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