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 "main/enums.h"
  30 #include "main/formats.h"
  31 #include "main/fbobject.h"
  32 #include "main/samplerobj.h"
  33 #include "main/framebuffer.h"
  34 #include "program/prog_parameter.h"
  35 #include "program/program.h"
  36 #include "intel_mipmap_tree.h"
  37 #include "intel_image.h"
  38 #include "compiler/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 gen_device_info *devinfo,
  45                                 const struct gl_program *prog,
  46                                 const struct brw_wm_prog_key *key,
  47                                 struct brw_wm_prog_data *prog_data)
  48 {
  49    /* Render targets implicitly start at surface index 0.  Even if there are
  50     * no color regions, we still perform an FB write to a null render target,
  51     * which will be surface 0.
  52     */
  53    uint32_t next_binding_table_offset = MAX2(key->nr_color_regions, 1);
  54
  55    next_binding_table_offset =
  56       brw_assign_common_binding_table_offsets(devinfo, prog, &prog_data->base,
  57                                               next_binding_table_offset);
  58
  59    if (prog->nir->info.outputs_read && !key->coherent_fb_fetch) {
  60       prog_data->binding_table.render_target_read_start =
  61          next_binding_table_offset;
  62       next_binding_table_offset += key->nr_color_regions;
  63    }
  64 }
  65
  66 static void
  67 brw_wm_debug_recompile(struct brw_context *brw, struct gl_program *prog,
  68                        const struct brw_wm_prog_key *key)
  69 {
  70    perf_debug("Recompiling fragment shader for program %d\n", prog->Id);
  71
  72    bool found = false;
  73    const struct brw_wm_prog_key *old_key =
  74       brw_find_previous_compile(&brw->cache, BRW_CACHE_FS_PROG,
  75                                 key->program_string_id);
  76
  77    if (!old_key) {
  78       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
  79       return;
  80    }
  81
  82    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
  83                       "depth write",
  84                       old_key->iz_lookup, key->iz_lookup);
  85    found |= key_debug(brw, "depth statistics",
  86                       old_key->stats_wm, key->stats_wm);
  87    found |= key_debug(brw, "flat shading",
  88                       old_key->flat_shade, key->flat_shade);
  89    found |= key_debug(brw, "number of color buffers",
  90                       old_key->nr_color_regions, key->nr_color_regions);
  91    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
  92                       old_key->replicate_alpha, key->replicate_alpha);
  93    found |= key_debug(brw, "fragment color clamping",
  94                       old_key->clamp_fragment_color, key->clamp_fragment_color);
  95    found |= key_debug(brw, "per-sample interpolation",
  96                       old_key->persample_interp, key->persample_interp);
  97    found |= key_debug(brw, "multisampled FBO",
  98                       old_key->multisample_fbo, key->multisample_fbo);
  99    found |= key_debug(brw, "frag coord adds sample pos",
 100                       old_key->frag_coord_adds_sample_pos,
 101                       key->frag_coord_adds_sample_pos);
 102    found |= key_debug(brw, "line smoothing",
 103                       old_key->line_aa, key->line_aa);
 104    found |= key_debug(brw, "high quality derivatives",
 105                       old_key->high_quality_derivatives,
 106                       key->high_quality_derivatives);
 107    found |= key_debug(brw, "force dual color blending",
 108                       old_key->force_dual_color_blend,
 109                       key->force_dual_color_blend);
 110    found |= key_debug(brw, "coherent fb fetch",
 111                       old_key->coherent_fb_fetch, key->coherent_fb_fetch);
 112
 113    found |= key_debug(brw, "input slots valid",
 114                       old_key->input_slots_valid, key->input_slots_valid);
 115    found |= key_debug(brw, "mrt alpha test function",
 116                       old_key->alpha_test_func, key->alpha_test_func);
 117    found |= key_debug(brw, "mrt alpha test reference value",
 118                       old_key->alpha_test_ref, key->alpha_test_ref);
 119
 120    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 121
 122    if (!found) {
 123       perf_debug("  Something else\n");
 124    }
 125 }
 126
 127 static bool
 128 brw_codegen_wm_prog(struct brw_context *brw,
 129                     struct brw_program *fp,
 130                     struct brw_wm_prog_key *key,
 131                     struct brw_vue_map *vue_map)
 132 {
 133    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 134    void *mem_ctx = ralloc_context(NULL);
 135    struct brw_wm_prog_data prog_data;
 136    const GLuint *program;
 137    bool start_busy = false;
 138    double start_time = 0;
 139
 140    memset(&prog_data, 0, sizeof(prog_data));
 141
 142    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 143    if (fp->program.is_arb_asm)
 144       prog_data.base.use_alt_mode = true;
 145
 146    assign_fs_binding_table_offsets(devinfo, &fp->program, key, &prog_data);
 147
 148    if (!fp->program.is_arb_asm) {
 149       brw_nir_setup_glsl_uniforms(mem_ctx, fp->program.nir, &fp->program,
 150                                   &prog_data.base, true);
 151       brw_nir_analyze_ubo_ranges(brw->screen->compiler, fp->program.nir,
 152                                  prog_data.base.ubo_ranges);
 153    } else {
 154       brw_nir_setup_arb_uniforms(mem_ctx, fp->program.nir, &fp->program,
 155                                  &prog_data.base);
 156
 157       if (unlikely(INTEL_DEBUG & DEBUG_WM))
 158          brw_dump_arb_asm("fragment", &fp->program);
 159    }
 160
 161    if (unlikely(brw->perf_debug)) {
 162       start_busy = (brw->batch.last_bo &&
 163                     brw_bo_busy(brw->batch.last_bo));
 164       start_time = get_time();
 165    }
 166
 167    int st_index8 = -1, st_index16 = -1;
 168    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 169       st_index8 = brw_get_shader_time_index(brw, &fp->program, ST_FS8,
 170                                             !fp->program.is_arb_asm);
 171       st_index16 = brw_get_shader_time_index(brw, &fp->program, ST_FS16,
 172                                              !fp->program.is_arb_asm);
 173    }
 174
 175    char *error_str = NULL;
 176    program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx,
 177                             key, &prog_data, fp->program.nir,
 178                             &fp->program, st_index8, st_index16,
 179                             true, false, vue_map,
 180                             &error_str);
 181
 182    if (program == NULL) {
 183       if (!fp->program.is_arb_asm) {
 184          fp->program.sh.data->LinkStatus = linking_failure;
 185          ralloc_strcat(&fp->program.sh.data->InfoLog, error_str);
 186       }
 187
 188       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 189
 190       ralloc_free(mem_ctx);
 191       return false;
 192    }
 193
 194    if (unlikely(brw->perf_debug)) {
 195       if (fp->compiled_once)
 196          brw_wm_debug_recompile(brw, &fp->program, key);
 197       fp->compiled_once = true;
 198
 199       if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
 200          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 201                     (get_time() - start_time) * 1000);
 202       }
 203    }
 204
 205    brw_alloc_stage_scratch(brw, &brw->wm.base, prog_data.base.total_scratch);
 206
 207    if (unlikely((INTEL_DEBUG & DEBUG_WM) && fp->program.is_arb_asm))
 208       fprintf(stderr, "\n");
 209
 210    /* The param and pull_param arrays will be freed by the shader cache. */
 211    ralloc_steal(NULL, prog_data.base.param);
 212    ralloc_steal(NULL, prog_data.base.pull_param);
 213    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 214                     key, sizeof(struct brw_wm_prog_key),
 215                     program, prog_data.base.program_size,
 216                     &prog_data, sizeof(prog_data),
 217                     &brw->wm.base.prog_offset, &brw->wm.base.prog_data);
 218
 219    ralloc_free(mem_ctx);
 220
 221    return true;
 222 }
 223
 224 bool
 225 brw_debug_recompile_sampler_key(struct brw_context *brw,
 226                                 const struct brw_sampler_prog_key_data *old_key,
 227                                 const struct brw_sampler_prog_key_data *key)
 228 {
 229    bool found = false;
 230
 231    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 232       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 233                          old_key->swizzles[i], key->swizzles[i]);
 234    }
 235    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 236                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 237    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 238                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 239    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 240                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 241    found |= key_debug(brw, "gather channel quirk on any texture unit",
 242                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 243    found |= key_debug(brw, "compressed multisample layout",
 244                       old_key->compressed_multisample_layout_mask,
 245                       key->compressed_multisample_layout_mask);
 246    found |= key_debug(brw, "16x msaa",
 247                       old_key->msaa_16,
 248                       key->msaa_16);
 249
 250    found |= key_debug(brw, "y_uv image bound",
 251                       old_key->y_uv_image_mask,
 252                       key->y_uv_image_mask);
 253    found |= key_debug(brw, "y_u_v image bound",
 254                       old_key->y_u_v_image_mask,
 255                       key->y_u_v_image_mask);
 256    found |= key_debug(brw, "yx_xuxv image bound",
 257                       old_key->yx_xuxv_image_mask,
 258                       key->yx_xuxv_image_mask);
 259    found |= key_debug(brw, "xy_uxvx image bound",
 260                       old_key->xy_uxvx_image_mask,
 261                       key->xy_uxvx_image_mask);
 262
 263
 264    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 265       found |= key_debug(brw, "textureGather workarounds",
 266                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 267    }
 268
 269    return found;
 270 }
 271
 272 static uint8_t
 273 gen6_gather_workaround(GLenum internalformat)
 274 {
 275    switch (internalformat) {
 276    case GL_R8I: return WA_SIGN | WA_8BIT;
 277    case GL_R8UI: return WA_8BIT;
 278    case GL_R16I: return WA_SIGN | WA_16BIT;
 279    case GL_R16UI: return WA_16BIT;
 280    default:
 281       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 282        * the surface state, there is no shader w/a required.
 283        */
 284       return 0;
 285    }
 286 }
 287
 288 void
 289 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 290                                    const struct gl_program *prog,
 291                                    struct brw_sampler_prog_key_data *key)
 292 {
 293    struct brw_context *brw = brw_context(ctx);
 294    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 295    GLbitfield mask = prog->SamplersUsed;
 296
 297    while (mask) {
 298       const int s = u_bit_scan(&mask);
 299
 300       key->swizzles[s] = SWIZZLE_NOOP;
 301
 302       int unit_id = prog->SamplerUnits[s];
 303       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 304
 305       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 306          const struct gl_texture_object *t = unit->_Current;
 307          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 308          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 309
 310          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 311             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 312              img->_BaseFormat == GL_DEPTH_STENCIL);
 313
 314          /* Haswell handles texture swizzling as surface format overrides
 315           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 316           */
 317          if (alpha_depth || (devinfo->gen < 8 && !devinfo->is_haswell))
 318             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 319
 320          if (devinfo->gen < 8 &&
 321              sampler->MinFilter != GL_NEAREST &&
 322              sampler->MagFilter != GL_NEAREST) {
 323             if (sampler->WrapS == GL_CLAMP)
 324                key->gl_clamp_mask[0] |= 1 << s;
 325             if (sampler->WrapT == GL_CLAMP)
 326                key->gl_clamp_mask[1] |= 1 << s;
 327             if (sampler->WrapR == GL_CLAMP)
 328                key->gl_clamp_mask[2] |= 1 << s;
 329          }
 330
 331          /* gather4 for RG32* is broken in multiple ways on Gen7. */
 332          if (devinfo->gen == 7 && prog->info.uses_texture_gather) {
 333             switch (img->InternalFormat) {
 334             case GL_RG32I:
 335             case GL_RG32UI: {
 336                /* We have to override the format to R32G32_FLOAT_LD.
 337                 * This means that SCS_ALPHA and SCS_ONE will return 0x3f8
 338                 * (1.0) rather than integer 1.  This needs shader hacks.
 339                 *
 340                 * On Ivybridge, we whack W (alpha) to ONE in our key's
 341                 * swizzle.  On Haswell, we look at the original texture
 342                 * swizzle, and use XYZW with channels overridden to ONE,
 343                 * leaving normal texture swizzling to SCS.
 344                 */
 345                unsigned src_swizzle =
 346                   devinfo->is_haswell ? t->_Swizzle : key->swizzles[s];
 347                for (int i = 0; i < 4; i++) {
 348                   unsigned src_comp = GET_SWZ(src_swizzle, i);
 349                   if (src_comp == SWIZZLE_ONE || src_comp == SWIZZLE_W) {
 350                      key->swizzles[i] &= ~(0x7 << (3 * i));
 351                      key->swizzles[i] |= SWIZZLE_ONE << (3 * i);
 352                   }
 353                }
 354                /* fallthrough */
 355             }
 356             case GL_RG32F:
 357                /* The channel select for green doesn't work - we have to
 358                 * request blue.  Haswell can use SCS for this, but Ivybridge
 359                 * needs a shader workaround.
 360                 */
 361                if (!devinfo->is_haswell)
 362                   key->gather_channel_quirk_mask |= 1 << s;
 363                break;
 364             }
 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 (devinfo->gen == 6 && prog->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 (intel_tex->mt->aux_usage == ISL_AUX_USAGE_MCS) {
 385             assert(devinfo->gen >= 7);
 386             assert(intel_tex->mt->surf.samples > 1);
 387             assert(intel_tex->mt->mcs_buf);
 388             assert(intel_tex->mt->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY);
 389             key->compressed_multisample_layout_mask |= 1 << s;
 390
 391             if (intel_tex->mt->surf.samples >= 16) {
 392                assert(devinfo->gen >= 9);
 393                key->msaa_16 |= 1 << s;
 394             }
 395          }
 396
 397          if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
 398             switch (intel_tex->planar_format->components) {
 399             case __DRI_IMAGE_COMPONENTS_Y_UV:
 400                key->y_uv_image_mask |= 1 << s;
 401                break;
 402             case __DRI_IMAGE_COMPONENTS_Y_U_V:
 403                key->y_u_v_image_mask |= 1 << s;
 404                break;
 405             case __DRI_IMAGE_COMPONENTS_Y_XUXV:
 406                key->yx_xuxv_image_mask |= 1 << s;
 407                break;
 408             case __DRI_IMAGE_COMPONENTS_Y_UXVX:
 409                key->xy_uxvx_image_mask |= 1 << s;
 410                break;
 411             default:
 412                break;
 413             }
 414          }
 415
 416       }
 417    }
 418 }
 419
 420 static bool
 421 brw_wm_state_dirty(const struct brw_context *brw)
 422 {
 423    return brw_state_dirty(brw,
 424                           _NEW_BUFFERS |
 425                           _NEW_COLOR |
 426                           _NEW_DEPTH |
 427                           _NEW_FRAG_CLAMP |
 428                           _NEW_HINT |
 429                           _NEW_LIGHT |
 430                           _NEW_LINE |
 431                           _NEW_MULTISAMPLE |
 432                           _NEW_POLYGON |
 433                           _NEW_STENCIL |
 434                           _NEW_TEXTURE,
 435                           BRW_NEW_FRAGMENT_PROGRAM |
 436                           BRW_NEW_REDUCED_PRIMITIVE |
 437                           BRW_NEW_STATS_WM |
 438                           BRW_NEW_VUE_MAP_GEOM_OUT);
 439 }
 440
 441 void
 442 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 443 {
 444    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 445    struct gl_context *ctx = &brw->ctx;
 446    /* BRW_NEW_FRAGMENT_PROGRAM */
 447    const struct gl_program *prog = brw->programs[MESA_SHADER_FRAGMENT];
 448    const struct brw_program *fp = brw_program_const(prog);
 449    GLuint lookup = 0;
 450    GLuint line_aa;
 451
 452    memset(key, 0, sizeof(*key));
 453
 454    /* Build the index for table lookup
 455     */
 456    if (devinfo->gen < 6) {
 457       /* _NEW_COLOR */
 458       if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) {
 459          lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 460       }
 461
 462       if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 463          lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 464       }
 465
 466       /* _NEW_DEPTH */
 467       if (ctx->Depth.Test)
 468          lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 469
 470       if (brw_depth_writes_enabled(brw))
 471          lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 472
 473       /* _NEW_STENCIL | _NEW_BUFFERS */
 474       if (brw->stencil_enabled) {
 475          lookup |= BRW_WM_IZ_STENCIL_TEST_ENABLE_BIT;
 476
 477          if (ctx->Stencil.WriteMask[0] ||
 478              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 479             lookup |= BRW_WM_IZ_STENCIL_WRITE_ENABLE_BIT;
 480       }
 481       key->iz_lookup = lookup;
 482    }
 483
 484    line_aa = BRW_WM_AA_NEVER;
 485
 486    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 487    if (ctx->Line.SmoothFlag) {
 488       if (brw->reduced_primitive == GL_LINES) {
 489          line_aa = BRW_WM_AA_ALWAYS;
 490       }
 491       else if (brw->reduced_primitive == GL_TRIANGLES) {
 492          if (ctx->Polygon.FrontMode == GL_LINE) {
 493             line_aa = BRW_WM_AA_SOMETIMES;
 494
 495             if (ctx->Polygon.BackMode == GL_LINE ||
 496                 (ctx->Polygon.CullFlag &&
 497                  ctx->Polygon.CullFaceMode == GL_BACK))
 498                line_aa = BRW_WM_AA_ALWAYS;
 499          }
 500          else if (ctx->Polygon.BackMode == GL_LINE) {
 501             line_aa = BRW_WM_AA_SOMETIMES;
 502
 503             if ((ctx->Polygon.CullFlag &&
 504                  ctx->Polygon.CullFaceMode == GL_FRONT))
 505                line_aa = BRW_WM_AA_ALWAYS;
 506          }
 507       }
 508    }
 509
 510    key->line_aa = line_aa;
 511
 512    /* _NEW_HINT */
 513    key->high_quality_derivatives =
 514       prog->info.uses_fddx_fddy &&
 515       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 516
 517    if (devinfo->gen < 6)
 518       key->stats_wm = brw->stats_wm;
 519
 520    /* _NEW_LIGHT */
 521    key->flat_shade =
 522       (prog->info.inputs_read & (VARYING_BIT_COL0 | VARYING_BIT_COL1)) &&
 523       (ctx->Light.ShadeModel == GL_FLAT);
 524
 525    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 526    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 527
 528    /* _NEW_TEXTURE */
 529    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 530
 531    /* _NEW_BUFFERS */
 532    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 533
 534    /* _NEW_COLOR */
 535    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 536       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 537
 538    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 539    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 540       (_mesa_is_alpha_test_enabled(ctx) ||
 541        _mesa_is_alpha_to_coverage_enabled(ctx));
 542
 543    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 544    /* Ignore sample qualifier while computing this flag. */
 545    if (ctx->Multisample.Enabled) {
 546       key->persample_interp =
 547          ctx->Multisample.SampleShading &&
 548          (ctx->Multisample.MinSampleShadingValue *
 549           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 550
 551       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 552    }
 553
 554    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 555    if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 556                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 557       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 558    }
 559
 560    /* _NEW_COLOR | _NEW_BUFFERS */
 561    /* Pre-gen6, the hardware alpha test always used each render
 562     * target's alpha to do alpha test, as opposed to render target 0's alpha
 563     * like GL requires.  Fix that by building the alpha test into the
 564     * shader, and we'll skip enabling the fixed function alpha test.
 565     */
 566    if (devinfo->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 567        ctx->Color.AlphaEnabled) {
 568       key->alpha_test_func = ctx->Color.AlphaFunc;
 569       key->alpha_test_ref = ctx->Color.AlphaRef;
 570    }
 571
 572    /* The unique fragment program ID */
 573    key->program_string_id = fp->id;
 574
 575    /* Whether reads from the framebuffer should behave coherently. */
 576    key->coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 577 }
 578
 579 void
 580 brw_upload_wm_prog(struct brw_context *brw)
 581 {
 582    struct brw_wm_prog_key key;
 583    struct brw_program *fp =
 584       (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
 585
 586    if (!brw_wm_state_dirty(brw))
 587       return;
 588
 589    brw_wm_populate_key(brw, &key);
 590
 591    if (brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 592                         &key, sizeof(key),
 593                         &brw->wm.base.prog_offset,
 594                         &brw->wm.base.prog_data))
 595       return;
 596
 597    if (brw_disk_cache_upload_program(brw, MESA_SHADER_FRAGMENT))
 598       return;
 599
 600    fp = (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
 601    fp->id = key.program_string_id;
 602
 603    MAYBE_UNUSED bool success = brw_codegen_wm_prog(brw, fp, &key,
 604                                                    &brw->vue_map_geom_out);
 605    assert(success);
 606 }
 607
 608 bool
 609 brw_fs_precompile(struct gl_context *ctx, struct gl_program *prog)
 610 {
 611    struct brw_context *brw = brw_context(ctx);
 612    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 613    struct brw_wm_prog_key key;
 614
 615    struct brw_program *bfp = brw_program(prog);
 616
 617    memset(&key, 0, sizeof(key));
 618
 619    uint64_t outputs_written = prog->info.outputs_written;
 620
 621    if (devinfo->gen < 6) {
 622       if (prog->info.fs.uses_discard)
 623          key.iz_lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 624
 625       if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 626          key.iz_lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 627
 628       /* Just assume depth testing. */
 629       key.iz_lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 630       key.iz_lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 631    }
 632
 633    if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 634                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 635       key.input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
 636    }
 637
 638    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 639
 640    key.nr_color_regions = _mesa_bitcount_64(outputs_written &
 641          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 642            BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
 643            BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 644
 645    key.program_string_id = bfp->id;
 646
 647    /* Whether reads from the framebuffer should behave coherently. */
 648    key.coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 649
 650    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 651    struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data;
 652
 653    struct brw_vue_map vue_map;
 654    if (devinfo->gen < 6) {
 655       brw_compute_vue_map(&brw->screen->devinfo, &vue_map,
 656                           prog->info.inputs_read | VARYING_BIT_POS,
 657                           false);
 658    }
 659
 660    bool success = brw_codegen_wm_prog(brw, bfp, &key, &vue_map);
 661
 662    brw->wm.base.prog_offset = old_prog_offset;
 663    brw->wm.base.prog_data = old_prog_data;
 664
 665    return success;
 666 }