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 uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
  73                                             unsigned mode)
  74 {
  75     struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
  76     uint32_t color_control = rs->color_control;
  77
  78     /* By default (see r300_state.c:r300_create_rs_state) color_control is
  79      * initialized to provoking the first vertex.
  80      *
  81      * Triangle fans must be reduced to the second vertex, not the first, in
  82      * Gallium flatshade-first mode, as per the GL spec.
  83      * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
  84      *
  85      * Quads never provoke correctly in flatshade-first mode. The first
  86      * vertex is never considered as provoking, so only the second, third,
  87      * and fourth vertices can be selected, and both "third" and "last" modes
  88      * select the fourth vertex. This is probably due to D3D lacking quads.
  89      *
  90      * Similarly, polygons reduce to the first, not the last, vertex, when in
  91      * "last" mode, and all other modes start from the second vertex.
  92      *
  93      * ~ C.
  94      */
  95
  96     if (rs->rs.flatshade_first) {
  97         switch (mode) {
  98             case PIPE_PRIM_TRIANGLE_FAN:
  99                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
 100                 break;
 101             case PIPE_PRIM_QUADS:
 102             case PIPE_PRIM_QUAD_STRIP:
 103             case PIPE_PRIM_POLYGON:
 104                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 105                 break;
 106             default:
 107                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
 108                 break;
 109         }
 110     } else {
 111         color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 112     }
 113
 114     return color_control;
 115 }
 116
 117 static void r300_emit_draw_arrays(struct r300_context *r300,
 118                                   unsigned mode,
 119                                   unsigned count)
 120 {
 121     CS_LOCALS(r300);
 122
 123     BEGIN_CS(8);
 124     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 125             r300_provoking_vertex_fixes(r300, mode));
 126     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
 127     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
 128     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 129     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 130            r300_translate_primitive(mode));
 131     END_CS;
 132 }
 133
 134 static void r300_emit_draw_elements(struct r300_context *r300,
 135                                     struct pipe_buffer* indexBuffer,
 136                                     unsigned indexSize,
 137                                     unsigned minIndex,
 138                                     unsigned maxIndex,
 139                                     unsigned mode,
 140                                     unsigned start,
 141                                     unsigned count)
 142 {
 143     uint32_t count_dwords;
 144     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
 145     CS_LOCALS(r300);
 146
 147     /* XXX most of these are stupid */
 148     assert(indexSize == 4 || indexSize == 2);
 149     assert((start * indexSize)  % 4 == 0);
 150     assert(offset_dwords == 0);
 151
 152     BEGIN_CS(14);
 153     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 154             r300_provoking_vertex_fixes(r300, mode));
 155     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex);
 156     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
 157     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 158     if (indexSize == 4) {
 159         count_dwords = count + start;
 160         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 161                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 162                r300_translate_primitive(mode));
 163     } else {
 164         count_dwords = (count + start + 1) / 2;
 165         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 166                r300_translate_primitive(mode));
 167     }
 168
 169     /* INDX_BUFFER is a truly special packet3.
 170      * Unlike most other packet3, where the offset is after the count,
 171      * the order is reversed, so the relocation ends up carrying the
 172      * size of the indexbuf instead of the offset.
 173      *
 174      * XXX Fix offset
 175      */
 176     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 177     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 178            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 179     OUT_CS(offset_dwords);
 180     OUT_CS_RELOC(indexBuffer, count_dwords,
 181         RADEON_GEM_DOMAIN_GTT, 0, 0);
 182
 183     END_CS;
 184 }
 185
 186
 187 static boolean r300_setup_vertex_buffers(struct r300_context *r300)
 188 {
 189     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
 190     struct pipe_vertex_element *velem = r300->vertex_element;
 191
 192 validate:
 193     for (int i = 0; i < r300->vertex_element_count; i++) {
 194         if (!r300->winsys->add_buffer(r300->winsys,
 195                 vbuf[velem[i].vertex_buffer_index].buffer,
 196             RADEON_GEM_DOMAIN_GTT, 0)) {
 197             r300->context.flush(&r300->context, 0, NULL);
 198             goto validate;
 199         }
 200     }
 201
 202     if (!r300->winsys->validate(r300->winsys)) {
 203         r300->context.flush(&r300->context, 0, NULL);
 204         return r300->winsys->validate(r300->winsys);
 205     }
 206
 207     return TRUE;
 208 }
 209
 210 /* This is the fast-path drawing & emission for HW TCL. */
 211 void r300_draw_range_elements(struct pipe_context* pipe,
 212                                  struct pipe_buffer* indexBuffer,
 213                                  unsigned indexSize,
 214                                  unsigned minIndex,
 215                                  unsigned maxIndex,
 216                                  unsigned mode,
 217                                  unsigned start,
 218                                  unsigned count)
 219 {
 220     struct r300_context* r300 = r300_context(pipe);
 221
 222     if (!u_trim_pipe_prim(mode, &count)) {
 223         return;
 224     }
 225
 226     if (count > 65535) {
 227        /* XXX: use aux/indices functions to split this into smaller
 228         * primitives.
 229         */
 230         return;
 231     }
 232
 233     r300_update_derived_state(r300);
 234
 235     if (!r300_setup_vertex_buffers(r300)) {
 236         return;
 237     }
 238
 239     if (!r300->winsys->add_buffer(r300->winsys, indexBuffer,
 240                                   RADEON_GEM_DOMAIN_GTT, 0)) {
 241         return;
 242     }
 243
 244     if (!r300->winsys->validate(r300->winsys)) {
 245         return;
 246     }
 247
 248     r300_emit_dirty_state(r300);
 249
 250     r300_emit_aos(r300, 0);
 251
 252     r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
 253                             mode, start, count);
 254 }
 255
 256 /* Simple helpers for context setup. Should probably be moved to util. */
 257 void r300_draw_elements(struct pipe_context* pipe,
 258                         struct pipe_buffer* indexBuffer,
 259                         unsigned indexSize, unsigned mode,
 260                         unsigned start, unsigned count)
 261 {
 262    pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
 263                              mode, start, count);
 264 }
 265
 266 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 267                          unsigned start, unsigned count)
 268 {
 269     struct r300_context* r300 = r300_context(pipe);
 270
 271     if (!u_trim_pipe_prim(mode, &count)) {
 272         return;
 273     }
 274
 275     if (count > 65535) {
 276         /* XXX: driver needs to handle this -- use the functions in
 277          * aux/indices to split this into several smaller primitives.
 278          */
 279         return;
 280     }
 281
 282     r300_update_derived_state(r300);
 283
 284     if (!r300_setup_vertex_buffers(r300)) {
 285         return;
 286     }
 287
 288     r300_emit_dirty_state(r300);
 289
 290     r300_emit_aos(r300, start);
 291
 292     r300_emit_draw_arrays(r300, mode, count);
 293 }
 294
 295 /****************************************************************************
 296  * The rest of this file is for SW TCL rendering only. Please be polite and *
 297  * keep these functions separated so that they are easier to locate. ~C.    *
 298  ***************************************************************************/
 299
 300 /* SW TCL arrays, using Draw. */
 301 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 302                                unsigned mode,
 303                                unsigned start,
 304                                unsigned count)
 305 {
 306     struct r300_context* r300 = r300_context(pipe);
 307     int i;
 308
 309     if (!u_trim_pipe_prim(mode, &count)) {
 310         return;
 311     }
 312
 313     for (i = 0; i < r300->vertex_buffer_count; i++) {
 314         void* buf = pipe_buffer_map(pipe->screen,
 315                                     r300->vertex_buffer[i].buffer,
 316                                     PIPE_BUFFER_USAGE_CPU_READ);
 317         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 318     }
 319
 320     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 321
 322     draw_set_mapped_constant_buffer(r300->draw,
 323                                     PIPE_SHADER_VERTEX,
 324                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 325                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 326                 (sizeof(float) * 4));
 327
 328     draw_arrays(r300->draw, mode, start, count);
 329
 330     for (i = 0; i < r300->vertex_buffer_count; i++) {
 331         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 332         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 333     }
 334 }
 335
 336 /* SW TCL elements, using Draw. */
 337 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 338                                        struct pipe_buffer* indexBuffer,
 339                                        unsigned indexSize,
 340                                        unsigned minIndex,
 341                                        unsigned maxIndex,
 342                                        unsigned mode,
 343                                        unsigned start,
 344                                        unsigned count)
 345 {
 346     struct r300_context* r300 = r300_context(pipe);
 347     int i;
 348     void* indices;
 349
 350     if (!u_trim_pipe_prim(mode, &count)) {
 351         return;
 352     }
 353
 354     for (i = 0; i < r300->vertex_buffer_count; i++) {
 355         void* buf = pipe_buffer_map(pipe->screen,
 356                                     r300->vertex_buffer[i].buffer,
 357                                     PIPE_BUFFER_USAGE_CPU_READ);
 358         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 359     }
 360
 361     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 362                               PIPE_BUFFER_USAGE_CPU_READ);
 363     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 364                                          minIndex, maxIndex, indices);
 365
 366     draw_set_mapped_constant_buffer(r300->draw,
 367                                     PIPE_SHADER_VERTEX,
 368             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 369             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 370                 (sizeof(float) * 4));
 371
 372     draw_arrays(r300->draw, mode, start, count);
 373
 374     for (i = 0; i < r300->vertex_buffer_count; i++) {
 375         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 376         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 377     }
 378
 379     pipe_buffer_unmap(pipe->screen, indexBuffer);
 380     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 381                                          start + count - 1, NULL);
 382 }
 383
 384 /* Object for rendering using Draw. */
 385 struct r300_render {
 386     /* Parent class */
 387     struct vbuf_render base;
 388
 389     /* Pipe context */
 390     struct r300_context* r300;
 391
 392     /* Vertex information */
 393     size_t vertex_size;
 394     unsigned prim;
 395     unsigned hwprim;
 396
 397     /* VBO */
 398     struct pipe_buffer* vbo;
 399     size_t vbo_size;
 400     size_t vbo_offset;
 401     size_t vbo_max_used;
 402     void * vbo_ptr;
 403 };
 404
 405 static INLINE struct r300_render*
 406 r300_render(struct vbuf_render* render)
 407 {
 408     return (struct r300_render*)render;
 409 }
 410
 411 static const struct vertex_info*
 412 r300_render_get_vertex_info(struct vbuf_render* render)
 413 {
 414     struct r300_render* r300render = r300_render(render);
 415     struct r300_context* r300 = r300render->r300;
 416
 417     r300_update_derived_state(r300);
 418
 419     return &r300->vertex_info->vinfo;
 420 }
 421
 422 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 423                                                    ushort vertex_size,
 424                                                    ushort count)
 425 {
 426     struct r300_render* r300render = r300_render(render);
 427     struct r300_context* r300 = r300render->r300;
 428     struct pipe_screen* screen = r300->context.screen;
 429     size_t size = (size_t)vertex_size * (size_t)count;
 430
 431     if (size + r300render->vbo_offset > r300render->vbo_size)
 432     {
 433         pipe_buffer_reference(&r300->vbo, NULL);
 434         r300render->vbo = pipe_buffer_create(screen,
 435                                              64,
 436                                              PIPE_BUFFER_USAGE_VERTEX,
 437                                              R300_MAX_VBO_SIZE);
 438         r300render->vbo_offset = 0;
 439         r300render->vbo_size = R300_MAX_VBO_SIZE;
 440     }
 441
 442     r300render->vertex_size = vertex_size;
 443     r300->vbo = r300render->vbo;
 444     r300->vbo_offset = r300render->vbo_offset;
 445
 446     return (r300render->vbo) ? TRUE : FALSE;
 447 }
 448
 449 static void* r300_render_map_vertices(struct vbuf_render* render)
 450 {
 451     struct r300_render* r300render = r300_render(render);
 452     struct pipe_screen* screen = r300render->r300->context.screen;
 453
 454     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 455                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 456
 457     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 458 }
 459
 460 static void r300_render_unmap_vertices(struct vbuf_render* render,
 461                                              ushort min,
 462                                              ushort max)
 463 {
 464     struct r300_render* r300render = r300_render(render);
 465     struct pipe_screen* screen = r300render->r300->context.screen;
 466     CS_LOCALS(r300render->r300);
 467     BEGIN_CS(2);
 468     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 469     END_CS;
 470
 471     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 472                                     r300render->vertex_size * (max + 1));
 473     pipe_buffer_unmap(screen, r300render->vbo);
 474 }
 475
 476 static void r300_render_release_vertices(struct vbuf_render* render)
 477 {
 478     struct r300_render* r300render = r300_render(render);
 479
 480     r300render->vbo_offset += r300render->vbo_max_used;
 481     r300render->vbo_max_used = 0;
 482 }
 483
 484 static boolean r300_render_set_primitive(struct vbuf_render* render,
 485                                                unsigned prim)
 486 {
 487     struct r300_render* r300render = r300_render(render);
 488
 489     r300render->prim = prim;
 490     r300render->hwprim = r300_translate_primitive(prim);
 491
 492     return TRUE;
 493 }
 494
 495 static void r300_render_draw_arrays(struct vbuf_render* render,
 496                                           unsigned start,
 497                                           unsigned count)
 498 {
 499     struct r300_render* r300render = r300_render(render);
 500     struct r300_context* r300 = r300render->r300;
 501
 502     CS_LOCALS(r300);
 503
 504     r300_emit_dirty_state(r300);
 505
 506     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 507
 508     BEGIN_CS(2);
 509     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 510     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 511            r300render->hwprim);
 512     END_CS;
 513 }
 514
 515 static void r300_render_draw(struct vbuf_render* render,
 516                                    const ushort* indices,
 517                                    uint count)
 518 {
 519     struct r300_render* r300render = r300_render(render);
 520     struct r300_context* r300 = r300render->r300;
 521     int i;
 522
 523     CS_LOCALS(r300);
 524
 525     r300_emit_dirty_state(r300);
 526
 527     BEGIN_CS(2 + (count+1)/2);
 528     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 529     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 530            r300render->hwprim);
 531     for (i = 0; i < count-1; i += 2) {
 532         OUT_CS(indices[i+1] << 16 | indices[i]);
 533     }
 534     if (count % 2) {
 535         OUT_CS(indices[count-1]);
 536     }
 537     END_CS;
 538 }
 539
 540 static void r300_render_destroy(struct vbuf_render* render)
 541 {
 542     FREE(render);
 543 }
 544
 545 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 546 {
 547     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 548
 549     r300render->r300 = r300;
 550
 551     /* XXX find real numbers plz */
 552     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 553     r300render->base.max_indices = 16 * 1024;
 554
 555     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 556     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 557     r300render->base.map_vertices = r300_render_map_vertices;
 558     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 559     r300render->base.set_primitive = r300_render_set_primitive;
 560     r300render->base.draw = r300_render_draw;
 561     r300render->base.draw_arrays = r300_render_draw_arrays;
 562     r300render->base.release_vertices = r300_render_release_vertices;
 563     r300render->base.destroy = r300_render_destroy;
 564
 565     r300render->vbo = NULL;
 566     r300render->vbo_size = 0;
 567     r300render->vbo_offset = 0;
 568
 569     return &r300render->base;
 570 }
 571
 572 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 573 {
 574     struct vbuf_render* render;
 575     struct draw_stage* stage;
 576
 577     render = r300_render_create(r300);
 578
 579     if (!render) {
 580         return NULL;
 581     }
 582
 583     stage = draw_vbuf_stage(r300->draw, render);
 584
 585     if (!stage) {
 586         render->destroy(render);
 587         return NULL;
 588     }
 589
 590     draw_set_render(r300->draw, render);
 591
 592     return stage;
 593 }