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" allocateLocal(MyCapability() "ptr",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" allocateLocal(MyCapability() "ptr",words) [R2];
101 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
103 SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_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 // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
123 // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
124 // normally doesn't. However, when we freeze a MUT_ARR_PTRS, we leave
125 // it on the mutable list for the GC to remove (removing something from
126 // the mutable list is not easy, because the mut_list is only singly-linked).
128 // So that we can tell whether a MUT_ARR_PTRS_FROZEN is on the mutable list,
129 // when we freeze it we set the info ptr to be MUT_ARR_PTRS_FROZEN0
130 // to indicate that it is still on the mutable list.
132 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
133 // either it is on a mut_list, or it isn't. We adopt the convention that
134 // the closure type is MUT_ARR_PTRS_FROZEN0 if it is on the mutable list,
135 // and MUT_ARR_PTRS_FROZEN otherwise. In fact it wouldn't matter if
136 // we put it on the mutable list more than once, but it would get scavenged
137 // multiple times during GC, which would be unnecessarily slow.
139 if (StgHeader_info(R1) != stg_MUT_ARR_PTRS_FROZEN0_info) {
140 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
141 foreign "C" recordMutableLock(R1 "ptr") [R1];
142 // must be done after SET_INFO, because it ASSERTs closure_MUTABLE()
145 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
150 /* -----------------------------------------------------------------------------
152 -------------------------------------------------------------------------- */
157 /* Args: R1 = initialisation value */
159 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);
161 mv = Hp - SIZEOF_StgMutVar + WDS(1);
162 SET_HDR(mv,stg_MUT_VAR_DIRTY_info,W_[CCCS]);
163 StgMutVar_var(mv) = R1;
168 atomicModifyMutVarzh_fast
171 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
173 /* If x is the current contents of the MutVar#, then
174 We want to make the new contents point to
178 and the return value is
182 obviously we can share (f x).
184 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
185 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
186 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
190 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
191 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
193 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
194 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
198 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
199 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
201 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
202 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
205 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
207 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);
209 #if defined(THREADED_RTS)
210 foreign "C" ACQUIRE_LOCK(atomic_modify_mutvar_mutex "ptr") [R1,R2];
213 x = StgMutVar_var(R1);
215 TICK_ALLOC_THUNK_2();
216 CCCS_ALLOC(THUNK_2_SIZE);
217 z = Hp - THUNK_2_SIZE + WDS(1);
218 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
219 LDV_RECORD_CREATE(z);
220 StgThunk_payload(z,0) = R2;
221 StgThunk_payload(z,1) = x;
223 TICK_ALLOC_THUNK_1();
224 CCCS_ALLOC(THUNK_1_SIZE);
225 y = z - THUNK_1_SIZE;
226 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
227 LDV_RECORD_CREATE(y);
228 StgThunk_payload(y,0) = z;
230 StgMutVar_var(R1) = y;
231 foreign "C" dirty_MUT_VAR(BaseReg "ptr", R1 "ptr") [R1];
233 TICK_ALLOC_THUNK_1();
234 CCCS_ALLOC(THUNK_1_SIZE);
235 r = y - THUNK_1_SIZE;
236 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
237 LDV_RECORD_CREATE(r);
238 StgThunk_payload(r,0) = z;
240 #if defined(THREADED_RTS)
241 foreign "C" RELEASE_LOCK(atomic_modify_mutvar_mutex "ptr") [];
247 /* -----------------------------------------------------------------------------
248 Weak Pointer Primitives
249 -------------------------------------------------------------------------- */
251 STRING(stg_weak_msg,"New weak pointer at %p\n")
257 R3 = finalizer (or NULL)
262 R3 = stg_NO_FINALIZER_closure;
265 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
267 w = Hp - SIZEOF_StgWeak + WDS(1);
268 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
271 StgWeak_value(w) = R2;
272 StgWeak_finalizer(w) = R3;
274 StgWeak_link(w) = W_[weak_ptr_list];
275 W_[weak_ptr_list] = w;
277 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
292 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
293 RET_NP(0,stg_NO_FINALIZER_closure);
299 // A weak pointer is inherently used, so we do not need to call
300 // LDV_recordDead_FILL_SLOP_DYNAMIC():
301 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
302 // or, LDV_recordDead():
303 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
304 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
305 // large as weak pointers, so there is no need to fill the slop, either.
306 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
310 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
312 SET_INFO(w,stg_DEAD_WEAK_info);
313 LDV_RECORD_CREATE(w);
315 f = StgWeak_finalizer(w);
316 StgDeadWeak_link(w) = StgWeak_link(w);
318 /* return the finalizer */
319 if (f == stg_NO_FINALIZER_closure) {
320 RET_NP(0,stg_NO_FINALIZER_closure);
332 if (GET_INFO(w) == stg_WEAK_info) {
334 val = StgWeak_value(w);
342 /* -----------------------------------------------------------------------------
343 Arbitrary-precision Integer operations.
345 There are some assumptions in this code that mp_limb_t == W_. This is
346 the case for all the platforms that GHC supports, currently.
347 -------------------------------------------------------------------------- */
351 /* arguments: R1 = Int# */
353 W_ val, s, p; /* to avoid aliasing */
356 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
358 p = Hp - SIZEOF_StgArrWords;
359 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
360 StgArrWords_words(p) = 1;
362 /* mpz_set_si is inlined here, makes things simpler */
375 /* returns (# size :: Int#,
384 /* arguments: R1 = Word# */
386 W_ val, s, p; /* to avoid aliasing */
390 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
392 p = Hp - SIZEOF_StgArrWords;
393 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
394 StgArrWords_words(p) = 1;
403 /* returns (# size :: Int#,
404 data :: ByteArray# #)
411 * 'long long' primops for converting to/from Integers.
414 #ifdef SUPPORT_LONG_LONGS
416 int64ToIntegerzh_fast
418 /* arguments: L1 = Int64# */
421 W_ hi, s, neg, words_needed, p;
426 if ( %ge(val,0x100000000::L_) || %le(val,-0x100000000::L_) ) {
429 // minimum is one word
433 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
434 NO_PTRS, int64ToIntegerzh_fast );
436 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
437 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
438 StgArrWords_words(p) = words_needed;
440 if ( %lt(val,0::L_) ) {
445 hi = TO_W_(val >> 32);
447 if ( words_needed == 2 ) {
452 if ( val != 0::L_ ) {
455 } else /* val==0 */ {
463 /* returns (# size :: Int#,
464 data :: ByteArray# #)
469 word64ToIntegerzh_fast
471 /* arguments: L1 = Word64# */
474 W_ hi, s, words_needed, p;
477 if ( val >= 0x100000000::L_ ) {
483 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
484 NO_PTRS, word64ToIntegerzh_fast );
486 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
487 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
488 StgArrWords_words(p) = words_needed;
490 hi = TO_W_(val >> 32);
491 if ( val >= 0x100000000::L_ ) {
496 if ( val != 0::L_ ) {
499 } else /* val==0 */ {
504 /* returns (# size :: Int#,
505 data :: ByteArray# #)
511 #endif /* SUPPORT_LONG_LONGS */
513 /* ToDo: this is shockingly inefficient */
518 bits8 [SIZEOF_MP_INT];
523 bits8 [SIZEOF_MP_INT];
528 bits8 [SIZEOF_MP_INT];
533 bits8 [SIZEOF_MP_INT];
538 #define FETCH_MP_TEMP(X) \
540 X = BaseReg + (OFFSET_StgRegTable_r ## X);
542 #define FETCH_MP_TEMP(X) /* Nothing */
545 #define GMP_TAKE2_RET1(name,mp_fun) \
550 FETCH_MP_TEMP(mp_tmp1); \
551 FETCH_MP_TEMP(mp_tmp2); \
552 FETCH_MP_TEMP(mp_result1) \
553 FETCH_MP_TEMP(mp_result2); \
555 /* call doYouWantToGC() */ \
556 MAYBE_GC(R2_PTR & R4_PTR, name); \
563 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
564 MP_INT__mp_size(mp_tmp1) = (s1); \
565 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
566 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
567 MP_INT__mp_size(mp_tmp2) = (s2); \
568 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
570 foreign "C" mpz_init(mp_result1 "ptr") []; \
572 /* Perform the operation */ \
573 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
575 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
576 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
579 #define GMP_TAKE1_RET1(name,mp_fun) \
584 FETCH_MP_TEMP(mp_tmp1); \
585 FETCH_MP_TEMP(mp_result1) \
587 /* call doYouWantToGC() */ \
588 MAYBE_GC(R2_PTR, name); \
593 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
594 MP_INT__mp_size(mp_tmp1) = (s1); \
595 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
597 foreign "C" mpz_init(mp_result1 "ptr") []; \
599 /* Perform the operation */ \
600 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr") []; \
602 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
603 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
606 #define GMP_TAKE2_RET2(name,mp_fun) \
611 FETCH_MP_TEMP(mp_tmp1); \
612 FETCH_MP_TEMP(mp_tmp2); \
613 FETCH_MP_TEMP(mp_result1) \
614 FETCH_MP_TEMP(mp_result2) \
616 /* call doYouWantToGC() */ \
617 MAYBE_GC(R2_PTR & R4_PTR, name); \
624 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
625 MP_INT__mp_size(mp_tmp1) = (s1); \
626 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
627 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
628 MP_INT__mp_size(mp_tmp2) = (s2); \
629 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
631 foreign "C" mpz_init(mp_result1 "ptr") []; \
632 foreign "C" mpz_init(mp_result2 "ptr") []; \
634 /* Perform the operation */ \
635 foreign "C" mp_fun(mp_result1 "ptr",mp_result2 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
637 RET_NPNP(TO_W_(MP_INT__mp_size(mp_result1)), \
638 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords, \
639 TO_W_(MP_INT__mp_size(mp_result2)), \
640 MP_INT__mp_d(mp_result2) - SIZEOF_StgArrWords); \
643 GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add)
644 GMP_TAKE2_RET1(minusIntegerzh_fast, mpz_sub)
645 GMP_TAKE2_RET1(timesIntegerzh_fast, mpz_mul)
646 GMP_TAKE2_RET1(gcdIntegerzh_fast, mpz_gcd)
647 GMP_TAKE2_RET1(quotIntegerzh_fast, mpz_tdiv_q)
648 GMP_TAKE2_RET1(remIntegerzh_fast, mpz_tdiv_r)
649 GMP_TAKE2_RET1(divExactIntegerzh_fast, mpz_divexact)
650 GMP_TAKE2_RET1(andIntegerzh_fast, mpz_and)
651 GMP_TAKE2_RET1(orIntegerzh_fast, mpz_ior)
652 GMP_TAKE2_RET1(xorIntegerzh_fast, mpz_xor)
653 GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com)
655 GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr)
656 GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr)
660 mp_tmp_w: W_; // NB. mp_tmp_w is really an here mp_limb_t
666 /* R1 = the first Int#; R2 = the second Int# */
668 FETCH_MP_TEMP(mp_tmp_w);
671 r = foreign "C" mpn_gcd_1(mp_tmp_w "ptr", 1, R2) [];
674 /* Result parked in R1, return via info-pointer at TOS */
675 jump %ENTRY_CODE(Sp(0));
681 /* R1 = s1; R2 = d1; R3 = the int */
682 R1 = foreign "C" mpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3) [];
684 /* Result parked in R1, return via info-pointer at TOS */
685 jump %ENTRY_CODE(Sp(0));
691 /* R1 = s1; R2 = d1; R3 = the int */
692 W_ usize, vsize, v_digit, u_digit;
698 // paraphrased from mpz_cmp_si() in the GMP sources
699 if (%gt(v_digit,0)) {
702 if (%lt(v_digit,0)) {
708 if (usize != vsize) {
710 jump %ENTRY_CODE(Sp(0));
715 jump %ENTRY_CODE(Sp(0));
718 u_digit = W_[BYTE_ARR_CTS(R2)];
720 if (u_digit == v_digit) {
722 jump %ENTRY_CODE(Sp(0));
725 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
731 jump %ENTRY_CODE(Sp(0));
736 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
737 W_ usize, vsize, size, up, vp;
740 // paraphrased from mpz_cmp() in the GMP sources
744 if (usize != vsize) {
746 jump %ENTRY_CODE(Sp(0));
751 jump %ENTRY_CODE(Sp(0));
754 if (%lt(usize,0)) { // NB. not <, which is unsigned
760 up = BYTE_ARR_CTS(R2);
761 vp = BYTE_ARR_CTS(R4);
763 cmp = foreign "C" mpn_cmp(up "ptr", vp "ptr", size) [];
765 if (cmp == 0 :: CInt) {
767 jump %ENTRY_CODE(Sp(0));
770 if (%lt(cmp,0 :: CInt) == %lt(usize,0)) {
775 /* Result parked in R1, return via info-pointer at TOS */
776 jump %ENTRY_CODE(Sp(0));
788 r = W_[R2 + SIZEOF_StgArrWords];
793 /* Result parked in R1, return via info-pointer at TOS */
795 jump %ENTRY_CODE(Sp(0));
807 r = W_[R2 + SIZEOF_StgArrWords];
812 /* Result parked in R1, return via info-pointer at TOS */
814 jump %ENTRY_CODE(Sp(0));
821 FETCH_MP_TEMP(mp_tmp1);
822 FETCH_MP_TEMP(mp_tmp_w);
824 /* arguments: F1 = Float# */
827 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
829 /* Be prepared to tell Lennart-coded __decodeFloat
830 where mantissa._mp_d can be put (it does not care about the rest) */
831 p = Hp - SIZEOF_StgArrWords;
832 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
833 StgArrWords_words(p) = 1;
834 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
836 /* Perform the operation */
837 foreign "C" __decodeFloat(mp_tmp1 "ptr",mp_tmp_w "ptr" ,arg) [];
839 /* returns: (Int# (expn), Int#, ByteArray#) */
840 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
843 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
844 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
850 FETCH_MP_TEMP(mp_tmp1);
851 FETCH_MP_TEMP(mp_tmp_w);
853 /* arguments: D1 = Double# */
856 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
858 /* Be prepared to tell Lennart-coded __decodeDouble
859 where mantissa.d can be put (it does not care about the rest) */
860 p = Hp - ARR_SIZE + WDS(1);
861 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
862 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
863 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
865 /* Perform the operation */
866 foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg) [];
868 /* returns: (Int# (expn), Int#, ByteArray#) */
869 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
872 /* -----------------------------------------------------------------------------
873 * Concurrency primitives
874 * -------------------------------------------------------------------------- */
878 /* args: R1 = closure to spark */
880 MAYBE_GC(R1_PTR, forkzh_fast);
886 "ptr" threadid = foreign "C" createIOThread( MyCapability() "ptr",
887 RtsFlags_GcFlags_initialStkSize(RtsFlags),
889 foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") [];
891 // switch at the earliest opportunity
892 CInt[context_switch] = 1 :: CInt;
899 /* args: R1 = cpu, R2 = closure to spark */
901 MAYBE_GC(R2_PTR, forkOnzh_fast);
909 "ptr" threadid = foreign "C" createIOThread( MyCapability() "ptr",
910 RtsFlags_GcFlags_initialStkSize(RtsFlags),
912 foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") [];
914 // switch at the earliest opportunity
915 CInt[context_switch] = 1 :: CInt;
922 jump stg_yield_noregs;
937 foreign "C" labelThread(R1 "ptr", R2 "ptr") [];
939 jump %ENTRY_CODE(Sp(0));
942 isCurrentThreadBoundzh_fast
946 r = foreign "C" isThreadBound(CurrentTSO) [];
951 /* -----------------------------------------------------------------------------
953 * -------------------------------------------------------------------------- */
957 #define IF_NOT_REG_R1(x)
960 #define IF_NOT_REG_R1(x) x
963 // Catch retry frame ------------------------------------------------------------
965 #define CATCH_RETRY_FRAME_ERROR(label) \
966 label { foreign "C" barf("catch_retry_frame incorrectly entered!"); }
968 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_0_ret)
969 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_1_ret)
970 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_2_ret)
971 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_3_ret)
972 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_4_ret)
973 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_5_ret)
974 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_6_ret)
975 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_7_ret)
977 #if MAX_VECTORED_RTN > 8
978 #error MAX_VECTORED_RTN has changed: please modify stg_catch_retry_frame too.
981 #if defined(PROFILING)
982 #define CATCH_RETRY_FRAME_BITMAP 7
983 #define CATCH_RETRY_FRAME_WORDS 6
985 #define CATCH_RETRY_FRAME_BITMAP 1
986 #define CATCH_RETRY_FRAME_WORDS 4
989 INFO_TABLE_RET(stg_catch_retry_frame,
990 CATCH_RETRY_FRAME_WORDS, CATCH_RETRY_FRAME_BITMAP,
992 stg_catch_retry_frame_0_ret,
993 stg_catch_retry_frame_1_ret,
994 stg_catch_retry_frame_2_ret,
995 stg_catch_retry_frame_3_ret,
996 stg_catch_retry_frame_4_ret,
997 stg_catch_retry_frame_5_ret,
998 stg_catch_retry_frame_6_ret,
999 stg_catch_retry_frame_7_ret)
1001 W_ r, frame, trec, outer;
1002 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1005 trec = StgTSO_trec(CurrentTSO);
1006 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1007 r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
1009 /* Succeeded (either first branch or second branch) */
1010 StgTSO_trec(CurrentTSO) = outer;
1011 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1012 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1013 jump %ENTRY_CODE(Sp(SP_OFF));
1015 /* Did not commit: retry */
1017 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1018 StgTSO_trec(CurrentTSO) = new_trec;
1019 if (StgCatchRetryFrame_running_alt_code(frame)) {
1020 R1 = StgCatchRetryFrame_alt_code(frame);
1022 R1 = StgCatchRetryFrame_first_code(frame);
1023 StgCatchRetryFrame_first_code_trec(frame) = new_trec;
1030 // Atomically frame -------------------------------------------------------------
1033 #define ATOMICALLY_FRAME_ERROR(label) \
1034 label { foreign "C" barf("atomically_frame incorrectly entered!"); }
1036 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_0_ret)
1037 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_1_ret)
1038 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_2_ret)
1039 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_3_ret)
1040 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_4_ret)
1041 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_5_ret)
1042 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_6_ret)
1043 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_7_ret)
1045 #if MAX_VECTORED_RTN > 8
1046 #error MAX_VECTORED_RTN has changed: please modify stg_atomically_frame too.
1049 #if defined(PROFILING)
1050 #define ATOMICALLY_FRAME_BITMAP 3
1051 #define ATOMICALLY_FRAME_WORDS 3
1053 #define ATOMICALLY_FRAME_BITMAP 0
1054 #define ATOMICALLY_FRAME_WORDS 1
1058 INFO_TABLE_RET(stg_atomically_frame,
1059 ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP,
1061 stg_atomically_frame_0_ret,
1062 stg_atomically_frame_1_ret,
1063 stg_atomically_frame_2_ret,
1064 stg_atomically_frame_3_ret,
1065 stg_atomically_frame_4_ret,
1066 stg_atomically_frame_5_ret,
1067 stg_atomically_frame_6_ret,
1068 stg_atomically_frame_7_ret)
1070 W_ frame, trec, valid;
1071 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1074 trec = StgTSO_trec(CurrentTSO);
1076 /* The TSO is not currently waiting: try to commit the transaction */
1077 valid = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") [];
1079 /* Transaction was valid: commit succeeded */
1080 StgTSO_trec(CurrentTSO) = NO_TREC;
1081 Sp = Sp + SIZEOF_StgAtomicallyFrame;
1082 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1083 jump %ENTRY_CODE(Sp(SP_OFF));
1085 /* Transaction was not valid: try again */
1086 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1087 StgTSO_trec(CurrentTSO) = trec;
1088 R1 = StgAtomicallyFrame_code(frame);
1093 INFO_TABLE_RET(stg_atomically_waiting_frame,
1094 ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP,
1096 stg_atomically_frame_0_ret,
1097 stg_atomically_frame_1_ret,
1098 stg_atomically_frame_2_ret,
1099 stg_atomically_frame_3_ret,
1100 stg_atomically_frame_4_ret,
1101 stg_atomically_frame_5_ret,
1102 stg_atomically_frame_6_ret,
1103 stg_atomically_frame_7_ret)
1105 W_ frame, trec, valid;
1106 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1110 /* The TSO is currently waiting: should we stop waiting? */
1111 valid = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") [];
1113 /* Previous attempt is still valid: no point trying again yet */
1114 IF_NOT_REG_R1(Sp_adj(-2);
1115 Sp(1) = stg_NO_FINALIZER_closure;
1116 Sp(0) = stg_ut_1_0_unreg_info;)
1117 jump stg_block_noregs;
1119 /* Previous attempt is no longer valid: try again */
1120 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1121 StgTSO_trec(CurrentTSO) = trec;
1122 StgHeader_info(frame) = stg_atomically_frame_info;
1123 R1 = StgAtomicallyFrame_code(frame);
1128 // STM catch frame --------------------------------------------------------------
1130 #define CATCH_STM_FRAME_ENTRY_TEMPLATE(label,ret) \
1133 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \
1134 Sp = Sp + SIZEOF_StgCatchSTMFrame; \
1135 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \
1145 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_0_ret,%RET_VEC(Sp(SP_OFF),0))
1146 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_1_ret,%RET_VEC(Sp(SP_OFF),1))
1147 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_2_ret,%RET_VEC(Sp(SP_OFF),2))
1148 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_3_ret,%RET_VEC(Sp(SP_OFF),3))
1149 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_4_ret,%RET_VEC(Sp(SP_OFF),4))
1150 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_5_ret,%RET_VEC(Sp(SP_OFF),5))
1151 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_6_ret,%RET_VEC(Sp(SP_OFF),6))
1152 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_7_ret,%RET_VEC(Sp(SP_OFF),7))
1154 #if MAX_VECTORED_RTN > 8
1155 #error MAX_VECTORED_RTN has changed: please modify stg_catch_stm_frame too.
1158 #if defined(PROFILING)
1159 #define CATCH_STM_FRAME_BITMAP 3
1160 #define CATCH_STM_FRAME_WORDS 3
1162 #define CATCH_STM_FRAME_BITMAP 0
1163 #define CATCH_STM_FRAME_WORDS 1
1166 /* Catch frames are very similar to update frames, but when entering
1167 * one we just pop the frame off the stack and perform the correct
1168 * kind of return to the activation record underneath us on the stack.
1171 INFO_TABLE_RET(stg_catch_stm_frame,
1172 CATCH_STM_FRAME_WORDS, CATCH_STM_FRAME_BITMAP,
1174 stg_catch_stm_frame_0_ret,
1175 stg_catch_stm_frame_1_ret,
1176 stg_catch_stm_frame_2_ret,
1177 stg_catch_stm_frame_3_ret,
1178 stg_catch_stm_frame_4_ret,
1179 stg_catch_stm_frame_5_ret,
1180 stg_catch_stm_frame_6_ret,
1181 stg_catch_stm_frame_7_ret)
1182 CATCH_STM_FRAME_ENTRY_TEMPLATE(,%ENTRY_CODE(Sp(SP_OFF)))
1185 // Primop definition ------------------------------------------------------------
1193 // stmStartTransaction may allocate
1194 MAYBE_GC (R1_PTR, atomicallyzh_fast);
1196 /* Args: R1 = m :: STM a */
1197 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);
1199 old_trec = StgTSO_trec(CurrentTSO);
1201 /* Nested transactions are not allowed; raise an exception */
1202 if (old_trec != NO_TREC) {
1203 R1 = GHCziIOBase_NestedAtomically_closure;
1207 /* Set up the atomically frame */
1208 Sp = Sp - SIZEOF_StgAtomicallyFrame;
1211 SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
1212 StgAtomicallyFrame_code(frame) = R1;
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;
1238 /* Apply R1 to the realworld token */
1249 // stmStartTransaction may allocate
1250 MAYBE_GC (R1_PTR & R2_PTR, catchRetryzh_fast);
1252 /* Args: R1 :: STM a */
1253 /* Args: R2 :: STM a */
1254 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);
1256 /* Start a nested transaction within which to run the first code */
1257 trec = StgTSO_trec(CurrentTSO);
1258 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2];
1259 StgTSO_trec(CurrentTSO) = new_trec;
1261 /* Set up the catch-retry frame */
1262 Sp = Sp - SIZEOF_StgCatchRetryFrame;
1265 SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
1266 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1267 StgCatchRetryFrame_first_code(frame) = R1;
1268 StgCatchRetryFrame_alt_code(frame) = R2;
1269 StgCatchRetryFrame_first_code_trec(frame) = new_trec;
1271 /* Apply R1 to the realworld token */
1284 MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate
1286 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
1288 trec = StgTSO_trec(CurrentTSO);
1289 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1290 StgTSO_sp(CurrentTSO) = Sp;
1291 frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") [];
1292 Sp = StgTSO_sp(CurrentTSO);
1295 if (frame_type == CATCH_RETRY_FRAME) {
1296 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
1297 ASSERT(outer != NO_TREC);
1298 if (!StgCatchRetryFrame_running_alt_code(frame)) {
1299 // Retry in the first code: try the alternative
1300 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1301 StgTSO_trec(CurrentTSO) = trec;
1302 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
1303 R1 = StgCatchRetryFrame_alt_code(frame);
1306 // Retry in the alternative code: propagate
1308 other_trec = StgCatchRetryFrame_first_code_trec(frame);
1309 r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", other_trec "ptr") [];
1311 r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
1313 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1316 // Merge between siblings succeeded: commit it back to enclosing transaction
1317 // and then propagate the retry
1318 StgTSO_trec(CurrentTSO) = outer;
1319 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1320 goto retry_pop_stack;
1322 // Merge failed: we musn't propagate the retry. Try both paths again.
1323 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1324 StgCatchRetryFrame_first_code_trec(frame) = trec;
1325 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1326 StgTSO_trec(CurrentTSO) = trec;
1327 R1 = StgCatchRetryFrame_first_code(frame);
1333 // We've reached the ATOMICALLY_FRAME: attempt to wait
1334 ASSERT(frame_type == ATOMICALLY_FRAME);
1335 ASSERT(outer == NO_TREC);
1336 r = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") [];
1338 // Transaction was valid: stmWait put us on the TVars' queues, we now block
1339 StgHeader_info(frame) = stg_atomically_waiting_frame_info;
1341 // Fix up the stack in the unregisterised case: the return convention is different.
1342 IF_NOT_REG_R1(Sp_adj(-2);
1343 Sp(1) = stg_NO_FINALIZER_closure;
1344 Sp(0) = stg_ut_1_0_unreg_info;)
1345 R3 = trec; // passing to stmWaitUnblock()
1346 jump stg_block_stmwait;
1348 // Transaction was not valid: retry immediately
1349 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1350 StgTSO_trec(CurrentTSO) = trec;
1351 R1 = StgAtomicallyFrame_code(frame);
1363 /* Args: R1 = initialisation value */
1365 MAYBE_GC (R1_PTR, newTVarzh_fast);
1367 "ptr" tv = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") [];
1378 /* Args: R1 = TVar closure */
1380 MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
1381 trec = StgTSO_trec(CurrentTSO);
1383 "ptr" result = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") [];
1395 /* Args: R1 = TVar closure */
1396 /* R2 = New value */
1398 MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
1399 trec = StgTSO_trec(CurrentTSO);
1402 foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") [];
1404 jump %ENTRY_CODE(Sp(0));
1408 /* -----------------------------------------------------------------------------
1411 * take & putMVar work as follows. Firstly, an important invariant:
1413 * If the MVar is full, then the blocking queue contains only
1414 * threads blocked on putMVar, and if the MVar is empty then the
1415 * blocking queue contains only threads blocked on takeMVar.
1418 * MVar empty : then add ourselves to the blocking queue
1419 * MVar full : remove the value from the MVar, and
1420 * blocking queue empty : return
1421 * blocking queue non-empty : perform the first blocked putMVar
1422 * from the queue, and wake up the
1423 * thread (MVar is now full again)
1425 * putMVar is just the dual of the above algorithm.
1427 * How do we "perform a putMVar"? Well, we have to fiddle around with
1428 * the stack of the thread waiting to do the putMVar. See
1429 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1430 * the stack layout, and the PerformPut and PerformTake macros below.
1432 * It is important that a blocked take or put is woken up with the
1433 * take/put already performed, because otherwise there would be a
1434 * small window of vulnerability where the thread could receive an
1435 * exception and never perform its take or put, and we'd end up with a
1438 * -------------------------------------------------------------------------- */
1442 /* args: R1 = MVar closure */
1444 if (GET_INFO(R1) == stg_EMPTY_MVAR_info) {
1456 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
1458 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1459 SET_HDR(mvar,stg_EMPTY_MVAR_info,W_[CCCS]);
1460 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1461 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1462 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1467 /* If R1 isn't available, pass it on the stack */
1469 #define PerformTake(tso, value) \
1470 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1471 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1473 #define PerformTake(tso, value) \
1474 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1475 W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info;
1478 #define PerformPut(tso,lval) \
1479 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1480 lval = W_[StgTSO_sp(tso) - WDS(1)];
1484 W_ mvar, val, info, tso;
1486 /* args: R1 = MVar closure */
1489 #if defined(THREADED_RTS)
1490 "ptr" info = foreign "C" lockClosure(mvar "ptr") [];
1492 info = GET_INFO(mvar);
1495 /* If the MVar is empty, put ourselves on its blocking queue,
1496 * and wait until we're woken up.
1498 if (info == stg_EMPTY_MVAR_info) {
1499 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1500 StgMVar_head(mvar) = CurrentTSO;
1502 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1504 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1505 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1506 StgTSO_block_info(CurrentTSO) = mvar;
1507 StgMVar_tail(mvar) = CurrentTSO;
1509 jump stg_block_takemvar;
1512 /* we got the value... */
1513 val = StgMVar_value(mvar);
1515 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1517 /* There are putMVar(s) waiting...
1518 * wake up the first thread on the queue
1520 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1522 /* actually perform the putMVar for the thread that we just woke up */
1523 tso = StgMVar_head(mvar);
1524 PerformPut(tso,StgMVar_value(mvar));
1525 foreign "C" dirtyTSO(tso "ptr") [];
1527 #if defined(GRAN) || defined(PAR)
1528 /* ToDo: check 2nd arg (mvar) is right */
1529 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar) [];
1530 StgMVar_head(mvar) = tso;
1532 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr",
1533 StgMVar_head(mvar) "ptr") [];
1534 StgMVar_head(mvar) = tso;
1537 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1538 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1541 #if defined(THREADED_RTS)
1542 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1548 /* No further putMVars, MVar is now empty */
1549 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1551 #if defined(THREADED_RTS)
1552 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1554 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1564 W_ mvar, val, info, tso;
1566 /* args: R1 = MVar closure */
1570 #if defined(THREADED_RTS)
1571 "ptr" info = foreign "C" lockClosure(mvar "ptr") [];
1573 info = GET_INFO(mvar);
1576 if (info == stg_EMPTY_MVAR_info) {
1577 #if defined(THREADED_RTS)
1578 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1580 /* HACK: we need a pointer to pass back,
1581 * so we abuse NO_FINALIZER_closure
1583 RET_NP(0, stg_NO_FINALIZER_closure);
1586 /* we got the value... */
1587 val = StgMVar_value(mvar);
1589 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1591 /* There are putMVar(s) waiting...
1592 * wake up the first thread on the queue
1594 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1596 /* actually perform the putMVar for the thread that we just woke up */
1597 tso = StgMVar_head(mvar);
1598 PerformPut(tso,StgMVar_value(mvar));
1599 foreign "C" dirtyTSO(tso "ptr") [];
1601 #if defined(GRAN) || defined(PAR)
1602 /* ToDo: check 2nd arg (mvar) is right */
1603 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr") [];
1604 StgMVar_head(mvar) = tso;
1606 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr",
1607 StgMVar_head(mvar) "ptr") [];
1608 StgMVar_head(mvar) = tso;
1611 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1612 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1614 #if defined(THREADED_RTS)
1615 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1620 /* No further putMVars, MVar is now empty */
1621 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1622 #if defined(THREADED_RTS)
1623 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1625 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1637 /* args: R1 = MVar, R2 = value */
1640 #if defined(THREADED_RTS)
1641 "ptr" info = foreign "C" lockClosure(mvar "ptr") [R2];
1643 info = GET_INFO(mvar);
1646 if (info == stg_FULL_MVAR_info) {
1647 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1648 StgMVar_head(mvar) = CurrentTSO;
1650 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1652 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1653 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1654 StgTSO_block_info(CurrentTSO) = mvar;
1655 StgMVar_tail(mvar) = CurrentTSO;
1657 jump stg_block_putmvar;
1660 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1662 /* There are takeMVar(s) waiting: wake up the first one
1664 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1666 /* actually perform the takeMVar */
1667 tso = StgMVar_head(mvar);
1668 PerformTake(tso, R2);
1669 foreign "C" dirtyTSO(tso "ptr") [];
1671 #if defined(GRAN) || defined(PAR)
1672 /* ToDo: check 2nd arg (mvar) is right */
1673 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") [];
1674 StgMVar_head(mvar) = tso;
1676 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") [];
1677 StgMVar_head(mvar) = tso;
1680 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1681 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1684 #if defined(THREADED_RTS)
1685 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1687 jump %ENTRY_CODE(Sp(0));
1691 /* No further takes, the MVar is now full. */
1692 StgMVar_value(mvar) = R2;
1694 #if defined(THREADED_RTS)
1695 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1697 SET_INFO(mvar,stg_FULL_MVAR_info);
1699 jump %ENTRY_CODE(Sp(0));
1702 /* ToDo: yield afterward for better communication performance? */
1710 /* args: R1 = MVar, R2 = value */
1713 #if defined(THREADED_RTS)
1714 "ptr" info = foreign "C" lockClosure(mvar "ptr") [R2];
1716 info = GET_INFO(mvar);
1719 if (info == stg_FULL_MVAR_info) {
1720 #if defined(THREADED_RTS)
1721 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1726 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1728 /* There are takeMVar(s) waiting: wake up the first one
1730 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1732 /* actually perform the takeMVar */
1733 tso = StgMVar_head(mvar);
1734 PerformTake(tso, R2);
1735 foreign "C" dirtyTSO(tso "ptr") [];
1737 #if defined(GRAN) || defined(PAR)
1738 /* ToDo: check 2nd arg (mvar) is right */
1739 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") [];
1740 StgMVar_head(mvar) = tso;
1742 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") [];
1743 StgMVar_head(mvar) = tso;
1746 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1747 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1750 #if defined(THREADED_RTS)
1751 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1756 /* No further takes, the MVar is now full. */
1757 StgMVar_value(mvar) = R2;
1759 #if defined(THREADED_RTS)
1760 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1762 SET_INFO(mvar,stg_FULL_MVAR_info);
1767 /* ToDo: yield afterward for better communication performance? */
1771 /* -----------------------------------------------------------------------------
1772 Stable pointer primitives
1773 ------------------------------------------------------------------------- */
1775 makeStableNamezh_fast
1779 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1781 index = foreign "C" lookupStableName(R1 "ptr") [];
1783 /* Is there already a StableName for this heap object?
1784 * stable_ptr_table is a pointer to an array of snEntry structs.
1786 if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) {
1787 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1788 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1789 StgStableName_sn(sn_obj) = index;
1790 snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj;
1792 sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry);
1799 makeStablePtrzh_fast
1803 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1804 "ptr" sp = foreign "C" getStablePtr(R1 "ptr") [];
1808 deRefStablePtrzh_fast
1810 /* Args: R1 = the stable ptr */
1813 r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry);
1817 /* -----------------------------------------------------------------------------
1818 Bytecode object primitives
1819 ------------------------------------------------------------------------- */
1830 W_ bco, bitmap_arr, bytes, words;
1833 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1836 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R4_PTR&R6_PTR, newBCOzh_fast );
1838 bco = Hp - bytes + WDS(1);
1839 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1841 StgBCO_instrs(bco) = R1;
1842 StgBCO_literals(bco) = R2;
1843 StgBCO_ptrs(bco) = R3;
1844 StgBCO_itbls(bco) = R4;
1845 StgBCO_arity(bco) = HALF_W_(R5);
1846 StgBCO_size(bco) = HALF_W_(words);
1848 // Copy the arity/bitmap info into the BCO
1852 if (i < StgArrWords_words(bitmap_arr)) {
1853 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1864 // R1 = the BCO# for the AP
1868 // This function is *only* used to wrap zero-arity BCOs in an
1869 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1870 // saturated and always points directly to a FUN or BCO.
1871 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
1872 StgBCO_arity(R1) == HALF_W_(0));
1874 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1875 TICK_ALLOC_UP_THK(0, 0);
1876 CCCS_ALLOC(SIZEOF_StgAP);
1878 ap = Hp - SIZEOF_StgAP + WDS(1);
1879 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1881 StgAP_n_args(ap) = HALF_W_(0);
1887 /* -----------------------------------------------------------------------------
1888 Thread I/O blocking primitives
1889 -------------------------------------------------------------------------- */
1891 /* Add a thread to the end of the blocked queue. (C-- version of the C
1892 * macro in Schedule.h).
1894 #define APPEND_TO_BLOCKED_QUEUE(tso) \
1895 ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \
1896 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
1897 W_[blocked_queue_hd] = tso; \
1899 StgTSO_link(W_[blocked_queue_tl]) = tso; \
1901 W_[blocked_queue_tl] = tso;
1907 foreign "C" barf("waitRead# on threaded RTS");
1910 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1911 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1912 StgTSO_block_info(CurrentTSO) = R1;
1913 // No locking - we're not going to use this interface in the
1914 // threaded RTS anyway.
1915 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1916 jump stg_block_noregs;
1924 foreign "C" barf("waitWrite# on threaded RTS");
1927 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1928 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1929 StgTSO_block_info(CurrentTSO) = R1;
1930 // No locking - we're not going to use this interface in the
1931 // threaded RTS anyway.
1932 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1933 jump stg_block_noregs;
1938 STRING(stg_delayzh_malloc_str, "delayzh_fast")
1941 #ifdef mingw32_HOST_OS
1949 foreign "C" barf("delay# on threaded RTS");
1952 /* args: R1 (microsecond delay amount) */
1953 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1954 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
1956 #ifdef mingw32_HOST_OS
1958 /* could probably allocate this on the heap instead */
1959 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1960 stg_delayzh_malloc_str);
1961 reqID = foreign "C" addDelayRequest(R1);
1962 StgAsyncIOResult_reqID(ares) = reqID;
1963 StgAsyncIOResult_len(ares) = 0;
1964 StgAsyncIOResult_errCode(ares) = 0;
1965 StgTSO_block_info(CurrentTSO) = ares;
1967 /* Having all async-blocked threads reside on the blocked_queue
1968 * simplifies matters, so change the status to OnDoProc put the
1969 * delayed thread on the blocked_queue.
1971 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1972 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1973 jump stg_block_async_void;
1978 time = foreign "C" getourtimeofday();
1979 target = (R1 / (TICK_MILLISECS*1000)) + time;
1980 StgTSO_block_info(CurrentTSO) = target;
1982 /* Insert the new thread in the sleeping queue. */
1984 t = W_[sleeping_queue];
1986 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
1992 StgTSO_link(CurrentTSO) = t;
1994 W_[sleeping_queue] = CurrentTSO;
1996 StgTSO_link(prev) = CurrentTSO;
1998 jump stg_block_noregs;
2000 #endif /* !THREADED_RTS */
2004 #ifdef mingw32_HOST_OS
2005 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
2012 foreign "C" barf("asyncRead# on threaded RTS");
2015 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2016 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2017 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
2019 /* could probably allocate this on the heap instead */
2020 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2021 stg_asyncReadzh_malloc_str)
2023 reqID = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr") [];
2024 StgAsyncIOResult_reqID(ares) = reqID;
2025 StgAsyncIOResult_len(ares) = 0;
2026 StgAsyncIOResult_errCode(ares) = 0;
2027 StgTSO_block_info(CurrentTSO) = ares;
2028 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2029 jump stg_block_async;
2033 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
2040 foreign "C" barf("asyncWrite# on threaded RTS");
2043 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2044 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2045 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2047 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2048 stg_asyncWritezh_malloc_str)
2050 reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr") [];
2052 StgAsyncIOResult_reqID(ares) = reqID;
2053 StgAsyncIOResult_len(ares) = 0;
2054 StgAsyncIOResult_errCode(ares) = 0;
2055 StgTSO_block_info(CurrentTSO) = ares;
2056 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2057 jump stg_block_async;
2061 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
2068 foreign "C" barf("asyncDoProc# on threaded RTS");
2071 /* args: R1 = proc, R2 = param */
2072 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2073 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2075 /* could probably allocate this on the heap instead */
2076 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2077 stg_asyncDoProczh_malloc_str)
2079 reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr") [];
2080 StgAsyncIOResult_reqID(ares) = reqID;
2081 StgAsyncIOResult_len(ares) = 0;
2082 StgAsyncIOResult_errCode(ares) = 0;
2083 StgTSO_block_info(CurrentTSO) = ares;
2084 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2085 jump stg_block_async;
2090 /* -----------------------------------------------------------------------------
2093 classes CCallable and CReturnable don't really exist, but the
2094 compiler insists on generating dictionaries containing references
2095 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
2096 for these. Some C compilers can't cope with zero-length static arrays,
2097 so we have to make these one element long.
2098 --------------------------------------------------------------------------- */
2101 GHC_ZCCCallable_static_info: W_ 0;
2105 GHC_ZCCReturnable_static_info: W_ 0;