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 "draw/draw_context.h"
  35 #include "draw/draw_private.h"
  36 #include "draw/draw_pt.h"
  37
  38
  39 #define CACHE_MAX 256
  40 #define FETCH_MAX 256
  41 #define DRAW_MAX (16*1024)
  42
  43 struct vcache_frontend {
  44    struct draw_pt_front_end base;
  45    struct draw_context *draw;
  46
  47    unsigned in[CACHE_MAX];
  48    ushort out[CACHE_MAX];
  49
  50    ushort draw_elts[DRAW_MAX];
  51    unsigned fetch_elts[FETCH_MAX];
  52
  53    unsigned draw_count;
  54    unsigned fetch_count;
  55    unsigned fetch_max;
  56
  57    struct draw_pt_middle_end *middle;
  58
  59    unsigned input_prim;
  60    unsigned output_prim;
  61
  62    unsigned middle_prim;
  63    unsigned opt;
  64 };
  65
  66 static INLINE void
  67 vcache_flush( struct vcache_frontend *vcache )
  68 {
  69    if (vcache->middle_prim != vcache->output_prim) {
  70       vcache->middle_prim = vcache->output_prim;
  71       vcache->middle->prepare( vcache->middle,
  72                                vcache->middle_prim,
  73                                vcache->opt,
  74                                &vcache->fetch_max );
  75    }
  76
  77    if (vcache->draw_count) {
  78       vcache->middle->run( vcache->middle,
  79                            vcache->fetch_elts,
  80                            vcache->fetch_count,
  81                            vcache->draw_elts,
  82                            vcache->draw_count );
  83    }
  84
  85    memset(vcache->in, ~0, sizeof(vcache->in));
  86    vcache->fetch_count = 0;
  87    vcache->draw_count = 0;
  88 }
  89
  90 static INLINE void
  91 vcache_check_flush( struct vcache_frontend *vcache )
  92 {
  93    if ( vcache->draw_count + 6 >= DRAW_MAX ||
  94         vcache->fetch_count + 4 >= FETCH_MAX )
  95    {
  96       vcache_flush( vcache );
  97    }
  98 }
  99
 100
 101 static INLINE void
 102 vcache_elt( struct vcache_frontend *vcache,
 103             unsigned felt,
 104             ushort flags )
 105 {
 106    unsigned idx = felt % CACHE_MAX;
 107
 108    if (vcache->in[idx] != felt) {
 109       assert(vcache->fetch_count < FETCH_MAX);
 110
 111       vcache->in[idx] = felt;
 112       vcache->out[idx] = (ushort)vcache->fetch_count;
 113       vcache->fetch_elts[vcache->fetch_count++] = felt;
 114    }
 115
 116    vcache->draw_elts[vcache->draw_count++] = vcache->out[idx] | flags;
 117 }
 118
 119
 120
 121 static INLINE void
 122 vcache_triangle( struct vcache_frontend *vcache,
 123                  unsigned i0,
 124                  unsigned i1,
 125                  unsigned i2 )
 126 {
 127    vcache_elt(vcache, i0, 0);
 128    vcache_elt(vcache, i1, 0);
 129    vcache_elt(vcache, i2, 0);
 130    vcache_check_flush(vcache);
 131 }
 132
 133
 134 static INLINE void
 135 vcache_triangle_flags( struct vcache_frontend *vcache,
 136                        ushort flags,
 137                        unsigned i0,
 138                        unsigned i1,
 139                        unsigned i2 )
 140 {
 141    vcache_elt(vcache, i0, flags);
 142    vcache_elt(vcache, i1, 0);
 143    vcache_elt(vcache, i2, 0);
 144    vcache_check_flush(vcache);
 145 }
 146
 147 static INLINE void
 148 vcache_line( struct vcache_frontend *vcache,
 149              unsigned i0,
 150              unsigned i1 )
 151 {
 152    vcache_elt(vcache, i0, 0);
 153    vcache_elt(vcache, i1, 0);
 154    vcache_check_flush(vcache);
 155 }
 156
 157
 158 static INLINE void
 159 vcache_line_flags( struct vcache_frontend *vcache,
 160                    ushort flags,
 161                    unsigned i0,
 162                    unsigned i1 )
 163 {
 164    vcache_elt(vcache, i0, flags);
 165    vcache_elt(vcache, i1, 0);
 166    vcache_check_flush(vcache);
 167 }
 168
 169
 170 static INLINE void
 171 vcache_point( struct vcache_frontend *vcache,
 172               unsigned i0 )
 173 {
 174    vcache_elt(vcache, i0, 0);
 175    vcache_check_flush(vcache);
 176 }
 177
 178 static INLINE void
 179 vcache_quad( struct vcache_frontend *vcache,
 180              unsigned i0,
 181              unsigned i1,
 182              unsigned i2,
 183              unsigned i3 )
 184 {
 185    vcache_triangle( vcache, i0, i1, i3 );
 186    vcache_triangle( vcache, i1, i2, i3 );
 187 }
 188
 189 static INLINE void
 190 vcache_ef_quad( struct vcache_frontend *vcache,
 191                 unsigned i0,
 192                 unsigned i1,
 193                 unsigned i2,
 194                 unsigned i3 )
 195 {
 196    if (vcache->draw->rasterizer->flatshade_first) {
 197       vcache_triangle_flags( vcache,
 198                              ( DRAW_PIPE_RESET_STIPPLE |
 199                                DRAW_PIPE_EDGE_FLAG_0 |
 200                                DRAW_PIPE_EDGE_FLAG_1 ),
 201                              i0, i1, i2 );
 202
 203       vcache_triangle_flags( vcache,
 204                              ( DRAW_PIPE_EDGE_FLAG_2 |
 205                                DRAW_PIPE_EDGE_FLAG_1 ),
 206                              i0, i2, i3 );
 207    }
 208    else {
 209       vcache_triangle_flags( vcache,
 210                              ( DRAW_PIPE_RESET_STIPPLE |
 211                                DRAW_PIPE_EDGE_FLAG_0 |
 212                                DRAW_PIPE_EDGE_FLAG_2 ),
 213                              i0, i1, i3 );
 214
 215       vcache_triangle_flags( vcache,
 216                              ( DRAW_PIPE_EDGE_FLAG_0 |
 217                                DRAW_PIPE_EDGE_FLAG_1 ),
 218                              i1, i2, i3 );
 219    }
 220 }
 221
 222 /* At least for now, we're back to using a template include file for
 223  * this.  The two paths aren't too different though - it may be
 224  * possible to reunify them.
 225  */
 226 #define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle_flags(vc,flags,i0,i1,i2)
 227 #define QUAD(vc,i0,i1,i2,i3)        vcache_ef_quad(vc,i0,i1,i2,i3)
 228 #define LINE(vc,flags,i0,i1)        vcache_line_flags(vc,flags,i0,i1)
 229 #define POINT(vc,i0)                vcache_point(vc,i0)
 230 #define FUNC vcache_run_extras
 231 #include "draw_pt_vcache_tmp.h"
 232
 233 #define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle(vc,i0,i1,i2)
 234 #define QUAD(vc,i0,i1,i2,i3)        vcache_quad(vc,i0,i1,i2,i3)
 235 #define LINE(vc,flags,i0,i1)        vcache_line(vc,i0,i1)
 236 #define POINT(vc,i0)                vcache_point(vc,i0)
 237 #define FUNC vcache_run
 238 #include "draw_pt_vcache_tmp.h"
 239
 240 static INLINE void
 241 rebase_uint_elts( const unsigned *src,
 242                   unsigned count,
 243                   int delta,
 244                   ushort *dest )
 245 {
 246    unsigned i;
 247
 248    for (i = 0; i < count; i++)
 249       dest[i] = (ushort)(src[i] + delta);
 250 }
 251
 252 static INLINE void
 253 rebase_ushort_elts( const ushort *src,
 254                     unsigned count,
 255                     int delta,
 256                                 ushort *dest )
 257 {
 258    unsigned i;
 259
 260    for (i = 0; i < count; i++)
 261       dest[i] = (ushort)(src[i] + delta);
 262 }
 263
 264 static INLINE void
 265 rebase_ubyte_elts( const ubyte *src,
 266                    unsigned count,
 267                    int delta,
 268                    ushort *dest )
 269 {
 270    unsigned i;
 271
 272    for (i = 0; i < count; i++)
 273       dest[i] = (ushort)(src[i] + delta);
 274 }
 275
 276
 277
 278 static INLINE void
 279 translate_uint_elts( const unsigned *src,
 280                      unsigned count,
 281                      ushort *dest )
 282 {
 283    unsigned i;
 284
 285    for (i = 0; i < count; i++)
 286       dest[i] = (ushort)(src[i]);
 287 }
 288
 289 static INLINE void
 290 translate_ushort_elts( const ushort *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_ubyte_elts( const ubyte *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
 312
 313
 314 #if 0
 315 static INLINE enum pipe_format
 316 format_from_get_elt( pt_elt_func get_elt )
 317 {
 318    switch (draw->pt.user.eltSize) {
 319    case 1: return PIPE_FORMAT_R8_UNORM;
 320    case 2: return PIPE_FORMAT_R16_UNORM;
 321    case 4: return PIPE_FORMAT_R32_UNORM;
 322    default: return PIPE_FORMAT_NONE;
 323    }
 324 }
 325 #endif
 326
 327 static INLINE void
 328 vcache_check_run( struct draw_pt_front_end *frontend,
 329                   pt_elt_func get_elt,
 330                   const void *elts,
 331                   unsigned draw_count )
 332 {
 333    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
 334    struct draw_context *draw = vcache->draw;
 335    unsigned min_index = draw->pt.user.min_index;
 336    unsigned max_index = draw->pt.user.max_index;
 337    unsigned index_size = draw->pt.user.eltSize;
 338    unsigned fetch_count = max_index + 1 - min_index;
 339    const ushort *transformed_elts;
 340    ushort *storage = NULL;
 341    boolean ok = FALSE;
 342
 343
 344    if (0) debug_printf("fetch_count %d fetch_max %d draw_count %d\n", fetch_count,
 345                        vcache->fetch_max,
 346                        draw_count);
 347
 348    if (max_index == 0xffffffff ||
 349        fetch_count > draw_count) {
 350       if (0) debug_printf("fail\n");
 351       goto fail;
 352    }
 353
 354    if (vcache->middle_prim != vcache->input_prim) {
 355       vcache->middle_prim = vcache->input_prim;
 356       vcache->middle->prepare( vcache->middle,
 357                                vcache->middle_prim,
 358                                vcache->opt,
 359                                &vcache->fetch_max );
 360    }
 361
 362
 363    if (min_index == 0 &&
 364        index_size == 2)
 365    {
 366       transformed_elts = (const ushort *)elts;
 367    }
 368    else
 369    {
 370       storage = MALLOC( draw_count * sizeof(ushort) );
 371       if (!storage)
 372          goto fail;
 373
 374       if (min_index == 0) {
 375          switch(index_size) {
 376          case 1:
 377             translate_ubyte_elts( (const ubyte *)elts,
 378                                   draw_count,
 379                                   storage );
 380             break;
 381
 382          case 2:
 383             translate_ushort_elts( (const ushort *)elts,
 384                                    draw_count,
 385                                    storage );
 386             break;
 387
 388          case 4:
 389             translate_uint_elts( (const uint *)elts,
 390                                  draw_count,
 391                                  storage );
 392             break;
 393
 394          default:
 395             assert(0);
 396             return;
 397          }
 398       }
 399       else {
 400          switch(index_size) {
 401          case 1:
 402             rebase_ubyte_elts( (const ubyte *)elts,
 403                                   draw_count,
 404                                   0 - (int)min_index,
 405                                   storage );
 406             break;
 407
 408          case 2:
 409             rebase_ushort_elts( (const ushort *)elts,
 410                                    draw_count,
 411                                    0 - (int)min_index,
 412                                    storage );
 413             break;
 414
 415          case 4:
 416             rebase_uint_elts( (const uint *)elts,
 417                                  draw_count,
 418                                  0 - (int)min_index,
 419                                  storage );
 420             break;
 421
 422          default:
 423             assert(0);
 424             return;
 425          }
 426       }
 427       transformed_elts = storage;
 428    }
 429
 430    if (fetch_count < UNDEFINED_VERTEX_ID)
 431       ok = vcache->middle->run_linear_elts( vcache->middle,
 432                                             min_index, /* start */
 433                                             fetch_count,
 434                                             transformed_elts,
 435                                             draw_count );
 436
 437    FREE(storage);
 438
 439    if (ok)
 440       return;
 441
 442    debug_printf("failed to execute atomic draw elts for %d/%d, splitting up\n",
 443                 fetch_count, draw_count);
 444
 445  fail:
 446    vcache_run( frontend, get_elt, elts, draw_count );
 447 }
 448
 449
 450
 451
 452 static void
 453 vcache_prepare( struct draw_pt_front_end *frontend,
 454                 unsigned prim,
 455                 struct draw_pt_middle_end *middle,
 456                 unsigned opt )
 457 {
 458    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
 459
 460    if (opt & PT_PIPELINE)
 461    {
 462       vcache->base.run = vcache_run_extras;
 463    }
 464    else
 465    {
 466       vcache->base.run = vcache_check_run;
 467    }
 468
 469    vcache->input_prim = prim;
 470    vcache->output_prim = draw_pt_reduced_prim(prim);
 471
 472    vcache->middle = middle;
 473    vcache->opt = opt;
 474
 475    /* Have to run prepare here, but try and guess a good prim for
 476     * doing so:
 477     */
 478    vcache->middle_prim = (opt & PT_PIPELINE) ? vcache->output_prim : vcache->input_prim;
 479    middle->prepare( middle, vcache->middle_prim, opt, &vcache->fetch_max );
 480 }
 481
 482
 483
 484
 485 static void
 486 vcache_finish( struct draw_pt_front_end *frontend )
 487 {
 488    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
 489    vcache->middle->finish( vcache->middle );
 490    vcache->middle = NULL;
 491 }
 492
 493 static void
 494 vcache_destroy( struct draw_pt_front_end *frontend )
 495 {
 496    FREE(frontend);
 497 }
 498
 499
 500 struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw )
 501 {
 502    struct vcache_frontend *vcache = CALLOC_STRUCT( vcache_frontend );
 503    if (vcache == NULL)
 504       return NULL;
 505
 506    vcache->base.prepare = vcache_prepare;
 507    vcache->base.run     = NULL;
 508    vcache->base.finish  = vcache_finish;
 509    vcache->base.destroy = vcache_destroy;
 510    vcache->draw = draw;
 511
 512    memset(vcache->in, ~0, sizeof(vcache->in));
 513
 514    return &vcache->base;
 515 }