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