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 start,
 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     in_map += start;
 375
 376     for (i = 0; i < count; i++) {
 377         *out_map = (unsigned short)*in_map;
 378         in_map++;
 379         out_map++;
 380     }
 381
 382     pipe_buffer_unmap(screen, *elts);
 383     pipe_buffer_unmap(screen, new_elts);
 384
 385     *elts = new_elts;
 386 }
 387
 388 /* This is the fast-path drawing & emission for HW TCL. */
 389 void r300_draw_range_elements(struct pipe_context* pipe,
 390                               struct pipe_buffer* indexBuffer,
 391                               unsigned indexSize,
 392                               unsigned minIndex,
 393                               unsigned maxIndex,
 394                               unsigned mode,
 395                               unsigned start,
 396                               unsigned count)
 397 {
 398     struct r300_context* r300 = r300_context(pipe);
 399     struct pipe_buffer* orgIndexBuffer = indexBuffer;
 400 #if defined(ENABLE_ALT_NUM_VERTS)
 401     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 402                             count > 65536;
 403 #else
 404     boolean alt_num_verts = FALSE;
 405 #endif
 406     unsigned short_count;
 407
 408     if (!u_trim_pipe_prim(mode, &count)) {
 409         return;
 410     }
 411
 412     if (indexSize == 1) {
 413         r300_shorten_ubyte_elts(r300, &indexBuffer, start, count);
 414         indexSize = 2;
 415         start = 0;
 416     }
 417
 418     r300_update_derived_state(r300);
 419
 420     r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count);
 421
 422     /* 128 dwords for emit_aos and emit_draw_elements */
 423     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 424     r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 425     r300_emit_dirty_state(r300);
 426     r300_emit_aos(r300, 0);
 427
 428     u_upload_flush(r300->upload_vb);
 429     u_upload_flush(r300->upload_ib);
 430     if (alt_num_verts || count <= 65535) {
 431         r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 432                                 maxIndex, mode, start, count);
 433     } else {
 434         do {
 435             short_count = MIN2(count, 65534);
 436             r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 437                                     maxIndex, mode, start, short_count);
 438
 439             start += short_count;
 440             count -= short_count;
 441
 442             /* 16 spare dwords are enough for emit_draw_elements. */
 443             if (count && r300_reserve_cs_space(r300, 16)) {
 444                 r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 445                 r300_emit_dirty_state(r300);
 446                 r300_emit_aos(r300, 0);
 447             }
 448         } while (count);
 449     }
 450
 451     if (indexBuffer != orgIndexBuffer) {
 452         pipe_buffer_reference( &indexBuffer, NULL );
 453     }
 454 }
 455
 456 /* Simple helpers for context setup. Should probably be moved to util. */
 457 void r300_draw_elements(struct pipe_context* pipe,
 458                         struct pipe_buffer* indexBuffer,
 459                         unsigned indexSize, unsigned mode,
 460                         unsigned start, unsigned count)
 461 {
 462     struct r300_context *r300 = r300_context(pipe);
 463
 464     pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0,
 465                               r300->vertex_buffer_max_index,
 466                               mode, start, count);
 467 }
 468
 469 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 470                       unsigned start, unsigned count)
 471 {
 472     struct r300_context* r300 = r300_context(pipe);
 473 #if defined(ENABLE_ALT_NUM_VERTS)
 474     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 475                             count > 65536;
 476 #else
 477     boolean alt_num_verts = FALSE;
 478 #endif
 479     unsigned short_count;
 480
 481     if (!u_trim_pipe_prim(mode, &count)) {
 482         return;
 483     }
 484
 485     r300_update_derived_state(r300);
 486
 487     if (immd_is_good_idea(r300, count)) {
 488         r300_emit_draw_arrays_immediate(r300, mode, start, count);
 489     } else {
 490         /* Make sure there are at least 128 spare dwords in the command buffer.
 491          * (most of it being consumed by emit_aos) */
 492         r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 493         r300_emit_buffer_validate(r300, TRUE, NULL);
 494         r300_emit_dirty_state(r300);
 495
 496         if (alt_num_verts || count <= 65535) {
 497             r300_emit_aos(r300, start);
 498             r300_emit_draw_arrays(r300, mode, count);
 499         } else {
 500             do {
 501                 short_count = MIN2(count, 65535);
 502                 r300_emit_aos(r300, start);
 503                 r300_emit_draw_arrays(r300, mode, short_count);
 504
 505                 start += short_count;
 506                 count -= short_count;
 507
 508                 /* Again, we emit both AOS and draw_arrays so there should be
 509                  * at least 128 spare dwords. */
 510                 if (count && r300_reserve_cs_space(r300, 128)) {
 511                     r300_emit_buffer_validate(r300, TRUE, NULL);
 512                     r300_emit_dirty_state(r300);
 513                 }
 514             } while (count);
 515         }
 516         u_upload_flush(r300->upload_vb);
 517     }
 518 }
 519
 520 /****************************************************************************
 521  * The rest of this file is for SW TCL rendering only. Please be polite and *
 522  * keep these functions separated so that they are easier to locate. ~C.    *
 523  ***************************************************************************/
 524
 525 /* SW TCL arrays, using Draw. */
 526 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 527                                unsigned mode,
 528                                unsigned start,
 529                                unsigned count)
 530 {
 531     struct r300_context* r300 = r300_context(pipe);
 532     int i;
 533
 534     if (!u_trim_pipe_prim(mode, &count)) {
 535         return;
 536     }
 537
 538     for (i = 0; i < r300->vertex_buffer_count; i++) {
 539         void* buf = pipe_buffer_map(pipe->screen,
 540                                     r300->vertex_buffer[i].buffer,
 541                                     PIPE_BUFFER_USAGE_CPU_READ);
 542         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 543     }
 544
 545     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 546
 547     draw_set_mapped_constant_buffer(r300->draw,
 548                                     PIPE_SHADER_VERTEX,
 549                                     0,
 550                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 551                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 552                 (sizeof(float) * 4));
 553
 554     draw_arrays(r300->draw, mode, start, count);
 555
 556     for (i = 0; i < r300->vertex_buffer_count; i++) {
 557         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 558         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 559     }
 560 }
 561
 562 /* SW TCL elements, using Draw. */
 563 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 564                                        struct pipe_buffer* indexBuffer,
 565                                        unsigned indexSize,
 566                                        unsigned minIndex,
 567                                        unsigned maxIndex,
 568                                        unsigned mode,
 569                                        unsigned start,
 570                                        unsigned count)
 571 {
 572     struct r300_context* r300 = r300_context(pipe);
 573     int i;
 574     void* indices;
 575
 576     if (!u_trim_pipe_prim(mode, &count)) {
 577         return;
 578     }
 579
 580     for (i = 0; i < r300->vertex_buffer_count; i++) {
 581         void* buf = pipe_buffer_map(pipe->screen,
 582                                     r300->vertex_buffer[i].buffer,
 583                                     PIPE_BUFFER_USAGE_CPU_READ);
 584         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 585     }
 586
 587     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 588                               PIPE_BUFFER_USAGE_CPU_READ);
 589     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 590                                          minIndex, maxIndex, indices);
 591
 592     draw_set_mapped_constant_buffer(r300->draw,
 593                                     PIPE_SHADER_VERTEX,
 594                                     0,
 595             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 596             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 597                 (sizeof(float) * 4));
 598
 599     draw_arrays(r300->draw, mode, start, count);
 600
 601     for (i = 0; i < r300->vertex_buffer_count; i++) {
 602         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 603         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 604     }
 605
 606     pipe_buffer_unmap(pipe->screen, indexBuffer);
 607     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 608                                          start + count - 1, NULL);
 609 }
 610
 611 /* Object for rendering using Draw. */
 612 struct r300_render {
 613     /* Parent class */
 614     struct vbuf_render base;
 615
 616     /* Pipe context */
 617     struct r300_context* r300;
 618
 619     /* Vertex information */
 620     size_t vertex_size;
 621     unsigned prim;
 622     unsigned hwprim;
 623
 624     /* VBO */
 625     struct pipe_buffer* vbo;
 626     size_t vbo_size;
 627     size_t vbo_offset;
 628     size_t vbo_max_used;
 629     void * vbo_ptr;
 630 };
 631
 632 static INLINE struct r300_render*
 633 r300_render(struct vbuf_render* render)
 634 {
 635     return (struct r300_render*)render;
 636 }
 637
 638 static const struct vertex_info*
 639 r300_render_get_vertex_info(struct vbuf_render* render)
 640 {
 641     struct r300_render* r300render = r300_render(render);
 642     struct r300_context* r300 = r300render->r300;
 643
 644     r300_update_derived_state(r300);
 645
 646     return &r300->vertex_info;
 647 }
 648
 649 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 650                                                    ushort vertex_size,
 651                                                    ushort count)
 652 {
 653     struct r300_render* r300render = r300_render(render);
 654     struct r300_context* r300 = r300render->r300;
 655     struct pipe_screen* screen = r300->context.screen;
 656     size_t size = (size_t)vertex_size * (size_t)count;
 657
 658     if (size + r300render->vbo_offset > r300render->vbo_size)
 659     {
 660         pipe_buffer_reference(&r300->vbo, NULL);
 661         r300render->vbo = pipe_buffer_create(screen,
 662                                              64,
 663                                              PIPE_BUFFER_USAGE_VERTEX,
 664                                              R300_MAX_VBO_SIZE);
 665         r300render->vbo_offset = 0;
 666         r300render->vbo_size = R300_MAX_VBO_SIZE;
 667     }
 668
 669     r300render->vertex_size = vertex_size;
 670     r300->vbo = r300render->vbo;
 671     r300->vbo_offset = r300render->vbo_offset;
 672
 673     return (r300render->vbo) ? TRUE : FALSE;
 674 }
 675
 676 static void* r300_render_map_vertices(struct vbuf_render* render)
 677 {
 678     struct r300_render* r300render = r300_render(render);
 679     struct pipe_screen* screen = r300render->r300->context.screen;
 680
 681     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 682                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 683
 684     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 685 }
 686
 687 static void r300_render_unmap_vertices(struct vbuf_render* render,
 688                                              ushort min,
 689                                              ushort max)
 690 {
 691     struct r300_render* r300render = r300_render(render);
 692     struct pipe_screen* screen = r300render->r300->context.screen;
 693     CS_LOCALS(r300render->r300);
 694     BEGIN_CS(2);
 695     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 696     END_CS;
 697
 698     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 699                                     r300render->vertex_size * (max + 1));
 700     pipe_buffer_unmap(screen, r300render->vbo);
 701 }
 702
 703 static void r300_render_release_vertices(struct vbuf_render* render)
 704 {
 705     struct r300_render* r300render = r300_render(render);
 706
 707     r300render->vbo_offset += r300render->vbo_max_used;
 708     r300render->vbo_max_used = 0;
 709 }
 710
 711 static boolean r300_render_set_primitive(struct vbuf_render* render,
 712                                                unsigned prim)
 713 {
 714     struct r300_render* r300render = r300_render(render);
 715
 716     r300render->prim = prim;
 717     r300render->hwprim = r300_translate_primitive(prim);
 718
 719     return TRUE;
 720 }
 721
 722 static void r300_render_draw_arrays(struct vbuf_render* render,
 723                                           unsigned start,
 724                                           unsigned count)
 725 {
 726     struct r300_render* r300render = r300_render(render);
 727     struct r300_context* r300 = r300render->r300;
 728
 729     CS_LOCALS(r300);
 730
 731     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2);
 732     r300_emit_buffer_validate(r300, FALSE, NULL);
 733     r300_emit_dirty_state(r300);
 734
 735     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 736
 737     BEGIN_CS(2);
 738     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 739     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 740            r300render->hwprim);
 741     END_CS;
 742 }
 743
 744 static void r300_render_draw(struct vbuf_render* render,
 745                                    const ushort* indices,
 746                                    uint count)
 747 {
 748     struct r300_render* r300render = r300_render(render);
 749     struct r300_context* r300 = r300render->r300;
 750     int i;
 751     unsigned dwords = 2 + (count+1)/2;
 752
 753     CS_LOCALS(r300);
 754
 755     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 756     r300_emit_buffer_validate(r300, FALSE, NULL);
 757     r300_emit_dirty_state(r300);
 758
 759     BEGIN_CS(dwords);
 760     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 761     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 762            r300render->hwprim);
 763     for (i = 0; i < count-1; i += 2) {
 764         OUT_CS(indices[i+1] << 16 | indices[i]);
 765     }
 766     if (count % 2) {
 767         OUT_CS(indices[count-1]);
 768     }
 769     END_CS;
 770 }
 771
 772 static void r300_render_destroy(struct vbuf_render* render)
 773 {
 774     FREE(render);
 775 }
 776
 777 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 778 {
 779     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 780
 781     r300render->r300 = r300;
 782
 783     /* XXX find real numbers plz */
 784     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 785     r300render->base.max_indices = 16 * 1024;
 786
 787     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 788     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 789     r300render->base.map_vertices = r300_render_map_vertices;
 790     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 791     r300render->base.set_primitive = r300_render_set_primitive;
 792     r300render->base.draw = r300_render_draw;
 793     r300render->base.draw_arrays = r300_render_draw_arrays;
 794     r300render->base.release_vertices = r300_render_release_vertices;
 795     r300render->base.destroy = r300_render_destroy;
 796
 797     r300render->vbo = NULL;
 798     r300render->vbo_size = 0;
 799     r300render->vbo_offset = 0;
 800
 801     return &r300render->base;
 802 }
 803
 804 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 805 {
 806     struct vbuf_render* render;
 807     struct draw_stage* stage;
 808
 809     render = r300_render_create(r300);
 810
 811     if (!render) {
 812         return NULL;
 813     }
 814
 815     stage = draw_vbuf_stage(r300->draw, render);
 816
 817     if (!stage) {
 818         render->destroy(render);
 819         return NULL;
 820     }
 821
 822     draw_set_render(r300->draw, render);
 823
 824     return stage;
 825 }