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
  40 #include "util/ralloc.h"
  41
  42 static void
  43 assign_fs_binding_table_offsets(const struct brw_device_info *devinfo,
  44                                 const struct gl_shader_program *shader_prog,
  45                                 const struct gl_program *prog,
  46                                 const struct brw_wm_prog_key *key,
  47                                 struct brw_wm_prog_data *prog_data)
  48 {
  49    uint32_t next_binding_table_offset = 0;
  50
  51    /* If there are no color regions, we still perform an FB write to a null
  52     * renderbuffer, which we place at surface index 0.
  53     */
  54    prog_data->binding_table.render_target_start = next_binding_table_offset;
  55    next_binding_table_offset += MAX2(key->nr_color_regions, 1);
  56
  57    brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
  58                                            shader_prog, prog, &prog_data->base,
  59                                            next_binding_table_offset);
  60 }
  61
  62 /**
  63  * All Mesa program -> GPU code generation goes through this function.
  64  * Depending on the instructions used (i.e. flow control instructions)
  65  * we'll use one of two code generators.
  66  */
  67 bool
  68 brw_codegen_wm_prog(struct brw_context *brw,
  69                     struct gl_shader_program *prog,
  70                     struct brw_fragment_program *fp,
  71                     struct brw_wm_prog_key *key)
  72 {
  73    struct gl_context *ctx = &brw->ctx;
  74    void *mem_ctx = ralloc_context(NULL);
  75    struct brw_wm_prog_data prog_data;
  76    const GLuint *program;
  77    struct brw_shader *fs = NULL;
  78    GLuint program_size;
  79    bool start_busy = false;
  80    double start_time = 0;
  81
  82    if (prog)
  83       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  84
  85    memset(&prog_data, 0, sizeof(prog_data));
  86
  87    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
  88    if (!prog)
  89       prog_data.base.use_alt_mode = true;
  90
  91    assign_fs_binding_table_offsets(brw->intelScreen->devinfo, prog,
  92                                    &fp->program.Base, key, &prog_data);
  93
  94    /* Allocate the references to the uniforms that will end up in the
  95     * prog_data associated with the compiled program, and which will be freed
  96     * by the state cache.
  97     */
  98    int param_count = fp->program.Base.nir->num_uniforms;
  99    if (fs)
 100       prog_data.base.nr_image_params = fs->base.NumImages;
 101    /* The backend also sometimes adds params for texture size. */
 102    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 103    prog_data.base.param =
 104       rzalloc_array(NULL, const gl_constant_value *, param_count);
 105    prog_data.base.pull_param =
 106       rzalloc_array(NULL, const gl_constant_value *, param_count);
 107    prog_data.base.image_param =
 108       rzalloc_array(NULL, struct brw_image_param,
 109                     prog_data.base.nr_image_params);
 110    prog_data.base.nr_params = param_count;
 111
 112    if (prog) {
 113       brw_nir_setup_glsl_uniforms(fp->program.Base.nir, prog, &fp->program.Base,
 114                                   &prog_data.base, true);
 115    } else {
 116       brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base,
 117                                  &prog_data.base);
 118    }
 119
 120    if (unlikely(brw->perf_debug)) {
 121       start_busy = (brw->batch.last_bo &&
 122                     drm_intel_bo_busy(brw->batch.last_bo));
 123       start_time = get_time();
 124    }
 125
 126    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 127       brw_dump_ir("fragment", prog, fs ? &fs->base : NULL, &fp->program.Base);
 128
 129    int st_index8 = -1, st_index16 = -1;
 130    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 131       st_index8 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS8);
 132       st_index16 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS16);
 133    }
 134
 135    char *error_str = NULL;
 136    program = brw_wm_fs_emit(brw->intelScreen->compiler, brw, mem_ctx,
 137                             key, &prog_data, fp->program.Base.nir,
 138                             &fp->program.Base, st_index8, st_index16,
 139                             brw->use_rep_send, &program_size, &error_str);
 140    if (program == NULL) {
 141       if (prog) {
 142          prog->LinkStatus = false;
 143          ralloc_strcat(&prog->InfoLog, error_str);
 144       }
 145
 146       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 147
 148       ralloc_free(mem_ctx);
 149       return false;
 150    }
 151
 152    if (unlikely(brw->perf_debug) && fs) {
 153       if (fs->compiled_once)
 154          brw_wm_debug_recompile(brw, prog, key);
 155       fs->compiled_once = true;
 156
 157       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 158          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 159                     (get_time() - start_time) * 1000);
 160       }
 161    }
 162
 163    if (prog_data.base.total_scratch) {
 164       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 165                          prog_data.base.total_scratch * brw->max_wm_threads);
 166    }
 167
 168    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 169       fprintf(stderr, "\n");
 170
 171    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 172                     key, sizeof(struct brw_wm_prog_key),
 173                     program, program_size,
 174                     &prog_data, sizeof(prog_data),
 175                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 176
 177    ralloc_free(mem_ctx);
 178
 179    return true;
 180 }
 181
 182 static bool
 183 key_debug(struct brw_context *brw, const char *name, int a, int b)
 184 {
 185    if (a != b) {
 186       perf_debug("  %s %d->%d\n", name, a, b);
 187       return true;
 188    } else {
 189       return false;
 190    }
 191 }
 192
 193 bool
 194 brw_debug_recompile_sampler_key(struct brw_context *brw,
 195                                 const struct brw_sampler_prog_key_data *old_key,
 196                                 const struct brw_sampler_prog_key_data *key)
 197 {
 198    bool found = false;
 199
 200    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 201       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 202                          old_key->swizzles[i], key->swizzles[i]);
 203    }
 204    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 205                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 206    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 207                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 208    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 209                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 210    found |= key_debug(brw, "gather channel quirk on any texture unit",
 211                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 212    found |= key_debug(brw, "compressed multisample layout",
 213                       old_key->compressed_multisample_layout_mask,
 214                       key->compressed_multisample_layout_mask);
 215
 216    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 217       found |= key_debug(brw, "textureGather workarounds",
 218                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 219    }
 220
 221    return found;
 222 }
 223
 224 void
 225 brw_wm_debug_recompile(struct brw_context *brw,
 226                        struct gl_shader_program *prog,
 227                        const struct brw_wm_prog_key *key)
 228 {
 229    struct brw_cache_item *c = NULL;
 230    const struct brw_wm_prog_key *old_key = NULL;
 231    bool found = false;
 232
 233    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 234
 235    for (unsigned int i = 0; i < brw->cache.size; i++) {
 236       for (c = brw->cache.items[i]; c; c = c->next) {
 237          if (c->cache_id == BRW_CACHE_FS_PROG) {
 238             old_key = c->key;
 239
 240             if (old_key->program_string_id == key->program_string_id)
 241                break;
 242          }
 243       }
 244       if (c)
 245          break;
 246    }
 247
 248    if (!c) {
 249       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 250       return;
 251    }
 252
 253    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 254                       "depth write",
 255                       old_key->iz_lookup, key->iz_lookup);
 256    found |= key_debug(brw, "depth statistics",
 257                       old_key->stats_wm, key->stats_wm);
 258    found |= key_debug(brw, "flat shading",
 259                       old_key->flat_shade, key->flat_shade);
 260    found |= key_debug(brw, "per-sample shading",
 261                       old_key->persample_shading, key->persample_shading);
 262    found |= key_debug(brw, "per-sample shading and 2x MSAA",
 263                       old_key->persample_2x, key->persample_2x);
 264    found |= key_debug(brw, "number of color buffers",
 265                       old_key->nr_color_regions, key->nr_color_regions);
 266    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 267                       old_key->replicate_alpha, key->replicate_alpha);
 268    found |= key_debug(brw, "rendering to FBO",
 269                       old_key->render_to_fbo, key->render_to_fbo);
 270    found |= key_debug(brw, "fragment color clamping",
 271                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 272    found |= key_debug(brw, "line smoothing",
 273                       old_key->line_aa, key->line_aa);
 274    found |= key_debug(brw, "renderbuffer height",
 275                       old_key->drawable_height, key->drawable_height);
 276    found |= key_debug(brw, "input slots valid",
 277                       old_key->input_slots_valid, key->input_slots_valid);
 278    found |= key_debug(brw, "mrt alpha test function",
 279                       old_key->alpha_test_func, key->alpha_test_func);
 280    found |= key_debug(brw, "mrt alpha test reference value",
 281                       old_key->alpha_test_ref, key->alpha_test_ref);
 282
 283    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 284
 285    if (!found) {
 286       perf_debug("  Something else\n");
 287    }
 288 }
 289
 290 static uint8_t
 291 gen6_gather_workaround(GLenum internalformat)
 292 {
 293    switch (internalformat) {
 294    case GL_R8I: return WA_SIGN | WA_8BIT;
 295    case GL_R8UI: return WA_8BIT;
 296    case GL_R16I: return WA_SIGN | WA_16BIT;
 297    case GL_R16UI: return WA_16BIT;
 298    default:
 299       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 300        * the surface state, there is no shader w/a required.
 301        */
 302       return 0;
 303    }
 304 }
 305
 306 void
 307 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 308                                    const struct gl_program *prog,
 309                                    unsigned sampler_count,
 310                                    struct brw_sampler_prog_key_data *key)
 311 {
 312    struct brw_context *brw = brw_context(ctx);
 313
 314    for (int s = 0; s < sampler_count; s++) {
 315       key->swizzles[s] = SWIZZLE_NOOP;
 316
 317       if (!(prog->SamplersUsed & (1 << s)))
 318          continue;
 319
 320       int unit_id = prog->SamplerUnits[s];
 321       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 322
 323       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 324          const struct gl_texture_object *t = unit->_Current;
 325          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 326          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 327
 328          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 329             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 330              img->_BaseFormat == GL_DEPTH_STENCIL);
 331
 332          /* Haswell handles texture swizzling as surface format overrides
 333           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 334           */
 335          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 336             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 337
 338          if (brw->gen < 8 &&
 339              sampler->MinFilter != GL_NEAREST &&
 340              sampler->MagFilter != GL_NEAREST) {
 341             if (sampler->WrapS == GL_CLAMP)
 342                key->gl_clamp_mask[0] |= 1 << s;
 343             if (sampler->WrapT == GL_CLAMP)
 344                key->gl_clamp_mask[1] |= 1 << s;
 345             if (sampler->WrapR == GL_CLAMP)
 346                key->gl_clamp_mask[2] |= 1 << s;
 347          }
 348
 349          /* gather4's channel select for green from RG32F is broken; requires
 350           * a shader w/a on IVB; fixable with just SCS on HSW.
 351           */
 352          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 353             if (img->InternalFormat == GL_RG32F)
 354                key->gather_channel_quirk_mask |= 1 << s;
 355          }
 356
 357          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 358           * UNORM/FLOAT instead and fix it in the shader.
 359           */
 360          if (brw->gen == 6 && prog->UsesGather) {
 361             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 362          }
 363
 364          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 365           * then we need to emit slightly different code to first sample the
 366           * MCS surface.
 367           */
 368          struct intel_texture_object *intel_tex =
 369             intel_texture_object((struct gl_texture_object *)t);
 370
 371          if (brw->gen >= 7 &&
 372              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
 373             key->compressed_multisample_layout_mask |= 1 << s;
 374          }
 375       }
 376    }
 377 }
 378
 379 static bool
 380 brw_wm_state_dirty (struct brw_context *brw)
 381 {
 382    return brw_state_dirty(brw,
 383                           _NEW_BUFFERS |
 384                           _NEW_COLOR |
 385                           _NEW_DEPTH |
 386                           _NEW_FRAG_CLAMP |
 387                           _NEW_HINT |
 388                           _NEW_LIGHT |
 389                           _NEW_LINE |
 390                           _NEW_MULTISAMPLE |
 391                           _NEW_POLYGON |
 392                           _NEW_STENCIL |
 393                           _NEW_TEXTURE,
 394                           BRW_NEW_FRAGMENT_PROGRAM |
 395                           BRW_NEW_REDUCED_PRIMITIVE |
 396                           BRW_NEW_STATS_WM |
 397                           BRW_NEW_VUE_MAP_GEOM_OUT);
 398 }
 399
 400 static void
 401 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 402 {
 403    struct gl_context *ctx = &brw->ctx;
 404    /* BRW_NEW_FRAGMENT_PROGRAM */
 405    const struct brw_fragment_program *fp =
 406       (struct brw_fragment_program *) brw->fragment_program;
 407    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 408    GLuint lookup = 0;
 409    GLuint line_aa;
 410    bool program_uses_dfdy = fp->program.UsesDFdy;
 411    const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 412
 413    memset(key, 0, sizeof(*key));
 414
 415    /* Build the index for table lookup
 416     */
 417    if (brw->gen < 6) {
 418       /* _NEW_COLOR */
 419       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 420          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 421
 422       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 423          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 424
 425       /* _NEW_DEPTH */
 426       if (ctx->Depth.Test)
 427          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 428
 429       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 430          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 431
 432       /* _NEW_STENCIL | _NEW_BUFFERS */
 433       if (ctx->Stencil._Enabled) {
 434          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 435
 436          if (ctx->Stencil.WriteMask[0] ||
 437              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 438             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 439       }
 440       key->iz_lookup = lookup;
 441    }
 442
 443    line_aa = AA_NEVER;
 444
 445    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 446    if (ctx->Line.SmoothFlag) {
 447       if (brw->reduced_primitive == GL_LINES) {
 448          line_aa = AA_ALWAYS;
 449       }
 450       else if (brw->reduced_primitive == GL_TRIANGLES) {
 451          if (ctx->Polygon.FrontMode == GL_LINE) {
 452             line_aa = AA_SOMETIMES;
 453
 454             if (ctx->Polygon.BackMode == GL_LINE ||
 455                 (ctx->Polygon.CullFlag &&
 456                  ctx->Polygon.CullFaceMode == GL_BACK))
 457                line_aa = AA_ALWAYS;
 458          }
 459          else if (ctx->Polygon.BackMode == GL_LINE) {
 460             line_aa = AA_SOMETIMES;
 461
 462             if ((ctx->Polygon.CullFlag &&
 463                  ctx->Polygon.CullFaceMode == GL_FRONT))
 464                line_aa = AA_ALWAYS;
 465          }
 466       }
 467    }
 468
 469    key->line_aa = line_aa;
 470
 471    /* _NEW_HINT */
 472    key->high_quality_derivatives =
 473       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 474
 475    if (brw->gen < 6)
 476       key->stats_wm = brw->stats_wm;
 477
 478    /* _NEW_LIGHT */
 479    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 480
 481    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 482    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 483
 484    /* _NEW_TEXTURE */
 485    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 486                                       &key->tex);
 487
 488    /* _NEW_BUFFERS */
 489    /*
 490     * Include the draw buffer origin and height so that we can calculate
 491     * fragment position values relative to the bottom left of the drawable,
 492     * from the incoming screen origin relative position we get as part of our
 493     * payload.
 494     *
 495     * This is only needed for the WM_WPOSXY opcode when the fragment program
 496     * uses the gl_FragCoord input.
 497     *
 498     * We could avoid recompiling by including this as a constant referenced by
 499     * our program, but if we were to do that it would also be nice to handle
 500     * getting that constant updated at batchbuffer submit time (when we
 501     * hold the lock and know where the buffer really is) rather than at emit
 502     * time when we don't hold the lock and are just guessing.  We could also
 503     * just avoid using this as key data if the program doesn't use
 504     * fragment.position.
 505     *
 506     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 507     * drawable height in order to invert the Y axis.
 508     */
 509    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 510       key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
 511    }
 512
 513    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 514       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 515    }
 516
 517    /* _NEW_BUFFERS */
 518    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 519
 520    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 521    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 522       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 523
 524    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 525    /* Ignore sample qualifier while computing this flag. */
 526    key->persample_shading =
 527       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
 528    if (key->persample_shading)
 529       key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
 530
 531    key->compute_pos_offset =
 532       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
 533       fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
 534
 535    key->compute_sample_id =
 536       multisample_fbo &&
 537       ctx->Multisample.Enabled &&
 538       (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
 539
 540    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 541    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 542                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 543       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 544
 545
 546    /* _NEW_COLOR | _NEW_BUFFERS */
 547    /* Pre-gen6, the hardware alpha test always used each render
 548     * target's alpha to do alpha test, as opposed to render target 0's alpha
 549     * like GL requires.  Fix that by building the alpha test into the
 550     * shader, and we'll skip enabling the fixed function alpha test.
 551     */
 552    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 553        ctx->Color.AlphaEnabled) {
 554       key->alpha_test_func = ctx->Color.AlphaFunc;
 555       key->alpha_test_ref = ctx->Color.AlphaRef;
 556    }
 557
 558    /* The unique fragment program ID */
 559    key->program_string_id = fp->id;
 560 }
 561
 562 void
 563 brw_upload_wm_prog(struct brw_context *brw)
 564 {
 565    struct gl_context *ctx = &brw->ctx;
 566    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 567    struct brw_wm_prog_key key;
 568    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 569       brw->fragment_program;
 570
 571    if (!brw_wm_state_dirty(brw))
 572       return;
 573
 574    brw_wm_populate_key(brw, &key);
 575
 576    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 577                          &key, sizeof(key),
 578                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 579       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 580       (void) success;
 581       assert(success);
 582    }
 583    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 584 }
 585
 586 bool
 587 brw_fs_precompile(struct gl_context *ctx,
 588                   struct gl_shader_program *shader_prog,
 589                   struct gl_program *prog)
 590 {
 591    struct brw_context *brw = brw_context(ctx);
 592    struct brw_wm_prog_key key;
 593
 594    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 595    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 596    bool program_uses_dfdy = fp->UsesDFdy;
 597
 598    memset(&key, 0, sizeof(key));
 599
 600    if (brw->gen < 6) {
 601       if (fp->UsesKill)
 602          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 603
 604       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 605          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 606
 607       /* Just assume depth testing. */
 608       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 609       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 610    }
 611
 612    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 613                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 614       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 615
 616    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 617
 618    if (fp->Base.InputsRead & VARYING_BIT_POS) {
 619       key.drawable_height = ctx->DrawBuffer->Height;
 620    }
 621
 622    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 623          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 624          BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 625
 626    if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 627       key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer) ||
 628                           key.nr_color_regions > 1;
 629    }
 630
 631    key.program_string_id = bfp->id;
 632
 633    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 634    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 635
 636    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 637
 638    brw->wm.base.prog_offset = old_prog_offset;
 639    brw->wm.prog_data = old_prog_data;
 640
 641    return success;
 642 }