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 #include "brw_context.h"
  27 #include "brw_wm.h"
  28 #include "brw_state.h"
  29 #include "main/enums.h"
  30 #include "main/formats.h"
  31 #include "main/fbobject.h"
  32 #include "main/samplerobj.h"
  33 #include "main/framebuffer.h"
  34 #include "program/prog_parameter.h"
  35 #include "program/program.h"
  36 #include "intel_mipmap_tree.h"
  37
  38 #include "util/ralloc.h"
  39
  40 /**
  41  * Return a bitfield where bit n is set if barycentric interpolation mode n
  42  * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
  43  */
  44 unsigned
  45 brw_compute_barycentric_interp_modes(struct brw_context *brw,
  46                                      bool shade_model_flat,
  47                                      bool persample_shading,
  48                                      const struct gl_fragment_program *fprog)
  49 {
  50    unsigned barycentric_interp_modes = 0;
  51    int attr;
  52
  53    /* Loop through all fragment shader inputs to figure out what interpolation
  54     * modes are in use, and set the appropriate bits in
  55     * barycentric_interp_modes.
  56     */
  57    for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
  58       enum glsl_interp_qualifier interp_qualifier =
  59          fprog->InterpQualifier[attr];
  60       bool is_centroid = (fprog->IsCentroid & BITFIELD64_BIT(attr)) &&
  61          !persample_shading;
  62       bool is_sample = (fprog->IsSample & BITFIELD64_BIT(attr)) ||
  63          persample_shading;
  64       bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
  65
  66       /* Ignore unused inputs. */
  67       if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
  68          continue;
  69
  70       /* Ignore WPOS and FACE, because they don't require interpolation. */
  71       if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
  72          continue;
  73
  74       /* Determine the set (or sets) of barycentric coordinates needed to
  75        * interpolate this variable.  Note that when
  76        * brw->needs_unlit_centroid_workaround is set, centroid interpolation
  77        * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
  78        * for lit pixels, so we need both sets of barycentric coordinates.
  79        */
  80       if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
  81          if (is_centroid) {
  82             barycentric_interp_modes |=
  83                1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
  84          } else if (is_sample) {
  85             barycentric_interp_modes |=
  86                1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC;
  87          }
  88          if ((!is_centroid && !is_sample) ||
  89              brw->needs_unlit_centroid_workaround) {
  90             barycentric_interp_modes |=
  91                1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
  92          }
  93       } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
  94                  (!(shade_model_flat && is_gl_Color) &&
  95                   interp_qualifier == INTERP_QUALIFIER_NONE)) {
  96          if (is_centroid) {
  97             barycentric_interp_modes |=
  98                1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
  99          } else if (is_sample) {
 100             barycentric_interp_modes |=
 101                1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC;
 102          }
 103          if ((!is_centroid && !is_sample) ||
 104              brw->needs_unlit_centroid_workaround) {
 105             barycentric_interp_modes |=
 106                1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
 107          }
 108       }
 109    }
 110
 111    return barycentric_interp_modes;
 112 }
 113
 114 static uint8_t
 115 computed_depth_mode(struct gl_fragment_program *fp)
 116 {
 117    if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 118       switch (fp->FragDepthLayout) {
 119       case FRAG_DEPTH_LAYOUT_NONE:
 120       case FRAG_DEPTH_LAYOUT_ANY:
 121          return BRW_PSCDEPTH_ON;
 122       case FRAG_DEPTH_LAYOUT_GREATER:
 123          return BRW_PSCDEPTH_ON_GE;
 124       case FRAG_DEPTH_LAYOUT_LESS:
 125          return BRW_PSCDEPTH_ON_LE;
 126       case FRAG_DEPTH_LAYOUT_UNCHANGED:
 127          return BRW_PSCDEPTH_OFF;
 128       }
 129    }
 130    return BRW_PSCDEPTH_OFF;
 131 }
 132
 133 bool
 134 brw_wm_prog_data_compare(const void *in_a, const void *in_b)
 135 {
 136    const struct brw_wm_prog_data *a = in_a;
 137    const struct brw_wm_prog_data *b = in_b;
 138
 139    /* Compare the base structure. */
 140    if (!brw_stage_prog_data_compare(&a->base, &b->base))
 141       return false;
 142
 143    /* Compare the rest of the structure. */
 144    const unsigned offset = sizeof(struct brw_stage_prog_data);
 145    if (memcmp(((char *) a) + offset, ((char *) b) + offset,
 146               sizeof(struct brw_wm_prog_data) - offset))
 147       return false;
 148
 149    return true;
 150 }
 151
 152 /**
 153  * All Mesa program -> GPU code generation goes through this function.
 154  * Depending on the instructions used (i.e. flow control instructions)
 155  * we'll use one of two code generators.
 156  */
 157 bool
 158 brw_codegen_wm_prog(struct brw_context *brw,
 159                     struct gl_shader_program *prog,
 160                     struct brw_fragment_program *fp,
 161                     struct brw_wm_prog_key *key)
 162 {
 163    struct gl_context *ctx = &brw->ctx;
 164    void *mem_ctx = ralloc_context(NULL);
 165    struct brw_wm_prog_data prog_data;
 166    const GLuint *program;
 167    struct gl_shader *fs = NULL;
 168    GLuint program_size;
 169
 170    if (prog)
 171       fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 172
 173    memset(&prog_data, 0, sizeof(prog_data));
 174    /* key->alpha_test_func means simulating alpha testing via discards,
 175     * so the shader definitely kills pixels.
 176     */
 177    prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
 178    prog_data.uses_omask =
 179       fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK);
 180    prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
 181
 182    prog_data.early_fragment_tests = fs && fs->EarlyFragmentTests;
 183
 184    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 185    if (!prog)
 186       prog_data.base.use_alt_mode = true;
 187
 188    /* Allocate the references to the uniforms that will end up in the
 189     * prog_data associated with the compiled program, and which will be freed
 190     * by the state cache.
 191     */
 192    int param_count;
 193    if (fs) {
 194       param_count = fs->num_uniform_components +
 195                     fs->NumImages * BRW_IMAGE_PARAM_SIZE;
 196       prog_data.base.nr_image_params = fs->NumImages;
 197    } else {
 198       param_count = fp->program.Base.Parameters->NumParameters * 4;
 199    }
 200    /* The backend also sometimes adds params for texture size. */
 201    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 202    prog_data.base.param =
 203       rzalloc_array(NULL, const gl_constant_value *, param_count);
 204    prog_data.base.pull_param =
 205       rzalloc_array(NULL, const gl_constant_value *, param_count);
 206    prog_data.base.image_param =
 207       rzalloc_array(NULL, struct brw_image_param,
 208                     prog_data.base.nr_image_params);
 209    prog_data.base.nr_params = param_count;
 210
 211    prog_data.barycentric_interp_modes =
 212       brw_compute_barycentric_interp_modes(brw, key->flat_shade,
 213                                            key->persample_shading,
 214                                            &fp->program);
 215
 216    program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
 217                             &fp->program, prog, &program_size);
 218    if (program == NULL) {
 219       ralloc_free(mem_ctx);
 220       return false;
 221    }
 222
 223    if (prog_data.base.total_scratch) {
 224       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 225                          prog_data.base.total_scratch * brw->max_wm_threads);
 226    }
 227
 228    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 229       fprintf(stderr, "\n");
 230
 231    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 232                     key, sizeof(struct brw_wm_prog_key),
 233                     program, program_size,
 234                     &prog_data, sizeof(prog_data),
 235                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 236
 237    ralloc_free(mem_ctx);
 238
 239    return true;
 240 }
 241
 242 static bool
 243 key_debug(struct brw_context *brw, const char *name, int a, int b)
 244 {
 245    if (a != b) {
 246       perf_debug("  %s %d->%d\n", name, a, b);
 247       return true;
 248    } else {
 249       return false;
 250    }
 251 }
 252
 253 bool
 254 brw_debug_recompile_sampler_key(struct brw_context *brw,
 255                                 const struct brw_sampler_prog_key_data *old_key,
 256                                 const struct brw_sampler_prog_key_data *key)
 257 {
 258    bool found = false;
 259
 260    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 261       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 262                          old_key->swizzles[i], key->swizzles[i]);
 263    }
 264    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 265                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 266    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 267                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 268    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 269                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 270    found |= key_debug(brw, "gather channel quirk on any texture unit",
 271                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 272    found |= key_debug(brw, "compressed multisample layout",
 273                       old_key->compressed_multisample_layout_mask,
 274                       key->compressed_multisample_layout_mask);
 275
 276    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 277       found |= key_debug(brw, "textureGather workarounds",
 278                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 279    }
 280
 281    return found;
 282 }
 283
 284 void
 285 brw_wm_debug_recompile(struct brw_context *brw,
 286                        struct gl_shader_program *prog,
 287                        const struct brw_wm_prog_key *key)
 288 {
 289    struct brw_cache_item *c = NULL;
 290    const struct brw_wm_prog_key *old_key = NULL;
 291    bool found = false;
 292
 293    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 294
 295    for (unsigned int i = 0; i < brw->cache.size; i++) {
 296       for (c = brw->cache.items[i]; c; c = c->next) {
 297          if (c->cache_id == BRW_CACHE_FS_PROG) {
 298             old_key = c->key;
 299
 300             if (old_key->program_string_id == key->program_string_id)
 301                break;
 302          }
 303       }
 304       if (c)
 305          break;
 306    }
 307
 308    if (!c) {
 309       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 310       return;
 311    }
 312
 313    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 314                       "depth write",
 315                       old_key->iz_lookup, key->iz_lookup);
 316    found |= key_debug(brw, "depth statistics",
 317                       old_key->stats_wm, key->stats_wm);
 318    found |= key_debug(brw, "flat shading",
 319                       old_key->flat_shade, key->flat_shade);
 320    found |= key_debug(brw, "per-sample shading",
 321                       old_key->persample_shading, key->persample_shading);
 322    found |= key_debug(brw, "per-sample shading and 2x MSAA",
 323                       old_key->persample_2x, key->persample_2x);
 324    found |= key_debug(brw, "number of color buffers",
 325                       old_key->nr_color_regions, key->nr_color_regions);
 326    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 327                       old_key->replicate_alpha, key->replicate_alpha);
 328    found |= key_debug(brw, "rendering to FBO",
 329                       old_key->render_to_fbo, key->render_to_fbo);
 330    found |= key_debug(brw, "fragment color clamping",
 331                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 332    found |= key_debug(brw, "line smoothing",
 333                       old_key->line_aa, key->line_aa);
 334    found |= key_debug(brw, "renderbuffer height",
 335                       old_key->drawable_height, key->drawable_height);
 336    found |= key_debug(brw, "input slots valid",
 337                       old_key->input_slots_valid, key->input_slots_valid);
 338    found |= key_debug(brw, "mrt alpha test function",
 339                       old_key->alpha_test_func, key->alpha_test_func);
 340    found |= key_debug(brw, "mrt alpha test reference value",
 341                       old_key->alpha_test_ref, key->alpha_test_ref);
 342
 343    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 344
 345    if (!found) {
 346       perf_debug("  Something else\n");
 347    }
 348 }
 349
 350 static uint8_t
 351 gen6_gather_workaround(GLenum internalformat)
 352 {
 353    switch (internalformat) {
 354    case GL_R8I: return WA_SIGN | WA_8BIT;
 355    case GL_R8UI: return WA_8BIT;
 356    case GL_R16I: return WA_SIGN | WA_16BIT;
 357    case GL_R16UI: return WA_16BIT;
 358    default:
 359       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 360        * the surface state, there is no shader w/a required.
 361        */
 362       return 0;
 363    }
 364 }
 365
 366 void
 367 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 368                                    const struct gl_program *prog,
 369                                    unsigned sampler_count,
 370                                    struct brw_sampler_prog_key_data *key)
 371 {
 372    struct brw_context *brw = brw_context(ctx);
 373
 374    for (int s = 0; s < sampler_count; s++) {
 375       key->swizzles[s] = SWIZZLE_NOOP;
 376
 377       if (!(prog->SamplersUsed & (1 << s)))
 378          continue;
 379
 380       int unit_id = prog->SamplerUnits[s];
 381       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 382
 383       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 384          const struct gl_texture_object *t = unit->_Current;
 385          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 386          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 387
 388          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 389             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 390              img->_BaseFormat == GL_DEPTH_STENCIL);
 391
 392          /* Haswell handles texture swizzling as surface format overrides
 393           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 394           */
 395          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 396             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 397
 398          if (brw->gen < 8 &&
 399              sampler->MinFilter != GL_NEAREST &&
 400              sampler->MagFilter != GL_NEAREST) {
 401             if (sampler->WrapS == GL_CLAMP)
 402                key->gl_clamp_mask[0] |= 1 << s;
 403             if (sampler->WrapT == GL_CLAMP)
 404                key->gl_clamp_mask[1] |= 1 << s;
 405             if (sampler->WrapR == GL_CLAMP)
 406                key->gl_clamp_mask[2] |= 1 << s;
 407          }
 408
 409          /* gather4's channel select for green from RG32F is broken; requires
 410           * a shader w/a on IVB; fixable with just SCS on HSW.
 411           */
 412          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 413             if (img->InternalFormat == GL_RG32F)
 414                key->gather_channel_quirk_mask |= 1 << s;
 415          }
 416
 417          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 418           * UNORM/FLOAT instead and fix it in the shader.
 419           */
 420          if (brw->gen == 6 && prog->UsesGather) {
 421             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 422          }
 423
 424          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 425           * then we need to emit slightly different code to first sample the
 426           * MCS surface.
 427           */
 428          struct intel_texture_object *intel_tex =
 429             intel_texture_object((struct gl_texture_object *)t);
 430
 431          if (brw->gen >= 7 &&
 432              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
 433             key->compressed_multisample_layout_mask |= 1 << s;
 434          }
 435       }
 436    }
 437 }
 438
 439 static bool
 440 brw_wm_state_dirty (struct brw_context *brw)
 441 {
 442    return brw_state_dirty(brw,
 443                           _NEW_BUFFERS |
 444                           _NEW_COLOR |
 445                           _NEW_DEPTH |
 446                           _NEW_FRAG_CLAMP |
 447                           _NEW_HINT |
 448                           _NEW_LIGHT |
 449                           _NEW_LINE |
 450                           _NEW_MULTISAMPLE |
 451                           _NEW_POLYGON |
 452                           _NEW_STENCIL |
 453                           _NEW_TEXTURE,
 454                           BRW_NEW_FRAGMENT_PROGRAM |
 455                           BRW_NEW_REDUCED_PRIMITIVE |
 456                           BRW_NEW_STATS_WM |
 457                           BRW_NEW_VUE_MAP_GEOM_OUT);
 458 }
 459
 460 static void
 461 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 462 {
 463    struct gl_context *ctx = &brw->ctx;
 464    /* BRW_NEW_FRAGMENT_PROGRAM */
 465    const struct brw_fragment_program *fp =
 466       (struct brw_fragment_program *) brw->fragment_program;
 467    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 468    GLuint lookup = 0;
 469    GLuint line_aa;
 470    bool program_uses_dfdy = fp->program.UsesDFdy;
 471    const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 472
 473    memset(key, 0, sizeof(*key));
 474
 475    /* Build the index for table lookup
 476     */
 477    if (brw->gen < 6) {
 478       /* _NEW_COLOR */
 479       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 480          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 481
 482       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 483          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 484
 485       /* _NEW_DEPTH */
 486       if (ctx->Depth.Test)
 487          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 488
 489       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 490          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 491
 492       /* _NEW_STENCIL | _NEW_BUFFERS */
 493       if (ctx->Stencil._Enabled) {
 494          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 495
 496          if (ctx->Stencil.WriteMask[0] ||
 497              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 498             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 499       }
 500       key->iz_lookup = lookup;
 501    }
 502
 503    line_aa = AA_NEVER;
 504
 505    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 506    if (ctx->Line.SmoothFlag) {
 507       if (brw->reduced_primitive == GL_LINES) {
 508          line_aa = AA_ALWAYS;
 509       }
 510       else if (brw->reduced_primitive == GL_TRIANGLES) {
 511          if (ctx->Polygon.FrontMode == GL_LINE) {
 512             line_aa = AA_SOMETIMES;
 513
 514             if (ctx->Polygon.BackMode == GL_LINE ||
 515                 (ctx->Polygon.CullFlag &&
 516                  ctx->Polygon.CullFaceMode == GL_BACK))
 517                line_aa = AA_ALWAYS;
 518          }
 519          else if (ctx->Polygon.BackMode == GL_LINE) {
 520             line_aa = AA_SOMETIMES;
 521
 522             if ((ctx->Polygon.CullFlag &&
 523                  ctx->Polygon.CullFaceMode == GL_FRONT))
 524                line_aa = AA_ALWAYS;
 525          }
 526       }
 527    }
 528
 529    key->line_aa = line_aa;
 530
 531    /* _NEW_HINT */
 532    key->high_quality_derivatives =
 533       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 534
 535    if (brw->gen < 6)
 536       key->stats_wm = brw->stats_wm;
 537
 538    /* _NEW_LIGHT */
 539    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 540
 541    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 542    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 543
 544    /* _NEW_TEXTURE */
 545    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 546                                       &key->tex);
 547
 548    /* _NEW_BUFFERS */
 549    /*
 550     * Include the draw buffer origin and height so that we can calculate
 551     * fragment position values relative to the bottom left of the drawable,
 552     * from the incoming screen origin relative position we get as part of our
 553     * payload.
 554     *
 555     * This is only needed for the WM_WPOSXY opcode when the fragment program
 556     * uses the gl_FragCoord input.
 557     *
 558     * We could avoid recompiling by including this as a constant referenced by
 559     * our program, but if we were to do that it would also be nice to handle
 560     * getting that constant updated at batchbuffer submit time (when we
 561     * hold the lock and know where the buffer really is) rather than at emit
 562     * time when we don't hold the lock and are just guessing.  We could also
 563     * just avoid using this as key data if the program doesn't use
 564     * fragment.position.
 565     *
 566     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 567     * drawable height in order to invert the Y axis.
 568     */
 569    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 570       key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
 571    }
 572
 573    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 574       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 575    }
 576
 577    /* _NEW_BUFFERS */
 578    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 579
 580    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 581    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 582       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 583
 584    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 585    /* Ignore sample qualifier while computing this flag. */
 586    key->persample_shading =
 587       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
 588    if (key->persample_shading)
 589       key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
 590
 591    key->compute_pos_offset =
 592       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
 593       fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
 594
 595    key->compute_sample_id =
 596       multisample_fbo &&
 597       ctx->Multisample.Enabled &&
 598       (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
 599
 600    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 601    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 602                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 603       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 604
 605
 606    /* _NEW_COLOR | _NEW_BUFFERS */
 607    /* Pre-gen6, the hardware alpha test always used each render
 608     * target's alpha to do alpha test, as opposed to render target 0's alpha
 609     * like GL requires.  Fix that by building the alpha test into the
 610     * shader, and we'll skip enabling the fixed function alpha test.
 611     */
 612    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 613        ctx->Color.AlphaEnabled) {
 614       key->alpha_test_func = ctx->Color.AlphaFunc;
 615       key->alpha_test_ref = ctx->Color.AlphaRef;
 616    }
 617
 618    /* The unique fragment program ID */
 619    key->program_string_id = fp->id;
 620 }
 621
 622 void
 623 brw_upload_wm_prog(struct brw_context *brw)
 624 {
 625    struct gl_context *ctx = &brw->ctx;
 626    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 627    struct brw_wm_prog_key key;
 628    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 629       brw->fragment_program;
 630
 631    if (!brw_wm_state_dirty(brw))
 632       return;
 633
 634    brw_wm_populate_key(brw, &key);
 635
 636    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 637                          &key, sizeof(key),
 638                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 639       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 640       (void) success;
 641       assert(success);
 642    }
 643    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 644 }