Parent Directory | Revision Log
Revision 1.3 - (view) (download)
1 : | chl | 1.2 | ;/***************************************************************************** |
2 : | Isibaar | 1.1 | ; * |
3 : | chl | 1.2 | ; * XVID MPEG-4 VIDEO CODEC |
4 : | ; * 3dnow 8x8 block-based halfpel interpolation | ||
5 : | Isibaar | 1.1 | ; * |
6 : | chl | 1.2 | ; * Copyright(C) 2002 Peter Ross <pross@xvid.org> |
7 : | ; * Copyright(C) 2002 Michael Militzer <michael@xvid.org> | ||
8 : | Isibaar | 1.1 | ; * |
9 : | edgomez | 1.3 | ; * This file is part of XviD, a free MPEG-4 video encoder/decoder |
10 : | chl | 1.2 | ; * |
11 : | edgomez | 1.3 | ; * XviD is free software; you can redistribute it and/or modify it |
12 : | ; * under the terms of the GNU General Public License as published by | ||
13 : | chl | 1.2 | ; * the Free Software Foundation; either version 2 of the License, or |
14 : | ; * (at your option) any later version. | ||
15 : | ; * | ||
16 : | ; * This program is distributed in the hope that it will be useful, | ||
17 : | ; * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
18 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
19 : | ; * GNU General Public License for more details. | ||
20 : | ; * | ||
21 : | ; * You should have received a copy of the GNU General Public License | ||
22 : | ; * along with this program; if not, write to the Free Software | ||
23 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
24 : | edgomez | 1.3 | ; * |
25 : | ; * Under section 8 of the GNU General Public License, the copyright | ||
26 : | ; * holders of XVID explicitly forbid distribution in the following | ||
27 : | ; * countries: | ||
28 : | ; * | ||
29 : | ; * - Japan | ||
30 : | ; * - United States of America | ||
31 : | ; * | ||
32 : | ; * Linking XviD statically or dynamically with other modules is making a | ||
33 : | ; * combined work based on XviD. Thus, the terms and conditions of the | ||
34 : | ; * GNU General Public License cover the whole combination. | ||
35 : | ; * | ||
36 : | ; * As a special exception, the copyright holders of XviD give you | ||
37 : | ; * permission to link XviD with independent modules that communicate with | ||
38 : | ; * XviD solely through the VFW1.1 and DShow interfaces, regardless of the | ||
39 : | ; * license terms of these independent modules, and to copy and distribute | ||
40 : | ; * the resulting combined work under terms of your choice, provided that | ||
41 : | ; * every copy of the combined work is accompanied by a complete copy of | ||
42 : | ; * the source code of XviD (the version of XviD used to produce the | ||
43 : | ; * combined work), being distributed under the terms of the GNU General | ||
44 : | ; * Public License plus this exception. An independent module is a module | ||
45 : | ; * which is not derived from or based on XviD. | ||
46 : | ; * | ||
47 : | ; * Note that people who make modified versions of XviD are not obligated | ||
48 : | ; * to grant this special exception for their modified versions; it is | ||
49 : | ; * their choice whether to do so. The GNU General Public License gives | ||
50 : | ; * permission to release a modified version without this exception; this | ||
51 : | ; * exception also makes it possible to release a modified version which | ||
52 : | ; * carries forward this exception. | ||
53 : | ; * | ||
54 : | ; * $Id$ | ||
55 : | Isibaar | 1.1 | ; * |
56 : | chl | 1.2 | ; ****************************************************************************/ |
57 : | Isibaar | 1.1 | |
58 : | bits 32 | ||
59 : | |||
60 : | %macro cglobal 1 | ||
61 : | %ifdef PREFIX | ||
62 : | global _%1 | ||
63 : | %define %1 _%1 | ||
64 : | %else | ||
65 : | global %1 | ||
66 : | %endif | ||
67 : | %endmacro | ||
68 : | |||
69 : | section .data | ||
70 : | |||
71 : | |||
72 : | align 16 | ||
73 : | |||
74 : | mmx_one | ||
75 : | times 8 db 1 | ||
76 : | |||
77 : | section .text | ||
78 : | |||
79 : | cglobal interpolate8x8_halfpel_h_3dn | ||
80 : | cglobal interpolate8x8_halfpel_v_3dn | ||
81 : | cglobal interpolate8x8_halfpel_hv_3dn | ||
82 : | |||
83 : | |||
84 : | ;=========================================================================== | ||
85 : | ; | ||
86 : | ; void interpolate8x8_halfpel_h_3dn(uint8_t * const dst, | ||
87 : | ; const uint8_t * const src, | ||
88 : | ; const uint32_t stride, | ||
89 : | ; const uint32_t rounding); | ||
90 : | ; | ||
91 : | ;=========================================================================== | ||
92 : | |||
93 : | %macro COPY_H_3DN_RND0 0 | ||
94 : | movq mm0, [eax] | ||
95 : | pavgusb mm0, [eax+1] | ||
96 : | movq mm1, [eax+edx] | ||
97 : | pavgusb mm1, [eax+edx+1] | ||
98 : | lea eax,[eax+2*edx] | ||
99 : | movq [ecx],mm0 | ||
100 : | movq [ecx+edx],mm1 | ||
101 : | %endmacro | ||
102 : | |||
103 : | %macro COPY_H_3DN_RND1 0 | ||
104 : | movq mm0, [eax] | ||
105 : | movq mm1, [eax+edx] | ||
106 : | movq mm4, mm0 | ||
107 : | movq mm5, mm1 | ||
108 : | movq mm2, [eax+1] | ||
109 : | movq mm3, [eax+edx+1] | ||
110 : | pavgusb mm0, mm2 | ||
111 : | pxor mm2, mm4 | ||
112 : | pavgusb mm1, mm3 | ||
113 : | lea eax,[eax+2*edx] | ||
114 : | pxor mm3, mm5 | ||
115 : | pand mm2, mm7 | ||
116 : | pand mm3, mm7 | ||
117 : | psubb mm0, mm2 | ||
118 : | movq [ecx], mm0 | ||
119 : | psubb mm1, mm3 | ||
120 : | movq [ecx+edx], mm1 | ||
121 : | %endmacro | ||
122 : | |||
123 : | align 16 | ||
124 : | interpolate8x8_halfpel_h_3dn: | ||
125 : | |||
126 : | mov eax, [esp+16] ; rounding | ||
127 : | mov ecx, [esp+ 4] ; Dst | ||
128 : | test eax,eax | ||
129 : | mov eax, [esp+ 8] ; Src | ||
130 : | mov edx, [esp+12] ; stride | ||
131 : | |||
132 : | jnz near .rounding1 | ||
133 : | |||
134 : | COPY_H_3DN_RND0 | ||
135 : | lea ecx,[ecx+2*edx] | ||
136 : | COPY_H_3DN_RND0 | ||
137 : | lea ecx,[ecx+2*edx] | ||
138 : | COPY_H_3DN_RND0 | ||
139 : | lea ecx,[ecx+2*edx] | ||
140 : | COPY_H_3DN_RND0 | ||
141 : | ret | ||
142 : | |||
143 : | .rounding1 | ||
144 : | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 | ||
145 : | movq mm7, [mmx_one] | ||
146 : | COPY_H_3DN_RND1 | ||
147 : | lea ecx, [ecx+2*edx] | ||
148 : | COPY_H_3DN_RND1 | ||
149 : | lea ecx,[ecx+2*edx] | ||
150 : | COPY_H_3DN_RND1 | ||
151 : | lea ecx,[ecx+2*edx] | ||
152 : | COPY_H_3DN_RND1 | ||
153 : | ret | ||
154 : | |||
155 : | |||
156 : | ;=========================================================================== | ||
157 : | ; | ||
158 : | ; void interpolate8x8_halfpel_v_3dn(uint8_t * const dst, | ||
159 : | ; const uint8_t * const src, | ||
160 : | ; const uint32_t stride, | ||
161 : | ; const uint32_t rounding); | ||
162 : | ; | ||
163 : | ;=========================================================================== | ||
164 : | |||
165 : | %macro COPY_V_3DN_RND0 0 | ||
166 : | movq mm0, [eax] | ||
167 : | movq mm1, [eax+edx] | ||
168 : | pavgusb mm0, mm1 | ||
169 : | pavgusb mm1, [eax+2*edx] | ||
170 : | lea eax,[eax+2*edx] | ||
171 : | movq [ecx],mm0 | ||
172 : | movq [ecx+edx],mm1 | ||
173 : | %endmacro | ||
174 : | |||
175 : | %macro COPY_V_3DN_RND1 0 | ||
176 : | movq mm0, mm2 | ||
177 : | movq mm1, [eax] | ||
178 : | movq mm2, [eax+edx] | ||
179 : | lea eax,[eax+2*edx] | ||
180 : | movq mm4, mm0 | ||
181 : | movq mm5, mm1 | ||
182 : | pavgusb mm0, mm1 | ||
183 : | pxor mm4, mm1 | ||
184 : | pavgusb mm1, mm2 | ||
185 : | pxor mm5, mm2 | ||
186 : | pand mm4, mm7 ; lsb's of (i^j)... | ||
187 : | pand mm5, mm7 ; lsb's of (i^j)... | ||
188 : | psubb mm0, mm4 ; ...are substracted from result of pavgusb | ||
189 : | movq [ecx], mm0 | ||
190 : | psubb mm1, mm5 ; ...are substracted from result of pavgusb | ||
191 : | movq [ecx+edx], mm1 | ||
192 : | %endmacro | ||
193 : | |||
194 : | align 16 | ||
195 : | interpolate8x8_halfpel_v_3dn: | ||
196 : | |||
197 : | mov eax, [esp+16] ; rounding | ||
198 : | mov ecx, [esp+ 4] ; Dst | ||
199 : | test eax,eax | ||
200 : | mov eax, [esp+ 8] ; Src | ||
201 : | mov edx, [esp+12] ; stride | ||
202 : | |||
203 : | ; we process 2 line at a time | ||
204 : | |||
205 : | jnz near .rounding1 | ||
206 : | |||
207 : | COPY_V_3DN_RND0 | ||
208 : | lea ecx, [ecx+2*edx] | ||
209 : | COPY_V_3DN_RND0 | ||
210 : | lea ecx, [ecx+2*edx] | ||
211 : | COPY_V_3DN_RND0 | ||
212 : | lea ecx, [ecx+2*edx] | ||
213 : | COPY_V_3DN_RND0 | ||
214 : | ret | ||
215 : | |||
216 : | .rounding1 | ||
217 : | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 | ||
218 : | movq mm7, [mmx_one] | ||
219 : | movq mm2, [eax] ; loop invariant | ||
220 : | add eax, edx | ||
221 : | |||
222 : | COPY_V_3DN_RND1 | ||
223 : | lea ecx,[ecx+2*edx] | ||
224 : | COPY_V_3DN_RND1 | ||
225 : | lea ecx,[ecx+2*edx] | ||
226 : | COPY_V_3DN_RND1 | ||
227 : | lea ecx,[ecx+2*edx] | ||
228 : | COPY_V_3DN_RND1 | ||
229 : | ret | ||
230 : | |||
231 : | |||
232 : | ;=========================================================================== | ||
233 : | ; | ||
234 : | ; void interpolate8x8_halfpel_hv_3dn(uint8_t * const dst, | ||
235 : | ; const uint8_t * const src, | ||
236 : | ; const uint32_t stride, | ||
237 : | ; const uint32_t rounding); | ||
238 : | ; | ||
239 : | ; | ||
240 : | ;=========================================================================== | ||
241 : | |||
242 : | ; The trick is to correct the result of 'pavgusb' with some combination of the | ||
243 : | ; lsb's of the 4 input values i,j,k,l, and their intermediate 'pavgusb' (s and t). | ||
244 : | ; The boolean relations are: | ||
245 : | ; (i+j+k+l+3)/4 = (s+t+1)/2 - (ij&kl)&st | ||
246 : | ; (i+j+k+l+2)/4 = (s+t+1)/2 - (ij|kl)&st | ||
247 : | ; (i+j+k+l+1)/4 = (s+t+1)/2 - (ij&kl)|st | ||
248 : | ; (i+j+k+l+0)/4 = (s+t+1)/2 - (ij|kl)|st | ||
249 : | ; with s=(i+j+1)/2, t=(k+l+1)/2, ij = i^j, kl = k^l, st = s^t. | ||
250 : | |||
251 : | ; Moreover, we process 2 lines at a times, for better overlapping (~15% faster). | ||
252 : | |||
253 : | %macro COPY_HV_3DN_RND0 0 | ||
254 : | lea eax,[eax+edx] | ||
255 : | |||
256 : | movq mm0, [eax] | ||
257 : | movq mm1, [eax+1] | ||
258 : | |||
259 : | movq mm6, mm0 | ||
260 : | pavgusb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step | ||
261 : | lea eax,[eax+edx] | ||
262 : | pxor mm1, mm6 ; mm1=(j^k). preserved for next step | ||
263 : | |||
264 : | por mm3, mm1 ; ij |= jk | ||
265 : | movq mm6, mm2 | ||
266 : | pxor mm6, mm0 ; mm6 = s^t | ||
267 : | pand mm3, mm6 ; (ij|jk) &= st | ||
268 : | pavgusb mm2, mm0 ; mm2 = (s+t+1)/2 | ||
269 : | pand mm3, mm7 ; mask lsb | ||
270 : | psubb mm2, mm3 ; apply. | ||
271 : | |||
272 : | movq [ecx], mm2 | ||
273 : | |||
274 : | movq mm2, [eax] | ||
275 : | movq mm3, [eax+1] | ||
276 : | movq mm6, mm2 | ||
277 : | pavgusb mm2, mm3 ; preserved for next iteration | ||
278 : | lea ecx,[ecx+edx] | ||
279 : | pxor mm3, mm6 ; preserved for next iteration | ||
280 : | |||
281 : | por mm1, mm3 | ||
282 : | movq mm6, mm0 | ||
283 : | pxor mm6, mm2 | ||
284 : | pand mm1, mm6 | ||
285 : | pavgusb mm0, mm2 | ||
286 : | |||
287 : | pand mm1, mm7 | ||
288 : | psubb mm0, mm1 | ||
289 : | |||
290 : | movq [ecx], mm0 | ||
291 : | %endmacro | ||
292 : | |||
293 : | %macro COPY_HV_3DN_RND1 0 | ||
294 : | lea eax,[eax+edx] | ||
295 : | |||
296 : | movq mm0, [eax] | ||
297 : | movq mm1, [eax+1] | ||
298 : | |||
299 : | movq mm6, mm0 | ||
300 : | pavgusb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step | ||
301 : | lea eax,[eax+edx] | ||
302 : | pxor mm1, mm6 ; mm1=(j^k). preserved for next step | ||
303 : | |||
304 : | pand mm3, mm1 | ||
305 : | movq mm6, mm2 | ||
306 : | pxor mm6, mm0 | ||
307 : | por mm3, mm6 | ||
308 : | pavgusb mm2, mm0 | ||
309 : | pand mm3, mm7 | ||
310 : | psubb mm2, mm3 | ||
311 : | |||
312 : | movq [ecx], mm2 | ||
313 : | |||
314 : | movq mm2, [eax] | ||
315 : | movq mm3, [eax+1] | ||
316 : | movq mm6, mm2 | ||
317 : | pavgusb mm2, mm3 ; preserved for next iteration | ||
318 : | lea ecx,[ecx+edx] | ||
319 : | pxor mm3, mm6 ; preserved for next iteration | ||
320 : | |||
321 : | pand mm1, mm3 | ||
322 : | movq mm6, mm0 | ||
323 : | pxor mm6, mm2 | ||
324 : | por mm1, mm6 | ||
325 : | pavgusb mm0, mm2 | ||
326 : | pand mm1, mm7 | ||
327 : | psubb mm0, mm1 | ||
328 : | |||
329 : | movq [ecx], mm0 | ||
330 : | %endmacro | ||
331 : | |||
332 : | align 16 | ||
333 : | interpolate8x8_halfpel_hv_3dn | ||
334 : | mov eax, [esp+16] ; rounding | ||
335 : | mov ecx, [esp+ 4] ; Dst | ||
336 : | test eax,eax | ||
337 : | mov eax, [esp+ 8] ; Src | ||
338 : | mov edx, [esp+12] ; stride | ||
339 : | |||
340 : | movq mm7, [mmx_one] | ||
341 : | |||
342 : | ; loop invariants: mm2=(i+j+1)/2 and mm3= i^j | ||
343 : | movq mm2, [eax] | ||
344 : | movq mm3, [eax+1] | ||
345 : | movq mm6, mm2 | ||
346 : | pavgusb mm2, mm3 | ||
347 : | pxor mm3, mm6 ; mm2/mm3 ready | ||
348 : | |||
349 : | jnz near .rounding1 | ||
350 : | |||
351 : | COPY_HV_3DN_RND0 | ||
352 : | add ecx, edx | ||
353 : | COPY_HV_3DN_RND0 | ||
354 : | add ecx, edx | ||
355 : | COPY_HV_3DN_RND0 | ||
356 : | add ecx, edx | ||
357 : | COPY_HV_3DN_RND0 | ||
358 : | ret | ||
359 : | |||
360 : | .rounding1 | ||
361 : | COPY_HV_3DN_RND1 | ||
362 : | add ecx, edx | ||
363 : | COPY_HV_3DN_RND1 | ||
364 : | add ecx, edx | ||
365 : | COPY_HV_3DN_RND1 | ||
366 : | add ecx, edx | ||
367 : | COPY_HV_3DN_RND1 | ||
368 : | ret |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |