1 |
207 |
jeremybenn |
/*******************************************************************************
|
2 |
|
|
*
|
3 |
|
|
* Copyright (c) 1993 Intel Corporation
|
4 |
|
|
*
|
5 |
|
|
* Intel hereby grants you permission to copy, modify, and distribute this
|
6 |
|
|
* software and its documentation. Intel grants this permission provided
|
7 |
|
|
* that the above copyright notice appears in all copies and that both the
|
8 |
|
|
* copyright notice and this permission notice appear in supporting
|
9 |
|
|
* documentation. In addition, Intel grants this permission provided that
|
10 |
|
|
* you prominently mark as "not part of the original" any modifications
|
11 |
|
|
* made to this software or documentation, and that the name of Intel
|
12 |
|
|
* Corporation not be used in advertising or publicity pertaining to
|
13 |
|
|
* distribution of the software or the documentation without specific,
|
14 |
|
|
* written prior permission.
|
15 |
|
|
*
|
16 |
|
|
* Intel Corporation provides this AS IS, WITHOUT ANY WARRANTY, EXPRESS OR
|
17 |
|
|
* IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTY OF MERCHANTABILITY
|
18 |
|
|
* OR FITNESS FOR A PARTICULAR PURPOSE. Intel makes no guarantee or
|
19 |
|
|
* representations regarding the use of, or the results of the use of,
|
20 |
|
|
* the software and documentation in terms of correctness, accuracy,
|
21 |
|
|
* reliability, currentness, or otherwise; and you rely on the software,
|
22 |
|
|
* documentation and results solely at your own risk.
|
23 |
|
|
*
|
24 |
|
|
* IN NO EVENT SHALL INTEL BE LIABLE FOR ANY LOSS OF USE, LOSS OF BUSINESS,
|
25 |
|
|
* LOSS OF PROFITS, INDIRECT, INCIDENTAL, SPECIAL OR CONSEQUENTIAL DAMAGES
|
26 |
|
|
* OF ANY KIND. IN NO EVENT SHALL INTEL'S TOTAL LIABILITY EXCEED THE SUM
|
27 |
|
|
* PAID TO INTEL FOR THE PRODUCT LICENSED HEREUNDER.
|
28 |
|
|
*
|
29 |
|
|
******************************************************************************/
|
30 |
|
|
|
31 |
|
|
.file "sncmp_ca.s"
|
32 |
|
|
#ifdef __PIC
|
33 |
|
|
.pic
|
34 |
|
|
#endif
|
35 |
|
|
#ifdef __PID
|
36 |
|
|
.pid
|
37 |
|
|
#endif
|
38 |
|
|
/*
|
39 |
|
|
* (c) copyright 1988,1993 Intel Corp., all rights reserved
|
40 |
|
|
*/
|
41 |
|
|
|
42 |
|
|
/*
|
43 |
|
|
procedure strncmp (optimized assembler version for the CA)
|
44 |
|
|
|
45 |
|
|
result = strncmp (src1_addr, src2_addr, max_bytes)
|
46 |
|
|
|
47 |
|
|
compare the null terminated string pointed to by src1_addr to
|
48 |
|
|
the string space pointed to by src2_addr. Return 0 iff the strings
|
49 |
|
|
are equal, -1 if src1_addr is lexicly less than src2_addr, and 1
|
50 |
|
|
if it is lexicly greater. Do not compare more than max_bytes bytes.
|
51 |
|
|
|
52 |
|
|
Undefined behavior will occur if the end of either source string
|
53 |
|
|
(i.e. the terminating null byte) is in the last word of the program's
|
54 |
|
|
allocated memory space. This is so because, in several cases, strncmp
|
55 |
|
|
will fetch ahead one word. Disallowing the fetch ahead would impose
|
56 |
|
|
a severe performance penalty.
|
57 |
|
|
|
58 |
|
|
This program handles five cases:
|
59 |
|
|
|
60 |
|
|
1) both arguments start on a word boundary
|
61 |
|
|
2) neither are word aligned, but they are offset by the same amount
|
62 |
|
|
3) source1 is word aligned, source2 is not
|
63 |
|
|
4) source2 is word aligned, source1 is not
|
64 |
|
|
5) neither is word aligned, and they are offset by differing amounts
|
65 |
|
|
|
66 |
|
|
At the time of this writing, only g0 thru g7 and g14 are available
|
67 |
|
|
for use in this leafproc; other registers would have to be saved and
|
68 |
|
|
restored. These nine registers are sufficient to implement the routine.
|
69 |
|
|
The registers are used as follows:
|
70 |
|
|
|
71 |
|
|
g0 original src1 ptr; extracted word; return result
|
72 |
|
|
g1 src2 ptr; 0xff -- byte extraction mask
|
73 |
|
|
g2 maximum number of bytes to compare
|
74 |
|
|
g3 src2 word ptr
|
75 |
|
|
Little endian:
|
76 |
|
|
g4 lsw of src1
|
77 |
|
|
g5 msw of src1
|
78 |
|
|
g6 src2 word
|
79 |
|
|
g7 src1 word ptr
|
80 |
|
|
Big endian:
|
81 |
|
|
g4 msw of src1
|
82 |
|
|
g5 lsw of src1
|
83 |
|
|
g6 src1 word ptr
|
84 |
|
|
g7 src2 word
|
85 |
|
|
g13 return address
|
86 |
|
|
g14 shift count
|
87 |
|
|
*/
|
88 |
|
|
|
89 |
|
|
#if __i960_BIG_ENDIAN__
|
90 |
|
|
#define MSW g4
|
91 |
|
|
#define LSW g5
|
92 |
|
|
#define SRC1 g6
|
93 |
|
|
#define SRC2 g7
|
94 |
|
|
#else
|
95 |
|
|
#define LSW g4
|
96 |
|
|
#define MSW g5
|
97 |
|
|
#define SRC2 g6
|
98 |
|
|
#define SRC1 g7
|
99 |
|
|
#endif
|
100 |
|
|
|
101 |
|
|
.globl _strncmp
|
102 |
|
|
.globl __strncmp
|
103 |
|
|
.leafproc _strncmp, __strncmp
|
104 |
|
|
.align 2
|
105 |
|
|
_strncmp:
|
106 |
|
|
#ifndef __PIC
|
107 |
|
|
lda Lrett,g14
|
108 |
|
|
#else
|
109 |
|
|
lda Lrett-(.+8)(ip),g14
|
110 |
|
|
#endif
|
111 |
|
|
__strncmp:
|
112 |
|
|
Lrestart:
|
113 |
|
|
notand g0,3,SRC1 # extract word addr of start of src1
|
114 |
|
|
lda (g14),g13 # preserve return address
|
115 |
|
|
cmpibge.f 0,g2,Lequal_exit # return equality if number of bytes to
|
116 |
|
|
/* compare is none. */
|
117 |
|
|
#if __i960_BIG_ENDIAN__
|
118 |
|
|
cmpo g0,SRC1 # check alignment of src1
|
119 |
|
|
#endif
|
120 |
|
|
ld (SRC1),LSW # fetch word with at least first byte of src1
|
121 |
|
|
notand g1,3,g3 # extract word addr of start of src2
|
122 |
|
|
ld 4(SRC1),MSW # fetch second word of src1
|
123 |
|
|
#if __i960_BIG_ENDIAN__
|
124 |
|
|
bne Lsrc1_unaligned # branch if src1 is unaligned
|
125 |
|
|
cmpo g3,g1 # check alignment of src2
|
126 |
|
|
ld (g3),SRC2 # fetch word with at least first byte of src2
|
127 |
|
|
shlo 3,g0,g14 # compute shift count for src1
|
128 |
|
|
subo g14,0,g14 # adjust shift count for big endian
|
129 |
|
|
lda 8(SRC1),SRC1 # advance src1 word addr
|
130 |
|
|
bne.f Lsrc2_unaligned # branch if src2 is NOT word aligned
|
131 |
|
|
|
132 |
|
|
/* src2 is word aligned */
|
133 |
|
|
|
134 |
|
|
mov LSW,g0
|
135 |
|
|
|
136 |
|
|
Lwloop2: # word comparing loop
|
137 |
|
|
cmpo SRC2,g0 # compare src1 and src2 words
|
138 |
|
|
lda 0xff000000,g1 # byte extraction mask
|
139 |
|
|
mov MSW,LSW # move msw of src1 to lsw
|
140 |
|
|
ld (SRC1),MSW # pre-fetch next msw of src1
|
141 |
|
|
addo 4,SRC1,SRC1 # post-increment src1 addr
|
142 |
|
|
lda 4(g3),g3 # pre-increment src2 addr
|
143 |
|
|
bne.f Lcloop # branch if src1 and src2 unequal
|
144 |
|
|
scanbyte 0,g0 # check for null byte in src1 word
|
145 |
|
|
ld (g3),SRC2 # pre-fetch next word of src2
|
146 |
|
|
mov LSW,g0 # extract word of src1
|
147 |
|
|
subi 4,g2,g2 # decrement maximum byte count
|
148 |
|
|
bo.f Lequal_exit # branch if null byte encountered
|
149 |
|
|
cmpibl.t 0,g2,Lwloop2 # branch if max_bytes not reached yet
|
150 |
|
|
|
151 |
|
|
b Lequal_exit # strings were equal up through max_bytes
|
152 |
|
|
|
153 |
|
|
Lsrc1_unaligned:
|
154 |
|
|
#endif
|
155 |
|
|
cmpo g3,g1 # check alignment of src2
|
156 |
|
|
ld (g3),SRC2 # fetch word with at least first byte of src2
|
157 |
|
|
shlo 3,g0,g14 # compute shift count for src1
|
158 |
|
|
#if __i960_BIG_ENDIAN__
|
159 |
|
|
subo g14,0,g14 # adjust shift count for big endian
|
160 |
|
|
#endif
|
161 |
|
|
eshro g14,g4,LSW # extract word of src1
|
162 |
|
|
lda 8(SRC1),SRC1 # advance src1 word addr
|
163 |
|
|
bne.f Lsrc2_unaligned # branch if src2 is NOT word aligned
|
164 |
|
|
|
165 |
|
|
/* at least src2 is word aligned */
|
166 |
|
|
|
167 |
|
|
mov LSW,g0
|
168 |
|
|
|
169 |
|
|
Lwloop: # word comparing loop
|
170 |
|
|
cmpo SRC2,g0 # compare src1 and src2 words
|
171 |
|
|
#if __i960_BIG_ENDIAN__
|
172 |
|
|
lda 0xff000000,g1 # byte extraction mask
|
173 |
|
|
#else
|
174 |
|
|
lda 0xff,g1 # byte extraction mask
|
175 |
|
|
#endif
|
176 |
|
|
mov MSW,LSW # move msw of src1 to lsw
|
177 |
|
|
ld (SRC1),MSW # pre-fetch next msw of src1
|
178 |
|
|
addo 4,SRC1,SRC1 # post-increment src1 addr
|
179 |
|
|
lda 4(g3),g3 # pre-increment src2 addr
|
180 |
|
|
bne.f Lcloop # branch if src1 and src2 unequal
|
181 |
|
|
scanbyte 0,g0 # check for null byte in src1 word
|
182 |
|
|
ld (g3),SRC2 # pre-fetch next word of src2
|
183 |
|
|
eshro g14,g4,g0 # extract word of src1
|
184 |
|
|
subi 4,g2,g2 # decrement maximum byte count
|
185 |
|
|
bo.f Lequal_exit # branch if null byte encountered
|
186 |
|
|
cmpibl.t 0,g2,Lwloop # branch if max_bytes not reached yet
|
187 |
|
|
|
188 |
|
|
b Lequal_exit # strings were equal up through max_bytes
|
189 |
|
|
|
190 |
|
|
Lcloop_setup: # setup for coming from Lsrc2_unaligned
|
191 |
|
|
mov LSW,g0 # restore extracted src1 word
|
192 |
|
|
#if __i960_BIG_ENDIAN__
|
193 |
|
|
lda 0xff000000,g1 # byte extraction mask
|
194 |
|
|
#else
|
195 |
|
|
lda 0xff,g1 # byte extraction mask
|
196 |
|
|
#endif
|
197 |
|
|
|
198 |
|
|
Lcloop: # character comparing loop
|
199 |
|
|
and SRC2,g1,g3 # extract next char of src2
|
200 |
|
|
and g0,g1,LSW # extract next char of src1
|
201 |
|
|
cmpobne.f LSW,g3,.diff # check for equality
|
202 |
|
|
cmpo 0,LSW # check for null byte
|
203 |
|
|
#if __i960_BIG_ENDIAN__
|
204 |
|
|
shro 8,g1,g1 # shift mask for next byte
|
205 |
|
|
#else
|
206 |
|
|
shlo 8,g1,g1 # shift mask for next byte
|
207 |
|
|
#endif
|
208 |
|
|
subi 1,g2,g2 # decrement character counter
|
209 |
|
|
bne.t Lcloop # branch if null not reached
|
210 |
|
|
|
211 |
|
|
/* words are equal up thru null byte */
|
212 |
|
|
|
213 |
|
|
Lequal_exit:
|
214 |
|
|
mov 0,g14 # conform to register conventions
|
215 |
|
|
lda 0,g0 # return zero, indicating equality
|
216 |
|
|
bx (g13) # return
|
217 |
|
|
|
218 |
|
|
Lrett:
|
219 |
|
|
ret
|
220 |
|
|
|
221 |
|
|
.diff:
|
222 |
|
|
mov 0,g14
|
223 |
|
|
bl Lless_than_exit
|
224 |
|
|
Lgreater_than_exit:
|
225 |
|
|
cmpibge.f 0,g2,Lequal_exit # branch if difference is beyond max_bytes
|
226 |
|
|
mov 1,g0
|
227 |
|
|
bx (g13) # g0 = 1 (src1 > src2)
|
228 |
|
|
Lless_than_exit:
|
229 |
|
|
cmpibge.f 0,g2,Lequal_exit # branch if difference is beyond max_bytes
|
230 |
|
|
subi 1,0,g0
|
231 |
|
|
bx (g13) # g0 = -1 (src1 < src2)
|
232 |
|
|
|
233 |
|
|
Lsrc2_unaligned:
|
234 |
|
|
notor g1,3,g14 # first step in computing new src1 ptr
|
235 |
|
|
ld 4(g3),SRC1 # fetch second word of src2
|
236 |
|
|
shlo 3,g1,MSW # compute shift count for src2
|
237 |
|
|
#if __i960_BIG_ENDIAN__
|
238 |
|
|
subo MSW,0,MSW # adjust shift count for big endian
|
239 |
|
|
#endif
|
240 |
|
|
eshro MSW,g6,SRC2 # extract word of src2
|
241 |
|
|
cmpo LSW,SRC2 # compare src1 and src2 words
|
242 |
|
|
lda 4(g3),g1 # set new src2 ptr
|
243 |
|
|
bne.f Lcloop_setup # first four bytes differ
|
244 |
|
|
scanbyte 0,LSW # check for null byte
|
245 |
|
|
subo g14,g0,g0 # second (final) step in computing new src1 ptr
|
246 |
|
|
addi g14,g2,g2 # compute new max_bytes too
|
247 |
|
|
lda (g13),g14 # prepare return pointer for Lrestart
|
248 |
|
|
bno.t Lrestart # if null byte not encountered, continue
|
249 |
|
|
/* with both string fetches shifted such that*/
|
250 |
|
|
/* src2 is now word aligned.*/
|
251 |
|
|
mov 0,g14 # conform to register conventions.
|
252 |
|
|
lda 0,g0 # return indicator of equality.
|
253 |
|
|
bx (g13)
|