diff options
Diffstat (limited to 'newlib/libc/machine/i960/strncat_ca.S')
-rw-r--r-- | newlib/libc/machine/i960/strncat_ca.S | 300 |
1 files changed, 300 insertions, 0 deletions
diff --git a/newlib/libc/machine/i960/strncat_ca.S b/newlib/libc/machine/i960/strncat_ca.S new file mode 100644 index 00000000000..45c5e171e90 --- /dev/null +++ b/newlib/libc/machine/i960/strncat_ca.S @@ -0,0 +1,300 @@ +/******************************************************************************* + * + * 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 "sncat_ca.s" +#ifdef __PIC + .pic +#endif +#ifdef __PID + .pid +#endif +/* + * (c) copyright 1988,1993 Intel Corp., all rights reserved + */ + +/* + procedure strncat (optimized assembler version for the CA) + + dest_addr = strncat (dest_addr, src_addr, max_bytes) + + append the null terminated string pointed to by src_addr to the null + terminated string pointed to by dest_addr. Return the original + dest_addr. If the source string is longer than max_bytes, then + append only max_bytes bytes, and tack on a null byte on the end + + This routine will fail if the source and destination string + overlap (in particular, if the end of the source is overlapped + by the beginning of the destination). The behavior is undefined. + This is acceptable according to the draft C standard. + + Undefined behavior will also occur if the end of the source string + (i.e. the terminating null byte) is in the last word of the program's + allocated memory space. This is so because, in several cases, strncat + will fetch ahead one word. Disallowing the fetch ahead would impose + a severe performance penalty. + + This program handles five cases: + + 1) both arguments start on a word boundary + 2) neither are word aligned, but they are offset by the same amount + 3) source is word aligned, destination is not + 4) destination is word aligned, source is not + 5) neither is word aligned, and they are offset by differing amounts + + 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 original dest ptr; not modified, so that it may be returned. + g1 src ptr; shift count + g2 max_bytes + g3 src ptr (word aligned) + g4 dest ptr (word aligned) + g5 0xff -- byte extraction mask + Little endian: + g6 lsw of double word for extraction of 4 bytes + g7 msw of double word for extraction of 4 bytes + Big endian: + g6 msw of double word for extraction of 4 bytes + g7 lsw of double word for extraction of 4 bytes + g13 return address + g14 byte extracted. +*/ + +#if __i960_BIG_ENDIAN__ +#define MSW g6 +#define LSW g7 +#else +#define LSW g6 +#define MSW g7 +#endif + + .globl _strncat + .globl __strncat + .leafproc _strncat, __strncat + .align 2 +_strncat: +#ifndef __PIC + lda Lrett,g14 +#else + lda Lrett-(.+8)(ip),g14 +#endif +__strncat: + notand g0,3,g4 # extract word addr of start of dest + lda (g14),g13 # preserve return address + cmpibge.f 0,g2,Lexit_code # Lexit if number of bytes to move is <= zero. + and g0,3,LSW # extract byte offset of dest + ld (g4),MSW # fetch word containing at least first byte + shlo 3,LSW,g14 # get shift count for making mask for first word + subi 1,0,LSW # mask initially all ones +#if __i960_BIG_ENDIAN__ + shro g14,LSW,LSW # get mask for bytes needed from first word +#else + shlo g14,LSW,LSW # get mask for bytes needed from first word +#endif + notor MSW,LSW,MSW # set unneeded bytes to all ones + lda 0xff,g5 # byte extraction mask +Lsearch_for_word_with_null: + scanbyte 0,MSW # check for null byte + lda 4(g4),g4 # post-increment dest word pointer + mov MSW,LSW # keep a copy of current word + ld (g4),MSW # fetch next word of dest + bno.t Lsearch_for_word_with_null # branch if null not found yet +#if __i960_BIG_ENDIAN__ + shro 24,LSW,g14 # extract byte +#else + and g5,LSW,g14 # extract byte +#endif + cmpo 0,g14 # branch if null is first byte of word + subo 4,g4,g4 # move dest word ptr to word with null + notand g1,3,g3 # extract word addr of start of src + bne.t Lsearch_for_null + +Lcase_14: + cmpo g1,g3 # check alignment of source + ld (g3),LSW # fetch first word of source + shlo 3,g1,g14 # compute shift count + lda 4(g3),g3 # post-increment src addr + bne.f Lcase_4 # branch if source is unaligned +Lcase_1: +Lcase_1_wloop: # word copying loop + cmpi g2,4 # check for fewer than four bytes to move + lda (LSW),g1 # keep a copy of the src word + bl.f Lcase_1_cloop # branch if fewer than four bytes to copy + scanbyte 0,g1 # check for null byte in src word + ld (g3),LSW # pre-fetch next word of src + addo 4,g3,g3 # post-increment src addr + bo.f Lcase_1_cloop # branch if word contains null byte + subi 4,g2,g2 # decrease max_byte count by the 4 bytes moved + st g1,(g4) # store word in dest string + addo 4,g4,g4 # post-increment dest addr + b Lcase_1_wloop + +Lcase_3_cloop: +Lcase_1_cloop: # character copying loop (max_bytes <= 3) + cmpdeci 0,g2,g2 # is max_bytes exhausted? +#if __i960_BIG_ENDIAN__ + rotate 8,g1,g1 # move next byte into position for extraction +#endif + and g5,g1,g14 # extract next char + be.f Lstore_null # if max_bytes is exhausted, store null and quit + cmpo 0,g14 # check for null byte + stob g14,(g4) # store the byte in dest +#if ! __i960_BIG_ENDIAN__ + shro 8,g1,g1 # move next byte into position for extraction +#endif + lda 1(g4),g4 # post-increment dest byte addr + bne.t Lcase_1_cloop # branch if null not reached + bx (g13) # Lexit (g14 == 0) + +Lstore_null: + mov 0,g14 # store null, and set g14 to zero + stob g14,(g4) + bx (g13) + + +Lsearch_for_null: +#if __i960_BIG_ENDIAN__ + shlo 8,LSW,LSW # check next byte + shro 24,LSW,g14 +#else + shlo 8,g5,g5 # move mask up to next byte + and g5,LSW,g14 # extract byte +#endif + lda 1(g4),g4 # move dest byte ptr to next byte + cmpobne.t 0,g14,Lsearch_for_null # branch if null is not yet found + +Lcase_235: + cmpo g1,g3 # check alignment of src + ld (g3),LSW # pre-fetch word with start of src + and 3,g1,g1 # compute shift count + lda 0xff,g5 # load mask for byte extraction + shlo 3,g1,g14 + lda 4(g3),g3 # post-increment src word counter + be.t Lcase_3 # branch if src is word aligned + and g4,3,MSW # extract byte offset for dest string + cmpo MSW,g1 # < indicates first word of dest has more bytes + /* than first word of source. */ + ld (g3),MSW # fetch second word of src +#if __i960_BIG_ENDIAN__ + subo g14,0,g14 # adjust shift count for big endian +#endif + eshro g14,g6,g5 # extract four bytes +#if __i960_BIG_ENDIAN__ + bge.f 1f +#else + bg.f 1f +#endif + mov MSW,LSW + lda 4(g3),g3 # move src word addr to second word boundary +1: + mov g5,MSW + lda 0xff,g5 + b Lcase_25 + +Lcase_3: # src is word aligned; dest is not + mov LSW,MSW # make copy of first word of src + lda 32,g14 # initialize shift count to zero (mod 32) +Lcase_25: + +Lcase_3_cloop_at_start: # character copying loop for start of dest str + cmpdeci 0,g2,g2 # is max_bytes exhausted? +#if __i960_BIG_ENDIAN__ + shro 24,MSW,g5 # extract next char +#else + and g5,MSW,g5 # extract next char +#endif + be.f Lstore_null # Lexit if max_bytes is exhausted + cmpo 0,g5 # check for null byte + stob g5,(g4) # store the byte in dest + addo 1,g4,g4 # post-increment dest ptr + lda 0xff,g5 # re-initialize byte extraction mask + notand g4,3,g1 # extract word address + be.t Lexit_code # Lexit if null byte reached + cmpo g1,g4 # have we reached word boundary in dest yet? +#if __i960_BIG_ENDIAN__ + lda -8(g14),g14 # augment the shift counter + rotate 8,MSW,MSW # move next byte into position for extraction +#else + lda 8(g14),g14 # augment the shift counter + shro 8,MSW,MSW # move next byte into position for extraction +#endif + bne.t Lcase_3_cloop_at_start # branch if reached word boundary? + +#if __i960_BIG_ENDIAN__ + cmpo 0,g14 + ld (g3),MSW # fetch msw of operand for double shift + bne Lcase_3_wloop # branch if src is still unaligned. + +Lcase_3_wloop2: + cmpi g2,4 # less than four bytes to move? + mov LSW,g1 # extract 4 bytes of src + lda 4(g3),g3 # post-increment src word addr + bl.f Lcase_3_cloop # branch if < four bytes left to move + scanbyte 0,g1 # check for null byte + mov MSW,LSW # move msw to lsw + ld (g3),MSW # pre-fetch msw of operand for double shift + bo.f Lcase_3_cloop # branch if word contains null byte + subi 4,g2,g2 # decrease max_byte count by the 4 bytes moved + st g1,(g4) # store 4 bytes to dest + addo 4,g4,g4 # post-increment dest ptr + b Lcase_3_wloop2 +Lcase_4: + subo g14,0,g14 # adjust shift count for big endian +#else +Lcase_4: +#endif + + ld (g3),MSW # fetch msw of operand for double shift + +Lcase_3_wloop: + cmpi g2,4 # less than four bytes to move? + eshro g14,g6,g1 # extract 4 bytes of src + lda 4(g3),g3 # post-increment src word addr + bl.f Lcase_3_cloop # branch if < four bytes left to move + scanbyte 0,g1 # check for null byte + mov MSW,LSW # move msw to lsw + ld (g3),MSW # pre-fetch msw of operand for double shift + bo.f Lcase_3_cloop # branch if word contains null byte + subi 4,g2,g2 # decrease max_byte count by the 4 bytes moved + st g1,(g4) # store 4 bytes to dest + addo 4,g4,g4 # post-increment dest ptr + b Lcase_3_wloop + + +Lexit_code: + mov 0,g14 # conform to register conventions + bx (g13) # g0 = addr of dest; g14 = 0 +Lrett: + ret + +/* end of strncat */ + |