src/mesa/drivers/dri/i965/brw_wm.c

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) 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  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   */
  31
  32 #include "brw_context.h"
  33 #include "brw_wm.h"
  34 #include "brw_state.h"
  35 #include "main/formats.h"
  36 #include "main/fbobject.h"
  37 #include "main/samplerobj.h"
  38 #include "program/prog_parameter.h"
  39
  40 #include "glsl/ralloc.h"
  41
  42 /** Return number of src args for given instruction */
  43 GLuint brw_wm_nr_args( GLuint opcode )
  44 {
  45    switch (opcode) {
  46    case WM_FRONTFACING:
  47    case WM_PIXELXY:
  48       return 0;
  49    case WM_CINTERP:
  50    case WM_WPOSXY:
  51    case WM_DELTAXY:
  52       return 1;
  53    case WM_LINTERP:
  54    case WM_PIXELW:
  55       return 2;
  56    case WM_FB_WRITE:
  57    case WM_PINTERP:
  58       return 3;
  59    default:
  60       assert(opcode < MAX_OPCODE);
  61       return _mesa_num_inst_src_regs(opcode);
  62    }
  63 }
  64
  65
  66 GLuint brw_wm_is_scalar_result( GLuint opcode )
  67 {
  68    switch (opcode) {
  69    case OPCODE_COS:
  70    case OPCODE_EX2:
  71    case OPCODE_LG2:
  72    case OPCODE_POW:
  73    case OPCODE_RCP:
  74    case OPCODE_RSQ:
  75    case OPCODE_SIN:
  76    case OPCODE_DP2:
  77    case OPCODE_DP3:
  78    case OPCODE_DP4:
  79    case OPCODE_DPH:
  80    case OPCODE_DST:
  81       return 1;
  82
  83    default:
  84       return 0;
  85    }
  86 }
  87
  88
  89 /**
  90  * Do GPU code generation for non-GLSL shader.  non-GLSL shaders have
  91  * no flow control instructions so we can more readily do SSA-style
  92  * optimizations.
  93  */
  94 static void
  95 brw_wm_non_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c)
  96 {
  97    /* Augment fragment program.  Add instructions for pre- and
  98     * post-fragment-program tasks such as interpolation and fogging.
  99     */
 100    brw_wm_pass_fp(c);
 101
 102    /* Translate to intermediate representation.  Build register usage
 103     * chains.
 104     */
 105    brw_wm_pass0(c);
 106
 107    /* Dead code removal.
 108     */
 109    brw_wm_pass1(c);
 110
 111    /* Register allocation.
 112     * Divide by two because we operate on 16 pixels at a time and require
 113     * two GRF entries for each logical shader register.
 114     */
 115    c->grf_limit = BRW_WM_MAX_GRF / 2;
 116
 117    brw_wm_pass2(c);
 118
 119    /* how many general-purpose registers are used */
 120    c->prog_data.reg_blocks = brw_register_blocks(c->max_wm_grf);
 121
 122    /* Emit GEN4 code.
 123     */
 124    brw_wm_emit(c);
 125 }
 126
 127
 128 /**
 129  * Return a bitfield where bit n is set if barycentric interpolation mode n
 130  * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
 131  */
 132 static unsigned
 133 brw_compute_barycentric_interp_modes(struct brw_context *brw,
 134                                      bool shade_model_flat,
 135                                      const struct gl_fragment_program *fprog)
 136 {
 137    unsigned barycentric_interp_modes = 0;
 138    int attr;
 139
 140    /* Loop through all fragment shader inputs to figure out what interpolation
 141     * modes are in use, and set the appropriate bits in
 142     * barycentric_interp_modes.
 143     */
 144    for (attr = 0; attr < FRAG_ATTRIB_MAX; ++attr) {
 145       enum glsl_interp_qualifier interp_qualifier =
 146          fprog->InterpQualifier[attr];
 147       bool is_centroid = fprog->IsCentroid & BITFIELD64_BIT(attr);
 148       bool is_gl_Color = attr == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_COL1;
 149
 150       /* Ignore unused inputs. */
 151       if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
 152          continue;
 153
 154       /* Ignore WPOS and FACE, because they don't require interpolation. */
 155       if (attr == FRAG_ATTRIB_WPOS || attr == FRAG_ATTRIB_FACE)
 156          continue;
 157
 158       /* Determine the set (or sets) of barycentric coordinates needed to
 159        * interpolate this variable.  Note that when
 160        * brw->needs_unlit_centroid_workaround is set, centroid interpolation
 161        * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
 162        * for lit pixels, so we need both sets of barycentric coordinates.
 163        */
 164       if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
 165          if (is_centroid) {
 166             barycentric_interp_modes |=
 167                1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
 168          }
 169          if (!is_centroid || brw->needs_unlit_centroid_workaround) {
 170             barycentric_interp_modes |=
 171                1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
 172          }
 173       } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
 174                  (!(shade_model_flat && is_gl_Color) &&
 175                   interp_qualifier == INTERP_QUALIFIER_NONE)) {
 176          if (is_centroid) {
 177             barycentric_interp_modes |=
 178                1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
 179          }
 180          if (!is_centroid || brw->needs_unlit_centroid_workaround) {
 181             barycentric_interp_modes |=
 182                1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
 183          }
 184       }
 185    }
 186
 187    return barycentric_interp_modes;
 188 }
 189
 190
 191 void
 192 brw_wm_payload_setup(struct brw_context *brw,
 193                      struct brw_wm_compile *c)
 194 {
 195    struct intel_context *intel = &brw->intel;
 196    bool uses_depth = (c->fp->program.Base.InputsRead &
 197                       (1 << FRAG_ATTRIB_WPOS)) != 0;
 198    unsigned barycentric_interp_modes = c->prog_data.barycentric_interp_modes;
 199    int i;
 200
 201    if (intel->gen >= 6) {
 202       /* R0-1: masks, pixel X/Y coordinates. */
 203       c->nr_payload_regs = 2;
 204       /* R2: only for 32-pixel dispatch.*/
 205
 206       /* R3-26: barycentric interpolation coordinates.  These appear in the
 207        * same order that they appear in the brw_wm_barycentric_interp_mode
 208        * enum.  Each set of coordinates occupies 2 registers if dispatch width
 209        * == 8 and 4 registers if dispatch width == 16.  Coordinates only
 210        * appear if they were enabled using the "Barycentric Interpolation
 211        * Mode" bits in WM_STATE.
 212        */
 213       for (i = 0; i < BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT; ++i) {
 214          if (barycentric_interp_modes & (1 << i)) {
 215             c->barycentric_coord_reg[i] = c->nr_payload_regs;
 216             c->nr_payload_regs += 2;
 217             if (c->dispatch_width == 16) {
 218                c->nr_payload_regs += 2;
 219             }
 220          }
 221       }
 222
 223       /* R27: interpolated depth if uses source depth */
 224       if (uses_depth) {
 225          c->source_depth_reg = c->nr_payload_regs;
 226          c->nr_payload_regs++;
 227          if (c->dispatch_width == 16) {
 228             /* R28: interpolated depth if not 8-wide. */
 229             c->nr_payload_regs++;
 230          }
 231       }
 232       /* R29: interpolated W set if GEN6_WM_USES_SOURCE_W.
 233        */
 234       if (uses_depth) {
 235          c->source_w_reg = c->nr_payload_regs;
 236          c->nr_payload_regs++;
 237          if (c->dispatch_width == 16) {
 238             /* R30: interpolated W if not 8-wide. */
 239             c->nr_payload_regs++;
 240          }
 241       }
 242       /* R31: MSAA position offsets. */
 243       /* R32-: bary for 32-pixel. */
 244       /* R58-59: interp W for 32-pixel. */
 245
 246       if (c->fp->program.Base.OutputsWritten &
 247           BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
 248          c->source_depth_to_render_target = true;
 249          c->computes_depth = true;
 250       }
 251    } else {
 252       brw_wm_lookup_iz(intel, c);
 253    }
 254 }
 255
 256 bool
 257 brw_wm_prog_data_compare(const void *in_a, const void *in_b,
 258                          int aux_size, const void *in_key)
 259 {
 260    const struct brw_wm_prog_data *a = in_a;
 261    const struct brw_wm_prog_data *b = in_b;
 262
 263    /* Compare all the struct up to the pointers. */
 264    if (memcmp(a, b, offsetof(struct brw_wm_prog_data, param)))
 265       return false;
 266
 267    if (memcmp(a->param, b->param, a->nr_params * sizeof(void *)))
 268       return false;
 269
 270    if (memcmp(a->pull_param, b->pull_param, a->nr_pull_params * sizeof(void *)))
 271       return false;
 272
 273    return true;
 274 }
 275
 276 void
 277 brw_wm_prog_data_free(const void *in_prog_data)
 278 {
 279    const struct brw_wm_prog_data *prog_data = in_prog_data;
 280
 281    ralloc_free((void *)prog_data->param);
 282    ralloc_free((void *)prog_data->pull_param);
 283 }
 284
 285 /**
 286  * All Mesa program -> GPU code generation goes through this function.
 287  * Depending on the instructions used (i.e. flow control instructions)
 288  * we'll use one of two code generators.
 289  */
 290 bool do_wm_prog(struct brw_context *brw,
 291                 struct gl_shader_program *prog,
 292                 struct brw_fragment_program *fp,
 293                 struct brw_wm_prog_key *key)
 294 {
 295    struct intel_context *intel = &brw->intel;
 296    struct brw_wm_compile *c;
 297    const GLuint *program;
 298    struct gl_shader *fs = NULL;
 299    GLuint program_size;
 300
 301    if (prog)
 302       fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 303
 304    c = brw->wm.compile_data;
 305    if (c == NULL) {
 306       brw->wm.compile_data = rzalloc(NULL, struct brw_wm_compile);
 307       c = brw->wm.compile_data;
 308       if (c == NULL) {
 309          /* Ouch - big out of memory problem.  Can't continue
 310           * without triggering a segfault, no way to signal,
 311           * so just return.
 312           */
 313          return false;
 314       }
 315    } else {
 316       void *instruction = c->instruction;
 317       void *prog_instructions = c->prog_instructions;
 318       void *vreg = c->vreg;
 319       void *refs = c->refs;
 320       memset(c, 0, sizeof(*brw->wm.compile_data));
 321       c->instruction = instruction;
 322       c->prog_instructions = prog_instructions;
 323       c->vreg = vreg;
 324       c->refs = refs;
 325    }
 326
 327    /* Allocate the references to the uniforms that will end up in the
 328     * prog_data associated with the compiled program, and which will be freed
 329     * by the state cache.
 330     */
 331    if (fs) {
 332       int param_count = fs->num_uniform_components;
 333       /* The backend also sometimes adds params for texture size. */
 334       param_count += 2 * BRW_MAX_TEX_UNIT;
 335
 336       c->prog_data.param = rzalloc_array(c, const float *, param_count);
 337       c->prog_data.pull_param = rzalloc_array(c, const float *, param_count);
 338    } else {
 339       /* brw_wm_pass0.c will also add references to 0.0 and 1.0 which are
 340        * uploaded as push parameters.
 341        */
 342       int param_count = (fp->program.Base.Parameters->NumParameters + 2) * 4;
 343       c->prog_data.param = rzalloc_array(c, const float *, param_count);
 344       /* The old backend never does pull constants. */
 345       c->prog_data.pull_param = NULL;
 346    }
 347
 348    memcpy(&c->key, key, sizeof(*key));
 349
 350    c->fp = fp;
 351    c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
 352
 353    brw_init_compile(brw, &c->func, c);
 354
 355    c->prog_data.barycentric_interp_modes =
 356       brw_compute_barycentric_interp_modes(brw, c->key.flat_shade,
 357                                            &fp->program);
 358
 359    if (prog && prog->_LinkedShaders[MESA_SHADER_FRAGMENT]) {
 360       if (!brw_wm_fs_emit(brw, c, prog))
 361          return false;
 362    } else {
 363       if (!c->instruction) {
 364          c->instruction = rzalloc_array(c, struct brw_wm_instruction, BRW_WM_MAX_INSN);
 365          c->prog_instructions = rzalloc_array(c, struct prog_instruction, BRW_WM_MAX_INSN);
 366          c->vreg = rzalloc_array(c, struct brw_wm_value, BRW_WM_MAX_VREG);
 367          c->refs = rzalloc_array(c, struct brw_wm_ref, BRW_WM_MAX_REF);
 368       }
 369
 370       /* Fallback for fixed function and ARB_fp shaders. */
 371       c->dispatch_width = 16;
 372       brw_wm_payload_setup(brw, c);
 373       brw_wm_non_glsl_emit(brw, c);
 374       c->prog_data.dispatch_width = 16;
 375    }
 376
 377    /* Scratch space is used for register spilling */
 378    if (c->last_scratch) {
 379       perf_debug("Fragment shader triggered register spilling.  "
 380                  "Try reducing the number of live scalar values to "
 381                  "improve performance.\n");
 382
 383       c->prog_data.total_scratch = brw_get_scratch_size(c->last_scratch);
 384
 385       brw_get_scratch_bo(intel, &brw->wm.scratch_bo,
 386                          c->prog_data.total_scratch * brw->max_wm_threads);
 387    }
 388
 389    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 390       fprintf(stderr, "\n");
 391
 392    /* get the program
 393     */
 394    program = brw_get_program(&c->func, &program_size);
 395
 396    brw_upload_cache(&brw->cache, BRW_WM_PROG,
 397                     &c->key, sizeof(c->key),
 398                     program, program_size,
 399                     &c->prog_data, sizeof(c->prog_data),
 400                     &brw->wm.prog_offset, &brw->wm.prog_data);
 401
 402    return true;
 403 }
 404
 405 static bool
 406 key_debug(const char *name, int a, int b)
 407 {
 408    if (a != b) {
 409       perf_debug("  %s %d->%d\n", name, a, b);
 410       return true;
 411    } else {
 412       return false;
 413    }
 414 }
 415
 416 bool
 417 brw_debug_recompile_sampler_key(const struct brw_sampler_prog_key_data *old_key,
 418                                 const struct brw_sampler_prog_key_data *key)
 419 {
 420    bool found = false;
 421
 422    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 423       found |= key_debug("EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 424                          old_key->swizzles[i], key->swizzles[i]);
 425    }
 426    found |= key_debug("GL_CLAMP enabled on any texture unit's 1st coordinate",
 427                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 428    found |= key_debug("GL_CLAMP enabled on any texture unit's 2nd coordinate",
 429                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 430    found |= key_debug("GL_CLAMP enabled on any texture unit's 3rd coordinate",
 431                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 432    found |= key_debug("GL_MESA_ycbcr texturing\n",
 433                       old_key->yuvtex_mask, key->yuvtex_mask);
 434    found |= key_debug("GL_MESA_ycbcr UV swapping\n",
 435                       old_key->yuvtex_swap_mask, key->yuvtex_swap_mask);
 436
 437    return found;
 438 }
 439
 440 void
 441 brw_wm_debug_recompile(struct brw_context *brw,
 442                        struct gl_shader_program *prog,
 443                        const struct brw_wm_prog_key *key)
 444 {
 445    struct brw_cache_item *c = NULL;
 446    const struct brw_wm_prog_key *old_key = NULL;
 447    bool found = false;
 448
 449    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 450
 451    for (unsigned int i = 0; i < brw->cache.size; i++) {
 452       for (c = brw->cache.items[i]; c; c = c->next) {
 453          if (c->cache_id == BRW_WM_PROG) {
 454             old_key = c->key;
 455
 456             if (old_key->program_string_id == key->program_string_id)
 457                break;
 458          }
 459       }
 460       if (c)
 461          break;
 462    }
 463
 464    if (!c) {
 465       perf_debug("  Didn't find previous compile in the shader cache for "
 466                  "debug\n");
 467       return;
 468    }
 469
 470    found |= key_debug("alphatest, computed depth, depth test, or depth write",
 471                       old_key->iz_lookup, key->iz_lookup);
 472    found |= key_debug("depth statistics", old_key->stats_wm, key->stats_wm);
 473    found |= key_debug("flat shading", old_key->flat_shade, key->flat_shade);
 474    found |= key_debug("number of color buffers", old_key->nr_color_regions, key->nr_color_regions);
 475    found |= key_debug("rendering to FBO", old_key->render_to_fbo, key->render_to_fbo);
 476    found |= key_debug("fragment color clamping", old_key->clamp_fragment_color, key->clamp_fragment_color);
 477    found |= key_debug("line smoothing", old_key->line_aa, key->line_aa);
 478    found |= key_debug("proj_attrib_mask", old_key->proj_attrib_mask, key->proj_attrib_mask);
 479    found |= key_debug("renderbuffer height", old_key->drawable_height, key->drawable_height);
 480    found |= key_debug("vertex shader outputs", old_key->vp_outputs_written, key->vp_outputs_written);
 481
 482    found |= brw_debug_recompile_sampler_key(&old_key->tex, &key->tex);
 483
 484    if (!found) {
 485       perf_debug("  Something else\n");
 486    }
 487 }
 488
 489 void
 490 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 491                                    const struct gl_program *prog,
 492                                    struct brw_sampler_prog_key_data *key)
 493 {
 494    for (int s = 0; s < MAX_SAMPLERS; s++) {
 495       key->swizzles[s] = SWIZZLE_NOOP;
 496
 497       if (!(prog->SamplersUsed & (1 << s)))
 498          continue;
 499
 500       int unit_id = prog->SamplerUnits[s];
 501       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 502
 503       if (unit->_ReallyEnabled && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 504          const struct gl_texture_object *t = unit->_Current;
 505          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 506          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 507          int swizzles[SWIZZLE_NIL + 1] = {
 508             SWIZZLE_X,
 509             SWIZZLE_Y,
 510             SWIZZLE_Z,
 511             SWIZZLE_W,
 512             SWIZZLE_ZERO,
 513             SWIZZLE_ONE,
 514             SWIZZLE_NIL
 515          };
 516
 517          if (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 518              img->_BaseFormat == GL_DEPTH_STENCIL) {
 519             /* We handle GL_DEPTH_TEXTURE_MODE here instead of as surface
 520              * format overrides because shadow comparison always returns the
 521              * result of the comparison in all channels anyway.
 522              */
 523             switch (t->DepthMode) {
 524             case GL_ALPHA:
 525                swizzles[0] = SWIZZLE_ZERO;
 526                swizzles[1] = SWIZZLE_ZERO;
 527                swizzles[2] = SWIZZLE_ZERO;
 528                swizzles[3] = SWIZZLE_X;
 529                break;
 530             case GL_LUMINANCE:
 531                swizzles[0] = SWIZZLE_X;
 532                swizzles[1] = SWIZZLE_X;
 533                swizzles[2] = SWIZZLE_X;
 534                swizzles[3] = SWIZZLE_ONE;
 535                break;
 536             case GL_INTENSITY:
 537                swizzles[0] = SWIZZLE_X;
 538                swizzles[1] = SWIZZLE_X;
 539                swizzles[2] = SWIZZLE_X;
 540                swizzles[3] = SWIZZLE_X;
 541                break;
 542             case GL_RED:
 543                swizzles[0] = SWIZZLE_X;
 544                swizzles[1] = SWIZZLE_ZERO;
 545                swizzles[2] = SWIZZLE_ZERO;
 546                swizzles[3] = SWIZZLE_ONE;
 547                break;
 548             }
 549          }
 550
 551          if (img->InternalFormat == GL_YCBCR_MESA) {
 552             key->yuvtex_mask |= 1 << s;
 553             if (img->TexFormat == MESA_FORMAT_YCBCR)
 554                 key->yuvtex_swap_mask |= 1 << s;
 555          }
 556
 557          key->swizzles[s] =
 558             MAKE_SWIZZLE4(swizzles[GET_SWZ(t->_Swizzle, 0)],
 559                           swizzles[GET_SWZ(t->_Swizzle, 1)],
 560                           swizzles[GET_SWZ(t->_Swizzle, 2)],
 561                           swizzles[GET_SWZ(t->_Swizzle, 3)]);
 562
 563          if (sampler->MinFilter != GL_NEAREST &&
 564              sampler->MagFilter != GL_NEAREST) {
 565             if (sampler->WrapS == GL_CLAMP)
 566                key->gl_clamp_mask[0] |= 1 << s;
 567             if (sampler->WrapT == GL_CLAMP)
 568                key->gl_clamp_mask[1] |= 1 << s;
 569             if (sampler->WrapR == GL_CLAMP)
 570                key->gl_clamp_mask[2] |= 1 << s;
 571          }
 572       }
 573    }
 574 }
 575
 576 static void brw_wm_populate_key( struct brw_context *brw,
 577                                  struct brw_wm_prog_key *key )
 578 {
 579    struct gl_context *ctx = &brw->intel.ctx;
 580    struct intel_context *intel = &brw->intel;
 581    /* BRW_NEW_FRAGMENT_PROGRAM */
 582    const struct brw_fragment_program *fp =
 583       (struct brw_fragment_program *)brw->fragment_program;
 584    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 585    GLuint lookup = 0;
 586    GLuint line_aa;
 587    bool program_uses_dfdy = fp->program.UsesDFdy;
 588
 589    memset(key, 0, sizeof(*key));
 590
 591    /* Build the index for table lookup
 592     */
 593    if (intel->gen < 6) {
 594       /* _NEW_COLOR */
 595       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 596          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 597
 598       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 599          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 600
 601       /* _NEW_DEPTH */
 602       if (ctx->Depth.Test)
 603          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 604
 605       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 606          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 607
 608       /* _NEW_STENCIL */
 609       if (ctx->Stencil._Enabled) {
 610          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 611
 612          if (ctx->Stencil.WriteMask[0] ||
 613              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 614             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 615       }
 616       key->iz_lookup = lookup;
 617    }
 618
 619    line_aa = AA_NEVER;
 620
 621    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 622    if (ctx->Line.SmoothFlag) {
 623       if (brw->intel.reduced_primitive == GL_LINES) {
 624          line_aa = AA_ALWAYS;
 625       }
 626       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 627          if (ctx->Polygon.FrontMode == GL_LINE) {
 628             line_aa = AA_SOMETIMES;
 629
 630             if (ctx->Polygon.BackMode == GL_LINE ||
 631                 (ctx->Polygon.CullFlag &&
 632                  ctx->Polygon.CullFaceMode == GL_BACK))
 633                line_aa = AA_ALWAYS;
 634          }
 635          else if (ctx->Polygon.BackMode == GL_LINE) {
 636             line_aa = AA_SOMETIMES;
 637
 638             if ((ctx->Polygon.CullFlag &&
 639                  ctx->Polygon.CullFaceMode == GL_FRONT))
 640                line_aa = AA_ALWAYS;
 641          }
 642       }
 643    }
 644
 645    key->line_aa = line_aa;
 646
 647    if (intel->gen < 6)
 648       key->stats_wm = brw->intel.stats_wm;
 649
 650    /* BRW_NEW_WM_INPUT_DIMENSIONS */
 651    /* Only set this for fixed function.  The optimization it enables isn't
 652     * useful for programs using shaders.
 653     */
 654    if (ctx->Shader.CurrentFragmentProgram)
 655       key->proj_attrib_mask = 0xffffffff;
 656    else
 657       key->proj_attrib_mask = brw->wm.input_size_masks[4-1];
 658
 659    /* _NEW_LIGHT */
 660    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 661
 662    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 663    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 664
 665    /* _NEW_TEXTURE */
 666    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 667
 668    /* _NEW_BUFFERS */
 669    /*
 670     * Include the draw buffer origin and height so that we can calculate
 671     * fragment position values relative to the bottom left of the drawable,
 672     * from the incoming screen origin relative position we get as part of our
 673     * payload.
 674     *
 675     * This is only needed for the WM_WPOSXY opcode when the fragment program
 676     * uses the gl_FragCoord input.
 677     *
 678     * We could avoid recompiling by including this as a constant referenced by
 679     * our program, but if we were to do that it would also be nice to handle
 680     * getting that constant updated at batchbuffer submit time (when we
 681     * hold the lock and know where the buffer really is) rather than at emit
 682     * time when we don't hold the lock and are just guessing.  We could also
 683     * just avoid using this as key data if the program doesn't use
 684     * fragment.position.
 685     *
 686     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 687     * drawable height in order to invert the Y axis.
 688     */
 689    if (fp->program.Base.InputsRead & FRAG_BIT_WPOS) {
 690       key->drawable_height = ctx->DrawBuffer->Height;
 691    }
 692
 693    if ((fp->program.Base.InputsRead & FRAG_BIT_WPOS) || program_uses_dfdy) {
 694       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 695    }
 696
 697    /* _NEW_BUFFERS */
 698    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 699   /* _NEW_MULTISAMPLE */
 700    key->sample_alpha_to_coverage = ctx->Multisample.SampleAlphaToCoverage;
 701
 702    /* CACHE_NEW_VS_PROG */
 703    if (intel->gen < 6)
 704       key->vp_outputs_written = brw->vs.prog_data->outputs_written;
 705
 706    /* The unique fragment program ID */
 707    key->program_string_id = fp->id;
 708 }
 709
 710
 711 static void
 712 brw_upload_wm_prog(struct brw_context *brw)
 713 {
 714    struct intel_context *intel = &brw->intel;
 715    struct gl_context *ctx = &intel->ctx;
 716    struct brw_wm_prog_key key;
 717    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 718       brw->fragment_program;
 719
 720    brw_wm_populate_key(brw, &key);
 721
 722    if (!brw_search_cache(&brw->cache, BRW_WM_PROG,
 723                          &key, sizeof(key),
 724                          &brw->wm.prog_offset, &brw->wm.prog_data)) {
 725       bool success = do_wm_prog(brw, ctx->Shader._CurrentFragmentProgram, fp,
 726                                 &key);
 727       (void) success;
 728       assert(success);
 729    }
 730 }
 731
 732
 733 const struct brw_tracked_state brw_wm_prog = {
 734    .dirty = {
 735       .mesa  = (_NEW_COLOR |
 736                 _NEW_DEPTH |
 737                 _NEW_STENCIL |
 738                 _NEW_POLYGON |
 739                 _NEW_LINE |
 740                 _NEW_LIGHT |
 741                 _NEW_FRAG_CLAMP |
 742                 _NEW_BUFFERS |
 743                 _NEW_TEXTURE |
 744                 _NEW_MULTISAMPLE),
 745       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 746                 BRW_NEW_WM_INPUT_DIMENSIONS |
 747                 BRW_NEW_REDUCED_PRIMITIVE),
 748       .cache = CACHE_NEW_VS_PROG,
 749    },
 750    .emit = brw_upload_wm_prog
 751 };
 752