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(count < (1 << 24));
 307
 308     maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
 309
 310     DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
 311         count, minIndex, maxIndex);
 312
 313     if (alt_num_verts) {
 314         BEGIN_CS(15);
 315         OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
 316     } else {
 317         BEGIN_CS(13);
 318     }
 319     OUT_CS_REG(R300_GA_COLOR_CONTROL,
 320             r300_provoking_vertex_fixes(r300, mode));
 321     OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
 322     OUT_CS(maxIndex);
 323     OUT_CS(minIndex);
 324     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
 325     if (indexSize == 4) {
 326         count_dwords = count;
 327         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 328                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
 329                r300_translate_primitive(mode) |
 330                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 331     } else {
 332         count_dwords = (count + 1) / 2;
 333         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 334                r300_translate_primitive(mode) |
 335                (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
 336     }
 337
 338     /* INDX_BUFFER is a truly special packet3.
 339      * Unlike most other packet3, where the offset is after the count,
 340      * the order is reversed, so the relocation ends up carrying the
 341      * size of the indexbuf instead of the offset.
 342      */
 343     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
 344     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
 345            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
 346     OUT_CS(offset_dwords << 2);
 347     OUT_CS_BUF_RELOC(indexBuffer, count_dwords,
 348                      RADEON_GEM_DOMAIN_GTT, 0, 0);
 349
 350     END_CS;
 351 }
 352
 353 static void r300_shorten_ubyte_elts(struct r300_context* r300,
 354                                     struct pipe_buffer** elts,
 355                                     unsigned start,
 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     in_map += start;
 374
 375     for (i = 0; i < count; i++) {
 376         *out_map = (unsigned short)*in_map;
 377         in_map++;
 378         out_map++;
 379     }
 380
 381     pipe_buffer_unmap(screen, *elts);
 382     pipe_buffer_unmap(screen, new_elts);
 383
 384     *elts = new_elts;
 385 }
 386
 387 static void r300_align_ushort_elts(struct r300_context *r300,
 388                                    struct pipe_buffer **elts,
 389                                    unsigned start, unsigned count)
 390 {
 391     struct pipe_screen* screen = r300->context.screen;
 392     struct pipe_buffer* new_elts;
 393     unsigned short *in_map;
 394     unsigned short *out_map;
 395
 396     new_elts = screen->buffer_create(screen, 32,
 397                                      PIPE_BUFFER_USAGE_INDEX |
 398                                      PIPE_BUFFER_USAGE_CPU_WRITE |
 399                                      PIPE_BUFFER_USAGE_GPU_READ,
 400                                      2 * count);
 401
 402     in_map = pipe_buffer_map(screen, *elts, PIPE_BUFFER_USAGE_CPU_READ);
 403     out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE);
 404
 405     memcpy(out_map, in_map+start, 2 * count);
 406
 407     pipe_buffer_unmap(screen, *elts);
 408     pipe_buffer_unmap(screen, new_elts);
 409
 410     *elts = new_elts;
 411 }
 412
 413 /* This is the fast-path drawing & emission for HW TCL. */
 414 void r300_draw_range_elements(struct pipe_context* pipe,
 415                               struct pipe_buffer* indexBuffer,
 416                               unsigned indexSize,
 417                               unsigned minIndex,
 418                               unsigned maxIndex,
 419                               unsigned mode,
 420                               unsigned start,
 421                               unsigned count)
 422 {
 423     struct r300_context* r300 = r300_context(pipe);
 424     struct pipe_buffer* orgIndexBuffer = indexBuffer;
 425 #if defined(ENABLE_ALT_NUM_VERTS)
 426     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 427                             count > 65536;
 428 #else
 429     boolean alt_num_verts = FALSE;
 430 #endif
 431     unsigned short_count;
 432
 433     if (!u_trim_pipe_prim(mode, &count)) {
 434         return;
 435     }
 436
 437     if (indexSize == 1) {
 438         r300_shorten_ubyte_elts(r300, &indexBuffer, start, count);
 439         indexSize = 2;
 440         start = 0;
 441     } else if (indexSize == 2 && start % 2 != 0) {
 442         r300_align_ushort_elts(r300, &indexBuffer, start, count);
 443         start = 0;
 444     }
 445
 446     r300_update_derived_state(r300);
 447
 448     r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count);
 449
 450     /* 128 dwords for emit_aos and emit_draw_elements */
 451     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 452     r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 453     r300_emit_dirty_state(r300);
 454     r300_emit_aos(r300, 0);
 455
 456     u_upload_flush(r300->upload_vb);
 457     u_upload_flush(r300->upload_ib);
 458     if (alt_num_verts || count <= 65535) {
 459         r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 460                                 maxIndex, mode, start, count);
 461     } else {
 462         do {
 463             short_count = MIN2(count, 65534);
 464             r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
 465                                     maxIndex, mode, start, short_count);
 466
 467             start += short_count;
 468             count -= short_count;
 469
 470             /* 16 spare dwords are enough for emit_draw_elements. */
 471             if (count && r300_reserve_cs_space(r300, 16)) {
 472                 r300_emit_buffer_validate(r300, TRUE, indexBuffer);
 473                 r300_emit_dirty_state(r300);
 474                 r300_emit_aos(r300, 0);
 475             }
 476         } while (count);
 477     }
 478
 479     if (indexBuffer != orgIndexBuffer) {
 480         pipe_buffer_reference( &indexBuffer, NULL );
 481     }
 482 }
 483
 484 /* Simple helpers for context setup. Should probably be moved to util. */
 485 void r300_draw_elements(struct pipe_context* pipe,
 486                         struct pipe_buffer* indexBuffer,
 487                         unsigned indexSize, unsigned mode,
 488                         unsigned start, unsigned count)
 489 {
 490     struct r300_context *r300 = r300_context(pipe);
 491
 492     pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0,
 493                               r300->vertex_buffer_max_index,
 494                               mode, start, count);
 495 }
 496
 497 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
 498                       unsigned start, unsigned count)
 499 {
 500     struct r300_context* r300 = r300_context(pipe);
 501 #if defined(ENABLE_ALT_NUM_VERTS)
 502     boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
 503                             count > 65536;
 504 #else
 505     boolean alt_num_verts = FALSE;
 506 #endif
 507     unsigned short_count;
 508
 509     if (!u_trim_pipe_prim(mode, &count)) {
 510         return;
 511     }
 512
 513     r300_update_derived_state(r300);
 514
 515     if (immd_is_good_idea(r300, count)) {
 516         r300_emit_draw_arrays_immediate(r300, mode, start, count);
 517     } else {
 518         /* Make sure there are at least 128 spare dwords in the command buffer.
 519          * (most of it being consumed by emit_aos) */
 520         r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
 521         r300_emit_buffer_validate(r300, TRUE, NULL);
 522         r300_emit_dirty_state(r300);
 523
 524         if (alt_num_verts || count <= 65535) {
 525             r300_emit_aos(r300, start);
 526             r300_emit_draw_arrays(r300, mode, count);
 527         } else {
 528             do {
 529                 short_count = MIN2(count, 65535);
 530                 r300_emit_aos(r300, start);
 531                 r300_emit_draw_arrays(r300, mode, short_count);
 532
 533                 start += short_count;
 534                 count -= short_count;
 535
 536                 /* Again, we emit both AOS and draw_arrays so there should be
 537                  * at least 128 spare dwords. */
 538                 if (count && r300_reserve_cs_space(r300, 128)) {
 539                     r300_emit_buffer_validate(r300, TRUE, NULL);
 540                     r300_emit_dirty_state(r300);
 541                 }
 542             } while (count);
 543         }
 544         u_upload_flush(r300->upload_vb);
 545     }
 546 }
 547
 548 /****************************************************************************
 549  * The rest of this file is for SW TCL rendering only. Please be polite and *
 550  * keep these functions separated so that they are easier to locate. ~C.    *
 551  ***************************************************************************/
 552
 553 /* SW TCL arrays, using Draw. */
 554 void r300_swtcl_draw_arrays(struct pipe_context* pipe,
 555                                unsigned mode,
 556                                unsigned start,
 557                                unsigned count)
 558 {
 559     struct r300_context* r300 = r300_context(pipe);
 560     int i;
 561
 562     if (!u_trim_pipe_prim(mode, &count)) {
 563         return;
 564     }
 565
 566     for (i = 0; i < r300->vertex_buffer_count; i++) {
 567         void* buf = pipe_buffer_map(pipe->screen,
 568                                     r300->vertex_buffer[i].buffer,
 569                                     PIPE_BUFFER_USAGE_CPU_READ);
 570         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 571     }
 572
 573     draw_set_mapped_element_buffer(r300->draw, 0, NULL);
 574
 575     draw_set_mapped_constant_buffer(r300->draw,
 576                                     PIPE_SHADER_VERTEX,
 577                                     0,
 578                                     r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 579                                     r300->shader_constants[PIPE_SHADER_VERTEX].count *
 580                 (sizeof(float) * 4));
 581
 582     draw_arrays(r300->draw, mode, start, count);
 583
 584     for (i = 0; i < r300->vertex_buffer_count; i++) {
 585         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 586         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 587     }
 588 }
 589
 590 /* SW TCL elements, using Draw. */
 591 void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
 592                                        struct pipe_buffer* indexBuffer,
 593                                        unsigned indexSize,
 594                                        unsigned minIndex,
 595                                        unsigned maxIndex,
 596                                        unsigned mode,
 597                                        unsigned start,
 598                                        unsigned count)
 599 {
 600     struct r300_context* r300 = r300_context(pipe);
 601     int i;
 602     void* indices;
 603
 604     if (!u_trim_pipe_prim(mode, &count)) {
 605         return;
 606     }
 607
 608     for (i = 0; i < r300->vertex_buffer_count; i++) {
 609         void* buf = pipe_buffer_map(pipe->screen,
 610                                     r300->vertex_buffer[i].buffer,
 611                                     PIPE_BUFFER_USAGE_CPU_READ);
 612         draw_set_mapped_vertex_buffer(r300->draw, i, buf);
 613     }
 614
 615     indices = pipe_buffer_map(pipe->screen, indexBuffer,
 616                               PIPE_BUFFER_USAGE_CPU_READ);
 617     draw_set_mapped_element_buffer_range(r300->draw, indexSize,
 618                                          minIndex, maxIndex, indices);
 619
 620     draw_set_mapped_constant_buffer(r300->draw,
 621                                     PIPE_SHADER_VERTEX,
 622                                     0,
 623             r300->shader_constants[PIPE_SHADER_VERTEX].constants,
 624             r300->shader_constants[PIPE_SHADER_VERTEX].count *
 625                 (sizeof(float) * 4));
 626
 627     draw_arrays(r300->draw, mode, start, count);
 628
 629     for (i = 0; i < r300->vertex_buffer_count; i++) {
 630         pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer);
 631         draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
 632     }
 633
 634     pipe_buffer_unmap(pipe->screen, indexBuffer);
 635     draw_set_mapped_element_buffer_range(r300->draw, 0, start,
 636                                          start + count - 1, NULL);
 637 }
 638
 639 /* Object for rendering using Draw. */
 640 struct r300_render {
 641     /* Parent class */
 642     struct vbuf_render base;
 643
 644     /* Pipe context */
 645     struct r300_context* r300;
 646
 647     /* Vertex information */
 648     size_t vertex_size;
 649     unsigned prim;
 650     unsigned hwprim;
 651
 652     /* VBO */
 653     struct pipe_buffer* vbo;
 654     size_t vbo_size;
 655     size_t vbo_offset;
 656     size_t vbo_max_used;
 657     void * vbo_ptr;
 658 };
 659
 660 static INLINE struct r300_render*
 661 r300_render(struct vbuf_render* render)
 662 {
 663     return (struct r300_render*)render;
 664 }
 665
 666 static const struct vertex_info*
 667 r300_render_get_vertex_info(struct vbuf_render* render)
 668 {
 669     struct r300_render* r300render = r300_render(render);
 670     struct r300_context* r300 = r300render->r300;
 671
 672     r300_update_derived_state(r300);
 673
 674     return &r300->vertex_info;
 675 }
 676
 677 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
 678                                                    ushort vertex_size,
 679                                                    ushort count)
 680 {
 681     struct r300_render* r300render = r300_render(render);
 682     struct r300_context* r300 = r300render->r300;
 683     struct pipe_screen* screen = r300->context.screen;
 684     size_t size = (size_t)vertex_size * (size_t)count;
 685
 686     if (size + r300render->vbo_offset > r300render->vbo_size)
 687     {
 688         pipe_buffer_reference(&r300->vbo, NULL);
 689         r300render->vbo = pipe_buffer_create(screen,
 690                                              64,
 691                                              PIPE_BUFFER_USAGE_VERTEX,
 692                                              R300_MAX_VBO_SIZE);
 693         r300render->vbo_offset = 0;
 694         r300render->vbo_size = R300_MAX_VBO_SIZE;
 695     }
 696
 697     r300render->vertex_size = vertex_size;
 698     r300->vbo = r300render->vbo;
 699     r300->vbo_offset = r300render->vbo_offset;
 700
 701     return (r300render->vbo) ? TRUE : FALSE;
 702 }
 703
 704 static void* r300_render_map_vertices(struct vbuf_render* render)
 705 {
 706     struct r300_render* r300render = r300_render(render);
 707     struct pipe_screen* screen = r300render->r300->context.screen;
 708
 709     r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo,
 710                                           PIPE_BUFFER_USAGE_CPU_WRITE);
 711
 712     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 713 }
 714
 715 static void r300_render_unmap_vertices(struct vbuf_render* render,
 716                                              ushort min,
 717                                              ushort max)
 718 {
 719     struct r300_render* r300render = r300_render(render);
 720     struct pipe_screen* screen = r300render->r300->context.screen;
 721     CS_LOCALS(r300render->r300);
 722     BEGIN_CS(2);
 723     OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max);
 724     END_CS;
 725
 726     r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
 727                                     r300render->vertex_size * (max + 1));
 728     pipe_buffer_unmap(screen, r300render->vbo);
 729 }
 730
 731 static void r300_render_release_vertices(struct vbuf_render* render)
 732 {
 733     struct r300_render* r300render = r300_render(render);
 734
 735     r300render->vbo_offset += r300render->vbo_max_used;
 736     r300render->vbo_max_used = 0;
 737 }
 738
 739 static boolean r300_render_set_primitive(struct vbuf_render* render,
 740                                                unsigned prim)
 741 {
 742     struct r300_render* r300render = r300_render(render);
 743
 744     r300render->prim = prim;
 745     r300render->hwprim = r300_translate_primitive(prim);
 746
 747     return TRUE;
 748 }
 749
 750 static void r300_render_draw_arrays(struct vbuf_render* render,
 751                                           unsigned start,
 752                                           unsigned count)
 753 {
 754     struct r300_render* r300render = r300_render(render);
 755     struct r300_context* r300 = r300render->r300;
 756
 757     CS_LOCALS(r300);
 758
 759     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2);
 760     r300_emit_buffer_validate(r300, FALSE, NULL);
 761     r300_emit_dirty_state(r300);
 762
 763     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
 764
 765     BEGIN_CS(2);
 766     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
 767     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
 768            r300render->hwprim);
 769     END_CS;
 770 }
 771
 772 static void r300_render_draw(struct vbuf_render* render,
 773                                    const ushort* indices,
 774                                    uint count)
 775 {
 776     struct r300_render* r300render = r300_render(render);
 777     struct r300_context* r300 = r300render->r300;
 778     int i;
 779     unsigned dwords = 2 + (count+1)/2;
 780
 781     CS_LOCALS(r300);
 782
 783     r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
 784     r300_emit_buffer_validate(r300, FALSE, NULL);
 785     r300_emit_dirty_state(r300);
 786
 787     BEGIN_CS(dwords);
 788     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
 789     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
 790            r300render->hwprim);
 791     for (i = 0; i < count-1; i += 2) {
 792         OUT_CS(indices[i+1] << 16 | indices[i]);
 793     }
 794     if (count % 2) {
 795         OUT_CS(indices[count-1]);
 796     }
 797     END_CS;
 798 }
 799
 800 static void r300_render_destroy(struct vbuf_render* render)
 801 {
 802     FREE(render);
 803 }
 804
 805 static struct vbuf_render* r300_render_create(struct r300_context* r300)
 806 {
 807     struct r300_render* r300render = CALLOC_STRUCT(r300_render);
 808
 809     r300render->r300 = r300;
 810
 811     /* XXX find real numbers plz */
 812     r300render->base.max_vertex_buffer_bytes = 128 * 1024;
 813     r300render->base.max_indices = 16 * 1024;
 814
 815     r300render->base.get_vertex_info = r300_render_get_vertex_info;
 816     r300render->base.allocate_vertices = r300_render_allocate_vertices;
 817     r300render->base.map_vertices = r300_render_map_vertices;
 818     r300render->base.unmap_vertices = r300_render_unmap_vertices;
 819     r300render->base.set_primitive = r300_render_set_primitive;
 820     r300render->base.draw = r300_render_draw;
 821     r300render->base.draw_arrays = r300_render_draw_arrays;
 822     r300render->base.release_vertices = r300_render_release_vertices;
 823     r300render->base.destroy = r300_render_destroy;
 824
 825     r300render->vbo = NULL;
 826     r300render->vbo_size = 0;
 827     r300render->vbo_offset = 0;
 828
 829     return &r300render->base;
 830 }
 831
 832 struct draw_stage* r300_draw_stage(struct r300_context* r300)
 833 {
 834     struct vbuf_render* render;
 835     struct draw_stage* stage;
 836
 837     render = r300_render_create(r300);
 838
 839     if (!render) {
 840         return NULL;
 841     }
 842
 843     stage = draw_vbuf_stage(r300->draw, render);
 844
 845     if (!stage) {
 846         render->destroy(render);
 847         return NULL;
 848     }
 849
 850     draw_set_render(r300->draw, render);
 851
 852     return stage;
 853 }