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 ((struct r300_rs_state*)r300->rs_state.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     struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
  82     uint32_t color_control = rs->color_control;
  83
  84     /* By default (see r300_state.c:r300_create_rs_state) color_control is
  85      * initialized to provoking the first vertex.
  86      *
  87      * Triangle fans must be reduced to the second vertex, not the first, in
  88      * Gallium flatshade-first mode, as per the GL spec.
  89      * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
  90      *
  91      * Quads never provoke correctly in flatshade-first mode. The first
  92      * vertex is never considered as provoking, so only the second, third,
  93      * and fourth vertices can be selected, and both "third" and "last" modes
  94      * select the fourth vertex. This is probably due to D3D lacking quads.
  95      *
  96      * Similarly, polygons reduce to the first, not the last, vertex, when in
  97      * "last" mode, and all other modes start from the second vertex.
  98      *
  99      * ~ C.
 100      */
 101
 102     if (rs->rs.flatshade_first) {
 103         switch (mode) {
 104             case PIPE_PRIM_TRIANGLE_FAN:
 105                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
 106                 break;
 107             case PIPE_PRIM_QUADS:
 108             case PIPE_PRIM_QUAD_STRIP:
 109             case PIPE_PRIM_POLYGON:
 110                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 111                 break;
 112             default:
 113                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
 114                 break;
 115         }
 116     } else {
 117         color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 118     }
 119
 120     return color_control;
 121 }
 122
 123 static void r300_emit_draw_arrays(struct r300_context *r300,
 124                                   unsigned mode,
 125                                   unsigned count)
 126 {
 127     CS_LOCALS(r300);
 128
 129     BEGIN_CS(8);
 130     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 131             r300_provoking_vertex_fixes(r300, mode));
 132     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
 133     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
 134     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 135     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 136            r300_translate_primitive(mode));
 137     END_CS;
 138 }
 139
 140 static void r300_emit_draw_elements(struct r300_context *r300,
 141                                     struct pipe_buffer* indexBuffer,
 142                                     unsigned indexSize,
 143                                     unsigned minIndex,
 144                                     unsigned maxIndex,
 145                                     unsigned mode,
 146                                     unsigned start,
 147                                     unsigned count)
 148 {
 149     uint32_t count_dwords;
 150     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
 151     CS_LOCALS(r300);
 152
 153     /* XXX most of these are stupid */
 154     assert(indexSize == 4 || indexSize == 2);
 155     assert((start * indexSize)  % 4 == 0);
 156     assert(offset_dwords == 0);
 157
 158     BEGIN_CS(14);
 159     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 160             r300_provoking_vertex_fixes(r300, mode));
 161     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex);
 162     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
 163     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 164     if (indexSize == 4) {
 165         count_dwords = count + start;
 166         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 167                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 168                r300_translate_primitive(mode));
 169     } else {
 170         count_dwords = (count + start + 1) / 2;
 171         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 172                r300_translate_primitive(mode));
 173     }
 174
 175     /* INDX_BUFFER is a truly special packet3.
 176      * Unlike most other packet3, where the offset is after the count,
 177      * the order is reversed, so the relocation ends up carrying the
 178      * size of the indexbuf instead of the offset.
 179      *
 180      * XXX Fix offset
 181      */
 182     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 183     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 184            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 185     OUT_CS(offset_dwords);
 186     OUT_CS_RELOC(indexBuffer, count_dwords,
 187         RADEON_GEM_DOMAIN_GTT, 0, 0);
 188
 189     END_CS;
 190 }
 191
 192
 193 static boolean r300_setup_vertex_buffers(struct r300_context *r300)
 194 {
 195     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
 196     struct pipe_vertex_element *velem = r300->vertex_element;
 197
 198 validate:
 199     for (int i = 0; i < r300->vertex_element_count; i++) {
 200         if (!r300->winsys->add_buffer(r300->winsys,
 201                 vbuf[velem[i].vertex_buffer_index].buffer,
 202             RADEON_GEM_DOMAIN_GTT, 0)) {
 203             r300->context.flush(&r300->context, 0, NULL);
 204             goto validate;
 205         }
 206     }
 207
 208     if (!r300->winsys->validate(r300->winsys)) {
 209         r300->context.flush(&r300->context, 0, NULL);
 210         return r300->winsys->validate(r300->winsys);
 211     }
 212
 213     return TRUE;
 214 }
 215
 216 /* This is the fast-path drawing & emission for HW TCL. */
 217 void r300_draw_range_elements(struct pipe_context* pipe,
 218                                  struct pipe_buffer* indexBuffer,
 219                                  unsigned indexSize,
 220                                  unsigned minIndex,
 221                                  unsigned maxIndex,
 222                                  unsigned mode,
 223                                  unsigned start,
 224                                  unsigned count)
 225 {
 226     struct r300_context* r300 = r300_context(pipe);
 227
 228     if (!u_trim_pipe_prim(mode, &count)) {
 229         return;
 230     }
 231
 232     if (count > 65535) {
 233        /* XXX: use aux/indices functions to split this into smaller
 234         * primitives.
 235         */
 236         return;
 237     }
 238
 239     if (r300_nothing_to_draw(r300)) {
 240         return;
 241     }
 242
 243     r300_update_derived_state(r300);
 244
 245     if (!r300_setup_vertex_buffers(r300)) {
 246         return;
 247     }
 248
 249     if (!r300->winsys->add_buffer(r300->winsys, indexBuffer,
 250                                   RADEON_GEM_DOMAIN_GTT, 0)) {
 251         return;
 252     }
 253
 254     if (!r300->winsys->validate(r300->winsys)) {
 255         return;
 256     }
 257
 258     r300_emit_dirty_state(r300);
 259
 260     r300_emit_aos(r300, 0);
 261
 262     r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
 263                             mode, start, count);
 264 }
 265
 266 /* Simple helpers for context setup. Should probably be moved to util. */
 267 void r300_draw_elements(struct pipe_context* pipe,
 268                         struct pipe_buffer* indexBuffer,
 269                         unsigned indexSize, unsigned mode,
 270                         unsigned start, unsigned count)
 271 {
 272    pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
 273                              mode, start, count);
 274 }
 275
 276 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 277                          unsigned start, unsigned count)
 278 {
 279     struct r300_context* r300 = r300_context(pipe);
 280
 281     if (!u_trim_pipe_prim(mode, &count)) {
 282         return;
 283     }
 284
 285     if (count > 65535) {
 286         /* XXX: driver needs to handle this -- use the functions in
 287          * aux/indices to split this into several smaller primitives.
 288          */
 289         return;
 290     }
 291
 292     if (r300_nothing_to_draw(r300)) {
 293         return;
 294     }
 295
 296     r300_update_derived_state(r300);
 297
 298     if (!r300_setup_vertex_buffers(r300)) {
 299         return;
 300     }
 301
 302     r300_emit_dirty_state(r300);
 303
 304     r300_emit_aos(r300, start);
 305
 306     r300_emit_draw_arrays(r300, mode, count);
 307 }
 308
 309 /****************************************************************************
 310  * The rest of this file is for SW TCL rendering only. Please be polite and *
 311  * keep these functions separated so that they are easier to locate. ~C.    *
 312  ***************************************************************************/
 313
 314 /* SW TCL arrays, using Draw. */
 315 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 316                                unsigned mode,
 317                                unsigned start,
 318                                unsigned count)
 319 {
 320     struct r300_context* r300 = r300_context(pipe);
 321     int i;
 322
 323     if (!u_trim_pipe_prim(mode, &count)) {
 324         return;
 325     }
 326
 327     if (r300_nothing_to_draw(r300)) {
 328         return;
 329     }
 330
 331     for (i = 0; i < r300->vertex_buffer_count; i++) {
 332         void* buf = pipe_buffer_map(pipe->screen,
 333                                     r300->vertex_buffer[i].buffer,
 334                                     PIPE_BUFFER_USAGE_CPU_READ);
 335         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 336     }
 337
 338     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 339
 340     draw_set_mapped_constant_buffer(r300->draw,
 341                                     PIPE_SHADER_VERTEX,
 342                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 343                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 344                 (sizeof(float) * 4));
 345
 346     draw_arrays(r300->draw, mode, start, count);
 347
 348     for (i = 0; i < r300->vertex_buffer_count; i++) {
 349         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 350         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 351     }
 352 }
 353
 354 /* SW TCL elements, using Draw. */
 355 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 356                                        struct pipe_buffer* indexBuffer,
 357                                        unsigned indexSize,
 358                                        unsigned minIndex,
 359                                        unsigned maxIndex,
 360                                        unsigned mode,
 361                                        unsigned start,
 362                                        unsigned count)
 363 {
 364     struct r300_context* r300 = r300_context(pipe);
 365     int i;
 366     void* indices;
 367
 368     if (!u_trim_pipe_prim(mode, &count)) {
 369         return;
 370     }
 371
 372     if (r300_nothing_to_draw(r300)) {
 373         return;
 374     }
 375
 376     for (i = 0; i < r300->vertex_buffer_count; i++) {
 377         void* buf = pipe_buffer_map(pipe->screen,
 378                                     r300->vertex_buffer[i].buffer,
 379                                     PIPE_BUFFER_USAGE_CPU_READ);
 380         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 381     }
 382
 383     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 384                               PIPE_BUFFER_USAGE_CPU_READ);
 385     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 386                                          minIndex, maxIndex, indices);
 387
 388     draw_set_mapped_constant_buffer(r300->draw,
 389                                     PIPE_SHADER_VERTEX,
 390             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 391             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 392                 (sizeof(float) * 4));
 393
 394     draw_arrays(r300->draw, mode, start, count);
 395
 396     for (i = 0; i < r300->vertex_buffer_count; i++) {
 397         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 398         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 399     }
 400
 401     pipe_buffer_unmap(pipe->screen, indexBuffer);
 402     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 403                                          start + count - 1, NULL);
 404 }
 405
 406 /* Object for rendering using Draw. */
 407 struct r300_render {
 408     /* Parent class */
 409     struct vbuf_render base;
 410
 411     /* Pipe context */
 412     struct r300_context* r300;
 413
 414     /* Vertex information */
 415     size_t vertex_size;
 416     unsigned prim;
 417     unsigned hwprim;
 418
 419     /* VBO */
 420     struct pipe_buffer* vbo;
 421     size_t vbo_size;
 422     size_t vbo_offset;
 423     size_t vbo_max_used;
 424     void * vbo_ptr;
 425 };
 426
 427 static INLINE struct r300_render*
 428 r300_render(struct vbuf_render* render)
 429 {
 430     return (struct r300_render*)render;
 431 }
 432
 433 static const struct vertex_info*
 434 r300_render_get_vertex_info(struct vbuf_render* render)
 435 {
 436     struct r300_render* r300render = r300_render(render);
 437     struct r300_context* r300 = r300render->r300;
 438
 439     r300_update_derived_state(r300);
 440
 441     return &r300->vertex_info->vinfo;
 442 }
 443
 444 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 445                                                    ushort vertex_size,
 446                                                    ushort count)
 447 {
 448     struct r300_render* r300render = r300_render(render);
 449     struct r300_context* r300 = r300render->r300;
 450     struct pipe_screen* screen = r300->context.screen;
 451     size_t size = (size_t)vertex_size * (size_t)count;
 452
 453     if (size + r300render->vbo_offset > r300render->vbo_size)
 454     {
 455         pipe_buffer_reference(&r300->vbo, NULL);
 456         r300render->vbo = pipe_buffer_create(screen,
 457                                              64,
 458                                              PIPE_BUFFER_USAGE_VERTEX,
 459                                              R300_MAX_VBO_SIZE);
 460         r300render->vbo_offset = 0;
 461         r300render->vbo_size = R300_MAX_VBO_SIZE;
 462     }
 463
 464     r300render->vertex_size = vertex_size;
 465     r300->vbo = r300render->vbo;
 466     r300->vbo_offset = r300render->vbo_offset;
 467
 468     return (r300render->vbo) ? TRUE : FALSE;
 469 }
 470
 471 static void* r300_render_map_vertices(struct vbuf_render* render)
 472 {
 473     struct r300_render* r300render = r300_render(render);
 474     struct pipe_screen* screen = r300render->r300->context.screen;
 475
 476     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 477                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 478
 479     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 480 }
 481
 482 static void r300_render_unmap_vertices(struct vbuf_render* render,
 483                                              ushort min,
 484                                              ushort max)
 485 {
 486     struct r300_render* r300render = r300_render(render);
 487     struct pipe_screen* screen = r300render->r300->context.screen;
 488     CS_LOCALS(r300render->r300);
 489     BEGIN_CS(2);
 490     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 491     END_CS;
 492
 493     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 494                                     r300render->vertex_size * (max + 1));
 495     pipe_buffer_unmap(screen, r300render->vbo);
 496 }
 497
 498 static void r300_render_release_vertices(struct vbuf_render* render)
 499 {
 500     struct r300_render* r300render = r300_render(render);
 501
 502     r300render->vbo_offset += r300render->vbo_max_used;
 503     r300render->vbo_max_used = 0;
 504 }
 505
 506 static boolean r300_render_set_primitive(struct vbuf_render* render,
 507                                                unsigned prim)
 508 {
 509     struct r300_render* r300render = r300_render(render);
 510
 511     r300render->prim = prim;
 512     r300render->hwprim = r300_translate_primitive(prim);
 513
 514     return TRUE;
 515 }
 516
 517 static void r300_render_draw_arrays(struct vbuf_render* render,
 518                                           unsigned start,
 519                                           unsigned count)
 520 {
 521     struct r300_render* r300render = r300_render(render);
 522     struct r300_context* r300 = r300render->r300;
 523
 524     CS_LOCALS(r300);
 525
 526     r300_emit_dirty_state(r300);
 527
 528     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 529
 530     BEGIN_CS(2);
 531     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 532     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 533            r300render->hwprim);
 534     END_CS;
 535 }
 536
 537 static void r300_render_draw(struct vbuf_render* render,
 538                                    const ushort* indices,
 539                                    uint count)
 540 {
 541     struct r300_render* r300render = r300_render(render);
 542     struct r300_context* r300 = r300render->r300;
 543     int i;
 544
 545     CS_LOCALS(r300);
 546
 547     r300_emit_dirty_state(r300);
 548
 549     BEGIN_CS(2 + (count+1)/2);
 550     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 551     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 552            r300render->hwprim);
 553     for (i = 0; i < count-1; i += 2) {
 554         OUT_CS(indices[i+1] << 16 | indices[i]);
 555     }
 556     if (count % 2) {
 557         OUT_CS(indices[count-1]);
 558     }
 559     END_CS;
 560 }
 561
 562 static void r300_render_destroy(struct vbuf_render* render)
 563 {
 564     FREE(render);
 565 }
 566
 567 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 568 {
 569     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 570
 571     r300render->r300 = r300;
 572
 573     /* XXX find real numbers plz */
 574     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 575     r300render->base.max_indices = 16 * 1024;
 576
 577     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 578     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 579     r300render->base.map_vertices = r300_render_map_vertices;
 580     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 581     r300render->base.set_primitive = r300_render_set_primitive;
 582     r300render->base.draw = r300_render_draw;
 583     r300render->base.draw_arrays = r300_render_draw_arrays;
 584     r300render->base.release_vertices = r300_render_release_vertices;
 585     r300render->base.destroy = r300_render_destroy;
 586
 587     r300render->vbo = NULL;
 588     r300render->vbo_size = 0;
 589     r300render->vbo_offset = 0;
 590
 591     return &r300render->base;
 592 }
 593
 594 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 595 {
 596     struct vbuf_render* render;
 597     struct draw_stage* stage;
 598
 599     render = r300_render_create(r300);
 600
 601     if (!render) {
 602         return NULL;
 603     }
 604
 605     stage = draw_vbuf_stage(r300->draw, render);
 606
 607     if (!stage) {
 608         render->destroy(render);
 609         return NULL;
 610     }
 611
 612     draw_set_render(r300->draw, render);
 613
 614     return stage;
 615 }