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    vcache_triangle( vcache, i0, i1, i3 );
 187    vcache_triangle( vcache, i1, i2, i3 );
 188 }
 189
 190 static INLINE void
 191 vcache_ef_quad( struct vcache_frontend *vcache,
 192                 unsigned i0,
 193                 unsigned i1,
 194                 unsigned i2,
 195                 unsigned i3 )
 196 {
 197    if (vcache->draw->rasterizer->flatshade_first) {
 198       vcache_triangle_flags( vcache,
 199                              ( DRAW_PIPE_RESET_STIPPLE |
 200                                DRAW_PIPE_EDGE_FLAG_0 |
 201                                DRAW_PIPE_EDGE_FLAG_1 ),
 202                              i0, i1, i2 );
 203
 204       vcache_triangle_flags( vcache,
 205                              ( DRAW_PIPE_EDGE_FLAG_2 |
 206                                DRAW_PIPE_EDGE_FLAG_1 ),
 207                              i0, i2, i3 );
 208    }
 209    else {
 210       vcache_triangle_flags( vcache,
 211                              ( DRAW_PIPE_RESET_STIPPLE |
 212                                DRAW_PIPE_EDGE_FLAG_0 |
 213                                DRAW_PIPE_EDGE_FLAG_2 ),
 214                              i0, i1, i3 );
 215
 216       vcache_triangle_flags( vcache,
 217                              ( DRAW_PIPE_EDGE_FLAG_0 |
 218                                DRAW_PIPE_EDGE_FLAG_1 ),
 219                              i1, i2, i3 );
 220    }
 221 }
 222
 223 /* At least for now, we're back to using a template include file for
 224  * this.  The two paths aren't too different though - it may be
 225  * possible to reunify them.
 226  */
 227 #define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle_flags(vc,flags,i0,i1,i2)
 228 #define QUAD(vc,i0,i1,i2,i3)        vcache_ef_quad(vc,i0,i1,i2,i3)
 229 #define LINE(vc,flags,i0,i1)        vcache_line_flags(vc,flags,i0,i1)
 230 #define POINT(vc,i0)                vcache_point(vc,i0)
 231 #define FUNC vcache_run_extras
 232 #include "draw_pt_vcache_tmp.h"
 233
 234 #define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle(vc,i0,i1,i2)
 235 #define QUAD(vc,i0,i1,i2,i3)        vcache_quad(vc,i0,i1,i2,i3)
 236 #define LINE(vc,flags,i0,i1)        vcache_line(vc,i0,i1)
 237 #define POINT(vc,i0)                vcache_point(vc,i0)
 238 #define FUNC vcache_run
 239 #include "draw_pt_vcache_tmp.h"
 240
 241 static INLINE void
 242 rebase_uint_elts( const unsigned *src,
 243                   unsigned count,
 244                   int delta,
 245                   ushort *dest )
 246 {
 247    unsigned i;
 248
 249    for (i = 0; i < count; i++)
 250       dest[i] = (ushort)(src[i] + delta);
 251 }
 252
 253 static INLINE void
 254 rebase_ushort_elts( const ushort *src,
 255                     unsigned count,
 256                     int delta,
 257                                 ushort *dest )
 258 {
 259    unsigned i;
 260
 261    for (i = 0; i < count; i++)
 262       dest[i] = (ushort)(src[i] + delta);
 263 }
 264
 265 static INLINE void
 266 rebase_ubyte_elts( const ubyte *src,
 267                    unsigned count,
 268                    int delta,
 269                    ushort *dest )
 270 {
 271    unsigned i;
 272
 273    for (i = 0; i < count; i++)
 274       dest[i] = (ushort)(src[i] + delta);
 275 }
 276
 277
 278
 279 static INLINE void
 280 translate_uint_elts( const unsigned *src,
 281                      unsigned count,
 282                      ushort *dest )
 283 {
 284    unsigned i;
 285
 286    for (i = 0; i < count; i++)
 287       dest[i] = (ushort)(src[i]);
 288 }
 289
 290 static INLINE void
 291 translate_ushort_elts( const ushort *src,
 292                        unsigned count,
 293                        ushort *dest )
 294 {
 295    unsigned i;
 296
 297    for (i = 0; i < count; i++)
 298       dest[i] = (ushort)(src[i]);
 299 }
 300
 301 static INLINE void
 302 translate_ubyte_elts( const ubyte *src,
 303                       unsigned count,
 304                       ushort *dest )
 305 {
 306    unsigned i;
 307
 308    for (i = 0; i < count; i++)
 309       dest[i] = (ushort)(src[i]);
 310 }
 311
 312
 313
 314
 315 #if 0
 316 static INLINE enum pipe_format
 317 format_from_get_elt( pt_elt_func get_elt )
 318 {
 319    switch (draw->pt.user.eltSize) {
 320    case 1: return PIPE_FORMAT_R8_UNORM;
 321    case 2: return PIPE_FORMAT_R16_UNORM;
 322    case 4: return PIPE_FORMAT_R32_UNORM;
 323    default: return PIPE_FORMAT_NONE;
 324    }
 325 }
 326 #endif
 327
 328 static INLINE void
 329 vcache_check_run( struct draw_pt_front_end *frontend,
 330                   pt_elt_func get_elt,
 331                   const void *elts,
 332                   unsigned draw_count )
 333 {
 334    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
 335    struct draw_context *draw = vcache->draw;
 336    unsigned min_index = draw->pt.user.min_index;
 337    unsigned max_index = draw->pt.user.max_index;
 338    unsigned index_size = draw->pt.user.eltSize;
 339    unsigned fetch_count = max_index + 1 - min_index;
 340    const ushort *transformed_elts;
 341    ushort *storage = NULL;
 342    boolean ok = FALSE;
 343
 344
 345    if (0) debug_printf("fetch_count %d fetch_max %d draw_count %d\n", fetch_count,
 346                        vcache->fetch_max,
 347                        draw_count);
 348
 349    if (max_index == 0xffffffff ||
 350        fetch_count > draw_count) {
 351       if (0) debug_printf("fail\n");
 352       goto fail;
 353    }
 354
 355    if (vcache->middle_prim != vcache->input_prim) {
 356       vcache->middle_prim = vcache->input_prim;
 357       vcache->middle->prepare( vcache->middle,
 358                                vcache->middle_prim,
 359                                vcache->opt,
 360                                &vcache->fetch_max );
 361    }
 362
 363
 364    if (min_index == 0 &&
 365        index_size == 2)
 366    {
 367       transformed_elts = (const ushort *)elts;
 368    }
 369    else
 370    {
 371       storage = MALLOC( draw_count * sizeof(ushort) );
 372       if (!storage)
 373          goto fail;
 374
 375       if (min_index == 0) {
 376          switch(index_size) {
 377          case 1:
 378             translate_ubyte_elts( (const ubyte *)elts,
 379                                   draw_count,
 380                                   storage );
 381             break;
 382
 383          case 2:
 384             translate_ushort_elts( (const ushort *)elts,
 385                                    draw_count,
 386                                    storage );
 387             break;
 388
 389          case 4:
 390             translate_uint_elts( (const uint *)elts,
 391                                  draw_count,
 392                                  storage );
 393             break;
 394
 395          default:
 396             assert(0);
 397             return;
 398          }
 399       }
 400       else {
 401          switch(index_size) {
 402          case 1:
 403             rebase_ubyte_elts( (const ubyte *)elts,
 404                                   draw_count,
 405                                   0 - (int)min_index,
 406                                   storage );
 407             break;
 408
 409          case 2:
 410             rebase_ushort_elts( (const ushort *)elts,
 411                                    draw_count,
 412                                    0 - (int)min_index,
 413                                    storage );
 414             break;
 415
 416          case 4:
 417             rebase_uint_elts( (const uint *)elts,
 418                                  draw_count,
 419                                  0 - (int)min_index,
 420                                  storage );
 421             break;
 422
 423          default:
 424             assert(0);
 425             return;
 426          }
 427       }
 428       transformed_elts = storage;
 429    }
 430
 431    if (fetch_count < UNDEFINED_VERTEX_ID)
 432       ok = vcache->middle->run_linear_elts( vcache->middle,
 433                                             min_index, /* start */
 434                                             fetch_count,
 435                                             transformed_elts,
 436                                             draw_count );
 437
 438    FREE(storage);
 439
 440    if (ok)
 441       return;
 442
 443    debug_printf("failed to execute atomic draw elts for %d/%d, splitting up\n",
 444                 fetch_count, draw_count);
 445
 446  fail:
 447    vcache_run( frontend, get_elt, elts, draw_count );
 448 }
 449
 450
 451
 452
 453 static void
 454 vcache_prepare( struct draw_pt_front_end *frontend,
 455                 unsigned prim,
 456                 struct draw_pt_middle_end *middle,
 457                 unsigned opt )
 458 {
 459    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
 460
 461    if (opt & PT_PIPELINE)
 462    {
 463       vcache->base.run = vcache_run_extras;
 464    }
 465    else
 466    {
 467       vcache->base.run = vcache_check_run;
 468    }
 469
 470    vcache->input_prim = prim;
 471    vcache->output_prim = u_reduced_prim(prim);
 472
 473    vcache->middle = middle;
 474    vcache->opt = opt;
 475
 476    /* Have to run prepare here, but try and guess a good prim for
 477     * doing so:
 478     */
 479    vcache->middle_prim = (opt & PT_PIPELINE) ? vcache->output_prim : vcache->input_prim;
 480    middle->prepare( middle, vcache->middle_prim, opt, &vcache->fetch_max );
 481 }
 482
 483
 484
 485
 486 static void
 487 vcache_finish( struct draw_pt_front_end *frontend )
 488 {
 489    struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
 490    vcache->middle->finish( vcache->middle );
 491    vcache->middle = NULL;
 492 }
 493
 494 static void
 495 vcache_destroy( struct draw_pt_front_end *frontend )
 496 {
 497    FREE(frontend);
 498 }
 499
 500
 501 struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw )
 502 {
 503    struct vcache_frontend *vcache = CALLOC_STRUCT( vcache_frontend );
 504    if (vcache == NULL)
 505       return NULL;
 506
 507    vcache->base.prepare = vcache_prepare;
 508    vcache->base.run     = NULL;
 509    vcache->base.finish  = vcache_finish;
 510    vcache->base.destroy = vcache_destroy;
 511    vcache->draw = draw;
 512
 513    memset(vcache->in, ~0, sizeof(vcache->in));
 514
 515    return &vcache->base;
 516 }