1 /* 2 * Copyright 2009 Corbin Simpson <MostAwesomeDude (at) gmail.com> 3 * Copyright 2010 Marek Olk <maraeo (at) 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 43 #include <limits.h> 44 45 #define IMMD_DWORDS 32 46 47 static uint32_t r300_translate_primitive(unsigned prim) 48 { 49 static const int prim_conv[] = { 50 R300_VAP_VF_CNTL__PRIM_POINTS, 51 R300_VAP_VF_CNTL__PRIM_LINES, 52 R300_VAP_VF_CNTL__PRIM_LINE_LOOP, 53 R300_VAP_VF_CNTL__PRIM_LINE_STRIP, 54 R300_VAP_VF_CNTL__PRIM_TRIANGLES, 55 R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP, 56 R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN, 57 R300_VAP_VF_CNTL__PRIM_QUADS, 58 R300_VAP_VF_CNTL__PRIM_QUAD_STRIP, 59 R300_VAP_VF_CNTL__PRIM_POLYGON, 60 -1, 61 -1, 62 -1, 63 -1 64 }; 65 unsigned hwprim = prim_conv[prim]; 66 67 assert(hwprim != -1); 68 return hwprim; 69 } 70 71 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, 72 unsigned mode) 73 { 74 struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state; 75 uint32_t color_control = rs->color_control; 76 77 /* By default (see r300_state.c:r300_create_rs_state) color_control is 78 * initialized to provoking the first vertex. 79 * 80 * Triangle fans must be reduced to the second vertex, not the first, in 81 * Gallium flatshade-first mode, as per the GL spec. 82 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt) 83 * 84 * Quads never provoke correctly in flatshade-first mode. The first 85 * vertex is never considered as provoking, so only the second, third, 86 * and fourth vertices can be selected, and both "third" and "last" modes 87 * select the fourth vertex. This is probably due to D3D lacking quads. 88 * 89 * Similarly, polygons reduce to the first, not the last, vertex, when in 90 * "last" mode, and all other modes start from the second vertex. 91 * 92 * ~ C. 93 */ 94 95 if (rs->rs.flatshade_first) { 96 switch (mode) { 97 case PIPE_PRIM_TRIANGLE_FAN: 98 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND; 99 break; 100 case PIPE_PRIM_QUADS: 101 case PIPE_PRIM_QUAD_STRIP: 102 case PIPE_PRIM_POLYGON: 103 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; 104 break; 105 default: 106 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST; 107 break; 108 } 109 } else { 110 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; 111 } 112 113 return color_control; 114 } 115 116 void r500_emit_index_bias(struct r300_context *r300, int index_bias) 117 { 118 CS_LOCALS(r300); 119 120 BEGIN_CS(2); 121 OUT_CS_REG(R500_VAP_INDEX_OFFSET, 122 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0)); 123 END_CS; 124 } 125 126 static void r300_emit_draw_init(struct r300_context *r300, unsigned mode, 127 unsigned max_index) 128 { 129 CS_LOCALS(r300); 130 131 assert(max_index < (1 << 24)); 132 133 BEGIN_CS(5); 134 OUT_CS_REG(R300_GA_COLOR_CONTROL, 135 r300_provoking_vertex_fixes(r300, mode)); 136 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 137 OUT_CS(max_index); 138 OUT_CS(0); 139 END_CS; 140 } 141 142 /* This function splits the index bias value into two parts: 143 * - buffer_offset: the value that can be safely added to buffer offsets 144 * in r300_emit_vertex_arrays (it must yield a positive offset when added to 145 * a vertex buffer offset) 146 * - index_offset: the value that must be manually subtracted from indices 147 * in an index buffer to achieve negative offsets. */ 148 static void r300_split_index_bias(struct r300_context *r300, int index_bias, 149 int *buffer_offset, int *index_offset) 150 { 151 struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer; 152 struct pipe_vertex_element *velem = r300->velems->velem; 153 unsigned i, size; 154 int max_neg_bias; 155 156 if (index_bias < 0) { 157 /* See how large index bias we may subtract. We must be careful 158 * here because negative buffer offsets are not allowed 159 * by the DRM API. */ 160 max_neg_bias = INT_MAX; 161 for (i = 0; i < r300->velems->count; i++) { 162 vb = &vbufs[velem[i].vertex_buffer_index]; 163 size = (vb->buffer_offset + velem[i].src_offset) / vb->stride; 164 max_neg_bias = MIN2(max_neg_bias, size); 165 } 166 167 /* Now set the minimum allowed value. */ 168 *buffer_offset = MAX2(-max_neg_bias, index_bias); 169 } else { 170 /* A positive index bias is OK. */ 171 *buffer_offset = index_bias; 172 } 173 174 *index_offset = index_bias - *buffer_offset; 175 } 176 177 enum r300_prepare_flags { 178 PREP_EMIT_STATES = (1 << 0), /* call emit_dirty_state and friends? */ 179 PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */ 180 PREP_EMIT_VARRAYS = (1 << 2), /* call emit_vertex_arrays? */ 181 PREP_EMIT_VARRAYS_SWTCL = (1 << 3), /* call emit_vertex_arrays_swtcl? */ 182 PREP_INDEXED = (1 << 4) /* is this draw_elements? */ 183 }; 184 185 /** 186 * Check if the requested number of dwords is available in the CS and 187 * if not, flush. 188 * \param r300 The context. 189 * \param flags See r300_prepare_flags. 190 * \param cs_dwords The number of dwords to reserve in CS. 191 * \return TRUE if the CS was flushed 192 */ 193 static boolean r300_reserve_cs_dwords(struct r300_context *r300, 194 enum r300_prepare_flags flags, 195 unsigned cs_dwords) 196 { 197 boolean flushed = FALSE; 198 boolean emit_states = flags & PREP_EMIT_STATES; 199 boolean emit_vertex_arrays = flags & PREP_EMIT_VARRAYS; 200 boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL; 201 202 /* Add dirty state, index offset, and AOS. */ 203 if (emit_states) 204 cs_dwords += r300_get_num_dirty_dwords(r300); 205 206 if (r300->screen->caps.is_r500) 207 cs_dwords += 2; /* emit_index_offset */ 208 209 if (emit_vertex_arrays) 210 cs_dwords += 55; /* emit_vertex_arrays */ 211 212 if (emit_vertex_arrays_swtcl) 213 cs_dwords += 7; /* emit_vertex_arrays_swtcl */ 214 215 cs_dwords += r300_get_num_cs_end_dwords(r300); 216 217 /* Reserve requested CS space. */ 218 if (!r300->rws->cs_check_space(r300->cs, cs_dwords)) { 219 r300_flush(&r300->context, PIPE_FLUSH_ASYNC, NULL); 220 flushed = TRUE; 221 } 222 223 return flushed; 224 } 225 226 /** 227 * Validate buffers and emit dirty state. 228 * \param r300 The context. 229 * \param flags See r300_prepare_flags. 230 * \param index_buffer The index buffer to validate. The parameter may be NULL. 231 * \param buffer_offset The offset passed to emit_vertex_arrays. 232 * \param index_bias The index bias to emit. 233 * \param instance_id Index of instance to render 234 * \return TRUE if rendering should be skipped 235 */ 236 static boolean r300_emit_states(struct r300_context *r300, 237 enum r300_prepare_flags flags, 238 struct pipe_resource *index_buffer, 239 int buffer_offset, 240 int index_bias, int instance_id) 241 { 242 boolean emit_states = flags & PREP_EMIT_STATES; 243 boolean emit_vertex_arrays = flags & PREP_EMIT_VARRAYS; 244 boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL; 245 boolean indexed = flags & PREP_INDEXED; 246 boolean validate_vbos = flags & PREP_VALIDATE_VBOS; 247 248 /* Validate buffers and emit dirty state if needed. */ 249 if (emit_states || (emit_vertex_arrays && validate_vbos)) { 250 if (!r300_emit_buffer_validate(r300, validate_vbos, 251 index_buffer)) { 252 fprintf(stderr, "r300: CS space validation failed. " 253 "(not enough memory?) Skipping rendering.\n"); 254 return FALSE; 255 } 256 } 257 258 if (emit_states) 259 r300_emit_dirty_state(r300); 260 261 if (r300->screen->caps.is_r500) { 262 if (r300->screen->caps.has_tcl) 263 r500_emit_index_bias(r300, index_bias); 264 else 265 r500_emit_index_bias(r300, 0); 266 } 267 268 if (emit_vertex_arrays && 269 (r300->vertex_arrays_dirty || 270 r300->vertex_arrays_indexed != indexed || 271 r300->vertex_arrays_offset != buffer_offset || 272 r300->vertex_arrays_instance_id != instance_id)) { 273 r300_emit_vertex_arrays(r300, buffer_offset, indexed, instance_id); 274 275 r300->vertex_arrays_dirty = FALSE; 276 r300->vertex_arrays_indexed = indexed; 277 r300->vertex_arrays_offset = buffer_offset; 278 r300->vertex_arrays_instance_id = instance_id; 279 } 280 281 if (emit_vertex_arrays_swtcl) 282 r300_emit_vertex_arrays_swtcl(r300, indexed); 283 284 return TRUE; 285 } 286 287 /** 288 * Check if the requested number of dwords is available in the CS and 289 * if not, flush. Then validate buffers and emit dirty state. 290 * \param r300 The context. 291 * \param flags See r300_prepare_flags. 292 * \param index_buffer The index buffer to validate. The parameter may be NULL. 293 * \param cs_dwords The number of dwords to reserve in CS. 294 * \param buffer_offset The offset passed to emit_vertex_arrays. 295 * \param index_bias The index bias to emit. 296 * \param instance_id The instance to render. 297 * \return TRUE if rendering should be skipped 298 */ 299 static boolean r300_prepare_for_rendering(struct r300_context *r300, 300 enum r300_prepare_flags flags, 301 struct pipe_resource *index_buffer, 302 unsigned cs_dwords, 303 int buffer_offset, 304 int index_bias, 305 int instance_id) 306 { 307 /* Make sure there is enough space in the command stream and emit states. */ 308 if (r300_reserve_cs_dwords(r300, flags, cs_dwords)) 309 flags |= PREP_EMIT_STATES; 310 311 return r300_emit_states(r300, flags, index_buffer, buffer_offset, 312 index_bias, instance_id); 313 } 314 315 static boolean immd_is_good_idea(struct r300_context *r300, 316 unsigned count) 317 { 318 if (DBG_ON(r300, DBG_NO_IMMD)) { 319 return FALSE; 320 } 321 322 if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) { 323 return FALSE; 324 } 325 326 /* Buffers can only be used for read by r300 (except query buffers, but 327 * those can't be bound by a state tracker as vertex buffers). */ 328 return TRUE; 329 } 330 331 /***************************************************************************** 332 * The HWTCL draw functions. * 333 ****************************************************************************/ 334 335 static void r300_draw_arrays_immediate(struct r300_context *r300, 336 const struct pipe_draw_info *info) 337 { 338 struct pipe_vertex_element* velem; 339 struct pipe_vertex_buffer* vbuf; 340 unsigned vertex_element_count = r300->velems->count; 341 unsigned i, v, vbi; 342 343 /* Size of the vertex, in dwords. */ 344 unsigned vertex_size = r300->velems->vertex_size_dwords; 345 346 /* The number of dwords for this draw operation. */ 347 unsigned dwords = 4 + info->count * vertex_size; 348 349 /* Size of the vertex element, in dwords. */ 350 unsigned size[PIPE_MAX_ATTRIBS]; 351 352 /* Stride to the same attrib in the next vertex in the vertex buffer, 353 * in dwords. */ 354 unsigned stride[PIPE_MAX_ATTRIBS]; 355 356 /* Mapped vertex buffers. */ 357 uint32_t* map[PIPE_MAX_ATTRIBS] = {0}; 358 uint32_t* mapelem[PIPE_MAX_ATTRIBS]; 359 360 CS_LOCALS(r300); 361 362 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1)) 363 return; 364 365 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */ 366 for (i = 0; i < vertex_element_count; i++) { 367 velem = &r300->velems->velem[i]; 368 size[i] = r300->velems->format_size[i] / 4; 369 vbi = velem->vertex_buffer_index; 370 vbuf = &r300->vertex_buffer[vbi]; 371 stride[i] = vbuf->stride / 4; 372 373 /* Map the buffer. */ 374 if (!map[vbi]) { 375 map[vbi] = (uint32_t*)r300->rws->buffer_map( 376 r300_resource(vbuf->buffer.resource)->buf, 377 r300->cs, PIPE_TRANSFER_READ | PIPE_TRANSFER_UNSYNCHRONIZED); 378 map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * info->start; 379 } 380 mapelem[i] = map[vbi] + (velem->src_offset / 4); 381 } 382 383 r300_emit_draw_init(r300, info->mode, info->count-1); 384 385 BEGIN_CS(dwords); 386 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); 387 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, info->count * vertex_size); 388 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (info->count << 16) | 389 r300_translate_primitive(info->mode)); 390 391 /* Emit vertices. */ 392 for (v = 0; v < info->count; v++) { 393 for (i = 0; i < vertex_element_count; i++) { 394 OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]); 395 } 396 } 397 END_CS; 398 } 399 400 static void r300_emit_draw_arrays(struct r300_context *r300, 401 unsigned mode, 402 unsigned count) 403 { 404 boolean alt_num_verts = count > 65535; 405 CS_LOCALS(r300); 406 407 if (count >= (1 << 24)) { 408 fprintf(stderr, "r300: Got a huge number of vertices: %i, " 409 "refusing to render.\n", count); 410 return; 411 } 412 413 r300_emit_draw_init(r300, mode, count-1); 414 415 BEGIN_CS(2 + (alt_num_verts ? 2 : 0)); 416 if (alt_num_verts) { 417 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); 418 } 419 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); 420 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | 421 r300_translate_primitive(mode) | 422 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 423 END_CS; 424 } 425 426 static void r300_emit_draw_elements(struct r300_context *r300, 427 struct pipe_resource* indexBuffer, 428 unsigned indexSize, 429 unsigned max_index, 430 unsigned mode, 431 unsigned start, 432 unsigned count, 433 uint16_t *imm_indices3) 434 { 435 uint32_t count_dwords, offset_dwords; 436 boolean alt_num_verts = count > 65535; 437 CS_LOCALS(r300); 438 439 if (count >= (1 << 24)) { 440 fprintf(stderr, "r300: Got a huge number of vertices: %i, " 441 "refusing to render (max_index: %i).\n", count, max_index); 442 return; 443 } 444 445 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, max %u\n", 446 count, max_index); 447 448 r300_emit_draw_init(r300, mode, max_index); 449 450 /* If start is odd, render the first triangle with indices embedded 451 * in the command stream. This will increase start by 3 and make it 452 * even. We can then proceed without a fallback. */ 453 if (indexSize == 2 && (start & 1) && 454 mode == PIPE_PRIM_TRIANGLES) { 455 BEGIN_CS(4); 456 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2); 457 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) | 458 R300_VAP_VF_CNTL__PRIM_TRIANGLES); 459 OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]); 460 OUT_CS(imm_indices3[2]); 461 END_CS; 462 463 start += 3; 464 count -= 3; 465 if (!count) 466 return; 467 } 468 469 offset_dwords = indexSize * start / sizeof(uint32_t); 470 471 BEGIN_CS(8 + (alt_num_verts ? 2 : 0)); 472 if (alt_num_verts) { 473 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count); 474 } 475 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0); 476 if (indexSize == 4) { 477 count_dwords = count; 478 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | 479 R300_VAP_VF_CNTL__INDEX_SIZE_32bit | 480 r300_translate_primitive(mode) | 481 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 482 } else { 483 count_dwords = (count + 1) / 2; 484 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | 485 r300_translate_primitive(mode) | 486 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0)); 487 } 488 489 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2); 490 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) | 491 (0 << R300_INDX_BUFFER_SKIP_SHIFT)); 492 OUT_CS(offset_dwords << 2); 493 OUT_CS(count_dwords); 494 OUT_CS_RELOC(r300_resource(indexBuffer)); 495 END_CS; 496 } 497 498 static void r300_draw_elements_immediate(struct r300_context *r300, 499 const struct pipe_draw_info *info) 500 { 501 const uint8_t *ptr1; 502 const uint16_t *ptr2; 503 const uint32_t *ptr4; 504 unsigned index_size = info->index_size; 505 unsigned i, count_dwords = index_size == 4 ? info->count : 506 (info->count + 1) / 2; 507 CS_LOCALS(r300); 508 509 /* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */ 510 if (!r300_prepare_for_rendering(r300, 511 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | 512 PREP_INDEXED, NULL, 2+count_dwords, 0, info->index_bias, -1)) 513 return; 514 515 r300_emit_draw_init(r300, info->mode, info->max_index); 516 517 BEGIN_CS(2 + count_dwords); 518 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords); 519 520 switch (index_size) { 521 case 1: 522 ptr1 = (uint8_t*)info->index.user; 523 ptr1 += info->start; 524 525 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) | 526 r300_translate_primitive(info->mode)); 527 528 if (info->index_bias && !r300->screen->caps.is_r500) { 529 for (i = 0; i < info->count-1; i += 2) 530 OUT_CS(((ptr1[i+1] + info->index_bias) << 16) | 531 (ptr1[i] + info->index_bias)); 532 533 if (info->count & 1) 534 OUT_CS(ptr1[i] + info->index_bias); 535 } else { 536 for (i = 0; i < info->count-1; i += 2) 537 OUT_CS(((ptr1[i+1]) << 16) | 538 (ptr1[i] )); 539 540 if (info->count & 1) 541 OUT_CS(ptr1[i]); 542 } 543 break; 544 545 case 2: 546 ptr2 = (uint16_t*)info->index.user; 547 ptr2 += info->start; 548 549 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) | 550 r300_translate_primitive(info->mode)); 551 552 if (info->index_bias && !r300->screen->caps.is_r500) { 553 for (i = 0; i < info->count-1; i += 2) 554 OUT_CS(((ptr2[i+1] + info->index_bias) << 16) | 555 (ptr2[i] + info->index_bias)); 556 557 if (info->count & 1) 558 OUT_CS(ptr2[i] + info->index_bias); 559 } else { 560 OUT_CS_TABLE(ptr2, count_dwords); 561 } 562 break; 563 564 case 4: 565 ptr4 = (uint32_t*)info->index.user; 566 ptr4 += info->start; 567 568 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) | 569 R300_VAP_VF_CNTL__INDEX_SIZE_32bit | 570 r300_translate_primitive(info->mode)); 571 572 if (info->index_bias && !r300->screen->caps.is_r500) { 573 for (i = 0; i < info->count; i++) 574 OUT_CS(ptr4[i] + info->index_bias); 575 } else { 576 OUT_CS_TABLE(ptr4, count_dwords); 577 } 578 break; 579 } 580 END_CS; 581 } 582 583 static void r300_draw_elements(struct r300_context *r300, 584 const struct pipe_draw_info *info, 585 int instance_id) 586 { 587 struct pipe_resource *indexBuffer = 588 info->has_user_indices ? NULL : info->index.resource; 589 unsigned indexSize = info->index_size; 590 struct pipe_resource* orgIndexBuffer = indexBuffer; 591 unsigned start = info->start; 592 unsigned count = info->count; 593 boolean alt_num_verts = r300->screen->caps.is_r500 && 594 count > 65536; 595 unsigned short_count; 596 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */ 597 uint16_t indices3[3]; 598 599 if (info->index_bias && !r300->screen->caps.is_r500) { 600 r300_split_index_bias(r300, info->index_bias, &buffer_offset, 601 &index_offset); 602 } 603 604 r300_translate_index_buffer(r300, info, &indexBuffer, 605 &indexSize, index_offset, &start, count); 606 607 /* Fallback for misaligned ushort indices. */ 608 if (indexSize == 2 && (start & 1) && indexBuffer) { 609 /* If we got here, then orgIndexBuffer == indexBuffer. */ 610 uint16_t *ptr = r300->rws->buffer_map(r300_resource(orgIndexBuffer)->buf, 611 r300->cs, 612 PIPE_TRANSFER_READ | 613 PIPE_TRANSFER_UNSYNCHRONIZED); 614 615 if (info->mode == PIPE_PRIM_TRIANGLES) { 616 memcpy(indices3, ptr + start, 6); 617 } else { 618 /* Copy the mapped index buffer directly to the upload buffer. 619 * The start index will be aligned simply from the fact that 620 * every sub-buffer in the upload buffer is aligned. */ 621 r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start, 622 count, (uint8_t*)ptr); 623 } 624 } else { 625 if (info->has_user_indices) 626 r300_upload_index_buffer(r300, &indexBuffer, indexSize, 627 &start, count, 628 info->index.user); 629 } 630 631 /* 19 dwords for emit_draw_elements. Give up if the function fails. */ 632 if (!r300_prepare_for_rendering(r300, 633 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | 634 PREP_INDEXED, indexBuffer, 19, buffer_offset, info->index_bias, 635 instance_id)) 636 goto done; 637 638 if (alt_num_verts || count <= 65535) { 639 r300_emit_draw_elements(r300, indexBuffer, indexSize, 640 info->max_index, info->mode, start, count, 641 indices3); 642 } else { 643 do { 644 /* The maximum must be divisible by 4 and 3, 645 * so that quad and triangle lists are split correctly. 646 * 647 * Strips, loops, and fans won't work. */ 648 short_count = MIN2(count, 65532); 649 650 r300_emit_draw_elements(r300, indexBuffer, indexSize, 651 info->max_index, 652 info->mode, start, short_count, indices3); 653 654 start += short_count; 655 count -= short_count; 656 657 /* 15 dwords for emit_draw_elements */ 658 if (count) { 659 if (!r300_prepare_for_rendering(r300, 660 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | PREP_INDEXED, 661 indexBuffer, 19, buffer_offset, info->index_bias, 662 instance_id)) 663 goto done; 664 } 665 } while (count); 666 } 667 668 done: 669 if (indexBuffer != orgIndexBuffer) { 670 pipe_resource_reference( &indexBuffer, NULL ); 671 } 672 } 673 674 static void r300_draw_arrays(struct r300_context *r300, 675 const struct pipe_draw_info *info, 676 int instance_id) 677 { 678 boolean alt_num_verts = r300->screen->caps.is_r500 && 679 info->count > 65536; 680 unsigned start = info->start; 681 unsigned count = info->count; 682 unsigned short_count; 683 684 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */ 685 if (!r300_prepare_for_rendering(r300, 686 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS, 687 NULL, 9, start, 0, instance_id)) 688 return; 689 690 if (alt_num_verts || count <= 65535) { 691 r300_emit_draw_arrays(r300, info->mode, count); 692 } else { 693 do { 694 /* The maximum must be divisible by 4 and 3, 695 * so that quad and triangle lists are split correctly. 696 * 697 * Strips, loops, and fans won't work. */ 698 short_count = MIN2(count, 65532); 699 r300_emit_draw_arrays(r300, info->mode, short_count); 700 701 start += short_count; 702 count -= short_count; 703 704 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */ 705 if (count) { 706 if (!r300_prepare_for_rendering(r300, 707 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS, NULL, 9, 708 start, 0, instance_id)) 709 return; 710 } 711 } while (count); 712 } 713 } 714 715 static void r300_draw_arrays_instanced(struct r300_context *r300, 716 const struct pipe_draw_info *info) 717 { 718 int i; 719 720 for (i = 0; i < info->instance_count; i++) 721 r300_draw_arrays(r300, info, i); 722 } 723 724 static void r300_draw_elements_instanced(struct r300_context *r300, 725 const struct pipe_draw_info *info) 726 { 727 int i; 728 729 for (i = 0; i < info->instance_count; i++) 730 r300_draw_elements(r300, info, i); 731 } 732 733 static unsigned r300_max_vertex_count(struct r300_context *r300) 734 { 735 unsigned i, nr = r300->velems->count; 736 struct pipe_vertex_element *velems = r300->velems->velem; 737 unsigned result = ~0; 738 739 for (i = 0; i < nr; i++) { 740 struct pipe_vertex_buffer *vb = 741 &r300->vertex_buffer[velems[i].vertex_buffer_index]; 742 unsigned size, max_count, value; 743 744 /* We're not interested in constant and per-instance attribs. */ 745 if (!vb->buffer.resource || 746 !vb->stride || 747 velems[i].instance_divisor) { 748 continue; 749 } 750 751 size = vb->buffer.resource->width0; 752 753 /* Subtract buffer_offset. */ 754 value = vb->buffer_offset; 755 if (value >= size) { 756 return 0; 757 } 758 size -= value; 759 760 /* Subtract src_offset. */ 761 value = velems[i].src_offset; 762 if (value >= size) { 763 return 0; 764 } 765 size -= value; 766 767 /* Subtract format_size. */ 768 value = r300->velems->format_size[i]; 769 if (value >= size) { 770 return 0; 771 } 772 size -= value; 773 774 /* Compute the max count. */ 775 max_count = 1 + size / vb->stride; 776 result = MIN2(result, max_count); 777 } 778 return result; 779 } 780 781 782 static void r300_draw_vbo(struct pipe_context* pipe, 783 const struct pipe_draw_info *dinfo) 784 { 785 struct r300_context* r300 = r300_context(pipe); 786 struct pipe_draw_info info = *dinfo; 787 788 if (r300->skip_rendering || 789 !u_trim_pipe_prim(info.mode, &info.count)) { 790 return; 791 } 792 793 r300_update_derived_state(r300); 794 795 /* Draw. */ 796 if (info.index_size) { 797 unsigned max_count = r300_max_vertex_count(r300); 798 799 if (!max_count) { 800 fprintf(stderr, "r300: Skipping a draw command. There is a buffer " 801 " which is too small to be used for rendering.\n"); 802 return; 803 } 804 805 if (max_count == ~0) { 806 /* There are no per-vertex vertex elements. Use the hardware maximum. */ 807 max_count = 0xffffff; 808 } 809 810 info.max_index = max_count - 1; 811 812 if (info.instance_count <= 1) { 813 if (info.count <= 8 && info.has_user_indices) { 814 r300_draw_elements_immediate(r300, &info); 815 } else { 816 r300_draw_elements(r300, &info, -1); 817 } 818 } else { 819 r300_draw_elements_instanced(r300, &info); 820 } 821 } else { 822 if (info.instance_count <= 1) { 823 if (immd_is_good_idea(r300, info.count)) { 824 r300_draw_arrays_immediate(r300, &info); 825 } else { 826 r300_draw_arrays(r300, &info, -1); 827 } 828 } else { 829 r300_draw_arrays_instanced(r300, &info); 830 } 831 } 832 } 833 834 /**************************************************************************** 835 * The rest of this file is for SW TCL rendering only. Please be polite and * 836 * keep these functions separated so that they are easier to locate. ~C. * 837 ***************************************************************************/ 838 839 /* SW TCL elements, using Draw. */ 840 static void r300_swtcl_draw_vbo(struct pipe_context* pipe, 841 const struct pipe_draw_info *info) 842 { 843 struct r300_context* r300 = r300_context(pipe); 844 845 if (r300->skip_rendering) { 846 return; 847 } 848 849 if (!u_trim_pipe_prim(info->mode, (unsigned*)&info->count)) 850 return; 851 852 if (info->index_size) { 853 draw_set_indexes(r300->draw, 854 info->has_user_indices ? 855 info->index.user : 856 r300_resource(info->index.resource)->malloced_buffer, 857 info->index_size, ~0); 858 } 859 860 r300_update_derived_state(r300); 861 862 draw_vbo(r300->draw, info); 863 draw_flush(r300->draw); 864 } 865 866 /* Object for rendering using Draw. */ 867 struct r300_render { 868 /* Parent class */ 869 struct vbuf_render base; 870 871 /* Pipe context */ 872 struct r300_context* r300; 873 874 /* Vertex information */ 875 size_t vertex_size; 876 unsigned prim; 877 unsigned hwprim; 878 879 /* VBO */ 880 size_t vbo_max_used; 881 uint8_t *vbo_ptr; 882 }; 883 884 static inline struct r300_render* 885 r300_render(struct vbuf_render* render) 886 { 887 return (struct r300_render*)render; 888 } 889 890 static const struct vertex_info* 891 r300_render_get_vertex_info(struct vbuf_render* render) 892 { 893 struct r300_render* r300render = r300_render(render); 894 struct r300_context* r300 = r300render->r300; 895 896 return &r300->vertex_info; 897 } 898 899 static boolean r300_render_allocate_vertices(struct vbuf_render* render, 900 ushort vertex_size, 901 ushort count) 902 { 903 struct r300_render* r300render = r300_render(render); 904 struct r300_context* r300 = r300render->r300; 905 struct radeon_winsys *rws = r300->rws; 906 size_t size = (size_t)vertex_size * (size_t)count; 907 908 DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size); 909 910 if (!r300->vbo || size + r300->draw_vbo_offset > r300->vbo->size) { 911 pb_reference(&r300->vbo, NULL); 912 r300->vbo = NULL; 913 r300render->vbo_ptr = NULL; 914 915 r300->vbo = rws->buffer_create(rws, 916 MAX2(R300_MAX_DRAW_VBO_SIZE, size), 917 R300_BUFFER_ALIGNMENT, 918 RADEON_DOMAIN_GTT, 0); 919 if (!r300->vbo) { 920 return FALSE; 921 } 922 r300->draw_vbo_offset = 0; 923 r300render->vbo_ptr = rws->buffer_map(r300->vbo, r300->cs, 924 PIPE_TRANSFER_WRITE); 925 } 926 927 r300render->vertex_size = vertex_size; 928 return TRUE; 929 } 930 931 static void* r300_render_map_vertices(struct vbuf_render* render) 932 { 933 struct r300_render* r300render = r300_render(render); 934 struct r300_context* r300 = r300render->r300; 935 936 DBG(r300, DBG_DRAW, "r300: render_map_vertices\n"); 937 938 assert(r300render->vbo_ptr); 939 return r300render->vbo_ptr + r300->draw_vbo_offset; 940 } 941 942 static void r300_render_unmap_vertices(struct vbuf_render* render, 943 ushort min, 944 ushort max) 945 { 946 struct r300_render* r300render = r300_render(render); 947 struct r300_context* r300 = r300render->r300; 948 949 DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n"); 950 951 r300render->vbo_max_used = MAX2(r300render->vbo_max_used, 952 r300render->vertex_size * (max + 1)); 953 } 954 955 static void r300_render_release_vertices(struct vbuf_render* render) 956 { 957 struct r300_render* r300render = r300_render(render); 958 struct r300_context* r300 = r300render->r300; 959 960 DBG(r300, DBG_DRAW, "r300: render_release_vertices\n"); 961 962 r300->draw_vbo_offset += r300render->vbo_max_used; 963 r300render->vbo_max_used = 0; 964 } 965 966 static void r300_render_set_primitive(struct vbuf_render* render, 967 enum pipe_prim_type prim) 968 { 969 struct r300_render* r300render = r300_render(render); 970 971 r300render->prim = prim; 972 r300render->hwprim = r300_translate_primitive(prim); 973 } 974 975 static void r300_render_draw_arrays(struct vbuf_render* render, 976 unsigned start, 977 unsigned count) 978 { 979 struct r300_render* r300render = r300_render(render); 980 struct r300_context* r300 = r300render->r300; 981 uint8_t* ptr; 982 unsigned i; 983 unsigned dwords = 6; 984 985 CS_LOCALS(r300); 986 (void) i; (void) ptr; 987 988 assert(start == 0); 989 assert(count < (1 << 16)); 990 991 DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count); 992 993 if (!r300_prepare_for_rendering(r300, 994 PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL, 995 NULL, dwords, 0, 0, -1)) { 996 return; 997 } 998 999 BEGIN_CS(dwords); 1000 OUT_CS_REG(R300_GA_COLOR_CONTROL, 1001 r300_provoking_vertex_fixes(r300, r300render->prim)); 1002 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1); 1003 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); 1004 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | 1005 r300render->hwprim); 1006 END_CS; 1007 } 1008 1009 static void r300_render_draw_elements(struct vbuf_render* render, 1010 const ushort* indices, 1011 uint count) 1012 { 1013 struct r300_render* r300render = r300_render(render); 1014 struct r300_context* r300 = r300render->r300; 1015 unsigned max_index = (r300->vbo->size - r300->draw_vbo_offset) / 1016 (r300render->r300->vertex_info.size * 4) - 1; 1017 struct pipe_resource *index_buffer = NULL; 1018 unsigned index_buffer_offset; 1019 1020 CS_LOCALS(r300); 1021 DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count); 1022 1023 u_upload_data(r300->uploader, 0, count * 2, 4, indices, 1024 &index_buffer_offset, &index_buffer); 1025 if (!index_buffer) { 1026 return; 1027 } 1028 1029 if (!r300_prepare_for_rendering(r300, 1030 PREP_EMIT_STATES | 1031 PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED, 1032 index_buffer, 12, 0, 0, -1)) { 1033 pipe_resource_reference(&index_buffer, NULL); 1034 return; 1035 } 1036 1037 BEGIN_CS(12); 1038 OUT_CS_REG(R300_GA_COLOR_CONTROL, 1039 r300_provoking_vertex_fixes(r300, r300render->prim)); 1040 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index); 1041 1042 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0); 1043 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) | 1044 r300render->hwprim); 1045 1046 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2); 1047 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2)); 1048 OUT_CS(index_buffer_offset); 1049 OUT_CS((count + 1) / 2); 1050 OUT_CS_RELOC(r300_resource(index_buffer)); 1051 END_CS; 1052 1053 pipe_resource_reference(&index_buffer, NULL); 1054 } 1055 1056 static void r300_render_destroy(struct vbuf_render* render) 1057 { 1058 FREE(render); 1059 } 1060 1061 static struct vbuf_render* r300_render_create(struct r300_context* r300) 1062 { 1063 struct r300_render* r300render = CALLOC_STRUCT(r300_render); 1064 1065 r300render->r300 = r300; 1066 1067 r300render->base.max_vertex_buffer_bytes = R300_MAX_DRAW_VBO_SIZE; 1068 r300render->base.max_indices = 16 * 1024; 1069 1070 r300render->base.get_vertex_info = r300_render_get_vertex_info; 1071 r300render->base.allocate_vertices = r300_render_allocate_vertices; 1072 r300render->base.map_vertices = r300_render_map_vertices; 1073 r300render->base.unmap_vertices = r300_render_unmap_vertices; 1074 r300render->base.set_primitive = r300_render_set_primitive; 1075 r300render->base.draw_elements = r300_render_draw_elements; 1076 r300render->base.draw_arrays = r300_render_draw_arrays; 1077 r300render->base.release_vertices = r300_render_release_vertices; 1078 r300render->base.destroy = r300_render_destroy; 1079 1080 return &r300render->base; 1081 } 1082 1083 struct draw_stage* r300_draw_stage(struct r300_context* r300) 1084 { 1085 struct vbuf_render* render; 1086 struct draw_stage* stage; 1087 1088 render = r300_render_create(r300); 1089 1090 if (!render) { 1091 return NULL; 1092 } 1093 1094 stage = draw_vbuf_stage(r300->draw, render); 1095 1096 if (!stage) { 1097 render->destroy(render); 1098 return NULL; 1099 } 1100 1101 draw_set_render(r300->draw, render); 1102 1103 return stage; 1104 } 1105 1106 /**************************************************************************** 1107 * End of SW TCL functions * 1108 ***************************************************************************/ 1109 1110 /* This functions is used to draw a rectangle for the blitter module. 1111 * 1112 * If we rendered a quad, the pixels on the main diagonal 1113 * would be computed and stored twice, which makes the clear/copy codepaths 1114 * somewhat inefficient. Instead we use a rectangular point sprite. */ 1115 void r300_blitter_draw_rectangle(struct blitter_context *blitter, 1116 void *vertex_elements_cso, 1117 blitter_get_vs_func get_vs, 1118 int x1, int y1, int x2, int y2, 1119 float depth, unsigned num_instances, 1120 enum blitter_attrib_type type, 1121 const union blitter_attrib *attrib) 1122 { 1123 struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter)); 1124 unsigned last_sprite_coord_enable = r300->sprite_coord_enable; 1125 unsigned width = x2 - x1; 1126 unsigned height = y2 - y1; 1127 unsigned vertex_size = 1128 type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4; 1129 unsigned dwords = 13 + vertex_size + 1130 (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY ? 7 : 0); 1131 static const union blitter_attrib zeros; 1132 CS_LOCALS(r300); 1133 1134 /* XXX workaround for a lockup in MSAA resolve on SWTCL chipsets, this 1135 * function most probably doesn't handle type=NONE correctly */ 1136 if ((!r300->screen->caps.has_tcl && type == UTIL_BLITTER_ATTRIB_NONE) || 1137 type == UTIL_BLITTER_ATTRIB_TEXCOORD_XYZW || 1138 num_instances > 1) { 1139 util_blitter_draw_rectangle(blitter, vertex_elements_cso, get_vs, 1140 x1, y1, x2, y2, 1141 depth, num_instances, type, attrib); 1142 return; 1143 } 1144 1145 if (r300->skip_rendering) 1146 return; 1147 1148 r300->context.bind_vertex_elements_state(&r300->context, vertex_elements_cso); 1149 r300->context.bind_vs_state(&r300->context, get_vs(blitter)); 1150 1151 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY) 1152 r300->sprite_coord_enable = 1; 1153 1154 r300_update_derived_state(r300); 1155 1156 /* Mark some states we don't care about as non-dirty. */ 1157 r300->viewport_state.dirty = FALSE; 1158 1159 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1)) 1160 goto done; 1161 1162 DBG(r300, DBG_DRAW, "r300: draw_rectangle\n"); 1163 1164 BEGIN_CS(dwords); 1165 /* Set up GA. */ 1166 OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16)); 1167 1168 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY) { 1169 /* Set up the GA to generate texcoords. */ 1170 OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE | 1171 (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT)); 1172 OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4); 1173 OUT_CS_32F(attrib->texcoord.x1); 1174 OUT_CS_32F(attrib->texcoord.y2); 1175 OUT_CS_32F(attrib->texcoord.x2); 1176 OUT_CS_32F(attrib->texcoord.y1); 1177 } 1178 1179 /* Set up VAP controls. */ 1180 OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE); 1181 OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT); 1182 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); 1183 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2); 1184 OUT_CS(1); 1185 OUT_CS(0); 1186 1187 /* Draw. */ 1188 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size); 1189 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) | 1190 R300_VAP_VF_CNTL__PRIM_POINTS); 1191 1192 OUT_CS_32F(x1 + width * 0.5f); 1193 OUT_CS_32F(y1 + height * 0.5f); 1194 OUT_CS_32F(depth); 1195 OUT_CS_32F(1); 1196 1197 if (vertex_size == 8) { 1198 if (!attrib) 1199 attrib = &zeros; 1200 OUT_CS_TABLE(attrib->color, 4); 1201 } 1202 END_CS; 1203 1204 done: 1205 /* Restore the state. */ 1206 r300_mark_atom_dirty(r300, &r300->rs_state); 1207 r300_mark_atom_dirty(r300, &r300->viewport_state); 1208 1209 r300->sprite_coord_enable = last_sprite_coord_enable; 1210 } 1211 1212 void r300_init_render_functions(struct r300_context *r300) 1213 { 1214 /* Set draw functions based on presence of HW TCL. */ 1215 if (r300->screen->caps.has_tcl) { 1216 r300->context.draw_vbo = r300_draw_vbo; 1217 } else { 1218 r300->context.draw_vbo = r300_swtcl_draw_vbo; 1219 } 1220 1221 /* Plug in the two-sided stencil reference value fallback if needed. */ 1222 if (!r300->screen->caps.is_r500) 1223 r300_plug_in_stencil_ref_fallback(r300); 1224 } 1225
