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 #include "brw_context.h"
  33 #include "brw_wm.h"
  34 #include "brw_state.h"
  35 #include "main/formats.h"
  36 #include "main/fbobject.h"
  37 #include "main/samplerobj.h"
  38 #include "program/prog_parameter.h"
  39
  40 #include "glsl/ralloc.h"
  41
  42 /** Return number of src args for given instruction */
  43 GLuint brw_wm_nr_args( GLuint opcode )
  44 {
  45    switch (opcode) {
  46    case WM_FRONTFACING:
  47    case WM_PIXELXY:
  48       return 0;
  49    case WM_CINTERP:
  50    case WM_WPOSXY:
  51    case WM_DELTAXY:
  52       return 1;
  53    case WM_LINTERP:
  54    case WM_PIXELW:
  55       return 2;
  56    case WM_FB_WRITE:
  57    case WM_PINTERP:
  58       return 3;
  59    default:
  60       assert(opcode < MAX_OPCODE);
  61       return _mesa_num_inst_src_regs(opcode);
  62    }
  63 }
  64
  65
  66 GLuint brw_wm_is_scalar_result( GLuint opcode )
  67 {
  68    switch (opcode) {
  69    case OPCODE_COS:
  70    case OPCODE_EX2:
  71    case OPCODE_LG2:
  72    case OPCODE_POW:
  73    case OPCODE_RCP:
  74    case OPCODE_RSQ:
  75    case OPCODE_SIN:
  76    case OPCODE_DP2:
  77    case OPCODE_DP3:
  78    case OPCODE_DP4:
  79    case OPCODE_DPH:
  80    case OPCODE_DST:
  81       return 1;
  82
  83    default:
  84       return 0;
  85    }
  86 }
  87
  88
  89 /**
  90  * Do GPU code generation for non-GLSL shader.  non-GLSL shaders have
  91  * no flow control instructions so we can more readily do SSA-style
  92  * optimizations.
  93  */
  94 static void
  95 brw_wm_non_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c)
  96 {
  97    /* Augment fragment program.  Add instructions for pre- and
  98     * post-fragment-program tasks such as interpolation and fogging.
  99     */
 100    brw_wm_pass_fp(c);
 101
 102    /* Translate to intermediate representation.  Build register usage
 103     * chains.
 104     */
 105    brw_wm_pass0(c);
 106
 107    /* Dead code removal.
 108     */
 109    brw_wm_pass1(c);
 110
 111    /* Register allocation.
 112     * Divide by two because we operate on 16 pixels at a time and require
 113     * two GRF entries for each logical shader register.
 114     */
 115    c->grf_limit = BRW_WM_MAX_GRF / 2;
 116
 117    brw_wm_pass2(c);
 118
 119    /* how many general-purpose registers are used */
 120    c->prog_data.reg_blocks = brw_register_blocks(c->max_wm_grf);
 121
 122    /* Emit GEN4 code.
 123     */
 124    brw_wm_emit(c);
 125 }
 126
 127
 128 /**
 129  * Return a bitfield where bit n is set if barycentric interpolation mode n
 130  * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
 131  */
 132 static unsigned
 133 brw_compute_barycentric_interp_modes(struct brw_context *brw,
 134                                      bool shade_model_flat,
 135                                      const struct gl_fragment_program *fprog)
 136 {
 137    unsigned barycentric_interp_modes = 0;
 138    int attr;
 139
 140    /* Loop through all fragment shader inputs to figure out what interpolation
 141     * modes are in use, and set the appropriate bits in
 142     * barycentric_interp_modes.
 143     */
 144    for (attr = 0; attr < FRAG_ATTRIB_MAX; ++attr) {
 145       enum glsl_interp_qualifier interp_qualifier =
 146          fprog->InterpQualifier[attr];
 147       bool is_centroid = fprog->IsCentroid & BITFIELD64_BIT(attr);
 148       bool is_gl_Color = attr == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_COL1;
 149
 150       /* Ignore unused inputs. */
 151       if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
 152          continue;
 153
 154       /* Ignore WPOS and FACE, because they don't require interpolation. */
 155       if (attr == FRAG_ATTRIB_WPOS || attr == FRAG_ATTRIB_FACE)
 156          continue;
 157
 158       /* Determine the set (or sets) of barycentric coordinates needed to
 159        * interpolate this variable.  Note that when
 160        * brw->needs_unlit_centroid_workaround is set, centroid interpolation
 161        * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
 162        * for lit pixels, so we need both sets of barycentric coordinates.
 163        */
 164       if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
 165          if (is_centroid) {
 166             barycentric_interp_modes |=
 167                1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
 168          }
 169          if (!is_centroid || brw->needs_unlit_centroid_workaround) {
 170             barycentric_interp_modes |=
 171                1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
 172          }
 173       } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
 174                  (!(shade_model_flat && is_gl_Color) &&
 175                   interp_qualifier == INTERP_QUALIFIER_NONE)) {
 176          if (is_centroid) {
 177             barycentric_interp_modes |=
 178                1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
 179          }
 180          if (!is_centroid || brw->needs_unlit_centroid_workaround) {
 181             barycentric_interp_modes |=
 182                1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
 183          }
 184       }
 185    }
 186
 187    return barycentric_interp_modes;
 188 }
 189
 190
 191 void
 192 brw_wm_payload_setup(struct brw_context *brw,
 193                      struct brw_wm_compile *c)
 194 {
 195    struct intel_context *intel = &brw->intel;
 196    bool uses_depth = (c->fp->program.Base.InputsRead &
 197                       (1 << FRAG_ATTRIB_WPOS)) != 0;
 198    unsigned barycentric_interp_modes = c->prog_data.barycentric_interp_modes;
 199    int i;
 200
 201    if (intel->gen >= 6) {
 202       /* R0-1: masks, pixel X/Y coordinates. */
 203       c->nr_payload_regs = 2;
 204       /* R2: only for 32-pixel dispatch.*/
 205
 206       /* R3-26: barycentric interpolation coordinates.  These appear in the
 207        * same order that they appear in the brw_wm_barycentric_interp_mode
 208        * enum.  Each set of coordinates occupies 2 registers if dispatch width
 209        * == 8 and 4 registers if dispatch width == 16.  Coordinates only
 210        * appear if they were enabled using the "Barycentric Interpolation
 211        * Mode" bits in WM_STATE.
 212        */
 213       for (i = 0; i < BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT; ++i) {
 214          if (barycentric_interp_modes & (1 << i)) {
 215             c->barycentric_coord_reg[i] = c->nr_payload_regs;
 216             c->nr_payload_regs += 2;
 217             if (c->dispatch_width == 16) {
 218                c->nr_payload_regs += 2;
 219             }
 220          }
 221       }
 222
 223       /* R27: interpolated depth if uses source depth */
 224       if (uses_depth) {
 225          c->source_depth_reg = c->nr_payload_regs;
 226          c->nr_payload_regs++;
 227          if (c->dispatch_width == 16) {
 228             /* R28: interpolated depth if not 8-wide. */
 229             c->nr_payload_regs++;
 230          }
 231       }
 232       /* R29: interpolated W set if GEN6_WM_USES_SOURCE_W.
 233        */
 234       if (uses_depth) {
 235          c->source_w_reg = c->nr_payload_regs;
 236          c->nr_payload_regs++;
 237          if (c->dispatch_width == 16) {
 238             /* R30: interpolated W if not 8-wide. */
 239             c->nr_payload_regs++;
 240          }
 241       }
 242       /* R31: MSAA position offsets. */
 243       /* R32-: bary for 32-pixel. */
 244       /* R58-59: interp W for 32-pixel. */
 245
 246       if (c->fp->program.Base.OutputsWritten &
 247           BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 248          c->source_depth_to_render_target = true;
 249          c->computes_depth = true;
 250       }
 251    } else {
 252       brw_wm_lookup_iz(intel, c);
 253    }
 254 }
 255
 256 bool
 257 brw_wm_prog_data_compare(const void *in_a, const void *in_b,
 258                          int aux_size, const void *in_key)
 259 {
 260    const struct brw_wm_prog_data *a = in_a;
 261    const struct brw_wm_prog_data *b = in_b;
 262
 263    /* Compare all the struct up to the pointers. */
 264    if (memcmp(a, b, offsetof(struct brw_wm_prog_data, param)))
 265       return false;
 266
 267    if (memcmp(a->param, b->param, a->nr_params * sizeof(void *)))
 268       return false;
 269
 270    if (memcmp(a->pull_param, b->pull_param, a->nr_pull_params * sizeof(void *)))
 271       return false;
 272
 273    return true;
 274 }
 275
 276 /**
 277  * All Mesa program -> GPU code generation goes through this function.
 278  * Depending on the instructions used (i.e. flow control instructions)
 279  * we'll use one of two code generators.
 280  */
 281 bool do_wm_prog(struct brw_context *brw,
 282                 struct gl_shader_program *prog,
 283                 struct brw_fragment_program *fp,
 284                 struct brw_wm_prog_key *key)
 285 {
 286    struct intel_context *intel = &brw->intel;
 287    struct brw_wm_compile *c;
 288    const GLuint *program;
 289    GLuint program_size;
 290
 291    c = brw->wm.compile_data;
 292    if (c == NULL) {
 293       brw->wm.compile_data = rzalloc(NULL, struct brw_wm_compile);
 294       c = brw->wm.compile_data;
 295       if (c == NULL) {
 296          /* Ouch - big out of memory problem.  Can't continue
 297           * without triggering a segfault, no way to signal,
 298           * so just return.
 299           */
 300          return false;
 301       }
 302    } else {
 303       void *instruction = c->instruction;
 304       void *prog_instructions = c->prog_instructions;
 305       void *vreg = c->vreg;
 306       void *refs = c->refs;
 307       memset(c, 0, sizeof(*brw->wm.compile_data));
 308       c->instruction = instruction;
 309       c->prog_instructions = prog_instructions;
 310       c->vreg = vreg;
 311       c->refs = refs;
 312    }
 313    memcpy(&c->key, key, sizeof(*key));
 314
 315    c->fp = fp;
 316    c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
 317
 318    brw_init_compile(brw, &c->func, c);
 319
 320    c->prog_data.barycentric_interp_modes =
 321       brw_compute_barycentric_interp_modes(brw, c->key.flat_shade,
 322                                            &fp->program);
 323
 324    if (prog && prog->_LinkedShaders[MESA_SHADER_FRAGMENT]) {
 325       if (!brw_wm_fs_emit(brw, c, prog))
 326          return false;
 327    } else {
 328       if (!c->instruction) {
 329          c->instruction = rzalloc_array(c, struct brw_wm_instruction, BRW_WM_MAX_INSN);
 330          c->prog_instructions = rzalloc_array(c, struct prog_instruction, BRW_WM_MAX_INSN);
 331          c->vreg = rzalloc_array(c, struct brw_wm_value, BRW_WM_MAX_VREG);
 332          c->refs = rzalloc_array(c, struct brw_wm_ref, BRW_WM_MAX_REF);
 333       }
 334
 335       /* Fallback for fixed function and ARB_fp shaders. */
 336       c->dispatch_width = 16;
 337       brw_wm_payload_setup(brw, c);
 338       brw_wm_non_glsl_emit(brw, c);
 339       c->prog_data.dispatch_width = 16;
 340    }
 341
 342    /* Scratch space is used for register spilling */
 343    if (c->last_scratch) {
 344       perf_debug("Fragment shader triggered register spilling.  "
 345                  "Try reducing the number of live scalar values to "
 346                  "improve performance.\n");
 347
 348       c->prog_data.total_scratch = brw_get_scratch_size(c->last_scratch);
 349
 350       brw_get_scratch_bo(intel, &brw->wm.scratch_bo,
 351                          c->prog_data.total_scratch * brw->max_wm_threads);
 352    }
 353
 354    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 355       fprintf(stderr, "\n");
 356
 357    /* get the program
 358     */
 359    program = brw_get_program(&c->func, &program_size);
 360
 361    brw_upload_cache(&brw->cache, BRW_WM_PROG,
 362                     &c->key, sizeof(c->key),
 363                     program, program_size,
 364                     &c->prog_data, sizeof(c->prog_data),
 365                     &brw->wm.prog_offset, &brw->wm.prog_data);
 366
 367    return true;
 368 }
 369
 370 static bool
 371 key_debug(const char *name, int a, int b)
 372 {
 373    if (a != b) {
 374       perf_debug("  %s %d->%d\n", name, a, b);
 375       return true;
 376    } else {
 377       return false;
 378    }
 379 }
 380
 381 bool
 382 brw_debug_recompile_sampler_key(const struct brw_sampler_prog_key_data *old_key,
 383                                 const struct brw_sampler_prog_key_data *key)
 384 {
 385    bool found = false;
 386
 387    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 388       found |= key_debug("EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 389                          old_key->swizzles[i], key->swizzles[i]);
 390    }
 391    found |= key_debug("GL_CLAMP enabled on any texture unit's 1st coordinate",
 392                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 393    found |= key_debug("GL_CLAMP enabled on any texture unit's 2nd coordinate",
 394                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 395    found |= key_debug("GL_CLAMP enabled on any texture unit's 3rd coordinate",
 396                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 397    found |= key_debug("GL_MESA_ycbcr texturing\n",
 398                       old_key->yuvtex_mask, key->yuvtex_mask);
 399    found |= key_debug("GL_MESA_ycbcr UV swapping\n",
 400                       old_key->yuvtex_swap_mask, key->yuvtex_swap_mask);
 401
 402    return found;
 403 }
 404
 405 void
 406 brw_wm_debug_recompile(struct brw_context *brw,
 407                        struct gl_shader_program *prog,
 408                        const struct brw_wm_prog_key *key)
 409 {
 410    struct brw_cache_item *c = NULL;
 411    const struct brw_wm_prog_key *old_key = NULL;
 412    bool found = false;
 413
 414    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 415
 416    for (unsigned int i = 0; i < brw->cache.size; i++) {
 417       for (c = brw->cache.items[i]; c; c = c->next) {
 418          if (c->cache_id == BRW_WM_PROG) {
 419             old_key = c->key;
 420
 421             if (old_key->program_string_id == key->program_string_id)
 422                break;
 423          }
 424       }
 425       if (c)
 426          break;
 427    }
 428
 429    if (!c) {
 430       perf_debug("  Didn't find previous compile in the shader cache for "
 431                  "debug\n");
 432       return;
 433    }
 434
 435    found |= key_debug("alphatest, computed depth, depth test, or depth write",
 436                       old_key->iz_lookup, key->iz_lookup);
 437    found |= key_debug("depth statistics", old_key->stats_wm, key->stats_wm);
 438    found |= key_debug("flat shading", old_key->flat_shade, key->flat_shade);
 439    found |= key_debug("number of color buffers", old_key->nr_color_regions, key->nr_color_regions);
 440    found |= key_debug("rendering to FBO", old_key->render_to_fbo, key->render_to_fbo);
 441    found |= key_debug("fragment color clamping", old_key->clamp_fragment_color, key->clamp_fragment_color);
 442    found |= key_debug("line smoothing", old_key->line_aa, key->line_aa);
 443    found |= key_debug("proj_attrib_mask", old_key->proj_attrib_mask, key->proj_attrib_mask);
 444    found |= key_debug("renderbuffer height", old_key->drawable_height, key->drawable_height);
 445    found |= key_debug("vertex shader outputs", old_key->vp_outputs_written, key->vp_outputs_written);
 446
 447    found |= brw_debug_recompile_sampler_key(&old_key->tex, &key->tex);
 448
 449    if (!found) {
 450       perf_debug("  Something else\n");
 451    }
 452 }
 453
 454 void
 455 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 456                                    const struct gl_program *prog,
 457                                    struct brw_sampler_prog_key_data *key)
 458 {
 459    for (int s = 0; s < MAX_SAMPLERS; s++) {
 460       key->swizzles[s] = SWIZZLE_NOOP;
 461
 462       if (!(prog->SamplersUsed & (1 << s)))
 463          continue;
 464
 465       int unit_id = prog->SamplerUnits[s];
 466       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 467
 468       if (unit->_ReallyEnabled && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 469          const struct gl_texture_object *t = unit->_Current;
 470          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 471          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 472          int swizzles[SWIZZLE_NIL + 1] = {
 473             SWIZZLE_X,
 474             SWIZZLE_Y,
 475             SWIZZLE_Z,
 476             SWIZZLE_W,
 477             SWIZZLE_ZERO,
 478             SWIZZLE_ONE,
 479             SWIZZLE_NIL
 480          };
 481
 482          if (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 483              img->_BaseFormat == GL_DEPTH_STENCIL) {
 484             /* We handle GL_DEPTH_TEXTURE_MODE here instead of as surface
 485              * format overrides because shadow comparison always returns the
 486              * result of the comparison in all channels anyway.
 487              */
 488             switch (t->DepthMode) {
 489             case GL_ALPHA:
 490                swizzles[0] = SWIZZLE_ZERO;
 491                swizzles[1] = SWIZZLE_ZERO;
 492                swizzles[2] = SWIZZLE_ZERO;
 493                swizzles[3] = SWIZZLE_X;
 494                break;
 495             case GL_LUMINANCE:
 496                swizzles[0] = SWIZZLE_X;
 497                swizzles[1] = SWIZZLE_X;
 498                swizzles[2] = SWIZZLE_X;
 499                swizzles[3] = SWIZZLE_ONE;
 500                break;
 501             case GL_INTENSITY:
 502                swizzles[0] = SWIZZLE_X;
 503                swizzles[1] = SWIZZLE_X;
 504                swizzles[2] = SWIZZLE_X;
 505                swizzles[3] = SWIZZLE_X;
 506                break;
 507             case GL_RED:
 508                swizzles[0] = SWIZZLE_X;
 509                swizzles[1] = SWIZZLE_ZERO;
 510                swizzles[2] = SWIZZLE_ZERO;
 511                swizzles[3] = SWIZZLE_ONE;
 512                break;
 513             }
 514          }
 515
 516          if (img->InternalFormat == GL_YCBCR_MESA) {
 517             key->yuvtex_mask |= 1 << s;
 518             if (img->TexFormat == MESA_FORMAT_YCBCR)
 519                 key->yuvtex_swap_mask |= 1 << s;
 520          }
 521
 522          key->swizzles[s] =
 523             MAKE_SWIZZLE4(swizzles[GET_SWZ(t->_Swizzle, 0)],
 524                           swizzles[GET_SWZ(t->_Swizzle, 1)],
 525                           swizzles[GET_SWZ(t->_Swizzle, 2)],
 526                           swizzles[GET_SWZ(t->_Swizzle, 3)]);
 527
 528          if (sampler->MinFilter != GL_NEAREST &&
 529              sampler->MagFilter != GL_NEAREST) {
 530             if (sampler->WrapS == GL_CLAMP)
 531                key->gl_clamp_mask[0] |= 1 << s;
 532             if (sampler->WrapT == GL_CLAMP)
 533                key->gl_clamp_mask[1] |= 1 << s;
 534             if (sampler->WrapR == GL_CLAMP)
 535                key->gl_clamp_mask[2] |= 1 << s;
 536          }
 537       }
 538    }
 539 }
 540
 541 static void brw_wm_populate_key( struct brw_context *brw,
 542                                  struct brw_wm_prog_key *key )
 543 {
 544    struct gl_context *ctx = &brw->intel.ctx;
 545    struct intel_context *intel = &brw->intel;
 546    /* BRW_NEW_FRAGMENT_PROGRAM */
 547    const struct brw_fragment_program *fp =
 548       (struct brw_fragment_program *)brw->fragment_program;
 549    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 550    GLuint lookup = 0;
 551    GLuint line_aa;
 552    bool program_uses_dfdy = fp->program.UsesDFdy;
 553
 554    memset(key, 0, sizeof(*key));
 555
 556    /* Build the index for table lookup
 557     */
 558    if (intel->gen < 6) {
 559       /* _NEW_COLOR */
 560       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 561          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 562
 563       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 564          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 565
 566       /* _NEW_DEPTH */
 567       if (ctx->Depth.Test)
 568          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 569
 570       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 571          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 572
 573       /* _NEW_STENCIL */
 574       if (ctx->Stencil._Enabled) {
 575          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 576
 577          if (ctx->Stencil.WriteMask[0] ||
 578              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 579             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 580       }
 581       key->iz_lookup = lookup;
 582    }
 583
 584    line_aa = AA_NEVER;
 585
 586    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 587    if (ctx->Line.SmoothFlag) {
 588       if (brw->intel.reduced_primitive == GL_LINES) {
 589          line_aa = AA_ALWAYS;
 590       }
 591       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 592          if (ctx->Polygon.FrontMode == GL_LINE) {
 593             line_aa = AA_SOMETIMES;
 594
 595             if (ctx->Polygon.BackMode == GL_LINE ||
 596                 (ctx->Polygon.CullFlag &&
 597                  ctx->Polygon.CullFaceMode == GL_BACK))
 598                line_aa = AA_ALWAYS;
 599          }
 600          else if (ctx->Polygon.BackMode == GL_LINE) {
 601             line_aa = AA_SOMETIMES;
 602
 603             if ((ctx->Polygon.CullFlag &&
 604                  ctx->Polygon.CullFaceMode == GL_FRONT))
 605                line_aa = AA_ALWAYS;
 606          }
 607       }
 608    }
 609
 610    key->line_aa = line_aa;
 611
 612    if (intel->gen < 6)
 613       key->stats_wm = brw->intel.stats_wm;
 614
 615    /* BRW_NEW_WM_INPUT_DIMENSIONS */
 616    /* Only set this for fixed function.  The optimization it enables isn't
 617     * useful for programs using shaders.
 618     */
 619    if (ctx->Shader.CurrentFragmentProgram)
 620       key->proj_attrib_mask = 0xffffffff;
 621    else
 622       key->proj_attrib_mask = brw->wm.input_size_masks[4-1];
 623
 624    /* _NEW_LIGHT */
 625    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 626
 627    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 628    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 629
 630    /* _NEW_TEXTURE */
 631    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 632
 633    /* _NEW_BUFFERS */
 634    /*
 635     * Include the draw buffer origin and height so that we can calculate
 636     * fragment position values relative to the bottom left of the drawable,
 637     * from the incoming screen origin relative position we get as part of our
 638     * payload.
 639     *
 640     * This is only needed for the WM_WPOSXY opcode when the fragment program
 641     * uses the gl_FragCoord input.
 642     *
 643     * We could avoid recompiling by including this as a constant referenced by
 644     * our program, but if we were to do that it would also be nice to handle
 645     * getting that constant updated at batchbuffer submit time (when we
 646     * hold the lock and know where the buffer really is) rather than at emit
 647     * time when we don't hold the lock and are just guessing.  We could also
 648     * just avoid using this as key data if the program doesn't use
 649     * fragment.position.
 650     *
 651     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 652     * drawable height in order to invert the Y axis.
 653     */
 654    if (fp->program.Base.InputsRead & FRAG_BIT_WPOS) {
 655       key->drawable_height = ctx->DrawBuffer->Height;
 656    }
 657
 658    if ((fp->program.Base.InputsRead & FRAG_BIT_WPOS) || program_uses_dfdy) {
 659       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 660    }
 661
 662    /* _NEW_BUFFERS */
 663    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 664   /* _NEW_MULTISAMPLE */
 665    key->sample_alpha_to_coverage = ctx->Multisample.SampleAlphaToCoverage;
 666
 667    /* CACHE_NEW_VS_PROG */
 668    if (intel->gen < 6)
 669       key->vp_outputs_written = brw->vs.prog_data->outputs_written;
 670
 671    /* The unique fragment program ID */
 672    key->program_string_id = fp->id;
 673 }
 674
 675
 676 static void
 677 brw_upload_wm_prog(struct brw_context *brw)
 678 {
 679    struct intel_context *intel = &brw->intel;
 680    struct gl_context *ctx = &intel->ctx;
 681    struct brw_wm_prog_key key;
 682    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 683       brw->fragment_program;
 684
 685    brw_wm_populate_key(brw, &key);
 686
 687    if (!brw_search_cache(&brw->cache, BRW_WM_PROG,
 688                          &key, sizeof(key),
 689                          &brw->wm.prog_offset, &brw->wm.prog_data)) {
 690       bool success = do_wm_prog(brw, ctx->Shader._CurrentFragmentProgram, fp,
 691                                 &key);
 692       (void) success;
 693       assert(success);
 694    }
 695 }
 696
 697
 698 const struct brw_tracked_state brw_wm_prog = {
 699    .dirty = {
 700       .mesa  = (_NEW_COLOR |
 701                 _NEW_DEPTH |
 702                 _NEW_STENCIL |
 703                 _NEW_POLYGON |
 704                 _NEW_LINE |
 705                 _NEW_LIGHT |
 706                 _NEW_FRAG_CLAMP |
 707                 _NEW_BUFFERS |
 708                 _NEW_TEXTURE |
 709                 _NEW_MULTISAMPLE),
 710       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 711                 BRW_NEW_WM_INPUT_DIMENSIONS |
 712                 BRW_NEW_REDUCED_PRIMITIVE),
 713       .cache = CACHE_NEW_VS_PROG,
 714    },
 715    .emit = brw_upload_wm_prog
 716 };
 717