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 gen_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    next_binding_table_offset =
  60       brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
  61                                               shader_prog, prog, &prog_data->base,
  62                                               next_binding_table_offset);
  63
  64    if (prog->nir->info->outputs_read && !key->coherent_fb_fetch) {
  65       prog_data->binding_table.render_target_read_start =
  66          next_binding_table_offset;
  67       next_binding_table_offset += key->nr_color_regions;
  68    }
  69 }
  70
  71 /**
  72  * All Mesa program -> GPU code generation goes through this function.
  73  * Depending on the instructions used (i.e. flow control instructions)
  74  * we'll use one of two code generators.
  75  */
  76 bool
  77 brw_codegen_wm_prog(struct brw_context *brw,
  78                     struct gl_shader_program *prog,
  79                     struct brw_program *fp,
  80                     struct brw_wm_prog_key *key,
  81                     struct brw_vue_map *vue_map)
  82 {
  83    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  84    struct gl_context *ctx = &brw->ctx;
  85    void *mem_ctx = ralloc_context(NULL);
  86    struct brw_wm_prog_data prog_data;
  87    const GLuint *program;
  88    struct brw_shader *fs = NULL;
  89    GLuint program_size;
  90    bool start_busy = false;
  91    double start_time = 0;
  92
  93    if (prog)
  94       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  95
  96    memset(&prog_data, 0, sizeof(prog_data));
  97
  98    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
  99    if (!prog)
 100       prog_data.base.use_alt_mode = true;
 101
 102    assign_fs_binding_table_offsets(devinfo, prog, &fp->program, key,
 103                                    &prog_data);
 104
 105    /* Allocate the references to the uniforms that will end up in the
 106     * prog_data associated with the compiled program, and which will be freed
 107     * by the state cache.
 108     */
 109    int param_count = fp->program.nir->num_uniforms / 4;
 110    if (fs)
 111       prog_data.base.nr_image_params = fs->base.NumImages;
 112    /* The backend also sometimes adds params for texture size. */
 113    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 114    prog_data.base.param =
 115       rzalloc_array(NULL, const gl_constant_value *, param_count);
 116    prog_data.base.pull_param =
 117       rzalloc_array(NULL, const gl_constant_value *, param_count);
 118    prog_data.base.image_param =
 119       rzalloc_array(NULL, struct brw_image_param,
 120                     prog_data.base.nr_image_params);
 121    prog_data.base.nr_params = param_count;
 122
 123    if (prog) {
 124       brw_nir_setup_glsl_uniforms(fp->program.nir, prog, &fp->program,
 125                                   &prog_data.base, true);
 126    } else {
 127       brw_nir_setup_arb_uniforms(fp->program.nir, &fp->program,
 128                                  &prog_data.base);
 129    }
 130
 131    if (unlikely(brw->perf_debug)) {
 132       start_busy = (brw->batch.last_bo &&
 133                     drm_intel_bo_busy(brw->batch.last_bo));
 134       start_time = get_time();
 135    }
 136
 137    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 138       brw_dump_ir("fragment", prog, fs ? &fs->base : NULL, &fp->program);
 139
 140    int st_index8 = -1, st_index16 = -1;
 141    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 142       st_index8 = brw_get_shader_time_index(brw, prog, &fp->program, ST_FS8);
 143       st_index16 = brw_get_shader_time_index(brw, prog, &fp->program, ST_FS16);
 144    }
 145
 146    char *error_str = NULL;
 147    program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx,
 148                             key, &prog_data, fp->program.nir,
 149                             &fp->program, st_index8, st_index16,
 150                             true, brw->use_rep_send, vue_map,
 151                             &program_size, &error_str);
 152
 153    if (program == NULL) {
 154       if (prog) {
 155          prog->LinkStatus = false;
 156          ralloc_strcat(&prog->InfoLog, error_str);
 157       }
 158
 159       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 160
 161       ralloc_free(mem_ctx);
 162       return false;
 163    }
 164
 165    if (unlikely(brw->perf_debug) && fs) {
 166       if (fs->compiled_once)
 167          brw_wm_debug_recompile(brw, prog, key);
 168       fs->compiled_once = true;
 169
 170       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 171          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 172                     (get_time() - start_time) * 1000);
 173       }
 174    }
 175
 176    brw_alloc_stage_scratch(brw, &brw->wm.base,
 177                            prog_data.base.total_scratch,
 178                            devinfo->max_wm_threads);
 179
 180    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 181       fprintf(stderr, "\n");
 182
 183    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 184                     key, sizeof(struct brw_wm_prog_key),
 185                     program, program_size,
 186                     &prog_data, sizeof(prog_data),
 187                     &brw->wm.base.prog_offset, &brw->wm.base.prog_data);
 188
 189    ralloc_free(mem_ctx);
 190
 191    return true;
 192 }
 193
 194 bool
 195 brw_debug_recompile_sampler_key(struct brw_context *brw,
 196                                 const struct brw_sampler_prog_key_data *old_key,
 197                                 const struct brw_sampler_prog_key_data *key)
 198 {
 199    bool found = false;
 200
 201    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 202       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 203                          old_key->swizzles[i], key->swizzles[i]);
 204    }
 205    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 206                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 207    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 208                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 209    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 210                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 211    found |= key_debug(brw, "gather channel quirk on any texture unit",
 212                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 213    found |= key_debug(brw, "compressed multisample layout",
 214                       old_key->compressed_multisample_layout_mask,
 215                       key->compressed_multisample_layout_mask);
 216    found |= key_debug(brw, "16x msaa",
 217                       old_key->msaa_16,
 218                       key->msaa_16);
 219
 220    found |= key_debug(brw, "y_uv image bound",
 221                       old_key->y_uv_image_mask,
 222                       key->y_uv_image_mask);
 223    found |= key_debug(brw, "y_u_v image bound",
 224                       old_key->y_u_v_image_mask,
 225                       key->y_u_v_image_mask);
 226    found |= key_debug(brw, "yx_xuxv image bound",
 227                       old_key->yx_xuxv_image_mask,
 228                       key->yx_xuxv_image_mask);
 229
 230    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 231       found |= key_debug(brw, "textureGather workarounds",
 232                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 233    }
 234
 235    return found;
 236 }
 237
 238 void
 239 brw_wm_debug_recompile(struct brw_context *brw,
 240                        struct gl_shader_program *prog,
 241                        const struct brw_wm_prog_key *key)
 242 {
 243    struct brw_cache_item *c = NULL;
 244    const struct brw_wm_prog_key *old_key = NULL;
 245    bool found = false;
 246
 247    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 248
 249    for (unsigned int i = 0; i < brw->cache.size; i++) {
 250       for (c = brw->cache.items[i]; c; c = c->next) {
 251          if (c->cache_id == BRW_CACHE_FS_PROG) {
 252             old_key = c->key;
 253
 254             if (old_key->program_string_id == key->program_string_id)
 255                break;
 256          }
 257       }
 258       if (c)
 259          break;
 260    }
 261
 262    if (!c) {
 263       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 264       return;
 265    }
 266
 267    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 268                       "depth write",
 269                       old_key->iz_lookup, key->iz_lookup);
 270    found |= key_debug(brw, "depth statistics",
 271                       old_key->stats_wm, key->stats_wm);
 272    found |= key_debug(brw, "flat shading",
 273                       old_key->flat_shade, key->flat_shade);
 274    found |= key_debug(brw, "per-sample interpolation",
 275                       old_key->persample_interp, key->persample_interp);
 276    found |= key_debug(brw, "number of color buffers",
 277                       old_key->nr_color_regions, key->nr_color_regions);
 278    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 279                       old_key->replicate_alpha, key->replicate_alpha);
 280    found |= key_debug(brw, "fragment color clamping",
 281                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 282    found |= key_debug(brw, "multisampled FBO",
 283                       old_key->multisample_fbo, key->multisample_fbo);
 284    found |= key_debug(brw, "line smoothing",
 285                       old_key->line_aa, key->line_aa);
 286    found |= key_debug(brw, "input slots valid",
 287                       old_key->input_slots_valid, key->input_slots_valid);
 288    found |= key_debug(brw, "mrt alpha test function",
 289                       old_key->alpha_test_func, key->alpha_test_func);
 290    found |= key_debug(brw, "mrt alpha test reference value",
 291                       old_key->alpha_test_ref, key->alpha_test_ref);
 292
 293    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 294
 295    if (!found) {
 296       perf_debug("  Something else\n");
 297    }
 298 }
 299
 300 static uint8_t
 301 gen6_gather_workaround(GLenum internalformat)
 302 {
 303    switch (internalformat) {
 304    case GL_R8I: return WA_SIGN | WA_8BIT;
 305    case GL_R8UI: return WA_8BIT;
 306    case GL_R16I: return WA_SIGN | WA_16BIT;
 307    case GL_R16UI: return WA_16BIT;
 308    default:
 309       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 310        * the surface state, there is no shader w/a required.
 311        */
 312       return 0;
 313    }
 314 }
 315
 316 void
 317 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 318                                    const struct gl_program *prog,
 319                                    struct brw_sampler_prog_key_data *key)
 320 {
 321    struct brw_context *brw = brw_context(ctx);
 322    GLbitfield mask = prog->SamplersUsed;
 323
 324    while (mask) {
 325       const int s = u_bit_scan(&mask);
 326
 327       key->swizzles[s] = SWIZZLE_NOOP;
 328
 329       int unit_id = prog->SamplerUnits[s];
 330       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 331
 332       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 333          const struct gl_texture_object *t = unit->_Current;
 334          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 335          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 336
 337          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 338             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 339              img->_BaseFormat == GL_DEPTH_STENCIL);
 340
 341          /* Haswell handles texture swizzling as surface format overrides
 342           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 343           */
 344          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 345             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 346
 347          if (brw->gen < 8 &&
 348              sampler->MinFilter != GL_NEAREST &&
 349              sampler->MagFilter != GL_NEAREST) {
 350             if (sampler->WrapS == GL_CLAMP)
 351                key->gl_clamp_mask[0] |= 1 << s;
 352             if (sampler->WrapT == GL_CLAMP)
 353                key->gl_clamp_mask[1] |= 1 << s;
 354             if (sampler->WrapR == GL_CLAMP)
 355                key->gl_clamp_mask[2] |= 1 << s;
 356          }
 357
 358          /* gather4's channel select for green from RG32F is broken; requires
 359           * a shader w/a on IVB; fixable with just SCS on HSW.
 360           */
 361          if (brw->gen == 7 && !brw->is_haswell &&
 362              prog->nir->info->uses_texture_gather) {
 363             if (img->InternalFormat == GL_RG32F)
 364                key->gather_channel_quirk_mask |= 1 << s;
 365          }
 366
 367          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 368           * UNORM/FLOAT instead and fix it in the shader.
 369           */
 370          if (brw->gen == 6 && prog->nir->info->uses_texture_gather) {
 371             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 372          }
 373
 374          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 375           * then we need to emit slightly different code to first sample the
 376           * MCS surface.
 377           */
 378          struct intel_texture_object *intel_tex =
 379             intel_texture_object((struct gl_texture_object *)t);
 380
 381          /* From gen9 onwards some single sampled buffers can also be
 382           * compressed. These don't need ld2dms sampling along with mcs fetch.
 383           */
 384          if (brw->gen >= 7 &&
 385              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS &&
 386              intel_tex->mt->num_samples > 1) {
 387             key->compressed_multisample_layout_mask |= 1 << s;
 388
 389             if (intel_tex->mt->num_samples >= 16) {
 390                assert(brw->gen >= 9);
 391                key->msaa_16 |= 1 << s;
 392             }
 393          }
 394
 395          if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
 396             switch (intel_tex->planar_format->components) {
 397             case __DRI_IMAGE_COMPONENTS_Y_UV:
 398                key->y_uv_image_mask |= 1 << s;
 399                break;
 400             case __DRI_IMAGE_COMPONENTS_Y_U_V:
 401                key->y_u_v_image_mask |= 1 << s;
 402                break;
 403             case __DRI_IMAGE_COMPONENTS_Y_XUXV:
 404                key->yx_xuxv_image_mask |= 1 << s;
 405                break;
 406             default:
 407                break;
 408             }
 409          }
 410
 411       }
 412    }
 413 }
 414
 415 static bool
 416 brw_wm_state_dirty(const struct brw_context *brw)
 417 {
 418    return brw_state_dirty(brw,
 419                           _NEW_BUFFERS |
 420                           _NEW_COLOR |
 421                           _NEW_DEPTH |
 422                           _NEW_FRAG_CLAMP |
 423                           _NEW_HINT |
 424                           _NEW_LIGHT |
 425                           _NEW_LINE |
 426                           _NEW_MULTISAMPLE |
 427                           _NEW_POLYGON |
 428                           _NEW_STENCIL |
 429                           _NEW_TEXTURE,
 430                           BRW_NEW_FRAGMENT_PROGRAM |
 431                           BRW_NEW_REDUCED_PRIMITIVE |
 432                           BRW_NEW_STATS_WM |
 433                           BRW_NEW_VUE_MAP_GEOM_OUT);
 434 }
 435
 436 void
 437 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 438 {
 439    struct gl_context *ctx = &brw->ctx;
 440    /* BRW_NEW_FRAGMENT_PROGRAM */
 441    const struct brw_program *fp = brw_program_const(brw->fragment_program);
 442    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 443    GLuint lookup = 0;
 444    GLuint line_aa;
 445
 446    memset(key, 0, sizeof(*key));
 447
 448    /* Build the index for table lookup
 449     */
 450    if (brw->gen < 6) {
 451       /* _NEW_COLOR */
 452       if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) {
 453          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 454       }
 455
 456       if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 457          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 458       }
 459
 460       /* _NEW_DEPTH */
 461       if (ctx->Depth.Test)
 462          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 463
 464       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 465          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 466
 467       /* _NEW_STENCIL | _NEW_BUFFERS */
 468       if (ctx->Stencil._Enabled) {
 469          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 470
 471          if (ctx->Stencil.WriteMask[0] ||
 472              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 473             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 474       }
 475       key->iz_lookup = lookup;
 476    }
 477
 478    line_aa = AA_NEVER;
 479
 480    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 481    if (ctx->Line.SmoothFlag) {
 482       if (brw->reduced_primitive == GL_LINES) {
 483          line_aa = AA_ALWAYS;
 484       }
 485       else if (brw->reduced_primitive == GL_TRIANGLES) {
 486          if (ctx->Polygon.FrontMode == GL_LINE) {
 487             line_aa = AA_SOMETIMES;
 488
 489             if (ctx->Polygon.BackMode == GL_LINE ||
 490                 (ctx->Polygon.CullFlag &&
 491                  ctx->Polygon.CullFaceMode == GL_BACK))
 492                line_aa = AA_ALWAYS;
 493          }
 494          else if (ctx->Polygon.BackMode == GL_LINE) {
 495             line_aa = AA_SOMETIMES;
 496
 497             if ((ctx->Polygon.CullFlag &&
 498                  ctx->Polygon.CullFaceMode == GL_FRONT))
 499                line_aa = AA_ALWAYS;
 500          }
 501       }
 502    }
 503
 504    key->line_aa = line_aa;
 505
 506    /* _NEW_HINT */
 507    key->high_quality_derivatives =
 508       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 509
 510    if (brw->gen < 6)
 511       key->stats_wm = brw->stats_wm;
 512
 513    /* _NEW_LIGHT */
 514    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 515
 516    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 517    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 518
 519    /* _NEW_TEXTURE */
 520    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 521
 522    /* _NEW_BUFFERS */
 523    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 524
 525    /* _NEW_COLOR */
 526    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 527       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 528
 529    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 530    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 531       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 532
 533    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 534    /* Ignore sample qualifier while computing this flag. */
 535    if (ctx->Multisample.Enabled) {
 536       key->persample_interp =
 537          ctx->Multisample.SampleShading &&
 538          (ctx->Multisample.MinSampleShadingValue *
 539           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 540
 541       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 542    }
 543
 544    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 545    if (brw->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 546                                          BRW_FS_VARYING_INPUT_MASK) > 16) {
 547       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 548    }
 549
 550
 551    /* _NEW_COLOR | _NEW_BUFFERS */
 552    /* Pre-gen6, the hardware alpha test always used each render
 553     * target's alpha to do alpha test, as opposed to render target 0's alpha
 554     * like GL requires.  Fix that by building the alpha test into the
 555     * shader, and we'll skip enabling the fixed function alpha test.
 556     */
 557    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 558        ctx->Color.AlphaEnabled) {
 559       key->alpha_test_func = ctx->Color.AlphaFunc;
 560       key->alpha_test_ref = ctx->Color.AlphaRef;
 561    }
 562
 563    /* The unique fragment program ID */
 564    key->program_string_id = fp->id;
 565
 566    /* Whether reads from the framebuffer should behave coherently. */
 567    key->coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 568 }
 569
 570 void
 571 brw_upload_wm_prog(struct brw_context *brw)
 572 {
 573    struct gl_context *ctx = &brw->ctx;
 574    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 575    struct brw_wm_prog_key key;
 576    struct brw_program *fp = (struct brw_program *) brw->fragment_program;
 577
 578    if (!brw_wm_state_dirty(brw))
 579       return;
 580
 581    brw_wm_populate_key(brw, &key);
 582
 583    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 584                          &key, sizeof(key),
 585                          &brw->wm.base.prog_offset,
 586                          &brw->wm.base.prog_data)) {
 587       bool success = brw_codegen_wm_prog(brw, current, fp, &key,
 588                                          &brw->vue_map_geom_out);
 589       (void) success;
 590       assert(success);
 591    }
 592 }
 593
 594 bool
 595 brw_fs_precompile(struct gl_context *ctx,
 596                   struct gl_shader_program *shader_prog,
 597                   struct gl_program *prog)
 598 {
 599    struct brw_context *brw = brw_context(ctx);
 600    struct brw_wm_prog_key key;
 601
 602    struct brw_program *bfp = brw_program(prog);
 603
 604    memset(&key, 0, sizeof(key));
 605
 606    uint64_t outputs_written = prog->info.outputs_written;
 607
 608    if (brw->gen < 6) {
 609       if (prog->info.fs.uses_discard)
 610          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 611
 612       if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 613          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 614
 615       /* Just assume depth testing. */
 616       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 617       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 618    }
 619
 620    if (brw->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 621                                          BRW_FS_VARYING_INPUT_MASK) > 16) {
 622       key.input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
 623    }
 624
 625    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 626
 627    key.nr_color_regions = _mesa_bitcount_64(outputs_written &
 628          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 629            BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
 630            BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 631
 632    key.program_string_id = bfp->id;
 633
 634    /* Whether reads from the framebuffer should behave coherently. */
 635    key.coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 636
 637    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 638    struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data;
 639
 640    struct brw_vue_map vue_map;
 641    if (brw->gen < 6) {
 642       brw_compute_vue_map(&brw->screen->devinfo, &vue_map,
 643                           prog->info.inputs_read | VARYING_BIT_POS,
 644                           false);
 645    }
 646
 647    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key, &vue_map);
 648
 649    brw->wm.base.prog_offset = old_prog_offset;
 650    brw->wm.base.prog_data = old_prog_data;
 651
 652    return success;
 653 }