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 "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 brw_device_info *devinfo,
  45                                 const struct gl_shader_program *shader_prog,
  46                                 const struct gl_program *prog,
  47                                 const struct brw_wm_prog_key *key,
  48                                 struct brw_wm_prog_data *prog_data)
  49 {
  50    uint32_t next_binding_table_offset = 0;
  51
  52    /* If there are no color regions, we still perform an FB write to a null
  53     * renderbuffer, which we place at surface index 0.
  54     */
  55    prog_data->binding_table.render_target_start = next_binding_table_offset;
  56    next_binding_table_offset += MAX2(key->nr_color_regions, 1);
  57
  58    brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
  59                                            shader_prog, prog, &prog_data->base,
  60                                            next_binding_table_offset);
  61 }
  62
  63 /**
  64  * All Mesa program -> GPU code generation goes through this function.
  65  * Depending on the instructions used (i.e. flow control instructions)
  66  * we'll use one of two code generators.
  67  */
  68 bool
  69 brw_codegen_wm_prog(struct brw_context *brw,
  70                     struct gl_shader_program *prog,
  71                     struct brw_fragment_program *fp,
  72                     struct brw_wm_prog_key *key)
  73 {
  74    struct gl_context *ctx = &brw->ctx;
  75    void *mem_ctx = ralloc_context(NULL);
  76    struct brw_wm_prog_data prog_data;
  77    const GLuint *program;
  78    struct brw_shader *fs = NULL;
  79    GLuint program_size;
  80    bool start_busy = false;
  81    double start_time = 0;
  82
  83    if (prog)
  84       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
  85
  86    memset(&prog_data, 0, sizeof(prog_data));
  87
  88    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
  89    if (!prog)
  90       prog_data.base.use_alt_mode = true;
  91
  92    assign_fs_binding_table_offsets(brw->intelScreen->devinfo, prog,
  93                                    &fp->program.Base, key, &prog_data);
  94
  95    /* Allocate the references to the uniforms that will end up in the
  96     * prog_data associated with the compiled program, and which will be freed
  97     * by the state cache.
  98     */
  99    int param_count = fp->program.Base.nir->num_uniforms;
 100    if (fs)
 101       prog_data.base.nr_image_params = fs->base.NumImages;
 102    /* The backend also sometimes adds params for texture size. */
 103    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 104    prog_data.base.param =
 105       rzalloc_array(NULL, const gl_constant_value *, param_count);
 106    prog_data.base.pull_param =
 107       rzalloc_array(NULL, const gl_constant_value *, param_count);
 108    prog_data.base.image_param =
 109       rzalloc_array(NULL, struct brw_image_param,
 110                     prog_data.base.nr_image_params);
 111    prog_data.base.nr_params = param_count;
 112
 113    if (prog) {
 114       brw_nir_setup_glsl_uniforms(fp->program.Base.nir, prog, &fp->program.Base,
 115                                   &prog_data.base, true);
 116    } else {
 117       brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base,
 118                                  &prog_data.base);
 119    }
 120
 121    if (unlikely(brw->perf_debug)) {
 122       start_busy = (brw->batch.last_bo &&
 123                     drm_intel_bo_busy(brw->batch.last_bo));
 124       start_time = get_time();
 125    }
 126
 127    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 128       brw_dump_ir("fragment", prog, fs ? &fs->base : NULL, &fp->program.Base);
 129
 130    int st_index8 = -1, st_index16 = -1;
 131    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 132       st_index8 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS8);
 133       st_index16 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS16);
 134    }
 135
 136    char *error_str = NULL;
 137    program = brw_compile_fs(brw->intelScreen->compiler, brw, mem_ctx,
 138                             key, &prog_data, fp->program.Base.nir,
 139                             &fp->program.Base, st_index8, st_index16,
 140                             brw->use_rep_send, &program_size, &error_str);
 141    if (program == NULL) {
 142       if (prog) {
 143          prog->LinkStatus = false;
 144          ralloc_strcat(&prog->InfoLog, error_str);
 145       }
 146
 147       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 148
 149       ralloc_free(mem_ctx);
 150       return false;
 151    }
 152
 153    if (unlikely(brw->perf_debug) && fs) {
 154       if (fs->compiled_once)
 155          brw_wm_debug_recompile(brw, prog, key);
 156       fs->compiled_once = true;
 157
 158       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 159          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 160                     (get_time() - start_time) * 1000);
 161       }
 162    }
 163
 164    if (prog_data.base.total_scratch) {
 165       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 166                          prog_data.base.total_scratch * brw->max_wm_threads);
 167    }
 168
 169    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 170       fprintf(stderr, "\n");
 171
 172    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 173                     key, sizeof(struct brw_wm_prog_key),
 174                     program, program_size,
 175                     &prog_data, sizeof(prog_data),
 176                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 177
 178    ralloc_free(mem_ctx);
 179
 180    return true;
 181 }
 182
 183 bool
 184 brw_debug_recompile_sampler_key(struct brw_context *brw,
 185                                 const struct brw_sampler_prog_key_data *old_key,
 186                                 const struct brw_sampler_prog_key_data *key)
 187 {
 188    bool found = false;
 189
 190    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 191       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 192                          old_key->swizzles[i], key->swizzles[i]);
 193    }
 194    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 195                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 196    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 197                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 198    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 199                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 200    found |= key_debug(brw, "gather channel quirk on any texture unit",
 201                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 202    found |= key_debug(brw, "compressed multisample layout",
 203                       old_key->compressed_multisample_layout_mask,
 204                       key->compressed_multisample_layout_mask);
 205    found |= key_debug(brw, "16x msaa",
 206                       old_key->msaa_16,
 207                       key->msaa_16);
 208
 209    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 210       found |= key_debug(brw, "textureGather workarounds",
 211                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 212    }
 213
 214    return found;
 215 }
 216
 217 void
 218 brw_wm_debug_recompile(struct brw_context *brw,
 219                        struct gl_shader_program *prog,
 220                        const struct brw_wm_prog_key *key)
 221 {
 222    struct brw_cache_item *c = NULL;
 223    const struct brw_wm_prog_key *old_key = NULL;
 224    bool found = false;
 225
 226    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 227
 228    for (unsigned int i = 0; i < brw->cache.size; i++) {
 229       for (c = brw->cache.items[i]; c; c = c->next) {
 230          if (c->cache_id == BRW_CACHE_FS_PROG) {
 231             old_key = c->key;
 232
 233             if (old_key->program_string_id == key->program_string_id)
 234                break;
 235          }
 236       }
 237       if (c)
 238          break;
 239    }
 240
 241    if (!c) {
 242       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 243       return;
 244    }
 245
 246    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 247                       "depth write",
 248                       old_key->iz_lookup, key->iz_lookup);
 249    found |= key_debug(brw, "depth statistics",
 250                       old_key->stats_wm, key->stats_wm);
 251    found |= key_debug(brw, "flat shading",
 252                       old_key->flat_shade, key->flat_shade);
 253    found |= key_debug(brw, "per-sample shading",
 254                       old_key->persample_shading, key->persample_shading);
 255    found |= key_debug(brw, "number of color buffers",
 256                       old_key->nr_color_regions, key->nr_color_regions);
 257    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 258                       old_key->replicate_alpha, key->replicate_alpha);
 259    found |= key_debug(brw, "rendering to FBO",
 260                       old_key->render_to_fbo, key->render_to_fbo);
 261    found |= key_debug(brw, "fragment color clamping",
 262                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 263    found |= key_debug(brw, "line smoothing",
 264                       old_key->line_aa, key->line_aa);
 265    found |= key_debug(brw, "renderbuffer height",
 266                       old_key->drawable_height, key->drawable_height);
 267    found |= key_debug(brw, "input slots valid",
 268                       old_key->input_slots_valid, key->input_slots_valid);
 269    found |= key_debug(brw, "mrt alpha test function",
 270                       old_key->alpha_test_func, key->alpha_test_func);
 271    found |= key_debug(brw, "mrt alpha test reference value",
 272                       old_key->alpha_test_ref, key->alpha_test_ref);
 273
 274    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 275
 276    if (!found) {
 277       perf_debug("  Something else\n");
 278    }
 279 }
 280
 281 static uint8_t
 282 gen6_gather_workaround(GLenum internalformat)
 283 {
 284    switch (internalformat) {
 285    case GL_R8I: return WA_SIGN | WA_8BIT;
 286    case GL_R8UI: return WA_8BIT;
 287    case GL_R16I: return WA_SIGN | WA_16BIT;
 288    case GL_R16UI: return WA_16BIT;
 289    default:
 290       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 291        * the surface state, there is no shader w/a required.
 292        */
 293       return 0;
 294    }
 295 }
 296
 297 void
 298 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 299                                    const struct gl_program *prog,
 300                                    unsigned sampler_count,
 301                                    struct brw_sampler_prog_key_data *key)
 302 {
 303    struct brw_context *brw = brw_context(ctx);
 304
 305    for (int s = 0; s < sampler_count; s++) {
 306       key->swizzles[s] = SWIZZLE_NOOP;
 307
 308       if (!(prog->SamplersUsed & (1 << s)))
 309          continue;
 310
 311       int unit_id = prog->SamplerUnits[s];
 312       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 313
 314       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 315          const struct gl_texture_object *t = unit->_Current;
 316          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 317          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 318
 319          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 320             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 321              img->_BaseFormat == GL_DEPTH_STENCIL);
 322
 323          /* Haswell handles texture swizzling as surface format overrides
 324           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 325           */
 326          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 327             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 328
 329          if (brw->gen < 8 &&
 330              sampler->MinFilter != GL_NEAREST &&
 331              sampler->MagFilter != GL_NEAREST) {
 332             if (sampler->WrapS == GL_CLAMP)
 333                key->gl_clamp_mask[0] |= 1 << s;
 334             if (sampler->WrapT == GL_CLAMP)
 335                key->gl_clamp_mask[1] |= 1 << s;
 336             if (sampler->WrapR == GL_CLAMP)
 337                key->gl_clamp_mask[2] |= 1 << s;
 338          }
 339
 340          /* gather4's channel select for green from RG32F is broken; requires
 341           * a shader w/a on IVB; fixable with just SCS on HSW.
 342           */
 343          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 344             if (img->InternalFormat == GL_RG32F)
 345                key->gather_channel_quirk_mask |= 1 << s;
 346          }
 347
 348          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 349           * UNORM/FLOAT instead and fix it in the shader.
 350           */
 351          if (brw->gen == 6 && prog->UsesGather) {
 352             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 353          }
 354
 355          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 356           * then we need to emit slightly different code to first sample the
 357           * MCS surface.
 358           */
 359          struct intel_texture_object *intel_tex =
 360             intel_texture_object((struct gl_texture_object *)t);
 361
 362          if (brw->gen >= 7 &&
 363              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
 364             key->compressed_multisample_layout_mask |= 1 << s;
 365
 366             if (intel_tex->mt->num_samples >= 16) {
 367                assert(brw->gen >= 9);
 368                key->msaa_16 |= 1 << s;
 369             }
 370          }
 371       }
 372    }
 373 }
 374
 375 static bool
 376 brw_wm_state_dirty (struct brw_context *brw)
 377 {
 378    return brw_state_dirty(brw,
 379                           _NEW_BUFFERS |
 380                           _NEW_COLOR |
 381                           _NEW_DEPTH |
 382                           _NEW_FRAG_CLAMP |
 383                           _NEW_HINT |
 384                           _NEW_LIGHT |
 385                           _NEW_LINE |
 386                           _NEW_MULTISAMPLE |
 387                           _NEW_POLYGON |
 388                           _NEW_STENCIL |
 389                           _NEW_TEXTURE,
 390                           BRW_NEW_FRAGMENT_PROGRAM |
 391                           BRW_NEW_REDUCED_PRIMITIVE |
 392                           BRW_NEW_STATS_WM |
 393                           BRW_NEW_VUE_MAP_GEOM_OUT);
 394 }
 395
 396 static void
 397 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 398 {
 399    struct gl_context *ctx = &brw->ctx;
 400    /* BRW_NEW_FRAGMENT_PROGRAM */
 401    const struct brw_fragment_program *fp =
 402       (struct brw_fragment_program *) brw->fragment_program;
 403    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 404    GLuint lookup = 0;
 405    GLuint line_aa;
 406    bool program_uses_dfdy = fp->program.UsesDFdy;
 407    const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 408
 409    memset(key, 0, sizeof(*key));
 410
 411    /* Build the index for table lookup
 412     */
 413    if (brw->gen < 6) {
 414       /* _NEW_COLOR */
 415       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 416          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 417
 418       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 419          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 420
 421       /* _NEW_DEPTH */
 422       if (ctx->Depth.Test)
 423          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 424
 425       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 426          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 427
 428       /* _NEW_STENCIL | _NEW_BUFFERS */
 429       if (ctx->Stencil._Enabled) {
 430          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 431
 432          if (ctx->Stencil.WriteMask[0] ||
 433              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 434             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 435       }
 436       key->iz_lookup = lookup;
 437    }
 438
 439    line_aa = AA_NEVER;
 440
 441    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 442    if (ctx->Line.SmoothFlag) {
 443       if (brw->reduced_primitive == GL_LINES) {
 444          line_aa = AA_ALWAYS;
 445       }
 446       else if (brw->reduced_primitive == GL_TRIANGLES) {
 447          if (ctx->Polygon.FrontMode == GL_LINE) {
 448             line_aa = AA_SOMETIMES;
 449
 450             if (ctx->Polygon.BackMode == GL_LINE ||
 451                 (ctx->Polygon.CullFlag &&
 452                  ctx->Polygon.CullFaceMode == GL_BACK))
 453                line_aa = AA_ALWAYS;
 454          }
 455          else if (ctx->Polygon.BackMode == GL_LINE) {
 456             line_aa = AA_SOMETIMES;
 457
 458             if ((ctx->Polygon.CullFlag &&
 459                  ctx->Polygon.CullFaceMode == GL_FRONT))
 460                line_aa = AA_ALWAYS;
 461          }
 462       }
 463    }
 464
 465    key->line_aa = line_aa;
 466
 467    /* _NEW_HINT */
 468    key->high_quality_derivatives =
 469       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 470
 471    if (brw->gen < 6)
 472       key->stats_wm = brw->stats_wm;
 473
 474    /* _NEW_LIGHT */
 475    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 476
 477    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 478    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 479
 480    /* _NEW_TEXTURE */
 481    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 482                                       &key->tex);
 483
 484    /* _NEW_BUFFERS */
 485    /*
 486     * Include the draw buffer origin and height so that we can calculate
 487     * fragment position values relative to the bottom left of the drawable,
 488     * from the incoming screen origin relative position we get as part of our
 489     * payload.
 490     *
 491     * This is only needed for the WM_WPOSXY opcode when the fragment program
 492     * uses the gl_FragCoord input.
 493     *
 494     * We could avoid recompiling by including this as a constant referenced by
 495     * our program, but if we were to do that it would also be nice to handle
 496     * getting that constant updated at batchbuffer submit time (when we
 497     * hold the lock and know where the buffer really is) rather than at emit
 498     * time when we don't hold the lock and are just guessing.  We could also
 499     * just avoid using this as key data if the program doesn't use
 500     * fragment.position.
 501     *
 502     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 503     * drawable height in order to invert the Y axis.
 504     */
 505    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 506       key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
 507    }
 508
 509    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 510       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 511    }
 512
 513    /* _NEW_BUFFERS */
 514    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 515
 516    /* _NEW_COLOR */
 517    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 518       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 519
 520    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 521    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 522       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 523
 524    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 525    /* Ignore sample qualifier while computing this flag. */
 526    key->persample_shading =
 527       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
 528
 529    key->compute_pos_offset =
 530       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
 531       fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
 532
 533    key->compute_sample_id =
 534       multisample_fbo &&
 535       ctx->Multisample.Enabled &&
 536       (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
 537
 538    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 539    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 540                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 541       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 542
 543
 544    /* _NEW_COLOR | _NEW_BUFFERS */
 545    /* Pre-gen6, the hardware alpha test always used each render
 546     * target's alpha to do alpha test, as opposed to render target 0's alpha
 547     * like GL requires.  Fix that by building the alpha test into the
 548     * shader, and we'll skip enabling the fixed function alpha test.
 549     */
 550    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 551        ctx->Color.AlphaEnabled) {
 552       key->alpha_test_func = ctx->Color.AlphaFunc;
 553       key->alpha_test_ref = ctx->Color.AlphaRef;
 554    }
 555
 556    /* The unique fragment program ID */
 557    key->program_string_id = fp->id;
 558 }
 559
 560 void
 561 brw_upload_wm_prog(struct brw_context *brw)
 562 {
 563    struct gl_context *ctx = &brw->ctx;
 564    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 565    struct brw_wm_prog_key key;
 566    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 567       brw->fragment_program;
 568
 569    if (!brw_wm_state_dirty(brw))
 570       return;
 571
 572    brw_wm_populate_key(brw, &key);
 573
 574    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 575                          &key, sizeof(key),
 576                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 577       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 578       (void) success;
 579       assert(success);
 580    }
 581    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 582 }
 583
 584 bool
 585 brw_fs_precompile(struct gl_context *ctx,
 586                   struct gl_shader_program *shader_prog,
 587                   struct gl_program *prog)
 588 {
 589    struct brw_context *brw = brw_context(ctx);
 590    struct brw_wm_prog_key key;
 591
 592    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 593    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 594    bool program_uses_dfdy = fp->UsesDFdy;
 595
 596    memset(&key, 0, sizeof(key));
 597
 598    if (brw->gen < 6) {
 599       if (fp->UsesKill)
 600          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 601
 602       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 603          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 604
 605       /* Just assume depth testing. */
 606       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 607       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 608    }
 609
 610    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 611                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 612       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 613
 614    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 615
 616    if (fp->Base.InputsRead & VARYING_BIT_POS) {
 617       key.drawable_height = ctx->DrawBuffer->Height;
 618    }
 619
 620    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 621          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 622          BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 623
 624    if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 625       key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer) ||
 626                           key.nr_color_regions > 1;
 627    }
 628
 629    key.program_string_id = bfp->id;
 630
 631    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 632    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 633
 634    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 635
 636    brw->wm.base.prog_offset = old_prog_offset;
 637    brw->wm.prog_data = old_prog_data;
 638
 639    return success;
 640 }