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