src/gallium/drivers/r300/r300_render.c

   1 /*
   2  * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com>
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
  22
  23 /* r300_render: Vertex and index buffer primitive emission. Contains both
  24  * HW TCL fastpath rendering, and SW TCL Draw-assisted rendering. */
  25
  26 #include "draw/draw_context.h"
  27 #include "draw/draw_vbuf.h"
  28
  29 #include "pipe/p_inlines.h"
  30
  31 #include "util/u_memory.h"
  32 #include "util/u_prim.h"
  33
  34 #include "r300_cs.h"
  35 #include "r300_context.h"
  36 #include "r300_emit.h"
  37 #include "r300_reg.h"
  38 #include "r300_render.h"
  39 #include "r300_state_derived.h"
  40
  41 /* r300_render: Vertex and index buffer primitive emission. */
  42 #define R300_MAX_VBO_SIZE  (1024 * 1024)
  43
  44 uint32_t r300_translate_primitive(unsigned prim)
  45 {
  46     switch (prim) {
  47         case PIPE_PRIM_POINTS:
  48             return R300_VAP_VF_CNTL__PRIM_POINTS;
  49         case PIPE_PRIM_LINES:
  50             return R300_VAP_VF_CNTL__PRIM_LINES;
  51         case PIPE_PRIM_LINE_LOOP:
  52             return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
  53         case PIPE_PRIM_LINE_STRIP:
  54             return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
  55         case PIPE_PRIM_TRIANGLES:
  56             return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
  57         case PIPE_PRIM_TRIANGLE_STRIP:
  58             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
  59         case PIPE_PRIM_TRIANGLE_FAN:
  60             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
  61         case PIPE_PRIM_QUADS:
  62             return R300_VAP_VF_CNTL__PRIM_QUADS;
  63         case PIPE_PRIM_QUAD_STRIP:
  64             return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
  65         case PIPE_PRIM_POLYGON:
  66             return R300_VAP_VF_CNTL__PRIM_POLYGON;
  67         default:
  68             return 0;
  69     }
  70 }
  71
  72 static boolean r300_nothing_to_draw(struct r300_context *r300)
  73 {
  74     return r300->rs_state->rs.scissor &&
  75            r300->scissor_state->scissor.empty_area;
  76 }
  77
  78 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
  79                                             unsigned mode)
  80 {
  81     uint32_t color_control = r300->rs_state->color_control;
  82
  83     /* By default (see r300_state.c:r300_create_rs_state) color_control is
  84      * initialized to provoking the first vertex.
  85      *
  86      * Triangle fans must be reduced to the second vertex, not the first, in
  87      * Gallium flatshade-first mode, as per the GL spec.
  88      * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
  89      *
  90      * Quads never provoke correctly in flatshade-first mode. The first
  91      * vertex is never considered as provoking, so only the second, third,
  92      * and fourth vertices can be selected, and both "third" and "last" modes
  93      * select the fourth vertex. This is probably due to D3D lacking quads.
  94      *
  95      * Similarly, polygons reduce to the first, not the last, vertex, when in
  96      * "last" mode, and all other modes start from the second vertex.
  97      *
  98      * ~ C.
  99      */
 100
 101     if (r300->rs_state->rs.flatshade_first) {
 102         switch (mode) {
 103             case PIPE_PRIM_TRIANGLE_FAN:
 104                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
 105                 break;
 106             case PIPE_PRIM_QUADS:
 107             case PIPE_PRIM_QUAD_STRIP:
 108             case PIPE_PRIM_POLYGON:
 109                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 110                 break;
 111             default:
 112                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
 113                 break;
 114         }
 115     } else {
 116         color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 117     }
 118
 119     return color_control;
 120 }
 121
 122 static void r300_emit_draw_arrays(struct r300_context *r300,
 123                                   unsigned mode,
 124                                   unsigned count)
 125 {
 126     CS_LOCALS(r300);
 127
 128     BEGIN_CS(8);
 129     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 130             r300_provoking_vertex_fixes(r300, mode));
 131     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
 132     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
 133     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 134     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 135            r300_translate_primitive(mode));
 136     END_CS;
 137 }
 138
 139 static void r300_emit_draw_elements(struct r300_context *r300,
 140                                     struct pipe_buffer* indexBuffer,
 141                                     unsigned indexSize,
 142                                     unsigned minIndex,
 143                                     unsigned maxIndex,
 144                                     unsigned mode,
 145                                     unsigned start,
 146                                     unsigned count)
 147 {
 148     uint32_t count_dwords;
 149     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
 150     CS_LOCALS(r300);
 151
 152     /* XXX most of these are stupid */
 153     assert(indexSize == 4 || indexSize == 2);
 154     assert((start * indexSize)  % 4 == 0);
 155     assert(offset_dwords == 0);
 156
 157     BEGIN_CS(14);
 158     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 159             r300_provoking_vertex_fixes(r300, mode));
 160     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex);
 161     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
 162     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 163     if (indexSize == 4) {
 164         count_dwords = count + start;
 165         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 166                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 167                r300_translate_primitive(mode));
 168     } else {
 169         count_dwords = (count + start + 1) / 2;
 170         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 171                r300_translate_primitive(mode));
 172     }
 173
 174     /* INDX_BUFFER is a truly special packet3.
 175      * Unlike most other packet3, where the offset is after the count,
 176      * the order is reversed, so the relocation ends up carrying the
 177      * size of the indexbuf instead of the offset.
 178      *
 179      * XXX Fix offset
 180      */
 181     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 182     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 183            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 184     OUT_CS(offset_dwords);
 185     OUT_CS_RELOC(indexBuffer, count_dwords,
 186         RADEON_GEM_DOMAIN_GTT, 0, 0);
 187
 188     END_CS;
 189 }
 190
 191
 192 static boolean r300_setup_vertex_buffers(struct r300_context *r300)
 193 {
 194     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
 195     struct pipe_vertex_element *velem = r300->vertex_element;
 196
 197 validate:
 198     for (int i = 0; i < r300->vertex_element_count; i++) {
 199         if (!r300->winsys->add_buffer(r300->winsys,
 200                 vbuf[velem[i].vertex_buffer_index].buffer,
 201             RADEON_GEM_DOMAIN_GTT, 0)) {
 202             r300->context.flush(&r300->context, 0, NULL);
 203             goto validate;
 204         }
 205     }
 206
 207     if (!r300->winsys->validate(r300->winsys)) {
 208         r300->context.flush(&r300->context, 0, NULL);
 209         return r300->winsys->validate(r300->winsys);
 210     }
 211
 212     return TRUE;
 213 }
 214
 215 /* This is the fast-path drawing & emission for HW TCL. */
 216 void r300_draw_range_elements(struct pipe_context* pipe,
 217                                  struct pipe_buffer* indexBuffer,
 218                                  unsigned indexSize,
 219                                  unsigned minIndex,
 220                                  unsigned maxIndex,
 221                                  unsigned mode,
 222                                  unsigned start,
 223                                  unsigned count)
 224 {
 225     struct r300_context* r300 = r300_context(pipe);
 226
 227     if (!u_trim_pipe_prim(mode, &count)) {
 228         return;
 229     }
 230
 231     if (count > 65535) {
 232        /* XXX: use aux/indices functions to split this into smaller
 233         * primitives.
 234         */
 235         return;
 236     }
 237
 238     if (r300_nothing_to_draw(r300)) {
 239         return;
 240     }
 241
 242     r300_update_derived_state(r300);
 243
 244     if (!r300_setup_vertex_buffers(r300)) {
 245         return;
 246     }
 247
 248     if (!r300->winsys->add_buffer(r300->winsys, indexBuffer,
 249                                   RADEON_GEM_DOMAIN_GTT, 0)) {
 250         return;
 251     }
 252
 253     if (!r300->winsys->validate(r300->winsys)) {
 254         return;
 255     }
 256
 257     r300_emit_dirty_state(r300);
 258
 259     r300_emit_aos(r300, 0);
 260
 261     r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
 262                             mode, start, count);
 263 }
 264
 265 /* Simple helpers for context setup. Should probably be moved to util. */
 266 void r300_draw_elements(struct pipe_context* pipe,
 267                         struct pipe_buffer* indexBuffer,
 268                         unsigned indexSize, unsigned mode,
 269                         unsigned start, unsigned count)
 270 {
 271    pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
 272                              mode, start, count);
 273 }
 274
 275 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 276                          unsigned start, unsigned count)
 277 {
 278     struct r300_context* r300 = r300_context(pipe);
 279
 280     if (!u_trim_pipe_prim(mode, &count)) {
 281         return;
 282     }
 283
 284     if (count > 65535) {
 285         /* XXX: driver needs to handle this -- use the functions in
 286          * aux/indices to split this into several smaller primitives.
 287          */
 288         return;
 289     }
 290
 291     if (r300_nothing_to_draw(r300)) {
 292         return;
 293     }
 294
 295     r300_update_derived_state(r300);
 296
 297     if (!r300_setup_vertex_buffers(r300)) {
 298         return;
 299     }
 300
 301     r300_emit_dirty_state(r300);
 302
 303     r300_emit_aos(r300, start);
 304
 305     r300_emit_draw_arrays(r300, mode, count);
 306 }
 307
 308 /****************************************************************************
 309  * The rest of this file is for SW TCL rendering only. Please be polite and *
 310  * keep these functions separated so that they are easier to locate. ~C.    *
 311  ***************************************************************************/
 312
 313 /* SW TCL arrays, using Draw. */
 314 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 315                                unsigned mode,
 316                                unsigned start,
 317                                unsigned count)
 318 {
 319     struct r300_context* r300 = r300_context(pipe);
 320     int i;
 321
 322     if (!u_trim_pipe_prim(mode, &count)) {
 323         return;
 324     }
 325
 326     if (r300_nothing_to_draw(r300)) {
 327         return;
 328     }
 329
 330     for (i = 0; i < r300->vertex_buffer_count; i++) {
 331         void* buf = pipe_buffer_map(pipe->screen,
 332                                     r300->vertex_buffer[i].buffer,
 333                                     PIPE_BUFFER_USAGE_CPU_READ);
 334         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 335     }
 336
 337     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 338
 339     draw_set_mapped_constant_buffer(r300->draw,
 340                                     PIPE_SHADER_VERTEX,
 341                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 342                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 343                 (sizeof(float) * 4));
 344
 345     draw_arrays(r300->draw, mode, start, count);
 346
 347     for (i = 0; i < r300->vertex_buffer_count; i++) {
 348         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 349         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 350     }
 351 }
 352
 353 /* SW TCL elements, using Draw. */
 354 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 355                                        struct pipe_buffer* indexBuffer,
 356                                        unsigned indexSize,
 357                                        unsigned minIndex,
 358                                        unsigned maxIndex,
 359                                        unsigned mode,
 360                                        unsigned start,
 361                                        unsigned count)
 362 {
 363     struct r300_context* r300 = r300_context(pipe);
 364     int i;
 365     void* indices;
 366
 367     if (!u_trim_pipe_prim(mode, &count)) {
 368         return;
 369     }
 370
 371     if (r300_nothing_to_draw(r300)) {
 372         return;
 373     }
 374
 375     for (i = 0; i < r300->vertex_buffer_count; i++) {
 376         void* buf = pipe_buffer_map(pipe->screen,
 377                                     r300->vertex_buffer[i].buffer,
 378                                     PIPE_BUFFER_USAGE_CPU_READ);
 379         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 380     }
 381
 382     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 383                               PIPE_BUFFER_USAGE_CPU_READ);
 384     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 385                                          minIndex, maxIndex, indices);
 386
 387     draw_set_mapped_constant_buffer(r300->draw,
 388                                     PIPE_SHADER_VERTEX,
 389             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 390             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 391                 (sizeof(float) * 4));
 392
 393     draw_arrays(r300->draw, mode, start, count);
 394
 395     for (i = 0; i < r300->vertex_buffer_count; i++) {
 396         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 397         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 398     }
 399
 400     pipe_buffer_unmap(pipe->screen, indexBuffer);
 401     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 402                                          start + count - 1, NULL);
 403 }
 404
 405 /* Object for rendering using Draw. */
 406 struct r300_render {
 407     /* Parent class */
 408     struct vbuf_render base;
 409
 410     /* Pipe context */
 411     struct r300_context* r300;
 412
 413     /* Vertex information */
 414     size_t vertex_size;
 415     unsigned prim;
 416     unsigned hwprim;
 417
 418     /* VBO */
 419     struct pipe_buffer* vbo;
 420     size_t vbo_size;
 421     size_t vbo_offset;
 422     size_t vbo_max_used;
 423     void * vbo_ptr;
 424 };
 425
 426 static INLINE struct r300_render*
 427 r300_render(struct vbuf_render* render)
 428 {
 429     return (struct r300_render*)render;
 430 }
 431
 432 static const struct vertex_info*
 433 r300_render_get_vertex_info(struct vbuf_render* render)
 434 {
 435     struct r300_render* r300render = r300_render(render);
 436     struct r300_context* r300 = r300render->r300;
 437
 438     r300_update_derived_state(r300);
 439
 440     return &r300->vertex_info->vinfo;
 441 }
 442
 443 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 444                                                    ushort vertex_size,
 445                                                    ushort count)
 446 {
 447     struct r300_render* r300render = r300_render(render);
 448     struct r300_context* r300 = r300render->r300;
 449     struct pipe_screen* screen = r300->context.screen;
 450     size_t size = (size_t)vertex_size * (size_t)count;
 451
 452     if (size + r300render->vbo_offset > r300render->vbo_size)
 453     {
 454         pipe_buffer_reference(&r300->vbo, NULL);
 455         r300render->vbo = pipe_buffer_create(screen,
 456                                              64,
 457                                              PIPE_BUFFER_USAGE_VERTEX,
 458                                              R300_MAX_VBO_SIZE);
 459         r300render->vbo_offset = 0;
 460         r300render->vbo_size = R300_MAX_VBO_SIZE;
 461     }
 462
 463     r300render->vertex_size = vertex_size;
 464     r300->vbo = r300render->vbo;
 465     r300->vbo_offset = r300render->vbo_offset;
 466
 467     return (r300render->vbo) ? TRUE : FALSE;
 468 }
 469
 470 static void* r300_render_map_vertices(struct vbuf_render* render)
 471 {
 472     struct r300_render* r300render = r300_render(render);
 473     struct pipe_screen* screen = r300render->r300->context.screen;
 474
 475     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 476                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 477
 478     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 479 }
 480
 481 static void r300_render_unmap_vertices(struct vbuf_render* render,
 482                                              ushort min,
 483                                              ushort max)
 484 {
 485     struct r300_render* r300render = r300_render(render);
 486     struct pipe_screen* screen = r300render->r300->context.screen;
 487     CS_LOCALS(r300render->r300);
 488     BEGIN_CS(2);
 489     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 490     END_CS;
 491
 492     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 493                                     r300render->vertex_size * (max + 1));
 494     pipe_buffer_unmap(screen, r300render->vbo);
 495 }
 496
 497 static void r300_render_release_vertices(struct vbuf_render* render)
 498 {
 499     struct r300_render* r300render = r300_render(render);
 500
 501     r300render->vbo_offset += r300render->vbo_max_used;
 502     r300render->vbo_max_used = 0;
 503 }
 504
 505 static boolean r300_render_set_primitive(struct vbuf_render* render,
 506                                                unsigned prim)
 507 {
 508     struct r300_render* r300render = r300_render(render);
 509
 510     r300render->prim = prim;
 511     r300render->hwprim = r300_translate_primitive(prim);
 512
 513     return TRUE;
 514 }
 515
 516 static void r300_render_draw_arrays(struct vbuf_render* render,
 517                                           unsigned start,
 518                                           unsigned count)
 519 {
 520     struct r300_render* r300render = r300_render(render);
 521     struct r300_context* r300 = r300render->r300;
 522
 523     CS_LOCALS(r300);
 524
 525     r300_emit_dirty_state(r300);
 526
 527     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 528
 529     BEGIN_CS(2);
 530     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 531     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 532            r300render->hwprim);
 533     END_CS;
 534 }
 535
 536 static void r300_render_draw(struct vbuf_render* render,
 537                                    const ushort* indices,
 538                                    uint count)
 539 {
 540     struct r300_render* r300render = r300_render(render);
 541     struct r300_context* r300 = r300render->r300;
 542     int i;
 543
 544     CS_LOCALS(r300);
 545
 546     r300_emit_dirty_state(r300);
 547
 548     BEGIN_CS(2 + (count+1)/2);
 549     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 550     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 551            r300render->hwprim);
 552     for (i = 0; i < count-1; i += 2) {
 553         OUT_CS(indices[i+1] << 16 | indices[i]);
 554     }
 555     if (count % 2) {
 556         OUT_CS(indices[count-1]);
 557     }
 558     END_CS;
 559 }
 560
 561 static void r300_render_destroy(struct vbuf_render* render)
 562 {
 563     FREE(render);
 564 }
 565
 566 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 567 {
 568     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 569
 570     r300render->r300 = r300;
 571
 572     /* XXX find real numbers plz */
 573     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 574     r300render->base.max_indices = 16 * 1024;
 575
 576     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 577     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 578     r300render->base.map_vertices = r300_render_map_vertices;
 579     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 580     r300render->base.set_primitive = r300_render_set_primitive;
 581     r300render->base.draw = r300_render_draw;
 582     r300render->base.draw_arrays = r300_render_draw_arrays;
 583     r300render->base.release_vertices = r300_render_release_vertices;
 584     r300render->base.destroy = r300_render_destroy;
 585
 586     r300render->vbo = NULL;
 587     r300render->vbo_size = 0;
 588     r300render->vbo_offset = 0;
 589
 590     return &r300render->base;
 591 }
 592
 593 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 594 {
 595     struct vbuf_render* render;
 596     struct draw_stage* stage;
 597
 598     render = r300_render_create(r300);
 599
 600     if (!render) {
 601         return NULL;
 602     }
 603
 604     stage = draw_vbuf_stage(r300->draw, render);
 605
 606     if (!stage) {
 607         render->destroy(render);
 608         return NULL;
 609     }
 610
 611     draw_set_render(r300->draw, render);
 612
 613     return stage;
 614 }