projects / mesa.git / blob
? search:


32374abb493af851953d932c319effdbaeccd53a
[mesa.git] / src / gallium / auxiliary / vl / vl_mpeg12_mc_renderer.c
1 /**************************************************************************
2 *
3 * Copyright 2009 Younes Manton.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28 #include "vl_mpeg12_mc_renderer.h"
29 #include <assert.h>
30 #include <pipe/p_context.h>
31 #include <pipe/p_inlines.h>
32 #include <util/u_format.h>
33 #include <util/u_math.h>
34 #include <util/u_memory.h>
35 #include <tgsi/tgsi_parse.h>
36 #include <tgsi/tgsi_build.h>
37 #include "vl_shader_build.h"
38
39 #define DEFAULT_BUF_ALIGNMENT 1
40 #define MACROBLOCK_WIDTH 16
41 #define MACROBLOCK_HEIGHT 16
42 #define BLOCK_WIDTH 8
43 #define BLOCK_HEIGHT 8
44 #define ZERO_BLOCK_NIL -1.0f
45 #define ZERO_BLOCK_IS_NIL(zb) ((zb).x < 0.0f)
46
47 struct vertex2f
48 {
49 float x, y;
50 };
51
52 struct vertex4f
53 {
54 float x, y, z, w;
55 };
56
57 struct vertex_shader_consts
58 {
59 struct vertex4f denorm;
60 };
61
62 struct fragment_shader_consts
63 {
64 struct vertex4f multiplier;
65 struct vertex4f div;
66 };
67
68 /*
69 * Muliplier renormalizes block samples from 16 bits to 12 bits.
70 * Divider is used when calculating Y % 2 for choosing top or bottom
71 * field for P or B macroblocks.
72 * TODO: Use immediates.
73 */
74 static const struct fragment_shader_consts fs_consts = {
75 {32767.0f / 255.0f, 32767.0f / 255.0f, 32767.0f / 255.0f, 0.0f},
76 {0.5f, 2.0f, 0.0f, 0.0f}
77 };
78
79 struct vert_stream_0
80 {
81 struct vertex2f pos;
82 struct vertex2f luma_tc;
83 struct vertex2f cb_tc;
84 struct vertex2f cr_tc;
85 };
86
87 enum MACROBLOCK_TYPE
88 {
89 MACROBLOCK_TYPE_INTRA,
90 MACROBLOCK_TYPE_FWD_FRAME_PRED,
91 MACROBLOCK_TYPE_FWD_FIELD_PRED,
92 MACROBLOCK_TYPE_BKWD_FRAME_PRED,
93 MACROBLOCK_TYPE_BKWD_FIELD_PRED,
94 MACROBLOCK_TYPE_BI_FRAME_PRED,
95 MACROBLOCK_TYPE_BI_FIELD_PRED,
96
97 NUM_MACROBLOCK_TYPES
98 };
99
100 static void
101 create_intra_vert_shader(struct vl_mpeg12_mc_renderer *r)
102 {
103 const unsigned max_tokens = 50;
104
105 struct pipe_shader_state vs;
106 struct tgsi_token *tokens;
107 struct tgsi_header *header;
108
109 struct tgsi_full_declaration decl;
110 struct tgsi_full_instruction inst;
111
112 unsigned ti;
113
114 unsigned i;
115
116 assert(r);
117
118 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
119 header = (struct tgsi_header *) &tokens[0];
120 *header = tgsi_build_header();
121 *(struct tgsi_processor *) &tokens[1] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
122
123 ti = 2;
124
125 /*
126 * decl i0 ; Vertex pos
127 * decl i1 ; Luma texcoords
128 * decl i2 ; Chroma Cb texcoords
129 * decl i3 ; Chroma Cr texcoords
130 */
131 for (i = 0; i < 4; i++) {
132 decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
133 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
134 }
135
136 /*
137 * decl o0 ; Vertex pos
138 * decl o1 ; Luma texcoords
139 * decl o2 ; Chroma Cb texcoords
140 * decl o3 ; Chroma Cr texcoords
141 */
142 for (i = 0; i < 4; i++) {
143 decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
144 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
145 }
146
147 /*
148 * mov o0, i0 ; Move input vertex pos to output
149 * mov o1, i1 ; Move input luma texcoords to output
150 * mov o2, i2 ; Move input chroma Cb texcoords to output
151 * mov o3, i3 ; Move input chroma Cr texcoords to output
152 */
153 for (i = 0; i < 4; ++i) {
154 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
155 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
156 }
157
158 /* end */
159 inst = vl_end();
160 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
161
162 assert(ti <= max_tokens);
163
164 vs.tokens = tokens;
165 r->i_vs = r->pipe->create_vs_state(r->pipe, &vs);
166 free(tokens);
167 }
168
169 static void
170 create_intra_frag_shader(struct vl_mpeg12_mc_renderer *r)
171 {
172 const unsigned max_tokens = 100;
173
174 struct pipe_shader_state fs;
175 struct tgsi_token *tokens;
176 struct tgsi_header *header;
177
178 struct tgsi_full_declaration decl;
179 struct tgsi_full_instruction inst;
180
181 unsigned ti;
182
183 unsigned i;
184
185 assert(r);
186
187 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
188 header = (struct tgsi_header *) &tokens[0];
189 *header = tgsi_build_header();
190 *(struct tgsi_processor *) &tokens[1] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
191
192 ti = 2;
193
194 /*
195 * decl i0 ; Luma texcoords
196 * decl i1 ; Chroma Cb texcoords
197 * decl i2 ; Chroma Cr texcoords
198 */
199 for (i = 0; i < 3; ++i) {
200 decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
201 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
202 }
203
204 /* decl c0 ; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
205 decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
206 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
207
208 /* decl o0 ; Fragment color */
209 decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
210 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
211
212 /* decl t0, t1 */
213 decl = vl_decl_temps(0, 1);
214 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
215
216 /*
217 * decl s0 ; Sampler for luma texture
218 * decl s1 ; Sampler for chroma Cb texture
219 * decl s2 ; Sampler for chroma Cr texture
220 */
221 for (i = 0; i < 3; ++i) {
222 decl = vl_decl_samplers(i, i);
223 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
224 }
225
226 /*
227 * tex2d t1, i0, s0 ; Read texel from luma texture
228 * mov t0.x, t1.x ; Move luma sample into .x component
229 * tex2d t1, i1, s1 ; Read texel from chroma Cb texture
230 * mov t0.y, t1.x ; Move Cb sample into .y component
231 * tex2d t1, i2, s2 ; Read texel from chroma Cr texture
232 * mov t0.z, t1.x ; Move Cr sample into .z component
233 */
234 for (i = 0; i < 3; ++i) {
235 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
236 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
237
238 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
239 inst.Src[0].Register.SwizzleX = TGSI_SWIZZLE_X;
240 inst.Src[0].Register.SwizzleY = TGSI_SWIZZLE_X;
241 inst.Src[0].Register.SwizzleZ = TGSI_SWIZZLE_X;
242 inst.Dst[0].Register.WriteMask = TGSI_WRITEMASK_X << i;
243 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
244 }
245
246 /* mul o0, t0, c0 ; Rescale texel to correct range */
247 inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
248 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
249
250 /* end */
251 inst = vl_end();
252 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
253
254 assert(ti <= max_tokens);
255
256 fs.tokens = tokens;
257 r->i_fs = r->pipe->create_fs_state(r->pipe, &fs);
258 free(tokens);
259 }
260
261 static void
262 create_frame_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
263 {
264 const unsigned max_tokens = 100;
265
266 struct pipe_shader_state vs;
267 struct tgsi_token *tokens;
268 struct tgsi_header *header;
269
270 struct tgsi_full_declaration decl;
271 struct tgsi_full_instruction inst;
272
273 unsigned ti;
274
275 unsigned i;
276
277 assert(r);
278
279 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
280 header = (struct tgsi_header *) &tokens[0];
281 *header = tgsi_build_header();
282 *(struct tgsi_processor *) &tokens[1] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
283
284 ti = 2;
285
286 /*
287 * decl i0 ; Vertex pos
288 * decl i1 ; Luma texcoords
289 * decl i2 ; Chroma Cb texcoords
290 * decl i3 ; Chroma Cr texcoords
291 * decl i4 ; Ref surface top field texcoords
292 * decl i5 ; Ref surface bottom field texcoords (unused, packed in the same stream)
293 */
294 for (i = 0; i < 6; i++) {
295 decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
296 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
297 }
298
299 /*
300 * decl o0 ; Vertex pos
301 * decl o1 ; Luma texcoords
302 * decl o2 ; Chroma Cb texcoords
303 * decl o3 ; Chroma Cr texcoords
304 * decl o4 ; Ref macroblock texcoords
305 */
306 for (i = 0; i < 5; i++) {
307 decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
308 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
309 }
310
311 /*
312 * mov o0, i0 ; Move input vertex pos to output
313 * mov o1, i1 ; Move input luma texcoords to output
314 * mov o2, i2 ; Move input chroma Cb texcoords to output
315 * mov o3, i3 ; Move input chroma Cr texcoords to output
316 */
317 for (i = 0; i < 4; ++i) {
318 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
319 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
320 }
321
322 /* add o4, i0, i4 ; Translate vertex pos by motion vec to form ref macroblock texcoords */
323 inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, 4);
324 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
325
326 /* end */
327 inst = vl_end();
328 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
329
330 assert(ti <= max_tokens);
331
332 vs.tokens = tokens;
333 r->p_vs[0] = r->pipe->create_vs_state(r->pipe, &vs);
334 free(tokens);
335 }
336
337 static void
338 create_field_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
339 {
340 assert(false);
341 }
342
343 static void
344 create_frame_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
345 {
346 const unsigned max_tokens = 100;
347
348 struct pipe_shader_state fs;
349 struct tgsi_token *tokens;
350 struct tgsi_header *header;
351
352 struct tgsi_full_declaration decl;
353 struct tgsi_full_instruction inst;
354
355 unsigned ti;
356
357 unsigned i;
358
359 assert(r);
360
361 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
362 header = (struct tgsi_header *) &tokens[0];
363 *header = tgsi_build_header();
364 *(struct tgsi_processor *) &tokens[1] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
365
366 ti = 2;
367
368 /*
369 * decl i0 ; Luma texcoords
370 * decl i1 ; Chroma Cb texcoords
371 * decl i2 ; Chroma Cr texcoords
372 * decl i3 ; Ref macroblock texcoords
373 */
374 for (i = 0; i < 4; ++i) {
375 decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
376 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
377 }
378
379 /* decl c0 ; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
380 decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
381 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
382
383 /* decl o0 ; Fragment color */
384 decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
385 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
386
387 /* decl t0, t1 */
388 decl = vl_decl_temps(0, 1);
389 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
390
391 /*
392 * decl s0 ; Sampler for luma texture
393 * decl s1 ; Sampler for chroma Cb texture
394 * decl s2 ; Sampler for chroma Cr texture
395 * decl s3 ; Sampler for ref surface texture
396 */
397 for (i = 0; i < 4; ++i) {
398 decl = vl_decl_samplers(i, i);
399 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
400 }
401
402 /*
403 * tex2d t1, i0, s0 ; Read texel from luma texture
404 * mov t0.x, t1.x ; Move luma sample into .x component
405 * tex2d t1, i1, s1 ; Read texel from chroma Cb texture
406 * mov t0.y, t1.x ; Move Cb sample into .y component
407 * tex2d t1, i2, s2 ; Read texel from chroma Cr texture
408 * mov t0.z, t1.x ; Move Cr sample into .z component
409 */
410 for (i = 0; i < 3; ++i) {
411 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
412 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
413
414 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
415 inst.Src[0].Register.SwizzleX = TGSI_SWIZZLE_X;
416 inst.Src[0].Register.SwizzleY = TGSI_SWIZZLE_X;
417 inst.Src[0].Register.SwizzleZ = TGSI_SWIZZLE_X;
418 inst.Dst[0].Register.WriteMask = TGSI_WRITEMASK_X << i;
419 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
420 }
421
422 /* mul t0, t0, c0 ; Rescale texel to correct range */
423 inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
424 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
425
426 /* tex2d t1, i3, s3 ; Read texel from ref macroblock */
427 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, 3, TGSI_FILE_SAMPLER, 3);
428 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
429
430 /* add o0, t0, t1 ; Add ref and differential to form final output */
431 inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
432 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
433
434 /* end */
435 inst = vl_end();
436 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
437
438 assert(ti <= max_tokens);
439
440 fs.tokens = tokens;
441 r->p_fs[0] = r->pipe->create_fs_state(r->pipe, &fs);
442 free(tokens);
443 }
444
445 static void
446 create_field_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
447 {
448 assert(false);
449 }
450
451 static void
452 create_frame_bi_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
453 {
454 const unsigned max_tokens = 100;
455
456 struct pipe_shader_state vs;
457 struct tgsi_token *tokens;
458 struct tgsi_header *header;
459
460 struct tgsi_full_declaration decl;
461 struct tgsi_full_instruction inst;
462
463 unsigned ti;
464
465 unsigned i;
466
467 assert(r);
468
469 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
470 header = (struct tgsi_header *) &tokens[0];
471 *header = tgsi_build_header();
472 *(struct tgsi_processor *) &tokens[1] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
473
474 ti = 2;
475
476 /*
477 * decl i0 ; Vertex pos
478 * decl i1 ; Luma texcoords
479 * decl i2 ; Chroma Cb texcoords
480 * decl i3 ; Chroma Cr texcoords
481 * decl i4 ; First ref macroblock top field texcoords
482 * decl i5 ; First ref macroblock bottom field texcoords (unused, packed in the same stream)
483 * decl i6 ; Second ref macroblock top field texcoords
484 * decl i7 ; Second ref macroblock bottom field texcoords (unused, packed in the same stream)
485 */
486 for (i = 0; i < 8; i++) {
487 decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
488 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
489 }
490
491 /*
492 * decl o0 ; Vertex pos
493 * decl o1 ; Luma texcoords
494 * decl o2 ; Chroma Cb texcoords
495 * decl o3 ; Chroma Cr texcoords
496 * decl o4 ; First ref macroblock texcoords
497 * decl o5 ; Second ref macroblock texcoords
498 */
499 for (i = 0; i < 6; i++) {
500 decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
501 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
502 }
503
504 /*
505 * mov o0, i0 ; Move input vertex pos to output
506 * mov o1, i1 ; Move input luma texcoords to output
507 * mov o2, i2 ; Move input chroma Cb texcoords to output
508 * mov o3, i3 ; Move input chroma Cr texcoords to output
509 */
510 for (i = 0; i < 4; ++i) {
511 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
512 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
513 }
514
515 /*
516 * add o4, i0, i4 ; Translate vertex pos by motion vec to form first ref macroblock texcoords
517 * add o5, i0, i6 ; Translate vertex pos by motion vec to form second ref macroblock texcoords
518 */
519 for (i = 0; i < 2; ++i) {
520 inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, (i + 2) * 2);
521 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
522 }
523
524 /* end */
525 inst = vl_end();
526 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
527
528 assert(ti <= max_tokens);
529
530 vs.tokens = tokens;
531 r->b_vs[0] = r->pipe->create_vs_state(r->pipe, &vs);
532 free(tokens);
533 }
534
535 static void
536 create_field_bi_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
537 {
538 assert(false);
539 }
540
541 static void
542 create_frame_bi_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
543 {
544 const unsigned max_tokens = 100;
545
546 struct pipe_shader_state fs;
547 struct tgsi_token *tokens;
548 struct tgsi_header *header;
549
550 struct tgsi_full_declaration decl;
551 struct tgsi_full_instruction inst;
552
553 unsigned ti;
554
555 unsigned i;
556
557 assert(r);
558
559 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
560 header = (struct tgsi_header *) &tokens[0];
561 *header = tgsi_build_header();
562 *(struct tgsi_processor *) &tokens[1] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
563
564 ti = 2;
565
566 /*
567 * decl i0 ; Luma texcoords
568 * decl i1 ; Chroma Cb texcoords
569 * decl i2 ; Chroma Cr texcoords
570 * decl i3 ; First ref macroblock texcoords
571 * decl i4 ; Second ref macroblock texcoords
572 */
573 for (i = 0; i < 5; ++i) {
574 decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
575 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
576 }
577
578 /*
579 * decl c0 ; Scaling factor, rescales 16-bit snorm to 9-bit snorm
580 * decl c1 ; Constant 1/2 in .x channel to use as weight to blend past and future texels
581 */
582 decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
583 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
584
585 /* decl o0 ; Fragment color */
586 decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
587 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
588
589 /* decl t0-t2 */
590 decl = vl_decl_temps(0, 2);
591 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
592
593 /*
594 * decl s0 ; Sampler for luma texture
595 * decl s1 ; Sampler for chroma Cb texture
596 * decl s2 ; Sampler for chroma Cr texture
597 * decl s3 ; Sampler for first ref surface texture
598 * decl s4 ; Sampler for second ref surface texture
599 */
600 for (i = 0; i < 5; ++i) {
601 decl = vl_decl_samplers(i, i);
602 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
603 }
604
605 /*
606 * tex2d t1, i0, s0 ; Read texel from luma texture
607 * mov t0.x, t1.x ; Move luma sample into .x component
608 * tex2d t1, i1, s1 ; Read texel from chroma Cb texture
609 * mov t0.y, t1.x ; Move Cb sample into .y component
610 * tex2d t1, i2, s2 ; Read texel from chroma Cr texture
611 * mov t0.z, t1.x ; Move Cr sample into .z component
612 */
613 for (i = 0; i < 3; ++i) {
614 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
615 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
616
617 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
618 inst.Src[0].Register.SwizzleX = TGSI_SWIZZLE_X;
619 inst.Src[0].Register.SwizzleY = TGSI_SWIZZLE_X;
620 inst.Src[0].Register.SwizzleZ = TGSI_SWIZZLE_X;
621 inst.Dst[0].Register.WriteMask = TGSI_WRITEMASK_X << i;
622 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
623 }
624
625 /* mul t0, t0, c0 ; Rescale texel to correct range */
626 inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
627 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
628
629 /*
630 * tex2d t1, i3, s3 ; Read texel from first ref macroblock
631 * tex2d t2, i4, s4 ; Read texel from second ref macroblock
632 */
633 for (i = 0; i < 2; ++i) {
634 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 3, TGSI_FILE_SAMPLER, i + 3);
635 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
636 }
637
638 /* lerp t1, c1.x, t1, t2 ; Blend past and future texels */
639 inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
640 inst.Src[0].Register.SwizzleX = TGSI_SWIZZLE_X;
641 inst.Src[0].Register.SwizzleY = TGSI_SWIZZLE_X;
642 inst.Src[0].Register.SwizzleZ = TGSI_SWIZZLE_X;
643 inst.Src[0].Register.SwizzleW = TGSI_SWIZZLE_X;
644 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
645
646 /* add o0, t0, t1 ; Add past/future ref and differential to form final output */
647 inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
648 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
649
650 /* end */
651 inst = vl_end();
652 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
653
654 assert(ti <= max_tokens);
655
656 fs.tokens = tokens;
657 r->b_fs[0] = r->pipe->create_fs_state(r->pipe, &fs);
658 free(tokens);
659 }
660
661 static void
662 create_field_bi_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
663 {
664 assert(false);
665 }
666
667 static void
668 xfer_buffers_map(struct vl_mpeg12_mc_renderer *r)
669 {
670 unsigned i;
671
672 assert(r);
673
674 for (i = 0; i < 3; ++i) {
675 r->tex_transfer[i] = r->pipe->screen->get_tex_transfer
676 (
677 r->pipe->screen, r->textures.all[i],
678 0, 0, 0, PIPE_TRANSFER_WRITE, 0, 0,
679 r->textures.all[i]->width0, r->textures.all[i]->height0
680 );
681
682 r->texels[i] = r->pipe->screen->transfer_map(r->pipe->screen, r->tex_transfer[i]);
683 }
684 }
685
686 static void
687 xfer_buffers_unmap(struct vl_mpeg12_mc_renderer *r)
688 {
689 unsigned i;
690
691 assert(r);
692
693 for (i = 0; i < 3; ++i) {
694 r->pipe->screen->transfer_unmap(r->pipe->screen, r->tex_transfer[i]);
695 r->pipe->screen->tex_transfer_destroy(r->tex_transfer[i]);
696 }
697 }
698
699 static bool
700 init_pipe_state(struct vl_mpeg12_mc_renderer *r)
701 {
702 struct pipe_sampler_state sampler;
703 unsigned filters[5];
704 unsigned i;
705
706 assert(r);
707
708 r->viewport.scale[0] = r->pot_buffers ?
709 util_next_power_of_two(r->picture_width) : r->picture_width;
710 r->viewport.scale[1] = r->pot_buffers ?
711 util_next_power_of_two(r->picture_height) : r->picture_height;
712 r->viewport.scale[2] = 1;
713 r->viewport.scale[3] = 1;
714 r->viewport.translate[0] = 0;
715 r->viewport.translate[1] = 0;
716 r->viewport.translate[2] = 0;
717 r->viewport.translate[3] = 0;
718
719 r->scissor.maxx = r->pot_buffers ?
720 util_next_power_of_two(r->picture_width) : r->picture_width;
721 r->scissor.maxy = r->pot_buffers ?
722 util_next_power_of_two(r->picture_height) : r->picture_height;
723
724 r->fb_state.width = r->pot_buffers ?
725 util_next_power_of_two(r->picture_width) : r->picture_width;
726 r->fb_state.height = r->pot_buffers ?
727 util_next_power_of_two(r->picture_height) : r->picture_height;
728 r->fb_state.nr_cbufs = 1;
729 r->fb_state.zsbuf = NULL;
730
731 /* Luma filter */
732 filters[0] = PIPE_TEX_FILTER_NEAREST;
733 /* Chroma filters */
734 if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_444 ||
735 r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE) {
736 filters[1] = PIPE_TEX_FILTER_NEAREST;
737 filters[2] = PIPE_TEX_FILTER_NEAREST;
738 }
739 else {
740 filters[1] = PIPE_TEX_FILTER_LINEAR;
741 filters[2] = PIPE_TEX_FILTER_LINEAR;
742 }
743 /* Fwd, bkwd ref filters */
744 filters[3] = PIPE_TEX_FILTER_LINEAR;
745 filters[4] = PIPE_TEX_FILTER_LINEAR;
746
747 for (i = 0; i < 5; ++i) {
748 sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
749 sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
750 sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
751 sampler.min_img_filter = filters[i];
752 sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
753 sampler.mag_img_filter = filters[i];
754 sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
755 sampler.compare_func = PIPE_FUNC_ALWAYS;
756 sampler.normalized_coords = 1;
757 /*sampler.prefilter = ; */
758 /*sampler.shadow_ambient = ; */
759 /*sampler.lod_bias = ; */
760 sampler.min_lod = 0;
761 /*sampler.max_lod = ; */
762 /*sampler.border_color[i] = ; */
763 /*sampler.max_anisotropy = ; */
764 r->samplers.all[i] = r->pipe->create_sampler_state(r->pipe, &sampler);
765 }
766
767 return true;
768 }
769
770 static void
771 cleanup_pipe_state(struct vl_mpeg12_mc_renderer *r)
772 {
773 unsigned i;
774
775 assert(r);
776
777 for (i = 0; i < 5; ++i)
778 r->pipe->delete_sampler_state(r->pipe, r->samplers.all[i]);
779 }
780
781 static bool
782 init_shaders(struct vl_mpeg12_mc_renderer *r)
783 {
784 assert(r);
785
786 create_intra_vert_shader(r);
787 create_intra_frag_shader(r);
788 create_frame_pred_vert_shader(r);
789 create_frame_pred_frag_shader(r);
790 create_frame_bi_pred_vert_shader(r);
791 create_frame_bi_pred_frag_shader(r);
792
793 return true;
794 }
795
796 static void
797 cleanup_shaders(struct vl_mpeg12_mc_renderer *r)
798 {
799 assert(r);
800
801 r->pipe->delete_vs_state(r->pipe, r->i_vs);
802 r->pipe->delete_fs_state(r->pipe, r->i_fs);
803 r->pipe->delete_vs_state(r->pipe, r->p_vs[0]);
804 r->pipe->delete_fs_state(r->pipe, r->p_fs[0]);
805 r->pipe->delete_vs_state(r->pipe, r->b_vs[0]);
806 r->pipe->delete_fs_state(r->pipe, r->b_fs[0]);
807 }
808
809 static bool
810 init_buffers(struct vl_mpeg12_mc_renderer *r)
811 {
812 struct pipe_texture template;
813
814 const unsigned mbw =
815 align(r->picture_width, MACROBLOCK_WIDTH) / MACROBLOCK_WIDTH;
816 const unsigned mbh =
817 align(r->picture_height, MACROBLOCK_HEIGHT) / MACROBLOCK_HEIGHT;
818
819 unsigned i;
820
821 assert(r);
822
823 r->macroblocks_per_batch =
824 mbw * (r->bufmode == VL_MPEG12_MC_RENDERER_BUFFER_PICTURE ? mbh : 1);
825 r->num_macroblocks = 0;
826 r->macroblock_buf = MALLOC(r->macroblocks_per_batch * sizeof(struct pipe_mpeg12_macroblock));
827
828 memset(&template, 0, sizeof(struct pipe_texture));
829 template.target = PIPE_TEXTURE_2D;
830 /* TODO: Accomodate HW that can't do this and also for cases when this isn't precise enough */
831 template.format = PIPE_FORMAT_R16_SNORM;
832 template.last_level = 0;
833 template.width0 = r->pot_buffers ?
834 util_next_power_of_two(r->picture_width) : r->picture_width;
835 template.height0 = r->pot_buffers ?
836 util_next_power_of_two(r->picture_height) : r->picture_height;
837 template.depth0 = 1;
838 util_format_get_block(template.format, &template.block);
839 template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_DYNAMIC;
840
841 r->textures.individual.y = r->pipe->screen->texture_create(r->pipe->screen, &template);
842
843 if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420) {
844 template.width0 = r->pot_buffers ?
845 util_next_power_of_two(r->picture_width / 2) :
846 r->picture_width / 2;
847 template.height0 = r->pot_buffers ?
848 util_next_power_of_two(r->picture_height / 2) :
849 r->picture_height / 2;
850 }
851 else if (r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_422)
852 template.height0 = r->pot_buffers ?
853 util_next_power_of_two(r->picture_height / 2) :
854 r->picture_height / 2;
855
856 r->textures.individual.cb =
857 r->pipe->screen->texture_create(r->pipe->screen, &template);
858 r->textures.individual.cr =
859 r->pipe->screen->texture_create(r->pipe->screen, &template);
860
861 r->vertex_bufs.individual.ycbcr.stride = sizeof(struct vertex2f) * 4;
862 r->vertex_bufs.individual.ycbcr.max_index = 24 * r->macroblocks_per_batch - 1;
863 r->vertex_bufs.individual.ycbcr.buffer_offset = 0;
864 r->vertex_bufs.individual.ycbcr.buffer = pipe_buffer_create
865 (
866 r->pipe->screen,
867 DEFAULT_BUF_ALIGNMENT,
868 PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
869 sizeof(struct vertex2f) * 4 * 24 * r->macroblocks_per_batch
870 );
871
872 for (i = 1; i < 3; ++i) {
873 r->vertex_bufs.all[i].stride = sizeof(struct vertex2f) * 2;
874 r->vertex_bufs.all[i].max_index = 24 * r->macroblocks_per_batch - 1;
875 r->vertex_bufs.all[i].buffer_offset = 0;
876 r->vertex_bufs.all[i].buffer = pipe_buffer_create
877 (
878 r->pipe->screen,
879 DEFAULT_BUF_ALIGNMENT,
880 PIPE_BUFFER_USAGE_VERTEX | PIPE_BUFFER_USAGE_DISCARD,
881 sizeof(struct vertex2f) * 2 * 24 * r->macroblocks_per_batch
882 );
883 }
884
885 /* Position element */
886 r->vertex_elems[0].src_offset = 0;
887 r->vertex_elems[0].vertex_buffer_index = 0;
888 r->vertex_elems[0].nr_components = 2;
889 r->vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
890
891 /* Luma, texcoord element */
892 r->vertex_elems[1].src_offset = sizeof(struct vertex2f);
893 r->vertex_elems[1].vertex_buffer_index = 0;
894 r->vertex_elems[1].nr_components = 2;
895 r->vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
896
897 /* Chroma Cr texcoord element */
898 r->vertex_elems[2].src_offset = sizeof(struct vertex2f) * 2;
899 r->vertex_elems[2].vertex_buffer_index = 0;
900 r->vertex_elems[2].nr_components = 2;
901 r->vertex_elems[2].src_format = PIPE_FORMAT_R32G32_FLOAT;
902
903 /* Chroma Cb texcoord element */
904 r->vertex_elems[3].src_offset = sizeof(struct vertex2f) * 3;
905 r->vertex_elems[3].vertex_buffer_index = 0;
906 r->vertex_elems[3].nr_components = 2;
907 r->vertex_elems[3].src_format = PIPE_FORMAT_R32G32_FLOAT;
908
909 /* First ref surface top field texcoord element */
910 r->vertex_elems[4].src_offset = 0;
911 r->vertex_elems[4].vertex_buffer_index = 1;
912 r->vertex_elems[4].nr_components = 2;
913 r->vertex_elems[4].src_format = PIPE_FORMAT_R32G32_FLOAT;
914
915 /* First ref surface bottom field texcoord element */
916 r->vertex_elems[5].src_offset = sizeof(struct vertex2f);
917 r->vertex_elems[5].vertex_buffer_index = 1;
918 r->vertex_elems[5].nr_components = 2;
919 r->vertex_elems[5].src_format = PIPE_FORMAT_R32G32_FLOAT;
920
921 /* Second ref surface top field texcoord element */
922 r->vertex_elems[6].src_offset = 0;
923 r->vertex_elems[6].vertex_buffer_index = 2;
924 r->vertex_elems[6].nr_components = 2;
925 r->vertex_elems[6].src_format = PIPE_FORMAT_R32G32_FLOAT;
926
927 /* Second ref surface bottom field texcoord element */
928 r->vertex_elems[7].src_offset = sizeof(struct vertex2f);
929 r->vertex_elems[7].vertex_buffer_index = 2;
930 r->vertex_elems[7].nr_components = 2;
931 r->vertex_elems[7].src_format = PIPE_FORMAT_R32G32_FLOAT;
932
933 r->vs_const_buf.buffer = pipe_buffer_create
934 (
935 r->pipe->screen,
936 DEFAULT_BUF_ALIGNMENT,
937 PIPE_BUFFER_USAGE_CONSTANT | PIPE_BUFFER_USAGE_DISCARD,
938 sizeof(struct vertex_shader_consts)
939 );
940
941 r->fs_const_buf.buffer = pipe_buffer_create
942 (
943 r->pipe->screen,
944 DEFAULT_BUF_ALIGNMENT,
945 PIPE_BUFFER_USAGE_CONSTANT, sizeof(struct fragment_shader_consts)
946 );
947
948 memcpy
949 (
950 pipe_buffer_map(r->pipe->screen, r->fs_const_buf.buffer, PIPE_BUFFER_USAGE_CPU_WRITE),
951 &fs_consts, sizeof(struct fragment_shader_consts)
952 );
953
954 pipe_buffer_unmap(r->pipe->screen, r->fs_const_buf.buffer);
955
956 return true;
957 }
958
959 static void
960 cleanup_buffers(struct vl_mpeg12_mc_renderer *r)
961 {
962 unsigned i;
963
964 assert(r);
965
966 pipe_buffer_reference(&r->vs_const_buf.buffer, NULL);
967 pipe_buffer_reference(&r->fs_const_buf.buffer, NULL);
968
969 for (i = 0; i < 3; ++i)
970 pipe_buffer_reference(&r->vertex_bufs.all[i].buffer, NULL);
971
972 for (i = 0; i < 3; ++i)
973 pipe_texture_reference(&r->textures.all[i], NULL);
974
975 FREE(r->macroblock_buf);
976 }
977
978 static enum MACROBLOCK_TYPE
979 get_macroblock_type(struct pipe_mpeg12_macroblock *mb)
980 {
981 assert(mb);
982
983 switch (mb->mb_type) {
984 case PIPE_MPEG12_MACROBLOCK_TYPE_INTRA:
985 return MACROBLOCK_TYPE_INTRA;
986 case PIPE_MPEG12_MACROBLOCK_TYPE_FWD:
987 return mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME ?
988 MACROBLOCK_TYPE_FWD_FRAME_PRED : MACROBLOCK_TYPE_FWD_FIELD_PRED;
989 case PIPE_MPEG12_MACROBLOCK_TYPE_BKWD:
990 return mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME ?
991 MACROBLOCK_TYPE_BKWD_FRAME_PRED : MACROBLOCK_TYPE_BKWD_FIELD_PRED;
992 case PIPE_MPEG12_MACROBLOCK_TYPE_BI:
993 return mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME ?
994 MACROBLOCK_TYPE_BI_FRAME_PRED : MACROBLOCK_TYPE_BI_FIELD_PRED;
995 default:
996 assert(0);
997 }
998
999 /* Unreachable */
1000 return -1;
1001 }
1002
1003 /* XXX: One of these days this will have to be killed with fire */
1004 #define SET_BLOCK(vb, cbp, mbx, mby, unitx, unity, ofsx, ofsy, hx, hy, lm, cbm, crm, use_zb, zb) \
1005 do { \
1006 (vb)[0].pos.x = (mbx) * (unitx) + (ofsx); (vb)[0].pos.y = (mby) * (unity) + (ofsy); \
1007 (vb)[1].pos.x = (mbx) * (unitx) + (ofsx); (vb)[1].pos.y = (mby) * (unity) + (ofsy) + (hy); \
1008 (vb)[2].pos.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[2].pos.y = (mby) * (unity) + (ofsy); \
1009 (vb)[3].pos.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[3].pos.y = (mby) * (unity) + (ofsy); \
1010 (vb)[4].pos.x = (mbx) * (unitx) + (ofsx); (vb)[4].pos.y = (mby) * (unity) + (ofsy) + (hy); \
1011 (vb)[5].pos.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[5].pos.y = (mby) * (unity) + (ofsy) + (hy); \
1012 \
1013 if (!use_zb || (cbp) & (lm)) \
1014 { \
1015 (vb)[0].luma_tc.x = (mbx) * (unitx) + (ofsx); (vb)[0].luma_tc.y = (mby) * (unity) + (ofsy); \
1016 (vb)[1].luma_tc.x = (mbx) * (unitx) + (ofsx); (vb)[1].luma_tc.y = (mby) * (unity) + (ofsy) + (hy); \
1017 (vb)[2].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[2].luma_tc.y = (mby) * (unity) + (ofsy); \
1018 (vb)[3].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[3].luma_tc.y = (mby) * (unity) + (ofsy); \
1019 (vb)[4].luma_tc.x = (mbx) * (unitx) + (ofsx); (vb)[4].luma_tc.y = (mby) * (unity) + (ofsy) + (hy); \
1020 (vb)[5].luma_tc.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[5].luma_tc.y = (mby) * (unity) + (ofsy) + (hy); \
1021 } \
1022 else \
1023 { \
1024 (vb)[0].luma_tc.x = (zb)[0].x; (vb)[0].luma_tc.y = (zb)[0].y; \
1025 (vb)[1].luma_tc.x = (zb)[0].x; (vb)[1].luma_tc.y = (zb)[0].y + (hy); \
1026 (vb)[2].luma_tc.x = (zb)[0].x + (hx); (vb)[2].luma_tc.y = (zb)[0].y; \
1027 (vb)[3].luma_tc.x = (zb)[0].x + (hx); (vb)[3].luma_tc.y = (zb)[0].y; \
1028 (vb)[4].luma_tc.x = (zb)[0].x; (vb)[4].luma_tc.y = (zb)[0].y + (hy); \
1029 (vb)[5].luma_tc.x = (zb)[0].x + (hx); (vb)[5].luma_tc.y = (zb)[0].y + (hy); \
1030 } \
1031 \
1032 if (!use_zb || (cbp) & (cbm)) \
1033 { \
1034 (vb)[0].cb_tc.x = (mbx) * (unitx) + (ofsx); (vb)[0].cb_tc.y = (mby) * (unity) + (ofsy); \
1035 (vb)[1].cb_tc.x = (mbx) * (unitx) + (ofsx); (vb)[1].cb_tc.y = (mby) * (unity) + (ofsy) + (hy); \
1036 (vb)[2].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[2].cb_tc.y = (mby) * (unity) + (ofsy); \
1037 (vb)[3].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[3].cb_tc.y = (mby) * (unity) + (ofsy); \
1038 (vb)[4].cb_tc.x = (mbx) * (unitx) + (ofsx); (vb)[4].cb_tc.y = (mby) * (unity) + (ofsy) + (hy); \
1039 (vb)[5].cb_tc.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[5].cb_tc.y = (mby) * (unity) + (ofsy) + (hy); \
1040 } \
1041 else \
1042 { \
1043 (vb)[0].cb_tc.x = (zb)[1].x; (vb)[0].cb_tc.y = (zb)[1].y; \
1044 (vb)[1].cb_tc.x = (zb)[1].x; (vb)[1].cb_tc.y = (zb)[1].y + (hy); \
1045 (vb)[2].cb_tc.x = (zb)[1].x + (hx); (vb)[2].cb_tc.y = (zb)[1].y; \
1046 (vb)[3].cb_tc.x = (zb)[1].x + (hx); (vb)[3].cb_tc.y = (zb)[1].y; \
1047 (vb)[4].cb_tc.x = (zb)[1].x; (vb)[4].cb_tc.y = (zb)[1].y + (hy); \
1048 (vb)[5].cb_tc.x = (zb)[1].x + (hx); (vb)[5].cb_tc.y = (zb)[1].y + (hy); \
1049 } \
1050 \
1051 if (!use_zb || (cbp) & (crm)) \
1052 { \
1053 (vb)[0].cr_tc.x = (mbx) * (unitx) + (ofsx); (vb)[0].cr_tc.y = (mby) * (unity) + (ofsy); \
1054 (vb)[1].cr_tc.x = (mbx) * (unitx) + (ofsx); (vb)[1].cr_tc.y = (mby) * (unity) + (ofsy) + (hy); \
1055 (vb)[2].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[2].cr_tc.y = (mby) * (unity) + (ofsy); \
1056 (vb)[3].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[3].cr_tc.y = (mby) * (unity) + (ofsy); \
1057 (vb)[4].cr_tc.x = (mbx) * (unitx) + (ofsx); (vb)[4].cr_tc.y = (mby) * (unity) + (ofsy) + (hy); \
1058 (vb)[5].cr_tc.x = (mbx) * (unitx) + (ofsx) + (hx); (vb)[5].cr_tc.y = (mby) * (unity) + (ofsy) + (hy); \
1059 } \
1060 else \
1061 { \
1062 (vb)[0].cr_tc.x = (zb)[2].x; (vb)[0].cr_tc.y = (zb)[2].y; \
1063 (vb)[1].cr_tc.x = (zb)[2].x; (vb)[1].cr_tc.y = (zb)[2].y + (hy); \
1064 (vb)[2].cr_tc.x = (zb)[2].x + (hx); (vb)[2].cr_tc.y = (zb)[2].y; \
1065 (vb)[3].cr_tc.x = (zb)[2].x + (hx); (vb)[3].cr_tc.y = (zb)[2].y; \
1066 (vb)[4].cr_tc.x = (zb)[2].x; (vb)[4].cr_tc.y = (zb)[2].y + (hy); \
1067 (vb)[5].cr_tc.x = (zb)[2].x + (hx); (vb)[5].cr_tc.y = (zb)[2].y + (hy); \
1068 } \
1069 } while (0)
1070
1071 static void
1072 gen_macroblock_verts(struct vl_mpeg12_mc_renderer *r,
1073 struct pipe_mpeg12_macroblock *mb, unsigned pos,
1074 struct vert_stream_0 *ycbcr_vb, struct vertex2f **ref_vb)
1075 {
1076 struct vertex2f mo_vec[2];
1077
1078 unsigned i;
1079
1080 assert(r);
1081 assert(mb);
1082 assert(ycbcr_vb);
1083 assert(pos < r->macroblocks_per_batch);
1084
1085 switch (mb->mb_type) {
1086 case PIPE_MPEG12_MACROBLOCK_TYPE_BI:
1087 {
1088 struct vertex2f *vb;
1089
1090 assert(ref_vb && ref_vb[1]);
1091
1092 vb = ref_vb[1] + pos * 2 * 24;
1093
1094 mo_vec[0].x = mb->pmv[0][1][0] * 0.5f * r->surface_tex_inv_size.x;
1095 mo_vec[0].y = mb->pmv[0][1][1] * 0.5f * r->surface_tex_inv_size.y;
1096
1097 if (mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FRAME) {
1098 for (i = 0; i < 24 * 2; i += 2) {
1099 vb[i].x = mo_vec[0].x;
1100 vb[i].y = mo_vec[0].y;
1101 }
1102 }
1103 else {
1104 mo_vec[1].x = mb->pmv[1][1][0] * 0.5f * r->surface_tex_inv_size.x;
1105 mo_vec[1].y = mb->pmv[1][1][1] * 0.5f * r->surface_tex_inv_size.y;
1106
1107 for (i = 0; i < 24 * 2; i += 2) {
1108 vb[i].x = mo_vec[0].x;
1109 vb[i].y = mo_vec[0].y;
1110 vb[i + 1].x = mo_vec[1].x;
1111 vb[i + 1].y = mo_vec[1].y;
1112 }
1113 }
1114
1115 /* fall-through */
1116 }
1117 case PIPE_MPEG12_MACROBLOCK_TYPE_FWD:
1118 case PIPE_MPEG12_MACROBLOCK_TYPE_BKWD:
1119 {
1120 struct vertex2f *vb;
1121
1122 assert(ref_vb && ref_vb[0]);
1123
1124 vb = ref_vb[0] + pos * 2 * 24;
1125
1126 if (mb->mb_type == PIPE_MPEG12_MACROBLOCK_TYPE_BKWD) {
1127 mo_vec[0].x = mb->pmv[0][1][0] * 0.5f * r->surface_tex_inv_size.x;
1128 mo_vec[0].y = mb->pmv[0][1][1] * 0.5f * r->surface_tex_inv_size.y;
1129
1130 if (mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FIELD) {
1131 mo_vec[1].x = mb->pmv[1][1][0] * 0.5f * r->surface_tex_inv_size.x;
1132 mo_vec[1].y = mb->pmv[1][1][1] * 0.5f * r->surface_tex_inv_size.y;
1133 }
1134 }
1135 else {
1136 mo_vec[0].x = mb->pmv[0][0][0] * 0.5f * r->surface_tex_inv_size.x;
1137 mo_vec[0].y = mb->pmv[0][0][1] * 0.5f * r->surface_tex_inv_size.y;
1138
1139 if (mb->mo_type == PIPE_MPEG12_MOTION_TYPE_FIELD) {
1140 mo_vec[1].x = mb->pmv[1][0][0] * 0.5f * r->surface_tex_inv_size.x;
1141 mo_vec[1].y = mb->pmv[1][0][1] * 0.5f * r->surface_tex_inv_size.y;
1142 }
1143 }
1144
1145 if (mb->mb_type == PIPE_MPEG12_MOTION_TYPE_FRAME) {
1146 for (i = 0; i < 24 * 2; i += 2) {
1147 vb[i].x = mo_vec[0].x;
1148 vb[i].y = mo_vec[0].y;
1149 }
1150 }
1151 else {
1152 for (i = 0; i < 24 * 2; i += 2) {
1153 vb[i].x = mo_vec[0].x;
1154 vb[i].y = mo_vec[0].y;
1155 vb[i + 1].x = mo_vec[1].x;
1156 vb[i + 1].y = mo_vec[1].y;
1157 }
1158 }
1159
1160 /* fall-through */
1161 }
1162 case PIPE_MPEG12_MACROBLOCK_TYPE_INTRA:
1163 {
1164 const struct vertex2f unit =
1165 {
1166 r->surface_tex_inv_size.x * MACROBLOCK_WIDTH,
1167 r->surface_tex_inv_size.y * MACROBLOCK_HEIGHT
1168 };
1169 const struct vertex2f half =
1170 {
1171 r->surface_tex_inv_size.x * (MACROBLOCK_WIDTH / 2),
1172 r->surface_tex_inv_size.y * (MACROBLOCK_HEIGHT / 2)
1173 };
1174 const bool use_zb = r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE;
1175
1176 struct vert_stream_0 *vb = ycbcr_vb + pos * 24;
1177
1178 SET_BLOCK(vb, mb->cbp, mb->mbx, mb->mby,
1179 unit.x, unit.y, 0, 0, half.x, half.y,
1180 32, 2, 1, use_zb, r->zero_block);
1181
1182 SET_BLOCK(vb + 6, mb->cbp, mb->mbx, mb->mby,
1183 unit.x, unit.y, half.x, 0, half.x, half.y,
1184 16, 2, 1, use_zb, r->zero_block);
1185
1186 SET_BLOCK(vb + 12, mb->cbp, mb->mbx, mb->mby,
1187 unit.x, unit.y, 0, half.y, half.x, half.y,
1188 8, 2, 1, use_zb, r->zero_block);
1189
1190 SET_BLOCK(vb + 18, mb->cbp, mb->mbx, mb->mby,
1191 unit.x, unit.y, half.x, half.y, half.x, half.y,
1192 4, 2, 1, use_zb, r->zero_block);
1193
1194 break;
1195 }
1196 default:
1197 assert(0);
1198 }
1199 }
1200
1201 static void
1202 gen_macroblock_stream(struct vl_mpeg12_mc_renderer *r,
1203 unsigned *num_macroblocks)
1204 {
1205 unsigned offset[NUM_MACROBLOCK_TYPES];
1206 struct vert_stream_0 *ycbcr_vb;
1207 struct vertex2f *ref_vb[2];
1208 unsigned i;
1209
1210 assert(r);
1211 assert(num_macroblocks);
1212
1213 for (i = 0; i < r->num_macroblocks; ++i) {
1214 enum MACROBLOCK_TYPE mb_type = get_macroblock_type(&r->macroblock_buf[i]);
1215 ++num_macroblocks[mb_type];
1216 }
1217
1218 offset[0] = 0;
1219
1220 for (i = 1; i < NUM_MACROBLOCK_TYPES; ++i)
1221 offset[i] = offset[i - 1] + num_macroblocks[i - 1];
1222
1223 ycbcr_vb = (struct vert_stream_0 *)pipe_buffer_map
1224 (
1225 r->pipe->screen,
1226 r->vertex_bufs.individual.ycbcr.buffer,
1227 PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
1228 );
1229
1230 for (i = 0; i < 2; ++i)
1231 ref_vb[i] = (struct vertex2f *)pipe_buffer_map
1232 (
1233 r->pipe->screen,
1234 r->vertex_bufs.individual.ref[i].buffer,
1235 PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
1236 );
1237
1238 for (i = 0; i < r->num_macroblocks; ++i) {
1239 enum MACROBLOCK_TYPE mb_type = get_macroblock_type(&r->macroblock_buf[i]);
1240
1241 gen_macroblock_verts(r, &r->macroblock_buf[i], offset[mb_type],
1242 ycbcr_vb, ref_vb);
1243
1244 ++offset[mb_type];
1245 }
1246
1247 pipe_buffer_unmap(r->pipe->screen, r->vertex_bufs.individual.ycbcr.buffer);
1248 for (i = 0; i < 2; ++i)
1249 pipe_buffer_unmap(r->pipe->screen, r->vertex_bufs.individual.ref[i].buffer);
1250 }
1251
1252 static void
1253 flush(struct vl_mpeg12_mc_renderer *r)
1254 {
1255 unsigned num_macroblocks[NUM_MACROBLOCK_TYPES] = { 0 };
1256 unsigned vb_start = 0;
1257 struct vertex_shader_consts *vs_consts;
1258 unsigned i;
1259
1260 assert(r);
1261 assert(r->num_macroblocks == r->macroblocks_per_batch);
1262
1263 gen_macroblock_stream(r, num_macroblocks);
1264
1265 r->fb_state.cbufs[0] = r->pipe->screen->get_tex_surface
1266 (
1267 r->pipe->screen, r->surface,
1268 0, 0, 0, PIPE_BUFFER_USAGE_GPU_WRITE
1269 );
1270
1271 r->pipe->set_framebuffer_state(r->pipe, &r->fb_state);
1272 r->pipe->set_viewport_state(r->pipe, &r->viewport);
1273 r->pipe->set_scissor_state(r->pipe, &r->scissor);
1274
1275 vs_consts = pipe_buffer_map
1276 (
1277 r->pipe->screen, r->vs_const_buf.buffer,
1278 PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
1279 );
1280
1281 vs_consts->denorm.x = r->surface->width0;
1282 vs_consts->denorm.y = r->surface->height0;
1283
1284 pipe_buffer_unmap(r->pipe->screen, r->vs_const_buf.buffer);
1285
1286 r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_VERTEX, 0,
1287 &r->vs_const_buf);
1288 r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_FRAGMENT, 0,
1289 &r->fs_const_buf);
1290
1291 if (num_macroblocks[MACROBLOCK_TYPE_INTRA] > 0) {
1292 r->pipe->set_vertex_buffers(r->pipe, 1, r->vertex_bufs.all);
1293 r->pipe->set_vertex_elements(r->pipe, 4, r->vertex_elems);
1294 r->pipe->set_fragment_sampler_textures(r->pipe, 3, r->textures.all);
1295 r->pipe->bind_fragment_sampler_states(r->pipe, 3, r->samplers.all);
1296 r->pipe->bind_vs_state(r->pipe, r->i_vs);
1297 r->pipe->bind_fs_state(r->pipe, r->i_fs);
1298
1299 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1300 num_macroblocks[MACROBLOCK_TYPE_INTRA] * 24);
1301 vb_start += num_macroblocks[MACROBLOCK_TYPE_INTRA] * 24;
1302 }
1303
1304 if (num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] > 0) {
1305 r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
1306 r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
1307 r->textures.individual.ref[0] = r->past;
1308 r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
1309 r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
1310 r->pipe->bind_vs_state(r->pipe, r->p_vs[0]);
1311 r->pipe->bind_fs_state(r->pipe, r->p_fs[0]);
1312
1313 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1314 num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] * 24);
1315 vb_start += num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] * 24;
1316 }
1317
1318 if (false /*num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] > 0 */ ) {
1319 r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
1320 r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
1321 r->textures.individual.ref[0] = r->past;
1322 r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
1323 r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
1324 r->pipe->bind_vs_state(r->pipe, r->p_vs[1]);
1325 r->pipe->bind_fs_state(r->pipe, r->p_fs[1]);
1326
1327 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1328 num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] * 24);
1329 vb_start += num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] * 24;
1330 }
1331
1332 if (num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] > 0) {
1333 r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
1334 r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
1335 r->textures.individual.ref[0] = r->future;
1336 r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
1337 r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
1338 r->pipe->bind_vs_state(r->pipe, r->p_vs[0]);
1339 r->pipe->bind_fs_state(r->pipe, r->p_fs[0]);
1340
1341 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1342 num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] * 24);
1343 vb_start += num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] * 24;
1344 }
1345
1346 if (false /*num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] > 0 */ ) {
1347 r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
1348 r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
1349 r->textures.individual.ref[0] = r->future;
1350 r->pipe->set_fragment_sampler_textures(r->pipe, 4, r->textures.all);
1351 r->pipe->bind_fragment_sampler_states(r->pipe, 4, r->samplers.all);
1352 r->pipe->bind_vs_state(r->pipe, r->p_vs[1]);
1353 r->pipe->bind_fs_state(r->pipe, r->p_fs[1]);
1354
1355 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1356 num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] * 24);
1357 vb_start += num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] * 24;
1358 }
1359
1360 if (num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] > 0) {
1361 r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
1362 r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
1363 r->textures.individual.ref[0] = r->past;
1364 r->textures.individual.ref[1] = r->future;
1365 r->pipe->set_fragment_sampler_textures(r->pipe, 5, r->textures.all);
1366 r->pipe->bind_fragment_sampler_states(r->pipe, 5, r->samplers.all);
1367 r->pipe->bind_vs_state(r->pipe, r->b_vs[0]);
1368 r->pipe->bind_fs_state(r->pipe, r->b_fs[0]);
1369
1370 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1371 num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] * 24);
1372 vb_start += num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] * 24;
1373 }
1374
1375 if (false /*num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] > 0 */ ) {
1376 r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
1377 r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
1378 r->textures.individual.ref[0] = r->past;
1379 r->textures.individual.ref[1] = r->future;
1380 r->pipe->set_fragment_sampler_textures(r->pipe, 5, r->textures.all);
1381 r->pipe->bind_fragment_sampler_states(r->pipe, 5, r->samplers.all);
1382 r->pipe->bind_vs_state(r->pipe, r->b_vs[1]);
1383 r->pipe->bind_fs_state(r->pipe, r->b_fs[1]);
1384
1385 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1386 num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] * 24);
1387 vb_start += num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] * 24;
1388 }
1389
1390 r->pipe->flush(r->pipe, PIPE_FLUSH_RENDER_CACHE, r->fence);
1391 pipe_surface_reference(&r->fb_state.cbufs[0], NULL);
1392
1393 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE)
1394 for (i = 0; i < 3; ++i)
1395 r->zero_block[i].x = ZERO_BLOCK_NIL;
1396
1397 r->num_macroblocks = 0;
1398 }
1399
1400 static void
1401 grab_frame_coded_block(short *src, short *dst, unsigned dst_pitch)
1402 {
1403 unsigned y;
1404
1405 assert(src);
1406 assert(dst);
1407
1408 for (y = 0; y < BLOCK_HEIGHT; ++y)
1409 memcpy(dst + y * dst_pitch, src + y * BLOCK_WIDTH, BLOCK_WIDTH * 2);
1410 }
1411
1412 static void
1413 grab_field_coded_block(short *src, short *dst, unsigned dst_pitch)
1414 {
1415 unsigned y;
1416
1417 assert(src);
1418 assert(dst);
1419
1420 for (y = 0; y < BLOCK_HEIGHT; ++y)
1421 memcpy(dst + y * dst_pitch * 2, src + y * BLOCK_WIDTH, BLOCK_WIDTH * 2);
1422 }
1423
1424 static void
1425 fill_zero_block(short *dst, unsigned dst_pitch)
1426 {
1427 unsigned y;
1428
1429 assert(dst);
1430
1431 for (y = 0; y < BLOCK_HEIGHT; ++y)
1432 memset(dst + y * dst_pitch, 0, BLOCK_WIDTH * 2);
1433 }
1434
1435 static void
1436 grab_blocks(struct vl_mpeg12_mc_renderer *r, unsigned mbx, unsigned mby,
1437 enum pipe_mpeg12_dct_type dct_type, unsigned cbp, short *blocks)
1438 {
1439 unsigned tex_pitch;
1440 short *texels;
1441 unsigned tb = 0, sb = 0;
1442 unsigned mbpx = mbx * MACROBLOCK_WIDTH, mbpy = mby * MACROBLOCK_HEIGHT;
1443 unsigned x, y;
1444
1445 assert(r);
1446 assert(blocks);
1447
1448 tex_pitch = r->tex_transfer[0]->stride / r->tex_transfer[0]->block.size;
1449 texels = r->texels[0] + mbpy * tex_pitch + mbpx;
1450
1451 for (y = 0; y < 2; ++y) {
1452 for (x = 0; x < 2; ++x, ++tb) {
1453 if ((cbp >> (5 - tb)) & 1) {
1454 if (dct_type == PIPE_MPEG12_DCT_TYPE_FRAME) {
1455 grab_frame_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT,
1456 texels + y * tex_pitch * BLOCK_WIDTH +
1457 x * BLOCK_WIDTH, tex_pitch);
1458 }
1459 else {
1460 grab_field_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT,
1461 texels + y * tex_pitch + x * BLOCK_WIDTH,
1462 tex_pitch);
1463 }
1464
1465 ++sb;
1466 }
1467 else if (r->eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE) {
1468 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL ||
1469 ZERO_BLOCK_IS_NIL(r->zero_block[0])) {
1470 fill_zero_block(texels + y * tex_pitch * BLOCK_WIDTH + x * BLOCK_WIDTH, tex_pitch);
1471 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE) {
1472 r->zero_block[0].x = (mbpx + x * 8) * r->surface_tex_inv_size.x;
1473 r->zero_block[0].y = (mbpy + y * 8) * r->surface_tex_inv_size.y;
1474 }
1475 }
1476 }
1477 }
1478 }
1479
1480 /* TODO: Implement 422, 444 */
1481 assert(r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
1482
1483 mbpx /= 2;
1484 mbpy /= 2;
1485
1486 for (tb = 0; tb < 2; ++tb) {
1487 tex_pitch = r->tex_transfer[tb + 1]->stride / r->tex_transfer[tb + 1]->block.size;
1488 texels = r->texels[tb + 1] + mbpy * tex_pitch + mbpx;
1489
1490 if ((cbp >> (1 - tb)) & 1) {
1491 grab_frame_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT, texels, tex_pitch);
1492 ++sb;
1493 }
1494 else if (r->eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE) {
1495 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL ||
1496 ZERO_BLOCK_IS_NIL(r->zero_block[tb + 1])) {
1497 fill_zero_block(texels, tex_pitch);
1498 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE) {
1499 r->zero_block[tb + 1].x = (mbpx << 1) * r->surface_tex_inv_size.x;
1500 r->zero_block[tb + 1].y = (mbpy << 1) * r->surface_tex_inv_size.y;
1501 }
1502 }
1503 }
1504 }
1505 }
1506
1507 static void
1508 grab_macroblock(struct vl_mpeg12_mc_renderer *r,
1509 struct pipe_mpeg12_macroblock *mb)
1510 {
1511 assert(r);
1512 assert(mb);
1513 assert(r->num_macroblocks < r->macroblocks_per_batch);
1514
1515 memcpy(&r->macroblock_buf[r->num_macroblocks], mb,
1516 sizeof(struct pipe_mpeg12_macroblock));
1517
1518 grab_blocks(r, mb->mbx, mb->mby, mb->dct_type, mb->cbp, mb->blocks);
1519
1520 ++r->num_macroblocks;
1521 }
1522
1523 bool
1524 vl_mpeg12_mc_renderer_init(struct vl_mpeg12_mc_renderer *renderer,
1525 struct pipe_context *pipe,
1526 unsigned picture_width,
1527 unsigned picture_height,
1528 enum pipe_video_chroma_format chroma_format,
1529 enum VL_MPEG12_MC_RENDERER_BUFFER_MODE bufmode,
1530 enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK eb_handling,
1531 bool pot_buffers)
1532 {
1533 unsigned i;
1534
1535 assert(renderer);
1536 assert(pipe);
1537 /* TODO: Implement other policies */
1538 assert(bufmode == VL_MPEG12_MC_RENDERER_BUFFER_PICTURE);
1539 /* TODO: Implement this */
1540 /* XXX: XFER_ALL sampling issue at block edges when using bilinear filtering */
1541 assert(eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE);
1542 /* TODO: Non-pot buffers untested, probably doesn't work without changes to texcoord generation, vert shader, etc */
1543 assert(pot_buffers);
1544
1545 memset(renderer, 0, sizeof(struct vl_mpeg12_mc_renderer));
1546
1547 renderer->pipe = pipe;
1548 renderer->picture_width = picture_width;
1549 renderer->picture_height = picture_height;
1550 renderer->chroma_format = chroma_format;
1551 renderer->bufmode = bufmode;
1552 renderer->eb_handling = eb_handling;
1553 renderer->pot_buffers = pot_buffers;
1554
1555 if (!init_pipe_state(renderer))
1556 return false;
1557 if (!init_shaders(renderer)) {
1558 cleanup_pipe_state(renderer);
1559 return false;
1560 }
1561 if (!init_buffers(renderer)) {
1562 cleanup_shaders(renderer);
1563 cleanup_pipe_state(renderer);
1564 return false;
1565 }
1566
1567 renderer->surface = NULL;
1568 renderer->past = NULL;
1569 renderer->future = NULL;
1570 for (i = 0; i < 3; ++i)
1571 renderer->zero_block[i].x = ZERO_BLOCK_NIL;
1572 renderer->num_macroblocks = 0;
1573
1574 xfer_buffers_map(renderer);
1575
1576 return true;
1577 }
1578
1579 void
1580 vl_mpeg12_mc_renderer_cleanup(struct vl_mpeg12_mc_renderer *renderer)
1581 {
1582 assert(renderer);
1583
1584 xfer_buffers_unmap(renderer);
1585
1586 cleanup_pipe_state(renderer);
1587 cleanup_shaders(renderer);
1588 cleanup_buffers(renderer);
1589 }
1590
1591 void
1592 vl_mpeg12_mc_renderer_render_macroblocks(struct vl_mpeg12_mc_renderer
1593 *renderer,
1594 struct pipe_texture *surface,
1595 struct pipe_texture *past,
1596 struct pipe_texture *future,
1597 unsigned num_macroblocks,
1598 struct pipe_mpeg12_macroblock
1599 *mpeg12_macroblocks,
1600 struct pipe_fence_handle **fence)
1601 {
1602 bool new_surface = false;
1603
1604 assert(renderer);
1605 assert(surface);
1606 assert(num_macroblocks);
1607 assert(mpeg12_macroblocks);
1608
1609 if (renderer->surface) {
1610 if (surface != renderer->surface) {
1611 if (renderer->num_macroblocks > 0) {
1612 xfer_buffers_unmap(renderer);
1613 flush(renderer);
1614 }
1615
1616 new_surface = true;
1617 }
1618
1619 /* If the surface we're rendering hasn't changed the ref frames shouldn't change. */
1620 assert(surface != renderer->surface || renderer->past == past);
1621 assert(surface != renderer->surface || renderer->future == future);
1622 }
1623 else
1624 new_surface = true;
1625
1626 if (new_surface) {
1627 renderer->surface = surface;
1628 renderer->past = past;
1629 renderer->future = future;
1630 renderer->fence = fence;
1631 renderer->surface_tex_inv_size.x = 1.0f / surface->width0;
1632 renderer->surface_tex_inv_size.y = 1.0f / surface->height0;
1633 }
1634
1635 while (num_macroblocks) {
1636 unsigned left_in_batch = renderer->macroblocks_per_batch - renderer->num_macroblocks;
1637 unsigned num_to_submit = MIN2(num_macroblocks, left_in_batch);
1638 unsigned i;
1639
1640 for (i = 0; i < num_to_submit; ++i) {
1641 assert(mpeg12_macroblocks[i].base.codec == PIPE_VIDEO_CODEC_MPEG12);
1642 grab_macroblock(renderer, &mpeg12_macroblocks[i]);
1643 }
1644
1645 num_macroblocks -= num_to_submit;
1646
1647 if (renderer->num_macroblocks == renderer->macroblocks_per_batch) {
1648 xfer_buffers_unmap(renderer);
1649 flush(renderer);
1650 xfer_buffers_map(renderer);
1651 /* Next time we get this surface it may have new ref frames */
1652 renderer->surface = NULL;
1653 }
1654 }
1655 }