libgomp/oacc-mem.c

   1 /* OpenACC Runtime initialization routines
   2
   3    Copyright (C) 2013-2018 Free Software Foundation, Inc.
   4
   5    Contributed by Mentor Embedded.
   6
   7    This file is part of the GNU Offloading and Multi Processing Library
   8    (libgomp).
   9
  10    Libgomp is free software; you can redistribute it and/or modify it
  11    under the terms of the GNU General Public License as published by
  12    the Free Software Foundation; either version 3, or (at your option)
  13    any later version.
  14
  15    Libgomp is distributed in the hope that it will be useful, but WITHOUT ANY
  16    WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
  17    FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  18    more details.
  19
  20    Under Section 7 of GPL version 3, you are granted additional
  21    permissions described in the GCC Runtime Library Exception, version
  22    3.1, as published by the Free Software Foundation.
  23
  24    You should have received a copy of the GNU General Public License and
  25    a copy of the GCC Runtime Library Exception along with this program;
  26    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
  27    <http://www.gnu.org/licenses/>.  */
  28
  29 #include "openacc.h"
  30 #include "config.h"
  31 #include "libgomp.h"
  32 #include "gomp-constants.h"
  33 #include "oacc-int.h"
  34 #include <stdint.h>
  35 #include <string.h>
  36 #include <assert.h>
  37
  38 /* Return block containing [H->S), or NULL if not contained.  The device lock
  39    for DEV must be locked on entry, and remains locked on exit.  */
  40
  41 static splay_tree_key
  42 lookup_host (struct gomp_device_descr *dev, void *h, size_t s)
  43 {
  44   struct splay_tree_key_s node;
  45   splay_tree_key key;
  46
  47   node.host_start = (uintptr_t) h;
  48   node.host_end = (uintptr_t) h + s;
  49
  50   key = splay_tree_lookup (&dev->mem_map, &node);
  51
  52   return key;
  53 }
  54
  55 /* Return block containing [D->S), or NULL if not contained.
  56    The list isn't ordered by device address, so we have to iterate
  57    over the whole array.  This is not expected to be a common
  58    operation.  The device lock associated with TGT must be locked on entry, and
  59    remains locked on exit.  */
  60
  61 static splay_tree_key
  62 lookup_dev (struct target_mem_desc *tgt, void *d, size_t s)
  63 {
  64   int i;
  65   struct target_mem_desc *t;
  66
  67   if (!tgt)
  68     return NULL;
  69
  70   for (t = tgt; t != NULL; t = t->prev)
  71     {
  72       if (t->tgt_start <= (uintptr_t) d && t->tgt_end >= (uintptr_t) d + s)
  73         break;
  74     }
  75
  76   if (!t)
  77     return NULL;
  78
  79   for (i = 0; i < t->list_count; i++)
  80     {
  81       void * offset;
  82
  83       splay_tree_key k = &t->array[i].key;
  84       offset = d - t->tgt_start + k->tgt_offset;
  85
  86       if (k->host_start + offset <= (void *) k->host_end)
  87         return k;
  88     }
  89
  90   return NULL;
  91 }
  92
  93 /* OpenACC is silent on how memory exhaustion is indicated.  We return
  94    NULL.  */
  95
  96 void *
  97 acc_malloc (size_t s)
  98 {
  99   if (!s)
 100     return NULL;
 101
 102   goacc_lazy_initialize ();
 103
 104   struct goacc_thread *thr = goacc_thread ();
 105
 106   assert (thr->dev);
 107
 108   if (thr->dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 109     return malloc (s);
 110
 111   return thr->dev->alloc_func (thr->dev->target_id, s);
 112 }
 113
 114 /* OpenACC 2.0a (3.2.16) doesn't specify what to do in the event
 115    the device address is mapped. We choose to check if it mapped,
 116    and if it is, to unmap it. */
 117 void
 118 acc_free (void *d)
 119 {
 120   splay_tree_key k;
 121
 122   if (!d)
 123     return;
 124
 125   struct goacc_thread *thr = goacc_thread ();
 126
 127   assert (thr && thr->dev);
 128
 129   struct gomp_device_descr *acc_dev = thr->dev;
 130
 131   if (acc_dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 132     return free (d);
 133
 134   gomp_mutex_lock (&acc_dev->lock);
 135
 136   /* We don't have to call lazy open here, as the ptr value must have
 137      been returned by acc_malloc.  It's not permitted to pass NULL in
 138      (unless you got that null from acc_malloc).  */
 139   if ((k = lookup_dev (acc_dev->openacc.data_environ, d, 1)))
 140     {
 141       void *offset;
 142
 143       offset = d - k->tgt->tgt_start + k->tgt_offset;
 144
 145       gomp_mutex_unlock (&acc_dev->lock);
 146
 147       acc_unmap_data ((void *)(k->host_start + offset));
 148     }
 149   else
 150     gomp_mutex_unlock (&acc_dev->lock);
 151
 152   if (!acc_dev->free_func (acc_dev->target_id, d))
 153     gomp_fatal ("error in freeing device memory in %s", __FUNCTION__);
 154 }
 155
 156 static void
 157 memcpy_tofrom_device (bool from, void *d, void *h, size_t s, int async,
 158                       const char *libfnname)
 159 {
 160   /* No need to call lazy open here, as the device pointer must have
 161      been obtained from a routine that did that.  */
 162   struct goacc_thread *thr = goacc_thread ();
 163
 164   assert (thr && thr->dev);
 165
 166   if (thr->dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 167     {
 168       if (from)
 169         memmove (h, d, s);
 170       else
 171         memmove (d, h, s);
 172       return;
 173     }
 174
 175   if (async > acc_async_sync)
 176     thr->dev->openacc.async_set_async_func (async);
 177
 178   bool ret = (from
 179               ? thr->dev->dev2host_func (thr->dev->target_id, h, d, s)
 180               : thr->dev->host2dev_func (thr->dev->target_id, d, h, s));
 181
 182   if (async > acc_async_sync)
 183     thr->dev->openacc.async_set_async_func (acc_async_sync);
 184
 185   if (!ret)
 186     gomp_fatal ("error in %s", libfnname);
 187 }
 188
 189 void
 190 acc_memcpy_to_device (void *d, void *h, size_t s)
 191 {
 192   memcpy_tofrom_device (false, d, h, s, acc_async_sync, __FUNCTION__);
 193 }
 194
 195 void
 196 acc_memcpy_to_device_async (void *d, void *h, size_t s, int async)
 197 {
 198   memcpy_tofrom_device (false, d, h, s, async, __FUNCTION__);
 199 }
 200
 201 void
 202 acc_memcpy_from_device (void *h, void *d, size_t s)
 203 {
 204   memcpy_tofrom_device (true, d, h, s, acc_async_sync, __FUNCTION__);
 205 }
 206
 207 void
 208 acc_memcpy_from_device_async (void *h, void *d, size_t s, int async)
 209 {
 210   memcpy_tofrom_device (true, d, h, s, async, __FUNCTION__);
 211 }
 212
 213 /* Return the device pointer that corresponds to host data H.  Or NULL
 214    if no mapping.  */
 215
 216 void *
 217 acc_deviceptr (void *h)
 218 {
 219   splay_tree_key n;
 220   void *d;
 221   void *offset;
 222
 223   goacc_lazy_initialize ();
 224
 225   struct goacc_thread *thr = goacc_thread ();
 226   struct gomp_device_descr *dev = thr->dev;
 227
 228   if (thr->dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 229     return h;
 230
 231   gomp_mutex_lock (&dev->lock);
 232
 233   n = lookup_host (dev, h, 1);
 234
 235   if (!n)
 236     {
 237       gomp_mutex_unlock (&dev->lock);
 238       return NULL;
 239     }
 240
 241   offset = h - n->host_start;
 242
 243   d = n->tgt->tgt_start + n->tgt_offset + offset;
 244
 245   gomp_mutex_unlock (&dev->lock);
 246
 247   return d;
 248 }
 249
 250 /* Return the host pointer that corresponds to device data D.  Or NULL
 251    if no mapping.  */
 252
 253 void *
 254 acc_hostptr (void *d)
 255 {
 256   splay_tree_key n;
 257   void *h;
 258   void *offset;
 259
 260   goacc_lazy_initialize ();
 261
 262   struct goacc_thread *thr = goacc_thread ();
 263   struct gomp_device_descr *acc_dev = thr->dev;
 264
 265   if (thr->dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 266     return d;
 267
 268   gomp_mutex_lock (&acc_dev->lock);
 269
 270   n = lookup_dev (acc_dev->openacc.data_environ, d, 1);
 271
 272   if (!n)
 273     {
 274       gomp_mutex_unlock (&acc_dev->lock);
 275       return NULL;
 276     }
 277
 278   offset = d - n->tgt->tgt_start + n->tgt_offset;
 279
 280   h = n->host_start + offset;
 281
 282   gomp_mutex_unlock (&acc_dev->lock);
 283
 284   return h;
 285 }
 286
 287 /* Return 1 if host data [H,+S] is present on the device.  */
 288
 289 int
 290 acc_is_present (void *h, size_t s)
 291 {
 292   splay_tree_key n;
 293
 294   if (!s || !h)
 295     return 0;
 296
 297   goacc_lazy_initialize ();
 298
 299   struct goacc_thread *thr = goacc_thread ();
 300   struct gomp_device_descr *acc_dev = thr->dev;
 301
 302   if (thr->dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 303     return h != NULL;
 304
 305   gomp_mutex_lock (&acc_dev->lock);
 306
 307   n = lookup_host (acc_dev, h, s);
 308
 309   if (n && ((uintptr_t)h < n->host_start
 310             || (uintptr_t)h + s > n->host_end
 311             || s > n->host_end - n->host_start))
 312     n = NULL;
 313
 314   gomp_mutex_unlock (&acc_dev->lock);
 315
 316   return n != NULL;
 317 }
 318
 319 /* Create a mapping for host [H,+S] -> device [D,+S] */
 320
 321 void
 322 acc_map_data (void *h, void *d, size_t s)
 323 {
 324   struct target_mem_desc *tgt = NULL;
 325   size_t mapnum = 1;
 326   void *hostaddrs = h;
 327   void *devaddrs = d;
 328   size_t sizes = s;
 329   unsigned short kinds = GOMP_MAP_ALLOC;
 330
 331   goacc_lazy_initialize ();
 332
 333   struct goacc_thread *thr = goacc_thread ();
 334   struct gomp_device_descr *acc_dev = thr->dev;
 335
 336   if (acc_dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 337     {
 338       if (d != h)
 339         gomp_fatal ("cannot map data on shared-memory system");
 340     }
 341   else
 342     {
 343       struct goacc_thread *thr = goacc_thread ();
 344
 345       if (!d || !h || !s)
 346         gomp_fatal ("[%p,+%d]->[%p,+%d] is a bad map",
 347                     (void *)h, (int)s, (void *)d, (int)s);
 348
 349       gomp_mutex_lock (&acc_dev->lock);
 350
 351       if (lookup_host (acc_dev, h, s))
 352         {
 353           gomp_mutex_unlock (&acc_dev->lock);
 354           gomp_fatal ("host address [%p, +%d] is already mapped", (void *)h,
 355                       (int)s);
 356         }
 357
 358       if (lookup_dev (thr->dev->openacc.data_environ, d, s))
 359         {
 360           gomp_mutex_unlock (&acc_dev->lock);
 361           gomp_fatal ("device address [%p, +%d] is already mapped", (void *)d,
 362                       (int)s);
 363         }
 364
 365       gomp_mutex_unlock (&acc_dev->lock);
 366
 367       tgt = gomp_map_vars (acc_dev, mapnum, &hostaddrs, &devaddrs, &sizes,
 368                            &kinds, true, GOMP_MAP_VARS_OPENACC);
 369       tgt->list[0].key->refcount = REFCOUNT_INFINITY;
 370     }
 371
 372   gomp_mutex_lock (&acc_dev->lock);
 373   tgt->prev = acc_dev->openacc.data_environ;
 374   acc_dev->openacc.data_environ = tgt;
 375   gomp_mutex_unlock (&acc_dev->lock);
 376 }
 377
 378 void
 379 acc_unmap_data (void *h)
 380 {
 381   struct goacc_thread *thr = goacc_thread ();
 382   struct gomp_device_descr *acc_dev = thr->dev;
 383
 384   /* No need to call lazy open, as the address must have been mapped.  */
 385
 386   /* This is a no-op on shared-memory targets.  */
 387   if (acc_dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 388     return;
 389
 390   size_t host_size;
 391
 392   gomp_mutex_lock (&acc_dev->lock);
 393
 394   splay_tree_key n = lookup_host (acc_dev, h, 1);
 395   struct target_mem_desc *t;
 396
 397   if (!n)
 398     {
 399       gomp_mutex_unlock (&acc_dev->lock);
 400       gomp_fatal ("%p is not a mapped block", (void *)h);
 401     }
 402
 403   host_size = n->host_end - n->host_start;
 404
 405   if (n->host_start != (uintptr_t) h)
 406     {
 407       gomp_mutex_unlock (&acc_dev->lock);
 408       gomp_fatal ("[%p,%d] surrounds %p",
 409                   (void *) n->host_start, (int) host_size, (void *) h);
 410     }
 411
 412   /* Mark for removal.  */
 413   n->refcount = 1;
 414
 415   t = n->tgt;
 416
 417   if (t->refcount == 2)
 418     {
 419       struct target_mem_desc *tp;
 420
 421       /* This is the last reference, so pull the descriptor off the
 422          chain. This avoids gomp_unmap_vars via gomp_unmap_tgt from
 423          freeing the device memory. */
 424       t->tgt_end = 0;
 425       t->to_free = 0;
 426
 427       for (tp = NULL, t = acc_dev->openacc.data_environ; t != NULL;
 428            tp = t, t = t->prev)
 429         if (n->tgt == t)
 430           {
 431             if (tp)
 432               tp->prev = t->prev;
 433             else
 434               acc_dev->openacc.data_environ = t->prev;
 435
 436             break;
 437           }
 438     }
 439
 440   gomp_mutex_unlock (&acc_dev->lock);
 441
 442   gomp_unmap_vars (t, true);
 443 }
 444
 445 #define FLAG_PRESENT (1 << 0)
 446 #define FLAG_CREATE (1 << 1)
 447 #define FLAG_COPY (1 << 2)
 448
 449 static void *
 450 present_create_copy (unsigned f, void *h, size_t s, int async)
 451 {
 452   void *d;
 453   splay_tree_key n;
 454
 455   if (!h || !s)
 456     gomp_fatal ("[%p,+%d] is a bad range", (void *)h, (int)s);
 457
 458   goacc_lazy_initialize ();
 459
 460   struct goacc_thread *thr = goacc_thread ();
 461   struct gomp_device_descr *acc_dev = thr->dev;
 462
 463   if (acc_dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 464     return h;
 465
 466   gomp_mutex_lock (&acc_dev->lock);
 467
 468   n = lookup_host (acc_dev, h, s);
 469   if (n)
 470     {
 471       /* Present. */
 472       d = (void *) (n->tgt->tgt_start + n->tgt_offset);
 473
 474       if (!(f & FLAG_PRESENT))
 475         {
 476           gomp_mutex_unlock (&acc_dev->lock);
 477           gomp_fatal ("[%p,+%d] already mapped to [%p,+%d]",
 478                       (void *)h, (int)s, (void *)d, (int)s);
 479         }
 480       if ((h + s) > (void *)n->host_end)
 481         {
 482           gomp_mutex_unlock (&acc_dev->lock);
 483           gomp_fatal ("[%p,+%d] not mapped", (void *)h, (int)s);
 484         }
 485
 486       if (n->refcount != REFCOUNT_INFINITY)
 487         {
 488           n->refcount++;
 489           n->dynamic_refcount++;
 490         }
 491       gomp_mutex_unlock (&acc_dev->lock);
 492     }
 493   else if (!(f & FLAG_CREATE))
 494     {
 495       gomp_mutex_unlock (&acc_dev->lock);
 496       gomp_fatal ("[%p,+%d] not mapped", (void *)h, (int)s);
 497     }
 498   else
 499     {
 500       struct target_mem_desc *tgt;
 501       size_t mapnum = 1;
 502       unsigned short kinds;
 503       void *hostaddrs = h;
 504
 505       if (f & FLAG_COPY)
 506         kinds = GOMP_MAP_TO;
 507       else
 508         kinds = GOMP_MAP_ALLOC;
 509
 510       gomp_mutex_unlock (&acc_dev->lock);
 511
 512       if (async > acc_async_sync)
 513         acc_dev->openacc.async_set_async_func (async);
 514
 515       tgt = gomp_map_vars (acc_dev, mapnum, &hostaddrs, NULL, &s, &kinds, true,
 516                            GOMP_MAP_VARS_OPENACC);
 517       /* Initialize dynamic refcount.  */
 518       tgt->list[0].key->dynamic_refcount = 1;
 519
 520       if (async > acc_async_sync)
 521         acc_dev->openacc.async_set_async_func (acc_async_sync);
 522
 523       gomp_mutex_lock (&acc_dev->lock);
 524
 525       d = tgt->to_free;
 526       tgt->prev = acc_dev->openacc.data_environ;
 527       acc_dev->openacc.data_environ = tgt;
 528
 529       gomp_mutex_unlock (&acc_dev->lock);
 530     }
 531
 532   return d;
 533 }
 534
 535 void *
 536 acc_create (void *h, size_t s)
 537 {
 538   return present_create_copy (FLAG_PRESENT | FLAG_CREATE, h, s, acc_async_sync);
 539 }
 540
 541 void
 542 acc_create_async (void *h, size_t s, int async)
 543 {
 544   present_create_copy (FLAG_PRESENT | FLAG_CREATE, h, s, async);
 545 }
 546
 547 /* acc_present_or_create used to be what acc_create is now.  */
 548 /* acc_pcreate is acc_present_or_create by a different name.  */
 549 #ifdef HAVE_ATTRIBUTE_ALIAS
 550 strong_alias (acc_create, acc_present_or_create)
 551 strong_alias (acc_create, acc_pcreate)
 552 #else
 553 void *
 554 acc_present_or_create (void *h, size_t s)
 555 {
 556   return acc_create (h, s);
 557 }
 558
 559 void *
 560 acc_pcreate (void *h, size_t s)
 561 {
 562   return acc_create (h, s);
 563 }
 564 #endif
 565
 566 void *
 567 acc_copyin (void *h, size_t s)
 568 {
 569   return present_create_copy (FLAG_PRESENT | FLAG_CREATE | FLAG_COPY, h, s,
 570                               acc_async_sync);
 571 }
 572
 573 void
 574 acc_copyin_async (void *h, size_t s, int async)
 575 {
 576   present_create_copy (FLAG_PRESENT | FLAG_CREATE | FLAG_COPY, h, s, async);
 577 }
 578
 579 /* acc_present_or_copyin used to be what acc_copyin is now.  */
 580 /* acc_pcopyin is acc_present_or_copyin by a different name.  */
 581 #ifdef HAVE_ATTRIBUTE_ALIAS
 582 strong_alias (acc_copyin, acc_present_or_copyin)
 583 strong_alias (acc_copyin, acc_pcopyin)
 584 #else
 585 void *
 586 acc_present_or_copyin (void *h, size_t s)
 587 {
 588   return acc_copyin (h, s);
 589 }
 590
 591 void *
 592 acc_pcopyin (void *h, size_t s)
 593 {
 594   return acc_copyin (h, s);
 595 }
 596 #endif
 597
 598 #define FLAG_COPYOUT  (1 << 0)
 599 #define FLAG_FINALIZE (1 << 1)
 600
 601 static void
 602 delete_copyout (unsigned f, void *h, size_t s, int async, const char *libfnname)
 603 {
 604   size_t host_size;
 605   splay_tree_key n;
 606   void *d;
 607   struct goacc_thread *thr = goacc_thread ();
 608   struct gomp_device_descr *acc_dev = thr->dev;
 609
 610   if (acc_dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 611     return;
 612
 613   gomp_mutex_lock (&acc_dev->lock);
 614
 615   n = lookup_host (acc_dev, h, s);
 616
 617   /* No need to call lazy open, as the data must already have been
 618      mapped.  */
 619
 620   if (!n)
 621     {
 622       gomp_mutex_unlock (&acc_dev->lock);
 623       gomp_fatal ("[%p,%d] is not mapped", (void *)h, (int)s);
 624     }
 625
 626   d = (void *) (n->tgt->tgt_start + n->tgt_offset
 627                 + (uintptr_t) h - n->host_start);
 628
 629   host_size = n->host_end - n->host_start;
 630
 631   if (n->host_start != (uintptr_t) h || host_size != s)
 632     {
 633       gomp_mutex_unlock (&acc_dev->lock);
 634       gomp_fatal ("[%p,%d] surrounds2 [%p,+%d]",
 635                   (void *) n->host_start, (int) host_size, (void *) h, (int) s);
 636     }
 637
 638   if (n->refcount == REFCOUNT_INFINITY)
 639     {
 640       n->refcount = 0;
 641       n->dynamic_refcount = 0;
 642     }
 643   if (n->refcount < n->dynamic_refcount)
 644     {
 645       gomp_mutex_unlock (&acc_dev->lock);
 646       gomp_fatal ("Dynamic reference counting assert fail\n");
 647     }
 648
 649   if (f & FLAG_FINALIZE)
 650     {
 651       n->refcount -= n->dynamic_refcount;
 652       n->dynamic_refcount = 0;
 653     }
 654   else if (n->dynamic_refcount)
 655     {
 656       n->dynamic_refcount--;
 657       n->refcount--;
 658     }
 659
 660   if (n->refcount == 0)
 661     {
 662       if (n->tgt->refcount == 2)
 663         {
 664           struct target_mem_desc *tp, *t;
 665           for (tp = NULL, t = acc_dev->openacc.data_environ; t != NULL;
 666                tp = t, t = t->prev)
 667             if (n->tgt == t)
 668               {
 669                 if (tp)
 670                   tp->prev = t->prev;
 671                 else
 672                   acc_dev->openacc.data_environ = t->prev;
 673                 break;
 674               }
 675         }
 676
 677       if (f & FLAG_COPYOUT)
 678         {
 679           if (async > acc_async_sync)
 680             acc_dev->openacc.async_set_async_func (async);
 681           acc_dev->dev2host_func (acc_dev->target_id, h, d, s);
 682           if (async > acc_async_sync)
 683             acc_dev->openacc.async_set_async_func (acc_async_sync);
 684         }
 685
 686       gomp_remove_var (acc_dev, n);
 687     }
 688
 689   gomp_mutex_unlock (&acc_dev->lock);
 690 }
 691
 692 void
 693 acc_delete (void *h , size_t s)
 694 {
 695   delete_copyout (0, h, s, acc_async_sync, __FUNCTION__);
 696 }
 697
 698 void
 699 acc_delete_async (void *h , size_t s, int async)
 700 {
 701   delete_copyout (0, h, s, async, __FUNCTION__);
 702 }
 703
 704 void
 705 acc_delete_finalize (void *h , size_t s)
 706 {
 707   delete_copyout (FLAG_FINALIZE, h, s, acc_async_sync, __FUNCTION__);
 708 }
 709
 710 void
 711 acc_delete_finalize_async (void *h , size_t s, int async)
 712 {
 713   delete_copyout (FLAG_FINALIZE, h, s, async, __FUNCTION__);
 714 }
 715
 716 void
 717 acc_copyout (void *h, size_t s)
 718 {
 719   delete_copyout (FLAG_COPYOUT, h, s, acc_async_sync, __FUNCTION__);
 720 }
 721
 722 void
 723 acc_copyout_async (void *h, size_t s, int async)
 724 {
 725   delete_copyout (FLAG_COPYOUT, h, s, async, __FUNCTION__);
 726 }
 727
 728 void
 729 acc_copyout_finalize (void *h, size_t s)
 730 {
 731   delete_copyout (FLAG_COPYOUT | FLAG_FINALIZE, h, s, acc_async_sync,
 732                   __FUNCTION__);
 733 }
 734
 735 void
 736 acc_copyout_finalize_async (void *h, size_t s, int async)
 737 {
 738   delete_copyout (FLAG_COPYOUT | FLAG_FINALIZE, h, s, async, __FUNCTION__);
 739 }
 740
 741 static void
 742 update_dev_host (int is_dev, void *h, size_t s, int async)
 743 {
 744   splay_tree_key n;
 745   void *d;
 746
 747   goacc_lazy_initialize ();
 748
 749   struct goacc_thread *thr = goacc_thread ();
 750   struct gomp_device_descr *acc_dev = thr->dev;
 751
 752   if (acc_dev->capabilities & GOMP_OFFLOAD_CAP_SHARED_MEM)
 753     return;
 754
 755   gomp_mutex_lock (&acc_dev->lock);
 756
 757   n = lookup_host (acc_dev, h, s);
 758
 759   if (!n)
 760     {
 761       gomp_mutex_unlock (&acc_dev->lock);
 762       gomp_fatal ("[%p,%d] is not mapped", h, (int)s);
 763     }
 764
 765   d = (void *) (n->tgt->tgt_start + n->tgt_offset
 766                 + (uintptr_t) h - n->host_start);
 767
 768   if (async > acc_async_sync)
 769     acc_dev->openacc.async_set_async_func (async);
 770
 771   if (is_dev)
 772     acc_dev->host2dev_func (acc_dev->target_id, d, h, s);
 773   else
 774     acc_dev->dev2host_func (acc_dev->target_id, h, d, s);
 775
 776   if (async > acc_async_sync)
 777     acc_dev->openacc.async_set_async_func (acc_async_sync);
 778
 779   gomp_mutex_unlock (&acc_dev->lock);
 780 }
 781
 782 void
 783 acc_update_device (void *h, size_t s)
 784 {
 785   update_dev_host (1, h, s, acc_async_sync);
 786 }
 787
 788 void
 789 acc_update_device_async (void *h, size_t s, int async)
 790 {
 791   update_dev_host (1, h, s, async);
 792 }
 793
 794 void
 795 acc_update_self (void *h, size_t s)
 796 {
 797   update_dev_host (0, h, s, acc_async_sync);
 798 }
 799
 800 void
 801 acc_update_self_async (void *h, size_t s, int async)
 802 {
 803   update_dev_host (0, h, s, async);
 804 }
 805
 806 void
 807 gomp_acc_insert_pointer (size_t mapnum, void **hostaddrs, size_t *sizes,
 808                          void *kinds)
 809 {
 810   struct target_mem_desc *tgt;
 811   struct goacc_thread *thr = goacc_thread ();
 812   struct gomp_device_descr *acc_dev = thr->dev;
 813
 814   if (acc_is_present (*hostaddrs, *sizes))
 815     {
 816       splay_tree_key n;
 817       gomp_mutex_lock (&acc_dev->lock);
 818       n = lookup_host (acc_dev, *hostaddrs, *sizes);
 819       gomp_mutex_unlock (&acc_dev->lock);
 820
 821       tgt = n->tgt;
 822       for (size_t i = 0; i < tgt->list_count; i++)
 823         if (tgt->list[i].key == n)
 824           {
 825             for (size_t j = 0; j < mapnum; j++)
 826               if (i + j < tgt->list_count && tgt->list[i + j].key)
 827                 {
 828                   tgt->list[i + j].key->refcount++;
 829                   tgt->list[i + j].key->dynamic_refcount++;
 830                 }
 831             return;
 832           }
 833       /* Should not reach here.  */
 834       gomp_fatal ("Dynamic refcount incrementing failed for pointer/pset");
 835     }
 836
 837   gomp_debug (0, "  %s: prepare mappings\n", __FUNCTION__);
 838   tgt = gomp_map_vars (acc_dev, mapnum, hostaddrs,
 839                        NULL, sizes, kinds, true, GOMP_MAP_VARS_OPENACC);
 840   gomp_debug (0, "  %s: mappings prepared\n", __FUNCTION__);
 841
 842   /* Initialize dynamic refcount.  */
 843   tgt->list[0].key->dynamic_refcount = 1;
 844
 845   gomp_mutex_lock (&acc_dev->lock);
 846   tgt->prev = acc_dev->openacc.data_environ;
 847   acc_dev->openacc.data_environ = tgt;
 848   gomp_mutex_unlock (&acc_dev->lock);
 849 }
 850
 851 void
 852 gomp_acc_remove_pointer (void *h, size_t s, bool force_copyfrom, int async,
 853                          int finalize, int mapnum)
 854 {
 855   struct goacc_thread *thr = goacc_thread ();
 856   struct gomp_device_descr *acc_dev = thr->dev;
 857   splay_tree_key n;
 858   struct target_mem_desc *t;
 859   int minrefs = (mapnum == 1) ? 2 : 3;
 860
 861   if (!acc_is_present (h, s))
 862     return;
 863
 864   gomp_mutex_lock (&acc_dev->lock);
 865
 866   n = lookup_host (acc_dev, h, 1);
 867
 868   if (!n)
 869     {
 870       gomp_mutex_unlock (&acc_dev->lock);
 871       gomp_fatal ("%p is not a mapped block", (void *)h);
 872     }
 873
 874   gomp_debug (0, "  %s: restore mappings\n", __FUNCTION__);
 875
 876   t = n->tgt;
 877
 878   if (n->refcount < n->dynamic_refcount)
 879     {
 880       gomp_mutex_unlock (&acc_dev->lock);
 881       gomp_fatal ("Dynamic reference counting assert fail\n");
 882     }
 883
 884   if (finalize)
 885     {
 886       n->refcount -= n->dynamic_refcount;
 887       n->dynamic_refcount = 0;
 888     }
 889   else if (n->dynamic_refcount)
 890     {
 891       n->dynamic_refcount--;
 892       n->refcount--;
 893     }
 894
 895   gomp_mutex_unlock (&acc_dev->lock);
 896
 897   if (n->refcount == 0)
 898     {
 899       if (t->refcount == minrefs)
 900         {
 901           /* This is the last reference, so pull the descriptor off the
 902              chain. This prevents gomp_unmap_vars via gomp_unmap_tgt from
 903              freeing the device memory. */
 904           struct target_mem_desc *tp;
 905           for (tp = NULL, t = acc_dev->openacc.data_environ; t != NULL;
 906                tp = t, t = t->prev)
 907             {
 908               if (n->tgt == t)
 909                 {
 910                   if (tp)
 911                     tp->prev = t->prev;
 912                   else
 913                     acc_dev->openacc.data_environ = t->prev;
 914                   break;
 915                 }
 916             }
 917         }
 918
 919       /* Set refcount to 1 to allow gomp_unmap_vars to unmap it.  */
 920       n->refcount = 1;
 921       t->refcount = minrefs;
 922       for (size_t i = 0; i < t->list_count; i++)
 923         if (t->list[i].key == n)
 924           {
 925             t->list[i].copy_from = force_copyfrom ? 1 : 0;
 926             break;
 927           }
 928
 929       /* If running synchronously, unmap immediately.  */
 930       if (async < acc_async_noval)
 931         gomp_unmap_vars (t, true);
 932       else
 933         t->device_descr->openacc.register_async_cleanup_func (t, async);
 934     }
 935
 936   gomp_mutex_unlock (&acc_dev->lock);
 937
 938   gomp_debug (0, "  %s: mappings restored\n", __FUNCTION__);
 939 }