r300_render.c revision 66d45567b4e2c6f2585789b68667e6c00b7567e1
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_context.h" 39#include "r300_screen_buffer.h" 40#include "r300_emit.h" 41#include "r300_reg.h" 42#include "r300_state_derived.h" 43 44#include <limits.h> 45 46#define IMMD_DWORDS 32 47 48static uint32_t r300_translate_primitive(unsigned prim) 49{ 50 switch (prim) { 51 case PIPE_PRIM_POINTS: 52 return R300_VAP_VF_CNTL__PRIM_POINTS; 53 case PIPE_PRIM_LINES: 54 return R300_VAP_VF_CNTL__PRIM_LINES; 55 case PIPE_PRIM_LINE_LOOP: 56 return R300_VAP_VF_CNTL__PRIM_LINE_LOOP; 57 case PIPE_PRIM_LINE_STRIP: 58 return R300_VAP_VF_CNTL__PRIM_LINE_STRIP; 59 case PIPE_PRIM_TRIANGLES: 60 return R300_VAP_VF_CNTL__PRIM_TRIANGLES; 61 case PIPE_PRIM_TRIANGLE_STRIP: 62 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP; 63 case PIPE_PRIM_TRIANGLE_FAN: 64 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN; 65 case PIPE_PRIM_QUADS: 66 return R300_VAP_VF_CNTL__PRIM_QUADS; 67 case PIPE_PRIM_QUAD_STRIP: 68 return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP; 69 case PIPE_PRIM_POLYGON: 70 return R300_VAP_VF_CNTL__PRIM_POLYGON; 71 default: 72 return 0; 73 } 74} 75 76static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, 77 unsigned mode) 78{ 79 struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state; 80 uint32_t color_control = rs->color_control; 81 82 /* By default (see r300_state.c:r300_create_rs_state) color_control is 83 * initialized to provoking the first vertex. 84 * 85 * Triangle fans must be reduced to the second vertex, not the first, in 86 * Gallium flatshade-first mode, as per the GL spec. 87 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt) 88 * 89 * Quads never provoke correctly in flatshade-first mode. The first 90 * vertex is never considered as provoking, so only the second, third, 91 * and fourth vertices can be selected, and both "third" and "last" modes 92 * select the fourth vertex. This is probably due to D3D lacking quads. 93 * 94 * Similarly, polygons reduce to the first, not the last, vertex, when in 95 * "last" mode, and all other modes start from the second vertex. 96 * 97 * ~ C. 98 */ 99 100 if (rs->rs.flatshade_first) { 101 switch (mode) { 102 case PIPE_PRIM_TRIANGLE_FAN: 103 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND; 104 break; 105 case PIPE_PRIM_QUADS: 106 case PIPE_PRIM_QUAD_STRIP: 107 case PIPE_PRIM_POLYGON: 108 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; 109 break; 110 default: 111 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST; 112 break; 113 } 114 } else { 115 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; 116 } 117 118 return color_control; 119} 120 121void r500_emit_index_bias(struct r300_context *r300, int index_bias) 122{ 123 CS_LOCALS(r300); 124 125 BEGIN_CS(2); 126 OUT_CS_REG(R500_VAP_INDEX_OFFSET, 127 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0)); 128 END_CS; 129} 130 131/* This function splits the index bias value into two parts: 132 * - buffer_offset: the value that can be safely added to buffer offsets 133 * in r300_emit_aos (it must yield a positive offset when added to 134 * a vertex buffer offset) 135 * - index_offset: the value that must be manually subtracted from indices 136 * in an index buffer to achieve negative offsets. */ 137static void r300_split_index_bias(struct r300_context *r300, int index_bias, 138 int *buffer_offset, int *index_offset) 139{ 140 struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer; 141 struct pipe_vertex_element *velem = r300->velems->velem; 142 unsigned i, size; 143 int max_neg_bias; 144 145 if (index_bias < 0) { 146 /* See how large index bias we may subtract. We must be careful 147 * here because negative buffer offsets are not allowed 148 * by the DRM API. */ 149 max_neg_bias = INT_MAX; 150 for (i = 0; i < r300->velems->count; i++) { 151 vb = &vbufs[velem[i].vertex_buffer_index]; 152 size = (vb->buffer_offset + velem[i].src_offset) / vb->stride; 153 max_neg_bias = MIN2(max_neg_bias, size); 154 } 155 156 /* Now set the minimum allowed value. */ 157 *buffer_offset = MAX2(-max_neg_bias, index_bias); 158 } else { 159 /* A positive index bias is OK. */ 160 *buffer_offset = index_bias; 161 } 162 163 *index_offset = index_bias - *buffer_offset; 164} 165 166enum r300_prepare_flags { 167 PREP_FIRST_DRAW = (1 << 0), /* call emit_dirty_state and friends? */ 168 PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */ 169 PREP_EMIT_AOS = (1 << 2), /* call emit_aos? */ 170 PREP_EMIT_AOS_SWTCL = (1 << 3), /* call emit_aos_swtcl? */ 171 PREP_INDEXED = (1 << 4) /* is this draw_elements? */ 172}; 173 174/** 175 * Check if the requested number of dwords is available in the CS and 176 * if not, flush. 177 * \param r300 The context. 178 * \param flags See r300_prepare_flags. 179 * \param cs_dwords The number of dwords to reserve in CS. 180 * \return TRUE if the CS was flushed 181 */ 182static boolean r300_reserve_cs_dwords(struct r300_context *r300, 183 enum r300_prepare_flags flags, 184 unsigned cs_dwords) 185{ 186 boolean flushed = FALSE; 187 boolean first_draw = flags & PREP_FIRST_DRAW; 188 boolean emit_aos = flags & PREP_EMIT_AOS; 189 boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL; 190 191 /* Add dirty state, index offset, and AOS. */ 192 if (first_draw) { 193 cs_dwords += r300_get_num_dirty_dwords(r300); 194 195 if (r300->screen->caps.index_bias_supported) 196 cs_dwords += 2; /* emit_index_offset */ 197 198 if (emit_aos) 199 cs_dwords += 55; /* emit_aos */ 200 201 if (emit_aos_swtcl) 202 cs_dwords += 7; /* emit_aos_swtcl */ 203 } 204 205 cs_dwords += r300_get_num_cs_end_dwords(r300); 206 207 /* Reserve requested CS space. */ 208 if (cs_dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) { 209 r300->context.flush(&r300->context, 0, NULL); 210 flushed = TRUE; 211 } 212 213 return flushed; 214} 215 216/** 217 * Validate buffers and emit dirty state. 218 * \param r300 The context. 219 * \param flags See r300_prepare_flags. 220 * \param index_buffer The index buffer to validate. The parameter may be NULL. 221 * \param aos_offset The offset passed to emit_aos. 222 * \param index_bias The index bias to emit. 223 * \return TRUE if rendering should be skipped 224 */ 225static boolean r300_emit_states(struct r300_context *r300, 226 enum r300_prepare_flags flags, 227 struct pipe_resource *index_buffer, 228 int aos_offset, 229 int index_bias) 230{ 231 boolean first_draw = flags & PREP_FIRST_DRAW; 232 boolean emit_aos = flags & PREP_EMIT_AOS; 233 boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL; 234 boolean indexed = flags & PREP_INDEXED; 235 236 /* Validate buffers and emit dirty state if needed. */ 237 if (first_draw) { 238 if (!r300_emit_buffer_validate(r300, flags & PREP_VALIDATE_VBOS, 239 index_buffer)) { 240 fprintf(stderr, "r300: CS space validation failed. " 241 "(not enough memory?) Skipping rendering.\n"); 242 return FALSE; 243 } 244 245 r300_emit_dirty_state(r300); 246 if (r300->screen->caps.index_bias_supported) { 247 if (r300->screen->caps.has_tcl) 248 r500_emit_index_bias(r300, index_bias); 249 else 250 r500_emit_index_bias(r300, 0); 251 } 252 253 if (emit_aos) 254 r300_emit_aos(r300, aos_offset, indexed); 255 256 if (emit_aos_swtcl) 257 r300_emit_aos_swtcl(r300, indexed); 258 } 259 260 return TRUE; 261} 262 263/** 264 * Check if the requested number of dwords is available in the CS and 265 * if not, flush. Then validate buffers and emit dirty state. 266 * \param r300 The context. 267 * \param flags See r300_prepare_flags. 268 * \param index_buffer The index buffer to validate. The parameter may be NULL. 269 * \param cs_dwords The number of dwords to reserve in CS. 270 * \param aos_offset The offset passed to emit_aos. 271 * \param index_bias The index bias to emit. 272 * \return TRUE if rendering should be skipped 273 */ 274static boolean r300_prepare_for_rendering(struct r300_context *r300, 275 enum r300_prepare_flags flags, 276 struct pipe_resource *index_buffer, 277 unsigned cs_dwords, 278 int aos_offset, 279 int index_bias) 280{ 281 if (r300_reserve_cs_dwords(r300, flags, cs_dwords)) 282 flags |= PREP_FIRST_DRAW; 283 284 return r300_emit_states(r300, flags, index_buffer, aos_offset, index_bias); 285} 286 287static boolean immd_is_good_idea(struct r300_context *r300, 288 unsigned count) 289{ 290 struct pipe_vertex_element* velem; 291 struct pipe_vertex_buffer* vbuf; 292 boolean checked[PIPE_MAX_ATTRIBS] = {0}; 293 unsigned vertex_element_count = r300->velems->count; 294 unsigned i, vbi; 295 296 if (DBG_ON(r300, DBG_NO_IMMD)) { 297 return FALSE; 298 } 299 300 if (r300->draw) { 301 return FALSE; 302 } 303 304 if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) { 305 return FALSE; 306 } 307 308 /* We shouldn't map buffers referenced by CS, busy buffers, 309 * and ones placed in VRAM. */ 310 for (i = 0; i < vertex_element_count; i++) { 311 velem = &r300->velems->velem[i]; 312 vbi = velem->vertex_buffer_index; 313 314 if (!checked[vbi]) { 315 vbuf = &r300->vertex_buffer[vbi]; 316 317 if (!(r300_buffer(vbuf->buffer)->domain & R300_DOMAIN_GTT)) { 318 return FALSE; 319 } 320 321 if (r300_buffer_is_referenced(&r300->context, 322 vbuf->buffer, 323 R300_REF_CS | R300_REF_HW)) { 324 /* It's a very bad idea to map it... */ 325 return FALSE; 326 } 327 checked[vbi] = TRUE; 328 } 329 } 330 return TRUE; 331} 332 333/***************************************************************************** 334 * The HWTCL draw functions. * 335 ****************************************************************************/ 336 337static void r300_emit_draw_arrays_immediate(struct r300_context *r300, 338 unsigned mode, 339 unsigned start, 340 unsigned count) 341{ 342 struct pipe_vertex_element* velem; 343 struct pipe_vertex_buffer* vbuf; 344 unsigned vertex_element_count = r300->velems->count; 345 unsigned i, v, vbi; 346 347 /* Size of the vertex, in dwords. */ 348 unsigned vertex_size = r300->velems->vertex_size_dwords; 349 350 /* The number of dwords for this draw operation. */ 351 unsigned dwords = 9 + count * vertex_size; 352 353 /* Size of the vertex element, in dwords. */ 354 unsigned size[PIPE_MAX_ATTRIBS]; 355 356 /* Stride to the same attrib in the next vertex in the vertex buffer, 357 * in dwords. */ 358 unsigned stride[PIPE_MAX_ATTRIBS]; 359 360 /* Mapped vertex buffers. */ 361 uint32_t* map[PIPE_MAX_ATTRIBS]; 362 uint32_t* mapelem[PIPE_MAX_ATTRIBS]; 363 struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {0}; 364 365 CS_LOCALS(r300); 366 367 if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0)) 368 return; 369 370 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */ 371 for (i = 0; i < vertex_element_count; i++) { 372 velem = &r300->velems->velem[i]; 373 size[i] = r300->velems->hw_format_size[i] / 4; 374 vbi = velem->vertex_buffer_index; 375 vbuf = &r300->vertex_buffer[vbi]; 376 stride[i] = vbuf->stride / 4; 377 378 /* Map the buffer. */ 379 if (!transfer[vbi]) { 380 map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context, 381 vbuf->buffer, 382 PIPE_TRANSFER_READ, 383 &transfer[vbi]); 384 map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * start; 385 } 386 mapelem[i] = map[vbi] + (velem->src_offset / 4); 387 } 388 389 BEGIN_CS(dwords); 390 OUT_CS_REG(R300_GA_COLOR_CONTROL, 391 r300_provoking_vertex_fixes(r300, mode)); 392 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); 393 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 394 OUT_CS(count - 1); 395 OUT_CS(0); 396 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size); 397 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) | 398 r300_translate_primitive(mode)); 399 400 /* Emit vertices. */ 401 for (v = 0; v < count; v++) { 402 for (i = 0; i < vertex_element_count; i++) { 403 OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]); 404 } 405 } 406 END_CS; 407 408 /* Unmap buffers. */ 409 for (i = 0; i < vertex_element_count; i++) { 410 vbi = r300->velems->velem[i].vertex_buffer_index; 411 412 if (transfer[vbi]) { 413 vbuf = &r300->vertex_buffer[vbi]; 414 pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]); 415 transfer[vbi] = NULL; 416 } 417 } 418} 419 420static void r300_emit_draw_arrays(struct r300_context *r300, 421 unsigned mode, 422 unsigned count) 423{ 424 boolean alt_num_verts = count > 65535; 425 CS_LOCALS(r300); 426 427 if (count >= (1 << 24)) { 428 fprintf(stderr, "r300: Got a huge number of vertices: %i, " 429 "refusing to render.\n", count); 430 return; 431 } 432 433 BEGIN_CS(7 + (alt_num_verts ? 2 : 0)); 434 if (alt_num_verts) { 435 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); 436 } 437 OUT_CS_REG(R300_GA_COLOR_CONTROL, 438 r300_provoking_vertex_fixes(r300, mode)); 439 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 440 OUT_CS(count - 1); 441 OUT_CS(0); 442 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); 443 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | 444 r300_translate_primitive(mode) | 445 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 446 END_CS; 447} 448 449static void r300_emit_draw_elements(struct r300_context *r300, 450 struct pipe_resource* indexBuffer, 451 unsigned indexSize, 452 unsigned minIndex, 453 unsigned maxIndex, 454 unsigned mode, 455 unsigned start, 456 unsigned count) 457{ 458 uint32_t count_dwords; 459 uint32_t offset_dwords = indexSize * start / sizeof(uint32_t); 460 boolean alt_num_verts = count > 65535; 461 CS_LOCALS(r300); 462 463 if (count >= (1 << 24)) { 464 fprintf(stderr, "r300: Got a huge number of vertices: %i, " 465 "refusing to render.\n", count); 466 return; 467 } 468 469 maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index); 470 471 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n", 472 count, minIndex, maxIndex); 473 474 BEGIN_CS(13 + (alt_num_verts ? 2 : 0)); 475 if (alt_num_verts) { 476 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); 477 } 478 OUT_CS_REG(R300_GA_COLOR_CONTROL, 479 r300_provoking_vertex_fixes(r300, mode)); 480 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 481 OUT_CS(maxIndex); 482 OUT_CS(minIndex); 483 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0); 484 if (indexSize == 4) { 485 count_dwords = count; 486 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | 487 R300_VAP_VF_CNTL__INDEX_SIZE_32bit | 488 r300_translate_primitive(mode) | 489 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 490 } else { 491 count_dwords = (count + 1) / 2; 492 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | 493 r300_translate_primitive(mode) | 494 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 495 } 496 497 /* INDX_BUFFER is a truly special packet3. 498 * Unlike most other packet3, where the offset is after the count, 499 * the order is reversed, so the relocation ends up carrying the 500 * size of the indexbuf instead of the offset. 501 */ 502 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2); 503 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) | 504 (0 << R300_INDX_BUFFER_SKIP_SHIFT)); 505 OUT_CS(offset_dwords << 2); 506 OUT_CS_BUF_RELOC(indexBuffer, count_dwords, 507 r300_buffer(indexBuffer)->domain, 0); 508 509 END_CS; 510} 511 512/* This is the fast-path drawing & emission for HW TCL. */ 513static void r300_draw_range_elements(struct pipe_context* pipe, 514 struct pipe_resource* indexBuffer, 515 unsigned indexSize, 516 int indexBias, 517 unsigned minIndex, 518 unsigned maxIndex, 519 unsigned mode, 520 unsigned start, 521 unsigned count) 522{ 523 struct r300_context* r300 = r300_context(pipe); 524 struct pipe_resource* orgIndexBuffer = indexBuffer; 525 boolean alt_num_verts = r300->screen->caps.is_r500 && 526 count > 65536 && 527 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); 528 unsigned short_count; 529 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */ 530 unsigned new_offset; 531 532 if (indexBias && !r300->screen->caps.index_bias_supported) { 533 r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset); 534 } 535 536 r300_translate_index_buffer(r300, &indexBuffer, &indexSize, index_offset, 537 &start, count); 538 539 r300_update_derived_state(r300); 540 r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count, &new_offset); 541 542 start = new_offset; 543 544 /* 15 dwords for emit_draw_elements. Give up if the function fails. */ 545 if (!r300_prepare_for_rendering(r300, 546 PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | 547 PREP_INDEXED, indexBuffer, 15, buffer_offset, indexBias)) 548 goto done; 549 550 if (alt_num_verts || count <= 65535) { 551 r300_emit_draw_elements(r300, indexBuffer, indexSize, 552 minIndex, maxIndex, mode, start, count); 553 } else { 554 do { 555 short_count = MIN2(count, 65534); 556 r300_emit_draw_elements(r300, indexBuffer, indexSize, 557 minIndex, maxIndex, 558 mode, start, short_count); 559 560 start += short_count; 561 count -= short_count; 562 563 /* 15 dwords for emit_draw_elements */ 564 if (count) { 565 if (!r300_prepare_for_rendering(r300, 566 PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, 567 indexBuffer, 15, buffer_offset, indexBias)) 568 goto done; 569 } 570 } while (count); 571 } 572 573done: 574 if (indexBuffer != orgIndexBuffer) { 575 pipe_resource_reference( &indexBuffer, NULL ); 576 } 577} 578 579static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, 580 unsigned start, unsigned count) 581{ 582 struct r300_context* r300 = r300_context(pipe); 583 boolean alt_num_verts = r300->screen->caps.is_r500 && 584 count > 65536 && 585 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); 586 unsigned short_count; 587 588 r300_update_derived_state(r300); 589 590 if (immd_is_good_idea(r300, count)) { 591 r300_emit_draw_arrays_immediate(r300, mode, start, count); 592 } else { 593 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */ 594 if (!r300_prepare_for_rendering(r300, 595 PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS, 596 NULL, 9, start, 0)) 597 return; 598 599 if (alt_num_verts || count <= 65535) { 600 r300_emit_draw_arrays(r300, mode, count); 601 } else { 602 do { 603 short_count = MIN2(count, 65535); 604 r300_emit_draw_arrays(r300, mode, short_count); 605 606 start += short_count; 607 count -= short_count; 608 609 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */ 610 if (count) { 611 if (!r300_prepare_for_rendering(r300, 612 PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9, 613 start, 0)) 614 return; 615 } 616 } while (count); 617 } 618 } 619} 620 621static void r300_draw_vbo(struct pipe_context* pipe, 622 const struct pipe_draw_info *info) 623{ 624 struct r300_context* r300 = r300_context(pipe); 625 unsigned count = info->count; 626 boolean translate = FALSE; 627 boolean indexed = info->indexed && r300->index_buffer.buffer; 628 unsigned start_indexed = 0; 629 630 if (r300->skip_rendering) { 631 return; 632 } 633 634 if (!u_trim_pipe_prim(info->mode, &count)) { 635 return; 636 } 637 638 /* Index buffer range checking. */ 639 if (indexed) { 640 assert(r300->index_buffer.offset % r300->index_buffer.index_size == 0); 641 642 /* Compute start for draw_elements, taking the offset into account. */ 643 start_indexed = 644 info->start + 645 (r300->index_buffer.offset / r300->index_buffer.index_size); 646 647 if ((start_indexed + count) * r300->index_buffer.index_size > 648 r300->index_buffer.buffer->width0) { 649 fprintf(stderr, "r300: Invalid index buffer range. Skipping rendering.\n"); 650 return; 651 } 652 } 653 654 /* Set up fallback for incompatible vertex layout if needed. */ 655 if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) { 656 r300_begin_vertex_translate(r300); 657 translate = TRUE; 658 } 659 660 if (indexed) { 661 r300_draw_range_elements(pipe, 662 r300->index_buffer.buffer, 663 r300->index_buffer.index_size, 664 info->index_bias, 665 info->min_index, 666 info->max_index, 667 info->mode, 668 start_indexed, 669 count); 670 } else { 671 r300_draw_arrays(pipe, 672 info->mode, 673 info->start, 674 count); 675 } 676 677 if (translate) { 678 r300_end_vertex_translate(r300); 679 } 680} 681 682/**************************************************************************** 683 * The rest of this file is for SW TCL rendering only. Please be polite and * 684 * keep these functions separated so that they are easier to locate. ~C. * 685 ***************************************************************************/ 686 687/* SW TCL elements, using Draw. */ 688static void r300_swtcl_draw_vbo(struct pipe_context* pipe, 689 const struct pipe_draw_info *info) 690{ 691 struct r300_context* r300 = r300_context(pipe); 692 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS]; 693 struct pipe_transfer *ib_transfer = NULL; 694 unsigned count = info->count; 695 int i; 696 void *indices = NULL; 697 boolean indexed = info->indexed && r300->index_buffer.buffer; 698 699 if (r300->skip_rendering) { 700 return; 701 } 702 703 if (!u_trim_pipe_prim(info->mode, &count)) { 704 return; 705 } 706 707 r300_update_derived_state(r300); 708 709 r300_reserve_cs_dwords(r300, 710 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | 711 (indexed ? PREP_INDEXED : 0), 712 indexed ? 256 : 6); 713 714 for (i = 0; i < r300->vertex_buffer_count; i++) { 715 if (r300->vertex_buffer[i].buffer) { 716 void *buf = pipe_buffer_map(pipe, 717 r300->vertex_buffer[i].buffer, 718 PIPE_TRANSFER_READ, 719 &vb_transfer[i]); 720 draw_set_mapped_vertex_buffer(r300->draw, i, buf); 721 } 722 } 723 724 if (indexed) { 725 indices = pipe_buffer_map(pipe, r300->index_buffer.buffer, 726 PIPE_TRANSFER_READ, &ib_transfer); 727 } 728 729 draw_set_mapped_index_buffer(r300->draw, indices); 730 731 r300->draw_vbo_locked = TRUE; 732 r300->draw_first_emitted = FALSE; 733 draw_vbo(r300->draw, info); 734 draw_flush(r300->draw); 735 r300->draw_vbo_locked = FALSE; 736 737 for (i = 0; i < r300->vertex_buffer_count; i++) { 738 if (r300->vertex_buffer[i].buffer) { 739 pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, 740 vb_transfer[i]); 741 draw_set_mapped_vertex_buffer(r300->draw, i, NULL); 742 } 743 } 744 745 if (indexed) { 746 pipe_buffer_unmap(pipe, r300->index_buffer.buffer, ib_transfer); 747 draw_set_mapped_index_buffer(r300->draw, NULL); 748 } 749} 750 751/* Object for rendering using Draw. */ 752struct r300_render { 753 /* Parent class */ 754 struct vbuf_render base; 755 756 /* Pipe context */ 757 struct r300_context* r300; 758 759 /* Vertex information */ 760 size_t vertex_size; 761 unsigned prim; 762 unsigned hwprim; 763 764 /* VBO */ 765 size_t vbo_max_used; 766 void * vbo_ptr; 767 768 struct pipe_transfer *vbo_transfer; 769}; 770 771static INLINE struct r300_render* 772r300_render(struct vbuf_render* render) 773{ 774 return (struct r300_render*)render; 775} 776 777static const struct vertex_info* 778r300_render_get_vertex_info(struct vbuf_render* render) 779{ 780 struct r300_render* r300render = r300_render(render); 781 struct r300_context* r300 = r300render->r300; 782 783 return &r300->vertex_info; 784} 785 786static boolean r300_render_allocate_vertices(struct vbuf_render* render, 787 ushort vertex_size, 788 ushort count) 789{ 790 struct r300_render* r300render = r300_render(render); 791 struct r300_context* r300 = r300render->r300; 792 struct pipe_screen* screen = r300->context.screen; 793 size_t size = (size_t)vertex_size * (size_t)count; 794 795 DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size); 796 797 if (size + r300->draw_vbo_offset > r300->draw_vbo_size) 798 { 799 pipe_resource_reference(&r300->vbo, NULL); 800 r300->vbo = pipe_buffer_create(screen, 801 PIPE_BIND_VERTEX_BUFFER, 802 R300_MAX_DRAW_VBO_SIZE); 803 r300->draw_vbo_offset = 0; 804 r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE; 805 } 806 807 r300render->vertex_size = vertex_size; 808 809 return (r300->vbo) ? TRUE : FALSE; 810} 811 812static void* r300_render_map_vertices(struct vbuf_render* render) 813{ 814 struct r300_render* r300render = r300_render(render); 815 struct r300_context* r300 = r300render->r300; 816 817 assert(!r300render->vbo_transfer); 818 819 DBG(r300, DBG_DRAW, "r300: render_map_vertices\n"); 820 821 r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context, 822 r300->vbo, 823 PIPE_TRANSFER_WRITE, 824 &r300render->vbo_transfer); 825 826 assert(r300render->vbo_ptr); 827 828 return ((uint8_t*)r300render->vbo_ptr + r300->draw_vbo_offset); 829} 830 831static void r300_render_unmap_vertices(struct vbuf_render* render, 832 ushort min, 833 ushort max) 834{ 835 struct r300_render* r300render = r300_render(render); 836 struct pipe_context* context = &r300render->r300->context; 837 struct r300_context* r300 = r300render->r300; 838 839 assert(r300render->vbo_transfer); 840 841 DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n"); 842 843 r300render->vbo_max_used = MAX2(r300render->vbo_max_used, 844 r300render->vertex_size * (max + 1)); 845 pipe_buffer_unmap(context, r300->vbo, r300render->vbo_transfer); 846 847 r300render->vbo_transfer = NULL; 848} 849 850static void r300_render_release_vertices(struct vbuf_render* render) 851{ 852 struct r300_render* r300render = r300_render(render); 853 struct r300_context* r300 = r300render->r300; 854 855 DBG(r300, DBG_DRAW, "r300: render_release_vertices\n"); 856 857 r300->draw_vbo_offset += r300render->vbo_max_used; 858 r300render->vbo_max_used = 0; 859} 860 861static boolean r300_render_set_primitive(struct vbuf_render* render, 862 unsigned prim) 863{ 864 struct r300_render* r300render = r300_render(render); 865 866 r300render->prim = prim; 867 r300render->hwprim = r300_translate_primitive(prim); 868 869 return TRUE; 870} 871 872static void r300_render_draw_arrays(struct vbuf_render* render, 873 unsigned start, 874 unsigned count) 875{ 876 struct r300_render* r300render = r300_render(render); 877 struct r300_context* r300 = r300render->r300; 878 uint8_t* ptr; 879 unsigned i; 880 unsigned dwords = 6; 881 882 CS_LOCALS(r300); 883 (void) i; (void) ptr; 884 885 DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count); 886 887 if (r300->draw_first_emitted) { 888 if (!r300_prepare_for_rendering(r300, 889 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL, 890 NULL, 6, 0, 0)) 891 return; 892 } else { 893 if (!r300_emit_states(r300, 894 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL, 895 NULL, 0, 0)) 896 return; 897 } 898 899 /* Uncomment to dump all VBOs rendered through this interface. 900 * Slow and noisy! 901 ptr = pipe_buffer_map(&r300render->r300->context, 902 r300render->vbo, PIPE_TRANSFER_READ, 903 &r300render->vbo_transfer); 904 905 for (i = 0; i < count; i++) { 906 printf("r300: Vertex %d\n", i); 907 draw_dump_emitted_vertex(&r300->vertex_info, ptr); 908 ptr += r300->vertex_info.size * 4; 909 printf("\n"); 910 } 911 912 pipe_buffer_unmap(&r300render->r300->context, r300render->vbo, 913 r300render->vbo_transfer); 914 */ 915 916 BEGIN_CS(dwords); 917 OUT_CS_REG(R300_GA_COLOR_CONTROL, 918 r300_provoking_vertex_fixes(r300, r300render->prim)); 919 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1); 920 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); 921 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | 922 r300render->hwprim); 923 END_CS; 924 925 r300->draw_first_emitted = TRUE; 926} 927 928static void r300_render_draw_elements(struct vbuf_render* render, 929 const ushort* indices, 930 uint count) 931{ 932 struct r300_render* r300render = r300_render(render); 933 struct r300_context* r300 = r300render->r300; 934 int i; 935 unsigned end_cs_dwords; 936 unsigned max_index = (r300->draw_vbo_size - r300->draw_vbo_offset) / 937 (r300render->r300->vertex_info.size * 4) - 1; 938 unsigned short_count; 939 unsigned free_dwords; 940 941 CS_LOCALS(r300); 942 DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count); 943 944 if (r300->draw_first_emitted) { 945 if (!r300_prepare_for_rendering(r300, 946 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, 947 NULL, 256, 0, 0)) 948 return; 949 } else { 950 if (!r300_emit_states(r300, 951 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, 952 NULL, 0, 0)) 953 return; 954 } 955 956 /* Below we manage the CS space manually because there may be more 957 * indices than it can fit in CS. */ 958 959 end_cs_dwords = r300_get_num_cs_end_dwords(r300); 960 961 while (count) { 962 free_dwords = R300_MAX_CMDBUF_DWORDS - r300->cs->cdw; 963 964 short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2); 965 966 BEGIN_CS(6 + (short_count+1)/2); 967 OUT_CS_REG(R300_GA_COLOR_CONTROL, 968 r300_provoking_vertex_fixes(r300, r300render->prim)); 969 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index); 970 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2); 971 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) | 972 r300render->hwprim); 973 for (i = 0; i < short_count-1; i += 2) { 974 OUT_CS(indices[i+1] << 16 | indices[i]); 975 } 976 if (short_count % 2) { 977 OUT_CS(indices[short_count-1]); 978 } 979 END_CS; 980 981 /* OK now subtract the emitted indices and see if we need to emit 982 * another draw packet. */ 983 indices += short_count; 984 count -= short_count; 985 986 if (count) { 987 if (!r300_prepare_for_rendering(r300, 988 PREP_EMIT_AOS_SWTCL | PREP_INDEXED, 989 NULL, 256, 0, 0)) 990 return; 991 992 end_cs_dwords = r300_get_num_cs_end_dwords(r300); 993 } 994 } 995 996 r300->draw_first_emitted = TRUE; 997} 998 999static void r300_render_destroy(struct vbuf_render* render) 1000{ 1001 FREE(render); 1002} 1003 1004static struct vbuf_render* r300_render_create(struct r300_context* r300) 1005{ 1006 struct r300_render* r300render = CALLOC_STRUCT(r300_render); 1007 1008 r300render->r300 = r300; 1009 1010 /* XXX find real numbers plz */ 1011 r300render->base.max_vertex_buffer_bytes = 128 * 1024; 1012 r300render->base.max_indices = 16 * 1024; 1013 1014 r300render->base.get_vertex_info = r300_render_get_vertex_info; 1015 r300render->base.allocate_vertices = r300_render_allocate_vertices; 1016 r300render->base.map_vertices = r300_render_map_vertices; 1017 r300render->base.unmap_vertices = r300_render_unmap_vertices; 1018 r300render->base.set_primitive = r300_render_set_primitive; 1019 r300render->base.draw_elements = r300_render_draw_elements; 1020 r300render->base.draw_arrays = r300_render_draw_arrays; 1021 r300render->base.release_vertices = r300_render_release_vertices; 1022 r300render->base.destroy = r300_render_destroy; 1023 1024 return &r300render->base; 1025} 1026 1027struct draw_stage* r300_draw_stage(struct r300_context* r300) 1028{ 1029 struct vbuf_render* render; 1030 struct draw_stage* stage; 1031 1032 render = r300_render_create(r300); 1033 1034 if (!render) { 1035 return NULL; 1036 } 1037 1038 stage = draw_vbuf_stage(r300->draw, render); 1039 1040 if (!stage) { 1041 render->destroy(render); 1042 return NULL; 1043 } 1044 1045 draw_set_render(r300->draw, render); 1046 1047 return stage; 1048} 1049 1050void r300_draw_flush_vbuf(struct r300_context *r300) 1051{ 1052 pipe_resource_reference(&r300->vbo, NULL); 1053 r300->draw_vbo_size = 0; 1054} 1055 1056/**************************************************************************** 1057 * End of SW TCL functions * 1058 ***************************************************************************/ 1059 1060/* This functions is used to draw a rectangle for the blitter module. 1061 * 1062 * If we rendered a quad, the pixels on the main diagonal 1063 * would be computed and stored twice, which makes the clear/copy codepaths 1064 * somewhat inefficient. Instead we use a rectangular point sprite. */ 1065static void r300_blitter_draw_rectangle(struct blitter_context *blitter, 1066 unsigned x1, unsigned y1, 1067 unsigned x2, unsigned y2, 1068 float depth, 1069 enum blitter_attrib_type type, 1070 const float attrib[4]) 1071{ 1072 struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter)); 1073 unsigned last_sprite_coord_enable = r300->sprite_coord_enable; 1074 unsigned width = x2 - x1; 1075 unsigned height = y2 - y1; 1076 unsigned vertex_size = 1077 type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4; 1078 unsigned dwords = 13 + vertex_size + 1079 (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0); 1080 const float zeros[4] = {0, 0, 0, 0}; 1081 CS_LOCALS(r300); 1082 1083 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) 1084 r300->sprite_coord_enable = 1; 1085 1086 r300_update_derived_state(r300); 1087 1088 /* Mark some states we don't care about as non-dirty. */ 1089 r300->clip_state.dirty = FALSE; 1090 r300->viewport_state.dirty = FALSE; 1091 1092 if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0)) 1093 goto done; 1094 1095 DBG(r300, DBG_DRAW, "r300: draw_rectangle\n"); 1096 1097 BEGIN_CS(dwords); 1098 /* Set up GA. */ 1099 OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16)); 1100 1101 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) { 1102 /* Set up the GA to generate texcoords. */ 1103 OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE | 1104 (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT)); 1105 OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4); 1106 OUT_CS_32F(attrib[0]); 1107 OUT_CS_32F(attrib[3]); 1108 OUT_CS_32F(attrib[2]); 1109 OUT_CS_32F(attrib[1]); 1110 } 1111 1112 /* Set up VAP controls. */ 1113 OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE); 1114 OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT); 1115 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); 1116 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 1117 OUT_CS(1); 1118 OUT_CS(0); 1119 1120 /* Draw. */ 1121 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size); 1122 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) | 1123 R300_VAP_VF_CNTL__PRIM_POINTS); 1124 1125 OUT_CS_32F(x1 + width * 0.5f); 1126 OUT_CS_32F(y1 + height * 0.5f); 1127 OUT_CS_32F(depth); 1128 OUT_CS_32F(1); 1129 1130 if (vertex_size == 8) { 1131 if (!attrib) 1132 attrib = zeros; 1133 OUT_CS_TABLE(attrib, 4); 1134 } 1135 END_CS; 1136 1137done: 1138 /* Restore the state. */ 1139 r300_mark_atom_dirty(r300, &r300->clip_state); 1140 r300_mark_atom_dirty(r300, &r300->rs_state); 1141 r300_mark_atom_dirty(r300, &r300->viewport_state); 1142 1143 r300->sprite_coord_enable = last_sprite_coord_enable; 1144} 1145 1146static void r300_resource_resolve(struct pipe_context* pipe, 1147 struct pipe_resource* dest, 1148 unsigned dst_layer, 1149 struct pipe_resource* src, 1150 unsigned src_layer) 1151{ 1152 struct r300_context* r300 = r300_context(pipe); 1153 struct pipe_surface* srcsurf, surf_tmpl; 1154 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state; 1155 float color[] = {0, 0, 0, 0}; 1156 1157 memset(&surf_tmpl, 0, sizeof(surf_tmpl)); 1158 surf_tmpl.format = src->format; 1159 surf_tmpl.usage = 0; /* not really a surface hence no bind flags */ 1160 surf_tmpl.u.tex.level = 0; /* msaa resources cannot have mipmaps */ 1161 surf_tmpl.u.tex.first_layer = src_layer; 1162 surf_tmpl.u.tex.last_layer = src_layer; 1163 srcsurf = pipe->create_surface(pipe, src, &surf_tmpl); 1164 surf_tmpl.format = dest->format; 1165 surf_tmpl.u.tex.first_layer = dst_layer; 1166 surf_tmpl.u.tex.last_layer = dst_layer; 1167 1168 DBG(r300, DBG_DRAW, "r300: Resolving resource...\n"); 1169 1170 /* Enable AA resolve. */ 1171 aa->dest = r300_surface(pipe->create_surface(pipe, dest, &surf_tmpl)); 1172 1173 aa->aaresolve_ctl = 1174 R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE | 1175 R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE; 1176 r300->aa_state.size = 12; 1177 r300_mark_atom_dirty(r300, &r300->aa_state); 1178 1179 /* Resolve the surface. */ 1180 r300->context.clear_render_target(pipe, 1181 srcsurf, color, 0, 0, src->width0, src->height0); 1182 1183 /* Disable AA resolve. */ 1184 aa->aaresolve_ctl = 0; 1185 r300->aa_state.size = 4; 1186 r300_mark_atom_dirty(r300, &r300->aa_state); 1187 1188 pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL); 1189 pipe_surface_reference((struct pipe_surface**)&aa->dest, NULL); 1190} 1191 1192void r300_init_render_functions(struct r300_context *r300) 1193{ 1194 /* Set draw functions based on presence of HW TCL. */ 1195 if (r300->screen->caps.has_tcl) { 1196 r300->context.draw_vbo = r300_draw_vbo; 1197 } else { 1198 r300->context.draw_vbo = r300_swtcl_draw_vbo; 1199 } 1200 1201 r300->context.resource_resolve = r300_resource_resolve; 1202 r300->blitter->draw_rectangle = r300_blitter_draw_rectangle; 1203 1204 /* Plug in the two-sided stencil reference value fallback if needed. */ 1205 if (!r300->screen->caps.is_r500) 1206 r300_plug_in_stencil_ref_fallback(r300); 1207} 1208