src/gallium/drivers/r300/r300_render.c

   1 /*
   2  * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com>
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
  22
  23 /* r300_render: Vertex and index buffer primitive emission. Contains both
  24  * HW TCL fastpath rendering, and SW TCL Draw-assisted rendering. */
  25
  26 #include "draw/draw_context.h"
  27 #include "draw/draw_vbuf.h"
  28
  29 #include "util/u_inlines.h"
  30
  31 #include "util/u_format.h"
  32 #include "util/u_memory.h"
  33 #include "util/u_prim.h"
  34
  35 #include "r300_cs.h"
  36 #include "r300_context.h"
  37 #include "r300_emit.h"
  38 #include "r300_reg.h"
  39 #include "r300_render.h"
  40 #include "r300_state_derived.h"
  41
  42 /* r300_render: Vertex and index buffer primitive emission. */
  43 #define R300_MAX_VBO_SIZE  (1024 * 1024)
  44
  45 /* XXX The DRM rejects VAP_ALT_NUM_VERTICES.. */
  46 //#define ENABLE_ALT_NUM_VERTS
  47
  48 uint32_t r300_translate_primitive(unsigned prim)
  49 {
  50     switch (prim) {
  51         case PIPE_PRIM_POINTS:
  52             return R300_VAP_VF_CNTL__PRIM_POINTS;
  53         case PIPE_PRIM_LINES:
  54             return R300_VAP_VF_CNTL__PRIM_LINES;
  55         case PIPE_PRIM_LINE_LOOP:
  56             return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
  57         case PIPE_PRIM_LINE_STRIP:
  58             return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
  59         case PIPE_PRIM_TRIANGLES:
  60             return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
  61         case PIPE_PRIM_TRIANGLE_STRIP:
  62             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
  63         case PIPE_PRIM_TRIANGLE_FAN:
  64             return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
  65         case PIPE_PRIM_QUADS:
  66             return R300_VAP_VF_CNTL__PRIM_QUADS;
  67         case PIPE_PRIM_QUAD_STRIP:
  68             return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
  69         case PIPE_PRIM_POLYGON:
  70             return R300_VAP_VF_CNTL__PRIM_POLYGON;
  71         default:
  72             return 0;
  73     }
  74 }
  75
  76 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
  77                                             unsigned mode)
  78 {
  79     struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
  80     uint32_t color_control = rs->color_control;
  81
  82     /* By default (see r300_state.c:r300_create_rs_state) color_control is
  83      * initialized to provoking the first vertex.
  84      *
  85      * Triangle fans must be reduced to the second vertex, not the first, in
  86      * Gallium flatshade-first mode, as per the GL spec.
  87      * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
  88      *
  89      * Quads never provoke correctly in flatshade-first mode. The first
  90      * vertex is never considered as provoking, so only the second, third,
  91      * and fourth vertices can be selected, and both "third" and "last" modes
  92      * select the fourth vertex. This is probably due to D3D lacking quads.
  93      *
  94      * Similarly, polygons reduce to the first, not the last, vertex, when in
  95      * "last" mode, and all other modes start from the second vertex.
  96      *
  97      * ~ C.
  98      */
  99
 100     if (rs->rs.flatshade_first) {
 101         switch (mode) {
 102             case PIPE_PRIM_TRIANGLE_FAN:
 103                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
 104                 break;
 105             case PIPE_PRIM_QUADS:
 106             case PIPE_PRIM_QUAD_STRIP:
 107             case PIPE_PRIM_POLYGON:
 108                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 109                 break;
 110             default:
 111                 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
 112                 break;
 113         }
 114     } else {
 115         color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
 116     }
 117
 118     return color_control;
 119 }
 120
 121 /* Check if the requested number of dwords is available in the CS and
 122  * if not, flush. Return TRUE if the flush occured. */
 123 static boolean r300_reserve_cs_space(struct r300_context *r300,
 124                                      unsigned dwords)
 125 {
 126     if (!r300->winsys->check_cs(r300->winsys, dwords)) {
 127         r300->context.flush(&r300->context, 0, NULL);
 128         return TRUE;
 129     }
 130     return FALSE;
 131 }
 132
 133 static boolean immd_is_good_idea(struct r300_context *r300,
 134                                       unsigned count)
 135 {
 136     return count <= 4;
 137 }
 138
 139 static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 140                                             unsigned mode,
 141                                             unsigned start,
 142                                             unsigned count)
 143 {
 144     struct pipe_vertex_element* velem;
 145     struct pipe_vertex_buffer* vbuf;
 146     unsigned vertex_element_count = r300->vertex_element_count;
 147     unsigned i, v, vbi, dw, elem_offset, dwords;
 148
 149     /* Size of the vertex, in dwords. */
 150     unsigned vertex_size = 0;
 151
 152     /* Offsets of the attribute, in dwords, from the start of the vertex. */
 153     unsigned offset[PIPE_MAX_ATTRIBS];
 154
 155     /* Size of the vertex element, in dwords. */
 156     unsigned size[PIPE_MAX_ATTRIBS];
 157
 158     /* Stride to the same attrib in the next vertex in the vertex buffer,
 159      * in dwords. */
 160     unsigned stride[PIPE_MAX_ATTRIBS] = {0};
 161
 162     /* Mapped vertex buffers. */
 163     uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
 164
 165     CS_LOCALS(r300);
 166
 167     /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
 168     for (i = 0; i < vertex_element_count; i++) {
 169         velem = &r300->vertex_element[i];
 170         offset[i] = velem->src_offset / 4;
 171         size[i] = util_format_get_blocksize(velem->src_format) / 4;
 172         vertex_size += size[i];
 173         vbi = velem->vertex_buffer_index;
 174
 175         /* Map the buffer. */
 176         if (!map[vbi]) {
 177             vbuf = &r300->vertex_buffer[vbi];
 178             map[vbi] = (uint32_t*)pipe_buffer_map(r300->context.screen,
 179                                                   vbuf->buffer,
 180                                                   PIPE_BUFFER_USAGE_CPU_READ);
 181             map[vbi] += vbuf->buffer_offset / 4;
 182             stride[vbi] = vbuf->stride / 4;
 183         }
 184     }
 185
 186     dwords = 10 + count * vertex_size;
 187
 188     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 189     r300_emit_buffer_validate(r300, FALSE, NULL);
 190     r300_emit_dirty_state(r300);
 191
 192     BEGIN_CS(dwords);
 193     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 194             r300_provoking_vertex_fixes(r300, mode));
 195     OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
 196     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
 197     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
 198     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
 199     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
 200             r300_translate_primitive(mode));
 201
 202     /* Emit vertices. */
 203     for (v = 0; v < count; v++) {
 204         for (i = 0; i < vertex_element_count; i++) {
 205             velem = &r300->vertex_element[i];
 206             vbi = velem->vertex_buffer_index;
 207             elem_offset = offset[i] + stride[vbi] * (v + start);
 208
 209             for (dw = 0; dw < size[i]; dw++) {
 210                 OUT_CS(map[vbi][elem_offset + dw]);
 211             }
 212         }
 213     }
 214     END_CS;
 215
 216     /* Unmap buffers. */
 217     for (i = 0; i < vertex_element_count; i++) {
 218         vbi = r300->vertex_element[i].vertex_buffer_index;
 219
 220         if (map[vbi]) {
 221             vbuf = &r300->vertex_buffer[vbi];
 222             pipe_buffer_unmap(r300->context.screen, vbuf->buffer);
 223             map[vbi] = NULL;
 224         }
 225     }
 226 }
 227
 228 static void r300_emit_draw_arrays(struct r300_context *r300,
 229                                   unsigned mode,
 230                                   unsigned count)
 231 {
 232 #if defined(ENABLE_ALT_NUM_VERTS)
 233     boolean alt_num_verts = count > 65535;
 234 #else
 235     boolean alt_num_verts = FALSE;
 236 #endif
 237     CS_LOCALS(r300);
 238
 239     if (alt_num_verts) {
 240         assert(count < (1 << 24));
 241         BEGIN_CS(10);
 242         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
 243     } else {
 244         BEGIN_CS(8);
 245     }
 246     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 247             r300_provoking_vertex_fixes(r300, mode));
 248     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
 249     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
 250     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 251     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 252            r300_translate_primitive(mode) |
 253            (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 254     END_CS;
 255 }
 256
 257 static void r300_emit_draw_elements(struct r300_context *r300,
 258                                     struct pipe_buffer* indexBuffer,
 259                                     unsigned indexSize,
 260                                     unsigned minIndex,
 261                                     unsigned maxIndex,
 262                                     unsigned mode,
 263                                     unsigned start,
 264                                     unsigned count)
 265 {
 266     uint32_t count_dwords;
 267     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
 268 #if defined(ENABLE_ALT_NUM_VERTS)
 269     boolean alt_num_verts = count > 65535;
 270 #else
 271     boolean alt_num_verts = FALSE;
 272 #endif
 273     CS_LOCALS(r300);
 274
 275     assert((start * indexSize)  % 4 == 0);
 276     assert(count < (1 << 24));
 277
 278     DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
 279         count, minIndex, maxIndex);
 280
 281     maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
 282
 283     if (alt_num_verts) {
 284         BEGIN_CS(16);
 285         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
 286     } else {
 287         BEGIN_CS(14);
 288     }
 289     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 290             r300_provoking_vertex_fixes(r300, mode));
 291     OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex);
 292     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
 293     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 294     if (indexSize == 4) {
 295         count_dwords = count;
 296         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 297                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 298                r300_translate_primitive(mode) |
 299                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 300     } else {
 301         count_dwords = (count + 1) / 2;
 302         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 303                r300_translate_primitive(mode) |
 304                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 305     }
 306
 307     /* INDX_BUFFER is a truly special packet3.
 308      * Unlike most other packet3, where the offset is after the count,
 309      * the order is reversed, so the relocation ends up carrying the
 310      * size of the indexbuf instead of the offset.
 311      */
 312     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 313     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 314            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 315     OUT_CS(offset_dwords << 2);
 316     OUT_CS_RELOC(indexBuffer, count_dwords,
 317         RADEON_GEM_DOMAIN_GTT, 0, 0);
 318
 319     END_CS;
 320 }
 321
 322 static void r300_shorten_ubyte_elts(struct r300_context* r300,
 323                                     struct pipe_buffer** elts,
 324                                     unsigned count)
 325 {
 326     struct pipe_screen* screen = r300->context.screen;
 327     struct pipe_buffer* new_elts;
 328     unsigned char *in_map;
 329     unsigned short *out_map;
 330     unsigned i;
 331
 332     new_elts = screen->buffer_create(screen, 32,
 333                                      PIPE_BUFFER_USAGE_INDEX |
 334                                      PIPE_BUFFER_USAGE_CPU_WRITE |
 335                                      PIPE_BUFFER_USAGE_GPU_READ,
 336                                      2 * count);
 337
 338     in_map = pipe_buffer_map(screen, *elts, PIPE_BUFFER_USAGE_CPU_READ);
 339     out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE);
 340
 341     for (i = 0; i < count; i++) {
 342         *out_map = (unsigned short)*in_map;
 343         in_map++;
 344         out_map++;
 345     }
 346
 347     pipe_buffer_unmap(screen, *elts);
 348     pipe_buffer_unmap(screen, new_elts);
 349
 350     *elts = new_elts;
 351 }
 352
 353 /* This is the fast-path drawing & emission for HW TCL. */
 354 void r300_draw_range_elements(struct pipe_context* pipe,
 355                               struct pipe_buffer* indexBuffer,
 356                               unsigned indexSize,
 357                               unsigned minIndex,
 358                               unsigned maxIndex,
 359                               unsigned mode,
 360                               unsigned start,
 361                               unsigned count)
 362 {
 363     struct r300_context* r300 = r300_context(pipe);
 364     struct pipe_buffer* orgIndexBuffer = indexBuffer;
 365 #if defined(ENABLE_ALT_NUM_VERTS)
 366     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 367                             count > 65536;
 368 #else
 369     boolean alt_num_verts = FALSE;
 370 #endif
 371     unsigned short_count;
 372
 373     if (!u_trim_pipe_prim(mode, &count)) {
 374         return;
 375     }
 376
 377     if (indexSize == 1) {
 378         r300_shorten_ubyte_elts(r300, &indexBuffer, count);
 379         indexSize = 2;
 380     }
 381
 382     r300_update_derived_state(r300);
 383
 384     /* 128 dwords for emit_aos and emit_draw_elements */
 385     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 386     r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 387     r300_emit_dirty_state(r300);
 388     r300_emit_aos(r300, 0);
 389
 390     if (alt_num_verts || count <= 65535) {
 391         r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 392                                 maxIndex, mode, start, count);
 393     } else {
 394         do {
 395             short_count = MIN2(count, 65534);
 396             r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 397                                     maxIndex, mode, start, short_count);
 398
 399             start += short_count;
 400             count -= short_count;
 401
 402             /* 16 spare dwords are enough for emit_draw_elements. */
 403             if (count && r300_reserve_cs_space(r300, 16)) {
 404                 r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 405                 r300_emit_dirty_state(r300);
 406                 r300_emit_aos(r300, 0);
 407             }
 408         } while (count);
 409     }
 410
 411     if (indexBuffer != orgIndexBuffer) {
 412         pipe->screen->buffer_destroy(indexBuffer);
 413     }
 414 }
 415
 416 /* Simple helpers for context setup. Should probably be moved to util. */
 417 void r300_draw_elements(struct pipe_context* pipe,
 418                         struct pipe_buffer* indexBuffer,
 419                         unsigned indexSize, unsigned mode,
 420                         unsigned start, unsigned count)
 421 {
 422     struct r300_context *r300 = r300_context(pipe);
 423
 424     pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0,
 425                               r300->vertex_buffer_max_index,
 426                               mode, start, count);
 427 }
 428
 429 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 430                       unsigned start, unsigned count)
 431 {
 432     struct r300_context* r300 = r300_context(pipe);
 433 #if defined(ENABLE_ALT_NUM_VERTS)
 434     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 435                             count > 65536;
 436 #else
 437     boolean alt_num_verts = FALSE;
 438 #endif
 439     unsigned short_count;
 440
 441     if (!u_trim_pipe_prim(mode, &count)) {
 442         return;
 443     }
 444
 445     r300_update_derived_state(r300);
 446
 447     if (immd_is_good_idea(r300, count)) {
 448         r300_emit_draw_arrays_immediate(r300, mode, start, count);
 449     } else {
 450         /* Make sure there are at least 128 spare dwords in the command buffer.
 451          * (most of it being consumed by emit_aos) */
 452         r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 453         r300_emit_buffer_validate(r300, TRUE, NULL);
 454         r300_emit_dirty_state(r300);
 455
 456         if (alt_num_verts || count <= 65535) {
 457             r300_emit_aos(r300, start);
 458             r300_emit_draw_arrays(r300, mode, count);
 459         } else {
 460             do {
 461                 short_count = MIN2(count, 65535);
 462                 r300_emit_aos(r300, start);
 463                 r300_emit_draw_arrays(r300, mode, short_count);
 464
 465                 start += short_count;
 466                 count -= short_count;
 467
 468                 /* Again, we emit both AOS and draw_arrays so there should be
 469                  * at least 128 spare dwords. */
 470                 if (count && r300_reserve_cs_space(r300, 128)) {
 471                     r300_emit_buffer_validate(r300, TRUE, NULL);
 472                     r300_emit_dirty_state(r300);
 473                 }
 474             } while (count);
 475         }
 476     }
 477 }
 478
 479 /****************************************************************************
 480  * The rest of this file is for SW TCL rendering only. Please be polite and *
 481  * keep these functions separated so that they are easier to locate. ~C.    *
 482  ***************************************************************************/
 483
 484 /* SW TCL arrays, using Draw. */
 485 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 486                                unsigned mode,
 487                                unsigned start,
 488                                unsigned count)
 489 {
 490     struct r300_context* r300 = r300_context(pipe);
 491     int i;
 492
 493     if (!u_trim_pipe_prim(mode, &count)) {
 494         return;
 495     }
 496
 497     for (i = 0; i < r300->vertex_buffer_count; i++) {
 498         void* buf = pipe_buffer_map(pipe->screen,
 499                                     r300->vertex_buffer[i].buffer,
 500                                     PIPE_BUFFER_USAGE_CPU_READ);
 501         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 502     }
 503
 504     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 505
 506     draw_set_mapped_constant_buffer(r300->draw,
 507                                     PIPE_SHADER_VERTEX,
 508                                     0,
 509                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 510                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 511                 (sizeof(float) * 4));
 512
 513     draw_arrays(r300->draw, mode, start, count);
 514
 515     for (i = 0; i < r300->vertex_buffer_count; i++) {
 516         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 517         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 518     }
 519 }
 520
 521 /* SW TCL elements, using Draw. */
 522 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 523                                        struct pipe_buffer* indexBuffer,
 524                                        unsigned indexSize,
 525                                        unsigned minIndex,
 526                                        unsigned maxIndex,
 527                                        unsigned mode,
 528                                        unsigned start,
 529                                        unsigned count)
 530 {
 531     struct r300_context* r300 = r300_context(pipe);
 532     int i;
 533     void* indices;
 534
 535     if (!u_trim_pipe_prim(mode, &count)) {
 536         return;
 537     }
 538
 539     for (i = 0; i < r300->vertex_buffer_count; i++) {
 540         void* buf = pipe_buffer_map(pipe->screen,
 541                                     r300->vertex_buffer[i].buffer,
 542                                     PIPE_BUFFER_USAGE_CPU_READ);
 543         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 544     }
 545
 546     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 547                               PIPE_BUFFER_USAGE_CPU_READ);
 548     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 549                                          minIndex, maxIndex, indices);
 550
 551     draw_set_mapped_constant_buffer(r300->draw,
 552                                     PIPE_SHADER_VERTEX,
 553                                     0,
 554             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 555             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 556                 (sizeof(float) * 4));
 557
 558     draw_arrays(r300->draw, mode, start, count);
 559
 560     for (i = 0; i < r300->vertex_buffer_count; i++) {
 561         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 562         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 563     }
 564
 565     pipe_buffer_unmap(pipe->screen, indexBuffer);
 566     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 567                                          start + count - 1, NULL);
 568 }
 569
 570 /* Object for rendering using Draw. */
 571 struct r300_render {
 572     /* Parent class */
 573     struct vbuf_render base;
 574
 575     /* Pipe context */
 576     struct r300_context* r300;
 577
 578     /* Vertex information */
 579     size_t vertex_size;
 580     unsigned prim;
 581     unsigned hwprim;
 582
 583     /* VBO */
 584     struct pipe_buffer* vbo;
 585     size_t vbo_size;
 586     size_t vbo_offset;
 587     size_t vbo_max_used;
 588     void * vbo_ptr;
 589 };
 590
 591 static INLINE struct r300_render*
 592 r300_render(struct vbuf_render* render)
 593 {
 594     return (struct r300_render*)render;
 595 }
 596
 597 static const struct vertex_info*
 598 r300_render_get_vertex_info(struct vbuf_render* render)
 599 {
 600     struct r300_render* r300render = r300_render(render);
 601     struct r300_context* r300 = r300render->r300;
 602
 603     r300_update_derived_state(r300);
 604
 605     return &r300->vertex_info;
 606 }
 607
 608 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 609                                                    ushort vertex_size,
 610                                                    ushort count)
 611 {
 612     struct r300_render* r300render = r300_render(render);
 613     struct r300_context* r300 = r300render->r300;
 614     struct pipe_screen* screen = r300->context.screen;
 615     size_t size = (size_t)vertex_size * (size_t)count;
 616
 617     if (size + r300render->vbo_offset > r300render->vbo_size)
 618     {
 619         pipe_buffer_reference(&r300->vbo, NULL);
 620         r300render->vbo = pipe_buffer_create(screen,
 621                                              64,
 622                                              PIPE_BUFFER_USAGE_VERTEX,
 623                                              R300_MAX_VBO_SIZE);
 624         r300render->vbo_offset = 0;
 625         r300render->vbo_size = R300_MAX_VBO_SIZE;
 626     }
 627
 628     r300render->vertex_size = vertex_size;
 629     r300->vbo = r300render->vbo;
 630     r300->vbo_offset = r300render->vbo_offset;
 631
 632     return (r300render->vbo) ? TRUE : FALSE;
 633 }
 634
 635 static void* r300_render_map_vertices(struct vbuf_render* render)
 636 {
 637     struct r300_render* r300render = r300_render(render);
 638     struct pipe_screen* screen = r300render->r300->context.screen;
 639
 640     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 641                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 642
 643     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 644 }
 645
 646 static void r300_render_unmap_vertices(struct vbuf_render* render,
 647                                              ushort min,
 648                                              ushort max)
 649 {
 650     struct r300_render* r300render = r300_render(render);
 651     struct pipe_screen* screen = r300render->r300->context.screen;
 652     CS_LOCALS(r300render->r300);
 653     BEGIN_CS(2);
 654     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 655     END_CS;
 656
 657     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 658                                     r300render->vertex_size * (max + 1));
 659     pipe_buffer_unmap(screen, r300render->vbo);
 660 }
 661
 662 static void r300_render_release_vertices(struct vbuf_render* render)
 663 {
 664     struct r300_render* r300render = r300_render(render);
 665
 666     r300render->vbo_offset += r300render->vbo_max_used;
 667     r300render->vbo_max_used = 0;
 668 }
 669
 670 static boolean r300_render_set_primitive(struct vbuf_render* render,
 671                                                unsigned prim)
 672 {
 673     struct r300_render* r300render = r300_render(render);
 674
 675     r300render->prim = prim;
 676     r300render->hwprim = r300_translate_primitive(prim);
 677
 678     return TRUE;
 679 }
 680
 681 static void r300_render_draw_arrays(struct vbuf_render* render,
 682                                           unsigned start,
 683                                           unsigned count)
 684 {
 685     struct r300_render* r300render = r300_render(render);
 686     struct r300_context* r300 = r300render->r300;
 687
 688     CS_LOCALS(r300);
 689
 690     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2);
 691     r300_emit_buffer_validate(r300, FALSE, NULL);
 692     r300_emit_dirty_state(r300);
 693
 694     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 695
 696     BEGIN_CS(2);
 697     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 698     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 699            r300render->hwprim);
 700     END_CS;
 701 }
 702
 703 static void r300_render_draw(struct vbuf_render* render,
 704                                    const ushort* indices,
 705                                    uint count)
 706 {
 707     struct r300_render* r300render = r300_render(render);
 708     struct r300_context* r300 = r300render->r300;
 709     int i;
 710     unsigned dwords = 2 + (count+1)/2;
 711
 712     CS_LOCALS(r300);
 713
 714     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 715     r300_emit_buffer_validate(r300, FALSE, NULL);
 716     r300_emit_dirty_state(r300);
 717
 718     BEGIN_CS(dwords);
 719     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 720     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 721            r300render->hwprim);
 722     for (i = 0; i < count-1; i += 2) {
 723         OUT_CS(indices[i+1] << 16 | indices[i]);
 724     }
 725     if (count % 2) {
 726         OUT_CS(indices[count-1]);
 727     }
 728     END_CS;
 729 }
 730
 731 static void r300_render_destroy(struct vbuf_render* render)
 732 {
 733     FREE(render);
 734 }
 735
 736 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 737 {
 738     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 739
 740     r300render->r300 = r300;
 741
 742     /* XXX find real numbers plz */
 743     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 744     r300render->base.max_indices = 16 * 1024;
 745
 746     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 747     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 748     r300render->base.map_vertices = r300_render_map_vertices;
 749     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 750     r300render->base.set_primitive = r300_render_set_primitive;
 751     r300render->base.draw = r300_render_draw;
 752     r300render->base.draw_arrays = r300_render_draw_arrays;
 753     r300render->base.release_vertices = r300_render_release_vertices;
 754     r300render->base.destroy = r300_render_destroy;
 755
 756     r300render->vbo = NULL;
 757     r300render->vbo_size = 0;
 758     r300render->vbo_offset = 0;
 759
 760     return &r300render->base;
 761 }
 762
 763 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 764 {
 765     struct vbuf_render* render;
 766     struct draw_stage* stage;
 767
 768     render = r300_render_create(r300);
 769
 770     if (!render) {
 771         return NULL;
 772     }
 773
 774     stage = draw_vbuf_stage(r300->draw, render);
 775
 776     if (!stage) {
 777         render->destroy(render);
 778         return NULL;
 779     }
 780
 781     draw_set_render(r300->draw, render);
 782
 783     return stage;
 784 }