r300_render.c revision 50db6dba65560c1fb598d495d7d4103019bbbea5
1/* 2 * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> 3 * Copyright 2010 Marek Olšák <maraeo@gmail.com> 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the "Software"), 7 * to deal in the Software without restriction, including without limitation 8 * on the rights to use, copy, modify, merge, publish, distribute, sub 9 * license, and/or sell copies of the Software, and to permit persons to whom 10 * the Software is furnished to do so, subject to the following conditions: 11 * 12 * The above copyright notice and this permission notice (including the next 13 * paragraph) shall be included in all copies or substantial portions of the 14 * 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 NON-INFRINGEMENT. IN NO EVENT SHALL 19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, 20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */ 23 24/* r300_render: Vertex and index buffer primitive emission. Contains both 25 * HW TCL fastpath rendering, and SW TCL Draw-assisted rendering. */ 26 27#include "draw/draw_context.h" 28#include "draw/draw_vbuf.h" 29 30#include "util/u_inlines.h" 31 32#include "util/u_format.h" 33#include "util/u_memory.h" 34#include "util/u_upload_mgr.h" 35#include "util/u_prim.h" 36 37#include "r300_cs.h" 38#include "r300_cb.h" 39#include "r300_context.h" 40#include "r300_screen_buffer.h" 41#include "r300_emit.h" 42#include "r300_reg.h" 43#include "r300_state_derived.h" 44 45#include <limits.h> 46 47#define IMMD_DWORDS 32 48 49static uint32_t r300_translate_primitive(unsigned prim) 50{ 51 switch (prim) { 52 case PIPE_PRIM_POINTS: 53 return R300_VAP_VF_CNTL__PRIM_POINTS; 54 case PIPE_PRIM_LINES: 55 return R300_VAP_VF_CNTL__PRIM_LINES; 56 case PIPE_PRIM_LINE_LOOP: 57 return R300_VAP_VF_CNTL__PRIM_LINE_LOOP; 58 case PIPE_PRIM_LINE_STRIP: 59 return R300_VAP_VF_CNTL__PRIM_LINE_STRIP; 60 case PIPE_PRIM_TRIANGLES: 61 return R300_VAP_VF_CNTL__PRIM_TRIANGLES; 62 case PIPE_PRIM_TRIANGLE_STRIP: 63 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP; 64 case PIPE_PRIM_TRIANGLE_FAN: 65 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN; 66 case PIPE_PRIM_QUADS: 67 return R300_VAP_VF_CNTL__PRIM_QUADS; 68 case PIPE_PRIM_QUAD_STRIP: 69 return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP; 70 case PIPE_PRIM_POLYGON: 71 return R300_VAP_VF_CNTL__PRIM_POLYGON; 72 default: 73 return 0; 74 } 75} 76 77static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, 78 unsigned mode) 79{ 80 struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state; 81 uint32_t color_control = rs->color_control; 82 83 /* By default (see r300_state.c:r300_create_rs_state) color_control is 84 * initialized to provoking the first vertex. 85 * 86 * Triangle fans must be reduced to the second vertex, not the first, in 87 * Gallium flatshade-first mode, as per the GL spec. 88 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt) 89 * 90 * Quads never provoke correctly in flatshade-first mode. The first 91 * vertex is never considered as provoking, so only the second, third, 92 * and fourth vertices can be selected, and both "third" and "last" modes 93 * select the fourth vertex. This is probably due to D3D lacking quads. 94 * 95 * Similarly, polygons reduce to the first, not the last, vertex, when in 96 * "last" mode, and all other modes start from the second vertex. 97 * 98 * ~ C. 99 */ 100 101 if (rs->rs.flatshade_first) { 102 switch (mode) { 103 case PIPE_PRIM_TRIANGLE_FAN: 104 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND; 105 break; 106 case PIPE_PRIM_QUADS: 107 case PIPE_PRIM_QUAD_STRIP: 108 case PIPE_PRIM_POLYGON: 109 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; 110 break; 111 default: 112 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST; 113 break; 114 } 115 } else { 116 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; 117 } 118 119 return color_control; 120} 121 122static boolean index_bias_supported(struct r300_context *r300) 123{ 124 return r300->screen->caps.is_r500 && 125 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); 126} 127 128static void r500_emit_index_bias(struct r300_context *r300, int index_bias) 129{ 130 CS_LOCALS(r300); 131 132 BEGIN_CS(2); 133 OUT_CS_REG(R500_VAP_INDEX_OFFSET, 134 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0)); 135 END_CS; 136} 137 138/* This function splits the index bias value into two parts: 139 * - buffer_offset: the value that can be safely added to buffer offsets 140 * in r300_emit_aos (it must yield a positive offset when added to 141 * a vertex buffer offset) 142 * - index_offset: the value that must be manually subtracted from indices 143 * in an index buffer to achieve negative offsets. */ 144static void r300_split_index_bias(struct r300_context *r300, int index_bias, 145 int *buffer_offset, int *index_offset) 146{ 147 struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer; 148 struct pipe_vertex_element *velem = r300->velems->velem; 149 unsigned i, size; 150 int max_neg_bias; 151 152 if (index_bias < 0) { 153 /* See how large index bias we may subtract. We must be careful 154 * here because negative buffer offsets are not allowed 155 * by the DRM API. */ 156 max_neg_bias = INT_MAX; 157 for (i = 0; i < r300->velems->count; i++) { 158 vb = &vbufs[velem[i].vertex_buffer_index]; 159 size = (vb->buffer_offset + velem[i].src_offset) / vb->stride; 160 max_neg_bias = MIN2(max_neg_bias, size); 161 } 162 163 /* Now set the minimum allowed value. */ 164 *buffer_offset = MAX2(-max_neg_bias, index_bias); 165 } else { 166 /* A positive index bias is OK. */ 167 *buffer_offset = index_bias; 168 } 169 170 *index_offset = index_bias - *buffer_offset; 171} 172 173enum r300_prepare_flags { 174 PREP_FIRST_DRAW = (1 << 0), /* call emit_dirty_state and friends? */ 175 PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */ 176 PREP_EMIT_AOS = (1 << 2), /* call emit_aos? */ 177 PREP_EMIT_AOS_SWTCL = (1 << 3), /* call emit_aos_swtcl? */ 178 PREP_INDEXED = (1 << 4) /* is this draw_elements? */ 179}; 180 181/** 182 * Check if the requested number of dwords is available in the CS and 183 * if not, flush. Then validate buffers and emit dirty state. 184 * \param r300 The context. 185 * \param flags See r300_prepare_flags. 186 * \param index_buffer The index buffer to validate. The parameter may be NULL. 187 * \param cs_dwords The number of dwords to reserve in CS. 188 * \param aos_offset The offset passed to emit_aos. 189 * \param index_bias The index bias to emit. 190 * \param end_cs_dwords The number of free dwords which must be available 191 * at the end of CS after drawing in case the CS space 192 * management is performed by a draw_* function manually. 193 * The parameter may be NULL. 194 */ 195static void r300_prepare_for_rendering(struct r300_context *r300, 196 enum r300_prepare_flags flags, 197 struct pipe_resource *index_buffer, 198 unsigned cs_dwords, 199 int aos_offset, 200 int index_bias, 201 unsigned *end_cs_dwords) 202{ 203 unsigned end_dwords = 0; 204 boolean flushed = FALSE; 205 boolean first_draw = flags & PREP_FIRST_DRAW; 206 boolean emit_aos = flags & PREP_EMIT_AOS; 207 boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL; 208 boolean indexed = flags & PREP_INDEXED; 209 boolean hw_index_bias = index_bias_supported(r300); 210 211 /* Add dirty state, index offset, and AOS. */ 212 if (first_draw) { 213 cs_dwords += r300_get_num_dirty_dwords(r300); 214 215 if (hw_index_bias) 216 cs_dwords += 2; /* emit_index_offset */ 217 218 if (emit_aos) 219 cs_dwords += 55; /* emit_aos */ 220 221 if (emit_aos_swtcl) 222 cs_dwords += 7; /* emit_aos_swtcl */ 223 } 224 225 /* Emitted in flush. */ 226 end_dwords += 26; /* emit_query_end */ 227 end_dwords += r300->hyperz_state.size; /* emit_hyperz_end */ 228 229 cs_dwords += end_dwords; 230 231 /* Reserve requested CS space. */ 232 if (!r300_check_cs(r300, cs_dwords)) { 233 r300->context.flush(&r300->context, 0, NULL); 234 flushed = TRUE; 235 } 236 237 /* Validate buffers and emit dirty state if needed. */ 238 if (first_draw || flushed) { 239 r300_emit_buffer_validate(r300, flags & PREP_VALIDATE_VBOS, index_buffer); 240 r300_emit_dirty_state(r300); 241 if (hw_index_bias) { 242 if (r300->screen->caps.has_tcl) 243 r500_emit_index_bias(r300, index_bias); 244 else 245 r500_emit_index_bias(r300, 0); 246 } 247 248 if (emit_aos) 249 r300_emit_aos(r300, aos_offset, indexed); 250 251 if (emit_aos_swtcl) 252 r300_emit_aos_swtcl(r300, indexed); 253 } 254 255 if (end_cs_dwords) 256 *end_cs_dwords = end_dwords; 257} 258 259static boolean immd_is_good_idea(struct r300_context *r300, 260 unsigned count) 261{ 262 struct pipe_vertex_element* velem; 263 struct pipe_vertex_buffer* vbuf; 264 boolean checked[PIPE_MAX_ATTRIBS] = {0}; 265 unsigned vertex_element_count = r300->velems->count; 266 unsigned i, vbi; 267 268 if (DBG_ON(r300, DBG_NO_IMMD)) { 269 return FALSE; 270 } 271 272 if (r300->draw) { 273 return FALSE; 274 } 275 276 if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) { 277 return FALSE; 278 } 279 280 /* We shouldn't map buffers referenced by CS, busy buffers, 281 * and ones placed in VRAM. */ 282 for (i = 0; i < vertex_element_count; i++) { 283 velem = &r300->velems->velem[i]; 284 vbi = velem->vertex_buffer_index; 285 286 if (!checked[vbi]) { 287 vbuf = &r300->vertex_buffer[vbi]; 288 289 if (!(r300_buffer(vbuf->buffer)->domain & R300_DOMAIN_GTT)) { 290 return FALSE; 291 } 292 293 if (r300_buffer_is_referenced(&r300->context, 294 vbuf->buffer, 295 R300_REF_CS | R300_REF_HW)) { 296 /* It's a very bad idea to map it... */ 297 return FALSE; 298 } 299 checked[vbi] = TRUE; 300 } 301 } 302 return TRUE; 303} 304 305/***************************************************************************** 306 * The HWTCL draw functions. * 307 ****************************************************************************/ 308 309static void r300_emit_draw_arrays_immediate(struct r300_context *r300, 310 unsigned mode, 311 unsigned start, 312 unsigned count) 313{ 314 struct pipe_vertex_element* velem; 315 struct pipe_vertex_buffer* vbuf; 316 unsigned vertex_element_count = r300->velems->count; 317 unsigned i, v, vbi, dwords; 318 319 /* Size of the vertex, in dwords. */ 320 unsigned vertex_size = r300->velems->vertex_size_dwords; 321 322 /* Size of the vertex element, in dwords. */ 323 unsigned size[PIPE_MAX_ATTRIBS]; 324 325 /* Stride to the same attrib in the next vertex in the vertex buffer, 326 * in dwords. */ 327 unsigned stride[PIPE_MAX_ATTRIBS]; 328 329 /* Mapped vertex buffers. */ 330 uint32_t* map[PIPE_MAX_ATTRIBS]; 331 uint32_t* mapelem[PIPE_MAX_ATTRIBS]; 332 struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {0}; 333 334 CB_LOCALS; 335 336 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */ 337 for (i = 0; i < vertex_element_count; i++) { 338 velem = &r300->velems->velem[i]; 339 size[i] = r300->velems->hw_format_size[i] / 4; 340 vbi = velem->vertex_buffer_index; 341 vbuf = &r300->vertex_buffer[vbi]; 342 stride[i] = vbuf->stride / 4; 343 344 /* Map the buffer. */ 345 if (!transfer[vbi]) { 346 map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context, 347 vbuf->buffer, 348 PIPE_TRANSFER_READ, 349 &transfer[vbi]); 350 map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * start; 351 } 352 mapelem[i] = map[vbi] + (velem->src_offset / 4); 353 } 354 355 dwords = 9 + count * vertex_size; 356 357 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0, NULL); 358 359 BEGIN_CS_AS_CB(r300, dwords); 360 OUT_CB_REG(R300_GA_COLOR_CONTROL, 361 r300_provoking_vertex_fixes(r300, mode)); 362 OUT_CB_REG(R300_VAP_VTX_SIZE, vertex_size); 363 OUT_CB_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 364 OUT_CB(count - 1); 365 OUT_CB(0); 366 OUT_CB_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size); 367 OUT_CB(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) | 368 r300_translate_primitive(mode)); 369 370 /* Emit vertices. */ 371 for (v = 0; v < count; v++) { 372 for (i = 0; i < vertex_element_count; i++) { 373 OUT_CB_TABLE(&mapelem[i][stride[i] * v], size[i]); 374 } 375 } 376 END_CB; 377 378 /* Unmap buffers. */ 379 for (i = 0; i < vertex_element_count; i++) { 380 vbi = r300->velems->velem[i].vertex_buffer_index; 381 382 if (transfer[vbi]) { 383 vbuf = &r300->vertex_buffer[vbi]; 384 pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]); 385 transfer[vbi] = NULL; 386 } 387 } 388} 389 390static void r300_emit_draw_arrays(struct r300_context *r300, 391 unsigned mode, 392 unsigned count) 393{ 394 boolean alt_num_verts = count > 65535; 395 CS_LOCALS(r300); 396 397 if (count >= (1 << 24)) { 398 fprintf(stderr, "r300: Got a huge number of vertices: %i, " 399 "refusing to render.\n", count); 400 return; 401 } 402 403 BEGIN_CS(7 + (alt_num_verts ? 2 : 0)); 404 if (alt_num_verts) { 405 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); 406 } 407 OUT_CS_REG(R300_GA_COLOR_CONTROL, 408 r300_provoking_vertex_fixes(r300, mode)); 409 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 410 OUT_CS(count - 1); 411 OUT_CS(0); 412 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); 413 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | 414 r300_translate_primitive(mode) | 415 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 416 END_CS; 417} 418 419static void r300_emit_draw_elements(struct r300_context *r300, 420 struct pipe_resource* indexBuffer, 421 unsigned indexSize, 422 unsigned minIndex, 423 unsigned maxIndex, 424 unsigned mode, 425 unsigned start, 426 unsigned count) 427{ 428 uint32_t count_dwords; 429 uint32_t offset_dwords = indexSize * start / sizeof(uint32_t); 430 boolean alt_num_verts = count > 65535; 431 CS_LOCALS(r300); 432 433 if (count >= (1 << 24)) { 434 fprintf(stderr, "r300: Got a huge number of vertices: %i, " 435 "refusing to render.\n", count); 436 return; 437 } 438 439 maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index); 440 441 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n", 442 count, minIndex, maxIndex); 443 444 BEGIN_CS(13 + (alt_num_verts ? 2 : 0)); 445 if (alt_num_verts) { 446 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); 447 } 448 OUT_CS_REG(R300_GA_COLOR_CONTROL, 449 r300_provoking_vertex_fixes(r300, mode)); 450 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 451 OUT_CS(maxIndex); 452 OUT_CS(minIndex); 453 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0); 454 if (indexSize == 4) { 455 count_dwords = count; 456 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | 457 R300_VAP_VF_CNTL__INDEX_SIZE_32bit | 458 r300_translate_primitive(mode) | 459 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 460 } else { 461 count_dwords = (count + 1) / 2; 462 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | 463 r300_translate_primitive(mode) | 464 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 465 } 466 467 /* INDX_BUFFER is a truly special packet3. 468 * Unlike most other packet3, where the offset is after the count, 469 * the order is reversed, so the relocation ends up carrying the 470 * size of the indexbuf instead of the offset. 471 */ 472 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2); 473 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) | 474 (0 << R300_INDX_BUFFER_SKIP_SHIFT)); 475 OUT_CS(offset_dwords << 2); 476 OUT_CS_BUF_RELOC(indexBuffer, count_dwords, 477 r300_buffer(indexBuffer)->domain, 0, 0); 478 479 END_CS; 480} 481 482/* This is the fast-path drawing & emission for HW TCL. */ 483static void r300_draw_range_elements(struct pipe_context* pipe, 484 struct pipe_resource* indexBuffer, 485 unsigned indexSize, 486 int indexBias, 487 unsigned minIndex, 488 unsigned maxIndex, 489 unsigned mode, 490 unsigned start, 491 unsigned count) 492{ 493 struct r300_context* r300 = r300_context(pipe); 494 struct pipe_resource* orgIndexBuffer = indexBuffer; 495 boolean alt_num_verts = r300->screen->caps.is_r500 && 496 count > 65536 && 497 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); 498 unsigned short_count; 499 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */ 500 boolean translate = FALSE; 501 502 if (r300->skip_rendering) { 503 return; 504 } 505 506 if (!u_trim_pipe_prim(mode, &count)) { 507 return; 508 } 509 510 /* Index buffer range checking. */ 511 if ((start + count) * indexSize > indexBuffer->width0) { 512 fprintf(stderr, "r300: Invalid index buffer range. Skipping rendering.\n"); 513 return; 514 } 515 516 /* Set up fallback for incompatible vertex layout if needed. */ 517 if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) { 518 r300_begin_vertex_translate(r300); 519 translate = TRUE; 520 } 521 522 if (indexBias && !index_bias_supported(r300)) { 523 r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset); 524 } 525 526 r300_translate_index_buffer(r300, &indexBuffer, &indexSize, index_offset, 527 &start, count); 528 529 r300_update_derived_state(r300); 530 r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count); 531 532 /* 15 dwords for emit_draw_elements */ 533 r300_prepare_for_rendering(r300, 534 PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, 535 indexBuffer, 15, buffer_offset, indexBias, NULL); 536 537 u_upload_flush(r300->upload_vb); 538 u_upload_flush(r300->upload_ib); 539 if (alt_num_verts || count <= 65535) { 540 r300_emit_draw_elements(r300, indexBuffer, indexSize, 541 minIndex, maxIndex, mode, start, count); 542 } else { 543 do { 544 short_count = MIN2(count, 65534); 545 r300_emit_draw_elements(r300, indexBuffer, indexSize, 546 minIndex, maxIndex, 547 mode, start, short_count); 548 549 start += short_count; 550 count -= short_count; 551 552 /* 15 dwords for emit_draw_elements */ 553 if (count) { 554 r300_prepare_for_rendering(r300, 555 PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, 556 indexBuffer, 15, buffer_offset, indexBias, NULL); 557 } 558 } while (count); 559 } 560 561 if (indexBuffer != orgIndexBuffer) { 562 pipe_resource_reference( &indexBuffer, NULL ); 563 } 564 565 if (translate) { 566 r300_end_vertex_translate(r300); 567 } 568} 569 570/* Simple helpers for context setup. Should probably be moved to util. */ 571static void r300_draw_elements(struct pipe_context* pipe, 572 struct pipe_resource* indexBuffer, 573 unsigned indexSize, int indexBias, unsigned mode, 574 unsigned start, unsigned count) 575{ 576 struct r300_context *r300 = r300_context(pipe); 577 578 pipe->draw_range_elements(pipe, indexBuffer, indexSize, indexBias, 579 0, r300->vertex_buffer_max_index, 580 mode, start, count); 581} 582 583static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, 584 unsigned start, unsigned count) 585{ 586 struct r300_context* r300 = r300_context(pipe); 587 boolean alt_num_verts = r300->screen->caps.is_r500 && 588 count > 65536 && 589 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); 590 unsigned short_count; 591 boolean translate = FALSE; 592 593 if (r300->skip_rendering) { 594 return; 595 } 596 597 if (!u_trim_pipe_prim(mode, &count)) { 598 return; 599 } 600 601 /* Set up fallback for incompatible vertex layout if needed. */ 602 if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) { 603 r300_begin_vertex_translate(r300); 604 translate = TRUE; 605 } 606 607 r300_update_derived_state(r300); 608 609 if (immd_is_good_idea(r300, count)) { 610 r300_emit_draw_arrays_immediate(r300, mode, start, count); 611 } else { 612 /* 9 spare dwords for emit_draw_arrays. */ 613 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS, 614 NULL, 9, start, 0, NULL); 615 616 if (alt_num_verts || count <= 65535) { 617 r300_emit_draw_arrays(r300, mode, count); 618 } else { 619 do { 620 short_count = MIN2(count, 65535); 621 r300_emit_draw_arrays(r300, mode, short_count); 622 623 start += short_count; 624 count -= short_count; 625 626 /* 9 spare dwords for emit_draw_arrays. */ 627 if (count) { 628 r300_prepare_for_rendering(r300, 629 PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9, 630 start, 0, NULL); 631 } 632 } while (count); 633 } 634 u_upload_flush(r300->upload_vb); 635 } 636 637 if (translate) { 638 r300_end_vertex_translate(r300); 639 } 640} 641 642/**************************************************************************** 643 * The rest of this file is for SW TCL rendering only. Please be polite and * 644 * keep these functions separated so that they are easier to locate. ~C. * 645 ***************************************************************************/ 646 647/* SW TCL arrays, using Draw. */ 648static void r300_swtcl_draw_arrays(struct pipe_context* pipe, 649 unsigned mode, 650 unsigned start, 651 unsigned count) 652{ 653 struct r300_context* r300 = r300_context(pipe); 654 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS]; 655 int i; 656 657 if (r300->skip_rendering) { 658 return; 659 } 660 661 if (!u_trim_pipe_prim(mode, &count)) { 662 return; 663 } 664 665 r300_update_derived_state(r300); 666 667 for (i = 0; i < r300->vertex_buffer_count; i++) { 668 void* buf = pipe_buffer_map(pipe, 669 r300->vertex_buffer[i].buffer, 670 PIPE_TRANSFER_READ, 671 &vb_transfer[i]); 672 draw_set_mapped_vertex_buffer(r300->draw, i, buf); 673 } 674 675 draw_set_mapped_element_buffer(r300->draw, 0, 0, NULL); 676 677 draw_arrays(r300->draw, mode, start, count); 678 679 /* XXX Not sure whether this is the best fix. 680 * It prevents CS from being rejected and weird assertion failures. */ 681 draw_flush(r300->draw); 682 683 for (i = 0; i < r300->vertex_buffer_count; i++) { 684 pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, 685 vb_transfer[i]); 686 draw_set_mapped_vertex_buffer(r300->draw, i, NULL); 687 } 688} 689 690/* SW TCL elements, using Draw. */ 691static void r300_swtcl_draw_range_elements(struct pipe_context* pipe, 692 struct pipe_resource* indexBuffer, 693 unsigned indexSize, 694 int indexBias, 695 unsigned minIndex, 696 unsigned maxIndex, 697 unsigned mode, 698 unsigned start, 699 unsigned count) 700{ 701 struct r300_context* r300 = r300_context(pipe); 702 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS]; 703 struct pipe_transfer *ib_transfer; 704 int i; 705 void* indices; 706 707 if (r300->skip_rendering) { 708 return; 709 } 710 711 if (!u_trim_pipe_prim(mode, &count)) { 712 return; 713 } 714 715 r300_update_derived_state(r300); 716 717 for (i = 0; i < r300->vertex_buffer_count; i++) { 718 void* buf = pipe_buffer_map(pipe, 719 r300->vertex_buffer[i].buffer, 720 PIPE_TRANSFER_READ, 721 &vb_transfer[i]); 722 draw_set_mapped_vertex_buffer(r300->draw, i, buf); 723 } 724 725 indices = pipe_buffer_map(pipe, indexBuffer, 726 PIPE_TRANSFER_READ, &ib_transfer); 727 draw_set_mapped_element_buffer_range(r300->draw, indexSize, indexBias, 728 minIndex, maxIndex, indices); 729 730 draw_arrays(r300->draw, mode, start, count); 731 732 /* XXX Not sure whether this is the best fix. 733 * It prevents CS from being rejected and weird assertion failures. */ 734 draw_flush(r300->draw); 735 736 for (i = 0; i < r300->vertex_buffer_count; i++) { 737 pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, 738 vb_transfer[i]); 739 draw_set_mapped_vertex_buffer(r300->draw, i, NULL); 740 } 741 742 pipe_buffer_unmap(pipe, indexBuffer, 743 ib_transfer); 744 draw_set_mapped_element_buffer_range(r300->draw, 0, 0, 745 start, start + count - 1, 746 NULL); 747} 748 749/* Object for rendering using Draw. */ 750struct r300_render { 751 /* Parent class */ 752 struct vbuf_render base; 753 754 /* Pipe context */ 755 struct r300_context* r300; 756 757 /* Vertex information */ 758 size_t vertex_size; 759 unsigned prim; 760 unsigned hwprim; 761 762 /* VBO */ 763 struct pipe_resource* vbo; 764 size_t vbo_size; 765 size_t vbo_offset; 766 size_t vbo_max_used; 767 void * vbo_ptr; 768 769 struct pipe_transfer *vbo_transfer; 770}; 771 772static INLINE struct r300_render* 773r300_render(struct vbuf_render* render) 774{ 775 return (struct r300_render*)render; 776} 777 778static const struct vertex_info* 779r300_render_get_vertex_info(struct vbuf_render* render) 780{ 781 struct r300_render* r300render = r300_render(render); 782 struct r300_context* r300 = r300render->r300; 783 784 return &r300->vertex_info; 785} 786 787static boolean r300_render_allocate_vertices(struct vbuf_render* render, 788 ushort vertex_size, 789 ushort count) 790{ 791 struct r300_render* r300render = r300_render(render); 792 struct r300_context* r300 = r300render->r300; 793 struct pipe_screen* screen = r300->context.screen; 794 size_t size = (size_t)vertex_size * (size_t)count; 795 796 if (size + r300render->vbo_offset > r300render->vbo_size) 797 { 798 pipe_resource_reference(&r300->vbo, NULL); 799 r300render->vbo = pipe_buffer_create(screen, 800 PIPE_BIND_VERTEX_BUFFER, 801 R300_MAX_DRAW_VBO_SIZE); 802 r300render->vbo_offset = 0; 803 r300render->vbo_size = R300_MAX_DRAW_VBO_SIZE; 804 } 805 806 r300render->vertex_size = vertex_size; 807 r300->vbo = r300render->vbo; 808 r300->vbo_offset = r300render->vbo_offset; 809 810 return (r300render->vbo) ? TRUE : FALSE; 811} 812 813static void* r300_render_map_vertices(struct vbuf_render* render) 814{ 815 struct r300_render* r300render = r300_render(render); 816 817 assert(!r300render->vbo_transfer); 818 819 r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context, 820 r300render->vbo, 821 PIPE_TRANSFER_WRITE, 822 &r300render->vbo_transfer); 823 824 return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset); 825} 826 827static void r300_render_unmap_vertices(struct vbuf_render* render, 828 ushort min, 829 ushort max) 830{ 831 struct r300_render* r300render = r300_render(render); 832 struct pipe_context* context = &r300render->r300->context; 833 834 assert(r300render->vbo_transfer); 835 836 r300render->vbo_max_used = MAX2(r300render->vbo_max_used, 837 r300render->vertex_size * (max + 1)); 838 pipe_buffer_unmap(context, r300render->vbo, r300render->vbo_transfer); 839 840 r300render->vbo_transfer = NULL; 841} 842 843static void r300_render_release_vertices(struct vbuf_render* render) 844{ 845 struct r300_render* r300render = r300_render(render); 846 847 r300render->vbo_offset += r300render->vbo_max_used; 848 r300render->vbo_max_used = 0; 849} 850 851static boolean r300_render_set_primitive(struct vbuf_render* render, 852 unsigned prim) 853{ 854 struct r300_render* r300render = r300_render(render); 855 856 r300render->prim = prim; 857 r300render->hwprim = r300_translate_primitive(prim); 858 859 return TRUE; 860} 861 862static void r300_render_draw_arrays(struct vbuf_render* render, 863 unsigned start, 864 unsigned count) 865{ 866 struct r300_render* r300render = r300_render(render); 867 struct r300_context* r300 = r300render->r300; 868 uint8_t* ptr; 869 unsigned i; 870 unsigned dwords = 6; 871 872 CS_LOCALS(r300); 873 (void) i; (void) ptr; 874 875 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL, 876 NULL, dwords, 0, 0, NULL); 877 878 DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count); 879 880 /* Uncomment to dump all VBOs rendered through this interface. 881 * Slow and noisy! 882 ptr = pipe_buffer_map(&r300render->r300->context, 883 r300render->vbo, PIPE_TRANSFER_READ, 884 &r300render->vbo_transfer); 885 886 for (i = 0; i < count; i++) { 887 printf("r300: Vertex %d\n", i); 888 draw_dump_emitted_vertex(&r300->vertex_info, ptr); 889 ptr += r300->vertex_info.size * 4; 890 printf("\n"); 891 } 892 893 pipe_buffer_unmap(&r300render->r300->context, r300render->vbo, 894 r300render->vbo_transfer); 895 */ 896 897 BEGIN_CS(dwords); 898 OUT_CS_REG(R300_GA_COLOR_CONTROL, 899 r300_provoking_vertex_fixes(r300, r300render->prim)); 900 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1); 901 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); 902 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | 903 r300render->hwprim); 904 END_CS; 905} 906 907static void r300_render_draw_elements(struct vbuf_render* render, 908 const ushort* indices, 909 uint count) 910{ 911 struct r300_render* r300render = r300_render(render); 912 struct r300_context* r300 = r300render->r300; 913 int i; 914 unsigned end_cs_dwords; 915 unsigned max_index = (r300render->vbo_size - r300render->vbo_offset) / 916 (r300render->r300->vertex_info.size * 4) - 1; 917 unsigned short_count; 918 unsigned free_dwords; 919 920 CS_LOCALS(r300); 921 DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count); 922 923 /* Reserve at least 256 dwords. 924 * 925 * Below we manage the CS space manually because there may be more 926 * indices than it can fit in CS. */ 927 r300_prepare_for_rendering(r300, 928 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, 929 NULL, 256, 0, 0, &end_cs_dwords); 930 931 while (count) { 932 free_dwords = r300->rws->get_cs_free_dwords(r300->rws); 933 934 short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2); 935 936 BEGIN_CS(6 + (short_count+1)/2); 937 OUT_CS_REG(R300_GA_COLOR_CONTROL, 938 r300_provoking_vertex_fixes(r300, r300render->prim)); 939 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index); 940 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2); 941 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) | 942 r300render->hwprim); 943 for (i = 0; i < short_count-1; i += 2) { 944 OUT_CS(indices[i+1] << 16 | indices[i]); 945 } 946 if (short_count % 2) { 947 OUT_CS(indices[short_count-1]); 948 } 949 END_CS; 950 951 /* OK now subtract the emitted indices and see if we need to emit 952 * another draw packet. */ 953 indices += short_count; 954 count -= short_count; 955 956 if (count) { 957 r300_prepare_for_rendering(r300, 958 PREP_EMIT_AOS_SWTCL | PREP_INDEXED, 959 NULL, 256, 0, 0, &end_cs_dwords); 960 } 961 } 962} 963 964static void r300_render_destroy(struct vbuf_render* render) 965{ 966 FREE(render); 967} 968 969static struct vbuf_render* r300_render_create(struct r300_context* r300) 970{ 971 struct r300_render* r300render = CALLOC_STRUCT(r300_render); 972 973 r300render->r300 = r300; 974 975 /* XXX find real numbers plz */ 976 r300render->base.max_vertex_buffer_bytes = 128 * 1024; 977 r300render->base.max_indices = 16 * 1024; 978 979 r300render->base.get_vertex_info = r300_render_get_vertex_info; 980 r300render->base.allocate_vertices = r300_render_allocate_vertices; 981 r300render->base.map_vertices = r300_render_map_vertices; 982 r300render->base.unmap_vertices = r300_render_unmap_vertices; 983 r300render->base.set_primitive = r300_render_set_primitive; 984 r300render->base.draw_elements = r300_render_draw_elements; 985 r300render->base.draw_arrays = r300_render_draw_arrays; 986 r300render->base.release_vertices = r300_render_release_vertices; 987 r300render->base.destroy = r300_render_destroy; 988 989 r300render->vbo = NULL; 990 r300render->vbo_size = 0; 991 r300render->vbo_offset = 0; 992 993 return &r300render->base; 994} 995 996struct draw_stage* r300_draw_stage(struct r300_context* r300) 997{ 998 struct vbuf_render* render; 999 struct draw_stage* stage; 1000 1001 render = r300_render_create(r300); 1002 1003 if (!render) { 1004 return NULL; 1005 } 1006 1007 stage = draw_vbuf_stage(r300->draw, render); 1008 1009 if (!stage) { 1010 render->destroy(render); 1011 return NULL; 1012 } 1013 1014 draw_set_render(r300->draw, render); 1015 1016 return stage; 1017} 1018 1019/**************************************************************************** 1020 * End of SW TCL functions * 1021 ***************************************************************************/ 1022 1023/* If we used a quad to draw a rectangle, the pixels on the main diagonal 1024 * would be computed and stored twice, which makes the clear/copy codepaths 1025 * somewhat inefficient. Instead we use a rectangular point sprite. */ 1026static void r300_blitter_draw_rectangle(struct blitter_context *blitter, 1027 unsigned x1, unsigned y1, 1028 unsigned x2, unsigned y2, 1029 float depth, 1030 enum blitter_attrib_type type, 1031 const float attrib[4]) 1032{ 1033 struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter)); 1034 unsigned last_sprite_coord_enable = r300->sprite_coord_enable; 1035 unsigned width = x2 - x1; 1036 unsigned height = y2 - y1; 1037 unsigned vertex_size = 1038 type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4; 1039 unsigned dwords = 13 + vertex_size + 1040 (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0); 1041 const float zeros[4] = {0, 0, 0, 0}; 1042 CB_LOCALS; 1043 1044 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) 1045 r300->sprite_coord_enable = 1; 1046 1047 r300_update_derived_state(r300); 1048 1049 /* Mark some states we don't care about as non-dirty. */ 1050 r300->clip_state.dirty = FALSE; 1051 r300->viewport_state.dirty = FALSE; 1052 1053 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0, NULL); 1054 1055 DBG(r300, DBG_DRAW, "r300: draw_rectangle\n"); 1056 1057 BEGIN_CS_AS_CB(r300, dwords); 1058 /* Set up GA. */ 1059 OUT_CB_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16)); 1060 1061 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) { 1062 /* Set up the GA to generate texcoords. */ 1063 OUT_CB_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE | 1064 (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT)); 1065 OUT_CB_REG_SEQ(R300_GA_POINT_S0, 4); 1066 OUT_CB_32F(attrib[0]); 1067 OUT_CB_32F(attrib[3]); 1068 OUT_CB_32F(attrib[2]); 1069 OUT_CB_32F(attrib[1]); 1070 } 1071 1072 /* Set up VAP controls. */ 1073 OUT_CB_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE); 1074 OUT_CB_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT); 1075 OUT_CB_REG(R300_VAP_VTX_SIZE, vertex_size); 1076 OUT_CB_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 1077 OUT_CB(1); 1078 OUT_CB(0); 1079 1080 /* Draw. */ 1081 OUT_CB_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size); 1082 OUT_CB(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) | 1083 R300_VAP_VF_CNTL__PRIM_POINTS); 1084 1085 OUT_CB_32F(x1 + width * 0.5f); 1086 OUT_CB_32F(y1 + height * 0.5f); 1087 OUT_CB_32F(depth); 1088 OUT_CB_32F(1); 1089 1090 if (vertex_size == 8) { 1091 if (!attrib) 1092 attrib = zeros; 1093 OUT_CB_TABLE(attrib, 4); 1094 } 1095 END_CB; 1096 1097 /* Restore the state. */ 1098 r300->clip_state.dirty = TRUE; 1099 r300->rs_state.dirty = TRUE; 1100 r300->viewport_state.dirty = TRUE; 1101 1102 r300->sprite_coord_enable = last_sprite_coord_enable; 1103} 1104 1105static void r300_resource_resolve(struct pipe_context* pipe, 1106 struct pipe_resource* dest, 1107 struct pipe_subresource subdest, 1108 struct pipe_resource* src, 1109 struct pipe_subresource subsrc) 1110{ 1111 struct r300_context* r300 = r300_context(pipe); 1112 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; 1113 struct pipe_surface* srcsurf = src->screen->get_tex_surface(src->screen, 1114 src, subsrc.face, subsrc.level, 0, 0); 1115 float color[] = {0, 0, 0, 0}; 1116 1117 DBG(r300, DBG_DRAW, "r300: Resolving resource...\n"); 1118 1119 /* Enable AA resolve. */ 1120 aa->dest = r300_surface( 1121 dest->screen->get_tex_surface(dest->screen, dest, subdest.face, 1122 subdest.level, 0, 0)); 1123 1124 aa->aaresolve_ctl = 1125 R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE | 1126 R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE; 1127 r300->aa_state.size = 12; 1128 r300->aa_state.dirty = TRUE; 1129 1130 /* Resolve the surface. */ 1131 r300->context.clear_render_target(pipe, 1132 srcsurf, color, 0, 0, src->width0, src->height0); 1133 1134 /* Disable AA resolve. */ 1135 aa->aaresolve_ctl = 0; 1136 r300->aa_state.size = 4; 1137 r300->aa_state.dirty = TRUE; 1138 1139 pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL); 1140 pipe_surface_reference((struct pipe_surface**)&aa->dest, NULL); 1141} 1142 1143void r300_init_render_functions(struct r300_context *r300) 1144{ 1145 /* Set generic functions. */ 1146 r300->context.draw_elements = r300_draw_elements; 1147 1148 /* Set draw functions based on presence of HW TCL. */ 1149 if (r300->screen->caps.has_tcl) { 1150 r300->context.draw_arrays = r300_draw_arrays; 1151 r300->context.draw_range_elements = r300_draw_range_elements; 1152 } else { 1153 r300->context.draw_arrays = r300_swtcl_draw_arrays; 1154 r300->context.draw_range_elements = r300_swtcl_draw_range_elements; 1155 } 1156 1157 r300->context.resource_resolve = r300_resource_resolve; 1158 r300->blitter->draw_rectangle = r300_blitter_draw_rectangle; 1159 1160 /* Plug in the two-sided stencil reference value fallback if needed. */ 1161 if (!r300->screen->caps.is_r500) 1162 r300_plug_in_stencil_ref_fallback(r300); 1163} 1164