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    struct gl_context *ctx = copy->ctx;
 175    const struct gl_client_array **saved_arrays = ctx->Array._DrawArrays;
 176    GLuint i;
 177
 178    /* Set some counters:
 179     */
 180    copy->dstib.count = copy->dstelt_nr;
 181
 182 #if 0
 183    dump_draw_info(copy->ctx,
 184                   copy->dstarray_ptr,
 185                   copy->dstprim,
 186                   copy->dstprim_nr,
 187                   &copy->dstib,
 188                   0,
 189                   copy->dstbuf_nr);
 190 #else
 191    (void) dump_draw_info;
 192 #endif
 193
 194    ctx->Array._DrawArrays = copy->dstarray_ptr;
 195
 196    copy->draw( ctx,
 197                copy->dstprim,
 198                copy->dstprim_nr,
 199                &copy->dstib,
 200                GL_TRUE,
 201                0,
 202                copy->dstbuf_nr - 1,
 203                NULL );
 204
 205    ctx->Array._DrawArrays = saved_arrays;
 206
 207    /* Reset all pointers:
 208     */
 209    copy->dstprim_nr = 0;
 210    copy->dstelt_nr = 0;
 211    copy->dstbuf_nr = 0;
 212    copy->dstptr = copy->dstbuf;
 213
 214    /* Clear the vertex cache:
 215     */
 216    for (i = 0; i < ELT_TABLE_SIZE; i++)
 217       copy->vert_cache[i].in = ~0;
 218 }
 219
 220
 221 /**
 222  * Called at begin of each primitive during replay.
 223  */
 224 static void
 225 begin( struct copy_context *copy, GLenum mode, GLboolean begin_flag )
 226 {
 227    struct _mesa_prim *prim = &copy->dstprim[copy->dstprim_nr];
 228
 229    prim->mode = mode;
 230    prim->begin = begin_flag;
 231    prim->num_instances = 1;
 232 }
 233
 234
 235 /**
 236  * Use a hashtable to attempt to identify recently-emitted vertices
 237  * and avoid re-emitting them.
 238  */
 239 static GLuint
 240 elt(struct copy_context *copy, GLuint elt_idx)
 241 {
 242    GLuint elt = copy->srcelt[elt_idx];
 243    GLuint slot = elt & (ELT_TABLE_SIZE-1);
 244
 245 /*    printf("elt %d\n", elt); */
 246
 247    /* Look up the incoming element in the vertex cache.  Re-emit if
 248     * necessary.
 249     */
 250    if (copy->vert_cache[slot].in != elt) {
 251       GLubyte *csr = copy->dstptr;
 252       GLuint i;
 253
 254 /*       printf("  --> emit to dstelt %d\n", copy->dstbuf_nr); */
 255
 256       for (i = 0; i < copy->nr_varying; i++) {
 257          const struct gl_client_array *srcarray = copy->varying[i].array;
 258          const GLubyte *srcptr = copy->varying[i].src_ptr + elt * srcarray->StrideB;
 259
 260          memcpy(csr, srcptr, copy->varying[i].size);
 261          csr += copy->varying[i].size;
 262
 263 #ifdef NAN_CHECK
 264          if (srcarray->Type == GL_FLOAT) {
 265             GLuint k;
 266             GLfloat *f = (GLfloat *) srcptr;
 267             for (k = 0; k < srcarray->Size; k++) {
 268                assert(!IS_INF_OR_NAN(f[k]));
 269                assert(f[k] <= 1.0e20 && f[k] >= -1.0e20);
 270             }
 271          }
 272 #endif
 273
 274          if (0)
 275          {
 276             const GLuint *f = (const GLuint *)srcptr;
 277             GLuint j;
 278             printf("  varying %d: ", i);
 279             for(j = 0; j < copy->varying[i].size / 4; j++)
 280                printf("%x ", f[j]);
 281             printf("\n");
 282          }
 283       }
 284
 285       copy->vert_cache[slot].in = elt;
 286       copy->vert_cache[slot].out = copy->dstbuf_nr++;
 287       copy->dstptr += copy->vertex_size;
 288
 289       assert(csr == copy->dstptr);
 290       assert(copy->dstptr == (copy->dstbuf +
 291                               copy->dstbuf_nr * copy->vertex_size));
 292    }
 293 /*    else */
 294 /*       printf("  --> reuse vertex\n"); */
 295
 296 /*    printf("  --> emit %d\n", copy->vert_cache[slot].out); */
 297    copy->dstelt[copy->dstelt_nr++] = copy->vert_cache[slot].out;
 298    return check_flush(copy);
 299 }
 300
 301
 302 /**
 303  * Called at end of each primitive during replay.
 304  */
 305 static void
 306 end( struct copy_context *copy, GLboolean end_flag )
 307 {
 308    struct _mesa_prim *prim = &copy->dstprim[copy->dstprim_nr];
 309
 310 /*    printf("end (%d)\n", end_flag); */
 311
 312    prim->end = end_flag;
 313    prim->count = copy->dstelt_nr - prim->start;
 314
 315    if (++copy->dstprim_nr == MAX_PRIM ||
 316        check_flush(copy))
 317       flush(copy);
 318 }
 319
 320
 321 static void
 322 replay_elts( struct copy_context *copy )
 323 {
 324    GLuint i, j, k;
 325    GLboolean split;
 326
 327    for (i = 0; i < copy->nr_prims; i++) {
 328       const struct _mesa_prim *prim = &copy->prim[i];
 329       const GLuint start = prim->start;
 330       GLuint first, incr;
 331
 332       switch (prim->mode) {
 333
 334       case GL_LINE_LOOP:
 335          /* Convert to linestrip and emit the final vertex explicitly,
 336           * but only in the resultant strip that requires it.
 337           */
 338          j = 0;
 339          while (j != prim->count) {
 340             begin(copy, GL_LINE_STRIP, prim->begin && j == 0);
 341
 342             for (split = GL_FALSE; j != prim->count && !split; j++)
 343                split = elt(copy, start + j);
 344
 345             if (j == prim->count) {
 346                /* Done, emit final line.  Split doesn't matter as
 347                 * it is always raised a bit early so we can emit
 348                 * the last verts if necessary!
 349                 */
 350                if (prim->end)
 351                   (void)elt(copy, start + 0);
 352
 353                end(copy, prim->end);
 354             }
 355             else {
 356                /* Wrap
 357                 */
 358                assert(split);
 359                end(copy, 0);
 360                j--;
 361             }
 362          }
 363          break;
 364
 365       case GL_TRIANGLE_FAN:
 366       case GL_POLYGON:
 367          j = 2;
 368          while (j != prim->count) {
 369             begin(copy, prim->mode, prim->begin && j == 0);
 370
 371             split = elt(copy, start+0);
 372             assert(!split);
 373
 374             split = elt(copy, start+j-1);
 375             assert(!split);
 376
 377             for (; j != prim->count && !split; j++)
 378                split = elt(copy, start+j);
 379
 380             end(copy, prim->end && j == prim->count);
 381
 382             if (j != prim->count) {
 383                /* Wrapped the primitive, need to repeat some vertices:
 384                 */
 385                j -= 1;
 386             }
 387          }
 388          break;
 389
 390       default:
 391          (void)split_prim_inplace(prim->mode, &first, &incr);
 392
 393          j = 0;
 394          while (j != prim->count) {
 395
 396             begin(copy, prim->mode, prim->begin && j == 0);
 397
 398             split = 0;
 399             for (k = 0; k < first; k++, j++)
 400                split |= elt(copy, start+j);
 401
 402             assert(!split);
 403
 404             for (; j != prim->count && !split; )
 405                for (k = 0; k < incr; k++, j++)
 406                   split |= elt(copy, start+j);
 407
 408             end(copy, prim->end && j == prim->count);
 409
 410             if (j != prim->count) {
 411                /* Wrapped the primitive, need to repeat some vertices:
 412                 */
 413                assert(j > first - incr);
 414                j -= (first - incr);
 415             }
 416          }
 417          break;
 418       }
 419    }
 420
 421    if (copy->dstprim_nr)
 422       flush(copy);
 423 }
 424
 425
 426 static void
 427 replay_init( struct copy_context *copy )
 428 {
 429    struct gl_context *ctx = copy->ctx;
 430    GLuint i;
 431    GLuint offset;
 432    const GLvoid *srcptr;
 433
 434    /* Make a list of varying attributes and their vbo's.  Also
 435     * calculate vertex size.
 436     */
 437    copy->vertex_size = 0;
 438    for (i = 0; i < VERT_ATTRIB_MAX; i++) {
 439       struct gl_buffer_object *vbo = copy->array[i]->BufferObj;
 440
 441       if (copy->array[i]->StrideB == 0) {
 442          copy->dstarray_ptr[i] = copy->array[i];
 443       }
 444       else {
 445          GLuint j = copy->nr_varying++;
 446
 447          copy->varying[j].attr = i;
 448          copy->varying[j].array = copy->array[i];
 449          copy->varying[j].size = attr_size(copy->array[i]);
 450          copy->vertex_size += attr_size(copy->array[i]);
 451
 452          if (_mesa_is_bufferobj(vbo) && !_mesa_bufferobj_mapped(vbo))
 453             ctx->Driver.MapBufferRange(ctx, 0, vbo->Size, GL_MAP_READ_BIT, vbo);
 454
 455          copy->varying[j].src_ptr = ADD_POINTERS(vbo->Pointer,
 456                                                  copy->array[i]->Ptr);
 457
 458          copy->dstarray_ptr[i] = &copy->varying[j].dstarray;
 459       }
 460    }
 461
 462    /* There must always be an index buffer.  Currently require the
 463     * caller convert non-indexed prims to indexed.  Could alternately
 464     * do it internally.
 465     */
 466    if (_mesa_is_bufferobj(copy->ib->obj) &&
 467        !_mesa_bufferobj_mapped(copy->ib->obj))
 468       ctx->Driver.MapBufferRange(ctx, 0, copy->ib->obj->Size, GL_MAP_READ_BIT,
 469                                  copy->ib->obj);
 470
 471    srcptr = (const GLubyte *) ADD_POINTERS(copy->ib->obj->Pointer,
 472                                            copy->ib->ptr);
 473
 474    switch (copy->ib->type) {
 475    case GL_UNSIGNED_BYTE:
 476       copy->translated_elt_buf = malloc(sizeof(GLuint) * copy->ib->count);
 477       copy->srcelt = copy->translated_elt_buf;
 478
 479       for (i = 0; i < copy->ib->count; i++)
 480          copy->translated_elt_buf[i] = ((const GLubyte *)srcptr)[i];
 481       break;
 482
 483    case GL_UNSIGNED_SHORT:
 484       copy->translated_elt_buf = malloc(sizeof(GLuint) * copy->ib->count);
 485       copy->srcelt = copy->translated_elt_buf;
 486
 487       for (i = 0; i < copy->ib->count; i++)
 488          copy->translated_elt_buf[i] = ((const GLushort *)srcptr)[i];
 489       break;
 490
 491    case GL_UNSIGNED_INT:
 492       copy->translated_elt_buf = NULL;
 493       copy->srcelt = (const GLuint *)srcptr;
 494       break;
 495    }
 496
 497    /* Figure out the maximum allowed vertex buffer size:
 498     */
 499    if (copy->vertex_size * copy->limits->max_verts <= copy->limits->max_vb_size) {
 500       copy->dstbuf_size = copy->limits->max_verts;
 501    }
 502    else {
 503       copy->dstbuf_size = copy->limits->max_vb_size / copy->vertex_size;
 504    }
 505
 506    /* Allocate an output vertex buffer:
 507     *
 508     * XXX:  This should be a VBO!
 509     */
 510    copy->dstbuf = malloc(copy->dstbuf_size * copy->vertex_size);
 511    copy->dstptr = copy->dstbuf;
 512
 513    /* Setup new vertex arrays to point into the output buffer:
 514     */
 515    for (offset = 0, i = 0; i < copy->nr_varying; i++) {
 516       const struct gl_client_array *src = copy->varying[i].array;
 517       struct gl_client_array *dst = &copy->varying[i].dstarray;
 518
 519       dst->Size = src->Size;
 520       dst->Type = src->Type;
 521       dst->Format = GL_RGBA;
 522       dst->Stride = copy->vertex_size;
 523       dst->StrideB = copy->vertex_size;
 524       dst->Ptr = copy->dstbuf + offset;
 525       dst->Enabled = GL_TRUE;
 526       dst->Normalized = src->Normalized;
 527       dst->Integer = src->Integer;
 528       dst->BufferObj = ctx->Shared->NullBufferObj;
 529       dst->_ElementSize = src->_ElementSize;
 530       dst->_MaxElement = copy->dstbuf_size; /* may be less! */
 531
 532       offset += copy->varying[i].size;
 533    }
 534
 535    /* Allocate an output element list:
 536     */
 537    copy->dstelt_size = MIN2(65536,
 538                             copy->ib->count * 2 + 3);
 539    copy->dstelt_size = MIN2(copy->dstelt_size,
 540                             copy->limits->max_indices);
 541    copy->dstelt = malloc(sizeof(GLuint) * copy->dstelt_size);
 542    copy->dstelt_nr = 0;
 543
 544    /* Setup the new index buffer to point to the allocated element
 545     * list:
 546     */
 547    copy->dstib.count = 0;       /* duplicates dstelt_nr */
 548    copy->dstib.type = GL_UNSIGNED_INT;
 549    copy->dstib.obj = ctx->Shared->NullBufferObj;
 550    copy->dstib.ptr = copy->dstelt;
 551 }
 552
 553
 554 /**
 555  * Free up everything allocated during split/replay.
 556  */
 557 static void
 558 replay_finish( struct copy_context *copy )
 559 {
 560    struct gl_context *ctx = copy->ctx;
 561    GLuint i;
 562
 563    /* Free our vertex and index buffers:
 564     */
 565    free(copy->translated_elt_buf);
 566    free(copy->dstbuf);
 567    free(copy->dstelt);
 568
 569    /* Unmap VBO's
 570     */
 571    for (i = 0; i < copy->nr_varying; i++) {
 572       struct gl_buffer_object *vbo = copy->varying[i].array->BufferObj;
 573       if (_mesa_is_bufferobj(vbo) && _mesa_bufferobj_mapped(vbo))
 574          ctx->Driver.UnmapBuffer(ctx, vbo);
 575    }
 576
 577    /* Unmap index buffer:
 578     */
 579    if (_mesa_is_bufferobj(copy->ib->obj) &&
 580        _mesa_bufferobj_mapped(copy->ib->obj)) {
 581       ctx->Driver.UnmapBuffer(ctx, copy->ib->obj);
 582    }
 583 }
 584
 585
 586 /**
 587  * Split VBO into smaller pieces, draw the pieces.
 588  */
 589 void vbo_split_copy( struct gl_context *ctx,
 590                      const struct gl_client_array *arrays[],
 591                      const struct _mesa_prim *prim,
 592                      GLuint nr_prims,
 593                      const struct _mesa_index_buffer *ib,
 594                      vbo_draw_func draw,
 595                      const struct split_limits *limits )
 596 {
 597    struct copy_context copy;
 598    GLuint i, this_nr_prims;
 599
 600    for (i = 0; i < nr_prims;) {
 601       /* Our SW TNL pipeline doesn't handle basevertex yet, so bind_indices
 602        * will rebase the elements to the basevertex, and we'll only
 603        * emit strings of prims with the same basevertex in one draw call.
 604        */
 605       for (this_nr_prims = 1; i + this_nr_prims < nr_prims;
 606            this_nr_prims++) {
 607          if (prim[i].basevertex != prim[i + this_nr_prims].basevertex)
 608             break;
 609       }
 610
 611       memset(&copy, 0, sizeof(copy));
 612
 613       /* Require indexed primitives:
 614        */
 615       assert(ib);
 616
 617       copy.ctx = ctx;
 618       copy.array = arrays;
 619       copy.prim = &prim[i];
 620       copy.nr_prims = this_nr_prims;
 621       copy.ib = ib;
 622       copy.draw = draw;
 623       copy.limits = limits;
 624
 625       /* Clear the vertex cache:
 626        */
 627       for (i = 0; i < ELT_TABLE_SIZE; i++)
 628          copy.vert_cache[i].in = ~0;
 629
 630       replay_init(&copy);
 631       replay_elts(&copy);
 632       replay_finish(&copy);
 633    }
 634 }