src/mesa/tnl/t_vertex_codegen.c

   1 /*
   2  * Copyright 2003 Tungsten Graphics, inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * on the rights to use, copy, modify, merge, publish, distribute, sub
   9  * license, and/or sell copies of the Software, and to permit persons to whom
  10  * the Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
  19  * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  20  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  21  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  22  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  23  *
  24  * Authors:
  25  *    Keith Whitwell <keithw@tungstengraphics.com>
  26  */
  27
  28 #include "glheader.h"
  29 #include "context.h"
  30 #include "colormac.h"
  31
  32 #include "t_context.h"
  33 #include "t_vertex.h"
  34
  35 #include "simple_list.h"
  36
  37 /* Another codegen scheme, hopefully portable to a few different
  38  * architectures without too much work.
  39  */
  40
  41
  42
  43
  44 static GLboolean emit_4f_viewport_4( struct tnl_clipspace_codegen *p )
  45 {
  46    return
  47       p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
  48       p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
  49       p->emit_mad(p, out(2), vp(10), in(2), vp(14)) &&
  50       p->emit_mov(p, out(3), in(3));
  51 }
  52
  53 static GLboolean emit_4f_viewport_3( struct tnl_clipspace_codegen *p )
  54 {
  55    return
  56       p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
  57       p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
  58       p->emit_mad(p, out(2), vp(10), in(2), vp(14)) &&
  59       p->emit_const(p, out(3), 1.0);
  60 }
  61
  62 static GLboolean emit_4f_viewport_2( struct tnl_clipspace_codegen *p )
  63 {
  64    return
  65       p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
  66       p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
  67       p->emit_mov(p, out(2), vp(14)) &&
  68       p->emit_const(p, out(3), 1.0);
  69 }
  70
  71 static GLboolean emit_4f_viewport_1( struct tnl_clipspace_codegen *p )
  72 {
  73    return
  74       p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
  75       p->emit_mov(p, out(1), vp(13)) &&
  76       p->emit_mov(p, out(2), vp(14)) &&
  77       p->emit_const(p, out(3), 1.0);
  78 }
  79
  80 static GLboolean emit_3f_viewport_3( struct tnl_clipspace_codegen *p )
  81 {
  82    return
  83       p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
  84       p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
  85       p->emit_mad(p, out(2), vp(10), in(2), vp(14));
  86 }
  87
  88 static GLboolean emit_3f_viewport_2( struct tnl_clipspace_codegen *p )
  89 {
  90    return
  91       p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
  92       p->emit_mad(p, out(1), vp(5), in(1), vp(13)) &&
  93       p->emit_mov(p, out(2), vp(14));
  94 }
  95
  96 static GLboolean emit_3f_viewport_1( struct tnl_clipspace_codegen *p )
  97 {
  98    return
  99       p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
 100       p->emit_mov(p, out(1), vp(13)) &&
 101       p->emit_mov(p, out(2), vp(14));
 102 }
 103
 104 static GLboolean emit_2f_viewport_2( struct tnl_clipspace_codegen *p )
 105 {
 106    return
 107       p->emit_mad(p, out(0), vp(0), in(0), vp(12)) &&
 108       p->emit_mad(p, out(1), vp(5), in(1), vp(13));
 109 }
 110
 111 static GLboolean emit_2f_viewport_1( struct tnl_clipspace_codegen *p )
 112 {
 113    return
 114       p->emit_mad(p, out(0), vp(0), in(0), vp(12));
 115 }
 116
 117
 118 static GLboolean emit_4f_4( struct tnl_clipspace_codegen *p )
 119 {
 120    return
 121       p->emit_mov(p, out(0), in(0)) &&
 122       p->emit_mov(p, out(1), in(1)) &&
 123       p->emit_mov(p, out(2), in(2)) &&
 124       p->emit_mov(p, out(3), in(3));
 125 }
 126
 127 static GLboolean emit_4f_3( struct tnl_clipspace_codegen *p )
 128 {
 129    return
 130       p->emit_mov(p, out(0),  in(0)) &&
 131       p->emit_mov(p, out(1),  in(1)) &&
 132       p->emit_mov(p, out(2),  in(2)) &&
 133       p->emit_const(p, out(3), 1.0);
 134 }
 135
 136 static GLboolean emit_4f_2( struct tnl_clipspace_codegen *p )
 137 {
 138    return
 139       p->emit_mov(p, out(0),  in(0)) &&
 140       p->emit_mov(p, out(1),  in(1)) &&
 141       p->emit_const(p, out(2), 0.0) &&
 142       p->emit_const(p, out(3), 1.0);
 143 }
 144
 145 static GLboolean emit_4f_1( struct tnl_clipspace_codegen *p )
 146 {
 147    return
 148       p->emit_mov(p, out(0),  in(0)) &&
 149       p->emit_const(p, out(1), 0.0) &&
 150       p->emit_const(p, out(2), 0.0) &&
 151       p->emit_const(p, out(3), 1.0);
 152 }
 153
 154 static GLboolean emit_3f_xyw_4( struct tnl_clipspace_codegen *p )
 155 {
 156    return
 157       p->emit_mov(p, out(0), in(0)) &&
 158       p->emit_mov(p, out(1), in(1)) &&
 159       p->emit_mov(p, out(2), in(3));
 160 }
 161
 162 static GLboolean emit_3f_xyw_err( struct tnl_clipspace_codegen *p )
 163 {
 164    assert(0);
 165    return GL_FALSE;
 166 }
 167
 168 static GLboolean emit_3f_3( struct tnl_clipspace_codegen *p )
 169 {
 170    return
 171       p->emit_mov(p, out(0), in(0)) &&
 172       p->emit_mov(p, out(1), in(1)) &&
 173       p->emit_mov(p, out(2), in(2));
 174 }
 175
 176 static GLboolean emit_3f_2( struct tnl_clipspace_codegen *p )
 177 {
 178    return
 179       p->emit_mov(p, out(0),   in(0)) &&
 180       p->emit_mov(p, out(1),   in(1)) &&
 181       p->emit_const(p, out(2), 0.0);
 182 }
 183
 184 static GLboolean emit_3f_1( struct tnl_clipspace_codegen *p )
 185 {
 186    return
 187       p->emit_mov(p, out(0),  in(0)) &&
 188       p->emit_const(p, out(1),  0.0) &&
 189       p->emit_const(p, out(2),  0.0);
 190 }
 191
 192
 193 static GLboolean emit_2f_2( struct tnl_clipspace_codegen *p )
 194 {
 195    return
 196       p->emit_mov(p, out(0),  in(0)) &&
 197       p->emit_mov(p, out(1),  in(1));
 198 }
 199
 200 static GLboolean emit_2f_1( struct tnl_clipspace_codegen *p )
 201 {
 202    return
 203       p->emit_mov(p, out(0),  in(0)) &&
 204       p->emit_const(p, out(1),  0.0);
 205 }
 206
 207 static GLboolean emit_1f_1( struct tnl_clipspace_codegen *p )
 208 {
 209    return
 210       p->emit_mov(p, out(0),  in(0));
 211 }
 212
 213 static GLboolean emit_4chan_4f_rgba_4( struct tnl_clipspace_codegen *p )
 214 {
 215    return
 216       p->emit_float_to_chan(p, out(0), in(0)) &&
 217       p->emit_float_to_chan(p, out(1), in(1)) &&
 218       p->emit_float_to_chan(p, out(2), in(2)) &&
 219       p->emit_float_to_chan(p, out(3), in(3));
 220 }
 221
 222 static GLboolean emit_4chan_4f_rgba_3( struct tnl_clipspace_codegen *p )
 223 {
 224    return
 225       p->emit_float_to_chan(p, out(0), in(0)) &&
 226       p->emit_float_to_chan(p, out(1), in(1)) &&
 227       p->emit_float_to_chan(p, out(2), in(2)) &&
 228       p->emit_const_chan(p, out(3), CHAN_MAX);
 229 }
 230
 231 static GLboolean emit_4chan_4f_rgba_2( struct tnl_clipspace_codegen *p )
 232 {
 233    return
 234       p->emit_float_to_chan(p, out(0), in(0)) &&
 235       p->emit_float_to_chan(p, out(1), in(1)) &&
 236       p->emit_const_chan(p, out(2), 0) &&
 237       p->emit_const_chan(p, out(3), CHAN_MAX);
 238 }
 239
 240 static GLboolean emit_4chan_4f_rgba_1( struct tnl_clipspace_codegen *p )
 241 {
 242    return
 243       p->emit_float_to_chan(p, out(0), in(0)) &&
 244       p->emit_const_chan(p, out(1), 0) &&
 245       p->emit_const_chan(p, out(2), 0) &&
 246       p->emit_const_chan(p, out(3), CHAN_MAX);
 247 }
 248
 249 static GLboolean emit_4ub_4f_rgba_4( struct tnl_clipspace_codegen *p )
 250 {
 251    return
 252       p->emit_float_to_ubyte(p, out(0), in(0)) &&
 253       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 254       p->emit_float_to_ubyte(p, out(2), in(2)) &&
 255       p->emit_float_to_ubyte(p, out(3), in(3));
 256 }
 257
 258 static GLboolean emit_4ub_4f_rgba_3( struct tnl_clipspace_codegen *p )
 259 {
 260    return
 261       p->emit_float_to_ubyte(p, out(0), in(0)) &&
 262       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 263       p->emit_float_to_ubyte(p, out(2), in(2)) &&
 264       p->emit_const_ubyte(p, out(3), 0xff);
 265 }
 266
 267 static GLboolean emit_4ub_4f_rgba_2( struct tnl_clipspace_codegen *p )
 268 {
 269    return
 270       p->emit_float_to_ubyte(p, out(0), in(0)) &&
 271       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 272       p->emit_const_ubyte(p, out(2), 0) &&
 273       p->emit_const_ubyte(p, out(3), 0xff);
 274 }
 275
 276 static GLboolean emit_4ub_4f_rgba_1( struct tnl_clipspace_codegen *p )
 277 {
 278    return
 279       p->emit_float_to_ubyte(p, out(0), in(0)) &&
 280       p->emit_const_ubyte(p, out(1), 0) &&
 281       p->emit_const_ubyte(p, out(2), 0) &&
 282       p->emit_const_ubyte(p, out(3), 0xff);
 283 }
 284
 285 static GLboolean emit_4ub_4f_bgra_4( struct tnl_clipspace_codegen *p )
 286 {
 287    return
 288       p->emit_float_to_ubyte(p, out(2), in(0)) &&
 289       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 290       p->emit_float_to_ubyte(p, out(0), in(2)) &&
 291       p->emit_float_to_ubyte(p, out(3), in(3));
 292 }
 293
 294 static GLboolean emit_4ub_4f_bgra_3( struct tnl_clipspace_codegen *p )
 295 {
 296    return
 297       p->emit_float_to_ubyte(p, out(2), in(0)) &&
 298       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 299       p->emit_float_to_ubyte(p, out(0), in(2)) &&
 300       p->emit_const_ubyte(p, out(3), 0xff);
 301 }
 302
 303 static GLboolean emit_4ub_4f_bgra_2( struct tnl_clipspace_codegen *p )
 304 {
 305    return
 306       p->emit_float_to_ubyte(p, out(2), in(0)) &&
 307       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 308       p->emit_const_ubyte(p, out(0), 0) &&
 309       p->emit_const_ubyte(p, out(3), 0xff);
 310 }
 311
 312 static GLboolean emit_4ub_4f_bgra_1( struct tnl_clipspace_codegen *p )
 313 {
 314    return
 315       p->emit_float_to_ubyte(p, out(2), in(0)) &&
 316       p->emit_const_ubyte(p, out(1), 0) &&
 317       p->emit_const_ubyte(p, out(0), 0) &&
 318       p->emit_const_ubyte(p, out(3), 0xff);
 319 }
 320
 321 static GLboolean emit_3ub_3f_rgb_3( struct tnl_clipspace_codegen *p )
 322 {
 323    return
 324       p->emit_float_to_ubyte(p, out(0), in(0)) &&
 325       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 326       p->emit_float_to_ubyte(p, out(2), in(2));
 327 }
 328
 329 static GLboolean emit_3ub_3f_rgb_2( struct tnl_clipspace_codegen *p )
 330 {
 331    return
 332       p->emit_float_to_ubyte(p, out(0), in(0)) &&
 333       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 334       p->emit_const_ubyte(p, out(2), 0);
 335 }
 336
 337 static GLboolean emit_3ub_3f_rgb_1( struct tnl_clipspace_codegen *p )
 338 {
 339    return
 340       p->emit_float_to_ubyte(p, out(0), in(0)) &&
 341       p->emit_const_ubyte(p, out(1), 0) &&
 342       p->emit_const_ubyte(p, out(2), 0);
 343 }
 344
 345 static GLboolean emit_3ub_3f_bgr_3( struct tnl_clipspace_codegen *p )
 346 {
 347    return
 348       p->emit_float_to_ubyte(p, out(2), in(0)) &&
 349       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 350       p->emit_float_to_ubyte(p, out(0), in(2));
 351 }
 352
 353 static GLboolean emit_3ub_3f_bgr_2( struct tnl_clipspace_codegen *p )
 354 {
 355    return
 356       p->emit_float_to_ubyte(p, out(2), in(0)) &&
 357       p->emit_float_to_ubyte(p, out(1), in(1)) &&
 358       p->emit_const_ubyte(p, out(0), 0);
 359 }
 360
 361 static GLboolean emit_3ub_3f_bgr_1( struct tnl_clipspace_codegen *p )
 362 {
 363    return
 364       p->emit_float_to_ubyte(p, out(2), in(0)) &&
 365       p->emit_const_ubyte(p, out(1), 0) &&
 366       p->emit_const_ubyte(p, out(0), 0);
 367 }
 368
 369
 370 static GLboolean emit_1ub_1f_1( struct tnl_clipspace_codegen *p )
 371 {
 372    return
 373       p->emit_float_to_ubyte(p, out(0), in(0));
 374 }
 375
 376
 377
 378
 379 static struct {
 380    const char *name;
 381    GLenum out_type;
 382    GLboolean need_vp;
 383    GLboolean (*emit[4])( struct tnl_clipspace_codegen * );
 384 } emit_info[EMIT_MAX] = {
 385
 386    { "1f", GL_FLOAT, GL_FALSE,
 387      { emit_1f_1, emit_1f_1, emit_1f_1, emit_1f_1 } },
 388
 389    { "2f", GL_FLOAT, GL_FALSE,
 390      { emit_2f_1, emit_2f_2, emit_2f_2, emit_2f_2 } },
 391
 392    { "3f", GL_FLOAT, GL_FALSE,
 393      { emit_3f_1, emit_3f_2, emit_3f_3, emit_3f_3 } },
 394
 395    { "4f", GL_FLOAT, GL_FALSE,
 396      { emit_4f_1, emit_4f_2, emit_4f_3, emit_4f_4 } },
 397
 398    { "2f_viewport", GL_FLOAT, GL_TRUE,
 399      { emit_2f_viewport_1, emit_2f_viewport_2, emit_2f_viewport_2,
 400        emit_2f_viewport_2 } },
 401
 402    { "3f_viewport", GL_FLOAT, GL_TRUE,
 403      { emit_3f_viewport_1, emit_3f_viewport_2, emit_3f_viewport_3,
 404        emit_3f_viewport_3 } },
 405
 406    { "4f_viewport", GL_FLOAT, GL_TRUE,
 407      { emit_4f_viewport_1, emit_4f_viewport_2, emit_4f_viewport_3,
 408        emit_4f_viewport_4 } },
 409
 410    { "3f_xyw", GL_FLOAT, GL_FALSE,
 411      { emit_3f_xyw_err, emit_3f_xyw_err, emit_3f_xyw_err,
 412        emit_3f_xyw_4 } },
 413
 414    { "1ub_1f", GL_UNSIGNED_BYTE, GL_FALSE,
 415      { emit_1ub_1f_1, emit_1ub_1f_1, emit_1ub_1f_1, emit_1ub_1f_1 } },
 416
 417    { "3ub_3f_rgb", GL_UNSIGNED_BYTE, GL_FALSE,
 418      { emit_3ub_3f_rgb_1, emit_3ub_3f_rgb_2, emit_3ub_3f_rgb_3,
 419        emit_3ub_3f_rgb_3 } },
 420
 421    { "3ub_3f_bgr", GL_UNSIGNED_BYTE, GL_FALSE,
 422      { emit_3ub_3f_bgr_1, emit_3ub_3f_bgr_2, emit_3ub_3f_bgr_3,
 423        emit_3ub_3f_bgr_3 } },
 424
 425    { "4ub_4f_rgba", GL_UNSIGNED_BYTE, GL_FALSE,
 426      { emit_4ub_4f_rgba_1, emit_4ub_4f_rgba_2, emit_4ub_4f_rgba_3,
 427        emit_4ub_4f_rgba_4 } },
 428
 429    { "4ub_4f_bgra", GL_UNSIGNED_BYTE, GL_FALSE,
 430      { emit_4ub_4f_bgra_1, emit_4ub_4f_bgra_2, emit_4ub_4f_bgra_3,
 431        emit_4ub_4f_bgra_4 } },
 432
 433    { "4chan_4f_rgba", CHAN_TYPE, GL_FALSE,
 434      { emit_4chan_4f_rgba_1, emit_4chan_4f_rgba_2, emit_4chan_4f_rgba_3,
 435        emit_4chan_4f_rgba_4 } },
 436
 437    { "pad", 0, 0,
 438      { 0, 0, 0, 0 } }
 439
 440 };
 441
 442
 443 /***********************************************************************
 444  * list(attrib, size) --> function
 445  *
 446  * Because of the dependence of size, this all has to take place after
 447  * the pipeline has been run.
 448  */
 449
 450 tnl_emit_func _tnl_codegen_emit( GLcontext *ctx )
 451 {
 452    struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
 453    struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
 454    struct tnl_clipspace_attr *a = vtx->attr;
 455    struct tnl_clipspace_codegen *p = &vtx->codegen;
 456    const GLuint count = vtx->attr_count;
 457    GLuint j;
 458
 459    /* Need a faster lookup, or is this linear scan of an MRU list good
 460     * enough?  MRU chosen based on the guess that consecutive VB's are
 461     * likely to be of the same format.  A hash of attributes and sizes
 462     * might be a better technique.
 463     *
 464     * With the vtx code now in place, it should be possible to track
 465     * changes to the sizes of input arrays (and state, of course) and
 466     * only invalidate this function when those sizes have changed.
 467     */
 468 #if 0
 469    foreach (l, p->codegen_list) {
 470       if (l->attr_count != count)
 471          continue;
 472
 473       /* Assumptions:
 474        *    a[j].vp will not change for a given attrib
 475        *    a[j].vertex_offset will not change nothing else has changed.
 476        */
 477       for (j = 0; j < count; j++)
 478          if (a[j].attrib != l->a[j].attrib ||
 479              a[j].sz != l->a[j].sz)
 480             break;
 481
 482       if (j == count) {
 483          move_to_head(l, p->codegen_list);
 484          return l->func;
 485       }
 486    }
 487 #endif
 488
 489    p->emit_header( p, vtx );
 490
 491    for (j = 0; j < count; j++) {
 492       GLuint sz = VB->AttribPtr[a[j].attrib]->size - 1;
 493       p->emit_attr_header( p, a, j,
 494                            emit_info[a[j].format].out_type,
 495                            emit_info[a[j].format].need_vp );
 496
 497       if (!emit_info[a[j].format].emit[sz]( p )) {
 498          fprintf(stderr, "codegen failed\n");
 499          return 0;
 500       }
 501
 502       p->emit_attr_footer( p );
 503    }
 504
 505    p->emit_footer( p );
 506
 507    return p->emit_store_func( p );
 508 }
 509