src/mesa/vbo/vbo_split_copy.c

   1
   2 /*
   3  * Mesa 3-D graphics library
   4  * Version:  6.5
   5  *
   6  * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
   7  *
   8  * Permission is hereby granted, free of charge, to any person obtaining a
   9  * copy of this software and associated documentation files (the "Software"),
  10  * to deal in the Software without restriction, including without limitation
  11  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  12  * and/or sell copies of the Software, and to permit persons to whom the
  13  * Software is furnished to do so, subject to the following conditions:
  14  *
  15  * The above copyright notice and this permission notice shall be included
  16  * in all copies or substantial portions of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  21  * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
  22  * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  23  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  *
  25  * Authors:
  26  *    Keith Whitwell <keith@tungstengraphics.com>
  27  */
  28
  29 /* Split indexed primitives with per-vertex copying.
  30  */
  31
  32 #include "main/glheader.h"
  33 #include "main/bufferobj.h"
  34 #include "main/imports.h"
  35 #include "main/image.h"
  36 #include "main/macros.h"
  37 #include "main/mtypes.h"
  38
  39 #include "vbo_split.h"
  40 #include "vbo.h"
  41
  42
  43 #define ELT_TABLE_SIZE 16
  44
  45 /**
  46  * Used for vertex-level splitting of indexed buffers.  Note that
  47  * non-indexed primitives may be converted to indexed in some cases
  48  * (eg loops, fans) in order to use this splitting path.
  49  */
  50 struct copy_context {
  51
  52    struct gl_context *ctx;
  53    const struct gl_client_array **array;
  54    const struct _mesa_prim *prim;
  55    GLuint nr_prims;
  56    const struct _mesa_index_buffer *ib;
  57    vbo_draw_func draw;
  58
  59    const struct split_limits *limits;
  60
  61    struct {
  62       GLuint attr;
  63       GLuint size;
  64       const struct gl_client_array *array;
  65       const GLubyte *src_ptr;
  66
  67       struct gl_client_array dstarray;
  68
  69    } varying[VERT_ATTRIB_MAX];
  70    GLuint nr_varying;
  71
  72    const struct gl_client_array *dstarray_ptr[VERT_ATTRIB_MAX];
  73    struct _mesa_index_buffer dstib;
  74
  75    GLuint *translated_elt_buf;
  76    const GLuint *srcelt;
  77
  78    /** A baby hash table to avoid re-emitting (some) duplicate
  79     * vertices when splitting indexed primitives.
  80     */
  81    struct {
  82       GLuint in;
  83       GLuint out;
  84    } vert_cache[ELT_TABLE_SIZE];
  85
  86    GLuint vertex_size;
  87    GLubyte *dstbuf;
  88    GLubyte *dstptr;     /**< dstptr == dstbuf + dstelt_max * vertsize */
  89    GLuint dstbuf_size;  /**< in vertices */
  90    GLuint dstbuf_nr;    /**< count of emitted vertices, also the largest value
  91                          * in dstelt.  Our MaxIndex.
  92                          */
  93
  94    GLuint *dstelt;
  95    GLuint dstelt_nr;
  96    GLuint dstelt_size;
  97
  98 #define MAX_PRIM 32
  99    struct _mesa_prim dstprim[MAX_PRIM];
 100    GLuint dstprim_nr;
 101
 102 };
 103
 104
 105 static GLuint attr_size( const struct gl_client_array *array )
 106 {
 107    return array->Size * _mesa_sizeof_type(array->Type);
 108 }
 109
 110
 111 /**
 112  * Starts returning true slightly before the buffer fills, to ensure
 113  * that there is sufficient room for any remaining vertices to finish
 114  * off the prim:
 115  */
 116 static GLboolean
 117 check_flush( struct copy_context *copy )
 118 {
 119    GLenum mode = copy->dstprim[copy->dstprim_nr].mode;
 120
 121    if (GL_TRIANGLE_STRIP == mode &&
 122        copy->dstelt_nr & 1) { /* see bug9962 */
 123        return GL_FALSE;
 124    }
 125
 126    if (copy->dstbuf_nr + 4 > copy->dstbuf_size)
 127       return GL_TRUE;
 128
 129    if (copy->dstelt_nr + 4 > copy->dstelt_size)
 130       return GL_TRUE;
 131
 132    return GL_FALSE;
 133 }
 134
 135
 136 /**
 137  * Dump the parameters/info for a vbo->draw() call.
 138  */
 139 static void
 140 dump_draw_info(struct gl_context *ctx,
 141                const struct gl_client_array **arrays,
 142                const struct _mesa_prim *prims,
 143                GLuint nr_prims,
 144                const struct _mesa_index_buffer *ib,
 145                GLuint min_index,
 146                GLuint max_index)
 147 {
 148    GLuint i, j;
 149
 150    printf("VBO Draw:\n");
 151    for (i = 0; i < nr_prims; i++) {
 152       printf("Prim %u of %u\n", i, nr_prims);
 153       printf("  Prim mode 0x%x\n", prims[i].mode);
 154       printf("  IB: %p\n", (void*) ib);
 155       for (j = 0; j < VERT_ATTRIB_MAX; j++) {
 156          printf("    array %d at %p:\n", j, (void*) arrays[j]);
 157          printf("      enabled %d, ptr %p, size %d, type 0x%x, stride %d\n",
 158                 arrays[j]->Enabled, arrays[j]->Ptr,
 159                 arrays[j]->Size, arrays[j]->Type, arrays[j]->StrideB);
 160          if (0) {
 161             GLint k = prims[i].start + prims[i].count - 1;
 162             GLfloat *last = (GLfloat *) (arrays[j]->Ptr + arrays[j]->Stride * k);
 163             printf("        last: %f %f %f\n",
 164                    last[0], last[1], last[2]);
 165          }
 166       }
 167    }
 168 }
 169
 170
 171 static void
 172 flush( struct copy_context *copy )
 173 {
 174    GLuint i;
 175
 176    /* Set some counters:
 177     */
 178    copy->dstib.count = copy->dstelt_nr;
 179
 180 #if 0
 181    dump_draw_info(copy->ctx,
 182                   copy->dstarray_ptr,
 183                   copy->dstprim,
 184                   copy->dstprim_nr,
 185                   &copy->dstib,
 186                   0,
 187                   copy->dstbuf_nr);
 188 #else
 189    (void) dump_draw_info;
 190 #endif
 191
 192    copy->draw( copy->ctx,
 193                copy->dstarray_ptr,
 194                copy->dstprim,
 195                copy->dstprim_nr,
 196                &copy->dstib,
 197                GL_TRUE,
 198                0,
 199                copy->dstbuf_nr - 1 );
 200
 201    /* Reset all pointers:
 202     */
 203    copy->dstprim_nr = 0;
 204    copy->dstelt_nr = 0;
 205    copy->dstbuf_nr = 0;
 206    copy->dstptr = copy->dstbuf;
 207
 208    /* Clear the vertex cache:
 209     */
 210    for (i = 0; i < ELT_TABLE_SIZE; i++)
 211       copy->vert_cache[i].in = ~0;
 212 }
 213
 214
 215 /**
 216  * Called at begin of each primitive during replay.
 217  */
 218 static void
 219 begin( struct copy_context *copy, GLenum mode, GLboolean begin_flag )
 220 {
 221    struct _mesa_prim *prim = &copy->dstprim[copy->dstprim_nr];
 222
 223    prim->mode = mode;
 224    prim->begin = begin_flag;
 225 }
 226
 227
 228 /**
 229  * Use a hashtable to attempt to identify recently-emitted vertices
 230  * and avoid re-emitting them.
 231  */
 232 static GLuint
 233 elt(struct copy_context *copy, GLuint elt_idx)
 234 {
 235    GLuint elt = copy->srcelt[elt_idx];
 236    GLuint slot = elt & (ELT_TABLE_SIZE-1);
 237
 238 /*    printf("elt %d\n", elt); */
 239
 240    /* Look up the incoming element in the vertex cache.  Re-emit if
 241     * necessary.
 242     */
 243    if (copy->vert_cache[slot].in != elt) {
 244       GLubyte *csr = copy->dstptr;
 245       GLuint i;
 246
 247 /*       printf("  --> emit to dstelt %d\n", copy->dstbuf_nr); */
 248
 249       for (i = 0; i < copy->nr_varying; i++) {
 250          const struct gl_client_array *srcarray = copy->varying[i].array;
 251          const GLubyte *srcptr = copy->varying[i].src_ptr + elt * srcarray->StrideB;
 252
 253          memcpy(csr, srcptr, copy->varying[i].size);
 254          csr += copy->varying[i].size;
 255
 256 #ifdef NAN_CHECK
 257          if (srcarray->Type == GL_FLOAT) {
 258             GLuint k;
 259             GLfloat *f = (GLfloat *) srcptr;
 260             for (k = 0; k < srcarray->Size; k++) {
 261                assert(!IS_INF_OR_NAN(f[k]));
 262                assert(f[k] <= 1.0e20 && f[k] >= -1.0e20);
 263             }
 264          }
 265 #endif
 266
 267          if (0)
 268          {
 269             const GLuint *f = (const GLuint *)srcptr;
 270             GLuint j;
 271             printf("  varying %d: ", i);
 272             for(j = 0; j < copy->varying[i].size / 4; j++)
 273                printf("%x ", f[j]);
 274             printf("\n");
 275          }
 276       }
 277
 278       copy->vert_cache[slot].in = elt;
 279       copy->vert_cache[slot].out = copy->dstbuf_nr++;
 280       copy->dstptr += copy->vertex_size;
 281
 282       assert(csr == copy->dstptr);
 283       assert(copy->dstptr == (copy->dstbuf +
 284                               copy->dstbuf_nr * copy->vertex_size));
 285    }
 286 /*    else */
 287 /*       printf("  --> reuse vertex\n"); */
 288
 289 /*    printf("  --> emit %d\n", copy->vert_cache[slot].out); */
 290    copy->dstelt[copy->dstelt_nr++] = copy->vert_cache[slot].out;
 291    return check_flush(copy);
 292 }
 293
 294
 295 /**
 296  * Called at end of each primitive during replay.
 297  */
 298 static void
 299 end( struct copy_context *copy, GLboolean end_flag )
 300 {
 301    struct _mesa_prim *prim = &copy->dstprim[copy->dstprim_nr];
 302
 303 /*    printf("end (%d)\n", end_flag); */
 304
 305    prim->end = end_flag;
 306    prim->count = copy->dstelt_nr - prim->start;
 307
 308    if (++copy->dstprim_nr == MAX_PRIM ||
 309        check_flush(copy))
 310       flush(copy);
 311 }
 312
 313
 314 static void
 315 replay_elts( struct copy_context *copy )
 316 {
 317    GLuint i, j, k;
 318    GLboolean split;
 319
 320    for (i = 0; i < copy->nr_prims; i++) {
 321       const struct _mesa_prim *prim = &copy->prim[i];
 322       const GLuint start = prim->start;
 323       GLuint first, incr;
 324
 325       switch (prim->mode) {
 326
 327       case GL_LINE_LOOP:
 328          /* Convert to linestrip and emit the final vertex explicitly,
 329           * but only in the resultant strip that requires it.
 330           */
 331          j = 0;
 332          while (j != prim->count) {
 333             begin(copy, GL_LINE_STRIP, prim->begin && j == 0);
 334
 335             for (split = GL_FALSE; j != prim->count && !split; j++)
 336                split = elt(copy, start + j);
 337
 338             if (j == prim->count) {
 339                /* Done, emit final line.  Split doesn't matter as
 340                 * it is always raised a bit early so we can emit
 341                 * the last verts if necessary!
 342                 */
 343                if (prim->end)
 344                   (void)elt(copy, start + 0);
 345
 346                end(copy, prim->end);
 347             }
 348             else {
 349                /* Wrap
 350                 */
 351                assert(split);
 352                end(copy, 0);
 353                j--;
 354             }
 355          }
 356          break;
 357
 358       case GL_TRIANGLE_FAN:
 359       case GL_POLYGON:
 360          j = 2;
 361          while (j != prim->count) {
 362             begin(copy, prim->mode, prim->begin && j == 0);
 363
 364             split = elt(copy, start+0);
 365             assert(!split);
 366
 367             split = elt(copy, start+j-1);
 368             assert(!split);
 369
 370             for (; j != prim->count && !split; j++)
 371                split = elt(copy, start+j);
 372
 373             end(copy, prim->end && j == prim->count);
 374
 375             if (j != prim->count) {
 376                /* Wrapped the primitive, need to repeat some vertices:
 377                 */
 378                j -= 1;
 379             }
 380          }
 381          break;
 382
 383       default:
 384          (void)split_prim_inplace(prim->mode, &first, &incr);
 385
 386          j = 0;
 387          while (j != prim->count) {
 388
 389             begin(copy, prim->mode, prim->begin && j == 0);
 390
 391             split = 0;
 392             for (k = 0; k < first; k++, j++)
 393                split |= elt(copy, start+j);
 394
 395             assert(!split);
 396
 397             for (; j != prim->count && !split; )
 398                for (k = 0; k < incr; k++, j++)
 399                   split |= elt(copy, start+j);
 400
 401             end(copy, prim->end && j == prim->count);
 402
 403             if (j != prim->count) {
 404                /* Wrapped the primitive, need to repeat some vertices:
 405                 */
 406                assert(j > first - incr);
 407                j -= (first - incr);
 408             }
 409          }
 410          break;
 411       }
 412    }
 413
 414    if (copy->dstprim_nr)
 415       flush(copy);
 416 }
 417
 418
 419 static void
 420 replay_init( struct copy_context *copy )
 421 {
 422    struct gl_context *ctx = copy->ctx;
 423    GLuint i;
 424    GLuint offset;
 425    const GLvoid *srcptr;
 426
 427    /* Make a list of varying attributes and their vbo's.  Also
 428     * calculate vertex size.
 429     */
 430    copy->vertex_size = 0;
 431    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 432       struct gl_buffer_object *vbo = copy->array[i]->BufferObj;
 433
 434       if (copy->array[i]->StrideB == 0) {
 435          copy->dstarray_ptr[i] = copy->array[i];
 436       }
 437       else {
 438          GLuint j = copy->nr_varying++;
 439
 440          copy->varying[j].attr = i;
 441          copy->varying[j].array = copy->array[i];
 442          copy->varying[j].size = attr_size(copy->array[i]);
 443          copy->vertex_size += attr_size(copy->array[i]);
 444
 445          if (_mesa_is_bufferobj(vbo) && !_mesa_bufferobj_mapped(vbo))
 446             ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY, vbo);
 447
 448          copy->varying[j].src_ptr = ADD_POINTERS(vbo->Pointer,
 449                                                  copy->array[i]->Ptr);
 450
 451          copy->dstarray_ptr[i] = &copy->varying[j].dstarray;
 452       }
 453    }
 454
 455    /* There must always be an index buffer.  Currently require the
 456     * caller convert non-indexed prims to indexed.  Could alternately
 457     * do it internally.
 458     */
 459    if (_mesa_is_bufferobj(copy->ib->obj) &&
 460        !_mesa_bufferobj_mapped(copy->ib->obj))
 461       ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY,
 462                             copy->ib->obj);
 463
 464    srcptr = (const GLubyte *) ADD_POINTERS(copy->ib->obj->Pointer,
 465                                            copy->ib->ptr);
 466
 467    switch (copy->ib->type) {
 468    case GL_UNSIGNED_BYTE:
 469       copy->translated_elt_buf = malloc(sizeof(GLuint) * copy->ib->count);
 470       copy->srcelt = copy->translated_elt_buf;
 471
 472       for (i = 0; i < copy->ib->count; i++)
 473          copy->translated_elt_buf[i] = ((const GLubyte *)srcptr)[i];
 474       break;
 475
 476    case GL_UNSIGNED_SHORT:
 477       copy->translated_elt_buf = malloc(sizeof(GLuint) * copy->ib->count);
 478       copy->srcelt = copy->translated_elt_buf;
 479
 480       for (i = 0; i < copy->ib->count; i++)
 481          copy->translated_elt_buf[i] = ((const GLushort *)srcptr)[i];
 482       break;
 483
 484    case GL_UNSIGNED_INT:
 485       copy->translated_elt_buf = NULL;
 486       copy->srcelt = (const GLuint *)srcptr;
 487       break;
 488    }
 489
 490    /* Figure out the maximum allowed vertex buffer size:
 491     */
 492    if (copy->vertex_size * copy->limits->max_verts <= copy->limits->max_vb_size) {
 493       copy->dstbuf_size = copy->limits->max_verts;
 494    }
 495    else {
 496       copy->dstbuf_size = copy->limits->max_vb_size / copy->vertex_size;
 497    }
 498
 499    /* Allocate an output vertex buffer:
 500     *
 501     * XXX:  This should be a VBO!
 502     */
 503    copy->dstbuf = malloc(copy->dstbuf_size * copy->vertex_size);
 504    copy->dstptr = copy->dstbuf;
 505
 506    /* Setup new vertex arrays to point into the output buffer:
 507     */
 508    for (offset = 0, i = 0; i < copy->nr_varying; i++) {
 509       const struct gl_client_array *src = copy->varying[i].array;
 510       struct gl_client_array *dst = &copy->varying[i].dstarray;
 511
 512       dst->Size = src->Size;
 513       dst->Type = src->Type;
 514       dst->Format = GL_RGBA;
 515       dst->Stride = copy->vertex_size;
 516       dst->StrideB = copy->vertex_size;
 517       dst->Ptr = copy->dstbuf + offset;
 518       dst->Enabled = GL_TRUE;
 519       dst->Normalized = src->Normalized;
 520       dst->BufferObj = ctx->Shared->NullBufferObj;
 521       dst->_MaxElement = copy->dstbuf_size; /* may be less! */
 522
 523       offset += copy->varying[i].size;
 524    }
 525
 526    /* Allocate an output element list:
 527     */
 528    copy->dstelt_size = MIN2(65536,
 529                             copy->ib->count * 2 + 3);
 530    copy->dstelt_size = MIN2(copy->dstelt_size,
 531                             copy->limits->max_indices);
 532    copy->dstelt = malloc(sizeof(GLuint) * copy->dstelt_size);
 533    copy->dstelt_nr = 0;
 534
 535    /* Setup the new index buffer to point to the allocated element
 536     * list:
 537     */
 538    copy->dstib.count = 0;       /* duplicates dstelt_nr */
 539    copy->dstib.type = GL_UNSIGNED_INT;
 540    copy->dstib.obj = ctx->Shared->NullBufferObj;
 541    copy->dstib.ptr = copy->dstelt;
 542 }
 543
 544
 545 /**
 546  * Free up everything allocated during split/replay.
 547  */
 548 static void
 549 replay_finish( struct copy_context *copy )
 550 {
 551    struct gl_context *ctx = copy->ctx;
 552    GLuint i;
 553
 554    /* Free our vertex and index buffers:
 555     */
 556    free(copy->translated_elt_buf);
 557    free(copy->dstbuf);
 558    free(copy->dstelt);
 559
 560    /* Unmap VBO's
 561     */
 562    for (i = 0; i < copy->nr_varying; i++) {
 563       struct gl_buffer_object *vbo = copy->varying[i].array->BufferObj;
 564       if (_mesa_is_bufferobj(vbo) && _mesa_bufferobj_mapped(vbo))
 565          ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, vbo);
 566    }
 567
 568    /* Unmap index buffer:
 569     */
 570    if (_mesa_is_bufferobj(copy->ib->obj) &&
 571        _mesa_bufferobj_mapped(copy->ib->obj)) {
 572       ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, copy->ib->obj);
 573    }
 574 }
 575
 576
 577 /**
 578  * Split VBO into smaller pieces, draw the pieces.
 579  */
 580 void vbo_split_copy( struct gl_context *ctx,
 581                      const struct gl_client_array *arrays[],
 582                      const struct _mesa_prim *prim,
 583                      GLuint nr_prims,
 584                      const struct _mesa_index_buffer *ib,
 585                      vbo_draw_func draw,
 586                      const struct split_limits *limits )
 587 {
 588    struct copy_context copy;
 589    GLuint i, this_nr_prims;
 590
 591    for (i = 0; i < nr_prims;) {
 592       /* Our SW TNL pipeline doesn't handle basevertex yet, so bind_indices
 593        * will rebase the elements to the basevertex, and we'll only
 594        * emit strings of prims with the same basevertex in one draw call.
 595        */
 596       for (this_nr_prims = 1; i + this_nr_prims < nr_prims;
 597            this_nr_prims++) {
 598          if (prim[i].basevertex != prim[i + this_nr_prims].basevertex)
 599             break;
 600       }
 601
 602       memset(&copy, 0, sizeof(copy));
 603
 604       /* Require indexed primitives:
 605        */
 606       assert(ib);
 607
 608       copy.ctx = ctx;
 609       copy.array = arrays;
 610       copy.prim = &prim[i];
 611       copy.nr_prims = this_nr_prims;
 612       copy.ib = ib;
 613       copy.draw = draw;
 614       copy.limits = limits;
 615
 616       /* Clear the vertex cache:
 617        */
 618       for (i = 0; i < ELT_TABLE_SIZE; i++)
 619          copy.vert_cache[i].in = ~0;
 620
 621       replay_init(&copy);
 622       replay_elts(&copy);
 623       replay_finish(&copy);
 624    }
 625 }