src/mesa/drivers/dri/i965/brw_wm.c

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a 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, sublicense, 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
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   */
  31
  32
  33 #include "brw_context.h"
  34 #include "brw_util.h"
  35 #include "brw_wm.h"
  36 #include "brw_state.h"
  37 #include "brw_hal.h"
  38
  39 #include "program.h"
  40 #include "program_instruction.h"
  41 #include "arbprogparse.h"
  42
  43
  44 GLuint brw_wm_nr_args( GLuint opcode )
  45 {
  46    switch (opcode) {
  47
  48    case WM_PIXELXY:
  49    case OPCODE_ABS:
  50    case OPCODE_FLR:
  51    case OPCODE_FRC:
  52    case OPCODE_SWZ:
  53    case OPCODE_MOV:
  54    case OPCODE_COS:
  55    case OPCODE_EX2:
  56    case OPCODE_LG2:
  57    case OPCODE_RCP:
  58    case OPCODE_RSQ:
  59    case OPCODE_SIN:
  60    case OPCODE_SCS:
  61    case OPCODE_TEX:
  62    case OPCODE_TXB:
  63    case OPCODE_TXP:
  64    case OPCODE_KIL:
  65    case OPCODE_LIT:
  66    case WM_CINTERP:
  67    case WM_WPOSXY:
  68       return 1;
  69
  70    case OPCODE_POW:
  71    case OPCODE_SUB:
  72    case OPCODE_SGE:
  73    case OPCODE_SLT:
  74    case OPCODE_ADD:
  75    case OPCODE_MAX:
  76    case OPCODE_MIN:
  77    case OPCODE_MUL:
  78    case OPCODE_XPD:
  79    case OPCODE_DP3:
  80    case OPCODE_DP4:
  81    case OPCODE_DPH:
  82    case OPCODE_DST:
  83    case WM_LINTERP:
  84    case WM_DELTAXY:
  85    case WM_PIXELW:
  86       return 2;
  87
  88    case WM_FB_WRITE:
  89    case WM_PINTERP:
  90    case OPCODE_MAD:
  91    case OPCODE_CMP:
  92    case OPCODE_LRP:
  93       return 3;
  94
  95    default:
  96       return 0;
  97    }
  98 }
  99
 100
 101 GLuint brw_wm_is_scalar_result( GLuint opcode )
 102 {
 103    switch (opcode) {
 104    case OPCODE_COS:
 105    case OPCODE_EX2:
 106    case OPCODE_LG2:
 107    case OPCODE_POW:
 108    case OPCODE_RCP:
 109    case OPCODE_RSQ:
 110    case OPCODE_SIN:
 111    case OPCODE_DP3:
 112    case OPCODE_DP4:
 113    case OPCODE_DPH:
 114    case OPCODE_DST:
 115       return 1;
 116
 117    default:
 118       return 0;
 119    }
 120 }
 121
 122
 123 static void brw_wm_pass_hal (struct brw_wm_compile *c)
 124 {
 125    static void (*hal_wm_pass) (struct brw_wm_compile *c);
 126    static GLboolean hal_tried;
 127
 128    if (!hal_tried)
 129    {
 130       hal_wm_pass = brw_hal_find_symbol ("intel_hal_wm_pass");
 131       hal_tried = 1;
 132    }
 133    if (hal_wm_pass)
 134       (*hal_wm_pass) (c);
 135 }
 136
 137 static void do_wm_prog( struct brw_context *brw,
 138                         struct brw_fragment_program *fp,
 139                         struct brw_wm_prog_key *key)
 140 {
 141    struct brw_wm_compile *c;
 142    const GLuint *program;
 143    GLuint program_size;
 144
 145    c = brw->wm.compile_data;
 146    if (c == NULL) {
 147      brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data));
 148      c = brw->wm.compile_data;
 149    } else {
 150      memset(c, 0, sizeof(*brw->wm.compile_data));
 151    }
 152    memcpy(&c->key, key, sizeof(*key));
 153
 154    c->fp = fp;
 155    c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
 156
 157
 158    /* Augment fragment program.  Add instructions for pre- and
 159     * post-fragment-program tasks such as interpolation and fogging.
 160     */
 161    brw_wm_pass_fp(c);
 162
 163    /* Translate to intermediate representation.  Build register usage
 164     * chains.
 165     */
 166    brw_wm_pass0(c);
 167
 168    /* Dead code removal.
 169     */
 170    brw_wm_pass1(c);
 171
 172    /* Hal optimization
 173     */
 174    brw_wm_pass_hal (c);
 175
 176    /* Register allocation.
 177     */
 178    c->grf_limit = BRW_WM_MAX_GRF/2;
 179
 180    /* This is where we start emitting gen4 code:
 181     */
 182    brw_init_compile(&c->func);
 183
 184    brw_wm_pass2(c);
 185
 186    c->prog_data.total_grf = c->max_wm_grf;
 187    if (c->last_scratch) {
 188       c->prog_data.total_scratch =
 189          c->last_scratch + 0x40;
 190    } else {
 191       c->prog_data.total_scratch = 0;
 192    }
 193
 194    /* Emit GEN4 code.
 195     */
 196    brw_wm_emit(c);
 197
 198    /* get the program
 199     */
 200    program = brw_get_program(&c->func, &program_size);
 201
 202    /*
 203     */
 204    brw->wm.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_WM_PROG],
 205                                               &c->key,
 206                                               sizeof(c->key),
 207                                               program,
 208                                               program_size,
 209                                               &c->prog_data,
 210                                               &brw->wm.prog_data );
 211 }
 212
 213
 214
 215 static void brw_wm_populate_key( struct brw_context *brw,
 216                                  struct brw_wm_prog_key *key )
 217 {
 218    /* BRW_NEW_FRAGMENT_PROGRAM */
 219    struct brw_fragment_program *fp =
 220       (struct brw_fragment_program *)brw->fragment_program;
 221    GLuint lookup = 0;
 222    GLuint line_aa;
 223    GLuint i;
 224
 225    memset(key, 0, sizeof(*key));
 226
 227    /* Build the index for table lookup
 228     */
 229    /* _NEW_COLOR */
 230    if (fp->program.UsesKill ||
 231        brw->attribs.Color->AlphaEnabled)
 232       lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 233
 234    if (fp->program.Base.OutputsWritten & (1<<FRAG_RESULT_DEPR))
 235       lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 236
 237    /* _NEW_DEPTH */
 238    if (brw->attribs.Depth->Test)
 239       lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 240
 241    if (brw->attribs.Depth->Test &&
 242        brw->attribs.Depth->Mask) /* ?? */
 243       lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 244
 245    /* _NEW_STENCIL */
 246    if (brw->attribs.Stencil->Enabled) {
 247       lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 248
 249       if (brw->attribs.Stencil->WriteMask[0] ||
 250           (brw->attribs.Stencil->TestTwoSide && brw->attribs.Stencil->WriteMask[1]))
 251          lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 252    }
 253
 254    /* XXX: when should this be disabled?
 255     */
 256    if (1)
 257       lookup |= IZ_EARLY_DEPTH_TEST_BIT;
 258
 259
 260    line_aa = AA_NEVER;
 261
 262    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 263    if (brw->attribs.Line->SmoothFlag) {
 264       if (brw->intel.reduced_primitive == GL_LINES) {
 265          line_aa = AA_ALWAYS;
 266       }
 267       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 268          if (brw->attribs.Polygon->FrontMode == GL_LINE) {
 269             line_aa = AA_SOMETIMES;
 270
 271             if (brw->attribs.Polygon->BackMode == GL_LINE ||
 272                 (brw->attribs.Polygon->CullFlag &&
 273                  brw->attribs.Polygon->CullFaceMode == GL_BACK))
 274                line_aa = AA_ALWAYS;
 275          }
 276          else if (brw->attribs.Polygon->BackMode == GL_LINE) {
 277             line_aa = AA_SOMETIMES;
 278
 279             if ((brw->attribs.Polygon->CullFlag &&
 280                  brw->attribs.Polygon->CullFaceMode == GL_FRONT))
 281                line_aa = AA_ALWAYS;
 282          }
 283       }
 284    }
 285
 286    brw_wm_lookup_iz(line_aa,
 287                     lookup,
 288                     key);
 289
 290
 291    /* BRW_NEW_WM_INPUT_DIMENSIONS */
 292    key->projtex_mask = brw->wm.input_size_masks[4-1];
 293
 294    /* _NEW_LIGHT */
 295    key->flat_shade = (brw->attribs.Light->ShadeModel == GL_FLAT);
 296
 297    /* _NEW_TEXTURE */
 298    for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
 299       const struct gl_texture_unit *unit = &brw->attribs.Texture->Unit[i];
 300       const struct gl_texture_object *t = unit->_Current;
 301
 302       if (unit->_ReallyEnabled) {
 303
 304          if (t->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB &&
 305              t->Image[0][t->BaseLevel]->_BaseFormat == GL_DEPTH_COMPONENT) {
 306             key->shadowtex_mask |= 1<<i;
 307          }
 308
 309          if (t->Image[0][t->BaseLevel]->InternalFormat == GL_YCBCR_MESA)
 310             key->yuvtex_mask |= 1<<i;
 311       }
 312    }
 313
 314
 315    /* Extra info:
 316     */
 317    key->program_string_id = fp->id;
 318
 319 }
 320
 321
 322 static void brw_upload_wm_prog( struct brw_context *brw )
 323 {
 324    struct brw_wm_prog_key key;
 325    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 326       brw->fragment_program;
 327
 328    brw_wm_populate_key(brw, &key);
 329
 330    /* Make an early check for the key.
 331     */
 332    if (brw_search_cache(&brw->cache[BRW_WM_PROG],
 333                         &key, sizeof(key),
 334                         &brw->wm.prog_data,
 335                         &brw->wm.prog_gs_offset))
 336       return;
 337
 338    do_wm_prog(brw, fp, &key);
 339 }
 340
 341
 342 /* See brw_wm.c:
 343  */
 344 const struct brw_tracked_state brw_wm_prog = {
 345    .dirty = {
 346       .mesa  = (_NEW_COLOR |
 347                 _NEW_DEPTH |
 348                 _NEW_STENCIL |
 349                 _NEW_POLYGON |
 350                 _NEW_LINE |
 351                 _NEW_LIGHT |
 352                 _NEW_TEXTURE),
 353       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 354                 BRW_NEW_WM_INPUT_DIMENSIONS |
 355                 BRW_NEW_REDUCED_PRIMITIVE),
 356       .cache = 0
 357    },
 358    .update = brw_upload_wm_prog
 359 };
 360