src/mesa/drivers/dri/i965/brw_wm.c

   1 /*
   2  * Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  * Intel funded Tungsten Graphics to
   4  * develop this 3D driver.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining
   7  * a copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sublicense, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial
  16  * portions of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  */
  26 #include "brw_context.h"
  27 #include "brw_wm.h"
  28 #include "brw_state.h"
  29 #include "brw_shader.h"
  30 #include "main/enums.h"
  31 #include "main/formats.h"
  32 #include "main/fbobject.h"
  33 #include "main/samplerobj.h"
  34 #include "main/framebuffer.h"
  35 #include "program/prog_parameter.h"
  36 #include "program/program.h"
  37 #include "intel_mipmap_tree.h"
  38 #include "intel_image.h"
  39 #include "brw_nir.h"
  40 #include "brw_program.h"
  41
  42 #include "util/ralloc.h"
  43
  44 static void
  45 assign_fs_binding_table_offsets(const struct gen_device_info *devinfo,
  46                                 const struct gl_shader_program *shader_prog,
  47                                 const struct gl_program *prog,
  48                                 const struct brw_wm_prog_key *key,
  49                                 struct brw_wm_prog_data *prog_data)
  50 {
  51    uint32_t next_binding_table_offset = 0;
  52
  53    /* If there are no color regions, we still perform an FB write to a null
  54     * renderbuffer, which we place at surface index 0.
  55     */
  56    prog_data->binding_table.render_target_start = next_binding_table_offset;
  57    next_binding_table_offset += MAX2(key->nr_color_regions, 1);
  58
  59    next_binding_table_offset =
  60       brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
  61                                               shader_prog, prog, &prog_data->base,
  62                                               next_binding_table_offset);
  63
  64    if (prog->nir->info->outputs_read && !key->coherent_fb_fetch) {
  65       prog_data->binding_table.render_target_read_start =
  66          next_binding_table_offset;
  67       next_binding_table_offset += key->nr_color_regions;
  68    }
  69 }
  70
  71 /**
  72  * All Mesa program -> GPU code generation goes through this function.
  73  * Depending on the instructions used (i.e. flow control instructions)
  74  * we'll use one of two code generators.
  75  */
  76 bool
  77 brw_codegen_wm_prog(struct brw_context *brw,
  78                     struct gl_shader_program *prog,
  79                     struct brw_program *fp,
  80                     struct brw_wm_prog_key *key,
  81                     struct brw_vue_map *vue_map)
  82 {
  83    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  84    struct gl_context *ctx = &brw->ctx;
  85    void *mem_ctx = ralloc_context(NULL);
  86    struct brw_wm_prog_data prog_data;
  87    const GLuint *program;
  88    struct brw_shader *fs = NULL;
  89    GLuint program_size;
  90    bool start_busy = false;
  91    double start_time = 0;
  92
  93    if (prog)
  94       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  95
  96    memset(&prog_data, 0, sizeof(prog_data));
  97
  98    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
  99    if (!prog)
 100       prog_data.base.use_alt_mode = true;
 101
 102    assign_fs_binding_table_offsets(devinfo, prog, &fp->program, key,
 103                                    &prog_data);
 104
 105    /* Allocate the references to the uniforms that will end up in the
 106     * prog_data associated with the compiled program, and which will be freed
 107     * by the state cache.
 108     */
 109    int param_count = fp->program.nir->num_uniforms / 4;
 110    prog_data.base.nr_image_params = fp->program.info.num_images;
 111    /* The backend also sometimes adds params for texture size. */
 112    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 113    prog_data.base.param =
 114       rzalloc_array(NULL, const gl_constant_value *, param_count);
 115    prog_data.base.pull_param =
 116       rzalloc_array(NULL, const gl_constant_value *, param_count);
 117    prog_data.base.image_param =
 118       rzalloc_array(NULL, struct brw_image_param,
 119                     prog_data.base.nr_image_params);
 120    prog_data.base.nr_params = param_count;
 121
 122    if (prog) {
 123       brw_nir_setup_glsl_uniforms(fp->program.nir, prog, &fp->program,
 124                                   &prog_data.base, true);
 125    } else {
 126       brw_nir_setup_arb_uniforms(fp->program.nir, &fp->program,
 127                                  &prog_data.base);
 128
 129       if (unlikely(INTEL_DEBUG & DEBUG_WM))
 130          brw_dump_arb_asm("fragment", &fp->program);
 131    }
 132
 133    if (unlikely(brw->perf_debug)) {
 134       start_busy = (brw->batch.last_bo &&
 135                     drm_intel_bo_busy(brw->batch.last_bo));
 136       start_time = get_time();
 137    }
 138
 139    int st_index8 = -1, st_index16 = -1;
 140    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 141       st_index8 = brw_get_shader_time_index(brw, prog, &fp->program, ST_FS8);
 142       st_index16 = brw_get_shader_time_index(brw, prog, &fp->program, ST_FS16);
 143    }
 144
 145    char *error_str = NULL;
 146    program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx,
 147                             key, &prog_data, fp->program.nir,
 148                             &fp->program, st_index8, st_index16,
 149                             true, brw->use_rep_send, vue_map,
 150                             &program_size, &error_str);
 151
 152    if (program == NULL) {
 153       if (prog) {
 154          prog->LinkStatus = false;
 155          ralloc_strcat(&prog->InfoLog, error_str);
 156       }
 157
 158       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 159
 160       ralloc_free(mem_ctx);
 161       return false;
 162    }
 163
 164    if (unlikely(brw->perf_debug) && fs) {
 165       if (fs->compiled_once)
 166          brw_wm_debug_recompile(brw, prog, key);
 167       fs->compiled_once = true;
 168
 169       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 170          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 171                     (get_time() - start_time) * 1000);
 172       }
 173    }
 174
 175    brw_alloc_stage_scratch(brw, &brw->wm.base,
 176                            prog_data.base.total_scratch,
 177                            devinfo->max_wm_threads);
 178
 179    if (unlikely((INTEL_DEBUG & DEBUG_WM) && !prog))
 180       fprintf(stderr, "\n");
 181
 182    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 183                     key, sizeof(struct brw_wm_prog_key),
 184                     program, program_size,
 185                     &prog_data, sizeof(prog_data),
 186                     &brw->wm.base.prog_offset, &brw->wm.base.prog_data);
 187
 188    ralloc_free(mem_ctx);
 189
 190    return true;
 191 }
 192
 193 bool
 194 brw_debug_recompile_sampler_key(struct brw_context *brw,
 195                                 const struct brw_sampler_prog_key_data *old_key,
 196                                 const struct brw_sampler_prog_key_data *key)
 197 {
 198    bool found = false;
 199
 200    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 201       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 202                          old_key->swizzles[i], key->swizzles[i]);
 203    }
 204    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 205                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 206    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 207                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 208    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 209                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 210    found |= key_debug(brw, "gather channel quirk on any texture unit",
 211                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 212    found |= key_debug(brw, "compressed multisample layout",
 213                       old_key->compressed_multisample_layout_mask,
 214                       key->compressed_multisample_layout_mask);
 215    found |= key_debug(brw, "16x msaa",
 216                       old_key->msaa_16,
 217                       key->msaa_16);
 218
 219    found |= key_debug(brw, "y_uv image bound",
 220                       old_key->y_uv_image_mask,
 221                       key->y_uv_image_mask);
 222    found |= key_debug(brw, "y_u_v image bound",
 223                       old_key->y_u_v_image_mask,
 224                       key->y_u_v_image_mask);
 225    found |= key_debug(brw, "yx_xuxv image bound",
 226                       old_key->yx_xuxv_image_mask,
 227                       key->yx_xuxv_image_mask);
 228
 229    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 230       found |= key_debug(brw, "textureGather workarounds",
 231                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 232    }
 233
 234    return found;
 235 }
 236
 237 void
 238 brw_wm_debug_recompile(struct brw_context *brw,
 239                        struct gl_shader_program *prog,
 240                        const struct brw_wm_prog_key *key)
 241 {
 242    struct brw_cache_item *c = NULL;
 243    const struct brw_wm_prog_key *old_key = NULL;
 244    bool found = false;
 245
 246    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 247
 248    for (unsigned int i = 0; i < brw->cache.size; i++) {
 249       for (c = brw->cache.items[i]; c; c = c->next) {
 250          if (c->cache_id == BRW_CACHE_FS_PROG) {
 251             old_key = c->key;
 252
 253             if (old_key->program_string_id == key->program_string_id)
 254                break;
 255          }
 256       }
 257       if (c)
 258          break;
 259    }
 260
 261    if (!c) {
 262       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 263       return;
 264    }
 265
 266    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 267                       "depth write",
 268                       old_key->iz_lookup, key->iz_lookup);
 269    found |= key_debug(brw, "depth statistics",
 270                       old_key->stats_wm, key->stats_wm);
 271    found |= key_debug(brw, "flat shading",
 272                       old_key->flat_shade, key->flat_shade);
 273    found |= key_debug(brw, "per-sample interpolation",
 274                       old_key->persample_interp, key->persample_interp);
 275    found |= key_debug(brw, "number of color buffers",
 276                       old_key->nr_color_regions, key->nr_color_regions);
 277    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 278                       old_key->replicate_alpha, key->replicate_alpha);
 279    found |= key_debug(brw, "fragment color clamping",
 280                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 281    found |= key_debug(brw, "multisampled FBO",
 282                       old_key->multisample_fbo, key->multisample_fbo);
 283    found |= key_debug(brw, "line smoothing",
 284                       old_key->line_aa, key->line_aa);
 285    found |= key_debug(brw, "input slots valid",
 286                       old_key->input_slots_valid, key->input_slots_valid);
 287    found |= key_debug(brw, "mrt alpha test function",
 288                       old_key->alpha_test_func, key->alpha_test_func);
 289    found |= key_debug(brw, "mrt alpha test reference value",
 290                       old_key->alpha_test_ref, key->alpha_test_ref);
 291
 292    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 293
 294    if (!found) {
 295       perf_debug("  Something else\n");
 296    }
 297 }
 298
 299 static uint8_t
 300 gen6_gather_workaround(GLenum internalformat)
 301 {
 302    switch (internalformat) {
 303    case GL_R8I: return WA_SIGN | WA_8BIT;
 304    case GL_R8UI: return WA_8BIT;
 305    case GL_R16I: return WA_SIGN | WA_16BIT;
 306    case GL_R16UI: return WA_16BIT;
 307    default:
 308       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 309        * the surface state, there is no shader w/a required.
 310        */
 311       return 0;
 312    }
 313 }
 314
 315 void
 316 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 317                                    const struct gl_program *prog,
 318                                    struct brw_sampler_prog_key_data *key)
 319 {
 320    struct brw_context *brw = brw_context(ctx);
 321    GLbitfield mask = prog->SamplersUsed;
 322
 323    while (mask) {
 324       const int s = u_bit_scan(&mask);
 325
 326       key->swizzles[s] = SWIZZLE_NOOP;
 327
 328       int unit_id = prog->SamplerUnits[s];
 329       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 330
 331       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 332          const struct gl_texture_object *t = unit->_Current;
 333          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 334          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 335
 336          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 337             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 338              img->_BaseFormat == GL_DEPTH_STENCIL);
 339
 340          /* Haswell handles texture swizzling as surface format overrides
 341           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 342           */
 343          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 344             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 345
 346          if (brw->gen < 8 &&
 347              sampler->MinFilter != GL_NEAREST &&
 348              sampler->MagFilter != GL_NEAREST) {
 349             if (sampler->WrapS == GL_CLAMP)
 350                key->gl_clamp_mask[0] |= 1 << s;
 351             if (sampler->WrapT == GL_CLAMP)
 352                key->gl_clamp_mask[1] |= 1 << s;
 353             if (sampler->WrapR == GL_CLAMP)
 354                key->gl_clamp_mask[2] |= 1 << s;
 355          }
 356
 357          /* gather4's channel select for green from RG32F is broken; requires
 358           * a shader w/a on IVB; fixable with just SCS on HSW.
 359           */
 360          if (brw->gen == 7 && !brw->is_haswell &&
 361              prog->nir->info->uses_texture_gather) {
 362             if (img->InternalFormat == GL_RG32F)
 363                key->gather_channel_quirk_mask |= 1 << s;
 364          }
 365
 366          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 367           * UNORM/FLOAT instead and fix it in the shader.
 368           */
 369          if (brw->gen == 6 && prog->nir->info->uses_texture_gather) {
 370             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 371          }
 372
 373          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 374           * then we need to emit slightly different code to first sample the
 375           * MCS surface.
 376           */
 377          struct intel_texture_object *intel_tex =
 378             intel_texture_object((struct gl_texture_object *)t);
 379
 380          /* From gen9 onwards some single sampled buffers can also be
 381           * compressed. These don't need ld2dms sampling along with mcs fetch.
 382           */
 383          if (brw->gen >= 7 &&
 384              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS &&
 385              intel_tex->mt->num_samples > 1) {
 386             key->compressed_multisample_layout_mask |= 1 << s;
 387
 388             if (intel_tex->mt->num_samples >= 16) {
 389                assert(brw->gen >= 9);
 390                key->msaa_16 |= 1 << s;
 391             }
 392          }
 393
 394          if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
 395             switch (intel_tex->planar_format->components) {
 396             case __DRI_IMAGE_COMPONENTS_Y_UV:
 397                key->y_uv_image_mask |= 1 << s;
 398                break;
 399             case __DRI_IMAGE_COMPONENTS_Y_U_V:
 400                key->y_u_v_image_mask |= 1 << s;
 401                break;
 402             case __DRI_IMAGE_COMPONENTS_Y_XUXV:
 403                key->yx_xuxv_image_mask |= 1 << s;
 404                break;
 405             default:
 406                break;
 407             }
 408          }
 409
 410       }
 411    }
 412 }
 413
 414 static bool
 415 brw_wm_state_dirty(const struct brw_context *brw)
 416 {
 417    return brw_state_dirty(brw,
 418                           _NEW_BUFFERS |
 419                           _NEW_COLOR |
 420                           _NEW_DEPTH |
 421                           _NEW_FRAG_CLAMP |
 422                           _NEW_HINT |
 423                           _NEW_LIGHT |
 424                           _NEW_LINE |
 425                           _NEW_MULTISAMPLE |
 426                           _NEW_POLYGON |
 427                           _NEW_STENCIL |
 428                           _NEW_TEXTURE,
 429                           BRW_NEW_FRAGMENT_PROGRAM |
 430                           BRW_NEW_REDUCED_PRIMITIVE |
 431                           BRW_NEW_STATS_WM |
 432                           BRW_NEW_VUE_MAP_GEOM_OUT);
 433 }
 434
 435 void
 436 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 437 {
 438    struct gl_context *ctx = &brw->ctx;
 439    /* BRW_NEW_FRAGMENT_PROGRAM */
 440    const struct brw_program *fp = brw_program_const(brw->fragment_program);
 441    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 442    GLuint lookup = 0;
 443    GLuint line_aa;
 444
 445    memset(key, 0, sizeof(*key));
 446
 447    /* Build the index for table lookup
 448     */
 449    if (brw->gen < 6) {
 450       /* _NEW_COLOR */
 451       if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) {
 452          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 453       }
 454
 455       if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 456          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 457       }
 458
 459       /* _NEW_DEPTH */
 460       if (ctx->Depth.Test)
 461          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 462
 463       if (brw_depth_writes_enabled(brw))
 464          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 465
 466       /* _NEW_STENCIL | _NEW_BUFFERS */
 467       if (ctx->Stencil._Enabled) {
 468          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 469
 470          if (ctx->Stencil.WriteMask[0] ||
 471              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 472             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 473       }
 474       key->iz_lookup = lookup;
 475    }
 476
 477    line_aa = AA_NEVER;
 478
 479    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 480    if (ctx->Line.SmoothFlag) {
 481       if (brw->reduced_primitive == GL_LINES) {
 482          line_aa = AA_ALWAYS;
 483       }
 484       else if (brw->reduced_primitive == GL_TRIANGLES) {
 485          if (ctx->Polygon.FrontMode == GL_LINE) {
 486             line_aa = AA_SOMETIMES;
 487
 488             if (ctx->Polygon.BackMode == GL_LINE ||
 489                 (ctx->Polygon.CullFlag &&
 490                  ctx->Polygon.CullFaceMode == GL_BACK))
 491                line_aa = AA_ALWAYS;
 492          }
 493          else if (ctx->Polygon.BackMode == GL_LINE) {
 494             line_aa = AA_SOMETIMES;
 495
 496             if ((ctx->Polygon.CullFlag &&
 497                  ctx->Polygon.CullFaceMode == GL_FRONT))
 498                line_aa = AA_ALWAYS;
 499          }
 500       }
 501    }
 502
 503    key->line_aa = line_aa;
 504
 505    /* _NEW_HINT */
 506    key->high_quality_derivatives =
 507       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 508
 509    if (brw->gen < 6)
 510       key->stats_wm = brw->stats_wm;
 511
 512    /* _NEW_LIGHT */
 513    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 514
 515    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 516    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 517
 518    /* _NEW_TEXTURE */
 519    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 520
 521    /* _NEW_BUFFERS */
 522    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 523
 524    /* _NEW_COLOR */
 525    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 526       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 527
 528    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 529    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 530       (_mesa_is_alpha_test_enabled(ctx) ||
 531        _mesa_is_alpha_to_coverage_enabled(ctx));
 532
 533    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 534    /* Ignore sample qualifier while computing this flag. */
 535    if (ctx->Multisample.Enabled) {
 536       key->persample_interp =
 537          ctx->Multisample.SampleShading &&
 538          (ctx->Multisample.MinSampleShadingValue *
 539           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 540
 541       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 542    }
 543
 544    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 545    if (brw->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 546                                          BRW_FS_VARYING_INPUT_MASK) > 16) {
 547       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 548    }
 549
 550    /* _NEW_COLOR | _NEW_BUFFERS */
 551    /* Pre-gen6, the hardware alpha test always used each render
 552     * target's alpha to do alpha test, as opposed to render target 0's alpha
 553     * like GL requires.  Fix that by building the alpha test into the
 554     * shader, and we'll skip enabling the fixed function alpha test.
 555     */
 556    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 557        ctx->Color.AlphaEnabled) {
 558       key->alpha_test_func = ctx->Color.AlphaFunc;
 559       key->alpha_test_ref = ctx->Color.AlphaRef;
 560    }
 561
 562    /* The unique fragment program ID */
 563    key->program_string_id = fp->id;
 564
 565    /* Whether reads from the framebuffer should behave coherently. */
 566    key->coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 567 }
 568
 569 void
 570 brw_upload_wm_prog(struct brw_context *brw)
 571 {
 572    struct gl_context *ctx = &brw->ctx;
 573    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 574    struct brw_wm_prog_key key;
 575    struct brw_program *fp = (struct brw_program *) brw->fragment_program;
 576
 577    if (!brw_wm_state_dirty(brw))
 578       return;
 579
 580    brw_wm_populate_key(brw, &key);
 581
 582    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 583                          &key, sizeof(key),
 584                          &brw->wm.base.prog_offset,
 585                          &brw->wm.base.prog_data)) {
 586       bool success = brw_codegen_wm_prog(brw, current, fp, &key,
 587                                          &brw->vue_map_geom_out);
 588       (void) success;
 589       assert(success);
 590    }
 591 }
 592
 593 bool
 594 brw_fs_precompile(struct gl_context *ctx,
 595                   struct gl_shader_program *shader_prog,
 596                   struct gl_program *prog)
 597 {
 598    struct brw_context *brw = brw_context(ctx);
 599    struct brw_wm_prog_key key;
 600
 601    struct brw_program *bfp = brw_program(prog);
 602
 603    memset(&key, 0, sizeof(key));
 604
 605    uint64_t outputs_written = prog->info.outputs_written;
 606
 607    if (brw->gen < 6) {
 608       if (prog->info.fs.uses_discard)
 609          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 610
 611       if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 612          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 613
 614       /* Just assume depth testing. */
 615       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 616       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 617    }
 618
 619    if (brw->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 620                                          BRW_FS_VARYING_INPUT_MASK) > 16) {
 621       key.input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
 622    }
 623
 624    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 625
 626    key.nr_color_regions = _mesa_bitcount_64(outputs_written &
 627          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 628            BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
 629            BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 630
 631    key.program_string_id = bfp->id;
 632
 633    /* Whether reads from the framebuffer should behave coherently. */
 634    key.coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 635
 636    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 637    struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data;
 638
 639    struct brw_vue_map vue_map;
 640    if (brw->gen < 6) {
 641       brw_compute_vue_map(&brw->screen->devinfo, &vue_map,
 642                           prog->info.inputs_read | VARYING_BIT_POS,
 643                           false);
 644    }
 645
 646    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key, &vue_map);
 647
 648    brw->wm.base.prog_offset = old_prog_offset;
 649    brw->wm.base.prog_data = old_prog_data;
 650
 651    return success;
 652 }