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 "intel_fbo.h"
  39 #include "compiler/brw_nir.h"
  40 #include "brw_program.h"
  41
  42 #include "util/ralloc.h"
  43 #include "util/u_math.h"
  44
  45 static void
  46 assign_fs_binding_table_offsets(const struct gen_device_info *devinfo,
  47                                 const struct gl_program *prog,
  48                                 const struct brw_wm_prog_key *key,
  49                                 struct brw_wm_prog_data *prog_data)
  50 {
  51    /* Render targets implicitly start at surface index 0.  Even if there are
  52     * no color regions, we still perform an FB write to a null render target,
  53     * which will be surface 0.
  54     */
  55    uint32_t next_binding_table_offset = MAX2(key->nr_color_regions, 1);
  56
  57    next_binding_table_offset =
  58       brw_assign_common_binding_table_offsets(devinfo, prog, &prog_data->base,
  59                                               next_binding_table_offset);
  60
  61    if (prog->nir->info.outputs_read && !key->coherent_fb_fetch) {
  62       prog_data->binding_table.render_target_read_start =
  63          next_binding_table_offset;
  64       next_binding_table_offset += key->nr_color_regions;
  65    }
  66 }
  67
  68 static void
  69 brw_wm_debug_recompile(struct brw_context *brw, struct gl_program *prog,
  70                        const struct brw_wm_prog_key *key)
  71 {
  72    perf_debug("Recompiling fragment shader for program %d\n", prog->Id);
  73
  74    bool found = false;
  75    const struct brw_wm_prog_key *old_key =
  76       brw_find_previous_compile(&brw->cache, BRW_CACHE_FS_PROG,
  77                                 key->program_string_id);
  78
  79    if (!old_key) {
  80       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
  81       return;
  82    }
  83
  84    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
  85                       "depth write",
  86                       old_key->iz_lookup, key->iz_lookup);
  87    found |= key_debug(brw, "depth statistics",
  88                       old_key->stats_wm, key->stats_wm);
  89    found |= key_debug(brw, "flat shading",
  90                       old_key->flat_shade, key->flat_shade);
  91    found |= key_debug(brw, "number of color buffers",
  92                       old_key->nr_color_regions, key->nr_color_regions);
  93    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
  94                       old_key->replicate_alpha, key->replicate_alpha);
  95    found |= key_debug(brw, "fragment color clamping",
  96                       old_key->clamp_fragment_color, key->clamp_fragment_color);
  97    found |= key_debug(brw, "per-sample interpolation",
  98                       old_key->persample_interp, key->persample_interp);
  99    found |= key_debug(brw, "multisampled FBO",
 100                       old_key->multisample_fbo, key->multisample_fbo);
 101    found |= key_debug(brw, "frag coord adds sample pos",
 102                       old_key->frag_coord_adds_sample_pos,
 103                       key->frag_coord_adds_sample_pos);
 104    found |= key_debug(brw, "line smoothing",
 105                       old_key->line_aa, key->line_aa);
 106    found |= key_debug(brw, "high quality derivatives",
 107                       old_key->high_quality_derivatives,
 108                       key->high_quality_derivatives);
 109    found |= key_debug(brw, "force dual color blending",
 110                       old_key->force_dual_color_blend,
 111                       key->force_dual_color_blend);
 112    found |= key_debug(brw, "coherent fb fetch",
 113                       old_key->coherent_fb_fetch, key->coherent_fb_fetch);
 114
 115    found |= key_debug(brw, "input slots valid",
 116                       old_key->input_slots_valid, key->input_slots_valid);
 117    found |= key_debug(brw, "mrt alpha test function",
 118                       old_key->alpha_test_func, key->alpha_test_func);
 119    found |= key_debug(brw, "mrt alpha test reference value",
 120                       old_key->alpha_test_ref, key->alpha_test_ref);
 121
 122    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 123
 124    if (!found) {
 125       perf_debug("  Something else\n");
 126    }
 127 }
 128
 129 static bool
 130 brw_codegen_wm_prog(struct brw_context *brw,
 131                     struct brw_program *fp,
 132                     struct brw_wm_prog_key *key,
 133                     struct brw_vue_map *vue_map)
 134 {
 135    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 136    void *mem_ctx = ralloc_context(NULL);
 137    struct brw_wm_prog_data prog_data;
 138    const GLuint *program;
 139    bool start_busy = false;
 140    double start_time = 0;
 141
 142    nir_shader *nir = nir_shader_clone(mem_ctx, fp->program.nir);
 143
 144    memset(&prog_data, 0, sizeof(prog_data));
 145
 146    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 147    if (fp->program.is_arb_asm)
 148       prog_data.base.use_alt_mode = true;
 149
 150    assign_fs_binding_table_offsets(devinfo, &fp->program, key, &prog_data);
 151
 152    if (!fp->program.is_arb_asm) {
 153       brw_nir_setup_glsl_uniforms(mem_ctx, nir, &fp->program,
 154                                   &prog_data.base, true);
 155       brw_nir_analyze_ubo_ranges(brw->screen->compiler, nir,
 156                                  NULL, prog_data.base.ubo_ranges);
 157    } else {
 158       brw_nir_setup_arb_uniforms(mem_ctx, nir, &fp->program, &prog_data.base);
 159
 160       if (unlikely(INTEL_DEBUG & DEBUG_WM))
 161          brw_dump_arb_asm("fragment", &fp->program);
 162    }
 163
 164    if (unlikely(brw->perf_debug)) {
 165       start_busy = (brw->batch.last_bo &&
 166                     brw_bo_busy(brw->batch.last_bo));
 167       start_time = get_time();
 168    }
 169
 170    int st_index8 = -1, st_index16 = -1, st_index32 = -1;
 171    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 172       st_index8 = brw_get_shader_time_index(brw, &fp->program, ST_FS8,
 173                                             !fp->program.is_arb_asm);
 174       st_index16 = brw_get_shader_time_index(brw, &fp->program, ST_FS16,
 175                                              !fp->program.is_arb_asm);
 176       st_index32 = brw_get_shader_time_index(brw, &fp->program, ST_FS32,
 177                                              !fp->program.is_arb_asm);
 178    }
 179
 180    char *error_str = NULL;
 181    program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx,
 182                             key, &prog_data, nir,
 183                             &fp->program, st_index8, st_index16, st_index32,
 184                             true, false, vue_map,
 185                             &error_str);
 186
 187    if (program == NULL) {
 188       if (!fp->program.is_arb_asm) {
 189          fp->program.sh.data->LinkStatus = LINKING_FAILURE;
 190          ralloc_strcat(&fp->program.sh.data->InfoLog, error_str);
 191       }
 192
 193       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 194
 195       ralloc_free(mem_ctx);
 196       return false;
 197    }
 198
 199    if (unlikely(brw->perf_debug)) {
 200       if (fp->compiled_once)
 201          brw_wm_debug_recompile(brw, &fp->program, key);
 202       fp->compiled_once = true;
 203
 204       if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
 205          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 206                     (get_time() - start_time) * 1000);
 207       }
 208    }
 209
 210    brw_alloc_stage_scratch(brw, &brw->wm.base, prog_data.base.total_scratch);
 211
 212    if (unlikely((INTEL_DEBUG & DEBUG_WM) && fp->program.is_arb_asm))
 213       fprintf(stderr, "\n");
 214
 215    /* The param and pull_param arrays will be freed by the shader cache. */
 216    ralloc_steal(NULL, prog_data.base.param);
 217    ralloc_steal(NULL, prog_data.base.pull_param);
 218    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 219                     key, sizeof(struct brw_wm_prog_key),
 220                     program, prog_data.base.program_size,
 221                     &prog_data, sizeof(prog_data),
 222                     &brw->wm.base.prog_offset, &brw->wm.base.prog_data);
 223
 224    ralloc_free(mem_ctx);
 225
 226    return true;
 227 }
 228
 229 bool
 230 brw_debug_recompile_sampler_key(struct brw_context *brw,
 231                                 const struct brw_sampler_prog_key_data *old_key,
 232                                 const struct brw_sampler_prog_key_data *key)
 233 {
 234    bool found = false;
 235
 236    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 237       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 238                          old_key->swizzles[i], key->swizzles[i]);
 239    }
 240    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 241                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 242    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 243                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 244    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 245                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 246    found |= key_debug(brw, "gather channel quirk on any texture unit",
 247                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 248    found |= key_debug(brw, "compressed multisample layout",
 249                       old_key->compressed_multisample_layout_mask,
 250                       key->compressed_multisample_layout_mask);
 251    found |= key_debug(brw, "16x msaa",
 252                       old_key->msaa_16,
 253                       key->msaa_16);
 254
 255    found |= key_debug(brw, "y_uv image bound",
 256                       old_key->y_uv_image_mask,
 257                       key->y_uv_image_mask);
 258    found |= key_debug(brw, "y_u_v image bound",
 259                       old_key->y_u_v_image_mask,
 260                       key->y_u_v_image_mask);
 261    found |= key_debug(brw, "yx_xuxv image bound",
 262                       old_key->yx_xuxv_image_mask,
 263                       key->yx_xuxv_image_mask);
 264    found |= key_debug(brw, "xy_uxvx image bound",
 265                       old_key->xy_uxvx_image_mask,
 266                       key->xy_uxvx_image_mask);
 267    found |= key_debug(brw, "ayuv image bound",
 268                       old_key->ayuv_image_mask,
 269                       key->ayuv_image_mask);
 270
 271
 272    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 273       found |= key_debug(brw, "textureGather workarounds",
 274                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 275    }
 276
 277    return found;
 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                                    struct brw_sampler_prog_key_data *key)
 300 {
 301    struct brw_context *brw = brw_context(ctx);
 302    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 303    GLbitfield mask = prog->SamplersUsed;
 304
 305    while (mask) {
 306       const int s = u_bit_scan(&mask);
 307
 308       key->swizzles[s] = SWIZZLE_NOOP;
 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 || (devinfo->gen < 8 && !devinfo->is_haswell))
 326             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 327
 328          if (devinfo->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 for RG32* is broken in multiple ways on Gen7. */
 340          if (devinfo->gen == 7 && prog->info.uses_texture_gather) {
 341             switch (img->InternalFormat) {
 342             case GL_RG32I:
 343             case GL_RG32UI: {
 344                /* We have to override the format to R32G32_FLOAT_LD.
 345                 * This means that SCS_ALPHA and SCS_ONE will return 0x3f8
 346                 * (1.0) rather than integer 1.  This needs shader hacks.
 347                 *
 348                 * On Ivybridge, we whack W (alpha) to ONE in our key's
 349                 * swizzle.  On Haswell, we look at the original texture
 350                 * swizzle, and use XYZW with channels overridden to ONE,
 351                 * leaving normal texture swizzling to SCS.
 352                 */
 353                unsigned src_swizzle =
 354                   devinfo->is_haswell ? t->_Swizzle : key->swizzles[s];
 355                for (int i = 0; i < 4; i++) {
 356                   unsigned src_comp = GET_SWZ(src_swizzle, i);
 357                   if (src_comp == SWIZZLE_ONE || src_comp == SWIZZLE_W) {
 358                      key->swizzles[i] &= ~(0x7 << (3 * i));
 359                      key->swizzles[i] |= SWIZZLE_ONE << (3 * i);
 360                   }
 361                }
 362                /* fallthrough */
 363             }
 364             case GL_RG32F:
 365                /* The channel select for green doesn't work - we have to
 366                 * request blue.  Haswell can use SCS for this, but Ivybridge
 367                 * needs a shader workaround.
 368                 */
 369                if (!devinfo->is_haswell)
 370                   key->gather_channel_quirk_mask |= 1 << s;
 371                break;
 372             }
 373          }
 374
 375          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 376           * UNORM/FLOAT instead and fix it in the shader.
 377           */
 378          if (devinfo->gen == 6 && prog->info.uses_texture_gather) {
 379             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 380          }
 381
 382          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 383           * then we need to emit slightly different code to first sample the
 384           * MCS surface.
 385           */
 386          struct intel_texture_object *intel_tex =
 387             intel_texture_object((struct gl_texture_object *)t);
 388
 389          /* From gen9 onwards some single sampled buffers can also be
 390           * compressed. These don't need ld2dms sampling along with mcs fetch.
 391           */
 392          if (intel_tex->mt->aux_usage == ISL_AUX_USAGE_MCS) {
 393             assert(devinfo->gen >= 7);
 394             assert(intel_tex->mt->surf.samples > 1);
 395             assert(intel_tex->mt->aux_buf);
 396             assert(intel_tex->mt->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY);
 397             key->compressed_multisample_layout_mask |= 1 << s;
 398
 399             if (intel_tex->mt->surf.samples >= 16) {
 400                assert(devinfo->gen >= 9);
 401                key->msaa_16 |= 1 << s;
 402             }
 403          }
 404
 405          if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
 406             switch (intel_tex->planar_format->components) {
 407             case __DRI_IMAGE_COMPONENTS_Y_UV:
 408                key->y_uv_image_mask |= 1 << s;
 409                break;
 410             case __DRI_IMAGE_COMPONENTS_Y_U_V:
 411                key->y_u_v_image_mask |= 1 << s;
 412                break;
 413             case __DRI_IMAGE_COMPONENTS_Y_XUXV:
 414                key->yx_xuxv_image_mask |= 1 << s;
 415                break;
 416             case __DRI_IMAGE_COMPONENTS_Y_UXVX:
 417                key->xy_uxvx_image_mask |= 1 << s;
 418                break;
 419             case __DRI_IMAGE_COMPONENTS_AYUV:
 420                key->ayuv_image_mask |= 1 << s;
 421                break;
 422             default:
 423                break;
 424             }
 425          }
 426
 427       }
 428    }
 429 }
 430
 431 static bool
 432 brw_wm_state_dirty(const struct brw_context *brw)
 433 {
 434    return brw_state_dirty(brw,
 435                           _NEW_BUFFERS |
 436                           _NEW_COLOR |
 437                           _NEW_DEPTH |
 438                           _NEW_FRAG_CLAMP |
 439                           _NEW_HINT |
 440                           _NEW_LIGHT |
 441                           _NEW_LINE |
 442                           _NEW_MULTISAMPLE |
 443                           _NEW_POLYGON |
 444                           _NEW_STENCIL |
 445                           _NEW_TEXTURE,
 446                           BRW_NEW_FRAGMENT_PROGRAM |
 447                           BRW_NEW_REDUCED_PRIMITIVE |
 448                           BRW_NEW_STATS_WM |
 449                           BRW_NEW_VUE_MAP_GEOM_OUT);
 450 }
 451
 452 void
 453 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 454 {
 455    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 456    struct gl_context *ctx = &brw->ctx;
 457    /* BRW_NEW_FRAGMENT_PROGRAM */
 458    const struct gl_program *prog = brw->programs[MESA_SHADER_FRAGMENT];
 459    const struct brw_program *fp = brw_program_const(prog);
 460    GLuint lookup = 0;
 461    GLuint line_aa;
 462
 463    memset(key, 0, sizeof(*key));
 464
 465    /* Build the index for table lookup
 466     */
 467    if (devinfo->gen < 6) {
 468       struct intel_renderbuffer *depth_irb =
 469          intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_DEPTH);
 470
 471       /* _NEW_COLOR */
 472       if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) {
 473          lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 474       }
 475
 476       if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 477          lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 478       }
 479
 480       /* _NEW_DEPTH */
 481       if (depth_irb && ctx->Depth.Test) {
 482          lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 483
 484          if (brw_depth_writes_enabled(brw))
 485             lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 486       }
 487
 488       /* _NEW_STENCIL | _NEW_BUFFERS */
 489       if (brw->stencil_enabled) {
 490          lookup |= BRW_WM_IZ_STENCIL_TEST_ENABLE_BIT;
 491
 492          if (ctx->Stencil.WriteMask[0] ||
 493              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 494             lookup |= BRW_WM_IZ_STENCIL_WRITE_ENABLE_BIT;
 495       }
 496       key->iz_lookup = lookup;
 497    }
 498
 499    line_aa = BRW_WM_AA_NEVER;
 500
 501    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 502    if (ctx->Line.SmoothFlag) {
 503       if (brw->reduced_primitive == GL_LINES) {
 504          line_aa = BRW_WM_AA_ALWAYS;
 505       }
 506       else if (brw->reduced_primitive == GL_TRIANGLES) {
 507          if (ctx->Polygon.FrontMode == GL_LINE) {
 508             line_aa = BRW_WM_AA_SOMETIMES;
 509
 510             if (ctx->Polygon.BackMode == GL_LINE ||
 511                 (ctx->Polygon.CullFlag &&
 512                  ctx->Polygon.CullFaceMode == GL_BACK))
 513                line_aa = BRW_WM_AA_ALWAYS;
 514          }
 515          else if (ctx->Polygon.BackMode == GL_LINE) {
 516             line_aa = BRW_WM_AA_SOMETIMES;
 517
 518             if ((ctx->Polygon.CullFlag &&
 519                  ctx->Polygon.CullFaceMode == GL_FRONT))
 520                line_aa = BRW_WM_AA_ALWAYS;
 521          }
 522       }
 523    }
 524
 525    key->line_aa = line_aa;
 526
 527    /* _NEW_HINT */
 528    key->high_quality_derivatives =
 529       prog->info.uses_fddx_fddy &&
 530       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 531
 532    if (devinfo->gen < 6)
 533       key->stats_wm = brw->stats_wm;
 534
 535    /* _NEW_LIGHT */
 536    key->flat_shade =
 537       (prog->info.inputs_read & (VARYING_BIT_COL0 | VARYING_BIT_COL1)) &&
 538       (ctx->Light.ShadeModel == GL_FLAT);
 539
 540    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 541    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 542
 543    /* _NEW_TEXTURE */
 544    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 545
 546    /* _NEW_BUFFERS */
 547    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 548
 549    /* _NEW_COLOR */
 550    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 551       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 552
 553    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 554    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 555       (_mesa_is_alpha_test_enabled(ctx) ||
 556        _mesa_is_alpha_to_coverage_enabled(ctx));
 557
 558    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 559    /* Ignore sample qualifier while computing this flag. */
 560    if (ctx->Multisample.Enabled) {
 561       key->persample_interp =
 562          ctx->Multisample.SampleShading &&
 563          (ctx->Multisample.MinSampleShadingValue *
 564           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 565
 566       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 567    }
 568
 569    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 570    if (devinfo->gen < 6 || util_bitcount64(prog->info.inputs_read &
 571                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 572       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 573    }
 574
 575    /* _NEW_COLOR | _NEW_BUFFERS */
 576    /* Pre-gen6, the hardware alpha test always used each render
 577     * target's alpha to do alpha test, as opposed to render target 0's alpha
 578     * like GL requires.  Fix that by building the alpha test into the
 579     * shader, and we'll skip enabling the fixed function alpha test.
 580     */
 581    if (devinfo->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 582        ctx->Color.AlphaEnabled) {
 583       key->alpha_test_func = ctx->Color.AlphaFunc;
 584       key->alpha_test_ref = ctx->Color.AlphaRef;
 585    }
 586
 587    /* The unique fragment program ID */
 588    key->program_string_id = fp->id;
 589
 590    /* Whether reads from the framebuffer should behave coherently. */
 591    key->coherent_fb_fetch = ctx->Extensions.EXT_shader_framebuffer_fetch;
 592 }
 593
 594 void
 595 brw_upload_wm_prog(struct brw_context *brw)
 596 {
 597    struct brw_wm_prog_key key;
 598    struct brw_program *fp =
 599       (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
 600
 601    if (!brw_wm_state_dirty(brw))
 602       return;
 603
 604    brw_wm_populate_key(brw, &key);
 605
 606    if (brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG, &key, sizeof(key),
 607                         &brw->wm.base.prog_offset, &brw->wm.base.prog_data,
 608                         true))
 609       return;
 610
 611    if (brw_disk_cache_upload_program(brw, MESA_SHADER_FRAGMENT))
 612       return;
 613
 614    fp = (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
 615    fp->id = key.program_string_id;
 616
 617    MAYBE_UNUSED bool success = brw_codegen_wm_prog(brw, fp, &key,
 618                                                    &brw->vue_map_geom_out);
 619    assert(success);
 620 }
 621
 622 void
 623 brw_wm_populate_default_key(const struct gen_device_info *devinfo,
 624                             struct brw_wm_prog_key *key,
 625                             struct gl_program *prog)
 626 {
 627    memset(key, 0, sizeof(*key));
 628
 629    uint64_t outputs_written = prog->info.outputs_written;
 630
 631    if (devinfo->gen < 6) {
 632       if (prog->info.fs.uses_discard)
 633          key->iz_lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 634
 635       if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 636          key->iz_lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 637
 638       /* Just assume depth testing. */
 639       key->iz_lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 640       key->iz_lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 641    }
 642
 643    if (devinfo->gen < 6 || util_bitcount64(prog->info.inputs_read &
 644                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 645       key->input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
 646    }
 647
 648    brw_setup_tex_for_precompile(devinfo, &key->tex, prog);
 649
 650    key->nr_color_regions = util_bitcount64(outputs_written &
 651          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 652            BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
 653            BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 654
 655    key->program_string_id = brw_program(prog)->id;
 656
 657    /* Whether reads from the framebuffer should behave coherently. */
 658    key->coherent_fb_fetch = devinfo->gen >= 9;
 659 }
 660
 661 bool
 662 brw_fs_precompile(struct gl_context *ctx, struct gl_program *prog)
 663 {
 664    struct brw_context *brw = brw_context(ctx);
 665    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 666    struct brw_wm_prog_key key;
 667
 668    struct brw_program *bfp = brw_program(prog);
 669
 670    brw_wm_populate_default_key(&brw->screen->devinfo, &key, prog);
 671
 672    /* check brw_wm_populate_default_key coherent_fb_fetch setting */
 673    assert(key.coherent_fb_fetch ==
 674           ctx->Extensions.EXT_shader_framebuffer_fetch);
 675
 676    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 677    struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data;
 678
 679    struct brw_vue_map vue_map;
 680    if (devinfo->gen < 6) {
 681       brw_compute_vue_map(&brw->screen->devinfo, &vue_map,
 682                           prog->info.inputs_read | VARYING_BIT_POS,
 683                           false);
 684    }
 685
 686    bool success = brw_codegen_wm_prog(brw, bfp, &key, &vue_map);
 687
 688    brw->wm.base.prog_offset = old_prog_offset;
 689    brw->wm.base.prog_data = old_prog_data;
 690
 691    return success;
 692 }