Blame sysdeps/i386/i686/multiarch/strchr-sse2.S

Packit 6c4009
/* strchr SSE2 without bsf
Packit 6c4009
   Copyright (C) 2011-2018 Free Software Foundation, Inc.
Packit 6c4009
   Contributed by Intel Corporation.
Packit 6c4009
   This file is part of the GNU C Library.
Packit 6c4009
Packit 6c4009
   The GNU C Library is free software; you can redistribute it and/or
Packit 6c4009
   modify it under the terms of the GNU Lesser General Public
Packit 6c4009
   License as published by the Free Software Foundation; either
Packit 6c4009
   version 2.1 of the License, or (at your option) any later version.
Packit 6c4009
Packit 6c4009
   The GNU C Library is distributed in the hope that it will be useful,
Packit 6c4009
   but WITHOUT ANY WARRANTY; without even the implied warranty of
Packit 6c4009
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
Packit 6c4009
   Lesser General Public License for more details.
Packit 6c4009
Packit 6c4009
   You should have received a copy of the GNU Lesser General Public
Packit 6c4009
   License along with the GNU C Library; if not, see
Packit 6c4009
   <http://www.gnu.org/licenses/>.  */
Packit 6c4009
Packit 6c4009
#if IS_IN (libc)
Packit 6c4009
Packit 6c4009
# include <sysdep.h>
Packit 6c4009
Packit 6c4009
# define CFI_PUSH(REG)	\
Packit 6c4009
	cfi_adjust_cfa_offset (4);	\
Packit 6c4009
	cfi_rel_offset (REG, 0)
Packit 6c4009
Packit 6c4009
# define CFI_POP(REG)	\
Packit 6c4009
	cfi_adjust_cfa_offset (-4);	\
Packit 6c4009
	cfi_restore (REG)
Packit 6c4009
Packit 6c4009
# define PUSH(REG) pushl REG; CFI_PUSH (REG)
Packit 6c4009
# define POP(REG) popl REG; CFI_POP (REG)
Packit 6c4009
Packit 6c4009
# define PARMS  8
Packit 6c4009
# define ENTRANCE PUSH(%edi)
Packit 6c4009
# define RETURN  POP(%edi); ret; CFI_PUSH(%edi);
Packit 6c4009
Packit 6c4009
# define STR1  PARMS
Packit 6c4009
# define STR2  STR1+4
Packit 6c4009
Packit 6c4009
	atom_text_section
Packit 6c4009
ENTRY (__strchr_sse2)
Packit 6c4009
Packit 6c4009
	ENTRANCE
Packit 6c4009
	mov	STR1(%esp), %ecx
Packit 6c4009
	movd	STR2(%esp), %xmm1
Packit 6c4009
Packit 6c4009
	pxor	%xmm2, %xmm2
Packit 6c4009
	mov	%ecx, %edi
Packit 6c4009
	punpcklbw %xmm1, %xmm1
Packit 6c4009
	punpcklbw %xmm1, %xmm1
Packit 6c4009
	/* ECX has OFFSET. */
Packit 6c4009
	and	$15, %ecx
Packit 6c4009
	pshufd	$0, %xmm1, %xmm1
Packit 6c4009
	je	L(loop)
Packit 6c4009
Packit 6c4009
/* Handle unaligned string.  */
Packit 6c4009
	and	$-16, %edi
Packit 6c4009
	movdqa	(%edi), %xmm0
Packit 6c4009
	pcmpeqb	%xmm0, %xmm2
Packit 6c4009
	pcmpeqb	%xmm1, %xmm0
Packit 6c4009
	/* Find where NULL is.  */
Packit 6c4009
	pmovmskb %xmm2, %edx
Packit 6c4009
	/* Check if there is a match.  */
Packit 6c4009
	pmovmskb %xmm0, %eax
Packit 6c4009
	/* Remove the leading bytes.  */
Packit 6c4009
	sarl	%cl, %edx
Packit 6c4009
	sarl	%cl, %eax
Packit 6c4009
	test	%eax, %eax
Packit 6c4009
	jz	L(unaligned_no_match)
Packit 6c4009
	/* Check which byte is a match.  */
Packit 6c4009
	/* Is there a NULL? */
Packit 6c4009
	add	%ecx, %edi
Packit 6c4009
	test	%edx, %edx
Packit 6c4009
	jz	L(match_case1)
Packit 6c4009
	jmp	L(match_case2)
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(unaligned_no_match):
Packit 6c4009
	test	%edx, %edx
Packit 6c4009
	jne	L(return_null)
Packit 6c4009
Packit 6c4009
	pxor	%xmm2, %xmm2
Packit 6c4009
	add	$16, %edi
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
/* Loop start on aligned string.  */
Packit 6c4009
L(loop):
Packit 6c4009
	movdqa	(%edi), %xmm0
Packit 6c4009
	pcmpeqb	%xmm0, %xmm2
Packit 6c4009
	pcmpeqb	%xmm1, %xmm0
Packit 6c4009
	pmovmskb %xmm2, %edx
Packit 6c4009
	pmovmskb %xmm0, %eax
Packit 6c4009
	test	%eax, %eax
Packit 6c4009
	jnz	L(matches)
Packit 6c4009
	test	%edx, %edx
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	add	$16, %edi
Packit 6c4009
Packit 6c4009
	movdqa	(%edi), %xmm0
Packit 6c4009
	pcmpeqb	%xmm0, %xmm2
Packit 6c4009
	pcmpeqb	%xmm1, %xmm0
Packit 6c4009
	pmovmskb %xmm2, %edx
Packit 6c4009
	pmovmskb %xmm0, %eax
Packit 6c4009
	test	%eax, %eax
Packit 6c4009
	jnz	L(matches)
Packit 6c4009
	test	%edx, %edx
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	add	$16, %edi
Packit 6c4009
Packit 6c4009
	movdqa	(%edi), %xmm0
Packit 6c4009
	pcmpeqb	%xmm0, %xmm2
Packit 6c4009
	pcmpeqb	%xmm1, %xmm0
Packit 6c4009
	pmovmskb %xmm2, %edx
Packit 6c4009
	pmovmskb %xmm0, %eax
Packit 6c4009
	test	%eax, %eax
Packit 6c4009
	jnz	L(matches)
Packit 6c4009
	test	%edx, %edx
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	add	$16, %edi
Packit 6c4009
Packit 6c4009
	movdqa	(%edi), %xmm0
Packit 6c4009
	pcmpeqb	%xmm0, %xmm2
Packit 6c4009
	pcmpeqb	%xmm1, %xmm0
Packit 6c4009
	pmovmskb %xmm2, %edx
Packit 6c4009
	pmovmskb %xmm0, %eax
Packit 6c4009
	test	%eax, %eax
Packit 6c4009
	jnz	L(matches)
Packit 6c4009
	test	%edx, %edx
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	add	$16, %edi
Packit 6c4009
	jmp	L(loop)
Packit 6c4009
Packit 6c4009
L(matches):
Packit 6c4009
	/* There is a match.  First find where NULL is.  */
Packit 6c4009
	test	%edx, %edx
Packit 6c4009
	jz	L(match_case1)
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(match_case2):
Packit 6c4009
	test	%al, %al
Packit 6c4009
	jz	L(match_higth_case2)
Packit 6c4009
Packit 6c4009
	mov	%al, %cl
Packit 6c4009
	and	$15, %cl
Packit 6c4009
	jnz	L(match_case2_4)
Packit 6c4009
Packit 6c4009
	mov	%dl, %ch
Packit 6c4009
	and	$15, %ch
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
Packit 6c4009
	test	$0x10, %al
Packit 6c4009
	jnz	L(Exit5)
Packit 6c4009
	test	$0x10, %dl
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	test	$0x20, %al
Packit 6c4009
	jnz	L(Exit6)
Packit 6c4009
	test	$0x20, %dl
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	test	$0x40, %al
Packit 6c4009
	jnz	L(Exit7)
Packit 6c4009
	test	$0x40, %dl
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	lea	7(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(match_case2_4):
Packit 6c4009
	test	$0x01, %al
Packit 6c4009
	jnz	L(Exit1)
Packit 6c4009
	test	$0x01, %dl
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	test	$0x02, %al
Packit 6c4009
	jnz	L(Exit2)
Packit 6c4009
	test	$0x02, %dl
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	test	$0x04, %al
Packit 6c4009
	jnz	L(Exit3)
Packit 6c4009
	test	$0x04, %dl
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	lea	3(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(match_higth_case2):
Packit 6c4009
	test	%dl, %dl
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
Packit 6c4009
	mov	%ah, %cl
Packit 6c4009
	and	$15, %cl
Packit 6c4009
	jnz	L(match_case2_12)
Packit 6c4009
Packit 6c4009
	mov	%dh, %ch
Packit 6c4009
	and	$15, %ch
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
Packit 6c4009
	test	$0x10, %ah
Packit 6c4009
	jnz	L(Exit13)
Packit 6c4009
	test	$0x10, %dh
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	test	$0x20, %ah
Packit 6c4009
	jnz	L(Exit14)
Packit 6c4009
	test	$0x20, %dh
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	test	$0x40, %ah
Packit 6c4009
	jnz	L(Exit15)
Packit 6c4009
	test	$0x40, %dh
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	lea	15(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(match_case2_12):
Packit 6c4009
	test	$0x01, %ah
Packit 6c4009
	jnz	L(Exit9)
Packit 6c4009
	test	$0x01, %dh
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	test	$0x02, %ah
Packit 6c4009
	jnz	L(Exit10)
Packit 6c4009
	test	$0x02, %dh
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	test	$0x04, %ah
Packit 6c4009
	jnz	L(Exit11)
Packit 6c4009
	test	$0x04, %dh
Packit 6c4009
	jnz	L(return_null)
Packit 6c4009
	lea	11(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(match_case1):
Packit 6c4009
	test	%al, %al
Packit 6c4009
	jz	L(match_higth_case1)
Packit 6c4009
Packit 6c4009
	test	$0x01, %al
Packit 6c4009
	jnz	L(Exit1)
Packit 6c4009
	test	$0x02, %al
Packit 6c4009
	jnz	L(Exit2)
Packit 6c4009
	test	$0x04, %al
Packit 6c4009
	jnz	L(Exit3)
Packit 6c4009
	test	$0x08, %al
Packit 6c4009
	jnz	L(Exit4)
Packit 6c4009
	test	$0x10, %al
Packit 6c4009
	jnz	L(Exit5)
Packit 6c4009
	test	$0x20, %al
Packit 6c4009
	jnz	L(Exit6)
Packit 6c4009
	test	$0x40, %al
Packit 6c4009
	jnz	L(Exit7)
Packit 6c4009
	lea	7(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(match_higth_case1):
Packit 6c4009
	test	$0x01, %ah
Packit 6c4009
	jnz	L(Exit9)
Packit 6c4009
	test	$0x02, %ah
Packit 6c4009
	jnz	L(Exit10)
Packit 6c4009
	test	$0x04, %ah
Packit 6c4009
	jnz	L(Exit11)
Packit 6c4009
	test	$0x08, %ah
Packit 6c4009
	jnz	L(Exit12)
Packit 6c4009
	test	$0x10, %ah
Packit 6c4009
	jnz	L(Exit13)
Packit 6c4009
	test	$0x20, %ah
Packit 6c4009
	jnz	L(Exit14)
Packit 6c4009
	test	$0x40, %ah
Packit 6c4009
	jnz	L(Exit15)
Packit 6c4009
	lea	15(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit1):
Packit 6c4009
	lea	(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit2):
Packit 6c4009
	lea	1(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit3):
Packit 6c4009
	lea	2(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit4):
Packit 6c4009
	lea	3(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit5):
Packit 6c4009
	lea	4(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit6):
Packit 6c4009
	lea	5(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit7):
Packit 6c4009
	lea	6(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit9):
Packit 6c4009
	lea	8(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit10):
Packit 6c4009
	lea	9(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit11):
Packit 6c4009
	lea	10(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit12):
Packit 6c4009
	lea	11(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit13):
Packit 6c4009
	lea	12(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit14):
Packit 6c4009
	lea	13(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
	.p2align 4
Packit 6c4009
L(Exit15):
Packit 6c4009
	lea	14(%edi), %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
/* Return NULL.  */
Packit 6c4009
	.p2align 4
Packit 6c4009
L(return_null):
Packit 6c4009
	xor	%eax, %eax
Packit 6c4009
	RETURN
Packit 6c4009
Packit 6c4009
END (__strchr_sse2)
Packit 6c4009
#endif