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

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics 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 <keithw@vmware.com>
  30   */
  31
  32 #include "brw_context.h"
  33 #include "brw_wm.h"
  34 #include "brw_state.h"
  35 #include "main/enums.h"
  36 #include "main/formats.h"
  37 #include "main/fbobject.h"
  38 #include "main/samplerobj.h"
  39 #include "program/prog_parameter.h"
  40 #include "program/program.h"
  41 #include "intel_mipmap_tree.h"
  42
  43 #include "util/ralloc.h"
  44
  45 /**
  46  * Return a bitfield where bit n is set if barycentric interpolation mode n
  47  * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
  48  */
  49 static unsigned
  50 brw_compute_barycentric_interp_modes(struct brw_context *brw,
  51                                      bool shade_model_flat,
  52                                      bool persample_shading,
  53                                      const struct gl_fragment_program *fprog)
  54 {
  55    unsigned barycentric_interp_modes = 0;
  56    int attr;
  57
  58    /* Loop through all fragment shader inputs to figure out what interpolation
  59     * modes are in use, and set the appropriate bits in
  60     * barycentric_interp_modes.
  61     */
  62    for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
  63       enum glsl_interp_qualifier interp_qualifier =
  64          fprog->InterpQualifier[attr];
  65       bool is_centroid = (fprog->IsCentroid & BITFIELD64_BIT(attr)) &&
  66          !persample_shading;
  67       bool is_sample = (fprog->IsSample & BITFIELD64_BIT(attr)) ||
  68          persample_shading;
  69       bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
  70
  71       /* Ignore unused inputs. */
  72       if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
  73          continue;
  74
  75       /* Ignore WPOS and FACE, because they don't require interpolation. */
  76       if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
  77          continue;
  78
  79       /* Determine the set (or sets) of barycentric coordinates needed to
  80        * interpolate this variable.  Note that when
  81        * brw->needs_unlit_centroid_workaround is set, centroid interpolation
  82        * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
  83        * for lit pixels, so we need both sets of barycentric coordinates.
  84        */
  85       if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
  86          if (is_centroid) {
  87             barycentric_interp_modes |=
  88                1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
  89          } else if (is_sample) {
  90             barycentric_interp_modes |=
  91                1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC;
  92          }
  93          if ((!is_centroid && !is_sample) ||
  94              brw->needs_unlit_centroid_workaround) {
  95             barycentric_interp_modes |=
  96                1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
  97          }
  98       } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
  99                  (!(shade_model_flat && is_gl_Color) &&
 100                   interp_qualifier == INTERP_QUALIFIER_NONE)) {
 101          if (is_centroid) {
 102             barycentric_interp_modes |=
 103                1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
 104          } else if (is_sample) {
 105             barycentric_interp_modes |=
 106                1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC;
 107          }
 108          if ((!is_centroid && !is_sample) ||
 109              brw->needs_unlit_centroid_workaround) {
 110             barycentric_interp_modes |=
 111                1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
 112          }
 113       }
 114    }
 115
 116    return barycentric_interp_modes;
 117 }
 118
 119 static uint8_t
 120 computed_depth_mode(struct gl_fragment_program *fp)
 121 {
 122    if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 123       switch (fp->FragDepthLayout) {
 124       case FRAG_DEPTH_LAYOUT_NONE:
 125       case FRAG_DEPTH_LAYOUT_ANY:
 126          return BRW_PSCDEPTH_ON;
 127       case FRAG_DEPTH_LAYOUT_GREATER:
 128          return BRW_PSCDEPTH_ON_GE;
 129       case FRAG_DEPTH_LAYOUT_LESS:
 130          return BRW_PSCDEPTH_ON_LE;
 131       case FRAG_DEPTH_LAYOUT_UNCHANGED:
 132          return BRW_PSCDEPTH_OFF;
 133       }
 134    }
 135    return BRW_PSCDEPTH_OFF;
 136 }
 137
 138 bool
 139 brw_wm_prog_data_compare(const void *in_a, const void *in_b)
 140 {
 141    const struct brw_wm_prog_data *a = in_a;
 142    const struct brw_wm_prog_data *b = in_b;
 143
 144    /* Compare the base structure. */
 145    if (!brw_stage_prog_data_compare(&a->base, &b->base))
 146       return false;
 147
 148    /* Compare the rest of the structure. */
 149    const unsigned offset = sizeof(struct brw_stage_prog_data);
 150    if (memcmp(((char *) a) + offset, ((char *) b) + offset,
 151               sizeof(struct brw_wm_prog_data) - offset))
 152       return false;
 153
 154    return true;
 155 }
 156
 157 /**
 158  * All Mesa program -> GPU code generation goes through this function.
 159  * Depending on the instructions used (i.e. flow control instructions)
 160  * we'll use one of two code generators.
 161  */
 162 bool do_wm_prog(struct brw_context *brw,
 163                 struct gl_shader_program *prog,
 164                 struct brw_fragment_program *fp,
 165                 struct brw_wm_prog_key *key)
 166 {
 167    struct gl_context *ctx = &brw->ctx;
 168    void *mem_ctx = ralloc_context(NULL);
 169    struct brw_wm_prog_data prog_data;
 170    const GLuint *program;
 171    struct gl_shader *fs = NULL;
 172    GLuint program_size;
 173
 174    if (prog)
 175       fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 176
 177    memset(&prog_data, 0, sizeof(prog_data));
 178    /* key->alpha_test_func means simulating alpha testing via discards,
 179     * so the shader definitely kills pixels.
 180     */
 181    prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
 182
 183    prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
 184
 185    /* Allocate the references to the uniforms that will end up in the
 186     * prog_data associated with the compiled program, and which will be freed
 187     * by the state cache.
 188     */
 189    int param_count;
 190    if (fs) {
 191       param_count = fs->num_uniform_components;
 192    } else {
 193       param_count = fp->program.Base.Parameters->NumParameters * 4;
 194    }
 195    /* The backend also sometimes adds params for texture size. */
 196    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 197    prog_data.base.param =
 198       rzalloc_array(NULL, const gl_constant_value *, param_count);
 199    prog_data.base.pull_param =
 200       rzalloc_array(NULL, const gl_constant_value *, param_count);
 201    prog_data.base.nr_params = param_count;
 202
 203    prog_data.barycentric_interp_modes =
 204       brw_compute_barycentric_interp_modes(brw, key->flat_shade,
 205                                            key->persample_shading,
 206                                            &fp->program);
 207
 208    program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
 209                             &fp->program, prog, &program_size);
 210    if (program == NULL) {
 211       ralloc_free(mem_ctx);
 212       return false;
 213    }
 214
 215    if (prog_data.base.total_scratch) {
 216       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 217                          prog_data.base.total_scratch * brw->max_wm_threads);
 218    }
 219
 220    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 221       fprintf(stderr, "\n");
 222
 223    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 224                     key, sizeof(struct brw_wm_prog_key),
 225                     program, program_size,
 226                     &prog_data, sizeof(prog_data),
 227                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 228
 229    ralloc_free(mem_ctx);
 230
 231    return true;
 232 }
 233
 234 static bool
 235 key_debug(struct brw_context *brw, const char *name, int a, int b)
 236 {
 237    if (a != b) {
 238       perf_debug("  %s %d->%d\n", name, a, b);
 239       return true;
 240    } else {
 241       return false;
 242    }
 243 }
 244
 245 bool
 246 brw_debug_recompile_sampler_key(struct brw_context *brw,
 247                                 const struct brw_sampler_prog_key_data *old_key,
 248                                 const struct brw_sampler_prog_key_data *key)
 249 {
 250    bool found = false;
 251
 252    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 253       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 254                          old_key->swizzles[i], key->swizzles[i]);
 255    }
 256    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 257                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 258    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 259                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 260    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 261                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 262    found |= key_debug(brw, "gather channel quirk on any texture unit",
 263                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 264
 265    return found;
 266 }
 267
 268 void
 269 brw_wm_debug_recompile(struct brw_context *brw,
 270                        struct gl_shader_program *prog,
 271                        const struct brw_wm_prog_key *key)
 272 {
 273    struct brw_cache_item *c = NULL;
 274    const struct brw_wm_prog_key *old_key = NULL;
 275    bool found = false;
 276
 277    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 278
 279    for (unsigned int i = 0; i < brw->cache.size; i++) {
 280       for (c = brw->cache.items[i]; c; c = c->next) {
 281          if (c->cache_id == BRW_CACHE_FS_PROG) {
 282             old_key = c->key;
 283
 284             if (old_key->program_string_id == key->program_string_id)
 285                break;
 286          }
 287       }
 288       if (c)
 289          break;
 290    }
 291
 292    if (!c) {
 293       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 294       return;
 295    }
 296
 297    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 298                       "depth write",
 299                       old_key->iz_lookup, key->iz_lookup);
 300    found |= key_debug(brw, "depth statistics",
 301                       old_key->stats_wm, key->stats_wm);
 302    found |= key_debug(brw, "flat shading",
 303                       old_key->flat_shade, key->flat_shade);
 304    found |= key_debug(brw, "per-sample shading",
 305                       old_key->persample_shading, key->persample_shading);
 306    found |= key_debug(brw, "per-sample shading and 2x MSAA",
 307                       old_key->persample_2x, key->persample_2x);
 308    found |= key_debug(brw, "number of color buffers",
 309                       old_key->nr_color_regions, key->nr_color_regions);
 310    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 311                       old_key->replicate_alpha, key->replicate_alpha);
 312    found |= key_debug(brw, "rendering to FBO",
 313                       old_key->render_to_fbo, key->render_to_fbo);
 314    found |= key_debug(brw, "fragment color clamping",
 315                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 316    found |= key_debug(brw, "line smoothing",
 317                       old_key->line_aa, key->line_aa);
 318    found |= key_debug(brw, "renderbuffer height",
 319                       old_key->drawable_height, key->drawable_height);
 320    found |= key_debug(brw, "input slots valid",
 321                       old_key->input_slots_valid, key->input_slots_valid);
 322    found |= key_debug(brw, "mrt alpha test function",
 323                       old_key->alpha_test_func, key->alpha_test_func);
 324    found |= key_debug(brw, "mrt alpha test reference value",
 325                       old_key->alpha_test_ref, key->alpha_test_ref);
 326
 327    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 328
 329    if (!found) {
 330       perf_debug("  Something else\n");
 331    }
 332 }
 333
 334 static uint8_t
 335 gen6_gather_workaround(GLenum internalformat)
 336 {
 337    switch (internalformat) {
 338       case GL_R8I: return WA_SIGN | WA_8BIT;
 339       case GL_R8UI: return WA_8BIT;
 340       case GL_R16I: return WA_SIGN | WA_16BIT;
 341       case GL_R16UI: return WA_16BIT;
 342       /* note that even though GL_R32I and GL_R32UI have format overrides
 343        * in the surface state, there is no shader w/a required */
 344       default: return 0;
 345    }
 346 }
 347
 348 void
 349 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 350                                    const struct gl_program *prog,
 351                                    unsigned sampler_count,
 352                                    struct brw_sampler_prog_key_data *key)
 353 {
 354    struct brw_context *brw = brw_context(ctx);
 355
 356    for (int s = 0; s < sampler_count; s++) {
 357       key->swizzles[s] = SWIZZLE_NOOP;
 358
 359       if (!(prog->SamplersUsed & (1 << s)))
 360          continue;
 361
 362       int unit_id = prog->SamplerUnits[s];
 363       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 364
 365       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 366          const struct gl_texture_object *t = unit->_Current;
 367          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 368          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 369
 370          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 371             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 372              img->_BaseFormat == GL_DEPTH_STENCIL);
 373
 374          /* Haswell handles texture swizzling as surface format overrides
 375           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 376           */
 377          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 378             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 379
 380          if (brw->gen < 8 &&
 381              sampler->MinFilter != GL_NEAREST &&
 382              sampler->MagFilter != GL_NEAREST) {
 383             if (sampler->WrapS == GL_CLAMP)
 384                key->gl_clamp_mask[0] |= 1 << s;
 385             if (sampler->WrapT == GL_CLAMP)
 386                key->gl_clamp_mask[1] |= 1 << s;
 387             if (sampler->WrapR == GL_CLAMP)
 388                key->gl_clamp_mask[2] |= 1 << s;
 389          }
 390
 391          /* gather4's channel select for green from RG32F is broken;
 392           * requires a shader w/a on IVB; fixable with just SCS on HSW. */
 393          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 394             if (img->InternalFormat == GL_RG32F)
 395                key->gather_channel_quirk_mask |= 1 << s;
 396          }
 397
 398          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 399           * UNORM/FLOAT instead and fix it in the shader.
 400           */
 401          if (brw->gen == 6 && prog->UsesGather) {
 402             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 403          }
 404
 405          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 406           * then we need to emit slightly different code to first sample the
 407           * MCS surface.
 408           */
 409          struct intel_texture_object *intel_tex =
 410             intel_texture_object((struct gl_texture_object *)t);
 411
 412          if (brw->gen >= 7 &&
 413              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
 414             key->compressed_multisample_layout_mask |= 1 << s;
 415          }
 416       }
 417    }
 418 }
 419
 420 static void brw_wm_populate_key( struct brw_context *brw,
 421                                  struct brw_wm_prog_key *key )
 422 {
 423    struct gl_context *ctx = &brw->ctx;
 424    /* BRW_NEW_FRAGMENT_PROGRAM */
 425    const struct brw_fragment_program *fp =
 426       (struct brw_fragment_program *)brw->fragment_program;
 427    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 428    GLuint lookup = 0;
 429    GLuint line_aa;
 430    bool program_uses_dfdy = fp->program.UsesDFdy;
 431    bool multisample_fbo = ctx->DrawBuffer->Visual.samples > 1;
 432
 433    memset(key, 0, sizeof(*key));
 434
 435    /* Build the index for table lookup
 436     */
 437    if (brw->gen < 6) {
 438       /* _NEW_COLOR */
 439       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 440          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 441
 442       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 443          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 444
 445       /* _NEW_DEPTH */
 446       if (ctx->Depth.Test)
 447          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 448
 449       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 450          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 451
 452       /* _NEW_STENCIL | _NEW_BUFFERS */
 453       if (ctx->Stencil._Enabled) {
 454          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 455
 456          if (ctx->Stencil.WriteMask[0] ||
 457              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 458             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 459       }
 460       key->iz_lookup = lookup;
 461    }
 462
 463    line_aa = AA_NEVER;
 464
 465    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 466    if (ctx->Line.SmoothFlag) {
 467       if (brw->reduced_primitive == GL_LINES) {
 468          line_aa = AA_ALWAYS;
 469       }
 470       else if (brw->reduced_primitive == GL_TRIANGLES) {
 471          if (ctx->Polygon.FrontMode == GL_LINE) {
 472             line_aa = AA_SOMETIMES;
 473
 474             if (ctx->Polygon.BackMode == GL_LINE ||
 475                 (ctx->Polygon.CullFlag &&
 476                  ctx->Polygon.CullFaceMode == GL_BACK))
 477                line_aa = AA_ALWAYS;
 478          }
 479          else if (ctx->Polygon.BackMode == GL_LINE) {
 480             line_aa = AA_SOMETIMES;
 481
 482             if ((ctx->Polygon.CullFlag &&
 483                  ctx->Polygon.CullFaceMode == GL_FRONT))
 484                line_aa = AA_ALWAYS;
 485          }
 486       }
 487    }
 488
 489    key->line_aa = line_aa;
 490
 491    /* _NEW_HINT */
 492    key->high_quality_derivatives =
 493       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 494
 495    if (brw->gen < 6)
 496       key->stats_wm = brw->stats_wm;
 497
 498    /* _NEW_LIGHT */
 499    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 500
 501    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 502    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 503
 504    /* _NEW_TEXTURE */
 505    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 506                                       &key->tex);
 507
 508    /* _NEW_BUFFERS */
 509    /*
 510     * Include the draw buffer origin and height so that we can calculate
 511     * fragment position values relative to the bottom left of the drawable,
 512     * from the incoming screen origin relative position we get as part of our
 513     * payload.
 514     *
 515     * This is only needed for the WM_WPOSXY opcode when the fragment program
 516     * uses the gl_FragCoord input.
 517     *
 518     * We could avoid recompiling by including this as a constant referenced by
 519     * our program, but if we were to do that it would also be nice to handle
 520     * getting that constant updated at batchbuffer submit time (when we
 521     * hold the lock and know where the buffer really is) rather than at emit
 522     * time when we don't hold the lock and are just guessing.  We could also
 523     * just avoid using this as key data if the program doesn't use
 524     * fragment.position.
 525     *
 526     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 527     * drawable height in order to invert the Y axis.
 528     */
 529    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 530       key->drawable_height = ctx->DrawBuffer->Height;
 531    }
 532
 533    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 534       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 535    }
 536
 537    /* _NEW_BUFFERS */
 538    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 539
 540    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 541    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 542       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 543
 544    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 545    /* Ignore sample qualifier while computing this flag. */
 546    key->persample_shading =
 547       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
 548    if (key->persample_shading)
 549       key->persample_2x = ctx->DrawBuffer->Visual.samples == 2;
 550
 551    key->compute_pos_offset =
 552       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
 553       fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
 554
 555    key->compute_sample_id =
 556       multisample_fbo &&
 557       ctx->Multisample.Enabled &&
 558       (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
 559
 560    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 561    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 562                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 563       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 564
 565
 566    /* _NEW_COLOR | _NEW_BUFFERS */
 567    /* Pre-gen6, the hardware alpha test always used each render
 568     * target's alpha to do alpha test, as opposed to render target 0's alpha
 569     * like GL requires.  Fix that by building the alpha test into the
 570     * shader, and we'll skip enabling the fixed function alpha test.
 571     */
 572    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 && ctx->Color.AlphaEnabled) {
 573       key->alpha_test_func = ctx->Color.AlphaFunc;
 574       key->alpha_test_ref = ctx->Color.AlphaRef;
 575    }
 576
 577    /* The unique fragment program ID */
 578    key->program_string_id = fp->id;
 579 }
 580
 581
 582 static void
 583 brw_upload_wm_prog(struct brw_context *brw)
 584 {
 585    struct gl_context *ctx = &brw->ctx;
 586    struct brw_wm_prog_key key;
 587    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 588       brw->fragment_program;
 589
 590    brw_wm_populate_key(brw, &key);
 591
 592    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 593                          &key, sizeof(key),
 594                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 595       bool success = do_wm_prog(brw, ctx->_Shader->_CurrentFragmentProgram, fp,
 596                                 &key);
 597       (void) success;
 598       assert(success);
 599    }
 600    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 601 }
 602
 603
 604 const struct brw_tracked_state brw_wm_prog = {
 605    .dirty = {
 606       .mesa  = _NEW_BUFFERS |
 607                _NEW_COLOR |
 608                _NEW_DEPTH |
 609                _NEW_FRAG_CLAMP |
 610                _NEW_HINT |
 611                _NEW_LIGHT |
 612                _NEW_LINE |
 613                _NEW_MULTISAMPLE |
 614                _NEW_POLYGON |
 615                _NEW_STENCIL |
 616                _NEW_TEXTURE,
 617       .brw   = BRW_NEW_FRAGMENT_PROGRAM |
 618                BRW_NEW_REDUCED_PRIMITIVE |
 619                BRW_NEW_STATS_WM |
 620                BRW_NEW_VUE_MAP_GEOM_OUT,
 621    },
 622    .emit = brw_upload_wm_prog
 623 };
 624