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 /**
  43  * Return a bitfield where bit n is set if barycentric interpolation mode n
  44  * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
  45  */
  46 static unsigned
  47 brw_compute_barycentric_interp_modes(struct brw_context *brw,
  48                                      bool shade_model_flat,
  49                                      const struct gl_fragment_program *fprog)
  50 {
  51    unsigned barycentric_interp_modes = 0;
  52    int attr;
  53
  54    /* Loop through all fragment shader inputs to figure out what interpolation
  55     * modes are in use, and set the appropriate bits in
  56     * barycentric_interp_modes.
  57     */
  58    for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
  59       enum glsl_interp_qualifier interp_qualifier =
  60          fprog->InterpQualifier[attr];
  61       bool is_centroid = fprog->IsCentroid & BITFIELD64_BIT(attr);
  62       bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
  63
  64       /* Ignore unused inputs. */
  65       if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
  66          continue;
  67
  68       /* Ignore WPOS and FACE, because they don't require interpolation. */
  69       if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
  70          continue;
  71
  72       /* Determine the set (or sets) of barycentric coordinates needed to
  73        * interpolate this variable.  Note that when
  74        * brw->needs_unlit_centroid_workaround is set, centroid interpolation
  75        * uses PIXEL interpolation for unlit pixels and CENTROID interpolation
  76        * for lit pixels, so we need both sets of barycentric coordinates.
  77        */
  78       if (interp_qualifier == INTERP_QUALIFIER_NOPERSPECTIVE) {
  79          if (is_centroid) {
  80             barycentric_interp_modes |=
  81                1 << BRW_WM_NONPERSPECTIVE_CENTROID_BARYCENTRIC;
  82          }
  83          if (!is_centroid || brw->needs_unlit_centroid_workaround) {
  84             barycentric_interp_modes |=
  85                1 << BRW_WM_NONPERSPECTIVE_PIXEL_BARYCENTRIC;
  86          }
  87       } else if (interp_qualifier == INTERP_QUALIFIER_SMOOTH ||
  88                  (!(shade_model_flat && is_gl_Color) &&
  89                   interp_qualifier == INTERP_QUALIFIER_NONE)) {
  90          if (is_centroid) {
  91             barycentric_interp_modes |=
  92                1 << BRW_WM_PERSPECTIVE_CENTROID_BARYCENTRIC;
  93          }
  94          if (!is_centroid || brw->needs_unlit_centroid_workaround) {
  95             barycentric_interp_modes |=
  96                1 << BRW_WM_PERSPECTIVE_PIXEL_BARYCENTRIC;
  97          }
  98       }
  99    }
 100
 101    return barycentric_interp_modes;
 102 }
 103
 104 bool
 105 brw_wm_prog_data_compare(const void *in_a, const void *in_b,
 106                          int aux_size, const void *in_key)
 107 {
 108    const struct brw_wm_prog_data *a = in_a;
 109    const struct brw_wm_prog_data *b = in_b;
 110
 111    /* Compare all the struct up to the pointers. */
 112    if (memcmp(a, b, offsetof(struct brw_wm_prog_data, param)))
 113       return false;
 114
 115    if (memcmp(a->param, b->param, a->nr_params * sizeof(void *)))
 116       return false;
 117
 118    if (memcmp(a->pull_param, b->pull_param, a->nr_pull_params * sizeof(void *)))
 119       return false;
 120
 121    return true;
 122 }
 123
 124 void
 125 brw_wm_prog_data_free(const void *in_prog_data)
 126 {
 127    const struct brw_wm_prog_data *prog_data = in_prog_data;
 128
 129    ralloc_free((void *)prog_data->param);
 130    ralloc_free((void *)prog_data->pull_param);
 131 }
 132
 133 /**
 134  * All Mesa program -> GPU code generation goes through this function.
 135  * Depending on the instructions used (i.e. flow control instructions)
 136  * we'll use one of two code generators.
 137  */
 138 bool do_wm_prog(struct brw_context *brw,
 139                 struct gl_shader_program *prog,
 140                 struct brw_fragment_program *fp,
 141                 struct brw_wm_prog_key *key)
 142 {
 143    struct intel_context *intel = &brw->intel;
 144    struct brw_wm_compile *c;
 145    const GLuint *program;
 146    struct gl_shader *fs = NULL;
 147    GLuint program_size;
 148
 149    if (prog)
 150       fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 151
 152    c = rzalloc(NULL, struct brw_wm_compile);
 153
 154    /* Allocate the references to the uniforms that will end up in the
 155     * prog_data associated with the compiled program, and which will be freed
 156     * by the state cache.
 157     */
 158    int param_count;
 159    if (fs) {
 160       param_count = fs->num_uniform_components;
 161    } else {
 162       param_count = fp->program.Base.Parameters->NumParameters * 4;
 163    }
 164    /* The backend also sometimes adds params for texture size. */
 165    param_count += 2 * BRW_MAX_TEX_UNIT;
 166    c->prog_data.param = rzalloc_array(NULL, const float *, param_count);
 167    c->prog_data.pull_param = rzalloc_array(NULL, const float *, param_count);
 168
 169    memcpy(&c->key, key, sizeof(*key));
 170
 171    c->prog_data.barycentric_interp_modes =
 172       brw_compute_barycentric_interp_modes(brw, c->key.flat_shade,
 173                                            &fp->program);
 174
 175    program = brw_wm_fs_emit(brw, c, &fp->program, prog, &program_size);
 176    if (program == NULL)
 177       return false;
 178
 179    /* Scratch space is used for register spilling */
 180    if (c->last_scratch) {
 181       perf_debug("Fragment shader triggered register spilling.  "
 182                  "Try reducing the number of live scalar values to "
 183                  "improve performance.\n");
 184
 185       c->prog_data.total_scratch = brw_get_scratch_size(c->last_scratch);
 186
 187       brw_get_scratch_bo(intel, &brw->wm.scratch_bo,
 188                          c->prog_data.total_scratch * brw->max_wm_threads);
 189    }
 190
 191    if (unlikely(INTEL_DEBUG & DEBUG_WM))
 192       fprintf(stderr, "\n");
 193
 194    brw_upload_cache(&brw->cache, BRW_WM_PROG,
 195                     &c->key, sizeof(c->key),
 196                     program, program_size,
 197                     &c->prog_data, sizeof(c->prog_data),
 198                     &brw->wm.prog_offset, &brw->wm.prog_data);
 199
 200    ralloc_free(c);
 201
 202    return true;
 203 }
 204
 205 static bool
 206 key_debug(struct intel_context *intel, const char *name, int a, int b)
 207 {
 208    if (a != b) {
 209       perf_debug("  %s %d->%d\n", name, a, b);
 210       return true;
 211    } else {
 212       return false;
 213    }
 214 }
 215
 216 bool
 217 brw_debug_recompile_sampler_key(struct intel_context *intel,
 218                                 const struct brw_sampler_prog_key_data *old_key,
 219                                 const struct brw_sampler_prog_key_data *key)
 220 {
 221    bool found = false;
 222
 223    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 224       found |= key_debug(intel, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 225                          old_key->swizzles[i], key->swizzles[i]);
 226    }
 227    found |= key_debug(intel, "GL_CLAMP enabled on any texture unit's 1st coordinate",
 228                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 229    found |= key_debug(intel, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
 230                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 231    found |= key_debug(intel, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
 232                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 233    found |= key_debug(intel, "GL_MESA_ycbcr texturing\n",
 234                       old_key->yuvtex_mask, key->yuvtex_mask);
 235    found |= key_debug(intel, "GL_MESA_ycbcr UV swapping\n",
 236                       old_key->yuvtex_swap_mask, key->yuvtex_swap_mask);
 237
 238    return found;
 239 }
 240
 241 void
 242 brw_wm_debug_recompile(struct brw_context *brw,
 243                        struct gl_shader_program *prog,
 244                        const struct brw_wm_prog_key *key)
 245 {
 246    struct intel_context *intel = &brw->intel;
 247    struct brw_cache_item *c = NULL;
 248    const struct brw_wm_prog_key *old_key = NULL;
 249    bool found = false;
 250
 251    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 252
 253    for (unsigned int i = 0; i < brw->cache.size; i++) {
 254       for (c = brw->cache.items[i]; c; c = c->next) {
 255          if (c->cache_id == BRW_WM_PROG) {
 256             old_key = c->key;
 257
 258             if (old_key->program_string_id == key->program_string_id)
 259                break;
 260          }
 261       }
 262       if (c)
 263          break;
 264    }
 265
 266    if (!c) {
 267       perf_debug("  Didn't find previous compile in the shader cache for debug\n");
 268       return;
 269    }
 270
 271    found |= key_debug(intel, "alphatest, computed depth, depth test, or "
 272                       "depth write",
 273                       old_key->iz_lookup, key->iz_lookup);
 274    found |= key_debug(intel, "depth statistics",
 275                       old_key->stats_wm, key->stats_wm);
 276    found |= key_debug(intel, "flat shading",
 277                       old_key->flat_shade, key->flat_shade);
 278    found |= key_debug(intel, "number of color buffers",
 279                       old_key->nr_color_regions, key->nr_color_regions);
 280    found |= key_debug(intel, "MRT alpha test or alpha-to-coverage",
 281                       old_key->replicate_alpha, key->replicate_alpha);
 282    found |= key_debug(intel, "rendering to FBO",
 283                       old_key->render_to_fbo, key->render_to_fbo);
 284    found |= key_debug(intel, "fragment color clamping",
 285                       old_key->clamp_fragment_color, key->clamp_fragment_color);
 286    found |= key_debug(intel, "line smoothing",
 287                       old_key->line_aa, key->line_aa);
 288    found |= key_debug(intel, "renderbuffer height",
 289                       old_key->drawable_height, key->drawable_height);
 290    found |= key_debug(intel, "input slots valid",
 291                       old_key->input_slots_valid, key->input_slots_valid);
 292
 293    found |= brw_debug_recompile_sampler_key(intel, &old_key->tex, &key->tex);
 294
 295    if (!found) {
 296       perf_debug("  Something else\n");
 297    }
 298 }
 299
 300 void
 301 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 302                                    const struct gl_program *prog,
 303                                    struct brw_sampler_prog_key_data *key)
 304 {
 305    struct intel_context *intel = intel_context(ctx);
 306
 307    for (int s = 0; s < MAX_SAMPLERS; s++) {
 308       key->swizzles[s] = SWIZZLE_NOOP;
 309
 310       if (!(prog->SamplersUsed & (1 << s)))
 311          continue;
 312
 313       int unit_id = prog->SamplerUnits[s];
 314       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 315
 316       if (unit->_ReallyEnabled && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 317          const struct gl_texture_object *t = unit->_Current;
 318          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 319          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 320
 321          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 322             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 323              img->_BaseFormat == GL_DEPTH_STENCIL);
 324
 325          /* Haswell handles texture swizzling as surface format overrides
 326           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 327           */
 328          if (!intel->is_haswell || alpha_depth)
 329             key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
 330
 331          if (img->InternalFormat == GL_YCBCR_MESA) {
 332             key->yuvtex_mask |= 1 << s;
 333             if (img->TexFormat == MESA_FORMAT_YCBCR)
 334                 key->yuvtex_swap_mask |= 1 << s;
 335          }
 336
 337          if (sampler->MinFilter != GL_NEAREST &&
 338              sampler->MagFilter != GL_NEAREST) {
 339             if (sampler->WrapS == GL_CLAMP)
 340                key->gl_clamp_mask[0] |= 1 << s;
 341             if (sampler->WrapT == GL_CLAMP)
 342                key->gl_clamp_mask[1] |= 1 << s;
 343             if (sampler->WrapR == GL_CLAMP)
 344                key->gl_clamp_mask[2] |= 1 << s;
 345          }
 346       }
 347    }
 348 }
 349
 350 static void brw_wm_populate_key( struct brw_context *brw,
 351                                  struct brw_wm_prog_key *key )
 352 {
 353    struct gl_context *ctx = &brw->intel.ctx;
 354    struct intel_context *intel = &brw->intel;
 355    /* BRW_NEW_FRAGMENT_PROGRAM */
 356    const struct brw_fragment_program *fp =
 357       (struct brw_fragment_program *)brw->fragment_program;
 358    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 359    GLuint lookup = 0;
 360    GLuint line_aa;
 361    bool program_uses_dfdy = fp->program.UsesDFdy;
 362
 363    memset(key, 0, sizeof(*key));
 364
 365    /* Build the index for table lookup
 366     */
 367    if (intel->gen < 6) {
 368       /* _NEW_COLOR */
 369       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 370          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 371
 372       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 373          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 374
 375       /* _NEW_DEPTH */
 376       if (ctx->Depth.Test)
 377          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 378
 379       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 380          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 381
 382       /* _NEW_STENCIL | _NEW_BUFFERS */
 383       if (ctx->Stencil._Enabled) {
 384          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 385
 386          if (ctx->Stencil.WriteMask[0] ||
 387              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 388             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 389       }
 390       key->iz_lookup = lookup;
 391    }
 392
 393    line_aa = AA_NEVER;
 394
 395    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 396    if (ctx->Line.SmoothFlag) {
 397       if (brw->intel.reduced_primitive == GL_LINES) {
 398          line_aa = AA_ALWAYS;
 399       }
 400       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 401          if (ctx->Polygon.FrontMode == GL_LINE) {
 402             line_aa = AA_SOMETIMES;
 403
 404             if (ctx->Polygon.BackMode == GL_LINE ||
 405                 (ctx->Polygon.CullFlag &&
 406                  ctx->Polygon.CullFaceMode == GL_BACK))
 407                line_aa = AA_ALWAYS;
 408          }
 409          else if (ctx->Polygon.BackMode == GL_LINE) {
 410             line_aa = AA_SOMETIMES;
 411
 412             if ((ctx->Polygon.CullFlag &&
 413                  ctx->Polygon.CullFaceMode == GL_FRONT))
 414                line_aa = AA_ALWAYS;
 415          }
 416       }
 417    }
 418
 419    key->line_aa = line_aa;
 420
 421    if (intel->gen < 6)
 422       key->stats_wm = brw->intel.stats_wm;
 423
 424    /* _NEW_LIGHT */
 425    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 426
 427    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 428    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 429
 430    /* _NEW_TEXTURE */
 431    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 432
 433    /* _NEW_BUFFERS */
 434    /*
 435     * Include the draw buffer origin and height so that we can calculate
 436     * fragment position values relative to the bottom left of the drawable,
 437     * from the incoming screen origin relative position we get as part of our
 438     * payload.
 439     *
 440     * This is only needed for the WM_WPOSXY opcode when the fragment program
 441     * uses the gl_FragCoord input.
 442     *
 443     * We could avoid recompiling by including this as a constant referenced by
 444     * our program, but if we were to do that it would also be nice to handle
 445     * getting that constant updated at batchbuffer submit time (when we
 446     * hold the lock and know where the buffer really is) rather than at emit
 447     * time when we don't hold the lock and are just guessing.  We could also
 448     * just avoid using this as key data if the program doesn't use
 449     * fragment.position.
 450     *
 451     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 452     * drawable height in order to invert the Y axis.
 453     */
 454    if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
 455       key->drawable_height = ctx->DrawBuffer->Height;
 456    }
 457
 458    if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
 459       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 460    }
 461
 462    /* _NEW_BUFFERS */
 463    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 464
 465    /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
 466    key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
 467       (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
 468
 469    /* BRW_NEW_VUE_MAP_GEOM_OUT */
 470    if (intel->gen < 6)
 471       key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
 472
 473    /* The unique fragment program ID */
 474    key->program_string_id = fp->id;
 475 }
 476
 477
 478 static void
 479 brw_upload_wm_prog(struct brw_context *brw)
 480 {
 481    struct intel_context *intel = &brw->intel;
 482    struct gl_context *ctx = &intel->ctx;
 483    struct brw_wm_prog_key key;
 484    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 485       brw->fragment_program;
 486
 487    brw_wm_populate_key(brw, &key);
 488
 489    if (!brw_search_cache(&brw->cache, BRW_WM_PROG,
 490                          &key, sizeof(key),
 491                          &brw->wm.prog_offset, &brw->wm.prog_data)) {
 492       bool success = do_wm_prog(brw, ctx->Shader._CurrentFragmentProgram, fp,
 493                                 &key);
 494       (void) success;
 495       assert(success);
 496    }
 497 }
 498
 499
 500 const struct brw_tracked_state brw_wm_prog = {
 501    .dirty = {
 502       .mesa  = (_NEW_COLOR |
 503                 _NEW_DEPTH |
 504                 _NEW_STENCIL |
 505                 _NEW_POLYGON |
 506                 _NEW_LINE |
 507                 _NEW_LIGHT |
 508                 _NEW_FRAG_CLAMP |
 509                 _NEW_BUFFERS |
 510                 _NEW_TEXTURE |
 511                 _NEW_MULTISAMPLE),
 512       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 513                 BRW_NEW_REDUCED_PRIMITIVE |
 514                 BRW_NEW_VUE_MAP_GEOM_OUT |
 515                 BRW_NEW_STATS_WM)
 516    },
 517    .emit = brw_upload_wm_prog
 518 };
 519