projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
u_vbuf_mgr: correctly obtain min/max_index for uploads and translate
[mesa.git] / src / gallium / auxiliary / util / u_vbuf_mgr.c
1 /**************************************************************************
2 *
3 * Copyright 2011 Marek Olšák <maraeo@gmail.com>
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 AUTHORS 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 "util/u_vbuf_mgr.h"
29
30 #include "util/u_format.h"
31 #include "util/u_inlines.h"
32 #include "util/u_memory.h"
33 #include "util/u_upload_mgr.h"
34 #include "translate/translate.h"
35 #include "translate/translate_cache.h"
36
37 struct u_vbuf_elements {
38 unsigned count;
39 struct pipe_vertex_element ve[PIPE_MAX_ATTRIBS];
40
41 unsigned src_format_size[PIPE_MAX_ATTRIBS];
42
43 /* If (velem[i].src_format != native_format[i]), the vertex buffer
44 * referenced by the vertex element cannot be used for rendering and
45 * its vertex data must be translated to native_format[i]. */
46 enum pipe_format native_format[PIPE_MAX_ATTRIBS];
47 unsigned native_format_size[PIPE_MAX_ATTRIBS];
48
49 /* This might mean two things:
50 * - src_format != native_format, as discussed above.
51 * - src_offset % 4 != 0 (if the caps don't allow such an offset). */
52 boolean incompatible_layout;
53 /* Per-element flags. */
54 boolean incompatible_layout_elem[PIPE_MAX_ATTRIBS];
55 };
56
57 struct u_vbuf_priv {
58 struct u_vbuf_mgr b;
59 struct pipe_context *pipe;
60 struct translate_cache *translate_cache;
61
62 /* Whether there is any user buffer. */
63 boolean any_user_vbs;
64
65 /* Vertex element state bound by the state tracker. */
66 void *saved_ve;
67 /* and its associated helper structure for this module. */
68 struct u_vbuf_elements *ve;
69
70 /* Vertex elements used for the translate fallback. */
71 struct pipe_vertex_element fallback_velems[PIPE_MAX_ATTRIBS];
72 /* If non-NULL, this is a vertex element state used for the translate
73 * fallback and therefore used for rendering too. */
74 void *fallback_ve;
75 /* The vertex buffer slot index where translated vertices have been
76 * stored in. */
77 unsigned translate_vb_slot;
78 /* When binding the fallback vertex element state, we don't want to
79 * change saved_ve and ve. This is set to TRUE in such cases. */
80 boolean ve_binding_lock;
81
82 /* Whether there is a buffer with a non-native layout. */
83 boolean incompatible_vb_layout;
84 /* Per-buffer flags. */
85 boolean incompatible_vb[PIPE_MAX_ATTRIBS];
86 };
87
88 static void u_vbuf_init_format_caps(struct u_vbuf_priv *mgr)
89 {
90 struct pipe_screen *screen = mgr->pipe->screen;
91
92 mgr->b.caps.format_fixed32 =
93 screen->is_format_supported(screen, PIPE_FORMAT_R32_FIXED, PIPE_BUFFER,
94 0, PIPE_BIND_VERTEX_BUFFER);
95
96 mgr->b.caps.format_float16 =
97 screen->is_format_supported(screen, PIPE_FORMAT_R16_FLOAT, PIPE_BUFFER,
98 0, PIPE_BIND_VERTEX_BUFFER);
99
100 mgr->b.caps.format_float64 =
101 screen->is_format_supported(screen, PIPE_FORMAT_R64_FLOAT, PIPE_BUFFER,
102 0, PIPE_BIND_VERTEX_BUFFER);
103
104 mgr->b.caps.format_norm32 =
105 screen->is_format_supported(screen, PIPE_FORMAT_R32_UNORM, PIPE_BUFFER,
106 0, PIPE_BIND_VERTEX_BUFFER) &&
107 screen->is_format_supported(screen, PIPE_FORMAT_R32_SNORM, PIPE_BUFFER,
108 0, PIPE_BIND_VERTEX_BUFFER);
109
110 mgr->b.caps.format_scaled32 =
111 screen->is_format_supported(screen, PIPE_FORMAT_R32_USCALED, PIPE_BUFFER,
112 0, PIPE_BIND_VERTEX_BUFFER) &&
113 screen->is_format_supported(screen, PIPE_FORMAT_R32_SSCALED, PIPE_BUFFER,
114 0, PIPE_BIND_VERTEX_BUFFER);
115 }
116
117 struct u_vbuf_mgr *
118 u_vbuf_create(struct pipe_context *pipe,
119 unsigned upload_buffer_size,
120 unsigned upload_buffer_alignment,
121 unsigned upload_buffer_bind,
122 enum u_fetch_alignment fetch_alignment)
123 {
124 struct u_vbuf_priv *mgr = CALLOC_STRUCT(u_vbuf_priv);
125
126 mgr->pipe = pipe;
127 mgr->translate_cache = translate_cache_create();
128 mgr->translate_vb_slot = ~0;
129
130 mgr->b.uploader = u_upload_create(pipe, upload_buffer_size,
131 upload_buffer_alignment,
132 upload_buffer_bind);
133
134 mgr->b.caps.fetch_dword_unaligned =
135 fetch_alignment == U_VERTEX_FETCH_BYTE_ALIGNED;
136
137 u_vbuf_init_format_caps(mgr);
138
139 return &mgr->b;
140 }
141
142 void u_vbuf_destroy(struct u_vbuf_mgr *mgrb)
143 {
144 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
145 unsigned i;
146
147 for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {
148 pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL);
149 }
150 for (i = 0; i < mgr->b.nr_real_vertex_buffers; i++) {
151 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL);
152 }
153
154 translate_cache_destroy(mgr->translate_cache);
155 u_upload_destroy(mgr->b.uploader);
156 FREE(mgr);
157 }
158
159
160 static void
161 u_vbuf_translate_begin(struct u_vbuf_priv *mgr,
162 int min_index, int max_index)
163 {
164 struct translate_key key;
165 struct translate_element *te;
166 unsigned tr_elem_index[PIPE_MAX_ATTRIBS];
167 struct translate *tr;
168 boolean vb_translated[PIPE_MAX_ATTRIBS] = {0};
169 uint8_t *vb_map[PIPE_MAX_ATTRIBS] = {0}, *out_map;
170 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0};
171 struct pipe_resource *out_buffer = NULL;
172 unsigned i, num_verts, out_offset;
173 boolean upload_flushed = FALSE;
174
175 memset(&key, 0, sizeof(key));
176 memset(tr_elem_index, 0xff, sizeof(tr_elem_index));
177
178 /* Initialize the translate key, i.e. the recipe how vertices should be
179 * translated. */
180 memset(&key, 0, sizeof key);
181 for (i = 0; i < mgr->ve->count; i++) {
182 enum pipe_format output_format = mgr->ve->native_format[i];
183 unsigned output_format_size = mgr->ve->native_format_size[i];
184
185 /* Check for support. */
186 if (!mgr->ve->incompatible_layout_elem[i] &&
187 !mgr->incompatible_vb[mgr->ve->ve[i].vertex_buffer_index]) {
188 continue;
189 }
190
191 /* Workaround for translate: output floats instead of halfs. */
192 switch (output_format) {
193 case PIPE_FORMAT_R16_FLOAT:
194 output_format = PIPE_FORMAT_R32_FLOAT;
195 output_format_size = 4;
196 break;
197 case PIPE_FORMAT_R16G16_FLOAT:
198 output_format = PIPE_FORMAT_R32G32_FLOAT;
199 output_format_size = 8;
200 break;
201 case PIPE_FORMAT_R16G16B16_FLOAT:
202 output_format = PIPE_FORMAT_R32G32B32_FLOAT;
203 output_format_size = 12;
204 break;
205 case PIPE_FORMAT_R16G16B16A16_FLOAT:
206 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
207 output_format_size = 16;
208 break;
209 default:;
210 }
211
212 /* Add this vertex element. */
213 te = &key.element[key.nr_elements];
214 /*te->type;
215 te->instance_divisor;*/
216 te->input_buffer = mgr->ve->ve[i].vertex_buffer_index;
217 te->input_format = mgr->ve->ve[i].src_format;
218 te->input_offset = mgr->ve->ve[i].src_offset;
219 te->output_format = output_format;
220 te->output_offset = key.output_stride;
221
222 key.output_stride += output_format_size;
223 vb_translated[mgr->ve->ve[i].vertex_buffer_index] = TRUE;
224 tr_elem_index[i] = key.nr_elements;
225 key.nr_elements++;
226 }
227
228 /* Get a translate object. */
229 tr = translate_cache_find(mgr->translate_cache, &key);
230
231 /* Map buffers we want to translate. */
232 for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {
233 if (vb_translated[i]) {
234 struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[i];
235
236 vb_map[i] = pipe_buffer_map(mgr->pipe, vb->buffer,
237 PIPE_TRANSFER_READ, &vb_transfer[i]);
238
239 tr->set_buffer(tr, i,
240 vb_map[i] + vb->buffer_offset + vb->stride * min_index,
241 vb->stride, ~0);
242 }
243 }
244
245 /* Create and map the output buffer. */
246 num_verts = max_index + 1 - min_index;
247
248 u_upload_alloc(mgr->b.uploader,
249 key.output_stride * min_index,
250 key.output_stride * num_verts,
251 &out_offset, &out_buffer, &upload_flushed,
252 (void**)&out_map);
253
254 out_offset -= key.output_stride * min_index;
255
256 /* Translate. */
257 tr->run(tr, 0, num_verts, 0, out_map);
258
259 /* Unmap all buffers. */
260 for (i = 0; i < mgr->b.nr_vertex_buffers; i++) {
261 if (vb_translated[i]) {
262 pipe_buffer_unmap(mgr->pipe, vb_transfer[i]);
263 }
264 }
265
266 /* Setup the new vertex buffer in the first free slot. */
267 mgr->translate_vb_slot = ~0;
268 for (i = 0; i < PIPE_MAX_ATTRIBS; i++) {
269 if (!mgr->b.vertex_buffer[i].buffer) {
270 mgr->translate_vb_slot = i;
271
272 if (i >= mgr->b.nr_vertex_buffers) {
273 mgr->b.nr_real_vertex_buffers = i+1;
274 }
275 break;
276 }
277 }
278
279 if (mgr->translate_vb_slot != ~0) {
280 /* Setup the new vertex buffer. */
281 pipe_resource_reference(
282 &mgr->b.real_vertex_buffer[mgr->translate_vb_slot].buffer, out_buffer);
283 mgr->b.real_vertex_buffer[mgr->translate_vb_slot].buffer_offset = out_offset;
284 mgr->b.real_vertex_buffer[mgr->translate_vb_slot].stride = key.output_stride;
285
286 /* Setup new vertex elements. */
287 for (i = 0; i < mgr->ve->count; i++) {
288 if (tr_elem_index[i] < key.nr_elements) {
289 te = &key.element[tr_elem_index[i]];
290 mgr->fallback_velems[i].instance_divisor = mgr->ve->ve[i].instance_divisor;
291 mgr->fallback_velems[i].src_format = te->output_format;
292 mgr->fallback_velems[i].src_offset = te->output_offset;
293 mgr->fallback_velems[i].vertex_buffer_index = mgr->translate_vb_slot;
294 } else {
295 memcpy(&mgr->fallback_velems[i], &mgr->ve->ve[i],
296 sizeof(struct pipe_vertex_element));
297 }
298 }
299
300 mgr->fallback_ve =
301 mgr->pipe->create_vertex_elements_state(mgr->pipe, mgr->ve->count,
302 mgr->fallback_velems);
303
304 /* Preserve saved_ve. */
305 mgr->ve_binding_lock = TRUE;
306 mgr->pipe->bind_vertex_elements_state(mgr->pipe, mgr->fallback_ve);
307 mgr->ve_binding_lock = FALSE;
308 }
309
310 pipe_resource_reference(&out_buffer, NULL);
311 }
312
313 static void u_vbuf_translate_end(struct u_vbuf_priv *mgr)
314 {
315 if (mgr->fallback_ve == NULL) {
316 return;
317 }
318
319 /* Restore vertex elements. */
320 /* Note that saved_ve will be overwritten in bind_vertex_elements_state. */
321 mgr->pipe->bind_vertex_elements_state(mgr->pipe, mgr->saved_ve);
322 mgr->pipe->delete_vertex_elements_state(mgr->pipe, mgr->fallback_ve);
323 mgr->fallback_ve = NULL;
324
325 /* Delete the now-unused VBO. */
326 pipe_resource_reference(&mgr->b.real_vertex_buffer[mgr->translate_vb_slot].buffer,
327 NULL);
328 mgr->translate_vb_slot = ~0;
329 mgr->b.nr_real_vertex_buffers = mgr->b.nr_vertex_buffers;
330 }
331
332 #define FORMAT_REPLACE(what, withwhat) \
333 case PIPE_FORMAT_##what: format = PIPE_FORMAT_##withwhat; break
334
335 struct u_vbuf_elements *
336 u_vbuf_create_vertex_elements(struct u_vbuf_mgr *mgrb,
337 unsigned count,
338 const struct pipe_vertex_element *attribs,
339 struct pipe_vertex_element *native_attribs)
340 {
341 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
342 unsigned i;
343 struct u_vbuf_elements *ve = CALLOC_STRUCT(u_vbuf_elements);
344
345 ve->count = count;
346
347 if (!count) {
348 return ve;
349 }
350
351 memcpy(ve->ve, attribs, sizeof(struct pipe_vertex_element) * count);
352 memcpy(native_attribs, attribs, sizeof(struct pipe_vertex_element) * count);
353
354 /* Set the best native format in case the original format is not
355 * supported. */
356 for (i = 0; i < count; i++) {
357 enum pipe_format format = ve->ve[i].src_format;
358
359 ve->src_format_size[i] = util_format_get_blocksize(format);
360
361 /* Choose a native format.
362 * For now we don't care about the alignment, that's going to
363 * be sorted out later. */
364 if (!mgr->b.caps.format_fixed32) {
365 switch (format) {
366 FORMAT_REPLACE(R32_FIXED, R32_FLOAT);
367 FORMAT_REPLACE(R32G32_FIXED, R32G32_FLOAT);
368 FORMAT_REPLACE(R32G32B32_FIXED, R32G32B32_FLOAT);
369 FORMAT_REPLACE(R32G32B32A32_FIXED, R32G32B32A32_FLOAT);
370 default:;
371 }
372 }
373 if (!mgr->b.caps.format_float16) {
374 switch (format) {
375 FORMAT_REPLACE(R16_FLOAT, R32_FLOAT);
376 FORMAT_REPLACE(R16G16_FLOAT, R32G32_FLOAT);
377 FORMAT_REPLACE(R16G16B16_FLOAT, R32G32B32_FLOAT);
378 FORMAT_REPLACE(R16G16B16A16_FLOAT, R32G32B32A32_FLOAT);
379 default:;
380 }
381 }
382 if (!mgr->b.caps.format_float64) {
383 switch (format) {
384 FORMAT_REPLACE(R64_FLOAT, R32_FLOAT);
385 FORMAT_REPLACE(R64G64_FLOAT, R32G32_FLOAT);
386 FORMAT_REPLACE(R64G64B64_FLOAT, R32G32B32_FLOAT);
387 FORMAT_REPLACE(R64G64B64A64_FLOAT, R32G32B32A32_FLOAT);
388 default:;
389 }
390 }
391 if (!mgr->b.caps.format_norm32) {
392 switch (format) {
393 FORMAT_REPLACE(R32_UNORM, R32_FLOAT);
394 FORMAT_REPLACE(R32G32_UNORM, R32G32_FLOAT);
395 FORMAT_REPLACE(R32G32B32_UNORM, R32G32B32_FLOAT);
396 FORMAT_REPLACE(R32G32B32A32_UNORM, R32G32B32A32_FLOAT);
397 FORMAT_REPLACE(R32_SNORM, R32_FLOAT);
398 FORMAT_REPLACE(R32G32_SNORM, R32G32_FLOAT);
399 FORMAT_REPLACE(R32G32B32_SNORM, R32G32B32_FLOAT);
400 FORMAT_REPLACE(R32G32B32A32_SNORM, R32G32B32A32_FLOAT);
401 default:;
402 }
403 }
404 if (!mgr->b.caps.format_scaled32) {
405 switch (format) {
406 FORMAT_REPLACE(R32_USCALED, R32_FLOAT);
407 FORMAT_REPLACE(R32G32_USCALED, R32G32_FLOAT);
408 FORMAT_REPLACE(R32G32B32_USCALED, R32G32B32_FLOAT);
409 FORMAT_REPLACE(R32G32B32A32_USCALED,R32G32B32A32_FLOAT);
410 FORMAT_REPLACE(R32_SSCALED, R32_FLOAT);
411 FORMAT_REPLACE(R32G32_SSCALED, R32G32_FLOAT);
412 FORMAT_REPLACE(R32G32B32_SSCALED, R32G32B32_FLOAT);
413 FORMAT_REPLACE(R32G32B32A32_SSCALED,R32G32B32A32_FLOAT);
414 default:;
415 }
416 }
417
418 native_attribs[i].src_format = format;
419 ve->native_format[i] = format;
420 ve->native_format_size[i] =
421 util_format_get_blocksize(ve->native_format[i]);
422
423 ve->incompatible_layout_elem[i] =
424 ve->ve[i].src_format != ve->native_format[i] ||
425 (!mgr->b.caps.fetch_dword_unaligned && ve->ve[i].src_offset % 4 != 0);
426 ve->incompatible_layout =
427 ve->incompatible_layout ||
428 ve->incompatible_layout_elem[i];
429 }
430
431 /* Align the formats to the size of DWORD if needed. */
432 if (!mgr->b.caps.fetch_dword_unaligned) {
433 for (i = 0; i < count; i++) {
434 ve->native_format_size[i] = align(ve->native_format_size[i], 4);
435 }
436 }
437
438 return ve;
439 }
440
441 void u_vbuf_bind_vertex_elements(struct u_vbuf_mgr *mgrb,
442 void *cso,
443 struct u_vbuf_elements *ve)
444 {
445 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
446
447 if (!cso) {
448 return;
449 }
450
451 if (!mgr->ve_binding_lock) {
452 mgr->saved_ve = cso;
453 mgr->ve = ve;
454 }
455 }
456
457 void u_vbuf_destroy_vertex_elements(struct u_vbuf_mgr *mgr,
458 struct u_vbuf_elements *ve)
459 {
460 FREE(ve);
461 }
462
463 void u_vbuf_set_vertex_buffers(struct u_vbuf_mgr *mgrb,
464 unsigned count,
465 const struct pipe_vertex_buffer *bufs)
466 {
467 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
468 unsigned i;
469
470 mgr->any_user_vbs = FALSE;
471 mgr->incompatible_vb_layout = FALSE;
472 memset(mgr->incompatible_vb, 0, sizeof(mgr->incompatible_vb));
473
474 if (!mgr->b.caps.fetch_dword_unaligned) {
475 /* Check if the strides and offsets are aligned to the size of DWORD. */
476 for (i = 0; i < count; i++) {
477 if (bufs[i].buffer) {
478 if (bufs[i].stride % 4 != 0 ||
479 bufs[i].buffer_offset % 4 != 0) {
480 mgr->incompatible_vb_layout = TRUE;
481 mgr->incompatible_vb[i] = TRUE;
482 }
483 }
484 }
485 }
486
487 for (i = 0; i < count; i++) {
488 const struct pipe_vertex_buffer *vb = &bufs[i];
489
490 pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, vb->buffer);
491
492 mgr->b.real_vertex_buffer[i].buffer_offset =
493 mgr->b.vertex_buffer[i].buffer_offset = vb->buffer_offset;
494
495 mgr->b.real_vertex_buffer[i].stride =
496 mgr->b.vertex_buffer[i].stride = vb->stride;
497
498 if (!vb->buffer) {
499 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL);
500 continue;
501 }
502
503 if (u_vbuf_resource(vb->buffer)->user_ptr) {
504 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL);
505 mgr->any_user_vbs = TRUE;
506 continue;
507 }
508
509 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, vb->buffer);
510 }
511
512 for (i = count; i < mgr->b.nr_vertex_buffers; i++) {
513 pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL);
514 }
515 for (i = count; i < mgr->b.nr_real_vertex_buffers; i++) {
516 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL);
517 }
518
519 mgr->b.nr_vertex_buffers = count;
520 mgr->b.nr_real_vertex_buffers = count;
521 }
522
523 void u_vbuf_set_index_buffer(struct u_vbuf_mgr *mgr,
524 const struct pipe_index_buffer *ib)
525 {
526 if (ib && ib->buffer) {
527 assert(ib->offset % ib->index_size == 0);
528 pipe_resource_reference(&mgr->index_buffer.buffer, ib->buffer);
529 mgr->index_buffer.offset = ib->offset;
530 mgr->index_buffer.index_size = ib->index_size;
531 } else {
532 pipe_resource_reference(&mgr->index_buffer.buffer, NULL);
533 }
534 }
535
536 static void
537 u_vbuf_upload_buffers(struct u_vbuf_priv *mgr,
538 int min_index, int max_index,
539 unsigned instance_count)
540 {
541 unsigned i;
542 unsigned count = max_index + 1 - min_index;
543 unsigned nr_velems = mgr->ve->count;
544 unsigned nr_vbufs = mgr->b.nr_vertex_buffers;
545 struct pipe_vertex_element *velems =
546 mgr->fallback_ve ? mgr->fallback_velems : mgr->ve->ve;
547 unsigned start_offset[PIPE_MAX_ATTRIBS];
548 unsigned end_offset[PIPE_MAX_ATTRIBS] = {0};
549
550 /* Determine how much data needs to be uploaded. */
551 for (i = 0; i < nr_velems; i++) {
552 struct pipe_vertex_element *velem = &velems[i];
553 unsigned index = velem->vertex_buffer_index;
554 struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[index];
555 unsigned instance_div, first, size;
556
557 /* Skip the buffer generated by translate. */
558 if (index == mgr->translate_vb_slot) {
559 continue;
560 }
561
562 assert(vb->buffer);
563
564 if (!u_vbuf_resource(vb->buffer)->user_ptr) {
565 continue;
566 }
567
568 instance_div = velem->instance_divisor;
569 first = vb->buffer_offset + velem->src_offset;
570
571 if (!vb->stride) {
572 /* Constant attrib. */
573 size = mgr->ve->src_format_size[i];
574 } else if (instance_div) {
575 /* Per-instance attrib. */
576 unsigned count = (instance_count + instance_div - 1) / instance_div;
577 size = vb->stride * (count - 1) + mgr->ve->src_format_size[i];
578 } else {
579 /* Per-vertex attrib. */
580 first += vb->stride * min_index;
581 size = vb->stride * (count - 1) + mgr->ve->src_format_size[i];
582 }
583
584 /* Update offsets. */
585 if (!end_offset[index]) {
586 start_offset[index] = first;
587 end_offset[index] = first + size;
588 } else {
589 if (first < start_offset[index])
590 start_offset[index] = first;
591 if (first + size > end_offset[index])
592 end_offset[index] = first + size;
593 }
594 }
595
596 /* Upload buffers. */
597 for (i = 0; i < nr_vbufs; i++) {
598 unsigned start, end = end_offset[i];
599 boolean flushed;
600 struct pipe_vertex_buffer *real_vb;
601 uint8_t *ptr;
602
603 if (!end) {
604 continue;
605 }
606
607 start = start_offset[i];
608 assert(start < end);
609
610 real_vb = &mgr->b.real_vertex_buffer[i];
611 ptr = u_vbuf_resource(mgr->b.vertex_buffer[i].buffer)->user_ptr;
612
613 u_upload_data(mgr->b.uploader, start, end - start, ptr + start,
614 &real_vb->buffer_offset, &real_vb->buffer, &flushed);
615
616 real_vb->buffer_offset -= start;
617 }
618 }
619
620 static void u_vbuf_compute_max_index(struct u_vbuf_priv *mgr)
621 {
622 unsigned i, nr = mgr->ve->count;
623
624 mgr->b.max_index = ~0;
625
626 for (i = 0; i < nr; i++) {
627 struct pipe_vertex_buffer *vb =
628 &mgr->b.vertex_buffer[mgr->ve->ve[i].vertex_buffer_index];
629 unsigned max_index, src_size, unused;
630
631 if (!vb->buffer ||
632 !vb->stride ||
633 u_vbuf_resource(vb->buffer)->user_ptr ||
634 mgr->ve->ve[i].instance_divisor) {
635 continue;
636 }
637
638 src_size = mgr->ve->ve[i].src_offset + mgr->ve->src_format_size[i];
639
640 /* If src_offset is greater than stride (which means it's a buffer
641 * offset rather than a vertex offset)... */
642 if (src_size >= vb->stride) {
643 unused = 0;
644 } else {
645 /* How many bytes is unused after the last vertex.
646 * width0 may be "count*stride - unused" and we have to compensate
647 * for that when dividing by stride. */
648 unused = vb->stride - src_size;
649 }
650
651 /* Compute the maximum index for this vertex element. */
652 max_index =
653 (vb->buffer->width0 - vb->buffer_offset + (unsigned)unused) /
654 vb->stride - 1;
655
656 mgr->b.max_index = MIN2(mgr->b.max_index, max_index);
657 }
658 }
659
660 static boolean u_vbuf_need_minmax_index(struct u_vbuf_priv *mgr)
661 {
662 unsigned i, nr = mgr->ve->count;
663
664 for (i = 0; i < nr; i++) {
665 struct pipe_vertex_buffer *vb;
666 unsigned index;
667
668 /* Per-instance attribs don't need min/max_index. */
669 if (mgr->ve->ve[i].instance_divisor) {
670 continue;
671 }
672
673 index = mgr->ve->ve[i].vertex_buffer_index;
674 vb = &mgr->b.vertex_buffer[index];
675
676 /* Constant attribs don't need min/max_index. */
677 if (!vb->stride) {
678 continue;
679 }
680
681 /* Per-vertex attribs need min/max_index. */
682 if (u_vbuf_resource(vb->buffer)->user_ptr ||
683 mgr->ve->incompatible_layout_elem[i] ||
684 mgr->incompatible_vb[index]) {
685 return TRUE;
686 }
687 }
688
689 return FALSE;
690 }
691
692 static void u_vbuf_get_minmax_index(struct pipe_context *pipe,
693 struct pipe_index_buffer *ib,
694 const struct pipe_draw_info *info,
695 int *out_min_index,
696 int *out_max_index)
697 {
698 struct pipe_transfer *transfer = NULL;
699 const void *indices;
700 unsigned i;
701 unsigned restart_index = info->restart_index;
702
703 if (u_vbuf_resource(ib->buffer)->user_ptr) {
704 indices = u_vbuf_resource(ib->buffer)->user_ptr +
705 ib->offset + info->start * ib->index_size;
706 } else {
707 indices = pipe_buffer_map_range(pipe, ib->buffer,
708 ib->offset + info->start * ib->index_size,
709 info->count * ib->index_size,
710 PIPE_TRANSFER_READ, &transfer);
711 }
712
713 switch (ib->index_size) {
714 case 4: {
715 const unsigned *ui_indices = (const unsigned*)indices;
716 unsigned max_ui = 0;
717 unsigned min_ui = ~0U;
718 if (info->primitive_restart) {
719 for (i = 0; i < info->count; i++) {
720 if (ui_indices[i] != restart_index) {
721 if (ui_indices[i] > max_ui) max_ui = ui_indices[i];
722 if (ui_indices[i] < min_ui) min_ui = ui_indices[i];
723 }
724 }
725 }
726 else {
727 for (i = 0; i < info->count; i++) {
728 if (ui_indices[i] > max_ui) max_ui = ui_indices[i];
729 if (ui_indices[i] < min_ui) min_ui = ui_indices[i];
730 }
731 }
732 *out_min_index = min_ui;
733 *out_max_index = max_ui;
734 break;
735 }
736 case 2: {
737 const unsigned short *us_indices = (const unsigned short*)indices;
738 unsigned max_us = 0;
739 unsigned min_us = ~0U;
740 if (info->primitive_restart) {
741 for (i = 0; i < info->count; i++) {
742 if (us_indices[i] != restart_index) {
743 if (us_indices[i] > max_us) max_us = us_indices[i];
744 if (us_indices[i] < min_us) min_us = us_indices[i];
745 }
746 }
747 }
748 else {
749 for (i = 0; i < info->count; i++) {
750 if (us_indices[i] > max_us) max_us = us_indices[i];
751 if (us_indices[i] < min_us) min_us = us_indices[i];
752 }
753 }
754 *out_min_index = min_us;
755 *out_max_index = max_us;
756 break;
757 }
758 case 1: {
759 const unsigned char *ub_indices = (const unsigned char*)indices;
760 unsigned max_ub = 0;
761 unsigned min_ub = ~0U;
762 if (info->primitive_restart) {
763 for (i = 0; i < info->count; i++) {
764 if (ub_indices[i] != restart_index) {
765 if (ub_indices[i] > max_ub) max_ub = ub_indices[i];
766 if (ub_indices[i] < min_ub) min_ub = ub_indices[i];
767 }
768 }
769 }
770 else {
771 for (i = 0; i < info->count; i++) {
772 if (ub_indices[i] > max_ub) max_ub = ub_indices[i];
773 if (ub_indices[i] < min_ub) min_ub = ub_indices[i];
774 }
775 }
776 *out_min_index = min_ub;
777 *out_max_index = max_ub;
778 break;
779 }
780 default:
781 assert(0);
782 }
783
784 if (transfer) {
785 pipe_buffer_unmap(pipe, transfer);
786 }
787 }
788
789 enum u_vbuf_return_flags
790 u_vbuf_draw_begin(struct u_vbuf_mgr *mgrb,
791 const struct pipe_draw_info *info)
792 {
793 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
794 int min_index, max_index;
795
796 u_vbuf_compute_max_index(mgr);
797
798 if (!mgr->incompatible_vb_layout &&
799 !mgr->ve->incompatible_layout &&
800 !mgr->any_user_vbs) {
801 return 0;
802 }
803
804 if (info->indexed) {
805 if (info->max_index != ~0) {
806 min_index = info->min_index + info->index_bias;
807 max_index = info->max_index + info->index_bias;
808 } else if (u_vbuf_need_minmax_index(mgr)) {
809 u_vbuf_get_minmax_index(mgr->pipe, &mgr->b.index_buffer, info,
810 &min_index, &max_index);
811 min_index += info->index_bias;
812 max_index += info->index_bias;
813 } else {
814 min_index = 0;
815 max_index = 0;
816 }
817 } else {
818 min_index = info->start;
819 max_index = info->start + info->count - 1;
820 }
821
822 /* Translate vertices with non-native layouts or formats. */
823 if (mgr->incompatible_vb_layout || mgr->ve->incompatible_layout) {
824 u_vbuf_translate_begin(mgr, min_index, max_index);
825 }
826
827 /* Upload user buffers. */
828 if (mgr->any_user_vbs) {
829 u_vbuf_upload_buffers(mgr, min_index, max_index, info->instance_count);
830 }
831 return U_VBUF_BUFFERS_UPDATED;
832 }
833
834 void u_vbuf_draw_end(struct u_vbuf_mgr *mgrb)
835 {
836 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb;
837
838 if (mgr->fallback_ve) {
839 u_vbuf_translate_end(mgr);
840 }
841 }