mirror of
				https://sourceware.org/git/glibc.git
				synced 2025-10-30 10:45:40 +03:00 
			
		
		
		
	
		
			
				
	
	
		
			279 lines
		
	
	
		
			11 KiB
		
	
	
	
		
			ArmAsm
		
	
	
	
	
	
			
		
		
	
	
			279 lines
		
	
	
		
			11 KiB
		
	
	
	
		
			ArmAsm
		
	
	
	
	
	
| /* strchrnul (str, chr) -- Return pointer to first occurrence of CHR in STR
 | |
|    or the final NUL byte.
 | |
|    For Intel 80x86, x>=3.
 | |
|    Copyright (C) 1994-2016 Free Software Foundation, Inc.
 | |
|    This file is part of the GNU C Library.
 | |
|    Contributed by Ulrich Drepper <drepper@gnu.org>
 | |
|    Some optimisations by Alan Modra <Alan@SPRI.Levels.UniSA.Edu.Au>
 | |
| 
 | |
|    The GNU C Library is free software; you can redistribute it and/or
 | |
|    modify it under the terms of the GNU Lesser General Public
 | |
|    License as published by the Free Software Foundation; either
 | |
|    version 2.1 of the License, or (at your option) any later version.
 | |
| 
 | |
|    The GNU C Library is distributed in the hope that it will be useful,
 | |
|    but WITHOUT ANY WARRANTY; without even the implied warranty of
 | |
|    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 | |
|    Lesser General Public License for more details.
 | |
| 
 | |
|    You should have received a copy of the GNU Lesser General Public
 | |
|    License along with the GNU C Library; if not, see
 | |
|    <http://www.gnu.org/licenses/>.  */
 | |
| 
 | |
| #include <sysdep.h>
 | |
| #include "asm-syntax.h"
 | |
| 
 | |
| #define PARMS	4+4	/* space for 1 saved reg */
 | |
| #define RTN	PARMS
 | |
| #define STR	RTN
 | |
| #define CHR	STR+4
 | |
| 
 | |
| 	.text
 | |
| ENTRY (__strchrnul)
 | |
| 
 | |
| 	pushl %edi		/* Save callee-safe registers used here.  */
 | |
| 	cfi_adjust_cfa_offset (4)
 | |
| 	cfi_rel_offset (edi, 0)
 | |
| 
 | |
| 	movl STR(%esp), %eax
 | |
| 	movl CHR(%esp), %edx
 | |
| 
 | |
| 	/* At the moment %edx contains CHR.  What we need for the
 | |
| 	   algorithm is CHR in all bytes of the dword.  Avoid
 | |
| 	   operations on 16 bit words because these require an
 | |
| 	   prefix byte (and one more cycle).  */
 | |
| 	movb %dl, %dh		/* now it is 0|0|c|c */
 | |
| 	movl %edx, %ecx
 | |
| 	shll $16, %edx		/* now it is c|c|0|0 */
 | |
| 	movw %cx, %dx		/* and finally c|c|c|c */
 | |
| 
 | |
| 	/* Before we start with the main loop we process single bytes
 | |
| 	   until the source pointer is aligned.  This has two reasons:
 | |
| 	   1. aligned 32-bit memory access is faster
 | |
| 	   and (more important)
 | |
| 	   2. we process in the main loop 32 bit in one step although
 | |
| 	      we don't know the end of the string.  But accessing at
 | |
| 	      4-byte alignment guarantees that we never access illegal
 | |
| 	      memory if this would not also be done by the trivial
 | |
| 	      implementation (this is because all processor inherent
 | |
| 	      boundaries are multiples of 4.  */
 | |
| 
 | |
| 	testb $3, %al		/* correctly aligned ? */
 | |
| 	jz L(11)		/* yes => begin loop */
 | |
| 	movb (%eax), %cl	/* load byte in question (we need it twice) */
 | |
| 	cmpb %cl, %dl		/* compare byte */
 | |
| 	je L(6)			/* target found => return */
 | |
| 	testb %cl, %cl		/* is NUL? */
 | |
| 	jz L(6)			/* yes => return NULL */
 | |
| 	incl %eax		/* increment pointer */
 | |
| 
 | |
| 	testb $3, %al		/* correctly aligned ? */
 | |
| 	jz L(11)		/* yes => begin loop */
 | |
| 	movb (%eax), %cl	/* load byte in question (we need it twice) */
 | |
| 	cmpb %cl, %dl		/* compare byte */
 | |
| 	je L(6)			/* target found => return */
 | |
| 	testb %cl, %cl		/* is NUL? */
 | |
| 	jz L(6)			/* yes => return NULL */
 | |
| 	incl %eax		/* increment pointer */
 | |
| 
 | |
| 	testb $3, %al		/* correctly aligned ? */
 | |
| 	jz L(11)		/* yes => begin loop */
 | |
| 	movb (%eax), %cl	/* load byte in question (we need it twice) */
 | |
| 	cmpb %cl, %dl		/* compare byte */
 | |
| 	je L(6)			/* target found => return */
 | |
| 	testb %cl, %cl		/* is NUL? */
 | |
| 	jz L(6)			/* yes => return NULL */
 | |
| 	incl %eax		/* increment pointer */
 | |
| 
 | |
| 	/* No we have reached alignment.  */
 | |
| 	jmp L(11)		/* begin loop */
 | |
| 
 | |
|       /* We exit the loop if adding MAGIC_BITS to LONGWORD fails to
 | |
| 	 change any of the hole bits of LONGWORD.
 | |
| 
 | |
| 	 1) Is this safe?  Will it catch all the zero bytes?
 | |
| 	 Suppose there is a byte with all zeros.  Any carry bits
 | |
| 	 propagating from its left will fall into the hole at its
 | |
| 	 least significant bit and stop.  Since there will be no
 | |
| 	 carry from its most significant bit, the LSB of the
 | |
| 	 byte to the left will be unchanged, and the zero will be
 | |
| 	 detected.
 | |
| 
 | |
| 	 2) Is this worthwhile?  Will it ignore everything except
 | |
| 	 zero bytes?  Suppose every byte of LONGWORD has a bit set
 | |
| 	 somewhere.  There will be a carry into bit 8.	If bit 8
 | |
| 	 is set, this will carry into bit 16.  If bit 8 is clear,
 | |
| 	 one of bits 9-15 must be set, so there will be a carry
 | |
| 	 into bit 16.  Similarly, there will be a carry into bit
 | |
| 	 24.  If one of bits 24-31 is set, there will be a carry
 | |
| 	 into bit 32 (=carry flag), so all of the hole bits will
 | |
| 	 be changed.
 | |
| 
 | |
| 	 3) But wait!  Aren't we looking for CHR, not zero?
 | |
| 	 Good point.  So what we do is XOR LONGWORD with a longword,
 | |
| 	 each of whose bytes is CHR.  This turns each byte that is CHR
 | |
| 	 into a zero.  */
 | |
| 
 | |
| 	/* Each round the main loop processes 16 bytes.  */
 | |
| 
 | |
| 	ALIGN(4)
 | |
| 
 | |
| L(1):	addl $16, %eax		/* adjust pointer for whole round */
 | |
| 
 | |
| L(11):	movl (%eax), %ecx	/* get word (= 4 bytes) in question */
 | |
| 	xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c
 | |
| 				   are now 0 */
 | |
| 	movl $0xfefefeff, %edi	/* magic value */
 | |
| 	addl %ecx, %edi		/* add the magic value to the word.  We get
 | |
| 				   carry bits reported for each byte which
 | |
| 				   is *not* CHR */
 | |
| 
 | |
| 	/* According to the algorithm we had to reverse the effect of the
 | |
| 	   XOR first and then test the overflow bits.  But because the
 | |
| 	   following XOR would destroy the carry flag and it would (in a
 | |
| 	   representation with more than 32 bits) not alter then last
 | |
| 	   overflow, we can now test this condition.  If no carry is signaled
 | |
| 	   no overflow must have occurred in the last byte => it was 0.	*/
 | |
| 	jnc L(7)
 | |
| 
 | |
| 	/* We are only interested in carry bits that change due to the
 | |
| 	   previous add, so remove original bits */
 | |
| 	xorl %ecx, %edi		/* ((word^charmask)+magic)^(word^charmask) */
 | |
| 
 | |
| 	/* Now test for the other three overflow bits.  */
 | |
| 	orl $0xfefefeff, %edi	/* set all non-carry bits */
 | |
| 	incl %edi		/* add 1: if one carry bit was *not* set
 | |
| 				   the addition will not result in 0.  */
 | |
| 
 | |
| 	/* If at least one byte of the word is CHR we don't get 0 in %edi.  */
 | |
| 	jnz L(7)		/* found it => return pointer */
 | |
| 
 | |
| 	/* Now we made sure the dword does not contain the character we are
 | |
| 	   looking for.  But because we deal with strings we have to check
 | |
| 	   for the end of string before testing the next dword.  */
 | |
| 
 | |
| 	xorl %edx, %ecx		/* restore original dword without reload */
 | |
| 	movl $0xfefefeff, %edi	/* magic value */
 | |
| 	addl %ecx, %edi		/* add the magic value to the word.  We get
 | |
| 				   carry bits reported for each byte which
 | |
| 				   is *not* 0 */
 | |
| 	jnc L(7)		/* highest byte is NUL => return NULL */
 | |
| 	xorl %ecx, %edi		/* (word+magic)^word */
 | |
| 	orl $0xfefefeff, %edi	/* set all non-carry bits */
 | |
| 	incl %edi		/* add 1: if one carry bit was *not* set
 | |
| 				   the addition will not result in 0.  */
 | |
| 	jnz L(7)		/* found NUL => return NULL */
 | |
| 
 | |
| 	movl 4(%eax), %ecx	/* get word (= 4 bytes) in question */
 | |
| 	xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c
 | |
| 				   are now 0 */
 | |
| 	movl $0xfefefeff, %edi	/* magic value */
 | |
| 	addl %ecx, %edi		/* add the magic value to the word.  We get
 | |
| 				   carry bits reported for each byte which
 | |
| 				   is *not* CHR */
 | |
| 	jnc L(71)		/* highest byte is CHR => return pointer */
 | |
| 	xorl %ecx, %edi		/* ((word^charmask)+magic)^(word^charmask) */
 | |
| 	orl $0xfefefeff, %edi	/* set all non-carry bits */
 | |
| 	incl %edi		/* add 1: if one carry bit was *not* set
 | |
| 				   the addition will not result in 0.  */
 | |
| 	jnz L(71)		/* found it => return pointer */
 | |
| 	xorl %edx, %ecx		/* restore original dword without reload */
 | |
| 	movl $0xfefefeff, %edi	/* magic value */
 | |
| 	addl %ecx, %edi		/* add the magic value to the word.  We get
 | |
| 				   carry bits reported for each byte which
 | |
| 				   is *not* 0 */
 | |
| 	jnc L(71)		/* highest byte is NUL => return NULL */
 | |
| 	xorl %ecx, %edi		/* (word+magic)^word */
 | |
| 	orl $0xfefefeff, %edi	/* set all non-carry bits */
 | |
| 	incl %edi		/* add 1: if one carry bit was *not* set
 | |
| 				   the addition will not result in 0.  */
 | |
| 	jnz L(71)		/* found NUL => return NULL */
 | |
| 
 | |
| 	movl 8(%eax), %ecx	/* get word (= 4 bytes) in question */
 | |
| 	xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c
 | |
| 				   are now 0 */
 | |
| 	movl $0xfefefeff, %edi	/* magic value */
 | |
| 	addl %ecx, %edi		/* add the magic value to the word.  We get
 | |
| 				   carry bits reported for each byte which
 | |
| 				   is *not* CHR */
 | |
| 	jnc L(72)		/* highest byte is CHR => return pointer */
 | |
| 	xorl %ecx, %edi		/* ((word^charmask)+magic)^(word^charmask) */
 | |
| 	orl $0xfefefeff, %edi	/* set all non-carry bits */
 | |
| 	incl %edi		/* add 1: if one carry bit was *not* set
 | |
| 				   the addition will not result in 0.  */
 | |
| 	jnz L(72)		/* found it => return pointer */
 | |
| 	xorl %edx, %ecx		/* restore original dword without reload */
 | |
| 	movl $0xfefefeff, %edi	/* magic value */
 | |
| 	addl %ecx, %edi		/* add the magic value to the word.  We get
 | |
| 				   carry bits reported for each byte which
 | |
| 				   is *not* 0 */
 | |
| 	jnc L(72)		/* highest byte is NUL => return NULL */
 | |
| 	xorl %ecx, %edi		/* (word+magic)^word */
 | |
| 	orl $0xfefefeff, %edi	/* set all non-carry bits */
 | |
| 	incl %edi		/* add 1: if one carry bit was *not* set
 | |
| 				   the addition will not result in 0.  */
 | |
| 	jnz L(72)		/* found NUL => return NULL */
 | |
| 
 | |
| 	movl 12(%eax), %ecx	/* get word (= 4 bytes) in question */
 | |
| 	xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c
 | |
| 				   are now 0 */
 | |
| 	movl $0xfefefeff, %edi	/* magic value */
 | |
| 	addl %ecx, %edi		/* add the magic value to the word.  We get
 | |
| 				   carry bits reported for each byte which
 | |
| 				   is *not* CHR */
 | |
| 	jnc L(73)		/* highest byte is CHR => return pointer */
 | |
| 	xorl %ecx, %edi		/* ((word^charmask)+magic)^(word^charmask) */
 | |
| 	orl $0xfefefeff, %edi	/* set all non-carry bits */
 | |
| 	incl %edi		/* add 1: if one carry bit was *not* set
 | |
| 				   the addition will not result in 0.  */
 | |
| 	jnz L(73)		/* found it => return pointer */
 | |
| 	xorl %edx, %ecx		/* restore original dword without reload */
 | |
| 	movl $0xfefefeff, %edi	/* magic value */
 | |
| 	addl %ecx, %edi		/* add the magic value to the word.  We get
 | |
| 				   carry bits reported for each byte which
 | |
| 				   is *not* 0 */
 | |
| 	jnc L(73)		/* highest byte is NUL => return NULL */
 | |
| 	xorl %ecx, %edi		/* (word+magic)^word */
 | |
| 	orl $0xfefefeff, %edi	/* set all non-carry bits */
 | |
| 	incl %edi		/* add 1: if one carry bit was *not* set
 | |
| 				   the addition will not result in 0.  */
 | |
| 	jz L(1)			/* no NUL found => restart loop */
 | |
| 
 | |
| L(73):	addl $4, %eax		/* adjust pointer */
 | |
| L(72):	addl $4, %eax
 | |
| L(71):	addl $4, %eax
 | |
| 
 | |
| 	/* We now scan for the byte in which the character was matched.
 | |
| 	   But we have to take care of the case that a NUL char is
 | |
| 	   found before this in the dword.  */
 | |
| 
 | |
| L(7):	testb %cl, %cl		/* is first byte CHR? */
 | |
| 	jz L(6)			/* yes => return pointer */
 | |
| 	cmpb %dl, %cl		/* is first byte NUL? */
 | |
| 	je L(6)			/* yes => return NULL */
 | |
| 	incl %eax		/* it's not in the first byte */
 | |
| 
 | |
| 	testb %ch, %ch		/* is second byte CHR? */
 | |
| 	jz L(6)			/* yes => return pointer */
 | |
| 	cmpb %dl, %ch		/* is second byte NUL? */
 | |
| 	je L(6)			/* yes => return NULL? */
 | |
| 	incl %eax		/* it's not in the second byte */
 | |
| 
 | |
| 	shrl $16, %ecx		/* make upper byte accessible */
 | |
| 	testb %cl, %cl		/* is third byte CHR? */
 | |
| 	jz L(6)			/* yes => return pointer */
 | |
| 	cmpb %dl, %cl		/* is third byte NUL? */
 | |
| 	je L(6)			/* yes => return NULL */
 | |
| 
 | |
| 	/* It must be in the fourth byte and it cannot be NUL.  */
 | |
| 	incl %eax
 | |
| 
 | |
| L(6):	popl %edi		/* restore saved register content */
 | |
| 	cfi_adjust_cfa_offset (-4)
 | |
| 	cfi_restore (edi)
 | |
| 
 | |
| 	ret
 | |
| END (__strchrnul)
 | |
| 
 | |
| weak_alias (__strchrnul, strchrnul)
 |