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_upload_mgr.h"
  34 #include "util/u_prim.h"
  35
  36 #include "r300_cs.h"
  37 #include "r300_context.h"
  38 #include "r300_screen_buffer.h"
  39 #include "r300_emit.h"
  40 #include "r300_reg.h"
  41 #include "r300_render.h"
  42 #include "r300_state_derived.h"
  43
  44 /* r300_render: Vertex and index buffer primitive emission. */
  45 #define R300_MAX_VBO_SIZE  (1024 * 1024)
  46
  47 /* XXX The DRM rejects VAP_ALT_NUM_VERTICES.. */
  48 //#define ENABLE_ALT_NUM_VERTS
  49
  50 uint32_t r300_translate_primitive(unsigned prim)
  51 {
  52     switch (prim) {
  53         case PIPE_PRIM_POINTS:
  54             return R300_VAP_VF_CNTL__PRIM_POINTS;
  55         case PIPE_PRIM_LINES:
  56             return R300_VAP_VF_CNTL__PRIM_LINES;
  57         case PIPE_PRIM_LINE_LOOP:
  58             return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
  59         case PIPE_PRIM_LINE_STRIP:
  60             return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
  61         case PIPE_PRIM_TRIANGLES:
  62             return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
  63         case PIPE_PRIM_TRIANGLE_STRIP:
  64             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
  65         case PIPE_PRIM_TRIANGLE_FAN:
  66             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
  67         case PIPE_PRIM_QUADS:
  68             return R300_VAP_VF_CNTL__PRIM_QUADS;
  69         case PIPE_PRIM_QUAD_STRIP:
  70             return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
  71         case PIPE_PRIM_POLYGON:
  72             return R300_VAP_VF_CNTL__PRIM_POLYGON;
  73         default:
  74             return 0;
  75     }
  76 }
  77
  78 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
  79                                             unsigned mode)
  80 {
  81     struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
  82     uint32_t color_control = rs->color_control;
  83
  84     /* By default (see r300_state.c:r300_create_rs_state) color_control is
  85      * initialized to provoking the first vertex.
  86      *
  87      * Triangle fans must be reduced to the second vertex, not the first, in
  88      * Gallium flatshade-first mode, as per the GL spec.
  89      * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
  90      *
  91      * Quads never provoke correctly in flatshade-first mode. The first
  92      * vertex is never considered as provoking, so only the second, third,
  93      * and fourth vertices can be selected, and both "third" and "last" modes
  94      * select the fourth vertex. This is probably due to D3D lacking quads.
  95      *
  96      * Similarly, polygons reduce to the first, not the last, vertex, when in
  97      * "last" mode, and all other modes start from the second vertex.
  98      *
  99      * ~ C.
 100      */
 101
 102     if (rs->rs.flatshade_first) {
 103         switch (mode) {
 104             case PIPE_PRIM_TRIANGLE_FAN:
 105                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
 106                 break;
 107             case PIPE_PRIM_QUADS:
 108             case PIPE_PRIM_QUAD_STRIP:
 109             case PIPE_PRIM_POLYGON:
 110                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 111                 break;
 112             default:
 113                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
 114                 break;
 115         }
 116     } else {
 117         color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 118     }
 119
 120     return color_control;
 121 }
 122
 123 /* Check if the requested number of dwords is available in the CS and
 124  * if not, flush. Return TRUE if the flush occured. */
 125 static boolean r300_reserve_cs_space(struct r300_context *r300,
 126                                      unsigned dwords)
 127 {
 128     if (!r300->rws->check_cs(r300->rws, dwords)) {
 129         r300->context.flush(&r300->context, 0, NULL);
 130         return TRUE;
 131     }
 132     return FALSE;
 133 }
 134
 135 static boolean immd_is_good_idea(struct r300_context *r300,
 136                                  unsigned count)
 137 {
 138     struct pipe_vertex_element* velem;
 139     struct pipe_vertex_buffer* vbuf;
 140     boolean checked[PIPE_MAX_ATTRIBS] = {0};
 141     unsigned vertex_element_count = r300->velems->count;
 142     unsigned i, vbi;
 143
 144     if (count > 4) {
 145         return FALSE;
 146     }
 147
 148     /* We shouldn't map buffers referenced by CS, busy buffers,
 149      * and ones placed in VRAM. */
 150     /* XXX Check for VRAM buffers. */
 151     for (i = 0; i < vertex_element_count; i++) {
 152         velem = &r300->velems->velem[i];
 153         vbi = velem->vertex_buffer_index;
 154
 155         if (!checked[vbi]) {
 156             vbuf = &r300->vertex_buffer[vbi];
 157
 158             if (r300_buffer_is_referenced(r300,
 159                                           vbuf->buffer)) {
 160                 /* It's a very bad idea to map it... */
 161                 return FALSE;
 162             }
 163             checked[vbi] = TRUE;
 164         }
 165     }
 166     return TRUE;
 167 }
 168
 169 static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 170                                             unsigned mode,
 171                                             unsigned start,
 172                                             unsigned count)
 173 {
 174     struct pipe_vertex_element* velem;
 175     struct pipe_vertex_buffer* vbuf;
 176     unsigned vertex_element_count = r300->velems->count;
 177     unsigned i, v, vbi, dw, elem_offset, dwords;
 178
 179     /* Size of the vertex, in dwords. */
 180     unsigned vertex_size = 0;
 181
 182     /* Offsets of the attribute, in dwords, from the start of the vertex. */
 183     unsigned offset[PIPE_MAX_ATTRIBS];
 184
 185     /* Size of the vertex element, in dwords. */
 186     unsigned size[PIPE_MAX_ATTRIBS];
 187
 188     /* Stride to the same attrib in the next vertex in the vertex buffer,
 189      * in dwords. */
 190     unsigned stride[PIPE_MAX_ATTRIBS] = {0};
 191
 192     /* Mapped vertex buffers. */
 193     uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
 194
 195     CS_LOCALS(r300);
 196
 197     /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
 198     for (i = 0; i < vertex_element_count; i++) {
 199         velem = &r300->velems->velem[i];
 200         offset[i] = velem->src_offset / 4;
 201         size[i] = util_format_get_blocksize(velem->src_format) / 4;
 202         vertex_size += size[i];
 203         vbi = velem->vertex_buffer_index;
 204
 205         /* Map the buffer. */
 206         if (!map[vbi]) {
 207             vbuf = &r300->vertex_buffer[vbi];
 208             map[vbi] = (uint32_t*)pipe_buffer_map(r300->context.screen,
 209                                                   vbuf->buffer,
 210                                                   PIPE_BUFFER_USAGE_CPU_READ);
 211             map[vbi] += vbuf->buffer_offset / 4;
 212             stride[vbi] = vbuf->stride / 4;
 213         }
 214     }
 215
 216     dwords = 9 + count * vertex_size;
 217
 218     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 219     r300_emit_buffer_validate(r300, FALSE, NULL);
 220     r300_emit_dirty_state(r300);
 221
 222     BEGIN_CS(dwords);
 223     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 224             r300_provoking_vertex_fixes(r300, mode));
 225     OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
 226     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 227     OUT_CS(count - 1);
 228     OUT_CS(0);
 229     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
 230     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
 231             r300_translate_primitive(mode));
 232
 233     /* Emit vertices. */
 234     for (v = 0; v < count; v++) {
 235         for (i = 0; i < vertex_element_count; i++) {
 236             velem = &r300->velems->velem[i];
 237             vbi = velem->vertex_buffer_index;
 238             elem_offset = offset[i] + stride[vbi] * (v + start);
 239
 240             for (dw = 0; dw < size[i]; dw++) {
 241                 OUT_CS(map[vbi][elem_offset + dw]);
 242             }
 243         }
 244     }
 245     END_CS;
 246
 247     /* Unmap buffers. */
 248     for (i = 0; i < vertex_element_count; i++) {
 249         vbi = r300->velems->velem[i].vertex_buffer_index;
 250
 251         if (map[vbi]) {
 252             vbuf = &r300->vertex_buffer[vbi];
 253             pipe_buffer_unmap(r300->context.screen, vbuf->buffer);
 254             map[vbi] = NULL;
 255         }
 256     }
 257 }
 258
 259 static void r300_emit_draw_arrays(struct r300_context *r300,
 260                                   unsigned mode,
 261                                   unsigned count)
 262 {
 263 #if defined(ENABLE_ALT_NUM_VERTS)
 264     boolean alt_num_verts = count > 65535;
 265 #else
 266     boolean alt_num_verts = FALSE;
 267 #endif
 268     CS_LOCALS(r300);
 269
 270     if (alt_num_verts) {
 271         assert(count < (1 << 24));
 272         BEGIN_CS(9);
 273         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
 274     } else {
 275         BEGIN_CS(7);
 276     }
 277     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 278             r300_provoking_vertex_fixes(r300, mode));
 279     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 280     OUT_CS(count - 1);
 281     OUT_CS(0);
 282     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 283     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 284            r300_translate_primitive(mode) |
 285            (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 286     END_CS;
 287 }
 288
 289 static void r300_emit_draw_elements(struct r300_context *r300,
 290                                     struct pipe_buffer* indexBuffer,
 291                                     unsigned indexSize,
 292                                     unsigned minIndex,
 293                                     unsigned maxIndex,
 294                                     unsigned mode,
 295                                     unsigned start,
 296                                     unsigned count)
 297 {
 298     uint32_t count_dwords;
 299     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
 300 #if defined(ENABLE_ALT_NUM_VERTS)
 301     boolean alt_num_verts = count > 65535;
 302 #else
 303     boolean alt_num_verts = FALSE;
 304 #endif
 305     CS_LOCALS(r300);
 306
 307     assert((start * indexSize) % 4 == 0);
 308     assert(count < (1 << 24));
 309
 310     maxIndex = MIN3(maxIndex, r300->vertex_buffer_max_index, count - minIndex);
 311
 312     DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
 313         count, minIndex, maxIndex);
 314
 315     if (alt_num_verts) {
 316         BEGIN_CS(15);
 317         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
 318     } else {
 319         BEGIN_CS(13);
 320     }
 321     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 322             r300_provoking_vertex_fixes(r300, mode));
 323     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 324     OUT_CS(maxIndex);
 325     OUT_CS(minIndex);
 326     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 327     if (indexSize == 4) {
 328         count_dwords = count;
 329         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 330                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 331                r300_translate_primitive(mode) |
 332                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 333     } else {
 334         count_dwords = (count + 1) / 2;
 335         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 336                r300_translate_primitive(mode) |
 337                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 338     }
 339
 340     /* INDX_BUFFER is a truly special packet3.
 341      * Unlike most other packet3, where the offset is after the count,
 342      * the order is reversed, so the relocation ends up carrying the
 343      * size of the indexbuf instead of the offset.
 344      */
 345     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 346     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 347            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 348     OUT_CS(offset_dwords << 2);
 349     OUT_CS_BUF_RELOC(indexBuffer, count_dwords,
 350                      RADEON_GEM_DOMAIN_GTT, 0, 0);
 351
 352     END_CS;
 353 }
 354
 355 static void r300_shorten_ubyte_elts(struct r300_context* r300,
 356                                     struct pipe_buffer** elts,
 357                                     unsigned count)
 358 {
 359     struct pipe_screen* screen = r300->context.screen;
 360     struct pipe_buffer* new_elts;
 361     unsigned char *in_map;
 362     unsigned short *out_map;
 363     unsigned i;
 364
 365     new_elts = screen->buffer_create(screen, 32,
 366                                      PIPE_BUFFER_USAGE_INDEX |
 367                                      PIPE_BUFFER_USAGE_CPU_WRITE |
 368                                      PIPE_BUFFER_USAGE_GPU_READ,
 369                                      2 * count);
 370
 371     in_map = pipe_buffer_map(screen, *elts, PIPE_BUFFER_USAGE_CPU_READ);
 372     out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE);
 373
 374     for (i = 0; i < count; i++) {
 375         *out_map = (unsigned short)*in_map;
 376         in_map++;
 377         out_map++;
 378     }
 379
 380     pipe_buffer_unmap(screen, *elts);
 381     pipe_buffer_unmap(screen, new_elts);
 382
 383     *elts = new_elts;
 384 }
 385
 386 /* This is the fast-path drawing & emission for HW TCL. */
 387 void r300_draw_range_elements(struct pipe_context* pipe,
 388                               struct pipe_buffer* indexBuffer,
 389                               unsigned indexSize,
 390                               unsigned minIndex,
 391                               unsigned maxIndex,
 392                               unsigned mode,
 393                               unsigned start,
 394                               unsigned count)
 395 {
 396     struct r300_context* r300 = r300_context(pipe);
 397     struct pipe_buffer* orgIndexBuffer = indexBuffer;
 398 #if defined(ENABLE_ALT_NUM_VERTS)
 399     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 400                             count > 65536;
 401 #else
 402     boolean alt_num_verts = FALSE;
 403 #endif
 404     unsigned short_count;
 405
 406     if (!u_trim_pipe_prim(mode, &count)) {
 407         return;
 408     }
 409
 410     if (indexSize == 1) {
 411         r300_shorten_ubyte_elts(r300, &indexBuffer, count);
 412         indexSize = 2;
 413     }
 414
 415     r300_update_derived_state(r300);
 416
 417     r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count);
 418
 419     /* 128 dwords for emit_aos and emit_draw_elements */
 420     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 421     r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 422     r300_emit_dirty_state(r300);
 423     r300_emit_aos(r300, 0);
 424
 425     u_upload_flush(r300->upload_vb);
 426     u_upload_flush(r300->upload_ib);
 427     if (alt_num_verts || count <= 65535) {
 428         r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 429                                 maxIndex, mode, start, count);
 430     } else {
 431         do {
 432             short_count = MIN2(count, 65534);
 433             r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 434                                     maxIndex, mode, start, short_count);
 435
 436             start += short_count;
 437             count -= short_count;
 438
 439             /* 16 spare dwords are enough for emit_draw_elements. */
 440             if (count && r300_reserve_cs_space(r300, 16)) {
 441                 r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 442                 r300_emit_dirty_state(r300);
 443                 r300_emit_aos(r300, 0);
 444             }
 445         } while (count);
 446     }
 447
 448     if (indexBuffer != orgIndexBuffer) {
 449         pipe_buffer_reference( &indexBuffer, NULL );
 450     }
 451 }
 452
 453 /* Simple helpers for context setup. Should probably be moved to util. */
 454 void r300_draw_elements(struct pipe_context* pipe,
 455                         struct pipe_buffer* indexBuffer,
 456                         unsigned indexSize, unsigned mode,
 457                         unsigned start, unsigned count)
 458 {
 459     struct r300_context *r300 = r300_context(pipe);
 460
 461     pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0,
 462                               r300->vertex_buffer_max_index,
 463                               mode, start, count);
 464 }
 465
 466 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 467                       unsigned start, unsigned count)
 468 {
 469     struct r300_context* r300 = r300_context(pipe);
 470 #if defined(ENABLE_ALT_NUM_VERTS)
 471     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 472                             count > 65536;
 473 #else
 474     boolean alt_num_verts = FALSE;
 475 #endif
 476     unsigned short_count;
 477
 478     if (!u_trim_pipe_prim(mode, &count)) {
 479         return;
 480     }
 481
 482     r300_update_derived_state(r300);
 483
 484     if (immd_is_good_idea(r300, count)) {
 485         r300_emit_draw_arrays_immediate(r300, mode, start, count);
 486     } else {
 487         /* Make sure there are at least 128 spare dwords in the command buffer.
 488          * (most of it being consumed by emit_aos) */
 489         r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 490         r300_emit_buffer_validate(r300, TRUE, NULL);
 491         r300_emit_dirty_state(r300);
 492
 493         if (alt_num_verts || count <= 65535) {
 494             r300_emit_aos(r300, start);
 495             r300_emit_draw_arrays(r300, mode, count);
 496         } else {
 497             do {
 498                 short_count = MIN2(count, 65535);
 499                 r300_emit_aos(r300, start);
 500                 r300_emit_draw_arrays(r300, mode, short_count);
 501
 502                 start += short_count;
 503                 count -= short_count;
 504
 505                 /* Again, we emit both AOS and draw_arrays so there should be
 506                  * at least 128 spare dwords. */
 507                 if (count && r300_reserve_cs_space(r300, 128)) {
 508                     r300_emit_buffer_validate(r300, TRUE, NULL);
 509                     r300_emit_dirty_state(r300);
 510                 }
 511             } while (count);
 512         }
 513         u_upload_flush(r300->upload_vb);
 514     }
 515 }
 516
 517 /****************************************************************************
 518  * The rest of this file is for SW TCL rendering only. Please be polite and *
 519  * keep these functions separated so that they are easier to locate. ~C.    *
 520  ***************************************************************************/
 521
 522 /* SW TCL arrays, using Draw. */
 523 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 524                                unsigned mode,
 525                                unsigned start,
 526                                unsigned count)
 527 {
 528     struct r300_context* r300 = r300_context(pipe);
 529     int i;
 530
 531     if (!u_trim_pipe_prim(mode, &count)) {
 532         return;
 533     }
 534
 535     for (i = 0; i < r300->vertex_buffer_count; i++) {
 536         void* buf = pipe_buffer_map(pipe->screen,
 537                                     r300->vertex_buffer[i].buffer,
 538                                     PIPE_BUFFER_USAGE_CPU_READ);
 539         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 540     }
 541
 542     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 543
 544     draw_set_mapped_constant_buffer(r300->draw,
 545                                     PIPE_SHADER_VERTEX,
 546                                     0,
 547                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 548                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 549                 (sizeof(float) * 4));
 550
 551     draw_arrays(r300->draw, mode, start, count);
 552
 553     for (i = 0; i < r300->vertex_buffer_count; i++) {
 554         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 555         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 556     }
 557 }
 558
 559 /* SW TCL elements, using Draw. */
 560 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 561                                        struct pipe_buffer* indexBuffer,
 562                                        unsigned indexSize,
 563                                        unsigned minIndex,
 564                                        unsigned maxIndex,
 565                                        unsigned mode,
 566                                        unsigned start,
 567                                        unsigned count)
 568 {
 569     struct r300_context* r300 = r300_context(pipe);
 570     int i;
 571     void* indices;
 572
 573     if (!u_trim_pipe_prim(mode, &count)) {
 574         return;
 575     }
 576
 577     for (i = 0; i < r300->vertex_buffer_count; i++) {
 578         void* buf = pipe_buffer_map(pipe->screen,
 579                                     r300->vertex_buffer[i].buffer,
 580                                     PIPE_BUFFER_USAGE_CPU_READ);
 581         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 582     }
 583
 584     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 585                               PIPE_BUFFER_USAGE_CPU_READ);
 586     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 587                                          minIndex, maxIndex, indices);
 588
 589     draw_set_mapped_constant_buffer(r300->draw,
 590                                     PIPE_SHADER_VERTEX,
 591                                     0,
 592             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 593             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 594                 (sizeof(float) * 4));
 595
 596     draw_arrays(r300->draw, mode, start, count);
 597
 598     for (i = 0; i < r300->vertex_buffer_count; i++) {
 599         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 600         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 601     }
 602
 603     pipe_buffer_unmap(pipe->screen, indexBuffer);
 604     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 605                                          start + count - 1, NULL);
 606 }
 607
 608 /* Object for rendering using Draw. */
 609 struct r300_render {
 610     /* Parent class */
 611     struct vbuf_render base;
 612
 613     /* Pipe context */
 614     struct r300_context* r300;
 615
 616     /* Vertex information */
 617     size_t vertex_size;
 618     unsigned prim;
 619     unsigned hwprim;
 620
 621     /* VBO */
 622     struct pipe_buffer* vbo;
 623     size_t vbo_size;
 624     size_t vbo_offset;
 625     size_t vbo_max_used;
 626     void * vbo_ptr;
 627 };
 628
 629 static INLINE struct r300_render*
 630 r300_render(struct vbuf_render* render)
 631 {
 632     return (struct r300_render*)render;
 633 }
 634
 635 static const struct vertex_info*
 636 r300_render_get_vertex_info(struct vbuf_render* render)
 637 {
 638     struct r300_render* r300render = r300_render(render);
 639     struct r300_context* r300 = r300render->r300;
 640
 641     r300_update_derived_state(r300);
 642
 643     return &r300->vertex_info;
 644 }
 645
 646 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 647                                                    ushort vertex_size,
 648                                                    ushort count)
 649 {
 650     struct r300_render* r300render = r300_render(render);
 651     struct r300_context* r300 = r300render->r300;
 652     struct pipe_screen* screen = r300->context.screen;
 653     size_t size = (size_t)vertex_size * (size_t)count;
 654
 655     if (size + r300render->vbo_offset > r300render->vbo_size)
 656     {
 657         pipe_buffer_reference(&r300->vbo, NULL);
 658         r300render->vbo = pipe_buffer_create(screen,
 659                                              64,
 660                                              PIPE_BUFFER_USAGE_VERTEX,
 661                                              R300_MAX_VBO_SIZE);
 662         r300render->vbo_offset = 0;
 663         r300render->vbo_size = R300_MAX_VBO_SIZE;
 664     }
 665
 666     r300render->vertex_size = vertex_size;
 667     r300->vbo = r300render->vbo;
 668     r300->vbo_offset = r300render->vbo_offset;
 669
 670     return (r300render->vbo) ? TRUE : FALSE;
 671 }
 672
 673 static void* r300_render_map_vertices(struct vbuf_render* render)
 674 {
 675     struct r300_render* r300render = r300_render(render);
 676     struct pipe_screen* screen = r300render->r300->context.screen;
 677
 678     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 679                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 680
 681     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 682 }
 683
 684 static void r300_render_unmap_vertices(struct vbuf_render* render,
 685                                              ushort min,
 686                                              ushort max)
 687 {
 688     struct r300_render* r300render = r300_render(render);
 689     struct pipe_screen* screen = r300render->r300->context.screen;
 690     CS_LOCALS(r300render->r300);
 691     BEGIN_CS(2);
 692     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 693     END_CS;
 694
 695     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 696                                     r300render->vertex_size * (max + 1));
 697     pipe_buffer_unmap(screen, r300render->vbo);
 698 }
 699
 700 static void r300_render_release_vertices(struct vbuf_render* render)
 701 {
 702     struct r300_render* r300render = r300_render(render);
 703
 704     r300render->vbo_offset += r300render->vbo_max_used;
 705     r300render->vbo_max_used = 0;
 706 }
 707
 708 static boolean r300_render_set_primitive(struct vbuf_render* render,
 709                                                unsigned prim)
 710 {
 711     struct r300_render* r300render = r300_render(render);
 712
 713     r300render->prim = prim;
 714     r300render->hwprim = r300_translate_primitive(prim);
 715
 716     return TRUE;
 717 }
 718
 719 static void r300_render_draw_arrays(struct vbuf_render* render,
 720                                           unsigned start,
 721                                           unsigned count)
 722 {
 723     struct r300_render* r300render = r300_render(render);
 724     struct r300_context* r300 = r300render->r300;
 725
 726     CS_LOCALS(r300);
 727
 728     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2);
 729     r300_emit_buffer_validate(r300, FALSE, NULL);
 730     r300_emit_dirty_state(r300);
 731
 732     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 733
 734     BEGIN_CS(2);
 735     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 736     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 737            r300render->hwprim);
 738     END_CS;
 739 }
 740
 741 static void r300_render_draw(struct vbuf_render* render,
 742                                    const ushort* indices,
 743                                    uint count)
 744 {
 745     struct r300_render* r300render = r300_render(render);
 746     struct r300_context* r300 = r300render->r300;
 747     int i;
 748     unsigned dwords = 2 + (count+1)/2;
 749
 750     CS_LOCALS(r300);
 751
 752     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 753     r300_emit_buffer_validate(r300, FALSE, NULL);
 754     r300_emit_dirty_state(r300);
 755
 756     BEGIN_CS(dwords);
 757     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 758     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 759            r300render->hwprim);
 760     for (i = 0; i < count-1; i += 2) {
 761         OUT_CS(indices[i+1] << 16 | indices[i]);
 762     }
 763     if (count % 2) {
 764         OUT_CS(indices[count-1]);
 765     }
 766     END_CS;
 767 }
 768
 769 static void r300_render_destroy(struct vbuf_render* render)
 770 {
 771     FREE(render);
 772 }
 773
 774 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 775 {
 776     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 777
 778     r300render->r300 = r300;
 779
 780     /* XXX find real numbers plz */
 781     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 782     r300render->base.max_indices = 16 * 1024;
 783
 784     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 785     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 786     r300render->base.map_vertices = r300_render_map_vertices;
 787     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 788     r300render->base.set_primitive = r300_render_set_primitive;
 789     r300render->base.draw = r300_render_draw;
 790     r300render->base.draw_arrays = r300_render_draw_arrays;
 791     r300render->base.release_vertices = r300_render_release_vertices;
 792     r300render->base.destroy = r300_render_destroy;
 793
 794     r300render->vbo = NULL;
 795     r300render->vbo_size = 0;
 796     r300render->vbo_offset = 0;
 797
 798     return &r300render->base;
 799 }
 800
 801 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 802 {
 803     struct vbuf_render* render;
 804     struct draw_stage* stage;
 805
 806     render = r300_render_create(r300);
 807
 808     if (!render) {
 809         return NULL;
 810     }
 811
 812     stage = draw_vbuf_stage(r300->draw, render);
 813
 814     if (!stage) {
 815         render->destroy(render);
 816         return NULL;
 817     }
 818
 819     draw_set_render(r300->draw, render);
 820
 821     return stage;
 822 }