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 "intel_image.h"
  39 #include "brw_nir.h"
  40 #include "brw_program.h"
  41
  42 #include "util/ralloc.h"
  43
  44 static void
  45 assign_fs_binding_table_offsets(const struct brw_device_info *devinfo,
  46                                 const struct gl_shader_program *shader_prog,
  47                                 const struct gl_program *prog,
  48                                 const struct brw_wm_prog_key *key,
  49                                 struct brw_wm_prog_data *prog_data)
  50 {
  51    uint32_t next_binding_table_offset = 0;
  52
  53    /* If there are no color regions, we still perform an FB write to a null
  54     * renderbuffer, which we place at surface index 0.
  55     */
  56    prog_data->binding_table.render_target_start = next_binding_table_offset;
  57    next_binding_table_offset += MAX2(key->nr_color_regions, 1);
  58
  59    brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
  60                                            shader_prog, prog, &prog_data->base,
  61                                            next_binding_table_offset);
  62 }
  63
  64 /**
  65  * All Mesa program -> GPU code generation goes through this function.
  66  * Depending on the instructions used (i.e. flow control instructions)
  67  * we'll use one of two code generators.
  68  */
  69 bool
  70 brw_codegen_wm_prog(struct brw_context *brw,
  71                     struct gl_shader_program *prog,
  72                     struct brw_fragment_program *fp,
  73                     struct brw_wm_prog_key *key)
  74 {
  75    struct gl_context *ctx = &brw->ctx;
  76    void *mem_ctx = ralloc_context(NULL);
  77    struct brw_wm_prog_data prog_data;
  78    const GLuint *program;
  79    struct brw_shader *fs = NULL;
  80    GLuint program_size;
  81    bool start_busy = false;
  82    double start_time = 0;
  83
  84    if (prog)
  85       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  86
  87    memset(&prog_data, 0, sizeof(prog_data));
  88
  89    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
  90    if (!prog)
  91       prog_data.base.use_alt_mode = true;
  92
  93    assign_fs_binding_table_offsets(brw->intelScreen->devinfo, prog,
  94                                    &fp->program.Base, key, &prog_data);
  95
  96    /* Allocate the references to the uniforms that will end up in the
  97     * prog_data associated with the compiled program, and which will be freed
  98     * by the state cache.
  99     */
 100    int param_count = fp->program.Base.nir->num_uniforms / 4;
 101    if (fs)
 102       prog_data.base.nr_image_params = fs->base.NumImages;
 103    /* The backend also sometimes adds params for texture size. */
 104    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 105    prog_data.base.param =
 106       rzalloc_array(NULL, const gl_constant_value *, param_count);
 107    prog_data.base.pull_param =
 108       rzalloc_array(NULL, const gl_constant_value *, param_count);
 109    prog_data.base.image_param =
 110       rzalloc_array(NULL, struct brw_image_param,
 111                     prog_data.base.nr_image_params);
 112    prog_data.base.nr_params = param_count;
 113
 114    if (prog) {
 115       brw_nir_setup_glsl_uniforms(fp->program.Base.nir, prog, &fp->program.Base,
 116                                   &prog_data.base, true);
 117    } else {
 118       brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base,
 119                                  &prog_data.base);
 120    }
 121
 122    if (unlikely(brw->perf_debug)) {
 123       start_busy = (brw->batch.last_bo &&
 124                     drm_intel_bo_busy(brw->batch.last_bo));
 125       start_time = get_time();
 126    }
 127
 128    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 129       brw_dump_ir("fragment", prog, fs ? &fs->base : NULL, &fp->program.Base);
 130
 131    int st_index8 = -1, st_index16 = -1;
 132    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 133       st_index8 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS8);
 134       st_index16 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS16);
 135    }
 136
 137    char *error_str = NULL;
 138    program = brw_compile_fs(brw->intelScreen->compiler, brw, mem_ctx,
 139                             key, &prog_data, fp->program.Base.nir,
 140                             &fp->program.Base, st_index8, st_index16,
 141                             true, brw->use_rep_send,
 142                             &program_size, &error_str);
 143    if (program == NULL) {
 144       if (prog) {
 145          prog->LinkStatus = false;
 146          ralloc_strcat(&prog->InfoLog, error_str);
 147       }
 148
 149       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 150
 151       ralloc_free(mem_ctx);
 152       return false;
 153    }
 154
 155    if (unlikely(brw->perf_debug) && fs) {
 156       if (fs->compiled_once)
 157          brw_wm_debug_recompile(brw, prog, key);
 158       fs->compiled_once = true;
 159
 160       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 161          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 162                     (get_time() - start_time) * 1000);
 163       }
 164    }
 165
 166    brw_alloc_stage_scratch(brw, &brw->wm.base,
 167                            prog_data.base.total_scratch,
 168                            brw->max_wm_threads);
 169
 170    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 171       fprintf(stderr, "\n");
 172
 173    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 174                     key, sizeof(struct brw_wm_prog_key),
 175                     program, program_size,
 176                     &prog_data, sizeof(prog_data),
 177                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 178
 179    ralloc_free(mem_ctx);
 180
 181    return true;
 182 }
 183
 184 bool
 185 brw_debug_recompile_sampler_key(struct brw_context *brw,
 186                                 const struct brw_sampler_prog_key_data *old_key,
 187                                 const struct brw_sampler_prog_key_data *key)
 188 {
 189    bool found = false;
 190
 191    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 192       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 193                          old_key->swizzles[i], key->swizzles[i]);
 194    }
 195    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 196                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 197    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 198                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 199    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 200                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 201    found |= key_debug(brw, "gather channel quirk on any texture unit",
 202                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 203    found |= key_debug(brw, "compressed multisample layout",
 204                       old_key->compressed_multisample_layout_mask,
 205                       key->compressed_multisample_layout_mask);
 206    found |= key_debug(brw, "16x msaa",
 207                       old_key->msaa_16,
 208                       key->msaa_16);
 209
 210    found |= key_debug(brw, "y_uv image bound",
 211                       old_key->y_uv_image_mask,
 212                       key->y_uv_image_mask);
 213    found |= key_debug(brw, "y_u_v image bound",
 214                       old_key->y_u_v_image_mask,
 215                       key->y_u_v_image_mask);
 216    found |= key_debug(brw, "yx_xuxv image bound",
 217                       old_key->yx_xuxv_image_mask,
 218                       key->yx_xuxv_image_mask);
 219
 220    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 221       found |= key_debug(brw, "textureGather workarounds",
 222                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 223    }
 224
 225    return found;
 226 }
 227
 228 void
 229 brw_wm_debug_recompile(struct brw_context *brw,
 230                        struct gl_shader_program *prog,
 231                        const struct brw_wm_prog_key *key)
 232 {
 233    struct brw_cache_item *c = NULL;
 234    const struct brw_wm_prog_key *old_key = NULL;
 235    bool found = false;
 236
 237    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 238
 239    for (unsigned int i = 0; i < brw->cache.size; i++) {
 240       for (c = brw->cache.items[i]; c; c = c->next) {
 241          if (c->cache_id == BRW_CACHE_FS_PROG) {
 242             old_key = c->key;
 243
 244             if (old_key->program_string_id == key->program_string_id)
 245                break;
 246          }
 247       }
 248       if (c)
 249          break;
 250    }
 251
 252    if (!c) {
 253       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 254       return;
 255    }
 256
 257    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 258                       "depth write",
 259                       old_key->iz_lookup, key->iz_lookup);
 260    found |= key_debug(brw, "depth statistics",
 261                       old_key->stats_wm, key->stats_wm);
 262    found |= key_debug(brw, "flat shading",
 263                       old_key->flat_shade, key->flat_shade);
 264    found |= key_debug(brw, "per-sample interpolation",
 265                       old_key->persample_interp, key->persample_interp);
 266    found |= key_debug(brw, "number of color buffers",
 267                       old_key->nr_color_regions, key->nr_color_regions);
 268    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 269                       old_key->replicate_alpha, key->replicate_alpha);
 270    found |= key_debug(brw, "fragment color clamping",
 271                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 272    found |= key_debug(brw, "multisampled FBO",
 273                       old_key->multisample_fbo, key->multisample_fbo);
 274    found |= key_debug(brw, "line smoothing",
 275                       old_key->line_aa, key->line_aa);
 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                                    struct brw_sampler_prog_key_data *key)
 310 {
 311    struct brw_context *brw = brw_context(ctx);
 312    GLbitfield mask = prog->SamplersUsed;
 313
 314    while (mask) {
 315       const int s = u_bit_scan(&mask);
 316
 317       key->swizzles[s] = SWIZZLE_NOOP;
 318
 319       int unit_id = prog->SamplerUnits[s];
 320       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 321
 322       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 323          const struct gl_texture_object *t = unit->_Current;
 324          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 325          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 326
 327          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 328             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 329              img->_BaseFormat == GL_DEPTH_STENCIL);
 330
 331          /* Haswell handles texture swizzling as surface format overrides
 332           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 333           */
 334          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 335             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 336
 337          if (brw->gen < 8 &&
 338              sampler->MinFilter != GL_NEAREST &&
 339              sampler->MagFilter != GL_NEAREST) {
 340             if (sampler->WrapS == GL_CLAMP)
 341                key->gl_clamp_mask[0] |= 1 << s;
 342             if (sampler->WrapT == GL_CLAMP)
 343                key->gl_clamp_mask[1] |= 1 << s;
 344             if (sampler->WrapR == GL_CLAMP)
 345                key->gl_clamp_mask[2] |= 1 << s;
 346          }
 347
 348          /* gather4's channel select for green from RG32F is broken; requires
 349           * a shader w/a on IVB; fixable with just SCS on HSW.
 350           */
 351          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 352             if (img->InternalFormat == GL_RG32F)
 353                key->gather_channel_quirk_mask |= 1 << s;
 354          }
 355
 356          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 357           * UNORM/FLOAT instead and fix it in the shader.
 358           */
 359          if (brw->gen == 6 && prog->UsesGather) {
 360             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 361          }
 362
 363          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 364           * then we need to emit slightly different code to first sample the
 365           * MCS surface.
 366           */
 367          struct intel_texture_object *intel_tex =
 368             intel_texture_object((struct gl_texture_object *)t);
 369
 370          /* From gen9 onwards some single sampled buffers can also be
 371           * compressed. These don't need ld2dms sampling along with mcs fetch.
 372           */
 373          if (brw->gen >= 7 &&
 374              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS &&
 375              intel_tex->mt->num_samples > 1) {
 376             key->compressed_multisample_layout_mask |= 1 << s;
 377
 378             if (intel_tex->mt->num_samples >= 16) {
 379                assert(brw->gen >= 9);
 380                key->msaa_16 |= 1 << s;
 381             }
 382          }
 383
 384          if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
 385             switch (intel_tex->planar_format->components) {
 386             case __DRI_IMAGE_COMPONENTS_Y_UV:
 387                key->y_uv_image_mask |= 1 << s;
 388                break;
 389             case __DRI_IMAGE_COMPONENTS_Y_U_V:
 390                key->y_u_v_image_mask |= 1 << s;
 391                break;
 392             case __DRI_IMAGE_COMPONENTS_Y_XUXV:
 393                key->yx_xuxv_image_mask |= 1 << s;
 394                break;
 395             default:
 396                break;
 397             }
 398          }
 399
 400       }
 401    }
 402 }
 403
 404 static bool
 405 brw_wm_state_dirty(const struct brw_context *brw)
 406 {
 407    return brw_state_dirty(brw,
 408                           _NEW_BUFFERS |
 409                           _NEW_COLOR |
 410                           _NEW_DEPTH |
 411                           _NEW_FRAG_CLAMP |
 412                           _NEW_HINT |
 413                           _NEW_LIGHT |
 414                           _NEW_LINE |
 415                           _NEW_MULTISAMPLE |
 416                           _NEW_POLYGON |
 417                           _NEW_STENCIL |
 418                           _NEW_TEXTURE,
 419                           BRW_NEW_FRAGMENT_PROGRAM |
 420                           BRW_NEW_REDUCED_PRIMITIVE |
 421                           BRW_NEW_STATS_WM |
 422                           BRW_NEW_VUE_MAP_GEOM_OUT);
 423 }
 424
 425 static void
 426 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 427 {
 428    struct gl_context *ctx = &brw->ctx;
 429    /* BRW_NEW_FRAGMENT_PROGRAM */
 430    const struct brw_fragment_program *fp =
 431       (struct brw_fragment_program *) brw->fragment_program;
 432    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 433    GLuint lookup = 0;
 434    GLuint line_aa;
 435
 436    memset(key, 0, sizeof(*key));
 437
 438    /* Build the index for table lookup
 439     */
 440    if (brw->gen < 6) {
 441       /* _NEW_COLOR */
 442       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 443          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 444
 445       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 446          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 447
 448       /* _NEW_DEPTH */
 449       if (ctx->Depth.Test)
 450          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 451
 452       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 453          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 454
 455       /* _NEW_STENCIL | _NEW_BUFFERS */
 456       if (ctx->Stencil._Enabled) {
 457          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 458
 459          if (ctx->Stencil.WriteMask[0] ||
 460              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 461             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 462       }
 463       key->iz_lookup = lookup;
 464    }
 465
 466    line_aa = AA_NEVER;
 467
 468    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 469    if (ctx->Line.SmoothFlag) {
 470       if (brw->reduced_primitive == GL_LINES) {
 471          line_aa = AA_ALWAYS;
 472       }
 473       else if (brw->reduced_primitive == GL_TRIANGLES) {
 474          if (ctx->Polygon.FrontMode == GL_LINE) {
 475             line_aa = AA_SOMETIMES;
 476
 477             if (ctx->Polygon.BackMode == GL_LINE ||
 478                 (ctx->Polygon.CullFlag &&
 479                  ctx->Polygon.CullFaceMode == GL_BACK))
 480                line_aa = AA_ALWAYS;
 481          }
 482          else if (ctx->Polygon.BackMode == GL_LINE) {
 483             line_aa = AA_SOMETIMES;
 484
 485             if ((ctx->Polygon.CullFlag &&
 486                  ctx->Polygon.CullFaceMode == GL_FRONT))
 487                line_aa = AA_ALWAYS;
 488          }
 489       }
 490    }
 491
 492    key->line_aa = line_aa;
 493
 494    /* _NEW_HINT */
 495    key->high_quality_derivatives =
 496       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 497
 498    if (brw->gen < 6)
 499       key->stats_wm = brw->stats_wm;
 500
 501    /* _NEW_LIGHT */
 502    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 503
 504    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 505    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 506
 507    /* _NEW_TEXTURE */
 508    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 509
 510    /* _NEW_BUFFERS */
 511    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 512
 513    /* _NEW_COLOR */
 514    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 515       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 516
 517    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 518    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 519       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 520
 521    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 522    /* Ignore sample qualifier while computing this flag. */
 523    if (ctx->Multisample.Enabled) {
 524       key->persample_interp =
 525          ctx->Multisample.SampleShading &&
 526          (ctx->Multisample.MinSampleShadingValue *
 527           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 528
 529       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 530    }
 531
 532    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 533    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 534                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 535       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 536
 537
 538    /* _NEW_COLOR | _NEW_BUFFERS */
 539    /* Pre-gen6, the hardware alpha test always used each render
 540     * target's alpha to do alpha test, as opposed to render target 0's alpha
 541     * like GL requires.  Fix that by building the alpha test into the
 542     * shader, and we'll skip enabling the fixed function alpha test.
 543     */
 544    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 545        ctx->Color.AlphaEnabled) {
 546       key->alpha_test_func = ctx->Color.AlphaFunc;
 547       key->alpha_test_ref = ctx->Color.AlphaRef;
 548    }
 549
 550    /* The unique fragment program ID */
 551    key->program_string_id = fp->id;
 552 }
 553
 554 void
 555 brw_upload_wm_prog(struct brw_context *brw)
 556 {
 557    struct gl_context *ctx = &brw->ctx;
 558    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 559    struct brw_wm_prog_key key;
 560    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 561       brw->fragment_program;
 562
 563    if (!brw_wm_state_dirty(brw))
 564       return;
 565
 566    brw_wm_populate_key(brw, &key);
 567
 568    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 569                          &key, sizeof(key),
 570                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 571       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 572       (void) success;
 573       assert(success);
 574    }
 575    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 576 }
 577
 578 bool
 579 brw_fs_precompile(struct gl_context *ctx,
 580                   struct gl_shader_program *shader_prog,
 581                   struct gl_program *prog)
 582 {
 583    struct brw_context *brw = brw_context(ctx);
 584    struct brw_wm_prog_key key;
 585
 586    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 587    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 588
 589    memset(&key, 0, sizeof(key));
 590
 591    if (brw->gen < 6) {
 592       if (fp->UsesKill)
 593          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 594
 595       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 596          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 597
 598       /* Just assume depth testing. */
 599       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 600       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 601    }
 602
 603    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 604                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 605       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 606
 607    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 608
 609    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 610          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 611          BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 612
 613    key.program_string_id = bfp->id;
 614
 615    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 616    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 617
 618    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 619
 620    brw->wm.base.prog_offset = old_prog_offset;
 621    brw->wm.prog_data = old_prog_data;
 622
 623    return success;
 624 }