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 < FRAG_ATTRIB_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 == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_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 == FRAG_ATTRIB_WPOS || attr == FRAG_ATTRIB_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    brw_init_compile(brw, &c->func, c);
 172
 173    c->prog_data.barycentric_interp_modes =
 174       brw_compute_barycentric_interp_modes(brw, c->key.flat_shade,
 175                                            &fp->program);
 176
 177    brw_wm_fs_emit(brw, c, &fp->program, prog);
 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    /* get the program
 195     */
 196    program = brw_get_program(&c->func, &program_size);
 197
 198    brw_upload_cache(&brw->cache, BRW_WM_PROG,
 199                     &c->key, sizeof(c->key),
 200                     program, program_size,
 201                     &c->prog_data, sizeof(c->prog_data),
 202                     &brw->wm.prog_offset, &brw->wm.prog_data);
 203
 204    ralloc_free(c);
 205
 206    return true;
 207 }
 208
 209 static bool
 210 key_debug(const char *name, int a, int b)
 211 {
 212    if (a != b) {
 213       perf_debug("  %s %d->%d\n", name, a, b);
 214       return true;
 215    } else {
 216       return false;
 217    }
 218 }
 219
 220 bool
 221 brw_debug_recompile_sampler_key(const struct brw_sampler_prog_key_data *old_key,
 222                                 const struct brw_sampler_prog_key_data *key)
 223 {
 224    bool found = false;
 225
 226    for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
 227       found |= key_debug("EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
 228                          old_key->swizzles[i], key->swizzles[i]);
 229    }
 230    found |= key_debug("GL_CLAMP enabled on any texture unit's 1st coordinate",
 231                       old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
 232    found |= key_debug("GL_CLAMP enabled on any texture unit's 2nd coordinate",
 233                       old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
 234    found |= key_debug("GL_CLAMP enabled on any texture unit's 3rd coordinate",
 235                       old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
 236    found |= key_debug("GL_MESA_ycbcr texturing\n",
 237                       old_key->yuvtex_mask, key->yuvtex_mask);
 238    found |= key_debug("GL_MESA_ycbcr UV swapping\n",
 239                       old_key->yuvtex_swap_mask, key->yuvtex_swap_mask);
 240
 241    return found;
 242 }
 243
 244 void
 245 brw_wm_debug_recompile(struct brw_context *brw,
 246                        struct gl_shader_program *prog,
 247                        const struct brw_wm_prog_key *key)
 248 {
 249    struct brw_cache_item *c = NULL;
 250    const struct brw_wm_prog_key *old_key = NULL;
 251    bool found = false;
 252
 253    perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
 254
 255    for (unsigned int i = 0; i < brw->cache.size; i++) {
 256       for (c = brw->cache.items[i]; c; c = c->next) {
 257          if (c->cache_id == BRW_WM_PROG) {
 258             old_key = c->key;
 259
 260             if (old_key->program_string_id == key->program_string_id)
 261                break;
 262          }
 263       }
 264       if (c)
 265          break;
 266    }
 267
 268    if (!c) {
 269       perf_debug("  Didn't find previous compile in the shader cache for "
 270                  "debug\n");
 271       return;
 272    }
 273
 274    found |= key_debug("alphatest, computed depth, depth test, or depth write",
 275                       old_key->iz_lookup, key->iz_lookup);
 276    found |= key_debug("depth statistics", old_key->stats_wm, key->stats_wm);
 277    found |= key_debug("flat shading", old_key->flat_shade, key->flat_shade);
 278    found |= key_debug("number of color buffers", old_key->nr_color_regions, key->nr_color_regions);
 279    found |= key_debug("sample alpha to coverage", old_key->sample_alpha_to_coverage, key->sample_alpha_to_coverage);
 280    found |= key_debug("rendering to FBO", old_key->render_to_fbo, key->render_to_fbo);
 281    found |= key_debug("fragment color clamping", old_key->clamp_fragment_color, key->clamp_fragment_color);
 282    found |= key_debug("line smoothing", old_key->line_aa, key->line_aa);
 283    found |= key_debug("proj_attrib_mask", old_key->proj_attrib_mask, key->proj_attrib_mask);
 284    found |= key_debug("renderbuffer height", old_key->drawable_height, key->drawable_height);
 285    found |= key_debug("vertex shader outputs", old_key->vp_outputs_written, key->vp_outputs_written);
 286
 287    found |= brw_debug_recompile_sampler_key(&old_key->tex, &key->tex);
 288
 289    if (!found) {
 290       perf_debug("  Something else\n");
 291    }
 292 }
 293
 294 void
 295 brw_populate_sampler_prog_key_data(struct gl_context *ctx,
 296                                    const struct gl_program *prog,
 297                                    struct brw_sampler_prog_key_data *key)
 298 {
 299    struct intel_context *intel = intel_context(ctx);
 300
 301    for (int s = 0; s < MAX_SAMPLERS; s++) {
 302       key->swizzles[s] = SWIZZLE_NOOP;
 303
 304       if (!(prog->SamplersUsed & (1 << s)))
 305          continue;
 306
 307       int unit_id = prog->SamplerUnits[s];
 308       const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
 309
 310       if (unit->_ReallyEnabled && unit->_Current->Target != GL_TEXTURE_BUFFER) {
 311          const struct gl_texture_object *t = unit->_Current;
 312          const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
 313          struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
 314
 315          const bool alpha_depth = t->DepthMode == GL_ALPHA &&
 316             (img->_BaseFormat == GL_DEPTH_COMPONENT ||
 317              img->_BaseFormat == GL_DEPTH_STENCIL);
 318
 319          /* Haswell handles texture swizzling as surface format overrides
 320           * (except for GL_ALPHA); all other platforms need MOVs in the shader.
 321           */
 322          if (!intel->is_haswell || alpha_depth)
 323             key->swizzles[s] = brw_get_texture_swizzle(t);
 324
 325          if (img->InternalFormat == GL_YCBCR_MESA) {
 326             key->yuvtex_mask |= 1 << s;
 327             if (img->TexFormat == MESA_FORMAT_YCBCR)
 328                 key->yuvtex_swap_mask |= 1 << s;
 329          }
 330
 331          if (sampler->MinFilter != GL_NEAREST &&
 332              sampler->MagFilter != GL_NEAREST) {
 333             if (sampler->WrapS == GL_CLAMP)
 334                key->gl_clamp_mask[0] |= 1 << s;
 335             if (sampler->WrapT == GL_CLAMP)
 336                key->gl_clamp_mask[1] |= 1 << s;
 337             if (sampler->WrapR == GL_CLAMP)
 338                key->gl_clamp_mask[2] |= 1 << s;
 339          }
 340       }
 341    }
 342 }
 343
 344 static void brw_wm_populate_key( struct brw_context *brw,
 345                                  struct brw_wm_prog_key *key )
 346 {
 347    struct gl_context *ctx = &brw->intel.ctx;
 348    struct intel_context *intel = &brw->intel;
 349    /* BRW_NEW_FRAGMENT_PROGRAM */
 350    const struct brw_fragment_program *fp =
 351       (struct brw_fragment_program *)brw->fragment_program;
 352    const struct gl_program *prog = (struct gl_program *) brw->fragment_program;
 353    GLuint lookup = 0;
 354    GLuint line_aa;
 355    bool program_uses_dfdy = fp->program.UsesDFdy;
 356
 357    memset(key, 0, sizeof(*key));
 358
 359    /* Build the index for table lookup
 360     */
 361    if (intel->gen < 6) {
 362       /* _NEW_COLOR */
 363       if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
 364          lookup |= IZ_PS_KILL_ALPHATEST_BIT;
 365
 366       if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
 367          lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
 368
 369       /* _NEW_DEPTH */
 370       if (ctx->Depth.Test)
 371          lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
 372
 373       if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
 374          lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
 375
 376       /* _NEW_STENCIL */
 377       if (ctx->Stencil._Enabled) {
 378          lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
 379
 380          if (ctx->Stencil.WriteMask[0] ||
 381              ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
 382             lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
 383       }
 384       key->iz_lookup = lookup;
 385    }
 386
 387    line_aa = AA_NEVER;
 388
 389    /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
 390    if (ctx->Line.SmoothFlag) {
 391       if (brw->intel.reduced_primitive == GL_LINES) {
 392          line_aa = AA_ALWAYS;
 393       }
 394       else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
 395          if (ctx->Polygon.FrontMode == GL_LINE) {
 396             line_aa = AA_SOMETIMES;
 397
 398             if (ctx->Polygon.BackMode == GL_LINE ||
 399                 (ctx->Polygon.CullFlag &&
 400                  ctx->Polygon.CullFaceMode == GL_BACK))
 401                line_aa = AA_ALWAYS;
 402          }
 403          else if (ctx->Polygon.BackMode == GL_LINE) {
 404             line_aa = AA_SOMETIMES;
 405
 406             if ((ctx->Polygon.CullFlag &&
 407                  ctx->Polygon.CullFaceMode == GL_FRONT))
 408                line_aa = AA_ALWAYS;
 409          }
 410       }
 411    }
 412
 413    key->line_aa = line_aa;
 414
 415    if (intel->gen < 6)
 416       key->stats_wm = brw->intel.stats_wm;
 417
 418    /* BRW_NEW_WM_INPUT_DIMENSIONS */
 419    /* Only set this for fixed function.  The optimization it enables isn't
 420     * useful for programs using shaders.
 421     */
 422    if (ctx->Shader.CurrentFragmentProgram)
 423       key->proj_attrib_mask = 0xffffffff;
 424    else
 425       key->proj_attrib_mask = brw->wm.input_size_masks[4-1];
 426
 427    /* _NEW_LIGHT */
 428    key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
 429
 430    /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
 431    key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
 432
 433    /* _NEW_TEXTURE */
 434    brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
 435
 436    /* _NEW_BUFFERS */
 437    /*
 438     * Include the draw buffer origin and height so that we can calculate
 439     * fragment position values relative to the bottom left of the drawable,
 440     * from the incoming screen origin relative position we get as part of our
 441     * payload.
 442     *
 443     * This is only needed for the WM_WPOSXY opcode when the fragment program
 444     * uses the gl_FragCoord input.
 445     *
 446     * We could avoid recompiling by including this as a constant referenced by
 447     * our program, but if we were to do that it would also be nice to handle
 448     * getting that constant updated at batchbuffer submit time (when we
 449     * hold the lock and know where the buffer really is) rather than at emit
 450     * time when we don't hold the lock and are just guessing.  We could also
 451     * just avoid using this as key data if the program doesn't use
 452     * fragment.position.
 453     *
 454     * For DRI2 the origin_x/y will always be (0,0) but we still need the
 455     * drawable height in order to invert the Y axis.
 456     */
 457    if (fp->program.Base.InputsRead & FRAG_BIT_WPOS) {
 458       key->drawable_height = ctx->DrawBuffer->Height;
 459    }
 460
 461    if ((fp->program.Base.InputsRead & FRAG_BIT_WPOS) || program_uses_dfdy) {
 462       key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
 463    }
 464
 465    /* _NEW_BUFFERS */
 466    key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
 467   /* _NEW_MULTISAMPLE */
 468    key->sample_alpha_to_coverage = ctx->Multisample.SampleAlphaToCoverage;
 469
 470    /* CACHE_NEW_VS_PROG */
 471    if (intel->gen < 6)
 472       key->vp_outputs_written = brw->vs.prog_data->outputs_written;
 473
 474    /* The unique fragment program ID */
 475    key->program_string_id = fp->id;
 476 }
 477
 478
 479 static void
 480 brw_upload_wm_prog(struct brw_context *brw)
 481 {
 482    struct intel_context *intel = &brw->intel;
 483    struct gl_context *ctx = &intel->ctx;
 484    struct brw_wm_prog_key key;
 485    struct brw_fragment_program *fp = (struct brw_fragment_program *)
 486       brw->fragment_program;
 487
 488    brw_wm_populate_key(brw, &key);
 489
 490    if (!brw_search_cache(&brw->cache, BRW_WM_PROG,
 491                          &key, sizeof(key),
 492                          &brw->wm.prog_offset, &brw->wm.prog_data)) {
 493       bool success = do_wm_prog(brw, ctx->Shader._CurrentFragmentProgram, fp,
 494                                 &key);
 495       (void) success;
 496       assert(success);
 497    }
 498 }
 499
 500
 501 const struct brw_tracked_state brw_wm_prog = {
 502    .dirty = {
 503       .mesa  = (_NEW_COLOR |
 504                 _NEW_DEPTH |
 505                 _NEW_STENCIL |
 506                 _NEW_POLYGON |
 507                 _NEW_LINE |
 508                 _NEW_LIGHT |
 509                 _NEW_FRAG_CLAMP |
 510                 _NEW_BUFFERS |
 511                 _NEW_TEXTURE |
 512                 _NEW_MULTISAMPLE),
 513       .brw   = (BRW_NEW_FRAGMENT_PROGRAM |
 514                 BRW_NEW_WM_INPUT_DIMENSIONS |
 515                 BRW_NEW_REDUCED_PRIMITIVE),
 516       .cache = CACHE_NEW_VS_PROG,
 517    },
 518    .emit = brw_upload_wm_prog
 519 };
 520