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    memset(&c, 0, sizeof(c));
 146    memcpy(&c.key, key, sizeof(*key));
 147
 148    c.fp = fp;
 149    c.env_param = brw->intel.ctx.FragmentProgram.Parameters;
 150
 151
 152    /* Augment fragment program.  Add instructions for pre- and
 153     * post-fragment-program tasks such as interpolation and fogging.
 154     */
 155    brw_wm_pass_fp(&c);
 156
 157    /* Translate to intermediate representation.  Build register usage
 158     * chains.
 159     */
 160    brw_wm_pass0(&c);
 161
 162    /* Dead code removal.
 163     */
 164    brw_wm_pass1(&c);
 165
 166    /* Hal optimization
 167     */
 168    brw_wm_pass_hal (&c);
 169
 170    /* Register allocation.
 171     */
 172    c.grf_limit = BRW_WM_MAX_GRF/2;
 173
 174    /* This is where we start emitting gen4 code:
 175     */
 176    brw_init_compile(&c.func);
 177
 178    brw_wm_pass2(&c);
 179
 180    c.prog_data.total_grf = c.max_wm_grf;
 181    c.prog_data.total_scratch = c.last_scratch ? c.last_scratch + 0x40 : 0;
 182
 183    /* Emit GEN4 code.
 184     */
 185    brw_wm_emit(&c);
 186
 187    /* get the program
 188     */
 189    program = brw_get_program(&c.func, &program_size);
 190
 191    /*
 192     */
 193    brw->wm.prog_gs_offset = brw_upload_cache( &brw->cache[BRW_WM_PROG],
 194                                               &c.key,
 195                                               sizeof(c.key),
 196                                               program,
 197                                               program_size,
 198                                               &c.prog_data,
 199                                               &brw->wm.prog_data );
 200 }
 201
 202
 203
 204 static void brw_wm_populate_key( struct brw_context *brw,
 205                                  struct brw_wm_prog_key *key )
 206 {
 207    /* BRW_NEW_FRAGMENT_PROGRAM */
 208    struct brw_fragment_program *fp =
 209       (struct brw_fragment_program *)brw->fragment_program;
 210    GLuint lookup = 0;
 211    GLuint line_aa;
 212    GLuint i;
 213
 214    memset(key, 0, sizeof(*key));
 215
 216    /* Build the index for table lookup
 217     */
 218    /* _NEW_COLOR */
 219    if (fp->program.UsesKill ||
 220        brw->attribs.Color->AlphaEnabled)
 221       lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 222
 223    if (fp->program.Base.OutputsWritten & (1<<FRAG_RESULT_DEPR))
 224       lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 225
 226    /* _NEW_DEPTH */
 227    if (brw->attribs.Depth->Test)
 228       lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 229
 230    if (brw->attribs.Depth->Test &&
 231        brw->attribs.Depth->Mask) /* ?? */
 232       lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 233
 234    /* _NEW_STENCIL */
 235    if (brw->attribs.Stencil->Enabled) {
 236       lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 237
 238       if (brw->attribs.Stencil->WriteMask[0] ||
 239           (brw->attribs.Stencil->TestTwoSide && brw->attribs.Stencil->WriteMask[1]))
 240          lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 241    }
 242
 243    /* XXX: when should this be disabled?
 244     */
 245    if (1)
 246       lookup |= IZ_EARLY_DEPTH_TEST_BIT;
 247
 248
 249    line_aa = AA_NEVER;
 250
 251    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 252    if (brw->attribs.Line->SmoothFlag) {
 253       if (brw->intel.reduced_primitive == GL_LINES) {
 254          line_aa = AA_ALWAYS;
 255       }
 256       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 257          if (brw->attribs.Polygon->FrontMode == GL_LINE) {
 258             line_aa = AA_SOMETIMES;
 259
 260             if (brw->attribs.Polygon->BackMode == GL_LINE ||
 261                 (brw->attribs.Polygon->CullFlag &&
 262                  brw->attribs.Polygon->CullFaceMode == GL_BACK))
 263                line_aa = AA_ALWAYS;
 264          }
 265          else if (brw->attribs.Polygon->BackMode == GL_LINE) {
 266             line_aa = AA_SOMETIMES;
 267
 268             if ((brw->attribs.Polygon->CullFlag &&
 269                  brw->attribs.Polygon->CullFaceMode == GL_FRONT))
 270                line_aa = AA_ALWAYS;
 271          }
 272       }
 273    }
 274
 275    brw_wm_lookup_iz(line_aa,
 276                     lookup,
 277                     key);
 278
 279
 280    /* BRW_NEW_WM_INPUT_DIMENSIONS */
 281    key->projtex_mask = brw->wm.input_size_masks[4-1];
 282
 283    /* _NEW_LIGHT */
 284    key->flat_shade = (brw->attribs.Light->ShadeModel == GL_FLAT);
 285
 286    /* _NEW_TEXTURE */
 287    for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
 288       const struct gl_texture_unit *unit = &brw->attribs.Texture->Unit[i];
 289       const struct gl_texture_object *t = unit->_Current;
 290
 291       if (unit->_ReallyEnabled) {
 292
 293          if (t->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB &&
 294              t->Image[0][t->BaseLevel]->_BaseFormat == GL_DEPTH_COMPONENT) {
 295             key->shadowtex_mask |= 1<<i;
 296          }
 297
 298          if (t->Image[0][t->BaseLevel]->InternalFormat == GL_YCBCR_MESA)
 299             key->yuvtex_mask |= 1<<i;
 300       }
 301    }
 302
 303
 304    /* Extra info:
 305     */
 306    key->program_string_id = fp->id;
 307
 308 }
 309
 310
 311 static void brw_upload_wm_prog( struct brw_context *brw )
 312 {
 313    struct brw_wm_prog_key key;
 314    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 315       brw->fragment_program;
 316
 317    brw_wm_populate_key(brw, &key);
 318
 319    /* Make an early check for the key.
 320     */
 321    if (brw_search_cache(&brw->cache[BRW_WM_PROG],
 322                         &key, sizeof(key),
 323                         &brw->wm.prog_data,
 324                         &brw->wm.prog_gs_offset))
 325       return;
 326
 327    do_wm_prog(brw, fp, &key);
 328 }
 329
 330
 331 /* See brw_wm.c:
 332  */
 333 const struct brw_tracked_state brw_wm_prog = {
 334    .dirty = {
 335       .mesa  = (_NEW_COLOR |
 336                 _NEW_DEPTH |
 337                 _NEW_STENCIL |
 338                 _NEW_POLYGON |
 339                 _NEW_LINE |
 340                 _NEW_LIGHT |
 341                 _NEW_TEXTURE),
 342       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 343                 BRW_NEW_WM_INPUT_DIMENSIONS |
 344                 BRW_NEW_REDUCED_PRIMITIVE),
 345       .cache = 0
 346    },
 347    .update = brw_upload_wm_prog
 348 };
 349