Mercurial > libavcodec.hg
annotate vc1dsp.c @ 5131:757efe20a3c8 libavcodec
Blackfin dct_quantize_bfin routine
2x performance boost in performing quantization for mpeg encoding
author | mhoffman |
---|---|
date | Tue, 12 Jun 2007 19:13:50 +0000 |
parents | a96d905dcbaa |
children | 2b72f9bc4f06 |
rev | line source |
---|---|
3526 | 1 /* |
2 * VC-1 and WMV3 decoder - DSP functions | |
3 * Copyright (c) 2006 Konstantin Shishkov | |
4 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3665
diff
changeset
|
5 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3665
diff
changeset
|
6 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3665
diff
changeset
|
7 * FFmpeg is free software; you can redistribute it and/or |
3526 | 8 * modify it under the terms of the GNU Lesser General Public |
9 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3665
diff
changeset
|
10 * version 2.1 of the License, or (at your option) any later version. |
3526 | 11 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3665
diff
changeset
|
12 * FFmpeg is distributed in the hope that it will be useful, |
3526 | 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 * Lesser General Public License for more details. | |
16 * | |
17 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3665
diff
changeset
|
18 * License along with FFmpeg; if not, write to the Free Software |
3526 | 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
20 * | |
21 */ | |
22 | |
23 /** | |
24 * @file vc1dsp.c | |
25 * VC-1 and WMV3 decoder | |
26 * | |
27 */ | |
28 | |
29 #include "dsputil.h" | |
30 | |
31 | |
4210 | 32 /** Apply overlap transform to horizontal edge |
3526 | 33 */ |
4239 | 34 static void vc1_v_overlap_c(uint8_t* src, int stride) |
3526 | 35 { |
36 int i; | |
37 int a, b, c, d; | |
4211 | 38 int d1, d2; |
4239 | 39 int rnd = 1; |
3526 | 40 for(i = 0; i < 8; i++) { |
41 a = src[-2*stride]; | |
42 b = src[-stride]; | |
43 c = src[0]; | |
44 d = src[stride]; | |
4211 | 45 d1 = (a - d + 3 + rnd) >> 3; |
46 d2 = (a - d + b - c + 4 - rnd) >> 3; | |
3526 | 47 |
4211 | 48 src[-2*stride] = a - d1; |
49 src[-stride] = b - d2; | |
50 src[0] = c + d2; | |
51 src[stride] = d + d1; | |
3526 | 52 src++; |
4239 | 53 rnd = !rnd; |
3526 | 54 } |
55 } | |
56 | |
4210 | 57 /** Apply overlap transform to vertical edge |
3526 | 58 */ |
4239 | 59 static void vc1_h_overlap_c(uint8_t* src, int stride) |
3526 | 60 { |
61 int i; | |
62 int a, b, c, d; | |
4211 | 63 int d1, d2; |
4239 | 64 int rnd = 1; |
3526 | 65 for(i = 0; i < 8; i++) { |
66 a = src[-2]; | |
67 b = src[-1]; | |
68 c = src[0]; | |
69 d = src[1]; | |
4211 | 70 d1 = (a - d + 3 + rnd) >> 3; |
71 d2 = (a - d + b - c + 4 - rnd) >> 3; | |
3526 | 72 |
4211 | 73 src[-2] = a - d1; |
74 src[-1] = b - d2; | |
75 src[0] = c + d2; | |
76 src[1] = d + d1; | |
3526 | 77 src += stride; |
4239 | 78 rnd = !rnd; |
3526 | 79 } |
80 } | |
81 | |
82 | |
83 /** Do inverse transform on 8x8 block | |
84 */ | |
85 static void vc1_inv_trans_8x8_c(DCTELEM block[64]) | |
86 { | |
87 int i; | |
88 register int t1,t2,t3,t4,t5,t6,t7,t8; | |
89 DCTELEM *src, *dst; | |
90 | |
91 src = block; | |
92 dst = block; | |
93 for(i = 0; i < 8; i++){ | |
94 t1 = 12 * (src[0] + src[4]); | |
95 t2 = 12 * (src[0] - src[4]); | |
96 t3 = 16 * src[2] + 6 * src[6]; | |
97 t4 = 6 * src[2] - 16 * src[6]; | |
98 | |
99 t5 = t1 + t3; | |
100 t6 = t2 + t4; | |
101 t7 = t2 - t4; | |
102 t8 = t1 - t3; | |
103 | |
104 t1 = 16 * src[1] + 15 * src[3] + 9 * src[5] + 4 * src[7]; | |
105 t2 = 15 * src[1] - 4 * src[3] - 16 * src[5] - 9 * src[7]; | |
106 t3 = 9 * src[1] - 16 * src[3] + 4 * src[5] + 15 * src[7]; | |
107 t4 = 4 * src[1] - 9 * src[3] + 15 * src[5] - 16 * src[7]; | |
108 | |
109 dst[0] = (t5 + t1 + 4) >> 3; | |
110 dst[1] = (t6 + t2 + 4) >> 3; | |
111 dst[2] = (t7 + t3 + 4) >> 3; | |
112 dst[3] = (t8 + t4 + 4) >> 3; | |
113 dst[4] = (t8 - t4 + 4) >> 3; | |
114 dst[5] = (t7 - t3 + 4) >> 3; | |
115 dst[6] = (t6 - t2 + 4) >> 3; | |
116 dst[7] = (t5 - t1 + 4) >> 3; | |
117 | |
118 src += 8; | |
119 dst += 8; | |
120 } | |
121 | |
122 src = block; | |
123 dst = block; | |
124 for(i = 0; i < 8; i++){ | |
125 t1 = 12 * (src[ 0] + src[32]); | |
126 t2 = 12 * (src[ 0] - src[32]); | |
127 t3 = 16 * src[16] + 6 * src[48]; | |
128 t4 = 6 * src[16] - 16 * src[48]; | |
129 | |
130 t5 = t1 + t3; | |
131 t6 = t2 + t4; | |
132 t7 = t2 - t4; | |
133 t8 = t1 - t3; | |
134 | |
135 t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56]; | |
136 t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56]; | |
137 t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56]; | |
138 t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56]; | |
139 | |
140 dst[ 0] = (t5 + t1 + 64) >> 7; | |
141 dst[ 8] = (t6 + t2 + 64) >> 7; | |
142 dst[16] = (t7 + t3 + 64) >> 7; | |
143 dst[24] = (t8 + t4 + 64) >> 7; | |
144 dst[32] = (t8 - t4 + 64 + 1) >> 7; | |
145 dst[40] = (t7 - t3 + 64 + 1) >> 7; | |
146 dst[48] = (t6 - t2 + 64 + 1) >> 7; | |
147 dst[56] = (t5 - t1 + 64 + 1) >> 7; | |
148 | |
149 src++; | |
150 dst++; | |
151 } | |
152 } | |
153 | |
154 /** Do inverse transform on 8x4 part of block | |
155 */ | |
156 static void vc1_inv_trans_8x4_c(DCTELEM block[64], int n) | |
157 { | |
158 int i; | |
159 register int t1,t2,t3,t4,t5,t6,t7,t8; | |
160 DCTELEM *src, *dst; | |
161 int off; | |
162 | |
163 off = n * 32; | |
164 src = block + off; | |
165 dst = block + off; | |
166 for(i = 0; i < 4; i++){ | |
167 t1 = 12 * (src[0] + src[4]); | |
168 t2 = 12 * (src[0] - src[4]); | |
169 t3 = 16 * src[2] + 6 * src[6]; | |
170 t4 = 6 * src[2] - 16 * src[6]; | |
171 | |
172 t5 = t1 + t3; | |
173 t6 = t2 + t4; | |
174 t7 = t2 - t4; | |
175 t8 = t1 - t3; | |
176 | |
177 t1 = 16 * src[1] + 15 * src[3] + 9 * src[5] + 4 * src[7]; | |
178 t2 = 15 * src[1] - 4 * src[3] - 16 * src[5] - 9 * src[7]; | |
179 t3 = 9 * src[1] - 16 * src[3] + 4 * src[5] + 15 * src[7]; | |
180 t4 = 4 * src[1] - 9 * src[3] + 15 * src[5] - 16 * src[7]; | |
181 | |
182 dst[0] = (t5 + t1 + 4) >> 3; | |
183 dst[1] = (t6 + t2 + 4) >> 3; | |
184 dst[2] = (t7 + t3 + 4) >> 3; | |
185 dst[3] = (t8 + t4 + 4) >> 3; | |
186 dst[4] = (t8 - t4 + 4) >> 3; | |
187 dst[5] = (t7 - t3 + 4) >> 3; | |
188 dst[6] = (t6 - t2 + 4) >> 3; | |
189 dst[7] = (t5 - t1 + 4) >> 3; | |
190 | |
191 src += 8; | |
192 dst += 8; | |
193 } | |
194 | |
195 src = block + off; | |
196 dst = block + off; | |
197 for(i = 0; i < 8; i++){ | |
198 t1 = 17 * (src[ 0] + src[16]); | |
199 t2 = 17 * (src[ 0] - src[16]); | |
200 t3 = 22 * src[ 8]; | |
201 t4 = 22 * src[24]; | |
202 t5 = 10 * src[ 8]; | |
203 t6 = 10 * src[24]; | |
204 | |
205 dst[ 0] = (t1 + t3 + t6 + 64) >> 7; | |
206 dst[ 8] = (t2 - t4 + t5 + 64) >> 7; | |
207 dst[16] = (t2 + t4 - t5 + 64) >> 7; | |
208 dst[24] = (t1 - t3 - t6 + 64) >> 7; | |
209 | |
210 src ++; | |
211 dst ++; | |
212 } | |
213 } | |
214 | |
215 /** Do inverse transform on 4x8 parts of block | |
216 */ | |
217 static void vc1_inv_trans_4x8_c(DCTELEM block[64], int n) | |
218 { | |
219 int i; | |
220 register int t1,t2,t3,t4,t5,t6,t7,t8; | |
221 DCTELEM *src, *dst; | |
222 int off; | |
223 | |
224 off = n * 4; | |
225 src = block + off; | |
226 dst = block + off; | |
227 for(i = 0; i < 8; i++){ | |
228 t1 = 17 * (src[0] + src[2]); | |
229 t2 = 17 * (src[0] - src[2]); | |
230 t3 = 22 * src[1]; | |
231 t4 = 22 * src[3]; | |
232 t5 = 10 * src[1]; | |
233 t6 = 10 * src[3]; | |
234 | |
235 dst[0] = (t1 + t3 + t6 + 4) >> 3; | |
236 dst[1] = (t2 - t4 + t5 + 4) >> 3; | |
237 dst[2] = (t2 + t4 - t5 + 4) >> 3; | |
238 dst[3] = (t1 - t3 - t6 + 4) >> 3; | |
239 | |
240 src += 8; | |
241 dst += 8; | |
242 } | |
243 | |
244 src = block + off; | |
245 dst = block + off; | |
246 for(i = 0; i < 4; i++){ | |
247 t1 = 12 * (src[ 0] + src[32]); | |
248 t2 = 12 * (src[ 0] - src[32]); | |
249 t3 = 16 * src[16] + 6 * src[48]; | |
250 t4 = 6 * src[16] - 16 * src[48]; | |
251 | |
252 t5 = t1 + t3; | |
253 t6 = t2 + t4; | |
254 t7 = t2 - t4; | |
255 t8 = t1 - t3; | |
256 | |
257 t1 = 16 * src[ 8] + 15 * src[24] + 9 * src[40] + 4 * src[56]; | |
258 t2 = 15 * src[ 8] - 4 * src[24] - 16 * src[40] - 9 * src[56]; | |
259 t3 = 9 * src[ 8] - 16 * src[24] + 4 * src[40] + 15 * src[56]; | |
260 t4 = 4 * src[ 8] - 9 * src[24] + 15 * src[40] - 16 * src[56]; | |
261 | |
262 dst[ 0] = (t5 + t1 + 64) >> 7; | |
263 dst[ 8] = (t6 + t2 + 64) >> 7; | |
264 dst[16] = (t7 + t3 + 64) >> 7; | |
265 dst[24] = (t8 + t4 + 64) >> 7; | |
266 dst[32] = (t8 - t4 + 64 + 1) >> 7; | |
267 dst[40] = (t7 - t3 + 64 + 1) >> 7; | |
268 dst[48] = (t6 - t2 + 64 + 1) >> 7; | |
269 dst[56] = (t5 - t1 + 64 + 1) >> 7; | |
270 | |
271 src++; | |
272 dst++; | |
273 } | |
274 } | |
275 | |
276 /** Do inverse transform on 4x4 part of block | |
277 */ | |
278 static void vc1_inv_trans_4x4_c(DCTELEM block[64], int n) | |
279 { | |
280 int i; | |
281 register int t1,t2,t3,t4,t5,t6; | |
282 DCTELEM *src, *dst; | |
283 int off; | |
284 | |
285 off = (n&1) * 4 + (n&2) * 16; | |
286 src = block + off; | |
287 dst = block + off; | |
288 for(i = 0; i < 4; i++){ | |
289 t1 = 17 * (src[0] + src[2]); | |
290 t2 = 17 * (src[0] - src[2]); | |
291 t3 = 22 * src[1]; | |
292 t4 = 22 * src[3]; | |
293 t5 = 10 * src[1]; | |
294 t6 = 10 * src[3]; | |
295 | |
296 dst[0] = (t1 + t3 + t6 + 4) >> 3; | |
297 dst[1] = (t2 - t4 + t5 + 4) >> 3; | |
298 dst[2] = (t2 + t4 - t5 + 4) >> 3; | |
299 dst[3] = (t1 - t3 - t6 + 4) >> 3; | |
300 | |
301 src += 8; | |
302 dst += 8; | |
303 } | |
304 | |
305 src = block + off; | |
306 dst = block + off; | |
307 for(i = 0; i < 4; i++){ | |
308 t1 = 17 * (src[ 0] + src[16]); | |
309 t2 = 17 * (src[ 0] - src[16]); | |
310 t3 = 22 * src[ 8]; | |
311 t4 = 22 * src[24]; | |
312 t5 = 10 * src[ 8]; | |
313 t6 = 10 * src[24]; | |
314 | |
315 dst[ 0] = (t1 + t3 + t6 + 64) >> 7; | |
316 dst[ 8] = (t2 - t4 + t5 + 64) >> 7; | |
317 dst[16] = (t2 + t4 - t5 + 64) >> 7; | |
318 dst[24] = (t1 - t3 - t6 + 64) >> 7; | |
319 | |
320 src ++; | |
321 dst ++; | |
322 } | |
323 } | |
324 | |
325 /* motion compensation functions */ | |
326 | |
327 /** Filter used to interpolate fractional pel values | |
328 */ | |
4283
d6f83e2f8804
rename always_inline to av_always_inline and move to common.h
mru
parents:
4239
diff
changeset
|
329 static av_always_inline int vc1_mspel_filter(const uint8_t *src, int stride, int mode, int r) |
3526 | 330 { |
331 switch(mode){ | |
332 case 0: //no shift | |
333 return src[0]; | |
334 case 1: // 1/4 shift | |
335 return (-4*src[-stride] + 53*src[0] + 18*src[stride] - 3*src[stride*2] + 32 - r) >> 6; | |
336 case 2: // 1/2 shift | |
337 return (-src[-stride] + 9*src[0] + 9*src[stride] - src[stride*2] + 8 - r) >> 4; | |
338 case 3: // 3/4 shift | |
339 return (-3*src[-stride] + 18*src[0] + 53*src[stride] - 4*src[stride*2] + 32 - r) >> 6; | |
340 } | |
341 return 0; //should not occur | |
342 } | |
343 | |
344 /** Function used to do motion compensation with bicubic interpolation | |
345 */ | |
346 static void vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride, int mode, int rnd) | |
347 { | |
348 int i, j; | |
349 uint8_t tmp[8*11], *tptr; | |
350 int m, r; | |
351 | |
352 m = (mode & 3); | |
353 r = rnd; | |
354 src -= stride; | |
355 tptr = tmp; | |
356 for(j = 0; j < 11; j++) { | |
357 for(i = 0; i < 8; i++) | |
4594 | 358 tptr[i] = av_clip_uint8(vc1_mspel_filter(src + i, 1, m, r)); |
3526 | 359 src += stride; |
360 tptr += 8; | |
361 } | |
362 r = 1 - rnd; | |
363 m = (mode >> 2) & 3; | |
364 | |
365 tptr = tmp + 8; | |
366 for(j = 0; j < 8; j++) { | |
367 for(i = 0; i < 8; i++) | |
4594 | 368 dst[i] = av_clip_uint8(vc1_mspel_filter(tptr + i, 8, m, r)); |
3526 | 369 dst += stride; |
370 tptr += 8; | |
371 } | |
372 } | |
373 | |
374 /* pixel functions - really are entry points to vc1_mspel_mc */ | |
375 | |
376 /* this one is defined in dsputil.c */ | |
377 void ff_put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src, int stride, int rnd); | |
378 | |
379 static void ff_put_vc1_mspel_mc10_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
380 vc1_mspel_mc(dst, src, stride, 0x1, rnd); | |
381 } | |
382 | |
383 static void ff_put_vc1_mspel_mc20_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
384 vc1_mspel_mc(dst, src, stride, 0x2, rnd); | |
385 } | |
386 | |
387 static void ff_put_vc1_mspel_mc30_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
388 vc1_mspel_mc(dst, src, stride, 0x3, rnd); | |
389 } | |
390 | |
391 static void ff_put_vc1_mspel_mc01_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
392 vc1_mspel_mc(dst, src, stride, 0x4, rnd); | |
393 } | |
394 | |
395 static void ff_put_vc1_mspel_mc11_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
396 vc1_mspel_mc(dst, src, stride, 0x5, rnd); | |
397 } | |
398 | |
399 static void ff_put_vc1_mspel_mc21_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
400 vc1_mspel_mc(dst, src, stride, 0x6, rnd); | |
401 } | |
402 | |
403 static void ff_put_vc1_mspel_mc31_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
404 vc1_mspel_mc(dst, src, stride, 0x7, rnd); | |
405 } | |
406 | |
407 static void ff_put_vc1_mspel_mc02_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
408 vc1_mspel_mc(dst, src, stride, 0x8, rnd); | |
409 } | |
410 | |
411 static void ff_put_vc1_mspel_mc12_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
412 vc1_mspel_mc(dst, src, stride, 0x9, rnd); | |
413 } | |
414 | |
415 static void ff_put_vc1_mspel_mc22_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
416 vc1_mspel_mc(dst, src, stride, 0xA, rnd); | |
417 } | |
418 | |
419 static void ff_put_vc1_mspel_mc32_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
420 vc1_mspel_mc(dst, src, stride, 0xB, rnd); | |
421 } | |
422 | |
423 static void ff_put_vc1_mspel_mc03_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
424 vc1_mspel_mc(dst, src, stride, 0xC, rnd); | |
425 } | |
426 | |
427 static void ff_put_vc1_mspel_mc13_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
428 vc1_mspel_mc(dst, src, stride, 0xD, rnd); | |
429 } | |
430 | |
431 static void ff_put_vc1_mspel_mc23_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
432 vc1_mspel_mc(dst, src, stride, 0xE, rnd); | |
433 } | |
434 | |
435 static void ff_put_vc1_mspel_mc33_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { | |
436 vc1_mspel_mc(dst, src, stride, 0xF, rnd); | |
437 } | |
438 | |
439 void ff_vc1dsp_init(DSPContext* dsp, AVCodecContext *avctx) { | |
440 dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c; | |
441 dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c; | |
442 dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c; | |
443 dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c; | |
444 dsp->vc1_h_overlap = vc1_h_overlap_c; | |
445 dsp->vc1_v_overlap = vc1_v_overlap_c; | |
446 | |
447 dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_vc1_mspel_mc00_c; | |
448 dsp->put_vc1_mspel_pixels_tab[ 1] = ff_put_vc1_mspel_mc10_c; | |
449 dsp->put_vc1_mspel_pixels_tab[ 2] = ff_put_vc1_mspel_mc20_c; | |
450 dsp->put_vc1_mspel_pixels_tab[ 3] = ff_put_vc1_mspel_mc30_c; | |
451 dsp->put_vc1_mspel_pixels_tab[ 4] = ff_put_vc1_mspel_mc01_c; | |
452 dsp->put_vc1_mspel_pixels_tab[ 5] = ff_put_vc1_mspel_mc11_c; | |
453 dsp->put_vc1_mspel_pixels_tab[ 6] = ff_put_vc1_mspel_mc21_c; | |
454 dsp->put_vc1_mspel_pixels_tab[ 7] = ff_put_vc1_mspel_mc31_c; | |
455 dsp->put_vc1_mspel_pixels_tab[ 8] = ff_put_vc1_mspel_mc02_c; | |
456 dsp->put_vc1_mspel_pixels_tab[ 9] = ff_put_vc1_mspel_mc12_c; | |
457 dsp->put_vc1_mspel_pixels_tab[10] = ff_put_vc1_mspel_mc22_c; | |
458 dsp->put_vc1_mspel_pixels_tab[11] = ff_put_vc1_mspel_mc32_c; | |
459 dsp->put_vc1_mspel_pixels_tab[12] = ff_put_vc1_mspel_mc03_c; | |
460 dsp->put_vc1_mspel_pixels_tab[13] = ff_put_vc1_mspel_mc13_c; | |
461 dsp->put_vc1_mspel_pixels_tab[14] = ff_put_vc1_mspel_mc23_c; | |
462 dsp->put_vc1_mspel_pixels_tab[15] = ff_put_vc1_mspel_mc33_c; | |
463 } |