mirror of
https://github.com/autc04/Retro68.git
synced 2024-11-28 21:49:33 +00:00
131 lines
4.7 KiB
ArmAsm
131 lines
4.7 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 "memchr.s"
|
|
#ifdef __PIC
|
|
.pic
|
|
#endif
|
|
#ifdef __PID
|
|
.pid
|
|
#endif
|
|
/*
|
|
* (c) copyright 1988,1993 Intel Corp., all rights reserved
|
|
*/
|
|
|
|
/*
|
|
procedure memchr (optimized assembler version for the 80960K series)
|
|
|
|
src_addr = memchr (src_addr, char, max_bytes)
|
|
|
|
searching from src_addr for a span of max_bytes bytes, return a
|
|
pointer to the first byte in the source array that contains the
|
|
indicated char. Return null if the char is not found.
|
|
|
|
Undefined behavior will occur if the last byte of the source array
|
|
is in the last two words of the program's allocated memory space.
|
|
This is so because memchr fetches ahead. Disallowing the fetch
|
|
ahead would impose a severe performance penalty.
|
|
|
|
Strategy:
|
|
|
|
Fetch the source array by words and scanbyte the words for the
|
|
char until either a word with the byte is found or max_bytes is
|
|
exhausted. In the former case, move through the word to find the
|
|
matching byte and return its memory address. In the latter case,
|
|
return zero (null).
|
|
|
|
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 array, and
|
|
quit when the source byte pointer passes that. Refining, actually
|
|
I calculate the address of the fifth byte after the last byte of
|
|
the source array, because the source byte pointer is ahead of the
|
|
actual examination point due to fetch ahead.
|
|
*/
|
|
|
|
.globl _memchr
|
|
.globl __memchr
|
|
.leafproc _memchr, __memchr
|
|
.align 2
|
|
_memchr:
|
|
#ifndef __PIC
|
|
lda Lrett,g14
|
|
#else
|
|
lda Lrett-(.+8)(ip),g14
|
|
#endif
|
|
__memchr:
|
|
|
|
mov g14,g13 # preserve return address
|
|
lda 0xff,g7 # byte extraction mask
|
|
and g1,g7,g1 # make char an 8-bit ordinal
|
|
mov 0,g14 # conform to register linkage standard
|
|
cmpibge 0,g2,Lnot_found # do nothing if max_bytes <= 0
|
|
addo 4,g0,g6 # post-increment src word pointer
|
|
addo g2,g6,g2 # compute ending address from start and len
|
|
ld (g0),g4 # fetch first word
|
|
shlo 8,g1,g3 # broadcast the char to four bytes
|
|
or g1,g3,g3
|
|
shlo 16,g3,g5
|
|
or g3,g5,g3
|
|
|
|
Lsearch_for_word_with_char:
|
|
mov g4,g5 # keep a copy of word
|
|
scanbyte g3,g5 # check for byte with char
|
|
ld (g6),g4 # fetch next word of src
|
|
bo Lsearch_for_char # branch if null found
|
|
addo 4,g6,g6 # post-increment src word pointer
|
|
cmpobge g2,g6,Lsearch_for_word_with_char # branch if max_bytes > 3
|
|
|
|
Lnot_found:
|
|
mov 0,g0 # char not found. Return null
|
|
bx (g13) # g0 = addr of char in src (or null); g14 = 0
|
|
Lrett:
|
|
ret
|
|
|
|
Lsearch_for_char:
|
|
cmpobe.f g6,g2,Lnot_found # quit if max_bytes exhausted
|
|
and g5,g7,g0 # extract byte
|
|
cmpo g1,g0 # is it char?
|
|
addo 1,g6,g6 # bump src byte ptr
|
|
shro 8,g5,g5 # shift word to position next byte
|
|
bne.t Lsearch_for_char
|
|
subo 5,g6,g0 # back up the byte pointer
|
|
bx (g13)
|
|
|
|
/* end of memchr */
|