u_vbuf.c revision c2cc630f2896175ff0f368d9199acbe24afb7e75
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.h" 29 30#include "util/u_dump.h" 31#include "util/u_format.h" 32#include "util/u_inlines.h" 33#include "util/u_memory.h" 34#include "util/u_upload_mgr.h" 35#include "translate/translate.h" 36#include "translate/translate_cache.h" 37#include "cso_cache/cso_cache.h" 38#include "cso_cache/cso_hash.h" 39 40struct u_vbuf_elements { 41 unsigned count; 42 struct pipe_vertex_element ve[PIPE_MAX_ATTRIBS]; 43 44 unsigned src_format_size[PIPE_MAX_ATTRIBS]; 45 46 /* If (velem[i].src_format != native_format[i]), the vertex buffer 47 * referenced by the vertex element cannot be used for rendering and 48 * its vertex data must be translated to native_format[i]. */ 49 enum pipe_format native_format[PIPE_MAX_ATTRIBS]; 50 unsigned native_format_size[PIPE_MAX_ATTRIBS]; 51 52 /* This might mean two things: 53 * - src_format != native_format, as discussed above. 54 * - src_offset % 4 != 0 (if the caps don't allow such an offset). */ 55 boolean incompatible_layout; 56 /* Per-element flags. */ 57 boolean incompatible_layout_elem[PIPE_MAX_ATTRIBS]; 58}; 59 60enum { 61 VB_VERTEX = 0, 62 VB_INSTANCE = 1, 63 VB_CONST = 2, 64 VB_NUM = 3 65}; 66 67struct u_vbuf_priv { 68 struct u_vbuf b; 69 struct pipe_context *pipe; 70 struct translate_cache *translate_cache; 71 struct cso_cache *cso_cache; 72 73 /* Vertex element state bound by the state tracker. */ 74 void *saved_ve; 75 /* and its associated helper structure for this module. */ 76 struct u_vbuf_elements *ve; 77 78 /* Vertex elements used for the translate fallback. */ 79 struct pipe_vertex_element fallback_velems[PIPE_MAX_ATTRIBS]; 80 /* If non-NULL, this is a vertex element state used for the translate 81 * fallback and therefore used for rendering too. */ 82 void *fallback_ve; 83 /* The vertex buffer slot index where translated vertices have been 84 * stored in. */ 85 unsigned fallback_vbs[VB_NUM]; 86 /* When binding the fallback vertex element state, we don't want to 87 * change saved_ve and ve. This is set to TRUE in such cases. */ 88 boolean ve_binding_lock; 89 90 /* Whether there is any user buffer. */ 91 boolean any_user_vbs; 92 /* Whether there is a buffer with a non-native layout. */ 93 boolean incompatible_vb_layout; 94 /* Per-buffer flags. */ 95 boolean incompatible_vb[PIPE_MAX_ATTRIBS]; 96}; 97 98static void u_vbuf_init_format_caps(struct u_vbuf_priv *mgr) 99{ 100 struct pipe_screen *screen = mgr->pipe->screen; 101 102 mgr->b.caps.format_fixed32 = 103 screen->is_format_supported(screen, PIPE_FORMAT_R32_FIXED, PIPE_BUFFER, 104 0, PIPE_BIND_VERTEX_BUFFER); 105 106 mgr->b.caps.format_float16 = 107 screen->is_format_supported(screen, PIPE_FORMAT_R16_FLOAT, PIPE_BUFFER, 108 0, PIPE_BIND_VERTEX_BUFFER); 109 110 mgr->b.caps.format_float64 = 111 screen->is_format_supported(screen, PIPE_FORMAT_R64_FLOAT, PIPE_BUFFER, 112 0, PIPE_BIND_VERTEX_BUFFER); 113 114 mgr->b.caps.format_norm32 = 115 screen->is_format_supported(screen, PIPE_FORMAT_R32_UNORM, PIPE_BUFFER, 116 0, PIPE_BIND_VERTEX_BUFFER) && 117 screen->is_format_supported(screen, PIPE_FORMAT_R32_SNORM, PIPE_BUFFER, 118 0, PIPE_BIND_VERTEX_BUFFER); 119 120 mgr->b.caps.format_scaled32 = 121 screen->is_format_supported(screen, PIPE_FORMAT_R32_USCALED, PIPE_BUFFER, 122 0, PIPE_BIND_VERTEX_BUFFER) && 123 screen->is_format_supported(screen, PIPE_FORMAT_R32_SSCALED, PIPE_BUFFER, 124 0, PIPE_BIND_VERTEX_BUFFER); 125} 126 127struct u_vbuf * 128u_vbuf_create(struct pipe_context *pipe, 129 unsigned upload_buffer_size, 130 unsigned upload_buffer_alignment, 131 unsigned upload_buffer_bind, 132 enum u_fetch_alignment fetch_alignment) 133{ 134 struct u_vbuf_priv *mgr = CALLOC_STRUCT(u_vbuf_priv); 135 136 mgr->pipe = pipe; 137 mgr->cso_cache = cso_cache_create(); 138 mgr->translate_cache = translate_cache_create(); 139 memset(mgr->fallback_vbs, ~0, sizeof(mgr->fallback_vbs)); 140 141 mgr->b.uploader = u_upload_create(pipe, upload_buffer_size, 142 upload_buffer_alignment, 143 upload_buffer_bind); 144 145 mgr->b.caps.fetch_dword_unaligned = 146 fetch_alignment == U_VERTEX_FETCH_BYTE_ALIGNED; 147 148 u_vbuf_init_format_caps(mgr); 149 150 return &mgr->b; 151} 152 153/* XXX I had to fork this off of cso_context. */ 154static void * 155u_vbuf_pipe_set_vertex_elements(struct u_vbuf_priv *mgr, 156 unsigned count, 157 const struct pipe_vertex_element *states) 158{ 159 unsigned key_size, hash_key; 160 struct cso_hash_iter iter; 161 void *handle; 162 struct cso_velems_state velems_state; 163 164 /* need to include the count into the stored state data too. */ 165 key_size = sizeof(struct pipe_vertex_element) * count + sizeof(unsigned); 166 velems_state.count = count; 167 memcpy(velems_state.velems, states, 168 sizeof(struct pipe_vertex_element) * count); 169 hash_key = cso_construct_key((void*)&velems_state, key_size); 170 iter = cso_find_state_template(mgr->cso_cache, hash_key, CSO_VELEMENTS, 171 (void*)&velems_state, key_size); 172 173 if (cso_hash_iter_is_null(iter)) { 174 struct cso_velements *cso = MALLOC_STRUCT(cso_velements); 175 memcpy(&cso->state, &velems_state, key_size); 176 cso->data = 177 mgr->pipe->create_vertex_elements_state(mgr->pipe, count, 178 &cso->state.velems[0]); 179 cso->delete_state = 180 (cso_state_callback)mgr->pipe->delete_vertex_elements_state; 181 cso->context = mgr->pipe; 182 183 iter = cso_insert_state(mgr->cso_cache, hash_key, CSO_VELEMENTS, cso); 184 handle = cso->data; 185 } else { 186 handle = ((struct cso_velements *)cso_hash_iter_data(iter))->data; 187 } 188 189 mgr->pipe->bind_vertex_elements_state(mgr->pipe, handle); 190 return handle; 191} 192 193void u_vbuf_destroy(struct u_vbuf *mgrb) 194{ 195 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb; 196 unsigned i; 197 198 for (i = 0; i < mgr->b.nr_vertex_buffers; i++) { 199 pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL); 200 } 201 for (i = 0; i < mgr->b.nr_real_vertex_buffers; i++) { 202 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL); 203 } 204 205 translate_cache_destroy(mgr->translate_cache); 206 u_upload_destroy(mgr->b.uploader); 207 cso_cache_delete(mgr->cso_cache); 208 FREE(mgr); 209} 210 211static void 212u_vbuf_translate_buffers(struct u_vbuf_priv *mgr, struct translate_key *key, 213 unsigned vb_mask, unsigned out_vb, 214 int start_vertex, unsigned num_vertices, 215 int start_index, unsigned num_indices, int min_index, 216 bool unroll_indices) 217{ 218 struct translate *tr; 219 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0}; 220 struct pipe_resource *out_buffer = NULL; 221 uint8_t *out_map; 222 unsigned i, out_offset; 223 224 /* Get a translate object. */ 225 tr = translate_cache_find(mgr->translate_cache, key); 226 227 /* Map buffers we want to translate. */ 228 for (i = 0; i < mgr->b.nr_vertex_buffers; i++) { 229 if (vb_mask & (1 << i)) { 230 struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[i]; 231 unsigned offset = vb->buffer_offset + vb->stride * start_vertex; 232 uint8_t *map; 233 234 if (u_vbuf_resource(vb->buffer)->user_ptr) { 235 map = u_vbuf_resource(vb->buffer)->user_ptr + offset; 236 } else { 237 unsigned size = vb->stride ? num_vertices * vb->stride 238 : sizeof(double)*4; 239 240 if (offset+size > vb->buffer->width0) { 241 size = vb->buffer->width0 - offset; 242 } 243 244 map = pipe_buffer_map_range(mgr->pipe, vb->buffer, offset, size, 245 PIPE_TRANSFER_READ, &vb_transfer[i]); 246 } 247 248 /* Subtract min_index so that indexing with the index buffer works. */ 249 if (unroll_indices) { 250 map -= vb->stride * min_index; 251 } 252 253 tr->set_buffer(tr, i, map, vb->stride, ~0); 254 } 255 } 256 257 /* Translate. */ 258 if (unroll_indices) { 259 struct pipe_index_buffer *ib = &mgr->b.index_buffer; 260 struct pipe_transfer *transfer = NULL; 261 unsigned offset = ib->offset + start_index * ib->index_size; 262 uint8_t *map; 263 264 assert(ib->buffer && ib->index_size); 265 266 if (u_vbuf_resource(ib->buffer)->user_ptr) { 267 map = u_vbuf_resource(ib->buffer)->user_ptr + offset; 268 } else { 269 map = pipe_buffer_map_range(mgr->pipe, ib->buffer, offset, 270 num_indices * ib->index_size, 271 PIPE_TRANSFER_READ, &transfer); 272 } 273 274 /* Create and map the output buffer. */ 275 u_upload_alloc(mgr->b.uploader, 0, 276 key->output_stride * num_indices, 277 &out_offset, &out_buffer, 278 (void**)&out_map); 279 280 switch (ib->index_size) { 281 case 4: 282 tr->run_elts(tr, (unsigned*)map, num_indices, 0, out_map); 283 break; 284 case 2: 285 tr->run_elts16(tr, (uint16_t*)map, num_indices, 0, out_map); 286 break; 287 case 1: 288 tr->run_elts8(tr, map, num_indices, 0, out_map); 289 break; 290 } 291 292 if (transfer) { 293 pipe_buffer_unmap(mgr->pipe, transfer); 294 } 295 } else { 296 /* Create and map the output buffer. */ 297 u_upload_alloc(mgr->b.uploader, 298 key->output_stride * start_vertex, 299 key->output_stride * num_vertices, 300 &out_offset, &out_buffer, 301 (void**)&out_map); 302 303 out_offset -= key->output_stride * start_vertex; 304 305 tr->run(tr, 0, num_vertices, 0, out_map); 306 } 307 308 /* Unmap all buffers. */ 309 for (i = 0; i < mgr->b.nr_vertex_buffers; i++) { 310 if (vb_transfer[i]) { 311 pipe_buffer_unmap(mgr->pipe, vb_transfer[i]); 312 } 313 } 314 315 /* Setup the new vertex buffer. */ 316 mgr->b.real_vertex_buffer[out_vb].buffer_offset = out_offset; 317 mgr->b.real_vertex_buffer[out_vb].stride = key->output_stride; 318 319 /* Move the buffer reference. */ 320 pipe_resource_reference( 321 &mgr->b.real_vertex_buffer[out_vb].buffer, NULL); 322 mgr->b.real_vertex_buffer[out_vb].buffer = out_buffer; 323} 324 325static boolean 326u_vbuf_translate_find_free_vb_slots(struct u_vbuf_priv *mgr, 327 unsigned mask[VB_NUM]) 328{ 329 unsigned i, type; 330 unsigned nr = mgr->ve->count; 331 boolean used_vb[PIPE_MAX_ATTRIBS] = {0}; 332 unsigned fallback_vbs[VB_NUM]; 333 334 memset(fallback_vbs, ~0, sizeof(fallback_vbs)); 335 336 /* Mark used vertex buffers as... used. */ 337 for (i = 0; i < nr; i++) { 338 if (!mgr->ve->incompatible_layout_elem[i]) { 339 unsigned index = mgr->ve->ve[i].vertex_buffer_index; 340 341 if (!mgr->incompatible_vb[index]) { 342 used_vb[index] = TRUE; 343 } 344 } 345 } 346 347 /* Find free slots for each type if needed. */ 348 i = 0; 349 for (type = 0; type < VB_NUM; type++) { 350 if (mask[type]) { 351 for (; i < PIPE_MAX_ATTRIBS; i++) { 352 if (!used_vb[i]) { 353 /*printf("found slot=%i for type=%i\n", i, type);*/ 354 fallback_vbs[type] = i; 355 i++; 356 if (i > mgr->b.nr_real_vertex_buffers) { 357 mgr->b.nr_real_vertex_buffers = i; 358 } 359 break; 360 } 361 } 362 if (i == PIPE_MAX_ATTRIBS) { 363 /* fail, reset the number to its original value */ 364 mgr->b.nr_real_vertex_buffers = mgr->b.nr_vertex_buffers; 365 return FALSE; 366 } 367 } 368 } 369 370 memcpy(mgr->fallback_vbs, fallback_vbs, sizeof(fallback_vbs)); 371 return TRUE; 372} 373 374static boolean 375u_vbuf_translate_begin(struct u_vbuf_priv *mgr, 376 int start_vertex, unsigned num_vertices, 377 int start_instance, unsigned num_instances, 378 int start_index, unsigned num_indices, int min_index, 379 bool unroll_indices) 380{ 381 unsigned mask[VB_NUM] = {0}; 382 struct translate_key key[VB_NUM]; 383 unsigned elem_index[VB_NUM][PIPE_MAX_ATTRIBS]; /* ... into key.elements */ 384 unsigned i, type; 385 386 int start[VB_NUM] = { 387 start_vertex, /* VERTEX */ 388 start_instance, /* INSTANCE */ 389 0 /* CONST */ 390 }; 391 392 unsigned num[VB_NUM] = { 393 num_vertices, /* VERTEX */ 394 num_instances, /* INSTANCE */ 395 1 /* CONST */ 396 }; 397 398 memset(key, 0, sizeof(key)); 399 memset(elem_index, ~0, sizeof(elem_index)); 400 401 /* See if there are vertex attribs of each type to translate and 402 * which ones. */ 403 for (i = 0; i < mgr->ve->count; i++) { 404 unsigned vb_index = mgr->ve->ve[i].vertex_buffer_index; 405 406 if (!mgr->b.vertex_buffer[vb_index].stride) { 407 if (!mgr->ve->incompatible_layout_elem[i] && 408 !mgr->incompatible_vb[vb_index]) { 409 continue; 410 } 411 mask[VB_CONST] |= 1 << vb_index; 412 } else if (mgr->ve->ve[i].instance_divisor) { 413 if (!mgr->ve->incompatible_layout_elem[i] && 414 !mgr->incompatible_vb[vb_index]) { 415 continue; 416 } 417 mask[VB_INSTANCE] |= 1 << vb_index; 418 } else { 419 if (!unroll_indices && 420 !mgr->ve->incompatible_layout_elem[i] && 421 !mgr->incompatible_vb[vb_index]) { 422 continue; 423 } 424 mask[VB_VERTEX] |= 1 << vb_index; 425 } 426 } 427 428 assert(mask[VB_VERTEX] || mask[VB_INSTANCE] || mask[VB_CONST]); 429 430 /* Find free vertex buffer slots. */ 431 if (!u_vbuf_translate_find_free_vb_slots(mgr, mask)) { 432 return FALSE; 433 } 434 435 /* Initialize the translate keys. */ 436 for (i = 0; i < mgr->ve->count; i++) { 437 struct translate_key *k; 438 struct translate_element *te; 439 unsigned bit, vb_index = mgr->ve->ve[i].vertex_buffer_index; 440 bit = 1 << vb_index; 441 442 if (!mgr->ve->incompatible_layout_elem[i] && 443 !mgr->incompatible_vb[vb_index] && 444 (!unroll_indices || !(mask[VB_VERTEX] & bit))) { 445 continue; 446 } 447 448 /* Set type to what we will translate. 449 * Whether vertex, instance, or constant attribs. */ 450 for (type = 0; type < VB_NUM; type++) { 451 if (mask[type] & bit) { 452 break; 453 } 454 } 455 assert(type < VB_NUM); 456 assert(translate_is_output_format_supported(mgr->ve->native_format[i])); 457 /*printf("velem=%i type=%i\n", i, type);*/ 458 459 /* Add the vertex element. */ 460 k = &key[type]; 461 elem_index[type][i] = k->nr_elements; 462 463 te = &k->element[k->nr_elements]; 464 te->type = TRANSLATE_ELEMENT_NORMAL; 465 te->instance_divisor = 0; 466 te->input_buffer = vb_index; 467 te->input_format = mgr->ve->ve[i].src_format; 468 te->input_offset = mgr->ve->ve[i].src_offset; 469 te->output_format = mgr->ve->native_format[i]; 470 te->output_offset = k->output_stride; 471 472 k->output_stride += mgr->ve->native_format_size[i]; 473 k->nr_elements++; 474 } 475 476 /* Translate buffers. */ 477 for (type = 0; type < VB_NUM; type++) { 478 if (key[type].nr_elements) { 479 u_vbuf_translate_buffers(mgr, &key[type], mask[type], 480 mgr->fallback_vbs[type], 481 start[type], num[type], 482 start_index, num_indices, min_index, 483 unroll_indices && type == VB_VERTEX); 484 485 /* Fixup the stride for constant attribs. */ 486 if (type == VB_CONST) { 487 mgr->b.real_vertex_buffer[mgr->fallback_vbs[VB_CONST]].stride = 0; 488 } 489 } 490 } 491 492 /* Setup new vertex elements. */ 493 for (i = 0; i < mgr->ve->count; i++) { 494 for (type = 0; type < VB_NUM; type++) { 495 if (elem_index[type][i] < key[type].nr_elements) { 496 struct translate_element *te = &key[type].element[elem_index[type][i]]; 497 mgr->fallback_velems[i].instance_divisor = mgr->ve->ve[i].instance_divisor; 498 mgr->fallback_velems[i].src_format = te->output_format; 499 mgr->fallback_velems[i].src_offset = te->output_offset; 500 mgr->fallback_velems[i].vertex_buffer_index = mgr->fallback_vbs[type]; 501 502 /* elem_index[type][i] can only be set for one type. */ 503 assert(type > VB_INSTANCE || elem_index[type+1][i] == ~0); 504 assert(type > VB_VERTEX || elem_index[type+2][i] == ~0); 505 break; 506 } 507 } 508 /* No translating, just copy the original vertex element over. */ 509 if (type == VB_NUM) { 510 memcpy(&mgr->fallback_velems[i], &mgr->ve->ve[i], 511 sizeof(struct pipe_vertex_element)); 512 } 513 } 514 515 /* Preserve saved_ve. */ 516 mgr->ve_binding_lock = TRUE; 517 mgr->fallback_ve = u_vbuf_pipe_set_vertex_elements(mgr, mgr->ve->count, 518 mgr->fallback_velems); 519 mgr->ve_binding_lock = FALSE; 520 return TRUE; 521} 522 523static void u_vbuf_translate_end(struct u_vbuf_priv *mgr) 524{ 525 unsigned i; 526 527 /* Restore vertex elements. */ 528 /* Note that saved_ve will be overwritten in bind_vertex_elements_state. */ 529 mgr->pipe->bind_vertex_elements_state(mgr->pipe, mgr->saved_ve); 530 mgr->fallback_ve = NULL; 531 532 /* Unreference the now-unused VBOs. */ 533 for (i = 0; i < VB_NUM; i++) { 534 unsigned vb = mgr->fallback_vbs[i]; 535 if (vb != ~0) { 536 pipe_resource_reference(&mgr->b.real_vertex_buffer[vb].buffer, NULL); 537 mgr->fallback_vbs[i] = ~0; 538 } 539 } 540 mgr->b.nr_real_vertex_buffers = mgr->b.nr_vertex_buffers; 541} 542 543#define FORMAT_REPLACE(what, withwhat) \ 544 case PIPE_FORMAT_##what: format = PIPE_FORMAT_##withwhat; break 545 546struct u_vbuf_elements * 547u_vbuf_create_vertex_elements(struct u_vbuf *mgrb, 548 unsigned count, 549 const struct pipe_vertex_element *attribs, 550 struct pipe_vertex_element *native_attribs) 551{ 552 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb; 553 unsigned i; 554 struct u_vbuf_elements *ve = CALLOC_STRUCT(u_vbuf_elements); 555 556 ve->count = count; 557 558 if (!count) { 559 return ve; 560 } 561 562 memcpy(ve->ve, attribs, sizeof(struct pipe_vertex_element) * count); 563 memcpy(native_attribs, attribs, sizeof(struct pipe_vertex_element) * count); 564 565 /* Set the best native format in case the original format is not 566 * supported. */ 567 for (i = 0; i < count; i++) { 568 enum pipe_format format = ve->ve[i].src_format; 569 570 ve->src_format_size[i] = util_format_get_blocksize(format); 571 572 /* Choose a native format. 573 * For now we don't care about the alignment, that's going to 574 * be sorted out later. */ 575 if (!mgr->b.caps.format_fixed32) { 576 switch (format) { 577 FORMAT_REPLACE(R32_FIXED, R32_FLOAT); 578 FORMAT_REPLACE(R32G32_FIXED, R32G32_FLOAT); 579 FORMAT_REPLACE(R32G32B32_FIXED, R32G32B32_FLOAT); 580 FORMAT_REPLACE(R32G32B32A32_FIXED, R32G32B32A32_FLOAT); 581 default:; 582 } 583 } 584 if (!mgr->b.caps.format_float16) { 585 switch (format) { 586 FORMAT_REPLACE(R16_FLOAT, R32_FLOAT); 587 FORMAT_REPLACE(R16G16_FLOAT, R32G32_FLOAT); 588 FORMAT_REPLACE(R16G16B16_FLOAT, R32G32B32_FLOAT); 589 FORMAT_REPLACE(R16G16B16A16_FLOAT, R32G32B32A32_FLOAT); 590 default:; 591 } 592 } 593 if (!mgr->b.caps.format_float64) { 594 switch (format) { 595 FORMAT_REPLACE(R64_FLOAT, R32_FLOAT); 596 FORMAT_REPLACE(R64G64_FLOAT, R32G32_FLOAT); 597 FORMAT_REPLACE(R64G64B64_FLOAT, R32G32B32_FLOAT); 598 FORMAT_REPLACE(R64G64B64A64_FLOAT, R32G32B32A32_FLOAT); 599 default:; 600 } 601 } 602 if (!mgr->b.caps.format_norm32) { 603 switch (format) { 604 FORMAT_REPLACE(R32_UNORM, R32_FLOAT); 605 FORMAT_REPLACE(R32G32_UNORM, R32G32_FLOAT); 606 FORMAT_REPLACE(R32G32B32_UNORM, R32G32B32_FLOAT); 607 FORMAT_REPLACE(R32G32B32A32_UNORM, R32G32B32A32_FLOAT); 608 FORMAT_REPLACE(R32_SNORM, R32_FLOAT); 609 FORMAT_REPLACE(R32G32_SNORM, R32G32_FLOAT); 610 FORMAT_REPLACE(R32G32B32_SNORM, R32G32B32_FLOAT); 611 FORMAT_REPLACE(R32G32B32A32_SNORM, R32G32B32A32_FLOAT); 612 default:; 613 } 614 } 615 if (!mgr->b.caps.format_scaled32) { 616 switch (format) { 617 FORMAT_REPLACE(R32_USCALED, R32_FLOAT); 618 FORMAT_REPLACE(R32G32_USCALED, R32G32_FLOAT); 619 FORMAT_REPLACE(R32G32B32_USCALED, R32G32B32_FLOAT); 620 FORMAT_REPLACE(R32G32B32A32_USCALED,R32G32B32A32_FLOAT); 621 FORMAT_REPLACE(R32_SSCALED, R32_FLOAT); 622 FORMAT_REPLACE(R32G32_SSCALED, R32G32_FLOAT); 623 FORMAT_REPLACE(R32G32B32_SSCALED, R32G32B32_FLOAT); 624 FORMAT_REPLACE(R32G32B32A32_SSCALED,R32G32B32A32_FLOAT); 625 default:; 626 } 627 } 628 629 native_attribs[i].src_format = format; 630 ve->native_format[i] = format; 631 ve->native_format_size[i] = 632 util_format_get_blocksize(ve->native_format[i]); 633 634 ve->incompatible_layout_elem[i] = 635 ve->ve[i].src_format != ve->native_format[i] || 636 (!mgr->b.caps.fetch_dword_unaligned && ve->ve[i].src_offset % 4 != 0); 637 ve->incompatible_layout = 638 ve->incompatible_layout || 639 ve->incompatible_layout_elem[i]; 640 } 641 642 /* Align the formats to the size of DWORD if needed. */ 643 if (!mgr->b.caps.fetch_dword_unaligned) { 644 for (i = 0; i < count; i++) { 645 ve->native_format_size[i] = align(ve->native_format_size[i], 4); 646 } 647 } 648 649 return ve; 650} 651 652void u_vbuf_bind_vertex_elements(struct u_vbuf *mgrb, 653 void *cso, 654 struct u_vbuf_elements *ve) 655{ 656 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb; 657 658 if (!cso) { 659 return; 660 } 661 662 if (!mgr->ve_binding_lock) { 663 mgr->saved_ve = cso; 664 mgr->ve = ve; 665 } 666} 667 668void u_vbuf_destroy_vertex_elements(struct u_vbuf *mgr, 669 struct u_vbuf_elements *ve) 670{ 671 FREE(ve); 672} 673 674void u_vbuf_set_vertex_buffers(struct u_vbuf *mgrb, 675 unsigned count, 676 const struct pipe_vertex_buffer *bufs) 677{ 678 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb; 679 unsigned i; 680 681 mgr->any_user_vbs = FALSE; 682 mgr->incompatible_vb_layout = FALSE; 683 memset(mgr->incompatible_vb, 0, sizeof(mgr->incompatible_vb)); 684 685 if (!mgr->b.caps.fetch_dword_unaligned) { 686 /* Check if the strides and offsets are aligned to the size of DWORD. */ 687 for (i = 0; i < count; i++) { 688 if (bufs[i].buffer) { 689 if (bufs[i].stride % 4 != 0 || 690 bufs[i].buffer_offset % 4 != 0) { 691 mgr->incompatible_vb_layout = TRUE; 692 mgr->incompatible_vb[i] = TRUE; 693 } 694 } 695 } 696 } 697 698 for (i = 0; i < count; i++) { 699 const struct pipe_vertex_buffer *vb = &bufs[i]; 700 701 pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, vb->buffer); 702 703 mgr->b.real_vertex_buffer[i].buffer_offset = 704 mgr->b.vertex_buffer[i].buffer_offset = vb->buffer_offset; 705 706 mgr->b.real_vertex_buffer[i].stride = 707 mgr->b.vertex_buffer[i].stride = vb->stride; 708 709 if (!vb->buffer || 710 mgr->incompatible_vb[i]) { 711 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL); 712 continue; 713 } 714 715 if (u_vbuf_resource(vb->buffer)->user_ptr) { 716 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL); 717 mgr->any_user_vbs = TRUE; 718 continue; 719 } 720 721 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, vb->buffer); 722 } 723 724 for (i = count; i < mgr->b.nr_vertex_buffers; i++) { 725 pipe_resource_reference(&mgr->b.vertex_buffer[i].buffer, NULL); 726 } 727 for (i = count; i < mgr->b.nr_real_vertex_buffers; i++) { 728 pipe_resource_reference(&mgr->b.real_vertex_buffer[i].buffer, NULL); 729 } 730 731 mgr->b.nr_vertex_buffers = count; 732 mgr->b.nr_real_vertex_buffers = count; 733} 734 735void u_vbuf_set_index_buffer(struct u_vbuf *mgr, 736 const struct pipe_index_buffer *ib) 737{ 738 if (ib && ib->buffer) { 739 assert(ib->offset % ib->index_size == 0); 740 pipe_resource_reference(&mgr->index_buffer.buffer, ib->buffer); 741 mgr->index_buffer.offset = ib->offset; 742 mgr->index_buffer.index_size = ib->index_size; 743 } else { 744 pipe_resource_reference(&mgr->index_buffer.buffer, NULL); 745 } 746} 747 748static void 749u_vbuf_upload_buffers(struct u_vbuf_priv *mgr, 750 int start_vertex, unsigned num_vertices, 751 int start_instance, unsigned num_instances) 752{ 753 unsigned i; 754 unsigned nr_velems = mgr->ve->count; 755 unsigned nr_vbufs = mgr->b.nr_vertex_buffers; 756 struct pipe_vertex_element *velems = 757 mgr->fallback_ve ? mgr->fallback_velems : mgr->ve->ve; 758 unsigned start_offset[PIPE_MAX_ATTRIBS]; 759 unsigned end_offset[PIPE_MAX_ATTRIBS] = {0}; 760 761 /* Determine how much data needs to be uploaded. */ 762 for (i = 0; i < nr_velems; i++) { 763 struct pipe_vertex_element *velem = &velems[i]; 764 unsigned index = velem->vertex_buffer_index; 765 struct pipe_vertex_buffer *vb = &mgr->b.vertex_buffer[index]; 766 unsigned instance_div, first, size; 767 768 /* Skip the buffers generated by translate. */ 769 if (index == mgr->fallback_vbs[VB_VERTEX] || 770 index == mgr->fallback_vbs[VB_INSTANCE] || 771 index == mgr->fallback_vbs[VB_CONST]) { 772 continue; 773 } 774 775 assert(vb->buffer); 776 777 if (!u_vbuf_resource(vb->buffer)->user_ptr) { 778 continue; 779 } 780 781 instance_div = velem->instance_divisor; 782 first = vb->buffer_offset + velem->src_offset; 783 784 if (!vb->stride) { 785 /* Constant attrib. */ 786 size = mgr->ve->src_format_size[i]; 787 } else if (instance_div) { 788 /* Per-instance attrib. */ 789 unsigned count = (num_instances + instance_div - 1) / instance_div; 790 first += vb->stride * start_instance; 791 size = vb->stride * (count - 1) + mgr->ve->src_format_size[i]; 792 } else { 793 /* Per-vertex attrib. */ 794 first += vb->stride * start_vertex; 795 size = vb->stride * (num_vertices - 1) + mgr->ve->src_format_size[i]; 796 } 797 798 /* Update offsets. */ 799 if (!end_offset[index]) { 800 start_offset[index] = first; 801 end_offset[index] = first + size; 802 } else { 803 if (first < start_offset[index]) 804 start_offset[index] = first; 805 if (first + size > end_offset[index]) 806 end_offset[index] = first + size; 807 } 808 } 809 810 /* Upload buffers. */ 811 for (i = 0; i < nr_vbufs; i++) { 812 unsigned start, end = end_offset[i]; 813 struct pipe_vertex_buffer *real_vb; 814 uint8_t *ptr; 815 816 if (!end) { 817 continue; 818 } 819 820 start = start_offset[i]; 821 assert(start < end); 822 823 real_vb = &mgr->b.real_vertex_buffer[i]; 824 ptr = u_vbuf_resource(mgr->b.vertex_buffer[i].buffer)->user_ptr; 825 826 u_upload_data(mgr->b.uploader, start, end - start, ptr + start, 827 &real_vb->buffer_offset, &real_vb->buffer); 828 829 real_vb->buffer_offset -= start; 830 } 831} 832 833unsigned u_vbuf_draw_max_vertex_count(struct u_vbuf *mgrb) 834{ 835 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb; 836 unsigned i, nr = mgr->ve->count; 837 struct pipe_vertex_element *velems = 838 mgr->fallback_ve ? mgr->fallback_velems : mgr->ve->ve; 839 unsigned result = ~0; 840 841 for (i = 0; i < nr; i++) { 842 struct pipe_vertex_buffer *vb = 843 &mgr->b.real_vertex_buffer[velems[i].vertex_buffer_index]; 844 unsigned size, max_count, value; 845 846 /* We're not interested in constant and per-instance attribs. */ 847 if (!vb->buffer || 848 !vb->stride || 849 velems[i].instance_divisor) { 850 continue; 851 } 852 853 size = vb->buffer->width0; 854 855 /* Subtract buffer_offset. */ 856 value = vb->buffer_offset; 857 if (value >= size) { 858 return 0; 859 } 860 size -= value; 861 862 /* Subtract src_offset. */ 863 value = velems[i].src_offset; 864 if (value >= size) { 865 return 0; 866 } 867 size -= value; 868 869 /* Subtract format_size. */ 870 value = mgr->ve->native_format_size[i]; 871 if (value >= size) { 872 return 0; 873 } 874 size -= value; 875 876 /* Compute the max count. */ 877 max_count = 1 + size / vb->stride; 878 result = MIN2(result, max_count); 879 } 880 return result; 881} 882 883static boolean u_vbuf_need_minmax_index(struct u_vbuf_priv *mgr) 884{ 885 unsigned i, nr = mgr->ve->count; 886 887 for (i = 0; i < nr; i++) { 888 struct pipe_vertex_buffer *vb; 889 unsigned index; 890 891 /* Per-instance attribs don't need min/max_index. */ 892 if (mgr->ve->ve[i].instance_divisor) { 893 continue; 894 } 895 896 index = mgr->ve->ve[i].vertex_buffer_index; 897 vb = &mgr->b.vertex_buffer[index]; 898 899 /* Constant attribs don't need min/max_index. */ 900 if (!vb->stride) { 901 continue; 902 } 903 904 /* Per-vertex attribs need min/max_index. */ 905 if (u_vbuf_resource(vb->buffer)->user_ptr || 906 mgr->ve->incompatible_layout_elem[i] || 907 mgr->incompatible_vb[index]) { 908 return TRUE; 909 } 910 } 911 912 return FALSE; 913} 914 915static void u_vbuf_get_minmax_index(struct pipe_context *pipe, 916 struct pipe_index_buffer *ib, 917 const struct pipe_draw_info *info, 918 int *out_min_index, 919 int *out_max_index) 920{ 921 struct pipe_transfer *transfer = NULL; 922 const void *indices; 923 unsigned i; 924 unsigned restart_index = info->restart_index; 925 926 if (u_vbuf_resource(ib->buffer)->user_ptr) { 927 indices = u_vbuf_resource(ib->buffer)->user_ptr + 928 ib->offset + info->start * ib->index_size; 929 } else { 930 indices = pipe_buffer_map_range(pipe, ib->buffer, 931 ib->offset + info->start * ib->index_size, 932 info->count * ib->index_size, 933 PIPE_TRANSFER_READ, &transfer); 934 } 935 936 switch (ib->index_size) { 937 case 4: { 938 const unsigned *ui_indices = (const unsigned*)indices; 939 unsigned max_ui = 0; 940 unsigned min_ui = ~0U; 941 if (info->primitive_restart) { 942 for (i = 0; i < info->count; i++) { 943 if (ui_indices[i] != restart_index) { 944 if (ui_indices[i] > max_ui) max_ui = ui_indices[i]; 945 if (ui_indices[i] < min_ui) min_ui = ui_indices[i]; 946 } 947 } 948 } 949 else { 950 for (i = 0; i < info->count; i++) { 951 if (ui_indices[i] > max_ui) max_ui = ui_indices[i]; 952 if (ui_indices[i] < min_ui) min_ui = ui_indices[i]; 953 } 954 } 955 *out_min_index = min_ui; 956 *out_max_index = max_ui; 957 break; 958 } 959 case 2: { 960 const unsigned short *us_indices = (const unsigned short*)indices; 961 unsigned max_us = 0; 962 unsigned min_us = ~0U; 963 if (info->primitive_restart) { 964 for (i = 0; i < info->count; i++) { 965 if (us_indices[i] != restart_index) { 966 if (us_indices[i] > max_us) max_us = us_indices[i]; 967 if (us_indices[i] < min_us) min_us = us_indices[i]; 968 } 969 } 970 } 971 else { 972 for (i = 0; i < info->count; i++) { 973 if (us_indices[i] > max_us) max_us = us_indices[i]; 974 if (us_indices[i] < min_us) min_us = us_indices[i]; 975 } 976 } 977 *out_min_index = min_us; 978 *out_max_index = max_us; 979 break; 980 } 981 case 1: { 982 const unsigned char *ub_indices = (const unsigned char*)indices; 983 unsigned max_ub = 0; 984 unsigned min_ub = ~0U; 985 if (info->primitive_restart) { 986 for (i = 0; i < info->count; i++) { 987 if (ub_indices[i] != restart_index) { 988 if (ub_indices[i] > max_ub) max_ub = ub_indices[i]; 989 if (ub_indices[i] < min_ub) min_ub = ub_indices[i]; 990 } 991 } 992 } 993 else { 994 for (i = 0; i < info->count; i++) { 995 if (ub_indices[i] > max_ub) max_ub = ub_indices[i]; 996 if (ub_indices[i] < min_ub) min_ub = ub_indices[i]; 997 } 998 } 999 *out_min_index = min_ub; 1000 *out_max_index = max_ub; 1001 break; 1002 } 1003 default: 1004 assert(0); 1005 *out_min_index = 0; 1006 *out_max_index = 0; 1007 } 1008 1009 if (transfer) { 1010 pipe_buffer_unmap(pipe, transfer); 1011 } 1012} 1013 1014enum u_vbuf_return_flags 1015u_vbuf_draw_begin(struct u_vbuf *mgrb, 1016 struct pipe_draw_info *info) 1017{ 1018 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb; 1019 int start_vertex, min_index; 1020 unsigned num_vertices; 1021 bool unroll_indices = false; 1022 1023 if (!mgr->incompatible_vb_layout && 1024 !mgr->ve->incompatible_layout && 1025 !mgr->any_user_vbs) { 1026 return 0; 1027 } 1028 1029 if (info->indexed) { 1030 int max_index; 1031 bool index_bounds_valid = false; 1032 1033 if (info->max_index != ~0) { 1034 min_index = info->min_index; 1035 max_index = info->max_index; 1036 index_bounds_valid = true; 1037 } else if (u_vbuf_need_minmax_index(mgr)) { 1038 u_vbuf_get_minmax_index(mgr->pipe, &mgr->b.index_buffer, info, 1039 &min_index, &max_index); 1040 index_bounds_valid = true; 1041 } 1042 1043 /* If the index bounds are valid, it means some upload or translation 1044 * of per-vertex attribs will be performed. */ 1045 if (index_bounds_valid) { 1046 assert(min_index <= max_index); 1047 1048 start_vertex = min_index + info->index_bias; 1049 num_vertices = max_index + 1 - min_index; 1050 1051 /* Primitive restart doesn't work when unrolling indices. 1052 * We would have to break this drawing operation into several ones. */ 1053 /* Use some heuristic to see if unrolling indices improves 1054 * performance. */ 1055 if (!info->primitive_restart && 1056 num_vertices > info->count*2 && 1057 num_vertices-info->count > 32) { 1058 /*printf("num_vertices=%i count=%i\n", num_vertices, info->count);*/ 1059 unroll_indices = true; 1060 } 1061 } else { 1062 /* Nothing to do for per-vertex attribs. */ 1063 start_vertex = 0; 1064 num_vertices = 0; 1065 min_index = 0; 1066 } 1067 } else { 1068 start_vertex = info->start; 1069 num_vertices = info->count; 1070 min_index = 0; 1071 } 1072 1073 /* Translate vertices with non-native layouts or formats. */ 1074 if (unroll_indices || 1075 mgr->incompatible_vb_layout || 1076 mgr->ve->incompatible_layout) { 1077 /* XXX check the return value */ 1078 u_vbuf_translate_begin(mgr, start_vertex, num_vertices, 1079 info->start_instance, info->instance_count, 1080 info->start, info->count, min_index, 1081 unroll_indices); 1082 } 1083 1084 /* Upload user buffers. */ 1085 if (mgr->any_user_vbs) { 1086 u_vbuf_upload_buffers(mgr, start_vertex, num_vertices, 1087 info->start_instance, info->instance_count); 1088 } 1089 1090 /* 1091 if (unroll_indices) { 1092 printf("unrolling indices: start_vertex = %i, num_vertices = %i\n", 1093 start_vertex, num_vertices); 1094 util_dump_draw_info(stdout, info); 1095 printf("\n"); 1096 } 1097 1098 unsigned i; 1099 for (i = 0; i < mgr->b.nr_vertex_buffers; i++) { 1100 printf("input %i: ", i); 1101 util_dump_vertex_buffer(stdout, mgr->b.vertex_buffer+i); 1102 printf("\n"); 1103 } 1104 for (i = 0; i < mgr->b.nr_real_vertex_buffers; i++) { 1105 printf("real %i: ", i); 1106 util_dump_vertex_buffer(stdout, mgr->b.real_vertex_buffer+i); 1107 printf("\n"); 1108 } 1109 */ 1110 1111 if (unroll_indices) { 1112 info->indexed = FALSE; 1113 info->index_bias = 0; 1114 info->min_index = 0; 1115 info->max_index = info->count - 1; 1116 info->start = 0; 1117 } 1118 1119 return U_VBUF_BUFFERS_UPDATED; 1120} 1121 1122void u_vbuf_draw_end(struct u_vbuf *mgrb) 1123{ 1124 struct u_vbuf_priv *mgr = (struct u_vbuf_priv*)mgrb; 1125 1126 if (mgr->fallback_ve) { 1127 u_vbuf_translate_end(mgr); 1128 } 1129} 1130