src/mesa/drivers/dri/intel/intel_regions.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * 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, sub license, 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 portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 /* Provide additional functionality on top of bufmgr buffers:
  29  *   - 2d semantics and blit operations
  30  *   - refcounting of buffers for multiple images in a buffer.
  31  *   - refcounting of buffer mappings.
  32  *   - some logic for moving the buffers to the best memory pools for
  33  *     given operations.
  34  *
  35  * Most of this is to make it easier to implement the fixed-layout
  36  * mipmap tree required by intel hardware in the face of GL's
  37  * programming interface where each image can be specifed in random
  38  * order and it isn't clear what layout the tree should have until the
  39  * last moment.
  40  */
  41
  42 #include <sys/ioctl.h>
  43 #include <errno.h>
  44
  45 #include "main/hash.h"
  46 #include "intel_context.h"
  47 #include "intel_regions.h"
  48 #include "intel_blit.h"
  49 #include "intel_buffer_objects.h"
  50 #include "intel_bufmgr.h"
  51 #include "intel_batchbuffer.h"
  52
  53 #define FILE_DEBUG_FLAG DEBUG_REGION
  54
  55 /* This should be set to the maximum backtrace size desired.
  56  * Set it to 0 to disable backtrace debugging.
  57  */
  58 #define DEBUG_BACKTRACE_SIZE 0
  59
  60 #if DEBUG_BACKTRACE_SIZE == 0
  61 /* Use the standard debug output */
  62 #define _DBG(...) DBG(__VA_ARGS__)
  63 #else
  64 /* Use backtracing debug output */
  65 #define _DBG(...) {debug_backtrace(); DBG(__VA_ARGS__);}
  66
  67 /* Backtracing debug support */
  68 #include <execinfo.h>
  69
  70 static void
  71 debug_backtrace(void)
  72 {
  73    void *trace[DEBUG_BACKTRACE_SIZE];
  74    char **strings = NULL;
  75    int traceSize;
  76    register int i;
  77
  78    traceSize = backtrace(trace, DEBUG_BACKTRACE_SIZE);
  79    strings = backtrace_symbols(trace, traceSize);
  80    if (strings == NULL) {
  81       DBG("no backtrace:");
  82       return;
  83    }
  84
  85    /* Spit out all the strings with a colon separator.  Ignore
  86     * the first, since we don't really care about the call
  87     * to debug_backtrace() itself.  Skip until the final "/" in
  88     * the trace to avoid really long lines.
  89     */
  90    for (i = 1; i < traceSize; i++) {
  91       char *p = strings[i], *slash = strings[i];
  92       while (*p) {
  93          if (*p++ == '/') {
  94             slash = p;
  95          }
  96       }
  97
  98       DBG("%s:", slash);
  99    }
 100
 101    /* Free up the memory, and we're done */
 102    free(strings);
 103 }
 104
 105 #endif
 106
 107
 108
 109 /* XXX: Thread safety?
 110  */
 111 GLubyte *
 112 intel_region_map(struct intel_context *intel, struct intel_region *region)
 113 {
 114    intel_flush(&intel->ctx);
 115
 116    _DBG("%s %p\n", __FUNCTION__, region);
 117    if (!region->map_refcount++) {
 118       if (region->pbo)
 119          intel_region_cow(intel, region);
 120
 121       if (region->tiling != I915_TILING_NONE)
 122          drm_intel_gem_bo_map_gtt(region->buffer);
 123       else
 124          drm_intel_bo_map(region->buffer, GL_TRUE);
 125       region->map = region->buffer->virtual;
 126    }
 127
 128    return region->map;
 129 }
 130
 131 void
 132 intel_region_unmap(struct intel_context *intel, struct intel_region *region)
 133 {
 134    _DBG("%s %p\n", __FUNCTION__, region);
 135    if (!--region->map_refcount) {
 136       if (region->tiling != I915_TILING_NONE)
 137          drm_intel_gem_bo_unmap_gtt(region->buffer);
 138       else
 139          drm_intel_bo_unmap(region->buffer);
 140       region->map = NULL;
 141    }
 142 }
 143
 144 static struct intel_region *
 145 intel_region_alloc_internal(struct intel_screen *screen,
 146                             GLuint cpp,
 147                             GLuint width, GLuint height, GLuint pitch,
 148                             uint32_t tiling, drm_intel_bo *buffer)
 149 {
 150    struct intel_region *region;
 151
 152    if (buffer == NULL) {
 153       _DBG("%s <-- NULL\n", __FUNCTION__);
 154       return NULL;
 155    }
 156
 157    region = calloc(sizeof(*region), 1);
 158    if (region == NULL)
 159       return region;
 160
 161    region->cpp = cpp;
 162    region->width = width;
 163    region->height = height;
 164    region->pitch = pitch;
 165    region->refcount = 1;
 166    region->buffer = buffer;
 167    region->tiling = tiling;
 168    region->screen = screen;
 169
 170    _DBG("%s <-- %p\n", __FUNCTION__, region);
 171    return region;
 172 }
 173
 174 struct intel_region *
 175 intel_region_alloc(struct intel_screen *screen,
 176                    uint32_t tiling,
 177                    GLuint cpp, GLuint width, GLuint height,
 178                    GLboolean expect_accelerated_upload)
 179 {
 180    drm_intel_bo *buffer;
 181    unsigned long flags = 0;
 182    unsigned long aligned_pitch;
 183
 184    if (expect_accelerated_upload)
 185       flags |= BO_ALLOC_FOR_RENDER;
 186
 187    buffer = drm_intel_bo_alloc_tiled(screen->bufmgr, "region",
 188                                      width, height, cpp,
 189                                      &tiling, &aligned_pitch, flags);
 190
 191    return intel_region_alloc_internal(screen, cpp, width, height,
 192                                       aligned_pitch / cpp, tiling, buffer);
 193 }
 194
 195 GLboolean
 196 intel_region_flink(struct intel_region *region, uint32_t *name)
 197 {
 198    if (region->name == 0) {
 199       if (drm_intel_bo_flink(region->buffer, &region->name))
 200          return GL_FALSE;
 201
 202       _mesa_HashInsert(region->screen->named_regions,
 203                        region->name, region);
 204    }
 205
 206    *name = region->name;
 207
 208    return GL_TRUE;
 209 }
 210
 211 struct intel_region *
 212 intel_region_alloc_for_handle(struct intel_screen *screen,
 213                               GLuint cpp,
 214                               GLuint width, GLuint height, GLuint pitch,
 215                               GLuint handle, const char *name)
 216 {
 217    struct intel_region *region, *dummy;
 218    drm_intel_bo *buffer;
 219    int ret;
 220    uint32_t bit_6_swizzle, tiling;
 221
 222    region = _mesa_HashLookup(screen->named_regions, handle);
 223    if (region != NULL) {
 224       dummy = NULL;
 225       if (region->width != width || region->height != height ||
 226           region->cpp != cpp || region->pitch != pitch) {
 227          fprintf(stderr,
 228                  "Region for name %d already exists but is not compatible\n",
 229                  handle);
 230          return NULL;
 231       }
 232       intel_region_reference(&dummy, region);
 233       return dummy;
 234    }
 235
 236    buffer = intel_bo_gem_create_from_name(screen->bufmgr, name, handle);
 237    if (buffer == NULL)
 238       return NULL;
 239    ret = drm_intel_bo_get_tiling(buffer, &tiling, &bit_6_swizzle);
 240    if (ret != 0) {
 241       fprintf(stderr, "Couldn't get tiling of buffer %d (%s): %s\n",
 242               handle, name, strerror(-ret));
 243       drm_intel_bo_unreference(buffer);
 244       return NULL;
 245    }
 246
 247    region = intel_region_alloc_internal(screen, cpp,
 248                                         width, height, pitch, tiling, buffer);
 249    if (region == NULL) {
 250       drm_intel_bo_unreference(buffer);
 251       return NULL;
 252    }
 253
 254    region->name = handle;
 255    _mesa_HashInsert(screen->named_regions, handle, region);
 256
 257    return region;
 258 }
 259
 260 void
 261 intel_region_reference(struct intel_region **dst, struct intel_region *src)
 262 {
 263    if (src)
 264       _DBG("%s %p %d\n", __FUNCTION__, src, src->refcount);
 265
 266    assert(*dst == NULL);
 267    if (src) {
 268       src->refcount++;
 269       *dst = src;
 270    }
 271 }
 272
 273 void
 274 intel_region_release(struct intel_region **region_handle)
 275 {
 276    struct intel_region *region = *region_handle;
 277
 278    if (region == NULL) {
 279       _DBG("%s NULL\n", __FUNCTION__);
 280       return;
 281    }
 282
 283    _DBG("%s %p %d\n", __FUNCTION__, region, region->refcount - 1);
 284
 285    ASSERT(region->refcount > 0);
 286    region->refcount--;
 287
 288    if (region->refcount == 0) {
 289       assert(region->map_refcount == 0);
 290
 291       if (region->pbo)
 292          region->pbo->region = NULL;
 293       region->pbo = NULL;
 294       drm_intel_bo_unreference(region->buffer);
 295
 296       if (region->name > 0)
 297          _mesa_HashRemove(region->screen->named_regions, region->name);
 298
 299       free(region);
 300    }
 301    *region_handle = NULL;
 302 }
 303
 304 /*
 305  * XXX Move this into core Mesa?
 306  */
 307 void
 308 _mesa_copy_rect(GLubyte * dst,
 309                 GLuint cpp,
 310                 GLuint dst_pitch,
 311                 GLuint dst_x,
 312                 GLuint dst_y,
 313                 GLuint width,
 314                 GLuint height,
 315                 const GLubyte * src,
 316                 GLuint src_pitch, GLuint src_x, GLuint src_y)
 317 {
 318    GLuint i;
 319
 320    dst_pitch *= cpp;
 321    src_pitch *= cpp;
 322    dst += dst_x * cpp;
 323    src += src_x * cpp;
 324    dst += dst_y * dst_pitch;
 325    src += src_y * src_pitch;
 326    width *= cpp;
 327
 328    if (width == dst_pitch && width == src_pitch)
 329       memcpy(dst, src, height * width);
 330    else {
 331       for (i = 0; i < height; i++) {
 332          memcpy(dst, src, width);
 333          dst += dst_pitch;
 334          src += src_pitch;
 335       }
 336    }
 337 }
 338
 339
 340 /* Upload data to a rectangular sub-region.  Lots of choices how to do this:
 341  *
 342  * - memcpy by span to current destination
 343  * - upload data as new buffer and blit
 344  *
 345  * Currently always memcpy.
 346  */
 347 void
 348 intel_region_data(struct intel_context *intel,
 349                   struct intel_region *dst,
 350                   GLuint dst_offset,
 351                   GLuint dstx, GLuint dsty,
 352                   const void *src, GLuint src_pitch,
 353                   GLuint srcx, GLuint srcy, GLuint width, GLuint height)
 354 {
 355    _DBG("%s\n", __FUNCTION__);
 356
 357    if (intel == NULL)
 358       return;
 359
 360    if (dst->pbo) {
 361       if (dstx == 0 &&
 362           dsty == 0 && width == dst->pitch && height == dst->height)
 363          intel_region_release_pbo(intel, dst);
 364       else
 365          intel_region_cow(intel, dst);
 366    }
 367
 368    intel_prepare_render(intel);
 369
 370    _mesa_copy_rect(intel_region_map(intel, dst) + dst_offset,
 371                    dst->cpp,
 372                    dst->pitch,
 373                    dstx, dsty, width, height, src, src_pitch, srcx, srcy);
 374
 375    intel_region_unmap(intel, dst);
 376 }
 377
 378 /* Copy rectangular sub-regions. Need better logic about when to
 379  * push buffers into AGP - will currently do so whenever possible.
 380  */
 381 GLboolean
 382 intel_region_copy(struct intel_context *intel,
 383                   struct intel_region *dst,
 384                   GLuint dst_offset,
 385                   GLuint dstx, GLuint dsty,
 386                   struct intel_region *src,
 387                   GLuint src_offset,
 388                   GLuint srcx, GLuint srcy, GLuint width, GLuint height,
 389                   GLboolean flip,
 390                   GLenum logicop)
 391 {
 392    uint32_t src_pitch = src->pitch;
 393
 394    _DBG("%s\n", __FUNCTION__);
 395
 396    if (intel == NULL)
 397       return GL_FALSE;
 398
 399    if (dst->pbo) {
 400       if (dstx == 0 &&
 401           dsty == 0 && width == dst->pitch && height == dst->height)
 402          intel_region_release_pbo(intel, dst);
 403       else
 404          intel_region_cow(intel, dst);
 405    }
 406
 407    assert(src->cpp == dst->cpp);
 408
 409    if (flip)
 410       src_pitch = -src_pitch;
 411
 412    return intelEmitCopyBlit(intel,
 413                             dst->cpp,
 414                             src_pitch, src->buffer, src_offset, src->tiling,
 415                             dst->pitch, dst->buffer, dst_offset, dst->tiling,
 416                             srcx, srcy, dstx, dsty, width, height,
 417                             logicop);
 418 }
 419
 420 /* Attach to a pbo, discarding our data.  Effectively zero-copy upload
 421  * the pbo's data.
 422  */
 423 void
 424 intel_region_attach_pbo(struct intel_context *intel,
 425                         struct intel_region *region,
 426                         struct intel_buffer_object *pbo)
 427 {
 428    drm_intel_bo *buffer;
 429
 430    if (region->pbo == pbo)
 431       return;
 432
 433    _DBG("%s %p %p\n", __FUNCTION__, region, pbo);
 434
 435    /* If there is already a pbo attached, break the cow tie now.
 436     * Don't call intel_region_release_pbo() as that would
 437     * unnecessarily allocate a new buffer we would have to immediately
 438     * discard.
 439     */
 440    if (region->pbo) {
 441       region->pbo->region = NULL;
 442       region->pbo = NULL;
 443    }
 444
 445    if (region->buffer) {
 446       drm_intel_bo_unreference(region->buffer);
 447       region->buffer = NULL;
 448    }
 449
 450    /* make sure pbo has a buffer of its own */
 451    buffer = intel_bufferobj_buffer(intel, pbo, INTEL_WRITE_FULL);
 452
 453    region->pbo = pbo;
 454    region->pbo->region = region;
 455    drm_intel_bo_reference(buffer);
 456    region->buffer = buffer;
 457    region->tiling = I915_TILING_NONE;
 458 }
 459
 460
 461 /* Break the COW tie to the pbo and allocate a new buffer.
 462  * The pbo gets to keep the data.
 463  */
 464 void
 465 intel_region_release_pbo(struct intel_context *intel,
 466                          struct intel_region *region)
 467 {
 468    _DBG("%s %p\n", __FUNCTION__, region);
 469    assert(region->buffer == region->pbo->buffer);
 470    region->pbo->region = NULL;
 471    region->pbo = NULL;
 472    drm_intel_bo_unreference(region->buffer);
 473    region->buffer = NULL;
 474
 475    region->buffer = drm_intel_bo_alloc(intel->bufmgr, "region",
 476                                        region->pitch * region->cpp *
 477                                        region->height,
 478                                        64);
 479 }
 480
 481 /* Break the COW tie to the pbo.  Both the pbo and the region end up
 482  * with a copy of the data.
 483  */
 484 void
 485 intel_region_cow(struct intel_context *intel, struct intel_region *region)
 486 {
 487    struct intel_buffer_object *pbo = region->pbo;
 488    GLboolean ok;
 489
 490    intel_region_release_pbo(intel, region);
 491
 492    assert(region->cpp * region->pitch * region->height == pbo->Base.Size);
 493
 494    _DBG("%s %p (%d bytes)\n", __FUNCTION__, region, pbo->Base.Size);
 495
 496    /* Now blit from the texture buffer to the new buffer:
 497     */
 498
 499    intel_prepare_render(intel);
 500    ok = intelEmitCopyBlit(intel,
 501                           region->cpp,
 502                           region->pitch, pbo->buffer, 0, region->tiling,
 503                           region->pitch, region->buffer, 0, region->tiling,
 504                           0, 0, 0, 0,
 505                           region->pitch, region->height,
 506                           GL_COPY);
 507    assert(ok);
 508 }
 509
 510 drm_intel_bo *
 511 intel_region_buffer(struct intel_context *intel,
 512                     struct intel_region *region, GLuint flag)
 513 {
 514    if (region->pbo) {
 515       if (flag == INTEL_WRITE_PART)
 516          intel_region_cow(intel, region);
 517       else if (flag == INTEL_WRITE_FULL)
 518          intel_region_release_pbo(intel, region);
 519    }
 520
 521    return region->buffer;
 522 }