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    bool start_busy = false;
 145    double start_time = 0;
 146
 147    memset(&prog_data, 0, sizeof(prog_data));
 148
 149    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 150    if (fp->program.is_arb_asm)
 151       prog_data.base.use_alt_mode = true;
 152
 153    assign_fs_binding_table_offsets(devinfo, &fp->program, key, &prog_data);
 154
 155    if (!fp->program.is_arb_asm) {
 156       brw_nir_setup_glsl_uniforms(mem_ctx, fp->program.nir, &fp->program,
 157                                   &prog_data.base, true);
 158       brw_nir_analyze_ubo_ranges(brw->screen->compiler, fp->program.nir,
 159                                  prog_data.base.ubo_ranges);
 160    } else {
 161       brw_nir_setup_arb_uniforms(mem_ctx, fp->program.nir, &fp->program,
 162                                  &prog_data.base);
 163
 164       if (unlikely(INTEL_DEBUG & DEBUG_WM))
 165          brw_dump_arb_asm("fragment", &fp->program);
 166    }
 167
 168    if (unlikely(brw->perf_debug)) {
 169       start_busy = (brw->batch.last_bo &&
 170                     brw_bo_busy(brw->batch.last_bo));
 171       start_time = get_time();
 172    }
 173
 174    int st_index8 = -1, st_index16 = -1;
 175    if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
 176       st_index8 = brw_get_shader_time_index(brw, &fp->program, ST_FS8,
 177                                             !fp->program.is_arb_asm);
 178       st_index16 = brw_get_shader_time_index(brw, &fp->program, ST_FS16,
 179                                              !fp->program.is_arb_asm);
 180    }
 181
 182    char *error_str = NULL;
 183    program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx,
 184                             key, &prog_data, fp->program.nir,
 185                             &fp->program, st_index8, st_index16,
 186                             true, false, vue_map,
 187                             &error_str);
 188
 189    if (program == NULL) {
 190       if (!fp->program.is_arb_asm) {
 191          fp->program.sh.data->LinkStatus = linking_failure;
 192          ralloc_strcat(&fp->program.sh.data->InfoLog, error_str);
 193       }
 194
 195       _mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
 196
 197       ralloc_free(mem_ctx);
 198       return false;
 199    }
 200
 201    if (unlikely(brw->perf_debug)) {
 202       if (fp->compiled_once)
 203          brw_wm_debug_recompile(brw, &fp->program, key);
 204       fp->compiled_once = true;
 205
 206       if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
 207          perf_debug("FS compile took %.03f ms and stalled the GPU\n",
 208                     (get_time() - start_time) * 1000);
 209       }
 210    }
 211
 212    brw_alloc_stage_scratch(brw, &brw->wm.base, prog_data.base.total_scratch);
 213
 214    if (unlikely((INTEL_DEBUG & DEBUG_WM) && fp->program.is_arb_asm))
 215       fprintf(stderr, "\n");
 216
 217    /* The param and pull_param arrays will be freed by the shader cache. */
 218    ralloc_steal(NULL, prog_data.base.param);
 219    ralloc_steal(NULL, prog_data.base.pull_param);
 220    brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
 221                     key, sizeof(struct brw_wm_prog_key),
 222                     program, prog_data.base.program_size,
 223                     &prog_data, sizeof(prog_data),
 224                     &brw->wm.base.prog_offset, &brw->wm.base.prog_data);
 225
 226    ralloc_free(mem_ctx);
 227
 228    return true;
 229 }
 230
 231 bool
 232 brw_debug_recompile_sampler_key(struct brw_context *brw,
 233                                 const struct brw_sampler_prog_key_data *old_key,
 234                                 const struct brw_sampler_prog_key_data *key)
 235 {
 236    bool found = false;
 237
 238    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 239       found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 240                          old_key->swizzles[i], key->swizzles[i]);
 241    }
 242    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 243                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 244    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 245                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 246    found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 247                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 248    found |= key_debug(brw, "gather channel quirk on any texture unit",
 249                       old_key->gather_channel_quirk_mask, key->gather_channel_quirk_mask);
 250    found |= key_debug(brw, "compressed multisample layout",
 251                       old_key->compressed_multisample_layout_mask,
 252                       key->compressed_multisample_layout_mask);
 253    found |= key_debug(brw, "16x msaa",
 254                       old_key->msaa_16,
 255                       key->msaa_16);
 256
 257    found |= key_debug(brw, "y_uv image bound",
 258                       old_key->y_uv_image_mask,
 259                       key->y_uv_image_mask);
 260    found |= key_debug(brw, "y_u_v image bound",
 261                       old_key->y_u_v_image_mask,
 262                       key->y_u_v_image_mask);
 263    found |= key_debug(brw, "yx_xuxv image bound",
 264                       old_key->yx_xuxv_image_mask,
 265                       key->yx_xuxv_image_mask);
 266    found |= key_debug(brw, "xy_uxvx image bound",
 267                       old_key->xy_uxvx_image_mask,
 268                       key->xy_uxvx_image_mask);
 269
 270
 271    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 272       found |= key_debug(brw, "textureGather workarounds",
 273                          old_key->gen6_gather_wa[i], key->gen6_gather_wa[i]);
 274    }
 275
 276    return found;
 277 }
 278
 279 static uint8_t
 280 gen6_gather_workaround(GLenum internalformat)
 281 {
 282    switch (internalformat) {
 283    case GL_R8I: return WA_SIGN | WA_8BIT;
 284    case GL_R8UI: return WA_8BIT;
 285    case GL_R16I: return WA_SIGN | WA_16BIT;
 286    case GL_R16UI: return WA_16BIT;
 287    default:
 288       /* Note that even though GL_R32I and GL_R32UI have format overrides in
 289        * the surface state, there is no shader w/a required.
 290        */
 291       return 0;
 292    }
 293 }
 294
 295 void
 296 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 297                                    const struct gl_program *prog,
 298                                    struct brw_sampler_prog_key_data *key)
 299 {
 300    struct brw_context *brw = brw_context(ctx);
 301    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 302    GLbitfield mask = prog->SamplersUsed;
 303
 304    while (mask) {
 305       const int s = u_bit_scan(&mask);
 306
 307       key->swizzles[s] = SWIZZLE_NOOP;
 308
 309       int unit_id = prog->SamplerUnits[s];
 310       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 311
 312       if (unit->_Current && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 313          const struct gl_texture_object *t = unit->_Current;
 314          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 315          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 316
 317          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 318             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 319              img->_BaseFormat == GL_DEPTH_STENCIL);
 320
 321          /* Haswell handles texture swizzling as surface format overrides
 322           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 323           */
 324          if (alpha_depth || (devinfo->gen < 8 && !devinfo->is_haswell))
 325             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 326
 327          if (devinfo->gen < 8 &&
 328              sampler->MinFilter != GL_NEAREST &&
 329              sampler->MagFilter != GL_NEAREST) {
 330             if (sampler->WrapS == GL_CLAMP)
 331                key->gl_clamp_mask[0] |= 1 << s;
 332             if (sampler->WrapT == GL_CLAMP)
 333                key->gl_clamp_mask[1] |= 1 << s;
 334             if (sampler->WrapR == GL_CLAMP)
 335                key->gl_clamp_mask[2] |= 1 << s;
 336          }
 337
 338          /* gather4 for RG32* is broken in multiple ways on Gen7. */
 339          if (devinfo->gen == 7 && prog->info.uses_texture_gather) {
 340             switch (img->InternalFormat) {
 341             case GL_RG32I:
 342             case GL_RG32UI: {
 343                /* We have to override the format to R32G32_FLOAT_LD.
 344                 * This means that SCS_ALPHA and SCS_ONE will return 0x3f8
 345                 * (1.0) rather than integer 1.  This needs shader hacks.
 346                 *
 347                 * On Ivybridge, we whack W (alpha) to ONE in our key's
 348                 * swizzle.  On Haswell, we look at the original texture
 349                 * swizzle, and use XYZW with channels overridden to ONE,
 350                 * leaving normal texture swizzling to SCS.
 351                 */
 352                unsigned src_swizzle =
 353                   devinfo->is_haswell ? t->_Swizzle : key->swizzles[s];
 354                for (int i = 0; i < 4; i++) {
 355                   unsigned src_comp = GET_SWZ(src_swizzle, i);
 356                   if (src_comp == SWIZZLE_ONE || src_comp == SWIZZLE_W) {
 357                      key->swizzles[i] &= ~(0x7 << (3 * i));
 358                      key->swizzles[i] |= SWIZZLE_ONE << (3 * i);
 359                   }
 360                }
 361                /* fallthrough */
 362             }
 363             case GL_RG32F:
 364                /* The channel select for green doesn't work - we have to
 365                 * request blue.  Haswell can use SCS for this, but Ivybridge
 366                 * needs a shader workaround.
 367                 */
 368                if (!devinfo->is_haswell)
 369                   key->gather_channel_quirk_mask |= 1 << s;
 370                break;
 371             }
 372          }
 373
 374          /* Gen6's gather4 is broken for UINT/SINT; we treat them as
 375           * UNORM/FLOAT instead and fix it in the shader.
 376           */
 377          if (devinfo->gen == 6 && prog->info.uses_texture_gather) {
 378             key->gen6_gather_wa[s] = gen6_gather_workaround(img->InternalFormat);
 379          }
 380
 381          /* If this is a multisample sampler, and uses the CMS MSAA layout,
 382           * then we need to emit slightly different code to first sample the
 383           * MCS surface.
 384           */
 385          struct intel_texture_object *intel_tex =
 386             intel_texture_object((struct gl_texture_object *)t);
 387
 388          /* From gen9 onwards some single sampled buffers can also be
 389           * compressed. These don't need ld2dms sampling along with mcs fetch.
 390           */
 391          if (intel_tex->mt->aux_usage == ISL_AUX_USAGE_MCS) {
 392             assert(devinfo->gen >= 7);
 393             assert(intel_tex->mt->surf.samples > 1);
 394             assert(intel_tex->mt->mcs_buf);
 395             assert(intel_tex->mt->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY);
 396             key->compressed_multisample_layout_mask |= 1 << s;
 397
 398             if (intel_tex->mt->surf.samples >= 16) {
 399                assert(devinfo->gen >= 9);
 400                key->msaa_16 |= 1 << s;
 401             }
 402          }
 403
 404          if (t->Target == GL_TEXTURE_EXTERNAL_OES && intel_tex->planar_format) {
 405             switch (intel_tex->planar_format->components) {
 406             case __DRI_IMAGE_COMPONENTS_Y_UV:
 407                key->y_uv_image_mask |= 1 << s;
 408                break;
 409             case __DRI_IMAGE_COMPONENTS_Y_U_V:
 410                key->y_u_v_image_mask |= 1 << s;
 411                break;
 412             case __DRI_IMAGE_COMPONENTS_Y_XUXV:
 413                key->yx_xuxv_image_mask |= 1 << s;
 414                break;
 415             case __DRI_IMAGE_COMPONENTS_Y_UXVX:
 416                key->xy_uxvx_image_mask |= 1 << s;
 417                break;
 418             default:
 419                break;
 420             }
 421          }
 422
 423       }
 424    }
 425 }
 426
 427 static bool
 428 brw_wm_state_dirty(const struct brw_context *brw)
 429 {
 430    return brw_state_dirty(brw,
 431                           _NEW_BUFFERS |
 432                           _NEW_COLOR |
 433                           _NEW_DEPTH |
 434                           _NEW_FRAG_CLAMP |
 435                           _NEW_HINT |
 436                           _NEW_LIGHT |
 437                           _NEW_LINE |
 438                           _NEW_MULTISAMPLE |
 439                           _NEW_POLYGON |
 440                           _NEW_STENCIL |
 441                           _NEW_TEXTURE,
 442                           BRW_NEW_FRAGMENT_PROGRAM |
 443                           BRW_NEW_REDUCED_PRIMITIVE |
 444                           BRW_NEW_STATS_WM |
 445                           BRW_NEW_VUE_MAP_GEOM_OUT);
 446 }
 447
 448 void
 449 brw_wm_populate_key(struct brw_context *brw, struct brw_wm_prog_key *key)
 450 {
 451    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 452    struct gl_context *ctx = &brw->ctx;
 453    /* BRW_NEW_FRAGMENT_PROGRAM */
 454    const struct gl_program *prog = brw->programs[MESA_SHADER_FRAGMENT];
 455    const struct brw_program *fp = brw_program_const(prog);
 456    GLuint lookup = 0;
 457    GLuint line_aa;
 458
 459    memset(key, 0, sizeof(*key));
 460
 461    /* Build the index for table lookup
 462     */
 463    if (devinfo->gen < 6) {
 464       /* _NEW_COLOR */
 465       if (prog->info.fs.uses_discard || ctx->Color.AlphaEnabled) {
 466          lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 467       }
 468
 469       if (prog->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 470          lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 471       }
 472
 473       /* _NEW_DEPTH */
 474       if (ctx->Depth.Test)
 475          lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 476
 477       if (brw_depth_writes_enabled(brw))
 478          lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 479
 480       /* _NEW_STENCIL | _NEW_BUFFERS */
 481       if (brw->stencil_enabled) {
 482          lookup |= BRW_WM_IZ_STENCIL_TEST_ENABLE_BIT;
 483
 484          if (ctx->Stencil.WriteMask[0] ||
 485              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 486             lookup |= BRW_WM_IZ_STENCIL_WRITE_ENABLE_BIT;
 487       }
 488       key->iz_lookup = lookup;
 489    }
 490
 491    line_aa = BRW_WM_AA_NEVER;
 492
 493    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 494    if (ctx->Line.SmoothFlag) {
 495       if (brw->reduced_primitive == GL_LINES) {
 496          line_aa = BRW_WM_AA_ALWAYS;
 497       }
 498       else if (brw->reduced_primitive == GL_TRIANGLES) {
 499          if (ctx->Polygon.FrontMode == GL_LINE) {
 500             line_aa = BRW_WM_AA_SOMETIMES;
 501
 502             if (ctx->Polygon.BackMode == GL_LINE ||
 503                 (ctx->Polygon.CullFlag &&
 504                  ctx->Polygon.CullFaceMode == GL_BACK))
 505                line_aa = BRW_WM_AA_ALWAYS;
 506          }
 507          else if (ctx->Polygon.BackMode == GL_LINE) {
 508             line_aa = BRW_WM_AA_SOMETIMES;
 509
 510             if ((ctx->Polygon.CullFlag &&
 511                  ctx->Polygon.CullFaceMode == GL_FRONT))
 512                line_aa = BRW_WM_AA_ALWAYS;
 513          }
 514       }
 515    }
 516
 517    key->line_aa = line_aa;
 518
 519    /* _NEW_HINT */
 520    key->high_quality_derivatives =
 521       prog->info.uses_fddx_fddy &&
 522       ctx->Hint.FragmentShaderDerivative == GL_NICEST;
 523
 524    if (devinfo->gen < 6)
 525       key->stats_wm = brw->stats_wm;
 526
 527    /* _NEW_LIGHT */
 528    key->flat_shade =
 529       (prog->info.inputs_read & (VARYING_BIT_COL0 | VARYING_BIT_COL1)) &&
 530       (ctx->Light.ShadeModel == GL_FLAT);
 531
 532    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 533    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 534
 535    /* _NEW_TEXTURE */
 536    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 537
 538    /* _NEW_BUFFERS */
 539    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 540
 541    /* _NEW_COLOR */
 542    key->force_dual_color_blend = brw->dual_color_blend_by_location &&
 543       (ctx->Color.BlendEnabled & 1) && ctx->Color.Blend[0]._UsesDualSrc;
 544
 545    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 546    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 547       (_mesa_is_alpha_test_enabled(ctx) ||
 548        _mesa_is_alpha_to_coverage_enabled(ctx));
 549
 550    /* _NEW_BUFFERS _NEW_MULTISAMPLE */
 551    /* Ignore sample qualifier while computing this flag. */
 552    if (ctx->Multisample.Enabled) {
 553       key->persample_interp =
 554          ctx->Multisample.SampleShading &&
 555          (ctx->Multisample.MinSampleShadingValue *
 556           _mesa_geometric_samples(ctx->DrawBuffer) > 1);
 557
 558       key->multisample_fbo = _mesa_geometric_samples(ctx->DrawBuffer) > 1;
 559    }
 560
 561    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 562    if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 563                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 564       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 565    }
 566
 567    /* _NEW_COLOR | _NEW_BUFFERS */
 568    /* Pre-gen6, the hardware alpha test always used each render
 569     * target's alpha to do alpha test, as opposed to render target 0's alpha
 570     * like GL requires.  Fix that by building the alpha test into the
 571     * shader, and we'll skip enabling the fixed function alpha test.
 572     */
 573    if (devinfo->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 574        ctx->Color.AlphaEnabled) {
 575       key->alpha_test_func = ctx->Color.AlphaFunc;
 576       key->alpha_test_ref = ctx->Color.AlphaRef;
 577    }
 578
 579    /* The unique fragment program ID */
 580    key->program_string_id = fp->id;
 581
 582    /* Whether reads from the framebuffer should behave coherently. */
 583    key->coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 584 }
 585
 586 void
 587 brw_upload_wm_prog(struct brw_context *brw)
 588 {
 589    struct brw_wm_prog_key key;
 590    struct brw_program *fp =
 591       (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
 592
 593    if (!brw_wm_state_dirty(brw))
 594       return;
 595
 596    brw_wm_populate_key(brw, &key);
 597
 598    if (brw_search_cache(&brw->cache, BRW_CACHE_FS_PROG,
 599                         &key, sizeof(key),
 600                         &brw->wm.base.prog_offset,
 601                         &brw->wm.base.prog_data))
 602       return;
 603
 604    if (brw_disk_cache_upload_program(brw, MESA_SHADER_FRAGMENT))
 605       return;
 606
 607    fp = (struct brw_program *) brw->programs[MESA_SHADER_FRAGMENT];
 608    fp->id = key.program_string_id;
 609
 610    MAYBE_UNUSED bool success = brw_codegen_wm_prog(brw, fp, &key,
 611                                                    &brw->vue_map_geom_out);
 612    assert(success);
 613 }
 614
 615 bool
 616 brw_fs_precompile(struct gl_context *ctx, struct gl_program *prog)
 617 {
 618    struct brw_context *brw = brw_context(ctx);
 619    const struct gen_device_info *devinfo = &brw->screen->devinfo;
 620    struct brw_wm_prog_key key;
 621
 622    struct brw_program *bfp = brw_program(prog);
 623
 624    memset(&key, 0, sizeof(key));
 625
 626    uint64_t outputs_written = prog->info.outputs_written;
 627
 628    if (devinfo->gen < 6) {
 629       if (prog->info.fs.uses_discard)
 630          key.iz_lookup |= BRW_WM_IZ_PS_KILL_ALPHATEST_BIT;
 631
 632       if (outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 633          key.iz_lookup |= BRW_WM_IZ_PS_COMPUTES_DEPTH_BIT;
 634
 635       /* Just assume depth testing. */
 636       key.iz_lookup |= BRW_WM_IZ_DEPTH_TEST_ENABLE_BIT;
 637       key.iz_lookup |= BRW_WM_IZ_DEPTH_WRITE_ENABLE_BIT;
 638    }
 639
 640    if (devinfo->gen < 6 || _mesa_bitcount_64(prog->info.inputs_read &
 641                                              BRW_FS_VARYING_INPUT_MASK) > 16) {
 642       key.input_slots_valid = prog->info.inputs_read | VARYING_BIT_POS;
 643    }
 644
 645    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 646
 647    key.nr_color_regions = _mesa_bitcount_64(outputs_written &
 648          ~(BITFIELD64_BIT(FRAG_RESULT_DEPTH) |
 649            BITFIELD64_BIT(FRAG_RESULT_STENCIL) |
 650            BITFIELD64_BIT(FRAG_RESULT_SAMPLE_MASK)));
 651
 652    key.program_string_id = bfp->id;
 653
 654    /* Whether reads from the framebuffer should behave coherently. */
 655    key.coherent_fb_fetch = ctx->Extensions.MESA_shader_framebuffer_fetch;
 656
 657    uint32_t old_prog_offset = brw->wm.base.prog_offset;
 658    struct brw_stage_prog_data *old_prog_data = brw->wm.base.prog_data;
 659
 660    struct brw_vue_map vue_map;
 661    if (devinfo->gen < 6) {
 662       brw_compute_vue_map(&brw->screen->devinfo, &vue_map,
 663                           prog->info.inputs_read | VARYING_BIT_POS,
 664                           false);
 665    }
 666
 667    bool success = brw_codegen_wm_prog(brw, bfp, &key, &vue_map);
 668
 669    brw->wm.base.prog_offset = old_prog_offset;
 670    brw->wm.base.prog_data = old_prog_data;
 671
 672    return success;
 673 }