Parent Directory | Revision Log
Revision 1.30 - (view) (download)
1 : | edgomez | 1.16 | /***************************************************************************** |
2 : | edgomez | 1.17 | * |
3 : | * XVID MPEG-4 VIDEO CODEC | ||
4 : | * - Native API implementation - | ||
5 : | * | ||
6 : | * This program is an implementation of a part of one or more MPEG-4 | ||
7 : | * Video tools as specified in ISO/IEC 14496-2 standard. Those intending | ||
8 : | * to use this software module in hardware or software products are | ||
9 : | * advised that its use may infringe existing patents or copyrights, and | ||
10 : | * any such use would be at such party's own risk. The original | ||
11 : | * developer of this software module and his/her company, and subsequent | ||
12 : | * editors and their companies, will have no liability for use of this | ||
13 : | * software or modifications or derivatives thereof. | ||
14 : | * | ||
15 : | * This program is free software ; you can redistribute it and/or modify | ||
16 : | * it under the terms of the GNU General Public License as published by | ||
17 : | * the Free Software Foundation ; either version 2 of the License, or | ||
18 : | * (at your option) any later version. | ||
19 : | * | ||
20 : | * This program is distributed in the hope that it will be useful, | ||
21 : | * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
22 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
23 : | * GNU General Public License for more details. | ||
24 : | * | ||
25 : | * You should have received a copy of the GNU General Public License | ||
26 : | * along with this program ; if not, write to the Free Software | ||
27 : | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
28 : | * | ||
29 : | ****************************************************************************/ | ||
30 : | chenm001 | 1.29 | |
31 : | edgomez | 1.16 | /***************************************************************************** |
32 : | edgomez | 1.17 | * |
33 : | * History | ||
34 : | * | ||
35 : | suxen_drol | 1.21 | * - 23.06.2002 added XVID_CPU_CHKONLY |
36 : | edgomez | 1.17 | * - 17.03.2002 Added interpolate8x8_halfpel_hv_xmm |
37 : | * - 22.12.2001 API change: added xvid_init() - Isibaar | ||
38 : | * - 16.12.2001 inital version; (c)2001 peter ross <pross@cs.rmit.edu.au> | ||
39 : | * | ||
40 : | ia64p | 1.30 | * $Id: xvid.c,v 1.29 2002/07/09 01:48:49 chenm001 Exp $ |
41 : | edgomez | 1.17 | * |
42 : | ****************************************************************************/ | ||
43 : | Isibaar | 1.1 | |
44 : | #include "xvid.h" | ||
45 : | #include "decoder.h" | ||
46 : | #include "encoder.h" | ||
47 : | #include "bitstream/cbp.h" | ||
48 : | #include "dct/idct.h" | ||
49 : | #include "dct/fdct.h" | ||
50 : | #include "image/colorspace.h" | ||
51 : | #include "image/interpolate8x8.h" | ||
52 : | #include "utils/mem_transfer.h" | ||
53 : | #include "quant/quant_h263.h" | ||
54 : | #include "quant/quant_mpeg4.h" | ||
55 : | ia64p | 1.30 | #include "motion/motion.h" |
56 : | Isibaar | 1.1 | #include "motion/sad.h" |
57 : | #include "utils/emms.h" | ||
58 : | #include "utils/timer.h" | ||
59 : | Isibaar | 1.9 | #include "bitstream/mbcoding.h" |
60 : | Isibaar | 1.1 | |
61 : | edgomez | 1.16 | /***************************************************************************** |
62 : | * XviD Init Entry point | ||
63 : | * | ||
64 : | * Well this function initialize all internal function pointers according | ||
65 : | * to the CPU features forced by the library client or autodetected (depending | ||
66 : | * on the XVID_CPU_FORCE flag). It also initializes vlc coding tables and all | ||
67 : | * image colorspace transformation tables. | ||
68 : | * | ||
69 : | * Returned value : XVID_ERR_OK | ||
70 : | * + API_VERSION in the input XVID_INIT_PARAM structure | ||
71 : | * + core build " " " " " | ||
72 : | * | ||
73 : | ****************************************************************************/ | ||
74 : | |||
75 : | edgomez | 1.15 | int |
76 : | xvid_init(void *handle, | ||
77 : | int opt, | ||
78 : | void *param1, | ||
79 : | void *param2) | ||
80 : | Isibaar | 1.1 | { |
81 : | int cpu_flags; | ||
82 : | XVID_INIT_PARAM *init_param; | ||
83 : | |||
84 : | init_param = (XVID_INIT_PARAM *) param1; | ||
85 : | |||
86 : | suxen_drol | 1.21 | /* Inform the client the API version */ |
87 : | init_param->api_version = API_VERSION; | ||
88 : | |||
89 : | /* Inform the client the core build - unused because we're still alpha */ | ||
90 : | init_param->core_build = 1000; | ||
91 : | |||
92 : | if ((init_param->cpu_flags & XVID_CPU_CHKONLY)) | ||
93 : | { | ||
94 : | chenm001 | 1.29 | init_param->cpu_flags = check_cpu_features(); |
95 : | suxen_drol | 1.21 | return XVID_ERR_OK; |
96 : | } | ||
97 : | |||
98 : | edgomez | 1.16 | /* Do we have to force CPU features ? */ |
99 : | if ((init_param->cpu_flags & XVID_CPU_FORCE) > 0) { | ||
100 : | Isibaar | 1.1 | cpu_flags = init_param->cpu_flags; |
101 : | edgomez | 1.16 | } else { |
102 : | Isibaar | 1.1 | |
103 : | chenm001 | 1.29 | cpu_flags = check_cpu_features(); |
104 : | Isibaar | 1.1 | init_param->cpu_flags = cpu_flags; |
105 : | } | ||
106 : | |||
107 : | edgomez | 1.16 | /* Initialize the function pointers */ |
108 : | Isibaar | 1.1 | idct_int32_init(); |
109 : | Isibaar | 1.9 | init_vlc_tables(); |
110 : | |||
111 : | edgomez | 1.16 | /* Fixed Point Forward/Inverse DCT transformations */ |
112 : | Isibaar | 1.1 | fdct = fdct_int32; |
113 : | idct = idct_int32; | ||
114 : | |||
115 : | edgomez | 1.16 | /* Only needed on PPC Altivec archs */ |
116 : | canard | 1.10 | sadInit = 0; |
117 : | edgomez | 1.15 | |
118 : | edgomez | 1.16 | /* Restore FPU context : emms_c is a nop functions */ |
119 : | Isibaar | 1.1 | emms = emms_c; |
120 : | |||
121 : | edgomez | 1.16 | /* Quantization functions */ |
122 : | quant_intra = quant_intra_c; | ||
123 : | Isibaar | 1.1 | dequant_intra = dequant_intra_c; |
124 : | edgomez | 1.16 | quant_inter = quant_inter_c; |
125 : | Isibaar | 1.1 | dequant_inter = dequant_inter_c; |
126 : | |||
127 : | edgomez | 1.16 | quant4_intra = quant4_intra_c; |
128 : | Isibaar | 1.1 | dequant4_intra = dequant4_intra_c; |
129 : | edgomez | 1.16 | quant4_inter = quant4_inter_c; |
130 : | Isibaar | 1.1 | dequant4_inter = dequant4_inter_c; |
131 : | |||
132 : | edgomez | 1.16 | /* Block transfer related functions */ |
133 : | Isibaar | 1.1 | transfer_8to16copy = transfer_8to16copy_c; |
134 : | transfer_16to8copy = transfer_16to8copy_c; | ||
135 : | edgomez | 1.16 | transfer_8to16sub = transfer_8to16sub_c; |
136 : | suxen_drol | 1.11 | transfer_8to16sub2 = transfer_8to16sub2_c; |
137 : | edgomez | 1.16 | transfer_16to8add = transfer_16to8add_c; |
138 : | transfer8x8_copy = transfer8x8_copy_c; | ||
139 : | Isibaar | 1.1 | |
140 : | edgomez | 1.16 | /* Image interpolation related functions */ |
141 : | interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_c; | ||
142 : | interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_c; | ||
143 : | Isibaar | 1.1 | interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_c; |
144 : | |||
145 : | edgomez | 1.16 | /* Initialize internal colorspace transformation tables */ |
146 : | Isibaar | 1.1 | colorspace_init(); |
147 : | |||
148 : | edgomez | 1.16 | /* All colorspace transformation functions User Format->YV12 */ |
149 : | Isibaar | 1.1 | rgb555_to_yv12 = rgb555_to_yv12_c; |
150 : | rgb565_to_yv12 = rgb565_to_yv12_c; | ||
151 : | edgomez | 1.16 | rgb24_to_yv12 = rgb24_to_yv12_c; |
152 : | rgb32_to_yv12 = rgb32_to_yv12_c; | ||
153 : | yuv_to_yv12 = yuv_to_yv12_c; | ||
154 : | yuyv_to_yv12 = yuyv_to_yv12_c; | ||
155 : | uyvy_to_yv12 = uyvy_to_yv12_c; | ||
156 : | Isibaar | 1.1 | |
157 : | edgomez | 1.16 | /* All colorspace transformation functions YV12->User format */ |
158 : | Isibaar | 1.1 | yv12_to_rgb555 = yv12_to_rgb555_c; |
159 : | yv12_to_rgb565 = yv12_to_rgb565_c; | ||
160 : | edgomez | 1.16 | yv12_to_rgb24 = yv12_to_rgb24_c; |
161 : | yv12_to_rgb32 = yv12_to_rgb32_c; | ||
162 : | yv12_to_yuv = yv12_to_yuv_c; | ||
163 : | yv12_to_yuyv = yv12_to_yuyv_c; | ||
164 : | yv12_to_uyvy = yv12_to_uyvy_c; | ||
165 : | Isibaar | 1.1 | |
166 : | edgomez | 1.16 | /* Functions used in motion estimation algorithms */ |
167 : | Isibaar | 1.1 | calc_cbp = calc_cbp_c; |
168 : | edgomez | 1.16 | sad16 = sad16_c; |
169 : | sad16bi = sad16bi_c; | ||
170 : | sad8 = sad8_c; | ||
171 : | dev16 = dev16_c; | ||
172 : | ia64p | 1.30 | Halfpel8_Refine = Halfpel8_Refine_c; |
173 : | Isibaar | 1.1 | |
174 : | #ifdef ARCH_X86 | ||
175 : | edgomez | 1.15 | if ((cpu_flags & XVID_CPU_MMX) > 0) { |
176 : | edgomez | 1.16 | |
177 : | /* Forward and Inverse Discrete Cosine Transformation functions */ | ||
178 : | Isibaar | 1.1 | fdct = fdct_mmx; |
179 : | idct = idct_mmx; | ||
180 : | |||
181 : | edgomez | 1.16 | /* To restore FPU context after mmx use */ |
182 : | Isibaar | 1.1 | emms = emms_mmx; |
183 : | |||
184 : | edgomez | 1.16 | /* Quantization related functions */ |
185 : | quant_intra = quant_intra_mmx; | ||
186 : | Isibaar | 1.1 | dequant_intra = dequant_intra_mmx; |
187 : | edgomez | 1.16 | quant_inter = quant_inter_mmx; |
188 : | Isibaar | 1.1 | dequant_inter = dequant_inter_mmx; |
189 : | |||
190 : | edgomez | 1.16 | quant4_intra = quant4_intra_mmx; |
191 : | Isibaar | 1.1 | dequant4_intra = dequant4_intra_mmx; |
192 : | edgomez | 1.16 | quant4_inter = quant4_inter_mmx; |
193 : | Isibaar | 1.1 | dequant4_inter = dequant4_inter_mmx; |
194 : | |||
195 : | edgomez | 1.16 | /* Block related functions */ |
196 : | Isibaar | 1.1 | transfer_8to16copy = transfer_8to16copy_mmx; |
197 : | transfer_16to8copy = transfer_16to8copy_mmx; | ||
198 : | edgomez | 1.16 | transfer_8to16sub = transfer_8to16sub_mmx; |
199 : | edgomez | 1.22 | transfer_8to16sub2 = transfer_8to16sub2_mmx; |
200 : | edgomez | 1.16 | transfer_16to8add = transfer_16to8add_mmx; |
201 : | transfer8x8_copy = transfer8x8_copy_mmx; | ||
202 : | edgomez | 1.22 | |
203 : | edgomez | 1.16 | |
204 : | /* Image Interpolation related functions */ | ||
205 : | interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_mmx; | ||
206 : | interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_mmx; | ||
207 : | Isibaar | 1.1 | interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx; |
208 : | |||
209 : | edgomez | 1.16 | /* Image RGB->YV12 related functions */ |
210 : | Isibaar | 1.1 | rgb24_to_yv12 = rgb24_to_yv12_mmx; |
211 : | rgb32_to_yv12 = rgb32_to_yv12_mmx; | ||
212 : | edgomez | 1.16 | yuv_to_yv12 = yuv_to_yv12_mmx; |
213 : | yuyv_to_yv12 = yuyv_to_yv12_mmx; | ||
214 : | uyvy_to_yv12 = uyvy_to_yv12_mmx; | ||
215 : | Isibaar | 1.1 | |
216 : | edgomez | 1.16 | /* Image YV12->RGB related functions */ |
217 : | Isibaar | 1.1 | yv12_to_rgb24 = yv12_to_rgb24_mmx; |
218 : | yv12_to_rgb32 = yv12_to_rgb32_mmx; | ||
219 : | edgomez | 1.16 | yv12_to_yuyv = yv12_to_yuyv_mmx; |
220 : | yv12_to_uyvy = yv12_to_uyvy_mmx; | ||
221 : | Isibaar | 1.1 | |
222 : | edgomez | 1.16 | /* Motion estimation related functions */ |
223 : | Isibaar | 1.1 | calc_cbp = calc_cbp_mmx; |
224 : | edgomez | 1.16 | sad16 = sad16_mmx; |
225 : | sad8 = sad8_mmx; | ||
226 : | dev16 = dev16_mmx; | ||
227 : | Isibaar | 1.1 | |
228 : | } | ||
229 : | |||
230 : | edgomez | 1.15 | if ((cpu_flags & XVID_CPU_MMXEXT) > 0) { |
231 : | edgomez | 1.16 | |
232 : | /* Inverse DCT */ | ||
233 : | Isibaar | 1.1 | idct = idct_xmm; |
234 : | edgomez | 1.16 | |
235 : | /* Interpolation */ | ||
236 : | interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_xmm; | ||
237 : | interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_xmm; | ||
238 : | h | 1.3 | interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm; |
239 : | Isibaar | 1.25 | |
240 : | chenm001 | 1.29 | /* Quantization */ |
241 : | dequant_intra = dequant_intra_xmm; | ||
242 : | dequant_inter = dequant_inter_xmm; | ||
243 : | |||
244 : | edgomez | 1.19 | /* Buffer transfer */ |
245 : | transfer_8to16sub2 = transfer_8to16sub2_xmm; | ||
246 : | edgomez | 1.16 | |
247 : | /* Colorspace transformation */ | ||
248 : | Isibaar | 1.1 | yuv_to_yv12 = yuv_to_yv12_xmm; |
249 : | |||
250 : | edgomez | 1.16 | /* ME functions */ |
251 : | Isibaar | 1.1 | sad16 = sad16_xmm; |
252 : | chenm001 | 1.29 | sad16bi = sad16bi_xmm; |
253 : | edgomez | 1.16 | sad8 = sad8_xmm; |
254 : | Isibaar | 1.1 | dev16 = dev16_xmm; |
255 : | |||
256 : | } | ||
257 : | |||
258 : | edgomez | 1.15 | if ((cpu_flags & XVID_CPU_3DNOW) > 0) { |
259 : | edgomez | 1.16 | |
260 : | /* Interpolation */ | ||
261 : | Isibaar | 1.1 | interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn; |
262 : | interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dn; | ||
263 : | h | 1.4 | interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn; |
264 : | Isibaar | 1.1 | } |
265 : | |||
266 : | edgomez | 1.15 | if ((cpu_flags & XVID_CPU_SSE2) > 0) { |
267 : | Isibaar | 1.14 | #ifdef EXPERIMENTAL_SSE2_CODE |
268 : | edgomez | 1.16 | |
269 : | chenm001 | 1.29 | calc_cbp = calc_cbp_sse2; |
270 : | |||
271 : | edgomez | 1.16 | /* Quantization */ |
272 : | quant_intra = quant_intra_sse2; | ||
273 : | Isibaar | 1.14 | dequant_intra = dequant_intra_sse2; |
274 : | edgomez | 1.16 | quant_inter = quant_inter_sse2; |
275 : | Isibaar | 1.14 | dequant_inter = dequant_inter_sse2; |
276 : | h | 1.13 | |
277 : | edgomez | 1.16 | /* ME */ |
278 : | sad16 = sad16_sse2; | ||
279 : | dev16 = dev16_sse2; | ||
280 : | |||
281 : | /* Forward and Inverse DCT */ | ||
282 : | idct = idct_sse2; | ||
283 : | Isibaar | 1.14 | fdct = fdct_sse2; |
284 : | #endif | ||
285 : | h | 1.12 | } |
286 : | edgomez | 1.16 | |
287 : | Isibaar | 1.1 | #endif |
288 : | Isibaar | 1.18 | |
289 : | #ifdef ARCH_IA64 | ||
290 : | if ((cpu_flags & XVID_CPU_IA64) > 0) { //use assembler routines? | ||
291 : | idct_ia64_init(); | ||
292 : | fdct = fdct_ia64; | ||
293 : | idct = idct_ia64; //not yet working, crashes | ||
294 : | interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_ia64; | ||
295 : | interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_ia64; | ||
296 : | interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_ia64; | ||
297 : | sad16 = sad16_ia64; | ||
298 : | sad16bi = sad16bi_ia64; | ||
299 : | sad8 = sad8_ia64; | ||
300 : | dev16 = dev16_ia64; | ||
301 : | ia64p | 1.30 | Halfpel8_Refine = Halfpel8_Refine_ia64; |
302 : | Isibaar | 1.18 | quant_intra = quant_intra_ia64; |
303 : | dequant_intra = dequant_intra_ia64; | ||
304 : | quant_inter = quant_inter_ia64; | ||
305 : | dequant_inter = dequant_inter_ia64; | ||
306 : | transfer_8to16copy = transfer_8to16copy_ia64; | ||
307 : | transfer_16to8copy = transfer_16to8copy_ia64; | ||
308 : | transfer_8to16sub = transfer_8to16sub_ia64; | ||
309 : | transfer_8to16sub2 = transfer_8to16sub2_ia64; | ||
310 : | transfer_16to8add = transfer_16to8add_ia64; | ||
311 : | transfer8x8_copy = transfer8x8_copy_ia64; | ||
312 : | DEBUG("Using IA-64 assembler routines.\n"); | ||
313 : | } | ||
314 : | #endif | ||
315 : | edgomez | 1.16 | |
316 : | canard | 1.5 | #ifdef ARCH_PPC |
317 : | canard | 1.6 | #ifdef ARCH_PPC_ALTIVEC |
318 : | calc_cbp = calc_cbp_altivec; | ||
319 : | canard | 1.7 | fdct = fdct_altivec; |
320 : | idct = idct_altivec; | ||
321 : | canard | 1.10 | sadInit = sadInit_altivec; |
322 : | canard | 1.8 | sad16 = sad16_altivec; |
323 : | sad8 = sad8_altivec; | ||
324 : | dev16 = dev16_altivec; | ||
325 : | canard | 1.6 | #else |
326 : | canard | 1.5 | calc_cbp = calc_cbp_ppc; |
327 : | canard | 1.6 | #endif |
328 : | canard | 1.5 | #endif |
329 : | Isibaar | 1.1 | |
330 : | return XVID_ERR_OK; | ||
331 : | } | ||
332 : | |||
333 : | edgomez | 1.16 | /***************************************************************************** |
334 : | * XviD Native decoder entry point | ||
335 : | * | ||
336 : | * This function is just a wrapper to all the option cases. | ||
337 : | * | ||
338 : | * Returned values : XVID_ERR_FAIL when opt is invalid | ||
339 : | * else returns the wrapped function result | ||
340 : | * | ||
341 : | ****************************************************************************/ | ||
342 : | |||
343 : | edgomez | 1.15 | int |
344 : | xvid_decore(void *handle, | ||
345 : | int opt, | ||
346 : | void *param1, | ||
347 : | void *param2) | ||
348 : | Isibaar | 1.1 | { |
349 : | edgomez | 1.15 | switch (opt) { |
350 : | case XVID_DEC_DECODE: | ||
351 : | return decoder_decode((DECODER *) handle, (XVID_DEC_FRAME *) param1); | ||
352 : | |||
353 : | case XVID_DEC_CREATE: | ||
354 : | chenm001 | 1.29 | return decoder_create((XVID_DEC_PARAM *) param1); |
355 : | edgomez | 1.15 | |
356 : | case XVID_DEC_DESTROY: | ||
357 : | return decoder_destroy((DECODER *) handle); | ||
358 : | Isibaar | 1.1 | |
359 : | default: | ||
360 : | edgomez | 1.15 | return XVID_ERR_FAIL; |
361 : | } | ||
362 : | Isibaar | 1.1 | } |
363 : | |||
364 : | edgomez | 1.16 | |
365 : | /***************************************************************************** | ||
366 : | * XviD Native encoder entry point | ||
367 : | * | ||
368 : | * This function is just a wrapper to all the option cases. | ||
369 : | * | ||
370 : | * Returned values : XVID_ERR_FAIL when opt is invalid | ||
371 : | * else returns the wrapped function result | ||
372 : | * | ||
373 : | ****************************************************************************/ | ||
374 : | Isibaar | 1.1 | |
375 : | edgomez | 1.15 | int |
376 : | xvid_encore(void *handle, | ||
377 : | int opt, | ||
378 : | void *param1, | ||
379 : | void *param2) | ||
380 : | Isibaar | 1.1 | { |
381 : | edgomez | 1.15 | switch (opt) { |
382 : | case XVID_ENC_ENCODE: | ||
383 : | suxen_drol | 1.20 | #ifdef BFRAMES |
384 : | if (((Encoder *) handle)->mbParam.max_bframes >= 0) | ||
385 : | return encoder_encode_bframes((Encoder *) handle, (XVID_ENC_FRAME *) param1, | ||
386 : | (XVID_ENC_STATS *) param2); | ||
387 : | else | ||
388 : | #endif | ||
389 : | edgomez | 1.15 | return encoder_encode((Encoder *) handle, (XVID_ENC_FRAME *) param1, |
390 : | (XVID_ENC_STATS *) param2); | ||
391 : | |||
392 : | case XVID_ENC_CREATE: | ||
393 : | return encoder_create((XVID_ENC_PARAM *) param1); | ||
394 : | |||
395 : | case XVID_ENC_DESTROY: | ||
396 : | return encoder_destroy((Encoder *) handle); | ||
397 : | Isibaar | 1.1 | |
398 : | default: | ||
399 : | edgomez | 1.15 | return XVID_ERR_FAIL; |
400 : | } | ||
401 : | Isibaar | 1.1 | } |
No admin address has been configured | ViewVC Help |
Powered by ViewVC 1.0.4 |