src/mesa/drivers/dri/i965/brw_state_upload.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
  33
  34 #include "brw_context.h"
  35 #include "brw_state.h"
  36 #include "intel_batchbuffer.h"
  37 #include "intel_buffers.h"
  38
  39 /* This is used to initialize brw->state.atoms[].  We could use this
  40  * list directly except for a single atom, brw_constant_buffer, which
  41  * has a .dirty value which changes according to the parameters of the
  42  * current fragment and vertex programs, and so cannot be a static
  43  * value.
  44  */
  45 static const struct brw_tracked_state *gen4_atoms[] =
  46 {
  47    &brw_check_fallback,
  48
  49    &brw_wm_input_sizes,
  50    &brw_vs_prog,
  51    &brw_gs_prog,
  52    &brw_clip_prog,
  53    &brw_sf_prog,
  54    &brw_wm_prog,
  55
  56    /* Once all the programs are done, we know how large urb entry
  57     * sizes need to be and can decide if we need to change the urb
  58     * layout.
  59     */
  60    &brw_curbe_offsets,
  61    &brw_recalculate_urb_fence,
  62
  63    &brw_cc_unit,
  64
  65    &brw_vs_constants, /* Before vs_surfaces and constant_buffer */
  66    &brw_wm_constants, /* Before wm_surfaces and constant_buffer */
  67
  68    &brw_vs_surfaces,            /* must do before unit */
  69    &brw_wm_constant_surface,    /* must do before wm surfaces/bind bo */
  70    &brw_wm_surfaces,            /* must do before samplers and unit */
  71    &brw_wm_binding_table,
  72    &brw_wm_samplers,
  73
  74    &brw_wm_unit,
  75    &brw_sf_vp,
  76    &brw_sf_unit,
  77    &brw_vs_unit,                /* always required, enabled or not */
  78    &brw_clip_unit,
  79    &brw_gs_unit,
  80
  81    /* Command packets:
  82     */
  83    &brw_invarient_state,
  84    &brw_state_base_address,
  85
  86    &brw_binding_table_pointers,
  87    &brw_blend_constant_color,
  88
  89    &brw_depthbuffer,
  90
  91    &brw_polygon_stipple,
  92    &brw_polygon_stipple_offset,
  93
  94    &brw_line_stipple,
  95    &brw_aa_line_parameters,
  96
  97    &brw_psp_urb_cbs,
  98
  99    &brw_drawing_rect,
 100    &brw_indices,
 101    &brw_index_buffer,
 102    &brw_vertices,
 103
 104    &brw_constant_buffer
 105 };
 106
 107 const struct brw_tracked_state *gen6_atoms[] =
 108 {
 109    &brw_check_fallback,
 110
 111    &brw_wm_input_sizes,
 112    &brw_vs_prog,
 113    &brw_gs_prog,
 114    &brw_wm_prog,
 115
 116    &gen6_clip_vp,
 117    &gen6_sf_vp,
 118
 119    /* Command packets: */
 120    &brw_invarient_state,
 121
 122    &gen6_viewport_state,        /* must do after *_vp stages */
 123
 124    &gen6_urb,
 125    &gen6_blend_state,           /* must do before cc unit */
 126    &gen6_color_calc_state,      /* must do before cc unit */
 127    &gen6_depth_stencil_state,   /* must do before cc unit */
 128    &gen6_cc_state_pointers,
 129
 130    &brw_vs_constants, /* Before vs_surfaces and constant_buffer */
 131    &brw_wm_constants, /* Before wm_surfaces and constant_buffer */
 132    &gen6_wm_constants, /* Before wm_surfaces and constant_buffer */
 133
 134    &brw_vs_surfaces,            /* must do before unit */
 135    &brw_wm_constant_surface,    /* must do before wm surfaces/bind bo */
 136    &brw_wm_surfaces,            /* must do before samplers and unit */
 137    &brw_wm_binding_table,
 138
 139    &brw_wm_samplers,
 140    &gen6_sampler_state,
 141
 142    &gen6_vs_state,
 143    &gen6_gs_state,
 144    &gen6_clip_state,
 145    &gen6_sf_state,
 146    &gen6_wm_state,
 147
 148    &gen6_scissor_state,
 149    &gen6_scissor_state_pointers,
 150
 151    &brw_state_base_address,
 152
 153    &gen6_binding_table_pointers,
 154
 155    &brw_depthbuffer,
 156
 157    &brw_polygon_stipple,
 158    &brw_polygon_stipple_offset,
 159
 160    &brw_line_stipple,
 161    &brw_aa_line_parameters,
 162
 163    &brw_drawing_rect,
 164
 165    &brw_indices,
 166    &brw_index_buffer,
 167    &brw_vertices,
 168 };
 169
 170 void brw_init_state( struct brw_context *brw )
 171 {
 172    brw_init_caches(brw);
 173 }
 174
 175
 176 void brw_destroy_state( struct brw_context *brw )
 177 {
 178    brw_destroy_caches(brw);
 179    brw_destroy_batch_cache(brw);
 180 }
 181
 182 /***********************************************************************
 183  */
 184
 185 static GLboolean check_state( const struct brw_state_flags *a,
 186                               const struct brw_state_flags *b )
 187 {
 188    return ((a->mesa & b->mesa) ||
 189            (a->brw & b->brw) ||
 190            (a->cache & b->cache));
 191 }
 192
 193 static void accumulate_state( struct brw_state_flags *a,
 194                               const struct brw_state_flags *b )
 195 {
 196    a->mesa |= b->mesa;
 197    a->brw |= b->brw;
 198    a->cache |= b->cache;
 199 }
 200
 201
 202 static void xor_states( struct brw_state_flags *result,
 203                              const struct brw_state_flags *a,
 204                               const struct brw_state_flags *b )
 205 {
 206    result->mesa = a->mesa ^ b->mesa;
 207    result->brw = a->brw ^ b->brw;
 208    result->cache = a->cache ^ b->cache;
 209 }
 210
 211 void
 212 brw_clear_validated_bos(struct brw_context *brw)
 213 {
 214    int i;
 215
 216    /* Clear the last round of validated bos */
 217    for (i = 0; i < brw->state.validated_bo_count; i++) {
 218       drm_intel_bo_unreference(brw->state.validated_bos[i]);
 219       brw->state.validated_bos[i] = NULL;
 220    }
 221    brw->state.validated_bo_count = 0;
 222 }
 223
 224 struct dirty_bit_map {
 225    uint32_t bit;
 226    char *name;
 227    uint32_t count;
 228 };
 229
 230 #define DEFINE_BIT(name) {name, #name, 0}
 231
 232 static struct dirty_bit_map mesa_bits[] = {
 233    DEFINE_BIT(_NEW_MODELVIEW),
 234    DEFINE_BIT(_NEW_PROJECTION),
 235    DEFINE_BIT(_NEW_TEXTURE_MATRIX),
 236    DEFINE_BIT(_NEW_ACCUM),
 237    DEFINE_BIT(_NEW_COLOR),
 238    DEFINE_BIT(_NEW_DEPTH),
 239    DEFINE_BIT(_NEW_EVAL),
 240    DEFINE_BIT(_NEW_FOG),
 241    DEFINE_BIT(_NEW_HINT),
 242    DEFINE_BIT(_NEW_LIGHT),
 243    DEFINE_BIT(_NEW_LINE),
 244    DEFINE_BIT(_NEW_PIXEL),
 245    DEFINE_BIT(_NEW_POINT),
 246    DEFINE_BIT(_NEW_POLYGON),
 247    DEFINE_BIT(_NEW_POLYGONSTIPPLE),
 248    DEFINE_BIT(_NEW_SCISSOR),
 249    DEFINE_BIT(_NEW_STENCIL),
 250    DEFINE_BIT(_NEW_TEXTURE),
 251    DEFINE_BIT(_NEW_TRANSFORM),
 252    DEFINE_BIT(_NEW_VIEWPORT),
 253    DEFINE_BIT(_NEW_PACKUNPACK),
 254    DEFINE_BIT(_NEW_ARRAY),
 255    DEFINE_BIT(_NEW_RENDERMODE),
 256    DEFINE_BIT(_NEW_BUFFERS),
 257    DEFINE_BIT(_NEW_MULTISAMPLE),
 258    DEFINE_BIT(_NEW_TRACK_MATRIX),
 259    DEFINE_BIT(_NEW_PROGRAM),
 260    DEFINE_BIT(_NEW_PROGRAM_CONSTANTS),
 261    {0, 0, 0}
 262 };
 263
 264 static struct dirty_bit_map brw_bits[] = {
 265    DEFINE_BIT(BRW_NEW_URB_FENCE),
 266    DEFINE_BIT(BRW_NEW_FRAGMENT_PROGRAM),
 267    DEFINE_BIT(BRW_NEW_VERTEX_PROGRAM),
 268    DEFINE_BIT(BRW_NEW_INPUT_DIMENSIONS),
 269    DEFINE_BIT(BRW_NEW_CURBE_OFFSETS),
 270    DEFINE_BIT(BRW_NEW_REDUCED_PRIMITIVE),
 271    DEFINE_BIT(BRW_NEW_PRIMITIVE),
 272    DEFINE_BIT(BRW_NEW_CONTEXT),
 273    DEFINE_BIT(BRW_NEW_WM_INPUT_DIMENSIONS),
 274    DEFINE_BIT(BRW_NEW_PSP),
 275    DEFINE_BIT(BRW_NEW_WM_SURFACES),
 276    DEFINE_BIT(BRW_NEW_BINDING_TABLE),
 277    DEFINE_BIT(BRW_NEW_INDICES),
 278    DEFINE_BIT(BRW_NEW_INDEX_BUFFER),
 279    DEFINE_BIT(BRW_NEW_VERTICES),
 280    DEFINE_BIT(BRW_NEW_BATCH),
 281    DEFINE_BIT(BRW_NEW_DEPTH_BUFFER),
 282    {0, 0, 0}
 283 };
 284
 285 static struct dirty_bit_map cache_bits[] = {
 286    DEFINE_BIT(CACHE_NEW_BLEND_STATE),
 287    DEFINE_BIT(CACHE_NEW_CC_VP),
 288    DEFINE_BIT(CACHE_NEW_CC_UNIT),
 289    DEFINE_BIT(CACHE_NEW_WM_PROG),
 290    DEFINE_BIT(CACHE_NEW_SAMPLER_DEFAULT_COLOR),
 291    DEFINE_BIT(CACHE_NEW_SAMPLER),
 292    DEFINE_BIT(CACHE_NEW_WM_UNIT),
 293    DEFINE_BIT(CACHE_NEW_SF_PROG),
 294    DEFINE_BIT(CACHE_NEW_SF_VP),
 295    DEFINE_BIT(CACHE_NEW_SF_UNIT),
 296    DEFINE_BIT(CACHE_NEW_VS_UNIT),
 297    DEFINE_BIT(CACHE_NEW_VS_PROG),
 298    DEFINE_BIT(CACHE_NEW_GS_UNIT),
 299    DEFINE_BIT(CACHE_NEW_GS_PROG),
 300    DEFINE_BIT(CACHE_NEW_CLIP_VP),
 301    DEFINE_BIT(CACHE_NEW_CLIP_UNIT),
 302    DEFINE_BIT(CACHE_NEW_CLIP_PROG),
 303    {0, 0, 0}
 304 };
 305
 306
 307 static void
 308 brw_update_dirty_count(struct dirty_bit_map *bit_map, int32_t bits)
 309 {
 310    int i;
 311
 312    for (i = 0; i < 32; i++) {
 313       if (bit_map[i].bit == 0)
 314          return;
 315
 316       if (bit_map[i].bit & bits)
 317          bit_map[i].count++;
 318    }
 319 }
 320
 321 static void
 322 brw_print_dirty_count(struct dirty_bit_map *bit_map, int32_t bits)
 323 {
 324    int i;
 325
 326    for (i = 0; i < 32; i++) {
 327       if (bit_map[i].bit == 0)
 328          return;
 329
 330       fprintf(stderr, "0x%08x: %12d (%s)\n",
 331               bit_map[i].bit, bit_map[i].count, bit_map[i].name);
 332    }
 333 }
 334
 335 /***********************************************************************
 336  * Emit all state:
 337  */
 338 void brw_validate_state( struct brw_context *brw )
 339 {
 340    struct gl_context *ctx = &brw->intel.ctx;
 341    struct intel_context *intel = &brw->intel;
 342    struct brw_state_flags *state = &brw->state.dirty;
 343    GLuint i;
 344    const struct brw_tracked_state **atoms;
 345    int num_atoms;
 346
 347    brw_clear_validated_bos(brw);
 348
 349    state->mesa |= brw->intel.NewGLState;
 350    brw->intel.NewGLState = 0;
 351
 352    brw_add_validated_bo(brw, intel->batch->buf);
 353
 354    if (intel->gen >= 6) {
 355       atoms = gen6_atoms;
 356       num_atoms = ARRAY_SIZE(gen6_atoms);
 357    } else {
 358       atoms = gen4_atoms;
 359       num_atoms = ARRAY_SIZE(gen4_atoms);
 360    }
 361
 362    if (brw->emit_state_always) {
 363       state->mesa |= ~0;
 364       state->brw |= ~0;
 365       state->cache |= ~0;
 366    }
 367
 368    if (brw->fragment_program != ctx->FragmentProgram._Current) {
 369       brw->fragment_program = ctx->FragmentProgram._Current;
 370       brw->state.dirty.brw |= BRW_NEW_FRAGMENT_PROGRAM;
 371    }
 372
 373    if (brw->vertex_program != ctx->VertexProgram._Current) {
 374       brw->vertex_program = ctx->VertexProgram._Current;
 375       brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM;
 376    }
 377
 378    if (state->mesa == 0 &&
 379        state->cache == 0 &&
 380        state->brw == 0)
 381       return;
 382
 383    if (brw->state.dirty.brw & BRW_NEW_CONTEXT)
 384       brw_clear_batch_cache(brw);
 385
 386    brw->intel.Fallback = GL_FALSE; /* boolean, not bitfield */
 387
 388    /* do prepare stage for all atoms */
 389    for (i = 0; i < num_atoms; i++) {
 390       const struct brw_tracked_state *atom = atoms[i];
 391
 392       if (brw->intel.Fallback)
 393          break;
 394
 395       if (check_state(state, &atom->dirty)) {
 396          if (atom->prepare) {
 397             atom->prepare(brw);
 398         }
 399       }
 400    }
 401
 402    intel_check_front_buffer_rendering(intel);
 403
 404    /* Make sure that the textures which are referenced by the current
 405     * brw fragment program are actually present/valid.
 406     * If this fails, we can experience GPU lock-ups.
 407     */
 408    {
 409       const struct brw_fragment_program *fp;
 410       fp = brw_fragment_program_const(brw->fragment_program);
 411       if (fp) {
 412          assert((fp->tex_units_used & ctx->Texture._EnabledUnits)
 413                 == fp->tex_units_used);
 414       }
 415    }
 416 }
 417
 418
 419 void brw_upload_state(struct brw_context *brw)
 420 {
 421    struct intel_context *intel = &brw->intel;
 422    struct brw_state_flags *state = &brw->state.dirty;
 423    int i;
 424    static int dirty_count = 0;
 425    const struct brw_tracked_state **atoms;
 426    int num_atoms;
 427
 428    if (intel->gen >= 6) {
 429       atoms = gen6_atoms;
 430       num_atoms = ARRAY_SIZE(gen6_atoms);
 431    } else {
 432       atoms = gen4_atoms;
 433       num_atoms = ARRAY_SIZE(gen4_atoms);
 434    }
 435
 436    brw_clear_validated_bos(brw);
 437
 438    if (unlikely(INTEL_DEBUG)) {
 439       /* Debug version which enforces various sanity checks on the
 440        * state flags which are generated and checked to help ensure
 441        * state atoms are ordered correctly in the list.
 442        */
 443       struct brw_state_flags examined, prev;
 444       memset(&examined, 0, sizeof(examined));
 445       prev = *state;
 446
 447       for (i = 0; i < num_atoms; i++) {
 448          const struct brw_tracked_state *atom = atoms[i];
 449          struct brw_state_flags generated;
 450
 451          assert(atom->dirty.mesa ||
 452                 atom->dirty.brw ||
 453                 atom->dirty.cache);
 454
 455          if (brw->intel.Fallback)
 456             break;
 457
 458          if (check_state(state, &atom->dirty)) {
 459             if (atom->emit) {
 460                atom->emit( brw );
 461             }
 462          }
 463
 464          accumulate_state(&examined, &atom->dirty);
 465
 466          /* generated = (prev ^ state)
 467           * if (examined & generated)
 468           *     fail;
 469           */
 470          xor_states(&generated, &prev, state);
 471          assert(!check_state(&examined, &generated));
 472          prev = *state;
 473       }
 474    }
 475    else {
 476       for (i = 0; i < num_atoms; i++) {
 477          const struct brw_tracked_state *atom = atoms[i];
 478
 479          if (brw->intel.Fallback)
 480             break;
 481
 482          if (check_state(state, &atom->dirty)) {
 483             if (atom->emit) {
 484                atom->emit( brw );
 485             }
 486          }
 487       }
 488    }
 489
 490    if (unlikely(INTEL_DEBUG & DEBUG_STATE)) {
 491       brw_update_dirty_count(mesa_bits, state->mesa);
 492       brw_update_dirty_count(brw_bits, state->brw);
 493       brw_update_dirty_count(cache_bits, state->cache);
 494       if (dirty_count++ % 1000 == 0) {
 495          brw_print_dirty_count(mesa_bits, state->mesa);
 496          brw_print_dirty_count(brw_bits, state->brw);
 497          brw_print_dirty_count(cache_bits, state->cache);
 498          fprintf(stderr, "\n");
 499       }
 500    }
 501
 502    if (!brw->intel.Fallback)
 503       memset(state, 0, sizeof(*state));
 504 }