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 "glsl/nir/nir.h"
  39
  40 #include "util/ralloc.h"
  41
  42 /**
  43  * Return a bitfield where bit n is set if barycentric interpolation mode n
  44  * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
  45  */
  46 static unsigned
  47 brw_compute_barycentric_interp_modes(struct brw_context *brw,
  48                                      bool shade_model_flat,
  49                                      bool persample_shading,
  50                                      const struct gl_fragment_program *fprog)
  51 {
  52    unsigned barycentric_interp_modes = 0;
  53    int attr;
  54
  55    /* Loop through all fragment shader inputs to figure out what interpolation
  56     * modes are in use, and set the appropriate bits in
  57     * barycentric_interp_modes.
  58     */
  59    for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
  60       enum glsl_interp_qualifier interp_qualifier =
  61          fprog->InterpQualifier[attr];
  62       bool is_centroid = (fprog->IsCentroid & BITFIELD64_BIT(attr)) &&
  63          !persample_shading;
  64       bool is_sample = (fprog->IsSample & BITFIELD64_BIT(attr)) ||
  65          persample_shading;
  66       bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
  67
  68       /* Ignore unused inputs. */
  69       if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
  70          continue;
  71
  72       /* Ignore WPOS and FACE, because they don't require interpolation. */
  73       if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
  74          continue;
  75
  76       /* Determine the set (or sets) of barycentric coordinates needed to
  77        * interpolate this variable.  Note that when
  78        * brw->needs_unlit_centroid_workaround is set, centroid interpolation
  79        * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
  80        * for lit pixels, so we need both sets of barycentric coordinates.
  81        */
  82       if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
  83          if (is_centroid) {
  84             barycentric_interp_modes |=
  85                1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
  86          } else if (is_sample) {
  87             barycentric_interp_modes |=
  88                1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC;
  89          }
  90          if ((!is_centroid && !is_sample) ||
  91              brw->needs_unlit_centroid_workaround) {
  92             barycentric_interp_modes |=
  93                1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
  94          }
  95       } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
  96                  (!(shade_model_flat && is_gl_Color) &&
  97                   interp_qualifier == INTERP_QUALIFIER_NONE)) {
  98          if (is_centroid) {
  99             barycentric_interp_modes |=
 100                1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
 101          } else if (is_sample) {
 102             barycentric_interp_modes |=
 103                1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC;
 104          }
 105          if ((!is_centroid && !is_sample) ||
 106              brw->needs_unlit_centroid_workaround) {
 107             barycentric_interp_modes |=
 108                1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
 109          }
 110       }
 111    }
 112
 113    return barycentric_interp_modes;
 114 }
 115
 116 static uint8_t
 117 computed_depth_mode(struct gl_fragment_program *fp)
 118 {
 119    if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 120       switch (fp->FragDepthLayout) {
 121       case FRAG_DEPTH_LAYOUT_NONE:
 122       case FRAG_DEPTH_LAYOUT_ANY:
 123          return BRW_PSCDEPTH_ON;
 124       case FRAG_DEPTH_LAYOUT_GREATER:
 125          return BRW_PSCDEPTH_ON_GE;
 126       case FRAG_DEPTH_LAYOUT_LESS:
 127          return BRW_PSCDEPTH_ON_LE;
 128       case FRAG_DEPTH_LAYOUT_UNCHANGED:
 129          return BRW_PSCDEPTH_OFF;
 130       }
 131    }
 132    return BRW_PSCDEPTH_OFF;
 133 }
 134
 135 static void
 136 assign_fs_binding_table_offsets(const struct brw_device_info *devinfo,
 137                                 const struct gl_shader_program *shader_prog,
 138                                 const struct gl_program *prog,
 139                                 const struct brw_wm_prog_key *key,
 140                                 struct brw_wm_prog_data *prog_data)
 141 {
 142    uint32_t next_binding_table_offset = 0;
 143
 144    /* If there are no color regions, we still perform an FB write to a null
 145     * renderbuffer, which we place at surface index 0.
 146     */
 147    prog_data->binding_table.render_target_start = next_binding_table_offset;
 148    next_binding_table_offset += MAX2(key->nr_color_regions, 1);
 149
 150    brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
 151                                            shader_prog, prog, &prog_data->base,
 152                                            next_binding_table_offset);
 153 }
 154
 155 /**
 156  * All Mesa program -> GPU code generation goes through this function.
 157  * Depending on the instructions used (i.e. flow control instructions)
 158  * we'll use one of two code generators.
 159  */
 160 bool
 161 brw_codegen_wm_prog(struct brw_context *brw,
 162                     struct gl_shader_program *prog,
 163                     struct brw_fragment_program *fp,
 164                     struct brw_wm_prog_key *key)
 165 {
 166    struct gl_context *ctx = &brw->ctx;
 167    void *mem_ctx = ralloc_context(NULL);
 168    struct brw_wm_prog_data prog_data;
 169    const GLuint *program;
 170    struct brw_shader *fs = NULL;
 171    GLuint program_size;
 172    bool start_busy = false;
 173    double start_time = 0;
 174
 175    if (prog)
 176       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 177
 178    memset(&prog_data, 0, sizeof(prog_data));
 179    /* key->alpha_test_func means simulating alpha testing via discards,
 180     * so the shader definitely kills pixels.
 181     */
 182    prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
 183    prog_data.uses_omask =
 184       fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK);
 185    prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
 186
 187    prog_data.early_fragment_tests = fs && fs->base.EarlyFragmentTests;
 188
 189    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 190    if (!prog)
 191       prog_data.base.use_alt_mode = true;
 192
 193    assign_fs_binding_table_offsets(brw->intelScreen->devinfo, prog,
 194                                    &fp->program.Base, key, &prog_data);
 195
 196    /* Allocate the references to the uniforms that will end up in the
 197     * prog_data associated with the compiled program, and which will be freed
 198     * by the state cache.
 199     */
 200    int param_count = fp->program.Base.nir->num_uniforms;
 201    if (fs)
 202       prog_data.base.nr_image_params = fs->base.NumImages;
 203    /* The backend also sometimes adds params for texture size. */
 204    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 205    prog_data.base.param =
 206       rzalloc_array(NULL, const gl_constant_value *, param_count);
 207    prog_data.base.pull_param =
 208       rzalloc_array(NULL, const gl_constant_value *, param_count);
 209    prog_data.base.image_param =
 210       rzalloc_array(NULL, struct brw_image_param,
 211                     prog_data.base.nr_image_params);
 212    prog_data.base.nr_params = param_count;
 213
 214    prog_data.barycentric_interp_modes =
 215       brw_compute_barycentric_interp_modes(brw, key->flat_shade,
 216                                            key->persample_shading,
 217                                            &fp->program);
 218
 219    if (unlikely(brw->perf_debug)) {
 220       start_busy = (brw->batch.last_bo &&
 221                     drm_intel_bo_busy(brw->batch.last_bo));
 222       start_time = get_time();
 223    }
 224
 225    program = brw_wm_fs_emit(brw, mem_ctx, key, &prog_data,
 226                             &fp->program, prog, &program_size);
 227    if (program == NULL) {
 228       ralloc_free(mem_ctx);
 229       return false;
 230    }
 231
 232    if (unlikely(brw->perf_debug) && fs) {
 233       if (fs->compiled_once)
 234          brw_wm_debug_recompile(brw, prog, key);
 235       fs->compiled_once = true;
 236
 237       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 238          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 239                     (get_time() - start_time) * 1000);
 240       }
 241    }
 242
 243    if (prog_data.base.total_scratch) {
 244       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 245                          prog_data.base.total_scratch * brw->max_wm_threads);
 246    }
 247
 248    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 249       fprintf(stderr, "\n");
 250
 251    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 252                     key, sizeof(struct brw_wm_prog_key),
 253                     program, program_size,
 254                     &prog_data, sizeof(prog_data),
 255                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 256
 257    ralloc_free(mem_ctx);
 258
 259    return true;
 260 }
 261
 262 static bool
 263 key_debug(struct brw_context *brw, const char *name, int a, int b)
 264 {
 265    if (a != b) {
 266       perf_debug("  %s %d->%d\n", name, a, b);
 267       return true;
 268    } else {
 269       return false;
 270    }
 271 }
 272
 273 bool
 274 brw_debug_recompile_sampler_key(struct brw_context *brw,
 275                                 const struct brw_sampler_prog_key_data *old_key,
 276                                 const struct brw_sampler_prog_key_data *key)
 277 {
 278    bool found = false;
 279
 280    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 281       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 282                          old_key->swizzles[i], key->swizzles[i]);
 283    }
 284    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 285                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 286    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 287                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 288    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 289                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 290    found |= key_debug(brw, "gather channel quirk on any texture unit",
 291                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 292    found |= key_debug(brw, "compressed multisample layout",
 293                       old_key->compressed_multisample_layout_mask,
 294                       key->compressed_multisample_layout_mask);
 295
 296    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 297       found |= key_debug(brw, "textureGather workarounds",
 298                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 299    }
 300
 301    return found;
 302 }
 303
 304 void
 305 brw_wm_debug_recompile(struct brw_context *brw,
 306                        struct gl_shader_program *prog,
 307                        const struct brw_wm_prog_key *key)
 308 {
 309    struct brw_cache_item *c = NULL;
 310    const struct brw_wm_prog_key *old_key = NULL;
 311    bool found = false;
 312
 313    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 314
 315    for (unsigned int i = 0; i < brw->cache.size; i++) {
 316       for (c = brw->cache.items[i]; c; c = c->next) {
 317          if (c->cache_id == BRW_CACHE_FS_PROG) {
 318             old_key = c->key;
 319
 320             if (old_key->program_string_id == key->program_string_id)
 321                break;
 322          }
 323       }
 324       if (c)
 325          break;
 326    }
 327
 328    if (!c) {
 329       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 330       return;
 331    }
 332
 333    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 334                       "depth write",
 335                       old_key->iz_lookup, key->iz_lookup);
 336    found |= key_debug(brw, "depth statistics",
 337                       old_key->stats_wm, key->stats_wm);
 338    found |= key_debug(brw, "flat shading",
 339                       old_key->flat_shade, key->flat_shade);
 340    found |= key_debug(brw, "per-sample shading",
 341                       old_key->persample_shading, key->persample_shading);
 342    found |= key_debug(brw, "per-sample shading and 2x MSAA",
 343                       old_key->persample_2x, key->persample_2x);
 344    found |= key_debug(brw, "number of color buffers",
 345                       old_key->nr_color_regions, key->nr_color_regions);
 346    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 347                       old_key->replicate_alpha, key->replicate_alpha);
 348    found |= key_debug(brw, "rendering to FBO",
 349                       old_key->render_to_fbo, key->render_to_fbo);
 350    found |= key_debug(brw, "fragment color clamping",
 351                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 352    found |= key_debug(brw, "line smoothing",
 353                       old_key->line_aa, key->line_aa);
 354    found |= key_debug(brw, "renderbuffer height",
 355                       old_key->drawable_height, key->drawable_height);
 356    found |= key_debug(brw, "input slots valid",
 357                       old_key->input_slots_valid, key->input_slots_valid);
 358    found |= key_debug(brw, "mrt alpha test function",
 359                       old_key->alpha_test_func, key->alpha_test_func);
 360    found |= key_debug(brw, "mrt alpha test reference value",
 361                       old_key->alpha_test_ref, key->alpha_test_ref);
 362
 363    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 364
 365    if (!found) {
 366       perf_debug("  Something else\n");
 367    }
 368 }
 369
 370 static uint8_t
 371 gen6_gather_workaround(GLenum internalformat)
 372 {
 373    switch (internalformat) {
 374    case GL_R8I: return WA_SIGN | WA_8BIT;
 375    case GL_R8UI: return WA_8BIT;
 376    case GL_R16I: return WA_SIGN | WA_16BIT;
 377    case GL_R16UI: return WA_16BIT;
 378    default:
 379       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 380        * the surface state, there is no shader w/a required.
 381        */
 382       return 0;
 383    }
 384 }
 385
 386 void
 387 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 388                                    const struct gl_program *prog,
 389                                    unsigned sampler_count,
 390                                    struct brw_sampler_prog_key_data *key)
 391 {
 392    struct brw_context *brw = brw_context(ctx);
 393
 394    for (int s = 0; s < sampler_count; s++) {
 395       key->swizzles[s] = SWIZZLE_NOOP;
 396
 397       if (!(prog->SamplersUsed & (1 << s)))
 398          continue;
 399
 400       int unit_id = prog->SamplerUnits[s];
 401       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 402
 403       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 404          const struct gl_texture_object *t = unit->_Current;
 405          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 406          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 407
 408          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 409             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 410              img->_BaseFormat == GL_DEPTH_STENCIL);
 411
 412          /* Haswell handles texture swizzling as surface format overrides
 413           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 414           */
 415          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 416             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 417
 418          if (brw->gen < 8 &&
 419              sampler->MinFilter != GL_NEAREST &&
 420              sampler->MagFilter != GL_NEAREST) {
 421             if (sampler->WrapS == GL_CLAMP)
 422                key->gl_clamp_mask[0] |= 1 << s;
 423             if (sampler->WrapT == GL_CLAMP)
 424                key->gl_clamp_mask[1] |= 1 << s;
 425             if (sampler->WrapR == GL_CLAMP)
 426                key->gl_clamp_mask[2] |= 1 << s;
 427          }
 428
 429          /* gather4's channel select for green from RG32F is broken; requires
 430           * a shader w/a on IVB; fixable with just SCS on HSW.
 431           */
 432          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 433             if (img->InternalFormat == GL_RG32F)
 434                key->gather_channel_quirk_mask |= 1 << s;
 435          }
 436
 437          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 438           * UNORM/FLOAT instead and fix it in the shader.
 439           */
 440          if (brw->gen == 6 && prog->UsesGather) {
 441             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 442          }
 443
 444          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 445           * then we need to emit slightly different code to first sample the
 446           * MCS surface.
 447           */
 448          struct intel_texture_object *intel_tex =
 449             intel_texture_object((struct gl_texture_object *)t);
 450
 451          if (brw->gen >= 7 &&
 452              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
 453             key->compressed_multisample_layout_mask |= 1 << s;
 454          }
 455       }
 456    }
 457 }
 458
 459 static bool
 460 brw_wm_state_dirty (struct brw_context *brw)
 461 {
 462    return brw_state_dirty(brw,
 463                           _NEW_BUFFERS |
 464                           _NEW_COLOR |
 465                           _NEW_DEPTH |
 466                           _NEW_FRAG_CLAMP |
 467                           _NEW_HINT |
 468                           _NEW_LIGHT |
 469                           _NEW_LINE |
 470                           _NEW_MULTISAMPLE |
 471                           _NEW_POLYGON |
 472                           _NEW_STENCIL |
 473                           _NEW_TEXTURE,
 474                           BRW_NEW_FRAGMENT_PROGRAM |
 475                           BRW_NEW_REDUCED_PRIMITIVE |
 476                           BRW_NEW_STATS_WM |
 477                           BRW_NEW_VUE_MAP_GEOM_OUT);
 478 }
 479
 480 static void
 481 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 482 {
 483    struct gl_context *ctx = &brw->ctx;
 484    /* BRW_NEW_FRAGMENT_PROGRAM */
 485    const struct brw_fragment_program *fp =
 486       (struct brw_fragment_program *) brw->fragment_program;
 487    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 488    GLuint lookup = 0;
 489    GLuint line_aa;
 490    bool program_uses_dfdy = fp->program.UsesDFdy;
 491    const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 492
 493    memset(key, 0, sizeof(*key));
 494
 495    /* Build the index for table lookup
 496     */
 497    if (brw->gen < 6) {
 498       /* _NEW_COLOR */
 499       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 500          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 501
 502       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 503          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 504
 505       /* _NEW_DEPTH */
 506       if (ctx->Depth.Test)
 507          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 508
 509       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 510          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 511
 512       /* _NEW_STENCIL | _NEW_BUFFERS */
 513       if (ctx->Stencil._Enabled) {
 514          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 515
 516          if (ctx->Stencil.WriteMask[0] ||
 517              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 518             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 519       }
 520       key->iz_lookup = lookup;
 521    }
 522
 523    line_aa = AA_NEVER;
 524
 525    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 526    if (ctx->Line.SmoothFlag) {
 527       if (brw->reduced_primitive == GL_LINES) {
 528          line_aa = AA_ALWAYS;
 529       }
 530       else if (brw->reduced_primitive == GL_TRIANGLES) {
 531          if (ctx->Polygon.FrontMode == GL_LINE) {
 532             line_aa = AA_SOMETIMES;
 533
 534             if (ctx->Polygon.BackMode == GL_LINE ||
 535                 (ctx->Polygon.CullFlag &&
 536                  ctx->Polygon.CullFaceMode == GL_BACK))
 537                line_aa = AA_ALWAYS;
 538          }
 539          else if (ctx->Polygon.BackMode == GL_LINE) {
 540             line_aa = AA_SOMETIMES;
 541
 542             if ((ctx->Polygon.CullFlag &&
 543                  ctx->Polygon.CullFaceMode == GL_FRONT))
 544                line_aa = AA_ALWAYS;
 545          }
 546       }
 547    }
 548
 549    key->line_aa = line_aa;
 550
 551    /* _NEW_HINT */
 552    key->high_quality_derivatives =
 553       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 554
 555    if (brw->gen < 6)
 556       key->stats_wm = brw->stats_wm;
 557
 558    /* _NEW_LIGHT */
 559    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 560
 561    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 562    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 563
 564    /* _NEW_TEXTURE */
 565    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 566                                       &key->tex);
 567
 568    /* _NEW_BUFFERS */
 569    /*
 570     * Include the draw buffer origin and height so that we can calculate
 571     * fragment position values relative to the bottom left of the drawable,
 572     * from the incoming screen origin relative position we get as part of our
 573     * payload.
 574     *
 575     * This is only needed for the WM_WPOSXY opcode when the fragment program
 576     * uses the gl_FragCoord input.
 577     *
 578     * We could avoid recompiling by including this as a constant referenced by
 579     * our program, but if we were to do that it would also be nice to handle
 580     * getting that constant updated at batchbuffer submit time (when we
 581     * hold the lock and know where the buffer really is) rather than at emit
 582     * time when we don't hold the lock and are just guessing.  We could also
 583     * just avoid using this as key data if the program doesn't use
 584     * fragment.position.
 585     *
 586     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 587     * drawable height in order to invert the Y axis.
 588     */
 589    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 590       key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
 591    }
 592
 593    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 594       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 595    }
 596
 597    /* _NEW_BUFFERS */
 598    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 599
 600    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 601    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 602       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 603
 604    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 605    /* Ignore sample qualifier while computing this flag. */
 606    key->persample_shading =
 607       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
 608    if (key->persample_shading)
 609       key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
 610
 611    key->compute_pos_offset =
 612       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
 613       fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
 614
 615    key->compute_sample_id =
 616       multisample_fbo &&
 617       ctx->Multisample.Enabled &&
 618       (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
 619
 620    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 621    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 622                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 623       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 624
 625
 626    /* _NEW_COLOR | _NEW_BUFFERS */
 627    /* Pre-gen6, the hardware alpha test always used each render
 628     * target's alpha to do alpha test, as opposed to render target 0's alpha
 629     * like GL requires.  Fix that by building the alpha test into the
 630     * shader, and we'll skip enabling the fixed function alpha test.
 631     */
 632    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 633        ctx->Color.AlphaEnabled) {
 634       key->alpha_test_func = ctx->Color.AlphaFunc;
 635       key->alpha_test_ref = ctx->Color.AlphaRef;
 636    }
 637
 638    /* The unique fragment program ID */
 639    key->program_string_id = fp->id;
 640 }
 641
 642 void
 643 brw_upload_wm_prog(struct brw_context *brw)
 644 {
 645    struct gl_context *ctx = &brw->ctx;
 646    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 647    struct brw_wm_prog_key key;
 648    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 649       brw->fragment_program;
 650
 651    if (!brw_wm_state_dirty(brw))
 652       return;
 653
 654    brw_wm_populate_key(brw, &key);
 655
 656    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 657                          &key, sizeof(key),
 658                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 659       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 660       (void) success;
 661       assert(success);
 662    }
 663    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 664 }
 665
 666 bool
 667 brw_fs_precompile(struct gl_context *ctx,
 668                   struct gl_shader_program *shader_prog,
 669                   struct gl_program *prog)
 670 {
 671    struct brw_context *brw = brw_context(ctx);
 672    struct brw_wm_prog_key key;
 673
 674    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 675    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 676    bool program_uses_dfdy = fp->UsesDFdy;
 677
 678    memset(&key, 0, sizeof(key));
 679
 680    if (brw->gen < 6) {
 681       if (fp->UsesKill)
 682          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 683
 684       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 685          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 686
 687       /* Just assume depth testing. */
 688       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 689       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 690    }
 691
 692    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 693                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 694       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 695
 696    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 697
 698    if (fp->Base.InputsRead & VARYING_BIT_POS) {
 699       key.drawable_height = ctx->DrawBuffer->Height;
 700    }
 701
 702    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 703          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 704          BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 705
 706    if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 707       key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer) ||
 708                           key.nr_color_regions > 1;
 709    }
 710
 711    key.program_string_id = bfp->id;
 712
 713    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 714    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 715
 716    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 717
 718    brw->wm.base.prog_offset = old_prog_offset;
 719    brw->wm.prog_data = old_prog_data;
 720
 721    return success;
 722 }