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