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