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                             true, brw->use_rep_send,
 141                             &program_size, &error_str);
 142    if (program == NULL) {
 143       if (prog) {
 144          prog->LinkStatus = false;
 145          ralloc_strcat(&prog->InfoLog, error_str);
 146       }
 147
 148       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 149
 150       ralloc_free(mem_ctx);
 151       return false;
 152    }
 153
 154    if (unlikely(brw->perf_debug) && fs) {
 155       if (fs->compiled_once)
 156          brw_wm_debug_recompile(brw, prog, key);
 157       fs->compiled_once = true;
 158
 159       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 160          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 161                     (get_time() - start_time) * 1000);
 162       }
 163    }
 164
 165    if (prog_data.base.total_scratch) {
 166       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 167                          prog_data.base.total_scratch * brw->max_wm_threads);
 168    }
 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    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 211       found |= key_debug(brw, "textureGather workarounds",
 212                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 213    }
 214
 215    return found;
 216 }
 217
 218 void
 219 brw_wm_debug_recompile(struct brw_context *brw,
 220                        struct gl_shader_program *prog,
 221                        const struct brw_wm_prog_key *key)
 222 {
 223    struct brw_cache_item *c = NULL;
 224    const struct brw_wm_prog_key *old_key = NULL;
 225    bool found = false;
 226
 227    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 228
 229    for (unsigned int i = 0; i < brw->cache.size; i++) {
 230       for (c = brw->cache.items[i]; c; c = c->next) {
 231          if (c->cache_id == BRW_CACHE_FS_PROG) {
 232             old_key = c->key;
 233
 234             if (old_key->program_string_id == key->program_string_id)
 235                break;
 236          }
 237       }
 238       if (c)
 239          break;
 240    }
 241
 242    if (!c) {
 243       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 244       return;
 245    }
 246
 247    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 248                       "depth write",
 249                       old_key->iz_lookup, key->iz_lookup);
 250    found |= key_debug(brw, "depth statistics",
 251                       old_key->stats_wm, key->stats_wm);
 252    found |= key_debug(brw, "flat shading",
 253                       old_key->flat_shade, key->flat_shade);
 254    found |= key_debug(brw, "per-sample interpolation",
 255                       old_key->persample_interp, key->persample_interp);
 256    found |= key_debug(brw, "number of color buffers",
 257                       old_key->nr_color_regions, key->nr_color_regions);
 258    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 259                       old_key->replicate_alpha, key->replicate_alpha);
 260    found |= key_debug(brw, "fragment color clamping",
 261                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 262    found |= key_debug(brw, "multisampled FBO",
 263                       old_key->multisample_fbo, key->multisample_fbo);
 264    found |= key_debug(brw, "line smoothing",
 265                       old_key->line_aa, key->line_aa);
 266    found |= key_debug(brw, "input slots valid",
 267                       old_key->input_slots_valid, key->input_slots_valid);
 268    found |= key_debug(brw, "mrt alpha test function",
 269                       old_key->alpha_test_func, key->alpha_test_func);
 270    found |= key_debug(brw, "mrt alpha test reference value",
 271                       old_key->alpha_test_ref, key->alpha_test_ref);
 272
 273    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 274
 275    if (!found) {
 276       perf_debug("  Something else\n");
 277    }
 278 }
 279
 280 static uint8_t
 281 gen6_gather_workaround(GLenum internalformat)
 282 {
 283    switch (internalformat) {
 284    case GL_R8I: return WA_SIGN | WA_8BIT;
 285    case GL_R8UI: return WA_8BIT;
 286    case GL_R16I: return WA_SIGN | WA_16BIT;
 287    case GL_R16UI: return WA_16BIT;
 288    default:
 289       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 290        * the surface state, there is no shader w/a required.
 291        */
 292       return 0;
 293    }
 294 }
 295
 296 void
 297 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 298                                    const struct gl_program *prog,
 299                                    unsigned sampler_count,
 300                                    struct brw_sampler_prog_key_data *key)
 301 {
 302    struct brw_context *brw = brw_context(ctx);
 303
 304    for (int s = 0; s < sampler_count; s++) {
 305       key->swizzles[s] = SWIZZLE_NOOP;
 306
 307       if (!(prog->SamplersUsed & (1 << s)))
 308          continue;
 309
 310       int unit_id = prog->SamplerUnits[s];
 311       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 312
 313       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 314          const struct gl_texture_object *t = unit->_Current;
 315          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 316          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 317
 318          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 319             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 320              img->_BaseFormat == GL_DEPTH_STENCIL);
 321
 322          /* Haswell handles texture swizzling as surface format overrides
 323           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 324           */
 325          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 326             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 327
 328          if (brw->gen < 8 &&
 329              sampler->MinFilter != GL_NEAREST &&
 330              sampler->MagFilter != GL_NEAREST) {
 331             if (sampler->WrapS == GL_CLAMP)
 332                key->gl_clamp_mask[0] |= 1 << s;
 333             if (sampler->WrapT == GL_CLAMP)
 334                key->gl_clamp_mask[1] |= 1 << s;
 335             if (sampler->WrapR == GL_CLAMP)
 336                key->gl_clamp_mask[2] |= 1 << s;
 337          }
 338
 339          /* gather4's channel select for green from RG32F is broken; requires
 340           * a shader w/a on IVB; fixable with just SCS on HSW.
 341           */
 342          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 343             if (img->InternalFormat == GL_RG32F)
 344                key->gather_channel_quirk_mask |= 1 << s;
 345          }
 346
 347          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 348           * UNORM/FLOAT instead and fix it in the shader.
 349           */
 350          if (brw->gen == 6 && prog->UsesGather) {
 351             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 352          }
 353
 354          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 355           * then we need to emit slightly different code to first sample the
 356           * MCS surface.
 357           */
 358          struct intel_texture_object *intel_tex =
 359             intel_texture_object((struct gl_texture_object *)t);
 360
 361          /* From gen9 onwards some single sampled buffers can also be
 362           * compressed. These don't need ld2dms sampling along with mcs fetch.
 363           */
 364          if (brw->gen >= 7 &&
 365              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS &&
 366              intel_tex->mt->num_samples > 1) {
 367             key->compressed_multisample_layout_mask |= 1 << s;
 368
 369             if (intel_tex->mt->num_samples >= 16) {
 370                assert(brw->gen >= 9);
 371                key->msaa_16 |= 1 << s;
 372             }
 373          }
 374       }
 375    }
 376 }
 377
 378 static bool
 379 brw_wm_state_dirty(const struct brw_context *brw)
 380 {
 381    return brw_state_dirty(brw,
 382                           _NEW_BUFFERS |
 383                           _NEW_COLOR |
 384                           _NEW_DEPTH |
 385                           _NEW_FRAG_CLAMP |
 386                           _NEW_HINT |
 387                           _NEW_LIGHT |
 388                           _NEW_LINE |
 389                           _NEW_MULTISAMPLE |
 390                           _NEW_POLYGON |
 391                           _NEW_STENCIL |
 392                           _NEW_TEXTURE,
 393                           BRW_NEW_FRAGMENT_PROGRAM |
 394                           BRW_NEW_REDUCED_PRIMITIVE |
 395                           BRW_NEW_STATS_WM |
 396                           BRW_NEW_VUE_MAP_GEOM_OUT);
 397 }
 398
 399 static void
 400 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 401 {
 402    struct gl_context *ctx = &brw->ctx;
 403    /* BRW_NEW_FRAGMENT_PROGRAM */
 404    const struct brw_fragment_program *fp =
 405       (struct brw_fragment_program *) brw->fragment_program;
 406    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 407    GLuint lookup = 0;
 408    GLuint line_aa;
 409
 410    memset(key, 0, sizeof(*key));
 411
 412    /* Build the index for table lookup
 413     */
 414    if (brw->gen < 6) {
 415       /* _NEW_COLOR */
 416       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 417          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 418
 419       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 420          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 421
 422       /* _NEW_DEPTH */
 423       if (ctx->Depth.Test)
 424          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 425
 426       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 427          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 428
 429       /* _NEW_STENCIL | _NEW_BUFFERS */
 430       if (ctx->Stencil._Enabled) {
 431          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 432
 433          if (ctx->Stencil.WriteMask[0] ||
 434              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 435             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 436       }
 437       key->iz_lookup = lookup;
 438    }
 439
 440    line_aa = AA_NEVER;
 441
 442    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 443    if (ctx->Line.SmoothFlag) {
 444       if (brw->reduced_primitive == GL_LINES) {
 445          line_aa = AA_ALWAYS;
 446       }
 447       else if (brw->reduced_primitive == GL_TRIANGLES) {
 448          if (ctx->Polygon.FrontMode == GL_LINE) {
 449             line_aa = AA_SOMETIMES;
 450
 451             if (ctx->Polygon.BackMode == GL_LINE ||
 452                 (ctx->Polygon.CullFlag &&
 453                  ctx->Polygon.CullFaceMode == GL_BACK))
 454                line_aa = AA_ALWAYS;
 455          }
 456          else if (ctx->Polygon.BackMode == GL_LINE) {
 457             line_aa = AA_SOMETIMES;
 458
 459             if ((ctx->Polygon.CullFlag &&
 460                  ctx->Polygon.CullFaceMode == GL_FRONT))
 461                line_aa = AA_ALWAYS;
 462          }
 463       }
 464    }
 465
 466    key->line_aa = line_aa;
 467
 468    /* _NEW_HINT */
 469    key->high_quality_derivatives =
 470       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 471
 472    if (brw->gen < 6)
 473       key->stats_wm = brw->stats_wm;
 474
 475    /* _NEW_LIGHT */
 476    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 477
 478    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 479    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 480
 481    /* _NEW_TEXTURE */
 482    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 483                                       &key->tex);
 484
 485    /* _NEW_BUFFERS */
 486    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 487
 488    /* _NEW_COLOR */
 489    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 490       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 491
 492    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 493    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 494       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 495
 496    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 497    /* Ignore sample qualifier while computing this flag. */
 498    if (ctx->Multisample.Enabled) {
 499       key->persample_interp =
 500          ctx->Multisample.SampleShading &&
 501          (ctx->Multisample.MinSampleShadingValue *
 502           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 503
 504       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 505    }
 506
 507    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 508    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 509                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 510       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 511
 512
 513    /* _NEW_COLOR | _NEW_BUFFERS */
 514    /* Pre-gen6, the hardware alpha test always used each render
 515     * target's alpha to do alpha test, as opposed to render target 0's alpha
 516     * like GL requires.  Fix that by building the alpha test into the
 517     * shader, and we'll skip enabling the fixed function alpha test.
 518     */
 519    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 520        ctx->Color.AlphaEnabled) {
 521       key->alpha_test_func = ctx->Color.AlphaFunc;
 522       key->alpha_test_ref = ctx->Color.AlphaRef;
 523    }
 524
 525    /* The unique fragment program ID */
 526    key->program_string_id = fp->id;
 527 }
 528
 529 void
 530 brw_upload_wm_prog(struct brw_context *brw)
 531 {
 532    struct gl_context *ctx = &brw->ctx;
 533    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 534    struct brw_wm_prog_key key;
 535    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 536       brw->fragment_program;
 537
 538    if (!brw_wm_state_dirty(brw))
 539       return;
 540
 541    brw_wm_populate_key(brw, &key);
 542
 543    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 544                          &key, sizeof(key),
 545                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 546       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 547       (void) success;
 548       assert(success);
 549    }
 550    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 551 }
 552
 553 bool
 554 brw_fs_precompile(struct gl_context *ctx,
 555                   struct gl_shader_program *shader_prog,
 556                   struct gl_program *prog)
 557 {
 558    struct brw_context *brw = brw_context(ctx);
 559    struct brw_wm_prog_key key;
 560
 561    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 562    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 563
 564    memset(&key, 0, sizeof(key));
 565
 566    if (brw->gen < 6) {
 567       if (fp->UsesKill)
 568          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 569
 570       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 571          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 572
 573       /* Just assume depth testing. */
 574       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 575       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 576    }
 577
 578    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 579                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 580       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 581
 582    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 583
 584    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 585          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 586          BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 587
 588    key.program_string_id = bfp->id;
 589
 590    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 591    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 592
 593    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 594
 595    brw->wm.base.prog_offset = old_prog_offset;
 596    brw->wm.prog_data = old_prog_data;
 597
 598    return success;
 599 }