|  | /* memchr (str, chr, len) -- Return pointer to first occurrence of CHR in STR | 
|  | less than LEN.  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.ai.mit.edu> | 
|  | Optimised a little by Alan Modra <Alan@SPRI.Levels.UniSA.Edu.Au> | 
|  | This version is developed using the same algorithm as the fast C | 
|  | version which carries the following introduction: | 
|  | Based on strlen implementation by Torbjorn Granlund (tege@sics.se), | 
|  | with help from Dan Sahlin (dan@sics.se) and | 
|  | commentary by Jim Blandy (jimb@ai.mit.edu); | 
|  | adaptation to memchr suggested by Dick Karpinski (dick@cca.ucsf.edu), | 
|  | and implemented by Roland McGrath (roland@ai.mit.edu). | 
|  |  | 
|  | 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+8		/* space for 2 saved regs */ | 
|  | #define RTN	PARMS | 
|  | #define STR	RTN | 
|  | #define CHR	STR+4 | 
|  | #define LEN	CHR+4 | 
|  |  | 
|  | .text | 
|  | ENTRY (__memchr) | 
|  |  | 
|  | /* Save callee-safe registers used in this function.  */ | 
|  | pushl %esi | 
|  | cfi_adjust_cfa_offset (4) | 
|  | pushl %edi | 
|  | cfi_adjust_cfa_offset (4) | 
|  | cfi_rel_offset (edi, 0) | 
|  |  | 
|  | /* Load parameters into registers.  */ | 
|  | movl STR(%esp), %eax	/* str: pointer to memory block.  */ | 
|  | movl CHR(%esp), %edx	/* c: byte we are looking for.  */ | 
|  | movl LEN(%esp), %esi	/* len: length of memory block.  */ | 
|  | cfi_rel_offset (esi, 4) | 
|  |  | 
|  | /* If my must not test more than three characters test | 
|  | them one by one.  This is especially true for 0.  */ | 
|  | cmpl $4, %esi | 
|  | jb L(3) | 
|  |  | 
|  | /* 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 c|c|0|0 */ | 
|  | movw %cx, %dx		/* And finally c|c|c|c */ | 
|  |  | 
|  | /* Better performance can be achieved if the word (32 | 
|  | bit) memory access is aligned on a four-byte-boundary. | 
|  | So process first bytes one by one until boundary is | 
|  | reached. Don't use a loop for better performance.  */ | 
|  |  | 
|  | testb $3, %al		/* correctly aligned ? */ | 
|  | je L(2)			/* yes => begin loop */ | 
|  | cmpb %dl, (%eax)	/* compare byte */ | 
|  | je L(9)			/* target found => return */ | 
|  | incl %eax		/* increment source pointer */ | 
|  | decl %esi		/* decrement length counter */ | 
|  | je L(4)			/* len==0 => return NULL */ | 
|  |  | 
|  | testb $3, %al		/* correctly aligned ? */ | 
|  | je L(2)			/* yes => begin loop */ | 
|  | cmpb %dl, (%eax)	/* compare byte */ | 
|  | je L(9)			/* target found => return */ | 
|  | incl %eax		/* increment source pointer */ | 
|  | decl %esi		/* decrement length counter */ | 
|  | je L(4)			/* len==0 => return NULL */ | 
|  |  | 
|  | testb $3, %al		/* correctly aligned ? */ | 
|  | je L(2)			/* yes => begin loop */ | 
|  | cmpb %dl, (%eax)	/* compare byte */ | 
|  | je L(9)			/* target found => return */ | 
|  | incl %eax		/* increment source pointer */ | 
|  | decl %esi		/* decrement length counter */ | 
|  | /* no test for len==0 here, because this is done in the | 
|  | loop head */ | 
|  | jmp L(2) | 
|  |  | 
|  | /* 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):	movl (%eax), %ecx	/* get word (= 4 bytes) in question */ | 
|  | movl $0xfefefeff, %edi	/* magic value */ | 
|  | xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c | 
|  | are now 0 */ | 
|  | addl %ecx, %edi		/* add the magic value to the word.  We get | 
|  | carry bits reported for each byte which | 
|  | is *not* 0 */ | 
|  |  | 
|  | /* 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(8) | 
|  |  | 
|  | /* 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(8)		/* found it => return pointer */ | 
|  |  | 
|  | /* This process is unfolded four times for better performance. | 
|  | we don't increment the source pointer each time.  Instead we | 
|  | use offsets and increment by 16 in each run of the loop.  But | 
|  | before probing for the matching byte we need some extra code | 
|  | (following LL(13) below).  Even the len can be compared with | 
|  | constants instead of decrementing each time.  */ | 
|  |  | 
|  | movl 4(%eax), %ecx	/* get word (= 4 bytes) in question */ | 
|  | movl $0xfefefeff, %edi	/* magic value */ | 
|  | xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c | 
|  | are now 0 */ | 
|  | 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 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(7)		/* found it => return pointer */ | 
|  |  | 
|  | movl 8(%eax), %ecx	/* get word (= 4 bytes) in question */ | 
|  | movl $0xfefefeff, %edi	/* magic value */ | 
|  | xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c | 
|  | are now 0 */ | 
|  | addl %ecx, %edi		/* add the magic value to the word.  We get | 
|  | carry bits reported for each byte which | 
|  | is *not* 0 */ | 
|  | jnc L(6)		/* 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(6)		/* found it => return pointer */ | 
|  |  | 
|  | movl 12(%eax), %ecx	/* get word (= 4 bytes) in question */ | 
|  | movl $0xfefefeff, %edi	/* magic value */ | 
|  | xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c | 
|  | are now 0 */ | 
|  | addl %ecx, %edi		/* add the magic value to the word.  We get | 
|  | carry bits reported for each byte which | 
|  | is *not* 0 */ | 
|  | jnc L(5)		/* 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(5)		/* found it => return pointer */ | 
|  |  | 
|  | /* Adjust both counters for a full round, i.e. 16 bytes.  */ | 
|  | addl $16, %eax | 
|  | L(2):	subl $16, %esi | 
|  | jae L(1)		/* Still more than 16 bytes remaining */ | 
|  |  | 
|  | /* Process remaining bytes separately.  */ | 
|  | cmpl $4-16, %esi	/* rest < 4 bytes? */ | 
|  | jb L(3)			/* yes, than test byte by byte */ | 
|  |  | 
|  | movl (%eax), %ecx	/* get word (= 4 bytes) in question */ | 
|  | movl $0xfefefeff, %edi	/* magic value */ | 
|  | xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c | 
|  | are now 0 */ | 
|  | addl %ecx, %edi		/* add the magic value to the word.  We get | 
|  | carry bits reported for each byte which | 
|  | is *not* 0 */ | 
|  | jnc L(8)		/* 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.  */ | 
|  | jne L(8)		/* found it => return pointer */ | 
|  | addl $4, %eax		/* adjust source pointer */ | 
|  |  | 
|  | cmpl $8-16, %esi	/* rest < 8 bytes? */ | 
|  | jb L(3)			/* yes, than test byte by byte */ | 
|  |  | 
|  | movl (%eax), %ecx	/* get word (= 4 bytes) in question */ | 
|  | movl $0xfefefeff, %edi	/* magic value */ | 
|  | xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c | 
|  | are now 0 */ | 
|  | addl %ecx, %edi		/* add the magic value to the word.  We get | 
|  | carry bits reported for each byte which | 
|  | is *not* 0 */ | 
|  | jnc L(8)		/* 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.  */ | 
|  | jne L(8)		/* found it => return pointer */ | 
|  | addl $4, %eax		/* adjust source pointer */ | 
|  |  | 
|  | cmpl $12-16, %esi	/* rest < 12 bytes? */ | 
|  | jb L(3)			/* yes, than test byte by byte */ | 
|  |  | 
|  | movl (%eax), %ecx	/* get word (= 4 bytes) in question */ | 
|  | movl $0xfefefeff, %edi	/* magic value */ | 
|  | xorl %edx, %ecx		/* XOR with word c|c|c|c => bytes of str == c | 
|  | are now 0 */ | 
|  | addl %ecx, %edi		/* add the magic value to the word.  We get | 
|  | carry bits reported for each byte which | 
|  | is *not* 0 */ | 
|  | jnc L(8)		/* 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.  */ | 
|  | jne L(8)		/* found it => return pointer */ | 
|  | addl $4, %eax		/* adjust source pointer */ | 
|  |  | 
|  | /* Check the remaining bytes one by one.  */ | 
|  | L(3):	andl $3, %esi		/* mask out uninteresting bytes */ | 
|  | jz L(4)			/* no remaining bytes => return NULL */ | 
|  |  | 
|  | cmpb %dl, (%eax)	/* compare byte with CHR */ | 
|  | je L(9)			/* equal, than return pointer */ | 
|  | incl %eax		/* increment source pointer */ | 
|  | decl %esi		/* decrement length */ | 
|  | jz L(4)			/* no remaining bytes => return NULL */ | 
|  |  | 
|  | cmpb %dl, (%eax)	/* compare byte with CHR */ | 
|  | je L(9)			/* equal, than return pointer */ | 
|  | incl %eax		/* increment source pointer */ | 
|  | decl %esi		/* decrement length */ | 
|  | jz L(4)			/* no remaining bytes => return NULL */ | 
|  |  | 
|  | cmpb %dl, (%eax)	/* compare byte with CHR */ | 
|  | je L(9)			/* equal, than return pointer */ | 
|  |  | 
|  | L(4):	/* no byte found => return NULL */ | 
|  | xorl %eax, %eax | 
|  | jmp L(9) | 
|  |  | 
|  | /* add missing source pointer increments */ | 
|  | L(5):	addl $4, %eax | 
|  | L(6):	addl $4, %eax | 
|  | L(7):	addl $4, %eax | 
|  |  | 
|  | /* Test for the matching byte in the word.  %ecx contains a NUL | 
|  | char in the byte which originally was the byte we are looking | 
|  | at.  */ | 
|  | L(8):	testb %cl, %cl		/* test first byte in dword */ | 
|  | jz L(9)			/* if zero => return pointer */ | 
|  | incl %eax		/* increment source pointer */ | 
|  |  | 
|  | testb %ch, %ch		/* test second byte in dword */ | 
|  | jz L(9)			/* if zero => return pointer */ | 
|  | incl %eax		/* increment source pointer */ | 
|  |  | 
|  | testl $0xff0000, %ecx	/* test third byte in dword */ | 
|  | jz L(9)			/* if zero => return pointer */ | 
|  | incl %eax		/* increment source pointer */ | 
|  |  | 
|  | /* No further test needed we we know it is one of the four bytes.  */ | 
|  | L(9):	popl %edi		/* pop saved registers */ | 
|  | cfi_adjust_cfa_offset (-4) | 
|  | cfi_restore (edi) | 
|  | popl %esi | 
|  | cfi_adjust_cfa_offset (-4) | 
|  | cfi_restore (esi) | 
|  |  | 
|  | ret | 
|  | END (__memchr) | 
|  |  | 
|  | weak_alias (__memchr, memchr) | 
|  | libc_hidden_builtin_def (memchr) |