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 > 10) {
 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, vbuf->buffer)) {
 159                 /* It's a very bad idea to map it... */
 160                 return FALSE;
 161             }
 162             checked[vbi] = TRUE;
 163         }
 164     }
 165     return TRUE;
 166 }
 167
 168 static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 169                                             unsigned mode,
 170                                             unsigned start,
 171                                             unsigned count)
 172 {
 173     struct pipe_vertex_element* velem;
 174     struct pipe_vertex_buffer* vbuf;
 175     unsigned vertex_element_count = r300->velems->count;
 176     unsigned i, v, vbi, dw, elem_offset, dwords;
 177
 178     /* Size of the vertex, in dwords. */
 179     unsigned vertex_size = 0;
 180
 181     /* Offsets of the attribute, in dwords, from the start of the vertex. */
 182     unsigned offset[PIPE_MAX_ATTRIBS];
 183
 184     /* Size of the vertex element, in dwords. */
 185     unsigned size[PIPE_MAX_ATTRIBS];
 186
 187     /* Stride to the same attrib in the next vertex in the vertex buffer,
 188      * in dwords. */
 189     unsigned stride[PIPE_MAX_ATTRIBS] = {0};
 190
 191     /* Mapped vertex buffers. */
 192     uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
 193
 194     CS_LOCALS(r300);
 195
 196     /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
 197     for (i = 0; i < vertex_element_count; i++) {
 198         velem = &r300->velems->velem[i];
 199         offset[i] = velem->src_offset / 4;
 200         size[i] = util_format_get_blocksize(velem->src_format) / 4;
 201         vertex_size += size[i];
 202         vbi = velem->vertex_buffer_index;
 203
 204         /* Map the buffer. */
 205         if (!map[vbi]) {
 206             vbuf = &r300->vertex_buffer[vbi];
 207             map[vbi] = (uint32_t*)pipe_buffer_map(r300->context.screen,
 208                                                   vbuf->buffer,
 209                                                   PIPE_BUFFER_USAGE_CPU_READ);
 210             map[vbi] += vbuf->buffer_offset / 4;
 211             stride[vbi] = vbuf->stride / 4;
 212         }
 213     }
 214
 215     dwords = 9 + count * vertex_size;
 216
 217     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 218     r300_emit_buffer_validate(r300, FALSE, NULL);
 219     r300_emit_dirty_state(r300);
 220
 221     BEGIN_CS(dwords);
 222     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 223             r300_provoking_vertex_fixes(r300, mode));
 224     OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
 225     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 226     OUT_CS(count - 1);
 227     OUT_CS(0);
 228     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
 229     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
 230             r300_translate_primitive(mode));
 231
 232     /* Emit vertices. */
 233     for (v = 0; v < count; v++) {
 234         for (i = 0; i < vertex_element_count; i++) {
 235             velem = &r300->velems->velem[i];
 236             vbi = velem->vertex_buffer_index;
 237             elem_offset = offset[i] + stride[vbi] * (v + start);
 238
 239             for (dw = 0; dw < size[i]; dw++) {
 240                 OUT_CS(map[vbi][elem_offset + dw]);
 241             }
 242         }
 243     }
 244     END_CS;
 245
 246     /* Unmap buffers. */
 247     for (i = 0; i < vertex_element_count; i++) {
 248         vbi = r300->velems->velem[i].vertex_buffer_index;
 249
 250         if (map[vbi]) {
 251             vbuf = &r300->vertex_buffer[vbi];
 252             pipe_buffer_unmap(r300->context.screen, vbuf->buffer);
 253             map[vbi] = NULL;
 254         }
 255     }
 256 }
 257
 258 static void r300_emit_draw_arrays(struct r300_context *r300,
 259                                   unsigned mode,
 260                                   unsigned count)
 261 {
 262 #if defined(ENABLE_ALT_NUM_VERTS)
 263     boolean alt_num_verts = count > 65535;
 264 #else
 265     boolean alt_num_verts = FALSE;
 266 #endif
 267     CS_LOCALS(r300);
 268
 269     if (alt_num_verts) {
 270         assert(count < (1 << 24));
 271         BEGIN_CS(9);
 272         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
 273     } else {
 274         BEGIN_CS(7);
 275     }
 276     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 277             r300_provoking_vertex_fixes(r300, mode));
 278     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 279     OUT_CS(count - 1);
 280     OUT_CS(0);
 281     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 282     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 283            r300_translate_primitive(mode) |
 284            (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 285     END_CS;
 286 }
 287
 288 static void r300_emit_draw_elements(struct r300_context *r300,
 289                                     struct pipe_buffer* indexBuffer,
 290                                     unsigned indexSize,
 291                                     unsigned minIndex,
 292                                     unsigned maxIndex,
 293                                     unsigned mode,
 294                                     unsigned start,
 295                                     unsigned count)
 296 {
 297     uint32_t count_dwords;
 298     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
 299 #if defined(ENABLE_ALT_NUM_VERTS)
 300     boolean alt_num_verts = count > 65535;
 301 #else
 302     boolean alt_num_verts = FALSE;
 303 #endif
 304     CS_LOCALS(r300);
 305
 306     assert((start * indexSize) % 4 == 0);
 307     assert(count < (1 << 24));
 308
 309     maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
 310
 311     DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
 312         count, minIndex, maxIndex);
 313
 314     if (alt_num_verts) {
 315         BEGIN_CS(15);
 316         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
 317     } else {
 318         BEGIN_CS(13);
 319     }
 320     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 321             r300_provoking_vertex_fixes(r300, mode));
 322     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 323     OUT_CS(maxIndex);
 324     OUT_CS(minIndex);
 325     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 326     if (indexSize == 4) {
 327         count_dwords = count;
 328         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 329                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 330                r300_translate_primitive(mode) |
 331                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 332     } else {
 333         count_dwords = (count + 1) / 2;
 334         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 335                r300_translate_primitive(mode) |
 336                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 337     }
 338
 339     /* INDX_BUFFER is a truly special packet3.
 340      * Unlike most other packet3, where the offset is after the count,
 341      * the order is reversed, so the relocation ends up carrying the
 342      * size of the indexbuf instead of the offset.
 343      */
 344     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 345     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 346            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 347     OUT_CS(offset_dwords << 2);
 348     OUT_CS_BUF_RELOC(indexBuffer, count_dwords,
 349                      RADEON_GEM_DOMAIN_GTT, 0, 0);
 350
 351     END_CS;
 352 }
 353
 354 static void r300_shorten_ubyte_elts(struct r300_context* r300,
 355                                     struct pipe_buffer** elts,
 356                                     unsigned count)
 357 {
 358     struct pipe_screen* screen = r300->context.screen;
 359     struct pipe_buffer* new_elts;
 360     unsigned char *in_map;
 361     unsigned short *out_map;
 362     unsigned i;
 363
 364     new_elts = screen->buffer_create(screen, 32,
 365                                      PIPE_BUFFER_USAGE_INDEX |
 366                                      PIPE_BUFFER_USAGE_CPU_WRITE |
 367                                      PIPE_BUFFER_USAGE_GPU_READ,
 368                                      2 * count);
 369
 370     in_map = pipe_buffer_map(screen, *elts, PIPE_BUFFER_USAGE_CPU_READ);
 371     out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE);
 372
 373     for (i = 0; i < count; i++) {
 374         *out_map = (unsigned short)*in_map;
 375         in_map++;
 376         out_map++;
 377     }
 378
 379     pipe_buffer_unmap(screen, *elts);
 380     pipe_buffer_unmap(screen, new_elts);
 381
 382     *elts = new_elts;
 383 }
 384
 385 /* This is the fast-path drawing & emission for HW TCL. */
 386 void r300_draw_range_elements(struct pipe_context* pipe,
 387                               struct pipe_buffer* indexBuffer,
 388                               unsigned indexSize,
 389                               unsigned minIndex,
 390                               unsigned maxIndex,
 391                               unsigned mode,
 392                               unsigned start,
 393                               unsigned count)
 394 {
 395     struct r300_context* r300 = r300_context(pipe);
 396     struct pipe_buffer* orgIndexBuffer = indexBuffer;
 397 #if defined(ENABLE_ALT_NUM_VERTS)
 398     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 399                             count > 65536;
 400 #else
 401     boolean alt_num_verts = FALSE;
 402 #endif
 403     unsigned short_count;
 404
 405     if (!u_trim_pipe_prim(mode, &count)) {
 406         return;
 407     }
 408
 409     if (indexSize == 1) {
 410         r300_shorten_ubyte_elts(r300, &indexBuffer, count);
 411         indexSize = 2;
 412     }
 413
 414     r300_update_derived_state(r300);
 415
 416     r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count);
 417
 418     /* 128 dwords for emit_aos and emit_draw_elements */
 419     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 420     r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 421     r300_emit_dirty_state(r300);
 422     r300_emit_aos(r300, 0);
 423
 424     u_upload_flush(r300->upload_vb);
 425     u_upload_flush(r300->upload_ib);
 426     if (alt_num_verts || count <= 65535) {
 427         r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 428                                 maxIndex, mode, start, count);
 429     } else {
 430         do {
 431             short_count = MIN2(count, 65534);
 432             r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 433                                     maxIndex, mode, start, short_count);
 434
 435             start += short_count;
 436             count -= short_count;
 437
 438             /* 16 spare dwords are enough for emit_draw_elements. */
 439             if (count && r300_reserve_cs_space(r300, 16)) {
 440                 r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 441                 r300_emit_dirty_state(r300);
 442                 r300_emit_aos(r300, 0);
 443             }
 444         } while (count);
 445     }
 446
 447     if (indexBuffer != orgIndexBuffer) {
 448         pipe_buffer_reference( &indexBuffer, NULL );
 449     }
 450 }
 451
 452 /* Simple helpers for context setup. Should probably be moved to util. */
 453 void r300_draw_elements(struct pipe_context* pipe,
 454                         struct pipe_buffer* indexBuffer,
 455                         unsigned indexSize, unsigned mode,
 456                         unsigned start, unsigned count)
 457 {
 458     struct r300_context *r300 = r300_context(pipe);
 459
 460     pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0,
 461                               r300->vertex_buffer_max_index,
 462                               mode, start, count);
 463 }
 464
 465 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 466                       unsigned start, unsigned count)
 467 {
 468     struct r300_context* r300 = r300_context(pipe);
 469 #if defined(ENABLE_ALT_NUM_VERTS)
 470     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 471                             count > 65536;
 472 #else
 473     boolean alt_num_verts = FALSE;
 474 #endif
 475     unsigned short_count;
 476
 477     if (!u_trim_pipe_prim(mode, &count)) {
 478         return;
 479     }
 480
 481     r300_update_derived_state(r300);
 482
 483     if (immd_is_good_idea(r300, count)) {
 484         r300_emit_draw_arrays_immediate(r300, mode, start, count);
 485     } else {
 486         /* Make sure there are at least 128 spare dwords in the command buffer.
 487          * (most of it being consumed by emit_aos) */
 488         r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 489         r300_emit_buffer_validate(r300, TRUE, NULL);
 490         r300_emit_dirty_state(r300);
 491
 492         if (alt_num_verts || count <= 65535) {
 493             r300_emit_aos(r300, start);
 494             r300_emit_draw_arrays(r300, mode, count);
 495         } else {
 496             do {
 497                 short_count = MIN2(count, 65535);
 498                 r300_emit_aos(r300, start);
 499                 r300_emit_draw_arrays(r300, mode, short_count);
 500
 501                 start += short_count;
 502                 count -= short_count;
 503
 504                 /* Again, we emit both AOS and draw_arrays so there should be
 505                  * at least 128 spare dwords. */
 506                 if (count && r300_reserve_cs_space(r300, 128)) {
 507                     r300_emit_buffer_validate(r300, TRUE, NULL);
 508                     r300_emit_dirty_state(r300);
 509                 }
 510             } while (count);
 511         }
 512         u_upload_flush(r300->upload_vb);
 513     }
 514 }
 515
 516 /****************************************************************************
 517  * The rest of this file is for SW TCL rendering only. Please be polite and *
 518  * keep these functions separated so that they are easier to locate. ~C.    *
 519  ***************************************************************************/
 520
 521 /* SW TCL arrays, using Draw. */
 522 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 523                                unsigned mode,
 524                                unsigned start,
 525                                unsigned count)
 526 {
 527     struct r300_context* r300 = r300_context(pipe);
 528     int i;
 529
 530     if (!u_trim_pipe_prim(mode, &count)) {
 531         return;
 532     }
 533
 534     for (i = 0; i < r300->vertex_buffer_count; i++) {
 535         void* buf = pipe_buffer_map(pipe->screen,
 536                                     r300->vertex_buffer[i].buffer,
 537                                     PIPE_BUFFER_USAGE_CPU_READ);
 538         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 539     }
 540
 541     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 542
 543     draw_set_mapped_constant_buffer(r300->draw,
 544                                     PIPE_SHADER_VERTEX,
 545                                     0,
 546                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 547                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 548                 (sizeof(float) * 4));
 549
 550     draw_arrays(r300->draw, mode, start, count);
 551
 552     for (i = 0; i < r300->vertex_buffer_count; i++) {
 553         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 554         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 555     }
 556 }
 557
 558 /* SW TCL elements, using Draw. */
 559 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 560                                        struct pipe_buffer* indexBuffer,
 561                                        unsigned indexSize,
 562                                        unsigned minIndex,
 563                                        unsigned maxIndex,
 564                                        unsigned mode,
 565                                        unsigned start,
 566                                        unsigned count)
 567 {
 568     struct r300_context* r300 = r300_context(pipe);
 569     int i;
 570     void* indices;
 571
 572     if (!u_trim_pipe_prim(mode, &count)) {
 573         return;
 574     }
 575
 576     for (i = 0; i < r300->vertex_buffer_count; i++) {
 577         void* buf = pipe_buffer_map(pipe->screen,
 578                                     r300->vertex_buffer[i].buffer,
 579                                     PIPE_BUFFER_USAGE_CPU_READ);
 580         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 581     }
 582
 583     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 584                               PIPE_BUFFER_USAGE_CPU_READ);
 585     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 586                                          minIndex, maxIndex, indices);
 587
 588     draw_set_mapped_constant_buffer(r300->draw,
 589                                     PIPE_SHADER_VERTEX,
 590                                     0,
 591             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 592             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 593                 (sizeof(float) * 4));
 594
 595     draw_arrays(r300->draw, mode, start, count);
 596
 597     for (i = 0; i < r300->vertex_buffer_count; i++) {
 598         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 599         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 600     }
 601
 602     pipe_buffer_unmap(pipe->screen, indexBuffer);
 603     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 604                                          start + count - 1, NULL);
 605 }
 606
 607 /* Object for rendering using Draw. */
 608 struct r300_render {
 609     /* Parent class */
 610     struct vbuf_render base;
 611
 612     /* Pipe context */
 613     struct r300_context* r300;
 614
 615     /* Vertex information */
 616     size_t vertex_size;
 617     unsigned prim;
 618     unsigned hwprim;
 619
 620     /* VBO */
 621     struct pipe_buffer* vbo;
 622     size_t vbo_size;
 623     size_t vbo_offset;
 624     size_t vbo_max_used;
 625     void * vbo_ptr;
 626 };
 627
 628 static INLINE struct r300_render*
 629 r300_render(struct vbuf_render* render)
 630 {
 631     return (struct r300_render*)render;
 632 }
 633
 634 static const struct vertex_info*
 635 r300_render_get_vertex_info(struct vbuf_render* render)
 636 {
 637     struct r300_render* r300render = r300_render(render);
 638     struct r300_context* r300 = r300render->r300;
 639
 640     r300_update_derived_state(r300);
 641
 642     return &r300->vertex_info;
 643 }
 644
 645 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 646                                                    ushort vertex_size,
 647                                                    ushort count)
 648 {
 649     struct r300_render* r300render = r300_render(render);
 650     struct r300_context* r300 = r300render->r300;
 651     struct pipe_screen* screen = r300->context.screen;
 652     size_t size = (size_t)vertex_size * (size_t)count;
 653
 654     if (size + r300render->vbo_offset > r300render->vbo_size)
 655     {
 656         pipe_buffer_reference(&r300->vbo, NULL);
 657         r300render->vbo = pipe_buffer_create(screen,
 658                                              64,
 659                                              PIPE_BUFFER_USAGE_VERTEX,
 660                                              R300_MAX_VBO_SIZE);
 661         r300render->vbo_offset = 0;
 662         r300render->vbo_size = R300_MAX_VBO_SIZE;
 663     }
 664
 665     r300render->vertex_size = vertex_size;
 666     r300->vbo = r300render->vbo;
 667     r300->vbo_offset = r300render->vbo_offset;
 668
 669     return (r300render->vbo) ? TRUE : FALSE;
 670 }
 671
 672 static void* r300_render_map_vertices(struct vbuf_render* render)
 673 {
 674     struct r300_render* r300render = r300_render(render);
 675     struct pipe_screen* screen = r300render->r300->context.screen;
 676
 677     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 678                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 679
 680     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 681 }
 682
 683 static void r300_render_unmap_vertices(struct vbuf_render* render,
 684                                              ushort min,
 685                                              ushort max)
 686 {
 687     struct r300_render* r300render = r300_render(render);
 688     struct pipe_screen* screen = r300render->r300->context.screen;
 689     CS_LOCALS(r300render->r300);
 690     BEGIN_CS(2);
 691     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 692     END_CS;
 693
 694     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 695                                     r300render->vertex_size * (max + 1));
 696     pipe_buffer_unmap(screen, r300render->vbo);
 697 }
 698
 699 static void r300_render_release_vertices(struct vbuf_render* render)
 700 {
 701     struct r300_render* r300render = r300_render(render);
 702
 703     r300render->vbo_offset += r300render->vbo_max_used;
 704     r300render->vbo_max_used = 0;
 705 }
 706
 707 static boolean r300_render_set_primitive(struct vbuf_render* render,
 708                                                unsigned prim)
 709 {
 710     struct r300_render* r300render = r300_render(render);
 711
 712     r300render->prim = prim;
 713     r300render->hwprim = r300_translate_primitive(prim);
 714
 715     return TRUE;
 716 }
 717
 718 static void r300_render_draw_arrays(struct vbuf_render* render,
 719                                           unsigned start,
 720                                           unsigned count)
 721 {
 722     struct r300_render* r300render = r300_render(render);
 723     struct r300_context* r300 = r300render->r300;
 724
 725     CS_LOCALS(r300);
 726
 727     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2);
 728     r300_emit_buffer_validate(r300, FALSE, NULL);
 729     r300_emit_dirty_state(r300);
 730
 731     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 732
 733     BEGIN_CS(2);
 734     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 735     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 736            r300render->hwprim);
 737     END_CS;
 738 }
 739
 740 static void r300_render_draw(struct vbuf_render* render,
 741                                    const ushort* indices,
 742                                    uint count)
 743 {
 744     struct r300_render* r300render = r300_render(render);
 745     struct r300_context* r300 = r300render->r300;
 746     int i;
 747     unsigned dwords = 2 + (count+1)/2;
 748
 749     CS_LOCALS(r300);
 750
 751     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 752     r300_emit_buffer_validate(r300, FALSE, NULL);
 753     r300_emit_dirty_state(r300);
 754
 755     BEGIN_CS(dwords);
 756     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 757     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 758            r300render->hwprim);
 759     for (i = 0; i < count-1; i += 2) {
 760         OUT_CS(indices[i+1] << 16 | indices[i]);
 761     }
 762     if (count % 2) {
 763         OUT_CS(indices[count-1]);
 764     }
 765     END_CS;
 766 }
 767
 768 static void r300_render_destroy(struct vbuf_render* render)
 769 {
 770     FREE(render);
 771 }
 772
 773 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 774 {
 775     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 776
 777     r300render->r300 = r300;
 778
 779     /* XXX find real numbers plz */
 780     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 781     r300render->base.max_indices = 16 * 1024;
 782
 783     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 784     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 785     r300render->base.map_vertices = r300_render_map_vertices;
 786     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 787     r300render->base.set_primitive = r300_render_set_primitive;
 788     r300render->base.draw = r300_render_draw;
 789     r300render->base.draw_arrays = r300_render_draw_arrays;
 790     r300render->base.release_vertices = r300_render_release_vertices;
 791     r300render->base.destroy = r300_render_destroy;
 792
 793     r300render->vbo = NULL;
 794     r300render->vbo_size = 0;
 795     r300render->vbo_offset = 0;
 796
 797     return &r300render->base;
 798 }
 799
 800 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 801 {
 802     struct vbuf_render* render;
 803     struct draw_stage* stage;
 804
 805     render = r300_render_create(r300);
 806
 807     if (!render) {
 808         return NULL;
 809     }
 810
 811     stage = draw_vbuf_stage(r300->draw, render);
 812
 813     if (!stage) {
 814         render->destroy(render);
 815         return NULL;
 816     }
 817
 818     draw_set_render(r300->draw, render);
 819
 820     return stage;
 821 }