src/mesa/drivers/dri/i965/intel_buffer_objects.c

   1 /*
   2  * Copyright 2003 VMware, Inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sublicense, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the
  14  * next paragraph) shall be included in all copies or substantial portions
  15  * of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  19  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  20  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  21  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  22  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  23  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  */
  25
  26 /**
  27  * @file intel_buffer_objects.c
  28  *
  29  * This provides core GL buffer object functionality.
  30  */
  31
  32 #include "main/imports.h"
  33 #include "main/mtypes.h"
  34 #include "main/macros.h"
  35 #include "main/bufferobj.h"
  36
  37 #include "brw_context.h"
  38 #include "intel_blit.h"
  39 #include "intel_buffer_objects.h"
  40 #include "intel_batchbuffer.h"
  41
  42 static void
  43 mark_buffer_gpu_usage(struct intel_buffer_object *intel_obj,
  44                                uint32_t offset, uint32_t size)
  45 {
  46    intel_obj->gpu_active_start = MIN2(intel_obj->gpu_active_start, offset);
  47    intel_obj->gpu_active_end = MAX2(intel_obj->gpu_active_end, offset + size);
  48 }
  49
  50 static void
  51 mark_buffer_inactive(struct intel_buffer_object *intel_obj)
  52 {
  53    intel_obj->gpu_active_start = ~0;
  54    intel_obj->gpu_active_end = 0;
  55 }
  56
  57 /** Allocates a new brw_bo to store the data for the buffer object. */
  58 static void
  59 alloc_buffer_object(struct brw_context *brw,
  60                     struct intel_buffer_object *intel_obj)
  61 {
  62    intel_obj->buffer = brw_bo_alloc(brw->bufmgr, "bufferobj",
  63                                           intel_obj->Base.Size, 64);
  64
  65    /* the buffer might be bound as a uniform buffer, need to update it
  66     */
  67    if (intel_obj->Base.UsageHistory & USAGE_UNIFORM_BUFFER)
  68       brw->ctx.NewDriverState |= BRW_NEW_UNIFORM_BUFFER;
  69    if (intel_obj->Base.UsageHistory & USAGE_SHADER_STORAGE_BUFFER)
  70       brw->ctx.NewDriverState |= BRW_NEW_UNIFORM_BUFFER;
  71    if (intel_obj->Base.UsageHistory & USAGE_TEXTURE_BUFFER)
  72       brw->ctx.NewDriverState |= BRW_NEW_TEXTURE_BUFFER;
  73    if (intel_obj->Base.UsageHistory & USAGE_ATOMIC_COUNTER_BUFFER)
  74       brw->ctx.NewDriverState |= BRW_NEW_ATOMIC_BUFFER;
  75
  76    mark_buffer_inactive(intel_obj);
  77 }
  78
  79 static void
  80 release_buffer(struct intel_buffer_object *intel_obj)
  81 {
  82    brw_bo_unreference(intel_obj->buffer);
  83    intel_obj->buffer = NULL;
  84 }
  85
  86 /**
  87  * The NewBufferObject() driver hook.
  88  *
  89  * Allocates a new intel_buffer_object structure and initializes it.
  90  *
  91  * There is some duplication between mesa's bufferobjects and our
  92  * bufmgr buffers.  Both have an integer handle and a hashtable to
  93  * lookup an opaque structure.  It would be nice if the handles and
  94  * internal structure where somehow shared.
  95  */
  96 static struct gl_buffer_object *
  97 brw_new_buffer_object(struct gl_context * ctx, GLuint name)
  98 {
  99    struct intel_buffer_object *obj = CALLOC_STRUCT(intel_buffer_object);
 100    if (!obj) {
 101       _mesa_error_no_memory(__func__);
 102    }
 103
 104    _mesa_initialize_buffer_object(ctx, &obj->Base, name);
 105
 106    obj->buffer = NULL;
 107
 108    return &obj->Base;
 109 }
 110
 111 /**
 112  * The DeleteBuffer() driver hook.
 113  *
 114  * Deletes a single OpenGL buffer object.  Used by glDeleteBuffers().
 115  */
 116 static void
 117 brw_delete_buffer(struct gl_context * ctx, struct gl_buffer_object *obj)
 118 {
 119    struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
 120
 121    assert(intel_obj);
 122
 123    /* Buffer objects are automatically unmapped when deleting according
 124     * to the spec, but Mesa doesn't do UnmapBuffer for us at context destroy
 125     * (though it does if you call glDeleteBuffers)
 126     */
 127    _mesa_buffer_unmap_all_mappings(ctx, obj);
 128
 129    brw_bo_unreference(intel_obj->buffer);
 130    _mesa_delete_buffer_object(ctx, obj);
 131 }
 132
 133
 134 /**
 135  * The BufferData() driver hook.
 136  *
 137  * Implements glBufferData(), which recreates a buffer object's data store
 138  * and populates it with the given data, if present.
 139  *
 140  * Any data that was previously stored in the buffer object is lost.
 141  *
 142  * \return true for success, false if out of memory
 143  */
 144 static GLboolean
 145 brw_buffer_data(struct gl_context *ctx,
 146                 GLenum target,
 147                 GLsizeiptrARB size,
 148                 const GLvoid *data,
 149                 GLenum usage,
 150                 GLbitfield storageFlags,
 151                 struct gl_buffer_object *obj)
 152 {
 153    struct brw_context *brw = brw_context(ctx);
 154    struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
 155
 156    /* Part of the ABI, but this function doesn't use it.
 157     */
 158    (void) target;
 159
 160    intel_obj->Base.Size = size;
 161    intel_obj->Base.Usage = usage;
 162    intel_obj->Base.StorageFlags = storageFlags;
 163
 164    assert(!obj->Mappings[MAP_USER].Pointer); /* Mesa should have unmapped it */
 165    assert(!obj->Mappings[MAP_INTERNAL].Pointer);
 166
 167    if (intel_obj->buffer != NULL)
 168       release_buffer(intel_obj);
 169
 170    if (size != 0) {
 171       alloc_buffer_object(brw, intel_obj);
 172       if (!intel_obj->buffer)
 173          return false;
 174
 175       if (data != NULL)
 176          brw_bo_subdata(intel_obj->buffer, 0, size, data);
 177    }
 178
 179    return true;
 180 }
 181
 182
 183 /**
 184  * The BufferSubData() driver hook.
 185  *
 186  * Implements glBufferSubData(), which replaces a portion of the data in a
 187  * buffer object.
 188  *
 189  * If the data range specified by (size + offset) extends beyond the end of
 190  * the buffer or if data is NULL, no copy is performed.
 191  */
 192 static void
 193 brw_buffer_subdata(struct gl_context *ctx,
 194                    GLintptrARB offset,
 195                    GLsizeiptrARB size,
 196                    const GLvoid *data,
 197                    struct gl_buffer_object *obj)
 198 {
 199    struct brw_context *brw = brw_context(ctx);
 200    struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
 201    bool busy;
 202
 203    if (size == 0)
 204       return;
 205
 206    assert(intel_obj);
 207
 208    /* See if we can unsynchronized write the data into the user's BO. This
 209     * avoids GPU stalls in unfortunately common user patterns (uploading
 210     * sequentially into a BO, with draw calls in between each upload).
 211     *
 212     * Once we've hit this path, we mark this GL BO as preferring stalling to
 213     * blits, so that we can hopefully hit this path again in the future
 214     * (otherwise, an app that might occasionally stall but mostly not will end
 215     * up with blitting all the time, at the cost of bandwidth)
 216     */
 217    if (offset + size <= intel_obj->gpu_active_start ||
 218        intel_obj->gpu_active_end <= offset) {
 219       if (brw->has_llc) {
 220          brw_bo_map_unsynchronized(intel_obj->buffer);
 221          memcpy(intel_obj->buffer->virtual + offset, data, size);
 222          brw_bo_unmap(intel_obj->buffer);
 223
 224          if (intel_obj->gpu_active_end > intel_obj->gpu_active_start)
 225             intel_obj->prefer_stall_to_blit = true;
 226          return;
 227       } else {
 228          perf_debug("BufferSubData could be unsynchronized, but !LLC doesn't support it yet\n");
 229       }
 230    }
 231
 232    busy =
 233       brw_bo_busy(intel_obj->buffer) ||
 234       brw_batch_references(&brw->batch, intel_obj->buffer);
 235
 236    if (busy) {
 237       if (size == intel_obj->Base.Size) {
 238          /* Replace the current busy bo so the subdata doesn't stall. */
 239          brw_bo_unreference(intel_obj->buffer);
 240          alloc_buffer_object(brw, intel_obj);
 241       } else if (!intel_obj->prefer_stall_to_blit) {
 242          perf_debug("Using a blit copy to avoid stalling on "
 243                     "glBufferSubData(%ld, %ld) (%ldkb) to a busy "
 244                     "(%d-%d) buffer object.\n",
 245                     (long)offset, (long)offset + size, (long)(size/1024),
 246                     intel_obj->gpu_active_start,
 247                     intel_obj->gpu_active_end);
 248          struct brw_bo *temp_bo =
 249             brw_bo_alloc(brw->bufmgr, "subdata temp", size, 64);
 250
 251          brw_bo_subdata(temp_bo, 0, size, data);
 252
 253          intel_emit_linear_blit(brw,
 254                                 intel_obj->buffer, offset,
 255                                 temp_bo, 0,
 256                                 size);
 257
 258          brw_bo_unreference(temp_bo);
 259          return;
 260       } else {
 261          perf_debug("Stalling on glBufferSubData(%ld, %ld) (%ldkb) to a busy "
 262                     "(%d-%d) buffer object.  Use glMapBufferRange() to "
 263                     "avoid this.\n",
 264                     (long)offset, (long)offset + size, (long)(size/1024),
 265                     intel_obj->gpu_active_start,
 266                     intel_obj->gpu_active_end);
 267          intel_batchbuffer_flush(brw);
 268       }
 269    }
 270
 271    brw_bo_subdata(intel_obj->buffer, offset, size, data);
 272    mark_buffer_inactive(intel_obj);
 273 }
 274
 275
 276 /**
 277  * The GetBufferSubData() driver hook.
 278  *
 279  * Implements glGetBufferSubData(), which copies a subrange of a buffer
 280  * object into user memory.
 281  */
 282 static void
 283 brw_get_buffer_subdata(struct gl_context *ctx,
 284                        GLintptrARB offset,
 285                        GLsizeiptrARB size,
 286                        GLvoid *data,
 287                        struct gl_buffer_object *obj)
 288 {
 289    struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
 290    struct brw_context *brw = brw_context(ctx);
 291
 292    assert(intel_obj);
 293    if (brw_batch_references(&brw->batch, intel_obj->buffer)) {
 294       intel_batchbuffer_flush(brw);
 295    }
 296    brw_bo_get_subdata(intel_obj->buffer, offset, size, data);
 297
 298    mark_buffer_inactive(intel_obj);
 299 }
 300
 301
 302 /**
 303  * The MapBufferRange() driver hook.
 304  *
 305  * This implements both glMapBufferRange() and glMapBuffer().
 306  *
 307  * The goal of this extension is to allow apps to accumulate their rendering
 308  * at the same time as they accumulate their buffer object.  Without it,
 309  * you'd end up blocking on execution of rendering every time you mapped
 310  * the buffer to put new data in.
 311  *
 312  * We support it in 3 ways: If unsynchronized, then don't bother
 313  * flushing the batchbuffer before mapping the buffer, which can save blocking
 314  * in many cases.  If we would still block, and they allow the whole buffer
 315  * to be invalidated, then just allocate a new buffer to replace the old one.
 316  * If not, and we'd block, and they allow the subrange of the buffer to be
 317  * invalidated, then we can make a new little BO, let them write into that,
 318  * and blit it into the real BO at unmap time.
 319  */
 320 static void *
 321 brw_map_buffer_range(struct gl_context *ctx,
 322                      GLintptr offset, GLsizeiptr length,
 323                      GLbitfield access, struct gl_buffer_object *obj,
 324                      gl_map_buffer_index index)
 325 {
 326    struct brw_context *brw = brw_context(ctx);
 327    struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
 328
 329    assert(intel_obj);
 330
 331    /* _mesa_MapBufferRange (GL entrypoint) sets these, but the vbo module also
 332     * internally uses our functions directly.
 333     */
 334    obj->Mappings[index].Offset = offset;
 335    obj->Mappings[index].Length = length;
 336    obj->Mappings[index].AccessFlags = access;
 337
 338    if (intel_obj->buffer == NULL) {
 339       obj->Mappings[index].Pointer = NULL;
 340       return NULL;
 341    }
 342
 343    /* If the access is synchronized (like a normal buffer mapping), then get
 344     * things flushed out so the later mapping syncs appropriately through GEM.
 345     * If the user doesn't care about existing buffer contents and mapping would
 346     * cause us to block, then throw out the old buffer.
 347     *
 348     * If they set INVALIDATE_BUFFER, we can pitch the current contents to
 349     * achieve the required synchronization.
 350     */
 351    if (!(access & GL_MAP_UNSYNCHRONIZED_BIT)) {
 352       if (brw_batch_references(&brw->batch, intel_obj->buffer)) {
 353          if (access & GL_MAP_INVALIDATE_BUFFER_BIT) {
 354             brw_bo_unreference(intel_obj->buffer);
 355             alloc_buffer_object(brw, intel_obj);
 356          } else {
 357             perf_debug("Stalling on the GPU for mapping a busy buffer "
 358                        "object\n");
 359             intel_batchbuffer_flush(brw);
 360          }
 361       } else if (brw_bo_busy(intel_obj->buffer) &&
 362                  (access & GL_MAP_INVALIDATE_BUFFER_BIT)) {
 363          brw_bo_unreference(intel_obj->buffer);
 364          alloc_buffer_object(brw, intel_obj);
 365       }
 366    }
 367
 368    /* If the user is mapping a range of an active buffer object but
 369     * doesn't require the current contents of that range, make a new
 370     * BO, and we'll copy what they put in there out at unmap or
 371     * FlushRange time.
 372     *
 373     * That is, unless they're looking for a persistent mapping -- we would
 374     * need to do blits in the MemoryBarrier call, and it's easier to just do a
 375     * GPU stall and do a mapping.
 376     */
 377    if (!(access & (GL_MAP_UNSYNCHRONIZED_BIT | GL_MAP_PERSISTENT_BIT)) &&
 378        (access & GL_MAP_INVALIDATE_RANGE_BIT) &&
 379        brw_bo_busy(intel_obj->buffer)) {
 380       /* Ensure that the base alignment of the allocation meets the alignment
 381        * guarantees the driver has advertised to the application.
 382        */
 383       const unsigned alignment = ctx->Const.MinMapBufferAlignment;
 384
 385       intel_obj->map_extra[index] = (uintptr_t) offset % alignment;
 386       intel_obj->range_map_bo[index] = brw_bo_alloc(brw->bufmgr,
 387                                                           "BO blit temp",
 388                                                           length +
 389                                                           intel_obj->map_extra[index],
 390                                                           alignment);
 391       if (brw->has_llc) {
 392          brw_bo_map(intel_obj->range_map_bo[index],
 393                           (access & GL_MAP_WRITE_BIT) != 0);
 394       } else {
 395          brw_bo_map_gtt(intel_obj->range_map_bo[index]);
 396       }
 397       obj->Mappings[index].Pointer =
 398          intel_obj->range_map_bo[index]->virtual + intel_obj->map_extra[index];
 399       return obj->Mappings[index].Pointer;
 400    }
 401
 402    if (access & GL_MAP_UNSYNCHRONIZED_BIT) {
 403       if (!brw->has_llc && brw->perf_debug &&
 404           brw_bo_busy(intel_obj->buffer)) {
 405          perf_debug("MapBufferRange with GL_MAP_UNSYNCHRONIZED_BIT stalling (it's actually synchronized on non-LLC platforms)\n");
 406       }
 407       brw_bo_map_unsynchronized(intel_obj->buffer);
 408    } else if (!brw->has_llc && (!(access & GL_MAP_READ_BIT) ||
 409                               (access & GL_MAP_PERSISTENT_BIT))) {
 410       brw_bo_map_gtt(intel_obj->buffer);
 411       mark_buffer_inactive(intel_obj);
 412    } else {
 413       brw_bo_map(intel_obj->buffer, (access & GL_MAP_WRITE_BIT) != 0);
 414       mark_buffer_inactive(intel_obj);
 415    }
 416
 417    obj->Mappings[index].Pointer = intel_obj->buffer->virtual + offset;
 418    return obj->Mappings[index].Pointer;
 419 }
 420
 421 /**
 422  * The FlushMappedBufferRange() driver hook.
 423  *
 424  * Implements glFlushMappedBufferRange(), which signifies that modifications
 425  * have been made to a range of a mapped buffer, and it should be flushed.
 426  *
 427  * This is only used for buffers mapped with GL_MAP_FLUSH_EXPLICIT_BIT.
 428  *
 429  * Ideally we'd use a BO to avoid taking up cache space for the temporary
 430  * data, but FlushMappedBufferRange may be followed by further writes to
 431  * the pointer, so we would have to re-map after emitting our blit, which
 432  * would defeat the point.
 433  */
 434 static void
 435 brw_flush_mapped_buffer_range(struct gl_context *ctx,
 436                               GLintptr offset, GLsizeiptr length,
 437                               struct gl_buffer_object *obj,
 438                               gl_map_buffer_index index)
 439 {
 440    struct brw_context *brw = brw_context(ctx);
 441    struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
 442
 443    assert(obj->Mappings[index].AccessFlags & GL_MAP_FLUSH_EXPLICIT_BIT);
 444
 445    /* If we gave a direct mapping of the buffer instead of using a temporary,
 446     * then there's nothing to do.
 447     */
 448    if (intel_obj->range_map_bo[index] == NULL)
 449       return;
 450
 451    if (length == 0)
 452       return;
 453
 454    /* Note that we're not unmapping our buffer while executing the blit.  We
 455     * need to have a mapping still at the end of this call, since the user
 456     * gets to make further modifications and glFlushMappedBufferRange() calls.
 457     * This is safe, because:
 458     *
 459     * - On LLC platforms, we're using a CPU mapping that's coherent with the
 460     *   GPU (except for the render caches), so the kernel doesn't need to do
 461     *   any flushing work for us except for what happens at batch exec time
 462     *   anyway.
 463     *
 464     * - On non-LLC platforms, we're using a GTT mapping that writes directly
 465     *   to system memory (except for the chipset cache that gets flushed at
 466     *   batch exec time).
 467     *
 468     * In both cases we don't need to stall for the previous blit to complete
 469     * so we can re-map (and we definitely don't want to, since that would be
 470     * slow): If the user edits a part of their buffer that's previously been
 471     * blitted, then our lack of synchoronization is fine, because either
 472     * they'll get some too-new data in the first blit and not do another blit
 473     * of that area (but in that case the results are undefined), or they'll do
 474     * another blit of that area and the complete newer data will land the
 475     * second time.
 476     */
 477    intel_emit_linear_blit(brw,
 478                           intel_obj->buffer,
 479                           obj->Mappings[index].Offset + offset,
 480                           intel_obj->range_map_bo[index],
 481                           intel_obj->map_extra[index] + offset,
 482                           length);
 483    mark_buffer_gpu_usage(intel_obj,
 484                          obj->Mappings[index].Offset + offset,
 485                          length);
 486 }
 487
 488
 489 /**
 490  * The UnmapBuffer() driver hook.
 491  *
 492  * Implements glUnmapBuffer().
 493  */
 494 static GLboolean
 495 brw_unmap_buffer(struct gl_context *ctx,
 496                  struct gl_buffer_object *obj,
 497                  gl_map_buffer_index index)
 498 {
 499    struct brw_context *brw = brw_context(ctx);
 500    struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
 501
 502    assert(intel_obj);
 503    assert(obj->Mappings[index].Pointer);
 504    if (intel_obj->range_map_bo[index] != NULL) {
 505       brw_bo_unmap(intel_obj->range_map_bo[index]);
 506
 507       if (!(obj->Mappings[index].AccessFlags & GL_MAP_FLUSH_EXPLICIT_BIT)) {
 508          intel_emit_linear_blit(brw,
 509                                 intel_obj->buffer, obj->Mappings[index].Offset,
 510                                 intel_obj->range_map_bo[index],
 511                                 intel_obj->map_extra[index],
 512                                 obj->Mappings[index].Length);
 513          mark_buffer_gpu_usage(intel_obj, obj->Mappings[index].Offset,
 514                                obj->Mappings[index].Length);
 515       }
 516
 517       /* Since we've emitted some blits to buffers that will (likely) be used
 518        * in rendering operations in other cache domains in this batch, emit a
 519        * flush.  Once again, we wish for a domain tracker in libdrm to cover
 520        * usage inside of a batchbuffer.
 521        */
 522       brw_emit_mi_flush(brw);
 523
 524       brw_bo_unreference(intel_obj->range_map_bo[index]);
 525       intel_obj->range_map_bo[index] = NULL;
 526    } else if (intel_obj->buffer != NULL) {
 527       brw_bo_unmap(intel_obj->buffer);
 528    }
 529    obj->Mappings[index].Pointer = NULL;
 530    obj->Mappings[index].Offset = 0;
 531    obj->Mappings[index].Length = 0;
 532
 533    return true;
 534 }
 535
 536 /**
 537  * Gets a pointer to the object's BO, and marks the given range as being used
 538  * on the GPU.
 539  *
 540  * Anywhere that uses buffer objects in the pipeline should be using this to
 541  * mark the range of the buffer that is being accessed by the pipeline.
 542  */
 543 struct brw_bo *
 544 intel_bufferobj_buffer(struct brw_context *brw,
 545                        struct intel_buffer_object *intel_obj,
 546                        uint32_t offset, uint32_t size)
 547 {
 548    /* This is needed so that things like transform feedback and texture buffer
 549     * objects that need a BO but don't want to check that they exist for
 550     * draw-time validation can just always get a BO from a GL buffer object.
 551     */
 552    if (intel_obj->buffer == NULL)
 553       alloc_buffer_object(brw, intel_obj);
 554
 555    mark_buffer_gpu_usage(intel_obj, offset, size);
 556
 557    return intel_obj->buffer;
 558 }
 559
 560 /**
 561  * The CopyBufferSubData() driver hook.
 562  *
 563  * Implements glCopyBufferSubData(), which copies a portion of one buffer
 564  * object's data to another.  Independent source and destination offsets
 565  * are allowed.
 566  */
 567 static void
 568 brw_copy_buffer_subdata(struct gl_context *ctx,
 569                         struct gl_buffer_object *src,
 570                         struct gl_buffer_object *dst,
 571                         GLintptr read_offset, GLintptr write_offset,
 572                         GLsizeiptr size)
 573 {
 574    struct brw_context *brw = brw_context(ctx);
 575    struct intel_buffer_object *intel_src = intel_buffer_object(src);
 576    struct intel_buffer_object *intel_dst = intel_buffer_object(dst);
 577    struct brw_bo *src_bo, *dst_bo;
 578
 579    if (size == 0)
 580       return;
 581
 582    dst_bo = intel_bufferobj_buffer(brw, intel_dst, write_offset, size);
 583    src_bo = intel_bufferobj_buffer(brw, intel_src, read_offset, size);
 584
 585    intel_emit_linear_blit(brw,
 586                           dst_bo, write_offset,
 587                           src_bo, read_offset, size);
 588
 589    /* Since we've emitted some blits to buffers that will (likely) be used
 590     * in rendering operations in other cache domains in this batch, emit a
 591     * flush.  Once again, we wish for a domain tracker in libdrm to cover
 592     * usage inside of a batchbuffer.
 593     */
 594    brw_emit_mi_flush(brw);
 595 }
 596
 597 void
 598 intelInitBufferObjectFuncs(struct dd_function_table *functions)
 599 {
 600    functions->NewBufferObject = brw_new_buffer_object;
 601    functions->DeleteBuffer = brw_delete_buffer;
 602    functions->BufferData = brw_buffer_data;
 603    functions->BufferSubData = brw_buffer_subdata;
 604    functions->GetBufferSubData = brw_get_buffer_subdata;
 605    functions->MapBufferRange = brw_map_buffer_range;
 606    functions->FlushMappedBufferRange = brw_flush_mapped_buffer_range;
 607    functions->UnmapBuffer = brw_unmap_buffer;
 608    functions->CopyBufferSubData = brw_copy_buffer_subdata;
 609 }