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_index_buffer(r300, indexBuffer, indexSize);
 187
 188     r300_emit_dirty_state(r300);
 189
 190     r300_emit_aos(r300, 0);
 191
 192     r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
 193                             mode, start, count);
 194
 195     return TRUE;
 196 }
 197
 198 /* Simple helpers for context setup. Should probably be moved to util. */
 199 boolean r300_draw_elements(struct pipe_context* pipe,
 200                            struct pipe_buffer* indexBuffer,
 201                            unsigned indexSize, unsigned mode,
 202                            unsigned start, unsigned count)
 203 {
 204     return pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
 205                                      mode, start, count);
 206 }
 207
 208 boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 209                          unsigned start, unsigned count)
 210 {
 211     struct r300_context* r300 = r300_context(pipe);
 212
 213     if (!u_trim_pipe_prim(mode, &count)) {
 214         return FALSE;
 215     }
 216
 217     if (count > 65535) {
 218         return FALSE;
 219     }
 220
 221     r300_update_derived_state(r300);
 222
 223     if (!r300_setup_vertex_buffers(r300)) {
 224         return FALSE;
 225     }
 226
 227     r300_emit_dirty_state(r300);
 228
 229     r300_emit_aos(r300, start);
 230
 231     r300_emit_draw_arrays(r300, mode, count);
 232
 233     return TRUE;
 234 }
 235
 236 /****************************************************************************
 237  * The rest of this file is for SW TCL rendering only. Please be polite and *
 238  * keep these functions separated so that they are easier to locate. ~C.    *
 239  ***************************************************************************/
 240
 241 /* SW TCL arrays, using Draw. */
 242 boolean r300_swtcl_draw_arrays(struct pipe_context* pipe,
 243                                unsigned mode,
 244                                unsigned start,
 245                                unsigned count)
 246 {
 247     struct r300_context* r300 = r300_context(pipe);
 248     int i;
 249
 250     if (!u_trim_pipe_prim(mode, &count)) {
 251         return FALSE;
 252     }
 253
 254     for (i = 0; i < r300->vertex_buffer_count; i++) {
 255         void* buf = pipe_buffer_map(pipe->screen,
 256                                     r300->vertex_buffer[i].buffer,
 257                                     PIPE_BUFFER_USAGE_CPU_READ);
 258         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 259     }
 260
 261     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 262
 263     draw_set_mapped_constant_buffer(r300->draw,
 264             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 265             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 266                 (sizeof(float) * 4));
 267
 268     draw_arrays(r300->draw, mode, start, count);
 269
 270     for (i = 0; i < r300->vertex_buffer_count; i++) {
 271         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 272         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 273     }
 274
 275     return TRUE;
 276 }
 277
 278 /* SW TCL elements, using Draw. */
 279 boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 280                                        struct pipe_buffer* indexBuffer,
 281                                        unsigned indexSize,
 282                                        unsigned minIndex,
 283                                        unsigned maxIndex,
 284                                        unsigned mode,
 285                                        unsigned start,
 286                                        unsigned count)
 287 {
 288     struct r300_context* r300 = r300_context(pipe);
 289     int i;
 290
 291     if (!u_trim_pipe_prim(mode, &count)) {
 292         return FALSE;
 293     }
 294
 295     for (i = 0; i < r300->vertex_buffer_count; i++) {
 296         void* buf = pipe_buffer_map(pipe->screen,
 297                                     r300->vertex_buffer[i].buffer,
 298                                     PIPE_BUFFER_USAGE_CPU_READ);
 299         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 300     }
 301
 302     void* indices = pipe_buffer_map(pipe->screen, indexBuffer,
 303                                     PIPE_BUFFER_USAGE_CPU_READ);
 304     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 305                                          minIndex, maxIndex, indices);
 306
 307     draw_set_mapped_constant_buffer(r300->draw,
 308             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 309             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 310                 (sizeof(float) * 4));
 311
 312     draw_arrays(r300->draw, mode, start, count);
 313
 314     for (i = 0; i < r300->vertex_buffer_count; i++) {
 315         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 316         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 317     }
 318
 319     pipe_buffer_unmap(pipe->screen, indexBuffer);
 320     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 321                                          start + count - 1, NULL);
 322
 323     return TRUE;
 324 }
 325
 326 /* Object for rendering using Draw. */
 327 struct r300_render {
 328     /* Parent class */
 329     struct vbuf_render base;
 330
 331     /* Pipe context */
 332     struct r300_context* r300;
 333
 334     /* Vertex information */
 335     size_t vertex_size;
 336     unsigned prim;
 337     unsigned hwprim;
 338
 339     /* VBO */
 340     struct pipe_buffer* vbo;
 341     size_t vbo_size;
 342     size_t vbo_offset;
 343     size_t vbo_max_used;
 344     void * vbo_ptr;
 345 };
 346
 347 static INLINE struct r300_render*
 348 r300_render(struct vbuf_render* render)
 349 {
 350     return (struct r300_render*)render;
 351 }
 352
 353 static const struct vertex_info*
 354 r300_render_get_vertex_info(struct vbuf_render* render)
 355 {
 356     struct r300_render* r300render = r300_render(render);
 357     struct r300_context* r300 = r300render->r300;
 358
 359     r300_update_derived_state(r300);
 360
 361     return &r300->vertex_info->vinfo;
 362 }
 363
 364 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 365                                                    ushort vertex_size,
 366                                                    ushort count)
 367 {
 368     struct r300_render* r300render = r300_render(render);
 369     struct r300_context* r300 = r300render->r300;
 370     struct pipe_screen* screen = r300->context.screen;
 371     size_t size = (size_t)vertex_size * (size_t)count;
 372
 373     if (size + r300render->vbo_offset > r300render->vbo_size)
 374     {
 375         pipe_buffer_reference(&r300->vbo, NULL);
 376         r300render->vbo = pipe_buffer_create(screen,
 377                                              64,
 378                                              PIPE_BUFFER_USAGE_VERTEX,
 379                                              R300_MAX_VBO_SIZE);
 380         r300render->vbo_offset = 0;
 381         r300render->vbo_size = R300_MAX_VBO_SIZE;
 382     }
 383
 384     r300render->vertex_size = vertex_size;
 385     r300->vbo = r300render->vbo;
 386     r300->vbo_offset = r300render->vbo_offset;
 387
 388     return (r300render->vbo) ? TRUE : FALSE;
 389 }
 390
 391 static void* r300_render_map_vertices(struct vbuf_render* render)
 392 {
 393     struct r300_render* r300render = r300_render(render);
 394     struct pipe_screen* screen = r300render->r300->context.screen;
 395
 396     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 397                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 398
 399     return (r300render->vbo_ptr + r300render->vbo_offset);
 400 }
 401
 402 static void r300_render_unmap_vertices(struct vbuf_render* render,
 403                                              ushort min,
 404                                              ushort max)
 405 {
 406     struct r300_render* r300render = r300_render(render);
 407     struct pipe_screen* screen = r300render->r300->context.screen;
 408     CS_LOCALS(r300render->r300);
 409     BEGIN_CS(2);
 410     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 411     END_CS;
 412
 413     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 414                                     r300render->vertex_size * (max + 1));
 415     pipe_buffer_unmap(screen, r300render->vbo);
 416 }
 417
 418 static void r300_render_release_vertices(struct vbuf_render* render)
 419 {
 420     struct r300_render* r300render = r300_render(render);
 421
 422     r300render->vbo_offset += r300render->vbo_max_used;
 423     r300render->vbo_max_used = 0;
 424 }
 425
 426 static boolean r300_render_set_primitive(struct vbuf_render* render,
 427                                                unsigned prim)
 428 {
 429     struct r300_render* r300render = r300_render(render);
 430
 431     r300render->prim = prim;
 432     r300render->hwprim = r300_translate_primitive(prim);
 433
 434     return TRUE;
 435 }
 436
 437 static void r300_render_draw_arrays(struct vbuf_render* render,
 438                                           unsigned start,
 439                                           unsigned count)
 440 {
 441     struct r300_render* r300render = r300_render(render);
 442     struct r300_context* r300 = r300render->r300;
 443
 444     CS_LOCALS(r300);
 445
 446     r300_emit_dirty_state(r300);
 447
 448     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 449
 450     BEGIN_CS(2);
 451     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 452     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 453            r300render->hwprim);
 454     END_CS;
 455 }
 456
 457 static void r300_render_draw(struct vbuf_render* render,
 458                                    const ushort* indices,
 459                                    uint count)
 460 {
 461     struct r300_render* r300render = r300_render(render);
 462     struct r300_context* r300 = r300render->r300;
 463     int i;
 464
 465     CS_LOCALS(r300);
 466
 467     r300_emit_dirty_state(r300);
 468
 469     BEGIN_CS(2 + (count+1)/2);
 470     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 471     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 472            r300render->hwprim);
 473     for (i = 0; i < count-1; i += 2) {
 474         OUT_CS(indices[i+1] << 16 | indices[i]);
 475     }
 476     if (count % 2) {
 477         OUT_CS(indices[count-1]);
 478     }
 479     END_CS;
 480 }
 481
 482 static void r300_render_destroy(struct vbuf_render* render)
 483 {
 484     FREE(render);
 485 }
 486
 487 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 488 {
 489     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 490
 491     r300render->r300 = r300;
 492
 493     /* XXX find real numbers plz */
 494     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 495     r300render->base.max_indices = 16 * 1024;
 496
 497     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 498     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 499     r300render->base.map_vertices = r300_render_map_vertices;
 500     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 501     r300render->base.set_primitive = r300_render_set_primitive;
 502     r300render->base.draw = r300_render_draw;
 503     r300render->base.draw_arrays = r300_render_draw_arrays;
 504     r300render->base.release_vertices = r300_render_release_vertices;
 505     r300render->base.destroy = r300_render_destroy;
 506
 507     r300render->vbo = NULL;
 508     r300render->vbo_size = 0;
 509     r300render->vbo_offset = 0;
 510
 511     return &r300render->base;
 512 }
 513
 514 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 515 {
 516     struct vbuf_render* render;
 517     struct draw_stage* stage;
 518
 519     render = r300_render_create(r300);
 520
 521     if (!render) {
 522         return NULL;
 523     }
 524
 525     stage = draw_vbuf_stage(r300->draw, render);
 526
 527     if (!stage) {
 528         render->destroy(render);
 529         return NULL;
 530     }
 531
 532     draw_set_render(r300->draw, render);
 533
 534     return stage;
 535 }