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 "brw_nir.h"
  39 #include "brw_program.h"
  40
  41 #include "util/ralloc.h"
  42
  43 static void
  44 assign_fs_binding_table_offsets(const struct brw_device_info *devinfo,
  45                                 const struct gl_shader_program *shader_prog,
  46                                 const struct gl_program *prog,
  47                                 const struct brw_wm_prog_key *key,
  48                                 struct brw_wm_prog_data *prog_data)
  49 {
  50    uint32_t next_binding_table_offset = 0;
  51
  52    /* If there are no color regions, we still perform an FB write to a null
  53     * renderbuffer, which we place at surface index 0.
  54     */
  55    prog_data->binding_table.render_target_start = next_binding_table_offset;
  56    next_binding_table_offset += MAX2(key->nr_color_regions, 1);
  57
  58    brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
  59                                            shader_prog, prog, &prog_data->base,
  60                                            next_binding_table_offset);
  61 }
  62
  63 /**
  64  * All Mesa program -> GPU code generation goes through this function.
  65  * Depending on the instructions used (i.e. flow control instructions)
  66  * we'll use one of two code generators.
  67  */
  68 bool
  69 brw_codegen_wm_prog(struct brw_context *brw,
  70                     struct gl_shader_program *prog,
  71                     struct brw_fragment_program *fp,
  72                     struct brw_wm_prog_key *key)
  73 {
  74    struct gl_context *ctx = &brw->ctx;
  75    void *mem_ctx = ralloc_context(NULL);
  76    struct brw_wm_prog_data prog_data;
  77    const GLuint *program;
  78    struct brw_shader *fs = NULL;
  79    GLuint program_size;
  80    bool start_busy = false;
  81    double start_time = 0;
  82
  83    if (prog)
  84       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  85
  86    memset(&prog_data, 0, sizeof(prog_data));
  87
  88    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
  89    if (!prog)
  90       prog_data.base.use_alt_mode = true;
  91
  92    assign_fs_binding_table_offsets(brw->intelScreen->devinfo, prog,
  93                                    &fp->program.Base, key, &prog_data);
  94
  95    /* Allocate the references to the uniforms that will end up in the
  96     * prog_data associated with the compiled program, and which will be freed
  97     * by the state cache.
  98     */
  99    int param_count = fp->program.Base.nir->num_uniforms;
 100    if (fs)
 101       prog_data.base.nr_image_params = fs->base.NumImages;
 102    /* The backend also sometimes adds params for texture size. */
 103    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 104    prog_data.base.param =
 105       rzalloc_array(NULL, const gl_constant_value *, param_count);
 106    prog_data.base.pull_param =
 107       rzalloc_array(NULL, const gl_constant_value *, param_count);
 108    prog_data.base.image_param =
 109       rzalloc_array(NULL, struct brw_image_param,
 110                     prog_data.base.nr_image_params);
 111    prog_data.base.nr_params = param_count;
 112
 113    if (prog) {
 114       brw_nir_setup_glsl_uniforms(fp->program.Base.nir, prog, &fp->program.Base,
 115                                   &prog_data.base, true);
 116    } else {
 117       brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base,
 118                                  &prog_data.base);
 119    }
 120
 121    if (unlikely(brw->perf_debug)) {
 122       start_busy = (brw->batch.last_bo &&
 123                     drm_intel_bo_busy(brw->batch.last_bo));
 124       start_time = get_time();
 125    }
 126
 127    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 128       brw_dump_ir("fragment", prog, fs ? &fs->base : NULL, &fp->program.Base);
 129
 130    int st_index8 = -1, st_index16 = -1;
 131    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 132       st_index8 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS8);
 133       st_index16 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS16);
 134    }
 135
 136    char *error_str = NULL;
 137    program = brw_compile_fs(brw->intelScreen->compiler, brw, mem_ctx,
 138                             key, &prog_data, fp->program.Base.nir,
 139                             &fp->program.Base, st_index8, st_index16,
 140                             brw->use_rep_send, &program_size, &error_str);
 141    if (program == NULL) {
 142       if (prog) {
 143          prog->LinkStatus = false;
 144          ralloc_strcat(&prog->InfoLog, error_str);
 145       }
 146
 147       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 148
 149       ralloc_free(mem_ctx);
 150       return false;
 151    }
 152
 153    if (unlikely(brw->perf_debug) && fs) {
 154       if (fs->compiled_once)
 155          brw_wm_debug_recompile(brw, prog, key);
 156       fs->compiled_once = true;
 157
 158       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 159          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 160                     (get_time() - start_time) * 1000);
 161       }
 162    }
 163
 164    if (prog_data.base.total_scratch) {
 165       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 166                          prog_data.base.total_scratch * brw->max_wm_threads);
 167    }
 168
 169    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 170       fprintf(stderr, "\n");
 171
 172    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 173                     key, sizeof(struct brw_wm_prog_key),
 174                     program, program_size,
 175                     &prog_data, sizeof(prog_data),
 176                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 177
 178    ralloc_free(mem_ctx);
 179
 180    return true;
 181 }
 182
 183 bool
 184 brw_debug_recompile_sampler_key(struct brw_context *brw,
 185                                 const struct brw_sampler_prog_key_data *old_key,
 186                                 const struct brw_sampler_prog_key_data *key)
 187 {
 188    bool found = false;
 189
 190    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 191       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 192                          old_key->swizzles[i], key->swizzles[i]);
 193    }
 194    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 195                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 196    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 197                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 198    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 199                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 200    found |= key_debug(brw, "gather channel quirk on any texture unit",
 201                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 202    found |= key_debug(brw, "compressed multisample layout",
 203                       old_key->compressed_multisample_layout_mask,
 204                       key->compressed_multisample_layout_mask);
 205    found |= key_debug(brw, "16x msaa",
 206                       old_key->msaa_16,
 207                       key->msaa_16);
 208
 209    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 210       found |= key_debug(brw, "textureGather workarounds",
 211                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 212    }
 213
 214    return found;
 215 }
 216
 217 void
 218 brw_wm_debug_recompile(struct brw_context *brw,
 219                        struct gl_shader_program *prog,
 220                        const struct brw_wm_prog_key *key)
 221 {
 222    struct brw_cache_item *c = NULL;
 223    const struct brw_wm_prog_key *old_key = NULL;
 224    bool found = false;
 225
 226    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 227
 228    for (unsigned int i = 0; i < brw->cache.size; i++) {
 229       for (c = brw->cache.items[i]; c; c = c->next) {
 230          if (c->cache_id == BRW_CACHE_FS_PROG) {
 231             old_key = c->key;
 232
 233             if (old_key->program_string_id == key->program_string_id)
 234                break;
 235          }
 236       }
 237       if (c)
 238          break;
 239    }
 240
 241    if (!c) {
 242       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 243       return;
 244    }
 245
 246    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 247                       "depth write",
 248                       old_key->iz_lookup, key->iz_lookup);
 249    found |= key_debug(brw, "depth statistics",
 250                       old_key->stats_wm, key->stats_wm);
 251    found |= key_debug(brw, "flat shading",
 252                       old_key->flat_shade, key->flat_shade);
 253    found |= key_debug(brw, "per-sample interpolation",
 254                       old_key->persample_interp, key->persample_interp);
 255    found |= key_debug(brw, "number of color buffers",
 256                       old_key->nr_color_regions, key->nr_color_regions);
 257    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 258                       old_key->replicate_alpha, key->replicate_alpha);
 259    found |= key_debug(brw, "rendering to FBO",
 260                       old_key->render_to_fbo, key->render_to_fbo);
 261    found |= key_debug(brw, "fragment color clamping",
 262                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 263    found |= key_debug(brw, "multisampled FBO",
 264                       old_key->multisample_fbo, key->multisample_fbo);
 265    found |= key_debug(brw, "line smoothing",
 266                       old_key->line_aa, key->line_aa);
 267    found |= key_debug(brw, "renderbuffer height",
 268                       old_key->drawable_height, key->drawable_height);
 269    found |= key_debug(brw, "input slots valid",
 270                       old_key->input_slots_valid, key->input_slots_valid);
 271    found |= key_debug(brw, "mrt alpha test function",
 272                       old_key->alpha_test_func, key->alpha_test_func);
 273    found |= key_debug(brw, "mrt alpha test reference value",
 274                       old_key->alpha_test_ref, key->alpha_test_ref);
 275
 276    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 277
 278    if (!found) {
 279       perf_debug("  Something else\n");
 280    }
 281 }
 282
 283 static uint8_t
 284 gen6_gather_workaround(GLenum internalformat)
 285 {
 286    switch (internalformat) {
 287    case GL_R8I: return WA_SIGN | WA_8BIT;
 288    case GL_R8UI: return WA_8BIT;
 289    case GL_R16I: return WA_SIGN | WA_16BIT;
 290    case GL_R16UI: return WA_16BIT;
 291    default:
 292       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 293        * the surface state, there is no shader w/a required.
 294        */
 295       return 0;
 296    }
 297 }
 298
 299 void
 300 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 301                                    const struct gl_program *prog,
 302                                    unsigned sampler_count,
 303                                    struct brw_sampler_prog_key_data *key)
 304 {
 305    struct brw_context *brw = brw_context(ctx);
 306
 307    for (int s = 0; s < sampler_count; s++) {
 308       key->swizzles[s] = SWIZZLE_NOOP;
 309
 310       if (!(prog->SamplersUsed & (1 << s)))
 311          continue;
 312
 313       int unit_id = prog->SamplerUnits[s];
 314       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 315
 316       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 317          const struct gl_texture_object *t = unit->_Current;
 318          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 319          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 320
 321          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 322             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 323              img->_BaseFormat == GL_DEPTH_STENCIL);
 324
 325          /* Haswell handles texture swizzling as surface format overrides
 326           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 327           */
 328          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 329             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 330
 331          if (brw->gen < 8 &&
 332              sampler->MinFilter != GL_NEAREST &&
 333              sampler->MagFilter != GL_NEAREST) {
 334             if (sampler->WrapS == GL_CLAMP)
 335                key->gl_clamp_mask[0] |= 1 << s;
 336             if (sampler->WrapT == GL_CLAMP)
 337                key->gl_clamp_mask[1] |= 1 << s;
 338             if (sampler->WrapR == GL_CLAMP)
 339                key->gl_clamp_mask[2] |= 1 << s;
 340          }
 341
 342          /* gather4's channel select for green from RG32F is broken; requires
 343           * a shader w/a on IVB; fixable with just SCS on HSW.
 344           */
 345          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 346             if (img->InternalFormat == GL_RG32F)
 347                key->gather_channel_quirk_mask |= 1 << s;
 348          }
 349
 350          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 351           * UNORM/FLOAT instead and fix it in the shader.
 352           */
 353          if (brw->gen == 6 && prog->UsesGather) {
 354             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 355          }
 356
 357          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 358           * then we need to emit slightly different code to first sample the
 359           * MCS surface.
 360           */
 361          struct intel_texture_object *intel_tex =
 362             intel_texture_object((struct gl_texture_object *)t);
 363
 364          /* From gen9 onwards some single sampled buffers can also be
 365           * compressed. These don't need ld2dms sampling along with mcs fetch.
 366           */
 367          if (brw->gen >= 7 &&
 368              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS &&
 369              intel_tex->mt->num_samples > 1) {
 370             key->compressed_multisample_layout_mask |= 1 << s;
 371
 372             if (intel_tex->mt->num_samples >= 16) {
 373                assert(brw->gen >= 9);
 374                key->msaa_16 |= 1 << s;
 375             }
 376          }
 377       }
 378    }
 379 }
 380
 381 static bool
 382 brw_wm_state_dirty (struct brw_context *brw)
 383 {
 384    return brw_state_dirty(brw,
 385                           _NEW_BUFFERS |
 386                           _NEW_COLOR |
 387                           _NEW_DEPTH |
 388                           _NEW_FRAG_CLAMP |
 389                           _NEW_HINT |
 390                           _NEW_LIGHT |
 391                           _NEW_LINE |
 392                           _NEW_MULTISAMPLE |
 393                           _NEW_POLYGON |
 394                           _NEW_STENCIL |
 395                           _NEW_TEXTURE,
 396                           BRW_NEW_FRAGMENT_PROGRAM |
 397                           BRW_NEW_REDUCED_PRIMITIVE |
 398                           BRW_NEW_STATS_WM |
 399                           BRW_NEW_VUE_MAP_GEOM_OUT);
 400 }
 401
 402 static void
 403 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 404 {
 405    struct gl_context *ctx = &brw->ctx;
 406    /* BRW_NEW_FRAGMENT_PROGRAM */
 407    const struct brw_fragment_program *fp =
 408       (struct brw_fragment_program *) brw->fragment_program;
 409    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 410    GLuint lookup = 0;
 411    GLuint line_aa;
 412    bool program_uses_dfdy = fp->program.UsesDFdy;
 413
 414    memset(key, 0, sizeof(*key));
 415
 416    /* Build the index for table lookup
 417     */
 418    if (brw->gen < 6) {
 419       /* _NEW_COLOR */
 420       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 421          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 422
 423       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 424          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 425
 426       /* _NEW_DEPTH */
 427       if (ctx->Depth.Test)
 428          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 429
 430       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 431          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 432
 433       /* _NEW_STENCIL | _NEW_BUFFERS */
 434       if (ctx->Stencil._Enabled) {
 435          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 436
 437          if (ctx->Stencil.WriteMask[0] ||
 438              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 439             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 440       }
 441       key->iz_lookup = lookup;
 442    }
 443
 444    line_aa = AA_NEVER;
 445
 446    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 447    if (ctx->Line.SmoothFlag) {
 448       if (brw->reduced_primitive == GL_LINES) {
 449          line_aa = AA_ALWAYS;
 450       }
 451       else if (brw->reduced_primitive == GL_TRIANGLES) {
 452          if (ctx->Polygon.FrontMode == GL_LINE) {
 453             line_aa = AA_SOMETIMES;
 454
 455             if (ctx->Polygon.BackMode == GL_LINE ||
 456                 (ctx->Polygon.CullFlag &&
 457                  ctx->Polygon.CullFaceMode == GL_BACK))
 458                line_aa = AA_ALWAYS;
 459          }
 460          else if (ctx->Polygon.BackMode == GL_LINE) {
 461             line_aa = AA_SOMETIMES;
 462
 463             if ((ctx->Polygon.CullFlag &&
 464                  ctx->Polygon.CullFaceMode == GL_FRONT))
 465                line_aa = AA_ALWAYS;
 466          }
 467       }
 468    }
 469
 470    key->line_aa = line_aa;
 471
 472    /* _NEW_HINT */
 473    key->high_quality_derivatives =
 474       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 475
 476    if (brw->gen < 6)
 477       key->stats_wm = brw->stats_wm;
 478
 479    /* _NEW_LIGHT */
 480    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 481
 482    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 483    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 484
 485    /* _NEW_TEXTURE */
 486    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 487                                       &key->tex);
 488
 489    /* _NEW_BUFFERS */
 490    /*
 491     * Include the draw buffer origin and height so that we can calculate
 492     * fragment position values relative to the bottom left of the drawable,
 493     * from the incoming screen origin relative position we get as part of our
 494     * payload.
 495     *
 496     * This is only needed for the WM_WPOSXY opcode when the fragment program
 497     * uses the gl_FragCoord input.
 498     *
 499     * We could avoid recompiling by including this as a constant referenced by
 500     * our program, but if we were to do that it would also be nice to handle
 501     * getting that constant updated at batchbuffer submit time (when we
 502     * hold the lock and know where the buffer really is) rather than at emit
 503     * time when we don't hold the lock and are just guessing.  We could also
 504     * just avoid using this as key data if the program doesn't use
 505     * fragment.position.
 506     *
 507     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 508     * drawable height in order to invert the Y axis.
 509     */
 510    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 511       key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
 512    }
 513
 514    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) ||
 515        program_uses_dfdy || prog->nir->info.uses_interp_var_at_offset) {
 516       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 517    }
 518
 519    /* _NEW_BUFFERS */
 520    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 521
 522    /* _NEW_COLOR */
 523    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 524       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 525
 526    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 527    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 528       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 529
 530    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 531    /* Ignore sample qualifier while computing this flag. */
 532    if (ctx->Multisample.Enabled) {
 533       key->persample_interp =
 534          ctx->Multisample.SampleShading &&
 535          (ctx->Multisample.MinSampleShadingValue *
 536           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 537
 538       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 539    }
 540
 541    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 542    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 543                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 544       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 545
 546
 547    /* _NEW_COLOR | _NEW_BUFFERS */
 548    /* Pre-gen6, the hardware alpha test always used each render
 549     * target's alpha to do alpha test, as opposed to render target 0's alpha
 550     * like GL requires.  Fix that by building the alpha test into the
 551     * shader, and we'll skip enabling the fixed function alpha test.
 552     */
 553    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 554        ctx->Color.AlphaEnabled) {
 555       key->alpha_test_func = ctx->Color.AlphaFunc;
 556       key->alpha_test_ref = ctx->Color.AlphaRef;
 557    }
 558
 559    /* The unique fragment program ID */
 560    key->program_string_id = fp->id;
 561 }
 562
 563 void
 564 brw_upload_wm_prog(struct brw_context *brw)
 565 {
 566    struct gl_context *ctx = &brw->ctx;
 567    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 568    struct brw_wm_prog_key key;
 569    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 570       brw->fragment_program;
 571
 572    if (!brw_wm_state_dirty(brw))
 573       return;
 574
 575    brw_wm_populate_key(brw, &key);
 576
 577    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 578                          &key, sizeof(key),
 579                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 580       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 581       (void) success;
 582       assert(success);
 583    }
 584    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 585 }
 586
 587 bool
 588 brw_fs_precompile(struct gl_context *ctx,
 589                   struct gl_shader_program *shader_prog,
 590                   struct gl_program *prog)
 591 {
 592    struct brw_context *brw = brw_context(ctx);
 593    struct brw_wm_prog_key key;
 594
 595    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 596    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 597    bool program_uses_dfdy = fp->UsesDFdy;
 598
 599    memset(&key, 0, sizeof(key));
 600
 601    if (brw->gen < 6) {
 602       if (fp->UsesKill)
 603          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 604
 605       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 606          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 607
 608       /* Just assume depth testing. */
 609       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 610       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 611    }
 612
 613    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 614                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 615       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 616
 617    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 618
 619    if (fp->Base.InputsRead & VARYING_BIT_POS) {
 620       key.drawable_height = ctx->DrawBuffer->Height;
 621    }
 622
 623    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 624          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 625          BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 626
 627    if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 628       key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer) ||
 629                           key.nr_color_regions > 1;
 630    }
 631
 632    key.program_string_id = bfp->id;
 633
 634    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 635    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 636
 637    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 638
 639    brw->wm.base.prog_offset = old_prog_offset;
 640    brw->wm.prog_data = old_prog_data;
 641
 642    return success;
 643 }