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 "util/u_inlines.h"
  30
  31 #include "util/u_format.h"
  32 #include "util/u_memory.h"
  33 #include "util/u_prim.h"
  34
  35 #include "r300_cs.h"
  36 #include "r300_context.h"
  37 #include "r300_emit.h"
  38 #include "r300_reg.h"
  39 #include "r300_render.h"
  40 #include "r300_state_derived.h"
  41
  42 /* r300_render: Vertex and index buffer primitive emission. */
  43 #define R300_MAX_VBO_SIZE  (1024 * 1024)
  44
  45 /* XXX The DRM rejects VAP_ALT_NUM_VERTICES.. */
  46 //#define ENABLE_ALT_NUM_VERTS
  47
  48 uint32_t r300_translate_primitive(unsigned prim)
  49 {
  50     switch (prim) {
  51         case PIPE_PRIM_POINTS:
  52             return R300_VAP_VF_CNTL__PRIM_POINTS;
  53         case PIPE_PRIM_LINES:
  54             return R300_VAP_VF_CNTL__PRIM_LINES;
  55         case PIPE_PRIM_LINE_LOOP:
  56             return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
  57         case PIPE_PRIM_LINE_STRIP:
  58             return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
  59         case PIPE_PRIM_TRIANGLES:
  60             return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
  61         case PIPE_PRIM_TRIANGLE_STRIP:
  62             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
  63         case PIPE_PRIM_TRIANGLE_FAN:
  64             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
  65         case PIPE_PRIM_QUADS:
  66             return R300_VAP_VF_CNTL__PRIM_QUADS;
  67         case PIPE_PRIM_QUAD_STRIP:
  68             return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
  69         case PIPE_PRIM_POLYGON:
  70             return R300_VAP_VF_CNTL__PRIM_POLYGON;
  71         default:
  72             return 0;
  73     }
  74 }
  75
  76 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
  77                                             unsigned mode)
  78 {
  79     struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
  80     uint32_t color_control = rs->color_control;
  81
  82     /* By default (see r300_state.c:r300_create_rs_state) color_control is
  83      * initialized to provoking the first vertex.
  84      *
  85      * Triangle fans must be reduced to the second vertex, not the first, in
  86      * Gallium flatshade-first mode, as per the GL spec.
  87      * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
  88      *
  89      * Quads never provoke correctly in flatshade-first mode. The first
  90      * vertex is never considered as provoking, so only the second, third,
  91      * and fourth vertices can be selected, and both "third" and "last" modes
  92      * select the fourth vertex. This is probably due to D3D lacking quads.
  93      *
  94      * Similarly, polygons reduce to the first, not the last, vertex, when in
  95      * "last" mode, and all other modes start from the second vertex.
  96      *
  97      * ~ C.
  98      */
  99
 100     if (rs->rs.flatshade_first) {
 101         switch (mode) {
 102             case PIPE_PRIM_TRIANGLE_FAN:
 103                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
 104                 break;
 105             case PIPE_PRIM_QUADS:
 106             case PIPE_PRIM_QUAD_STRIP:
 107             case PIPE_PRIM_POLYGON:
 108                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 109                 break;
 110             default:
 111                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
 112                 break;
 113         }
 114     } else {
 115         color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 116     }
 117
 118     return color_control;
 119 }
 120
 121 /* Check if the requested number of dwords is available in the CS and
 122  * if not, flush. Return TRUE if the flush occured. */
 123 static boolean r300_reserve_cs_space(struct r300_context *r300,
 124                                      unsigned dwords)
 125 {
 126     if (!r300->winsys->check_cs(r300->winsys, dwords)) {
 127         r300->context.flush(&r300->context, 0, NULL);
 128         return TRUE;
 129     }
 130     return FALSE;
 131 }
 132
 133 static boolean immd_is_good_idea(struct r300_context *r300,
 134                                       unsigned count)
 135 {
 136     return count <= 4;
 137 }
 138
 139 static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 140                                             unsigned mode,
 141                                             unsigned start,
 142                                             unsigned count)
 143 {
 144     struct pipe_vertex_element* velem;
 145     struct pipe_vertex_buffer* vbuf;
 146     unsigned vertex_element_count = r300->velems->count;
 147     unsigned i, v, vbi, dw, elem_offset, dwords;
 148
 149     /* Size of the vertex, in dwords. */
 150     unsigned vertex_size = 0;
 151
 152     /* Offsets of the attribute, in dwords, from the start of the vertex. */
 153     unsigned offset[PIPE_MAX_ATTRIBS];
 154
 155     /* Size of the vertex element, in dwords. */
 156     unsigned size[PIPE_MAX_ATTRIBS];
 157
 158     /* Stride to the same attrib in the next vertex in the vertex buffer,
 159      * in dwords. */
 160     unsigned stride[PIPE_MAX_ATTRIBS] = {0};
 161
 162     /* Mapped vertex buffers. */
 163     uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
 164
 165     CS_LOCALS(r300);
 166
 167     /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
 168     for (i = 0; i < vertex_element_count; i++) {
 169         velem = &r300->velems->velem[i];
 170         offset[i] = velem->src_offset / 4;
 171         size[i] = util_format_get_blocksize(velem->src_format) / 4;
 172         vertex_size += size[i];
 173         vbi = velem->vertex_buffer_index;
 174
 175         /* Map the buffer. */
 176         if (!map[vbi]) {
 177             vbuf = &r300->vertex_buffer[vbi];
 178             map[vbi] = (uint32_t*)pipe_buffer_map(r300->context.screen,
 179                                                   vbuf->buffer,
 180                                                   PIPE_BUFFER_USAGE_CPU_READ);
 181             map[vbi] += vbuf->buffer_offset / 4;
 182             stride[vbi] = vbuf->stride / 4;
 183         }
 184     }
 185
 186     dwords = 9 + count * vertex_size;
 187
 188     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 189     r300_emit_buffer_validate(r300, FALSE, NULL);
 190     r300_emit_dirty_state(r300);
 191
 192     BEGIN_CS(dwords);
 193     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 194             r300_provoking_vertex_fixes(r300, mode));
 195     OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
 196     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 197     OUT_CS(count - 1);
 198     OUT_CS(0);
 199     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
 200     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
 201             r300_translate_primitive(mode));
 202
 203     /* Emit vertices. */
 204     for (v = 0; v < count; v++) {
 205         for (i = 0; i < vertex_element_count; i++) {
 206             velem = &r300->velems->velem[i];
 207             vbi = velem->vertex_buffer_index;
 208             elem_offset = offset[i] + stride[vbi] * (v + start);
 209
 210             for (dw = 0; dw < size[i]; dw++) {
 211                 OUT_CS(map[vbi][elem_offset + dw]);
 212             }
 213         }
 214     }
 215     END_CS;
 216
 217     /* Unmap buffers. */
 218     for (i = 0; i < vertex_element_count; i++) {
 219         vbi = r300->velems->velem[i].vertex_buffer_index;
 220
 221         if (map[vbi]) {
 222             vbuf = &r300->vertex_buffer[vbi];
 223             pipe_buffer_unmap(r300->context.screen, vbuf->buffer);
 224             map[vbi] = NULL;
 225         }
 226     }
 227 }
 228
 229 static void r300_emit_draw_arrays(struct r300_context *r300,
 230                                   unsigned mode,
 231                                   unsigned count)
 232 {
 233 #if defined(ENABLE_ALT_NUM_VERTS)
 234     boolean alt_num_verts = count > 65535;
 235 #else
 236     boolean alt_num_verts = FALSE;
 237 #endif
 238     CS_LOCALS(r300);
 239
 240     if (alt_num_verts) {
 241         assert(count < (1 << 24));
 242         BEGIN_CS(9);
 243         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
 244     } else {
 245         BEGIN_CS(7);
 246     }
 247     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 248             r300_provoking_vertex_fixes(r300, mode));
 249     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 250     OUT_CS(count - 1);
 251     OUT_CS(0);
 252     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 253     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 254            r300_translate_primitive(mode) |
 255            (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 256     END_CS;
 257 }
 258
 259 static void r300_emit_draw_elements(struct r300_context *r300,
 260                                     struct pipe_buffer* indexBuffer,
 261                                     unsigned indexSize,
 262                                     unsigned minIndex,
 263                                     unsigned maxIndex,
 264                                     unsigned mode,
 265                                     unsigned start,
 266                                     unsigned count)
 267 {
 268     uint32_t count_dwords;
 269     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
 270 #if defined(ENABLE_ALT_NUM_VERTS)
 271     boolean alt_num_verts = count > 65535;
 272 #else
 273     boolean alt_num_verts = FALSE;
 274 #endif
 275     CS_LOCALS(r300);
 276
 277     assert((start * indexSize)  % 4 == 0);
 278     assert(count < (1 << 24));
 279
 280     DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
 281         count, minIndex, maxIndex);
 282
 283     maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
 284
 285     if (alt_num_verts) {
 286         BEGIN_CS(15);
 287         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
 288     } else {
 289         BEGIN_CS(13);
 290     }
 291     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 292             r300_provoking_vertex_fixes(r300, mode));
 293     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 294     OUT_CS(maxIndex);
 295     OUT_CS(minIndex);
 296     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 297     if (indexSize == 4) {
 298         count_dwords = count;
 299         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 300                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 301                r300_translate_primitive(mode) |
 302                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 303     } else {
 304         count_dwords = (count + 1) / 2;
 305         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 306                r300_translate_primitive(mode) |
 307                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 308     }
 309
 310     /* INDX_BUFFER is a truly special packet3.
 311      * Unlike most other packet3, where the offset is after the count,
 312      * the order is reversed, so the relocation ends up carrying the
 313      * size of the indexbuf instead of the offset.
 314      */
 315     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 316     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 317            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 318     OUT_CS(offset_dwords << 2);
 319     OUT_CS_RELOC(indexBuffer, count_dwords,
 320         RADEON_GEM_DOMAIN_GTT, 0, 0);
 321
 322     END_CS;
 323 }
 324
 325 static void r300_shorten_ubyte_elts(struct r300_context* r300,
 326                                     struct pipe_buffer** elts,
 327                                     unsigned count)
 328 {
 329     struct pipe_screen* screen = r300->context.screen;
 330     struct pipe_buffer* new_elts;
 331     unsigned char *in_map;
 332     unsigned short *out_map;
 333     unsigned i;
 334
 335     new_elts = screen->buffer_create(screen, 32,
 336                                      PIPE_BUFFER_USAGE_INDEX |
 337                                      PIPE_BUFFER_USAGE_CPU_WRITE |
 338                                      PIPE_BUFFER_USAGE_GPU_READ,
 339                                      2 * count);
 340
 341     in_map = pipe_buffer_map(screen, *elts, PIPE_BUFFER_USAGE_CPU_READ);
 342     out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE);
 343
 344     for (i = 0; i < count; i++) {
 345         *out_map = (unsigned short)*in_map;
 346         in_map++;
 347         out_map++;
 348     }
 349
 350     pipe_buffer_unmap(screen, *elts);
 351     pipe_buffer_unmap(screen, new_elts);
 352
 353     *elts = new_elts;
 354 }
 355
 356 /* This is the fast-path drawing & emission for HW TCL. */
 357 void r300_draw_range_elements(struct pipe_context* pipe,
 358                               struct pipe_buffer* indexBuffer,
 359                               unsigned indexSize,
 360                               unsigned minIndex,
 361                               unsigned maxIndex,
 362                               unsigned mode,
 363                               unsigned start,
 364                               unsigned count)
 365 {
 366     struct r300_context* r300 = r300_context(pipe);
 367     struct pipe_buffer* orgIndexBuffer = indexBuffer;
 368 #if defined(ENABLE_ALT_NUM_VERTS)
 369     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 370                             count > 65536;
 371 #else
 372     boolean alt_num_verts = FALSE;
 373 #endif
 374     unsigned short_count;
 375
 376     if (!u_trim_pipe_prim(mode, &count)) {
 377         return;
 378     }
 379
 380     if (indexSize == 1) {
 381         r300_shorten_ubyte_elts(r300, &indexBuffer, count);
 382         indexSize = 2;
 383     }
 384
 385     r300_update_derived_state(r300);
 386
 387     /* 128 dwords for emit_aos and emit_draw_elements */
 388     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 389     r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 390     r300_emit_dirty_state(r300);
 391     r300_emit_aos(r300, 0);
 392
 393     if (alt_num_verts || count <= 65535) {
 394         r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 395                                 maxIndex, mode, start, count);
 396     } else {
 397         do {
 398             short_count = MIN2(count, 65534);
 399             r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 400                                     maxIndex, mode, start, short_count);
 401
 402             start += short_count;
 403             count -= short_count;
 404
 405             /* 16 spare dwords are enough for emit_draw_elements. */
 406             if (count && r300_reserve_cs_space(r300, 16)) {
 407                 r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 408                 r300_emit_dirty_state(r300);
 409                 r300_emit_aos(r300, 0);
 410             }
 411         } while (count);
 412     }
 413
 414     if (indexBuffer != orgIndexBuffer) {
 415         pipe->screen->buffer_destroy(indexBuffer);
 416     }
 417 }
 418
 419 /* Simple helpers for context setup. Should probably be moved to util. */
 420 void r300_draw_elements(struct pipe_context* pipe,
 421                         struct pipe_buffer* indexBuffer,
 422                         unsigned indexSize, unsigned mode,
 423                         unsigned start, unsigned count)
 424 {
 425     struct r300_context *r300 = r300_context(pipe);
 426
 427     pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0,
 428                               r300->vertex_buffer_max_index,
 429                               mode, start, count);
 430 }
 431
 432 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 433                       unsigned start, unsigned count)
 434 {
 435     struct r300_context* r300 = r300_context(pipe);
 436 #if defined(ENABLE_ALT_NUM_VERTS)
 437     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 438                             count > 65536;
 439 #else
 440     boolean alt_num_verts = FALSE;
 441 #endif
 442     unsigned short_count;
 443
 444     if (!u_trim_pipe_prim(mode, &count)) {
 445         return;
 446     }
 447
 448     r300_update_derived_state(r300);
 449
 450     if (immd_is_good_idea(r300, count)) {
 451         r300_emit_draw_arrays_immediate(r300, mode, start, count);
 452     } else {
 453         /* Make sure there are at least 128 spare dwords in the command buffer.
 454          * (most of it being consumed by emit_aos) */
 455         r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 456         r300_emit_buffer_validate(r300, TRUE, NULL);
 457         r300_emit_dirty_state(r300);
 458
 459         if (alt_num_verts || count <= 65535) {
 460             r300_emit_aos(r300, start);
 461             r300_emit_draw_arrays(r300, mode, count);
 462         } else {
 463             do {
 464                 short_count = MIN2(count, 65535);
 465                 r300_emit_aos(r300, start);
 466                 r300_emit_draw_arrays(r300, mode, short_count);
 467
 468                 start += short_count;
 469                 count -= short_count;
 470
 471                 /* Again, we emit both AOS and draw_arrays so there should be
 472                  * at least 128 spare dwords. */
 473                 if (count && r300_reserve_cs_space(r300, 128)) {
 474                     r300_emit_buffer_validate(r300, TRUE, NULL);
 475                     r300_emit_dirty_state(r300);
 476                 }
 477             } while (count);
 478         }
 479     }
 480 }
 481
 482 /****************************************************************************
 483  * The rest of this file is for SW TCL rendering only. Please be polite and *
 484  * keep these functions separated so that they are easier to locate. ~C.    *
 485  ***************************************************************************/
 486
 487 /* SW TCL arrays, using Draw. */
 488 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 489                                unsigned mode,
 490                                unsigned start,
 491                                unsigned count)
 492 {
 493     struct r300_context* r300 = r300_context(pipe);
 494     int i;
 495
 496     if (!u_trim_pipe_prim(mode, &count)) {
 497         return;
 498     }
 499
 500     for (i = 0; i < r300->vertex_buffer_count; i++) {
 501         void* buf = pipe_buffer_map(pipe->screen,
 502                                     r300->vertex_buffer[i].buffer,
 503                                     PIPE_BUFFER_USAGE_CPU_READ);
 504         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 505     }
 506
 507     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 508
 509     draw_set_mapped_constant_buffer(r300->draw,
 510                                     PIPE_SHADER_VERTEX,
 511                                     0,
 512                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 513                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 514                 (sizeof(float) * 4));
 515
 516     draw_arrays(r300->draw, mode, start, count);
 517
 518     for (i = 0; i < r300->vertex_buffer_count; i++) {
 519         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 520         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 521     }
 522 }
 523
 524 /* SW TCL elements, using Draw. */
 525 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 526                                        struct pipe_buffer* indexBuffer,
 527                                        unsigned indexSize,
 528                                        unsigned minIndex,
 529                                        unsigned maxIndex,
 530                                        unsigned mode,
 531                                        unsigned start,
 532                                        unsigned count)
 533 {
 534     struct r300_context* r300 = r300_context(pipe);
 535     int i;
 536     void* indices;
 537
 538     if (!u_trim_pipe_prim(mode, &count)) {
 539         return;
 540     }
 541
 542     for (i = 0; i < r300->vertex_buffer_count; i++) {
 543         void* buf = pipe_buffer_map(pipe->screen,
 544                                     r300->vertex_buffer[i].buffer,
 545                                     PIPE_BUFFER_USAGE_CPU_READ);
 546         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 547     }
 548
 549     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 550                               PIPE_BUFFER_USAGE_CPU_READ);
 551     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 552                                          minIndex, maxIndex, indices);
 553
 554     draw_set_mapped_constant_buffer(r300->draw,
 555                                     PIPE_SHADER_VERTEX,
 556                                     0,
 557             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 558             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 559                 (sizeof(float) * 4));
 560
 561     draw_arrays(r300->draw, mode, start, count);
 562
 563     for (i = 0; i < r300->vertex_buffer_count; i++) {
 564         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 565         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 566     }
 567
 568     pipe_buffer_unmap(pipe->screen, indexBuffer);
 569     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 570                                          start + count - 1, NULL);
 571 }
 572
 573 /* Object for rendering using Draw. */
 574 struct r300_render {
 575     /* Parent class */
 576     struct vbuf_render base;
 577
 578     /* Pipe context */
 579     struct r300_context* r300;
 580
 581     /* Vertex information */
 582     size_t vertex_size;
 583     unsigned prim;
 584     unsigned hwprim;
 585
 586     /* VBO */
 587     struct pipe_buffer* vbo;
 588     size_t vbo_size;
 589     size_t vbo_offset;
 590     size_t vbo_max_used;
 591     void * vbo_ptr;
 592 };
 593
 594 static INLINE struct r300_render*
 595 r300_render(struct vbuf_render* render)
 596 {
 597     return (struct r300_render*)render;
 598 }
 599
 600 static const struct vertex_info*
 601 r300_render_get_vertex_info(struct vbuf_render* render)
 602 {
 603     struct r300_render* r300render = r300_render(render);
 604     struct r300_context* r300 = r300render->r300;
 605
 606     r300_update_derived_state(r300);
 607
 608     return &r300->vertex_info;
 609 }
 610
 611 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 612                                                    ushort vertex_size,
 613                                                    ushort count)
 614 {
 615     struct r300_render* r300render = r300_render(render);
 616     struct r300_context* r300 = r300render->r300;
 617     struct pipe_screen* screen = r300->context.screen;
 618     size_t size = (size_t)vertex_size * (size_t)count;
 619
 620     if (size + r300render->vbo_offset > r300render->vbo_size)
 621     {
 622         pipe_buffer_reference(&r300->vbo, NULL);
 623         r300render->vbo = pipe_buffer_create(screen,
 624                                              64,
 625                                              PIPE_BUFFER_USAGE_VERTEX,
 626                                              R300_MAX_VBO_SIZE);
 627         r300render->vbo_offset = 0;
 628         r300render->vbo_size = R300_MAX_VBO_SIZE;
 629     }
 630
 631     r300render->vertex_size = vertex_size;
 632     r300->vbo = r300render->vbo;
 633     r300->vbo_offset = r300render->vbo_offset;
 634
 635     return (r300render->vbo) ? TRUE : FALSE;
 636 }
 637
 638 static void* r300_render_map_vertices(struct vbuf_render* render)
 639 {
 640     struct r300_render* r300render = r300_render(render);
 641     struct pipe_screen* screen = r300render->r300->context.screen;
 642
 643     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 644                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 645
 646     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 647 }
 648
 649 static void r300_render_unmap_vertices(struct vbuf_render* render,
 650                                              ushort min,
 651                                              ushort max)
 652 {
 653     struct r300_render* r300render = r300_render(render);
 654     struct pipe_screen* screen = r300render->r300->context.screen;
 655     CS_LOCALS(r300render->r300);
 656     BEGIN_CS(2);
 657     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 658     END_CS;
 659
 660     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 661                                     r300render->vertex_size * (max + 1));
 662     pipe_buffer_unmap(screen, r300render->vbo);
 663 }
 664
 665 static void r300_render_release_vertices(struct vbuf_render* render)
 666 {
 667     struct r300_render* r300render = r300_render(render);
 668
 669     r300render->vbo_offset += r300render->vbo_max_used;
 670     r300render->vbo_max_used = 0;
 671 }
 672
 673 static boolean r300_render_set_primitive(struct vbuf_render* render,
 674                                                unsigned prim)
 675 {
 676     struct r300_render* r300render = r300_render(render);
 677
 678     r300render->prim = prim;
 679     r300render->hwprim = r300_translate_primitive(prim);
 680
 681     return TRUE;
 682 }
 683
 684 static void r300_render_draw_arrays(struct vbuf_render* render,
 685                                           unsigned start,
 686                                           unsigned count)
 687 {
 688     struct r300_render* r300render = r300_render(render);
 689     struct r300_context* r300 = r300render->r300;
 690
 691     CS_LOCALS(r300);
 692
 693     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2);
 694     r300_emit_buffer_validate(r300, FALSE, NULL);
 695     r300_emit_dirty_state(r300);
 696
 697     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 698
 699     BEGIN_CS(2);
 700     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 701     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 702            r300render->hwprim);
 703     END_CS;
 704 }
 705
 706 static void r300_render_draw(struct vbuf_render* render,
 707                                    const ushort* indices,
 708                                    uint count)
 709 {
 710     struct r300_render* r300render = r300_render(render);
 711     struct r300_context* r300 = r300render->r300;
 712     int i;
 713     unsigned dwords = 2 + (count+1)/2;
 714
 715     CS_LOCALS(r300);
 716
 717     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 718     r300_emit_buffer_validate(r300, FALSE, NULL);
 719     r300_emit_dirty_state(r300);
 720
 721     BEGIN_CS(dwords);
 722     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 723     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 724            r300render->hwprim);
 725     for (i = 0; i < count-1; i += 2) {
 726         OUT_CS(indices[i+1] << 16 | indices[i]);
 727     }
 728     if (count % 2) {
 729         OUT_CS(indices[count-1]);
 730     }
 731     END_CS;
 732 }
 733
 734 static void r300_render_destroy(struct vbuf_render* render)
 735 {
 736     FREE(render);
 737 }
 738
 739 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 740 {
 741     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 742
 743     r300render->r300 = r300;
 744
 745     /* XXX find real numbers plz */
 746     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 747     r300render->base.max_indices = 16 * 1024;
 748
 749     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 750     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 751     r300render->base.map_vertices = r300_render_map_vertices;
 752     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 753     r300render->base.set_primitive = r300_render_set_primitive;
 754     r300render->base.draw = r300_render_draw;
 755     r300render->base.draw_arrays = r300_render_draw_arrays;
 756     r300render->base.release_vertices = r300_render_release_vertices;
 757     r300render->base.destroy = r300_render_destroy;
 758
 759     r300render->vbo = NULL;
 760     r300render->vbo_size = 0;
 761     r300render->vbo_offset = 0;
 762
 763     return &r300render->base;
 764 }
 765
 766 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 767 {
 768     struct vbuf_render* render;
 769     struct draw_stage* stage;
 770
 771     render = r300_render_create(r300);
 772
 773     if (!render) {
 774         return NULL;
 775     }
 776
 777     stage = draw_vbuf_stage(r300->draw, render);
 778
 779     if (!stage) {
 780         render->destroy(render);
 781         return NULL;
 782     }
 783
 784     draw_set_render(r300->draw, render);
 785
 786     return stage;
 787 }