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 "main/enums.h"
  30 #include "main/formats.h"
  31 #include "main/fbobject.h"
  32 #include "main/samplerobj.h"
  33 #include "main/framebuffer.h"
  34 #include "program/prog_parameter.h"
  35 #include "program/program.h"
  36 #include "intel_mipmap_tree.h"
  37 #include "intel_image.h"
  38 #include "compiler/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 gen_device_info *devinfo,
  45                                 const struct gl_program *prog,
  46                                 const struct brw_wm_prog_key *key,
  47                                 struct brw_wm_prog_data *prog_data)
  48 {
  49    uint32_t next_binding_table_offset = 0;
  50
  51    /* If there are no color regions, we still perform an FB write to a null
  52     * renderbuffer, which we place at surface index 0.
  53     */
  54    prog_data->binding_table.render_target_start = next_binding_table_offset;
  55    next_binding_table_offset += MAX2(key->nr_color_regions, 1);
  56
  57    next_binding_table_offset =
  58       brw_assign_common_binding_table_offsets(devinfo, prog, &prog_data->base,
  59                                               next_binding_table_offset);
  60
  61    if (prog->nir->info.outputs_read && !key->coherent_fb_fetch) {
  62       prog_data->binding_table.render_target_read_start =
  63          next_binding_table_offset;
  64       next_binding_table_offset += key->nr_color_regions;
  65    }
  66 }
  67
  68 static void
  69 brw_wm_debug_recompile(struct brw_context *brw, struct gl_program *prog,
  70                        const struct brw_wm_prog_key *key)
  71 {
  72    perf_debug("Recompiling fragment shader for program %d\n", prog->Id);
  73
  74    bool found = false;
  75    const struct brw_wm_prog_key *old_key =
  76       brw_find_previous_compile(&brw->cache, BRW_CACHE_FS_PROG,
  77                                 key->program_string_id);
  78
  79    if (!old_key) {
  80       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
  81       return;
  82    }
  83
  84    found |= key_debug(brw, "alphatest, computed depth, depth test, or "
  85                       "depth write",
  86                       old_key->iz_lookup, key->iz_lookup);
  87    found |= key_debug(brw, "depth statistics",
  88                       old_key->stats_wm, key->stats_wm);
  89    found |= key_debug(brw, "flat shading",
  90                       old_key->flat_shade, key->flat_shade);
  91    found |= key_debug(brw, "number of color buffers",
  92                       old_key->nr_color_regions, key->nr_color_regions);
  93    found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
  94                       old_key->replicate_alpha, key->replicate_alpha);
  95    found |= key_debug(brw, "fragment color clamping",
  96                       old_key->clamp_fragment_color, key->clamp_fragment_color);
  97    found |= key_debug(brw, "per-sample interpolation",
  98                       old_key->persample_interp, key->persample_interp);
  99    found |= key_debug(brw, "multisampled FBO",
 100                       old_key->multisample_fbo, key->multisample_fbo);
 101    found |= key_debug(brw, "frag coord adds sample pos",
 102                       old_key->frag_coord_adds_sample_pos,
 103                       key->frag_coord_adds_sample_pos);
 104    found |= key_debug(brw, "line smoothing",
 105                       old_key->line_aa, key->line_aa);
 106    found |= key_debug(brw, "high quality derivatives",
 107                       old_key->high_quality_derivatives,
 108                       key->high_quality_derivatives);
 109    found |= key_debug(brw, "force dual color blending",
 110                       old_key->force_dual_color_blend,
 111                       key->force_dual_color_blend);
 112    found |= key_debug(brw, "coherent fb fetch",
 113                       old_key->coherent_fb_fetch, key->coherent_fb_fetch);
 114
 115    found |= key_debug(brw, "input slots valid",
 116                       old_key->input_slots_valid, key->input_slots_valid);
 117    found |= key_debug(brw, "mrt alpha test function",
 118                       old_key->alpha_test_func, key->alpha_test_func);
 119    found |= key_debug(brw, "mrt alpha test reference value",
 120                       old_key->alpha_test_ref, key->alpha_test_ref);
 121
 122    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 123
 124    if (!found) {
 125       perf_debug("  Something else\n");
 126    }
 127 }
 128
 129 /**
 130  * All Mesa program -> GPU code generation goes through this function.
 131  * Depending on the instructions used (i.e. flow control instructions)
 132  * we'll use one of two code generators.
 133  */
 134 static bool
 135 brw_codegen_wm_prog(struct brw_context *brw,
 136                     struct brw_program *fp,
 137                     struct brw_wm_prog_key *key,
 138                     struct brw_vue_map *vue_map)
 139 {
 140    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 141    void *mem_ctx = ralloc_context(NULL);
 142    struct brw_wm_prog_data prog_data;
 143    const GLuint *program;
 144    GLuint program_size;
 145    bool start_busy = false;
 146    double start_time = 0;
 147
 148    memset(&prog_data, 0, sizeof(prog_data));
 149
 150    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 151    if (fp->program.is_arb_asm)
 152       prog_data.base.use_alt_mode = true;
 153
 154    assign_fs_binding_table_offsets(devinfo, &fp->program, key, &prog_data);
 155
 156    if (!fp->program.is_arb_asm) {
 157       brw_nir_setup_glsl_uniforms(mem_ctx, fp->program.nir, &fp->program,
 158                                   &prog_data.base, true);
 159       brw_nir_analyze_ubo_ranges(brw->screen->compiler, fp->program.nir,
 160                                  prog_data.base.ubo_ranges);
 161    } else {
 162       brw_nir_setup_arb_uniforms(mem_ctx, fp->program.nir, &fp->program,
 163                                  &prog_data.base);
 164
 165       if (unlikely(INTEL_DEBUG & DEBUG_WM))
 166          brw_dump_arb_asm("fragment", &fp->program);
 167    }
 168
 169    if (unlikely(brw->perf_debug)) {
 170       start_busy = (brw->batch.last_bo &&
 171                     brw_bo_busy(brw->batch.last_bo));
 172       start_time = get_time();
 173    }
 174
 175    int st_index8 = -1, st_index16 = -1;
 176    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 177       st_index8 = brw_get_shader_time_index(brw, &fp->program, ST_FS8,
 178                                             !fp->program.is_arb_asm);
 179       st_index16 = brw_get_shader_time_index(brw, &fp->program, ST_FS16,
 180                                              !fp->program.is_arb_asm);
 181    }
 182
 183    char *error_str = NULL;
 184    program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx,
 185                             key, &prog_data, fp->program.nir,
 186                             &fp->program, st_index8, st_index16,
 187                             true, false, vue_map,
 188                             &program_size, &error_str);
 189
 190    if (program == NULL) {
 191       if (!fp->program.is_arb_asm) {
 192          fp->program.sh.data->LinkStatus = linking_failure;
 193          ralloc_strcat(&fp->program.sh.data->InfoLog, error_str);
 194       }
 195
 196       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 197
 198       ralloc_free(mem_ctx);
 199       return false;
 200    }
 201
 202    if (unlikely(brw->perf_debug)) {
 203       if (fp->compiled_once)
 204          brw_wm_debug_recompile(brw, &fp->program, key);
 205       fp->compiled_once = true;
 206
 207       if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
 208          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 209                     (get_time() - start_time) * 1000);
 210       }
 211    }
 212
 213    brw_alloc_stage_scratch(brw, &brw->wm.base,
 214                            prog_data.base.total_scratch,
 215                            devinfo->max_wm_threads);
 216
 217    if (unlikely((INTEL_DEBUG & DEBUG_WM) && fp->program.is_arb_asm))
 218       fprintf(stderr, "\n");
 219
 220    /* The param and pull_param arrays will be freed by the shader cache. */
 221    ralloc_steal(NULL, prog_data.base.param);
 222    ralloc_steal(NULL, prog_data.base.pull_param);
 223    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 224                     key, sizeof(struct brw_wm_prog_key),
 225                     program, program_size,
 226                     &prog_data, sizeof(prog_data),
 227                     &brw->wm.base.prog_offset, &brw->wm.base.prog_data);
 228
 229    ralloc_free(mem_ctx);
 230
 231    return true;
 232 }
 233
 234 bool
 235 brw_debug_recompile_sampler_key(struct brw_context *brw,
 236                                 const struct brw_sampler_prog_key_data *old_key,
 237                                 const struct brw_sampler_prog_key_data *key)
 238 {
 239    bool found = false;
 240
 241    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 242       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 243                          old_key->swizzles[i], key->swizzles[i]);
 244    }
 245    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 246                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 247    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 248                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 249    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 250                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 251    found |= key_debug(brw, "gather channel quirk on any texture unit",
 252                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 253    found |= key_debug(brw, "compressed multisample layout",
 254                       old_key->compressed_multisample_layout_mask,
 255                       key->compressed_multisample_layout_mask);
 256    found |= key_debug(brw, "16x msaa",
 257                       old_key->msaa_16,
 258                       key->msaa_16);
 259
 260    found |= key_debug(brw, "y_uv image bound",
 261                       old_key->y_uv_image_mask,
 262                       key->y_uv_image_mask);
 263    found |= key_debug(brw, "y_u_v image bound",
 264                       old_key->y_u_v_image_mask,
 265                       key->y_u_v_image_mask);
 266    found |= key_debug(brw, "yx_xuxv image bound",
 267                       old_key->yx_xuxv_image_mask,
 268                       key->yx_xuxv_image_mask);
 269    found |= key_debug(brw, "xy_uxvx image bound",
 270                       old_key->xy_uxvx_image_mask,
 271                       key->xy_uxvx_image_mask);
 272
 273
 274    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 275       found |= key_debug(brw, "textureGather workarounds",
 276                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 277    }
 278
 279    return found;
 280 }
 281
 282 static uint8_t
 283 gen6_gather_workaround(GLenum internalformat)
 284 {
 285    switch (internalformat) {
 286    case GL_R8I: return WA_SIGN | WA_8BIT;
 287    case GL_R8UI: return WA_8BIT;
 288    case GL_R16I: return WA_SIGN | WA_16BIT;
 289    case GL_R16UI: return WA_16BIT;
 290    default:
 291       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 292        * the surface state, there is no shader w/a required.
 293        */
 294       return 0;
 295    }
 296 }
 297
 298 void
 299 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 300                                    const struct gl_program *prog,
 301                                    struct brw_sampler_prog_key_data *key)
 302 {
 303    struct brw_context *brw = brw_context(ctx);
 304    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 305    GLbitfield mask = prog->SamplersUsed;
 306
 307    while (mask) {
 308       const int s = u_bit_scan(&mask);
 309
 310       key->swizzles[s] = SWIZZLE_NOOP;
 311
 312       int unit_id = prog->SamplerUnits[s];
 313       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 314
 315       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 316          const struct gl_texture_object *t = unit->_Current;
 317          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 318          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 319
 320          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 321             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 322              img->_BaseFormat == GL_DEPTH_STENCIL);
 323
 324          /* Haswell handles texture swizzling as surface format overrides
 325           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 326           */
 327          if (alpha_depth || (devinfo->gen < 8 && !devinfo->is_haswell))
 328             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 329
 330          if (devinfo->gen < 8 &&
 331              sampler->MinFilter != GL_NEAREST &&
 332              sampler->MagFilter != GL_NEAREST) {
 333             if (sampler->WrapS == GL_CLAMP)
 334                key->gl_clamp_mask[0] |= 1 << s;
 335             if (sampler->WrapT == GL_CLAMP)
 336                key->gl_clamp_mask[1] |= 1 << s;
 337             if (sampler->WrapR == GL_CLAMP)
 338                key->gl_clamp_mask[2] |= 1 << s;
 339          }
 340
 341          /* gather4 for RG32* is broken in multiple ways on Gen7. */
 342          if (devinfo->gen == 7 && prog->nir->info.uses_texture_gather) {
 343             switch (img->InternalFormat) {
 344             case GL_RG32I:
 345             case GL_RG32UI: {
 346                /* We have to override the format to R32G32_FLOAT_LD.
 347                 * This means that SCS_ALPHA and SCS_ONE will return 0x3f8
 348                 * (1.0) rather than integer 1.  This needs shader hacks.
 349                 *
 350                 * On Ivybridge, we whack W (alpha) to ONE in our key's
 351                 * swizzle.  On Haswell, we look at the original texture
 352                 * swizzle, and use XYZW with channels overridden to ONE,
 353                 * leaving normal texture swizzling to SCS.
 354                 */
 355                unsigned src_swizzle =
 356                   devinfo->is_haswell ? t->_Swizzle : key->swizzles[s];
 357                for (int i = 0; i < 4; i++) {
 358                   unsigned src_comp = GET_SWZ(src_swizzle, i);
 359                   if (src_comp == SWIZZLE_ONE || src_comp == SWIZZLE_W) {
 360                      key->swizzles[i] &= ~(0x7 << (3 * i));
 361                      key->swizzles[i] |= SWIZZLE_ONE << (3 * i);
 362                   }
 363                }
 364                /* fallthrough */
 365             }
 366             case GL_RG32F:
 367                /* The channel select for green doesn't work - we have to
 368                 * request blue.  Haswell can use SCS for this, but Ivybridge
 369                 * needs a shader workaround.
 370                 */
 371                if (!devinfo->is_haswell)
 372                   key->gather_channel_quirk_mask |= 1 << s;
 373                break;
 374             }
 375          }
 376
 377          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 378           * UNORM/FLOAT instead and fix it in the shader.
 379           */
 380          if (devinfo->gen == 6 && prog->nir->info.uses_texture_gather) {
 381             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 382          }
 383
 384          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 385           * then we need to emit slightly different code to first sample the
 386           * MCS surface.
 387           */
 388          struct intel_texture_object *intel_tex =
 389             intel_texture_object((struct gl_texture_object *)t);
 390
 391          /* From gen9 onwards some single sampled buffers can also be
 392           * compressed. These don't need ld2dms sampling along with mcs fetch.
 393           */
 394          if (intel_tex->mt->aux_usage == ISL_AUX_USAGE_MCS) {
 395             assert(devinfo->gen >= 7);
 396             assert(intel_tex->mt->surf.samples > 1);
 397             assert(intel_tex->mt->mcs_buf);
 398             assert(intel_tex->mt->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY);
 399             key->compressed_multisample_layout_mask |= 1 << s;
 400
 401             if (intel_tex->mt->surf.samples >= 16) {
 402                assert(devinfo->gen >= 9);
 403                key->msaa_16 |= 1 << s;
 404             }
 405          }
 406
 407          if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
 408             switch (intel_tex->planar_format->components) {
 409             case __DRI_IMAGE_COMPONENTS_Y_UV:
 410                key->y_uv_image_mask |= 1 << s;
 411                break;
 412             case __DRI_IMAGE_COMPONENTS_Y_U_V:
 413                key->y_u_v_image_mask |= 1 << s;
 414                break;
 415             case __DRI_IMAGE_COMPONENTS_Y_XUXV:
 416                key->yx_xuxv_image_mask |= 1 << s;
 417                break;
 418             case __DRI_IMAGE_COMPONENTS_Y_UXVX:
 419                key->xy_uxvx_image_mask |= 1 << s;
 420                break;
 421             default:
 422                break;
 423             }
 424          }
 425
 426       }
 427    }
 428 }
 429
 430 static bool
 431 brw_wm_state_dirty(const struct brw_context *brw)
 432 {
 433    return brw_state_dirty(brw,
 434                           _NEW_BUFFERS |
 435                           _NEW_COLOR |
 436                           _NEW_DEPTH |
 437                           _NEW_FRAG_CLAMP |
 438                           _NEW_HINT |
 439                           _NEW_LIGHT |
 440                           _NEW_LINE |
 441                           _NEW_MULTISAMPLE |
 442                           _NEW_POLYGON |
 443                           _NEW_STENCIL |
 444                           _NEW_TEXTURE,
 445                           BRW_NEW_FRAGMENT_PROGRAM |
 446                           BRW_NEW_REDUCED_PRIMITIVE |
 447                           BRW_NEW_STATS_WM |
 448                           BRW_NEW_VUE_MAP_GEOM_OUT);
 449 }
 450
 451 void
 452 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 453 {
 454    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 455    struct gl_context *ctx = &brw->ctx;
 456    /* BRW_NEW_FRAGMENT_PROGRAM */
 457    const struct gl_program *prog = brw->programs[MESA_SHADER_FRAGMENT];
 458    const struct brw_program *fp = brw_program_const(prog);
 459    GLuint lookup = 0;
 460    GLuint line_aa;
 461
 462    memset(key, 0, sizeof(*key));
 463
 464    /* Build the index for table lookup
 465     */
 466    if (devinfo->gen < 6) {
 467       /* _NEW_COLOR */
 468       if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) {
 469          lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 470       }
 471
 472       if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 473          lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 474       }
 475
 476       /* _NEW_DEPTH */
 477       if (ctx->Depth.Test)
 478          lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 479
 480       if (brw_depth_writes_enabled(brw))
 481          lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 482
 483       /* _NEW_STENCIL | _NEW_BUFFERS */
 484       if (brw->stencil_enabled) {
 485          lookup |= BRW_WM_IZ_STENCIL_TEST_ENABLE_BIT;
 486
 487          if (ctx->Stencil.WriteMask[0] ||
 488              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 489             lookup |= BRW_WM_IZ_STENCIL_WRITE_ENABLE_BIT;
 490       }
 491       key->iz_lookup = lookup;
 492    }
 493
 494    line_aa = BRW_WM_AA_NEVER;
 495
 496    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 497    if (ctx->Line.SmoothFlag) {
 498       if (brw->reduced_primitive == GL_LINES) {
 499          line_aa = BRW_WM_AA_ALWAYS;
 500       }
 501       else if (brw->reduced_primitive == GL_TRIANGLES) {
 502          if (ctx->Polygon.FrontMode == GL_LINE) {
 503             line_aa = BRW_WM_AA_SOMETIMES;
 504
 505             if (ctx->Polygon.BackMode == GL_LINE ||
 506                 (ctx->Polygon.CullFlag &&
 507                  ctx->Polygon.CullFaceMode == GL_BACK))
 508                line_aa = BRW_WM_AA_ALWAYS;
 509          }
 510          else if (ctx->Polygon.BackMode == GL_LINE) {
 511             line_aa = BRW_WM_AA_SOMETIMES;
 512
 513             if ((ctx->Polygon.CullFlag &&
 514                  ctx->Polygon.CullFaceMode == GL_FRONT))
 515                line_aa = BRW_WM_AA_ALWAYS;
 516          }
 517       }
 518    }
 519
 520    key->line_aa = line_aa;
 521
 522    /* _NEW_HINT */
 523    key->high_quality_derivatives =
 524       prog->info.uses_fddx_fddy &&
 525       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 526
 527    if (devinfo->gen < 6)
 528       key->stats_wm = brw->stats_wm;
 529
 530    /* _NEW_LIGHT */
 531    key->flat_shade =
 532       (prog->info.inputs_read & (VARYING_BIT_COL0 | VARYING_BIT_COL1)) &&
 533       (ctx->Light.ShadeModel == GL_FLAT);
 534
 535    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 536    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 537
 538    /* _NEW_TEXTURE */
 539    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 540
 541    /* _NEW_BUFFERS */
 542    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 543
 544    /* _NEW_COLOR */
 545    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 546       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 547
 548    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 549    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 550       (_mesa_is_alpha_test_enabled(ctx) ||
 551        _mesa_is_alpha_to_coverage_enabled(ctx));
 552
 553    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 554    /* Ignore sample qualifier while computing this flag. */
 555    if (ctx->Multisample.Enabled) {
 556       key->persample_interp =
 557          ctx->Multisample.SampleShading &&
 558          (ctx->Multisample.MinSampleShadingValue *
 559           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 560
 561       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 562    }
 563
 564    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 565    if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 566                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 567       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 568    }
 569
 570    /* _NEW_COLOR | _NEW_BUFFERS */
 571    /* Pre-gen6, the hardware alpha test always used each render
 572     * target's alpha to do alpha test, as opposed to render target 0's alpha
 573     * like GL requires.  Fix that by building the alpha test into the
 574     * shader, and we'll skip enabling the fixed function alpha test.
 575     */
 576    if (devinfo->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 577        ctx->Color.AlphaEnabled) {
 578       key->alpha_test_func = ctx->Color.AlphaFunc;
 579       key->alpha_test_ref = ctx->Color.AlphaRef;
 580    }
 581
 582    /* The unique fragment program ID */
 583    key->program_string_id = fp->id;
 584
 585    /* Whether reads from the framebuffer should behave coherently. */
 586    key->coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 587 }
 588
 589 void
 590 brw_upload_wm_prog(struct brw_context *brw)
 591 {
 592    struct brw_wm_prog_key key;
 593    struct brw_program *fp =
 594       (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
 595
 596    if (!brw_wm_state_dirty(brw))
 597       return;
 598
 599    brw_wm_populate_key(brw, &key);
 600
 601    if (!brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 602                          &key, sizeof(key),
 603                          &brw->wm.base.prog_offset,
 604                          &brw->wm.base.prog_data)) {
 605       bool success = brw_codegen_wm_prog(brw, fp, &key,
 606                                          &brw->vue_map_geom_out);
 607       (void) success;
 608       assert(success);
 609    }
 610 }
 611
 612 bool
 613 brw_fs_precompile(struct gl_context *ctx, struct gl_program *prog)
 614 {
 615    struct brw_context *brw = brw_context(ctx);
 616    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 617    struct brw_wm_prog_key key;
 618
 619    struct brw_program *bfp = brw_program(prog);
 620
 621    memset(&key, 0, sizeof(key));
 622
 623    uint64_t outputs_written = prog->info.outputs_written;
 624
 625    if (devinfo->gen < 6) {
 626       if (prog->info.fs.uses_discard)
 627          key.iz_lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 628
 629       if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 630          key.iz_lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 631
 632       /* Just assume depth testing. */
 633       key.iz_lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 634       key.iz_lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 635    }
 636
 637    if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 638                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 639       key.input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
 640    }
 641
 642    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 643
 644    key.nr_color_regions = _mesa_bitcount_64(outputs_written &
 645          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 646            BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
 647            BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 648
 649    key.program_string_id = bfp->id;
 650
 651    /* Whether reads from the framebuffer should behave coherently. */
 652    key.coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 653
 654    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 655    struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data;
 656
 657    struct brw_vue_map vue_map;
 658    if (devinfo->gen < 6) {
 659       brw_compute_vue_map(&brw->screen->devinfo, &vue_map,
 660                           prog->info.inputs_read | VARYING_BIT_POS,
 661                           false);
 662    }
 663
 664    bool success = brw_codegen_wm_prog(brw, bfp, &key, &vue_map);
 665
 666    brw->wm.base.prog_offset = old_prog_offset;
 667    brw->wm.base.prog_data = old_prog_data;
 668
 669    return success;
 670 }