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 #include "glsl/nir/nir.h"
  39
  40 #include "util/ralloc.h"
  41
  42 /**
  43  * Return a bitfield where bit n is set if barycentric interpolation mode n
  44  * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
  45  */
  46 static unsigned
  47 brw_compute_barycentric_interp_modes(struct brw_context *brw,
  48                                      bool shade_model_flat,
  49                                      bool persample_shading,
  50                                      const struct gl_fragment_program *fprog)
  51 {
  52    unsigned barycentric_interp_modes = 0;
  53    int attr;
  54
  55    /* Loop through all fragment shader inputs to figure out what interpolation
  56     * modes are in use, and set the appropriate bits in
  57     * barycentric_interp_modes.
  58     */
  59    for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
  60       enum glsl_interp_qualifier interp_qualifier =
  61          fprog->InterpQualifier[attr];
  62       bool is_centroid = (fprog->IsCentroid & BITFIELD64_BIT(attr)) &&
  63          !persample_shading;
  64       bool is_sample = (fprog->IsSample & BITFIELD64_BIT(attr)) ||
  65          persample_shading;
  66       bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
  67
  68       /* Ignore unused inputs. */
  69       if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
  70          continue;
  71
  72       /* Ignore WPOS and FACE, because they don't require interpolation. */
  73       if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
  74          continue;
  75
  76       /* Determine the set (or sets) of barycentric coordinates needed to
  77        * interpolate this variable.  Note that when
  78        * brw->needs_unlit_centroid_workaround is set, centroid interpolation
  79        * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
  80        * for lit pixels, so we need both sets of barycentric coordinates.
  81        */
  82       if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
  83          if (is_centroid) {
  84             barycentric_interp_modes |=
  85                1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
  86          } else if (is_sample) {
  87             barycentric_interp_modes |=
  88                1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC;
  89          }
  90          if ((!is_centroid && !is_sample) ||
  91              brw->needs_unlit_centroid_workaround) {
  92             barycentric_interp_modes |=
  93                1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
  94          }
  95       } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
  96                  (!(shade_model_flat && is_gl_Color) &&
  97                   interp_qualifier == INTERP_QUALIFIER_NONE)) {
  98          if (is_centroid) {
  99             barycentric_interp_modes |=
 100                1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
 101          } else if (is_sample) {
 102             barycentric_interp_modes |=
 103                1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC;
 104          }
 105          if ((!is_centroid && !is_sample) ||
 106              brw->needs_unlit_centroid_workaround) {
 107             barycentric_interp_modes |=
 108                1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
 109          }
 110       }
 111    }
 112
 113    return barycentric_interp_modes;
 114 }
 115
 116 static uint8_t
 117 computed_depth_mode(struct gl_fragment_program *fp)
 118 {
 119    if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 120       switch (fp->FragDepthLayout) {
 121       case FRAG_DEPTH_LAYOUT_NONE:
 122       case FRAG_DEPTH_LAYOUT_ANY:
 123          return BRW_PSCDEPTH_ON;
 124       case FRAG_DEPTH_LAYOUT_GREATER:
 125          return BRW_PSCDEPTH_ON_GE;
 126       case FRAG_DEPTH_LAYOUT_LESS:
 127          return BRW_PSCDEPTH_ON_LE;
 128       case FRAG_DEPTH_LAYOUT_UNCHANGED:
 129          return BRW_PSCDEPTH_OFF;
 130       }
 131    }
 132    return BRW_PSCDEPTH_OFF;
 133 }
 134
 135 /**
 136  * All Mesa program -> GPU code generation goes through this function.
 137  * Depending on the instructions used (i.e. flow control instructions)
 138  * we'll use one of two code generators.
 139  */
 140 bool
 141 brw_codegen_wm_prog(struct brw_context *brw,
 142                     struct gl_shader_program *prog,
 143                     struct brw_fragment_program *fp,
 144                     struct brw_wm_prog_key *key)
 145 {
 146    struct gl_context *ctx = &brw->ctx;
 147    void *mem_ctx = ralloc_context(NULL);
 148    struct brw_wm_prog_data prog_data;
 149    const GLuint *program;
 150    struct brw_shader *fs = NULL;
 151    GLuint program_size;
 152    bool start_busy = false;
 153    double start_time = 0;
 154
 155    if (prog)
 156       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 157
 158    memset(&prog_data, 0, sizeof(prog_data));
 159    /* key->alpha_test_func means simulating alpha testing via discards,
 160     * so the shader definitely kills pixels.
 161     */
 162    prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
 163    prog_data.uses_omask =
 164       fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK);
 165    prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
 166
 167    prog_data.early_fragment_tests = fs && fs->base.EarlyFragmentTests;
 168
 169    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 170    if (!prog)
 171       prog_data.base.use_alt_mode = true;
 172
 173    /* Allocate the references to the uniforms that will end up in the
 174     * prog_data associated with the compiled program, and which will be freed
 175     * by the state cache.
 176     */
 177    int param_count = fp->program.Base.nir->num_uniforms;
 178    if (fs)
 179       prog_data.base.nr_image_params = fs->base.NumImages;
 180    /* The backend also sometimes adds params for texture size. */
 181    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 182    prog_data.base.param =
 183       rzalloc_array(NULL, const gl_constant_value *, param_count);
 184    prog_data.base.pull_param =
 185       rzalloc_array(NULL, const gl_constant_value *, param_count);
 186    prog_data.base.image_param =
 187       rzalloc_array(NULL, struct brw_image_param,
 188                     prog_data.base.nr_image_params);
 189    prog_data.base.nr_params = param_count;
 190
 191    prog_data.barycentric_interp_modes =
 192       brw_compute_barycentric_interp_modes(brw, key->flat_shade,
 193                                            key->persample_shading,
 194                                            &fp->program);
 195
 196    if (unlikely(brw->perf_debug)) {
 197       start_busy = (brw->batch.last_bo &&
 198                     drm_intel_bo_busy(brw->batch.last_bo));
 199       start_time = get_time();
 200    }
 201
 202    program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
 203                             &fp->program, prog, &program_size);
 204    if (program == NULL) {
 205       ralloc_free(mem_ctx);
 206       return false;
 207    }
 208
 209    if (unlikely(brw->perf_debug) && fs) {
 210       if (fs->compiled_once)
 211          brw_wm_debug_recompile(brw, prog, key);
 212       fs->compiled_once = true;
 213
 214       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 215          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 216                     (get_time() - start_time) * 1000);
 217       }
 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    default:
 356       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 357        * the surface state, there is no shader w/a required.
 358        */
 359       return 0;
 360    }
 361 }
 362
 363 void
 364 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 365                                    const struct gl_program *prog,
 366                                    unsigned sampler_count,
 367                                    struct brw_sampler_prog_key_data *key)
 368 {
 369    struct brw_context *brw = brw_context(ctx);
 370
 371    for (int s = 0; s < sampler_count; s++) {
 372       key->swizzles[s] = SWIZZLE_NOOP;
 373
 374       if (!(prog->SamplersUsed & (1 << s)))
 375          continue;
 376
 377       int unit_id = prog->SamplerUnits[s];
 378       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 379
 380       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 381          const struct gl_texture_object *t = unit->_Current;
 382          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 383          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 384
 385          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 386             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 387              img->_BaseFormat == GL_DEPTH_STENCIL);
 388
 389          /* Haswell handles texture swizzling as surface format overrides
 390           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 391           */
 392          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 393             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 394
 395          if (brw->gen < 8 &&
 396              sampler->MinFilter != GL_NEAREST &&
 397              sampler->MagFilter != GL_NEAREST) {
 398             if (sampler->WrapS == GL_CLAMP)
 399                key->gl_clamp_mask[0] |= 1 << s;
 400             if (sampler->WrapT == GL_CLAMP)
 401                key->gl_clamp_mask[1] |= 1 << s;
 402             if (sampler->WrapR == GL_CLAMP)
 403                key->gl_clamp_mask[2] |= 1 << s;
 404          }
 405
 406          /* gather4's channel select for green from RG32F is broken; requires
 407           * a shader w/a on IVB; fixable with just SCS on HSW.
 408           */
 409          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 410             if (img->InternalFormat == GL_RG32F)
 411                key->gather_channel_quirk_mask |= 1 << s;
 412          }
 413
 414          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 415           * UNORM/FLOAT instead and fix it in the shader.
 416           */
 417          if (brw->gen == 6 && prog->UsesGather) {
 418             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 419          }
 420
 421          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 422           * then we need to emit slightly different code to first sample the
 423           * MCS surface.
 424           */
 425          struct intel_texture_object *intel_tex =
 426             intel_texture_object((struct gl_texture_object *)t);
 427
 428          if (brw->gen >= 7 &&
 429              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
 430             key->compressed_multisample_layout_mask |= 1 << s;
 431          }
 432       }
 433    }
 434 }
 435
 436 static bool
 437 brw_wm_state_dirty (struct brw_context *brw)
 438 {
 439    return brw_state_dirty(brw,
 440                           _NEW_BUFFERS |
 441                           _NEW_COLOR |
 442                           _NEW_DEPTH |
 443                           _NEW_FRAG_CLAMP |
 444                           _NEW_HINT |
 445                           _NEW_LIGHT |
 446                           _NEW_LINE |
 447                           _NEW_MULTISAMPLE |
 448                           _NEW_POLYGON |
 449                           _NEW_STENCIL |
 450                           _NEW_TEXTURE,
 451                           BRW_NEW_FRAGMENT_PROGRAM |
 452                           BRW_NEW_REDUCED_PRIMITIVE |
 453                           BRW_NEW_STATS_WM |
 454                           BRW_NEW_VUE_MAP_GEOM_OUT);
 455 }
 456
 457 static void
 458 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 459 {
 460    struct gl_context *ctx = &brw->ctx;
 461    /* BRW_NEW_FRAGMENT_PROGRAM */
 462    const struct brw_fragment_program *fp =
 463       (struct brw_fragment_program *) brw->fragment_program;
 464    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 465    GLuint lookup = 0;
 466    GLuint line_aa;
 467    bool program_uses_dfdy = fp->program.UsesDFdy;
 468    const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 469
 470    memset(key, 0, sizeof(*key));
 471
 472    /* Build the index for table lookup
 473     */
 474    if (brw->gen < 6) {
 475       /* _NEW_COLOR */
 476       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 477          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 478
 479       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 480          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 481
 482       /* _NEW_DEPTH */
 483       if (ctx->Depth.Test)
 484          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 485
 486       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 487          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 488
 489       /* _NEW_STENCIL | _NEW_BUFFERS */
 490       if (ctx->Stencil._Enabled) {
 491          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 492
 493          if (ctx->Stencil.WriteMask[0] ||
 494              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 495             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 496       }
 497       key->iz_lookup = lookup;
 498    }
 499
 500    line_aa = AA_NEVER;
 501
 502    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 503    if (ctx->Line.SmoothFlag) {
 504       if (brw->reduced_primitive == GL_LINES) {
 505          line_aa = AA_ALWAYS;
 506       }
 507       else if (brw->reduced_primitive == GL_TRIANGLES) {
 508          if (ctx->Polygon.FrontMode == GL_LINE) {
 509             line_aa = AA_SOMETIMES;
 510
 511             if (ctx->Polygon.BackMode == GL_LINE ||
 512                 (ctx->Polygon.CullFlag &&
 513                  ctx->Polygon.CullFaceMode == GL_BACK))
 514                line_aa = AA_ALWAYS;
 515          }
 516          else if (ctx->Polygon.BackMode == GL_LINE) {
 517             line_aa = AA_SOMETIMES;
 518
 519             if ((ctx->Polygon.CullFlag &&
 520                  ctx->Polygon.CullFaceMode == GL_FRONT))
 521                line_aa = AA_ALWAYS;
 522          }
 523       }
 524    }
 525
 526    key->line_aa = line_aa;
 527
 528    /* _NEW_HINT */
 529    key->high_quality_derivatives =
 530       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 531
 532    if (brw->gen < 6)
 533       key->stats_wm = brw->stats_wm;
 534
 535    /* _NEW_LIGHT */
 536    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 537
 538    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 539    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 540
 541    /* _NEW_TEXTURE */
 542    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 543                                       &key->tex);
 544
 545    /* _NEW_BUFFERS */
 546    /*
 547     * Include the draw buffer origin and height so that we can calculate
 548     * fragment position values relative to the bottom left of the drawable,
 549     * from the incoming screen origin relative position we get as part of our
 550     * payload.
 551     *
 552     * This is only needed for the WM_WPOSXY opcode when the fragment program
 553     * uses the gl_FragCoord input.
 554     *
 555     * We could avoid recompiling by including this as a constant referenced by
 556     * our program, but if we were to do that it would also be nice to handle
 557     * getting that constant updated at batchbuffer submit time (when we
 558     * hold the lock and know where the buffer really is) rather than at emit
 559     * time when we don't hold the lock and are just guessing.  We could also
 560     * just avoid using this as key data if the program doesn't use
 561     * fragment.position.
 562     *
 563     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 564     * drawable height in order to invert the Y axis.
 565     */
 566    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 567       key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
 568    }
 569
 570    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 571       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 572    }
 573
 574    /* _NEW_BUFFERS */
 575    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 576
 577    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 578    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 579       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 580
 581    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 582    /* Ignore sample qualifier while computing this flag. */
 583    key->persample_shading =
 584       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
 585    if (key->persample_shading)
 586       key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
 587
 588    key->compute_pos_offset =
 589       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
 590       fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
 591
 592    key->compute_sample_id =
 593       multisample_fbo &&
 594       ctx->Multisample.Enabled &&
 595       (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
 596
 597    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 598    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 599                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 600       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 601
 602
 603    /* _NEW_COLOR | _NEW_BUFFERS */
 604    /* Pre-gen6, the hardware alpha test always used each render
 605     * target's alpha to do alpha test, as opposed to render target 0's alpha
 606     * like GL requires.  Fix that by building the alpha test into the
 607     * shader, and we'll skip enabling the fixed function alpha test.
 608     */
 609    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 610        ctx->Color.AlphaEnabled) {
 611       key->alpha_test_func = ctx->Color.AlphaFunc;
 612       key->alpha_test_ref = ctx->Color.AlphaRef;
 613    }
 614
 615    /* The unique fragment program ID */
 616    key->program_string_id = fp->id;
 617 }
 618
 619 void
 620 brw_upload_wm_prog(struct brw_context *brw)
 621 {
 622    struct gl_context *ctx = &brw->ctx;
 623    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 624    struct brw_wm_prog_key key;
 625    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 626       brw->fragment_program;
 627
 628    if (!brw_wm_state_dirty(brw))
 629       return;
 630
 631    brw_wm_populate_key(brw, &key);
 632
 633    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 634                          &key, sizeof(key),
 635                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 636       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 637       (void) success;
 638       assert(success);
 639    }
 640    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 641 }
 642
 643 bool
 644 brw_fs_precompile(struct gl_context *ctx,
 645                   struct gl_shader_program *shader_prog,
 646                   struct gl_program *prog)
 647 {
 648    struct brw_context *brw = brw_context(ctx);
 649    struct brw_wm_prog_key key;
 650
 651    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 652    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 653    bool program_uses_dfdy = fp->UsesDFdy;
 654
 655    memset(&key, 0, sizeof(key));
 656
 657    if (brw->gen < 6) {
 658       if (fp->UsesKill)
 659          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 660
 661       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 662          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 663
 664       /* Just assume depth testing. */
 665       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 666       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 667    }
 668
 669    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 670                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 671       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 672
 673    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 674
 675    if (fp->Base.InputsRead & VARYING_BIT_POS) {
 676       key.drawable_height = ctx->DrawBuffer->Height;
 677    }
 678
 679    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 680          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 681          BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 682
 683    if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 684       key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer) ||
 685                           key.nr_color_regions > 1;
 686    }
 687
 688    key.program_string_id = bfp->id;
 689
 690    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 691    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 692
 693    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 694
 695    brw->wm.base.prog_offset = old_prog_offset;
 696    brw->wm.prog_data = old_prog_data;
 697
 698    return success;
 699 }