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 #include "r300_vbo.h"
  41
  42 /* r300_render: Vertex and index buffer primitive emission. */
  43 #define R300_MAX_VBO_SIZE  (1024 * 1024)
  44
  45 uint32_t r300_translate_primitive(unsigned prim)
  46 {
  47     switch (prim) {
  48         case PIPE_PRIM_POINTS:
  49             return R300_VAP_VF_CNTL__PRIM_POINTS;
  50         case PIPE_PRIM_LINES:
  51             return R300_VAP_VF_CNTL__PRIM_LINES;
  52         case PIPE_PRIM_LINE_LOOP:
  53             return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
  54         case PIPE_PRIM_LINE_STRIP:
  55             return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
  56         case PIPE_PRIM_TRIANGLES:
  57             return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
  58         case PIPE_PRIM_TRIANGLE_STRIP:
  59             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
  60         case PIPE_PRIM_TRIANGLE_FAN:
  61             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
  62         case PIPE_PRIM_QUADS:
  63             return R300_VAP_VF_CNTL__PRIM_QUADS;
  64         case PIPE_PRIM_QUAD_STRIP:
  65             return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
  66         case PIPE_PRIM_POLYGON:
  67             return R300_VAP_VF_CNTL__PRIM_POLYGON;
  68         default:
  69             return 0;
  70     }
  71 }
  72
  73 static boolean r300_nothing_to_draw(struct r300_context *r300)
  74 {
  75     return r300->rs_state->rs.scissor &&
  76            r300->scissor_state->scissor.empty_area;
  77 }
  78
  79 static void r300_emit_draw_arrays(struct r300_context *r300,
  80                                   unsigned mode,
  81                                   unsigned count)
  82 {
  83     CS_LOCALS(r300);
  84
  85     BEGIN_CS(4);
  86     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count);
  87     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
  88     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
  89            r300_translate_primitive(mode));
  90     END_CS;
  91 }
  92
  93 static void r300_emit_draw_elements(struct r300_context *r300,
  94                                     struct pipe_buffer* indexBuffer,
  95                                     unsigned indexSize,
  96                                     unsigned minIndex,
  97                                     unsigned maxIndex,
  98                                     unsigned mode,
  99                                     unsigned start,
 100                                     unsigned count)
 101 {
 102     uint32_t count_dwords;
 103     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
 104     CS_LOCALS(r300);
 105
 106     /* XXX most of these are stupid */
 107     assert(indexSize == 4 || indexSize == 2);
 108     assert((start * indexSize)  % 4 == 0);
 109     assert(offset_dwords == 0);
 110
 111     BEGIN_CS(10);
 112     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
 113     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 114     if (indexSize == 4) {
 115         count_dwords = count + start;
 116         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 117                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 118                r300_translate_primitive(mode));
 119     } else {
 120         count_dwords = (count + start + 1) / 2;
 121         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 122                r300_translate_primitive(mode));
 123     }
 124
 125     /* INDX_BUFFER is a truly special packet3.
 126      * Unlike most other packet3, where the offset is after the count,
 127      * the order is reversed, so the relocation ends up carrying the
 128      * size of the indexbuf instead of the offset.
 129      *
 130      * XXX Fix offset
 131      */
 132     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 133     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 134            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 135     OUT_CS(offset_dwords);
 136     OUT_CS_RELOC(indexBuffer, count_dwords,
 137         RADEON_GEM_DOMAIN_GTT, 0, 0);
 138
 139     END_CS;
 140 }
 141
 142
 143 static boolean r300_setup_vertex_buffers(struct r300_context *r300)
 144 {
 145     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
 146     struct pipe_vertex_element *velem = r300->vertex_element;
 147
 148 validate:
 149     for (int i = 0; i < r300->vertex_element_count; i++) {
 150         if (!r300->winsys->add_buffer(r300->winsys,
 151                 vbuf[velem[i].vertex_buffer_index].buffer,
 152             RADEON_GEM_DOMAIN_GTT, 0)) {
 153             r300->context.flush(&r300->context, 0, NULL);
 154             goto validate;
 155         }
 156     }
 157
 158     if (!r300->winsys->validate(r300->winsys)) {
 159         r300->context.flush(&r300->context, 0, NULL);
 160         return r300->winsys->validate(r300->winsys);
 161     }
 162
 163     return TRUE;
 164 }
 165
 166 /* This is the fast-path drawing & emission for HW TCL. */
 167 boolean r300_draw_range_elements(struct pipe_context* pipe,
 168                                  struct pipe_buffer* indexBuffer,
 169                                  unsigned indexSize,
 170                                  unsigned minIndex,
 171                                  unsigned maxIndex,
 172                                  unsigned mode,
 173                                  unsigned start,
 174                                  unsigned count)
 175 {
 176     struct r300_context* r300 = r300_context(pipe);
 177
 178     if (!u_trim_pipe_prim(mode, &count)) {
 179         return FALSE;
 180     }
 181
 182
 183     if (count > 65535) {
 184         return FALSE;
 185     }
 186
 187     if (r300_nothing_to_draw(r300)) {
 188         return TRUE;
 189     }
 190
 191     r300_update_derived_state(r300);
 192
 193     if (!r300_setup_vertex_buffers(r300)) {
 194         return FALSE;
 195     }
 196
 197     setup_index_buffer(r300, indexBuffer, indexSize);
 198
 199     r300_emit_dirty_state(r300);
 200
 201     r300_emit_aos(r300, 0);
 202
 203     r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
 204                             mode, start, count);
 205
 206     return TRUE;
 207 }
 208
 209 /* Simple helpers for context setup. Should probably be moved to util. */
 210 boolean r300_draw_elements(struct pipe_context* pipe,
 211                            struct pipe_buffer* indexBuffer,
 212                            unsigned indexSize, unsigned mode,
 213                            unsigned start, unsigned count)
 214 {
 215     return pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
 216                                      mode, start, count);
 217 }
 218
 219 boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 220                          unsigned start, unsigned count)
 221 {
 222     struct r300_context* r300 = r300_context(pipe);
 223
 224     if (!u_trim_pipe_prim(mode, &count)) {
 225         return FALSE;
 226     }
 227
 228     if (count > 65535) {
 229         return FALSE;
 230     }
 231
 232     if (r300_nothing_to_draw(r300)) {
 233         return TRUE;
 234     }
 235
 236     r300_update_derived_state(r300);
 237
 238     if (!r300_setup_vertex_buffers(r300)) {
 239         return FALSE;
 240     }
 241
 242     r300_emit_dirty_state(r300);
 243
 244     r300_emit_aos(r300, start);
 245
 246     r300_emit_draw_arrays(r300, mode, count);
 247
 248     return TRUE;
 249 }
 250
 251 /****************************************************************************
 252  * The rest of this file is for SW TCL rendering only. Please be polite and *
 253  * keep these functions separated so that they are easier to locate. ~C.    *
 254  ***************************************************************************/
 255
 256 /* SW TCL arrays, using Draw. */
 257 boolean r300_swtcl_draw_arrays(struct pipe_context* pipe,
 258                                unsigned mode,
 259                                unsigned start,
 260                                unsigned count)
 261 {
 262     struct r300_context* r300 = r300_context(pipe);
 263     int i;
 264
 265     if (!u_trim_pipe_prim(mode, &count)) {
 266         return FALSE;
 267     }
 268
 269     if (r300_nothing_to_draw(r300)) {
 270         return TRUE;
 271     }
 272
 273     for (i = 0; i < r300->vertex_buffer_count; i++) {
 274         void* buf = pipe_buffer_map(pipe->screen,
 275                                     r300->vertex_buffer[i].buffer,
 276                                     PIPE_BUFFER_USAGE_CPU_READ);
 277         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 278     }
 279
 280     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 281
 282     draw_set_mapped_constant_buffer(r300->draw,
 283             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 284             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 285                 (sizeof(float) * 4));
 286
 287     draw_arrays(r300->draw, mode, start, count);
 288
 289     for (i = 0; i < r300->vertex_buffer_count; i++) {
 290         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 291         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 292     }
 293
 294     return TRUE;
 295 }
 296
 297 /* SW TCL elements, using Draw. */
 298 boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 299                                        struct pipe_buffer* indexBuffer,
 300                                        unsigned indexSize,
 301                                        unsigned minIndex,
 302                                        unsigned maxIndex,
 303                                        unsigned mode,
 304                                        unsigned start,
 305                                        unsigned count)
 306 {
 307     struct r300_context* r300 = r300_context(pipe);
 308     int i;
 309
 310     if (!u_trim_pipe_prim(mode, &count)) {
 311         return FALSE;
 312     }
 313
 314     if (r300_nothing_to_draw(r300)) {
 315         return TRUE;
 316     }
 317
 318     for (i = 0; i < r300->vertex_buffer_count; i++) {
 319         void* buf = pipe_buffer_map(pipe->screen,
 320                                     r300->vertex_buffer[i].buffer,
 321                                     PIPE_BUFFER_USAGE_CPU_READ);
 322         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 323     }
 324
 325     void* indices = pipe_buffer_map(pipe->screen, indexBuffer,
 326                                     PIPE_BUFFER_USAGE_CPU_READ);
 327     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 328                                          minIndex, maxIndex, indices);
 329
 330     draw_set_mapped_constant_buffer(r300->draw,
 331             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 332             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 333                 (sizeof(float) * 4));
 334
 335     draw_arrays(r300->draw, mode, start, count);
 336
 337     for (i = 0; i < r300->vertex_buffer_count; i++) {
 338         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 339         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 340     }
 341
 342     pipe_buffer_unmap(pipe->screen, indexBuffer);
 343     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 344                                          start + count - 1, NULL);
 345
 346     return TRUE;
 347 }
 348
 349 /* Object for rendering using Draw. */
 350 struct r300_render {
 351     /* Parent class */
 352     struct vbuf_render base;
 353
 354     /* Pipe context */
 355     struct r300_context* r300;
 356
 357     /* Vertex information */
 358     size_t vertex_size;
 359     unsigned prim;
 360     unsigned hwprim;
 361
 362     /* VBO */
 363     struct pipe_buffer* vbo;
 364     size_t vbo_size;
 365     size_t vbo_offset;
 366     size_t vbo_max_used;
 367     void * vbo_ptr;
 368 };
 369
 370 static INLINE struct r300_render*
 371 r300_render(struct vbuf_render* render)
 372 {
 373     return (struct r300_render*)render;
 374 }
 375
 376 static const struct vertex_info*
 377 r300_render_get_vertex_info(struct vbuf_render* render)
 378 {
 379     struct r300_render* r300render = r300_render(render);
 380     struct r300_context* r300 = r300render->r300;
 381
 382     r300_update_derived_state(r300);
 383
 384     return &r300->vertex_info->vinfo;
 385 }
 386
 387 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 388                                                    ushort vertex_size,
 389                                                    ushort count)
 390 {
 391     struct r300_render* r300render = r300_render(render);
 392     struct r300_context* r300 = r300render->r300;
 393     struct pipe_screen* screen = r300->context.screen;
 394     size_t size = (size_t)vertex_size * (size_t)count;
 395
 396     if (size + r300render->vbo_offset > r300render->vbo_size)
 397     {
 398         pipe_buffer_reference(&r300->vbo, NULL);
 399         r300render->vbo = pipe_buffer_create(screen,
 400                                              64,
 401                                              PIPE_BUFFER_USAGE_VERTEX,
 402                                              R300_MAX_VBO_SIZE);
 403         r300render->vbo_offset = 0;
 404         r300render->vbo_size = R300_MAX_VBO_SIZE;
 405     }
 406
 407     r300render->vertex_size = vertex_size;
 408     r300->vbo = r300render->vbo;
 409     r300->vbo_offset = r300render->vbo_offset;
 410
 411     return (r300render->vbo) ? TRUE : FALSE;
 412 }
 413
 414 static void* r300_render_map_vertices(struct vbuf_render* render)
 415 {
 416     struct r300_render* r300render = r300_render(render);
 417     struct pipe_screen* screen = r300render->r300->context.screen;
 418
 419     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 420                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 421
 422     return (r300render->vbo_ptr + r300render->vbo_offset);
 423 }
 424
 425 static void r300_render_unmap_vertices(struct vbuf_render* render,
 426                                              ushort min,
 427                                              ushort max)
 428 {
 429     struct r300_render* r300render = r300_render(render);
 430     struct pipe_screen* screen = r300render->r300->context.screen;
 431     CS_LOCALS(r300render->r300);
 432     BEGIN_CS(2);
 433     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 434     END_CS;
 435
 436     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 437                                     r300render->vertex_size * (max + 1));
 438     pipe_buffer_unmap(screen, r300render->vbo);
 439 }
 440
 441 static void r300_render_release_vertices(struct vbuf_render* render)
 442 {
 443     struct r300_render* r300render = r300_render(render);
 444
 445     r300render->vbo_offset += r300render->vbo_max_used;
 446     r300render->vbo_max_used = 0;
 447 }
 448
 449 static boolean r300_render_set_primitive(struct vbuf_render* render,
 450                                                unsigned prim)
 451 {
 452     struct r300_render* r300render = r300_render(render);
 453
 454     r300render->prim = prim;
 455     r300render->hwprim = r300_translate_primitive(prim);
 456
 457     return TRUE;
 458 }
 459
 460 static void r300_render_draw_arrays(struct vbuf_render* render,
 461                                           unsigned start,
 462                                           unsigned count)
 463 {
 464     struct r300_render* r300render = r300_render(render);
 465     struct r300_context* r300 = r300render->r300;
 466
 467     CS_LOCALS(r300);
 468
 469     r300_emit_dirty_state(r300);
 470
 471     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 472
 473     BEGIN_CS(2);
 474     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 475     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 476            r300render->hwprim);
 477     END_CS;
 478 }
 479
 480 static void r300_render_draw(struct vbuf_render* render,
 481                                    const ushort* indices,
 482                                    uint count)
 483 {
 484     struct r300_render* r300render = r300_render(render);
 485     struct r300_context* r300 = r300render->r300;
 486     int i;
 487
 488     CS_LOCALS(r300);
 489
 490     r300_emit_dirty_state(r300);
 491
 492     BEGIN_CS(2 + (count+1)/2);
 493     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 494     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 495            r300render->hwprim);
 496     for (i = 0; i < count-1; i += 2) {
 497         OUT_CS(indices[i+1] << 16 | indices[i]);
 498     }
 499     if (count % 2) {
 500         OUT_CS(indices[count-1]);
 501     }
 502     END_CS;
 503 }
 504
 505 static void r300_render_destroy(struct vbuf_render* render)
 506 {
 507     FREE(render);
 508 }
 509
 510 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 511 {
 512     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 513
 514     r300render->r300 = r300;
 515
 516     /* XXX find real numbers plz */
 517     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 518     r300render->base.max_indices = 16 * 1024;
 519
 520     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 521     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 522     r300render->base.map_vertices = r300_render_map_vertices;
 523     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 524     r300render->base.set_primitive = r300_render_set_primitive;
 525     r300render->base.draw = r300_render_draw;
 526     r300render->base.draw_arrays = r300_render_draw_arrays;
 527     r300render->base.release_vertices = r300_render_release_vertices;
 528     r300render->base.destroy = r300_render_destroy;
 529
 530     r300render->vbo = NULL;
 531     r300render->vbo_size = 0;
 532     r300render->vbo_offset = 0;
 533
 534     return &r300render->base;
 535 }
 536
 537 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 538 {
 539     struct vbuf_render* render;
 540     struct draw_stage* stage;
 541
 542     render = r300_render_create(r300);
 543
 544     if (!render) {
 545         return NULL;
 546     }
 547
 548     stage = draw_vbuf_stage(r300->draw, render);
 549
 550     if (!stage) {
 551         render->destroy(render);
 552         return NULL;
 553     }
 554
 555     draw_set_render(r300->draw, render);
 556
 557     return stage;
 558 }