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);
319 StgDeadWeak_link(w) = StgWeak_link(w);
321 /* return the finalizer */
322 if (f == stg_NO_FINALIZER_closure) {
323 RET_NP(0,stg_NO_FINALIZER_closure);
335 if (GET_INFO(w) == stg_WEAK_info) {
337 val = StgWeak_value(w);
345 /* -----------------------------------------------------------------------------
346 Arbitrary-precision Integer operations.
348 There are some assumptions in this code that mp_limb_t == W_. This is
349 the case for all the platforms that GHC supports, currently.
350 -------------------------------------------------------------------------- */
354 /* arguments: R1 = Int# */
356 W_ val, s, p; /* to avoid aliasing */
359 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
361 p = Hp - SIZEOF_StgArrWords;
362 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
363 StgArrWords_words(p) = 1;
365 /* mpz_set_si is inlined here, makes things simpler */
378 /* returns (# size :: Int#,
387 /* arguments: R1 = Word# */
389 W_ val, s, p; /* to avoid aliasing */
393 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
395 p = Hp - SIZEOF_StgArrWords;
396 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
397 StgArrWords_words(p) = 1;
406 /* returns (# size :: Int#,
407 data :: ByteArray# #)
414 * 'long long' primops for converting to/from Integers.
417 #ifdef SUPPORT_LONG_LONGS
419 int64ToIntegerzh_fast
421 /* arguments: L1 = Int64# */
424 W_ hi, s, neg, words_needed, p;
429 if ( %ge(val,0x100000000::L_) || %le(val,-0x100000000::L_) ) {
432 // minimum is one word
436 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
437 NO_PTRS, int64ToIntegerzh_fast );
439 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
440 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
441 StgArrWords_words(p) = words_needed;
443 if ( %lt(val,0::L_) ) {
448 hi = TO_W_(val >> 32);
450 if ( words_needed == 2 ) {
455 if ( val != 0::L_ ) {
458 } else /* val==0 */ {
466 /* returns (# size :: Int#,
467 data :: ByteArray# #)
472 word64ToIntegerzh_fast
474 /* arguments: L1 = Word64# */
477 W_ hi, s, words_needed, p;
480 if ( val >= 0x100000000::L_ ) {
486 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
487 NO_PTRS, word64ToIntegerzh_fast );
489 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
490 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
491 StgArrWords_words(p) = words_needed;
493 hi = TO_W_(val >> 32);
494 if ( val >= 0x100000000::L_ ) {
499 if ( val != 0::L_ ) {
502 } else /* val==0 */ {
507 /* returns (# size :: Int#,
508 data :: ByteArray# #)
514 #endif /* SUPPORT_LONG_LONGS */
516 /* ToDo: this is shockingly inefficient */
520 bits8 [SIZEOF_MP_INT];
525 bits8 [SIZEOF_MP_INT];
530 bits8 [SIZEOF_MP_INT];
535 bits8 [SIZEOF_MP_INT];
538 #define GMP_TAKE2_RET1(name,mp_fun) \
543 /* call doYouWantToGC() */ \
544 MAYBE_GC(R2_PTR & R4_PTR, name); \
551 MP_INT__mp_alloc(mp_tmp1) = StgArrWords_words(d1); \
552 MP_INT__mp_size(mp_tmp1) = (s1); \
553 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
554 MP_INT__mp_alloc(mp_tmp2) = StgArrWords_words(d2); \
555 MP_INT__mp_size(mp_tmp2) = (s2); \
556 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
558 foreign "C" mpz_init(result1); \
560 /* Perform the operation */ \
561 foreign "C" mp_fun(result1,mp_tmp1,mp_tmp2); \
563 RET_NP(MP_INT__mp_size(result1), \
564 MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \
567 #define GMP_TAKE1_RET1(name,mp_fun) \
572 /* call doYouWantToGC() */ \
573 MAYBE_GC(R2_PTR, name); \
578 MP_INT__mp_alloc(mp_tmp1) = StgArrWords_words(d1); \
579 MP_INT__mp_size(mp_tmp1) = (s1); \
580 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
582 foreign "C" mpz_init(result1); \
584 /* Perform the operation */ \
585 foreign "C" mp_fun(result1,mp_tmp1); \
587 RET_NP(MP_INT__mp_size(result1), \
588 MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \
591 #define GMP_TAKE2_RET2(name,mp_fun) \
596 /* call doYouWantToGC() */ \
597 MAYBE_GC(R2_PTR & R4_PTR, name); \
604 MP_INT__mp_alloc(mp_tmp1) = StgArrWords_words(d1); \
605 MP_INT__mp_size(mp_tmp1) = (s1); \
606 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
607 MP_INT__mp_alloc(mp_tmp2) = StgArrWords_words(d2); \
608 MP_INT__mp_size(mp_tmp2) = (s2); \
609 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
611 foreign "C" mpz_init(result1); \
612 foreign "C" mpz_init(result2); \
614 /* Perform the operation */ \
615 foreign "C" mp_fun(result1,result2,mp_tmp1,mp_tmp2); \
617 RET_NPNP(MP_INT__mp_size(result1), \
618 MP_INT__mp_d(result1) - SIZEOF_StgArrWords, \
619 MP_INT__mp_size(result2), \
620 MP_INT__mp_d(result2) - SIZEOF_StgArrWords); \
623 GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add)
624 GMP_TAKE2_RET1(minusIntegerzh_fast, mpz_sub)
625 GMP_TAKE2_RET1(timesIntegerzh_fast, mpz_mul)
626 GMP_TAKE2_RET1(gcdIntegerzh_fast, mpz_gcd)
627 GMP_TAKE2_RET1(quotIntegerzh_fast, mpz_tdiv_q)
628 GMP_TAKE2_RET1(remIntegerzh_fast, mpz_tdiv_r)
629 GMP_TAKE2_RET1(divExactIntegerzh_fast, mpz_divexact)
630 GMP_TAKE2_RET1(andIntegerzh_fast, mpz_and)
631 GMP_TAKE2_RET1(orIntegerzh_fast, mpz_ior)
632 GMP_TAKE2_RET1(xorIntegerzh_fast, mpz_xor)
633 GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com)
635 GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr)
636 GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr)
639 aa: W_; // NB. aa is really an mp_limb_t
644 /* R1 = the first Int#; R2 = the second Int# */
648 r = foreign "C" mpn_gcd_1(aa, 1, R2);
651 /* Result parked in R1, return via info-pointer at TOS */
652 jump %ENTRY_CODE(Sp(0));
658 /* R1 = s1; R2 = d1; R3 = the int */
659 R1 = foreign "C" mpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3);
661 /* Result parked in R1, return via info-pointer at TOS */
662 jump %ENTRY_CODE(Sp(0));
668 /* R1 = s1; R2 = d1; R3 = the int */
669 W_ usize, vsize, v_digit, u_digit;
675 // paraphrased from mpz_cmp_si() in the GMP sources
676 if (%gt(v_digit,0)) {
679 if (%lt(v_digit,0)) {
685 if (usize != vsize) {
687 jump %ENTRY_CODE(Sp(0));
692 jump %ENTRY_CODE(Sp(0));
695 u_digit = W_[BYTE_ARR_CTS(R2)];
697 if (u_digit == v_digit) {
699 jump %ENTRY_CODE(Sp(0));
702 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
708 jump %ENTRY_CODE(Sp(0));
713 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
714 W_ usize, vsize, size, up, vp;
717 // paraphrased from mpz_cmp() in the GMP sources
721 if (usize != vsize) {
723 jump %ENTRY_CODE(Sp(0));
728 jump %ENTRY_CODE(Sp(0));
731 if (%lt(usize,0)) { // NB. not <, which is unsigned
737 up = BYTE_ARR_CTS(R2);
738 vp = BYTE_ARR_CTS(R4);
740 cmp = foreign "C" mpn_cmp(up "ptr", vp "ptr", size);
744 jump %ENTRY_CODE(Sp(0));
747 if (%lt(cmp,0) == %lt(usize,0)) {
752 /* Result parked in R1, return via info-pointer at TOS */
753 jump %ENTRY_CODE(Sp(0));
765 r = W_[R2 + SIZEOF_StgArrWords];
770 /* Result parked in R1, return via info-pointer at TOS */
772 jump %ENTRY_CODE(Sp(0));
784 r = W_[R2 + SIZEOF_StgArrWords];
789 /* Result parked in R1, return via info-pointer at TOS */
791 jump %ENTRY_CODE(Sp(0));
803 /* arguments: F1 = Float# */
806 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
808 /* Be prepared to tell Lennart-coded __decodeFloat
809 where mantissa._mp_d can be put (it does not care about the rest) */
810 p = Hp - SIZEOF_StgArrWords;
811 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
812 StgArrWords_words(p) = 1;
813 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
815 /* Perform the operation */
816 foreign "C" __decodeFloat(mp_tmp1,exponent,arg);
818 /* returns: (Int# (expn), Int#, ByteArray#) */
819 RET_NNP(W_[exponent], MP_INT__mp_size(mp_tmp1), p);
822 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
823 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
830 /* arguments: D1 = Double# */
833 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
835 /* Be prepared to tell Lennart-coded __decodeDouble
836 where mantissa.d can be put (it does not care about the rest) */
837 p = Hp - ARR_SIZE + WDS(1);
838 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
839 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
840 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
842 /* Perform the operation */
843 foreign "C" __decodeDouble(mp_tmp1,exponent,arg);
845 /* returns: (Int# (expn), Int#, ByteArray#) */
846 RET_NNP(W_[exponent], MP_INT__mp_size(mp_tmp1), p);
849 /* -----------------------------------------------------------------------------
850 * Concurrency primitives
851 * -------------------------------------------------------------------------- */
855 /* args: R1 = closure to spark */
857 MAYBE_GC(R1_PTR, forkzh_fast);
859 // create it right now, return ThreadID in R1
860 "ptr" R1 = foreign "C" createIOThread( RtsFlags_GcFlags_initialStkSize(RtsFlags),
862 foreign "C" scheduleThread(R1 "ptr");
864 // switch at the earliest opportunity
865 CInt[context_switch] = 1;
872 jump stg_yield_noregs;
887 foreign "C" labelThread(R1 "ptr", R2 "ptr");
889 jump %ENTRY_CODE(Sp(0));
892 isCurrentThreadBoundzh_fast
896 r = foreign "C" isThreadBound(CurrentTSO);
900 /* -----------------------------------------------------------------------------
903 * take & putMVar work as follows. Firstly, an important invariant:
905 * If the MVar is full, then the blocking queue contains only
906 * threads blocked on putMVar, and if the MVar is empty then the
907 * blocking queue contains only threads blocked on takeMVar.
910 * MVar empty : then add ourselves to the blocking queue
911 * MVar full : remove the value from the MVar, and
912 * blocking queue empty : return
913 * blocking queue non-empty : perform the first blocked putMVar
914 * from the queue, and wake up the
915 * thread (MVar is now full again)
917 * putMVar is just the dual of the above algorithm.
919 * How do we "perform a putMVar"? Well, we have to fiddle around with
920 * the stack of the thread waiting to do the putMVar. See
921 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
922 * the stack layout, and the PerformPut and PerformTake macros below.
924 * It is important that a blocked take or put is woken up with the
925 * take/put already performed, because otherwise there would be a
926 * small window of vulnerability where the thread could receive an
927 * exception and never perform its take or put, and we'd end up with a
930 * -------------------------------------------------------------------------- */
934 /* args: R1 = MVar closure */
936 if (GET_INFO(R1) == stg_EMPTY_MVAR_info) {
948 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
950 mvar = Hp - SIZEOF_StgMVar + WDS(1);
951 SET_HDR(mvar,stg_EMPTY_MVAR_info,W_[CCCS]);
952 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
953 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
954 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
959 /* If R1 isn't available, pass it on the stack */
961 #define PerformTake(tso, value) \
962 W_[StgTSO_sp(tso) + WDS(1)] = value; \
963 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
965 #define PerformTake(tso, value) \
966 W_[StgTSO_sp(tso) + WDS(1)] = value; \
967 W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info;
970 #define PerformPut(tso,lval) \
971 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
972 lval = W_[StgTSO_sp(tso) - WDS(1)];
977 W_ mvar, val, info, tso;
979 /* args: R1 = MVar closure */
982 info = GET_INFO(mvar);
984 /* If the MVar is empty, put ourselves on its blocking queue,
985 * and wait until we're woken up.
987 if (info == stg_EMPTY_MVAR_info) {
988 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
989 StgMVar_head(mvar) = CurrentTSO;
991 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
993 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
994 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
995 StgTSO_block_info(CurrentTSO) = mvar;
996 StgMVar_tail(mvar) = CurrentTSO;
998 jump stg_block_takemvar;
1001 /* we got the value... */
1002 val = StgMVar_value(mvar);
1004 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1006 /* There are putMVar(s) waiting...
1007 * wake up the first thread on the queue
1009 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1011 /* actually perform the putMVar for the thread that we just woke up */
1012 tso = StgMVar_head(mvar);
1013 PerformPut(tso,StgMVar_value(mvar));
1015 #if defined(GRAN) || defined(PAR)
1016 /* ToDo: check 2nd arg (mvar) is right */
1017 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar);
1018 StgMVar_head(mvar) = tso;
1020 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1021 StgMVar_head(mvar) = tso;
1023 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1024 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1030 /* No further putMVars, MVar is now empty */
1032 /* do this last... we might have locked the MVar in the SMP case,
1033 * and writing the info pointer will unlock it.
1035 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1036 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1044 W_ mvar, val, info, tso;
1046 /* args: R1 = MVar closure */
1050 info = GET_INFO(mvar);
1052 if (info == stg_EMPTY_MVAR_info) {
1053 /* HACK: we need a pointer to pass back,
1054 * so we abuse NO_FINALIZER_closure
1056 RET_NP(0, stg_NO_FINALIZER_closure);
1059 /* we got the value... */
1060 val = StgMVar_value(mvar);
1062 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1063 /* There are putMVar(s) waiting...
1064 * wake up the first thread on the queue
1066 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1068 /* actually perform the putMVar for the thread that we just woke up */
1069 tso = StgMVar_head(mvar);
1070 PerformPut(tso,StgMVar_value(mvar));
1072 #if defined(GRAN) || defined(PAR)
1073 /* ToDo: check 2nd arg (mvar) is right */
1074 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr");
1075 StgMVar_head(mvar) = tso;
1077 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1078 StgMVar_head(mvar) = tso;
1081 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1082 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1087 /* No further putMVars, MVar is now empty */
1088 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1090 /* do this last... we might have locked the MVar in the SMP case,
1091 * and writing the info pointer will unlock it.
1093 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1104 /* args: R1 = MVar, R2 = value */
1107 info = GET_INFO(mvar);
1109 if (info == stg_FULL_MVAR_info) {
1110 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1111 StgMVar_head(mvar) = CurrentTSO;
1113 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1115 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1116 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1117 StgTSO_block_info(CurrentTSO) = mvar;
1118 StgMVar_tail(mvar) = CurrentTSO;
1120 jump stg_block_putmvar;
1123 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1124 /* There are takeMVar(s) waiting: wake up the first one
1126 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1128 /* actually perform the takeMVar */
1129 tso = StgMVar_head(mvar);
1130 PerformTake(tso, R2);
1132 #if defined(GRAN) || defined(PAR)
1133 /* ToDo: check 2nd arg (mvar) is right */
1134 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1135 StgMVar_head(mvar) = tso;
1137 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1138 StgMVar_head(mvar) = tso;
1141 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1142 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1145 jump %ENTRY_CODE(Sp(0));
1149 /* No further takes, the MVar is now full. */
1150 StgMVar_value(mvar) = R2;
1151 /* unlocks the MVar in the SMP case */
1152 SET_INFO(mvar,stg_FULL_MVAR_info);
1153 jump %ENTRY_CODE(Sp(0));
1156 /* ToDo: yield afterward for better communication performance? */
1164 /* args: R1 = MVar, R2 = value */
1167 info = GET_INFO(mvar);
1169 if (info == stg_FULL_MVAR_info) {
1173 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1174 /* There are takeMVar(s) waiting: wake up the first one
1176 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1178 /* actually perform the takeMVar */
1179 tso = StgMVar_head(mvar);
1180 PerformTake(tso, R2);
1182 #if defined(GRAN) || defined(PAR)
1183 /* ToDo: check 2nd arg (mvar) is right */
1184 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1185 StgMVar_head(mvar) = tso;
1187 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1188 StgMVar_head(mvar) = tso;
1191 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1192 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1195 jump %ENTRY_CODE(Sp(0));
1199 /* No further takes, the MVar is now full. */
1200 StgMVar_value(mvar) = R2;
1201 /* unlocks the MVar in the SMP case */
1202 SET_INFO(mvar,stg_FULL_MVAR_info);
1203 jump %ENTRY_CODE(Sp(0));
1206 /* ToDo: yield afterward for better communication performance? */
1210 /* -----------------------------------------------------------------------------
1211 Stable pointer primitives
1212 ------------------------------------------------------------------------- */
1214 makeStableNamezh_fast
1218 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1220 index = foreign "C" lookupStableName(R1 "ptr");
1222 /* Is there already a StableName for this heap object?
1223 * stable_ptr_table is an array of snEntry structs.
1225 if ( snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry) == NULL ) {
1226 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1227 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1228 StgStableName_sn(sn_obj) = index;
1229 snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry) = sn_obj;
1231 sn_obj = snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry);
1238 makeStablePtrzh_fast
1242 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1243 "ptr" sp = foreign "C" getStablePtr(R1 "ptr");
1247 deRefStablePtrzh_fast
1249 /* Args: R1 = the stable ptr */
1252 r = snEntry_addr(stable_ptr_table + sp*SIZEOF_snEntry);
1256 /* -----------------------------------------------------------------------------
1257 Bytecode object primitives
1258 ------------------------------------------------------------------------- */
1269 W_ bco, bitmap_arr, bytes, words;
1272 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1275 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R4_PTR&R6_PTR, newBCOzh_fast );
1277 bco = Hp - bytes + WDS(1);
1278 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1280 StgBCO_instrs(bco) = R1;
1281 StgBCO_literals(bco) = R2;
1282 StgBCO_ptrs(bco) = R3;
1283 StgBCO_itbls(bco) = R4;
1284 StgBCO_arity(bco) = HALF_W_(R5);
1285 StgBCO_size(bco) = HALF_W_(words);
1287 // Copy the arity/bitmap info into the BCO
1291 if (i < StgArrWords_words(bitmap_arr)) {
1292 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1303 // R1 = the BCO# for the AP
1307 // This function is *only* used to wrap zero-arity BCOs in an
1308 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1309 // saturated and always points directly to a FUN or BCO.
1310 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == BCO::I16 &&
1311 StgBCO_arity(R1) == 0::I16);
1313 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1314 TICK_ALLOC_UP_THK(0, 0);
1315 CCCS_ALLOC(SIZEOF_StgAP);
1317 ap = Hp - SIZEOF_StgAP + WDS(1);
1318 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1320 StgAP_n_args(ap) = 0::I16;
1326 /* -----------------------------------------------------------------------------
1327 Thread I/O blocking primitives
1328 -------------------------------------------------------------------------- */
1330 /* Add a thread to the end of the blocked queue. (C-- version of the C
1331 * macro in Schedule.h).
1333 #define APPEND_TO_BLOCKED_QUEUE(tso) \
1334 ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \
1335 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
1336 W_[blocked_queue_hd] = tso; \
1338 StgTSO_link(W_[blocked_queue_tl]) = tso; \
1340 W_[blocked_queue_tl] = tso;
1345 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1346 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1347 StgTSO_block_info(CurrentTSO) = R1;
1348 // No locking - we're not going to use this interface in the
1349 // threaded RTS anyway.
1350 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1351 jump stg_block_noregs;
1357 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1358 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1359 StgTSO_block_info(CurrentTSO) = R1;
1360 // No locking - we're not going to use this interface in the
1361 // threaded RTS anyway.
1362 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1363 jump stg_block_noregs;
1367 STRING(stg_delayzh_malloc_str, "delayzh_fast")
1370 #ifdef mingw32_TARGET_OS
1377 /* args: R1 (microsecond delay amount) */
1378 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1379 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
1381 #ifdef mingw32_TARGET_OS
1383 /* could probably allocate this on the heap instead */
1384 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1385 stg_delayzh_malloc_str);
1386 reqID = foreign "C" addDelayRequest(R1);
1387 StgAsyncIOResult_reqID(ares) = reqID;
1388 StgAsyncIOResult_len(ares) = 0;
1389 StgAsyncIOResult_errCode(ares) = 0;
1390 StgTSO_block_info(CurrentTSO) = ares;
1392 /* Having all async-blocked threads reside on the blocked_queue
1393 * simplifies matters, so change the status to OnDoProc put the
1394 * delayed thread on the blocked_queue.
1396 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1397 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1402 time = foreign "C" getourtimeofday();
1403 target = (R1 / (TICK_MILLISECS*1000)) + TO_W_(time);
1404 StgTSO_block_info(CurrentTSO) = target;
1406 /* Insert the new thread in the sleeping queue. */
1408 t = W_[sleeping_queue];
1410 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
1416 StgTSO_link(CurrentTSO) = t;
1418 W_[sleeping_queue] = CurrentTSO;
1420 StgTSO_link(prev) = CurrentTSO;
1424 jump stg_block_noregs;
1428 #ifdef mingw32_TARGET_OS
1429 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
1435 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1436 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1437 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1439 /* could probably allocate this on the heap instead */
1440 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1441 stg_asyncReadzh_malloc_str);
1442 reqID = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr");
1443 StgAsyncIOResult_reqID(ares) = reqID;
1444 StgAsyncIOResult_len(ares) = 0;
1445 StgAsyncIOResult_errCode(ares) = 0;
1446 StgTSO_block_info(CurrentTSO) = ares;
1447 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1448 jump stg_block_async;
1451 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
1457 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1458 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1459 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1461 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1462 stg_asyncWritezh_malloc_str);
1463 reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr");
1465 StgAsyncIOResult_reqID(ares) = reqID;
1466 StgAsyncIOResult_len(ares) = 0;
1467 StgAsyncIOResult_errCode(ares) = 0;
1468 StgTSO_block_info(CurrentTSO) = ares;
1469 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1470 jump stg_block_async;
1473 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
1479 /* args: R1 = proc, R2 = param */
1480 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1481 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1483 /* could probably allocate this on the heap instead */
1484 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1485 stg_asyncDoProczh_malloc_str);
1486 reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr");
1487 StgAsyncIOResult_reqID(ares) = reqID;
1488 StgAsyncIOResult_len(ares) = 0;
1489 StgAsyncIOResult_errCode(ares) = 0;
1490 StgTSO_block_info(CurrentTSO) = ares;
1491 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1492 jump stg_block_async;
1496 /* -----------------------------------------------------------------------------
1499 classes CCallable and CReturnable don't really exist, but the
1500 compiler insists on generating dictionaries containing references
1501 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
1502 for these. Some C compilers can't cope with zero-length static arrays,
1503 so we have to make these one element long.
1504 --------------------------------------------------------------------------- */
1507 GHC_ZCCCallable_static_info: W_ 0;
1511 GHC_ZCCReturnable_static_info: W_ 0;