projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
Merge branch 'mesa_7_6_branch'
[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_math.h>
33 #include <util/u_memory.h>
34 #include <tgsi/tgsi_parse.h>
35 #include <tgsi/tgsi_build.h>
36 #include "vl_shader_build.h"
37
38 #define DEFAULT_BUF_ALIGNMENT 1
39 #define MACROBLOCK_WIDTH 16
40 #define MACROBLOCK_HEIGHT 16
41 #define BLOCK_WIDTH 8
42 #define BLOCK_HEIGHT 8
43 #define ZERO_BLOCK_NIL -1.0f
44 #define ZERO_BLOCK_IS_NIL(zb) ((zb).x < 0.0f)
45
46 struct vertex2f
47 {
48 float x, y;
49 };
50
51 struct vertex4f
52 {
53 float x, y, z, w;
54 };
55
56 struct vertex_shader_consts
57 {
58 struct vertex4f denorm;
59 };
60
61 struct fragment_shader_consts
62 {
63 struct vertex4f multiplier;
64 struct vertex4f div;
65 };
66
67 /*
68 * Muliplier renormalizes block samples from 16 bits to 12 bits.
69 * Divider is used when calculating Y % 2 for choosing top or bottom
70 * field for P or B macroblocks.
71 * TODO: Use immediates.
72 */
73 static const struct fragment_shader_consts fs_consts = {
74 {32767.0f / 255.0f, 32767.0f / 255.0f, 32767.0f / 255.0f, 0.0f},
75 {0.5f, 2.0f, 0.0f, 0.0f}
76 };
77
78 struct vert_stream_0
79 {
80 struct vertex2f pos;
81 struct vertex2f luma_tc;
82 struct vertex2f cb_tc;
83 struct vertex2f cr_tc;
84 };
85
86 enum MACROBLOCK_TYPE
87 {
88 MACROBLOCK_TYPE_INTRA,
89 MACROBLOCK_TYPE_FWD_FRAME_PRED,
90 MACROBLOCK_TYPE_FWD_FIELD_PRED,
91 MACROBLOCK_TYPE_BKWD_FRAME_PRED,
92 MACROBLOCK_TYPE_BKWD_FIELD_PRED,
93 MACROBLOCK_TYPE_BI_FRAME_PRED,
94 MACROBLOCK_TYPE_BI_FIELD_PRED,
95
96 NUM_MACROBLOCK_TYPES
97 };
98
99 static void
100 create_intra_vert_shader(struct vl_mpeg12_mc_renderer *r)
101 {
102 const unsigned max_tokens = 50;
103
104 struct pipe_shader_state vs;
105 struct tgsi_token *tokens;
106 struct tgsi_header *header;
107
108 struct tgsi_full_declaration decl;
109 struct tgsi_full_instruction inst;
110
111 unsigned ti;
112
113 unsigned i;
114
115 assert(r);
116
117 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
118 *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
119 header = (struct tgsi_header *) &tokens[1];
120 *header = tgsi_build_header();
121 *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
122
123 ti = 3;
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 *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
189 header = (struct tgsi_header *) &tokens[1];
190 *header = tgsi_build_header();
191 *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
192
193 ti = 3;
194
195 /*
196 * decl i0 ; Luma texcoords
197 * decl i1 ; Chroma Cb texcoords
198 * decl i2 ; Chroma Cr texcoords
199 */
200 for (i = 0; i < 3; ++i) {
201 decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
202 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
203 }
204
205 /* decl c0 ; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
206 decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
207 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
208
209 /* decl o0 ; Fragment color */
210 decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
211 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
212
213 /* decl t0, t1 */
214 decl = vl_decl_temps(0, 1);
215 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
216
217 /*
218 * decl s0 ; Sampler for luma texture
219 * decl s1 ; Sampler for chroma Cb texture
220 * decl s2 ; Sampler for chroma Cr texture
221 */
222 for (i = 0; i < 3; ++i) {
223 decl = vl_decl_samplers(i, i);
224 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
225 }
226
227 /*
228 * tex2d t1, i0, s0 ; Read texel from luma texture
229 * mov t0.x, t1.x ; Move luma sample into .x component
230 * tex2d t1, i1, s1 ; Read texel from chroma Cb texture
231 * mov t0.y, t1.x ; Move Cb sample into .y component
232 * tex2d t1, i2, s2 ; Read texel from chroma Cr texture
233 * mov t0.z, t1.x ; Move Cr sample into .z component
234 */
235 for (i = 0; i < 3; ++i) {
236 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
237 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
238
239 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
240 inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
241 inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
242 inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
243 inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
244 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
245 }
246
247 /* mul o0, t0, c0 ; Rescale texel to correct range */
248 inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
249 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
250
251 /* end */
252 inst = vl_end();
253 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
254
255 assert(ti <= max_tokens);
256
257 fs.tokens = tokens;
258 r->i_fs = r->pipe->create_fs_state(r->pipe, &fs);
259 free(tokens);
260 }
261
262 static void
263 create_frame_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
264 {
265 const unsigned max_tokens = 100;
266
267 struct pipe_shader_state vs;
268 struct tgsi_token *tokens;
269 struct tgsi_header *header;
270
271 struct tgsi_full_declaration decl;
272 struct tgsi_full_instruction inst;
273
274 unsigned ti;
275
276 unsigned i;
277
278 assert(r);
279
280 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
281 *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
282 header = (struct tgsi_header *) &tokens[1];
283 *header = tgsi_build_header();
284 *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
285
286 ti = 3;
287
288 /*
289 * decl i0 ; Vertex pos
290 * decl i1 ; Luma texcoords
291 * decl i2 ; Chroma Cb texcoords
292 * decl i3 ; Chroma Cr texcoords
293 * decl i4 ; Ref surface top field texcoords
294 * decl i5 ; Ref surface bottom field texcoords (unused, packed in the same stream)
295 */
296 for (i = 0; i < 6; i++) {
297 decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
298 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
299 }
300
301 /*
302 * decl o0 ; Vertex pos
303 * decl o1 ; Luma texcoords
304 * decl o2 ; Chroma Cb texcoords
305 * decl o3 ; Chroma Cr texcoords
306 * decl o4 ; Ref macroblock texcoords
307 */
308 for (i = 0; i < 5; i++) {
309 decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
310 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
311 }
312
313 /*
314 * mov o0, i0 ; Move input vertex pos to output
315 * mov o1, i1 ; Move input luma texcoords to output
316 * mov o2, i2 ; Move input chroma Cb texcoords to output
317 * mov o3, i3 ; Move input chroma Cr texcoords to output
318 */
319 for (i = 0; i < 4; ++i) {
320 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
321 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
322 }
323
324 /* add o4, i0, i4 ; Translate vertex pos by motion vec to form ref macroblock texcoords */
325 inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, 4);
326 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
327
328 /* end */
329 inst = vl_end();
330 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
331
332 assert(ti <= max_tokens);
333
334 vs.tokens = tokens;
335 r->p_vs[0] = r->pipe->create_vs_state(r->pipe, &vs);
336 free(tokens);
337 }
338
339 static void
340 create_field_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
341 {
342 assert(false);
343 }
344
345 static void
346 create_frame_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
347 {
348 const unsigned max_tokens = 100;
349
350 struct pipe_shader_state fs;
351 struct tgsi_token *tokens;
352 struct tgsi_header *header;
353
354 struct tgsi_full_declaration decl;
355 struct tgsi_full_instruction inst;
356
357 unsigned ti;
358
359 unsigned i;
360
361 assert(r);
362
363 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
364 *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
365 header = (struct tgsi_header *) &tokens[1];
366 *header = tgsi_build_header();
367 *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
368
369 ti = 3;
370
371 /*
372 * decl i0 ; Luma texcoords
373 * decl i1 ; Chroma Cb texcoords
374 * decl i2 ; Chroma Cr texcoords
375 * decl i3 ; Ref macroblock texcoords
376 */
377 for (i = 0; i < 4; ++i) {
378 decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
379 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
380 }
381
382 /* decl c0 ; Scaling factor, rescales 16-bit snorm to 9-bit snorm */
383 decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 0);
384 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
385
386 /* decl o0 ; Fragment color */
387 decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
388 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
389
390 /* decl t0, t1 */
391 decl = vl_decl_temps(0, 1);
392 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
393
394 /*
395 * decl s0 ; Sampler for luma texture
396 * decl s1 ; Sampler for chroma Cb texture
397 * decl s2 ; Sampler for chroma Cr texture
398 * decl s3 ; Sampler for ref surface texture
399 */
400 for (i = 0; i < 4; ++i) {
401 decl = vl_decl_samplers(i, i);
402 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
403 }
404
405 /*
406 * tex2d t1, i0, s0 ; Read texel from luma texture
407 * mov t0.x, t1.x ; Move luma sample into .x component
408 * tex2d t1, i1, s1 ; Read texel from chroma Cb texture
409 * mov t0.y, t1.x ; Move Cb sample into .y component
410 * tex2d t1, i2, s2 ; Read texel from chroma Cr texture
411 * mov t0.z, t1.x ; Move Cr sample into .z component
412 */
413 for (i = 0; i < 3; ++i) {
414 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
415 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
416
417 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
418 inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
419 inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
420 inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
421 inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
422 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
423 }
424
425 /* mul t0, t0, c0 ; Rescale texel to correct range */
426 inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
427 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
428
429 /* tex2d t1, i3, s3 ; Read texel from ref macroblock */
430 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, 3, TGSI_FILE_SAMPLER, 3);
431 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
432
433 /* add o0, t0, t1 ; Add ref and differential to form final output */
434 inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
435 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
436
437 /* end */
438 inst = vl_end();
439 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
440
441 assert(ti <= max_tokens);
442
443 fs.tokens = tokens;
444 r->p_fs[0] = r->pipe->create_fs_state(r->pipe, &fs);
445 free(tokens);
446 }
447
448 static void
449 create_field_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
450 {
451 assert(false);
452 }
453
454 static void
455 create_frame_bi_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
456 {
457 const unsigned max_tokens = 100;
458
459 struct pipe_shader_state vs;
460 struct tgsi_token *tokens;
461 struct tgsi_header *header;
462
463 struct tgsi_full_declaration decl;
464 struct tgsi_full_instruction inst;
465
466 unsigned ti;
467
468 unsigned i;
469
470 assert(r);
471
472 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
473 *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
474 header = (struct tgsi_header *) &tokens[1];
475 *header = tgsi_build_header();
476 *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_VERTEX, header);
477
478 ti = 3;
479
480 /*
481 * decl i0 ; Vertex pos
482 * decl i1 ; Luma texcoords
483 * decl i2 ; Chroma Cb texcoords
484 * decl i3 ; Chroma Cr texcoords
485 * decl i4 ; First ref macroblock top field texcoords
486 * decl i5 ; First ref macroblock bottom field texcoords (unused, packed in the same stream)
487 * decl i6 ; Second ref macroblock top field texcoords
488 * decl i7 ; Second ref macroblock bottom field texcoords (unused, packed in the same stream)
489 */
490 for (i = 0; i < 8; i++) {
491 decl = vl_decl_input(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
492 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
493 }
494
495 /*
496 * decl o0 ; Vertex pos
497 * decl o1 ; Luma texcoords
498 * decl o2 ; Chroma Cb texcoords
499 * decl o3 ; Chroma Cr texcoords
500 * decl o4 ; First ref macroblock texcoords
501 * decl o5 ; Second ref macroblock texcoords
502 */
503 for (i = 0; i < 6; i++) {
504 decl = vl_decl_output(i == 0 ? TGSI_SEMANTIC_POSITION : TGSI_SEMANTIC_GENERIC, i, i, i);
505 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
506 }
507
508 /*
509 * mov o0, i0 ; Move input vertex pos to output
510 * mov o1, i1 ; Move input luma texcoords to output
511 * mov o2, i2 ; Move input chroma Cb texcoords to output
512 * mov o3, i3 ; Move input chroma Cr texcoords to output
513 */
514 for (i = 0; i < 4; ++i) {
515 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_OUTPUT, i, TGSI_FILE_INPUT, i);
516 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
517 }
518
519 /*
520 * add o4, i0, i4 ; Translate vertex pos by motion vec to form first ref macroblock texcoords
521 * add o5, i0, i6 ; Translate vertex pos by motion vec to form second ref macroblock texcoords
522 */
523 for (i = 0; i < 2; ++i) {
524 inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, i + 4, TGSI_FILE_INPUT, 0, TGSI_FILE_INPUT, (i + 2) * 2);
525 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
526 }
527
528 /* end */
529 inst = vl_end();
530 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
531
532 assert(ti <= max_tokens);
533
534 vs.tokens = tokens;
535 r->b_vs[0] = r->pipe->create_vs_state(r->pipe, &vs);
536 free(tokens);
537 }
538
539 static void
540 create_field_bi_pred_vert_shader(struct vl_mpeg12_mc_renderer *r)
541 {
542 assert(false);
543 }
544
545 static void
546 create_frame_bi_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
547 {
548 const unsigned max_tokens = 100;
549
550 struct pipe_shader_state fs;
551 struct tgsi_token *tokens;
552 struct tgsi_header *header;
553
554 struct tgsi_full_declaration decl;
555 struct tgsi_full_instruction inst;
556
557 unsigned ti;
558
559 unsigned i;
560
561 assert(r);
562
563 tokens = (struct tgsi_token *) malloc(max_tokens * sizeof(struct tgsi_token));
564 *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
565 header = (struct tgsi_header *) &tokens[1];
566 *header = tgsi_build_header();
567 *(struct tgsi_processor *) &tokens[2] = tgsi_build_processor(TGSI_PROCESSOR_FRAGMENT, header);
568
569 ti = 3;
570
571 /*
572 * decl i0 ; Luma texcoords
573 * decl i1 ; Chroma Cb texcoords
574 * decl i2 ; Chroma Cr texcoords
575 * decl i3 ; First ref macroblock texcoords
576 * decl i4 ; Second ref macroblock texcoords
577 */
578 for (i = 0; i < 5; ++i) {
579 decl = vl_decl_interpolated_input(TGSI_SEMANTIC_GENERIC, i + 1, i, i, TGSI_INTERPOLATE_LINEAR);
580 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
581 }
582
583 /*
584 * decl c0 ; Scaling factor, rescales 16-bit snorm to 9-bit snorm
585 * decl c1 ; Constant 1/2 in .x channel to use as weight to blend past and future texels
586 */
587 decl = vl_decl_constants(TGSI_SEMANTIC_GENERIC, 0, 0, 1);
588 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
589
590 /* decl o0 ; Fragment color */
591 decl = vl_decl_output(TGSI_SEMANTIC_COLOR, 0, 0, 0);
592 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
593
594 /* decl t0-t2 */
595 decl = vl_decl_temps(0, 2);
596 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
597
598 /*
599 * decl s0 ; Sampler for luma texture
600 * decl s1 ; Sampler for chroma Cb texture
601 * decl s2 ; Sampler for chroma Cr texture
602 * decl s3 ; Sampler for first ref surface texture
603 * decl s4 ; Sampler for second ref surface texture
604 */
605 for (i = 0; i < 5; ++i) {
606 decl = vl_decl_samplers(i, i);
607 ti += tgsi_build_full_declaration(&decl, &tokens[ti], header, max_tokens - ti);
608 }
609
610 /*
611 * tex2d t1, i0, s0 ; Read texel from luma texture
612 * mov t0.x, t1.x ; Move luma sample into .x component
613 * tex2d t1, i1, s1 ; Read texel from chroma Cb texture
614 * mov t0.y, t1.x ; Move Cb sample into .y component
615 * tex2d t1, i2, s2 ; Read texel from chroma Cr texture
616 * mov t0.z, t1.x ; Move Cr sample into .z component
617 */
618 for (i = 0; i < 3; ++i) {
619 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_INPUT, i, TGSI_FILE_SAMPLER, i);
620 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
621
622 inst = vl_inst2(TGSI_OPCODE_MOV, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
623 inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
624 inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
625 inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
626 inst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X << i;
627 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
628 }
629
630 /* mul t0, t0, c0 ; Rescale texel to correct range */
631 inst = vl_inst3(TGSI_OPCODE_MUL, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_CONSTANT, 0);
632 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
633
634 /*
635 * tex2d t1, i3, s3 ; Read texel from first ref macroblock
636 * tex2d t2, i4, s4 ; Read texel from second ref macroblock
637 */
638 for (i = 0; i < 2; ++i) {
639 inst = vl_tex(TGSI_TEXTURE_2D, TGSI_FILE_TEMPORARY, i + 1, TGSI_FILE_INPUT, i + 3, TGSI_FILE_SAMPLER, i + 3);
640 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
641 }
642
643 /* lerp t1, c1.x, t1, t2 ; Blend past and future texels */
644 inst = vl_inst4(TGSI_OPCODE_LRP, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_CONSTANT, 1, TGSI_FILE_TEMPORARY, 1, TGSI_FILE_TEMPORARY, 2);
645 inst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
646 inst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
647 inst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_X;
648 inst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_X;
649 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
650
651 /* add o0, t0, t1 ; Add past/future ref and differential to form final output */
652 inst = vl_inst3(TGSI_OPCODE_ADD, TGSI_FILE_OUTPUT, 0, TGSI_FILE_TEMPORARY, 0, TGSI_FILE_TEMPORARY, 1);
653 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
654
655 /* end */
656 inst = vl_end();
657 ti += tgsi_build_full_instruction(&inst, &tokens[ti], header, max_tokens - ti);
658
659 assert(ti <= max_tokens);
660
661 fs.tokens = tokens;
662 r->b_fs[0] = r->pipe->create_fs_state(r->pipe, &fs);
663 free(tokens);
664 }
665
666 static void
667 create_field_bi_pred_frag_shader(struct vl_mpeg12_mc_renderer *r)
668 {
669 assert(false);
670 }
671
672 static void
673 xfer_buffers_map(struct vl_mpeg12_mc_renderer *r)
674 {
675 unsigned i;
676
677 assert(r);
678
679 for (i = 0; i < 3; ++i) {
680 r->tex_transfer[i] = r->pipe->screen->get_tex_transfer
681 (
682 r->pipe->screen, r->textures.all[i],
683 0, 0, 0, PIPE_TRANSFER_WRITE, 0, 0,
684 r->textures.all[i]->width[0], r->textures.all[i]->height[0]
685 );
686
687 r->texels[i] = r->pipe->screen->transfer_map(r->pipe->screen, r->tex_transfer[i]);
688 }
689 }
690
691 static void
692 xfer_buffers_unmap(struct vl_mpeg12_mc_renderer *r)
693 {
694 unsigned i;
695
696 assert(r);
697
698 for (i = 0; i < 3; ++i) {
699 r->pipe->screen->transfer_unmap(r->pipe->screen, r->tex_transfer[i]);
700 r->pipe->screen->tex_transfer_destroy(r->tex_transfer[i]);
701 }
702 }
703
704 static bool
705 init_pipe_state(struct vl_mpeg12_mc_renderer *r)
706 {
707 struct pipe_sampler_state sampler;
708 unsigned filters[5];
709 unsigned i;
710
711 assert(r);
712
713 r->viewport.scale[0] = r->pot_buffers ?
714 util_next_power_of_two(r->picture_width) : r->picture_width;
715 r->viewport.scale[1] = r->pot_buffers ?
716 util_next_power_of_two(r->picture_height) : r->picture_height;
717 r->viewport.scale[2] = 1;
718 r->viewport.scale[3] = 1;
719 r->viewport.translate[0] = 0;
720 r->viewport.translate[1] = 0;
721 r->viewport.translate[2] = 0;
722 r->viewport.translate[3] = 0;
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.width[0] = r->pot_buffers ?
834 util_next_power_of_two(r->picture_width) : r->picture_width;
835 template.height[0] = r->pot_buffers ?
836 util_next_power_of_two(r->picture_height) : r->picture_height;
837 template.depth[0] = 1;
838 pf_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.width[0] = r->pot_buffers ?
845 util_next_power_of_two(r->picture_width / 2) :
846 r->picture_width / 2;
847 template.height[0] = 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.height[0] = 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
1274 vs_consts = pipe_buffer_map
1275 (
1276 r->pipe->screen, r->vs_const_buf.buffer,
1277 PIPE_BUFFER_USAGE_CPU_WRITE | PIPE_BUFFER_USAGE_DISCARD
1278 );
1279
1280 vs_consts->denorm.x = r->surface->width[0];
1281 vs_consts->denorm.y = r->surface->height[0];
1282
1283 pipe_buffer_unmap(r->pipe->screen, r->vs_const_buf.buffer);
1284
1285 r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_VERTEX, 0,
1286 &r->vs_const_buf);
1287 r->pipe->set_constant_buffer(r->pipe, PIPE_SHADER_FRAGMENT, 0,
1288 &r->fs_const_buf);
1289
1290 if (num_macroblocks[MACROBLOCK_TYPE_INTRA] > 0) {
1291 r->pipe->set_vertex_buffers(r->pipe, 1, r->vertex_bufs.all);
1292 r->pipe->set_vertex_elements(r->pipe, 4, r->vertex_elems);
1293 r->pipe->set_sampler_textures(r->pipe, 3, r->textures.all);
1294 r->pipe->bind_sampler_states(r->pipe, 3, r->samplers.all);
1295 r->pipe->bind_vs_state(r->pipe, r->i_vs);
1296 r->pipe->bind_fs_state(r->pipe, r->i_fs);
1297
1298 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1299 num_macroblocks[MACROBLOCK_TYPE_INTRA] * 24);
1300 vb_start += num_macroblocks[MACROBLOCK_TYPE_INTRA] * 24;
1301 }
1302
1303 if (num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] > 0) {
1304 r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
1305 r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
1306 r->textures.individual.ref[0] = r->past;
1307 r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
1308 r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
1309 r->pipe->bind_vs_state(r->pipe, r->p_vs[0]);
1310 r->pipe->bind_fs_state(r->pipe, r->p_fs[0]);
1311
1312 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1313 num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] * 24);
1314 vb_start += num_macroblocks[MACROBLOCK_TYPE_FWD_FRAME_PRED] * 24;
1315 }
1316
1317 if (false /*num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] > 0 */ ) {
1318 r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
1319 r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
1320 r->textures.individual.ref[0] = r->past;
1321 r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
1322 r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
1323 r->pipe->bind_vs_state(r->pipe, r->p_vs[1]);
1324 r->pipe->bind_fs_state(r->pipe, r->p_fs[1]);
1325
1326 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1327 num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] * 24);
1328 vb_start += num_macroblocks[MACROBLOCK_TYPE_FWD_FIELD_PRED] * 24;
1329 }
1330
1331 if (num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] > 0) {
1332 r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
1333 r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
1334 r->textures.individual.ref[0] = r->future;
1335 r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
1336 r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
1337 r->pipe->bind_vs_state(r->pipe, r->p_vs[0]);
1338 r->pipe->bind_fs_state(r->pipe, r->p_fs[0]);
1339
1340 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1341 num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] * 24);
1342 vb_start += num_macroblocks[MACROBLOCK_TYPE_BKWD_FRAME_PRED] * 24;
1343 }
1344
1345 if (false /*num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] > 0 */ ) {
1346 r->pipe->set_vertex_buffers(r->pipe, 2, r->vertex_bufs.all);
1347 r->pipe->set_vertex_elements(r->pipe, 6, r->vertex_elems);
1348 r->textures.individual.ref[0] = r->future;
1349 r->pipe->set_sampler_textures(r->pipe, 4, r->textures.all);
1350 r->pipe->bind_sampler_states(r->pipe, 4, r->samplers.all);
1351 r->pipe->bind_vs_state(r->pipe, r->p_vs[1]);
1352 r->pipe->bind_fs_state(r->pipe, r->p_fs[1]);
1353
1354 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1355 num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] * 24);
1356 vb_start += num_macroblocks[MACROBLOCK_TYPE_BKWD_FIELD_PRED] * 24;
1357 }
1358
1359 if (num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] > 0) {
1360 r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
1361 r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
1362 r->textures.individual.ref[0] = r->past;
1363 r->textures.individual.ref[1] = r->future;
1364 r->pipe->set_sampler_textures(r->pipe, 5, r->textures.all);
1365 r->pipe->bind_sampler_states(r->pipe, 5, r->samplers.all);
1366 r->pipe->bind_vs_state(r->pipe, r->b_vs[0]);
1367 r->pipe->bind_fs_state(r->pipe, r->b_fs[0]);
1368
1369 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1370 num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] * 24);
1371 vb_start += num_macroblocks[MACROBLOCK_TYPE_BI_FRAME_PRED] * 24;
1372 }
1373
1374 if (false /*num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] > 0 */ ) {
1375 r->pipe->set_vertex_buffers(r->pipe, 3, r->vertex_bufs.all);
1376 r->pipe->set_vertex_elements(r->pipe, 8, r->vertex_elems);
1377 r->textures.individual.ref[0] = r->past;
1378 r->textures.individual.ref[1] = r->future;
1379 r->pipe->set_sampler_textures(r->pipe, 5, r->textures.all);
1380 r->pipe->bind_sampler_states(r->pipe, 5, r->samplers.all);
1381 r->pipe->bind_vs_state(r->pipe, r->b_vs[1]);
1382 r->pipe->bind_fs_state(r->pipe, r->b_fs[1]);
1383
1384 r->pipe->draw_arrays(r->pipe, PIPE_PRIM_TRIANGLES, vb_start,
1385 num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] * 24);
1386 vb_start += num_macroblocks[MACROBLOCK_TYPE_BI_FIELD_PRED] * 24;
1387 }
1388
1389 r->pipe->flush(r->pipe, PIPE_FLUSH_RENDER_CACHE, r->fence);
1390 pipe_surface_reference(&r->fb_state.cbufs[0], NULL);
1391
1392 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE)
1393 for (i = 0; i < 3; ++i)
1394 r->zero_block[i].x = ZERO_BLOCK_NIL;
1395
1396 r->num_macroblocks = 0;
1397 }
1398
1399 static void
1400 grab_frame_coded_block(short *src, short *dst, unsigned dst_pitch)
1401 {
1402 unsigned y;
1403
1404 assert(src);
1405 assert(dst);
1406
1407 for (y = 0; y < BLOCK_HEIGHT; ++y)
1408 memcpy(dst + y * dst_pitch, src + y * BLOCK_WIDTH, BLOCK_WIDTH * 2);
1409 }
1410
1411 static void
1412 grab_field_coded_block(short *src, short *dst, unsigned dst_pitch)
1413 {
1414 unsigned y;
1415
1416 assert(src);
1417 assert(dst);
1418
1419 for (y = 0; y < BLOCK_HEIGHT; ++y)
1420 memcpy(dst + y * dst_pitch * 2, src + y * BLOCK_WIDTH, BLOCK_WIDTH * 2);
1421 }
1422
1423 static void
1424 fill_zero_block(short *dst, unsigned dst_pitch)
1425 {
1426 unsigned y;
1427
1428 assert(dst);
1429
1430 for (y = 0; y < BLOCK_HEIGHT; ++y)
1431 memset(dst + y * dst_pitch, 0, BLOCK_WIDTH * 2);
1432 }
1433
1434 static void
1435 grab_blocks(struct vl_mpeg12_mc_renderer *r, unsigned mbx, unsigned mby,
1436 enum pipe_mpeg12_dct_type dct_type, unsigned cbp, short *blocks)
1437 {
1438 unsigned tex_pitch;
1439 short *texels;
1440 unsigned tb = 0, sb = 0;
1441 unsigned mbpx = mbx * MACROBLOCK_WIDTH, mbpy = mby * MACROBLOCK_HEIGHT;
1442 unsigned x, y;
1443
1444 assert(r);
1445 assert(blocks);
1446
1447 tex_pitch = r->tex_transfer[0]->stride / r->tex_transfer[0]->block.size;
1448 texels = r->texels[0] + mbpy * tex_pitch + mbpx;
1449
1450 for (y = 0; y < 2; ++y) {
1451 for (x = 0; x < 2; ++x, ++tb) {
1452 if ((cbp >> (5 - tb)) & 1) {
1453 if (dct_type == PIPE_MPEG12_DCT_TYPE_FRAME) {
1454 grab_frame_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT,
1455 texels + y * tex_pitch * BLOCK_WIDTH +
1456 x * BLOCK_WIDTH, tex_pitch);
1457 }
1458 else {
1459 grab_field_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT,
1460 texels + y * tex_pitch + x * BLOCK_WIDTH,
1461 tex_pitch);
1462 }
1463
1464 ++sb;
1465 }
1466 else if (r->eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE) {
1467 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL ||
1468 ZERO_BLOCK_IS_NIL(r->zero_block[0])) {
1469 fill_zero_block(texels + y * tex_pitch * BLOCK_WIDTH + x * BLOCK_WIDTH, tex_pitch);
1470 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE) {
1471 r->zero_block[0].x = (mbpx + x * 8) * r->surface_tex_inv_size.x;
1472 r->zero_block[0].y = (mbpy + y * 8) * r->surface_tex_inv_size.y;
1473 }
1474 }
1475 }
1476 }
1477 }
1478
1479 /* TODO: Implement 422, 444 */
1480 assert(r->chroma_format == PIPE_VIDEO_CHROMA_FORMAT_420);
1481
1482 mbpx /= 2;
1483 mbpy /= 2;
1484
1485 for (tb = 0; tb < 2; ++tb) {
1486 tex_pitch = r->tex_transfer[tb + 1]->stride / r->tex_transfer[tb + 1]->block.size;
1487 texels = r->texels[tb + 1] + mbpy * tex_pitch + mbpx;
1488
1489 if ((cbp >> (1 - tb)) & 1) {
1490 grab_frame_coded_block(blocks + sb * BLOCK_WIDTH * BLOCK_HEIGHT, texels, tex_pitch);
1491 ++sb;
1492 }
1493 else if (r->eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE) {
1494 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ALL ||
1495 ZERO_BLOCK_IS_NIL(r->zero_block[tb + 1])) {
1496 fill_zero_block(texels, tex_pitch);
1497 if (r->eb_handling == VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_ONE) {
1498 r->zero_block[tb + 1].x = (mbpx << 1) * r->surface_tex_inv_size.x;
1499 r->zero_block[tb + 1].y = (mbpy << 1) * r->surface_tex_inv_size.y;
1500 }
1501 }
1502 }
1503 }
1504 }
1505
1506 static void
1507 grab_macroblock(struct vl_mpeg12_mc_renderer *r,
1508 struct pipe_mpeg12_macroblock *mb)
1509 {
1510 assert(r);
1511 assert(mb);
1512 assert(r->num_macroblocks < r->macroblocks_per_batch);
1513
1514 memcpy(&r->macroblock_buf[r->num_macroblocks], mb,
1515 sizeof(struct pipe_mpeg12_macroblock));
1516
1517 grab_blocks(r, mb->mbx, mb->mby, mb->dct_type, mb->cbp, mb->blocks);
1518
1519 ++r->num_macroblocks;
1520 }
1521
1522 bool
1523 vl_mpeg12_mc_renderer_init(struct vl_mpeg12_mc_renderer *renderer,
1524 struct pipe_context *pipe,
1525 unsigned picture_width,
1526 unsigned picture_height,
1527 enum pipe_video_chroma_format chroma_format,
1528 enum VL_MPEG12_MC_RENDERER_BUFFER_MODE bufmode,
1529 enum VL_MPEG12_MC_RENDERER_EMPTY_BLOCK eb_handling,
1530 bool pot_buffers)
1531 {
1532 unsigned i;
1533
1534 assert(renderer);
1535 assert(pipe);
1536 /* TODO: Implement other policies */
1537 assert(bufmode == VL_MPEG12_MC_RENDERER_BUFFER_PICTURE);
1538 /* TODO: Implement this */
1539 /* XXX: XFER_ALL sampling issue at block edges when using bilinear filtering */
1540 assert(eb_handling != VL_MPEG12_MC_RENDERER_EMPTY_BLOCK_XFER_NONE);
1541 /* TODO: Non-pot buffers untested, probably doesn't work without changes to texcoord generation, vert shader, etc */
1542 assert(pot_buffers);
1543
1544 memset(renderer, 0, sizeof(struct vl_mpeg12_mc_renderer));
1545
1546 renderer->pipe = pipe;
1547 renderer->picture_width = picture_width;
1548 renderer->picture_height = picture_height;
1549 renderer->chroma_format = chroma_format;
1550 renderer->bufmode = bufmode;
1551 renderer->eb_handling = eb_handling;
1552 renderer->pot_buffers = pot_buffers;
1553
1554 if (!init_pipe_state(renderer))
1555 return false;
1556 if (!init_shaders(renderer)) {
1557 cleanup_pipe_state(renderer);
1558 return false;
1559 }
1560 if (!init_buffers(renderer)) {
1561 cleanup_shaders(renderer);
1562 cleanup_pipe_state(renderer);
1563 return false;
1564 }
1565
1566 renderer->surface = NULL;
1567 renderer->past = NULL;
1568 renderer->future = NULL;
1569 for (i = 0; i < 3; ++i)
1570 renderer->zero_block[i].x = ZERO_BLOCK_NIL;
1571 renderer->num_macroblocks = 0;
1572
1573 xfer_buffers_map(renderer);
1574
1575 return true;
1576 }
1577
1578 void
1579 vl_mpeg12_mc_renderer_cleanup(struct vl_mpeg12_mc_renderer *renderer)
1580 {
1581 assert(renderer);
1582
1583 xfer_buffers_unmap(renderer);
1584
1585 cleanup_pipe_state(renderer);
1586 cleanup_shaders(renderer);
1587 cleanup_buffers(renderer);
1588 }
1589
1590 void
1591 vl_mpeg12_mc_renderer_render_macroblocks(struct vl_mpeg12_mc_renderer
1592 *renderer,
1593 struct pipe_texture *surface,
1594 struct pipe_texture *past,
1595 struct pipe_texture *future,
1596 unsigned num_macroblocks,
1597 struct pipe_mpeg12_macroblock
1598 *mpeg12_macroblocks,
1599 struct pipe_fence_handle **fence)
1600 {
1601 bool new_surface = false;
1602
1603 assert(renderer);
1604 assert(surface);
1605 assert(num_macroblocks);
1606 assert(mpeg12_macroblocks);
1607
1608 if (renderer->surface) {
1609 if (surface != renderer->surface) {
1610 if (renderer->num_macroblocks > 0) {
1611 xfer_buffers_unmap(renderer);
1612 flush(renderer);
1613 }
1614
1615 new_surface = true;
1616 }
1617
1618 /* If the surface we're rendering hasn't changed the ref frames shouldn't change. */
1619 assert(surface != renderer->surface || renderer->past == past);
1620 assert(surface != renderer->surface || renderer->future == future);
1621 }
1622 else
1623 new_surface = true;
1624
1625 if (new_surface) {
1626 renderer->surface = surface;
1627 renderer->past = past;
1628 renderer->future = future;
1629 renderer->fence = fence;
1630 renderer->surface_tex_inv_size.x = 1.0f / surface->width[0];
1631 renderer->surface_tex_inv_size.y = 1.0f / surface->height[0];
1632 }
1633
1634 while (num_macroblocks) {
1635 unsigned left_in_batch = renderer->macroblocks_per_batch - renderer->num_macroblocks;
1636 unsigned num_to_submit = MIN2(num_macroblocks, left_in_batch);
1637 unsigned i;
1638
1639 for (i = 0; i < num_to_submit; ++i) {
1640 assert(mpeg12_macroblocks[i].base.codec == PIPE_VIDEO_CODEC_MPEG12);
1641 grab_macroblock(renderer, &mpeg12_macroblocks[i]);
1642 }
1643
1644 num_macroblocks -= num_to_submit;
1645
1646 if (renderer->num_macroblocks == renderer->macroblocks_per_batch) {
1647 xfer_buffers_unmap(renderer);
1648 flush(renderer);
1649 xfer_buffers_map(renderer);
1650 /* Next time we get this surface it may have new ref frames */
1651 renderer->surface = NULL;
1652 }
1653 }
1654 }