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