Parent Directory | Revision Log
Revision 1.9 - (view) (download)
1 : | edgomez | 1.5 | ;/***************************************************************************** |
2 : | Isibaar | 1.1 | ; * |
3 : | edgomez | 1.5 | ; * XVID MPEG-4 VIDEO CODEC |
4 : | ; * - 3dnow 8x8 block-based halfpel interpolation - | ||
5 : | Isibaar | 1.1 | ; * |
6 : | edgomez | 1.5 | ; * Copyright(C) 2001 Peter Ross <pross@xvid.org> |
7 : | ; * 2002 Michael Militzer <isibaar@xvid.org> | ||
8 : | ; * 2002 Pascal Massimino <skal@planet-d.net> | ||
9 : | Isibaar | 1.1 | ; * |
10 : | edgomez | 1.5 | ; * This program is free software ; you can redistribute it and/or modify |
11 : | ; * it under the terms of the GNU General Public License as published by | ||
12 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
13 : | ; * (at your option) any later version. | ||
14 : | edgomez | 1.3 | ; * |
15 : | edgomez | 1.5 | ; * This program is distributed in the hope that it will be useful, |
16 : | ; * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
17 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
18 : | ; * GNU General Public License for more details. | ||
19 : | Isibaar | 1.1 | ; * |
20 : | edgomez | 1.5 | ; * You should have received a copy of the GNU General Public License |
21 : | ; * along with this program ; if not, write to the Free Software | ||
22 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
23 : | edgomez | 1.4 | ; * |
24 : | edgomez | 1.5 | ; ****************************************************************************/ |
25 : | edgomez | 1.4 | |
26 : | edgomez | 1.5 | BITS 32 |
27 : | Isibaar | 1.1 | |
28 : | edgomez | 1.5 | %macro cglobal 1 |
29 : | Isibaar | 1.1 | %ifdef PREFIX |
30 : | edgomez | 1.7 | %ifdef MARK_FUNCS |
31 : | edgomez | 1.8 | global _%1:function %1.endfunc-%1 |
32 : | %define %1 _%1:function %1.endfunc-%1 | ||
33 : | edgomez | 1.7 | %else |
34 : | global _%1 | ||
35 : | %define %1 _%1 | ||
36 : | %endif | ||
37 : | Isibaar | 1.1 | %else |
38 : | edgomez | 1.7 | %ifdef MARK_FUNCS |
39 : | edgomez | 1.8 | global %1:function %1.endfunc-%1 |
40 : | edgomez | 1.7 | %else |
41 : | global %1 | ||
42 : | %endif | ||
43 : | Isibaar | 1.1 | %endif |
44 : | %endmacro | ||
45 : | |||
46 : | edgomez | 1.5 | ;============================================================================= |
47 : | ; Read Only data | ||
48 : | ;============================================================================= | ||
49 : | |||
50 : | %ifdef FORMAT_COFF | ||
51 : | edgomez | 1.6 | SECTION .rodata |
52 : | edgomez | 1.5 | %else |
53 : | edgomez | 1.6 | SECTION .rodata align=16 |
54 : | edgomez | 1.5 | %endif |
55 : | |||
56 : | ALIGN 16 | ||
57 : | mmx_one: | ||
58 : | times 8 db 1 | ||
59 : | |||
60 : | ;============================================================================= | ||
61 : | ; Code | ||
62 : | ;============================================================================= | ||
63 : | Isibaar | 1.1 | |
64 : | edgomez | 1.5 | SECTION .text |
65 : | Isibaar | 1.1 | |
66 : | cglobal interpolate8x8_halfpel_h_3dn | ||
67 : | cglobal interpolate8x8_halfpel_v_3dn | ||
68 : | cglobal interpolate8x8_halfpel_hv_3dn | ||
69 : | |||
70 : | suxen_drol | 1.9 | cglobal interpolate8x4_halfpel_h_3dn |
71 : | cglobal interpolate8x4_halfpel_v_3dn | ||
72 : | cglobal interpolate8x4_halfpel_hv_3dn | ||
73 : | |||
74 : | edgomez | 1.5 | ;----------------------------------------------------------------------------- |
75 : | Isibaar | 1.1 | ; |
76 : | ; void interpolate8x8_halfpel_h_3dn(uint8_t * const dst, | ||
77 : | edgomez | 1.5 | ; const uint8_t * const src, |
78 : | ; const uint32_t stride, | ||
79 : | ; const uint32_t rounding); | ||
80 : | Isibaar | 1.1 | ; |
81 : | edgomez | 1.5 | ;----------------------------------------------------------------------------- |
82 : | Isibaar | 1.1 | |
83 : | %macro COPY_H_3DN_RND0 0 | ||
84 : | edgomez | 1.5 | movq mm0, [eax] |
85 : | Isibaar | 1.1 | pavgusb mm0, [eax+1] |
86 : | edgomez | 1.5 | movq mm1, [eax+edx] |
87 : | Isibaar | 1.1 | pavgusb mm1, [eax+edx+1] |
88 : | edgomez | 1.5 | lea eax, [eax+2*edx] |
89 : | movq [ecx], mm0 | ||
90 : | movq [ecx+edx], mm1 | ||
91 : | Isibaar | 1.1 | %endmacro |
92 : | |||
93 : | %macro COPY_H_3DN_RND1 0 | ||
94 : | movq mm0, [eax] | ||
95 : | movq mm1, [eax+edx] | ||
96 : | movq mm4, mm0 | ||
97 : | movq mm5, mm1 | ||
98 : | edgomez | 1.5 | movq mm2, [eax+1] |
99 : | Isibaar | 1.1 | movq mm3, [eax+edx+1] |
100 : | pavgusb mm0, mm2 | ||
101 : | pxor mm2, mm4 | ||
102 : | pavgusb mm1, mm3 | ||
103 : | edgomez | 1.5 | lea eax, [eax+2*edx] |
104 : | Isibaar | 1.1 | pxor mm3, mm5 |
105 : | pand mm2, mm7 | ||
106 : | pand mm3, mm7 | ||
107 : | psubb mm0, mm2 | ||
108 : | movq [ecx], mm0 | ||
109 : | psubb mm1, mm3 | ||
110 : | movq [ecx+edx], mm1 | ||
111 : | %endmacro | ||
112 : | |||
113 : | edgomez | 1.5 | ALIGN 16 |
114 : | Isibaar | 1.1 | interpolate8x8_halfpel_h_3dn: |
115 : | |||
116 : | mov eax, [esp+16] ; rounding | ||
117 : | mov ecx, [esp+ 4] ; Dst | ||
118 : | edgomez | 1.5 | test eax, eax |
119 : | Isibaar | 1.1 | mov eax, [esp+ 8] ; Src |
120 : | mov edx, [esp+12] ; stride | ||
121 : | |||
122 : | jnz near .rounding1 | ||
123 : | |||
124 : | COPY_H_3DN_RND0 | ||
125 : | edgomez | 1.5 | lea ecx, [ecx+2*edx] |
126 : | Isibaar | 1.1 | COPY_H_3DN_RND0 |
127 : | edgomez | 1.5 | lea ecx, [ecx+2*edx] |
128 : | Isibaar | 1.1 | COPY_H_3DN_RND0 |
129 : | edgomez | 1.5 | lea ecx, [ecx+2*edx] |
130 : | Isibaar | 1.1 | COPY_H_3DN_RND0 |
131 : | ret | ||
132 : | |||
133 : | .rounding1 | ||
134 : | edgomez | 1.5 | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 |
135 : | Isibaar | 1.1 | movq mm7, [mmx_one] |
136 : | COPY_H_3DN_RND1 | ||
137 : | lea ecx, [ecx+2*edx] | ||
138 : | COPY_H_3DN_RND1 | ||
139 : | edgomez | 1.5 | lea ecx, [ecx+2*edx] |
140 : | Isibaar | 1.1 | COPY_H_3DN_RND1 |
141 : | edgomez | 1.5 | lea ecx, [ecx+2*edx] |
142 : | Isibaar | 1.1 | COPY_H_3DN_RND1 |
143 : | ret | ||
144 : | edgomez | 1.8 | .endfunc |
145 : | Isibaar | 1.1 | |
146 : | |||
147 : | edgomez | 1.5 | ;----------------------------------------------------------------------------- |
148 : | Isibaar | 1.1 | ; |
149 : | ; void interpolate8x8_halfpel_v_3dn(uint8_t * const dst, | ||
150 : | edgomez | 1.5 | ; const uint8_t * const src, |
151 : | ; const uint32_t stride, | ||
152 : | ; const uint32_t rounding); | ||
153 : | Isibaar | 1.1 | ; |
154 : | edgomez | 1.5 | ;----------------------------------------------------------------------------- |
155 : | Isibaar | 1.1 | |
156 : | %macro COPY_V_3DN_RND0 0 | ||
157 : | edgomez | 1.5 | movq mm0, [eax] |
158 : | movq mm1, [eax+edx] | ||
159 : | Isibaar | 1.1 | pavgusb mm0, mm1 |
160 : | pavgusb mm1, [eax+2*edx] | ||
161 : | edgomez | 1.5 | lea eax, [eax+2*edx] |
162 : | movq [ecx], mm0 | ||
163 : | movq [ecx+edx], mm1 | ||
164 : | Isibaar | 1.1 | %endmacro |
165 : | |||
166 : | %macro COPY_V_3DN_RND1 0 | ||
167 : | movq mm0, mm2 | ||
168 : | movq mm1, [eax] | ||
169 : | movq mm2, [eax+edx] | ||
170 : | edgomez | 1.5 | lea eax, [eax+2*edx] |
171 : | Isibaar | 1.1 | movq mm4, mm0 |
172 : | movq mm5, mm1 | ||
173 : | pavgusb mm0, mm1 | ||
174 : | edgomez | 1.5 | pxor mm4, mm1 |
175 : | Isibaar | 1.1 | pavgusb mm1, mm2 |
176 : | pxor mm5, mm2 | ||
177 : | edgomez | 1.5 | pand mm4, mm7 ; lsb's of (i^j)... |
178 : | pand mm5, mm7 ; lsb's of (i^j)... | ||
179 : | psubb mm0, mm4 ; ...are substracted from result of pavgusb | ||
180 : | Isibaar | 1.1 | movq [ecx], mm0 |
181 : | edgomez | 1.5 | psubb mm1, mm5 ; ...are substracted from result of pavgusb |
182 : | Isibaar | 1.1 | movq [ecx+edx], mm1 |
183 : | %endmacro | ||
184 : | |||
185 : | edgomez | 1.5 | ALIGN 16 |
186 : | Isibaar | 1.1 | interpolate8x8_halfpel_v_3dn: |
187 : | |||
188 : | mov eax, [esp+16] ; rounding | ||
189 : | mov ecx, [esp+ 4] ; Dst | ||
190 : | test eax,eax | ||
191 : | mov eax, [esp+ 8] ; Src | ||
192 : | mov edx, [esp+12] ; stride | ||
193 : | |||
194 : | ; we process 2 line at a time | ||
195 : | |||
196 : | jnz near .rounding1 | ||
197 : | |||
198 : | COPY_V_3DN_RND0 | ||
199 : | lea ecx, [ecx+2*edx] | ||
200 : | COPY_V_3DN_RND0 | ||
201 : | lea ecx, [ecx+2*edx] | ||
202 : | COPY_V_3DN_RND0 | ||
203 : | lea ecx, [ecx+2*edx] | ||
204 : | COPY_V_3DN_RND0 | ||
205 : | ret | ||
206 : | |||
207 : | .rounding1 | ||
208 : | edgomez | 1.5 | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 |
209 : | Isibaar | 1.1 | movq mm7, [mmx_one] |
210 : | edgomez | 1.5 | movq mm2, [eax] ; loop invariant |
211 : | Isibaar | 1.1 | add eax, edx |
212 : | |||
213 : | COPY_V_3DN_RND1 | ||
214 : | edgomez | 1.5 | lea ecx, [ecx+2*edx] |
215 : | Isibaar | 1.1 | COPY_V_3DN_RND1 |
216 : | edgomez | 1.5 | lea ecx, [ecx+2*edx] |
217 : | Isibaar | 1.1 | COPY_V_3DN_RND1 |
218 : | edgomez | 1.5 | lea ecx, [ecx+2*edx] |
219 : | Isibaar | 1.1 | COPY_V_3DN_RND1 |
220 : | ret | ||
221 : | edgomez | 1.8 | .endfunc |
222 : | Isibaar | 1.1 | |
223 : | |||
224 : | edgomez | 1.5 | ;----------------------------------------------------------------------------- |
225 : | Isibaar | 1.1 | ; |
226 : | ; void interpolate8x8_halfpel_hv_3dn(uint8_t * const dst, | ||
227 : | edgomez | 1.5 | ; const uint8_t * const src, |
228 : | ; const uint32_t stride, | ||
229 : | ; const uint32_t rounding); | ||
230 : | Isibaar | 1.1 | ; |
231 : | ; | ||
232 : | edgomez | 1.5 | ;----------------------------------------------------------------------------- |
233 : | Isibaar | 1.1 | |
234 : | ; The trick is to correct the result of 'pavgusb' with some combination of the | ||
235 : | ; lsb's of the 4 input values i,j,k,l, and their intermediate 'pavgusb' (s and t). | ||
236 : | ; The boolean relations are: | ||
237 : | edgomez | 1.5 | ; (i+j+k+l+3)/4 = (s+t+1)/2 - (ij&kl)&st |
238 : | Isibaar | 1.1 | ; (i+j+k+l+2)/4 = (s+t+1)/2 - (ij|kl)&st |
239 : | ; (i+j+k+l+1)/4 = (s+t+1)/2 - (ij&kl)|st | ||
240 : | ; (i+j+k+l+0)/4 = (s+t+1)/2 - (ij|kl)|st | ||
241 : | ; with s=(i+j+1)/2, t=(k+l+1)/2, ij = i^j, kl = k^l, st = s^t. | ||
242 : | |||
243 : | ; Moreover, we process 2 lines at a times, for better overlapping (~15% faster). | ||
244 : | |||
245 : | %macro COPY_HV_3DN_RND0 0 | ||
246 : | edgomez | 1.5 | lea eax, [eax+edx] |
247 : | |||
248 : | movq mm0, [eax] | ||
249 : | movq mm1, [eax+1] | ||
250 : | |||
251 : | movq mm6, mm0 | ||
252 : | pavgusb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step | ||
253 : | lea eax, [eax+edx] | ||
254 : | pxor mm1, mm6 ; mm1=(j^k). preserved for next step | ||
255 : | |||
256 : | por mm3, mm1 ; ij |= jk | ||
257 : | movq mm6, mm2 | ||
258 : | pxor mm6, mm0 ; mm6 = s^t | ||
259 : | pand mm3, mm6 ; (ij|jk) &= st | ||
260 : | pavgusb mm2, mm0 ; mm2 = (s+t+1)/2 | ||
261 : | pand mm3, mm7 ; mask lsb | ||
262 : | psubb mm2, mm3 ; apply. | ||
263 : | Isibaar | 1.1 | |
264 : | edgomez | 1.5 | movq [ecx], mm2 |
265 : | Isibaar | 1.1 | |
266 : | edgomez | 1.5 | movq mm2, [eax] |
267 : | movq mm3, [eax+1] | ||
268 : | movq mm6, mm2 | ||
269 : | pavgusb mm2, mm3 ; preserved for next iteration | ||
270 : | lea ecx, [ecx+edx] | ||
271 : | pxor mm3, mm6 ; preserved for next iteration | ||
272 : | |||
273 : | por mm1, mm3 | ||
274 : | movq mm6, mm0 | ||
275 : | pxor mm6, mm2 | ||
276 : | pand mm1, mm6 | ||
277 : | pavgusb mm0, mm2 | ||
278 : | Isibaar | 1.1 | |
279 : | edgomez | 1.5 | pand mm1, mm7 |
280 : | psubb mm0, mm1 | ||
281 : | Isibaar | 1.1 | |
282 : | edgomez | 1.5 | movq [ecx], mm0 |
283 : | Isibaar | 1.1 | %endmacro |
284 : | |||
285 : | %macro COPY_HV_3DN_RND1 0 | ||
286 : | edgomez | 1.5 | lea eax,[eax+edx] |
287 : | |||
288 : | movq mm0, [eax] | ||
289 : | movq mm1, [eax+1] | ||
290 : | |||
291 : | movq mm6, mm0 | ||
292 : | pavgusb mm0, mm1 ; mm0=(j+k+1)/2. preserved for next step | ||
293 : | lea eax, [eax+edx] | ||
294 : | pxor mm1, mm6 ; mm1=(j^k). preserved for next step | ||
295 : | |||
296 : | pand mm3, mm1 | ||
297 : | movq mm6, mm2 | ||
298 : | pxor mm6, mm0 | ||
299 : | por mm3, mm6 | ||
300 : | pavgusb mm2, mm0 | ||
301 : | pand mm3, mm7 | ||
302 : | psubb mm2, mm3 | ||
303 : | Isibaar | 1.1 | |
304 : | edgomez | 1.5 | movq [ecx], mm2 |
305 : | Isibaar | 1.1 | |
306 : | edgomez | 1.5 | movq mm2, [eax] |
307 : | movq mm3, [eax+1] | ||
308 : | movq mm6, mm2 | ||
309 : | pavgusb mm2, mm3 ; preserved for next iteration | ||
310 : | lea ecx, [ecx+edx] | ||
311 : | pxor mm3, mm6 ; preserved for next iteration | ||
312 : | |||
313 : | pand mm1, mm3 | ||
314 : | movq mm6, mm0 | ||
315 : | pxor mm6, mm2 | ||
316 : | por mm1, mm6 | ||
317 : | pavgusb mm0, mm2 | ||
318 : | pand mm1, mm7 | ||
319 : | psubb mm0, mm1 | ||
320 : | Isibaar | 1.1 | |
321 : | edgomez | 1.5 | movq [ecx], mm0 |
322 : | Isibaar | 1.1 | %endmacro |
323 : | |||
324 : | edgomez | 1.5 | ALIGN 16 |
325 : | Isibaar | 1.1 | interpolate8x8_halfpel_hv_3dn |
326 : | mov eax, [esp+16] ; rounding | ||
327 : | mov ecx, [esp+ 4] ; Dst | ||
328 : | edgomez | 1.5 | test eax, eax |
329 : | Isibaar | 1.1 | mov eax, [esp+ 8] ; Src |
330 : | mov edx, [esp+12] ; stride | ||
331 : | |||
332 : | movq mm7, [mmx_one] | ||
333 : | |||
334 : | ; loop invariants: mm2=(i+j+1)/2 and mm3= i^j | ||
335 : | movq mm2, [eax] | ||
336 : | movq mm3, [eax+1] | ||
337 : | movq mm6, mm2 | ||
338 : | pavgusb mm2, mm3 | ||
339 : | edgomez | 1.5 | pxor mm3, mm6 ; mm2/mm3 ready |
340 : | Isibaar | 1.1 | |
341 : | jnz near .rounding1 | ||
342 : | |||
343 : | COPY_HV_3DN_RND0 | ||
344 : | add ecx, edx | ||
345 : | COPY_HV_3DN_RND0 | ||
346 : | add ecx, edx | ||
347 : | COPY_HV_3DN_RND0 | ||
348 : | add ecx, edx | ||
349 : | COPY_HV_3DN_RND0 | ||
350 : | ret | ||
351 : | |||
352 : | .rounding1 | ||
353 : | COPY_HV_3DN_RND1 | ||
354 : | add ecx, edx | ||
355 : | COPY_HV_3DN_RND1 | ||
356 : | add ecx, edx | ||
357 : | COPY_HV_3DN_RND1 | ||
358 : | add ecx, edx | ||
359 : | COPY_HV_3DN_RND1 | ||
360 : | edgomez | 1.5 | ret |
361 : | edgomez | 1.8 | .endfunc |
362 : | |||
363 : | suxen_drol | 1.9 | ;----------------------------------------------------------------------------- |
364 : | ; | ||
365 : | ; void interpolate8x4_halfpel_h_3dn(uint8_t * const dst, | ||
366 : | ; const uint8_t * const src, | ||
367 : | ; const uint32_t stride, | ||
368 : | ; const uint32_t rounding); | ||
369 : | ; | ||
370 : | ;----------------------------------------------------------------------------- | ||
371 : | |||
372 : | ALIGN 16 | ||
373 : | interpolate8x4_halfpel_h_3dn: | ||
374 : | |||
375 : | mov eax, [esp+16] ; rounding | ||
376 : | mov ecx, [esp+ 4] ; Dst | ||
377 : | test eax, eax | ||
378 : | mov eax, [esp+ 8] ; Src | ||
379 : | mov edx, [esp+12] ; stride | ||
380 : | |||
381 : | jnz near .rounding1 | ||
382 : | |||
383 : | COPY_H_3DN_RND0 | ||
384 : | lea ecx, [ecx+2*edx] | ||
385 : | COPY_H_3DN_RND0 | ||
386 : | ret | ||
387 : | |||
388 : | .rounding1 | ||
389 : | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 | ||
390 : | movq mm7, [mmx_one] | ||
391 : | COPY_H_3DN_RND1 | ||
392 : | lea ecx, [ecx+2*edx] | ||
393 : | COPY_H_3DN_RND1 | ||
394 : | ret | ||
395 : | .endfunc | ||
396 : | |||
397 : | |||
398 : | ;----------------------------------------------------------------------------- | ||
399 : | ; | ||
400 : | ; void interpolate8x4_halfpel_v_3dn(uint8_t * const dst, | ||
401 : | ; const uint8_t * const src, | ||
402 : | ; const uint32_t stride, | ||
403 : | ; const uint32_t rounding); | ||
404 : | ; | ||
405 : | ;----------------------------------------------------------------------------- | ||
406 : | |||
407 : | ALIGN 16 | ||
408 : | interpolate8x4_halfpel_v_3dn: | ||
409 : | |||
410 : | mov eax, [esp+16] ; rounding | ||
411 : | mov ecx, [esp+ 4] ; Dst | ||
412 : | test eax,eax | ||
413 : | mov eax, [esp+ 8] ; Src | ||
414 : | mov edx, [esp+12] ; stride | ||
415 : | |||
416 : | ; we process 2 line at a time | ||
417 : | |||
418 : | jnz near .rounding1 | ||
419 : | |||
420 : | COPY_V_3DN_RND0 | ||
421 : | lea ecx, [ecx+2*edx] | ||
422 : | COPY_V_3DN_RND0 | ||
423 : | ret | ||
424 : | |||
425 : | .rounding1 | ||
426 : | ; we use: (i+j)/2 = ( i+j+1 )/2 - (i^j)&1 | ||
427 : | movq mm7, [mmx_one] | ||
428 : | movq mm2, [eax] ; loop invariant | ||
429 : | add eax, edx | ||
430 : | |||
431 : | COPY_V_3DN_RND1 | ||
432 : | lea ecx, [ecx+2*edx] | ||
433 : | COPY_V_3DN_RND1 | ||
434 : | ret | ||
435 : | .endfunc | ||
436 : | |||
437 : | |||
438 : | ;----------------------------------------------------------------------------- | ||
439 : | ; | ||
440 : | ; void interpolate8x4_halfpel_hv_3dn(uint8_t * const dst, | ||
441 : | ; const uint8_t * const src, | ||
442 : | ; const uint32_t stride, | ||
443 : | ; const uint32_t rounding); | ||
444 : | ; | ||
445 : | ; | ||
446 : | ;----------------------------------------------------------------------------- | ||
447 : | |||
448 : | ; The trick is to correct the result of 'pavgusb' with some combination of the | ||
449 : | ; lsb's of the 4 input values i,j,k,l, and their intermediate 'pavgusb' (s and t). | ||
450 : | ; The boolean relations are: | ||
451 : | ; (i+j+k+l+3)/4 = (s+t+1)/2 - (ij&kl)&st | ||
452 : | ; (i+j+k+l+2)/4 = (s+t+1)/2 - (ij|kl)&st | ||
453 : | ; (i+j+k+l+1)/4 = (s+t+1)/2 - (ij&kl)|st | ||
454 : | ; (i+j+k+l+0)/4 = (s+t+1)/2 - (ij|kl)|st | ||
455 : | ; with s=(i+j+1)/2, t=(k+l+1)/2, ij = i^j, kl = k^l, st = s^t. | ||
456 : | |||
457 : | ALIGN 16 | ||
458 : | interpolate8x4_halfpel_hv_3dn | ||
459 : | mov eax, [esp+16] ; rounding | ||
460 : | mov ecx, [esp+ 4] ; Dst | ||
461 : | test eax, eax | ||
462 : | mov eax, [esp+ 8] ; Src | ||
463 : | mov edx, [esp+12] ; stride | ||
464 : | |||
465 : | movq mm7, [mmx_one] | ||
466 : | |||
467 : | ; loop invariants: mm2=(i+j+1)/2 and mm3= i^j | ||
468 : | movq mm2, [eax] | ||
469 : | movq mm3, [eax+1] | ||
470 : | movq mm6, mm2 | ||
471 : | pavgusb mm2, mm3 | ||
472 : | pxor mm3, mm6 ; mm2/mm3 ready | ||
473 : | |||
474 : | jnz near .rounding1 | ||
475 : | |||
476 : | COPY_HV_3DN_RND0 | ||
477 : | add ecx, edx | ||
478 : | COPY_HV_3DN_RND0 | ||
479 : | ret | ||
480 : | |||
481 : | .rounding1 | ||
482 : | COPY_HV_3DN_RND1 | ||
483 : | add ecx, edx | ||
484 : | COPY_HV_3DN_RND1 | ||
485 : | ret | ||
486 : | .endfunc | ||
487 : |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |