PrimOps.cmm 52.4 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
/*-----------------------------------------------------------------------------
  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.
 */

54
stg_newByteArrayzh
55 56
{
    W_ words, payload_words, n, p;
57
    MAYBE_GC(NO_PTRS,stg_newByteArrayzh);
58 59 60
    n = R1;
    payload_words = ROUNDUP_BYTES_TO_WDS(n);
    words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
61
    ("ptr" p) = foreign "C" allocate(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
stg_newPinnedByteArrayzh
72
{
73
    W_ words, bytes, payload_words, p;
74

75
    MAYBE_GC(NO_PTRS,stg_newPinnedByteArrayzh);
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
    ("ptr" p) = foreign "C" allocatePinned(MyCapability() "ptr", words) [];
Simon Marlow's avatar
Simon Marlow committed
89 90
    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
    p = p + ((-p - SIZEOF_StgArrWords) & BA_MASK);

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

100
stg_newAlignedPinnedByteArrayzh
Simon Marlow's avatar
Simon Marlow committed
101
{
102
    W_ words, bytes, payload_words, p, alignment;
Simon Marlow's avatar
Simon Marlow committed
103

104
    MAYBE_GC(NO_PTRS,stg_newAlignedPinnedByteArrayzh);
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(MyCapability() "ptr", 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

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

133
stg_newArrayzh
134 135 136 137 138
{
    W_ words, n, init, arr, p;
    /* Args: R1 = words, R2 = initialisation value */

    n = R1;
139
    MAYBE_GC(R2_PTR,stg_newArrayzh);
140 141

    words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
142
    ("ptr" arr) = foreign "C" allocate(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
    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);
}

161
stg_unsafeThawArrayzh
162 163 164 165 166 167 168 169
{
  // 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
  }
}

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

196
stg_newMutVarzh
197 198 199 200
{
    W_ mv;
    /* Args: R1 = initialisation value */

201
    ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, stg_newMutVarzh);
202 203

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

210
stg_atomicModifyMutVarzh
211
{
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
#define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
#endif

#define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)

249
   HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, stg_atomicModifyMutVarzh);
250

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
   RET_P(r);
}

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

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

298
stg_mkWeakzh
299 300 301 302 303 304 305 306 307 308 309
{
  /* R1 = key
     R2 = value
     R3 = finalizer (or NULL)
  */
  W_ w;

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

310
  ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, stg_mkWeakzh );
311 312 313 314

  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
  ACQUIRE_LOCK(sm_mutex);
325 326
  StgWeak_link(w)	= W_[weak_ptr_list];
  W_[weak_ptr_list] 	= w;
327
  RELEASE_LOCK(sm_mutex);
328

329
  IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
330 331 332 333

  RET_P(w);
}

334
stg_mkWeakForeignEnvzh
335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353
{
  /* 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;

354
  ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR, stg_mkWeakForeignEnvzh );
355 356 357 358 359 360

  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;
361
  ("ptr" p)     = foreign "C" allocate(MyCapability() "ptr", words) [];
362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379

  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;

380
  ACQUIRE_LOCK(sm_mutex);
381 382
  StgWeak_link(w)   = W_[weak_ptr_list];
  W_[weak_ptr_list] = w;
383
  RELEASE_LOCK(sm_mutex);
384 385 386 387 388

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

  RET_P(w);
}
389

390
stg_finalizzeWeakzh
391 392 393
{
  /* R1 = weak ptr
   */
394
  W_ w, f, arr;
395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421

  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);

422 423 424
  f   = StgWeak_finalizer(w);
  arr = StgWeak_cfinalizer(w);

425
  StgDeadWeak_link(w) = StgWeak_link(w);
426

427 428 429 430 431 432 433
  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)) [];
  }

434 435 436 437 438 439 440 441
  /* return the finalizer */
  if (f == stg_NO_FINALIZER_closure) {
      RET_NP(0,stg_NO_FINALIZER_closure);
  } else {
      RET_NP(1,f);
  }
}

442
stg_deRefWeakzh
443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458
{
  /* 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);
}

/* -----------------------------------------------------------------------------
459
   Floating point operations.
460 461
   -------------------------------------------------------------------------- */

462
stg_decodeFloatzuIntzh
463 464 465
{ 
    W_ p;
    F_ arg;
466 467 468
    W_ mp_tmp1;
    W_ mp_tmp_w;

469
    STK_CHK_GEN( WDS(2), NO_PTRS, stg_decodeFloatzuIntzh );
470 471 472

    mp_tmp1  = Sp - WDS(1);
    mp_tmp_w = Sp - WDS(2);
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]);
}

484
stg_decodeDoublezu2Intzh
485 486 487
{ 
    D_ arg;
    W_ p;
488 489 490 491 492
    W_ mp_tmp1;
    W_ mp_tmp2;
    W_ mp_result1;
    W_ mp_result2;

493
    STK_CHK_GEN( WDS(4), NO_PTRS, stg_decodeDoublezu2Intzh );
494 495 496 497 498

    mp_tmp1    = Sp - WDS(1);
    mp_tmp2    = Sp - WDS(2);
    mp_result1 = Sp - WDS(3);
    mp_result2 = Sp - WDS(4);
499 500 501 502 503

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

    /* Perform the operation */
504 505 506 507 508 509 510
    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]);
511 512
}

513 514 515 516
/* -----------------------------------------------------------------------------
 * Concurrency primitives
 * -------------------------------------------------------------------------- */

517
stg_forkzh
518 519
{
  /* args: R1 = closure to spark */
520

521
  MAYBE_GC(R1_PTR, stg_forkzh);
522

523 524 525 526
  W_ closure;
  W_ threadid;
  closure = R1;

527
  ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr", 
528 529
				RtsFlags_GcFlags_initialStkSize(RtsFlags), 
				closure "ptr") [];
530 531 532 533 534 535

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

536 537
  foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") [];

538 539
  // context switch soon, but not immediately: we don't want every
  // forkIO to force a context-switch.
540
  Capability_context_switch(MyCapability()) = 1 :: CInt;
541 542 543 544
  
  RET_P(threadid);
}

545
stg_forkOnzh
546 547 548
{
  /* args: R1 = cpu, R2 = closure to spark */

549
  MAYBE_GC(R2_PTR, stg_forkOnzh);
550 551 552 553 554 555 556

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

557
  ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr", 
558
				RtsFlags_GcFlags_initialStkSize(RtsFlags), 
559
				closure "ptr") [];
560 561 562 563 564 565

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

566
  foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") [];
567

568 569
  // context switch soon, but not immediately: we don't want every
  // forkIO to force a context-switch.
570
  Capability_context_switch(MyCapability()) = 1 :: CInt;
571
  
572
  RET_P(threadid);
573 574
}

575
stg_yieldzh
576 577 578 579
{
  jump stg_yield_noregs;
}

580
stg_myThreadIdzh
581 582 583 584 585
{
  /* no args. */
  RET_P(CurrentTSO);
}

586
stg_labelThreadzh
587 588 589 590 591
{
  /* args: 
	R1 = ThreadId#
	R2 = Addr# */
#ifdef DEBUG
592
  foreign "C" labelThread(R1 "ptr", R2 "ptr") [];
593 594 595 596
#endif
  jump %ENTRY_CODE(Sp(0));
}

597
stg_isCurrentThreadBoundzh
598 599 600
{
  /* no args */
  W_ r;
601
  (r) = foreign "C" isThreadBound(CurrentTSO) [];
602 603 604
  RET_N(r);
}

605
stg_threadStatuszh
606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637
{
    /* 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);
}
638 639 640 641 642 643 644 645 646

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

#define SP_OFF 0

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

647
INFO_TABLE_RET(stg_catch_retry_frame, CATCH_RETRY_FRAME,
648
#if defined(PROFILING)
649
  W_ unused1, W_ unused2,
650
#endif
651
  W_ unused3, P_ unused4, P_ unused5)
652 653 654 655 656
{
   W_ r, frame, trec, outer;

   frame = Sp;
   trec = StgTSO_trec(CurrentTSO);
657
   outer  = StgTRecHeader_enclosing_trec(trec);
658
   (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
659
   if (r != 0) {
660 661 662 663 664
     /* 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
665
     /* Did not commit: re-execute */
666
     W_ new_trec;
667
     ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
668
     StgTSO_trec(CurrentTSO) = new_trec;
Simon Marlow's avatar
Simon Marlow committed
669
     if (StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
670 671 672 673
       R1 = StgCatchRetryFrame_alt_code(frame);
     } else {
       R1 = StgCatchRetryFrame_first_code(frame);
     }
674
     jump stg_ap_v_fast;
675 676
   }
}
677 678


Simon Marlow's avatar
Simon Marlow committed
679
// Atomically frame ------------------------------------------------------------
680

681
INFO_TABLE_RET(stg_atomically_frame, ATOMICALLY_FRAME,
682
#if defined(PROFILING)
683
  W_ unused1, W_ unused2,
684
#endif
685
  P_ code, P_ next_invariant_to_check, P_ result)
686
{
tharris@microsoft.com's avatar
tharris@microsoft.com committed
687
  W_ frame, trec, valid, next_invariant, q, outer;
688

689 690 691
  frame  = Sp;
  trec   = StgTSO_trec(CurrentTSO);
  result = R1;
692
  outer  = StgTRecHeader_enclosing_trec(trec);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
693 694 695

  if (outer == NO_TREC) {
    /* First time back at the atomically frame -- pick up invariants */
696
    ("ptr" q) = foreign "C" stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr") [];
tharris@microsoft.com's avatar
tharris@microsoft.com committed
697
    StgAtomicallyFrame_next_invariant_to_check(frame) = q;
698
    StgAtomicallyFrame_result(frame) = result;
699 700

  } else {
tharris@microsoft.com's avatar
tharris@microsoft.com committed
701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717
    /* 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 */
718
    ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [];
719
    StgTSO_trec(CurrentTSO) = trec;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
720 721 722

    next_invariant = StgInvariantCheckQueue_invariant(q);
    R1 = StgAtomicInvariant_code(next_invariant);
723
    jump stg_ap_v_fast;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
724 725 726 727

  } else {

    /* We've got no more invariants to check, try to commit */
728
    (valid) = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") [];
tharris@microsoft.com's avatar
tharris@microsoft.com committed
729 730 731
    if (valid != 0) {
      /* Transaction was valid: commit succeeded */
      StgTSO_trec(CurrentTSO) = NO_TREC;
732
      R1 = StgAtomicallyFrame_result(frame);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
733 734 735 736
      Sp = Sp + SIZEOF_StgAtomicallyFrame;
      jump %ENTRY_CODE(Sp(SP_OFF));
    } else {
      /* Transaction was not valid: try again */
737
      ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
tharris@microsoft.com's avatar
tharris@microsoft.com committed
738 739 740 741 742
      StgTSO_trec(CurrentTSO) = trec;
      StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
      R1 = StgAtomicallyFrame_code(frame);
      jump stg_ap_v_fast;
    }
743 744 745
  }
}

746 747 748 749
INFO_TABLE_RET(stg_atomically_waiting_frame, ATOMICALLY_FRAME,
#if defined(PROFILING)
  W_ unused1, W_ unused2,
#endif
750
  P_ code, P_ next_invariant_to_check, P_ result)
751 752 753 754 755 756
{
  W_ frame, trec, valid;

  frame = Sp;

  /* The TSO is currently waiting: should we stop waiting? */
757
  (valid) = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
758
  if (valid != 0) {
759 760 761 762
    /* Previous attempt is still valid: no point trying again yet */
    jump stg_block_noregs;
  } else {
    /* Previous attempt is no longer valid: try again */
763
    ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
764 765 766
    StgTSO_trec(CurrentTSO) = trec;
    StgHeader_info(frame) = stg_atomically_frame_info;
    R1 = StgAtomicallyFrame_code(frame);
767
    jump stg_ap_v_fast;
768
  }
769
}
770 771 772 773 774 775 776 777 778 779

// 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.
 */

780 781 782 783
INFO_TABLE_RET(stg_catch_stm_frame, CATCH_STM_FRAME,
#if defined(PROFILING)
  W_ unused1, W_ unused2,
#endif
784
  P_ unused3, P_ unused4)
Simon Marlow's avatar
Simon Marlow committed
785 786 787 788
   {
      W_ r, frame, trec, outer;
      frame = Sp;
      trec = StgTSO_trec(CurrentTSO);
789
      outer  = StgTRecHeader_enclosing_trec(trec);
790
      (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
791 792 793 794 795 796 797 798
      if (r != 0) {
        /* Commit succeeded */
        StgTSO_trec(CurrentTSO) = outer;
        Sp = Sp + SIZEOF_StgCatchSTMFrame;
        jump Sp(SP_OFF);
      } else {
        /* Commit failed */
        W_ new_trec;
799
        ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
800 801 802 803 804
        StgTSO_trec(CurrentTSO) = new_trec;
        R1 = StgCatchSTMFrame_code(frame);
        jump stg_ap_v_fast;
      }
   }
805 806 807 808


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

809
stg_atomicallyzh
810 811 812 813 814
{
  W_ frame;
  W_ old_trec;
  W_ new_trec;
  
815
  // stmStartTransaction may allocate
816
  MAYBE_GC (R1_PTR, stg_atomicallyzh); 
817

818
  /* Args: R1 = m :: STM a */
819
  STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, stg_atomicallyzh);
820

821 822 823 824
  old_trec = StgTSO_trec(CurrentTSO);

  /* Nested transactions are not allowed; raise an exception */
  if (old_trec != NO_TREC) {
825
     R1 = base_ControlziExceptionziBase_nestedAtomically_closure;
826
     jump stg_raisezh;
827 828
  }

829 830 831 832
  /* Set up the atomically frame */
  Sp = Sp - SIZEOF_StgAtomicallyFrame;
  frame = Sp;

833
  SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
834
  StgAtomicallyFrame_code(frame) = R1;
835
  StgAtomicallyFrame_result(frame) = NO_TREC;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
836
  StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
837 838

  /* Start the memory transcation */
839
  ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1];
840 841 842
  StgTSO_trec(CurrentTSO) = new_trec;

  /* Apply R1 to the realworld token */
843
  jump stg_ap_v_fast;
844 845 846
}


847
stg_catchSTMzh
848 849 850 851 852
{
  W_ frame;
  
  /* Args: R1 :: STM a */
  /* Args: R2 :: Exception -> STM a */
853
  STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, stg_catchSTMzh);
854 855 856 857 858

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

859
  SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
860
  StgCatchSTMFrame_handler(frame) = R2;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
861 862 863 864 865 866
  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);
867
  ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", cur_trec "ptr");
tharris@microsoft.com's avatar
tharris@microsoft.com committed
868
  StgTSO_trec(CurrentTSO) = new_trec;
869 870

  /* Apply R1 to the realworld token */
871
  jump stg_ap_v_fast;
872 873 874
}


875
stg_catchRetryzh
876 877 878 879 880
{
  W_ frame;
  W_ new_trec;
  W_ trec;

881
  // stmStartTransaction may allocate
882
  MAYBE_GC (R1_PTR & R2_PTR, stg_catchRetryzh); 
883

884 885
  /* Args: R1 :: STM a */
  /* Args: R2 :: STM a */
886
  STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, stg_catchRetryzh);
887 888 889

  /* Start a nested transaction within which to run the first code */
  trec = StgTSO_trec(CurrentTSO);
890
  ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2];
891 892 893 894 895 896
  StgTSO_trec(CurrentTSO) = new_trec;

  /* Set up the catch-retry frame */
  Sp = Sp - SIZEOF_StgCatchRetryFrame;
  frame = Sp;
  
897
  SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
898
  StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
899 900 901 902
  StgCatchRetryFrame_first_code(frame) = R1;
  StgCatchRetryFrame_alt_code(frame) = R2;

  /* Apply R1 to the realworld token */
903
  jump stg_ap_v_fast;
904 905 906
}


907
stg_retryzh
908 909 910 911 912 913 914
{
  W_ frame_type;
  W_ frame;
  W_ trec;
  W_ outer;
  W_ r;

915
  MAYBE_GC (NO_PTRS, stg_retryzh); // STM operations may allocate
916 917 918 919

  // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
retry_pop_stack:
  StgTSO_sp(CurrentTSO) = Sp;
920
  (frame_type) = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") [];
921 922
  Sp = StgTSO_sp(CurrentTSO);
  frame = Sp;
tharris@microsoft.com's avatar
tharris@microsoft.com committed
923
  trec = StgTSO_trec(CurrentTSO);
924
  outer  = StgTRecHeader_enclosing_trec(trec);
925 926 927 928

  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
929 930 931
    // 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
932
    if (!StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
tharris@microsoft.com's avatar
tharris@microsoft.com committed
933
      // Retry in the first branch: try the alternative
934
      ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
935
      StgTSO_trec(CurrentTSO) = trec;
936
      StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
937
      R1 = StgCatchRetryFrame_alt_code(frame);
938
      jump stg_ap_v_fast;
939
    } else {
tharris@microsoft.com's avatar
tharris@microsoft.com committed
940 941 942 943
      // Retry in the alternative code: propagate the retry
      StgTSO_trec(CurrentTSO) = outer;
      Sp = Sp + SIZEOF_StgCatchRetryFrame;
      goto retry_pop_stack;
944 945 946 947 948
    }
  }

  // We've reached the ATOMICALLY_FRAME: attempt to wait 
  ASSERT(frame_type == ATOMICALLY_FRAME);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
949 950 951 952 953 954 955
  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;
956
    StgTSO_trec(CurrentTSO) = trec;
957
    outer  = StgTRecHeader_enclosing_trec(trec);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
958
  }
959
  ASSERT(outer == NO_TREC);
tharris@microsoft.com's avatar
tharris@microsoft.com committed
960

961
  (r) = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") [];
Simon Marlow's avatar
Simon Marlow committed
962
  if (r != 0) {
963
    // Transaction was valid: stmWait put us on the TVars' queues, we now block
964
    StgHeader_info(frame) = stg_atomically_waiting_frame_info;
965
    Sp = frame;
966
    // Fix up the stack in the unregisterised case: the return convention is different.
967 968
    R3 = trec; // passing to stmWaitUnblock()
    jump stg_block_stmwait;
969 970
  } else {
    // Transaction was not valid: retry immediately
971
    ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
972 973 974
    StgTSO_trec(CurrentTSO) = trec;
    R1 = StgAtomicallyFrame_code(frame);
    Sp = frame;
975
    jump stg_ap_v_fast;
976 977 978 979
  }
}


980
stg_checkzh
tharris@microsoft.com's avatar
tharris@microsoft.com committed
981 982 983 984
{
  W_ trec, closure;

  /* Args: R1 = invariant closure */
985
  MAYBE_GC (R1_PTR, stg_checkzh); 
tharris@microsoft.com's avatar
tharris@microsoft.com committed
986 987 988 989 990 991 992 993 994 995 996

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

  jump %ENTRY_CODE(Sp(0));
}


997
stg_newTVarzh
998 999
{
  W_ tv;
1000
  W_ new_value;
1001 1002 1003

  /* Args: R1 = initialisation value */

1004
  MAYBE_GC (R1_PTR, stg_newTVarzh); 
1005
  new_value = R1;
1006
  ("ptr" tv) = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") [];
1007 1008 1009 1010
  RET_P(tv);
}


1011
stg_readTVarzh
1012 1013 1014 1015 1016 1017 1018
{
  W_ trec;
  W_ tvar;
  W_ result;

  /* Args: R1 = TVar closure */

1019
  MAYBE_GC (R1_PTR, stg_readTVarzh); // Call to stmReadTVar may allocate
1020 1021
  trec = StgTSO_trec(CurrentTSO);
  tvar = R1;
1022
  ("ptr" result) = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") [];
1023 1024 1025 1026

  RET_P(result);
}

1027
stg_readTVarIOzh
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037
{
    W_ result;

again:
    result = StgTVar_current_value(R1);
    if (%INFO_PTR(result) == stg_TREC_HEADER_info) {
        goto again;
    }
    RET_P(result);
}
1038

1039
stg_writeTVarzh
1040 1041 1042 1043 1044 1045 1046 1047
{
  W_ trec;
  W_ tvar;
  W_ new_value;
  
  /* Args: R1 = TVar closure */
  /*       R2 = New value    */

1048
  MAYBE_GC (R1_PTR & R2_PTR, stg_writeTVarzh); // Call to stmWriteTVar may allocate
1049 1050 1051
  trec = StgTSO_trec(CurrentTSO);
  tvar = R1;
  new_value = R2;
1052
  foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") [];
1053 1054 1055 1056 1057

  jump %ENTRY_CODE(Sp(0));
}


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 1086 1087 1088 1089
/* -----------------------------------------------------------------------------
 * 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.
 *
 * -------------------------------------------------------------------------- */

1090
stg_isEmptyMVarzh
1091 1092 1093
{
    /* args: R1 = MVar closure */

1094
    if (StgMVar_value(R1) == stg_END_TSO_QUEUE_closure) {
1095
	RET_N(1);
1096 1097
    } else {
	RET_N(0);
simonmar's avatar