PrimOps.cmm 52.1 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
/* -----------------------------------------------------------------------------
 *
 * (c) The GHC Team, 1998-2004
 *
 * 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.
 *
 * Entry convention: the entry convention for a primop is that all the
 * args are in Stg registers (R1, R2, etc.).  This is to make writing
 * the primops easier.  (see compiler/codeGen/CgCallConv.hs).
 *
 * Return convention: results from a primop are generally returned
 * using the ordinary unboxed tuple return convention.  The C-- parser
 * implements the RET_xxxx() macros to perform unboxed-tuple returns
 * based on the prevailing return convention.
 *
 * 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"

30
#ifdef __PIC__
31 32
import pthread_mutex_lock;
import pthread_mutex_unlock;
33
#endif
34
import base_ControlziExceptionziBase_nestedAtomically_closure;
35 36
import EnterCriticalSection;
import LeaveCriticalSection;
37
import ghczmprim_GHCziBool_False_closure;
38

39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
/*-----------------------------------------------------------------------------
  Array Primitives

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

  The size arg is always passed in R1, and the result returned in R1.

  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.
 */

newByteArrayzh_fast
{
    W_ words, payload_words, n, p;
    MAYBE_GC(NO_PTRS,newByteArrayzh_fast);
    n = R1;
    payload_words = ROUNDUP_BYTES_TO_WDS(n);
    words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
61
    ("ptr" p) = foreign "C" allocateLocal(MyCapability() "ptr",words) [];
62 63 64 65 66 67
    TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
    SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
    StgArrWords_words(p) = payload_words;
    RET_P(p);
}

Simon Marlow's avatar
Simon Marlow committed
68 69 70
#define BA_ALIGN 16
#define BA_MASK  (BA_ALIGN-1)

71 72
newPinnedByteArrayzh_fast
{
73
    W_ words, bytes, payload_words, p;
74 75

    MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast);
76 77 78 79 80 81 82 83 84 85 86
    bytes = R1;
    /* 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
87 88 89 90

    ("ptr" p) = foreign "C" allocatePinned(words) [];
    TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);

91 92
    /* Now we need to move p forward so that the payload is aligned
       to BA_ALIGN bytes: */
Simon Marlow's avatar
Simon Marlow committed
93 94 95 96 97 98 99 100 101
    p = p + ((-p - SIZEOF_StgArrWords) & BA_MASK);

    SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
    StgArrWords_words(p) = payload_words;
    RET_P(p);
}

newAlignedPinnedByteArrayzh_fast
{
102
    W_ words, bytes, payload_words, p, alignment;
Simon Marlow's avatar
Simon Marlow committed
103 104

    MAYBE_GC(NO_PTRS,newAlignedPinnedByteArrayzh_fast);
105 106
    bytes = R1;
    alignment = R2;
Simon Marlow's avatar
Simon Marlow committed
107

108 109
    /* 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
110

111 112 113 114 115 116 117 118
    /* 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);
119

120
    ("ptr" p) = foreign "C" allocatePinned(words) [];
121 122
    TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);

123 124 125 126
    /* 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));
127 128 129 130 131 132 133 134 135 136 137 138 139 140 141

    SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
    StgArrWords_words(p) = payload_words;
    RET_P(p);
}

newArrayzh_fast
{
    W_ words, n, init, arr, p;
    /* Args: R1 = words, R2 = initialisation value */

    n = R1;
    MAYBE_GC(R2_PTR,newArrayzh_fast);

    words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
142
    ("ptr" arr) = foreign "C" allocateLocal(MyCapability() "ptr",words) [R2];
143 144
    TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);

145
    SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, W_[CCCS]);
146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169
    StgMutArrPtrs_ptrs(arr) = n;

    // Initialise all elements of the the array with the value in R2
    init = R2;
    p = arr + SIZEOF_StgMutArrPtrs;
  for:
    if (p < arr + WDS(words)) {
	W_[p] = init;
	p = p + WDS(1);
	goto for;
    }

    RET_P(arr);
}

unsafeThawArrayzh_fast
{
  // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
  //
  // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN 
  // 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
  // the mutable list is not easy, because the mut_list is only singly-linked).
  // 
170
  // So that we can tell whether a MUT_ARR_PTRS_FROZEN is on the mutable list,
171 172 173
  // 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.
  //
174 175
  // 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
176
  // the closure type is MUT_ARR_PTRS_FROZEN0 if it is on the mutable list,
177 178 179
  // 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.
180
  //
181
  if (StgHeader_info(R1) != stg_MUT_ARR_PTRS_FROZEN0_info) {
182
	SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
183
	recordMutable(R1, R1);
184 185 186
	// must be done after SET_INFO, because it ASSERTs closure_MUTABLE()
	RET_P(R1);
  } else {
187
	SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
188
	RET_P(R1);
189 190 191 192 193 194 195 196 197 198 199 200 201 202 203
  }
}

/* -----------------------------------------------------------------------------
   MutVar primitives
   -------------------------------------------------------------------------- */

newMutVarzh_fast
{
    W_ mv;
    /* Args: R1 = initialisation value */

    ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);

    mv = Hp - SIZEOF_StgMutVar + WDS(1);
204
    SET_HDR(mv,stg_MUT_VAR_DIRTY_info,W_[CCCS]);
205 206 207 208 209 210 211
    StgMutVar_var(mv) = R1;
    
    RET_P(mv);
}

atomicModifyMutVarzh_fast
{
212
    W_ mv, f, z, x, y, r, h;
213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231
    /* Args: R1 :: MutVar#,  R2 :: a -> (a,b) */

    /* If x is the current contents of the MutVar#, then 
       We want to make the new contents point to

         (sel_0 (f x))
 
       and the return value is
	 
	 (sel_1 (f x))

        obviously we can share (f x).

         z = [stg_ap_2 f x]  (max (HS + 2) MIN_UPD_SIZE)
	 y = [stg_sel_0 z]   (max (HS + 1) MIN_UPD_SIZE)
         r = [stg_sel_1 z]   (max (HS + 1) MIN_UPD_SIZE)
    */

#if MIN_UPD_SIZE > 1
232
#define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
233 234
#define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
#else
235
#define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
236 237 238 239
#define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
#endif

#if MIN_UPD_SIZE > 2
240
#define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
241 242
#define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
#else
243
#define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
244 245 246 247 248 249 250
#define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
#endif

#define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)

   HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);

251 252
   mv = R1;
   f = R2;
253 254 255 256 257 258

   TICK_ALLOC_THUNK_2();
   CCCS_ALLOC(THUNK_2_SIZE);
   z = Hp - THUNK_2_SIZE + WDS(1);
   SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
   LDV_RECORD_CREATE(z);
259
   StgThunk_payload(z,0) = f;
260 261 262 263 264 265

   TICK_ALLOC_THUNK_1();
   CCCS_ALLOC(THUNK_1_SIZE);
   y = z - THUNK_1_SIZE;
   SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
   LDV_RECORD_CREATE(y);
266
   StgThunk_payload(y,0) = z;
267 268 269 270 271 272

   TICK_ALLOC_THUNK_1();
   CCCS_ALLOC(THUNK_1_SIZE);
   r = y - THUNK_1_SIZE;
   SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
   LDV_RECORD_CREATE(r);
273 274
   StgThunk_payload(r,0) = z;

275 276 277 278 279 280 281 282
 retry:
   x = StgMutVar_var(mv);
   StgThunk_payload(z,1) = x;
#ifdef THREADED_RTS
   (h) = foreign "C" cas(mv + SIZEOF_StgHeader + OFFSET_StgMutVar_var, x, y) [];
   if (h != x) { goto retry; }
#else
   StgMutVar_var(mv) = y;
283
#endif
284

285 286 287 288
   if (GET_INFO(mv) == stg_MUT_VAR_CLEAN_info) {
     foreign "C" dirty_MUT_VAR(BaseReg "ptr", mv "ptr") [];
   }

289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314
   RET_P(r);
}

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

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

mkWeakzh_fast
{
  /* R1 = key
     R2 = value
     R3 = finalizer (or NULL)
  */
  W_ w;

  if (R3 == NULL) {
    R3 = stg_NO_FINALIZER_closure;
  }

  ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );

  w = Hp - SIZEOF_StgWeak + WDS(1);
  SET_HDR(w, stg_WEAK_info, W_[CCCS]);

315 316 317 318 319 320 321 322
  // We don't care about cfinalizer here.
  // Should StgWeak_cfinalizer(w) be stg_NO_FINALIZER_closure or
  // something else?

  StgWeak_key(w)        = R1;
  StgWeak_value(w)      = R2;
  StgWeak_finalizer(w)  = R3;
  StgWeak_cfinalizer(w) = stg_NO_FINALIZER_closure;
323 324 325 326

  StgWeak_link(w)	= W_[weak_ptr_list];
  W_[weak_ptr_list] 	= w;

327
  IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
328 329 330 331

  RET_P(w);
}

332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351
mkWeakForeignEnvzh_fast
{
  /* R1 = key
     R2 = value
     R3 = finalizer
     R4 = pointer
     R5 = has environment (0 or 1)
     R6 = environment
  */
  W_ w, payload_words, words, p;

  W_ key, val, fptr, ptr, flag, eptr;

  key  = R1;
  val  = R2;
  fptr = R3;
  ptr  = R4;
  flag = R5;
  eptr = R6;

352
  ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR, mkWeakForeignEnvzh_fast );
353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384

  w = Hp - SIZEOF_StgWeak + WDS(1);
  SET_HDR(w, stg_WEAK_info, W_[CCCS]);

  payload_words = 4;
  words         = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
  ("ptr" p)     = foreign "C" allocateLocal(MyCapability() "ptr", words) [];

  TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
  SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);

  StgArrWords_words(p)     = payload_words;
  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;

  StgWeak_link(w)   = W_[weak_ptr_list];
  W_[weak_ptr_list] = w;

  IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);

  RET_P(w);
}
385 386 387 388 389

finalizzeWeakzh_fast
{
  /* R1 = weak ptr
   */
390
  W_ w, f, arr;
391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417

  w = R1;

  // already dead?
  if (GET_INFO(w) == stg_DEAD_WEAK_info) {
      RET_NP(0,stg_NO_FINALIZER_closure);
  }

  // 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));
  // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as 
  // 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);

418 419 420
  f   = StgWeak_finalizer(w);
  arr = StgWeak_cfinalizer(w);

421
  StgDeadWeak_link(w) = StgWeak_link(w);
422

423 424 425 426 427 428 429
  if (arr != stg_NO_FINALIZER_closure) {
    foreign "C" runCFinalizer(StgArrWords_payload(arr,0),
                              StgArrWords_payload(arr,1),
                              StgArrWords_payload(arr,2),
                              StgArrWords_payload(arr,3)) [];
  }

430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454
  /* return the finalizer */
  if (f == stg_NO_FINALIZER_closure) {
      RET_NP(0,stg_NO_FINALIZER_closure);
  } else {
      RET_NP(1,f);
  }
}

deRefWeakzh_fast
{
  /* R1 = weak ptr */
  W_ w, code, val;

  w = R1;
  if (GET_INFO(w) == stg_WEAK_info) {
    code = 1;
    val = StgWeak_value(w);
  } else {
    code = 0;
    val = w;
  }
  RET_NP(code,val);
}

/* -----------------------------------------------------------------------------
455
   Floating point operations.
456 457
   -------------------------------------------------------------------------- */

458 459 460 461
decodeFloatzuIntzh_fast
{ 
    W_ p;
    F_ arg;
462 463 464 465 466 467 468
    W_ mp_tmp1;
    W_ mp_tmp_w;

    STK_CHK_GEN( WDS(2), NO_PTRS, decodeFloatzuIntzh_fast );

    mp_tmp1  = Sp - WDS(1);
    mp_tmp_w = Sp - WDS(2);
469 470 471 472 473 474 475 476 477 478 479 480 481 482 483
    
    /* arguments: F1 = Float# */
    arg = F1;
    
    /* Perform the operation */
    foreign "C" __decodeFloat_Int(mp_tmp1 "ptr", mp_tmp_w "ptr", arg) [];
    
    /* returns: (Int# (mantissa), Int# (exponent)) */
    RET_NN(W_[mp_tmp1], W_[mp_tmp_w]);
}

decodeDoublezu2Intzh_fast
{ 
    D_ arg;
    W_ p;
484 485 486 487 488 489 490 491 492 493 494
    W_ mp_tmp1;
    W_ mp_tmp2;
    W_ mp_result1;
    W_ mp_result2;

    STK_CHK_GEN( WDS(4), NO_PTRS, decodeDoublezu2Intzh_fast );

    mp_tmp1    = Sp - WDS(1);
    mp_tmp2    = Sp - WDS(2);
    mp_result1 = Sp - WDS(3);
    mp_result2 = Sp - WDS(4);
495 496 497 498 499

    /* arguments: D1 = Double# */
    arg = D1;

    /* Perform the operation */
500 501 502 503 504 505 506
    foreign "C" __decodeDouble_2Int(mp_tmp1 "ptr", mp_tmp2 "ptr",
                                    mp_result1 "ptr", mp_result2 "ptr",
                                    arg) [];

    /* returns:
       (Int# (mant sign), Word# (mant high), Word# (mant low), Int# (expn)) */
    RET_NNNN(W_[mp_tmp1], W_[mp_tmp2], W_[mp_result1], W_[mp_result2]);
507 508
}

509 510 511 512 513 514 515
/* -----------------------------------------------------------------------------
 * Concurrency primitives
 * -------------------------------------------------------------------------- */

forkzh_fast
{
  /* args: R1 = closure to spark */
516

517 518
  MAYBE_GC(R1_PTR, forkzh_fast);

519 520 521 522
  W_ closure;
  W_ threadid;
  closure = R1;

523
  ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr", 
524 525
				RtsFlags_GcFlags_initialStkSize(RtsFlags), 
				closure "ptr") [];
526 527 528 529 530 531

  /* start blocked if the current thread is blocked */
  StgTSO_flags(threadid) = 
     StgTSO_flags(threadid) |  (StgTSO_flags(CurrentTSO) & 
                                (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));

532 533
  foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") [];

534 535
  // context switch soon, but not immediately: we don't want every
  // forkIO to force a context-switch.
536
  Capability_context_switch(MyCapability()) = 1 :: CInt;
537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552
  
  RET_P(threadid);
}

forkOnzh_fast
{
  /* args: R1 = cpu, R2 = closure to spark */

  MAYBE_GC(R2_PTR, forkOnzh_fast);

  W_ cpu;
  W_ closure;
  W_ threadid;
  cpu = R1;
  closure = R2;

553
  ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr", 
554
				RtsFlags_GcFlags_initialStkSize(RtsFlags), 
555
				closure "ptr") [];
556 557 558 559 560 561

  /* start blocked if the current thread is blocked */
  StgTSO_flags(threadid) = 
     StgTSO_flags(threadid) |  (StgTSO_flags(CurrentTSO) & 
                                (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));

562
  foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") [];
563

564 565
  // context switch soon, but not immediately: we don't want every
  // forkIO to force a context-switch.
566
  Capability_context_switch(MyCapability()) = 1 :: CInt;
567
  
568
  RET_P(threadid);
569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587
}

yieldzh_fast
{
  jump stg_yield_noregs;
}

myThreadIdzh_fast
{
  /* no args. */
  RET_P(CurrentTSO);
}

labelThreadzh_fast
{
  /* args: 
	R1 = ThreadId#
	R2 = Addr# */
#ifdef DEBUG
588
  foreign "C" labelThread(R1 "ptr", R2 "ptr") [];
589 590 591 592 593 594 595 596
#endif
  jump %ENTRY_CODE(Sp(0));
}

isCurrentThreadBoundzh_fast
{
  /* no args */
  W_ r;
597
  (r) = foreign "C" isThreadBound(CurrentTSO) [];
598 599 600
  RET_N(r);
}

601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633
threadStatuszh_fast
{
    /* args: R1 :: ThreadId# */
    W_ tso;
    W_ why_blocked;
    W_ what_next;
    W_ ret;

    tso = R1;
    loop:
      if (TO_W_(StgTSO_what_next(tso)) == ThreadRelocated) {
          tso = StgTSO__link(tso);
          goto loop;
      }

    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;
        }
    }
    RET_N(ret);
}
634 635 636 637 638 639 640 641 642

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

#define SP_OFF 0

// Catch retry frame ------------------------------------------------------------

643
INFO_TABLE_RET(stg_catch_retry_frame, CATCH_RETRY_FRAME,
644
#if defined(PROFILING)
645
  W_ unused1, W_ unused2,
646
#endif
647
  W_ unused3, P_ unused4, P_ unused5)
648 649 650 651 652
{
   W_ r, frame, trec, outer;

   frame = Sp;
   trec = StgTSO_trec(CurrentTSO);
653 654
   ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
   (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
655
   if (r != 0) {
656 657 658 659 660
     /* Succeeded (either first branch or second branch) */
     StgTSO_trec(CurrentTSO) = outer;
     Sp = Sp + SIZEOF_StgCatchRetryFrame;
     jump %ENTRY_CODE(Sp(SP_OFF));
   } else {
tharris@microsoft.com's avatar
tharris@microsoft.com committed
661
     /* Did not commit: re-execute */
662
     W_ new_trec;
663
     ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
664
     StgTSO_trec(CurrentTSO) = new_trec;
Simon Marlow's avatar
Simon Marlow committed
665
     if (StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
666 667 668 669
       R1 = StgCatchRetryFrame_alt_code(frame);
     } else {
       R1 = StgCatchRetryFrame_first_code(frame);
     }
670
     jump stg_ap_v_fast;
671 672
   }
}
673 674


Simon Marlow's avatar
Simon Marlow committed
675
// Atomically frame ------------------------------------------------------------
676

677
INFO_TABLE_RET(stg_atomically_frame, ATOMICALLY_FRAME,
678
#if defined(PROFILING)
679
  W_ unused1, W_ unused2,
680
#endif
681
  P_ unused3, P_ unused4)
682
{
tharris@microsoft.com's avatar
tharris@microsoft.com committed
683
  W_ frame, trec, valid, next_invariant, q, outer;
684

685 686
  frame = Sp;
  trec = StgTSO_trec(CurrentTSO);
687
  ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
tharris@microsoft.com's avatar
tharris@microsoft.com committed
688 689 690

  if (outer == NO_TREC) {
    /* First time back at the atomically frame -- pick up invariants */
691
    ("ptr" q) = foreign "C" stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr") [];
tharris@microsoft.com's avatar
tharris@microsoft.com committed
692
    StgAtomicallyFrame_next_invariant_to_check(frame) = q;
693 694

  } else {
tharris@microsoft.com's avatar
tharris@microsoft.com committed
695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711
    /* 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;
    q = StgAtomicallyFrame_next_invariant_to_check(frame);
    StgInvariantCheckQueue_my_execution(q) = trec;
    foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
    /* Don't free trec -- it's linked from q and will be stashed in the
     * invariant if we eventually commit. */
    q = StgInvariantCheckQueue_next_queue_entry(q);
    StgAtomicallyFrame_next_invariant_to_check(frame) = q;
    trec = outer;
  }

  q = StgAtomicallyFrame_next_invariant_to_check(frame);

  if (q != END_INVARIANT_CHECK_QUEUE) {
    /* We can't commit yet: another invariant to check */
712
    ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [];
713
    StgTSO_trec(CurrentTSO) = trec;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
714 715 716

    next_invariant = StgInvariantCheckQueue_invariant(q);
    R1 = StgAtomicInvariant_code(next_invariant);
717
    jump stg_ap_v_fast;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
718 719 720 721

  } else {

    /* We've got no more invariants to check, try to commit */
722
    (valid) = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") [];
tharris@microsoft.com's avatar
tharris@microsoft.com committed
723 724 725 726 727 728 729
    if (valid != 0) {
      /* Transaction was valid: commit succeeded */
      StgTSO_trec(CurrentTSO) = NO_TREC;
      Sp = Sp + SIZEOF_StgAtomicallyFrame;
      jump %ENTRY_CODE(Sp(SP_OFF));
    } else {
      /* Transaction was not valid: try again */
730
      ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
tharris@microsoft.com's avatar
tharris@microsoft.com committed
731 732 733 734 735
      StgTSO_trec(CurrentTSO) = trec;
      StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
      R1 = StgAtomicallyFrame_code(frame);
      jump stg_ap_v_fast;
    }
736 737 738
  }
}

739 740 741 742
INFO_TABLE_RET(stg_atomically_waiting_frame, ATOMICALLY_FRAME,
#if defined(PROFILING)
  W_ unused1, W_ unused2,
#endif
743
  P_ unused3, P_ unused4)
744 745 746 747 748 749
{
  W_ frame, trec, valid;

  frame = Sp;

  /* The TSO is currently waiting: should we stop waiting? */
750
  (valid) = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
751
  if (valid != 0) {
752 753 754 755
    /* Previous attempt is still valid: no point trying again yet */
    jump stg_block_noregs;
  } else {
    /* Previous attempt is no longer valid: try again */
756
    ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
757 758 759
    StgTSO_trec(CurrentTSO) = trec;
    StgHeader_info(frame) = stg_atomically_frame_info;
    R1 = StgAtomicallyFrame_code(frame);
760
    jump stg_ap_v_fast;
761
  }
762
}
763 764 765 766 767 768 769 770 771 772

// STM catch frame --------------------------------------------------------------

#define SP_OFF 0

/* 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.
 */

773 774 775 776
INFO_TABLE_RET(stg_catch_stm_frame, CATCH_STM_FRAME,
#if defined(PROFILING)
  W_ unused1, W_ unused2,
#endif
777
  P_ unused3, P_ unused4)
Simon Marlow's avatar
Simon Marlow committed
778 779 780 781
   {
      W_ r, frame, trec, outer;
      frame = Sp;
      trec = StgTSO_trec(CurrentTSO);
782 783
      ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
      (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
784 785 786 787 788 789 790 791
      if (r != 0) {
        /* Commit succeeded */
        StgTSO_trec(CurrentTSO) = outer;
        Sp = Sp + SIZEOF_StgCatchSTMFrame;
        jump Sp(SP_OFF);
      } else {
        /* Commit failed */
        W_ new_trec;
792
        ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
793 794 795 796 797
        StgTSO_trec(CurrentTSO) = new_trec;
        R1 = StgCatchSTMFrame_code(frame);
        jump stg_ap_v_fast;
      }
   }
798 799 800 801 802 803 804 805 806 807


// Primop definition ------------------------------------------------------------

atomicallyzh_fast
{
  W_ frame;
  W_ old_trec;
  W_ new_trec;
  
808 809 810
  // stmStartTransaction may allocate
  MAYBE_GC (R1_PTR, atomicallyzh_fast); 

811 812 813
  /* Args: R1 = m :: STM a */
  STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);

814 815 816 817
  old_trec = StgTSO_trec(CurrentTSO);

  /* Nested transactions are not allowed; raise an exception */
  if (old_trec != NO_TREC) {
818
     R1 = base_ControlziExceptionziBase_nestedAtomically_closure;
819 820 821
     jump raisezh_fast;
  }

822 823 824 825
  /* Set up the atomically frame */
  Sp = Sp - SIZEOF_StgAtomicallyFrame;
  frame = Sp;

826
  SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
827
  StgAtomicallyFrame_code(frame) = R1;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
828
  StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
829 830

  /* Start the memory transcation */
831
  ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1];
832 833 834
  StgTSO_trec(CurrentTSO) = new_trec;

  /* Apply R1 to the realworld token */
835
  jump stg_ap_v_fast;
836 837 838 839 840 841 842 843 844 845 846 847 848 849 850
}


catchSTMzh_fast
{
  W_ frame;
  
  /* Args: R1 :: STM a */
  /* Args: R2 :: Exception -> STM a */
  STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast);

  /* Set up the catch frame */
  Sp = Sp - SIZEOF_StgCatchSTMFrame;
  frame = Sp;

851
  SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
852
  StgCatchSTMFrame_handler(frame) = R2;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
853 854 855 856 857 858
  StgCatchSTMFrame_code(frame) = R1;

  /* Start a nested transaction to run the body of the try block in */
  W_ cur_trec;  
  W_ new_trec;
  cur_trec = StgTSO_trec(CurrentTSO);
859
  ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", cur_trec "ptr");
tharris@microsoft.com's avatar
tharris@microsoft.com committed
860
  StgTSO_trec(CurrentTSO) = new_trec;
861 862

  /* Apply R1 to the realworld token */
863
  jump stg_ap_v_fast;
864 865 866 867 868 869 870 871 872
}


catchRetryzh_fast
{
  W_ frame;
  W_ new_trec;
  W_ trec;

873 874 875
  // stmStartTransaction may allocate
  MAYBE_GC (R1_PTR & R2_PTR, catchRetryzh_fast); 

876 877 878 879 880 881
  /* Args: R1 :: STM a */
  /* Args: R2 :: STM a */
  STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);

  /* Start a nested transaction within which to run the first code */
  trec = StgTSO_trec(CurrentTSO);
882
  ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2];
883 884 885 886 887 888
  StgTSO_trec(CurrentTSO) = new_trec;

  /* Set up the catch-retry frame */
  Sp = Sp - SIZEOF_StgCatchRetryFrame;
  frame = Sp;
  
889
  SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
890
  StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
891 892 893 894
  StgCatchRetryFrame_first_code(frame) = R1;
  StgCatchRetryFrame_alt_code(frame) = R2;

  /* Apply R1 to the realworld token */
895
  jump stg_ap_v_fast;
896 897 898 899 900 901 902 903 904 905 906
}


retryzh_fast
{
  W_ frame_type;
  W_ frame;
  W_ trec;
  W_ outer;
  W_ r;

907
  MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate
908 909 910 911

  // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
retry_pop_stack:
  StgTSO_sp(CurrentTSO) = Sp;
912
  (frame_type) = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") [];
913 914
  Sp = StgTSO_sp(CurrentTSO);
  frame = Sp;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
915
  trec = StgTSO_trec(CurrentTSO);
916
  ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
917 918 919 920

  if (frame_type == CATCH_RETRY_FRAME) {
    // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
    ASSERT(outer != NO_TREC);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
921 922 923
    // Abort the transaction attempting the current branch
    foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
    foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
924
    if (!StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
tharris@microsoft.com's avatar
tharris@microsoft.com committed
925
      // Retry in the first branch: try the alternative
926
      ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
927
      StgTSO_trec(CurrentTSO) = trec;
928
      StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
929
      R1 = StgCatchRetryFrame_alt_code(frame);
930
      jump stg_ap_v_fast;
931
    } else {
tharris@microsoft.com's avatar
tharris@microsoft.com committed
932 933 934 935
      // Retry in the alternative code: propagate the retry
      StgTSO_trec(CurrentTSO) = outer;
      Sp = Sp + SIZEOF_StgCatchRetryFrame;
      goto retry_pop_stack;
936 937 938 939 940
    }
  }

  // We've reached the ATOMICALLY_FRAME: attempt to wait 
  ASSERT(frame_type == ATOMICALLY_FRAME);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
941 942 943 944 945 946 947
  if (outer != NO_TREC) {
    // We called retry while checking invariants, so abort the current
    // invariant check (merging its TVar accesses into the parents read
    // set so we'll wait on them)
    foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
    foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
    trec = outer;
948 949
    StgTSO_trec(CurrentTSO) = trec;
    ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
tharris@microsoft.com's avatar
tharris@microsoft.com committed
950
  }
951
  ASSERT(outer == NO_TREC);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
952

953
  (r) = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
954
  if (r != 0) {
955
    // Transaction was valid: stmWait put us on the TVars' queues, we now block
956
    StgHeader_info(frame) = stg_atomically_waiting_frame_info;
957
    Sp = frame;
958
    // Fix up the stack in the unregisterised case: the return convention is different.
959 960
    R3 = trec; // passing to stmWaitUnblock()
    jump stg_block_stmwait;
961 962
  } else {
    // Transaction was not valid: retry immediately
963
    ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
964 965 966
    StgTSO_trec(CurrentTSO) = trec;
    R1 = StgAtomicallyFrame_code(frame);
    Sp = frame;
967
    jump stg_ap_v_fast;
968 969 970 971
  }
}


tharris@microsoft.com's avatar
tharris@microsoft.com committed
972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988
checkzh_fast
{
  W_ trec, closure;

  /* Args: R1 = invariant closure */
  MAYBE_GC (R1_PTR, checkzh_fast); 

  trec = StgTSO_trec(CurrentTSO);
  closure = R1;
  foreign "C" stmAddInvariantToCheck(MyCapability() "ptr", 
                                     trec "ptr",
                                     closure "ptr") [];

  jump %ENTRY_CODE(Sp(0));
}


989 990 991
newTVarzh_fast
{
  W_ tv;
992
  W_ new_value;
993 994 995

  /* Args: R1 = initialisation value */

996 997
  MAYBE_GC (R1_PTR, newTVarzh_fast); 
  new_value = R1;
998
  ("ptr" tv) = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") [];
999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013
  RET_P(tv);
}


readTVarzh_fast
{
  W_ trec;
  W_ tvar;
  W_ result;

  /* Args: R1 = TVar closure */

  MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
  trec = StgTSO_trec(CurrentTSO);
  tvar = R1;
1014
  ("ptr" result) = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") [];
1015 1016 1017 1018

  RET_P(result);
}

1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029
readTVarIOzh_fast
{
    W_ result;

again:
    result = StgTVar_current_value(R1);
    if (%INFO_PTR(result) == stg_TREC_HEADER_info) {
        goto again;
    }
    RET_P(result);
}
1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043

writeTVarzh_fast
{
  W_ trec;
  W_ tvar;
  W_ new_value;
  
  /* Args: R1 = TVar closure */
  /*       R2 = New value    */

  MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
  trec = StgTSO_trec(CurrentTSO);
  tvar = R1;
  new_value = R2;
1044
  foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") [];
1045 1046 1047 1048 1049

  jump %ENTRY_CODE(Sp(0));
}


1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085
/* -----------------------------------------------------------------------------
 * MVar primitives
 *
 * take & putMVar work as follows.  Firstly, an important invariant:
 *
 *    If the MVar is full, then the blocking queue contains only
 *    threads blocked on putMVar, and if the MVar is empty then the
 *    blocking queue contains only threads blocked on takeMVar.
 *
 * takeMvar:
 *    MVar empty : then add ourselves to the blocking queue
 *    MVar full  : remove the value from the MVar, and
 *                 blocking queue empty     : return
 *                 blocking queue non-empty : perform the first blocked putMVar
 *                                            from the queue, and wake up the
 *                                            thread (MVar is now full again)
 *
 * putMVar is just the dual of the above algorithm.
 *
 * How do we "perform a putMVar"?  Well, we have to fiddle around with
 * the stack of the thread waiting to do the putMVar.  See
 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
 * the stack layout, and the PerformPut and PerformTake macros below.
 *
 * It is important that a blocked take or put is woken up with the
 * take/put already performed, because otherwise there would be a
 * small window of vulnerability where the thread could receive an
 * exception and never perform its take or put, and we'd end up with a
 * deadlock.
 *
 * -------------------------------------------------------------------------- */

isEmptyMVarzh_fast
{
    /* args: R1 = MVar closure */

1086
    if (StgMVar_value(R1) == stg_END_TSO_QUEUE_closure) {
1087
	RET_N(1);
1088 1089
    } else {
	RET_N(0);
1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100
    }
}

newMVarzh_fast
{
    /* args: none */
    W_ mvar;

    ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
  
    mvar = Hp - SIZEOF_StgMVar + WDS(1);
1101 1102
    SET_HDR(mvar,stg_MVAR_DIRTY_info,W_[CCCS]);
        // MVARs start dirty: generation 0 has no mutable list
1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124
    StgMVar_head(mvar)  = stg_END_TSO_QUEUE_closure;
    StgMVar_tail(mvar)  = stg_END_TSO_QUEUE_closure;
    StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
    RET_P(mvar);
}


#define PerformTake(tso, value)				\
    W_[StgTSO_sp(tso) + WDS(1)] = value;		\
    W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;

#define PerformPut(tso,lval)			\
    StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3);	\
    lval = W_[StgTSO_sp(tso) - WDS(1)];

takeMVarzh_fast
{
    W_ mvar, val, info, tso;

    /* args: R1 = MVar closure */
    mvar = R1;

1125
#if defined(THREADED_RTS)
1126
    ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1127
#else
1128
    info = GET_INFO(mvar);
1129
#endif
1130 1131
        
    if (info == stg_MVAR_CLEAN_info) {
1132
        foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr") [];
1133
    }
1134 1135 1136 1137

    /* If the MVar is empty, put ourselves on its blocking queue,
     * and wait until we're woken up.
     */
1138
    if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1139 1140 1141
	if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
	    StgMVar_head(mvar) = CurrentTSO;
	} else {
1142 1143 1144
            foreign "C" setTSOLink(MyCapability() "ptr", 
                                   StgMVar_tail(mvar) "ptr",
                                   CurrentTSO) [];
1145
	}
1146
	StgTSO__link(CurrentTSO)       = stg_END_TSO_QUEUE_closure;
1147
	StgTSO_block_info(CurrentTSO)  = mvar;
1148 1149 1150 1151
        // write barrier for throwTo(), which looks at block_info
        // if why_blocked==BlockedOnMVar.
        prim %write_barrier() [];
	StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1152 1153
	StgMVar_tail(mvar) = CurrentTSO;
	
1154
        R1 = mvar;
1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170
	jump stg_block_takemvar;
  }

  /* we got the value... */
  val = StgMVar_value(mvar);

  if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
  {
      /* There are putMVar(s) waiting... 
       * wake up the first thread on the queue
       */
      ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);

      /* actually perform the putMVar for the thread that we just woke up */
      tso = StgMVar_head(mvar);
      PerformPut(tso,StgMVar_value(mvar));
1171

Simon Marlow's avatar
Simon Marlow committed
1172
      if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1173
          foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1174
      }
1175

1176 1177
      ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr", 
                                            StgMVar_head(mvar) "ptr", 1) [];
1178
      StgMVar_head(mvar) = tso;
1179

1180 1181 1182
      if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
	  StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
      }
1183

1184
#if defined(THREADED_RTS)
1185 1186 1187
      unlockClosure(mvar, stg_MVAR_DIRTY_info);
#else
      SET_INFO(mvar,stg_MVAR_DIRTY_info);
1188
#endif
1189 1190 1191 1192 1193 1194
      RET_P(val);
  } 
  else
  {
      /* No further putMVars, MVar is now empty */
      StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1195
 
1196
#if defined(THREADED_RTS)
1197
      unlockClosure(mvar, stg_MVAR_DIRTY_info);
1198
#else
1199
      SET_INFO(mvar,stg_MVAR_DIRTY_info);
1200
#endif
1201

1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214
      RET_P(val);
  }
}


tryTakeMVarzh_fast
{
    W_ mvar, val, info, tso;

    /* args: R1 = MVar closure */

    mvar = R1;

1215
#if defined(THREADED_RTS)
1216
    ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1217
#else
1218
    info = GET_INFO(mvar);
1219
#endif
1220

1221
    if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1222
#if defined(THREADED_RTS)
1223
        unlockClosure(mvar, info);
1224
#endif
1225 1226 1227 1228 1229 1230
	/* HACK: we need a pointer to pass back, 
	 * so we abuse NO_FINALIZER_closure
	 */
	RET_NP(0, stg_NO_FINALIZER_closure);
    }

1231
    if (info == stg_MVAR_CLEAN_info) {
1232
        foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1233 1234
    }

1235 1236 1237 1238
    /* we got the value... */
    val = StgMVar_value(mvar);

    if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1239

1240 1241 1242 1243 1244 1245 1246 1247
	/* There are putMVar(s) waiting... 
	 * wake up the first thread on the queue
	 */
	ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);

	/* actually perform the putMVar for the thread that we just woke up */
	tso = StgMVar_head(mvar);
	PerformPut(tso,StgMVar_value(mvar));
Simon Marlow's avatar
Simon Marlow committed
1248
        if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1249
            foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1250
        }
1251

1252 1253
        ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr", 
                                              StgMVar_head(mvar) "ptr", 1) [];