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    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  83    struct gl_context *ctx = &brw->ctx;
  84    void *mem_ctx = ralloc_context(NULL);
  85    struct brw_wm_prog_data prog_data;
  86    const GLuint *program;
  87    struct brw_shader *fs = NULL;
  88    GLuint program_size;
  89    bool start_busy = false;
  90    double start_time = 0;
  91
  92    if (prog)
  93       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  94
  95    memset(&prog_data, 0, sizeof(prog_data));
  96
  97    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
  98    if (!prog)
  99       prog_data.base.use_alt_mode = true;
 100
 101    assign_fs_binding_table_offsets(devinfo, prog,
 102                                    &fp->program.Base, key, &prog_data);
 103
 104    /* Allocate the references to the uniforms that will end up in the
 105     * prog_data associated with the compiled program, and which will be freed
 106     * by the state cache.
 107     */
 108    int param_count = fp->program.Base.nir->num_uniforms / 4;
 109    if (fs)
 110       prog_data.base.nr_image_params = fs->base.NumImages;
 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.Base.nir, prog, &fp->program.Base,
 124                                   &prog_data.base, true);
 125    } else {
 126       brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base,
 127                                  &prog_data.base);
 128    }
 129
 130    if (unlikely(brw->perf_debug)) {
 131       start_busy = (brw->batch.last_bo &&
 132                     drm_intel_bo_busy(brw->batch.last_bo));
 133       start_time = get_time();
 134    }
 135
 136    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 137       brw_dump_ir("fragment", prog, fs ? &fs->base : NULL, &fp->program.Base);
 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.Base, ST_FS8);
 142       st_index16 = brw_get_shader_time_index(brw, prog, &fp->program.Base, 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.Base.nir,
 148                             &fp->program.Base, st_index8, st_index16,
 149                             true, brw->use_rep_send,
 150                             &program_size, &error_str);
 151    if (program == NULL) {
 152       if (prog) {
 153          prog->LinkStatus = false;
 154          ralloc_strcat(&prog->InfoLog, error_str);
 155       }
 156
 157       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 158
 159       ralloc_free(mem_ctx);
 160       return false;
 161    }
 162
 163    if (unlikely(brw->perf_debug) && fs) {
 164       if (fs->compiled_once)
 165          brw_wm_debug_recompile(brw, prog, key);
 166       fs->compiled_once = true;
 167
 168       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 169          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 170                     (get_time() - start_time) * 1000);
 171       }
 172    }
 173
 174    brw_alloc_stage_scratch(brw, &brw->wm.base,
 175                            prog_data.base.total_scratch,
 176                            devinfo->max_wm_threads);
 177
 178    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 179       fprintf(stderr, "\n");
 180
 181    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 182                     key, sizeof(struct brw_wm_prog_key),
 183                     program, program_size,
 184                     &prog_data, sizeof(prog_data),
 185                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 186
 187    ralloc_free(mem_ctx);
 188
 189    return true;
 190 }
 191
 192 bool
 193 brw_debug_recompile_sampler_key(struct brw_context *brw,
 194                                 const struct brw_sampler_prog_key_data *old_key,
 195                                 const struct brw_sampler_prog_key_data *key)
 196 {
 197    bool found = false;
 198
 199    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 200       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 201                          old_key->swizzles[i], key->swizzles[i]);
 202    }
 203    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 204                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 205    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 206                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 207    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 208                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 209    found |= key_debug(brw, "gather channel quirk on any texture unit",
 210                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 211    found |= key_debug(brw, "compressed multisample layout",
 212                       old_key->compressed_multisample_layout_mask,
 213                       key->compressed_multisample_layout_mask);
 214    found |= key_debug(brw, "16x msaa",
 215                       old_key->msaa_16,
 216                       key->msaa_16);
 217
 218    found |= key_debug(brw, "y_uv image bound",
 219                       old_key->y_uv_image_mask,
 220                       key->y_uv_image_mask);
 221    found |= key_debug(brw, "y_u_v image bound",
 222                       old_key->y_u_v_image_mask,
 223                       key->y_u_v_image_mask);
 224    found |= key_debug(brw, "yx_xuxv image bound",
 225                       old_key->yx_xuxv_image_mask,
 226                       key->yx_xuxv_image_mask);
 227
 228    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 229       found |= key_debug(brw, "textureGather workarounds",
 230                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 231    }
 232
 233    return found;
 234 }
 235
 236 void
 237 brw_wm_debug_recompile(struct brw_context *brw,
 238                        struct gl_shader_program *prog,
 239                        const struct brw_wm_prog_key *key)
 240 {
 241    struct brw_cache_item *c = NULL;
 242    const struct brw_wm_prog_key *old_key = NULL;
 243    bool found = false;
 244
 245    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 246
 247    for (unsigned int i = 0; i < brw->cache.size; i++) {
 248       for (c = brw->cache.items[i]; c; c = c->next) {
 249          if (c->cache_id == BRW_CACHE_FS_PROG) {
 250             old_key = c->key;
 251
 252             if (old_key->program_string_id == key->program_string_id)
 253                break;
 254          }
 255       }
 256       if (c)
 257          break;
 258    }
 259
 260    if (!c) {
 261       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 262       return;
 263    }
 264
 265    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 266                       "depth write",
 267                       old_key->iz_lookup, key->iz_lookup);
 268    found |= key_debug(brw, "depth statistics",
 269                       old_key->stats_wm, key->stats_wm);
 270    found |= key_debug(brw, "flat shading",
 271                       old_key->flat_shade, key->flat_shade);
 272    found |= key_debug(brw, "per-sample interpolation",
 273                       old_key->persample_interp, key->persample_interp);
 274    found |= key_debug(brw, "number of color buffers",
 275                       old_key->nr_color_regions, key->nr_color_regions);
 276    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 277                       old_key->replicate_alpha, key->replicate_alpha);
 278    found |= key_debug(brw, "fragment color clamping",
 279                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 280    found |= key_debug(brw, "multisampled FBO",
 281                       old_key->multisample_fbo, key->multisample_fbo);
 282    found |= key_debug(brw, "line smoothing",
 283                       old_key->line_aa, key->line_aa);
 284    found |= key_debug(brw, "input slots valid",
 285                       old_key->input_slots_valid, key->input_slots_valid);
 286    found |= key_debug(brw, "mrt alpha test function",
 287                       old_key->alpha_test_func, key->alpha_test_func);
 288    found |= key_debug(brw, "mrt alpha test reference value",
 289                       old_key->alpha_test_ref, key->alpha_test_ref);
 290
 291    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 292
 293    if (!found) {
 294       perf_debug("  Something else\n");
 295    }
 296 }
 297
 298 static uint8_t
 299 gen6_gather_workaround(GLenum internalformat)
 300 {
 301    switch (internalformat) {
 302    case GL_R8I: return WA_SIGN | WA_8BIT;
 303    case GL_R8UI: return WA_8BIT;
 304    case GL_R16I: return WA_SIGN | WA_16BIT;
 305    case GL_R16UI: return WA_16BIT;
 306    default:
 307       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 308        * the surface state, there is no shader w/a required.
 309        */
 310       return 0;
 311    }
 312 }
 313
 314 void
 315 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 316                                    const struct gl_program *prog,
 317                                    struct brw_sampler_prog_key_data *key)
 318 {
 319    struct brw_context *brw = brw_context(ctx);
 320    GLbitfield mask = prog->SamplersUsed;
 321
 322    while (mask) {
 323       const int s = u_bit_scan(&mask);
 324
 325       key->swizzles[s] = SWIZZLE_NOOP;
 326
 327       int unit_id = prog->SamplerUnits[s];
 328       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 329
 330       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 331          const struct gl_texture_object *t = unit->_Current;
 332          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 333          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 334
 335          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 336             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 337              img->_BaseFormat == GL_DEPTH_STENCIL);
 338
 339          /* Haswell handles texture swizzling as surface format overrides
 340           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 341           */
 342          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 343             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 344
 345          if (brw->gen < 8 &&
 346              sampler->MinFilter != GL_NEAREST &&
 347              sampler->MagFilter != GL_NEAREST) {
 348             if (sampler->WrapS == GL_CLAMP)
 349                key->gl_clamp_mask[0] |= 1 << s;
 350             if (sampler->WrapT == GL_CLAMP)
 351                key->gl_clamp_mask[1] |= 1 << s;
 352             if (sampler->WrapR == GL_CLAMP)
 353                key->gl_clamp_mask[2] |= 1 << s;
 354          }
 355
 356          /* gather4's channel select for green from RG32F is broken; requires
 357           * a shader w/a on IVB; fixable with just SCS on HSW.
 358           */
 359          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 360             if (img->InternalFormat == GL_RG32F)
 361                key->gather_channel_quirk_mask |= 1 << s;
 362          }
 363
 364          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 365           * UNORM/FLOAT instead and fix it in the shader.
 366           */
 367          if (brw->gen == 6 && prog->UsesGather) {
 368             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 369          }
 370
 371          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 372           * then we need to emit slightly different code to first sample the
 373           * MCS surface.
 374           */
 375          struct intel_texture_object *intel_tex =
 376             intel_texture_object((struct gl_texture_object *)t);
 377
 378          /* From gen9 onwards some single sampled buffers can also be
 379           * compressed. These don't need ld2dms sampling along with mcs fetch.
 380           */
 381          if (brw->gen >= 7 &&
 382              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS &&
 383              intel_tex->mt->num_samples > 1) {
 384             key->compressed_multisample_layout_mask |= 1 << s;
 385
 386             if (intel_tex->mt->num_samples >= 16) {
 387                assert(brw->gen >= 9);
 388                key->msaa_16 |= 1 << s;
 389             }
 390          }
 391
 392          if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
 393             switch (intel_tex->planar_format->components) {
 394             case __DRI_IMAGE_COMPONENTS_Y_UV:
 395                key->y_uv_image_mask |= 1 << s;
 396                break;
 397             case __DRI_IMAGE_COMPONENTS_Y_U_V:
 398                key->y_u_v_image_mask |= 1 << s;
 399                break;
 400             case __DRI_IMAGE_COMPONENTS_Y_XUXV:
 401                key->yx_xuxv_image_mask |= 1 << s;
 402                break;
 403             default:
 404                break;
 405             }
 406          }
 407
 408       }
 409    }
 410 }
 411
 412 static bool
 413 brw_wm_state_dirty(const struct brw_context *brw)
 414 {
 415    return brw_state_dirty(brw,
 416                           _NEW_BUFFERS |
 417                           _NEW_COLOR |
 418                           _NEW_DEPTH |
 419                           _NEW_FRAG_CLAMP |
 420                           _NEW_HINT |
 421                           _NEW_LIGHT |
 422                           _NEW_LINE |
 423                           _NEW_MULTISAMPLE |
 424                           _NEW_POLYGON |
 425                           _NEW_STENCIL |
 426                           _NEW_TEXTURE,
 427                           BRW_NEW_FRAGMENT_PROGRAM |
 428                           BRW_NEW_REDUCED_PRIMITIVE |
 429                           BRW_NEW_STATS_WM |
 430                           BRW_NEW_VUE_MAP_GEOM_OUT);
 431 }
 432
 433 static void
 434 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 435 {
 436    struct gl_context *ctx = &brw->ctx;
 437    /* BRW_NEW_FRAGMENT_PROGRAM */
 438    const struct brw_fragment_program *fp =
 439       (struct brw_fragment_program *) brw->fragment_program;
 440    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 441    GLuint lookup = 0;
 442    GLuint line_aa;
 443
 444    memset(key, 0, sizeof(*key));
 445
 446    /* Build the index for table lookup
 447     */
 448    if (brw->gen < 6) {
 449       /* _NEW_COLOR */
 450       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 451          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 452
 453       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 454          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 455
 456       /* _NEW_DEPTH */
 457       if (ctx->Depth.Test)
 458          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 459
 460       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 461          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 462
 463       /* _NEW_STENCIL | _NEW_BUFFERS */
 464       if (ctx->Stencil._Enabled) {
 465          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 466
 467          if (ctx->Stencil.WriteMask[0] ||
 468              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 469             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 470       }
 471       key->iz_lookup = lookup;
 472    }
 473
 474    line_aa = AA_NEVER;
 475
 476    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 477    if (ctx->Line.SmoothFlag) {
 478       if (brw->reduced_primitive == GL_LINES) {
 479          line_aa = AA_ALWAYS;
 480       }
 481       else if (brw->reduced_primitive == GL_TRIANGLES) {
 482          if (ctx->Polygon.FrontMode == GL_LINE) {
 483             line_aa = AA_SOMETIMES;
 484
 485             if (ctx->Polygon.BackMode == GL_LINE ||
 486                 (ctx->Polygon.CullFlag &&
 487                  ctx->Polygon.CullFaceMode == GL_BACK))
 488                line_aa = AA_ALWAYS;
 489          }
 490          else if (ctx->Polygon.BackMode == GL_LINE) {
 491             line_aa = AA_SOMETIMES;
 492
 493             if ((ctx->Polygon.CullFlag &&
 494                  ctx->Polygon.CullFaceMode == GL_FRONT))
 495                line_aa = AA_ALWAYS;
 496          }
 497       }
 498    }
 499
 500    key->line_aa = line_aa;
 501
 502    /* _NEW_HINT */
 503    key->high_quality_derivatives =
 504       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 505
 506    if (brw->gen < 6)
 507       key->stats_wm = brw->stats_wm;
 508
 509    /* _NEW_LIGHT */
 510    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 511
 512    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 513    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 514
 515    /* _NEW_TEXTURE */
 516    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 517
 518    /* _NEW_BUFFERS */
 519    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 520
 521    /* _NEW_COLOR */
 522    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 523       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 524
 525    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 526    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 527       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 528
 529    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 530    /* Ignore sample qualifier while computing this flag. */
 531    if (ctx->Multisample.Enabled) {
 532       key->persample_interp =
 533          ctx->Multisample.SampleShading &&
 534          (ctx->Multisample.MinSampleShadingValue *
 535           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 536
 537       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 538    }
 539
 540    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 541    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 542                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 543       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 544
 545
 546    /* _NEW_COLOR | _NEW_BUFFERS */
 547    /* Pre-gen6, the hardware alpha test always used each render
 548     * target's alpha to do alpha test, as opposed to render target 0's alpha
 549     * like GL requires.  Fix that by building the alpha test into the
 550     * shader, and we'll skip enabling the fixed function alpha test.
 551     */
 552    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 553        ctx->Color.AlphaEnabled) {
 554       key->alpha_test_func = ctx->Color.AlphaFunc;
 555       key->alpha_test_ref = ctx->Color.AlphaRef;
 556    }
 557
 558    /* The unique fragment program ID */
 559    key->program_string_id = fp->id;
 560
 561    /* Whether reads from the framebuffer should behave coherently. */
 562    key->coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 563 }
 564
 565 void
 566 brw_upload_wm_prog(struct brw_context *brw)
 567 {
 568    struct gl_context *ctx = &brw->ctx;
 569    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 570    struct brw_wm_prog_key key;
 571    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 572       brw->fragment_program;
 573
 574    if (!brw_wm_state_dirty(brw))
 575       return;
 576
 577    brw_wm_populate_key(brw, &key);
 578
 579    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 580                          &key, sizeof(key),
 581                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 582       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 583       (void) success;
 584       assert(success);
 585    }
 586    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 587 }
 588
 589 bool
 590 brw_fs_precompile(struct gl_context *ctx,
 591                   struct gl_shader_program *shader_prog,
 592                   struct gl_program *prog)
 593 {
 594    struct brw_context *brw = brw_context(ctx);
 595    struct brw_wm_prog_key key;
 596
 597    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 598    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 599
 600    memset(&key, 0, sizeof(key));
 601
 602    if (brw->gen < 6) {
 603       if (fp->UsesKill)
 604          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 605
 606       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 607          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 608
 609       /* Just assume depth testing. */
 610       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 611       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 612    }
 613
 614    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 615                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 616       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 617
 618    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 619
 620    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 621          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 622            BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
 623            BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 624
 625    key.program_string_id = bfp->id;
 626
 627    /* Whether reads from the framebuffer should behave coherently. */
 628    key.coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 629
 630    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 631    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 632
 633    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 634
 635    brw->wm.base.prog_offset = old_prog_offset;
 636    brw->wm.prog_data = old_prog_data;
 637
 638    return success;
 639 }