Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
4358 | Serge | 1 | /* |
2 | * Mesa 3-D graphics library |
||
3 | * |
||
4 | * Copyright (C) 2012-2013 LunarG, Inc. |
||
5 | * |
||
6 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
7 | * copy of this software and associated documentation files (the "Software"), |
||
8 | * to deal in the Software without restriction, including without limitation |
||
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
||
10 | * and/or sell copies of the Software, and to permit persons to whom the |
||
11 | * Software is furnished to do so, subject to the following conditions: |
||
12 | * |
||
13 | * The above copyright notice and this permission notice shall be included |
||
14 | * in all copies or substantial portions of the Software. |
||
15 | * |
||
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
||
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
||
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
||
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING |
||
21 | * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER |
||
22 | * DEALINGS IN THE SOFTWARE. |
||
23 | * |
||
24 | * Authors: |
||
25 | * Chia-I Wu |
||
26 | */ |
||
27 | |||
28 | #include "intel_winsys.h" |
||
29 | |||
30 | #include "ilo_3d_pipeline.h" |
||
31 | #include "ilo_context.h" |
||
32 | #include "ilo_cp.h" |
||
33 | #include "ilo_query.h" |
||
34 | #include "ilo_shader.h" |
||
35 | #include "ilo_state.h" |
||
36 | #include "ilo_3d.h" |
||
37 | |||
38 | static void |
||
39 | process_query_for_occlusion_counter(struct ilo_3d *hw3d, |
||
40 | struct ilo_query *q) |
||
41 | { |
||
42 | uint64_t *vals, depth_count = 0; |
||
43 | int i; |
||
44 | |||
45 | /* in pairs */ |
||
46 | assert(q->reg_read % 2 == 0); |
||
47 | |||
48 | intel_bo_map(q->bo, false); |
||
49 | vals = intel_bo_get_virtual(q->bo); |
||
50 | for (i = 1; i < q->reg_read; i += 2) |
||
51 | depth_count += vals[i] - vals[i - 1]; |
||
52 | intel_bo_unmap(q->bo); |
||
53 | |||
54 | /* accumulate so that the query can be resumed if wanted */ |
||
55 | q->data.u64 += depth_count; |
||
56 | q->reg_read = 0; |
||
57 | } |
||
58 | |||
59 | static uint64_t |
||
60 | timestamp_to_ns(uint64_t timestamp) |
||
61 | { |
||
62 | /* see ilo_get_timestamp() */ |
||
63 | return (timestamp & 0xffffffff) * 80; |
||
64 | } |
||
65 | |||
66 | static void |
||
67 | process_query_for_timestamp(struct ilo_3d *hw3d, struct ilo_query *q) |
||
68 | { |
||
69 | uint64_t *vals, timestamp; |
||
70 | |||
71 | assert(q->reg_read == 1); |
||
72 | |||
73 | intel_bo_map(q->bo, false); |
||
74 | vals = intel_bo_get_virtual(q->bo); |
||
75 | timestamp = vals[0]; |
||
76 | intel_bo_unmap(q->bo); |
||
77 | |||
78 | q->data.u64 = timestamp_to_ns(timestamp); |
||
79 | q->reg_read = 0; |
||
80 | } |
||
81 | |||
82 | static void |
||
83 | process_query_for_time_elapsed(struct ilo_3d *hw3d, struct ilo_query *q) |
||
84 | { |
||
85 | uint64_t *vals, elapsed = 0; |
||
86 | int i; |
||
87 | |||
88 | /* in pairs */ |
||
89 | assert(q->reg_read % 2 == 0); |
||
90 | |||
91 | intel_bo_map(q->bo, false); |
||
92 | vals = intel_bo_get_virtual(q->bo); |
||
93 | |||
94 | for (i = 1; i < q->reg_read; i += 2) |
||
95 | elapsed += vals[i] - vals[i - 1]; |
||
96 | |||
97 | intel_bo_unmap(q->bo); |
||
98 | |||
99 | /* accumulate so that the query can be resumed if wanted */ |
||
100 | q->data.u64 += timestamp_to_ns(elapsed); |
||
101 | q->reg_read = 0; |
||
102 | } |
||
103 | |||
104 | static void |
||
105 | ilo_3d_resume_queries(struct ilo_3d *hw3d) |
||
106 | { |
||
107 | struct ilo_query *q; |
||
108 | |||
109 | /* resume occlusion queries */ |
||
110 | LIST_FOR_EACH_ENTRY(q, &hw3d->occlusion_queries, list) { |
||
111 | /* accumulate the result if the bo is alreay full */ |
||
112 | if (q->reg_read >= q->reg_total) |
||
113 | process_query_for_occlusion_counter(hw3d, q); |
||
114 | |||
115 | ilo_3d_pipeline_emit_write_depth_count(hw3d->pipeline, |
||
116 | q->bo, q->reg_read++); |
||
117 | } |
||
118 | |||
119 | /* resume timer queries */ |
||
120 | LIST_FOR_EACH_ENTRY(q, &hw3d->time_elapsed_queries, list) { |
||
121 | /* accumulate the result if the bo is alreay full */ |
||
122 | if (q->reg_read >= q->reg_total) |
||
123 | process_query_for_time_elapsed(hw3d, q); |
||
124 | |||
125 | ilo_3d_pipeline_emit_write_timestamp(hw3d->pipeline, |
||
126 | q->bo, q->reg_read++); |
||
127 | } |
||
128 | } |
||
129 | |||
130 | static void |
||
131 | ilo_3d_pause_queries(struct ilo_3d *hw3d) |
||
132 | { |
||
133 | struct ilo_query *q; |
||
134 | |||
135 | /* pause occlusion queries */ |
||
136 | LIST_FOR_EACH_ENTRY(q, &hw3d->occlusion_queries, list) { |
||
137 | assert(q->reg_read < q->reg_total); |
||
138 | ilo_3d_pipeline_emit_write_depth_count(hw3d->pipeline, |
||
139 | q->bo, q->reg_read++); |
||
140 | } |
||
141 | |||
142 | /* pause timer queries */ |
||
143 | LIST_FOR_EACH_ENTRY(q, &hw3d->time_elapsed_queries, list) { |
||
144 | assert(q->reg_read < q->reg_total); |
||
145 | ilo_3d_pipeline_emit_write_timestamp(hw3d->pipeline, |
||
146 | q->bo, q->reg_read++); |
||
147 | } |
||
148 | } |
||
149 | |||
150 | static void |
||
151 | ilo_3d_release_render_ring(struct ilo_cp *cp, void *data) |
||
152 | { |
||
153 | struct ilo_3d *hw3d = data; |
||
154 | |||
155 | ilo_3d_pause_queries(hw3d); |
||
156 | } |
||
157 | |||
158 | static void |
||
159 | ilo_3d_own_render_ring(struct ilo_3d *hw3d) |
||
160 | { |
||
161 | ilo_cp_set_ring(hw3d->cp, ILO_CP_RING_RENDER); |
||
162 | |||
163 | if (ilo_cp_set_owner(hw3d->cp, &hw3d->owner, hw3d->owner_reserve)) |
||
164 | ilo_3d_resume_queries(hw3d); |
||
165 | } |
||
166 | |||
167 | /** |
||
168 | * Begin a query. |
||
169 | */ |
||
170 | void |
||
171 | ilo_3d_begin_query(struct ilo_context *ilo, struct ilo_query *q) |
||
172 | { |
||
173 | struct ilo_3d *hw3d = ilo->hw3d; |
||
174 | |||
175 | ilo_3d_own_render_ring(hw3d); |
||
176 | |||
177 | switch (q->type) { |
||
178 | case PIPE_QUERY_OCCLUSION_COUNTER: |
||
179 | /* reserve some space for pausing the query */ |
||
180 | q->reg_cmd_size = ilo_3d_pipeline_estimate_size(hw3d->pipeline, |
||
181 | ILO_3D_PIPELINE_WRITE_DEPTH_COUNT, NULL); |
||
182 | hw3d->owner_reserve += q->reg_cmd_size; |
||
183 | ilo_cp_set_owner(hw3d->cp, &hw3d->owner, hw3d->owner_reserve); |
||
184 | |||
185 | q->data.u64 = 0; |
||
186 | |||
187 | if (ilo_query_alloc_bo(q, 2, -1, hw3d->cp->winsys)) { |
||
188 | /* XXX we should check the aperture size */ |
||
189 | ilo_3d_pipeline_emit_write_depth_count(hw3d->pipeline, |
||
190 | q->bo, q->reg_read++); |
||
191 | |||
192 | list_add(&q->list, &hw3d->occlusion_queries); |
||
193 | } |
||
194 | break; |
||
195 | case PIPE_QUERY_TIMESTAMP: |
||
196 | /* nop */ |
||
197 | break; |
||
198 | case PIPE_QUERY_TIME_ELAPSED: |
||
199 | /* reserve some space for pausing the query */ |
||
200 | q->reg_cmd_size = ilo_3d_pipeline_estimate_size(hw3d->pipeline, |
||
201 | ILO_3D_PIPELINE_WRITE_TIMESTAMP, NULL); |
||
202 | hw3d->owner_reserve += q->reg_cmd_size; |
||
203 | ilo_cp_set_owner(hw3d->cp, &hw3d->owner, hw3d->owner_reserve); |
||
204 | |||
205 | q->data.u64 = 0; |
||
206 | |||
207 | if (ilo_query_alloc_bo(q, 2, -1, hw3d->cp->winsys)) { |
||
208 | /* XXX we should check the aperture size */ |
||
209 | ilo_3d_pipeline_emit_write_timestamp(hw3d->pipeline, |
||
210 | q->bo, q->reg_read++); |
||
211 | |||
212 | list_add(&q->list, &hw3d->time_elapsed_queries); |
||
213 | } |
||
214 | break; |
||
215 | case PIPE_QUERY_PRIMITIVES_GENERATED: |
||
216 | q->data.u64 = 0; |
||
217 | list_add(&q->list, &hw3d->prim_generated_queries); |
||
218 | break; |
||
219 | case PIPE_QUERY_PRIMITIVES_EMITTED: |
||
220 | q->data.u64 = 0; |
||
221 | list_add(&q->list, &hw3d->prim_emitted_queries); |
||
222 | break; |
||
223 | default: |
||
224 | assert(!"unknown query type"); |
||
225 | break; |
||
226 | } |
||
227 | } |
||
228 | |||
229 | /** |
||
230 | * End a query. |
||
231 | */ |
||
232 | void |
||
233 | ilo_3d_end_query(struct ilo_context *ilo, struct ilo_query *q) |
||
234 | { |
||
235 | struct ilo_3d *hw3d = ilo->hw3d; |
||
236 | |||
237 | ilo_3d_own_render_ring(hw3d); |
||
238 | |||
239 | switch (q->type) { |
||
240 | case PIPE_QUERY_OCCLUSION_COUNTER: |
||
241 | list_del(&q->list); |
||
242 | |||
243 | assert(q->reg_read < q->reg_total); |
||
244 | hw3d->owner_reserve -= q->reg_cmd_size; |
||
245 | ilo_cp_set_owner(hw3d->cp, &hw3d->owner, hw3d->owner_reserve); |
||
246 | ilo_3d_pipeline_emit_write_depth_count(hw3d->pipeline, |
||
247 | q->bo, q->reg_read++); |
||
248 | break; |
||
249 | case PIPE_QUERY_TIMESTAMP: |
||
250 | q->data.u64 = 0; |
||
251 | |||
252 | if (ilo_query_alloc_bo(q, 1, 1, hw3d->cp->winsys)) { |
||
253 | ilo_3d_pipeline_emit_write_timestamp(hw3d->pipeline, |
||
254 | q->bo, q->reg_read++); |
||
255 | } |
||
256 | break; |
||
257 | case PIPE_QUERY_TIME_ELAPSED: |
||
258 | list_del(&q->list); |
||
259 | |||
260 | assert(q->reg_read < q->reg_total); |
||
261 | hw3d->owner_reserve -= q->reg_cmd_size; |
||
262 | ilo_cp_set_owner(hw3d->cp, &hw3d->owner, hw3d->owner_reserve); |
||
263 | ilo_3d_pipeline_emit_write_timestamp(hw3d->pipeline, |
||
264 | q->bo, q->reg_read++); |
||
265 | break; |
||
266 | case PIPE_QUERY_PRIMITIVES_GENERATED: |
||
267 | case PIPE_QUERY_PRIMITIVES_EMITTED: |
||
268 | list_del(&q->list); |
||
269 | break; |
||
270 | default: |
||
271 | assert(!"unknown query type"); |
||
272 | break; |
||
273 | } |
||
274 | } |
||
275 | |||
276 | /** |
||
277 | * Process the raw query data. |
||
278 | */ |
||
279 | void |
||
280 | ilo_3d_process_query(struct ilo_context *ilo, struct ilo_query *q) |
||
281 | { |
||
282 | struct ilo_3d *hw3d = ilo->hw3d; |
||
283 | |||
284 | switch (q->type) { |
||
285 | case PIPE_QUERY_OCCLUSION_COUNTER: |
||
286 | if (q->bo) |
||
287 | process_query_for_occlusion_counter(hw3d, q); |
||
288 | break; |
||
289 | case PIPE_QUERY_TIMESTAMP: |
||
290 | if (q->bo) |
||
291 | process_query_for_timestamp(hw3d, q); |
||
292 | break; |
||
293 | case PIPE_QUERY_TIME_ELAPSED: |
||
294 | if (q->bo) |
||
295 | process_query_for_time_elapsed(hw3d, q); |
||
296 | break; |
||
297 | case PIPE_QUERY_PRIMITIVES_GENERATED: |
||
298 | case PIPE_QUERY_PRIMITIVES_EMITTED: |
||
299 | break; |
||
300 | default: |
||
301 | assert(!"unknown query type"); |
||
302 | break; |
||
303 | } |
||
304 | } |
||
305 | |||
306 | /** |
||
307 | * Hook for CP new-batch. |
||
308 | */ |
||
309 | void |
||
310 | ilo_3d_cp_flushed(struct ilo_3d *hw3d) |
||
311 | { |
||
312 | if (ilo_debug & ILO_DEBUG_3D) |
||
313 | ilo_3d_pipeline_dump(hw3d->pipeline); |
||
314 | |||
315 | /* invalidate the pipeline */ |
||
316 | ilo_3d_pipeline_invalidate(hw3d->pipeline, |
||
317 | ILO_3D_PIPELINE_INVALIDATE_BATCH_BO | |
||
318 | ILO_3D_PIPELINE_INVALIDATE_STATE_BO); |
||
319 | if (!hw3d->cp->render_ctx) { |
||
320 | ilo_3d_pipeline_invalidate(hw3d->pipeline, |
||
321 | ILO_3D_PIPELINE_INVALIDATE_HW); |
||
322 | } |
||
323 | |||
324 | hw3d->new_batch = true; |
||
325 | } |
||
326 | |||
327 | /** |
||
328 | * Create a 3D context. |
||
329 | */ |
||
330 | struct ilo_3d * |
||
331 | ilo_3d_create(struct ilo_cp *cp, const struct ilo_dev_info *dev) |
||
332 | { |
||
333 | struct ilo_3d *hw3d; |
||
334 | |||
335 | hw3d = CALLOC_STRUCT(ilo_3d); |
||
336 | if (!hw3d) |
||
337 | return NULL; |
||
338 | |||
339 | hw3d->cp = cp; |
||
340 | hw3d->owner.release_callback = ilo_3d_release_render_ring; |
||
341 | hw3d->owner.release_data = hw3d; |
||
342 | |||
343 | hw3d->new_batch = true; |
||
344 | |||
345 | list_inithead(&hw3d->occlusion_queries); |
||
346 | list_inithead(&hw3d->time_elapsed_queries); |
||
347 | list_inithead(&hw3d->prim_generated_queries); |
||
348 | list_inithead(&hw3d->prim_emitted_queries); |
||
349 | |||
350 | hw3d->pipeline = ilo_3d_pipeline_create(cp, dev); |
||
351 | if (!hw3d->pipeline) { |
||
352 | FREE(hw3d); |
||
353 | return NULL; |
||
354 | } |
||
355 | |||
356 | return hw3d; |
||
357 | } |
||
358 | |||
359 | /** |
||
360 | * Destroy a 3D context. |
||
361 | */ |
||
362 | void |
||
363 | ilo_3d_destroy(struct ilo_3d *hw3d) |
||
364 | { |
||
365 | ilo_3d_pipeline_destroy(hw3d->pipeline); |
||
366 | |||
367 | if (hw3d->kernel.bo) |
||
368 | intel_bo_unreference(hw3d->kernel.bo); |
||
369 | |||
370 | FREE(hw3d); |
||
371 | } |
||
372 | |||
373 | static bool |
||
374 | draw_vbo(struct ilo_3d *hw3d, const struct ilo_context *ilo, |
||
375 | int *prim_generated, int *prim_emitted) |
||
376 | { |
||
377 | bool need_flush = false; |
||
378 | int max_len; |
||
379 | |||
380 | ilo_3d_own_render_ring(hw3d); |
||
381 | |||
382 | if (!hw3d->new_batch) { |
||
383 | /* |
||
384 | * Without a better tracking mechanism, when the framebuffer changes, we |
||
385 | * have to assume that the old framebuffer may be sampled from. If that |
||
386 | * happens in the middle of a batch buffer, we need to insert manual |
||
387 | * flushes. |
||
388 | */ |
||
389 | need_flush = (ilo->dirty & ILO_DIRTY_FB); |
||
390 | |||
391 | /* same to SO target changes */ |
||
392 | need_flush |= (ilo->dirty & ILO_DIRTY_SO); |
||
393 | } |
||
394 | |||
395 | /* make sure there is enough room first */ |
||
396 | max_len = ilo_3d_pipeline_estimate_size(hw3d->pipeline, |
||
397 | ILO_3D_PIPELINE_DRAW, ilo); |
||
398 | if (need_flush) { |
||
399 | max_len += ilo_3d_pipeline_estimate_size(hw3d->pipeline, |
||
400 | ILO_3D_PIPELINE_FLUSH, NULL); |
||
401 | } |
||
402 | |||
403 | if (max_len > ilo_cp_space(hw3d->cp)) { |
||
404 | ilo_cp_flush(hw3d->cp); |
||
405 | need_flush = false; |
||
406 | assert(max_len <= ilo_cp_space(hw3d->cp)); |
||
407 | } |
||
408 | |||
409 | if (need_flush) |
||
410 | ilo_3d_pipeline_emit_flush(hw3d->pipeline); |
||
411 | |||
412 | return ilo_3d_pipeline_emit_draw(hw3d->pipeline, ilo, |
||
413 | prim_generated, prim_emitted); |
||
414 | } |
||
415 | |||
416 | static void |
||
417 | update_prim_count(struct ilo_3d *hw3d, int generated, int emitted) |
||
418 | { |
||
419 | struct ilo_query *q; |
||
420 | |||
421 | LIST_FOR_EACH_ENTRY(q, &hw3d->prim_generated_queries, list) |
||
422 | q->data.u64 += generated; |
||
423 | |||
424 | LIST_FOR_EACH_ENTRY(q, &hw3d->prim_emitted_queries, list) |
||
425 | q->data.u64 += emitted; |
||
426 | } |
||
427 | |||
428 | bool |
||
429 | ilo_3d_pass_render_condition(struct ilo_context *ilo) |
||
430 | { |
||
431 | struct ilo_3d *hw3d = ilo->hw3d; |
||
432 | uint64_t result; |
||
433 | bool wait; |
||
434 | |||
435 | if (!hw3d->render_condition.query) |
||
436 | return true; |
||
437 | |||
438 | switch (hw3d->render_condition.mode) { |
||
439 | case PIPE_RENDER_COND_WAIT: |
||
440 | case PIPE_RENDER_COND_BY_REGION_WAIT: |
||
441 | wait = true; |
||
442 | break; |
||
443 | case PIPE_RENDER_COND_NO_WAIT: |
||
444 | case PIPE_RENDER_COND_BY_REGION_NO_WAIT: |
||
445 | default: |
||
446 | wait = false; |
||
447 | break; |
||
448 | } |
||
449 | |||
450 | if (ilo->base.get_query_result(&ilo->base, hw3d->render_condition.query, |
||
451 | wait, (union pipe_query_result *) &result)) |
||
452 | return (!result == hw3d->render_condition.cond); |
||
453 | else |
||
454 | return true; |
||
455 | } |
||
456 | |||
457 | #define UPDATE_MIN2(a, b) (a) = MIN2((a), (b)) |
||
458 | #define UPDATE_MAX2(a, b) (a) = MAX2((a), (b)) |
||
459 | |||
460 | /** |
||
461 | * \see find_sub_primitives() from core mesa |
||
462 | */ |
||
463 | static int |
||
464 | ilo_find_sub_primitives(const void *elements, unsigned element_size, |
||
465 | const struct pipe_draw_info *orig_info, |
||
466 | struct pipe_draw_info *info) |
||
467 | { |
||
468 | const unsigned max_prims = orig_info->count - orig_info->start; |
||
469 | unsigned i, cur_start, cur_count; |
||
470 | int scan_index; |
||
471 | unsigned scan_num; |
||
472 | |||
473 | cur_start = orig_info->start; |
||
474 | cur_count = 0; |
||
475 | scan_num = 0; |
||
476 | |||
477 | #define IB_INDEX_READ(TYPE, INDEX) (((const TYPE *) elements)[INDEX]) |
||
478 | |||
479 | #define SCAN_ELEMENTS(TYPE) \ |
||
480 | info[scan_num] = *orig_info; \ |
||
481 | info[scan_num].primitive_restart = false; \ |
||
482 | for (i = orig_info->start; i < orig_info->count; i++) { \ |
||
483 | scan_index = IB_INDEX_READ(TYPE, i); \ |
||
484 | if (scan_index == orig_info->restart_index) { \ |
||
485 | if (cur_count > 0) { \ |
||
486 | assert(scan_num < max_prims); \ |
||
487 | info[scan_num].start = cur_start; \ |
||
488 | info[scan_num].count = cur_count; \ |
||
489 | scan_num++; \ |
||
490 | info[scan_num] = *orig_info; \ |
||
491 | info[scan_num].primitive_restart = false; \ |
||
492 | } \ |
||
493 | cur_start = i + 1; \ |
||
494 | cur_count = 0; \ |
||
495 | } \ |
||
496 | else { \ |
||
497 | UPDATE_MIN2(info[scan_num].min_index, scan_index); \ |
||
498 | UPDATE_MAX2(info[scan_num].max_index, scan_index); \ |
||
499 | cur_count++; \ |
||
500 | } \ |
||
501 | } \ |
||
502 | if (cur_count > 0) { \ |
||
503 | assert(scan_num < max_prims); \ |
||
504 | info[scan_num].start = cur_start; \ |
||
505 | info[scan_num].count = cur_count; \ |
||
506 | scan_num++; \ |
||
507 | } |
||
508 | |||
509 | switch (element_size) { |
||
510 | case 1: |
||
511 | SCAN_ELEMENTS(uint8_t); |
||
512 | break; |
||
513 | case 2: |
||
514 | SCAN_ELEMENTS(uint16_t); |
||
515 | break; |
||
516 | case 4: |
||
517 | SCAN_ELEMENTS(uint32_t); |
||
518 | break; |
||
519 | default: |
||
520 | assert(0 && "bad index_size in find_sub_primitives()"); |
||
521 | } |
||
522 | |||
523 | #undef SCAN_ELEMENTS |
||
524 | |||
525 | return scan_num; |
||
526 | } |
||
527 | |||
528 | static inline bool |
||
529 | ilo_check_restart_index(const struct ilo_context *ilo, unsigned restart_index) |
||
530 | { |
||
531 | /* |
||
532 | * Haswell (GEN(7.5)) supports an arbitrary cut index, check everything |
||
533 | * older. |
||
534 | */ |
||
535 | if (ilo->dev->gen >= ILO_GEN(7.5)) |
||
536 | return true; |
||
537 | |||
538 | /* Note: indices must be unsigned byte, unsigned short or unsigned int */ |
||
539 | switch (ilo->ib.index_size) { |
||
540 | case 1: |
||
541 | return ((restart_index & 0xff) == 0xff); |
||
542 | break; |
||
543 | case 2: |
||
544 | return ((restart_index & 0xffff) == 0xffff); |
||
545 | break; |
||
546 | case 4: |
||
547 | return (restart_index == 0xffffffff); |
||
548 | break; |
||
549 | } |
||
550 | return false; |
||
551 | } |
||
552 | |||
553 | static inline bool |
||
554 | ilo_check_restart_prim_type(const struct ilo_context *ilo, unsigned prim) |
||
555 | { |
||
556 | switch (prim) { |
||
557 | case PIPE_PRIM_POINTS: |
||
558 | case PIPE_PRIM_LINES: |
||
559 | case PIPE_PRIM_LINE_STRIP: |
||
560 | case PIPE_PRIM_TRIANGLES: |
||
561 | case PIPE_PRIM_TRIANGLE_STRIP: |
||
562 | /* All 965 GEN graphics support a cut index for these primitive types */ |
||
563 | return true; |
||
564 | break; |
||
565 | |||
566 | case PIPE_PRIM_LINE_LOOP: |
||
567 | case PIPE_PRIM_POLYGON: |
||
568 | case PIPE_PRIM_QUAD_STRIP: |
||
569 | case PIPE_PRIM_QUADS: |
||
570 | case PIPE_PRIM_TRIANGLE_FAN: |
||
571 | if (ilo->dev->gen >= ILO_GEN(7.5)) { |
||
572 | /* Haswell and newer parts can handle these prim types. */ |
||
573 | return true; |
||
574 | } |
||
575 | break; |
||
576 | } |
||
577 | |||
578 | return false; |
||
579 | } |
||
580 | |||
581 | /* |
||
582 | * Handle VBOs using primitive restart. |
||
583 | * Verify that restart index and primitive type can be handled by the HW. |
||
584 | * Return true if this routine did the rendering |
||
585 | * Return false if this routine did NOT render because restart can be handled |
||
586 | * in HW. |
||
587 | */ |
||
588 | static void |
||
589 | ilo_draw_vbo_with_sw_restart(struct pipe_context *pipe, |
||
590 | const struct pipe_draw_info *info) |
||
591 | { |
||
592 | struct ilo_context *ilo = ilo_context(pipe); |
||
593 | struct pipe_draw_info *restart_info = NULL; |
||
594 | int sub_prim_count = 1; |
||
595 | |||
596 | /* |
||
597 | * We have to break up the primitive into chunks manually |
||
598 | * Worst case, every other index could be a restart index so |
||
599 | * need to have space for that many primitives |
||
600 | */ |
||
601 | restart_info = MALLOC(((info->count + 1) / 2) * sizeof(*info)); |
||
602 | if (NULL == restart_info) { |
||
603 | /* If we can't get memory for this, bail out */ |
||
604 | ilo_err("%s:%d - Out of memory", __FILE__, __LINE__); |
||
605 | return; |
||
606 | } |
||
607 | |||
608 | if (ilo->ib.buffer) { |
||
609 | struct pipe_transfer *transfer; |
||
610 | const void *map; |
||
611 | |||
612 | map = pipe_buffer_map(pipe, ilo->ib.buffer, |
||
613 | PIPE_TRANSFER_READ, &transfer); |
||
614 | |||
615 | sub_prim_count = ilo_find_sub_primitives(map + ilo->ib.offset, |
||
616 | ilo->ib.index_size, info, restart_info); |
||
617 | |||
618 | pipe_buffer_unmap(pipe, transfer); |
||
619 | } |
||
620 | else { |
||
621 | sub_prim_count = ilo_find_sub_primitives(ilo->ib.user_buffer, |
||
622 | ilo->ib.index_size, info, restart_info); |
||
623 | } |
||
624 | |||
625 | info = restart_info; |
||
626 | |||
627 | while (sub_prim_count > 0) { |
||
628 | pipe->draw_vbo(pipe, info); |
||
629 | |||
630 | sub_prim_count--; |
||
631 | info++; |
||
632 | } |
||
633 | |||
634 | FREE(restart_info); |
||
635 | } |
||
636 | |||
637 | static bool |
||
638 | upload_shaders(struct ilo_3d *hw3d, struct ilo_shader_cache *shc) |
||
639 | { |
||
640 | bool incremental = true; |
||
641 | int upload; |
||
642 | |||
643 | upload = ilo_shader_cache_upload(shc, |
||
644 | NULL, hw3d->kernel.used, incremental); |
||
645 | if (!upload) |
||
646 | return true; |
||
647 | |||
648 | /* |
||
649 | * Allocate a new bo. When this is a new batch, assume the bo is still in |
||
650 | * use by the previous batch and force allocation. |
||
651 | * |
||
652 | * Does it help to make shader cache upload with unsynchronized mapping, |
||
653 | * and remove the check for new batch here? |
||
654 | */ |
||
655 | if (hw3d->kernel.used + upload > hw3d->kernel.size || hw3d->new_batch) { |
||
656 | unsigned new_size = (hw3d->kernel.size) ? |
||
657 | hw3d->kernel.size : (8 * 1024); |
||
658 | |||
659 | while (hw3d->kernel.used + upload > new_size) |
||
660 | new_size *= 2; |
||
661 | |||
662 | if (hw3d->kernel.bo) |
||
663 | intel_bo_unreference(hw3d->kernel.bo); |
||
664 | |||
665 | hw3d->kernel.bo = intel_winsys_alloc_buffer(hw3d->cp->winsys, |
||
666 | "kernel bo", new_size, 0); |
||
667 | if (!hw3d->kernel.bo) { |
||
668 | ilo_err("failed to allocate kernel bo\n"); |
||
669 | return false; |
||
670 | } |
||
671 | |||
672 | hw3d->kernel.used = 0; |
||
673 | hw3d->kernel.size = new_size; |
||
674 | incremental = false; |
||
675 | |||
676 | assert(new_size >= ilo_shader_cache_upload(shc, |
||
677 | NULL, hw3d->kernel.used, incremental)); |
||
678 | |||
679 | ilo_3d_pipeline_invalidate(hw3d->pipeline, |
||
680 | ILO_3D_PIPELINE_INVALIDATE_KERNEL_BO); |
||
681 | } |
||
682 | |||
683 | upload = ilo_shader_cache_upload(shc, |
||
684 | hw3d->kernel.bo, hw3d->kernel.used, incremental); |
||
685 | if (upload < 0) { |
||
686 | ilo_err("failed to upload shaders\n"); |
||
687 | return false; |
||
688 | } |
||
689 | |||
690 | hw3d->kernel.used += upload; |
||
691 | |||
692 | assert(hw3d->kernel.used <= hw3d->kernel.size); |
||
693 | |||
694 | return true; |
||
695 | } |
||
696 | |||
697 | static void |
||
698 | ilo_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info) |
||
699 | { |
||
700 | struct ilo_context *ilo = ilo_context(pipe); |
||
701 | struct ilo_3d *hw3d = ilo->hw3d; |
||
702 | int prim_generated, prim_emitted; |
||
703 | |||
704 | if (!ilo_3d_pass_render_condition(ilo)) |
||
705 | return; |
||
706 | |||
707 | if (info->primitive_restart && info->indexed) { |
||
708 | /* |
||
709 | * Want to draw an indexed primitive using primitive restart |
||
710 | * Check that HW can handle the request and fall to SW if not. |
||
711 | */ |
||
712 | if (!ilo_check_restart_index(ilo, info->restart_index) || |
||
713 | !ilo_check_restart_prim_type(ilo, info->mode)) { |
||
714 | ilo_draw_vbo_with_sw_restart(pipe, info); |
||
715 | return; |
||
716 | } |
||
717 | } |
||
718 | |||
719 | ilo_finalize_3d_states(ilo, info); |
||
720 | |||
721 | if (!upload_shaders(hw3d, ilo->shader_cache)) |
||
722 | return; |
||
723 | |||
724 | /* If draw_vbo ever fails, return immediately. */ |
||
725 | if (!draw_vbo(hw3d, ilo, &prim_generated, &prim_emitted)) |
||
726 | return; |
||
727 | |||
728 | /* clear dirty status */ |
||
729 | ilo->dirty = 0x0; |
||
730 | hw3d->new_batch = false; |
||
731 | |||
732 | /* avoid dangling pointer reference */ |
||
733 | ilo->draw = NULL; |
||
734 | |||
735 | update_prim_count(hw3d, prim_generated, prim_emitted); |
||
736 | |||
737 | if (ilo_debug & ILO_DEBUG_NOCACHE) |
||
738 | ilo_3d_pipeline_emit_flush(hw3d->pipeline); |
||
739 | } |
||
740 | |||
741 | static void |
||
742 | ilo_render_condition(struct pipe_context *pipe, |
||
743 | struct pipe_query *query, |
||
744 | boolean condition, |
||
745 | uint mode) |
||
746 | { |
||
747 | struct ilo_context *ilo = ilo_context(pipe); |
||
748 | struct ilo_3d *hw3d = ilo->hw3d; |
||
749 | |||
750 | /* reference count? */ |
||
751 | hw3d->render_condition.query = query; |
||
752 | hw3d->render_condition.mode = mode; |
||
753 | hw3d->render_condition.cond = condition; |
||
754 | } |
||
755 | |||
756 | static void |
||
757 | ilo_texture_barrier(struct pipe_context *pipe) |
||
758 | { |
||
759 | struct ilo_context *ilo = ilo_context(pipe); |
||
760 | struct ilo_3d *hw3d = ilo->hw3d; |
||
761 | |||
762 | if (ilo->cp->ring != ILO_CP_RING_RENDER) |
||
763 | return; |
||
764 | |||
765 | ilo_3d_pipeline_emit_flush(hw3d->pipeline); |
||
766 | |||
767 | /* don't know why */ |
||
768 | if (ilo->dev->gen >= ILO_GEN(7)) |
||
769 | ilo_cp_flush(hw3d->cp); |
||
770 | } |
||
771 | |||
772 | static void |
||
773 | ilo_get_sample_position(struct pipe_context *pipe, |
||
774 | unsigned sample_count, |
||
775 | unsigned sample_index, |
||
776 | float *out_value) |
||
777 | { |
||
778 | struct ilo_context *ilo = ilo_context(pipe); |
||
779 | struct ilo_3d *hw3d = ilo->hw3d; |
||
780 | |||
781 | ilo_3d_pipeline_get_sample_position(hw3d->pipeline, |
||
782 | sample_count, sample_index, |
||
783 | &out_value[0], &out_value[1]); |
||
784 | } |
||
785 | |||
786 | /** |
||
787 | * Initialize 3D-related functions. |
||
788 | */ |
||
789 | void |
||
790 | ilo_init_3d_functions(struct ilo_context *ilo) |
||
791 | { |
||
792 | ilo->base.draw_vbo = ilo_draw_vbo; |
||
793 | ilo->base.render_condition = ilo_render_condition; |
||
794 | ilo->base.texture_barrier = ilo_texture_barrier; |
||
795 | ilo->base.get_sample_position = ilo_get_sample_position; |
||
796 | }=>>>>>=>>>>>>> |