src/gallium/auxiliary/draw/draw_pt_vcache.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * 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
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28  /*
  29   * Authors:
  30   *   Keith Whitwell <keith@tungstengraphics.com>
  31   */
  32
  33 #include "util/u_memory.h"
  34 #include "util/u_prim.h"
  35 #include "draw/draw_context.h"
  36 #include "draw/draw_private.h"
  37 #include "draw/draw_pt.h"
  38
  39
  40 #define CACHE_MAX 256
  41 #define FETCH_MAX 256
  42 #define DRAW_MAX (16*1024)
  43
  44 struct vcache_frontend {
  45    struct draw_pt_front_end base;
  46    struct draw_context *draw;
  47
  48    unsigned in[CACHE_MAX];
  49    ushort out[CACHE_MAX];
  50
  51    ushort draw_elts[DRAW_MAX];
  52    unsigned fetch_elts[FETCH_MAX];
  53
  54    unsigned draw_count;
  55    unsigned fetch_count;
  56    unsigned fetch_max;
  57
  58    struct draw_pt_middle_end *middle;
  59
  60    unsigned input_prim;
  61    unsigned output_prim;
  62
  63    unsigned middle_prim;
  64    unsigned opt;
  65 };
  66
  67 static INLINE void
  68 vcache_flush( struct vcache_frontend *vcache )
  69 {
  70    if (vcache->middle_prim != vcache->output_prim) {
  71       vcache->middle_prim = vcache->output_prim;
  72       vcache->middle->prepare( vcache->middle,
  73                                vcache->middle_prim,
  74                                vcache->opt,
  75                                &vcache->fetch_max );
  76    }
  77
  78    if (vcache->draw_count) {
  79       vcache->middle->run( vcache->middle,
  80                            vcache->fetch_elts,
  81                            vcache->fetch_count,
  82                            vcache->draw_elts,
  83                            vcache->draw_count );
  84    }
  85
  86    memset(vcache->in, ~0, sizeof(vcache->in));
  87    vcache->fetch_count = 0;
  88    vcache->draw_count = 0;
  89 }
  90
  91 static INLINE void
  92 vcache_check_flush( struct vcache_frontend *vcache )
  93 {
  94    if ( vcache->draw_count + 6 >= DRAW_MAX ||
  95         vcache->fetch_count + 4 >= FETCH_MAX )
  96    {
  97       vcache_flush( vcache );
  98    }
  99 }
 100
 101
 102 static INLINE void
 103 vcache_elt( struct vcache_frontend *vcache,
 104             unsigned felt,
 105             ushort flags )
 106 {
 107    unsigned idx = felt % CACHE_MAX;
 108
 109    if (vcache->in[idx] != felt) {
 110       assert(vcache->fetch_count < FETCH_MAX);
 111
 112       vcache->in[idx] = felt;
 113       vcache->out[idx] = (ushort)vcache->fetch_count;
 114       vcache->fetch_elts[vcache->fetch_count++] = felt;
 115    }
 116
 117    vcache->draw_elts[vcache->draw_count++] = vcache->out[idx] | flags;
 118 }
 119
 120
 121
 122 static INLINE void
 123 vcache_triangle( struct vcache_frontend *vcache,
 124                  unsigned i0,
 125                  unsigned i1,
 126                  unsigned i2 )
 127 {
 128    vcache_elt(vcache, i0, 0);
 129    vcache_elt(vcache, i1, 0);
 130    vcache_elt(vcache, i2, 0);
 131    vcache_check_flush(vcache);
 132 }
 133
 134
 135 static INLINE void
 136 vcache_triangle_flags( struct vcache_frontend *vcache,
 137                        ushort flags,
 138                        unsigned i0,
 139                        unsigned i1,
 140                        unsigned i2 )
 141 {
 142    vcache_elt(vcache, i0, flags);
 143    vcache_elt(vcache, i1, 0);
 144    vcache_elt(vcache, i2, 0);
 145    vcache_check_flush(vcache);
 146 }
 147
 148 static INLINE void
 149 vcache_line( struct vcache_frontend *vcache,
 150              unsigned i0,
 151              unsigned i1 )
 152 {
 153    vcache_elt(vcache, i0, 0);
 154    vcache_elt(vcache, i1, 0);
 155    vcache_check_flush(vcache);
 156 }
 157
 158
 159 static INLINE void
 160 vcache_line_flags( struct vcache_frontend *vcache,
 161                    ushort flags,
 162                    unsigned i0,
 163                    unsigned i1 )
 164 {
 165    vcache_elt(vcache, i0, flags);
 166    vcache_elt(vcache, i1, 0);
 167    vcache_check_flush(vcache);
 168 }
 169
 170
 171 static INLINE void
 172 vcache_point( struct vcache_frontend *vcache,
 173               unsigned i0 )
 174 {
 175    vcache_elt(vcache, i0, 0);
 176    vcache_check_flush(vcache);
 177 }
 178
 179 static INLINE void
 180 vcache_quad( struct vcache_frontend *vcache,
 181              unsigned i0,
 182              unsigned i1,
 183              unsigned i2,
 184              unsigned i3 )
 185 {
 186    if (vcache->draw->rasterizer->flatshade_first) {
 187       /* pass last quad vertex as first triangle vertex */
 188       vcache_triangle( vcache, i3, i0, i1 );
 189       vcache_triangle( vcache, i3, i1, i2 );
 190    }
 191    else {
 192       /* pass last quad vertex as last triangle vertex */
 193       vcache_triangle( vcache, i0, i1, i3 );
 194       vcache_triangle( vcache, i1, i2, i3 );
 195    }
 196 }
 197
 198 static INLINE void
 199 vcache_ef_quad( struct vcache_frontend *vcache,
 200                 unsigned i0,
 201                 unsigned i1,
 202                 unsigned i2,
 203                 unsigned i3 )
 204 {
 205    if (vcache->draw->rasterizer->flatshade_first) {
 206       /* pass last quad vertex as first triangle vertex */
 207       vcache_triangle_flags( vcache,
 208                              ( DRAW_PIPE_RESET_STIPPLE |
 209                                DRAW_PIPE_EDGE_FLAG_0 |
 210                                DRAW_PIPE_EDGE_FLAG_1 ),
 211                              i3, i0, i1 );
 212
 213       vcache_triangle_flags( vcache,
 214                              ( DRAW_PIPE_EDGE_FLAG_1 |
 215                                DRAW_PIPE_EDGE_FLAG_2 ),
 216                              i3, i1, i2 );
 217    }
 218    else {
 219       /* pass last quad vertex as last triangle vertex */
 220       vcache_triangle_flags( vcache,
 221                              ( DRAW_PIPE_RESET_STIPPLE |
 222                                DRAW_PIPE_EDGE_FLAG_0 |
 223                                DRAW_PIPE_EDGE_FLAG_2 ),
 224                              i0, i1, i3 );
 225
 226       vcache_triangle_flags( vcache,
 227                              ( DRAW_PIPE_EDGE_FLAG_0 |
 228                                DRAW_PIPE_EDGE_FLAG_1 ),
 229                              i1, i2, i3 );
 230    }
 231 }
 232
 233 /* At least for now, we're back to using a template include file for
 234  * this.  The two paths aren't too different though - it may be
 235  * possible to reunify them.
 236  */
 237 #define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle_flags(vc,flags,i0,i1,i2)
 238 #define QUAD(vc,i0,i1,i2,i3)        vcache_ef_quad(vc,i0,i1,i2,i3)
 239 #define LINE(vc,flags,i0,i1)        vcache_line_flags(vc,flags,i0,i1)
 240 #define POINT(vc,i0)                vcache_point(vc,i0)
 241 #define FUNC vcache_run_extras
 242 #include "draw_pt_vcache_tmp.h"
 243
 244 #define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle(vc,i0,i1,i2)
 245 #define QUAD(vc,i0,i1,i2,i3)        vcache_quad(vc,i0,i1,i2,i3)
 246 #define LINE(vc,flags,i0,i1)        vcache_line(vc,i0,i1)
 247 #define POINT(vc,i0)                vcache_point(vc,i0)
 248 #define FUNC vcache_run
 249 #include "draw_pt_vcache_tmp.h"
 250
 251 static INLINE void
 252 rebase_uint_elts( const unsigned *src,
 253                   unsigned count,
 254                   int delta,
 255                   ushort *dest )
 256 {
 257    unsigned i;
 258
 259    for (i = 0; i < count; i++)
 260       dest[i] = (ushort)(src[i] + delta);
 261 }
 262
 263 static INLINE void
 264 rebase_ushort_elts( const ushort *src,
 265                     unsigned count,
 266                     int delta,
 267                                 ushort *dest )
 268 {
 269    unsigned i;
 270
 271    for (i = 0; i < count; i++)
 272       dest[i] = (ushort)(src[i] + delta);
 273 }
 274
 275 static INLINE void
 276 rebase_ubyte_elts( const ubyte *src,
 277                    unsigned count,
 278                    int delta,
 279                    ushort *dest )
 280 {
 281    unsigned i;
 282
 283    for (i = 0; i < count; i++)
 284       dest[i] = (ushort)(src[i] + delta);
 285 }
 286
 287
 288
 289 static INLINE void
 290 translate_uint_elts( const unsigned *src,
 291                      unsigned count,
 292                      ushort *dest )
 293 {
 294    unsigned i;
 295
 296    for (i = 0; i < count; i++)
 297       dest[i] = (ushort)(src[i]);
 298 }
 299
 300 static INLINE void
 301 translate_ushort_elts( const ushort *src,
 302                        unsigned count,
 303                        ushort *dest )
 304 {
 305    unsigned i;
 306
 307    for (i = 0; i < count; i++)
 308       dest[i] = (ushort)(src[i]);
 309 }
 310
 311 static INLINE void
 312 translate_ubyte_elts( const ubyte *src,
 313                       unsigned count,
 314                       ushort *dest )
 315 {
 316    unsigned i;
 317
 318    for (i = 0; i < count; i++)
 319       dest[i] = (ushort)(src[i]);
 320 }
 321
 322
 323
 324
 325 #if 0
 326 static INLINE enum pipe_format
 327 format_from_get_elt( pt_elt_func get_elt )
 328 {
 329    switch (draw->pt.user.eltSize) {
 330    case 1: return PIPE_FORMAT_R8_UNORM;
 331    case 2: return PIPE_FORMAT_R16_UNORM;
 332    case 4: return PIPE_FORMAT_R32_UNORM;
 333    default: return PIPE_FORMAT_NONE;
 334    }
 335 }
 336 #endif
 337
 338 static INLINE void
 339 vcache_check_run( struct draw_pt_front_end *frontend,
 340                   pt_elt_func get_elt,
 341                   const void *elts,
 342                   int elt_bias,
 343                   unsigned draw_count )
 344 {
 345    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
 346    struct draw_context *draw = vcache->draw;
 347    unsigned min_index = draw->pt.user.min_index;
 348    unsigned max_index = draw->pt.user.max_index;
 349    unsigned index_size = draw->pt.user.eltSize;
 350    unsigned fetch_count = max_index + 1 - min_index;
 351    const ushort *transformed_elts;
 352    ushort *storage = NULL;
 353    boolean ok = FALSE;
 354
 355
 356    if (0) debug_printf("fetch_count %d fetch_max %d draw_count %d\n", fetch_count,
 357                        vcache->fetch_max,
 358                        draw_count);
 359
 360    if (elt_bias + max_index >= DRAW_PIPE_MAX_VERTICES ||
 361        fetch_count >= UNDEFINED_VERTEX_ID ||
 362        fetch_count > draw_count) {
 363       if (0) debug_printf("fail\n");
 364       goto fail;
 365    }
 366
 367    if (vcache->middle_prim != vcache->input_prim) {
 368       vcache->middle_prim = vcache->input_prim;
 369       vcache->middle->prepare( vcache->middle,
 370                                vcache->middle_prim,
 371                                vcache->opt,
 372                                &vcache->fetch_max );
 373    }
 374
 375
 376    assert((elt_bias >= 0 && min_index + elt_bias >= min_index) ||
 377           (elt_bias <  0 && min_index + elt_bias <  min_index));
 378
 379    if (min_index == 0 &&
 380        index_size == 2)
 381    {
 382       transformed_elts = (const ushort *)elts;
 383    }
 384    else
 385    {
 386       storage = MALLOC( draw_count * sizeof(ushort) );
 387       if (!storage)
 388          goto fail;
 389
 390       if (min_index == 0) {
 391          switch(index_size) {
 392          case 1:
 393             translate_ubyte_elts( (const ubyte *)elts,
 394                                   draw_count,
 395                                   storage );
 396             break;
 397
 398          case 2:
 399             translate_ushort_elts( (const ushort *)elts,
 400                                    draw_count,
 401                                    storage );
 402             break;
 403
 404          case 4:
 405             translate_uint_elts( (const uint *)elts,
 406                                  draw_count,
 407                                  storage );
 408             break;
 409
 410          default:
 411             assert(0);
 412             FREE(storage);
 413             return;
 414          }
 415       }
 416       else {
 417          switch(index_size) {
 418          case 1:
 419             rebase_ubyte_elts( (const ubyte *)elts,
 420                                   draw_count,
 421                                   0 - (int)min_index,
 422                                   storage );
 423             break;
 424
 425          case 2:
 426             rebase_ushort_elts( (const ushort *)elts,
 427                                    draw_count,
 428                                    0 - (int)min_index,
 429                                    storage );
 430             break;
 431
 432          case 4:
 433             rebase_uint_elts( (const uint *)elts,
 434                                  draw_count,
 435                                  0 - (int)min_index,
 436                                  storage );
 437             break;
 438
 439          default:
 440             assert(0);
 441             FREE(storage);
 442             return;
 443          }
 444       }
 445       transformed_elts = storage;
 446    }
 447
 448    if (fetch_count < UNDEFINED_VERTEX_ID)
 449       ok = vcache->middle->run_linear_elts( vcache->middle,
 450                                             min_index + elt_bias, /* start */
 451                                             fetch_count,
 452                                             transformed_elts,
 453                                             draw_count );
 454
 455    FREE(storage);
 456
 457    if (ok)
 458       return;
 459
 460    debug_printf("failed to execute atomic draw elts for %d/%d, splitting up\n",
 461                 fetch_count, draw_count);
 462
 463  fail:
 464    vcache_run( frontend, get_elt, elts, elt_bias, draw_count );
 465 }
 466
 467
 468
 469
 470 static void
 471 vcache_prepare( struct draw_pt_front_end *frontend,
 472                 unsigned prim,
 473                 struct draw_pt_middle_end *middle,
 474                 unsigned opt )
 475 {
 476    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
 477
 478    if (opt & PT_PIPELINE)
 479    {
 480       vcache->base.run = vcache_run_extras;
 481    }
 482    else
 483    {
 484       vcache->base.run = vcache_check_run;
 485    }
 486
 487    vcache->input_prim = prim;
 488    vcache->output_prim = u_reduced_prim(prim);
 489
 490    vcache->middle = middle;
 491    vcache->opt = opt;
 492
 493    /* Have to run prepare here, but try and guess a good prim for
 494     * doing so:
 495     */
 496    vcache->middle_prim = (opt & PT_PIPELINE) ? vcache->output_prim : vcache->input_prim;
 497    middle->prepare( middle, vcache->middle_prim, opt, &vcache->fetch_max );
 498 }
 499
 500
 501
 502
 503 static void
 504 vcache_finish( struct draw_pt_front_end *frontend )
 505 {
 506    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
 507    vcache->middle->finish( vcache->middle );
 508    vcache->middle = NULL;
 509 }
 510
 511 static void
 512 vcache_destroy( struct draw_pt_front_end *frontend )
 513 {
 514    FREE(frontend);
 515 }
 516
 517
 518 struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw )
 519 {
 520    struct vcache_frontend *vcache = CALLOC_STRUCT( vcache_frontend );
 521    if (vcache == NULL)
 522       return NULL;
 523
 524    vcache->base.prepare = vcache_prepare;
 525    vcache->base.run     = NULL;
 526    vcache->base.finish  = vcache_finish;
 527    vcache->base.destroy = vcache_destroy;
 528    vcache->draw = draw;
 529
 530    memset(vcache->in, ~0, sizeof(vcache->in));
 531
 532    return &vcache->base;
 533 }