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 uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
  73                                             unsigned mode)
  74 {
  75     struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
  76     uint32_t color_control = rs->color_control;
  77
  78     /* By default (see r300_state.c:r300_create_rs_state) color_control is
  79      * initialized to provoking the first vertex.
  80      *
  81      * Triangle fans must be reduced to the second vertex, not the first, in
  82      * Gallium flatshade-first mode, as per the GL spec.
  83      * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
  84      *
  85      * Quads never provoke correctly in flatshade-first mode. The first
  86      * vertex is never considered as provoking, so only the second, third,
  87      * and fourth vertices can be selected, and both "third" and "last" modes
  88      * select the fourth vertex. This is probably due to D3D lacking quads.
  89      *
  90      * Similarly, polygons reduce to the first, not the last, vertex, when in
  91      * "last" mode, and all other modes start from the second vertex.
  92      *
  93      * ~ C.
  94      */
  95
  96     if (rs->rs.flatshade_first) {
  97         switch (mode) {
  98             case PIPE_PRIM_TRIANGLE_FAN:
  99                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
 100                 break;
 101             case PIPE_PRIM_QUADS:
 102             case PIPE_PRIM_QUAD_STRIP:
 103             case PIPE_PRIM_POLYGON:
 104                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 105                 break;
 106             default:
 107                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
 108                 break;
 109         }
 110     } else {
 111         color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 112     }
 113
 114     return color_control;
 115 }
 116
 117 static void r300_emit_draw_immediate(struct r300_context *r300,
 118                                      unsigned mode,
 119                                      unsigned start,
 120                                      unsigned count)
 121 {
 122     struct pipe_buffer* vbo = r300->vertex_buffer[0].buffer;
 123     unsigned vertex_size = r300->vertex_buffer[0].stride / sizeof(float);
 124     unsigned i;
 125     uint32_t* map;
 126     CS_LOCALS(r300);
 127
 128     map = (uint32_t*)pipe_buffer_map_range(r300->context.screen, vbo,
 129             start * vertex_size, count * vertex_size,
 130             PIPE_BUFFER_USAGE_CPU_READ);
 131
 132     BEGIN_CS(10 + count * vertex_size);
 133     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 134             r300_provoking_vertex_fixes(r300, mode));
 135     OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
 136     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
 137     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
 138     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
 139     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
 140             r300_translate_primitive(mode));
 141     //debug_printf("r300: Immd %d verts, %d attrs\n", count, vertex_size);
 142     for (i = 0; i < count * vertex_size; i++) {
 143         if (i % vertex_size == 0) {
 144             //debug_printf("r300: -- vert --\n");
 145         }
 146         //debug_printf("r300: 0x%08x\n", *map);
 147         OUT_CS(*map);
 148         map++;
 149     }
 150     END_CS;
 151
 152     pipe_buffer_unmap(r300->context.screen, vbo);
 153 }
 154
 155 static void r300_emit_draw_arrays(struct r300_context *r300,
 156                                   unsigned mode,
 157                                   unsigned count)
 158 {
 159     CS_LOCALS(r300);
 160
 161     BEGIN_CS(8);
 162     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 163             r300_provoking_vertex_fixes(r300, mode));
 164     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
 165     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
 166     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 167     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 168            r300_translate_primitive(mode));
 169     END_CS;
 170 }
 171
 172 static void r300_emit_draw_elements(struct r300_context *r300,
 173                                     struct pipe_buffer* indexBuffer,
 174                                     unsigned indexSize,
 175                                     unsigned minIndex,
 176                                     unsigned maxIndex,
 177                                     unsigned mode,
 178                                     unsigned start,
 179                                     unsigned count)
 180 {
 181     uint32_t count_dwords;
 182     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
 183     CS_LOCALS(r300);
 184
 185     /* XXX most of these are stupid */
 186     assert(indexSize == 4 || indexSize == 2);
 187     assert((start * indexSize)  % 4 == 0);
 188     assert(offset_dwords == 0);
 189
 190     BEGIN_CS(14);
 191     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 192             r300_provoking_vertex_fixes(r300, mode));
 193     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex);
 194     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
 195     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 196     if (indexSize == 4) {
 197         count_dwords = count + start;
 198         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 199                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 200                r300_translate_primitive(mode));
 201     } else {
 202         count_dwords = (count + start + 1) / 2;
 203         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 204                r300_translate_primitive(mode));
 205     }
 206
 207     /* INDX_BUFFER is a truly special packet3.
 208      * Unlike most other packet3, where the offset is after the count,
 209      * the order is reversed, so the relocation ends up carrying the
 210      * size of the indexbuf instead of the offset.
 211      *
 212      * XXX Fix offset
 213      */
 214     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 215     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 216            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 217     OUT_CS(offset_dwords);
 218     OUT_CS_RELOC(indexBuffer, count_dwords,
 219         RADEON_GEM_DOMAIN_GTT, 0, 0);
 220
 221     END_CS;
 222 }
 223
 224
 225 static boolean r300_setup_vertex_buffers(struct r300_context *r300)
 226 {
 227     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
 228     struct pipe_vertex_element *velem = r300->vertex_element;
 229
 230 validate:
 231     for (int i = 0; i < r300->vertex_element_count; i++) {
 232         if (!r300->winsys->add_buffer(r300->winsys,
 233                 vbuf[velem[i].vertex_buffer_index].buffer,
 234             RADEON_GEM_DOMAIN_GTT, 0)) {
 235             r300->context.flush(&r300->context, 0, NULL);
 236             goto validate;
 237         }
 238     }
 239
 240     if (!r300->winsys->validate(r300->winsys)) {
 241         r300->context.flush(&r300->context, 0, NULL);
 242         return r300->winsys->validate(r300->winsys);
 243     }
 244
 245     return TRUE;
 246 }
 247
 248 static void r300_shorten_ubyte_elts(struct r300_context* r300,
 249                                     struct pipe_buffer** elts,
 250                                     unsigned count)
 251 {
 252     struct pipe_screen* screen = r300->context.screen;
 253     struct pipe_buffer* new_elts;
 254     unsigned char *in_map;
 255     unsigned short *out_map;
 256     unsigned i;
 257
 258     new_elts = screen->buffer_create(screen, 32,
 259                                      PIPE_BUFFER_USAGE_INDEX |
 260                                      PIPE_BUFFER_USAGE_CPU_WRITE |
 261                                      PIPE_BUFFER_USAGE_GPU_READ,
 262                                      2 * count);
 263
 264     in_map = pipe_buffer_map(screen, *elts, PIPE_BUFFER_USAGE_CPU_READ);
 265     out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE);
 266
 267     for (i = 0; i < count; i++) {
 268         *out_map = (unsigned short)*in_map;
 269         in_map++;
 270         out_map++;
 271     }
 272
 273     pipe_buffer_unmap(screen, *elts);
 274     pipe_buffer_unmap(screen, new_elts);
 275
 276     *elts = new_elts;
 277 }
 278
 279 /* This is the fast-path drawing & emission for HW TCL. */
 280 void r300_draw_range_elements(struct pipe_context* pipe,
 281                               struct pipe_buffer* indexBuffer,
 282                               unsigned indexSize,
 283                               unsigned minIndex,
 284                               unsigned maxIndex,
 285                               unsigned mode,
 286                               unsigned start,
 287                               unsigned count)
 288 {
 289     struct r300_context* r300 = r300_context(pipe);
 290     struct pipe_buffer* orgIndexBuffer = indexBuffer;
 291
 292     if (!u_trim_pipe_prim(mode, &count)) {
 293         return;
 294     }
 295
 296     if (count > 65535) {
 297        /* XXX: use aux/indices functions to split this into smaller
 298         * primitives.
 299         */
 300         return;
 301     }
 302
 303     r300_update_derived_state(r300);
 304
 305     if (!r300_setup_vertex_buffers(r300)) {
 306         return;
 307     }
 308
 309     if (indexSize == 1) {
 310         r300_shorten_ubyte_elts(r300, &indexBuffer, count);
 311         indexSize = 2;
 312     }
 313
 314     if (!r300->winsys->add_buffer(r300->winsys, indexBuffer,
 315                                   RADEON_GEM_DOMAIN_GTT, 0)) {
 316         goto cleanup;
 317     }
 318
 319     if (!r300->winsys->validate(r300->winsys)) {
 320         goto cleanup;
 321     }
 322
 323     r300_emit_dirty_state(r300);
 324
 325     r300_emit_aos(r300, 0);
 326
 327     r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
 328                             mode, start, count);
 329
 330 cleanup:
 331     if (indexBuffer != orgIndexBuffer) {
 332         pipe->screen->buffer_destroy(indexBuffer);
 333     }
 334 }
 335
 336 /* Simple helpers for context setup. Should probably be moved to util. */
 337 void r300_draw_elements(struct pipe_context* pipe,
 338                         struct pipe_buffer* indexBuffer,
 339                         unsigned indexSize, unsigned mode,
 340                         unsigned start, unsigned count)
 341 {
 342    pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
 343                              mode, start, count);
 344 }
 345
 346 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 347                       unsigned start, unsigned count)
 348 {
 349     struct r300_context* r300 = r300_context(pipe);
 350
 351     if (!u_trim_pipe_prim(mode, &count)) {
 352         return;
 353     }
 354
 355     if (count > 65535) {
 356         /* XXX: driver needs to handle this -- use the functions in
 357          * aux/indices to split this into several smaller primitives.
 358          */
 359         return;
 360     }
 361
 362     r300_update_derived_state(r300);
 363
 364     if (!r300_setup_vertex_buffers(r300)) {
 365         return;
 366     }
 367
 368     r300_emit_dirty_state(r300);
 369
 370     if (FALSE && count <= 4 && r300->vertex_buffer_count == 1) {
 371         r300_emit_draw_immediate(r300, mode, start, count);
 372     } else {
 373         r300_emit_aos(r300, start);
 374         r300_emit_draw_arrays(r300, mode, count);
 375     }
 376 }
 377
 378 /****************************************************************************
 379  * The rest of this file is for SW TCL rendering only. Please be polite and *
 380  * keep these functions separated so that they are easier to locate. ~C.    *
 381  ***************************************************************************/
 382
 383 /* SW TCL arrays, using Draw. */
 384 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 385                                unsigned mode,
 386                                unsigned start,
 387                                unsigned count)
 388 {
 389     struct r300_context* r300 = r300_context(pipe);
 390     int i;
 391
 392     if (!u_trim_pipe_prim(mode, &count)) {
 393         return;
 394     }
 395
 396     for (i = 0; i < r300->vertex_buffer_count; i++) {
 397         void* buf = pipe_buffer_map(pipe->screen,
 398                                     r300->vertex_buffer[i].buffer,
 399                                     PIPE_BUFFER_USAGE_CPU_READ);
 400         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 401     }
 402
 403     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 404
 405     draw_set_mapped_constant_buffer(r300->draw,
 406                                     PIPE_SHADER_VERTEX,
 407                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 408                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 409                 (sizeof(float) * 4));
 410
 411     draw_arrays(r300->draw, mode, start, count);
 412
 413     for (i = 0; i < r300->vertex_buffer_count; i++) {
 414         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 415         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 416     }
 417 }
 418
 419 /* SW TCL elements, using Draw. */
 420 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 421                                        struct pipe_buffer* indexBuffer,
 422                                        unsigned indexSize,
 423                                        unsigned minIndex,
 424                                        unsigned maxIndex,
 425                                        unsigned mode,
 426                                        unsigned start,
 427                                        unsigned count)
 428 {
 429     struct r300_context* r300 = r300_context(pipe);
 430     int i;
 431     void* indices;
 432
 433     if (!u_trim_pipe_prim(mode, &count)) {
 434         return;
 435     }
 436
 437     for (i = 0; i < r300->vertex_buffer_count; i++) {
 438         void* buf = pipe_buffer_map(pipe->screen,
 439                                     r300->vertex_buffer[i].buffer,
 440                                     PIPE_BUFFER_USAGE_CPU_READ);
 441         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 442     }
 443
 444     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 445                               PIPE_BUFFER_USAGE_CPU_READ);
 446     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 447                                          minIndex, maxIndex, indices);
 448
 449     draw_set_mapped_constant_buffer(r300->draw,
 450                                     PIPE_SHADER_VERTEX,
 451             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 452             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 453                 (sizeof(float) * 4));
 454
 455     draw_arrays(r300->draw, mode, start, count);
 456
 457     for (i = 0; i < r300->vertex_buffer_count; i++) {
 458         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 459         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 460     }
 461
 462     pipe_buffer_unmap(pipe->screen, indexBuffer);
 463     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 464                                          start + count - 1, NULL);
 465 }
 466
 467 /* Object for rendering using Draw. */
 468 struct r300_render {
 469     /* Parent class */
 470     struct vbuf_render base;
 471
 472     /* Pipe context */
 473     struct r300_context* r300;
 474
 475     /* Vertex information */
 476     size_t vertex_size;
 477     unsigned prim;
 478     unsigned hwprim;
 479
 480     /* VBO */
 481     struct pipe_buffer* vbo;
 482     size_t vbo_size;
 483     size_t vbo_offset;
 484     size_t vbo_max_used;
 485     void * vbo_ptr;
 486 };
 487
 488 static INLINE struct r300_render*
 489 r300_render(struct vbuf_render* render)
 490 {
 491     return (struct r300_render*)render;
 492 }
 493
 494 static const struct vertex_info*
 495 r300_render_get_vertex_info(struct vbuf_render* render)
 496 {
 497     struct r300_render* r300render = r300_render(render);
 498     struct r300_context* r300 = r300render->r300;
 499
 500     r300_update_derived_state(r300);
 501
 502     return &r300->vertex_info->vinfo;
 503 }
 504
 505 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 506                                                    ushort vertex_size,
 507                                                    ushort count)
 508 {
 509     struct r300_render* r300render = r300_render(render);
 510     struct r300_context* r300 = r300render->r300;
 511     struct pipe_screen* screen = r300->context.screen;
 512     size_t size = (size_t)vertex_size * (size_t)count;
 513
 514     if (size + r300render->vbo_offset > r300render->vbo_size)
 515     {
 516         pipe_buffer_reference(&r300->vbo, NULL);
 517         r300render->vbo = pipe_buffer_create(screen,
 518                                              64,
 519                                              PIPE_BUFFER_USAGE_VERTEX,
 520                                              R300_MAX_VBO_SIZE);
 521         r300render->vbo_offset = 0;
 522         r300render->vbo_size = R300_MAX_VBO_SIZE;
 523     }
 524
 525     r300render->vertex_size = vertex_size;
 526     r300->vbo = r300render->vbo;
 527     r300->vbo_offset = r300render->vbo_offset;
 528
 529     return (r300render->vbo) ? TRUE : FALSE;
 530 }
 531
 532 static void* r300_render_map_vertices(struct vbuf_render* render)
 533 {
 534     struct r300_render* r300render = r300_render(render);
 535     struct pipe_screen* screen = r300render->r300->context.screen;
 536
 537     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 538                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 539
 540     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 541 }
 542
 543 static void r300_render_unmap_vertices(struct vbuf_render* render,
 544                                              ushort min,
 545                                              ushort max)
 546 {
 547     struct r300_render* r300render = r300_render(render);
 548     struct pipe_screen* screen = r300render->r300->context.screen;
 549     CS_LOCALS(r300render->r300);
 550     BEGIN_CS(2);
 551     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 552     END_CS;
 553
 554     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 555                                     r300render->vertex_size * (max + 1));
 556     pipe_buffer_unmap(screen, r300render->vbo);
 557 }
 558
 559 static void r300_render_release_vertices(struct vbuf_render* render)
 560 {
 561     struct r300_render* r300render = r300_render(render);
 562
 563     r300render->vbo_offset += r300render->vbo_max_used;
 564     r300render->vbo_max_used = 0;
 565 }
 566
 567 static boolean r300_render_set_primitive(struct vbuf_render* render,
 568                                                unsigned prim)
 569 {
 570     struct r300_render* r300render = r300_render(render);
 571
 572     r300render->prim = prim;
 573     r300render->hwprim = r300_translate_primitive(prim);
 574
 575     return TRUE;
 576 }
 577
 578 static void r300_render_draw_arrays(struct vbuf_render* render,
 579                                           unsigned start,
 580                                           unsigned count)
 581 {
 582     struct r300_render* r300render = r300_render(render);
 583     struct r300_context* r300 = r300render->r300;
 584
 585     CS_LOCALS(r300);
 586
 587     r300_emit_dirty_state(r300);
 588
 589     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 590
 591     BEGIN_CS(2);
 592     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 593     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 594            r300render->hwprim);
 595     END_CS;
 596 }
 597
 598 static void r300_render_draw(struct vbuf_render* render,
 599                                    const ushort* indices,
 600                                    uint count)
 601 {
 602     struct r300_render* r300render = r300_render(render);
 603     struct r300_context* r300 = r300render->r300;
 604     int i;
 605
 606     CS_LOCALS(r300);
 607
 608     r300_emit_dirty_state(r300);
 609
 610     BEGIN_CS(2 + (count+1)/2);
 611     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 612     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 613            r300render->hwprim);
 614     for (i = 0; i < count-1; i += 2) {
 615         OUT_CS(indices[i+1] << 16 | indices[i]);
 616     }
 617     if (count % 2) {
 618         OUT_CS(indices[count-1]);
 619     }
 620     END_CS;
 621 }
 622
 623 static void r300_render_destroy(struct vbuf_render* render)
 624 {
 625     FREE(render);
 626 }
 627
 628 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 629 {
 630     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 631
 632     r300render->r300 = r300;
 633
 634     /* XXX find real numbers plz */
 635     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 636     r300render->base.max_indices = 16 * 1024;
 637
 638     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 639     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 640     r300render->base.map_vertices = r300_render_map_vertices;
 641     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 642     r300render->base.set_primitive = r300_render_set_primitive;
 643     r300render->base.draw = r300_render_draw;
 644     r300render->base.draw_arrays = r300_render_draw_arrays;
 645     r300render->base.release_vertices = r300_render_release_vertices;
 646     r300render->base.destroy = r300_render_destroy;
 647
 648     r300render->vbo = NULL;
 649     r300render->vbo_size = 0;
 650     r300render->vbo_offset = 0;
 651
 652     return &r300render->base;
 653 }
 654
 655 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 656 {
 657     struct vbuf_render* render;
 658     struct draw_stage* stage;
 659
 660     render = r300_render_create(r300);
 661
 662     if (!render) {
 663         return NULL;
 664     }
 665
 666     stage = draw_vbuf_stage(r300->draw, render);
 667
 668     if (!stage) {
 669         render->destroy(render);
 670         return NULL;
 671     }
 672
 673     draw_set_render(r300->draw, render);
 674
 675     return stage;
 676 }