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 /**
 257  * All Mesa program -> GPU code generation goes through this function.
 258  * Depending on the instructions used (i.e. flow control instructions)
 259  * we'll use one of two code generators.
 260  */
 261 bool do_wm_prog(struct brw_context *brw,
 262                 struct gl_shader_program *prog,
 263                 struct brw_fragment_program *fp,
 264                 struct brw_wm_prog_key *key)
 265 {
 266    struct intel_context *intel = &brw->intel;
 267    struct brw_wm_compile *c;
 268    const GLuint *program;
 269    GLuint program_size;
 270
 271    c = brw->wm.compile_data;
 272    if (c == NULL) {
 273       brw->wm.compile_data = rzalloc(NULL, struct brw_wm_compile);
 274       c = brw->wm.compile_data;
 275       if (c == NULL) {
 276          /* Ouch - big out of memory problem.  Can't continue
 277           * without triggering a segfault, no way to signal,
 278           * so just return.
 279           */
 280          return false;
 281       }
 282    } else {
 283       void *instruction = c->instruction;
 284       void *prog_instructions = c->prog_instructions;
 285       void *vreg = c->vreg;
 286       void *refs = c->refs;
 287       memset(c, 0, sizeof(*brw->wm.compile_data));
 288       c->instruction = instruction;
 289       c->prog_instructions = prog_instructions;
 290       c->vreg = vreg;
 291       c->refs = refs;
 292    }
 293    memcpy(&c->key, key, sizeof(*key));
 294
 295    c->fp = fp;
 296    c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
 297
 298    brw_init_compile(brw, &c->func, c);
 299
 300    c->prog_data.barycentric_interp_modes =
 301       brw_compute_barycentric_interp_modes(brw, c->key.flat_shade,
 302                                            &fp->program);
 303
 304    if (prog && prog->_LinkedShaders[MESA_SHADER_FRAGMENT]) {
 305       if (!brw_wm_fs_emit(brw, c, prog))
 306          return false;
 307    } else {
 308       if (!c->instruction) {
 309          c->instruction = rzalloc_array(c, struct brw_wm_instruction, BRW_WM_MAX_INSN);
 310          c->prog_instructions = rzalloc_array(c, struct prog_instruction, BRW_WM_MAX_INSN);
 311          c->vreg = rzalloc_array(c, struct brw_wm_value, BRW_WM_MAX_VREG);
 312          c->refs = rzalloc_array(c, struct brw_wm_ref, BRW_WM_MAX_REF);
 313       }
 314
 315       /* Fallback for fixed function and ARB_fp shaders. */
 316       c->dispatch_width = 16;
 317       brw_wm_payload_setup(brw, c);
 318       brw_wm_non_glsl_emit(brw, c);
 319       c->prog_data.dispatch_width = 16;
 320    }
 321
 322    /* Scratch space is used for register spilling */
 323    if (c->last_scratch) {
 324       perf_debug("Fragment shader triggered register spilling.  "
 325                  "Try reducing the number of live scalar values to "
 326                  "improve performance.\n");
 327
 328       c->prog_data.total_scratch = brw_get_scratch_size(c->last_scratch);
 329
 330       brw_get_scratch_bo(intel, &brw->wm.scratch_bo,
 331                          c->prog_data.total_scratch * brw->max_wm_threads);
 332    }
 333
 334    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 335       fprintf(stderr, "\n");
 336
 337    /* get the program
 338     */
 339    program = brw_get_program(&c->func, &program_size);
 340
 341    brw_upload_cache(&brw->cache, BRW_WM_PROG,
 342                     &c->key, sizeof(c->key),
 343                     program, program_size,
 344                     &c->prog_data, sizeof(c->prog_data),
 345                     &brw->wm.prog_offset, &brw->wm.prog_data);
 346
 347    return true;
 348 }
 349
 350 static bool
 351 key_debug(const char *name, int a, int b)
 352 {
 353    if (a != b) {
 354       perf_debug("  %s %d->%d\n", name, a, b);
 355       return true;
 356    } else {
 357       return false;
 358    }
 359 }
 360
 361 bool
 362 brw_debug_recompile_sampler_key(const struct brw_sampler_prog_key_data *old_key,
 363                                 const struct brw_sampler_prog_key_data *key)
 364 {
 365    bool found = false;
 366
 367    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 368       found |= key_debug("EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 369                          old_key->swizzles[i], key->swizzles[i]);
 370    }
 371    found |= key_debug("GL_CLAMP enabled on any texture unit's 1st coordinate",
 372                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 373    found |= key_debug("GL_CLAMP enabled on any texture unit's 2nd coordinate",
 374                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 375    found |= key_debug("GL_CLAMP enabled on any texture unit's 3rd coordinate",
 376                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 377    found |= key_debug("GL_MESA_ycbcr texturing\n",
 378                       old_key->yuvtex_mask, key->yuvtex_mask);
 379    found |= key_debug("GL_MESA_ycbcr UV swapping\n",
 380                       old_key->yuvtex_swap_mask, key->yuvtex_swap_mask);
 381
 382    return found;
 383 }
 384
 385 void
 386 brw_wm_debug_recompile(struct brw_context *brw,
 387                        struct gl_shader_program *prog,
 388                        const struct brw_wm_prog_key *key)
 389 {
 390    struct brw_cache_item *c = NULL;
 391    const struct brw_wm_prog_key *old_key = NULL;
 392    bool found = false;
 393
 394    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 395
 396    for (unsigned int i = 0; i < brw->cache.size; i++) {
 397       for (c = brw->cache.items[i]; c; c = c->next) {
 398          if (c->cache_id == BRW_WM_PROG) {
 399             old_key = c->key;
 400
 401             if (old_key->program_string_id == key->program_string_id)
 402                break;
 403          }
 404       }
 405       if (c)
 406          break;
 407    }
 408
 409    if (!c) {
 410       perf_debug("  Didn't find previous compile in the shader cache for "
 411                  "debug\n");
 412       return;
 413    }
 414
 415    found |= key_debug("alphatest, computed depth, depth test, or depth write",
 416                       old_key->iz_lookup, key->iz_lookup);
 417    found |= key_debug("depth statistics", old_key->stats_wm, key->stats_wm);
 418    found |= key_debug("flat shading", old_key->flat_shade, key->flat_shade);
 419    found |= key_debug("number of color buffers", old_key->nr_color_regions, key->nr_color_regions);
 420    found |= key_debug("rendering to FBO", old_key->render_to_fbo, key->render_to_fbo);
 421    found |= key_debug("fragment color clamping", old_key->clamp_fragment_color, key->clamp_fragment_color);
 422    found |= key_debug("line smoothing", old_key->line_aa, key->line_aa);
 423    found |= key_debug("proj_attrib_mask", old_key->proj_attrib_mask, key->proj_attrib_mask);
 424    found |= key_debug("renderbuffer height", old_key->drawable_height, key->drawable_height);
 425    found |= key_debug("vertex shader outputs", old_key->vp_outputs_written, key->vp_outputs_written);
 426
 427    found |= brw_debug_recompile_sampler_key(&old_key->tex, &key->tex);
 428
 429    if (!found) {
 430       perf_debug("  Something else\n");
 431    }
 432 }
 433
 434 void
 435 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 436                                    const struct gl_program *prog,
 437                                    struct brw_sampler_prog_key_data *key)
 438 {
 439    for (int s = 0; s < MAX_SAMPLERS; s++) {
 440       key->swizzles[s] = SWIZZLE_NOOP;
 441
 442       if (!(prog->SamplersUsed & (1 << s)))
 443          continue;
 444
 445       int unit_id = prog->SamplerUnits[s];
 446       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 447
 448       if (unit->_ReallyEnabled && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 449          const struct gl_texture_object *t = unit->_Current;
 450          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 451          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 452          int swizzles[SWIZZLE_NIL + 1] = {
 453             SWIZZLE_X,
 454             SWIZZLE_Y,
 455             SWIZZLE_Z,
 456             SWIZZLE_W,
 457             SWIZZLE_ZERO,
 458             SWIZZLE_ONE,
 459             SWIZZLE_NIL
 460          };
 461
 462          if (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 463              img->_BaseFormat == GL_DEPTH_STENCIL) {
 464             /* We handle GL_DEPTH_TEXTURE_MODE here instead of as surface
 465              * format overrides because shadow comparison always returns the
 466              * result of the comparison in all channels anyway.
 467              */
 468             switch (t->DepthMode) {
 469             case GL_ALPHA:
 470                swizzles[0] = SWIZZLE_ZERO;
 471                swizzles[1] = SWIZZLE_ZERO;
 472                swizzles[2] = SWIZZLE_ZERO;
 473                swizzles[3] = SWIZZLE_X;
 474                break;
 475             case GL_LUMINANCE:
 476                swizzles[0] = SWIZZLE_X;
 477                swizzles[1] = SWIZZLE_X;
 478                swizzles[2] = SWIZZLE_X;
 479                swizzles[3] = SWIZZLE_ONE;
 480                break;
 481             case GL_INTENSITY:
 482                swizzles[0] = SWIZZLE_X;
 483                swizzles[1] = SWIZZLE_X;
 484                swizzles[2] = SWIZZLE_X;
 485                swizzles[3] = SWIZZLE_X;
 486                break;
 487             case GL_RED:
 488                swizzles[0] = SWIZZLE_X;
 489                swizzles[1] = SWIZZLE_ZERO;
 490                swizzles[2] = SWIZZLE_ZERO;
 491                swizzles[3] = SWIZZLE_ONE;
 492                break;
 493             }
 494          }
 495
 496          if (img->InternalFormat == GL_YCBCR_MESA) {
 497             key->yuvtex_mask |= 1 << s;
 498             if (img->TexFormat == MESA_FORMAT_YCBCR)
 499                 key->yuvtex_swap_mask |= 1 << s;
 500          }
 501
 502          key->swizzles[s] =
 503             MAKE_SWIZZLE4(swizzles[GET_SWZ(t->_Swizzle, 0)],
 504                           swizzles[GET_SWZ(t->_Swizzle, 1)],
 505                           swizzles[GET_SWZ(t->_Swizzle, 2)],
 506                           swizzles[GET_SWZ(t->_Swizzle, 3)]);
 507
 508          if (sampler->MinFilter != GL_NEAREST &&
 509              sampler->MagFilter != GL_NEAREST) {
 510             if (sampler->WrapS == GL_CLAMP)
 511                key->gl_clamp_mask[0] |= 1 << s;
 512             if (sampler->WrapT == GL_CLAMP)
 513                key->gl_clamp_mask[1] |= 1 << s;
 514             if (sampler->WrapR == GL_CLAMP)
 515                key->gl_clamp_mask[2] |= 1 << s;
 516          }
 517       }
 518    }
 519 }
 520
 521 static void brw_wm_populate_key( struct brw_context *brw,
 522                                  struct brw_wm_prog_key *key )
 523 {
 524    struct gl_context *ctx = &brw->intel.ctx;
 525    struct intel_context *intel = &brw->intel;
 526    /* BRW_NEW_FRAGMENT_PROGRAM */
 527    const struct brw_fragment_program *fp =
 528       (struct brw_fragment_program *)brw->fragment_program;
 529    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 530    GLuint lookup = 0;
 531    GLuint line_aa;
 532    bool program_uses_dfdy = fp->program.UsesDFdy;
 533
 534    memset(key, 0, sizeof(*key));
 535
 536    /* Build the index for table lookup
 537     */
 538    if (intel->gen < 6) {
 539       /* _NEW_COLOR */
 540       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 541          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 542
 543       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 544          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 545
 546       /* _NEW_DEPTH */
 547       if (ctx->Depth.Test)
 548          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 549
 550       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 551          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 552
 553       /* _NEW_STENCIL */
 554       if (ctx->Stencil._Enabled) {
 555          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 556
 557          if (ctx->Stencil.WriteMask[0] ||
 558              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 559             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 560       }
 561       key->iz_lookup = lookup;
 562    }
 563
 564    line_aa = AA_NEVER;
 565
 566    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 567    if (ctx->Line.SmoothFlag) {
 568       if (brw->intel.reduced_primitive == GL_LINES) {
 569          line_aa = AA_ALWAYS;
 570       }
 571       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 572          if (ctx->Polygon.FrontMode == GL_LINE) {
 573             line_aa = AA_SOMETIMES;
 574
 575             if (ctx->Polygon.BackMode == GL_LINE ||
 576                 (ctx->Polygon.CullFlag &&
 577                  ctx->Polygon.CullFaceMode == GL_BACK))
 578                line_aa = AA_ALWAYS;
 579          }
 580          else if (ctx->Polygon.BackMode == GL_LINE) {
 581             line_aa = AA_SOMETIMES;
 582
 583             if ((ctx->Polygon.CullFlag &&
 584                  ctx->Polygon.CullFaceMode == GL_FRONT))
 585                line_aa = AA_ALWAYS;
 586          }
 587       }
 588    }
 589
 590    key->line_aa = line_aa;
 591
 592    if (intel->gen < 6)
 593       key->stats_wm = brw->intel.stats_wm;
 594
 595    /* BRW_NEW_WM_INPUT_DIMENSIONS */
 596    /* Only set this for fixed function.  The optimization it enables isn't
 597     * useful for programs using shaders.
 598     */
 599    if (ctx->Shader.CurrentFragmentProgram)
 600       key->proj_attrib_mask = 0xffffffff;
 601    else
 602       key->proj_attrib_mask = brw->wm.input_size_masks[4-1];
 603
 604    /* _NEW_LIGHT */
 605    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 606
 607    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 608    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 609
 610    /* _NEW_TEXTURE */
 611    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 612
 613    /* _NEW_BUFFERS */
 614    /*
 615     * Include the draw buffer origin and height so that we can calculate
 616     * fragment position values relative to the bottom left of the drawable,
 617     * from the incoming screen origin relative position we get as part of our
 618     * payload.
 619     *
 620     * This is only needed for the WM_WPOSXY opcode when the fragment program
 621     * uses the gl_FragCoord input.
 622     *
 623     * We could avoid recompiling by including this as a constant referenced by
 624     * our program, but if we were to do that it would also be nice to handle
 625     * getting that constant updated at batchbuffer submit time (when we
 626     * hold the lock and know where the buffer really is) rather than at emit
 627     * time when we don't hold the lock and are just guessing.  We could also
 628     * just avoid using this as key data if the program doesn't use
 629     * fragment.position.
 630     *
 631     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 632     * drawable height in order to invert the Y axis.
 633     */
 634    if (fp->program.Base.InputsRead & FRAG_BIT_WPOS) {
 635       key->drawable_height = ctx->DrawBuffer->Height;
 636    }
 637
 638    if ((fp->program.Base.InputsRead & FRAG_BIT_WPOS) || program_uses_dfdy) {
 639       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 640    }
 641
 642    /* _NEW_BUFFERS */
 643    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 644   /* _NEW_MULTISAMPLE */
 645    key->sample_alpha_to_coverage = ctx->Multisample.SampleAlphaToCoverage;
 646
 647    /* CACHE_NEW_VS_PROG */
 648    if (intel->gen < 6)
 649       key->vp_outputs_written = brw->vs.prog_data->outputs_written;
 650
 651    /* The unique fragment program ID */
 652    key->program_string_id = fp->id;
 653 }
 654
 655
 656 static void
 657 brw_upload_wm_prog(struct brw_context *brw)
 658 {
 659    struct intel_context *intel = &brw->intel;
 660    struct gl_context *ctx = &intel->ctx;
 661    struct brw_wm_prog_key key;
 662    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 663       brw->fragment_program;
 664
 665    brw_wm_populate_key(brw, &key);
 666
 667    if (!brw_search_cache(&brw->cache, BRW_WM_PROG,
 668                          &key, sizeof(key),
 669                          &brw->wm.prog_offset, &brw->wm.prog_data)) {
 670       bool success = do_wm_prog(brw, ctx->Shader._CurrentFragmentProgram, fp,
 671                                 &key);
 672       (void) success;
 673       assert(success);
 674    }
 675 }
 676
 677
 678 const struct brw_tracked_state brw_wm_prog = {
 679    .dirty = {
 680       .mesa  = (_NEW_COLOR |
 681                 _NEW_DEPTH |
 682                 _NEW_STENCIL |
 683                 _NEW_POLYGON |
 684                 _NEW_LINE |
 685                 _NEW_LIGHT |
 686                 _NEW_FRAG_CLAMP |
 687                 _NEW_BUFFERS |
 688                 _NEW_TEXTURE),
 689       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 690                 BRW_NEW_WM_INPUT_DIMENSIONS |
 691                 BRW_NEW_REDUCED_PRIMITIVE),
 692       .cache = CACHE_NEW_VS_PROG,
 693    },
 694    .emit = brw_upload_wm_prog
 695 };
 696