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
  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(const struct brw_device_info *devinfo,
  48                                      bool shade_model_flat,
  49                                      bool persample_shading,
  50                                      nir_shader *shader)
  51 {
  52    unsigned barycentric_interp_modes = 0;
  53
  54    nir_foreach_variable(var, &shader->inputs) {
  55       enum glsl_interp_qualifier interp_qualifier = var->data.interpolation;
  56       bool is_centroid = var->data.centroid && !persample_shading;
  57       bool is_sample = var->data.sample || persample_shading;
  58       bool is_gl_Color = (var->data.location == VARYING_SLOT_COL0) ||
  59                          (var->data.location == VARYING_SLOT_COL1);
  60
  61       /* Ignore WPOS and FACE, because they don't require interpolation. */
  62       if (var->data.location == VARYING_SLOT_POS ||
  63           var->data.location == VARYING_SLOT_FACE)
  64          continue;
  65
  66       /* Determine the set (or sets) of barycentric coordinates needed to
  67        * interpolate this variable.  Note that when
  68        * brw->needs_unlit_centroid_workaround is set, centroid interpolation
  69        * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
  70        * for lit pixels, so we need both sets of barycentric coordinates.
  71        */
  72       if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
  73          if (is_centroid) {
  74             barycentric_interp_modes |=
  75                1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
  76          } else if (is_sample) {
  77             barycentric_interp_modes |=
  78                1 << BRW_WM_NONPERSPECTIVE_SAMPLE_BARYCENTRIC;
  79          }
  80          if ((!is_centroid && !is_sample) ||
  81              devinfo->needs_unlit_centroid_workaround) {
  82             barycentric_interp_modes |=
  83                1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
  84          }
  85       } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
  86                  (!(shade_model_flat && is_gl_Color) &&
  87                   interp_qualifier == INTERP_QUALIFIER_NONE)) {
  88          if (is_centroid) {
  89             barycentric_interp_modes |=
  90                1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
  91          } else if (is_sample) {
  92             barycentric_interp_modes |=
  93                1 << BRW_WM_PERSPECTIVE_SAMPLE_BARYCENTRIC;
  94          }
  95          if ((!is_centroid && !is_sample) ||
  96              devinfo->needs_unlit_centroid_workaround) {
  97             barycentric_interp_modes |=
  98                1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
  99          }
 100       }
 101    }
 102
 103    return barycentric_interp_modes;
 104 }
 105
 106 static uint8_t
 107 computed_depth_mode(struct gl_fragment_program *fp)
 108 {
 109    if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 110       switch (fp->FragDepthLayout) {
 111       case FRAG_DEPTH_LAYOUT_NONE:
 112       case FRAG_DEPTH_LAYOUT_ANY:
 113          return BRW_PSCDEPTH_ON;
 114       case FRAG_DEPTH_LAYOUT_GREATER:
 115          return BRW_PSCDEPTH_ON_GE;
 116       case FRAG_DEPTH_LAYOUT_LESS:
 117          return BRW_PSCDEPTH_ON_LE;
 118       case FRAG_DEPTH_LAYOUT_UNCHANGED:
 119          return BRW_PSCDEPTH_OFF;
 120       }
 121    }
 122    return BRW_PSCDEPTH_OFF;
 123 }
 124
 125 static void
 126 assign_fs_binding_table_offsets(const struct brw_device_info *devinfo,
 127                                 const struct gl_shader_program *shader_prog,
 128                                 const struct gl_program *prog,
 129                                 const struct brw_wm_prog_key *key,
 130                                 struct brw_wm_prog_data *prog_data)
 131 {
 132    uint32_t next_binding_table_offset = 0;
 133
 134    /* If there are no color regions, we still perform an FB write to a null
 135     * renderbuffer, which we place at surface index 0.
 136     */
 137    prog_data->binding_table.render_target_start = next_binding_table_offset;
 138    next_binding_table_offset += MAX2(key->nr_color_regions, 1);
 139
 140    brw_assign_common_binding_table_offsets(MESA_SHADER_FRAGMENT, devinfo,
 141                                            shader_prog, prog, &prog_data->base,
 142                                            next_binding_table_offset);
 143 }
 144
 145 /**
 146  * All Mesa program -> GPU code generation goes through this function.
 147  * Depending on the instructions used (i.e. flow control instructions)
 148  * we'll use one of two code generators.
 149  */
 150 bool
 151 brw_codegen_wm_prog(struct brw_context *brw,
 152                     struct gl_shader_program *prog,
 153                     struct brw_fragment_program *fp,
 154                     struct brw_wm_prog_key *key)
 155 {
 156    struct gl_context *ctx = &brw->ctx;
 157    void *mem_ctx = ralloc_context(NULL);
 158    struct brw_wm_prog_data prog_data;
 159    const GLuint *program;
 160    struct brw_shader *fs = NULL;
 161    GLuint program_size;
 162    bool start_busy = false;
 163    double start_time = 0;
 164
 165    if (prog)
 166       fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 167
 168    memset(&prog_data, 0, sizeof(prog_data));
 169    /* key->alpha_test_func means simulating alpha testing via discards,
 170     * so the shader definitely kills pixels.
 171     */
 172    prog_data.uses_kill = fp->program.UsesKill || key->alpha_test_func;
 173    prog_data.uses_omask =
 174       fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK);
 175    prog_data.computed_depth_mode = computed_depth_mode(&fp->program);
 176
 177    prog_data.early_fragment_tests = fs && fs->base.EarlyFragmentTests;
 178
 179    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 180    if (!prog)
 181       prog_data.base.use_alt_mode = true;
 182
 183    assign_fs_binding_table_offsets(brw->intelScreen->devinfo, prog,
 184                                    &fp->program.Base, key, &prog_data);
 185
 186    /* Allocate the references to the uniforms that will end up in the
 187     * prog_data associated with the compiled program, and which will be freed
 188     * by the state cache.
 189     */
 190    int param_count = fp->program.Base.nir->num_uniforms;
 191    if (fs)
 192       prog_data.base.nr_image_params = fs->base.NumImages;
 193    /* The backend also sometimes adds params for texture size. */
 194    param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
 195    prog_data.base.param =
 196       rzalloc_array(NULL, const gl_constant_value *, param_count);
 197    prog_data.base.pull_param =
 198       rzalloc_array(NULL, const gl_constant_value *, param_count);
 199    prog_data.base.image_param =
 200       rzalloc_array(NULL, struct brw_image_param,
 201                     prog_data.base.nr_image_params);
 202    prog_data.base.nr_params = param_count;
 203
 204    if (prog) {
 205       brw_nir_setup_glsl_uniforms(fp->program.Base.nir, prog, &fp->program.Base,
 206                                   &prog_data.base, true);
 207    } else {
 208       brw_nir_setup_arb_uniforms(fp->program.Base.nir, &fp->program.Base,
 209                                  &prog_data.base);
 210    }
 211
 212    prog_data.barycentric_interp_modes =
 213       brw_compute_barycentric_interp_modes(brw->intelScreen->devinfo,
 214                                            key->flat_shade,
 215                                            key->persample_shading,
 216                                            fp->program.Base.nir);
 217
 218    if (unlikely(brw->perf_debug)) {
 219       start_busy = (brw->batch.last_bo &&
 220                     drm_intel_bo_busy(brw->batch.last_bo));
 221       start_time = get_time();
 222    }
 223
 224    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 225       brw_dump_ir("fragment", prog, fs ? &fs->base : NULL, &fp->program.Base);
 226
 227    int st_index8 = -1, st_index16 = -1;
 228    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 229       st_index8 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS8);
 230       st_index16 = brw_get_shader_time_index(brw, prog, &fp->program.Base, ST_FS16);
 231    }
 232
 233    char *error_str = NULL;
 234    program = brw_wm_fs_emit(brw->intelScreen->compiler, brw, mem_ctx,
 235                             key, &prog_data, fp->program.Base.nir,
 236                             &fp->program.Base, st_index8, st_index16,
 237                             brw->use_rep_send, &program_size, &error_str);
 238    if (program == NULL) {
 239       if (prog) {
 240          prog->LinkStatus = false;
 241          ralloc_strcat(&prog->InfoLog, error_str);
 242       }
 243
 244       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 245
 246       ralloc_free(mem_ctx);
 247       return false;
 248    }
 249
 250    if (unlikely(brw->perf_debug) && fs) {
 251       if (fs->compiled_once)
 252          brw_wm_debug_recompile(brw, prog, key);
 253       fs->compiled_once = true;
 254
 255       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 256          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 257                     (get_time() - start_time) * 1000);
 258       }
 259    }
 260
 261    if (prog_data.base.total_scratch) {
 262       brw_get_scratch_bo(brw, &brw->wm.base.scratch_bo,
 263                          prog_data.base.total_scratch * brw->max_wm_threads);
 264    }
 265
 266    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 267       fprintf(stderr, "\n");
 268
 269    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 270                     key, sizeof(struct brw_wm_prog_key),
 271                     program, program_size,
 272                     &prog_data, sizeof(prog_data),
 273                     &brw->wm.base.prog_offset, &brw->wm.prog_data);
 274
 275    ralloc_free(mem_ctx);
 276
 277    return true;
 278 }
 279
 280 static bool
 281 key_debug(struct brw_context *brw, const char *name, int a, int b)
 282 {
 283    if (a != b) {
 284       perf_debug("  %s %d->%d\n", name, a, b);
 285       return true;
 286    } else {
 287       return false;
 288    }
 289 }
 290
 291 bool
 292 brw_debug_recompile_sampler_key(struct brw_context *brw,
 293                                 const struct brw_sampler_prog_key_data *old_key,
 294                                 const struct brw_sampler_prog_key_data *key)
 295 {
 296    bool found = false;
 297
 298    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 299       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 300                          old_key->swizzles[i], key->swizzles[i]);
 301    }
 302    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 303                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 304    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 305                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 306    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 307                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 308    found |= key_debug(brw, "gather channel quirk on any texture unit",
 309                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 310    found |= key_debug(brw, "compressed multisample layout",
 311                       old_key->compressed_multisample_layout_mask,
 312                       key->compressed_multisample_layout_mask);
 313
 314    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 315       found |= key_debug(brw, "textureGather workarounds",
 316                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 317    }
 318
 319    return found;
 320 }
 321
 322 void
 323 brw_wm_debug_recompile(struct brw_context *brw,
 324                        struct gl_shader_program *prog,
 325                        const struct brw_wm_prog_key *key)
 326 {
 327    struct brw_cache_item *c = NULL;
 328    const struct brw_wm_prog_key *old_key = NULL;
 329    bool found = false;
 330
 331    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 332
 333    for (unsigned int i = 0; i < brw->cache.size; i++) {
 334       for (c = brw->cache.items[i]; c; c = c->next) {
 335          if (c->cache_id == BRW_CACHE_FS_PROG) {
 336             old_key = c->key;
 337
 338             if (old_key->program_string_id == key->program_string_id)
 339                break;
 340          }
 341       }
 342       if (c)
 343          break;
 344    }
 345
 346    if (!c) {
 347       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 348       return;
 349    }
 350
 351    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
 352                       "depth write",
 353                       old_key->iz_lookup, key->iz_lookup);
 354    found |= key_debug(brw, "depth statistics",
 355                       old_key->stats_wm, key->stats_wm);
 356    found |= key_debug(brw, "flat shading",
 357                       old_key->flat_shade, key->flat_shade);
 358    found |= key_debug(brw, "per-sample shading",
 359                       old_key->persample_shading, key->persample_shading);
 360    found |= key_debug(brw, "per-sample shading and 2x MSAA",
 361                       old_key->persample_2x, key->persample_2x);
 362    found |= key_debug(brw, "number of color buffers",
 363                       old_key->nr_color_regions, key->nr_color_regions);
 364    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
 365                       old_key->replicate_alpha, key->replicate_alpha);
 366    found |= key_debug(brw, "rendering to FBO",
 367                       old_key->render_to_fbo, key->render_to_fbo);
 368    found |= key_debug(brw, "fragment color clamping",
 369                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 370    found |= key_debug(brw, "line smoothing",
 371                       old_key->line_aa, key->line_aa);
 372    found |= key_debug(brw, "renderbuffer height",
 373                       old_key->drawable_height, key->drawable_height);
 374    found |= key_debug(brw, "input slots valid",
 375                       old_key->input_slots_valid, key->input_slots_valid);
 376    found |= key_debug(brw, "mrt alpha test function",
 377                       old_key->alpha_test_func, key->alpha_test_func);
 378    found |= key_debug(brw, "mrt alpha test reference value",
 379                       old_key->alpha_test_ref, key->alpha_test_ref);
 380
 381    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 382
 383    if (!found) {
 384       perf_debug("  Something else\n");
 385    }
 386 }
 387
 388 static uint8_t
 389 gen6_gather_workaround(GLenum internalformat)
 390 {
 391    switch (internalformat) {
 392    case GL_R8I: return WA_SIGN | WA_8BIT;
 393    case GL_R8UI: return WA_8BIT;
 394    case GL_R16I: return WA_SIGN | WA_16BIT;
 395    case GL_R16UI: return WA_16BIT;
 396    default:
 397       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 398        * the surface state, there is no shader w/a required.
 399        */
 400       return 0;
 401    }
 402 }
 403
 404 void
 405 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 406                                    const struct gl_program *prog,
 407                                    unsigned sampler_count,
 408                                    struct brw_sampler_prog_key_data *key)
 409 {
 410    struct brw_context *brw = brw_context(ctx);
 411
 412    for (int s = 0; s < sampler_count; s++) {
 413       key->swizzles[s] = SWIZZLE_NOOP;
 414
 415       if (!(prog->SamplersUsed & (1 << s)))
 416          continue;
 417
 418       int unit_id = prog->SamplerUnits[s];
 419       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 420
 421       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 422          const struct gl_texture_object *t = unit->_Current;
 423          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 424          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 425
 426          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 427             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 428              img->_BaseFormat == GL_DEPTH_STENCIL);
 429
 430          /* Haswell handles texture swizzling as surface format overrides
 431           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 432           */
 433          if (alpha_depth || (brw->gen < 8 && !brw->is_haswell))
 434             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 435
 436          if (brw->gen < 8 &&
 437              sampler->MinFilter != GL_NEAREST &&
 438              sampler->MagFilter != GL_NEAREST) {
 439             if (sampler->WrapS == GL_CLAMP)
 440                key->gl_clamp_mask[0] |= 1 << s;
 441             if (sampler->WrapT == GL_CLAMP)
 442                key->gl_clamp_mask[1] |= 1 << s;
 443             if (sampler->WrapR == GL_CLAMP)
 444                key->gl_clamp_mask[2] |= 1 << s;
 445          }
 446
 447          /* gather4's channel select for green from RG32F is broken; requires
 448           * a shader w/a on IVB; fixable with just SCS on HSW.
 449           */
 450          if (brw->gen == 7 && !brw->is_haswell && prog->UsesGather) {
 451             if (img->InternalFormat == GL_RG32F)
 452                key->gather_channel_quirk_mask |= 1 << s;
 453          }
 454
 455          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 456           * UNORM/FLOAT instead and fix it in the shader.
 457           */
 458          if (brw->gen == 6 && prog->UsesGather) {
 459             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 460          }
 461
 462          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 463           * then we need to emit slightly different code to first sample the
 464           * MCS surface.
 465           */
 466          struct intel_texture_object *intel_tex =
 467             intel_texture_object((struct gl_texture_object *)t);
 468
 469          if (brw->gen >= 7 &&
 470              intel_tex->mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
 471             key->compressed_multisample_layout_mask |= 1 << s;
 472          }
 473       }
 474    }
 475 }
 476
 477 static bool
 478 brw_wm_state_dirty (struct brw_context *brw)
 479 {
 480    return brw_state_dirty(brw,
 481                           _NEW_BUFFERS |
 482                           _NEW_COLOR |
 483                           _NEW_DEPTH |
 484                           _NEW_FRAG_CLAMP |
 485                           _NEW_HINT |
 486                           _NEW_LIGHT |
 487                           _NEW_LINE |
 488                           _NEW_MULTISAMPLE |
 489                           _NEW_POLYGON |
 490                           _NEW_STENCIL |
 491                           _NEW_TEXTURE,
 492                           BRW_NEW_FRAGMENT_PROGRAM |
 493                           BRW_NEW_REDUCED_PRIMITIVE |
 494                           BRW_NEW_STATS_WM |
 495                           BRW_NEW_VUE_MAP_GEOM_OUT);
 496 }
 497
 498 static void
 499 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 500 {
 501    struct gl_context *ctx = &brw->ctx;
 502    /* BRW_NEW_FRAGMENT_PROGRAM */
 503    const struct brw_fragment_program *fp =
 504       (struct brw_fragment_program *) brw->fragment_program;
 505    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 506    GLuint lookup = 0;
 507    GLuint line_aa;
 508    bool program_uses_dfdy = fp->program.UsesDFdy;
 509    const bool multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 510
 511    memset(key, 0, sizeof(*key));
 512
 513    /* Build the index for table lookup
 514     */
 515    if (brw->gen < 6) {
 516       /* _NEW_COLOR */
 517       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 518          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 519
 520       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 521          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 522
 523       /* _NEW_DEPTH */
 524       if (ctx->Depth.Test)
 525          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 526
 527       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 528          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 529
 530       /* _NEW_STENCIL | _NEW_BUFFERS */
 531       if (ctx->Stencil._Enabled) {
 532          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 533
 534          if (ctx->Stencil.WriteMask[0] ||
 535              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 536             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 537       }
 538       key->iz_lookup = lookup;
 539    }
 540
 541    line_aa = AA_NEVER;
 542
 543    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 544    if (ctx->Line.SmoothFlag) {
 545       if (brw->reduced_primitive == GL_LINES) {
 546          line_aa = AA_ALWAYS;
 547       }
 548       else if (brw->reduced_primitive == GL_TRIANGLES) {
 549          if (ctx->Polygon.FrontMode == GL_LINE) {
 550             line_aa = AA_SOMETIMES;
 551
 552             if (ctx->Polygon.BackMode == GL_LINE ||
 553                 (ctx->Polygon.CullFlag &&
 554                  ctx->Polygon.CullFaceMode == GL_BACK))
 555                line_aa = AA_ALWAYS;
 556          }
 557          else if (ctx->Polygon.BackMode == GL_LINE) {
 558             line_aa = AA_SOMETIMES;
 559
 560             if ((ctx->Polygon.CullFlag &&
 561                  ctx->Polygon.CullFaceMode == GL_FRONT))
 562                line_aa = AA_ALWAYS;
 563          }
 564       }
 565    }
 566
 567    key->line_aa = line_aa;
 568
 569    /* _NEW_HINT */
 570    key->high_quality_derivatives =
 571       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 572
 573    if (brw->gen < 6)
 574       key->stats_wm = brw->stats_wm;
 575
 576    /* _NEW_LIGHT */
 577    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 578
 579    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 580    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 581
 582    /* _NEW_TEXTURE */
 583    brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
 584                                       &key->tex);
 585
 586    /* _NEW_BUFFERS */
 587    /*
 588     * Include the draw buffer origin and height so that we can calculate
 589     * fragment position values relative to the bottom left of the drawable,
 590     * from the incoming screen origin relative position we get as part of our
 591     * payload.
 592     *
 593     * This is only needed for the WM_WPOSXY opcode when the fragment program
 594     * uses the gl_FragCoord input.
 595     *
 596     * We could avoid recompiling by including this as a constant referenced by
 597     * our program, but if we were to do that it would also be nice to handle
 598     * getting that constant updated at batchbuffer submit time (when we
 599     * hold the lock and know where the buffer really is) rather than at emit
 600     * time when we don't hold the lock and are just guessing.  We could also
 601     * just avoid using this as key data if the program doesn't use
 602     * fragment.position.
 603     *
 604     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 605     * drawable height in order to invert the Y axis.
 606     */
 607    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 608       key->drawable_height = _mesa_geometric_height(ctx->DrawBuffer);
 609    }
 610
 611    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 612       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 613    }
 614
 615    /* _NEW_BUFFERS */
 616    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 617
 618    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 619    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 620       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 621
 622    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 623    /* Ignore sample qualifier while computing this flag. */
 624    key->persample_shading =
 625       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
 626    if (key->persample_shading)
 627       key->persample_2x = _mesa_geometric_samples(ctx->DrawBuffer) == 2;
 628
 629    key->compute_pos_offset =
 630       _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
 631       fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
 632
 633    key->compute_sample_id =
 634       multisample_fbo &&
 635       ctx->Multisample.Enabled &&
 636       (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
 637
 638    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 639    if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
 640                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 641       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 642
 643
 644    /* _NEW_COLOR | _NEW_BUFFERS */
 645    /* Pre-gen6, the hardware alpha test always used each render
 646     * target's alpha to do alpha test, as opposed to render target 0's alpha
 647     * like GL requires.  Fix that by building the alpha test into the
 648     * shader, and we'll skip enabling the fixed function alpha test.
 649     */
 650    if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 651        ctx->Color.AlphaEnabled) {
 652       key->alpha_test_func = ctx->Color.AlphaFunc;
 653       key->alpha_test_ref = ctx->Color.AlphaRef;
 654    }
 655
 656    /* The unique fragment program ID */
 657    key->program_string_id = fp->id;
 658 }
 659
 660 void
 661 brw_upload_wm_prog(struct brw_context *brw)
 662 {
 663    struct gl_context *ctx = &brw->ctx;
 664    struct gl_shader_program *current = ctx->_Shader->_CurrentFragmentProgram;
 665    struct brw_wm_prog_key key;
 666    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 667       brw->fragment_program;
 668
 669    if (!brw_wm_state_dirty(brw))
 670       return;
 671
 672    brw_wm_populate_key(brw, &key);
 673
 674    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 675                          &key, sizeof(key),
 676                          &brw->wm.base.prog_offset, &brw->wm.prog_data)) {
 677       bool success = brw_codegen_wm_prog(brw, current, fp, &key);
 678       (void) success;
 679       assert(success);
 680    }
 681    brw->wm.base.prog_data = &brw->wm.prog_data->base;
 682 }
 683
 684 bool
 685 brw_fs_precompile(struct gl_context *ctx,
 686                   struct gl_shader_program *shader_prog,
 687                   struct gl_program *prog)
 688 {
 689    struct brw_context *brw = brw_context(ctx);
 690    struct brw_wm_prog_key key;
 691
 692    struct gl_fragment_program *fp = (struct gl_fragment_program *) prog;
 693    struct brw_fragment_program *bfp = brw_fragment_program(fp);
 694    bool program_uses_dfdy = fp->UsesDFdy;
 695
 696    memset(&key, 0, sizeof(key));
 697
 698    if (brw->gen < 6) {
 699       if (fp->UsesKill)
 700          key.iz_lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 701
 702       if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 703          key.iz_lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 704
 705       /* Just assume depth testing. */
 706       key.iz_lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 707       key.iz_lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 708    }
 709
 710    if (brw->gen < 6 || _mesa_bitcount_64(fp->Base.InputsRead &
 711                                          BRW_FS_VARYING_INPUT_MASK) > 16)
 712       key.input_slots_valid = fp->Base.InputsRead | VARYING_BIT_POS;
 713
 714    brw_setup_tex_for_precompile(brw, &key.tex, &fp->Base);
 715
 716    if (fp->Base.InputsRead & VARYING_BIT_POS) {
 717       key.drawable_height = ctx->DrawBuffer->Height;
 718    }
 719
 720    key.nr_color_regions = _mesa_bitcount_64(fp->Base.OutputsWritten &
 721          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 722          BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 723
 724    if ((fp->Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 725       key.render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer) ||
 726                           key.nr_color_regions > 1;
 727    }
 728
 729    key.program_string_id = bfp->id;
 730
 731    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 732    struct brw_wm_prog_data *old_prog_data = brw->wm.prog_data;
 733
 734    bool success = brw_codegen_wm_prog(brw, shader_prog, bfp, &key);
 735
 736    brw->wm.base.prog_offset = old_prog_offset;
 737    brw->wm.prog_data = old_prog_data;
 738
 739    return success;
 740 }