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 "brw_shader.h"
  30 #include "main/enums.h"
  31 #include "main/formats.h"
  32 #include "main/fbobject.h"
  33 #include "main/samplerobj.h"
  34 #include "main/framebuffer.h"
  35 #include "program/prog_parameter.h"
  36 #include "program/program.h"
  37 #include "intel_mipmap_tree.h"
  38
  39 #include "util/ralloc.h"
  40
  41 /**
  42  * Return a bitfield where bit n is set if barycentric interpolation mode n
  43  * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
  44  */
  45 static unsigned
  46 brw_compute_barycentric_interp_modes(struct brw_context *brw,
  47                                      bool shade_model_flat,
  48                                      bool persample_shading,
  49                                      const struct gl_fragment_program *fprog)
  50 {
  51    unsigned barycentric_interp_modes = 0;
  52    int attr;
  53
  54    /* Loop through all fragment shader inputs to figure out what interpolation
  55     * modes are in use, and set the appropriate bits in
  56     * barycentric_interp_modes.
  57     */
  58    for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
  59       enum glsl_interp_qualifier interp_qualifier =
  60          fprog->InterpQualifier[attr];
  61       bool is_centroid = (fprog->IsCentroid & BITFIELD64_BIT(attr)) &&
  62          !persample_shading;
  63       bool is_sample = (fprog->IsSample & BITFIELD64_BIT(attr)) ||
  64          persample_shading;
  65       bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
  66
  67       /* Ignore unused inputs. */
  68       if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
  69          continue;
  70
  71       /* Ignore WPOS and FACE, because they don't require interpolation. */
  72       if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
  73          continue;
  74
  75       /* Determine the set (or sets) of barycentric coordinates needed to
  76        * interpolate this variable.  Note that when
  77        * brw->needs_unlit_centroid_workaround is set, centroid interpolation
  78        * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
  79        * for lit pixels, so we need both sets of barycentric coordinates.
  80        */
  81       if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
  82          if (is_centroid) {
  83             barycentric_interp_modes |=
  84                1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
  85          } else if (is_sample) {
  86             barycentric_interp_modes |=
  87                1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC;
  88          }
  89          if ((!is_centroid && !is_sample) ||
  90              brw->needs_unlit_centroid_workaround) {
  91             barycentric_interp_modes |=
  92                1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
  93          }
  94       } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
  95                  (!(shade_model_flat && is_gl_Color) &&
  96                   interp_qualifier == INTERP_QUALIFIER_NONE)) {
  97          if (is_centroid) {
  98             barycentric_interp_modes |=
  99                1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
 100          } else if (is_sample) {
 101             barycentric_interp_modes |=
 102                1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC;
 103          }
 104          if ((!is_centroid && !is_sample) ||
 105              brw->needs_unlit_centroid_workaround) {
 106             barycentric_interp_modes |=
 107                1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
 108          }
 109       }
 110    }
 111
 112    return barycentric_interp_modes;
 113 }
 114
 115 static uint8_t
 116 computed_depth_mode(struct gl_fragment_program *fp)
 117 {
 118    if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 119       switch (fp->FragDepthLayout) {
 120       case FRAG_DEPTH_LAYOUT_NONE:
 121       case FRAG_DEPTH_LAYOUT_ANY:
 122          return BRW_PSCDEPTH_ON;
 123       case FRAG_DEPTH_LAYOUT_GREATER:
 124          return BRW_PSCDEPTH_ON_GE;
 125       case FRAG_DEPTH_LAYOUT_LESS:
 126          return BRW_PSCDEPTH_ON_LE;
 127       case FRAG_DEPTH_LAYOUT_UNCHANGED:
 128          return BRW_PSCDEPTH_OFF;
 129       }
 130    }
 131    return BRW_PSCDEPTH_OFF;
 132 }
 133
 134 bool
 135 brw_wm_prog_data_compare(const void *in_a, const void *in_b)
 136 {
 137    const struct brw_wm_prog_data *a = in_a;
 138    const struct brw_wm_prog_data *b = in_b;
 139
 140    /* Compare the base structure. */
 141    if (!brw_stage_prog_data_compare(&a->base, &b->base))
 142       return false;
 143
 144    /* Compare the rest of the structure. */
 145    const unsigned offset = sizeof(struct brw_stage_prog_data);
 146    if (memcmp(((char *) a) + offset, ((char *) b) + offset,
 147               sizeof(struct brw_wm_prog_data) - offset))
 148       return false;
 149
 150    return true;
 151 }
 152
 153 /**
 154  * All Mesa program -> GPU code generation goes through this function.
 155  * Depending on the instructions used (i.e. flow control instructions)
 156  * we'll use one of two code generators.
 157  */
 158 bool
 159 brw_codegen_wm_prog(struct brw_context *brw,
 160                     struct gl_shader_program *prog,
 161                     struct brw_fragment_program *fp,
 162                     struct brw_wm_prog_key *key)
 163 {
 164    struct gl_context *ctx = &brw->ctx;
 165    void *mem_ctx = ralloc_context(NULL);
 166    struct brw_wm_prog_data prog_data;
 167    const GLuint *program;
 168    struct gl_shader *fs = NULL;
 169    GLuint program_size;
 170
 171    if (prog)
 172       fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 173
 174    memset(&prog_data, 0, sizeof(prog_data));
 175    /* key->alpha_test_func means simulating alpha testing via discards,
 176     * so the shader definitely kills pixels.
 177     */
 178    prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
 179    prog_data.uses_omask =
 180       fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK);
 181    prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
 182
 183    prog_data.early_fragment_tests = fs && fs->EarlyFragmentTests;
 184
 185    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 186    if (!prog)
 187       prog_data.base.use_alt_mode = true;
 188
 189    /* Allocate the references to the uniforms that will end up in the
 190     * prog_data associated with the compiled program, and which will be freed
 191     * by the state cache.
 192     */
 193    int param_count;
 194    if (fs) {
 195       param_count = fs->num_uniform_components +
 196                     fs->NumImages * BRW_IMAGE_PARAM_SIZE;
 197       prog_data.base.nr_image_params = fs->NumImages;
 198    } else {
 199       param_count = fp->program.Base.Parameters->NumParameters * 4;
 200    }
 201    /* The backend also sometimes adds params for texture size. */
 202    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 203    prog_data.base.param =
 204       rzalloc_array(NULL, const gl_constant_value *, param_count);
 205    prog_data.base.pull_param =
 206       rzalloc_array(NULL, const gl_constant_value *, param_count);
 207    prog_data.base.image_param =
 208       rzalloc_array(NULL, struct brw_image_param,
 209                     prog_data.base.nr_image_params);
 210    prog_data.base.nr_params = param_count;
 211
 212    prog_data.barycentric_interp_modes =
 213       brw_compute_barycentric_interp_modes(brw, key->flat_shade,
 214                                            key->persample_shading,
 215                                            &fp->program);
 216
 217    program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
 218                             &fp->program, prog, &program_size);
 219    if (program == NULL) {
 220       ralloc_free(mem_ctx);
 221       return false;
 222    }
 223
 224    if (prog_data.base.total_scratch) {
 225       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 226                          prog_data.base.total_scratch * brw->max_wm_threads);
 227    }
 228
 229    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 230       fprintf(stderr, "\n");
 231
 232    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 233                     key, sizeof(struct brw_wm_prog_key),
 234                     program, program_size,
 235                     &prog_data, sizeof(prog_data),
 236                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 237
 238    ralloc_free(mem_ctx);
 239
 240    return true;
 241 }
 242
 243 static bool
 244 key_debug(struct brw_context *brw, const char *name, int a, int b)
 245 {
 246    if (a != b) {
 247       perf_debug("  %s %d->%d\n", name, a, b);
 248       return true;
 249    } else {
 250       return false;
 251    }
 252 }
 253
 254 bool
 255 brw_debug_recompile_sampler_key(struct brw_context *brw,
 256                                 const struct brw_sampler_prog_key_data *old_key,
 257                                 const struct brw_sampler_prog_key_data *key)
 258 {
 259    bool found = false;
 260
 261    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 262       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 263                          old_key->swizzles[i], key->swizzles[i]);
 264    }
 265    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 266                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 267    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 268                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 269    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 270                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 271    found |= key_debug(brw, "gather channel quirk on any texture unit",
 272                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 273    found |= key_debug(brw, "compressed multisample layout",
 274                       old_key->compressed_multisample_layout_mask,
 275                       key->compressed_multisample_layout_mask);
 276
 277    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 278       found |= key_debug(brw, "textureGather workarounds",
 279                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 280    }
 281
 282    return found;
 283 }
 284
 285 void
 286 brw_wm_debug_recompile(struct brw_context *brw,
 287                        struct gl_shader_program *prog,
 288                        const struct brw_wm_prog_key *key)
 289 {
 290    struct brw_cache_item *c = NULL;
 291    const struct brw_wm_prog_key *old_key = NULL;
 292    bool found = false;
 293
 294    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 295
 296    for (unsigned int i = 0; i < brw->cache.size; i++) {
 297       for (c = brw->cache.items[i]; c; c = c->next) {
 298          if (c->cache_id == BRW_CACHE_FS_PROG) {
 299             old_key = c->key;
 300
 301             if (old_key->program_string_id == key->program_string_id)
 302                break;
 303          }
 304       }
 305       if (c)
 306          break;
 307    }
 308
 309    if (!c) {
 310       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 311       return;
 312    }
 313
 314    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 315                       "depth write",
 316                       old_key->iz_lookup, key->iz_lookup);
 317    found |= key_debug(brw, "depth statistics",
 318                       old_key->stats_wm, key->stats_wm);
 319    found |= key_debug(brw, "flat shading",
 320                       old_key->flat_shade, key->flat_shade);
 321    found |= key_debug(brw, "per-sample shading",
 322                       old_key->persample_shading, key->persample_shading);
 323    found |= key_debug(brw, "per-sample shading and 2x MSAA",
 324                       old_key->persample_2x, key->persample_2x);
 325    found |= key_debug(brw, "number of color buffers",
 326                       old_key->nr_color_regions, key->nr_color_regions);
 327    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 328                       old_key->replicate_alpha, key->replicate_alpha);
 329    found |= key_debug(brw, "rendering to FBO",
 330                       old_key->render_to_fbo, key->render_to_fbo);
 331    found |= key_debug(brw, "fragment color clamping",
 332                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 333    found |= key_debug(brw, "line smoothing",
 334                       old_key->line_aa, key->line_aa);
 335    found |= key_debug(brw, "renderbuffer height",
 336                       old_key->drawable_height, key->drawable_height);
 337    found |= key_debug(brw, "input slots valid",
 338                       old_key->input_slots_valid, key->input_slots_valid);
 339    found |= key_debug(brw, "mrt alpha test function",
 340                       old_key->alpha_test_func, key->alpha_test_func);
 341    found |= key_debug(brw, "mrt alpha test reference value",
 342                       old_key->alpha_test_ref, key->alpha_test_ref);
 343
 344    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 345
 346    if (!found) {
 347       perf_debug("  Something else\n");
 348    }
 349 }
 350
 351 static uint8_t
 352 gen6_gather_workaround(GLenum internalformat)
 353 {
 354    switch (internalformat) {
 355    case GL_R8I: return WA_SIGN | WA_8BIT;
 356    case GL_R8UI: return WA_8BIT;
 357    case GL_R16I: return WA_SIGN | WA_16BIT;
 358    case GL_R16UI: return WA_16BIT;
 359    default:
 360       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 361        * the surface state, there is no shader w/a required.
 362        */
 363       return 0;
 364    }
 365 }
 366
 367 void
 368 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 369                                    const struct gl_program *prog,
 370                                    unsigned sampler_count,
 371                                    struct brw_sampler_prog_key_data *key)
 372 {
 373    struct brw_context *brw = brw_context(ctx);
 374
 375    for (int s = 0; s < sampler_count; s++) {
 376       key->swizzles[s] = SWIZZLE_NOOP;
 377
 378       if (!(prog->SamplersUsed & (1 << s)))
 379          continue;
 380
 381       int unit_id = prog->SamplerUnits[s];
 382       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 383
 384       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 385          const struct gl_texture_object *t = unit->_Current;
 386          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 387          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 388
 389          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 390             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 391              img->_BaseFormat == GL_DEPTH_STENCIL);
 392
 393          /* Haswell handles texture swizzling as surface format overrides
 394           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 395           */
 396          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 397             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 398
 399          if (brw->gen < 8 &&
 400              sampler->MinFilter != GL_NEAREST &&
 401              sampler->MagFilter != GL_NEAREST) {
 402             if (sampler->WrapS == GL_CLAMP)
 403                key->gl_clamp_mask[0] |= 1 << s;
 404             if (sampler->WrapT == GL_CLAMP)
 405                key->gl_clamp_mask[1] |= 1 << s;
 406             if (sampler->WrapR == GL_CLAMP)
 407                key->gl_clamp_mask[2] |= 1 << s;
 408          }
 409
 410          /* gather4's channel select for green from RG32F is broken; requires
 411           * a shader w/a on IVB; fixable with just SCS on HSW.
 412           */
 413          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 414             if (img->InternalFormat == GL_RG32F)
 415                key->gather_channel_quirk_mask |= 1 << s;
 416          }
 417
 418          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 419           * UNORM/FLOAT instead and fix it in the shader.
 420           */
 421          if (brw->gen == 6 && prog->UsesGather) {
 422             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 423          }
 424
 425          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 426           * then we need to emit slightly different code to first sample the
 427           * MCS surface.
 428           */
 429          struct intel_texture_object *intel_tex =
 430             intel_texture_object((struct gl_texture_object *)t);
 431
 432          if (brw->gen >= 7 &&
 433              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
 434             key->compressed_multisample_layout_mask |= 1 << s;
 435          }
 436       }
 437    }
 438 }
 439
 440 static bool
 441 brw_wm_state_dirty (struct brw_context *brw)
 442 {
 443    return brw_state_dirty(brw,
 444                           _NEW_BUFFERS |
 445                           _NEW_COLOR |
 446                           _NEW_DEPTH |
 447                           _NEW_FRAG_CLAMP |
 448                           _NEW_HINT |
 449                           _NEW_LIGHT |
 450                           _NEW_LINE |
 451                           _NEW_MULTISAMPLE |
 452                           _NEW_POLYGON |
 453                           _NEW_STENCIL |
 454                           _NEW_TEXTURE,
 455                           BRW_NEW_FRAGMENT_PROGRAM |
 456                           BRW_NEW_REDUCED_PRIMITIVE |
 457                           BRW_NEW_STATS_WM |
 458                           BRW_NEW_VUE_MAP_GEOM_OUT);
 459 }
 460
 461 static void
 462 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 463 {
 464    struct gl_context *ctx = &brw->ctx;
 465    /* BRW_NEW_FRAGMENT_PROGRAM */
 466    const struct brw_fragment_program *fp =
 467       (struct brw_fragment_program *) brw->fragment_program;
 468    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 469    GLuint lookup = 0;
 470    GLuint line_aa;
 471    bool program_uses_dfdy = fp->program.UsesDFdy;
 472    const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 473
 474    memset(key, 0, sizeof(*key));
 475
 476    /* Build the index for table lookup
 477     */
 478    if (brw->gen < 6) {
 479       /* _NEW_COLOR */
 480       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 481          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 482
 483       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 484          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 485
 486       /* _NEW_DEPTH */
 487       if (ctx->Depth.Test)
 488          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 489
 490       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 491          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 492
 493       /* _NEW_STENCIL | _NEW_BUFFERS */
 494       if (ctx->Stencil._Enabled) {
 495          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 496
 497          if (ctx->Stencil.WriteMask[0] ||
 498              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 499             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 500       }
 501       key->iz_lookup = lookup;
 502    }
 503
 504    line_aa = AA_NEVER;
 505
 506    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 507    if (ctx->Line.SmoothFlag) {
 508       if (brw->reduced_primitive == GL_LINES) {
 509          line_aa = AA_ALWAYS;
 510       }
 511       else if (brw->reduced_primitive == GL_TRIANGLES) {
 512          if (ctx->Polygon.FrontMode == GL_LINE) {
 513             line_aa = AA_SOMETIMES;
 514
 515             if (ctx->Polygon.BackMode == GL_LINE ||
 516                 (ctx->Polygon.CullFlag &&
 517                  ctx->Polygon.CullFaceMode == GL_BACK))
 518                line_aa = AA_ALWAYS;
 519          }
 520          else if (ctx->Polygon.BackMode == GL_LINE) {
 521             line_aa = AA_SOMETIMES;
 522
 523             if ((ctx->Polygon.CullFlag &&
 524                  ctx->Polygon.CullFaceMode == GL_FRONT))
 525                line_aa = AA_ALWAYS;
 526          }
 527       }
 528    }
 529
 530    key->line_aa = line_aa;
 531
 532    /* _NEW_HINT */
 533    key->high_quality_derivatives =
 534       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 535
 536    if (brw->gen < 6)
 537       key->stats_wm = brw->stats_wm;
 538
 539    /* _NEW_LIGHT */
 540    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 541
 542    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 543    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 544
 545    /* _NEW_TEXTURE */
 546    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 547                                       &key->tex);
 548
 549    /* _NEW_BUFFERS */
 550    /*
 551     * Include the draw buffer origin and height so that we can calculate
 552     * fragment position values relative to the bottom left of the drawable,
 553     * from the incoming screen origin relative position we get as part of our
 554     * payload.
 555     *
 556     * This is only needed for the WM_WPOSXY opcode when the fragment program
 557     * uses the gl_FragCoord input.
 558     *
 559     * We could avoid recompiling by including this as a constant referenced by
 560     * our program, but if we were to do that it would also be nice to handle
 561     * getting that constant updated at batchbuffer submit time (when we
 562     * hold the lock and know where the buffer really is) rather than at emit
 563     * time when we don't hold the lock and are just guessing.  We could also
 564     * just avoid using this as key data if the program doesn't use
 565     * fragment.position.
 566     *
 567     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 568     * drawable height in order to invert the Y axis.
 569     */
 570    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 571       key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
 572    }
 573
 574    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 575       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 576    }
 577
 578    /* _NEW_BUFFERS */
 579    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 580
 581    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 582    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 583       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 584
 585    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 586    /* Ignore sample qualifier while computing this flag. */
 587    key->persample_shading =
 588       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
 589    if (key->persample_shading)
 590       key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
 591
 592    key->compute_pos_offset =
 593       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
 594       fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
 595
 596    key->compute_sample_id =
 597       multisample_fbo &&
 598       ctx->Multisample.Enabled &&
 599       (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
 600
 601    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 602    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 603                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 604       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 605
 606
 607    /* _NEW_COLOR | _NEW_BUFFERS */
 608    /* Pre-gen6, the hardware alpha test always used each render
 609     * target's alpha to do alpha test, as opposed to render target 0's alpha
 610     * like GL requires.  Fix that by building the alpha test into the
 611     * shader, and we'll skip enabling the fixed function alpha test.
 612     */
 613    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 614        ctx->Color.AlphaEnabled) {
 615       key->alpha_test_func = ctx->Color.AlphaFunc;
 616       key->alpha_test_ref = ctx->Color.AlphaRef;
 617    }
 618
 619    /* The unique fragment program ID */
 620    key->program_string_id = fp->id;
 621 }
 622
 623 void
 624 brw_upload_wm_prog(struct brw_context *brw)
 625 {
 626    struct gl_context *ctx = &brw->ctx;
 627    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 628    struct brw_wm_prog_key key;
 629    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 630       brw->fragment_program;
 631
 632    if (!brw_wm_state_dirty(brw))
 633       return;
 634
 635    brw_wm_populate_key(brw, &key);
 636
 637    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 638                          &key, sizeof(key),
 639                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 640       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 641       (void) success;
 642       assert(success);
 643    }
 644    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 645 }
 646
 647 bool
 648 brw_fs_precompile(struct gl_context *ctx,
 649                   struct gl_shader_program *shader_prog,
 650                   struct gl_program *prog)
 651 {
 652    struct brw_context *brw = brw_context(ctx);
 653    struct brw_wm_prog_key key;
 654
 655    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 656    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 657    bool program_uses_dfdy = fp->UsesDFdy;
 658
 659    memset(&key, 0, sizeof(key));
 660
 661    if (brw->gen < 6) {
 662       if (fp->UsesKill)
 663          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 664
 665       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 666          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 667
 668       /* Just assume depth testing. */
 669       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 670       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 671    }
 672
 673    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 674                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 675       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 676
 677    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 678
 679    if (fp->Base.InputsRead & VARYING_BIT_POS) {
 680       key.drawable_height = ctx->DrawBuffer->Height;
 681    }
 682
 683    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 684          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 685          BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 686
 687    if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 688       key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer) ||
 689                           key.nr_color_regions > 1;
 690    }
 691
 692    key.program_string_id = bfp->id;
 693
 694    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 695    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 696
 697    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 698
 699    brw->wm.base.prog_offset = old_prog_offset;
 700    brw->wm.prog_data = old_prog_data;
 701
 702    return success;
 703 }