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 * ---------------------------------------------------------------------------*/
30 /*-----------------------------------------------------------------------------
33 Basically just new*Array - the others are all inline macros.
35 The size arg is always passed in R1, and the result returned in R1.
37 The slow entry point is for returning from a heap check, the saved
38 size argument must be re-loaded from the stack.
39 -------------------------------------------------------------------------- */
41 /* for objects that are *less* than the size of a word, make sure we
42 * round up to the nearest word for the size of the array.
47 W_ words, payload_words, n, p;
48 MAYBE_GC(NO_PTRS,newByteArrayzh_fast);
50 payload_words = ROUNDUP_BYTES_TO_WDS(n);
51 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
52 "ptr" p = foreign "C" allocate(words);
53 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
54 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
55 StgArrWords_words(p) = payload_words;
59 newPinnedByteArrayzh_fast
61 W_ words, payload_words, n, p;
63 MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast);
65 payload_words = ROUNDUP_BYTES_TO_WDS(n);
67 // We want an 8-byte aligned array. allocatePinned() gives us
68 // 8-byte aligned memory by default, but we want to align the
69 // *goods* inside the ArrWords object, so we have to check the
70 // size of the ArrWords header and adjust our size accordingly.
71 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
72 if ((SIZEOF_StgArrWords & 7) != 0) {
76 "ptr" p = foreign "C" allocatePinned(words);
77 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
79 // Again, if the ArrWords header isn't a multiple of 8 bytes, we
80 // have to push the object forward one word so that the goods
81 // fall on an 8-byte boundary.
82 if ((SIZEOF_StgArrWords & 7) != 0) {
86 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
87 StgArrWords_words(p) = payload_words;
93 W_ words, n, init, arr, p;
94 /* Args: R1 = words, R2 = initialisation value */
97 MAYBE_GC(R2_PTR,newArrayzh_fast);
99 words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
100 "ptr" arr = foreign "C" allocate(words);
101 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
103 SET_HDR(arr, stg_MUT_ARR_PTRS_info, W_[CCCS]);
104 StgMutArrPtrs_ptrs(arr) = n;
106 // Initialise all elements of the the array with the value in R2
108 p = arr + SIZEOF_StgMutArrPtrs;
110 if (p < arr + WDS(words)) {
119 unsafeThawArrayzh_fast
121 SET_INFO(R1,stg_MUT_ARR_PTRS_info);
123 // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
125 // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
126 // normally doesn't. However, when we freeze a MUT_ARR_PTRS, we leave
127 // it on the mutable list for the GC to remove (removing something from
128 // the mutable list is not easy, because the mut_list is only singly-linked).
130 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
131 // either it is on a mut_list, or it isn't. We adopt the convention that
132 // the mut_link field is NULL if it isn't on a mut_list, and the GC
133 // maintains this invariant.
135 if (StgMutClosure_mut_link(R1) == NULL) {
136 foreign "C" recordMutable(R1 "ptr");
142 /* -----------------------------------------------------------------------------
144 -------------------------------------------------------------------------- */
149 /* Args: R1 = initialisation value */
151 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);
153 mv = Hp - SIZEOF_StgMutVar + WDS(1);
154 SET_HDR(mv,stg_MUT_VAR_info,W_[CCCS]);
155 StgMutVar_var(mv) = R1;
160 atomicModifyMutVarzh_fast
163 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
165 /* If x is the current contents of the MutVar#, then
166 We want to make the new contents point to
170 and the return value is
174 obviously we can share (f x).
176 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
177 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
178 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
182 #define THUNK_1_SIZE (SIZEOF_StgHeader + WDS(MIN_UPD_SIZE))
183 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
185 #define THUNK_1_SIZE (SIZEOF_StgHeader + WDS(1))
186 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
190 #define THUNK_2_SIZE (SIZEOF_StgHeader + WDS(MIN_UPD_SIZE))
191 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
193 #define THUNK_2_SIZE (SIZEOF_StgHeader + WDS(2))
194 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
197 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
199 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);
201 x = StgMutVar_var(R1);
203 TICK_ALLOC_THUNK_2();
204 CCCS_ALLOC(THUNK_2_SIZE);
205 z = Hp - THUNK_2_SIZE + WDS(1);
206 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
207 LDV_RECORD_CREATE(z);
208 StgClosure_payload(z,0) = R2;
209 StgClosure_payload(z,1) = x;
211 TICK_ALLOC_THUNK_1();
212 CCCS_ALLOC(THUNK_1_SIZE);
213 y = z - THUNK_1_SIZE;
214 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
215 LDV_RECORD_CREATE(y);
216 StgClosure_payload(y,0) = z;
218 StgMutVar_var(R1) = y;
220 TICK_ALLOC_THUNK_1();
221 CCCS_ALLOC(THUNK_1_SIZE);
222 r = y - THUNK_1_SIZE;
223 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
224 LDV_RECORD_CREATE(r);
225 StgClosure_payload(r,0) = z;
230 /* -----------------------------------------------------------------------------
231 Foreign Object Primitives
232 -------------------------------------------------------------------------- */
236 /* R1 = ptr to foreign object,
240 ALLOC_PRIM( SIZEOF_StgForeignObj, NO_PTRS, mkForeignObjzh_fast);
242 result = Hp - SIZEOF_StgForeignObj + WDS(1);
243 SET_HDR(result,stg_FOREIGN_info,W_[CCCS]);
244 StgForeignObj_data(result) = R1;
246 /* returns (# s#, ForeignObj# #) */
250 /* -----------------------------------------------------------------------------
251 Weak Pointer Primitives
252 -------------------------------------------------------------------------- */
254 STRING(stg_weak_msg,"New weak pointer at %p\n")
260 R3 = finalizer (or NULL)
265 R3 = stg_NO_FINALIZER_closure;
268 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
270 w = Hp - SIZEOF_StgWeak + WDS(1);
271 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
274 StgWeak_value(w) = R2;
275 StgWeak_finalizer(w) = R3;
277 StgWeak_link(w) = W_[weak_ptr_list];
278 W_[weak_ptr_list] = w;
280 IF_DEBUG(weak, foreign "C" fprintf(stderr,stg_weak_msg,w));
295 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
296 RET_NP(0,stg_NO_FINALIZER_closure);
302 // A weak pointer is inherently used, so we do not need to call
303 // LDV_recordDead_FILL_SLOP_DYNAMIC():
304 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
305 // or, LDV_recordDead():
306 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
307 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
308 // large as weak pointers, so there is no need to fill the slop, either.
309 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
313 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
315 SET_INFO(w,stg_DEAD_WEAK_info);
316 LDV_RECORD_CREATE(w);
318 f = StgWeak_finalizer(w);
320 /* return the finalizer */
321 if (f == stg_NO_FINALIZER_closure) {
322 RET_NP(0,stg_NO_FINALIZER_closure);
334 if (GET_INFO(w) == stg_WEAK_info) {
336 val = StgWeak_value(w);
344 /* -----------------------------------------------------------------------------
345 Arbitrary-precision Integer operations.
347 There are some assumptions in this code that mp_limb_t == W_. This is
348 the case for all the platforms that GHC supports, currently.
349 -------------------------------------------------------------------------- */
353 /* arguments: R1 = Int# */
355 W_ val, s, p; /* to avoid aliasing */
358 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
360 p = Hp - SIZEOF_StgArrWords;
361 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
362 StgArrWords_words(p) = 1;
364 /* mpz_set_si is inlined here, makes things simpler */
377 /* returns (# size :: Int#,
386 /* arguments: R1 = Word# */
388 W_ val, s, p; /* to avoid aliasing */
392 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
394 p = Hp - SIZEOF_StgArrWords;
395 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
396 StgArrWords_words(p) = 1;
405 /* returns (# size :: Int#,
406 data :: ByteArray# #)
413 * 'long long' primops for converting to/from Integers.
416 #ifdef SUPPORT_LONG_LONGS
418 int64ToIntegerzh_fast
420 /* arguments: L1 = Int64# */
423 W_ hi, s, neg, words_needed, p;
428 if ( %ge(val,0x100000000::L_) || %le(val,-0x100000000::L_) ) {
431 // minimum is one word
435 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
436 NO_PTRS, int64ToIntegerzh_fast );
438 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
439 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
440 StgArrWords_words(p) = words_needed;
442 if ( %lt(val,0::L_) ) {
447 hi = TO_W_(val >> 32);
449 if ( words_needed == 2 ) {
454 if ( val != 0::L_ ) {
457 } else /* val==0 */ {
465 /* returns (# size :: Int#,
466 data :: ByteArray# #)
471 word64ToIntegerzh_fast
473 /* arguments: L1 = Word64# */
476 W_ hi, s, words_needed, p;
479 if ( val >= 0x100000000::L_ ) {
485 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
486 NO_PTRS, word64ToIntegerzh_fast );
488 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
489 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
490 StgArrWords_words(p) = words_needed;
492 hi = TO_W_(val >> 32);
493 if ( val >= 0x100000000::L_ ) {
498 if ( val != 0::L_ ) {
501 } else /* val==0 */ {
506 /* returns (# size :: Int#,
507 data :: ByteArray# #)
513 #endif /* SUPPORT_LONG_LONGS */
515 /* ToDo: this is shockingly inefficient */
519 bits8 [SIZEOF_MP_INT];
524 bits8 [SIZEOF_MP_INT];
529 bits8 [SIZEOF_MP_INT];
534 bits8 [SIZEOF_MP_INT];
537 #define GMP_TAKE2_RET1(name,mp_fun) \
542 /* call doYouWantToGC() */ \
543 MAYBE_GC(R2_PTR & R4_PTR, name); \
550 MP_INT__mp_alloc(mp_tmp1) = StgArrWords_words(d1); \
551 MP_INT__mp_size(mp_tmp1) = (s1); \
552 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
553 MP_INT__mp_alloc(mp_tmp2) = StgArrWords_words(d2); \
554 MP_INT__mp_size(mp_tmp2) = (s2); \
555 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
557 foreign "C" mpz_init(result1); \
559 /* Perform the operation */ \
560 foreign "C" mp_fun(result1,mp_tmp1,mp_tmp2); \
562 RET_NP(MP_INT__mp_size(result1), \
563 MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \
566 #define GMP_TAKE1_RET1(name,mp_fun) \
571 /* call doYouWantToGC() */ \
572 MAYBE_GC(R2_PTR, name); \
577 MP_INT__mp_alloc(mp_tmp1) = StgArrWords_words(d1); \
578 MP_INT__mp_size(mp_tmp1) = (s1); \
579 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
581 foreign "C" mpz_init(result1); \
583 /* Perform the operation */ \
584 foreign "C" mp_fun(result1,mp_tmp1); \
586 RET_NP(MP_INT__mp_size(result1), \
587 MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \
590 #define GMP_TAKE2_RET2(name,mp_fun) \
595 /* call doYouWantToGC() */ \
596 MAYBE_GC(R2_PTR & R4_PTR, name); \
603 MP_INT__mp_alloc(mp_tmp1) = StgArrWords_words(d1); \
604 MP_INT__mp_size(mp_tmp1) = (s1); \
605 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
606 MP_INT__mp_alloc(mp_tmp2) = StgArrWords_words(d2); \
607 MP_INT__mp_size(mp_tmp2) = (s2); \
608 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
610 foreign "C" mpz_init(result1); \
611 foreign "C" mpz_init(result2); \
613 /* Perform the operation */ \
614 foreign "C" mp_fun(result1,result2,mp_tmp1,mp_tmp2); \
616 RET_NPNP(MP_INT__mp_size(result1), \
617 MP_INT__mp_d(result1) - SIZEOF_StgArrWords, \
618 MP_INT__mp_size(result2), \
619 MP_INT__mp_d(result2) - SIZEOF_StgArrWords); \
622 GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add)
623 GMP_TAKE2_RET1(minusIntegerzh_fast, mpz_sub)
624 GMP_TAKE2_RET1(timesIntegerzh_fast, mpz_mul)
625 GMP_TAKE2_RET1(gcdIntegerzh_fast, mpz_gcd)
626 GMP_TAKE2_RET1(quotIntegerzh_fast, mpz_tdiv_q)
627 GMP_TAKE2_RET1(remIntegerzh_fast, mpz_tdiv_r)
628 GMP_TAKE2_RET1(divExactIntegerzh_fast, mpz_divexact)
629 GMP_TAKE2_RET1(andIntegerzh_fast, mpz_and)
630 GMP_TAKE2_RET1(orIntegerzh_fast, mpz_ior)
631 GMP_TAKE2_RET1(xorIntegerzh_fast, mpz_xor)
632 GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com)
634 GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr)
635 GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr)
638 aa: W_; // NB. aa is really an mp_limb_t
643 /* R1 = the first Int#; R2 = the second Int# */
647 r = foreign "C" mpn_gcd_1(aa, 1, R2);
650 /* Result parked in R1, return via info-pointer at TOS */
651 jump %ENTRY_CODE(Sp(0));
657 /* R1 = s1; R2 = d1; R3 = the int */
658 R1 = foreign "C" mpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3);
660 /* Result parked in R1, return via info-pointer at TOS */
661 jump %ENTRY_CODE(Sp(0));
667 /* R1 = s1; R2 = d1; R3 = the int */
668 W_ usize, vsize, v_digit, u_digit;
674 // paraphrased from mpz_cmp_si() in the GMP sources
675 if (%gt(v_digit,0)) {
678 if (%lt(v_digit,0)) {
684 if (usize != vsize) {
686 jump %ENTRY_CODE(Sp(0));
691 jump %ENTRY_CODE(Sp(0));
694 u_digit = W_[BYTE_ARR_CTS(R2)];
696 if (u_digit == v_digit) {
698 jump %ENTRY_CODE(Sp(0));
701 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
707 jump %ENTRY_CODE(Sp(0));
712 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
713 W_ usize, vsize, size, up, vp;
716 // paraphrased from mpz_cmp() in the GMP sources
720 if (usize != vsize) {
722 jump %ENTRY_CODE(Sp(0));
727 jump %ENTRY_CODE(Sp(0));
730 if (%lt(usize,0)) { // NB. not <, which is unsigned
736 up = BYTE_ARR_CTS(R2);
737 vp = BYTE_ARR_CTS(R4);
739 cmp = foreign "C" mpn_cmp(up "ptr", vp "ptr", size);
743 jump %ENTRY_CODE(Sp(0));
746 if (%lt(cmp,0) == %lt(usize,0)) {
751 /* Result parked in R1, return via info-pointer at TOS */
752 jump %ENTRY_CODE(Sp(0));
764 r = W_[R2 + SIZEOF_StgArrWords];
769 /* Result parked in R1, return via info-pointer at TOS */
771 jump %ENTRY_CODE(Sp(0));
783 r = W_[R2 + SIZEOF_StgArrWords];
788 /* Result parked in R1, return via info-pointer at TOS */
790 jump %ENTRY_CODE(Sp(0));
802 /* arguments: F1 = Float# */
805 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
807 /* Be prepared to tell Lennart-coded __decodeFloat
808 where mantissa._mp_d can be put (it does not care about the rest) */
809 p = Hp - SIZEOF_StgArrWords;
810 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
811 StgArrWords_words(p) = 1;
812 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
814 /* Perform the operation */
815 foreign "C" __decodeFloat(mp_tmp1,exponent,arg);
817 /* returns: (Int# (expn), Int#, ByteArray#) */
818 RET_NNP(W_[exponent], MP_INT__mp_size(mp_tmp1), p);
821 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
822 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
829 /* arguments: D1 = Double# */
832 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
834 /* Be prepared to tell Lennart-coded __decodeDouble
835 where mantissa.d can be put (it does not care about the rest) */
836 p = Hp - ARR_SIZE + WDS(1);
837 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
838 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
839 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
841 /* Perform the operation */
842 foreign "C" __decodeDouble(mp_tmp1,exponent,arg);
844 /* returns: (Int# (expn), Int#, ByteArray#) */
845 RET_NNP(W_[exponent], MP_INT__mp_size(mp_tmp1), p);
848 /* -----------------------------------------------------------------------------
849 * Concurrency primitives
850 * -------------------------------------------------------------------------- */
854 /* args: R1 = closure to spark */
856 MAYBE_GC(R1_PTR, forkzh_fast);
858 // create it right now, return ThreadID in R1
859 "ptr" R1 = foreign "C" createIOThread( RtsFlags_GcFlags_initialStkSize(RtsFlags),
861 foreign "C" scheduleThread(R1 "ptr");
863 // switch at the earliest opportunity
864 CInt[context_switch] = 1;
871 jump stg_yield_noregs;
886 foreign "C" labelThread(R1 "ptr", R2 "ptr");
888 jump %ENTRY_CODE(Sp(0));
891 isCurrentThreadBoundzh_fast
895 r = foreign "C" isThreadBound(CurrentTSO);
899 /* -----------------------------------------------------------------------------
902 * take & putMVar work as follows. Firstly, an important invariant:
904 * If the MVar is full, then the blocking queue contains only
905 * threads blocked on putMVar, and if the MVar is empty then the
906 * blocking queue contains only threads blocked on takeMVar.
909 * MVar empty : then add ourselves to the blocking queue
910 * MVar full : remove the value from the MVar, and
911 * blocking queue empty : return
912 * blocking queue non-empty : perform the first blocked putMVar
913 * from the queue, and wake up the
914 * thread (MVar is now full again)
916 * putMVar is just the dual of the above algorithm.
918 * How do we "perform a putMVar"? Well, we have to fiddle around with
919 * the stack of the thread waiting to do the putMVar. See
920 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
921 * the stack layout, and the PerformPut and PerformTake macros below.
923 * It is important that a blocked take or put is woken up with the
924 * take/put already performed, because otherwise there would be a
925 * small window of vulnerability where the thread could receive an
926 * exception and never perform its take or put, and we'd end up with a
929 * -------------------------------------------------------------------------- */
933 /* args: R1 = MVar closure */
935 if (GET_INFO(R1) == stg_EMPTY_MVAR_info) {
947 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
949 mvar = Hp - SIZEOF_StgMVar + WDS(1);
950 SET_HDR(mvar,stg_EMPTY_MVAR_info,W_[CCCS]);
951 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
952 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
953 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
958 /* If R1 isn't available, pass it on the stack */
960 #define PerformTake(tso, value) \
961 W_[StgTSO_sp(tso) + WDS(1)] = value; \
962 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
964 #define PerformTake(tso, value) \
965 W_[StgTSO_sp(tso) + WDS(1)] = value; \
966 W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info;
969 #define PerformPut(tso,lval) \
970 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
971 lval = W_[StgTSO_sp(tso) - WDS(1)];
976 W_ mvar, val, info, tso;
978 /* args: R1 = MVar closure */
981 info = GET_INFO(mvar);
983 /* If the MVar is empty, put ourselves on its blocking queue,
984 * and wait until we're woken up.
986 if (info == stg_EMPTY_MVAR_info) {
987 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
988 StgMVar_head(mvar) = CurrentTSO;
990 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
992 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
993 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
994 StgTSO_block_info(CurrentTSO) = mvar;
995 StgMVar_tail(mvar) = CurrentTSO;
997 jump stg_block_takemvar;
1000 /* we got the value... */
1001 val = StgMVar_value(mvar);
1003 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1005 /* There are putMVar(s) waiting...
1006 * wake up the first thread on the queue
1008 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1010 /* actually perform the putMVar for the thread that we just woke up */
1011 tso = StgMVar_head(mvar);
1012 PerformPut(tso,StgMVar_value(mvar));
1014 #if defined(GRAN) || defined(PAR)
1015 /* ToDo: check 2nd arg (mvar) is right */
1016 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar);
1017 StgMVar_head(mvar) = tso;
1019 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1020 StgMVar_head(mvar) = tso;
1022 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1023 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1029 /* No further putMVars, MVar is now empty */
1031 /* do this last... we might have locked the MVar in the SMP case,
1032 * and writing the info pointer will unlock it.
1034 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1035 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1043 W_ mvar, val, info, tso;
1045 /* args: R1 = MVar closure */
1049 info = GET_INFO(mvar);
1051 if (info == stg_EMPTY_MVAR_info) {
1052 /* HACK: we need a pointer to pass back,
1053 * so we abuse NO_FINALIZER_closure
1055 RET_NP(0, stg_NO_FINALIZER_closure);
1058 /* we got the value... */
1059 val = StgMVar_value(mvar);
1061 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1062 /* There are putMVar(s) waiting...
1063 * wake up the first thread on the queue
1065 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1067 /* actually perform the putMVar for the thread that we just woke up */
1068 tso = StgMVar_head(mvar);
1069 PerformPut(tso,StgMVar_value(mvar));
1071 #if defined(GRAN) || defined(PAR)
1072 /* ToDo: check 2nd arg (mvar) is right */
1073 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr");
1074 StgMVar_head(mvar) = tso;
1076 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1077 StgMVar_head(mvar) = tso;
1080 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1081 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1086 /* No further putMVars, MVar is now empty */
1087 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1089 /* do this last... we might have locked the MVar in the SMP case,
1090 * and writing the info pointer will unlock it.
1092 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1103 /* args: R1 = MVar, R2 = value */
1106 info = GET_INFO(mvar);
1108 if (info == stg_FULL_MVAR_info) {
1109 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1110 StgMVar_head(mvar) = CurrentTSO;
1112 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1114 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1115 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1116 StgTSO_block_info(CurrentTSO) = mvar;
1117 StgMVar_tail(mvar) = CurrentTSO;
1119 jump stg_block_putmvar;
1122 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1123 /* There are takeMVar(s) waiting: wake up the first one
1125 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1127 /* actually perform the takeMVar */
1128 tso = StgMVar_head(mvar);
1129 PerformTake(tso, R2);
1131 #if defined(GRAN) || defined(PAR)
1132 /* ToDo: check 2nd arg (mvar) is right */
1133 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1134 StgMVar_head(mvar) = tso;
1136 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1137 StgMVar_head(mvar) = tso;
1140 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1141 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1144 jump %ENTRY_CODE(Sp(0));
1148 /* No further takes, the MVar is now full. */
1149 StgMVar_value(mvar) = R2;
1150 /* unlocks the MVar in the SMP case */
1151 SET_INFO(mvar,stg_FULL_MVAR_info);
1152 jump %ENTRY_CODE(Sp(0));
1155 /* ToDo: yield afterward for better communication performance? */
1163 /* args: R1 = MVar, R2 = value */
1166 info = GET_INFO(mvar);
1168 if (info == stg_FULL_MVAR_info) {
1172 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1173 /* There are takeMVar(s) waiting: wake up the first one
1175 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1177 /* actually perform the takeMVar */
1178 tso = StgMVar_head(mvar);
1179 PerformTake(tso, R2);
1181 #if defined(GRAN) || defined(PAR)
1182 /* ToDo: check 2nd arg (mvar) is right */
1183 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1184 StgMVar_head(mvar) = tso;
1186 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1187 StgMVar_head(mvar) = tso;
1190 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1191 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1194 jump %ENTRY_CODE(Sp(0));
1198 /* No further takes, the MVar is now full. */
1199 StgMVar_value(mvar) = R2;
1200 /* unlocks the MVar in the SMP case */
1201 SET_INFO(mvar,stg_FULL_MVAR_info);
1202 jump %ENTRY_CODE(Sp(0));
1205 /* ToDo: yield afterward for better communication performance? */
1209 /* -----------------------------------------------------------------------------
1210 Stable pointer primitives
1211 ------------------------------------------------------------------------- */
1213 makeStableNamezh_fast
1217 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1219 index = foreign "C" lookupStableName(R1 "ptr");
1221 /* Is there already a StableName for this heap object?
1222 * stable_ptr_table is an array of snEntry structs.
1224 if ( snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry) == NULL ) {
1225 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1226 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1227 StgStableName_sn(sn_obj) = index;
1228 snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry) = sn_obj;
1230 sn_obj = snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry);
1237 makeStablePtrzh_fast
1241 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1242 "ptr" sp = foreign "C" getStablePtr(R1 "ptr");
1246 deRefStablePtrzh_fast
1248 /* Args: R1 = the stable ptr */
1251 r = snEntry_addr(stable_ptr_table + sp*SIZEOF_snEntry);
1255 /* -----------------------------------------------------------------------------
1256 Bytecode object primitives
1257 ------------------------------------------------------------------------- */
1268 W_ bco, bitmap_arr, bytes, words;
1271 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1274 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R4_PTR&R6_PTR, newBCOzh_fast );
1276 bco = Hp - bytes + WDS(1);
1277 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1279 StgBCO_instrs(bco) = R1;
1280 StgBCO_literals(bco) = R2;
1281 StgBCO_ptrs(bco) = R3;
1282 StgBCO_itbls(bco) = R4;
1283 StgBCO_arity(bco) = HALF_W_(R5);
1284 StgBCO_size(bco) = HALF_W_(words);
1286 // Copy the arity/bitmap info into the BCO
1290 if (i < StgArrWords_words(bitmap_arr)) {
1291 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1302 // R1 = the BCO# for the AP
1306 // This function is *only* used to wrap zero-arity BCOs in an
1307 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1308 // saturated and always points directly to a FUN or BCO.
1309 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == BCO::I16 &&
1310 StgBCO_arity(R1) == 0::I16);
1312 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1313 TICK_ALLOC_UP_THK(0, 0);
1314 CCCS_ALLOC(SIZEOF_StgAP);
1316 ap = Hp - SIZEOF_StgAP + WDS(1);
1317 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1319 StgAP_n_args(ap) = 0::I16;
1325 /* -----------------------------------------------------------------------------
1326 Thread I/O blocking primitives
1327 -------------------------------------------------------------------------- */
1329 /* Add a thread to the end of the blocked queue. (C-- version of the C
1330 * macro in Schedule.h).
1332 #define APPEND_TO_BLOCKED_QUEUE(tso) \
1333 ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \
1334 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
1335 W_[blocked_queue_hd] = tso; \
1337 StgTSO_link(W_[blocked_queue_tl]) = tso; \
1339 W_[blocked_queue_tl] = tso;
1344 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1345 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1346 StgTSO_block_info(CurrentTSO) = R1;
1347 // No locking - we're not going to use this interface in the
1348 // threaded RTS anyway.
1349 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1350 jump stg_block_noregs;
1356 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1357 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1358 StgTSO_block_info(CurrentTSO) = R1;
1359 // No locking - we're not going to use this interface in the
1360 // threaded RTS anyway.
1361 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1362 jump stg_block_noregs;
1366 STRING(stg_delayzh_malloc_str, "delayzh_fast")
1369 #ifdef mingw32_TARGET_OS
1376 /* args: R1 (microsecond delay amount) */
1377 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1378 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
1380 #ifdef mingw32_TARGET_OS
1382 /* could probably allocate this on the heap instead */
1383 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1384 stg_delayzh_malloc_str);
1385 reqID = foreign "C" addDelayRequest(R1);
1386 StgAsyncIOResult_reqID(ares) = reqID;
1387 StgAsyncIOResult_len(ares) = 0;
1388 StgAsyncIOResult_errCode(ares) = 0;
1389 StgTSO_block_info(CurrentTSO) = ares;
1391 /* Having all async-blocked threads reside on the blocked_queue
1392 * simplifies matters, so change the status to OnDoProc put the
1393 * delayed thread on the blocked_queue.
1395 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1396 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1401 time = foreign "C" getourtimeofday();
1402 target = (R1 / (TICK_MILLISECS*1000)) + TO_W_(time);
1403 StgTSO_block_info(CurrentTSO) = target;
1405 /* Insert the new thread in the sleeping queue. */
1407 t = W_[sleeping_queue];
1409 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
1415 StgTSO_link(CurrentTSO) = t;
1417 W_[sleeping_queue] = CurrentTSO;
1419 StgTSO_link(prev) = CurrentTSO;
1423 jump stg_block_noregs;
1427 #ifdef mingw32_TARGET_OS
1428 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
1434 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1435 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1436 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1438 /* could probably allocate this on the heap instead */
1439 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1440 stg_asyncReadzh_malloc_str);
1441 reqID = foreign "C" addIORequest(R1,FALSE,R2,R3,R4);
1442 StgAsyncIOResult_reqID(ares) = reqID;
1443 StgAsyncIOResult_len(ares) = 0;
1444 StgAsyncIOResult_errCode(ares) = 0;
1445 StgTSO_block_info(CurrentTSO) = ares;
1446 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1447 jump stg_block_async;
1450 STRING(asyncWritezh_malloc_str, "asyncWritezh_fast")
1456 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1457 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1458 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1460 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1461 asyncWritezh_malloc_str);
1462 reqID = foreign "C" addIORequest(R1,TRUE,R2,R3,R4);
1464 StgAsyncIOResult_reqID(ares) = reqID;
1465 StgAsyncIOResult_len(ares) = 0;
1466 StgAsyncIOResult_errCode(ares) = 0;
1467 StgTSO_block_info(CurrentTSO) = ares;
1468 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1469 jump stg_block_async;
1472 STRING(asyncDoProczh_malloc_str, "asyncDoProczh_fast")
1478 /* args: R1 = proc, R2 = param */
1479 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1480 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1482 /* could probably allocate this on the heap instead */
1483 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1484 asyncDoProczh_malloc_str);
1485 reqID = foreign "C" addDoProcRequest(R1,R2);
1486 StgAsyncIOResult_reqID(ares) = reqID;
1487 StgAsyncIOResult_len(ares) = 0;
1488 StgAsyncIOResult_errCode(ares) = 0;
1489 StgTSO_block_info(CurrentTSO) = ares;
1490 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1491 jump stg_block_async;
1495 /* -----------------------------------------------------------------------------
1498 classes CCallable and CReturnable don't really exist, but the
1499 compiler insists on generating dictionaries containing references
1500 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
1501 for these. Some C compilers can't cope with zero-length static arrays,
1502 so we have to make these one element long.
1503 --------------------------------------------------------------------------- */
1506 GHC_ZCCCallable_static_info: W_ 0;
1510 GHC_ZCCReturnable_static_info: W_ 0;