]>
Commit | Line | Data |
---|---|---|
28f540f4 | 1 | /* _memcopy.c -- subroutines for memory copy functions. |
568035b7 | 2 | Copyright (C) 1991-2013 Free Software Foundation, Inc. |
41bdb6e2 | 3 | This file is part of the GNU C Library. |
28f540f4 RM |
4 | Contributed by Torbjorn Granlund (tege@sics.se). |
5 | ||
01c901a5 | 6 | The GNU C Library is free software; you can redistribute it and/or |
41bdb6e2 AJ |
7 | modify it under the terms of the GNU Lesser General Public |
8 | License as published by the Free Software Foundation; either | |
9 | version 2.1 of the License, or (at your option) any later version. | |
28f540f4 | 10 | |
01c901a5 UD |
11 | The GNU C Library is distributed in the hope that it will be useful, |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
41bdb6e2 | 14 | Lesser General Public License for more details. |
28f540f4 | 15 | |
41bdb6e2 | 16 | You should have received a copy of the GNU Lesser General Public |
59ba27a6 PE |
17 | License along with the GNU C Library; if not, see |
18 | <http://www.gnu.org/licenses/>. */ | |
28f540f4 RM |
19 | |
20 | /* BE VERY CAREFUL IF YOU CHANGE THIS CODE...! */ | |
21 | ||
28f540f4 RM |
22 | #include <stddef.h> |
23 | #include <memcopy.h> | |
24 | ||
25 | /* _wordcopy_fwd_aligned -- Copy block beginning at SRCP to | |
26 | block beginning at DSTP with LEN `op_t' words (not LEN bytes!). | |
27 | Both SRCP and DSTP should be aligned for memory operations on `op_t's. */ | |
28 | ||
29 | void | |
01c901a5 UD |
30 | _wordcopy_fwd_aligned (dstp, srcp, len) |
31 | long int dstp; | |
32 | long int srcp; | |
33 | size_t len; | |
28f540f4 RM |
34 | { |
35 | op_t a0, a1; | |
36 | ||
37 | switch (len % 8) | |
38 | { | |
39 | case 2: | |
40 | a0 = ((op_t *) srcp)[0]; | |
41 | srcp -= 6 * OPSIZ; | |
42 | dstp -= 7 * OPSIZ; | |
43 | len += 6; | |
44 | goto do1; | |
45 | case 3: | |
46 | a1 = ((op_t *) srcp)[0]; | |
47 | srcp -= 5 * OPSIZ; | |
48 | dstp -= 6 * OPSIZ; | |
49 | len += 5; | |
50 | goto do2; | |
51 | case 4: | |
52 | a0 = ((op_t *) srcp)[0]; | |
53 | srcp -= 4 * OPSIZ; | |
54 | dstp -= 5 * OPSIZ; | |
55 | len += 4; | |
56 | goto do3; | |
57 | case 5: | |
58 | a1 = ((op_t *) srcp)[0]; | |
59 | srcp -= 3 * OPSIZ; | |
60 | dstp -= 4 * OPSIZ; | |
61 | len += 3; | |
62 | goto do4; | |
63 | case 6: | |
64 | a0 = ((op_t *) srcp)[0]; | |
65 | srcp -= 2 * OPSIZ; | |
66 | dstp -= 3 * OPSIZ; | |
67 | len += 2; | |
68 | goto do5; | |
69 | case 7: | |
70 | a1 = ((op_t *) srcp)[0]; | |
71 | srcp -= 1 * OPSIZ; | |
72 | dstp -= 2 * OPSIZ; | |
73 | len += 1; | |
74 | goto do6; | |
01c901a5 | 75 | |
28f540f4 RM |
76 | case 0: |
77 | if (OP_T_THRES <= 3 * OPSIZ && len == 0) | |
78 | return; | |
79 | a0 = ((op_t *) srcp)[0]; | |
80 | srcp -= 0 * OPSIZ; | |
81 | dstp -= 1 * OPSIZ; | |
82 | goto do7; | |
83 | case 1: | |
84 | a1 = ((op_t *) srcp)[0]; | |
85 | srcp -=-1 * OPSIZ; | |
86 | dstp -= 0 * OPSIZ; | |
87 | len -= 1; | |
88 | if (OP_T_THRES <= 3 * OPSIZ && len == 0) | |
89 | goto do0; | |
90 | goto do8; /* No-op. */ | |
91 | } | |
92 | ||
93 | do | |
94 | { | |
95 | do8: | |
96 | a0 = ((op_t *) srcp)[0]; | |
97 | ((op_t *) dstp)[0] = a1; | |
98 | do7: | |
99 | a1 = ((op_t *) srcp)[1]; | |
100 | ((op_t *) dstp)[1] = a0; | |
101 | do6: | |
102 | a0 = ((op_t *) srcp)[2]; | |
103 | ((op_t *) dstp)[2] = a1; | |
104 | do5: | |
105 | a1 = ((op_t *) srcp)[3]; | |
106 | ((op_t *) dstp)[3] = a0; | |
107 | do4: | |
108 | a0 = ((op_t *) srcp)[4]; | |
109 | ((op_t *) dstp)[4] = a1; | |
110 | do3: | |
111 | a1 = ((op_t *) srcp)[5]; | |
112 | ((op_t *) dstp)[5] = a0; | |
113 | do2: | |
114 | a0 = ((op_t *) srcp)[6]; | |
115 | ((op_t *) dstp)[6] = a1; | |
116 | do1: | |
117 | a1 = ((op_t *) srcp)[7]; | |
118 | ((op_t *) dstp)[7] = a0; | |
119 | ||
120 | srcp += 8 * OPSIZ; | |
121 | dstp += 8 * OPSIZ; | |
122 | len -= 8; | |
123 | } | |
124 | while (len != 0); | |
125 | ||
126 | /* This is the right position for do0. Please don't move | |
127 | it into the loop. */ | |
128 | do0: | |
129 | ((op_t *) dstp)[0] = a1; | |
130 | } | |
131 | ||
132 | /* _wordcopy_fwd_dest_aligned -- Copy block beginning at SRCP to | |
133 | block beginning at DSTP with LEN `op_t' words (not LEN bytes!). | |
134 | DSTP should be aligned for memory operations on `op_t's, but SRCP must | |
135 | *not* be aligned. */ | |
136 | ||
137 | void | |
01c901a5 UD |
138 | _wordcopy_fwd_dest_aligned (dstp, srcp, len) |
139 | long int dstp; | |
140 | long int srcp; | |
141 | size_t len; | |
28f540f4 RM |
142 | { |
143 | op_t a0, a1, a2, a3; | |
144 | int sh_1, sh_2; | |
145 | ||
146 | /* Calculate how to shift a word read at the memory operation | |
147 | aligned srcp to make it aligned for copy. */ | |
148 | ||
149 | sh_1 = 8 * (srcp % OPSIZ); | |
150 | sh_2 = 8 * OPSIZ - sh_1; | |
151 | ||
152 | /* Make SRCP aligned by rounding it down to the beginning of the `op_t' | |
153 | it points in the middle of. */ | |
154 | srcp &= -OPSIZ; | |
155 | ||
156 | switch (len % 4) | |
157 | { | |
158 | case 2: | |
159 | a1 = ((op_t *) srcp)[0]; | |
160 | a2 = ((op_t *) srcp)[1]; | |
161 | srcp -= 1 * OPSIZ; | |
162 | dstp -= 3 * OPSIZ; | |
163 | len += 2; | |
164 | goto do1; | |
165 | case 3: | |
166 | a0 = ((op_t *) srcp)[0]; | |
167 | a1 = ((op_t *) srcp)[1]; | |
168 | srcp -= 0 * OPSIZ; | |
169 | dstp -= 2 * OPSIZ; | |
170 | len += 1; | |
171 | goto do2; | |
172 | case 0: | |
173 | if (OP_T_THRES <= 3 * OPSIZ && len == 0) | |
174 | return; | |
175 | a3 = ((op_t *) srcp)[0]; | |
176 | a0 = ((op_t *) srcp)[1]; | |
177 | srcp -=-1 * OPSIZ; | |
178 | dstp -= 1 * OPSIZ; | |
179 | len += 0; | |
180 | goto do3; | |
181 | case 1: | |
182 | a2 = ((op_t *) srcp)[0]; | |
183 | a3 = ((op_t *) srcp)[1]; | |
184 | srcp -=-2 * OPSIZ; | |
185 | dstp -= 0 * OPSIZ; | |
186 | len -= 1; | |
187 | if (OP_T_THRES <= 3 * OPSIZ && len == 0) | |
188 | goto do0; | |
189 | goto do4; /* No-op. */ | |
190 | } | |
191 | ||
192 | do | |
193 | { | |
194 | do4: | |
195 | a0 = ((op_t *) srcp)[0]; | |
196 | ((op_t *) dstp)[0] = MERGE (a2, sh_1, a3, sh_2); | |
197 | do3: | |
198 | a1 = ((op_t *) srcp)[1]; | |
199 | ((op_t *) dstp)[1] = MERGE (a3, sh_1, a0, sh_2); | |
200 | do2: | |
201 | a2 = ((op_t *) srcp)[2]; | |
202 | ((op_t *) dstp)[2] = MERGE (a0, sh_1, a1, sh_2); | |
203 | do1: | |
204 | a3 = ((op_t *) srcp)[3]; | |
205 | ((op_t *) dstp)[3] = MERGE (a1, sh_1, a2, sh_2); | |
206 | ||
207 | srcp += 4 * OPSIZ; | |
208 | dstp += 4 * OPSIZ; | |
209 | len -= 4; | |
210 | } | |
211 | while (len != 0); | |
212 | ||
213 | /* This is the right position for do0. Please don't move | |
214 | it into the loop. */ | |
215 | do0: | |
216 | ((op_t *) dstp)[0] = MERGE (a2, sh_1, a3, sh_2); | |
217 | } | |
218 | ||
219 | /* _wordcopy_bwd_aligned -- Copy block finishing right before | |
220 | SRCP to block finishing right before DSTP with LEN `op_t' words | |
221 | (not LEN bytes!). Both SRCP and DSTP should be aligned for memory | |
222 | operations on `op_t's. */ | |
223 | ||
224 | void | |
01c901a5 UD |
225 | _wordcopy_bwd_aligned (dstp, srcp, len) |
226 | long int dstp; | |
227 | long int srcp; | |
228 | size_t len; | |
28f540f4 RM |
229 | { |
230 | op_t a0, a1; | |
231 | ||
232 | switch (len % 8) | |
233 | { | |
234 | case 2: | |
235 | srcp -= 2 * OPSIZ; | |
236 | dstp -= 1 * OPSIZ; | |
237 | a0 = ((op_t *) srcp)[1]; | |
238 | len += 6; | |
239 | goto do1; | |
240 | case 3: | |
241 | srcp -= 3 * OPSIZ; | |
242 | dstp -= 2 * OPSIZ; | |
243 | a1 = ((op_t *) srcp)[2]; | |
244 | len += 5; | |
245 | goto do2; | |
246 | case 4: | |
247 | srcp -= 4 * OPSIZ; | |
248 | dstp -= 3 * OPSIZ; | |
249 | a0 = ((op_t *) srcp)[3]; | |
250 | len += 4; | |
251 | goto do3; | |
252 | case 5: | |
253 | srcp -= 5 * OPSIZ; | |
254 | dstp -= 4 * OPSIZ; | |
255 | a1 = ((op_t *) srcp)[4]; | |
256 | len += 3; | |
257 | goto do4; | |
258 | case 6: | |
259 | srcp -= 6 * OPSIZ; | |
260 | dstp -= 5 * OPSIZ; | |
261 | a0 = ((op_t *) srcp)[5]; | |
262 | len += 2; | |
263 | goto do5; | |
264 | case 7: | |
265 | srcp -= 7 * OPSIZ; | |
266 | dstp -= 6 * OPSIZ; | |
267 | a1 = ((op_t *) srcp)[6]; | |
268 | len += 1; | |
269 | goto do6; | |
01c901a5 | 270 | |
28f540f4 RM |
271 | case 0: |
272 | if (OP_T_THRES <= 3 * OPSIZ && len == 0) | |
273 | return; | |
274 | srcp -= 8 * OPSIZ; | |
275 | dstp -= 7 * OPSIZ; | |
276 | a0 = ((op_t *) srcp)[7]; | |
277 | goto do7; | |
278 | case 1: | |
279 | srcp -= 9 * OPSIZ; | |
280 | dstp -= 8 * OPSIZ; | |
281 | a1 = ((op_t *) srcp)[8]; | |
282 | len -= 1; | |
283 | if (OP_T_THRES <= 3 * OPSIZ && len == 0) | |
284 | goto do0; | |
285 | goto do8; /* No-op. */ | |
286 | } | |
287 | ||
288 | do | |
289 | { | |
290 | do8: | |
291 | a0 = ((op_t *) srcp)[7]; | |
292 | ((op_t *) dstp)[7] = a1; | |
293 | do7: | |
294 | a1 = ((op_t *) srcp)[6]; | |
295 | ((op_t *) dstp)[6] = a0; | |
296 | do6: | |
297 | a0 = ((op_t *) srcp)[5]; | |
298 | ((op_t *) dstp)[5] = a1; | |
299 | do5: | |
300 | a1 = ((op_t *) srcp)[4]; | |
301 | ((op_t *) dstp)[4] = a0; | |
302 | do4: | |
303 | a0 = ((op_t *) srcp)[3]; | |
304 | ((op_t *) dstp)[3] = a1; | |
305 | do3: | |
306 | a1 = ((op_t *) srcp)[2]; | |
307 | ((op_t *) dstp)[2] = a0; | |
308 | do2: | |
309 | a0 = ((op_t *) srcp)[1]; | |
310 | ((op_t *) dstp)[1] = a1; | |
311 | do1: | |
312 | a1 = ((op_t *) srcp)[0]; | |
313 | ((op_t *) dstp)[0] = a0; | |
314 | ||
315 | srcp -= 8 * OPSIZ; | |
316 | dstp -= 8 * OPSIZ; | |
317 | len -= 8; | |
318 | } | |
319 | while (len != 0); | |
320 | ||
321 | /* This is the right position for do0. Please don't move | |
322 | it into the loop. */ | |
323 | do0: | |
324 | ((op_t *) dstp)[7] = a1; | |
325 | } | |
326 | ||
327 | /* _wordcopy_bwd_dest_aligned -- Copy block finishing right | |
328 | before SRCP to block finishing right before DSTP with LEN `op_t' | |
329 | words (not LEN bytes!). DSTP should be aligned for memory | |
330 | operations on `op_t', but SRCP must *not* be aligned. */ | |
331 | ||
332 | void | |
01c901a5 UD |
333 | _wordcopy_bwd_dest_aligned (dstp, srcp, len) |
334 | long int dstp; | |
335 | long int srcp; | |
336 | size_t len; | |
28f540f4 RM |
337 | { |
338 | op_t a0, a1, a2, a3; | |
339 | int sh_1, sh_2; | |
340 | ||
341 | /* Calculate how to shift a word read at the memory operation | |
342 | aligned srcp to make it aligned for copy. */ | |
343 | ||
344 | sh_1 = 8 * (srcp % OPSIZ); | |
345 | sh_2 = 8 * OPSIZ - sh_1; | |
346 | ||
347 | /* Make srcp aligned by rounding it down to the beginning of the op_t | |
348 | it points in the middle of. */ | |
349 | srcp &= -OPSIZ; | |
350 | srcp += OPSIZ; | |
351 | ||
352 | switch (len % 4) | |
353 | { | |
354 | case 2: | |
355 | srcp -= 3 * OPSIZ; | |
356 | dstp -= 1 * OPSIZ; | |
357 | a2 = ((op_t *) srcp)[2]; | |
358 | a1 = ((op_t *) srcp)[1]; | |
359 | len += 2; | |
360 | goto do1; | |
361 | case 3: | |
362 | srcp -= 4 * OPSIZ; | |
363 | dstp -= 2 * OPSIZ; | |
364 | a3 = ((op_t *) srcp)[3]; | |
365 | a2 = ((op_t *) srcp)[2]; | |
366 | len += 1; | |
367 | goto do2; | |
368 | case 0: | |
369 | if (OP_T_THRES <= 3 * OPSIZ && len == 0) | |
370 | return; | |
371 | srcp -= 5 * OPSIZ; | |
372 | dstp -= 3 * OPSIZ; | |
373 | a0 = ((op_t *) srcp)[4]; | |
374 | a3 = ((op_t *) srcp)[3]; | |
375 | goto do3; | |
376 | case 1: | |
377 | srcp -= 6 * OPSIZ; | |
378 | dstp -= 4 * OPSIZ; | |
379 | a1 = ((op_t *) srcp)[5]; | |
380 | a0 = ((op_t *) srcp)[4]; | |
381 | len -= 1; | |
382 | if (OP_T_THRES <= 3 * OPSIZ && len == 0) | |
383 | goto do0; | |
384 | goto do4; /* No-op. */ | |
385 | } | |
386 | ||
387 | do | |
388 | { | |
389 | do4: | |
390 | a3 = ((op_t *) srcp)[3]; | |
391 | ((op_t *) dstp)[3] = MERGE (a0, sh_1, a1, sh_2); | |
392 | do3: | |
393 | a2 = ((op_t *) srcp)[2]; | |
394 | ((op_t *) dstp)[2] = MERGE (a3, sh_1, a0, sh_2); | |
395 | do2: | |
396 | a1 = ((op_t *) srcp)[1]; | |
397 | ((op_t *) dstp)[1] = MERGE (a2, sh_1, a3, sh_2); | |
398 | do1: | |
399 | a0 = ((op_t *) srcp)[0]; | |
400 | ((op_t *) dstp)[0] = MERGE (a1, sh_1, a2, sh_2); | |
401 | ||
402 | srcp -= 4 * OPSIZ; | |
403 | dstp -= 4 * OPSIZ; | |
404 | len -= 4; | |
405 | } | |
406 | while (len != 0); | |
407 | ||
408 | /* This is the right position for do0. Please don't move | |
409 | it into the loop. */ | |
410 | do0: | |
411 | ((op_t *) dstp)[3] = MERGE (a0, sh_1, a1, sh_2); | |
412 | } |