PrimOps.cmm 62.2 KB
Newer Older
1
/* -*- tab-width: 8 -*- */
2 3
/* -----------------------------------------------------------------------------
 *
4
 * (c) The GHC Team, 1998-2012
5 6 7 8 9 10 11 12 13
 *
 * Out-of-line primitive operations
 *
 * This file contains the implementations of all the primitive
 * operations ("primops") which are not expanded inline.  See
 * ghc/compiler/prelude/primops.txt.pp for a list of all the primops;
 * this file contains code for most of those with the attribute
 * out_of_line=True.
 *
14 15 16
 * Entry convention: the entry convention for a primop is the
 * NativeNodeCall convention, and the return convention is
 * NativeReturn.  (see compiler/cmm/CmmCallConv.hs)
17 18 19 20 21 22 23 24 25
 *
 * This file is written in a subset of C--, extended with various
 * features specific to GHC.  It is compiled by GHC directly.  For the
 * syntax of .cmm files, see the parser in ghc/compiler/cmm/CmmParse.y.
 *
 * ---------------------------------------------------------------------------*/

#include "Cmm.h"

26
#ifdef __PIC__
27 28
import pthread_mutex_lock;
import pthread_mutex_unlock;
29
#endif
30
import base_ControlziExceptionziBase_nestedAtomically_closure;
31 32
import EnterCriticalSection;
import LeaveCriticalSection;
Ian Lynagh's avatar
Ian Lynagh committed
33
import ghczmprim_GHCziTypes_False_closure;
34
#if defined(USE_MINIINTERPRETER) || !defined(mingw32_HOST_OS)
35
import sm_mutex;
36
#endif
37

38 39 40 41 42 43 44 45 46 47 48 49 50
/*-----------------------------------------------------------------------------
  Array Primitives

  Basically just new*Array - the others are all inline macros.

  The slow entry point is for returning from a heap check, the saved
  size argument must be re-loaded from the stack.
  -------------------------------------------------------------------------- */

/* for objects that are *less* than the size of a word, make sure we
 * round up to the nearest word for the size of the array.
 */

51
stg_newByteArrayzh ( W_ n )
52
{
53 54 55 56 57
    W_ words, payload_words;
    gcptr p;

    MAYBE_GC_N(stg_newByteArrayzh, n);

58 59
    payload_words = ROUNDUP_BYTES_TO_WDS(n);
    words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
60
    ("ptr" p) = ccall allocate(MyCapability() "ptr",words);
61
    TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
62
    SET_HDR(p, stg_ARR_WORDS_info, CCCS);
63
    StgArrWords_bytes(p) = n;
64
    return (p);
65 66
}

Simon Marlow's avatar
Simon Marlow committed
67 68 69
#define BA_ALIGN 16
#define BA_MASK  (BA_ALIGN-1)

70
stg_newPinnedByteArrayzh ( W_ n )
71
{
72 73 74 75
    W_ words, bytes, payload_words;
    gcptr p;

    MAYBE_GC_N(stg_newPinnedByteArrayzh, n);
76

77
    bytes = n;
78 79 80 81 82 83 84 85 86 87
    /* payload_words is what we will tell the profiler we had to allocate */
    payload_words = ROUNDUP_BYTES_TO_WDS(bytes);
    /* When we actually allocate memory, we need to allow space for the
       header: */
    bytes = bytes + SIZEOF_StgArrWords;
    /* And we want to align to BA_ALIGN bytes, so we need to allow space
       to shift up to BA_ALIGN - 1 bytes: */
    bytes = bytes + BA_ALIGN - 1;
    /* Now we convert to a number of words: */
    words = ROUNDUP_BYTES_TO_WDS(bytes);
Simon Marlow's avatar
Simon Marlow committed
88

89
    ("ptr" p) = ccall allocatePinned(MyCapability() "ptr", words);
Simon Marlow's avatar
Simon Marlow committed
90 91
    TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);

92 93
    /* Now we need to move p forward so that the payload is aligned
       to BA_ALIGN bytes: */
Simon Marlow's avatar
Simon Marlow committed
94 95
    p = p + ((-p - SIZEOF_StgArrWords) & BA_MASK);

96
    SET_HDR(p, stg_ARR_WORDS_info, CCCS);
97
    StgArrWords_bytes(p) = n;
98
    return (p);
Simon Marlow's avatar
Simon Marlow committed
99 100
}

101
stg_newAlignedPinnedByteArrayzh ( W_ n, W_ alignment )
Simon Marlow's avatar
Simon Marlow committed
102
{
103 104
    W_ words, bytes, payload_words;
    gcptr p;
Simon Marlow's avatar
Simon Marlow committed
105

106
    again: MAYBE_GC(again);
Simon Marlow's avatar
Simon Marlow committed
107

108 109 110 111 112
    /* we always supply at least word-aligned memory, so there's no
       need to allow extra space for alignment if the requirement is less
       than a word.  This also prevents mischief with alignment == 0. */
    if (alignment <= SIZEOF_W) { alignment = 1; }

113 114
    bytes = n;

115 116
    /* payload_words is what we will tell the profiler we had to allocate */
    payload_words = ROUNDUP_BYTES_TO_WDS(bytes);
Simon Marlow's avatar
Simon Marlow committed
117

118 119 120 121 122 123 124 125
    /* When we actually allocate memory, we need to allow space for the
       header: */
    bytes = bytes + SIZEOF_StgArrWords;
    /* And we want to align to <alignment> bytes, so we need to allow space
       to shift up to <alignment - 1> bytes: */
    bytes = bytes + alignment - 1;
    /* Now we convert to a number of words: */
    words = ROUNDUP_BYTES_TO_WDS(bytes);
126

127
    ("ptr" p) = ccall allocatePinned(MyCapability() "ptr", words);
128 129
    TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);

130 131 132 133
    /* Now we need to move p forward so that the payload is aligned
       to <alignment> bytes. Note that we are assuming that
       <alignment> is a power of 2, which is technically not guaranteed */
    p = p + ((-p - SIZEOF_StgArrWords) & (alignment - 1));
134

135
    SET_HDR(p, stg_ARR_WORDS_info, CCCS);
136
    StgArrWords_bytes(p) = n;
137
    return (p);
138 139
}

140
stg_newArrayzh ( W_ n /* words */, gcptr init )
141
{
142 143
    W_ words, size;
    gcptr p, arr;
144

145
    again: MAYBE_GC(again);
146

147 148 149 150 151
    // the mark area contains one byte for each 2^MUT_ARR_PTRS_CARD_BITS words
    // in the array, making sure we round up, and then rounding up to a whole
    // number of words.
    size = n + mutArrPtrsCardWords(n);
    words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + size;
152
    ("ptr" arr) = ccall allocate(MyCapability() "ptr",words);
153 154
    TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);

155
    SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, CCCS);
156
    StgMutArrPtrs_ptrs(arr) = n;
157
    StgMutArrPtrs_size(arr) = size;
158 159 160 161 162

    // Initialise all elements of the the array with the value in R2
    p = arr + SIZEOF_StgMutArrPtrs;
  for:
    if (p < arr + WDS(words)) {
ian@well-typed.com's avatar
ian@well-typed.com committed
163 164 165
        W_[p] = init;
        p = p + WDS(1);
        goto for;
166
    }
167 168 169
    // Initialise the mark bits with 0
  for2:
    if (p < arr + WDS(size)) {
ian@well-typed.com's avatar
ian@well-typed.com committed
170 171 172
        W_[p] = 0;
        p = p + WDS(1);
        goto for2;
173
    }
174

175
    return (arr);
176 177
}

178
stg_unsafeThawArrayzh ( gcptr arr )
179 180 181
{
  // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
  //
ian@well-typed.com's avatar
ian@well-typed.com committed
182
  // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
183 184
  // normally doesn't.  However, when we freeze a MUT_ARR_PTRS, we leave
  // it on the mutable list for the GC to remove (removing something from
185
  // the mutable list is not easy).
ian@well-typed.com's avatar
ian@well-typed.com committed
186
  //
187
  // So that we can tell whether a MUT_ARR_PTRS_FROZEN is on the mutable list,
188 189 190
  // when we freeze it we set the info ptr to be MUT_ARR_PTRS_FROZEN0
  // to indicate that it is still on the mutable list.
  //
191 192
  // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
  // either it is on a mut_list, or it isn't.  We adopt the convention that
193
  // the closure type is MUT_ARR_PTRS_FROZEN0 if it is on the mutable list,
194 195 196
  // and MUT_ARR_PTRS_FROZEN otherwise.  In fact it wouldn't matter if
  // we put it on the mutable list more than once, but it would get scavenged
  // multiple times during GC, which would be unnecessarily slow.
197
  //
198 199 200
  if (StgHeader_info(arr) != stg_MUT_ARR_PTRS_FROZEN0_info) {
        SET_INFO(arr,stg_MUT_ARR_PTRS_DIRTY_info);
        recordMutable(arr);
ian@well-typed.com's avatar
ian@well-typed.com committed
201 202
        // must be done after SET_INFO, because it ASSERTs closure_MUTABLE()
        return (arr);
203
  } else {
ian@well-typed.com's avatar
ian@well-typed.com committed
204 205
        SET_INFO(arr,stg_MUT_ARR_PTRS_DIRTY_info);
        return (arr);
206 207 208
  }
}

209
stg_newArrayArrayzh ( W_ n /* words */ )
210
{
211 212
    W_ words, size;
    gcptr p, arr;
213

214
    MAYBE_GC_N(stg_newArrayArrayzh, n);
215 216 217 218 219 220

    // the mark area contains one byte for each 2^MUT_ARR_PTRS_CARD_BITS words
    // in the array, making sure we round up, and then rounding up to a whole
    // number of words.
    size = n + mutArrPtrsCardWords(n);
    words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + size;
221
    ("ptr" arr) = ccall allocate(MyCapability() "ptr",words);
222 223 224 225 226 227 228 229 230 231
    TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);

    SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, W_[CCCS]);
    StgMutArrPtrs_ptrs(arr) = n;
    StgMutArrPtrs_size(arr) = size;

    // Initialise all elements of the array with a pointer to the new array
    p = arr + SIZEOF_StgMutArrPtrs;
  for:
    if (p < arr + WDS(words)) {
ian@well-typed.com's avatar
ian@well-typed.com committed
232 233 234
        W_[p] = arr;
        p = p + WDS(1);
        goto for;
235 236 237 238
    }
    // Initialise the mark bits with 0
  for2:
    if (p < arr + WDS(size)) {
ian@well-typed.com's avatar
ian@well-typed.com committed
239 240 241
        W_[p] = 0;
        p = p + WDS(1);
        goto for2;
242 243
    }

244
    return (arr);
245 246
}

pumpkin's avatar
pumpkin committed
247

248 249 250 251
/* -----------------------------------------------------------------------------
   MutVar primitives
   -------------------------------------------------------------------------- */

252
stg_newMutVarzh ( gcptr init )
253 254 255
{
    W_ mv;

256
    ALLOC_PRIM_P (SIZEOF_StgMutVar, stg_newMutVarzh, init);
257 258

    mv = Hp - SIZEOF_StgMutVar + WDS(1);
259
    SET_HDR(mv,stg_MUT_VAR_DIRTY_info,CCCS);
260
    StgMutVar_var(mv) = init;
ian@well-typed.com's avatar
ian@well-typed.com committed
261

262
    return (mv);
263 264
}

265
stg_casMutVarzh ( gcptr mv, gcptr old, gcptr new )
Simon Marlow's avatar
Simon Marlow committed
266 267
 /* MutVar# s a -> a -> a -> State# s -> (# State#, Int#, a #) */
{
268
    gcptr h;
Simon Marlow's avatar
Simon Marlow committed
269

270 271
    (h) = ccall cas(mv + SIZEOF_StgHeader + OFFSET_StgMutVar_var,
                          old, new);
Simon Marlow's avatar
Simon Marlow committed
272
    if (h != old) {
273
        return (1,h);
Simon Marlow's avatar
Simon Marlow committed
274
    } else {
275
        if (GET_INFO(mv) == stg_MUT_VAR_CLEAN_info) {
276
           ccall dirty_MUT_VAR(BaseReg "ptr", mv "ptr");
277
        }
278
        return (0,h);
Simon Marlow's avatar
Simon Marlow committed
279 280 281
    }
}

282
stg_atomicModifyMutVarzh ( gcptr mv, gcptr f )
283
{
284
    W_ z, x, y, r, h;
285

ian@well-typed.com's avatar
ian@well-typed.com committed
286
    /* If x is the current contents of the MutVar#, then
287 288 289
       We want to make the new contents point to

         (sel_0 (f x))
ian@well-typed.com's avatar
ian@well-typed.com committed
290

291
       and the return value is
ian@well-typed.com's avatar
ian@well-typed.com committed
292 293

         (sel_1 (f x))
294 295 296 297

        obviously we can share (f x).

         z = [stg_ap_2 f x]  (max (HS + 2) MIN_UPD_SIZE)
ian@well-typed.com's avatar
ian@well-typed.com committed
298
         y = [stg_sel_0 z]   (max (HS + 1) MIN_UPD_SIZE)
299 300 301 302
         r = [stg_sel_1 z]   (max (HS + 1) MIN_UPD_SIZE)
    */

#if MIN_UPD_SIZE > 1
303
#define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
304 305
#define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
#else
306
#define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
307 308 309 310
#define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
#endif

#if MIN_UPD_SIZE > 2
311
#define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
312 313
#define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
#else
314
#define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
315 316 317 318 319
#define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
#endif

#define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)

320
   HP_CHK_GEN_TICKY(SIZE);
321 322 323 324

   TICK_ALLOC_THUNK_2();
   CCCS_ALLOC(THUNK_2_SIZE);
   z = Hp - THUNK_2_SIZE + WDS(1);
325
   SET_HDR(z, stg_ap_2_upd_info, CCCS);
326
   LDV_RECORD_CREATE(z);
327
   StgThunk_payload(z,0) = f;
328 329 330 331

   TICK_ALLOC_THUNK_1();
   CCCS_ALLOC(THUNK_1_SIZE);
   y = z - THUNK_1_SIZE;
332
   SET_HDR(y, stg_sel_0_upd_info, CCCS);
333
   LDV_RECORD_CREATE(y);
334
   StgThunk_payload(y,0) = z;
335 336 337 338

   TICK_ALLOC_THUNK_1();
   CCCS_ALLOC(THUNK_1_SIZE);
   r = y - THUNK_1_SIZE;
339
   SET_HDR(r, stg_sel_1_upd_info, CCCS);
340
   LDV_RECORD_CREATE(r);
341 342
   StgThunk_payload(r,0) = z;

343 344 345 346
 retry:
   x = StgMutVar_var(mv);
   StgThunk_payload(z,1) = x;
#ifdef THREADED_RTS
347
   (h) = ccall cas(mv + SIZEOF_StgHeader + OFFSET_StgMutVar_var, x, y);
348 349 350
   if (h != x) { goto retry; }
#else
   StgMutVar_var(mv) = y;
351
#endif
352

353
   if (GET_INFO(mv) == stg_MUT_VAR_CLEAN_info) {
354
     ccall dirty_MUT_VAR(BaseReg "ptr", mv "ptr");
355 356
   }

357
   return (r);
358 359 360 361 362 363 364 365
}

/* -----------------------------------------------------------------------------
   Weak Pointer Primitives
   -------------------------------------------------------------------------- */

STRING(stg_weak_msg,"New weak pointer at %p\n")

366 367 368
stg_mkWeakzh ( gcptr key,
               gcptr value,
               gcptr finalizer /* or stg_NO_FINALIZER_closure */ )
369
{
370
  gcptr w;
371

372
  ALLOC_PRIM (SIZEOF_StgWeak)
373 374

  w = Hp - SIZEOF_StgWeak + WDS(1);
375
  SET_HDR(w, stg_WEAK_info, CCCS);
376

377 378 379 380
  // We don't care about cfinalizer here.
  // Should StgWeak_cfinalizer(w) be stg_NO_FINALIZER_closure or
  // something else?

381 382 383
  StgWeak_key(w)        = key;
  StgWeak_value(w)      = value;
  StgWeak_finalizer(w)  = finalizer;
384
  StgWeak_cfinalizer(w) = stg_NO_FINALIZER_closure;
385

386
  ACQUIRE_LOCK(sm_mutex);
ian@well-typed.com's avatar
ian@well-typed.com committed
387 388
  StgWeak_link(w)       = W_[weak_ptr_list];
  W_[weak_ptr_list]     = w;
389
  RELEASE_LOCK(sm_mutex);
390

391
  IF_DEBUG(weak, ccall debugBelch(stg_weak_msg,w));
392

393
  return (w);
394 395
}

396
stg_mkWeakNoFinalizzerzh ( gcptr key, gcptr value )
397
{
398
  jump stg_mkWeakzh (key, value, stg_NO_FINALIZER_closure);
399 400
}

401 402 403 404 405 406
stg_mkWeakForeignEnvzh ( gcptr key,
                         gcptr val,
                         W_ fptr,   // finalizer
                         W_ ptr,
                         W_ flag,   // has environment (0 or 1)
                         W_ eptr )
407
{
408 409 410 411
  W_ payload_words, words;
  gcptr w, p;

  ALLOC_PRIM (SIZEOF_StgWeak);
412 413

  w = Hp - SIZEOF_StgWeak + WDS(1);
414
  SET_HDR(w, stg_WEAK_info, CCCS);
415 416 417

  payload_words = 4;
  words         = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
418
  ("ptr" p)     = ccall allocate(MyCapability() "ptr", words);
419 420

  TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
421
  SET_HDR(p, stg_ARR_WORDS_info, CCCS);
422

423
  StgArrWords_bytes(p)     = WDS(payload_words);
424 425 426 427 428 429 430 431 432 433 434 435 436
  StgArrWords_payload(p,0) = fptr;
  StgArrWords_payload(p,1) = ptr;
  StgArrWords_payload(p,2) = eptr;
  StgArrWords_payload(p,3) = flag;

  // We don't care about the value here.
  // Should StgWeak_value(w) be stg_NO_FINALIZER_closure or something else?

  StgWeak_key(w)        = key;
  StgWeak_value(w)      = val;
  StgWeak_finalizer(w)  = stg_NO_FINALIZER_closure;
  StgWeak_cfinalizer(w) = p;

437
  ACQUIRE_LOCK(sm_mutex);
438 439
  StgWeak_link(w)   = W_[weak_ptr_list];
  W_[weak_ptr_list] = w;
440
  RELEASE_LOCK(sm_mutex);
441

442
  IF_DEBUG(weak, ccall debugBelch(stg_weak_msg,w));
443

444
  return (w);
445
}
446

447
stg_finalizzeWeakzh ( gcptr w )
448
{
449
  gcptr f, arr;
450 451 452

  // already dead?
  if (GET_INFO(w) == stg_DEAD_WEAK_info) {
453
      return (0,stg_NO_FINALIZER_closure);
454 455 456 457 458 459 460 461 462 463
  }

  // kill it
#ifdef PROFILING
  // @LDV profiling
  // A weak pointer is inherently used, so we do not need to call
  // LDV_recordDead_FILL_SLOP_DYNAMIC():
  //    LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
  // or, LDV_recordDead():
  //    LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
ian@well-typed.com's avatar
ian@well-typed.com committed
464
  // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
465 466 467 468 469 470 471 472 473 474
  // large as weak pointers, so there is no need to fill the slop, either.
  // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
#endif

  //
  // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
  //
  SET_INFO(w,stg_DEAD_WEAK_info);
  LDV_RECORD_CREATE(w);

475 476 477
  f   = StgWeak_finalizer(w);
  arr = StgWeak_cfinalizer(w);

478
  StgDeadWeak_link(w) = StgWeak_link(w);
479

480
  if (arr != stg_NO_FINALIZER_closure) {
481
    ccall runCFinalizer(StgArrWords_payload(arr,0),
482 483
                              StgArrWords_payload(arr,1),
                              StgArrWords_payload(arr,2),
484
                              StgArrWords_payload(arr,3));
485 486
  }

487 488
  /* return the finalizer */
  if (f == stg_NO_FINALIZER_closure) {
489
      return (0,stg_NO_FINALIZER_closure);
490
  } else {
491
      return (1,f);
492 493 494
  }
}

495
stg_deRefWeakzh ( gcptr w )
496
{
497 498
  W_ code;
  gcptr val;
499 500 501 502 503 504 505 506

  if (GET_INFO(w) == stg_WEAK_info) {
    code = 1;
    val = StgWeak_value(w);
  } else {
    code = 0;
    val = w;
  }
507
  return (code,val);
508 509 510
}

/* -----------------------------------------------------------------------------
511
   Floating point operations.
512 513
   -------------------------------------------------------------------------- */

514
stg_decodeFloatzuIntzh ( F_ arg )
ian@well-typed.com's avatar
ian@well-typed.com committed
515
{
516
    W_ p;
517 518 519
    W_ mp_tmp1;
    W_ mp_tmp_w;

520
    STK_CHK_GEN_N (WDS(2));
521 522 523

    mp_tmp1  = Sp - WDS(1);
    mp_tmp_w = Sp - WDS(2);
ian@well-typed.com's avatar
ian@well-typed.com committed
524

525
    /* Perform the operation */
526
    ccall __decodeFloat_Int(mp_tmp1 "ptr", mp_tmp_w "ptr", arg);
ian@well-typed.com's avatar
ian@well-typed.com committed
527

528
    /* returns: (Int# (mantissa), Int# (exponent)) */
529
    return (W_[mp_tmp1], W_[mp_tmp_w]);
530 531
}

532
stg_decodeDoublezu2Intzh ( D_ arg )
ian@well-typed.com's avatar
ian@well-typed.com committed
533
{
534
    W_ p;
535 536 537 538 539
    W_ mp_tmp1;
    W_ mp_tmp2;
    W_ mp_result1;
    W_ mp_result2;

540
    STK_CHK_GEN_N (WDS(4));
541 542 543 544 545

    mp_tmp1    = Sp - WDS(1);
    mp_tmp2    = Sp - WDS(2);
    mp_result1 = Sp - WDS(3);
    mp_result2 = Sp - WDS(4);
546 547

    /* Perform the operation */
548
    ccall __decodeDouble_2Int(mp_tmp1 "ptr", mp_tmp2 "ptr",
549
                                    mp_result1 "ptr", mp_result2 "ptr",
550
                                    arg);
551 552 553

    /* returns:
       (Int# (mant sign), Word# (mant high), Word# (mant low), Int# (expn)) */
554
    return (W_[mp_tmp1], W_[mp_tmp2], W_[mp_result1], W_[mp_result2]);
555 556
}

557 558 559 560
/* -----------------------------------------------------------------------------
 * Concurrency primitives
 * -------------------------------------------------------------------------- */

561
stg_forkzh ( gcptr closure )
562
{
563
  MAYBE_GC_P(stg_forkzh, closure);
564

565
  gcptr threadid;
566

ian@well-typed.com's avatar
ian@well-typed.com committed
567 568
  ("ptr" threadid) = ccall createIOThread( MyCapability() "ptr",
                                RtsFlags_GcFlags_initialStkSize(RtsFlags),
569
                                closure "ptr");
570 571

  /* start blocked if the current thread is blocked */
572
  StgTSO_flags(threadid) = %lobits16(
ian@well-typed.com's avatar
ian@well-typed.com committed
573
     TO_W_(StgTSO_flags(threadid)) |
574
     TO_W_(StgTSO_flags(CurrentTSO)) & (TSO_BLOCKEX | TSO_INTERRUPTIBLE));
575

576
  ccall scheduleThread(MyCapability() "ptr", threadid "ptr");
577

578 579
  // context switch soon, but not immediately: we don't want every
  // forkIO to force a context-switch.
580
  Capability_context_switch(MyCapability()) = 1 :: CInt;
ian@well-typed.com's avatar
ian@well-typed.com committed
581

582
  return (threadid);
583 584
}

585
stg_forkOnzh ( W_ cpu, gcptr closure )
586
{
587
again: MAYBE_GC(again);
588

589
  gcptr threadid;
590

ian@well-typed.com's avatar
ian@well-typed.com committed
591 592
  ("ptr" threadid) = ccall createIOThread( MyCapability() "ptr",
                                RtsFlags_GcFlags_initialStkSize(RtsFlags),
593
                                closure "ptr");
594 595

  /* start blocked if the current thread is blocked */
596
  StgTSO_flags(threadid) = %lobits16(
ian@well-typed.com's avatar
ian@well-typed.com committed
597
     TO_W_(StgTSO_flags(threadid)) |
598
     TO_W_(StgTSO_flags(CurrentTSO)) & (TSO_BLOCKEX | TSO_INTERRUPTIBLE));
599

600
  ccall scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr");
601

602 603
  // context switch soon, but not immediately: we don't want every
  // forkIO to force a context-switch.
604
  Capability_context_switch(MyCapability()) = 1 :: CInt;
ian@well-typed.com's avatar
ian@well-typed.com committed
605

606
  return (threadid);
607 608
}

609
stg_yieldzh ()
610
{
611 612 613 614 615
  // when we yield to the scheduler, we have to tell it to put the
  // current thread to the back of the queue by setting the
  // context_switch flag.  If we don't do this, it will run the same
  // thread again.
  Capability_context_switch(MyCapability()) = 1 :: CInt;
616
  jump stg_yield_noregs();
617 618
}

619
stg_myThreadIdzh ()
620
{
621
  return (CurrentTSO);
622 623
}

624
stg_labelThreadzh ( gcptr threadid, W_ addr )
625
{
626
#if defined(DEBUG) || defined(TRACING) || defined(DTRACE)
627
  ccall labelThread(MyCapability() "ptr", threadid "ptr", addr "ptr");
628
#endif
629
  return ();
630 631
}

632
stg_isCurrentThreadBoundzh (/* no args */)
633 634
{
  W_ r;
635 636
  (r) = ccall isThreadBound(CurrentTSO);
  return (r);
637 638
}

639
stg_threadStatuszh ( gcptr tso )
640 641 642
{
    W_ why_blocked;
    W_ what_next;
643
    W_ ret, cap, locked;
644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660

    what_next   = TO_W_(StgTSO_what_next(tso));
    why_blocked = TO_W_(StgTSO_why_blocked(tso));
    // Note: these two reads are not atomic, so they might end up
    // being inconsistent.  It doesn't matter, since we
    // only return one or the other.  If we wanted to return the
    // contents of block_info too, then we'd have to do some synchronisation.

    if (what_next == ThreadComplete) {
        ret = 16;  // NB. magic, matches up with GHC.Conc.threadStatus
    } else {
        if (what_next == ThreadKilled) {
            ret = 17;
        } else {
            ret = why_blocked;
        }
    }
661 662 663 664 665 666 667 668 669

    cap = TO_W_(Capability_no(StgTSO_cap(tso)));

    if ((TO_W_(StgTSO_flags(tso)) & TSO_LOCKED) != 0) {
        locked = 1;
    } else {
        locked = 0;
    }

670
    return (ret,cap,locked);
671
}
672 673 674 675 676

/* -----------------------------------------------------------------------------
 * TVar primitives
 * -------------------------------------------------------------------------- */

677 678 679
// Catch retry frame -----------------------------------------------------------

#define CATCH_RETRY_FRAME_FIELDS(w_,p_,info_ptr,        \
Simon Marlow's avatar
Simon Marlow committed
680
                                 p1, p2,                \
681 682 683 684
                                 running_alt_code,      \
                                 first_code,            \
                                 alt_code)              \
  w_ info_ptr,                                          \
Simon Marlow's avatar
Simon Marlow committed
685
  PROF_HDR_FIELDS(w_,p1,p2)                             \
686 687 688
  w_ running_alt_code,                                  \
  p_ first_code,                                        \
  p_ alt_code
689 690


691
INFO_TABLE_RET(stg_catch_retry_frame, CATCH_RETRY_FRAME,
692
               CATCH_RETRY_FRAME_FIELDS(W_,P_,
Simon Marlow's avatar
Simon Marlow committed
693
                                        info_ptr, p1, p2,
694 695 696 697
                                        running_alt_code,
                                        first_code,
                                        alt_code))
    return (P_ ret)
698
{
699 700
    W_ r;
    gcptr trec, outer, arg;
701

702 703 704 705 706 707 708 709 710 711 712 713 714 715 716
    trec = StgTSO_trec(CurrentTSO);
    outer  = StgTRecHeader_enclosing_trec(trec);
    (r) = ccall stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr");
    if (r != 0) {
        // Succeeded (either first branch or second branch)
        StgTSO_trec(CurrentTSO) = outer;
        return (ret);
    } else {
        // Did not commit: re-execute
        P_ new_trec;
        ("ptr" new_trec) = ccall stmStartTransaction(MyCapability() "ptr",
                                                           outer "ptr");
        StgTSO_trec(CurrentTSO) = new_trec;
        if (running_alt_code != 0) {
            jump stg_ap_v_fast
Simon Marlow's avatar
Simon Marlow committed
717
                (CATCH_RETRY_FRAME_FIELDS(,,info_ptr, p1, p2,
718 719 720 721 722 723
                                          running_alt_code,
                                          first_code,
                                          alt_code))
                (alt_code);
        } else {
            jump stg_ap_v_fast
Simon Marlow's avatar
Simon Marlow committed
724
                (CATCH_RETRY_FRAME_FIELDS(,,info_ptr, p1, p2,
725 726 727 728 729 730 731
                                          running_alt_code,
                                          first_code,
                                          alt_code))
                (first_code);
        }
    }
}
732

Simon Marlow's avatar
Simon Marlow committed
733
// Atomically frame ------------------------------------------------------------
734

735
// This must match StgAtomicallyFrame in Closures.h
Simon Marlow's avatar
Simon Marlow committed
736
#define ATOMICALLY_FRAME_FIELDS(w_,p_,info_ptr,p1,p2,code,next,result)  \
737
  w_ info_ptr,                                                          \
Simon Marlow's avatar
Simon Marlow committed
738
  PROF_HDR_FIELDS(w_,p1,p2)                                             \
739 740 741 742 743
  p_ code,                                                              \
  p_ next,                                                              \
  p_ result


744
INFO_TABLE_RET(stg_atomically_frame, ATOMICALLY_FRAME,
745 746
               // layout of the frame, and bind the field names
               ATOMICALLY_FRAME_FIELDS(W_,P_,
Simon Marlow's avatar
Simon Marlow committed
747
                                       info_ptr, p1, p2,
748 749 750 751
                                       code,
                                       next_invariant,
                                       frame_result))
    return (P_ result) // value returned to the frame
752
{
753 754
  W_ valid;
  gcptr trec, outer, next_invariant, q;
755

756
  trec   = StgTSO_trec(CurrentTSO);
757
  outer  = StgTRecHeader_enclosing_trec(trec);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
758 759 760

  if (outer == NO_TREC) {
    /* First time back at the atomically frame -- pick up invariants */
761 762 763
    ("ptr" next_invariant) =
        ccall stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr");
    frame_result = result;
764 765

  } else {
tharris@microsoft.com's avatar
tharris@microsoft.com committed
766 767 768
    /* Second/subsequent time back at the atomically frame -- abort the
     * tx that's checking the invariant and move on to the next one */
    StgTSO_trec(CurrentTSO) = outer;
769 770
    StgInvariantCheckQueue_my_execution(next_invariant) = trec;
    ccall stmAbortTransaction(MyCapability() "ptr", trec "ptr");
tharris@microsoft.com's avatar
tharris@microsoft.com committed
771 772
    /* Don't free trec -- it's linked from q and will be stashed in the
     * invariant if we eventually commit. */
773 774
    next_invariant =
       StgInvariantCheckQueue_next_queue_entry(next_invariant);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
775 776 777
    trec = outer;
  }

778
  if (next_invariant != END_INVARIANT_CHECK_QUEUE) {
tharris@microsoft.com's avatar
tharris@microsoft.com committed
779
    /* We can't commit yet: another invariant to check */
780
    ("ptr" trec) = ccall stmStartTransaction(MyCapability() "ptr", trec "ptr");
781
    StgTSO_trec(CurrentTSO) = trec;
782 783
    q = StgInvariantCheckQueue_invariant(next_invariant);
    jump stg_ap_v_fast
Simon Marlow's avatar
Simon Marlow committed
784 785
        (ATOMICALLY_FRAME_FIELDS(,,info_ptr,p1,p2,
                                 code,next_invariant,frame_result))
786
        (StgAtomicInvariant_code(q));
tharris@microsoft.com's avatar
tharris@microsoft.com committed
787 788 789 790

  } else {

    /* We've got no more invariants to check, try to commit */
791
    (valid) = ccall stmCommitTransaction(MyCapability() "ptr", trec "ptr");
tharris@microsoft.com's avatar
tharris@microsoft.com committed
792 793 794
    if (valid != 0) {
      /* Transaction was valid: commit succeeded */
      StgTSO_trec(CurrentTSO) = NO_TREC;
795
      return (frame_result);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
796 797
    } else {
      /* Transaction was not valid: try again */
798
      ("ptr" trec) = ccall stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr");
tharris@microsoft.com's avatar
tharris@microsoft.com committed
799
      StgTSO_trec(CurrentTSO) = trec;
800 801 802 803 804
      next_invariant = END_INVARIANT_CHECK_QUEUE;

      jump stg_ap_v_fast
          // push the StgAtomicallyFrame again: the code generator is
          // clever enough to only assign the fields that have changed.
Simon Marlow's avatar
Simon Marlow committed
805 806
          (ATOMICALLY_FRAME_FIELDS(,,info_ptr,p1,p2,
                                   code,next_invariant,frame_result))
807
          (code);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
808
    }
809 810 811
  }
}

812

813
INFO_TABLE_RET(stg_atomically_waiting_frame, ATOMICALLY_FRAME,
814 815
               // layout of the frame, and bind the field names
               ATOMICALLY_FRAME_FIELDS(W_,P_,
Simon Marlow's avatar
Simon Marlow committed
816
                                       info_ptr, p1, p2,
817 818 819 820
                                       code,
                                       next_invariant,
                                       frame_result))
    return (/* no return values */)
821
{
822
  W_ trec, valid;
823 824

  /* The TSO is currently waiting: should we stop waiting? */
825
  (valid) = ccall stmReWait(MyCapability() "ptr", CurrentTSO "ptr");
Simon Marlow's avatar
Simon Marlow committed
826
  if (valid != 0) {
827 828
      /* Previous attempt is still valid: no point trying again yet */
      jump stg_block_noregs
Simon Marlow's avatar
Simon Marlow committed
829
          (ATOMICALLY_FRAME_FIELDS(,,info_ptr, p1, p2,
830 831
                                   code,next_invariant,frame_result))
          ();
832 833
  } else {
    /* Previous attempt is no longer valid: try again */
834
    ("ptr" trec) = ccall stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr");
835
    StgTSO_trec(CurrentTSO) = trec;
836 837 838

    // change the frame header to stg_atomically_frame_info
    jump stg_ap_v_fast
Simon Marlow's avatar
Simon Marlow committed
839
        (ATOMICALLY_FRAME_FIELDS(,,stg_atomically_frame_info, p1, p2,
840 841
                                 code,next_invariant,frame_result))
        (code);
842
  }
843
}
844

845
// STM catch frame -------------------------------------------------------------
846 847 848 849 850 851

/* Catch frames are very similar to update frames, but when entering
 * one we just pop the frame off the stack and perform the correct
 * kind of return to the activation record underneath us on the stack.
 */

Simon Marlow's avatar
Simon Marlow committed
852
#define CATCH_STM_FRAME_FIELDS(w_,p_,info_ptr,p1,p2,code,handler) \
853
  w_ info_ptr,                                                  \
Simon Marlow's avatar
Simon Marlow committed
854
  PROF_HDR_FIELDS(w_,p1,p2)                                     \
855 856 857
  p_ code,                                                      \
  p_ handler

858
INFO_TABLE_RET(stg_catch_stm_frame, CATCH_STM_FRAME,
859
               // layout of the frame, and bind the field names
Simon Marlow's avatar
Simon Marlow committed
860
               CATCH_STM_FRAME_FIELDS(W_,P_,info_ptr,p1,p2,code,handler))
861 862 863 864 865 866 867 868
    return (P_ ret)
{
    W_ r, trec, outer;

    trec = StgTSO_trec(CurrentTSO);
    outer  = StgTRecHeader_enclosing_trec(trec);
    (r) = ccall stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr");
    if (r != 0) {
Simon Marlow's avatar
Simon Marlow committed
869 870
        /* Commit succeeded */
        StgTSO_trec(CurrentTSO) = outer;
871 872
        return (ret);
    } else {