1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2004
5 * Out-of-line primitive operations
7 * This file contains the implementations of all the primitive
8 * operations ("primops") which are not expanded inline. See
9 * ghc/compiler/prelude/primops.txt.pp for a list of all the primops;
10 * this file contains code for most of those with the attribute
13 * Entry convention: the entry convention for a primop is that all the
14 * args are in Stg registers (R1, R2, etc.). This is to make writing
15 * the primops easier. (see compiler/codeGen/CgCallConv.hs).
17 * Return convention: results from a primop are generally returned
18 * using the ordinary unboxed tuple return convention. The C-- parser
19 * implements the RET_xxxx() macros to perform unboxed-tuple returns
20 * based on the prevailing return convention.
22 * This file is written in a subset of C--, extended with various
23 * features specific to GHC. It is compiled by GHC directly. For the
24 * syntax of .cmm files, see the parser in ghc/compiler/cmm/CmmParse.y.
26 * ---------------------------------------------------------------------------*/
31 #ifndef mingw32_HOST_OS
41 import __gmpz_tdiv_qr;
42 import __gmpz_fdiv_qr;
43 import __gmpz_divexact;
49 import base_GHCziIOBase_NestedAtomically_closure;
50 import pthread_mutex_lock;
51 import pthread_mutex_unlock;
53 import EnterCriticalSection;
54 import LeaveCriticalSection;
56 /*-----------------------------------------------------------------------------
59 Basically just new*Array - the others are all inline macros.
61 The size arg is always passed in R1, and the result returned in R1.
63 The slow entry point is for returning from a heap check, the saved
64 size argument must be re-loaded from the stack.
65 -------------------------------------------------------------------------- */
67 /* for objects that are *less* than the size of a word, make sure we
68 * round up to the nearest word for the size of the array.
73 W_ words, payload_words, n, p;
74 MAYBE_GC(NO_PTRS,newByteArrayzh_fast);
76 payload_words = ROUNDUP_BYTES_TO_WDS(n);
77 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
78 ("ptr" p) = foreign "C" allocateLocal(MyCapability() "ptr",words) [];
79 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
80 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
81 StgArrWords_words(p) = payload_words;
85 newPinnedByteArrayzh_fast
87 W_ words, payload_words, n, p;
89 MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast);
91 payload_words = ROUNDUP_BYTES_TO_WDS(n);
93 // We want an 8-byte aligned array. allocatePinned() gives us
94 // 8-byte aligned memory by default, but we want to align the
95 // *goods* inside the ArrWords object, so we have to check the
96 // size of the ArrWords header and adjust our size accordingly.
97 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
98 if ((SIZEOF_StgArrWords & 7) != 0) {
102 ("ptr" p) = foreign "C" allocatePinned(words) [];
103 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
105 // Again, if the ArrWords header isn't a multiple of 8 bytes, we
106 // have to push the object forward one word so that the goods
107 // fall on an 8-byte boundary.
108 if ((SIZEOF_StgArrWords & 7) != 0) {
112 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
113 StgArrWords_words(p) = payload_words;
119 W_ words, n, init, arr, p;
120 /* Args: R1 = words, R2 = initialisation value */
123 MAYBE_GC(R2_PTR,newArrayzh_fast);
125 words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
126 ("ptr" arr) = foreign "C" allocateLocal(MyCapability() "ptr",words) [R2];
127 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
129 SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, W_[CCCS]);
130 StgMutArrPtrs_ptrs(arr) = n;
132 // Initialise all elements of the the array with the value in R2
134 p = arr + SIZEOF_StgMutArrPtrs;
136 if (p < arr + WDS(words)) {
145 unsafeThawArrayzh_fast
147 // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
149 // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
150 // normally doesn't. However, when we freeze a MUT_ARR_PTRS, we leave
151 // it on the mutable list for the GC to remove (removing something from
152 // the mutable list is not easy, because the mut_list is only singly-linked).
154 // So that we can tell whether a MUT_ARR_PTRS_FROZEN is on the mutable list,
155 // when we freeze it we set the info ptr to be MUT_ARR_PTRS_FROZEN0
156 // to indicate that it is still on the mutable list.
158 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
159 // either it is on a mut_list, or it isn't. We adopt the convention that
160 // the closure type is MUT_ARR_PTRS_FROZEN0 if it is on the mutable list,
161 // and MUT_ARR_PTRS_FROZEN otherwise. In fact it wouldn't matter if
162 // we put it on the mutable list more than once, but it would get scavenged
163 // multiple times during GC, which would be unnecessarily slow.
165 if (StgHeader_info(R1) != stg_MUT_ARR_PTRS_FROZEN0_info) {
166 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
167 recordMutable(R1, R1);
168 // must be done after SET_INFO, because it ASSERTs closure_MUTABLE()
171 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
176 /* -----------------------------------------------------------------------------
178 -------------------------------------------------------------------------- */
183 /* Args: R1 = initialisation value */
185 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);
187 mv = Hp - SIZEOF_StgMutVar + WDS(1);
188 SET_HDR(mv,stg_MUT_VAR_DIRTY_info,W_[CCCS]);
189 StgMutVar_var(mv) = R1;
194 atomicModifyMutVarzh_fast
197 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
199 /* If x is the current contents of the MutVar#, then
200 We want to make the new contents point to
204 and the return value is
208 obviously we can share (f x).
210 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
211 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
212 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
216 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
217 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
219 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
220 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
224 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
225 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
227 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
228 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
231 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
233 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);
235 #if defined(THREADED_RTS)
236 ACQUIRE_LOCK(atomic_modify_mutvar_mutex "ptr") [R1,R2];
239 x = StgMutVar_var(R1);
241 TICK_ALLOC_THUNK_2();
242 CCCS_ALLOC(THUNK_2_SIZE);
243 z = Hp - THUNK_2_SIZE + WDS(1);
244 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
245 LDV_RECORD_CREATE(z);
246 StgThunk_payload(z,0) = R2;
247 StgThunk_payload(z,1) = x;
249 TICK_ALLOC_THUNK_1();
250 CCCS_ALLOC(THUNK_1_SIZE);
251 y = z - THUNK_1_SIZE;
252 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
253 LDV_RECORD_CREATE(y);
254 StgThunk_payload(y,0) = z;
256 StgMutVar_var(R1) = y;
257 foreign "C" dirty_MUT_VAR(BaseReg "ptr", R1 "ptr") [R1];
259 TICK_ALLOC_THUNK_1();
260 CCCS_ALLOC(THUNK_1_SIZE);
261 r = y - THUNK_1_SIZE;
262 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
263 LDV_RECORD_CREATE(r);
264 StgThunk_payload(r,0) = z;
266 #if defined(THREADED_RTS)
267 RELEASE_LOCK(atomic_modify_mutvar_mutex "ptr") [];
273 /* -----------------------------------------------------------------------------
274 Weak Pointer Primitives
275 -------------------------------------------------------------------------- */
277 STRING(stg_weak_msg,"New weak pointer at %p\n")
283 R3 = finalizer (or NULL)
288 R3 = stg_NO_FINALIZER_closure;
291 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
293 w = Hp - SIZEOF_StgWeak + WDS(1);
294 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
297 StgWeak_value(w) = R2;
298 StgWeak_finalizer(w) = R3;
300 StgWeak_link(w) = W_[weak_ptr_list];
301 W_[weak_ptr_list] = w;
303 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
318 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
319 RET_NP(0,stg_NO_FINALIZER_closure);
325 // A weak pointer is inherently used, so we do not need to call
326 // LDV_recordDead_FILL_SLOP_DYNAMIC():
327 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
328 // or, LDV_recordDead():
329 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
330 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
331 // large as weak pointers, so there is no need to fill the slop, either.
332 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
336 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
338 SET_INFO(w,stg_DEAD_WEAK_info);
339 LDV_RECORD_CREATE(w);
341 f = StgWeak_finalizer(w);
342 StgDeadWeak_link(w) = StgWeak_link(w);
344 /* return the finalizer */
345 if (f == stg_NO_FINALIZER_closure) {
346 RET_NP(0,stg_NO_FINALIZER_closure);
358 if (GET_INFO(w) == stg_WEAK_info) {
360 val = StgWeak_value(w);
368 /* -----------------------------------------------------------------------------
369 Arbitrary-precision Integer operations.
371 There are some assumptions in this code that mp_limb_t == W_. This is
372 the case for all the platforms that GHC supports, currently.
373 -------------------------------------------------------------------------- */
377 /* arguments: R1 = Int# */
379 W_ val, s, p; /* to avoid aliasing */
382 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
384 p = Hp - SIZEOF_StgArrWords;
385 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
386 StgArrWords_words(p) = 1;
388 /* mpz_set_si is inlined here, makes things simpler */
401 /* returns (# size :: Int#,
410 /* arguments: R1 = Word# */
412 W_ val, s, p; /* to avoid aliasing */
416 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
418 p = Hp - SIZEOF_StgArrWords;
419 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
420 StgArrWords_words(p) = 1;
429 /* returns (# size :: Int#,
430 data :: ByteArray# #)
437 * 'long long' primops for converting to/from Integers.
440 #ifdef SUPPORT_LONG_LONGS
442 int64ToIntegerzh_fast
444 /* arguments: L1 = Int64# */
447 W_ hi, lo, s, neg, words_needed, p;
452 hi = TO_W_(val >> 32);
455 if ( hi != 0 && hi != 0xFFFFFFFF ) {
458 // minimum is one word
462 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
463 NO_PTRS, int64ToIntegerzh_fast );
465 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
466 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
467 StgArrWords_words(p) = words_needed;
479 if ( words_needed == 2 ) {
487 } else /* val==0 */ {
495 /* returns (# size :: Int#,
496 data :: ByteArray# #)
500 word64ToIntegerzh_fast
502 /* arguments: L1 = Word64# */
505 W_ hi, lo, s, words_needed, p;
508 hi = TO_W_(val >> 32);
517 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
518 NO_PTRS, word64ToIntegerzh_fast );
520 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
521 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
522 StgArrWords_words(p) = words_needed;
532 } else /* val==0 */ {
537 /* returns (# size :: Int#,
538 data :: ByteArray# #)
545 #endif /* SUPPORT_LONG_LONGS */
547 /* ToDo: this is shockingly inefficient */
552 bits8 [SIZEOF_MP_INT];
557 bits8 [SIZEOF_MP_INT];
562 bits8 [SIZEOF_MP_INT];
567 bits8 [SIZEOF_MP_INT];
572 #define FETCH_MP_TEMP(X) \
574 X = BaseReg + (OFFSET_StgRegTable_r ## X);
576 #define FETCH_MP_TEMP(X) /* Nothing */
579 #define GMP_TAKE2_RET1(name,mp_fun) \
584 FETCH_MP_TEMP(mp_tmp1); \
585 FETCH_MP_TEMP(mp_tmp2); \
586 FETCH_MP_TEMP(mp_result1) \
587 FETCH_MP_TEMP(mp_result2); \
589 /* call doYouWantToGC() */ \
590 MAYBE_GC(R2_PTR & R4_PTR, name); \
597 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
598 MP_INT__mp_size(mp_tmp1) = (s1); \
599 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
600 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
601 MP_INT__mp_size(mp_tmp2) = (s2); \
602 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
604 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
606 /* Perform the operation */ \
607 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
609 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
610 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
613 #define GMP_TAKE1_RET1(name,mp_fun) \
618 FETCH_MP_TEMP(mp_tmp1); \
619 FETCH_MP_TEMP(mp_result1) \
621 /* call doYouWantToGC() */ \
622 MAYBE_GC(R2_PTR, name); \
627 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
628 MP_INT__mp_size(mp_tmp1) = (s1); \
629 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
631 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
633 /* Perform the operation */ \
634 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr") []; \
636 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
637 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
640 #define GMP_TAKE2_RET2(name,mp_fun) \
645 FETCH_MP_TEMP(mp_tmp1); \
646 FETCH_MP_TEMP(mp_tmp2); \
647 FETCH_MP_TEMP(mp_result1) \
648 FETCH_MP_TEMP(mp_result2) \
650 /* call doYouWantToGC() */ \
651 MAYBE_GC(R2_PTR & R4_PTR, name); \
658 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
659 MP_INT__mp_size(mp_tmp1) = (s1); \
660 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
661 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
662 MP_INT__mp_size(mp_tmp2) = (s2); \
663 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
665 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
666 foreign "C" __gmpz_init(mp_result2 "ptr") []; \
668 /* Perform the operation */ \
669 foreign "C" mp_fun(mp_result1 "ptr",mp_result2 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
671 RET_NPNP(TO_W_(MP_INT__mp_size(mp_result1)), \
672 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords, \
673 TO_W_(MP_INT__mp_size(mp_result2)), \
674 MP_INT__mp_d(mp_result2) - SIZEOF_StgArrWords); \
677 GMP_TAKE2_RET1(plusIntegerzh_fast, __gmpz_add)
678 GMP_TAKE2_RET1(minusIntegerzh_fast, __gmpz_sub)
679 GMP_TAKE2_RET1(timesIntegerzh_fast, __gmpz_mul)
680 GMP_TAKE2_RET1(gcdIntegerzh_fast, __gmpz_gcd)
681 GMP_TAKE2_RET1(quotIntegerzh_fast, __gmpz_tdiv_q)
682 GMP_TAKE2_RET1(remIntegerzh_fast, __gmpz_tdiv_r)
683 GMP_TAKE2_RET1(divExactIntegerzh_fast, __gmpz_divexact)
684 GMP_TAKE2_RET1(andIntegerzh_fast, __gmpz_and)
685 GMP_TAKE2_RET1(orIntegerzh_fast, __gmpz_ior)
686 GMP_TAKE2_RET1(xorIntegerzh_fast, __gmpz_xor)
687 GMP_TAKE1_RET1(complementIntegerzh_fast, __gmpz_com)
689 GMP_TAKE2_RET2(quotRemIntegerzh_fast, __gmpz_tdiv_qr)
690 GMP_TAKE2_RET2(divModIntegerzh_fast, __gmpz_fdiv_qr)
694 mp_tmp_w: W_; // NB. mp_tmp_w is really an here mp_limb_t
700 /* R1 = the first Int#; R2 = the second Int# */
702 FETCH_MP_TEMP(mp_tmp_w);
705 (r) = foreign "C" __gmpn_gcd_1(mp_tmp_w "ptr", 1, R2) [];
708 /* Result parked in R1, return via info-pointer at TOS */
709 jump %ENTRY_CODE(Sp(0));
715 /* R1 = s1; R2 = d1; R3 = the int */
717 (s1) = foreign "C" __gmpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3) [];
720 /* Result parked in R1, return via info-pointer at TOS */
721 jump %ENTRY_CODE(Sp(0));
727 /* R1 = s1; R2 = d1; R3 = the int */
728 W_ usize, vsize, v_digit, u_digit;
734 // paraphrased from __gmpz_cmp_si() in the GMP sources
735 if (%gt(v_digit,0)) {
738 if (%lt(v_digit,0)) {
744 if (usize != vsize) {
746 jump %ENTRY_CODE(Sp(0));
751 jump %ENTRY_CODE(Sp(0));
754 u_digit = W_[BYTE_ARR_CTS(R2)];
756 if (u_digit == v_digit) {
758 jump %ENTRY_CODE(Sp(0));
761 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
767 jump %ENTRY_CODE(Sp(0));
772 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
773 W_ usize, vsize, size, up, vp;
776 // paraphrased from __gmpz_cmp() in the GMP sources
780 if (usize != vsize) {
782 jump %ENTRY_CODE(Sp(0));
787 jump %ENTRY_CODE(Sp(0));
790 if (%lt(usize,0)) { // NB. not <, which is unsigned
796 up = BYTE_ARR_CTS(R2);
797 vp = BYTE_ARR_CTS(R4);
799 (cmp) = foreign "C" __gmpn_cmp(up "ptr", vp "ptr", size) [];
801 if (cmp == 0 :: CInt) {
803 jump %ENTRY_CODE(Sp(0));
806 if (%lt(cmp,0 :: CInt) == %lt(usize,0)) {
811 /* Result parked in R1, return via info-pointer at TOS */
812 jump %ENTRY_CODE(Sp(0));
824 r = W_[R2 + SIZEOF_StgArrWords];
829 /* Result parked in R1, return via info-pointer at TOS */
831 jump %ENTRY_CODE(Sp(0));
843 r = W_[R2 + SIZEOF_StgArrWords];
848 /* Result parked in R1, return via info-pointer at TOS */
850 jump %ENTRY_CODE(Sp(0));
857 FETCH_MP_TEMP(mp_tmp1);
858 FETCH_MP_TEMP(mp_tmp_w);
860 /* arguments: F1 = Float# */
863 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
865 /* Be prepared to tell Lennart-coded __decodeFloat
866 where mantissa._mp_d can be put (it does not care about the rest) */
867 p = Hp - SIZEOF_StgArrWords;
868 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
869 StgArrWords_words(p) = 1;
870 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
872 /* Perform the operation */
873 foreign "C" __decodeFloat(mp_tmp1 "ptr",mp_tmp_w "ptr" ,arg) [];
875 /* returns: (Int# (expn), Int#, ByteArray#) */
876 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
879 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
880 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
886 FETCH_MP_TEMP(mp_tmp1);
887 FETCH_MP_TEMP(mp_tmp_w);
889 /* arguments: D1 = Double# */
892 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
894 /* Be prepared to tell Lennart-coded __decodeDouble
895 where mantissa.d can be put (it does not care about the rest) */
896 p = Hp - ARR_SIZE + WDS(1);
897 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
898 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
899 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
901 /* Perform the operation */
902 foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg) [];
904 /* returns: (Int# (expn), Int#, ByteArray#) */
905 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
908 /* -----------------------------------------------------------------------------
909 * Concurrency primitives
910 * -------------------------------------------------------------------------- */
914 /* args: R1 = closure to spark */
916 MAYBE_GC(R1_PTR, forkzh_fast);
922 ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr",
923 RtsFlags_GcFlags_initialStkSize(RtsFlags),
926 /* start blocked if the current thread is blocked */
927 StgTSO_flags(threadid) =
928 StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) &
929 (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));
931 foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") [];
933 // switch at the earliest opportunity
934 CInt[context_switch] = 1 :: CInt;
941 /* args: R1 = cpu, R2 = closure to spark */
943 MAYBE_GC(R2_PTR, forkOnzh_fast);
951 ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr",
952 RtsFlags_GcFlags_initialStkSize(RtsFlags),
955 /* start blocked if the current thread is blocked */
956 StgTSO_flags(threadid) =
957 StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) &
958 (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));
960 foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") [];
962 // switch at the earliest opportunity
963 CInt[context_switch] = 1 :: CInt;
970 jump stg_yield_noregs;
985 foreign "C" labelThread(R1 "ptr", R2 "ptr") [];
987 jump %ENTRY_CODE(Sp(0));
990 isCurrentThreadBoundzh_fast
994 (r) = foreign "C" isThreadBound(CurrentTSO) [];
999 /* -----------------------------------------------------------------------------
1001 * -------------------------------------------------------------------------- */
1005 #define IF_NOT_REG_R1(x)
1008 #define IF_NOT_REG_R1(x) x
1011 // Catch retry frame ------------------------------------------------------------
1013 INFO_TABLE_RET(stg_catch_retry_frame, CATCH_RETRY_FRAME,
1014 #if defined(PROFILING)
1015 W_ unused1, W_ unused2,
1017 W_ unused3, "ptr" W_ unused4, "ptr" W_ unused5)
1019 W_ r, frame, trec, outer;
1020 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1023 trec = StgTSO_trec(CurrentTSO);
1024 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1025 (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
1027 /* Succeeded (either first branch or second branch) */
1028 StgTSO_trec(CurrentTSO) = outer;
1029 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1030 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1031 jump %ENTRY_CODE(Sp(SP_OFF));
1033 /* Did not commit: re-execute */
1035 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1036 StgTSO_trec(CurrentTSO) = new_trec;
1037 if (StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
1038 R1 = StgCatchRetryFrame_alt_code(frame);
1040 R1 = StgCatchRetryFrame_first_code(frame);
1047 // Atomically frame ------------------------------------------------------------
1049 INFO_TABLE_RET(stg_atomically_frame, ATOMICALLY_FRAME,
1050 #if defined(PROFILING)
1051 W_ unused1, W_ unused2,
1053 "ptr" W_ unused3, "ptr" W_ unused4)
1055 W_ frame, trec, valid, next_invariant, q, outer;
1056 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1059 trec = StgTSO_trec(CurrentTSO);
1060 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1062 if (outer == NO_TREC) {
1063 /* First time back at the atomically frame -- pick up invariants */
1064 ("ptr" q) = foreign "C" stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr") [];
1065 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
1068 /* Second/subsequent time back at the atomically frame -- abort the
1069 * tx that's checking the invariant and move on to the next one */
1070 StgTSO_trec(CurrentTSO) = outer;
1071 q = StgAtomicallyFrame_next_invariant_to_check(frame);
1072 StgInvariantCheckQueue_my_execution(q) = trec;
1073 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1074 /* Don't free trec -- it's linked from q and will be stashed in the
1075 * invariant if we eventually commit. */
1076 q = StgInvariantCheckQueue_next_queue_entry(q);
1077 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
1081 q = StgAtomicallyFrame_next_invariant_to_check(frame);
1083 if (q != END_INVARIANT_CHECK_QUEUE) {
1084 /* We can't commit yet: another invariant to check */
1085 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [];
1086 StgTSO_trec(CurrentTSO) = trec;
1088 next_invariant = StgInvariantCheckQueue_invariant(q);
1089 R1 = StgAtomicInvariant_code(next_invariant);
1094 /* We've got no more invariants to check, try to commit */
1095 (valid) = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") [];
1097 /* Transaction was valid: commit succeeded */
1098 StgTSO_trec(CurrentTSO) = NO_TREC;
1099 Sp = Sp + SIZEOF_StgAtomicallyFrame;
1100 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1101 jump %ENTRY_CODE(Sp(SP_OFF));
1103 /* Transaction was not valid: try again */
1104 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1105 StgTSO_trec(CurrentTSO) = trec;
1106 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
1107 R1 = StgAtomicallyFrame_code(frame);
1113 INFO_TABLE_RET(stg_atomically_waiting_frame, ATOMICALLY_FRAME,
1114 #if defined(PROFILING)
1115 W_ unused1, W_ unused2,
1117 "ptr" W_ unused3, "ptr" W_ unused4)
1119 W_ frame, trec, valid;
1120 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1124 /* The TSO is currently waiting: should we stop waiting? */
1125 (valid) = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") [];
1127 /* Previous attempt is still valid: no point trying again yet */
1128 IF_NOT_REG_R1(Sp_adj(-2);
1129 Sp(1) = stg_NO_FINALIZER_closure;
1130 Sp(0) = stg_ut_1_0_unreg_info;)
1131 jump stg_block_noregs;
1133 /* Previous attempt is no longer valid: try again */
1134 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1135 StgTSO_trec(CurrentTSO) = trec;
1136 StgHeader_info(frame) = stg_atomically_frame_info;
1137 R1 = StgAtomicallyFrame_code(frame);
1142 // STM catch frame --------------------------------------------------------------
1150 /* Catch frames are very similar to update frames, but when entering
1151 * one we just pop the frame off the stack and perform the correct
1152 * kind of return to the activation record underneath us on the stack.
1155 INFO_TABLE_RET(stg_catch_stm_frame, CATCH_STM_FRAME,
1156 #if defined(PROFILING)
1157 W_ unused1, W_ unused2,
1159 "ptr" W_ unused3, "ptr" W_ unused4)
1161 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1162 W_ r, frame, trec, outer;
1164 trec = StgTSO_trec(CurrentTSO);
1165 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1166 (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
1168 /* Commit succeeded */
1169 StgTSO_trec(CurrentTSO) = outer;
1170 Sp = Sp + SIZEOF_StgCatchSTMFrame;
1171 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1176 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1177 StgTSO_trec(CurrentTSO) = new_trec;
1178 R1 = StgCatchSTMFrame_code(frame);
1184 // Primop definition ------------------------------------------------------------
1192 // stmStartTransaction may allocate
1193 MAYBE_GC (R1_PTR, atomicallyzh_fast);
1195 /* Args: R1 = m :: STM a */
1196 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);
1198 old_trec = StgTSO_trec(CurrentTSO);
1200 /* Nested transactions are not allowed; raise an exception */
1201 if (old_trec != NO_TREC) {
1202 R1 = base_GHCziIOBase_NestedAtomically_closure;
1206 /* Set up the atomically frame */
1207 Sp = Sp - SIZEOF_StgAtomicallyFrame;
1210 SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
1211 StgAtomicallyFrame_code(frame) = R1;
1212 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
1214 /* Start the memory transcation */
1215 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1];
1216 StgTSO_trec(CurrentTSO) = new_trec;
1218 /* Apply R1 to the realworld token */
1227 /* Args: R1 :: STM a */
1228 /* Args: R2 :: Exception -> STM a */
1229 STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast);
1231 /* Set up the catch frame */
1232 Sp = Sp - SIZEOF_StgCatchSTMFrame;
1235 SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
1236 StgCatchSTMFrame_handler(frame) = R2;
1237 StgCatchSTMFrame_code(frame) = R1;
1239 /* Start a nested transaction to run the body of the try block in */
1242 cur_trec = StgTSO_trec(CurrentTSO);
1243 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", cur_trec "ptr");
1244 StgTSO_trec(CurrentTSO) = new_trec;
1246 /* Apply R1 to the realworld token */
1257 // stmStartTransaction may allocate
1258 MAYBE_GC (R1_PTR & R2_PTR, catchRetryzh_fast);
1260 /* Args: R1 :: STM a */
1261 /* Args: R2 :: STM a */
1262 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);
1264 /* Start a nested transaction within which to run the first code */
1265 trec = StgTSO_trec(CurrentTSO);
1266 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2];
1267 StgTSO_trec(CurrentTSO) = new_trec;
1269 /* Set up the catch-retry frame */
1270 Sp = Sp - SIZEOF_StgCatchRetryFrame;
1273 SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
1274 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1275 StgCatchRetryFrame_first_code(frame) = R1;
1276 StgCatchRetryFrame_alt_code(frame) = R2;
1278 /* Apply R1 to the realworld token */
1291 MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate
1293 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
1295 StgTSO_sp(CurrentTSO) = Sp;
1296 (frame_type) = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") [];
1297 Sp = StgTSO_sp(CurrentTSO);
1299 trec = StgTSO_trec(CurrentTSO);
1300 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1302 if (frame_type == CATCH_RETRY_FRAME) {
1303 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
1304 ASSERT(outer != NO_TREC);
1305 // Abort the transaction attempting the current branch
1306 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1307 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
1308 if (!StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
1309 // Retry in the first branch: try the alternative
1310 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1311 StgTSO_trec(CurrentTSO) = trec;
1312 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
1313 R1 = StgCatchRetryFrame_alt_code(frame);
1316 // Retry in the alternative code: propagate the retry
1317 StgTSO_trec(CurrentTSO) = outer;
1318 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1319 goto retry_pop_stack;
1323 // We've reached the ATOMICALLY_FRAME: attempt to wait
1324 ASSERT(frame_type == ATOMICALLY_FRAME);
1325 if (outer != NO_TREC) {
1326 // We called retry while checking invariants, so abort the current
1327 // invariant check (merging its TVar accesses into the parents read
1328 // set so we'll wait on them)
1329 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1330 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
1332 StgTSO_trec(CurrentTSO) = trec;
1333 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1335 ASSERT(outer == NO_TREC);
1337 (r) = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") [];
1339 // Transaction was valid: stmWait put us on the TVars' queues, we now block
1340 StgHeader_info(frame) = stg_atomically_waiting_frame_info;
1342 // Fix up the stack in the unregisterised case: the return convention is different.
1343 IF_NOT_REG_R1(Sp_adj(-2);
1344 Sp(1) = stg_NO_FINALIZER_closure;
1345 Sp(0) = stg_ut_1_0_unreg_info;)
1346 R3 = trec; // passing to stmWaitUnblock()
1347 jump stg_block_stmwait;
1349 // Transaction was not valid: retry immediately
1350 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1351 StgTSO_trec(CurrentTSO) = trec;
1352 R1 = StgAtomicallyFrame_code(frame);
1363 /* Args: R1 = invariant closure */
1364 MAYBE_GC (R1_PTR, checkzh_fast);
1366 trec = StgTSO_trec(CurrentTSO);
1368 foreign "C" stmAddInvariantToCheck(MyCapability() "ptr",
1372 jump %ENTRY_CODE(Sp(0));
1381 /* Args: R1 = initialisation value */
1383 MAYBE_GC (R1_PTR, newTVarzh_fast);
1385 ("ptr" tv) = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") [];
1396 /* Args: R1 = TVar closure */
1398 MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
1399 trec = StgTSO_trec(CurrentTSO);
1401 ("ptr" result) = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") [];
1413 /* Args: R1 = TVar closure */
1414 /* R2 = New value */
1416 MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
1417 trec = StgTSO_trec(CurrentTSO);
1420 foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") [];
1422 jump %ENTRY_CODE(Sp(0));
1426 /* -----------------------------------------------------------------------------
1429 * take & putMVar work as follows. Firstly, an important invariant:
1431 * If the MVar is full, then the blocking queue contains only
1432 * threads blocked on putMVar, and if the MVar is empty then the
1433 * blocking queue contains only threads blocked on takeMVar.
1436 * MVar empty : then add ourselves to the blocking queue
1437 * MVar full : remove the value from the MVar, and
1438 * blocking queue empty : return
1439 * blocking queue non-empty : perform the first blocked putMVar
1440 * from the queue, and wake up the
1441 * thread (MVar is now full again)
1443 * putMVar is just the dual of the above algorithm.
1445 * How do we "perform a putMVar"? Well, we have to fiddle around with
1446 * the stack of the thread waiting to do the putMVar. See
1447 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1448 * the stack layout, and the PerformPut and PerformTake macros below.
1450 * It is important that a blocked take or put is woken up with the
1451 * take/put already performed, because otherwise there would be a
1452 * small window of vulnerability where the thread could receive an
1453 * exception and never perform its take or put, and we'd end up with a
1456 * -------------------------------------------------------------------------- */
1460 /* args: R1 = MVar closure */
1462 if (StgMVar_value(R1) == stg_END_TSO_QUEUE_closure) {
1474 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
1476 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1477 SET_HDR(mvar,stg_MVAR_DIRTY_info,W_[CCCS]);
1478 // MVARs start dirty: generation 0 has no mutable list
1479 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1480 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1481 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1486 /* If R1 isn't available, pass it on the stack */
1488 #define PerformTake(tso, value) \
1489 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1490 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1492 #define PerformTake(tso, value) \
1493 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1494 W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info;
1497 #define PerformPut(tso,lval) \
1498 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1499 lval = W_[StgTSO_sp(tso) - WDS(1)];
1503 W_ mvar, val, info, tso;
1505 /* args: R1 = MVar closure */
1508 #if defined(THREADED_RTS)
1509 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1511 info = GET_INFO(mvar);
1514 if (info == stg_MVAR_CLEAN_info) {
1515 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1518 /* If the MVar is empty, put ourselves on its blocking queue,
1519 * and wait until we're woken up.
1521 if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1522 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1523 StgMVar_head(mvar) = CurrentTSO;
1525 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1527 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1528 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1529 StgTSO_block_info(CurrentTSO) = mvar;
1530 StgMVar_tail(mvar) = CurrentTSO;
1532 jump stg_block_takemvar;
1535 /* we got the value... */
1536 val = StgMVar_value(mvar);
1538 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1540 /* There are putMVar(s) waiting...
1541 * wake up the first thread on the queue
1543 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1545 /* actually perform the putMVar for the thread that we just woke up */
1546 tso = StgMVar_head(mvar);
1547 PerformPut(tso,StgMVar_value(mvar));
1550 #if defined(GRAN) || defined(PAR)
1551 /* ToDo: check 2nd arg (mvar) is right */
1552 ("ptr" tso) = foreign "C" unblockOne(StgMVar_head(mvar),mvar) [];
1553 StgMVar_head(mvar) = tso;
1555 ("ptr" tso) = foreign "C" unblockOne(MyCapability() "ptr",
1556 StgMVar_head(mvar) "ptr") [];
1557 StgMVar_head(mvar) = tso;
1560 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1561 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1564 #if defined(THREADED_RTS)
1565 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1567 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1573 /* No further putMVars, MVar is now empty */
1574 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1576 #if defined(THREADED_RTS)
1577 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1579 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1589 W_ mvar, val, info, tso;
1591 /* args: R1 = MVar closure */
1595 #if defined(THREADED_RTS)
1596 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1598 info = GET_INFO(mvar);
1601 if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1602 #if defined(THREADED_RTS)
1603 unlockClosure(mvar, info);
1605 /* HACK: we need a pointer to pass back,
1606 * so we abuse NO_FINALIZER_closure
1608 RET_NP(0, stg_NO_FINALIZER_closure);
1611 if (info == stg_MVAR_CLEAN_info) {
1612 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1615 /* we got the value... */
1616 val = StgMVar_value(mvar);
1618 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1620 /* There are putMVar(s) waiting...
1621 * wake up the first thread on the queue
1623 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1625 /* actually perform the putMVar for the thread that we just woke up */
1626 tso = StgMVar_head(mvar);
1627 PerformPut(tso,StgMVar_value(mvar));
1630 #if defined(GRAN) || defined(PAR)
1631 /* ToDo: check 2nd arg (mvar) is right */
1632 ("ptr" tso) = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr") [];
1633 StgMVar_head(mvar) = tso;
1635 ("ptr" tso) = foreign "C" unblockOne(MyCapability() "ptr",
1636 StgMVar_head(mvar) "ptr") [];
1637 StgMVar_head(mvar) = tso;
1640 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1641 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1643 #if defined(THREADED_RTS)
1644 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1646 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1651 /* No further putMVars, MVar is now empty */
1652 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1653 #if defined(THREADED_RTS)
1654 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1656 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1668 /* args: R1 = MVar, R2 = value */
1671 #if defined(THREADED_RTS)
1672 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [R2];
1674 info = GET_INFO(mvar);
1677 if (info == stg_MVAR_CLEAN_info) {
1678 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1681 if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) {
1682 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1683 StgMVar_head(mvar) = CurrentTSO;
1685 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1687 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1688 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1689 StgTSO_block_info(CurrentTSO) = mvar;
1690 StgMVar_tail(mvar) = CurrentTSO;
1692 jump stg_block_putmvar;
1695 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1697 /* There are takeMVar(s) waiting: wake up the first one
1699 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1701 /* actually perform the takeMVar */
1702 tso = StgMVar_head(mvar);
1703 PerformTake(tso, R2);
1706 #if defined(GRAN) || defined(PAR)
1707 /* ToDo: check 2nd arg (mvar) is right */
1708 ("ptr" tso) = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") [];
1709 StgMVar_head(mvar) = tso;
1711 ("ptr" tso) = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") [];
1712 StgMVar_head(mvar) = tso;
1715 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1716 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1719 #if defined(THREADED_RTS)
1720 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1722 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1724 jump %ENTRY_CODE(Sp(0));
1728 /* No further takes, the MVar is now full. */
1729 StgMVar_value(mvar) = R2;
1731 #if defined(THREADED_RTS)
1732 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1734 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1736 jump %ENTRY_CODE(Sp(0));
1739 /* ToDo: yield afterward for better communication performance? */
1747 /* args: R1 = MVar, R2 = value */
1750 #if defined(THREADED_RTS)
1751 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [R2];
1753 info = GET_INFO(mvar);
1756 if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) {
1757 #if defined(THREADED_RTS)
1758 unlockClosure(mvar, info);
1763 if (info == stg_MVAR_CLEAN_info) {
1764 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1767 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1769 /* There are takeMVar(s) waiting: wake up the first one
1771 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1773 /* actually perform the takeMVar */
1774 tso = StgMVar_head(mvar);
1775 PerformTake(tso, R2);
1778 #if defined(GRAN) || defined(PAR)
1779 /* ToDo: check 2nd arg (mvar) is right */
1780 ("ptr" tso) = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") [];
1781 StgMVar_head(mvar) = tso;
1783 ("ptr" tso) = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") [];
1784 StgMVar_head(mvar) = tso;
1787 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1788 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1791 #if defined(THREADED_RTS)
1792 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1794 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1799 /* No further takes, the MVar is now full. */
1800 StgMVar_value(mvar) = R2;
1802 #if defined(THREADED_RTS)
1803 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1805 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1810 /* ToDo: yield afterward for better communication performance? */
1814 /* -----------------------------------------------------------------------------
1815 Stable pointer primitives
1816 ------------------------------------------------------------------------- */
1818 makeStableNamezh_fast
1822 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1824 (index) = foreign "C" lookupStableName(R1 "ptr") [];
1826 /* Is there already a StableName for this heap object?
1827 * stable_ptr_table is a pointer to an array of snEntry structs.
1829 if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) {
1830 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1831 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1832 StgStableName_sn(sn_obj) = index;
1833 snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj;
1835 sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry);
1842 makeStablePtrzh_fast
1846 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1847 ("ptr" sp) = foreign "C" getStablePtr(R1 "ptr") [];
1851 deRefStablePtrzh_fast
1853 /* Args: R1 = the stable ptr */
1856 r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry);
1860 /* -----------------------------------------------------------------------------
1861 Bytecode object primitives
1862 ------------------------------------------------------------------------- */
1872 W_ bco, bitmap_arr, bytes, words;
1876 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1879 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R5_PTR, newBCOzh_fast );
1881 bco = Hp - bytes + WDS(1);
1882 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1884 StgBCO_instrs(bco) = R1;
1885 StgBCO_literals(bco) = R2;
1886 StgBCO_ptrs(bco) = R3;
1887 StgBCO_arity(bco) = HALF_W_(R4);
1888 StgBCO_size(bco) = HALF_W_(words);
1890 // Copy the arity/bitmap info into the BCO
1894 if (i < StgArrWords_words(bitmap_arr)) {
1895 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1906 // R1 = the BCO# for the AP
1910 // This function is *only* used to wrap zero-arity BCOs in an
1911 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1912 // saturated and always points directly to a FUN or BCO.
1913 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
1914 StgBCO_arity(R1) == HALF_W_(0));
1916 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1917 TICK_ALLOC_UP_THK(0, 0);
1918 CCCS_ALLOC(SIZEOF_StgAP);
1920 ap = Hp - SIZEOF_StgAP + WDS(1);
1921 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1923 StgAP_n_args(ap) = HALF_W_(0);
1929 unpackClosurezh_fast
1931 /* args: R1 = closure to analyze */
1932 // TODO: Consider the absence of ptrs or nonptrs as a special case ?
1934 W_ info, ptrs, nptrs, p, ptrs_arr, nptrs_arr;
1935 info = %GET_STD_INFO(UNTAG(R1));
1937 // Some closures have non-standard layout, so we omit those here.
1939 type = TO_W_(%INFO_TYPE(info));
1940 switch [0 .. N_CLOSURE_TYPES] type {
1941 case THUNK_SELECTOR : {
1946 case THUNK, THUNK_1_0, THUNK_0_1, THUNK_2_0, THUNK_1_1,
1947 THUNK_0_2, THUNK_STATIC, AP, PAP, AP_STACK, BCO : {
1953 ptrs = TO_W_(%INFO_PTRS(info));
1954 nptrs = TO_W_(%INFO_NPTRS(info));
1959 W_ ptrs_arr_sz, nptrs_arr_sz;
1960 nptrs_arr_sz = SIZEOF_StgArrWords + WDS(nptrs);
1961 ptrs_arr_sz = SIZEOF_StgMutArrPtrs + WDS(ptrs);
1963 ALLOC_PRIM (ptrs_arr_sz + nptrs_arr_sz, R1_PTR, unpackClosurezh_fast);
1968 ptrs_arr = Hp - nptrs_arr_sz - ptrs_arr_sz + WDS(1);
1969 nptrs_arr = Hp - nptrs_arr_sz + WDS(1);
1971 SET_HDR(ptrs_arr, stg_MUT_ARR_PTRS_FROZEN_info, W_[CCCS]);
1972 StgMutArrPtrs_ptrs(ptrs_arr) = ptrs;
1976 W_[ptrs_arr + SIZEOF_StgMutArrPtrs + WDS(p)] = StgClosure_payload(clos,p);
1981 SET_HDR(nptrs_arr, stg_ARR_WORDS_info, W_[CCCS]);
1982 StgArrWords_words(nptrs_arr) = nptrs;
1986 W_[BYTE_ARR_CTS(nptrs_arr) + WDS(p)] = StgClosure_payload(clos, p+ptrs);
1990 RET_NPP(info, ptrs_arr, nptrs_arr);
1993 /* -----------------------------------------------------------------------------
1994 Thread I/O blocking primitives
1995 -------------------------------------------------------------------------- */
1997 /* Add a thread to the end of the blocked queue. (C-- version of the C
1998 * macro in Schedule.h).
2000 #define APPEND_TO_BLOCKED_QUEUE(tso) \
2001 ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \
2002 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
2003 W_[blocked_queue_hd] = tso; \
2005 StgTSO_link(W_[blocked_queue_tl]) = tso; \
2007 W_[blocked_queue_tl] = tso;
2013 foreign "C" barf("waitRead# on threaded RTS") never returns;
2016 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2017 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
2018 StgTSO_block_info(CurrentTSO) = R1;
2019 // No locking - we're not going to use this interface in the
2020 // threaded RTS anyway.
2021 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2022 jump stg_block_noregs;
2030 foreign "C" barf("waitWrite# on threaded RTS") never returns;
2033 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2034 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2035 StgTSO_block_info(CurrentTSO) = R1;
2036 // No locking - we're not going to use this interface in the
2037 // threaded RTS anyway.
2038 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2039 jump stg_block_noregs;
2044 STRING(stg_delayzh_malloc_str, "delayzh_fast")
2047 #ifdef mingw32_HOST_OS
2055 foreign "C" barf("delay# on threaded RTS") never returns;
2058 /* args: R1 (microsecond delay amount) */
2059 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2060 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
2062 #ifdef mingw32_HOST_OS
2064 /* could probably allocate this on the heap instead */
2065 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2066 stg_delayzh_malloc_str);
2067 (reqID) = foreign "C" addDelayRequest(R1);
2068 StgAsyncIOResult_reqID(ares) = reqID;
2069 StgAsyncIOResult_len(ares) = 0;
2070 StgAsyncIOResult_errCode(ares) = 0;
2071 StgTSO_block_info(CurrentTSO) = ares;
2073 /* Having all async-blocked threads reside on the blocked_queue
2074 * simplifies matters, so change the status to OnDoProc put the
2075 * delayed thread on the blocked_queue.
2077 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2078 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2079 jump stg_block_async_void;
2085 (time) = foreign "C" getourtimeofday() [R1];
2086 divisor = TO_W_(RtsFlags_MiscFlags_tickInterval(RtsFlags))*1000;
2087 target = ((R1 + divisor - 1) / divisor) /* divide rounding up */
2088 + time + 1; /* Add 1 as getourtimeofday rounds down */
2089 StgTSO_block_info(CurrentTSO) = target;
2091 /* Insert the new thread in the sleeping queue. */
2093 t = W_[sleeping_queue];
2095 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
2101 StgTSO_link(CurrentTSO) = t;
2103 W_[sleeping_queue] = CurrentTSO;
2105 StgTSO_link(prev) = CurrentTSO;
2107 jump stg_block_noregs;
2109 #endif /* !THREADED_RTS */
2113 #ifdef mingw32_HOST_OS
2114 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
2121 foreign "C" barf("asyncRead# on threaded RTS") never returns;
2124 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2125 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2126 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
2128 /* could probably allocate this on the heap instead */
2129 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2130 stg_asyncReadzh_malloc_str)
2132 (reqID) = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr") [];
2133 StgAsyncIOResult_reqID(ares) = reqID;
2134 StgAsyncIOResult_len(ares) = 0;
2135 StgAsyncIOResult_errCode(ares) = 0;
2136 StgTSO_block_info(CurrentTSO) = ares;
2137 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2138 jump stg_block_async;
2142 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
2149 foreign "C" barf("asyncWrite# on threaded RTS") never returns;
2152 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2153 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2154 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2156 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2157 stg_asyncWritezh_malloc_str)
2159 (reqID) = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr") [];
2161 StgAsyncIOResult_reqID(ares) = reqID;
2162 StgAsyncIOResult_len(ares) = 0;
2163 StgAsyncIOResult_errCode(ares) = 0;
2164 StgTSO_block_info(CurrentTSO) = ares;
2165 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2166 jump stg_block_async;
2170 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
2177 foreign "C" barf("asyncDoProc# on threaded RTS") never returns;
2180 /* args: R1 = proc, R2 = param */
2181 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2182 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2184 /* could probably allocate this on the heap instead */
2185 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2186 stg_asyncDoProczh_malloc_str)
2188 (reqID) = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr") [];
2189 StgAsyncIOResult_reqID(ares) = reqID;
2190 StgAsyncIOResult_len(ares) = 0;
2191 StgAsyncIOResult_errCode(ares) = 0;
2192 StgTSO_block_info(CurrentTSO) = ares;
2193 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2194 jump stg_block_async;
2199 // noDuplicate# tries to ensure that none of the thunks under
2200 // evaluation by the current thread are also under evaluation by
2201 // another thread. It relies on *both* threads doing noDuplicate#;
2202 // the second one will get blocked if they are duplicating some work.
2205 SAVE_THREAD_STATE();
2206 ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16);
2207 foreign "C" threadPaused (MyCapability() "ptr", CurrentTSO "ptr") [];
2209 if (StgTSO_what_next(CurrentTSO) == ThreadKilled::I16) {
2210 jump stg_threadFinished;
2212 LOAD_THREAD_STATE();
2213 ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16);
2214 jump %ENTRY_CODE(Sp(0));
2218 getApStackValzh_fast
2220 W_ ap_stack, offset, val, ok;
2222 /* args: R1 = AP_STACK, R2 = offset */
2226 if (%INFO_PTR(ap_stack) == stg_AP_STACK_info) {
2228 val = StgAP_STACK_payload(ap_stack,offset);