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, st_index32 = -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       st_index32 = brw_get_shader_time_index(brw, &fp->program, ST_FS32,
 174                                              !fp->program.is_arb_asm);
 175    }
 176
 177    char *error_str = NULL;
 178    program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx,
 179                             key, &prog_data, fp->program.nir,
 180                             &fp->program, st_index8, st_index16, st_index32,
 181                             true, false, vue_map,
 182                             &error_str);
 183
 184    if (program == NULL) {
 185       if (!fp->program.is_arb_asm) {
 186          fp->program.sh.data->LinkStatus = LINKING_FAILURE;
 187          ralloc_strcat(&fp->program.sh.data->InfoLog, error_str);
 188       }
 189
 190       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 191
 192       ralloc_free(mem_ctx);
 193       return false;
 194    }
 195
 196    if (unlikely(brw->perf_debug)) {
 197       if (fp->compiled_once)
 198          brw_wm_debug_recompile(brw, &fp->program, key);
 199       fp->compiled_once = true;
 200
 201       if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
 202          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 203                     (get_time() - start_time) * 1000);
 204       }
 205    }
 206
 207    brw_alloc_stage_scratch(brw, &brw->wm.base, prog_data.base.total_scratch);
 208
 209    if (unlikely((INTEL_DEBUG & DEBUG_WM) && fp->program.is_arb_asm))
 210       fprintf(stderr, "\n");
 211
 212    /* The param and pull_param arrays will be freed by the shader cache. */
 213    ralloc_steal(NULL, prog_data.base.param);
 214    ralloc_steal(NULL, prog_data.base.pull_param);
 215    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 216                     key, sizeof(struct brw_wm_prog_key),
 217                     program, prog_data.base.program_size,
 218                     &prog_data, sizeof(prog_data),
 219                     &brw->wm.base.prog_offset, &brw->wm.base.prog_data);
 220
 221    ralloc_free(mem_ctx);
 222
 223    return true;
 224 }
 225
 226 bool
 227 brw_debug_recompile_sampler_key(struct brw_context *brw,
 228                                 const struct brw_sampler_prog_key_data *old_key,
 229                                 const struct brw_sampler_prog_key_data *key)
 230 {
 231    bool found = false;
 232
 233    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 234       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 235                          old_key->swizzles[i], key->swizzles[i]);
 236    }
 237    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 238                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 239    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 240                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 241    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 242                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 243    found |= key_debug(brw, "gather channel quirk on any texture unit",
 244                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 245    found |= key_debug(brw, "compressed multisample layout",
 246                       old_key->compressed_multisample_layout_mask,
 247                       key->compressed_multisample_layout_mask);
 248    found |= key_debug(brw, "16x msaa",
 249                       old_key->msaa_16,
 250                       key->msaa_16);
 251
 252    found |= key_debug(brw, "y_uv image bound",
 253                       old_key->y_uv_image_mask,
 254                       key->y_uv_image_mask);
 255    found |= key_debug(brw, "y_u_v image bound",
 256                       old_key->y_u_v_image_mask,
 257                       key->y_u_v_image_mask);
 258    found |= key_debug(brw, "yx_xuxv image bound",
 259                       old_key->yx_xuxv_image_mask,
 260                       key->yx_xuxv_image_mask);
 261    found |= key_debug(brw, "xy_uxvx image bound",
 262                       old_key->xy_uxvx_image_mask,
 263                       key->xy_uxvx_image_mask);
 264
 265
 266    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 267       found |= key_debug(brw, "textureGather workarounds",
 268                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 269    }
 270
 271    return found;
 272 }
 273
 274 static uint8_t
 275 gen6_gather_workaround(GLenum internalformat)
 276 {
 277    switch (internalformat) {
 278    case GL_R8I: return WA_SIGN | WA_8BIT;
 279    case GL_R8UI: return WA_8BIT;
 280    case GL_R16I: return WA_SIGN | WA_16BIT;
 281    case GL_R16UI: return WA_16BIT;
 282    default:
 283       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 284        * the surface state, there is no shader w/a required.
 285        */
 286       return 0;
 287    }
 288 }
 289
 290 void
 291 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 292                                    const struct gl_program *prog,
 293                                    struct brw_sampler_prog_key_data *key)
 294 {
 295    struct brw_context *brw = brw_context(ctx);
 296    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 297    GLbitfield mask = prog->SamplersUsed;
 298
 299    while (mask) {
 300       const int s = u_bit_scan(&mask);
 301
 302       key->swizzles[s] = SWIZZLE_NOOP;
 303
 304       int unit_id = prog->SamplerUnits[s];
 305       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 306
 307       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 308          const struct gl_texture_object *t = unit->_Current;
 309          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 310          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 311
 312          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 313             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 314              img->_BaseFormat == GL_DEPTH_STENCIL);
 315
 316          /* Haswell handles texture swizzling as surface format overrides
 317           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 318           */
 319          if (alpha_depth || (devinfo->gen < 8 && !devinfo->is_haswell))
 320             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 321
 322          if (devinfo->gen < 8 &&
 323              sampler->MinFilter != GL_NEAREST &&
 324              sampler->MagFilter != GL_NEAREST) {
 325             if (sampler->WrapS == GL_CLAMP)
 326                key->gl_clamp_mask[0] |= 1 << s;
 327             if (sampler->WrapT == GL_CLAMP)
 328                key->gl_clamp_mask[1] |= 1 << s;
 329             if (sampler->WrapR == GL_CLAMP)
 330                key->gl_clamp_mask[2] |= 1 << s;
 331          }
 332
 333          /* gather4 for RG32* is broken in multiple ways on Gen7. */
 334          if (devinfo->gen == 7 && prog->info.uses_texture_gather) {
 335             switch (img->InternalFormat) {
 336             case GL_RG32I:
 337             case GL_RG32UI: {
 338                /* We have to override the format to R32G32_FLOAT_LD.
 339                 * This means that SCS_ALPHA and SCS_ONE will return 0x3f8
 340                 * (1.0) rather than integer 1.  This needs shader hacks.
 341                 *
 342                 * On Ivybridge, we whack W (alpha) to ONE in our key's
 343                 * swizzle.  On Haswell, we look at the original texture
 344                 * swizzle, and use XYZW with channels overridden to ONE,
 345                 * leaving normal texture swizzling to SCS.
 346                 */
 347                unsigned src_swizzle =
 348                   devinfo->is_haswell ? t->_Swizzle : key->swizzles[s];
 349                for (int i = 0; i < 4; i++) {
 350                   unsigned src_comp = GET_SWZ(src_swizzle, i);
 351                   if (src_comp == SWIZZLE_ONE || src_comp == SWIZZLE_W) {
 352                      key->swizzles[i] &= ~(0x7 << (3 * i));
 353                      key->swizzles[i] |= SWIZZLE_ONE << (3 * i);
 354                   }
 355                }
 356                /* fallthrough */
 357             }
 358             case GL_RG32F:
 359                /* The channel select for green doesn't work - we have to
 360                 * request blue.  Haswell can use SCS for this, but Ivybridge
 361                 * needs a shader workaround.
 362                 */
 363                if (!devinfo->is_haswell)
 364                   key->gather_channel_quirk_mask |= 1 << s;
 365                break;
 366             }
 367          }
 368
 369          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 370           * UNORM/FLOAT instead and fix it in the shader.
 371           */
 372          if (devinfo->gen == 6 && prog->info.uses_texture_gather) {
 373             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 374          }
 375
 376          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 377           * then we need to emit slightly different code to first sample the
 378           * MCS surface.
 379           */
 380          struct intel_texture_object *intel_tex =
 381             intel_texture_object((struct gl_texture_object *)t);
 382
 383          /* From gen9 onwards some single sampled buffers can also be
 384           * compressed. These don't need ld2dms sampling along with mcs fetch.
 385           */
 386          if (intel_tex->mt->aux_usage == ISL_AUX_USAGE_MCS) {
 387             assert(devinfo->gen >= 7);
 388             assert(intel_tex->mt->surf.samples > 1);
 389             assert(intel_tex->mt->aux_buf);
 390             assert(intel_tex->mt->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY);
 391             key->compressed_multisample_layout_mask |= 1 << s;
 392
 393             if (intel_tex->mt->surf.samples >= 16) {
 394                assert(devinfo->gen >= 9);
 395                key->msaa_16 |= 1 << s;
 396             }
 397          }
 398
 399          if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
 400             switch (intel_tex->planar_format->components) {
 401             case __DRI_IMAGE_COMPONENTS_Y_UV:
 402                key->y_uv_image_mask |= 1 << s;
 403                break;
 404             case __DRI_IMAGE_COMPONENTS_Y_U_V:
 405                key->y_u_v_image_mask |= 1 << s;
 406                break;
 407             case __DRI_IMAGE_COMPONENTS_Y_XUXV:
 408                key->yx_xuxv_image_mask |= 1 << s;
 409                break;
 410             case __DRI_IMAGE_COMPONENTS_Y_UXVX:
 411                key->xy_uxvx_image_mask |= 1 << s;
 412                break;
 413             default:
 414                break;
 415             }
 416          }
 417
 418       }
 419    }
 420 }
 421
 422 static bool
 423 brw_wm_state_dirty(const struct brw_context *brw)
 424 {
 425    return brw_state_dirty(brw,
 426                           _NEW_BUFFERS |
 427                           _NEW_COLOR |
 428                           _NEW_DEPTH |
 429                           _NEW_FRAG_CLAMP |
 430                           _NEW_HINT |
 431                           _NEW_LIGHT |
 432                           _NEW_LINE |
 433                           _NEW_MULTISAMPLE |
 434                           _NEW_POLYGON |
 435                           _NEW_STENCIL |
 436                           _NEW_TEXTURE,
 437                           BRW_NEW_FRAGMENT_PROGRAM |
 438                           BRW_NEW_REDUCED_PRIMITIVE |
 439                           BRW_NEW_STATS_WM |
 440                           BRW_NEW_VUE_MAP_GEOM_OUT);
 441 }
 442
 443 void
 444 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 445 {
 446    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 447    struct gl_context *ctx = &brw->ctx;
 448    /* BRW_NEW_FRAGMENT_PROGRAM */
 449    const struct gl_program *prog = brw->programs[MESA_SHADER_FRAGMENT];
 450    const struct brw_program *fp = brw_program_const(prog);
 451    GLuint lookup = 0;
 452    GLuint line_aa;
 453
 454    memset(key, 0, sizeof(*key));
 455
 456    /* Build the index for table lookup
 457     */
 458    if (devinfo->gen < 6) {
 459       /* _NEW_COLOR */
 460       if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) {
 461          lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 462       }
 463
 464       if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 465          lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 466       }
 467
 468       /* _NEW_DEPTH */
 469       if (ctx->Depth.Test)
 470          lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 471
 472       if (brw_depth_writes_enabled(brw))
 473          lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 474
 475       /* _NEW_STENCIL | _NEW_BUFFERS */
 476       if (brw->stencil_enabled) {
 477          lookup |= BRW_WM_IZ_STENCIL_TEST_ENABLE_BIT;
 478
 479          if (ctx->Stencil.WriteMask[0] ||
 480              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 481             lookup |= BRW_WM_IZ_STENCIL_WRITE_ENABLE_BIT;
 482       }
 483       key->iz_lookup = lookup;
 484    }
 485
 486    line_aa = BRW_WM_AA_NEVER;
 487
 488    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 489    if (ctx->Line.SmoothFlag) {
 490       if (brw->reduced_primitive == GL_LINES) {
 491          line_aa = BRW_WM_AA_ALWAYS;
 492       }
 493       else if (brw->reduced_primitive == GL_TRIANGLES) {
 494          if (ctx->Polygon.FrontMode == GL_LINE) {
 495             line_aa = BRW_WM_AA_SOMETIMES;
 496
 497             if (ctx->Polygon.BackMode == GL_LINE ||
 498                 (ctx->Polygon.CullFlag &&
 499                  ctx->Polygon.CullFaceMode == GL_BACK))
 500                line_aa = BRW_WM_AA_ALWAYS;
 501          }
 502          else if (ctx->Polygon.BackMode == GL_LINE) {
 503             line_aa = BRW_WM_AA_SOMETIMES;
 504
 505             if ((ctx->Polygon.CullFlag &&
 506                  ctx->Polygon.CullFaceMode == GL_FRONT))
 507                line_aa = BRW_WM_AA_ALWAYS;
 508          }
 509       }
 510    }
 511
 512    key->line_aa = line_aa;
 513
 514    /* _NEW_HINT */
 515    key->high_quality_derivatives =
 516       prog->info.uses_fddx_fddy &&
 517       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 518
 519    if (devinfo->gen < 6)
 520       key->stats_wm = brw->stats_wm;
 521
 522    /* _NEW_LIGHT */
 523    key->flat_shade =
 524       (prog->info.inputs_read & (VARYING_BIT_COL0 | VARYING_BIT_COL1)) &&
 525       (ctx->Light.ShadeModel == GL_FLAT);
 526
 527    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 528    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 529
 530    /* _NEW_TEXTURE */
 531    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 532
 533    /* _NEW_BUFFERS */
 534    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 535
 536    /* _NEW_COLOR */
 537    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 538       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 539
 540    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 541    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 542       (_mesa_is_alpha_test_enabled(ctx) ||
 543        _mesa_is_alpha_to_coverage_enabled(ctx));
 544
 545    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 546    /* Ignore sample qualifier while computing this flag. */
 547    if (ctx->Multisample.Enabled) {
 548       key->persample_interp =
 549          ctx->Multisample.SampleShading &&
 550          (ctx->Multisample.MinSampleShadingValue *
 551           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 552
 553       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 554    }
 555
 556    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 557    if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 558                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 559       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 560    }
 561
 562    /* _NEW_COLOR | _NEW_BUFFERS */
 563    /* Pre-gen6, the hardware alpha test always used each render
 564     * target's alpha to do alpha test, as opposed to render target 0's alpha
 565     * like GL requires.  Fix that by building the alpha test into the
 566     * shader, and we'll skip enabling the fixed function alpha test.
 567     */
 568    if (devinfo->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 569        ctx->Color.AlphaEnabled) {
 570       key->alpha_test_func = ctx->Color.AlphaFunc;
 571       key->alpha_test_ref = ctx->Color.AlphaRef;
 572    }
 573
 574    /* The unique fragment program ID */
 575    key->program_string_id = fp->id;
 576
 577    /* Whether reads from the framebuffer should behave coherently. */
 578    key->coherent_fb_fetch = ctx->Extensions.EXT_shader_framebuffer_fetch;
 579 }
 580
 581 void
 582 brw_upload_wm_prog(struct brw_context *brw)
 583 {
 584    struct brw_wm_prog_key key;
 585    struct brw_program *fp =
 586       (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
 587
 588    if (!brw_wm_state_dirty(brw))
 589       return;
 590
 591    brw_wm_populate_key(brw, &key);
 592
 593    if (brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 594                         &key, sizeof(key),
 595                         &brw->wm.base.prog_offset,
 596                         &brw->wm.base.prog_data))
 597       return;
 598
 599    if (brw_disk_cache_upload_program(brw, MESA_SHADER_FRAGMENT))
 600       return;
 601
 602    fp = (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
 603    fp->id = key.program_string_id;
 604
 605    MAYBE_UNUSED bool success = brw_codegen_wm_prog(brw, fp, &key,
 606                                                    &brw->vue_map_geom_out);
 607    assert(success);
 608 }
 609
 610 void
 611 brw_wm_populate_default_key(const struct gen_device_info *devinfo,
 612                             struct brw_wm_prog_key *key,
 613                             struct gl_program *prog)
 614 {
 615    memset(key, 0, sizeof(*key));
 616
 617    uint64_t outputs_written = prog->info.outputs_written;
 618
 619    if (devinfo->gen < 6) {
 620       if (prog->info.fs.uses_discard)
 621          key->iz_lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 622
 623       if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 624          key->iz_lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 625
 626       /* Just assume depth testing. */
 627       key->iz_lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 628       key->iz_lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 629    }
 630
 631    if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 632                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 633       key->input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
 634    }
 635
 636    brw_setup_tex_for_precompile(devinfo, &key->tex, prog);
 637
 638    key->nr_color_regions = _mesa_bitcount_64(outputs_written &
 639          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 640            BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
 641            BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 642
 643    key->program_string_id = brw_program(prog)->id;
 644
 645    /* Whether reads from the framebuffer should behave coherently. */
 646    key->coherent_fb_fetch = devinfo->gen >= 9;
 647 }
 648
 649 bool
 650 brw_fs_precompile(struct gl_context *ctx, struct gl_program *prog)
 651 {
 652    struct brw_context *brw = brw_context(ctx);
 653    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 654    struct brw_wm_prog_key key;
 655
 656    struct brw_program *bfp = brw_program(prog);
 657
 658    brw_wm_populate_default_key(&brw->screen->devinfo, &key, prog);
 659
 660    /* check brw_wm_populate_default_key coherent_fb_fetch setting */
 661    assert(key.coherent_fb_fetch ==
 662           ctx->Extensions.EXT_shader_framebuffer_fetch);
 663
 664    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 665    struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data;
 666
 667    struct brw_vue_map vue_map;
 668    if (devinfo->gen < 6) {
 669       brw_compute_vue_map(&brw->screen->devinfo, &vue_map,
 670                           prog->info.inputs_read | VARYING_BIT_POS,
 671                           false);
 672    }
 673
 674    bool success = brw_codegen_wm_prog(brw, bfp, &key, &vue_map);
 675
 676    brw->wm.base.prog_offset = old_prog_offset;
 677    brw->wm.base.prog_data = old_prog_data;
 678
 679    return success;
 680 }