Retro68/gcc/newlib/libc/machine/i960/memchr_ca.S
Wolfgang Thaller ec13cc9ce7 fix newlib
2018-12-29 09:59:36 +01:00

154 lines
5.2 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_ca.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 CA)
src_addr = memchr (src_addr, char, max_bytes)
searching from src_addr for max_bytes bytes, return a pointer to the
first byte that contains the indicated byte in the source string.
Return null if the byte is not found.
Undefined behavior will occur if the end of the source string (i.e.
the terminating null byte) is in the last two words of the program's
allocated memory space. This is so because, in several cases, memchr
will fetch ahead. Disallowing the fetch ahead would impose a severe
performance penalty.
This program handles two cases:
1) the argument starts on a word boundary
2) the argument doesn't start on a word boundary
At the time of this writing, only g0 thru g7 and g13 are available
for use in this leafproc; other registers would have to be saved and
restored. These nine registers, plus tricky use of g14 are sufficient
to implement the routine. The registers are used as follows:
g0 src ptr; upon return it is a pointer to the matching byte, or null
g1 char to seek
g2 maximum number of bytes to check
g3 char to seek, broadcast to all four bytes
g4 word of the source string
g5 copy of the word
g6 mask to avoid unimportant bytes in first word
g7 byte extraction mask
g13 return address
g14
*/
.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
lda 0,g14 # conform to register linkage standard
cmpibge.f 0,g2,Lnot_found # do nothing if max_bytes <= 0
addo g0,g2,g2 # compute ending address from start and len
and g0,3,g6 # extract byte offset of src
notand g0,3,g0 # extract word addr of start of src
shlo 8,g1,g3 # broadcast the char to four bytes
ld (g0),g4 # fetch word containing at least first byte
or g1,g3,g3
shlo 16,g3,g5
cmpo g1,g7 # is char being sought 0xff?
or g5,g3,g3
shlo 3,g6,g6 # get shift count for making mask for first word
subi 1,0,g5 # mask initially all ones
#if __i960_BIG_ENDIAN__
shro g6,g5,g5 # get mask for bytes needed from first word
#else
shlo g6,g5,g5 # get mask for bytes needed from first word
#endif
notor g4,g5,g4 # set unneeded bytes to all ones
be.f Lsearch_for_0xff # branch if seeking 0xff
Lsearch_for_word_with_char:
scanbyte g3,g4 # check for byte with char
lda 4(g0),g0 # pre-increment src word pointer
mov g4,g5 # keep a copy of word
ld (g0),g4 # fetch next word of src
bo.f Lsearch_for_char # branch if null found
cmpoble.t g0,g2,Lsearch_for_word_with_char # branch if not null
Lnot_found:
mov 0,g0 # char not found. Return null
Lexit_code:
bx (g13) # g0 = addr of char in src (or null); g14 = 0
Lrett:
ret
Lsearch_for_char:
subo 4,g0,g0 # back up the byte pointer
Lsearch_for_char.a:
cmpobe.f g0,g2,Lnot_found # quit if max_bytes exhausted
#if __i960_BIG_ENDIAN__
rotate 8,g5,g5 # shift word to position next byte
#endif
and g5,g7,g6 # extract byte
cmpo g1,g6 # is it char?
lda 1(g0),g0 # bump src byte ptr
#if ! __i960_BIG_ENDIAN__
shro 8,g5,g5 # shift word to position next byte
#endif
bne.t Lsearch_for_char.a
subo 1,g0,g0 # back up the byte pointer
b Lexit_code
Lsearch_for_0xff:
lda 0xf0f0f0f0,g6 # make first comparison mask for char=-1 case.
or g6,g5,g6
and g4,g6,g4 # make unimportant bytes of first word 0x0f
b Lsearch_for_word_with_char
/* end of memchr */