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
 163 brw_codegen_wm_prog(struct brw_context *brw,
 164                     struct gl_shader_program *prog,
 165                     struct brw_fragment_program *fp,
 166                     struct brw_wm_prog_key *key)
 167 {
 168    struct gl_context *ctx = &brw->ctx;
 169    void *mem_ctx = ralloc_context(NULL);
 170    struct brw_wm_prog_data prog_data;
 171    const GLuint *program;
 172    struct gl_shader *fs = NULL;
 173    GLuint program_size;
 174
 175    if (prog)
 176       fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 177
 178    memset(&prog_data, 0, sizeof(prog_data));
 179    /* key->alpha_test_func means simulating alpha testing via discards,
 180     * so the shader definitely kills pixels.
 181     */
 182    prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
 183
 184    prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
 185
 186    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 187    if (!prog)
 188       prog_data.base.use_alt_mode = true;
 189
 190    /* Allocate the references to the uniforms that will end up in the
 191     * prog_data associated with the compiled program, and which will be freed
 192     * by the state cache.
 193     */
 194    int param_count;
 195    if (fs) {
 196       param_count = fs->num_uniform_components;
 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.nr_params = param_count;
 207
 208    prog_data.barycentric_interp_modes =
 209       brw_compute_barycentric_interp_modes(brw, key->flat_shade,
 210                                            key->persample_shading,
 211                                            &fp->program);
 212
 213    program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
 214                             &fp->program, prog, &program_size);
 215    if (program == NULL) {
 216       ralloc_free(mem_ctx);
 217       return false;
 218    }
 219
 220    if (prog_data.base.total_scratch) {
 221       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 222                          prog_data.base.total_scratch * brw->max_wm_threads);
 223    }
 224
 225    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 226       fprintf(stderr, "\n");
 227
 228    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 229                     key, sizeof(struct brw_wm_prog_key),
 230                     program, program_size,
 231                     &prog_data, sizeof(prog_data),
 232                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 233
 234    ralloc_free(mem_ctx);
 235
 236    return true;
 237 }
 238
 239 static bool
 240 key_debug(struct brw_context *brw, const char *name, int a, int b)
 241 {
 242    if (a != b) {
 243       perf_debug("  %s %d->%d\n", name, a, b);
 244       return true;
 245    } else {
 246       return false;
 247    }
 248 }
 249
 250 bool
 251 brw_debug_recompile_sampler_key(struct brw_context *brw,
 252                                 const struct brw_sampler_prog_key_data *old_key,
 253                                 const struct brw_sampler_prog_key_data *key)
 254 {
 255    bool found = false;
 256
 257    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 258       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 259                          old_key->swizzles[i], key->swizzles[i]);
 260    }
 261    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 262                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 263    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 264                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 265    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 266                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 267    found |= key_debug(brw, "gather channel quirk on any texture unit",
 268                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 269    found |= key_debug(brw, "compressed multisample layout",
 270                       old_key->compressed_multisample_layout_mask,
 271                       key->compressed_multisample_layout_mask);
 272
 273    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 274       found |= key_debug(brw, "textureGather workarounds",
 275                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 276    }
 277
 278    return found;
 279 }
 280
 281 void
 282 brw_wm_debug_recompile(struct brw_context *brw,
 283                        struct gl_shader_program *prog,
 284                        const struct brw_wm_prog_key *key)
 285 {
 286    struct brw_cache_item *c = NULL;
 287    const struct brw_wm_prog_key *old_key = NULL;
 288    bool found = false;
 289
 290    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 291
 292    for (unsigned int i = 0; i < brw->cache.size; i++) {
 293       for (c = brw->cache.items[i]; c; c = c->next) {
 294          if (c->cache_id == BRW_CACHE_FS_PROG) {
 295             old_key = c->key;
 296
 297             if (old_key->program_string_id == key->program_string_id)
 298                break;
 299          }
 300       }
 301       if (c)
 302          break;
 303    }
 304
 305    if (!c) {
 306       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 307       return;
 308    }
 309
 310    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 311                       "depth write",
 312                       old_key->iz_lookup, key->iz_lookup);
 313    found |= key_debug(brw, "depth statistics",
 314                       old_key->stats_wm, key->stats_wm);
 315    found |= key_debug(brw, "flat shading",
 316                       old_key->flat_shade, key->flat_shade);
 317    found |= key_debug(brw, "per-sample shading",
 318                       old_key->persample_shading, key->persample_shading);
 319    found |= key_debug(brw, "per-sample shading and 2x MSAA",
 320                       old_key->persample_2x, key->persample_2x);
 321    found |= key_debug(brw, "number of color buffers",
 322                       old_key->nr_color_regions, key->nr_color_regions);
 323    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 324                       old_key->replicate_alpha, key->replicate_alpha);
 325    found |= key_debug(brw, "rendering to FBO",
 326                       old_key->render_to_fbo, key->render_to_fbo);
 327    found |= key_debug(brw, "fragment color clamping",
 328                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 329    found |= key_debug(brw, "line smoothing",
 330                       old_key->line_aa, key->line_aa);
 331    found |= key_debug(brw, "renderbuffer height",
 332                       old_key->drawable_height, key->drawable_height);
 333    found |= key_debug(brw, "input slots valid",
 334                       old_key->input_slots_valid, key->input_slots_valid);
 335    found |= key_debug(brw, "mrt alpha test function",
 336                       old_key->alpha_test_func, key->alpha_test_func);
 337    found |= key_debug(brw, "mrt alpha test reference value",
 338                       old_key->alpha_test_ref, key->alpha_test_ref);
 339
 340    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 341
 342    if (!found) {
 343       perf_debug("  Something else\n");
 344    }
 345 }
 346
 347 static uint8_t
 348 gen6_gather_workaround(GLenum internalformat)
 349 {
 350    switch (internalformat) {
 351       case GL_R8I: return WA_SIGN | WA_8BIT;
 352       case GL_R8UI: return WA_8BIT;
 353       case GL_R16I: return WA_SIGN | WA_16BIT;
 354       case GL_R16UI: return WA_16BIT;
 355       /* note that even though GL_R32I and GL_R32UI have format overrides
 356        * in the surface state, there is no shader w/a required */
 357       default: return 0;
 358    }
 359 }
 360
 361 void
 362 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 363                                    const struct gl_program *prog,
 364                                    unsigned sampler_count,
 365                                    struct brw_sampler_prog_key_data *key)
 366 {
 367    struct brw_context *brw = brw_context(ctx);
 368
 369    for (int s = 0; s < sampler_count; s++) {
 370       key->swizzles[s] = SWIZZLE_NOOP;
 371
 372       if (!(prog->SamplersUsed & (1 << s)))
 373          continue;
 374
 375       int unit_id = prog->SamplerUnits[s];
 376       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 377
 378       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 379          const struct gl_texture_object *t = unit->_Current;
 380          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 381          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 382
 383          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 384             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 385              img->_BaseFormat == GL_DEPTH_STENCIL);
 386
 387          /* Haswell handles texture swizzling as surface format overrides
 388           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 389           */
 390          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 391             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 392
 393          if (brw->gen < 8 &&
 394              sampler->MinFilter != GL_NEAREST &&
 395              sampler->MagFilter != GL_NEAREST) {
 396             if (sampler->WrapS == GL_CLAMP)
 397                key->gl_clamp_mask[0] |= 1 << s;
 398             if (sampler->WrapT == GL_CLAMP)
 399                key->gl_clamp_mask[1] |= 1 << s;
 400             if (sampler->WrapR == GL_CLAMP)
 401                key->gl_clamp_mask[2] |= 1 << s;
 402          }
 403
 404          /* gather4's channel select for green from RG32F is broken;
 405           * requires a shader w/a on IVB; fixable with just SCS on HSW. */
 406          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 407             if (img->InternalFormat == GL_RG32F)
 408                key->gather_channel_quirk_mask |= 1 << s;
 409          }
 410
 411          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 412           * UNORM/FLOAT instead and fix it in the shader.
 413           */
 414          if (brw->gen == 6 && prog->UsesGather) {
 415             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 416          }
 417
 418          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 419           * then we need to emit slightly different code to first sample the
 420           * MCS surface.
 421           */
 422          struct intel_texture_object *intel_tex =
 423             intel_texture_object((struct gl_texture_object *)t);
 424
 425          if (brw->gen >= 7 &&
 426              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
 427             key->compressed_multisample_layout_mask |= 1 << s;
 428          }
 429       }
 430    }
 431 }
 432
 433 static bool
 434 brw_wm_state_dirty (struct brw_context *brw)
 435 {
 436    return brw_state_dirty(brw,
 437                           _NEW_BUFFERS |
 438                           _NEW_COLOR |
 439                           _NEW_DEPTH |
 440                           _NEW_FRAG_CLAMP |
 441                           _NEW_HINT |
 442                           _NEW_LIGHT |
 443                           _NEW_LINE |
 444                           _NEW_MULTISAMPLE |
 445                           _NEW_POLYGON |
 446                           _NEW_STENCIL |
 447                           _NEW_TEXTURE,
 448                           BRW_NEW_FRAGMENT_PROGRAM |
 449                           BRW_NEW_REDUCED_PRIMITIVE |
 450                           BRW_NEW_STATS_WM |
 451                           BRW_NEW_VUE_MAP_GEOM_OUT);
 452 }
 453
 454 static void brw_wm_populate_key( struct brw_context *brw,
 455                                  struct brw_wm_prog_key *key )
 456 {
 457    struct gl_context *ctx = &brw->ctx;
 458    /* BRW_NEW_FRAGMENT_PROGRAM */
 459    const struct brw_fragment_program *fp =
 460       (struct brw_fragment_program *)brw->fragment_program;
 461    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 462    GLuint lookup = 0;
 463    GLuint line_aa;
 464    bool program_uses_dfdy = fp->program.UsesDFdy;
 465    bool multisample_fbo = ctx->DrawBuffer->Visual.samples > 1;
 466
 467    memset(key, 0, sizeof(*key));
 468
 469    /* Build the index for table lookup
 470     */
 471    if (brw->gen < 6) {
 472       /* _NEW_COLOR */
 473       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 474          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 475
 476       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 477          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 478
 479       /* _NEW_DEPTH */
 480       if (ctx->Depth.Test)
 481          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 482
 483       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 484          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 485
 486       /* _NEW_STENCIL | _NEW_BUFFERS */
 487       if (ctx->Stencil._Enabled) {
 488          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 489
 490          if (ctx->Stencil.WriteMask[0] ||
 491              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 492             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 493       }
 494       key->iz_lookup = lookup;
 495    }
 496
 497    line_aa = AA_NEVER;
 498
 499    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 500    if (ctx->Line.SmoothFlag) {
 501       if (brw->reduced_primitive == GL_LINES) {
 502          line_aa = AA_ALWAYS;
 503       }
 504       else if (brw->reduced_primitive == GL_TRIANGLES) {
 505          if (ctx->Polygon.FrontMode == GL_LINE) {
 506             line_aa = AA_SOMETIMES;
 507
 508             if (ctx->Polygon.BackMode == GL_LINE ||
 509                 (ctx->Polygon.CullFlag &&
 510                  ctx->Polygon.CullFaceMode == GL_BACK))
 511                line_aa = AA_ALWAYS;
 512          }
 513          else if (ctx->Polygon.BackMode == GL_LINE) {
 514             line_aa = AA_SOMETIMES;
 515
 516             if ((ctx->Polygon.CullFlag &&
 517                  ctx->Polygon.CullFaceMode == GL_FRONT))
 518                line_aa = AA_ALWAYS;
 519          }
 520       }
 521    }
 522
 523    key->line_aa = line_aa;
 524
 525    /* _NEW_HINT */
 526    key->high_quality_derivatives =
 527       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 528
 529    if (brw->gen < 6)
 530       key->stats_wm = brw->stats_wm;
 531
 532    /* _NEW_LIGHT */
 533    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 534
 535    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 536    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 537
 538    /* _NEW_TEXTURE */
 539    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 540                                       &key->tex);
 541
 542    /* _NEW_BUFFERS */
 543    /*
 544     * Include the draw buffer origin and height so that we can calculate
 545     * fragment position values relative to the bottom left of the drawable,
 546     * from the incoming screen origin relative position we get as part of our
 547     * payload.
 548     *
 549     * This is only needed for the WM_WPOSXY opcode when the fragment program
 550     * uses the gl_FragCoord input.
 551     *
 552     * We could avoid recompiling by including this as a constant referenced by
 553     * our program, but if we were to do that it would also be nice to handle
 554     * getting that constant updated at batchbuffer submit time (when we
 555     * hold the lock and know where the buffer really is) rather than at emit
 556     * time when we don't hold the lock and are just guessing.  We could also
 557     * just avoid using this as key data if the program doesn't use
 558     * fragment.position.
 559     *
 560     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 561     * drawable height in order to invert the Y axis.
 562     */
 563    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 564       key->drawable_height = ctx->DrawBuffer->Height;
 565    }
 566
 567    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 568       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 569    }
 570
 571    /* _NEW_BUFFERS */
 572    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 573
 574    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 575    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 576       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 577
 578    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 579    /* Ignore sample qualifier while computing this flag. */
 580    key->persample_shading =
 581       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
 582    if (key->persample_shading)
 583       key->persample_2x = ctx->DrawBuffer->Visual.samples == 2;
 584
 585    key->compute_pos_offset =
 586       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
 587       fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
 588
 589    key->compute_sample_id =
 590       multisample_fbo &&
 591       ctx->Multisample.Enabled &&
 592       (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
 593
 594    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 595    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 596                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 597       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 598
 599
 600    /* _NEW_COLOR | _NEW_BUFFERS */
 601    /* Pre-gen6, the hardware alpha test always used each render
 602     * target's alpha to do alpha test, as opposed to render target 0's alpha
 603     * like GL requires.  Fix that by building the alpha test into the
 604     * shader, and we'll skip enabling the fixed function alpha test.
 605     */
 606    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 && ctx->Color.AlphaEnabled) {
 607       key->alpha_test_func = ctx->Color.AlphaFunc;
 608       key->alpha_test_ref = ctx->Color.AlphaRef;
 609    }
 610
 611    /* The unique fragment program ID */
 612    key->program_string_id = fp->id;
 613 }
 614
 615 void
 616 brw_upload_wm_prog(struct brw_context *brw)
 617 {
 618    struct gl_context *ctx = &brw->ctx;
 619    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 620    struct brw_wm_prog_key key;
 621    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 622       brw->fragment_program;
 623
 624    if (!brw_wm_state_dirty(brw))
 625       return;
 626
 627    brw_wm_populate_key(brw, &key);
 628
 629    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 630                          &key, sizeof(key),
 631                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 632       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 633       (void) success;
 634       assert(success);
 635    }
 636    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 637 }