129 lines
4.5 KiB
ArmAsm
129 lines
4.5 KiB
ArmAsm
/*******************************************************************************
|
|
*
|
|
* Copyright (c) 1993 Intel Corporation
|
|
*
|
|
* Intel hereby grants you permission to copy, modify, and distribute this
|
|
* software and its documentation. Intel grants this permission provided
|
|
* that the above copyright notice appears in all copies and that both the
|
|
* copyright notice and this permission notice appear in supporting
|
|
* documentation. In addition, Intel grants this permission provided that
|
|
* you prominently mark as "not part of the original" any modifications
|
|
* made to this software or documentation, and that the name of Intel
|
|
* Corporation not be used in advertising or publicity pertaining to
|
|
* distribution of the software or the documentation without specific,
|
|
* written prior permission.
|
|
*
|
|
* Intel Corporation provides this AS IS, WITHOUT ANY WARRANTY, EXPRESS OR
|
|
* IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTY OF MERCHANTABILITY
|
|
* OR FITNESS FOR A PARTICULAR PURPOSE. Intel makes no guarantee or
|
|
* representations regarding the use of, or the results of the use of,
|
|
* the software and documentation in terms of correctness, accuracy,
|
|
* reliability, currentness, or otherwise; and you rely on the software,
|
|
* documentation and results solely at your own risk.
|
|
*
|
|
* IN NO EVENT SHALL INTEL BE LIABLE FOR ANY LOSS OF USE, LOSS OF BUSINESS,
|
|
* LOSS OF PROFITS, INDIRECT, INCIDENTAL, SPECIAL OR CONSEQUENTIAL DAMAGES
|
|
* OF ANY KIND. IN NO EVENT SHALL INTEL'S TOTAL LIABILITY EXCEED THE SUM
|
|
* PAID TO INTEL FOR THE PRODUCT LICENSED HEREUNDER.
|
|
*
|
|
******************************************************************************/
|
|
|
|
.file "memcmp.s"
|
|
#ifdef __PIC
|
|
.pic
|
|
#endif
|
|
#ifdef __PID
|
|
.pid
|
|
#endif
|
|
/*
|
|
* (c) copyright 1988,1993 Intel Corp., all rights reserved
|
|
*/
|
|
/*
|
|
procedure memcmp (optimized assembler version for the 80960K series)
|
|
|
|
result = memcmp (src1_addr, src2_addr, max_bytes)
|
|
|
|
compare the byte array pointed to by src1_addr to the byte array
|
|
pointed to by src2_addr. Return 0 iff the arrays are equal, -1 iff
|
|
src1_addr is lexicographically less than src2_addr, and 1 iff it is
|
|
lexicographically greater. Do not compare more than max_bytes bytes.
|
|
|
|
Undefined behavior will occur if the end of either source array
|
|
is in the last two words of the program's allocated memory space.
|
|
This is so because memcmp fetches ahead. Disallowing the fetch ahead
|
|
would impose a severe performance penalty.
|
|
|
|
Strategy:
|
|
|
|
Fetch the source strings by words and compare the words until either
|
|
a differing word is found or max_bytes is exhausted. In the former
|
|
case, move through the words to find the differing byte and return
|
|
plus or minus one, appropriately. In the latter case, return zero
|
|
(equality).
|
|
|
|
Tactics:
|
|
|
|
1) Do NOT try to fetch the words in a word aligned manner because,
|
|
in my judgement, the performance degradation experienced due to
|
|
non-aligned accesses does NOT outweigh the time and complexity added
|
|
by the preamble that would be necessary to assure alignment. This
|
|
is supported by the intuition that most source arrays (even more
|
|
true of most big source arrays) will be word aligned to begin with.
|
|
|
|
2) Rather than decrementing max_bytes to zero, I calculate the
|
|
address of the byte after the last byte of the source_1 array, and
|
|
quit when the source byte pointer passes that.
|
|
*/
|
|
|
|
.globl _memcmp
|
|
.globl __memcmp
|
|
.leafproc _memcmp,__memcmp
|
|
.align 2
|
|
|
|
_memcmp:
|
|
#ifndef __PIC
|
|
lda .Lrett,g14
|
|
#else
|
|
lda .Lrett-(.+8)(ip),g14
|
|
#endif
|
|
__memcmp:
|
|
mov g14,g13 # preserve return address
|
|
ldconst 0,g14 # conform to register conventions
|
|
cmpibge 0,g2,Lequal_exit # quit if max_bytes <= 0
|
|
addo g0,g2,g2 # calculate byte addr of byte after last in src1
|
|
|
|
.Lwloop:
|
|
cmpo g0,g2
|
|
ld (g0), g5 # fetch word of source_1
|
|
bge Lequal_exit # quit (equal) if max_bytes exhausted
|
|
ld (g1), g3 # fetch word of source_2
|
|
addo 4,g0,g0 # post-increment source_1 byte ptr
|
|
addo 4,g1,g1 # post-increment source_2 byte ptr
|
|
cmpobe g5,g3,.Lwloop # branch if source words are equal
|
|
|
|
ldconst 0xff,g4 # byte extraction mask
|
|
subo 4,g0,g0 # back up src1 pointer
|
|
|
|
.Lcloop: and g4,g5,g7 # extract and compare individual bytes
|
|
and g4,g3,g6
|
|
cmpobne g7,g6,.diff # branch if they are different
|
|
shlo 8,g4,g4 # position mask for next extraction
|
|
addo 1,g0,g0
|
|
cmpobl g0,g2,.Lcloop # quit if max_bytes is exhausted
|
|
|
|
Lequal_exit:
|
|
mov 0,g0
|
|
bx (g13)
|
|
.Lrett:
|
|
ret
|
|
|
|
.diff: bl .neg # arrays differ at current byte.
|
|
/* return 1 or -1 appropriately */
|
|
mov 1,g0
|
|
bx (g13)
|
|
.neg: subi 1,0,g0
|
|
.Lexit:
|
|
bx (g13)
|
|
|
|
/* end or memcmp */
|