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    prim->num_instances = 1;
 226 }
 227
 228
 229 /**
 230  * Use a hashtable to attempt to identify recently-emitted vertices
 231  * and avoid re-emitting them.
 232  */
 233 static GLuint
 234 elt(struct copy_context *copy, GLuint elt_idx)
 235 {
 236    GLuint elt = copy->srcelt[elt_idx];
 237    GLuint slot = elt & (ELT_TABLE_SIZE-1);
 238
 239 /*    printf("elt %d\n", elt); */
 240
 241    /* Look up the incoming element in the vertex cache.  Re-emit if
 242     * necessary.
 243     */
 244    if (copy->vert_cache[slot].in != elt) {
 245       GLubyte *csr = copy->dstptr;
 246       GLuint i;
 247
 248 /*       printf("  --> emit to dstelt %d\n", copy->dstbuf_nr); */
 249
 250       for (i = 0; i < copy->nr_varying; i++) {
 251          const struct gl_client_array *srcarray = copy->varying[i].array;
 252          const GLubyte *srcptr = copy->varying[i].src_ptr + elt * srcarray->StrideB;
 253
 254          memcpy(csr, srcptr, copy->varying[i].size);
 255          csr += copy->varying[i].size;
 256
 257 #ifdef NAN_CHECK
 258          if (srcarray->Type == GL_FLOAT) {
 259             GLuint k;
 260             GLfloat *f = (GLfloat *) srcptr;
 261             for (k = 0; k < srcarray->Size; k++) {
 262                assert(!IS_INF_OR_NAN(f[k]));
 263                assert(f[k] <= 1.0e20 && f[k] >= -1.0e20);
 264             }
 265          }
 266 #endif
 267
 268          if (0)
 269          {
 270             const GLuint *f = (const GLuint *)srcptr;
 271             GLuint j;
 272             printf("  varying %d: ", i);
 273             for(j = 0; j < copy->varying[i].size / 4; j++)
 274                printf("%x ", f[j]);
 275             printf("\n");
 276          }
 277       }
 278
 279       copy->vert_cache[slot].in = elt;
 280       copy->vert_cache[slot].out = copy->dstbuf_nr++;
 281       copy->dstptr += copy->vertex_size;
 282
 283       assert(csr == copy->dstptr);
 284       assert(copy->dstptr == (copy->dstbuf +
 285                               copy->dstbuf_nr * copy->vertex_size));
 286    }
 287 /*    else */
 288 /*       printf("  --> reuse vertex\n"); */
 289
 290 /*    printf("  --> emit %d\n", copy->vert_cache[slot].out); */
 291    copy->dstelt[copy->dstelt_nr++] = copy->vert_cache[slot].out;
 292    return check_flush(copy);
 293 }
 294
 295
 296 /**
 297  * Called at end of each primitive during replay.
 298  */
 299 static void
 300 end( struct copy_context *copy, GLboolean end_flag )
 301 {
 302    struct _mesa_prim *prim = &copy->dstprim[copy->dstprim_nr];
 303
 304 /*    printf("end (%d)\n", end_flag); */
 305
 306    prim->end = end_flag;
 307    prim->count = copy->dstelt_nr - prim->start;
 308
 309    if (++copy->dstprim_nr == MAX_PRIM ||
 310        check_flush(copy))
 311       flush(copy);
 312 }
 313
 314
 315 static void
 316 replay_elts( struct copy_context *copy )
 317 {
 318    GLuint i, j, k;
 319    GLboolean split;
 320
 321    for (i = 0; i < copy->nr_prims; i++) {
 322       const struct _mesa_prim *prim = &copy->prim[i];
 323       const GLuint start = prim->start;
 324       GLuint first, incr;
 325
 326       switch (prim->mode) {
 327
 328       case GL_LINE_LOOP:
 329          /* Convert to linestrip and emit the final vertex explicitly,
 330           * but only in the resultant strip that requires it.
 331           */
 332          j = 0;
 333          while (j != prim->count) {
 334             begin(copy, GL_LINE_STRIP, prim->begin && j == 0);
 335
 336             for (split = GL_FALSE; j != prim->count && !split; j++)
 337                split = elt(copy, start + j);
 338
 339             if (j == prim->count) {
 340                /* Done, emit final line.  Split doesn't matter as
 341                 * it is always raised a bit early so we can emit
 342                 * the last verts if necessary!
 343                 */
 344                if (prim->end)
 345                   (void)elt(copy, start + 0);
 346
 347                end(copy, prim->end);
 348             }
 349             else {
 350                /* Wrap
 351                 */
 352                assert(split);
 353                end(copy, 0);
 354                j--;
 355             }
 356          }
 357          break;
 358
 359       case GL_TRIANGLE_FAN:
 360       case GL_POLYGON:
 361          j = 2;
 362          while (j != prim->count) {
 363             begin(copy, prim->mode, prim->begin && j == 0);
 364
 365             split = elt(copy, start+0);
 366             assert(!split);
 367
 368             split = elt(copy, start+j-1);
 369             assert(!split);
 370
 371             for (; j != prim->count && !split; j++)
 372                split = elt(copy, start+j);
 373
 374             end(copy, prim->end && j == prim->count);
 375
 376             if (j != prim->count) {
 377                /* Wrapped the primitive, need to repeat some vertices:
 378                 */
 379                j -= 1;
 380             }
 381          }
 382          break;
 383
 384       default:
 385          (void)split_prim_inplace(prim->mode, &first, &incr);
 386
 387          j = 0;
 388          while (j != prim->count) {
 389
 390             begin(copy, prim->mode, prim->begin && j == 0);
 391
 392             split = 0;
 393             for (k = 0; k < first; k++, j++)
 394                split |= elt(copy, start+j);
 395
 396             assert(!split);
 397
 398             for (; j != prim->count && !split; )
 399                for (k = 0; k < incr; k++, j++)
 400                   split |= elt(copy, start+j);
 401
 402             end(copy, prim->end && j == prim->count);
 403
 404             if (j != prim->count) {
 405                /* Wrapped the primitive, need to repeat some vertices:
 406                 */
 407                assert(j > first - incr);
 408                j -= (first - incr);
 409             }
 410          }
 411          break;
 412       }
 413    }
 414
 415    if (copy->dstprim_nr)
 416       flush(copy);
 417 }
 418
 419
 420 static void
 421 replay_init( struct copy_context *copy )
 422 {
 423    struct gl_context *ctx = copy->ctx;
 424    GLuint i;
 425    GLuint offset;
 426    const GLvoid *srcptr;
 427
 428    /* Make a list of varying attributes and their vbo's.  Also
 429     * calculate vertex size.
 430     */
 431    copy->vertex_size = 0;
 432    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 433       struct gl_buffer_object *vbo = copy->array[i]->BufferObj;
 434
 435       if (copy->array[i]->StrideB == 0) {
 436          copy->dstarray_ptr[i] = copy->array[i];
 437       }
 438       else {
 439          GLuint j = copy->nr_varying++;
 440
 441          copy->varying[j].attr = i;
 442          copy->varying[j].array = copy->array[i];
 443          copy->varying[j].size = attr_size(copy->array[i]);
 444          copy->vertex_size += attr_size(copy->array[i]);
 445
 446          if (_mesa_is_bufferobj(vbo) && !_mesa_bufferobj_mapped(vbo))
 447             ctx->Driver.MapBufferRange(ctx, 0, vbo->Size, GL_MAP_READ_BIT, vbo);
 448
 449          copy->varying[j].src_ptr = ADD_POINTERS(vbo->Pointer,
 450                                                  copy->array[i]->Ptr);
 451
 452          copy->dstarray_ptr[i] = &copy->varying[j].dstarray;
 453       }
 454    }
 455
 456    /* There must always be an index buffer.  Currently require the
 457     * caller convert non-indexed prims to indexed.  Could alternately
 458     * do it internally.
 459     */
 460    if (_mesa_is_bufferobj(copy->ib->obj) &&
 461        !_mesa_bufferobj_mapped(copy->ib->obj))
 462       ctx->Driver.MapBufferRange(ctx, 0, copy->ib->obj->Size, GL_MAP_READ_BIT,
 463                                  copy->ib->obj);
 464
 465    srcptr = (const GLubyte *) ADD_POINTERS(copy->ib->obj->Pointer,
 466                                            copy->ib->ptr);
 467
 468    switch (copy->ib->type) {
 469    case GL_UNSIGNED_BYTE:
 470       copy->translated_elt_buf = malloc(sizeof(GLuint) * copy->ib->count);
 471       copy->srcelt = copy->translated_elt_buf;
 472
 473       for (i = 0; i < copy->ib->count; i++)
 474          copy->translated_elt_buf[i] = ((const GLubyte *)srcptr)[i];
 475       break;
 476
 477    case GL_UNSIGNED_SHORT:
 478       copy->translated_elt_buf = malloc(sizeof(GLuint) * copy->ib->count);
 479       copy->srcelt = copy->translated_elt_buf;
 480
 481       for (i = 0; i < copy->ib->count; i++)
 482          copy->translated_elt_buf[i] = ((const GLushort *)srcptr)[i];
 483       break;
 484
 485    case GL_UNSIGNED_INT:
 486       copy->translated_elt_buf = NULL;
 487       copy->srcelt = (const GLuint *)srcptr;
 488       break;
 489    }
 490
 491    /* Figure out the maximum allowed vertex buffer size:
 492     */
 493    if (copy->vertex_size * copy->limits->max_verts <= copy->limits->max_vb_size) {
 494       copy->dstbuf_size = copy->limits->max_verts;
 495    }
 496    else {
 497       copy->dstbuf_size = copy->limits->max_vb_size / copy->vertex_size;
 498    }
 499
 500    /* Allocate an output vertex buffer:
 501     *
 502     * XXX:  This should be a VBO!
 503     */
 504    copy->dstbuf = malloc(copy->dstbuf_size * copy->vertex_size);
 505    copy->dstptr = copy->dstbuf;
 506
 507    /* Setup new vertex arrays to point into the output buffer:
 508     */
 509    for (offset = 0, i = 0; i < copy->nr_varying; i++) {
 510       const struct gl_client_array *src = copy->varying[i].array;
 511       struct gl_client_array *dst = &copy->varying[i].dstarray;
 512
 513       dst->Size = src->Size;
 514       dst->Type = src->Type;
 515       dst->Format = GL_RGBA;
 516       dst->Stride = copy->vertex_size;
 517       dst->StrideB = copy->vertex_size;
 518       dst->Ptr = copy->dstbuf + offset;
 519       dst->Enabled = GL_TRUE;
 520       dst->Normalized = src->Normalized;
 521       dst->BufferObj = ctx->Shared->NullBufferObj;
 522       dst->_ElementSize = src->_ElementSize;
 523       dst->_MaxElement = copy->dstbuf_size; /* may be less! */
 524
 525       offset += copy->varying[i].size;
 526    }
 527
 528    /* Allocate an output element list:
 529     */
 530    copy->dstelt_size = MIN2(65536,
 531                             copy->ib->count * 2 + 3);
 532    copy->dstelt_size = MIN2(copy->dstelt_size,
 533                             copy->limits->max_indices);
 534    copy->dstelt = malloc(sizeof(GLuint) * copy->dstelt_size);
 535    copy->dstelt_nr = 0;
 536
 537    /* Setup the new index buffer to point to the allocated element
 538     * list:
 539     */
 540    copy->dstib.count = 0;       /* duplicates dstelt_nr */
 541    copy->dstib.type = GL_UNSIGNED_INT;
 542    copy->dstib.obj = ctx->Shared->NullBufferObj;
 543    copy->dstib.ptr = copy->dstelt;
 544 }
 545
 546
 547 /**
 548  * Free up everything allocated during split/replay.
 549  */
 550 static void
 551 replay_finish( struct copy_context *copy )
 552 {
 553    struct gl_context *ctx = copy->ctx;
 554    GLuint i;
 555
 556    /* Free our vertex and index buffers:
 557     */
 558    free(copy->translated_elt_buf);
 559    free(copy->dstbuf);
 560    free(copy->dstelt);
 561
 562    /* Unmap VBO's
 563     */
 564    for (i = 0; i < copy->nr_varying; i++) {
 565       struct gl_buffer_object *vbo = copy->varying[i].array->BufferObj;
 566       if (_mesa_is_bufferobj(vbo) && _mesa_bufferobj_mapped(vbo))
 567          ctx->Driver.UnmapBuffer(ctx, vbo);
 568    }
 569
 570    /* Unmap index buffer:
 571     */
 572    if (_mesa_is_bufferobj(copy->ib->obj) &&
 573        _mesa_bufferobj_mapped(copy->ib->obj)) {
 574       ctx->Driver.UnmapBuffer(ctx, copy->ib->obj);
 575    }
 576 }
 577
 578
 579 /**
 580  * Split VBO into smaller pieces, draw the pieces.
 581  */
 582 void vbo_split_copy( struct gl_context *ctx,
 583                      const struct gl_client_array *arrays[],
 584                      const struct _mesa_prim *prim,
 585                      GLuint nr_prims,
 586                      const struct _mesa_index_buffer *ib,
 587                      vbo_draw_func draw,
 588                      const struct split_limits *limits )
 589 {
 590    struct copy_context copy;
 591    GLuint i, this_nr_prims;
 592
 593    for (i = 0; i < nr_prims;) {
 594       /* Our SW TNL pipeline doesn't handle basevertex yet, so bind_indices
 595        * will rebase the elements to the basevertex, and we'll only
 596        * emit strings of prims with the same basevertex in one draw call.
 597        */
 598       for (this_nr_prims = 1; i + this_nr_prims < nr_prims;
 599            this_nr_prims++) {
 600          if (prim[i].basevertex != prim[i + this_nr_prims].basevertex)
 601             break;
 602       }
 603
 604       memset(&copy, 0, sizeof(copy));
 605
 606       /* Require indexed primitives:
 607        */
 608       assert(ib);
 609
 610       copy.ctx = ctx;
 611       copy.array = arrays;
 612       copy.prim = &prim[i];
 613       copy.nr_prims = this_nr_prims;
 614       copy.ib = ib;
 615       copy.draw = draw;
 616       copy.limits = limits;
 617
 618       /* Clear the vertex cache:
 619        */
 620       for (i = 0; i < ELT_TABLE_SIZE; i++)
 621          copy.vert_cache[i].in = ~0;
 622
 623       replay_init(&copy);
 624       replay_elts(&copy);
 625       replay_finish(&copy);
 626    }
 627 }