, %k4
; AVX512F-32-NEXT: kmovd %edi, %k7
; AVX512F-32-NEXT: kxorq %k7, %k4, %k4 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k4
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $49, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $49, %k4, %k4
; AVX512F-32-NEXT: kxorq %k3, %k4, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k4, %k3
; AVX512F-32-NEXT: kshiftrq $15, %k3, %k4 ; AVX512F-32-NEXT: kshiftrq $15, %k3, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k7 ; AVX512F-32-NEXT: kmovd %esi, %k7
; AVX512F-32-NEXT: kxorq %k7, %k4, %k4 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k4
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $48, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $48, %k4, %k4
; AVX512F-32-NEXT: kxorq %k3, %k4, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k4, %k3
; AVX512F-32-NEXT: kshiftrq $16, %k3, %k4 ; AVX512F-32-NEXT: kshiftrq $16, %k3, %k4
; AVX512F-32-NEXT: kmovd %eax, %k7 ; AVX512F-32-NEXT: kmovd %eax, %k7
; AVX512F-32-NEXT: kxorq %k7, %k4, %k4 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k4
; AVX512F-32-NEXT: kmovd %edx, %k7 ; AVX512F-32-NEXT: kmovd %edx, %k7
; AVX512F-32-NEXT: movl %esi, %edx ; AVX512F-32-NEXT: movl %ebp, %edx
; AVX512F-32-NEXT: shrl $24, %edx ; AVX512F-32-NEXT: shrl $24, %edx
; AVX512F-32-NEXT: # kill: def %al killed %al killed %eax def %eax ; AVX512F-32-NEXT: # kill: def %al killed %al killed %eax def %eax
; AVX512F-32-NEXT: shrb $7, %al ; AVX512F-32-NEXT: shrb $7, %al
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $47, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $47, %k4, %k4
; AVX512F-32-NEXT: kxorq %k3, %k4, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k4, %k3
; AVX512F-32-NEXT: kshiftrq $17, %k3, %k4 ; AVX512F-32-NEXT: kshiftrq $17, %k3, %k4
; AVX512F-32-NEXT: kxorq %k5, %k4, %k4 ; AVX512F-32-NEXT: kxorq %k5, %k4, %k4
Show All 23 Lines
; AVX512F-32-NEXT: kshiftrq $20, %k1, %k6 ; AVX512F-32-NEXT: kshiftrq $20, %k1, %k6
; AVX512F-32-NEXT: kxorq %k2, %k6, %k6 ; AVX512F-32-NEXT: kxorq %k2, %k6, %k6
; AVX512F-32-NEXT: kmovd %edx, %k2 ; AVX512F-32-NEXT: kmovd %edx, %k2
; AVX512F-32-NEXT: movl %ebx, %eax ; AVX512F-32-NEXT: movl %ebx, %eax
; AVX512F-32-NEXT: andb $15, %al ; AVX512F-32-NEXT: andb $15, %al
; AVX512F-32-NEXT: shrb $3, %cl ; AVX512F-32-NEXT: shrb $3, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6 ; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $43, %k6, %k6 ; AVX512F-32-NEXT: kshiftrq $43, %k6, %k6
; AVX512F-32-NEXT: kxorq %k1, %k6, %k1
; AVX512F-32-NEXT: kshiftrq $21, %k1, %k6
; AVX512F-32-NEXT: kxorq %k0, %k6, %k6
; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %esi, %ecx
; AVX512F-32-NEXT: shrl $29, %ecx
; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $42, %k6, %k6
; AVX512F-32-NEXT: kxorq %k1, %k6, %k6 ; AVX512F-32-NEXT: kxorq %k1, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $22, %k6, %k1 ; AVX512F-32-NEXT: kshiftrq $21, %k6, %k1
; AVX512F-32-NEXT: kxorq %k7, %k1, %k7 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kmovd %ecx, %k1 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: movl %ebp, %ecx
; AVX512F-32-NEXT: shrl $29, %ecx
; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $42, %k0, %k0
; AVX512F-32-NEXT: kxorq %k6, %k0, %k6
; AVX512F-32-NEXT: kshiftrq $22, %k6, %k0
; AVX512F-32-NEXT: kxorq %k7, %k0, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %ebx, %edx ; AVX512F-32-NEXT: movl %ebx, %edx
; AVX512F-32-NEXT: andb $2, %dl ; AVX512F-32-NEXT: andb $2, %dl
; AVX512F-32-NEXT: shrb %dl ; AVX512F-32-NEXT: shrb %dl
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $41, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $41, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $23, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $23, %k6, %k7
; AVX512F-32-NEXT: kxorq %k5, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k5, %k7, %k7
Show All 14 Lines
; AVX512F-32-NEXT: kxorq %k7, %k3, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k3, %k7
; AVX512F-32-NEXT: kshiftrq $25, %k7, %k3 ; AVX512F-32-NEXT: kshiftrq $25, %k7, %k3
; AVX512F-32-NEXT: kxorq %k4, %k3, %k4 ; AVX512F-32-NEXT: kxorq %k4, %k3, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k3 ; AVX512F-32-NEXT: kmovd %ecx, %k3
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: shrb $4, %cl ; AVX512F-32-NEXT: shrb $4, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $38, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $38, %k4, %k4
; AVX512F-32-NEXT: kxorq %k7, %k4, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $26, %k7, %k4 ; AVX512F-32-NEXT: kshiftrq $26, %k4, %k7
; AVX512F-32-NEXT: kxorq %k2, %k4, %k2 ; AVX512F-32-NEXT: kxorq %k2, %k7, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k4 ; AVX512F-32-NEXT: kmovd %ecx, %k2
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: shrb $5, %cl ; AVX512F-32-NEXT: shrb $5, %cl
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: movl %esi, %edx ; AVX512F-32-NEXT: movl %ebp, %edx
; AVX512F-32-NEXT: shrl $28, %edx ; AVX512F-32-NEXT: shrl $28, %edx
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $37, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $37, %k7, %k7
; AVX512F-32-NEXT: kxorq %k7, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k4, %k7, %k4
; AVX512F-32-NEXT: kshiftrq $27, %k2, %k7 ; AVX512F-32-NEXT: kshiftrq $27, %k4, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k7, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $36, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $36, %k1, %k1
; AVX512F-32-NEXT: kxorq %k2, %k0, %k2 ; AVX512F-32-NEXT: kxorq %k4, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $28, %k2, %k0 ; AVX512F-32-NEXT: kshiftrq $28, %k1, %k4
; AVX512F-32-NEXT: kmovd %edx, %k7 ; AVX512F-32-NEXT: kmovd %edx, %k7
; AVX512F-32-NEXT: kxorq %k7, %k0, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k0 ; AVX512F-32-NEXT: kmovd %ecx, %k4
; AVX512F-32-NEXT: movl %ebx, %edx ; AVX512F-32-NEXT: movl %ebx, %edx
; AVX512F-32-NEXT: shrb $6, %dl ; AVX512F-32-NEXT: shrb $6, %dl
; AVX512F-32-NEXT: movl %esi, %ecx ; AVX512F-32-NEXT: movl %ebp, %ecx
; AVX512F-32-NEXT: shrl $30, %ecx ; AVX512F-32-NEXT: shrl $31, %ecx
; AVX512F-32-NEXT: movl %ebp, %esi
; AVX512F-32-NEXT: shrl $30, %esi
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $35, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $35, %k7, %k7
; AVX512F-32-NEXT: kxorq %k2, %k7, %k2
; AVX512F-32-NEXT: kshiftrq $29, %k2, %k7
; AVX512F-32-NEXT: kxorq %k1, %k7, %k1 ; AVX512F-32-NEXT: kxorq %k1, %k7, %k1
; AVX512F-32-NEXT: kmovd %ecx, %k7 ; AVX512F-32-NEXT: kshiftrq $29, %k1, %k7
; AVX512F-32-NEXT: movl %esi, %ecx ; AVX512F-32-NEXT: kxorq %k0, %k7, %k0
; AVX512F-32-NEXT: shrl $31, %ecx ; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $34, %k0, %k0
; AVX512F-32-NEXT: kxorq %k1, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $30, %k0, %k1
; AVX512F-32-NEXT: kmovd %esi, %k7
; AVX512F-32-NEXT: kxorq %k7, %k1, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $34, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $33, %k1, %k1
; AVX512F-32-NEXT: kxorq %k2, %k1, %k1 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $30, %k1, %k2 ; AVX512F-32-NEXT: kshiftrq $31, %k0, %k1
; AVX512F-32-NEXT: kxorq %k7, %k2, %k2
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $33, %k2, %k2
; AVX512F-32-NEXT: kxorq %k1, %k2, %k1
; AVX512F-32-NEXT: kshiftrq $31, %k1, %k2
; AVX512F-32-NEXT: kmovd %ecx, %k7 ; AVX512F-32-NEXT: kmovd %ecx, %k7
; AVX512F-32-NEXT: kxorq %k7, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k7, %k1, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $32, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $32, %k1, %k1
; AVX512F-32-NEXT: kxorq %k1, %k2, %k1 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $32, %k1, %k2 ; AVX512F-32-NEXT: kshiftrq $32, %k0, %k1
; AVX512F-32-NEXT: kmovd %ebx, %k7 ; AVX512F-32-NEXT: kmovd %ebx, %k7
; AVX512F-32-NEXT: kxorq %k7, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k7, %k1, %k1
; AVX512F-32-NEXT: kmovd %edx, %k7 ; AVX512F-32-NEXT: kmovd %edx, %k7
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: shrb $7, %cl ; AVX512F-32-NEXT: shrb $7, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $31, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $31, %k1, %k1
; AVX512F-32-NEXT: kxorq %k1, %k2, %k1 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $33, %k1, %k2 ; AVX512F-32-NEXT: kshiftrq $33, %k0, %k1
; AVX512F-32-NEXT: kxorq %k5, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k5, %k1, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $30, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $30, %k1, %k1
; AVX512F-32-NEXT: kxorq %k1, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $34, %k2, %k1 ; AVX512F-32-NEXT: kshiftrq $34, %k0, %k1
; AVX512F-32-NEXT: kxorq %k6, %k1, %k5 ; AVX512F-32-NEXT: kxorq %k6, %k1, %k5
; AVX512F-32-NEXT: kmovd %ecx, %k6 ; AVX512F-32-NEXT: kmovd %ecx, %k6
; AVX512F-32-NEXT: movb %bh, %cl ; AVX512F-32-NEXT: movb %bh, %cl
; AVX512F-32-NEXT: kmovd %ecx, %k1 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: andb $2, %cl ; AVX512F-32-NEXT: andb $2, %cl
; AVX512F-32-NEXT: shrb %cl ; AVX512F-32-NEXT: shrb %cl
; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5 ; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $29, %k5, %k5 ; AVX512F-32-NEXT: kshiftrq $29, %k5, %k5
; AVX512F-32-NEXT: kxorq %k2, %k5, %k5 ; AVX512F-32-NEXT: kxorq %k0, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $35, %k5, %k2 ; AVX512F-32-NEXT: kshiftrq $35, %k5, %k0
; AVX512F-32-NEXT: kxorq %k3, %k2, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k0, %k3
; AVX512F-32-NEXT: kmovd %ecx, %k2 ; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $2, %al ; AVX512F-32-NEXT: shrb $2, %al
; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3 ; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $28, %k3, %k3 ; AVX512F-32-NEXT: kshiftrq $28, %k3, %k3
; AVX512F-32-NEXT: kxorq %k5, %k3, %k5 ; AVX512F-32-NEXT: kxorq %k5, %k3, %k5
; AVX512F-32-NEXT: kshiftrq $36, %k5, %k3 ; AVX512F-32-NEXT: kshiftrq $36, %k5, %k3
; AVX512F-32-NEXT: kxorq %k4, %k3, %k4 ; AVX512F-32-NEXT: kxorq %k2, %k3, %k2
; AVX512F-32-NEXT: kmovd %eax, %k3 ; AVX512F-32-NEXT: kmovd %eax, %k3
; AVX512F-32-NEXT: movl %ebx, %eax ; AVX512F-32-NEXT: movl %ebx, %eax
; AVX512F-32-NEXT: shrl $16, %eax ; AVX512F-32-NEXT: shrl $16, %eax
; AVX512F-32-NEXT: shrb $3, %cl ; AVX512F-32-NEXT: shrb $3, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $27, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $27, %k2, %k2
; AVX512F-32-NEXT: kxorq %k5, %k4, %k5 ; AVX512F-32-NEXT: kxorq %k5, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $37, %k5, %k4 ; AVX512F-32-NEXT: kshiftrq $37, %k2, %k5
; AVX512F-32-NEXT: kxorq %k0, %k4, %k0 ; AVX512F-32-NEXT: kxorq %k4, %k5, %k5
; AVX512F-32-NEXT: kmovd %ecx, %k4 ; AVX512F-32-NEXT: kmovd %ecx, %k4
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: shrl $13, %ecx ; AVX512F-32-NEXT: shrl $13, %ecx
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $26, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $26, %k5, %k5
; AVX512F-32-NEXT: kxorq %k5, %k0, %k0 ; AVX512F-32-NEXT: kxorq %k2, %k5, %k2
; AVX512F-32-NEXT: kshiftrq $38, %k0, %k5 ; AVX512F-32-NEXT: kshiftrq $38, %k2, %k5
; AVX512F-32-NEXT: kxorq %k7, %k5, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k5, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k5 ; AVX512F-32-NEXT: kmovd %ecx, %k5
; AVX512F-32-NEXT: movl %eax, %edx ; AVX512F-32-NEXT: movl %eax, %edx
; AVX512F-32-NEXT: andb $2, %dl ; AVX512F-32-NEXT: andb $2, %dl
; AVX512F-32-NEXT: shrb %dl ; AVX512F-32-NEXT: shrb %dl
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $25, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $25, %k7, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k2, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $39, %k7, %k0 ; AVX512F-32-NEXT: kshiftrq $39, %k7, %k2
; AVX512F-32-NEXT: kxorq %k6, %k0, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k2, %k6
; AVX512F-32-NEXT: kmovd %edx, %k0 ; AVX512F-32-NEXT: kmovd %edx, %k2
; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp) # 8-byte Spill
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: andb $15, %cl ; AVX512F-32-NEXT: andb $15, %cl
; AVX512F-32-NEXT: movl %ecx, %edx ; AVX512F-32-NEXT: movl %ecx, %edx
; AVX512F-32-NEXT: shrb $2, %cl ; AVX512F-32-NEXT: shrb $2, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6 ; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $24, %k6, %k6 ; AVX512F-32-NEXT: kshiftrq $24, %k6, %k6
; AVX512F-32-NEXT: kxorq %k7, %k6, %k6 ; AVX512F-32-NEXT: kxorq %k7, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $40, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $40, %k6, %k7
; AVX512F-32-NEXT: kxorq %k1, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k1, %k7, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k1 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: kmovq %k1, {{[0-9]+}}(%esp) # 8-byte Spill
; AVX512F-32-NEXT: movzwl %bx, %ecx
; AVX512F-32-NEXT: movl %ecx, %esi
; AVX512F-32-NEXT: movl %ecx, %edi
; AVX512F-32-NEXT: shrl $12, %ecx ; AVX512F-32-NEXT: shrl $12, %ecx
; AVX512F-32-NEXT: andl $15, %ecx
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $23, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $23, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $41, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $41, %k6, %k7
; AVX512F-32-NEXT: kxorq %k2, %k7, %k2 ; AVX512F-32-NEXT: kxorq %k0, %k7, %k0
; AVX512F-32-NEXT: kmovd %ecx, %k0 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: shrl $14, %edi
; AVX512F-32-NEXT: shrl $14, %ecx ; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: andl $3, %ecx ; AVX512F-32-NEXT: kshiftrq $22, %k0, %k0
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k6, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $22, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $42, %k0, %k6
; AVX512F-32-NEXT: kxorq %k6, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $42, %k2, %k6
; AVX512F-32-NEXT: kxorq %k3, %k6, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k6, %k3
; AVX512F-32-NEXT: kmovd %ecx, %k7 ; AVX512F-32-NEXT: kmovd %edi, %k7
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: shrl $15, %esi
; AVX512F-32-NEXT: shrl $15, %ecx
; AVX512F-32-NEXT: andl $1, %ecx
; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3 ; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $21, %k3, %k3 ; AVX512F-32-NEXT: kshiftrq $21, %k3, %k3
; AVX512F-32-NEXT: kxorq %k2, %k3, %k2 ; AVX512F-32-NEXT: kxorq %k0, %k3, %k0
; AVX512F-32-NEXT: kshiftrq $43, %k2, %k3 ; AVX512F-32-NEXT: kshiftrq $43, %k0, %k3
; AVX512F-32-NEXT: kxorq %k4, %k3, %k3 ; AVX512F-32-NEXT: kxorq %k4, %k3, %k3
; AVX512F-32-NEXT: kmovd %ecx, %k6 ; AVX512F-32-NEXT: kmovd %esi, %k6
; AVX512F-32-NEXT: shrb $3, %dl ; AVX512F-32-NEXT: shrb $3, %dl
; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3 ; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $20, %k3, %k3 ; AVX512F-32-NEXT: kshiftrq $20, %k3, %k3
; AVX512F-32-NEXT: kxorq %k2, %k3, %k3 ; AVX512F-32-NEXT: kxorq %k0, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $44, %k3, %k2 ; AVX512F-32-NEXT: kshiftrq $44, %k3, %k0
; AVX512F-32-NEXT: kxorq %k0, %k2, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k0, %k1
; AVX512F-32-NEXT: kmovd %edx, %k2 ; AVX512F-32-NEXT: kmovd %edx, %k0
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $4, %cl ; AVX512F-32-NEXT: shrb $4, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $19, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $19, %k1, %k1
; AVX512F-32-NEXT: kxorq %k3, %k0, %k0 ; AVX512F-32-NEXT: kxorq %k3, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $45, %k0, %k3 ; AVX512F-32-NEXT: kshiftrq $45, %k1, %k3
; AVX512F-32-NEXT: kxorq %k5, %k3, %k4 ; AVX512F-32-NEXT: kxorq %k5, %k3, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k3 ; AVX512F-32-NEXT: kmovd %ecx, %k3
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $5, %cl ; AVX512F-32-NEXT: shrb $5, %cl
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $18, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $18, %k4, %k4
; AVX512F-32-NEXT: kxorq %k0, %k4, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k4, %k1
; AVX512F-32-NEXT: kshiftrq $46, %k0, %k4 ; AVX512F-32-NEXT: kshiftrq $46, %k1, %k4
; AVX512F-32-NEXT: kxorq %k7, %k4, %k5 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k5
; AVX512F-32-NEXT: kmovd %ecx, %k4 ; AVX512F-32-NEXT: kmovd %ecx, %k4
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $6, %cl ; AVX512F-32-NEXT: shrb $6, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5 ; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $17, %k5, %k5 ; AVX512F-32-NEXT: kshiftrq $17, %k5, %k5
; AVX512F-32-NEXT: kxorq %k0, %k5, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k5, %k1
; AVX512F-32-NEXT: kshiftrq $47, %k0, %k5 ; AVX512F-32-NEXT: kshiftrq $47, %k1, %k5
; AVX512F-32-NEXT: kxorq %k6, %k5, %k5 ; AVX512F-32-NEXT: kxorq %k6, %k5, %k5
; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5 ; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $16, %k5, %k5 ; AVX512F-32-NEXT: kshiftrq $16, %k5, %k5
; AVX512F-32-NEXT: kxorq %k0, %k5, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k5, %k1
; AVX512F-32-NEXT: kshiftrq $48, %k0, %k5 ; AVX512F-32-NEXT: kshiftrq $48, %k1, %k5
; AVX512F-32-NEXT: kmovd %eax, %k6 ; AVX512F-32-NEXT: kmovd %eax, %k6
; AVX512F-32-NEXT: kxorq %k6, %k5, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k5, %k6
; AVX512F-32-NEXT: kmovd %ecx, %k5 ; AVX512F-32-NEXT: kmovd %ecx, %k5
; AVX512F-32-NEXT: movl %ebx, %edx ; AVX512F-32-NEXT: movl %ebx, %edx
; AVX512F-32-NEXT: shrl $24, %edx ; AVX512F-32-NEXT: shrl $24, %edx
; AVX512F-32-NEXT: # kill: def %al killed %al killed %eax def %eax ; AVX512F-32-NEXT: # kill: def %al killed %al killed %eax def %eax
; AVX512F-32-NEXT: shrb $7, %al ; AVX512F-32-NEXT: shrb $7, %al
; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6 ; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $15, %k6, %k6 ; AVX512F-32-NEXT: kshiftrq $15, %k6, %k6
; AVX512F-32-NEXT: kxorq %k0, %k6, %k6 ; AVX512F-32-NEXT: kxorq %k1, %k6, %k1
; AVX512F-32-NEXT: kshiftrq $49, %k6, %k0 ; AVX512F-32-NEXT: kshiftrq $49, %k1, %k6
; AVX512F-32-NEXT: kmovq {{[0-9]+}}(%esp), %k7 # 8-byte Reload ; AVX512F-32-NEXT: kxorq %k2, %k6, %k6
; AVX512F-32-NEXT: kxorq %k7, %k0, %k7 ; AVX512F-32-NEXT: kmovd %eax, %k2
; AVX512F-32-NEXT: kmovd %eax, %k0
; AVX512F-32-NEXT: movl %edx, %eax ; AVX512F-32-NEXT: movl %edx, %eax
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $14, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $14, %k6, %k6
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k1, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $50, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $50, %k6, %k1
; AVX512F-32-NEXT: kxorq %k1, %k7, %k7 ; AVX512F-32-NEXT: kmovq {{[0-9]+}}(%esp), %k7 # 8-byte Reload
; AVX512F-32-NEXT: kxorq %k7, %k1, %k7
; AVX512F-32-NEXT: kmovd %edx, %k1 ; AVX512F-32-NEXT: kmovd %edx, %k1
; AVX512F-32-NEXT: # kill: def %dl killed %dl killed %edx def %edx ; AVX512F-32-NEXT: # kill: def %dl killed %dl killed %edx def %edx
; AVX512F-32-NEXT: andb $15, %dl ; AVX512F-32-NEXT: andb $15, %dl
; AVX512F-32-NEXT: andb $2, %al ; AVX512F-32-NEXT: andb $2, %al
; AVX512F-32-NEXT: shrb %al ; AVX512F-32-NEXT: shrb %al
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $13, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $13, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $51, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $51, %k6, %k7
; AVX512F-32-NEXT: kxorq %k2, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k0, %k7, %k7
; AVX512F-32-NEXT: kmovd %eax, %k2 ; AVX512F-32-NEXT: kmovd %eax, %k0
; AVX512F-32-NEXT: movl %edx, %eax ; AVX512F-32-NEXT: movl %edx, %eax
; AVX512F-32-NEXT: shrb $2, %dl ; AVX512F-32-NEXT: shrb $2, %dl
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $12, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $12, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $52, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $52, %k6, %k7
; AVX512F-32-NEXT: kxorq %k3, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k3, %k7, %k7
; AVX512F-32-NEXT: kmovd %edx, %k3 ; AVX512F-32-NEXT: kmovd %edx, %k3
Show All 12 Lines
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $54, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $54, %k6, %k7
; AVX512F-32-NEXT: kxorq %k5, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k5, %k7, %k7
; AVX512F-32-NEXT: kmovd %eax, %k5 ; AVX512F-32-NEXT: kmovd %eax, %k5
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $9, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $9, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $55, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $55, %k6, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k0 ; AVX512F-32-NEXT: kxorq %k2, %k7, %k2
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $8, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $8, %k2, %k2
; AVX512F-32-NEXT: kxorq %k6, %k0, %k0 ; AVX512F-32-NEXT: kxorq %k6, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $56, %k0, %k6 ; AVX512F-32-NEXT: kshiftrq $56, %k2, %k6
; AVX512F-32-NEXT: kxorq %k1, %k6, %k1 ; AVX512F-32-NEXT: kxorq %k1, %k6, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $7, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $7, %k1, %k1
; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $57, %k0, %k1
; AVX512F-32-NEXT: kxorq %k2, %k1, %k1 ; AVX512F-32-NEXT: kxorq %k2, %k1, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $57, %k1, %k2
; AVX512F-32-NEXT: kshiftrq $6, %k1, %k1 ; AVX512F-32-NEXT: kxorq %k0, %k2, %k0
; AVX512F-32-NEXT: kxorq %k0, %k1, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $6, %k0, %k0
; AVX512F-32-NEXT: kxorq %k1, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $58, %k0, %k1 ; AVX512F-32-NEXT: kshiftrq $58, %k0, %k1
; AVX512F-32-NEXT: kxorq %k3, %k1, %k1 ; AVX512F-32-NEXT: kxorq %k3, %k1, %k1
; AVX512F-32-NEXT: movl %ebx, %eax ; AVX512F-32-NEXT: movl %ebx, %eax
; AVX512F-32-NEXT: shrl $28, %eax ; AVX512F-32-NEXT: shrl $28, %eax
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $5, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $5, %k1, %k1
; AVX512F-32-NEXT: kxorq %k0, %k1, %k0 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $59, %k0, %k1 ; AVX512F-32-NEXT: kshiftrq $59, %k0, %k1
▲ Show 20 Lines • Show All 46 Lines • ▼ Show 20 Lines
; AVX512F-32-NEXT: vpcmpleb %zmm0, %zmm1, %k0 {%k1} ; AVX512F-32-NEXT: vpcmpleb %zmm0, %zmm1, %k0 {%k1}
; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp) ; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax ; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax
; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx ; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx
; AVX512F-32-NEXT: vpcmpgtb %zmm1, %zmm0, %k0 {%k1} ; AVX512F-32-NEXT: vpcmpgtb %zmm1, %zmm0, %k0 {%k1}
; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp) ; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax ; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax
; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx ; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx
; AVX512F-32-NEXT: addl %esi, %eax ; AVX512F-32-NEXT: addl %ebp, %eax
; AVX512F-32-NEXT: adcl %ebx, %edx ; AVX512F-32-NEXT: adcl %ebx, %edx
; AVX512F-32-NEXT: addl $60, %esp ; AVX512F-32-NEXT: addl $60, %esp
; AVX512F-32-NEXT: popl %esi ; AVX512F-32-NEXT: popl %esi
; AVX512F-32-NEXT: popl %edi
; AVX512F-32-NEXT: popl %ebx ; AVX512F-32-NEXT: popl %ebx
; AVX512F-32-NEXT: popl %ebp
; AVX512F-32-NEXT: vzeroupper ; AVX512F-32-NEXT: vzeroupper
; AVX512F-32-NEXT: retl ; AVX512F-32-NEXT: retl
%res0 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 0, i64 %mask) %res0 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 0, i64 %mask)
%res1 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 1, i64 %mask) %res1 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 1, i64 %mask)
%ret1 = add i64 %res0, %res1 %ret1 = add i64 %res0, %res1
%res2 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 2, i64 %mask) %res2 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 2, i64 %mask)
%ret2 = add i64 %ret1, %res2 %ret2 = add i64 %ret1, %res2
%res3 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 3, i64 %mask) %res3 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 3, i64 %mask)
▲ Show 20 Lines • Show All 112 Lines • ▼ Show 20 Lines
; AVX512BW-NEXT: kmovq %k0, %rax ; AVX512BW-NEXT: kmovq %k0, %rax
; AVX512BW-NEXT: addq %rdx, %rax ; AVX512BW-NEXT: addq %rdx, %rax
; AVX512BW-NEXT: addq %rdi, %rax ; AVX512BW-NEXT: addq %rdi, %rax
; AVX512BW-NEXT: vzeroupper ; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq ; AVX512BW-NEXT: retq
; ;
; AVX512F-32-LABEL: test_mask_x86_avx512_ucmp_b_512: ; AVX512F-32-LABEL: test_mask_x86_avx512_ucmp_b_512:
; AVX512F-32: # %bb.0: ; AVX512F-32: # %bb.0:
; AVX512F-32-NEXT: pushl %ebx ; AVX512F-32-NEXT: pushl %ebp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 8 ; AVX512F-32-NEXT: .cfi_def_cfa_offset 8
; AVX512F-32-NEXT: pushl %esi ; AVX512F-32-NEXT: pushl %ebx
; AVX512F-32-NEXT: .cfi_def_cfa_offset 12 ; AVX512F-32-NEXT: .cfi_def_cfa_offset 12
; AVX512F-32-NEXT: pushl %edi
; AVX512F-32-NEXT: .cfi_def_cfa_offset 16
; AVX512F-32-NEXT: pushl %esi
; AVX512F-32-NEXT: .cfi_def_cfa_offset 20
; AVX512F-32-NEXT: subl $60, %esp ; AVX512F-32-NEXT: subl $60, %esp
; AVX512F-32-NEXT: .cfi_def_cfa_offset 72 ; AVX512F-32-NEXT: .cfi_def_cfa_offset 80
; AVX512F-32-NEXT: .cfi_offset %esi, -12 ; AVX512F-32-NEXT: .cfi_offset %esi, -20
; AVX512F-32-NEXT: .cfi_offset %ebx, -8 ; AVX512F-32-NEXT: .cfi_offset %edi, -16
; AVX512F-32-NEXT: .cfi_offset %ebx, -12
; AVX512F-32-NEXT: .cfi_offset %ebp, -8
; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %ebx ; AVX512F-32-NEXT: movl {{[0-9]+}}(%esp), %ebx
; AVX512F-32-NEXT: movl %ebx, %eax ; AVX512F-32-NEXT: movl %ebx, %eax
; AVX512F-32-NEXT: shrl $16, %eax ; AVX512F-32-NEXT: shrl $16, %eax
; AVX512F-32-NEXT: movl %ebx, %edx ; AVX512F-32-NEXT: movl %ebx, %edx
; AVX512F-32-NEXT: andb $15, %dl ; AVX512F-32-NEXT: andb $15, %dl
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: andb $2, %cl ; AVX512F-32-NEXT: andb $2, %cl
; AVX512F-32-NEXT: shrb %cl ; AVX512F-32-NEXT: shrb %cl
Show All 23 Lines
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $62, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $62, %k1, %k1
; AVX512F-32-NEXT: kxorq %k5, %k1, %k7 ; AVX512F-32-NEXT: kxorq %k5, %k1, %k7
; AVX512F-32-NEXT: kshiftrq $2, %k7, %k1 ; AVX512F-32-NEXT: kshiftrq $2, %k7, %k1
; AVX512F-32-NEXT: kxorq %k2, %k1, %k2 ; AVX512F-32-NEXT: kxorq %k2, %k1, %k2
; AVX512F-32-NEXT: kmovd %ecx, %k5 ; AVX512F-32-NEXT: kmovd %ecx, %k5
; AVX512F-32-NEXT: movb %bh, %cl ; AVX512F-32-NEXT: movb %bh, %cl
; AVX512F-32-NEXT: kmovd %ecx, %k1 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: movl %ebx, %esi ; AVX512F-32-NEXT: movl %ebx, %ebp
; AVX512F-32-NEXT: andb $2, %cl ; AVX512F-32-NEXT: andb $2, %cl
; AVX512F-32-NEXT: shrb %cl ; AVX512F-32-NEXT: shrb %cl
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $61, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $61, %k2, %k2
; AVX512F-32-NEXT: kxorq %k7, %k2, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k2, %k7
; AVX512F-32-NEXT: kshiftrq $3, %k7, %k2 ; AVX512F-32-NEXT: kshiftrq $3, %k7, %k2
; AVX512F-32-NEXT: kxorq %k0, %k2, %k0 ; AVX512F-32-NEXT: kxorq %k0, %k2, %k0
; AVX512F-32-NEXT: kmovd %ecx, %k2 ; AVX512F-32-NEXT: kmovd %ecx, %k2
Show All 9 Lines
; AVX512F-32-NEXT: andb $15, %dl ; AVX512F-32-NEXT: andb $15, %dl
; AVX512F-32-NEXT: shrb $3, %cl ; AVX512F-32-NEXT: shrb $3, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $59, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $59, %k7, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k0, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $5, %k7, %k0 ; AVX512F-32-NEXT: kshiftrq $5, %k7, %k0
; AVX512F-32-NEXT: kxorq %k4, %k0, %k4 ; AVX512F-32-NEXT: kxorq %k4, %k0, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k0 ; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %esi, %ecx ; AVX512F-32-NEXT: movl %ebp, %ecx
; AVX512F-32-NEXT: shrl $13, %ecx ; AVX512F-32-NEXT: shrl $13, %ecx
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $58, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $58, %k4, %k4
; AVX512F-32-NEXT: kxorq %k7, %k4, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k7
; AVX512F-32-NEXT: kshiftrq $6, %k7, %k4 ; AVX512F-32-NEXT: kshiftrq $6, %k7, %k4
; AVX512F-32-NEXT: kxorq %k6, %k4, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k4, %k6
; AVX512F-32-NEXT: kmovd %ecx, %k4 ; AVX512F-32-NEXT: kmovd %ecx, %k4
Show All 28 Lines
; AVX512F-32-NEXT: kshiftrq $54, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $54, %k2, %k2
; AVX512F-32-NEXT: kxorq %k7, %k2, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k2, %k7
; AVX512F-32-NEXT: kshiftrq $10, %k7, %k2 ; AVX512F-32-NEXT: kshiftrq $10, %k7, %k2
; AVX512F-32-NEXT: kxorq %k3, %k2, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k2, %k3
; AVX512F-32-NEXT: kmovd %ecx, %k2 ; AVX512F-32-NEXT: kmovd %ecx, %k2
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $5, %cl ; AVX512F-32-NEXT: shrb $5, %cl
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: movl %esi, %edx ; AVX512F-32-NEXT: movzwl %bp, %edx
; AVX512F-32-NEXT: movl %edx, %esi
; AVX512F-32-NEXT: movl %edx, %edi
; AVX512F-32-NEXT: shrl $12, %edx ; AVX512F-32-NEXT: shrl $12, %edx
; AVX512F-32-NEXT: andl $15, %edx
; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3 ; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $53, %k3, %k3 ; AVX512F-32-NEXT: kshiftrq $53, %k3, %k3
; AVX512F-32-NEXT: kxorq %k7, %k3, %k3 ; AVX512F-32-NEXT: kxorq %k7, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $11, %k3, %k7 ; AVX512F-32-NEXT: kshiftrq $11, %k3, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k0 ; AVX512F-32-NEXT: kxorq %k0, %k7, %k0
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $52, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $52, %k0, %k0
; AVX512F-32-NEXT: kxorq %k3, %k0, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k0, %k3
; AVX512F-32-NEXT: kshiftrq $12, %k3, %k0 ; AVX512F-32-NEXT: kshiftrq $12, %k3, %k0
; AVX512F-32-NEXT: kmovd %edx, %k7 ; AVX512F-32-NEXT: kmovd %edx, %k7
; AVX512F-32-NEXT: kxorq %k7, %k0, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k0, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k0 ; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %eax, %edx ; AVX512F-32-NEXT: movl %eax, %edx
; AVX512F-32-NEXT: shrb $6, %dl ; AVX512F-32-NEXT: shrb $6, %dl
; AVX512F-32-NEXT: movl %esi, %ecx ; AVX512F-32-NEXT: shrl $15, %esi
; AVX512F-32-NEXT: shrl $14, %ecx ; AVX512F-32-NEXT: shrl $14, %edi
; AVX512F-32-NEXT: andl $3, %ecx
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $51, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $51, %k7, %k7
; AVX512F-32-NEXT: kxorq %k3, %k7, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k7, %k3
; AVX512F-32-NEXT: kshiftrq $13, %k3, %k7 ; AVX512F-32-NEXT: kshiftrq $13, %k3, %k7
; AVX512F-32-NEXT: kxorq %k4, %k7, %k4 ; AVX512F-32-NEXT: kxorq %k4, %k7, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k7
; AVX512F-32-NEXT: movl %esi, %ecx
; AVX512F-32-NEXT: shrl $15, %ecx
; AVX512F-32-NEXT: andl $1, %ecx
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $50, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $50, %k4, %k4
; AVX512F-32-NEXT: kxorq %k3, %k4, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k4, %k3
; AVX512F-32-NEXT: kshiftrq $14, %k3, %k4 ; AVX512F-32-NEXT: kshiftrq $14, %k3, %k4
; AVX512F-32-NEXT: kmovd %edi, %k7
; AVX512F-32-NEXT: kxorq %k7, %k4, %k4 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k4
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $49, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $49, %k4, %k4
; AVX512F-32-NEXT: kxorq %k3, %k4, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k4, %k3
; AVX512F-32-NEXT: kshiftrq $15, %k3, %k4 ; AVX512F-32-NEXT: kshiftrq $15, %k3, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k7 ; AVX512F-32-NEXT: kmovd %esi, %k7
; AVX512F-32-NEXT: kxorq %k7, %k4, %k4 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k4
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $48, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $48, %k4, %k4
; AVX512F-32-NEXT: kxorq %k3, %k4, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k4, %k3
; AVX512F-32-NEXT: kshiftrq $16, %k3, %k4 ; AVX512F-32-NEXT: kshiftrq $16, %k3, %k4
; AVX512F-32-NEXT: kmovd %eax, %k7 ; AVX512F-32-NEXT: kmovd %eax, %k7
; AVX512F-32-NEXT: kxorq %k7, %k4, %k4 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k4
; AVX512F-32-NEXT: kmovd %edx, %k7 ; AVX512F-32-NEXT: kmovd %edx, %k7
; AVX512F-32-NEXT: movl %esi, %edx ; AVX512F-32-NEXT: movl %ebp, %edx
; AVX512F-32-NEXT: shrl $24, %edx ; AVX512F-32-NEXT: shrl $24, %edx
; AVX512F-32-NEXT: # kill: def %al killed %al killed %eax def %eax ; AVX512F-32-NEXT: # kill: def %al killed %al killed %eax def %eax
; AVX512F-32-NEXT: shrb $7, %al ; AVX512F-32-NEXT: shrb $7, %al
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $47, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $47, %k4, %k4
; AVX512F-32-NEXT: kxorq %k3, %k4, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k4, %k3
; AVX512F-32-NEXT: kshiftrq $17, %k3, %k4 ; AVX512F-32-NEXT: kshiftrq $17, %k3, %k4
; AVX512F-32-NEXT: kxorq %k5, %k4, %k4 ; AVX512F-32-NEXT: kxorq %k5, %k4, %k4
Show All 23 Lines
; AVX512F-32-NEXT: kshiftrq $20, %k1, %k6 ; AVX512F-32-NEXT: kshiftrq $20, %k1, %k6
; AVX512F-32-NEXT: kxorq %k2, %k6, %k6 ; AVX512F-32-NEXT: kxorq %k2, %k6, %k6
; AVX512F-32-NEXT: kmovd %edx, %k2 ; AVX512F-32-NEXT: kmovd %edx, %k2
; AVX512F-32-NEXT: movl %ebx, %eax ; AVX512F-32-NEXT: movl %ebx, %eax
; AVX512F-32-NEXT: andb $15, %al ; AVX512F-32-NEXT: andb $15, %al
; AVX512F-32-NEXT: shrb $3, %cl ; AVX512F-32-NEXT: shrb $3, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6 ; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $43, %k6, %k6 ; AVX512F-32-NEXT: kshiftrq $43, %k6, %k6
; AVX512F-32-NEXT: kxorq %k1, %k6, %k1
; AVX512F-32-NEXT: kshiftrq $21, %k1, %k6
; AVX512F-32-NEXT: kxorq %k0, %k6, %k6
; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %esi, %ecx
; AVX512F-32-NEXT: shrl $29, %ecx
; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $42, %k6, %k6
; AVX512F-32-NEXT: kxorq %k1, %k6, %k6 ; AVX512F-32-NEXT: kxorq %k1, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $22, %k6, %k1 ; AVX512F-32-NEXT: kshiftrq $21, %k6, %k1
; AVX512F-32-NEXT: kxorq %k7, %k1, %k7 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kmovd %ecx, %k1 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: movl %ebp, %ecx
; AVX512F-32-NEXT: shrl $29, %ecx
; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $42, %k0, %k0
; AVX512F-32-NEXT: kxorq %k6, %k0, %k6
; AVX512F-32-NEXT: kshiftrq $22, %k6, %k0
; AVX512F-32-NEXT: kxorq %k7, %k0, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %ebx, %edx ; AVX512F-32-NEXT: movl %ebx, %edx
; AVX512F-32-NEXT: andb $2, %dl ; AVX512F-32-NEXT: andb $2, %dl
; AVX512F-32-NEXT: shrb %dl ; AVX512F-32-NEXT: shrb %dl
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $41, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $41, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $23, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $23, %k6, %k7
; AVX512F-32-NEXT: kxorq %k5, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k5, %k7, %k7
Show All 14 Lines
; AVX512F-32-NEXT: kxorq %k7, %k3, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k3, %k7
; AVX512F-32-NEXT: kshiftrq $25, %k7, %k3 ; AVX512F-32-NEXT: kshiftrq $25, %k7, %k3
; AVX512F-32-NEXT: kxorq %k4, %k3, %k4 ; AVX512F-32-NEXT: kxorq %k4, %k3, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k3 ; AVX512F-32-NEXT: kmovd %ecx, %k3
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: shrb $4, %cl ; AVX512F-32-NEXT: shrb $4, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $38, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $38, %k4, %k4
; AVX512F-32-NEXT: kxorq %k7, %k4, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $26, %k7, %k4 ; AVX512F-32-NEXT: kshiftrq $26, %k4, %k7
; AVX512F-32-NEXT: kxorq %k2, %k4, %k2 ; AVX512F-32-NEXT: kxorq %k2, %k7, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k4 ; AVX512F-32-NEXT: kmovd %ecx, %k2
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: shrb $5, %cl ; AVX512F-32-NEXT: shrb $5, %cl
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: movl %esi, %edx ; AVX512F-32-NEXT: movl %ebp, %edx
; AVX512F-32-NEXT: shrl $28, %edx ; AVX512F-32-NEXT: shrl $28, %edx
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $37, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $37, %k7, %k7
; AVX512F-32-NEXT: kxorq %k7, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k4, %k7, %k4
; AVX512F-32-NEXT: kshiftrq $27, %k2, %k7 ; AVX512F-32-NEXT: kshiftrq $27, %k4, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k7, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $36, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $36, %k1, %k1
; AVX512F-32-NEXT: kxorq %k2, %k0, %k2 ; AVX512F-32-NEXT: kxorq %k4, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $28, %k2, %k0 ; AVX512F-32-NEXT: kshiftrq $28, %k1, %k4
; AVX512F-32-NEXT: kmovd %edx, %k7 ; AVX512F-32-NEXT: kmovd %edx, %k7
; AVX512F-32-NEXT: kxorq %k7, %k0, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k0 ; AVX512F-32-NEXT: kmovd %ecx, %k4
; AVX512F-32-NEXT: movl %ebx, %edx ; AVX512F-32-NEXT: movl %ebx, %edx
; AVX512F-32-NEXT: shrb $6, %dl ; AVX512F-32-NEXT: shrb $6, %dl
; AVX512F-32-NEXT: movl %esi, %ecx ; AVX512F-32-NEXT: movl %ebp, %ecx
; AVX512F-32-NEXT: shrl $30, %ecx ; AVX512F-32-NEXT: shrl $31, %ecx
; AVX512F-32-NEXT: movl %ebp, %esi
; AVX512F-32-NEXT: shrl $30, %esi
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $35, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $35, %k7, %k7
; AVX512F-32-NEXT: kxorq %k2, %k7, %k2
; AVX512F-32-NEXT: kshiftrq $29, %k2, %k7
; AVX512F-32-NEXT: kxorq %k1, %k7, %k1 ; AVX512F-32-NEXT: kxorq %k1, %k7, %k1
; AVX512F-32-NEXT: kmovd %ecx, %k7 ; AVX512F-32-NEXT: kshiftrq $29, %k1, %k7
; AVX512F-32-NEXT: movl %esi, %ecx ; AVX512F-32-NEXT: kxorq %k0, %k7, %k0
; AVX512F-32-NEXT: shrl $31, %ecx ; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $34, %k0, %k0
; AVX512F-32-NEXT: kxorq %k1, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $30, %k0, %k1
; AVX512F-32-NEXT: kmovd %esi, %k7
; AVX512F-32-NEXT: kxorq %k7, %k1, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $34, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $33, %k1, %k1
; AVX512F-32-NEXT: kxorq %k2, %k1, %k1 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $30, %k1, %k2 ; AVX512F-32-NEXT: kshiftrq $31, %k0, %k1
; AVX512F-32-NEXT: kxorq %k7, %k2, %k2
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $33, %k2, %k2
; AVX512F-32-NEXT: kxorq %k1, %k2, %k1
; AVX512F-32-NEXT: kshiftrq $31, %k1, %k2
; AVX512F-32-NEXT: kmovd %ecx, %k7 ; AVX512F-32-NEXT: kmovd %ecx, %k7
; AVX512F-32-NEXT: kxorq %k7, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k7, %k1, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $32, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $32, %k1, %k1
; AVX512F-32-NEXT: kxorq %k1, %k2, %k1 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $32, %k1, %k2 ; AVX512F-32-NEXT: kshiftrq $32, %k0, %k1
; AVX512F-32-NEXT: kmovd %ebx, %k7 ; AVX512F-32-NEXT: kmovd %ebx, %k7
; AVX512F-32-NEXT: kxorq %k7, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k7, %k1, %k1
; AVX512F-32-NEXT: kmovd %edx, %k7 ; AVX512F-32-NEXT: kmovd %edx, %k7
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: shrb $7, %cl ; AVX512F-32-NEXT: shrb $7, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $31, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $31, %k1, %k1
; AVX512F-32-NEXT: kxorq %k1, %k2, %k1 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $33, %k1, %k2 ; AVX512F-32-NEXT: kshiftrq $33, %k0, %k1
; AVX512F-32-NEXT: kxorq %k5, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k5, %k1, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $30, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $30, %k1, %k1
; AVX512F-32-NEXT: kxorq %k1, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $34, %k2, %k1 ; AVX512F-32-NEXT: kshiftrq $34, %k0, %k1
; AVX512F-32-NEXT: kxorq %k6, %k1, %k5 ; AVX512F-32-NEXT: kxorq %k6, %k1, %k5
; AVX512F-32-NEXT: kmovd %ecx, %k6 ; AVX512F-32-NEXT: kmovd %ecx, %k6
; AVX512F-32-NEXT: movb %bh, %cl ; AVX512F-32-NEXT: movb %bh, %cl
; AVX512F-32-NEXT: kmovd %ecx, %k1 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: andb $2, %cl ; AVX512F-32-NEXT: andb $2, %cl
; AVX512F-32-NEXT: shrb %cl ; AVX512F-32-NEXT: shrb %cl
; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5 ; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $29, %k5, %k5 ; AVX512F-32-NEXT: kshiftrq $29, %k5, %k5
; AVX512F-32-NEXT: kxorq %k2, %k5, %k5 ; AVX512F-32-NEXT: kxorq %k0, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $35, %k5, %k2 ; AVX512F-32-NEXT: kshiftrq $35, %k5, %k0
; AVX512F-32-NEXT: kxorq %k3, %k2, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k0, %k3
; AVX512F-32-NEXT: kmovd %ecx, %k2 ; AVX512F-32-NEXT: kmovd %ecx, %k0
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $2, %al ; AVX512F-32-NEXT: shrb $2, %al
; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3 ; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $28, %k3, %k3 ; AVX512F-32-NEXT: kshiftrq $28, %k3, %k3
; AVX512F-32-NEXT: kxorq %k5, %k3, %k5 ; AVX512F-32-NEXT: kxorq %k5, %k3, %k5
; AVX512F-32-NEXT: kshiftrq $36, %k5, %k3 ; AVX512F-32-NEXT: kshiftrq $36, %k5, %k3
; AVX512F-32-NEXT: kxorq %k4, %k3, %k4 ; AVX512F-32-NEXT: kxorq %k2, %k3, %k2
; AVX512F-32-NEXT: kmovd %eax, %k3 ; AVX512F-32-NEXT: kmovd %eax, %k3
; AVX512F-32-NEXT: movl %ebx, %eax ; AVX512F-32-NEXT: movl %ebx, %eax
; AVX512F-32-NEXT: shrl $16, %eax ; AVX512F-32-NEXT: shrl $16, %eax
; AVX512F-32-NEXT: shrb $3, %cl ; AVX512F-32-NEXT: shrb $3, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $27, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $27, %k2, %k2
; AVX512F-32-NEXT: kxorq %k5, %k4, %k5 ; AVX512F-32-NEXT: kxorq %k5, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $37, %k5, %k4 ; AVX512F-32-NEXT: kshiftrq $37, %k2, %k5
; AVX512F-32-NEXT: kxorq %k0, %k4, %k0 ; AVX512F-32-NEXT: kxorq %k4, %k5, %k5
; AVX512F-32-NEXT: kmovd %ecx, %k4 ; AVX512F-32-NEXT: kmovd %ecx, %k4
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: movl %ebx, %ecx
; AVX512F-32-NEXT: shrl $13, %ecx ; AVX512F-32-NEXT: shrl $13, %ecx
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $26, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $26, %k5, %k5
; AVX512F-32-NEXT: kxorq %k5, %k0, %k0 ; AVX512F-32-NEXT: kxorq %k2, %k5, %k2
; AVX512F-32-NEXT: kshiftrq $38, %k0, %k5 ; AVX512F-32-NEXT: kshiftrq $38, %k2, %k5
; AVX512F-32-NEXT: kxorq %k7, %k5, %k7 ; AVX512F-32-NEXT: kxorq %k7, %k5, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k5 ; AVX512F-32-NEXT: kmovd %ecx, %k5
; AVX512F-32-NEXT: movl %eax, %edx ; AVX512F-32-NEXT: movl %eax, %edx
; AVX512F-32-NEXT: andb $2, %dl ; AVX512F-32-NEXT: andb $2, %dl
; AVX512F-32-NEXT: shrb %dl ; AVX512F-32-NEXT: shrb %dl
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $25, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $25, %k7, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k2, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $39, %k7, %k0 ; AVX512F-32-NEXT: kshiftrq $39, %k7, %k2
; AVX512F-32-NEXT: kxorq %k6, %k0, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k2, %k6
; AVX512F-32-NEXT: kmovd %edx, %k0 ; AVX512F-32-NEXT: kmovd %edx, %k2
; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp) # 8-byte Spill
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: andb $15, %cl ; AVX512F-32-NEXT: andb $15, %cl
; AVX512F-32-NEXT: movl %ecx, %edx ; AVX512F-32-NEXT: movl %ecx, %edx
; AVX512F-32-NEXT: shrb $2, %cl ; AVX512F-32-NEXT: shrb $2, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6 ; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $24, %k6, %k6 ; AVX512F-32-NEXT: kshiftrq $24, %k6, %k6
; AVX512F-32-NEXT: kxorq %k7, %k6, %k6 ; AVX512F-32-NEXT: kxorq %k7, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $40, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $40, %k6, %k7
; AVX512F-32-NEXT: kxorq %k1, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k1, %k7, %k7
; AVX512F-32-NEXT: kmovd %ecx, %k1 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: kmovq %k1, {{[0-9]+}}(%esp) # 8-byte Spill
; AVX512F-32-NEXT: movzwl %bx, %ecx
; AVX512F-32-NEXT: movl %ecx, %esi
; AVX512F-32-NEXT: movl %ecx, %edi
; AVX512F-32-NEXT: shrl $12, %ecx ; AVX512F-32-NEXT: shrl $12, %ecx
; AVX512F-32-NEXT: andl $15, %ecx
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $23, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $23, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $41, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $41, %k6, %k7
; AVX512F-32-NEXT: kxorq %k2, %k7, %k2 ; AVX512F-32-NEXT: kxorq %k0, %k7, %k0
; AVX512F-32-NEXT: kmovd %ecx, %k0 ; AVX512F-32-NEXT: kmovd %ecx, %k1
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: shrl $14, %edi
; AVX512F-32-NEXT: shrl $14, %ecx ; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: andl $3, %ecx ; AVX512F-32-NEXT: kshiftrq $22, %k0, %k0
; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2 ; AVX512F-32-NEXT: kxorq %k6, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $22, %k2, %k2 ; AVX512F-32-NEXT: kshiftrq $42, %k0, %k6
; AVX512F-32-NEXT: kxorq %k6, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $42, %k2, %k6
; AVX512F-32-NEXT: kxorq %k3, %k6, %k3 ; AVX512F-32-NEXT: kxorq %k3, %k6, %k3
; AVX512F-32-NEXT: kmovd %ecx, %k7 ; AVX512F-32-NEXT: kmovd %edi, %k7
; AVX512F-32-NEXT: movl %ebx, %ecx ; AVX512F-32-NEXT: shrl $15, %esi
; AVX512F-32-NEXT: shrl $15, %ecx
; AVX512F-32-NEXT: andl $1, %ecx
; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3 ; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $21, %k3, %k3 ; AVX512F-32-NEXT: kshiftrq $21, %k3, %k3
; AVX512F-32-NEXT: kxorq %k2, %k3, %k2 ; AVX512F-32-NEXT: kxorq %k0, %k3, %k0
; AVX512F-32-NEXT: kshiftrq $43, %k2, %k3 ; AVX512F-32-NEXT: kshiftrq $43, %k0, %k3
; AVX512F-32-NEXT: kxorq %k4, %k3, %k3 ; AVX512F-32-NEXT: kxorq %k4, %k3, %k3
; AVX512F-32-NEXT: kmovd %ecx, %k6 ; AVX512F-32-NEXT: kmovd %esi, %k6
; AVX512F-32-NEXT: shrb $3, %dl ; AVX512F-32-NEXT: shrb $3, %dl
; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3 ; AVX512F-32-NEXT: kshiftlq $63, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $20, %k3, %k3 ; AVX512F-32-NEXT: kshiftrq $20, %k3, %k3
; AVX512F-32-NEXT: kxorq %k2, %k3, %k3 ; AVX512F-32-NEXT: kxorq %k0, %k3, %k3
; AVX512F-32-NEXT: kshiftrq $44, %k3, %k2 ; AVX512F-32-NEXT: kshiftrq $44, %k3, %k0
; AVX512F-32-NEXT: kxorq %k0, %k2, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k0, %k1
; AVX512F-32-NEXT: kmovd %edx, %k2 ; AVX512F-32-NEXT: kmovd %edx, %k0
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $4, %cl ; AVX512F-32-NEXT: shrb $4, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $19, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $19, %k1, %k1
; AVX512F-32-NEXT: kxorq %k3, %k0, %k0 ; AVX512F-32-NEXT: kxorq %k3, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $45, %k0, %k3 ; AVX512F-32-NEXT: kshiftrq $45, %k1, %k3
; AVX512F-32-NEXT: kxorq %k5, %k3, %k4 ; AVX512F-32-NEXT: kxorq %k5, %k3, %k4
; AVX512F-32-NEXT: kmovd %ecx, %k3 ; AVX512F-32-NEXT: kmovd %ecx, %k3
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $5, %cl ; AVX512F-32-NEXT: shrb $5, %cl
; AVX512F-32-NEXT: andb $1, %cl ; AVX512F-32-NEXT: andb $1, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4 ; AVX512F-32-NEXT: kshiftlq $63, %k4, %k4
; AVX512F-32-NEXT: kshiftrq $18, %k4, %k4 ; AVX512F-32-NEXT: kshiftrq $18, %k4, %k4
; AVX512F-32-NEXT: kxorq %k0, %k4, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k4, %k1
; AVX512F-32-NEXT: kshiftrq $46, %k0, %k4 ; AVX512F-32-NEXT: kshiftrq $46, %k1, %k4
; AVX512F-32-NEXT: kxorq %k7, %k4, %k5 ; AVX512F-32-NEXT: kxorq %k7, %k4, %k5
; AVX512F-32-NEXT: kmovd %ecx, %k4 ; AVX512F-32-NEXT: kmovd %ecx, %k4
; AVX512F-32-NEXT: movl %eax, %ecx ; AVX512F-32-NEXT: movl %eax, %ecx
; AVX512F-32-NEXT: shrb $6, %cl ; AVX512F-32-NEXT: shrb $6, %cl
; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5 ; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $17, %k5, %k5 ; AVX512F-32-NEXT: kshiftrq $17, %k5, %k5
; AVX512F-32-NEXT: kxorq %k0, %k5, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k5, %k1
; AVX512F-32-NEXT: kshiftrq $47, %k0, %k5 ; AVX512F-32-NEXT: kshiftrq $47, %k1, %k5
; AVX512F-32-NEXT: kxorq %k6, %k5, %k5 ; AVX512F-32-NEXT: kxorq %k6, %k5, %k5
; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5 ; AVX512F-32-NEXT: kshiftlq $63, %k5, %k5
; AVX512F-32-NEXT: kshiftrq $16, %k5, %k5 ; AVX512F-32-NEXT: kshiftrq $16, %k5, %k5
; AVX512F-32-NEXT: kxorq %k0, %k5, %k0 ; AVX512F-32-NEXT: kxorq %k1, %k5, %k1
; AVX512F-32-NEXT: kshiftrq $48, %k0, %k5 ; AVX512F-32-NEXT: kshiftrq $48, %k1, %k5
; AVX512F-32-NEXT: kmovd %eax, %k6 ; AVX512F-32-NEXT: kmovd %eax, %k6
; AVX512F-32-NEXT: kxorq %k6, %k5, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k5, %k6
; AVX512F-32-NEXT: kmovd %ecx, %k5 ; AVX512F-32-NEXT: kmovd %ecx, %k5
; AVX512F-32-NEXT: movl %ebx, %edx ; AVX512F-32-NEXT: movl %ebx, %edx
; AVX512F-32-NEXT: shrl $24, %edx ; AVX512F-32-NEXT: shrl $24, %edx
; AVX512F-32-NEXT: # kill: def %al killed %al killed %eax def %eax ; AVX512F-32-NEXT: # kill: def %al killed %al killed %eax def %eax
; AVX512F-32-NEXT: shrb $7, %al ; AVX512F-32-NEXT: shrb $7, %al
; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6 ; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $15, %k6, %k6 ; AVX512F-32-NEXT: kshiftrq $15, %k6, %k6
; AVX512F-32-NEXT: kxorq %k0, %k6, %k6 ; AVX512F-32-NEXT: kxorq %k1, %k6, %k1
; AVX512F-32-NEXT: kshiftrq $49, %k6, %k0 ; AVX512F-32-NEXT: kshiftrq $49, %k1, %k6
; AVX512F-32-NEXT: kmovq {{[0-9]+}}(%esp), %k7 # 8-byte Reload ; AVX512F-32-NEXT: kxorq %k2, %k6, %k6
; AVX512F-32-NEXT: kxorq %k7, %k0, %k7 ; AVX512F-32-NEXT: kmovd %eax, %k2
; AVX512F-32-NEXT: kmovd %eax, %k0
; AVX512F-32-NEXT: movl %edx, %eax ; AVX512F-32-NEXT: movl %edx, %eax
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $14, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $14, %k6, %k6
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k1, %k6, %k6
; AVX512F-32-NEXT: kshiftrq $50, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $50, %k6, %k1
; AVX512F-32-NEXT: kxorq %k1, %k7, %k7 ; AVX512F-32-NEXT: kmovq {{[0-9]+}}(%esp), %k7 # 8-byte Reload
; AVX512F-32-NEXT: kxorq %k7, %k1, %k7
; AVX512F-32-NEXT: kmovd %edx, %k1 ; AVX512F-32-NEXT: kmovd %edx, %k1
; AVX512F-32-NEXT: # kill: def %dl killed %dl killed %edx def %edx ; AVX512F-32-NEXT: # kill: def %dl killed %dl killed %edx def %edx
; AVX512F-32-NEXT: andb $15, %dl ; AVX512F-32-NEXT: andb $15, %dl
; AVX512F-32-NEXT: andb $2, %al ; AVX512F-32-NEXT: andb $2, %al
; AVX512F-32-NEXT: shrb %al ; AVX512F-32-NEXT: shrb %al
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $13, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $13, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $51, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $51, %k6, %k7
; AVX512F-32-NEXT: kxorq %k2, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k0, %k7, %k7
; AVX512F-32-NEXT: kmovd %eax, %k2 ; AVX512F-32-NEXT: kmovd %eax, %k0
; AVX512F-32-NEXT: movl %edx, %eax ; AVX512F-32-NEXT: movl %edx, %eax
; AVX512F-32-NEXT: shrb $2, %dl ; AVX512F-32-NEXT: shrb $2, %dl
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $12, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $12, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $52, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $52, %k6, %k7
; AVX512F-32-NEXT: kxorq %k3, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k3, %k7, %k7
; AVX512F-32-NEXT: kmovd %edx, %k3 ; AVX512F-32-NEXT: kmovd %edx, %k3
Show All 12 Lines
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $54, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $54, %k6, %k7
; AVX512F-32-NEXT: kxorq %k5, %k7, %k7 ; AVX512F-32-NEXT: kxorq %k5, %k7, %k7
; AVX512F-32-NEXT: kmovd %eax, %k5 ; AVX512F-32-NEXT: kmovd %eax, %k5
; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7 ; AVX512F-32-NEXT: kshiftlq $63, %k7, %k7
; AVX512F-32-NEXT: kshiftrq $9, %k7, %k7 ; AVX512F-32-NEXT: kshiftrq $9, %k7, %k7
; AVX512F-32-NEXT: kxorq %k6, %k7, %k6 ; AVX512F-32-NEXT: kxorq %k6, %k7, %k6
; AVX512F-32-NEXT: kshiftrq $55, %k6, %k7 ; AVX512F-32-NEXT: kshiftrq $55, %k6, %k7
; AVX512F-32-NEXT: kxorq %k0, %k7, %k0 ; AVX512F-32-NEXT: kxorq %k2, %k7, %k2
; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $8, %k0, %k0 ; AVX512F-32-NEXT: kshiftrq $8, %k2, %k2
; AVX512F-32-NEXT: kxorq %k6, %k0, %k0 ; AVX512F-32-NEXT: kxorq %k6, %k2, %k2
; AVX512F-32-NEXT: kshiftrq $56, %k0, %k6 ; AVX512F-32-NEXT: kshiftrq $56, %k2, %k6
; AVX512F-32-NEXT: kxorq %k1, %k6, %k1 ; AVX512F-32-NEXT: kxorq %k1, %k6, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $7, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $7, %k1, %k1
; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $57, %k0, %k1
; AVX512F-32-NEXT: kxorq %k2, %k1, %k1 ; AVX512F-32-NEXT: kxorq %k2, %k1, %k1
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $57, %k1, %k2
; AVX512F-32-NEXT: kshiftrq $6, %k1, %k1 ; AVX512F-32-NEXT: kxorq %k0, %k2, %k0
; AVX512F-32-NEXT: kxorq %k0, %k1, %k0 ; AVX512F-32-NEXT: kshiftlq $63, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $6, %k0, %k0
; AVX512F-32-NEXT: kxorq %k1, %k0, %k0
; AVX512F-32-NEXT: kshiftrq $58, %k0, %k1 ; AVX512F-32-NEXT: kshiftrq $58, %k0, %k1
; AVX512F-32-NEXT: kxorq %k3, %k1, %k1 ; AVX512F-32-NEXT: kxorq %k3, %k1, %k1
; AVX512F-32-NEXT: movl %ebx, %eax ; AVX512F-32-NEXT: movl %ebx, %eax
; AVX512F-32-NEXT: shrl $28, %eax ; AVX512F-32-NEXT: shrl $28, %eax
; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1 ; AVX512F-32-NEXT: kshiftlq $63, %k1, %k1
; AVX512F-32-NEXT: kshiftrq $5, %k1, %k1 ; AVX512F-32-NEXT: kshiftrq $5, %k1, %k1
; AVX512F-32-NEXT: kxorq %k0, %k1, %k0 ; AVX512F-32-NEXT: kxorq %k0, %k1, %k0
; AVX512F-32-NEXT: kshiftrq $59, %k0, %k1 ; AVX512F-32-NEXT: kshiftrq $59, %k0, %k1
▲ Show 20 Lines • Show All 46 Lines • ▼ Show 20 Lines
; AVX512F-32-NEXT: vpcmpnltub %zmm1, %zmm0, %k0 {%k1} ; AVX512F-32-NEXT: vpcmpnltub %zmm1, %zmm0, %k0 {%k1}
; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp) ; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax ; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax
; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx ; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx
; AVX512F-32-NEXT: vpcmpnleub %zmm1, %zmm0, %k0 {%k1} ; AVX512F-32-NEXT: vpcmpnleub %zmm1, %zmm0, %k0 {%k1}
; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp) ; AVX512F-32-NEXT: kmovq %k0, {{[0-9]+}}(%esp)
; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax ; AVX512F-32-NEXT: addl {{[0-9]+}}(%esp), %eax
; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx ; AVX512F-32-NEXT: adcl {{[0-9]+}}(%esp), %edx
; AVX512F-32-NEXT: addl %esi, %eax ; AVX512F-32-NEXT: addl %ebp, %eax
; AVX512F-32-NEXT: adcl %ebx, %edx ; AVX512F-32-NEXT: adcl %ebx, %edx
; AVX512F-32-NEXT: addl $60, %esp ; AVX512F-32-NEXT: addl $60, %esp
; AVX512F-32-NEXT: popl %esi ; AVX512F-32-NEXT: popl %esi
; AVX512F-32-NEXT: popl %edi
; AVX512F-32-NEXT: popl %ebx ; AVX512F-32-NEXT: popl %ebx
; AVX512F-32-NEXT: popl %ebp
; AVX512F-32-NEXT: vzeroupper ; AVX512F-32-NEXT: vzeroupper
; AVX512F-32-NEXT: retl ; AVX512F-32-NEXT: retl
%res0 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 0, i64 %mask) %res0 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 0, i64 %mask)
%res1 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 1, i64 %mask) %res1 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 1, i64 %mask)
%ret1 = add i64 %res0, %res1 %ret1 = add i64 %res0, %res1
%res2 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 2, i64 %mask) %res2 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 2, i64 %mask)
%ret2 = add i64 %ret1, %res2 %ret2 = add i64 %ret1, %res2
%res3 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 3, i64 %mask) %res3 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i32 3, i64 %mask)
▲ Show 20 Lines • Show All 570 Lines • Show Last 20 Lines