r300_render.c revision cb17f5ee752d07d82e9b079c6bda9d89e51c7108
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 46static uint32_t r300_translate_primitive(unsigned prim) 47{ 48 switch (prim) { 49 case PIPE_PRIM_POINTS: 50 return R300_VAP_VF_CNTL__PRIM_POINTS; 51 case PIPE_PRIM_LINES: 52 return R300_VAP_VF_CNTL__PRIM_LINES; 53 case PIPE_PRIM_LINE_LOOP: 54 return R300_VAP_VF_CNTL__PRIM_LINE_LOOP; 55 case PIPE_PRIM_LINE_STRIP: 56 return R300_VAP_VF_CNTL__PRIM_LINE_STRIP; 57 case PIPE_PRIM_TRIANGLES: 58 return R300_VAP_VF_CNTL__PRIM_TRIANGLES; 59 case PIPE_PRIM_TRIANGLE_STRIP: 60 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP; 61 case PIPE_PRIM_TRIANGLE_FAN: 62 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN; 63 case PIPE_PRIM_QUADS: 64 return R300_VAP_VF_CNTL__PRIM_QUADS; 65 case PIPE_PRIM_QUAD_STRIP: 66 return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP; 67 case PIPE_PRIM_POLYGON: 68 return R300_VAP_VF_CNTL__PRIM_POLYGON; 69 default: 70 return 0; 71 } 72} 73 74static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, 75 unsigned mode) 76{ 77 struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state; 78 uint32_t color_control = rs->color_control; 79 80 /* By default (see r300_state.c:r300_create_rs_state) color_control is 81 * initialized to provoking the first vertex. 82 * 83 * Triangle fans must be reduced to the second vertex, not the first, in 84 * Gallium flatshade-first mode, as per the GL spec. 85 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt) 86 * 87 * Quads never provoke correctly in flatshade-first mode. The first 88 * vertex is never considered as provoking, so only the second, third, 89 * and fourth vertices can be selected, and both "third" and "last" modes 90 * select the fourth vertex. This is probably due to D3D lacking quads. 91 * 92 * Similarly, polygons reduce to the first, not the last, vertex, when in 93 * "last" mode, and all other modes start from the second vertex. 94 * 95 * ~ C. 96 */ 97 98 if (rs->rs.flatshade_first) { 99 switch (mode) { 100 case PIPE_PRIM_TRIANGLE_FAN: 101 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND; 102 break; 103 case PIPE_PRIM_QUADS: 104 case PIPE_PRIM_QUAD_STRIP: 105 case PIPE_PRIM_POLYGON: 106 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; 107 break; 108 default: 109 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST; 110 break; 111 } 112 } else { 113 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; 114 } 115 116 return color_control; 117} 118 119static boolean index_bias_supported(struct r300_context *r300) 120{ 121 return r300->screen->caps.is_r500 && 122 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); 123} 124 125static void r500_emit_index_bias(struct r300_context *r300, int index_bias) 126{ 127 CS_LOCALS(r300); 128 129 BEGIN_CS(2); 130 OUT_CS_REG(R500_VAP_INDEX_OFFSET, 131 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0)); 132 END_CS; 133} 134 135/* This function splits the index bias value into two parts: 136 * - buffer_offset: the value that can be safely added to buffer offsets 137 * in r300_emit_aos (it must yield a positive offset when added to 138 * a vertex buffer offset) 139 * - index_offset: the value that must be manually subtracted from indices 140 * in an index buffer to achieve negative offsets. */ 141static void r300_split_index_bias(struct r300_context *r300, int index_bias, 142 int *buffer_offset, int *index_offset) 143{ 144 struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer; 145 struct pipe_vertex_element *velem = r300->velems->velem; 146 unsigned i, size; 147 int max_neg_bias; 148 149 if (index_bias < 0) { 150 /* See how large index bias we may subtract. We must be careful 151 * here because negative buffer offsets are not allowed 152 * by the DRM API. */ 153 max_neg_bias = INT_MAX; 154 for (i = 0; i < r300->velems->count; i++) { 155 vb = &vbufs[velem[i].vertex_buffer_index]; 156 size = (vb->buffer_offset + velem[i].src_offset) / vb->stride; 157 max_neg_bias = MIN2(max_neg_bias, size); 158 } 159 160 /* Now set the minimum allowed value. */ 161 *buffer_offset = MAX2(-max_neg_bias, index_bias); 162 } else { 163 /* A positive index bias is OK. */ 164 *buffer_offset = index_bias; 165 } 166 167 *index_offset = index_bias - *buffer_offset; 168} 169 170enum r300_prepare_flags { 171 PREP_FIRST_DRAW = (1 << 0), /* call emit_dirty_state and friends? */ 172 PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */ 173 PREP_EMIT_AOS = (1 << 2), /* call emit_aos? */ 174 PREP_EMIT_AOS_SWTCL = (1 << 3), /* call emit_aos_swtcl? */ 175 PREP_INDEXED = (1 << 4) /* is this draw_elements? */ 176}; 177 178/** 179 * Check if the requested number of dwords is available in the CS and 180 * if not, flush. Then validate buffers and emit dirty state. 181 * \param r300 The context. 182 * \param flags See r300_prepare_flags. 183 * \param index_buffer The index buffer to validate. The parameter may be NULL. 184 * \param cs_dwords The number of dwords to reserve in CS. 185 * \param aos_offset The offset passed to emit_aos. 186 * \param index_bias The index bias to emit. 187 * \param end_cs_dwords The number of free dwords which must be available 188 * at the end of CS after drawing in case the CS space 189 * management is performed by a draw_* function manually. 190 * The parameter may be NULL. 191 */ 192static void r300_prepare_for_rendering(struct r300_context *r300, 193 enum r300_prepare_flags flags, 194 struct pipe_resource *index_buffer, 195 unsigned cs_dwords, 196 int aos_offset, 197 int index_bias, 198 unsigned *end_cs_dwords) 199{ 200 unsigned end_dwords = 0; 201 boolean flushed = FALSE; 202 boolean first_draw = flags & PREP_FIRST_DRAW; 203 boolean emit_aos = flags & PREP_EMIT_AOS; 204 boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL; 205 boolean indexed = flags & PREP_INDEXED; 206 boolean hw_index_bias = index_bias_supported(r300); 207 208 /* Add dirty state, index offset, and AOS. */ 209 if (first_draw) { 210 cs_dwords += r300_get_num_dirty_dwords(r300); 211 212 if (hw_index_bias) 213 cs_dwords += 2; /* emit_index_offset */ 214 215 if (emit_aos) 216 cs_dwords += 55; /* emit_aos */ 217 218 if (emit_aos_swtcl) 219 cs_dwords += 7; /* emit_aos_swtcl */ 220 } 221 222 /* Emitted in flush. */ 223 end_dwords += 26; /* emit_query_end */ 224 225 cs_dwords += end_dwords; 226 227 /* Reserve requested CS space. */ 228 if (!r300_check_cs(r300, cs_dwords)) { 229 r300->context.flush(&r300->context, 0, NULL); 230 flushed = TRUE; 231 } 232 233 /* Validate buffers and emit dirty state if needed. */ 234 if (first_draw || flushed) { 235 r300_emit_buffer_validate(r300, flags & PREP_VALIDATE_VBOS, index_buffer); 236 r300_emit_dirty_state(r300); 237 if (hw_index_bias) { 238 if (r300->screen->caps.has_tcl) 239 r500_emit_index_bias(r300, index_bias); 240 else 241 r500_emit_index_bias(r300, 0); 242 } 243 244 if (emit_aos) 245 r300_emit_aos(r300, aos_offset, indexed); 246 247 if (emit_aos_swtcl) 248 r300_emit_aos_swtcl(r300, indexed); 249 } 250 251 if (end_cs_dwords) 252 *end_cs_dwords = end_dwords; 253} 254 255static boolean immd_is_good_idea(struct r300_context *r300, 256 unsigned count) 257{ 258 struct pipe_vertex_element* velem; 259 struct pipe_vertex_buffer* vbuf; 260 boolean checked[PIPE_MAX_ATTRIBS] = {0}; 261 unsigned vertex_element_count = r300->velems->count; 262 unsigned i, vbi; 263 264 if (DBG_ON(r300, DBG_NO_IMMD)) { 265 return FALSE; 266 } 267 268 if (r300->draw) { 269 return FALSE; 270 } 271 272 if (count > 10) { 273 return FALSE; 274 } 275 276 /* We shouldn't map buffers referenced by CS, busy buffers, 277 * and ones placed in VRAM. */ 278 /* XXX Check for VRAM buffers. */ 279 for (i = 0; i < vertex_element_count; i++) { 280 velem = &r300->velems->velem[i]; 281 vbi = velem->vertex_buffer_index; 282 283 if (!checked[vbi]) { 284 vbuf = &r300->vertex_buffer[vbi]; 285 286 if (r300_buffer_is_referenced(&r300->context, 287 vbuf->buffer, 288 R300_REF_CS | R300_REF_HW)) { 289 /* It's a very bad idea to map it... */ 290 return FALSE; 291 } 292 checked[vbi] = TRUE; 293 } 294 } 295 return TRUE; 296} 297 298/***************************************************************************** 299 * The emission of draw packets for r500. Older GPUs may use these functions * 300 * after resolving fallback issues (e.g. stencil ref two-sided). * 301 ****************************************************************************/ 302 303static void r300_emit_draw_arrays_immediate(struct r300_context *r300, 304 unsigned mode, 305 unsigned start, 306 unsigned count) 307{ 308 struct pipe_vertex_element* velem; 309 struct pipe_vertex_buffer* vbuf; 310 unsigned vertex_element_count = r300->velems->count; 311 unsigned i, v, vbi, dw, elem_offset, dwords; 312 313 /* Size of the vertex, in dwords. */ 314 unsigned vertex_size = 0; 315 316 /* Offsets of the attribute, in dwords, from the start of the vertex. */ 317 unsigned offset[PIPE_MAX_ATTRIBS]; 318 319 /* Size of the vertex element, in dwords. */ 320 unsigned size[PIPE_MAX_ATTRIBS]; 321 322 /* Stride to the same attrib in the next vertex in the vertex buffer, 323 * in dwords. */ 324 unsigned stride[PIPE_MAX_ATTRIBS] = {0}; 325 326 /* Mapped vertex buffers. */ 327 uint32_t* map[PIPE_MAX_ATTRIBS] = {0}; 328 struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {NULL}; 329 330 CS_LOCALS(r300); 331 332 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */ 333 for (i = 0; i < vertex_element_count; i++) { 334 velem = &r300->velems->velem[i]; 335 offset[i] = velem->src_offset / 4; 336 size[i] = r300->velems->hw_format_size[i] / 4; 337 vertex_size += size[i]; 338 vbi = velem->vertex_buffer_index; 339 340 /* Map the buffer. */ 341 if (!map[vbi]) { 342 vbuf = &r300->vertex_buffer[vbi]; 343 map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context, 344 vbuf->buffer, 345 PIPE_TRANSFER_READ, 346 &transfer[vbi]); 347 map[vbi] += vbuf->buffer_offset / 4; 348 stride[vbi] = vbuf->stride / 4; 349 } 350 } 351 352 dwords = 9 + count * vertex_size; 353 354 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0, NULL); 355 356 BEGIN_CS(dwords); 357 OUT_CS_REG(R300_GA_COLOR_CONTROL, 358 r300_provoking_vertex_fixes(r300, mode)); 359 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); 360 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 361 OUT_CS(count - 1); 362 OUT_CS(0); 363 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size); 364 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) | 365 r300_translate_primitive(mode)); 366 367 /* Emit vertices. */ 368 for (v = 0; v < count; v++) { 369 for (i = 0; i < vertex_element_count; i++) { 370 velem = &r300->velems->velem[i]; 371 vbi = velem->vertex_buffer_index; 372 elem_offset = offset[i] + stride[vbi] * (v + start); 373 374 for (dw = 0; dw < size[i]; dw++) { 375 OUT_CS(map[vbi][elem_offset + dw]); 376 } 377 } 378 } 379 END_CS; 380 381 /* Unmap buffers. */ 382 for (i = 0; i < vertex_element_count; i++) { 383 vbi = r300->velems->velem[i].vertex_buffer_index; 384 385 if (map[vbi]) { 386 vbuf = &r300->vertex_buffer[vbi]; 387 pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]); 388 map[vbi] = NULL; 389 } 390 } 391} 392 393static void r300_emit_draw_arrays(struct r300_context *r300, 394 unsigned mode, 395 unsigned count) 396{ 397 boolean alt_num_verts = count > 65535; 398 CS_LOCALS(r300); 399 400 if (count >= (1 << 24)) { 401 fprintf(stderr, "r300: Got a huge number of vertices: %i, " 402 "refusing to render.\n", count); 403 return; 404 } 405 406 BEGIN_CS(7 + (alt_num_verts ? 2 : 0)); 407 if (alt_num_verts) { 408 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); 409 } 410 OUT_CS_REG(R300_GA_COLOR_CONTROL, 411 r300_provoking_vertex_fixes(r300, mode)); 412 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 413 OUT_CS(count - 1); 414 OUT_CS(0); 415 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); 416 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | 417 r300_translate_primitive(mode) | 418 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 419 END_CS; 420} 421 422static void r300_emit_draw_elements(struct r300_context *r300, 423 struct pipe_resource* indexBuffer, 424 unsigned indexSize, 425 unsigned minIndex, 426 unsigned maxIndex, 427 unsigned mode, 428 unsigned start, 429 unsigned count) 430{ 431 uint32_t count_dwords; 432 uint32_t offset_dwords = indexSize * start / sizeof(uint32_t); 433 boolean alt_num_verts = count > 65535; 434 CS_LOCALS(r300); 435 436 if (count >= (1 << 24)) { 437 fprintf(stderr, "r300: Got a huge number of vertices: %i, " 438 "refusing to render.\n", count); 439 return; 440 } 441 442 maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index); 443 444 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n", 445 count, minIndex, maxIndex); 446 447 BEGIN_CS(13 + (alt_num_verts ? 2 : 0)); 448 if (alt_num_verts) { 449 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); 450 } 451 OUT_CS_REG(R300_GA_COLOR_CONTROL, 452 r300_provoking_vertex_fixes(r300, mode)); 453 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 454 OUT_CS(maxIndex); 455 OUT_CS(minIndex); 456 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0); 457 if (indexSize == 4) { 458 count_dwords = count; 459 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | 460 R300_VAP_VF_CNTL__INDEX_SIZE_32bit | 461 r300_translate_primitive(mode) | 462 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 463 } else { 464 count_dwords = (count + 1) / 2; 465 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | 466 r300_translate_primitive(mode) | 467 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 468 } 469 470 /* INDX_BUFFER is a truly special packet3. 471 * Unlike most other packet3, where the offset is after the count, 472 * the order is reversed, so the relocation ends up carrying the 473 * size of the indexbuf instead of the offset. 474 */ 475 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2); 476 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) | 477 (0 << R300_INDX_BUFFER_SKIP_SHIFT)); 478 OUT_CS(offset_dwords << 2); 479 OUT_CS_BUF_RELOC(indexBuffer, count_dwords, 480 r300_buffer(indexBuffer)->domain, 0, 0); 481 482 END_CS; 483} 484 485/* This is the fast-path drawing & emission for HW TCL. */ 486static void r300_draw_range_elements(struct pipe_context* pipe, 487 struct pipe_resource* indexBuffer, 488 unsigned indexSize, 489 int indexBias, 490 unsigned minIndex, 491 unsigned maxIndex, 492 unsigned mode, 493 unsigned start, 494 unsigned count) 495{ 496 struct r300_context* r300 = r300_context(pipe); 497 struct pipe_resource* orgIndexBuffer = indexBuffer; 498 boolean alt_num_verts = r300->screen->caps.is_r500 && 499 count > 65536 && 500 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); 501 unsigned short_count; 502 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */ 503 boolean translate = FALSE; 504 505 if (r300->skip_rendering) { 506 return; 507 } 508 509 if (!u_trim_pipe_prim(mode, &count)) { 510 return; 511 } 512 513 /* Set up fallback for incompatible vertex layout if needed. */ 514 if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) { 515 r300_begin_vertex_translate(r300); 516 translate = TRUE; 517 } 518 519 if (indexBias && !index_bias_supported(r300)) { 520 r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset); 521 } 522 523 r300_translate_index_buffer(r300, &indexBuffer, &indexSize, index_offset, 524 &start, count); 525 526 r300_update_derived_state(r300); 527 r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count); 528 529 /* 15 dwords for emit_draw_elements */ 530 r300_prepare_for_rendering(r300, 531 PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, 532 indexBuffer, 15, buffer_offset, indexBias, NULL); 533 534 u_upload_flush(r300->upload_vb); 535 u_upload_flush(r300->upload_ib); 536 if (alt_num_verts || count <= 65535) { 537 r300_emit_draw_elements(r300, indexBuffer, indexSize, 538 minIndex, maxIndex, mode, start, count); 539 } else { 540 do { 541 short_count = MIN2(count, 65534); 542 r300_emit_draw_elements(r300, indexBuffer, indexSize, 543 minIndex, maxIndex, 544 mode, start, short_count); 545 546 start += short_count; 547 count -= short_count; 548 549 /* 15 dwords for emit_draw_elements */ 550 if (count) { 551 r300_prepare_for_rendering(r300, 552 PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED, 553 indexBuffer, 15, buffer_offset, indexBias, NULL); 554 } 555 } while (count); 556 } 557 558 if (indexBuffer != orgIndexBuffer) { 559 pipe_resource_reference( &indexBuffer, NULL ); 560 } 561 562 if (translate) { 563 r300_end_vertex_translate(r300); 564 } 565} 566 567/* Simple helpers for context setup. Should probably be moved to util. */ 568static void r300_draw_elements(struct pipe_context* pipe, 569 struct pipe_resource* indexBuffer, 570 unsigned indexSize, int indexBias, unsigned mode, 571 unsigned start, unsigned count) 572{ 573 struct r300_context *r300 = r300_context(pipe); 574 575 pipe->draw_range_elements(pipe, indexBuffer, indexSize, indexBias, 576 0, r300->vertex_buffer_max_index, 577 mode, start, count); 578} 579 580static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, 581 unsigned start, unsigned count) 582{ 583 struct r300_context* r300 = r300_context(pipe); 584 boolean alt_num_verts = r300->screen->caps.is_r500 && 585 count > 65536 && 586 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0); 587 unsigned short_count; 588 boolean translate = FALSE; 589 590 if (r300->skip_rendering) { 591 return; 592 } 593 594 if (!u_trim_pipe_prim(mode, &count)) { 595 return; 596 } 597 598 /* Set up fallback for incompatible vertex layout if needed. */ 599 if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) { 600 r300_begin_vertex_translate(r300); 601 translate = TRUE; 602 } 603 604 r300_update_derived_state(r300); 605 606 if (immd_is_good_idea(r300, count)) { 607 r300_emit_draw_arrays_immediate(r300, mode, start, count); 608 } else { 609 /* 9 spare dwords for emit_draw_arrays. */ 610 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS, 611 NULL, 9, start, 0, NULL); 612 613 if (alt_num_verts || count <= 65535) { 614 r300_emit_draw_arrays(r300, mode, count); 615 } else { 616 do { 617 short_count = MIN2(count, 65535); 618 r300_emit_draw_arrays(r300, mode, short_count); 619 620 start += short_count; 621 count -= short_count; 622 623 /* 9 spare dwords for emit_draw_arrays. */ 624 if (count) { 625 r300_prepare_for_rendering(r300, 626 PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9, 627 start, 0, NULL); 628 } 629 } while (count); 630 } 631 u_upload_flush(r300->upload_vb); 632 } 633 634 if (translate) { 635 r300_end_vertex_translate(r300); 636 } 637} 638 639/**************************************************************************** 640 * The rest of this file is for SW TCL rendering only. Please be polite and * 641 * keep these functions separated so that they are easier to locate. ~C. * 642 ***************************************************************************/ 643 644/* SW TCL arrays, using Draw. */ 645static void r300_swtcl_draw_arrays(struct pipe_context* pipe, 646 unsigned mode, 647 unsigned start, 648 unsigned count) 649{ 650 struct r300_context* r300 = r300_context(pipe); 651 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS]; 652 int i; 653 654 if (r300->skip_rendering) { 655 return; 656 } 657 658 if (!u_trim_pipe_prim(mode, &count)) { 659 return; 660 } 661 662 r300_update_derived_state(r300); 663 664 for (i = 0; i < r300->vertex_buffer_count; i++) { 665 void* buf = pipe_buffer_map(pipe, 666 r300->vertex_buffer[i].buffer, 667 PIPE_TRANSFER_READ, 668 &vb_transfer[i]); 669 draw_set_mapped_vertex_buffer(r300->draw, i, buf); 670 } 671 672 draw_set_mapped_element_buffer(r300->draw, 0, 0, NULL); 673 674 draw_arrays(r300->draw, mode, start, count); 675 676 /* XXX Not sure whether this is the best fix. 677 * It prevents CS from being rejected and weird assertion failures. */ 678 draw_flush(r300->draw); 679 680 for (i = 0; i < r300->vertex_buffer_count; i++) { 681 pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, 682 vb_transfer[i]); 683 draw_set_mapped_vertex_buffer(r300->draw, i, NULL); 684 } 685} 686 687/* SW TCL elements, using Draw. */ 688static void r300_swtcl_draw_range_elements(struct pipe_context* pipe, 689 struct pipe_resource* indexBuffer, 690 unsigned indexSize, 691 int indexBias, 692 unsigned minIndex, 693 unsigned maxIndex, 694 unsigned mode, 695 unsigned start, 696 unsigned count) 697{ 698 struct r300_context* r300 = r300_context(pipe); 699 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS]; 700 struct pipe_transfer *ib_transfer; 701 int i; 702 void* indices; 703 704 if (r300->skip_rendering) { 705 return; 706 } 707 708 if (!u_trim_pipe_prim(mode, &count)) { 709 return; 710 } 711 712 r300_update_derived_state(r300); 713 714 for (i = 0; i < r300->vertex_buffer_count; i++) { 715 void* buf = pipe_buffer_map(pipe, 716 r300->vertex_buffer[i].buffer, 717 PIPE_TRANSFER_READ, 718 &vb_transfer[i]); 719 draw_set_mapped_vertex_buffer(r300->draw, i, buf); 720 } 721 722 indices = pipe_buffer_map(pipe, indexBuffer, 723 PIPE_TRANSFER_READ, &ib_transfer); 724 draw_set_mapped_element_buffer_range(r300->draw, indexSize, indexBias, 725 minIndex, maxIndex, indices); 726 727 draw_arrays(r300->draw, mode, start, count); 728 729 /* XXX Not sure whether this is the best fix. 730 * It prevents CS from being rejected and weird assertion failures. */ 731 draw_flush(r300->draw); 732 733 for (i = 0; i < r300->vertex_buffer_count; i++) { 734 pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer, 735 vb_transfer[i]); 736 draw_set_mapped_vertex_buffer(r300->draw, i, NULL); 737 } 738 739 pipe_buffer_unmap(pipe, indexBuffer, 740 ib_transfer); 741 draw_set_mapped_element_buffer_range(r300->draw, 0, 0, 742 start, start + count - 1, 743 NULL); 744} 745 746/* Object for rendering using Draw. */ 747struct r300_render { 748 /* Parent class */ 749 struct vbuf_render base; 750 751 /* Pipe context */ 752 struct r300_context* r300; 753 754 /* Vertex information */ 755 size_t vertex_size; 756 unsigned prim; 757 unsigned hwprim; 758 759 /* VBO */ 760 struct pipe_resource* vbo; 761 size_t vbo_size; 762 size_t vbo_offset; 763 size_t vbo_max_used; 764 void * vbo_ptr; 765 766 struct pipe_transfer *vbo_transfer; 767}; 768 769static INLINE struct r300_render* 770r300_render(struct vbuf_render* render) 771{ 772 return (struct r300_render*)render; 773} 774 775static const struct vertex_info* 776r300_render_get_vertex_info(struct vbuf_render* render) 777{ 778 struct r300_render* r300render = r300_render(render); 779 struct r300_context* r300 = r300render->r300; 780 781 return &r300->vertex_info; 782} 783 784static boolean r300_render_allocate_vertices(struct vbuf_render* render, 785 ushort vertex_size, 786 ushort count) 787{ 788 struct r300_render* r300render = r300_render(render); 789 struct r300_context* r300 = r300render->r300; 790 struct pipe_screen* screen = r300->context.screen; 791 size_t size = (size_t)vertex_size * (size_t)count; 792 793 if (size + r300render->vbo_offset > r300render->vbo_size) 794 { 795 pipe_resource_reference(&r300->vbo, NULL); 796 r300render->vbo = pipe_buffer_create(screen, 797 PIPE_BIND_VERTEX_BUFFER, 798 R300_MAX_DRAW_VBO_SIZE); 799 r300render->vbo_offset = 0; 800 r300render->vbo_size = R300_MAX_DRAW_VBO_SIZE; 801 } 802 803 r300render->vertex_size = vertex_size; 804 r300->vbo = r300render->vbo; 805 r300->vbo_offset = r300render->vbo_offset; 806 807 return (r300render->vbo) ? TRUE : FALSE; 808} 809 810static void* r300_render_map_vertices(struct vbuf_render* render) 811{ 812 struct r300_render* r300render = r300_render(render); 813 814 assert(!r300render->vbo_transfer); 815 816 r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context, 817 r300render->vbo, 818 PIPE_TRANSFER_WRITE, 819 &r300render->vbo_transfer); 820 821 return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset); 822} 823 824static void r300_render_unmap_vertices(struct vbuf_render* render, 825 ushort min, 826 ushort max) 827{ 828 struct r300_render* r300render = r300_render(render); 829 struct pipe_context* context = &r300render->r300->context; 830 831 assert(r300render->vbo_transfer); 832 833 r300render->vbo_max_used = MAX2(r300render->vbo_max_used, 834 r300render->vertex_size * (max + 1)); 835 pipe_buffer_unmap(context, r300render->vbo, r300render->vbo_transfer); 836 837 r300render->vbo_transfer = NULL; 838} 839 840static void r300_render_release_vertices(struct vbuf_render* render) 841{ 842 struct r300_render* r300render = r300_render(render); 843 844 r300render->vbo_offset += r300render->vbo_max_used; 845 r300render->vbo_max_used = 0; 846} 847 848static boolean r300_render_set_primitive(struct vbuf_render* render, 849 unsigned prim) 850{ 851 struct r300_render* r300render = r300_render(render); 852 853 r300render->prim = prim; 854 r300render->hwprim = r300_translate_primitive(prim); 855 856 return TRUE; 857} 858 859static void r300_render_draw_arrays(struct vbuf_render* render, 860 unsigned start, 861 unsigned count) 862{ 863 struct r300_render* r300render = r300_render(render); 864 struct r300_context* r300 = r300render->r300; 865 uint8_t* ptr; 866 unsigned i; 867 unsigned dwords = 6; 868 869 CS_LOCALS(r300); 870 871 (void) i; (void) ptr; 872 873 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL, 874 NULL, dwords, 0, 0, NULL); 875 876 DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count); 877 878 /* Uncomment to dump all VBOs rendered through this interface. 879 * Slow and noisy! 880 ptr = pipe_buffer_map(&r300render->r300->context, 881 r300render->vbo, PIPE_TRANSFER_READ, 882 &r300render->vbo_transfer); 883 884 for (i = 0; i < count; i++) { 885 printf("r300: Vertex %d\n", i); 886 draw_dump_emitted_vertex(&r300->vertex_info, ptr); 887 ptr += r300->vertex_info.size * 4; 888 printf("\n"); 889 } 890 891 pipe_buffer_unmap(&r300render->r300->context, r300render->vbo, 892 r300render->vbo_transfer); 893 */ 894 895 BEGIN_CS(dwords); 896 OUT_CS_REG(R300_GA_COLOR_CONTROL, 897 r300_provoking_vertex_fixes(r300, r300render->prim)); 898 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1); 899 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); 900 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | 901 r300render->hwprim); 902 END_CS; 903} 904 905static void r300_render_draw_elements(struct vbuf_render* render, 906 const ushort* indices, 907 uint count) 908{ 909 struct r300_render* r300render = r300_render(render); 910 struct r300_context* r300 = r300render->r300; 911 int i; 912 unsigned end_cs_dwords; 913 unsigned max_index = (r300render->vbo_size - r300render->vbo_offset) / 914 (r300render->r300->vertex_info.size * 4) - 1; 915 unsigned short_count; 916 struct r300_cs_info cs_info; 917 918 CS_LOCALS(r300); 919 920 /* Reserve at least 256 dwords. 921 * 922 * Below we manage the CS space manually because there may be more 923 * indices than it can fit in CS. */ 924 r300_prepare_for_rendering(r300, 925 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED, 926 NULL, 256, 0, 0, &end_cs_dwords); 927 928 while (count) { 929 r300->rws->get_cs_info(r300->rws, &cs_info); 930 931 short_count = MIN2(count, (cs_info.free - end_cs_dwords - 6) * 2); 932 933 BEGIN_CS(6 + (short_count+1)/2); 934 OUT_CS_REG(R300_GA_COLOR_CONTROL, 935 r300_provoking_vertex_fixes(r300, r300render->prim)); 936 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index); 937 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2); 938 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) | 939 r300render->hwprim); 940 for (i = 0; i < short_count-1; i += 2) { 941 OUT_CS(indices[i+1] << 16 | indices[i]); 942 } 943 if (short_count % 2) { 944 OUT_CS(indices[short_count-1]); 945 } 946 END_CS; 947 948 /* OK now subtract the emitted indices and see if we need to emit 949 * another draw packet. */ 950 indices += short_count; 951 count -= short_count; 952 953 if (count) { 954 r300_prepare_for_rendering(r300, 955 PREP_EMIT_AOS_SWTCL | PREP_INDEXED, 956 NULL, 256, 0, 0, &end_cs_dwords); 957 } 958 } 959} 960 961static void r300_render_destroy(struct vbuf_render* render) 962{ 963 FREE(render); 964} 965 966static struct vbuf_render* r300_render_create(struct r300_context* r300) 967{ 968 struct r300_render* r300render = CALLOC_STRUCT(r300_render); 969 970 r300render->r300 = r300; 971 972 /* XXX find real numbers plz */ 973 r300render->base.max_vertex_buffer_bytes = 128 * 1024; 974 r300render->base.max_indices = 16 * 1024; 975 976 r300render->base.get_vertex_info = r300_render_get_vertex_info; 977 r300render->base.allocate_vertices = r300_render_allocate_vertices; 978 r300render->base.map_vertices = r300_render_map_vertices; 979 r300render->base.unmap_vertices = r300_render_unmap_vertices; 980 r300render->base.set_primitive = r300_render_set_primitive; 981 r300render->base.draw_elements = r300_render_draw_elements; 982 r300render->base.draw_arrays = r300_render_draw_arrays; 983 r300render->base.release_vertices = r300_render_release_vertices; 984 r300render->base.destroy = r300_render_destroy; 985 986 r300render->vbo = NULL; 987 r300render->vbo_size = 0; 988 r300render->vbo_offset = 0; 989 990 return &r300render->base; 991} 992 993struct draw_stage* r300_draw_stage(struct r300_context* r300) 994{ 995 struct vbuf_render* render; 996 struct draw_stage* stage; 997 998 render = r300_render_create(r300); 999 1000 if (!render) { 1001 return NULL; 1002 } 1003 1004 stage = draw_vbuf_stage(r300->draw, render); 1005 1006 if (!stage) { 1007 render->destroy(render); 1008 return NULL; 1009 } 1010 1011 draw_set_render(r300->draw, render); 1012 1013 return stage; 1014} 1015 1016void r300_init_render_functions(struct r300_context *r300) 1017{ 1018 /* Set generic functions. */ 1019 r300->context.draw_elements = r300_draw_elements; 1020 1021 /* Set draw functions based on presence of HW TCL. */ 1022 if (r300->screen->caps.has_tcl) { 1023 r300->context.draw_arrays = r300_draw_arrays; 1024 r300->context.draw_range_elements = r300_draw_range_elements; 1025 } else { 1026 r300->context.draw_arrays = r300_swtcl_draw_arrays; 1027 r300->context.draw_range_elements = r300_swtcl_draw_range_elements; 1028 } 1029 1030 /* Plug in the two-sided stencil reference value fallback if needed. */ 1031 if (!r300->screen->caps.is_r500) 1032 r300_plug_in_stencil_ref_fallback(r300); 1033} 1034