Parent Directory | Revision Log
Revision 1790 - (view) (download)
1 : | edgomez | 1382 | ;/**************************************************************************** |
2 : | edgomez | 851 | ; * |
3 : | edgomez | 1382 | ; * XVID MPEG-4 VIDEO CODEC |
4 : | ; * - 8<->16 bit transfer functions - | ||
5 : | edgomez | 851 | ; * |
6 : | edgomez | 1382 | ; * Copyright (C) 2002 Jaan Kalda |
7 : | edgomez | 851 | ; * |
8 : | edgomez | 1382 | ; * This program is free software ; you can redistribute it and/or modify |
9 : | ; * it under the terms of the GNU General Public License as published by | ||
10 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
11 : | ; * (at your option) any later version. | ||
12 : | edgomez | 851 | ; * |
13 : | edgomez | 1382 | ; * This program is distributed in the hope that it will be useful, |
14 : | ; * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
15 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
16 : | ; * GNU General Public License for more details. | ||
17 : | edgomez | 851 | ; * |
18 : | edgomez | 1382 | ; * You should have received a copy of the GNU General Public License |
19 : | ; * along with this program ; if not, write to the Free Software | ||
20 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
21 : | edgomez | 851 | ; * |
22 : | Isibaar | 1790 | ; * $Id: mem_transfer_3dne.asm,v 1.9 2008-08-19 09:06:48 Isibaar Exp $ |
23 : | edgomez | 1382 | ; * |
24 : | ; ***************************************************************************/ | ||
25 : | edgomez | 851 | |
26 : | edgomez | 1382 | ; these 3dne functions are compatible with iSSE, but are optimized specifically |
27 : | ; for K7 pipelines | ||
28 : | edgomez | 851 | |
29 : | edgomez | 1382 | BITS 32 |
30 : | edgomez | 851 | |
31 : | edgomez | 1382 | %macro cglobal 1 |
32 : | edgomez | 851 | %ifdef PREFIX |
33 : | edgomez | 1535 | %ifdef MARK_FUNCS |
34 : | edgomez | 1540 | global _%1:function %1.endfunc-%1 |
35 : | %define %1 _%1:function %1.endfunc-%1 | ||
36 : | edgomez | 1535 | %else |
37 : | global _%1 | ||
38 : | %define %1 _%1 | ||
39 : | %endif | ||
40 : | edgomez | 851 | %else |
41 : | edgomez | 1535 | %ifdef MARK_FUNCS |
42 : | edgomez | 1540 | global %1:function %1.endfunc-%1 |
43 : | edgomez | 1535 | %else |
44 : | global %1 | ||
45 : | %endif | ||
46 : | edgomez | 851 | %endif |
47 : | %endmacro | ||
48 : | edgomez | 1382 | |
49 : | ;============================================================================= | ||
50 : | ; Read only data | ||
51 : | ;============================================================================= | ||
52 : | |||
53 : | %ifdef FORMAT_COFF | ||
54 : | edgomez | 1519 | SECTION .rodata |
55 : | edgomez | 1382 | %else |
56 : | edgomez | 1519 | SECTION .rodata align=16 |
57 : | edgomez | 1382 | %endif |
58 : | |||
59 : | ALIGN 8 | ||
60 : | mm_zero: | ||
61 : | dd 0,0 | ||
62 : | ;============================================================================= | ||
63 : | ; Macros | ||
64 : | ;============================================================================= | ||
65 : | |||
66 : | edgomez | 851 | %macro nop4 0 |
67 : | edgomez | 1382 | db 08Dh, 074h, 026h, 0 |
68 : | edgomez | 851 | %endmacro |
69 : | |||
70 : | edgomez | 1382 | ;============================================================================= |
71 : | ; Code | ||
72 : | ;============================================================================= | ||
73 : | edgomez | 851 | |
74 : | edgomez | 1382 | SECTION .text |
75 : | |||
76 : | edgomez | 851 | cglobal transfer_8to16copy_3dne |
77 : | cglobal transfer_16to8copy_3dne | ||
78 : | cglobal transfer_8to16sub_3dne | ||
79 : | cglobal transfer_8to16subro_3dne | ||
80 : | cglobal transfer_8to16sub2_3dne | ||
81 : | cglobal transfer_16to8add_3dne | ||
82 : | cglobal transfer8x8_copy_3dne | ||
83 : | suxen_drol | 1632 | cglobal transfer8x4_copy_3dne |
84 : | edgomez | 851 | |
85 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
86 : | edgomez | 851 | ; |
87 : | ; void transfer_8to16copy_3dne(int16_t * const dst, | ||
88 : | ; const uint8_t * const src, | ||
89 : | ; uint32_t stride); | ||
90 : | ; | ||
91 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
92 : | edgomez | 851 | |
93 : | edgomez | 1382 | ALIGN 16 |
94 : | edgomez | 851 | transfer_8to16copy_3dne: |
95 : | |||
96 : | mov eax, [esp+ 8] ; Src | ||
97 : | mov edx, [esp+12] ; Stride | ||
98 : | mov ecx, [esp+ 4] ; Dst | ||
99 : | punpcklbw mm0, [byte eax] | ||
100 : | punpcklbw mm1, [eax+4] | ||
101 : | edgomez | 1382 | movq mm2, [eax+edx] |
102 : | movq mm3, [eax+edx] | ||
103 : | pxor mm7, mm7 | ||
104 : | lea eax, [eax+2*edx] | ||
105 : | punpcklbw mm2, mm7 | ||
106 : | punpckhbw mm3, mm7 | ||
107 : | psrlw mm0, 8 | ||
108 : | psrlw mm1, 8 | ||
109 : | edgomez | 851 | punpcklbw mm4, [eax] |
110 : | punpcklbw mm5, [eax+edx+4] | ||
111 : | movq [byte ecx+0*64], mm0 | ||
112 : | movq [ecx+0*64+8], mm1 | ||
113 : | punpcklbw mm6, [eax+edx] | ||
114 : | punpcklbw mm7, [eax+4] | ||
115 : | edgomez | 1382 | lea eax, [byte eax+2*edx] |
116 : | psrlw mm4, 8 | ||
117 : | psrlw mm5, 8 | ||
118 : | edgomez | 851 | punpcklbw mm0, [eax] |
119 : | punpcklbw mm1, [eax+edx+4] | ||
120 : | movq [ecx+0*64+16], mm2 | ||
121 : | movq [ecx+0*64+24], mm3 | ||
122 : | edgomez | 1382 | psrlw mm6, 8 |
123 : | psrlw mm7, 8 | ||
124 : | edgomez | 851 | punpcklbw mm2, [eax+edx] |
125 : | punpcklbw mm3, [eax+4] | ||
126 : | edgomez | 1382 | lea eax, [byte eax+2*edx] |
127 : | edgomez | 851 | movq [byte ecx+0*64+32], mm4 |
128 : | movq [ecx+0*64+56], mm5 | ||
129 : | edgomez | 1382 | psrlw mm0, 8 |
130 : | psrlw mm1, 8 | ||
131 : | edgomez | 851 | punpcklbw mm4, [eax] |
132 : | punpcklbw mm5, [eax+edx+4] | ||
133 : | movq [byte ecx+0*64+48], mm6 | ||
134 : | movq [ecx+0*64+40], mm7 | ||
135 : | edgomez | 1382 | psrlw mm2, 8 |
136 : | psrlw mm3, 8 | ||
137 : | edgomez | 851 | punpcklbw mm6, [eax+edx] |
138 : | punpcklbw mm7, [eax+4] | ||
139 : | movq [byte ecx+1*64], mm0 | ||
140 : | movq [ecx+1*64+24], mm1 | ||
141 : | edgomez | 1382 | psrlw mm4, 8 |
142 : | psrlw mm5, 8 | ||
143 : | edgomez | 851 | movq [ecx+1*64+16], mm2 |
144 : | movq [ecx+1*64+8], mm3 | ||
145 : | edgomez | 1382 | psrlw mm6, 8 |
146 : | psrlw mm7, 8 | ||
147 : | edgomez | 851 | movq [byte ecx+1*64+32], mm4 |
148 : | edgomez | 1382 | movq [ecx+1*64+56], mm5 |
149 : | edgomez | 851 | movq [byte ecx+1*64+48], mm6 |
150 : | movq [ecx+1*64+40], mm7 | ||
151 : | edgomez | 1382 | ret |
152 : | edgomez | 1540 | .endfunc |
153 : | edgomez | 851 | |
154 : | edgomez | 1382 | |
155 : | ;----------------------------------------------------------------------------- | ||
156 : | edgomez | 851 | ; |
157 : | ; void transfer_16to8copy_3dne(uint8_t * const dst, | ||
158 : | ; const int16_t * const src, | ||
159 : | ; uint32_t stride); | ||
160 : | ; | ||
161 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
162 : | edgomez | 851 | |
163 : | edgomez | 1382 | ALIGN 16 |
164 : | edgomez | 851 | transfer_16to8copy_3dne: |
165 : | |||
166 : | mov eax, [esp+ 8] ; Src | ||
167 : | mov ecx, [esp+ 4] ; Dst | ||
168 : | mov edx, [esp+12] ; Stride | ||
169 : | |||
170 : | movq mm0, [byte eax+0*32] | ||
171 : | edgomez | 1382 | packuswb mm0, [eax+0*32+8] |
172 : | edgomez | 851 | movq mm1, [eax+0*32+16] |
173 : | edgomez | 1382 | packuswb mm1, [eax+0*32+24] |
174 : | edgomez | 851 | movq mm5, [eax+2*32+16] |
175 : | movq mm2, [eax+1*32] | ||
176 : | edgomez | 1382 | packuswb mm2, [eax+1*32+8] |
177 : | edgomez | 851 | movq mm3, [eax+1*32+16] |
178 : | packuswb mm3, [eax+1*32+24] | ||
179 : | edgomez | 1382 | movq mm6, [eax+3*32] |
180 : | edgomez | 851 | movq mm4, [eax+2*32] |
181 : | edgomez | 1382 | packuswb mm4, [eax+2*32+8] |
182 : | edgomez | 851 | packuswb mm5, [eax+2*32+24] |
183 : | movq mm7, [eax+3*32+16] | ||
184 : | packuswb mm7, [eax+3*32+24] | ||
185 : | edgomez | 1382 | packuswb mm6, [eax+3*32+8] |
186 : | edgomez | 851 | movq [ecx], mm0 |
187 : | edgomez | 1382 | lea eax, [3*edx] |
188 : | add eax, ecx | ||
189 : | edgomez | 851 | movq [ecx+edx], mm1 |
190 : | movq [ecx+2*edx], mm2 | ||
191 : | movq [byte eax], mm3 | ||
192 : | movq [ecx+4*edx], mm4 | ||
193 : | edgomez | 1382 | lea ecx, [byte ecx+4*edx] |
194 : | edgomez | 851 | movq [eax+2*edx], mm5 |
195 : | movq [eax+4*edx], mm7 | ||
196 : | movq [ecx+2*edx], mm6 | ||
197 : | ret | ||
198 : | edgomez | 1540 | .endfunc |
199 : | edgomez | 851 | |
200 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
201 : | edgomez | 851 | ; |
202 : | ; void transfer_8to16sub_3dne(int16_t * const dct, | ||
203 : | ; uint8_t * const cur, | ||
204 : | ; const uint8_t * const ref, | ||
205 : | ; const uint32_t stride); | ||
206 : | ; | ||
207 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
208 : | edgomez | 851 | |
209 : | ; when second argument == 1, reference (ebx) block is to current (eax) | ||
210 : | %macro COPY_8_TO_16_SUB 2 | ||
211 : | movq mm1, [eax] ; cur | ||
212 : | movq mm0, mm1 | ||
213 : | movq mm4, [ecx] ; ref | ||
214 : | movq mm6, mm4 | ||
215 : | %if %2 == 1 | ||
216 : | movq [eax], mm4 | ||
217 : | %endif | ||
218 : | punpckhbw mm1, mm7 | ||
219 : | punpckhbw mm6, mm7 | ||
220 : | punpcklbw mm4, mm7 | ||
221 : | edgomez | 1382 | ALIGN 8 |
222 : | edgomez | 851 | movq mm2, [byte eax+edx] |
223 : | punpcklbw mm0, mm7 | ||
224 : | movq mm3, [byte eax+edx] | ||
225 : | punpcklbw mm2, mm7 | ||
226 : | movq mm5, [byte ecx+edx] ; ref | ||
227 : | punpckhbw mm3, mm7 | ||
228 : | %if %2 == 1 | ||
229 : | movq [byte eax+edx], mm5 | ||
230 : | %endif | ||
231 : | psubsw mm1, mm6 | ||
232 : | |||
233 : | movq mm6, mm5 | ||
234 : | psubsw mm0, mm4 | ||
235 : | edgomez | 1382 | %if (%1 < 3) |
236 : | edgomez | 851 | lea eax,[eax+2*edx] |
237 : | lea ecx,[ecx+2*edx] | ||
238 : | %else | ||
239 : | mov ecx,[esp] | ||
240 : | add esp,byte 4 | ||
241 : | %endif | ||
242 : | movq [edi+%1*32+ 8], mm1 | ||
243 : | movq [byte edi+%1*32+ 0], mm0 ; dst | ||
244 : | punpcklbw mm5, mm7 | ||
245 : | punpckhbw mm6, mm7 | ||
246 : | psubsw mm2, mm5 | ||
247 : | psubsw mm3, mm6 | ||
248 : | movq [edi+%1*32+16], mm2 | ||
249 : | movq [edi+%1*32+24], mm3 | ||
250 : | %endmacro | ||
251 : | |||
252 : | edgomez | 1382 | ALIGN 16 |
253 : | edgomez | 851 | transfer_8to16sub_3dne: |
254 : | mov eax, [esp + 8] ; Cur | ||
255 : | mov ecx, [esp +12] ; Ref | ||
256 : | push edi | ||
257 : | suxen_drol | 1787 | mov edx, [esp+4+16] ; Stride |
258 : | edgomez | 851 | mov edi, [esp+4+ 4] ; Dst |
259 : | pxor mm7, mm7 | ||
260 : | nop | ||
261 : | edgomez | 1382 | ALIGN 4 |
262 : | edgomez | 851 | COPY_8_TO_16_SUB 0, 1 |
263 : | COPY_8_TO_16_SUB 1, 1 | ||
264 : | COPY_8_TO_16_SUB 2, 1 | ||
265 : | COPY_8_TO_16_SUB 3, 1 | ||
266 : | edgomez | 1382 | mov edi, ecx |
267 : | edgomez | 851 | ret |
268 : | edgomez | 1540 | .endfunc |
269 : | edgomez | 851 | |
270 : | edgomez | 1382 | ALIGN 16 |
271 : | edgomez | 851 | transfer_8to16subro_3dne: |
272 : | mov eax, [esp + 8] ; Cur | ||
273 : | mov ecx, [esp +12] ; Ref | ||
274 : | push edi | ||
275 : | suxen_drol | 1787 | mov edx, [esp+4+16] ; Stride |
276 : | edgomez | 851 | mov edi, [esp+4+ 4] ; Dst |
277 : | pxor mm7, mm7 | ||
278 : | nop | ||
279 : | edgomez | 1382 | ALIGN 4 |
280 : | edgomez | 851 | COPY_8_TO_16_SUB 0, 0 |
281 : | COPY_8_TO_16_SUB 1, 0 | ||
282 : | COPY_8_TO_16_SUB 2, 0 | ||
283 : | COPY_8_TO_16_SUB 3, 0 | ||
284 : | edgomez | 1382 | mov edi, ecx |
285 : | edgomez | 851 | ret |
286 : | edgomez | 1540 | .endfunc |
287 : | edgomez | 851 | |
288 : | |||
289 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
290 : | edgomez | 851 | ; |
291 : | ; void transfer_8to16sub2_3dne(int16_t * const dct, | ||
292 : | ; uint8_t * const cur, | ||
293 : | ; const uint8_t * ref1, | ||
294 : | ; const uint8_t * ref2, | ||
295 : | ; const uint32_t stride) | ||
296 : | ; | ||
297 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
298 : | edgomez | 851 | |
299 : | %macro COPY_8_TO_16_SUB2_SSE 1 | ||
300 : | edgomez | 1382 | db 0Fh, 6Fh, 44h, 20h, 00 ;movq mm0, [byte eax] ; cur |
301 : | edgomez | 851 | punpcklbw mm0, mm7 |
302 : | movq mm2, [byte eax+edx] | ||
303 : | punpcklbw mm2, mm7 | ||
304 : | db 0Fh, 6Fh, 4ch, 20h, 00 ;movq mm1, [byte eax] | ||
305 : | punpckhbw mm1, mm7 | ||
306 : | movq mm3, [byte eax+edx] | ||
307 : | punpckhbw mm3, mm7 | ||
308 : | edgomez | 1382 | |
309 : | edgomez | 851 | movq mm4, [byte ebx] ; ref1 |
310 : | pavgb mm4, [byte esi] ; ref2 | ||
311 : | edgomez | 1382 | movq [eax], mm4 |
312 : | edgomez | 851 | movq mm5, [ebx+edx] ; ref |
313 : | pavgb mm5, [esi+edx] ; ref2 | ||
314 : | edgomez | 1382 | movq [eax+edx], mm5 |
315 : | edgomez | 851 | movq mm6, mm4 |
316 : | punpcklbw mm4, mm7 | ||
317 : | punpckhbw mm6, mm7 | ||
318 : | %if (%1 < 3) | ||
319 : | lea esi,[esi+2*edx] | ||
320 : | lea ebx,[byte ebx+2*edx] | ||
321 : | lea eax,[eax+2*edx] | ||
322 : | %else | ||
323 : | mov esi,[esp] | ||
324 : | mov ebx,[esp+4] | ||
325 : | add esp,byte 8 | ||
326 : | edgomez | 1382 | %endif |
327 : | edgomez | 851 | psubsw mm0, mm4 |
328 : | psubsw mm1, mm6 | ||
329 : | movq mm6, mm5 | ||
330 : | punpcklbw mm5, mm7 | ||
331 : | punpckhbw mm6, mm7 | ||
332 : | psubsw mm2, mm5 | ||
333 : | psubsw mm3, mm6 | ||
334 : | movq [byte ecx+%1*32+ 0], mm0 ; dst | ||
335 : | movq [ecx+%1*32+ 8], mm1 | ||
336 : | movq [ecx+%1*32+16], mm2 | ||
337 : | movq [ecx+%1*32+24], mm3 | ||
338 : | %endmacro | ||
339 : | |||
340 : | edgomez | 1382 | ALIGN 16 |
341 : | edgomez | 851 | transfer_8to16sub2_3dne: |
342 : | mov edx, [esp +20] ; Stride | ||
343 : | mov ecx, [esp + 4] ; Dst | ||
344 : | mov eax, [esp + 8] ; Cur | ||
345 : | push ebx | ||
346 : | lea ebp,[byte ebp] | ||
347 : | mov ebx, [esp+4+12] ; Ref1 | ||
348 : | push esi | ||
349 : | pxor mm7, mm7 | ||
350 : | mov esi, [esp+8+16] ; Ref2 | ||
351 : | edgomez | 1382 | nop4 |
352 : | edgomez | 851 | COPY_8_TO_16_SUB2_SSE 0 |
353 : | COPY_8_TO_16_SUB2_SSE 1 | ||
354 : | COPY_8_TO_16_SUB2_SSE 2 | ||
355 : | COPY_8_TO_16_SUB2_SSE 3 | ||
356 : | |||
357 : | ret | ||
358 : | edgomez | 1540 | .endfunc |
359 : | edgomez | 851 | |
360 : | |||
361 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
362 : | edgomez | 851 | ; |
363 : | ; void transfer_16to8add_3dne(uint8_t * const dst, | ||
364 : | ; const int16_t * const src, | ||
365 : | ; uint32_t stride); | ||
366 : | ; | ||
367 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
368 : | edgomez | 851 | |
369 : | %macro COPY_16_TO_8_ADD 1 | ||
370 : | db 0Fh, 6Fh, 44h, 21h, 00 ;movq mm0, [byte ecx] | ||
371 : | punpcklbw mm0, mm7 | ||
372 : | movq mm2, [byte ecx+edx] | ||
373 : | punpcklbw mm2, mm7 | ||
374 : | db 0Fh, 6Fh, 4ch, 21h, 00 ;movq mm1, [byte ecx] | ||
375 : | punpckhbw mm1, mm7 | ||
376 : | movq mm3, [byte ecx+edx] | ||
377 : | punpckhbw mm3, mm7 | ||
378 : | paddsw mm0, [byte eax+%1*32+ 0] | ||
379 : | paddsw mm1, [eax+%1*32+ 8] | ||
380 : | paddsw mm2, [eax+%1*32+16] | ||
381 : | paddsw mm3, [eax+%1*32+24] | ||
382 : | packuswb mm0, mm1 | ||
383 : | packuswb mm2, mm3 | ||
384 : | edgomez | 1382 | mov esp,esp |
385 : | edgomez | 851 | movq [byte ecx], mm0 |
386 : | movq [ecx+edx], mm2 | ||
387 : | %endmacro | ||
388 : | |||
389 : | |||
390 : | edgomez | 1382 | ALIGN 16 |
391 : | edgomez | 851 | transfer_16to8add_3dne: |
392 : | mov ecx, [esp+ 4] ; Dst | ||
393 : | mov edx, [esp+12] ; Stride | ||
394 : | mov eax, [esp+ 8] ; Src | ||
395 : | pxor mm7, mm7 | ||
396 : | nop | ||
397 : | |||
398 : | COPY_16_TO_8_ADD 0 | ||
399 : | lea ecx,[byte ecx+2*edx] | ||
400 : | COPY_16_TO_8_ADD 1 | ||
401 : | lea ecx,[byte ecx+2*edx] | ||
402 : | COPY_16_TO_8_ADD 2 | ||
403 : | lea ecx,[byte ecx+2*edx] | ||
404 : | COPY_16_TO_8_ADD 3 | ||
405 : | ret | ||
406 : | edgomez | 1540 | .endfunc |
407 : | edgomez | 851 | |
408 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
409 : | edgomez | 851 | ; |
410 : | ; void transfer8x8_copy_3dne(uint8_t * const dst, | ||
411 : | ; const uint8_t * const src, | ||
412 : | ; const uint32_t stride); | ||
413 : | ; | ||
414 : | ; | ||
415 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
416 : | edgomez | 851 | |
417 : | %macro COPY_8_TO_8 0 | ||
418 : | movq mm0, [byte eax] | ||
419 : | movq mm1, [eax+edx] | ||
420 : | movq [byte ecx], mm0 | ||
421 : | lea eax,[byte eax+2*edx] | ||
422 : | movq [ecx+edx], mm1 | ||
423 : | %endmacro | ||
424 : | |||
425 : | edgomez | 1382 | ALIGN 16 |
426 : | edgomez | 851 | transfer8x8_copy_3dne: |
427 : | mov eax, [esp+ 8] ; Src | ||
428 : | mov edx, [esp+12] ; Stride | ||
429 : | mov ecx, [esp+ 4] ; Dst | ||
430 : | |||
431 : | COPY_8_TO_8 | ||
432 : | lea ecx,[byte ecx+2*edx] | ||
433 : | COPY_8_TO_8 | ||
434 : | lea ecx,[byte ecx+2*edx] | ||
435 : | COPY_8_TO_8 | ||
436 : | lea ecx,[byte ecx+2*edx] | ||
437 : | COPY_8_TO_8 | ||
438 : | ret | ||
439 : | edgomez | 1540 | .endfunc |
440 : | |||
441 : | suxen_drol | 1632 | ;----------------------------------------------------------------------------- |
442 : | ; | ||
443 : | ; void transfer8x4_copy_3dne(uint8_t * const dst, | ||
444 : | ; const uint8_t * const src, | ||
445 : | ; const uint32_t stride); | ||
446 : | ; | ||
447 : | ; | ||
448 : | ;----------------------------------------------------------------------------- | ||
449 : | |||
450 : | ALIGN 16 | ||
451 : | transfer8x4_copy_3dne: | ||
452 : | mov eax, [esp+ 8] ; Src | ||
453 : | mov edx, [esp+12] ; Stride | ||
454 : | mov ecx, [esp+ 4] ; Dst | ||
455 : | |||
456 : | COPY_8_TO_8 | ||
457 : | lea ecx,[byte ecx+2*edx] | ||
458 : | COPY_8_TO_8 | ||
459 : | ret | ||
460 : | .endfunc | ||
461 : | |||
462 : | Isibaar | 1790 | |
463 : | %ifidn __OUTPUT_FORMAT__,elf | ||
464 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
465 : | %endif | ||
466 : |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |