1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2004
5 * Out-of-line primitive operations
7 * This file contains the implementations of all the primitive
8 * operations ("primops") which are not expanded inline. See
9 * ghc/compiler/prelude/primops.txt.pp for a list of all the primops;
10 * this file contains code for most of those with the attribute
13 * Entry convention: the entry convention for a primop is that all the
14 * args are in Stg registers (R1, R2, etc.). This is to make writing
15 * the primops easier. (see compiler/codeGen/CgCallConv.hs).
17 * Return convention: results from a primop are generally returned
18 * using the ordinary unboxed tuple return convention. The C-- parser
19 * implements the RET_xxxx() macros to perform unboxed-tuple returns
20 * based on the prevailing return convention.
22 * This file is written in a subset of C--, extended with various
23 * features specific to GHC. It is compiled by GHC directly. For the
24 * syntax of .cmm files, see the parser in ghc/compiler/cmm/CmmParse.y.
26 * ---------------------------------------------------------------------------*/
31 #ifndef mingw32_HOST_OS
41 import __gmpz_tdiv_qr;
42 import __gmpz_fdiv_qr;
43 import __gmpz_divexact;
49 import pthread_mutex_lock;
50 import pthread_mutex_unlock;
52 import base_ControlziExceptionziBase_nestedAtomically_closure;
53 import EnterCriticalSection;
54 import LeaveCriticalSection;
56 /*-----------------------------------------------------------------------------
59 Basically just new*Array - the others are all inline macros.
61 The size arg is always passed in R1, and the result returned in R1.
63 The slow entry point is for returning from a heap check, the saved
64 size argument must be re-loaded from the stack.
65 -------------------------------------------------------------------------- */
67 /* for objects that are *less* than the size of a word, make sure we
68 * round up to the nearest word for the size of the array.
73 W_ words, payload_words, n, p;
74 MAYBE_GC(NO_PTRS,newByteArrayzh_fast);
76 payload_words = ROUNDUP_BYTES_TO_WDS(n);
77 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
78 ("ptr" p) = foreign "C" allocateLocal(MyCapability() "ptr",words) [];
79 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
80 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
81 StgArrWords_words(p) = payload_words;
85 newPinnedByteArrayzh_fast
87 W_ words, payload_words, n, p;
89 MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast);
91 payload_words = ROUNDUP_BYTES_TO_WDS(n);
93 // We want an 8-byte aligned array. allocatePinned() gives us
94 // 8-byte aligned memory by default, but we want to align the
95 // *goods* inside the ArrWords object, so we have to check the
96 // size of the ArrWords header and adjust our size accordingly.
97 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
98 if ((SIZEOF_StgArrWords & 7) != 0) {
102 ("ptr" p) = foreign "C" allocatePinned(words) [];
103 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
105 // Again, if the ArrWords header isn't a multiple of 8 bytes, we
106 // have to push the object forward one word so that the goods
107 // fall on an 8-byte boundary.
108 if ((SIZEOF_StgArrWords & 7) != 0) {
112 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
113 StgArrWords_words(p) = payload_words;
119 W_ words, n, init, arr, p;
120 /* Args: R1 = words, R2 = initialisation value */
123 MAYBE_GC(R2_PTR,newArrayzh_fast);
125 words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
126 ("ptr" arr) = foreign "C" allocateLocal(MyCapability() "ptr",words) [R2];
127 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
129 SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, W_[CCCS]);
130 StgMutArrPtrs_ptrs(arr) = n;
132 // Initialise all elements of the the array with the value in R2
134 p = arr + SIZEOF_StgMutArrPtrs;
136 if (p < arr + WDS(words)) {
145 unsafeThawArrayzh_fast
147 // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
149 // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
150 // normally doesn't. However, when we freeze a MUT_ARR_PTRS, we leave
151 // it on the mutable list for the GC to remove (removing something from
152 // the mutable list is not easy, because the mut_list is only singly-linked).
154 // So that we can tell whether a MUT_ARR_PTRS_FROZEN is on the mutable list,
155 // when we freeze it we set the info ptr to be MUT_ARR_PTRS_FROZEN0
156 // to indicate that it is still on the mutable list.
158 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
159 // either it is on a mut_list, or it isn't. We adopt the convention that
160 // the closure type is MUT_ARR_PTRS_FROZEN0 if it is on the mutable list,
161 // and MUT_ARR_PTRS_FROZEN otherwise. In fact it wouldn't matter if
162 // we put it on the mutable list more than once, but it would get scavenged
163 // multiple times during GC, which would be unnecessarily slow.
165 if (StgHeader_info(R1) != stg_MUT_ARR_PTRS_FROZEN0_info) {
166 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
167 recordMutable(R1, R1);
168 // must be done after SET_INFO, because it ASSERTs closure_MUTABLE()
171 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
176 /* -----------------------------------------------------------------------------
178 -------------------------------------------------------------------------- */
183 /* Args: R1 = initialisation value */
185 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);
187 mv = Hp - SIZEOF_StgMutVar + WDS(1);
188 SET_HDR(mv,stg_MUT_VAR_DIRTY_info,W_[CCCS]);
189 StgMutVar_var(mv) = R1;
194 atomicModifyMutVarzh_fast
197 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
199 /* If x is the current contents of the MutVar#, then
200 We want to make the new contents point to
204 and the return value is
208 obviously we can share (f x).
210 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
211 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
212 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
216 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
217 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
219 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
220 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
224 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
225 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
227 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
228 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
231 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
233 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);
235 #if defined(THREADED_RTS)
236 ACQUIRE_LOCK(atomic_modify_mutvar_mutex "ptr") [R1,R2];
239 x = StgMutVar_var(R1);
241 TICK_ALLOC_THUNK_2();
242 CCCS_ALLOC(THUNK_2_SIZE);
243 z = Hp - THUNK_2_SIZE + WDS(1);
244 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
245 LDV_RECORD_CREATE(z);
246 StgThunk_payload(z,0) = R2;
247 StgThunk_payload(z,1) = x;
249 TICK_ALLOC_THUNK_1();
250 CCCS_ALLOC(THUNK_1_SIZE);
251 y = z - THUNK_1_SIZE;
252 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
253 LDV_RECORD_CREATE(y);
254 StgThunk_payload(y,0) = z;
256 StgMutVar_var(R1) = y;
257 foreign "C" dirty_MUT_VAR(BaseReg "ptr", R1 "ptr") [R1];
259 TICK_ALLOC_THUNK_1();
260 CCCS_ALLOC(THUNK_1_SIZE);
261 r = y - THUNK_1_SIZE;
262 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
263 LDV_RECORD_CREATE(r);
264 StgThunk_payload(r,0) = z;
266 #if defined(THREADED_RTS)
267 RELEASE_LOCK(atomic_modify_mutvar_mutex "ptr") [];
273 /* -----------------------------------------------------------------------------
274 Weak Pointer Primitives
275 -------------------------------------------------------------------------- */
277 STRING(stg_weak_msg,"New weak pointer at %p\n")
283 R3 = finalizer (or NULL)
288 R3 = stg_NO_FINALIZER_closure;
291 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
293 w = Hp - SIZEOF_StgWeak + WDS(1);
294 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
297 StgWeak_value(w) = R2;
298 StgWeak_finalizer(w) = R3;
300 StgWeak_link(w) = W_[weak_ptr_list];
301 W_[weak_ptr_list] = w;
303 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
318 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
319 RET_NP(0,stg_NO_FINALIZER_closure);
325 // A weak pointer is inherently used, so we do not need to call
326 // LDV_recordDead_FILL_SLOP_DYNAMIC():
327 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
328 // or, LDV_recordDead():
329 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
330 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
331 // large as weak pointers, so there is no need to fill the slop, either.
332 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
336 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
338 SET_INFO(w,stg_DEAD_WEAK_info);
339 LDV_RECORD_CREATE(w);
341 f = StgWeak_finalizer(w);
342 StgDeadWeak_link(w) = StgWeak_link(w);
344 /* return the finalizer */
345 if (f == stg_NO_FINALIZER_closure) {
346 RET_NP(0,stg_NO_FINALIZER_closure);
358 if (GET_INFO(w) == stg_WEAK_info) {
360 val = StgWeak_value(w);
368 /* -----------------------------------------------------------------------------
369 Arbitrary-precision Integer operations.
371 There are some assumptions in this code that mp_limb_t == W_. This is
372 the case for all the platforms that GHC supports, currently.
373 -------------------------------------------------------------------------- */
377 /* arguments: R1 = Int# */
379 W_ val, s, p; /* to avoid aliasing */
382 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
384 p = Hp - SIZEOF_StgArrWords;
385 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
386 StgArrWords_words(p) = 1;
388 /* mpz_set_si is inlined here, makes things simpler */
401 /* returns (# size :: Int#,
410 /* arguments: R1 = Word# */
412 W_ val, s, p; /* to avoid aliasing */
416 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
418 p = Hp - SIZEOF_StgArrWords;
419 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
420 StgArrWords_words(p) = 1;
429 /* returns (# size :: Int#,
430 data :: ByteArray# #)
437 * 'long long' primops for converting to/from Integers.
440 #ifdef SUPPORT_LONG_LONGS
442 int64ToIntegerzh_fast
444 /* arguments: L1 = Int64# */
447 W_ hi, lo, s, neg, words_needed, p;
452 hi = TO_W_(val >> 32);
455 if ( hi == 0 || (hi == 0xFFFFFFFF && lo != 0) ) {
456 // minimum is one word
462 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
463 NO_PTRS, int64ToIntegerzh_fast );
465 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
466 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
467 StgArrWords_words(p) = words_needed;
479 if ( words_needed == 2 ) {
487 } else /* val==0 */ {
495 /* returns (# size :: Int#,
496 data :: ByteArray# #)
500 word64ToIntegerzh_fast
502 /* arguments: L1 = Word64# */
505 W_ hi, lo, s, words_needed, p;
508 hi = TO_W_(val >> 32);
517 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
518 NO_PTRS, word64ToIntegerzh_fast );
520 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
521 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
522 StgArrWords_words(p) = words_needed;
532 } else /* val==0 */ {
537 /* returns (# size :: Int#,
538 data :: ByteArray# #)
545 #endif /* SUPPORT_LONG_LONGS */
547 /* ToDo: this is shockingly inefficient */
552 bits8 [SIZEOF_MP_INT];
557 bits8 [SIZEOF_MP_INT];
562 bits8 [SIZEOF_MP_INT];
567 bits8 [SIZEOF_MP_INT];
572 #define FETCH_MP_TEMP(X) \
574 X = BaseReg + (OFFSET_StgRegTable_r ## X);
576 #define FETCH_MP_TEMP(X) /* Nothing */
579 #define GMP_TAKE2_RET1(name,mp_fun) \
584 FETCH_MP_TEMP(mp_tmp1); \
585 FETCH_MP_TEMP(mp_tmp2); \
586 FETCH_MP_TEMP(mp_result1) \
587 FETCH_MP_TEMP(mp_result2); \
589 /* call doYouWantToGC() */ \
590 MAYBE_GC(R2_PTR & R4_PTR, name); \
597 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
598 MP_INT__mp_size(mp_tmp1) = (s1); \
599 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
600 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
601 MP_INT__mp_size(mp_tmp2) = (s2); \
602 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
604 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
606 /* Perform the operation */ \
607 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
609 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
610 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
613 #define GMP_TAKE1_RET1(name,mp_fun) \
618 FETCH_MP_TEMP(mp_tmp1); \
619 FETCH_MP_TEMP(mp_result1) \
621 /* call doYouWantToGC() */ \
622 MAYBE_GC(R2_PTR, name); \
627 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
628 MP_INT__mp_size(mp_tmp1) = (s1); \
629 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
631 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
633 /* Perform the operation */ \
634 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr") []; \
636 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
637 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
640 #define GMP_TAKE2_RET2(name,mp_fun) \
645 FETCH_MP_TEMP(mp_tmp1); \
646 FETCH_MP_TEMP(mp_tmp2); \
647 FETCH_MP_TEMP(mp_result1) \
648 FETCH_MP_TEMP(mp_result2) \
650 /* call doYouWantToGC() */ \
651 MAYBE_GC(R2_PTR & R4_PTR, name); \
658 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
659 MP_INT__mp_size(mp_tmp1) = (s1); \
660 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
661 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
662 MP_INT__mp_size(mp_tmp2) = (s2); \
663 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
665 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
666 foreign "C" __gmpz_init(mp_result2 "ptr") []; \
668 /* Perform the operation */ \
669 foreign "C" mp_fun(mp_result1 "ptr",mp_result2 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
671 RET_NPNP(TO_W_(MP_INT__mp_size(mp_result1)), \
672 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords, \
673 TO_W_(MP_INT__mp_size(mp_result2)), \
674 MP_INT__mp_d(mp_result2) - SIZEOF_StgArrWords); \
677 GMP_TAKE2_RET1(plusIntegerzh_fast, __gmpz_add)
678 GMP_TAKE2_RET1(minusIntegerzh_fast, __gmpz_sub)
679 GMP_TAKE2_RET1(timesIntegerzh_fast, __gmpz_mul)
680 GMP_TAKE2_RET1(gcdIntegerzh_fast, __gmpz_gcd)
681 GMP_TAKE2_RET1(quotIntegerzh_fast, __gmpz_tdiv_q)
682 GMP_TAKE2_RET1(remIntegerzh_fast, __gmpz_tdiv_r)
683 GMP_TAKE2_RET1(divExactIntegerzh_fast, __gmpz_divexact)
684 GMP_TAKE2_RET1(andIntegerzh_fast, __gmpz_and)
685 GMP_TAKE2_RET1(orIntegerzh_fast, __gmpz_ior)
686 GMP_TAKE2_RET1(xorIntegerzh_fast, __gmpz_xor)
687 GMP_TAKE1_RET1(complementIntegerzh_fast, __gmpz_com)
689 GMP_TAKE2_RET2(quotRemIntegerzh_fast, __gmpz_tdiv_qr)
690 GMP_TAKE2_RET2(divModIntegerzh_fast, __gmpz_fdiv_qr)
694 mp_tmp_w: W_; // NB. mp_tmp_w is really an here mp_limb_t
700 /* R1 = the first Int#; R2 = the second Int# */
702 FETCH_MP_TEMP(mp_tmp_w);
705 (r) = foreign "C" __gmpn_gcd_1(mp_tmp_w "ptr", 1, R2) [];
708 /* Result parked in R1, return via info-pointer at TOS */
709 jump %ENTRY_CODE(Sp(0));
715 /* R1 = s1; R2 = d1; R3 = the int */
717 (s1) = foreign "C" __gmpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3) [];
720 /* Result parked in R1, return via info-pointer at TOS */
721 jump %ENTRY_CODE(Sp(0));
727 /* R1 = s1; R2 = d1; R3 = the int */
728 W_ usize, vsize, v_digit, u_digit;
734 // paraphrased from __gmpz_cmp_si() in the GMP sources
735 if (%gt(v_digit,0)) {
738 if (%lt(v_digit,0)) {
744 if (usize != vsize) {
746 jump %ENTRY_CODE(Sp(0));
751 jump %ENTRY_CODE(Sp(0));
754 u_digit = W_[BYTE_ARR_CTS(R2)];
756 if (u_digit == v_digit) {
758 jump %ENTRY_CODE(Sp(0));
761 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
767 jump %ENTRY_CODE(Sp(0));
772 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
773 W_ usize, vsize, size, up, vp;
776 // paraphrased from __gmpz_cmp() in the GMP sources
780 if (usize != vsize) {
782 jump %ENTRY_CODE(Sp(0));
787 jump %ENTRY_CODE(Sp(0));
790 if (%lt(usize,0)) { // NB. not <, which is unsigned
796 up = BYTE_ARR_CTS(R2);
797 vp = BYTE_ARR_CTS(R4);
799 (cmp) = foreign "C" __gmpn_cmp(up "ptr", vp "ptr", size) [];
801 if (cmp == 0 :: CInt) {
803 jump %ENTRY_CODE(Sp(0));
806 if (%lt(cmp,0 :: CInt) == %lt(usize,0)) {
811 /* Result parked in R1, return via info-pointer at TOS */
812 jump %ENTRY_CODE(Sp(0));
824 r = W_[R2 + SIZEOF_StgArrWords];
829 /* Result parked in R1, return via info-pointer at TOS */
831 jump %ENTRY_CODE(Sp(0));
843 r = W_[R2 + SIZEOF_StgArrWords];
848 /* Result parked in R1, return via info-pointer at TOS */
850 jump %ENTRY_CODE(Sp(0));
857 FETCH_MP_TEMP(mp_tmp1);
858 FETCH_MP_TEMP(mp_tmp_w);
860 /* arguments: F1 = Float# */
863 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
865 /* Be prepared to tell Lennart-coded __decodeFloat
866 where mantissa._mp_d can be put (it does not care about the rest) */
867 p = Hp - SIZEOF_StgArrWords;
868 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
869 StgArrWords_words(p) = 1;
870 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
872 /* Perform the operation */
873 foreign "C" __decodeFloat(mp_tmp1 "ptr",mp_tmp_w "ptr" ,arg) [];
875 /* returns: (Int# (expn), Int#, ByteArray#) */
876 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
879 decodeFloatzuIntzh_fast
883 FETCH_MP_TEMP(mp_tmp1);
884 FETCH_MP_TEMP(mp_tmp_w);
886 /* arguments: F1 = Float# */
889 /* Perform the operation */
890 foreign "C" __decodeFloat_Int(mp_tmp1 "ptr", mp_tmp_w "ptr", arg) [];
892 /* returns: (Int# (mantissa), Int# (exponent)) */
893 RET_NN(W_[mp_tmp1], W_[mp_tmp_w]);
896 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
897 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
903 FETCH_MP_TEMP(mp_tmp1);
904 FETCH_MP_TEMP(mp_tmp_w);
906 /* arguments: D1 = Double# */
909 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
911 /* Be prepared to tell Lennart-coded __decodeDouble
912 where mantissa.d can be put (it does not care about the rest) */
913 p = Hp - ARR_SIZE + WDS(1);
914 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
915 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
916 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
918 /* Perform the operation */
919 foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg) [];
921 /* returns: (Int# (expn), Int#, ByteArray#) */
922 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
925 decodeDoublezu2Intzh_fast
929 FETCH_MP_TEMP(mp_tmp1);
930 FETCH_MP_TEMP(mp_tmp2);
931 FETCH_MP_TEMP(mp_result1);
932 FETCH_MP_TEMP(mp_result2);
934 /* arguments: D1 = Double# */
937 /* Perform the operation */
938 foreign "C" __decodeDouble_2Int(mp_tmp1 "ptr", mp_tmp2 "ptr",
939 mp_result1 "ptr", mp_result2 "ptr",
943 (Int# (mant sign), Word# (mant high), Word# (mant low), Int# (expn)) */
944 RET_NNNN(W_[mp_tmp1], W_[mp_tmp2], W_[mp_result1], W_[mp_result2]);
947 /* -----------------------------------------------------------------------------
948 * Concurrency primitives
949 * -------------------------------------------------------------------------- */
953 /* args: R1 = closure to spark */
955 MAYBE_GC(R1_PTR, forkzh_fast);
961 ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr",
962 RtsFlags_GcFlags_initialStkSize(RtsFlags),
965 /* start blocked if the current thread is blocked */
966 StgTSO_flags(threadid) =
967 StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) &
968 (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));
970 foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") [];
972 // switch at the earliest opportunity
973 Capability_context_switch(MyCapability()) = 1 :: CInt;
980 /* args: R1 = cpu, R2 = closure to spark */
982 MAYBE_GC(R2_PTR, forkOnzh_fast);
990 ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr",
991 RtsFlags_GcFlags_initialStkSize(RtsFlags),
994 /* start blocked if the current thread is blocked */
995 StgTSO_flags(threadid) =
996 StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) &
997 (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));
999 foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") [];
1001 // switch at the earliest opportunity
1002 Capability_context_switch(MyCapability()) = 1 :: CInt;
1009 jump stg_yield_noregs;
1024 foreign "C" labelThread(R1 "ptr", R2 "ptr") [];
1026 jump %ENTRY_CODE(Sp(0));
1029 isCurrentThreadBoundzh_fast
1033 (r) = foreign "C" isThreadBound(CurrentTSO) [];
1039 /* args: R1 :: ThreadId# */
1047 if (TO_W_(StgTSO_what_next(tso)) == ThreadRelocated) {
1048 tso = StgTSO__link(tso);
1052 what_next = TO_W_(StgTSO_what_next(tso));
1053 why_blocked = TO_W_(StgTSO_why_blocked(tso));
1054 // Note: these two reads are not atomic, so they might end up
1055 // being inconsistent. It doesn't matter, since we
1056 // only return one or the other. If we wanted to return the
1057 // contents of block_info too, then we'd have to do some synchronisation.
1059 if (what_next == ThreadComplete) {
1060 ret = 16; // NB. magic, matches up with GHC.Conc.threadStatus
1062 if (what_next == ThreadKilled) {
1071 /* -----------------------------------------------------------------------------
1073 * -------------------------------------------------------------------------- */
1077 // Catch retry frame ------------------------------------------------------------
1079 INFO_TABLE_RET(stg_catch_retry_frame, CATCH_RETRY_FRAME,
1080 #if defined(PROFILING)
1081 W_ unused1, W_ unused2,
1083 W_ unused3, "ptr" W_ unused4, "ptr" W_ unused5)
1085 W_ r, frame, trec, outer;
1088 trec = StgTSO_trec(CurrentTSO);
1089 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1090 (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
1092 /* Succeeded (either first branch or second branch) */
1093 StgTSO_trec(CurrentTSO) = outer;
1094 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1095 jump %ENTRY_CODE(Sp(SP_OFF));
1097 /* Did not commit: re-execute */
1099 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1100 StgTSO_trec(CurrentTSO) = new_trec;
1101 if (StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
1102 R1 = StgCatchRetryFrame_alt_code(frame);
1104 R1 = StgCatchRetryFrame_first_code(frame);
1111 // Atomically frame ------------------------------------------------------------
1113 INFO_TABLE_RET(stg_atomically_frame, ATOMICALLY_FRAME,
1114 #if defined(PROFILING)
1115 W_ unused1, W_ unused2,
1117 "ptr" W_ unused3, "ptr" W_ unused4)
1119 W_ frame, trec, valid, next_invariant, q, outer;
1122 trec = StgTSO_trec(CurrentTSO);
1123 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1125 if (outer == NO_TREC) {
1126 /* First time back at the atomically frame -- pick up invariants */
1127 ("ptr" q) = foreign "C" stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr") [];
1128 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
1131 /* Second/subsequent time back at the atomically frame -- abort the
1132 * tx that's checking the invariant and move on to the next one */
1133 StgTSO_trec(CurrentTSO) = outer;
1134 q = StgAtomicallyFrame_next_invariant_to_check(frame);
1135 StgInvariantCheckQueue_my_execution(q) = trec;
1136 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1137 /* Don't free trec -- it's linked from q and will be stashed in the
1138 * invariant if we eventually commit. */
1139 q = StgInvariantCheckQueue_next_queue_entry(q);
1140 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
1144 q = StgAtomicallyFrame_next_invariant_to_check(frame);
1146 if (q != END_INVARIANT_CHECK_QUEUE) {
1147 /* We can't commit yet: another invariant to check */
1148 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [];
1149 StgTSO_trec(CurrentTSO) = trec;
1151 next_invariant = StgInvariantCheckQueue_invariant(q);
1152 R1 = StgAtomicInvariant_code(next_invariant);
1157 /* We've got no more invariants to check, try to commit */
1158 (valid) = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") [];
1160 /* Transaction was valid: commit succeeded */
1161 StgTSO_trec(CurrentTSO) = NO_TREC;
1162 Sp = Sp + SIZEOF_StgAtomicallyFrame;
1163 jump %ENTRY_CODE(Sp(SP_OFF));
1165 /* Transaction was not valid: try again */
1166 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1167 StgTSO_trec(CurrentTSO) = trec;
1168 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
1169 R1 = StgAtomicallyFrame_code(frame);
1175 INFO_TABLE_RET(stg_atomically_waiting_frame, ATOMICALLY_FRAME,
1176 #if defined(PROFILING)
1177 W_ unused1, W_ unused2,
1179 "ptr" W_ unused3, "ptr" W_ unused4)
1181 W_ frame, trec, valid;
1185 /* The TSO is currently waiting: should we stop waiting? */
1186 (valid) = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") [];
1188 /* Previous attempt is still valid: no point trying again yet */
1189 jump stg_block_noregs;
1191 /* Previous attempt is no longer valid: try again */
1192 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1193 StgTSO_trec(CurrentTSO) = trec;
1194 StgHeader_info(frame) = stg_atomically_frame_info;
1195 R1 = StgAtomicallyFrame_code(frame);
1200 // STM catch frame --------------------------------------------------------------
1204 /* Catch frames are very similar to update frames, but when entering
1205 * one we just pop the frame off the stack and perform the correct
1206 * kind of return to the activation record underneath us on the stack.
1209 INFO_TABLE_RET(stg_catch_stm_frame, CATCH_STM_FRAME,
1210 #if defined(PROFILING)
1211 W_ unused1, W_ unused2,
1213 "ptr" W_ unused3, "ptr" W_ unused4)
1215 W_ r, frame, trec, outer;
1217 trec = StgTSO_trec(CurrentTSO);
1218 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1219 (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
1221 /* Commit succeeded */
1222 StgTSO_trec(CurrentTSO) = outer;
1223 Sp = Sp + SIZEOF_StgCatchSTMFrame;
1228 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1229 StgTSO_trec(CurrentTSO) = new_trec;
1230 R1 = StgCatchSTMFrame_code(frame);
1236 // Primop definition ------------------------------------------------------------
1244 // stmStartTransaction may allocate
1245 MAYBE_GC (R1_PTR, atomicallyzh_fast);
1247 /* Args: R1 = m :: STM a */
1248 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);
1250 old_trec = StgTSO_trec(CurrentTSO);
1252 /* Nested transactions are not allowed; raise an exception */
1253 if (old_trec != NO_TREC) {
1254 R1 = base_ControlziExceptionziBase_nestedAtomically_closure;
1258 /* Set up the atomically frame */
1259 Sp = Sp - SIZEOF_StgAtomicallyFrame;
1262 SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
1263 StgAtomicallyFrame_code(frame) = R1;
1264 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
1266 /* Start the memory transcation */
1267 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1];
1268 StgTSO_trec(CurrentTSO) = new_trec;
1270 /* Apply R1 to the realworld token */
1279 /* Args: R1 :: STM a */
1280 /* Args: R2 :: Exception -> STM a */
1281 STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast);
1283 /* Set up the catch frame */
1284 Sp = Sp - SIZEOF_StgCatchSTMFrame;
1287 SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
1288 StgCatchSTMFrame_handler(frame) = R2;
1289 StgCatchSTMFrame_code(frame) = R1;
1291 /* Start a nested transaction to run the body of the try block in */
1294 cur_trec = StgTSO_trec(CurrentTSO);
1295 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", cur_trec "ptr");
1296 StgTSO_trec(CurrentTSO) = new_trec;
1298 /* Apply R1 to the realworld token */
1309 // stmStartTransaction may allocate
1310 MAYBE_GC (R1_PTR & R2_PTR, catchRetryzh_fast);
1312 /* Args: R1 :: STM a */
1313 /* Args: R2 :: STM a */
1314 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);
1316 /* Start a nested transaction within which to run the first code */
1317 trec = StgTSO_trec(CurrentTSO);
1318 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2];
1319 StgTSO_trec(CurrentTSO) = new_trec;
1321 /* Set up the catch-retry frame */
1322 Sp = Sp - SIZEOF_StgCatchRetryFrame;
1325 SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
1326 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1327 StgCatchRetryFrame_first_code(frame) = R1;
1328 StgCatchRetryFrame_alt_code(frame) = R2;
1330 /* Apply R1 to the realworld token */
1343 MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate
1345 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
1347 StgTSO_sp(CurrentTSO) = Sp;
1348 (frame_type) = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") [];
1349 Sp = StgTSO_sp(CurrentTSO);
1351 trec = StgTSO_trec(CurrentTSO);
1352 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1354 if (frame_type == CATCH_RETRY_FRAME) {
1355 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
1356 ASSERT(outer != NO_TREC);
1357 // Abort the transaction attempting the current branch
1358 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1359 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
1360 if (!StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
1361 // Retry in the first branch: try the alternative
1362 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1363 StgTSO_trec(CurrentTSO) = trec;
1364 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
1365 R1 = StgCatchRetryFrame_alt_code(frame);
1368 // Retry in the alternative code: propagate the retry
1369 StgTSO_trec(CurrentTSO) = outer;
1370 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1371 goto retry_pop_stack;
1375 // We've reached the ATOMICALLY_FRAME: attempt to wait
1376 ASSERT(frame_type == ATOMICALLY_FRAME);
1377 if (outer != NO_TREC) {
1378 // We called retry while checking invariants, so abort the current
1379 // invariant check (merging its TVar accesses into the parents read
1380 // set so we'll wait on them)
1381 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1382 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
1384 StgTSO_trec(CurrentTSO) = trec;
1385 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1387 ASSERT(outer == NO_TREC);
1389 (r) = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") [];
1391 // Transaction was valid: stmWait put us on the TVars' queues, we now block
1392 StgHeader_info(frame) = stg_atomically_waiting_frame_info;
1394 // Fix up the stack in the unregisterised case: the return convention is different.
1395 R3 = trec; // passing to stmWaitUnblock()
1396 jump stg_block_stmwait;
1398 // Transaction was not valid: retry immediately
1399 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1400 StgTSO_trec(CurrentTSO) = trec;
1401 R1 = StgAtomicallyFrame_code(frame);
1412 /* Args: R1 = invariant closure */
1413 MAYBE_GC (R1_PTR, checkzh_fast);
1415 trec = StgTSO_trec(CurrentTSO);
1417 foreign "C" stmAddInvariantToCheck(MyCapability() "ptr",
1421 jump %ENTRY_CODE(Sp(0));
1430 /* Args: R1 = initialisation value */
1432 MAYBE_GC (R1_PTR, newTVarzh_fast);
1434 ("ptr" tv) = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") [];
1445 /* Args: R1 = TVar closure */
1447 MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
1448 trec = StgTSO_trec(CurrentTSO);
1450 ("ptr" result) = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") [];
1462 /* Args: R1 = TVar closure */
1463 /* R2 = New value */
1465 MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
1466 trec = StgTSO_trec(CurrentTSO);
1469 foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") [];
1471 jump %ENTRY_CODE(Sp(0));
1475 /* -----------------------------------------------------------------------------
1478 * take & putMVar work as follows. Firstly, an important invariant:
1480 * If the MVar is full, then the blocking queue contains only
1481 * threads blocked on putMVar, and if the MVar is empty then the
1482 * blocking queue contains only threads blocked on takeMVar.
1485 * MVar empty : then add ourselves to the blocking queue
1486 * MVar full : remove the value from the MVar, and
1487 * blocking queue empty : return
1488 * blocking queue non-empty : perform the first blocked putMVar
1489 * from the queue, and wake up the
1490 * thread (MVar is now full again)
1492 * putMVar is just the dual of the above algorithm.
1494 * How do we "perform a putMVar"? Well, we have to fiddle around with
1495 * the stack of the thread waiting to do the putMVar. See
1496 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1497 * the stack layout, and the PerformPut and PerformTake macros below.
1499 * It is important that a blocked take or put is woken up with the
1500 * take/put already performed, because otherwise there would be a
1501 * small window of vulnerability where the thread could receive an
1502 * exception and never perform its take or put, and we'd end up with a
1505 * -------------------------------------------------------------------------- */
1509 /* args: R1 = MVar closure */
1511 if (StgMVar_value(R1) == stg_END_TSO_QUEUE_closure) {
1523 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
1525 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1526 SET_HDR(mvar,stg_MVAR_DIRTY_info,W_[CCCS]);
1527 // MVARs start dirty: generation 0 has no mutable list
1528 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1529 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1530 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1535 #define PerformTake(tso, value) \
1536 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1537 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1539 #define PerformPut(tso,lval) \
1540 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1541 lval = W_[StgTSO_sp(tso) - WDS(1)];
1545 W_ mvar, val, info, tso;
1547 /* args: R1 = MVar closure */
1550 #if defined(THREADED_RTS)
1551 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1553 info = GET_INFO(mvar);
1556 if (info == stg_MVAR_CLEAN_info) {
1557 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr") [];
1560 /* If the MVar is empty, put ourselves on its blocking queue,
1561 * and wait until we're woken up.
1563 if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1564 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1565 StgMVar_head(mvar) = CurrentTSO;
1567 foreign "C" setTSOLink(MyCapability() "ptr",
1568 StgMVar_tail(mvar) "ptr",
1571 StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1572 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1573 StgTSO_block_info(CurrentTSO) = mvar;
1574 StgMVar_tail(mvar) = CurrentTSO;
1577 jump stg_block_takemvar;
1580 /* we got the value... */
1581 val = StgMVar_value(mvar);
1583 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1585 /* There are putMVar(s) waiting...
1586 * wake up the first thread on the queue
1588 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1590 /* actually perform the putMVar for the thread that we just woke up */
1591 tso = StgMVar_head(mvar);
1592 PerformPut(tso,StgMVar_value(mvar));
1594 if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1595 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1598 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1599 StgMVar_head(mvar) "ptr", 1) [];
1600 StgMVar_head(mvar) = tso;
1602 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1603 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1606 #if defined(THREADED_RTS)
1607 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1609 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1615 /* No further putMVars, MVar is now empty */
1616 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1618 #if defined(THREADED_RTS)
1619 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1621 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1631 W_ mvar, val, info, tso;
1633 /* args: R1 = MVar closure */
1637 #if defined(THREADED_RTS)
1638 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1640 info = GET_INFO(mvar);
1643 if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1644 #if defined(THREADED_RTS)
1645 unlockClosure(mvar, info);
1647 /* HACK: we need a pointer to pass back,
1648 * so we abuse NO_FINALIZER_closure
1650 RET_NP(0, stg_NO_FINALIZER_closure);
1653 if (info == stg_MVAR_CLEAN_info) {
1654 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1657 /* we got the value... */
1658 val = StgMVar_value(mvar);
1660 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1662 /* There are putMVar(s) waiting...
1663 * wake up the first thread on the queue
1665 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1667 /* actually perform the putMVar for the thread that we just woke up */
1668 tso = StgMVar_head(mvar);
1669 PerformPut(tso,StgMVar_value(mvar));
1670 if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1671 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1674 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1675 StgMVar_head(mvar) "ptr", 1) [];
1676 StgMVar_head(mvar) = tso;
1678 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1679 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1681 #if defined(THREADED_RTS)
1682 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1684 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1689 /* No further putMVars, MVar is now empty */
1690 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1691 #if defined(THREADED_RTS)
1692 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1694 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1704 W_ mvar, val, info, tso;
1706 /* args: R1 = MVar, R2 = value */
1710 #if defined(THREADED_RTS)
1711 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1713 info = GET_INFO(mvar);
1716 if (info == stg_MVAR_CLEAN_info) {
1717 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1720 if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) {
1721 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1722 StgMVar_head(mvar) = CurrentTSO;
1724 foreign "C" setTSOLink(MyCapability() "ptr",
1725 StgMVar_tail(mvar) "ptr",
1728 StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1729 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1730 StgTSO_block_info(CurrentTSO) = mvar;
1731 StgMVar_tail(mvar) = CurrentTSO;
1735 jump stg_block_putmvar;
1738 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1740 /* There are takeMVar(s) waiting: wake up the first one
1742 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1744 /* actually perform the takeMVar */
1745 tso = StgMVar_head(mvar);
1746 PerformTake(tso, val);
1747 if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1748 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1751 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1752 StgMVar_head(mvar) "ptr", 1) [];
1753 StgMVar_head(mvar) = tso;
1755 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1756 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1759 #if defined(THREADED_RTS)
1760 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1762 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1764 jump %ENTRY_CODE(Sp(0));
1768 /* No further takes, the MVar is now full. */
1769 StgMVar_value(mvar) = val;
1771 #if defined(THREADED_RTS)
1772 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1774 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1776 jump %ENTRY_CODE(Sp(0));
1779 /* ToDo: yield afterward for better communication performance? */
1787 /* args: R1 = MVar, R2 = value */
1790 #if defined(THREADED_RTS)
1791 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [R2];
1793 info = GET_INFO(mvar);
1796 if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) {
1797 #if defined(THREADED_RTS)
1798 unlockClosure(mvar, info);
1803 if (info == stg_MVAR_CLEAN_info) {
1804 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1807 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1809 /* There are takeMVar(s) waiting: wake up the first one
1811 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1813 /* actually perform the takeMVar */
1814 tso = StgMVar_head(mvar);
1815 PerformTake(tso, R2);
1816 if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1817 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1820 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1821 StgMVar_head(mvar) "ptr", 1) [];
1822 StgMVar_head(mvar) = tso;
1824 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1825 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1828 #if defined(THREADED_RTS)
1829 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1831 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1836 /* No further takes, the MVar is now full. */
1837 StgMVar_value(mvar) = R2;
1839 #if defined(THREADED_RTS)
1840 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1842 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1847 /* ToDo: yield afterward for better communication performance? */
1851 /* -----------------------------------------------------------------------------
1852 Stable pointer primitives
1853 ------------------------------------------------------------------------- */
1855 makeStableNamezh_fast
1859 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1861 (index) = foreign "C" lookupStableName(R1 "ptr") [];
1863 /* Is there already a StableName for this heap object?
1864 * stable_ptr_table is a pointer to an array of snEntry structs.
1866 if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) {
1867 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1868 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1869 StgStableName_sn(sn_obj) = index;
1870 snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj;
1872 sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry);
1879 makeStablePtrzh_fast
1883 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1884 ("ptr" sp) = foreign "C" getStablePtr(R1 "ptr") [];
1888 deRefStablePtrzh_fast
1890 /* Args: R1 = the stable ptr */
1893 r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry);
1897 /* -----------------------------------------------------------------------------
1898 Bytecode object primitives
1899 ------------------------------------------------------------------------- */
1909 W_ bco, bitmap_arr, bytes, words;
1913 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1916 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R5_PTR, newBCOzh_fast );
1918 bco = Hp - bytes + WDS(1);
1919 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1921 StgBCO_instrs(bco) = R1;
1922 StgBCO_literals(bco) = R2;
1923 StgBCO_ptrs(bco) = R3;
1924 StgBCO_arity(bco) = HALF_W_(R4);
1925 StgBCO_size(bco) = HALF_W_(words);
1927 // Copy the arity/bitmap info into the BCO
1931 if (i < StgArrWords_words(bitmap_arr)) {
1932 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1943 // R1 = the BCO# for the AP
1947 // This function is *only* used to wrap zero-arity BCOs in an
1948 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1949 // saturated and always points directly to a FUN or BCO.
1950 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
1951 StgBCO_arity(R1) == HALF_W_(0));
1953 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1954 TICK_ALLOC_UP_THK(0, 0);
1955 CCCS_ALLOC(SIZEOF_StgAP);
1957 ap = Hp - SIZEOF_StgAP + WDS(1);
1958 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1960 StgAP_n_args(ap) = HALF_W_(0);
1966 unpackClosurezh_fast
1968 /* args: R1 = closure to analyze */
1969 // TODO: Consider the absence of ptrs or nonptrs as a special case ?
1971 W_ info, ptrs, nptrs, p, ptrs_arr, nptrs_arr;
1972 info = %GET_STD_INFO(UNTAG(R1));
1974 // Some closures have non-standard layout, so we omit those here.
1976 type = TO_W_(%INFO_TYPE(info));
1977 switch [0 .. N_CLOSURE_TYPES] type {
1978 case THUNK_SELECTOR : {
1983 case THUNK, THUNK_1_0, THUNK_0_1, THUNK_2_0, THUNK_1_1,
1984 THUNK_0_2, THUNK_STATIC, AP, PAP, AP_STACK, BCO : {
1990 ptrs = TO_W_(%INFO_PTRS(info));
1991 nptrs = TO_W_(%INFO_NPTRS(info));
1996 W_ ptrs_arr_sz, nptrs_arr_sz;
1997 nptrs_arr_sz = SIZEOF_StgArrWords + WDS(nptrs);
1998 ptrs_arr_sz = SIZEOF_StgMutArrPtrs + WDS(ptrs);
2000 ALLOC_PRIM (ptrs_arr_sz + nptrs_arr_sz, R1_PTR, unpackClosurezh_fast);
2005 ptrs_arr = Hp - nptrs_arr_sz - ptrs_arr_sz + WDS(1);
2006 nptrs_arr = Hp - nptrs_arr_sz + WDS(1);
2008 SET_HDR(ptrs_arr, stg_MUT_ARR_PTRS_FROZEN_info, W_[CCCS]);
2009 StgMutArrPtrs_ptrs(ptrs_arr) = ptrs;
2013 W_[ptrs_arr + SIZEOF_StgMutArrPtrs + WDS(p)] = StgClosure_payload(clos,p);
2018 SET_HDR(nptrs_arr, stg_ARR_WORDS_info, W_[CCCS]);
2019 StgArrWords_words(nptrs_arr) = nptrs;
2023 W_[BYTE_ARR_CTS(nptrs_arr) + WDS(p)] = StgClosure_payload(clos, p+ptrs);
2027 RET_NPP(info, ptrs_arr, nptrs_arr);
2030 /* -----------------------------------------------------------------------------
2031 Thread I/O blocking primitives
2032 -------------------------------------------------------------------------- */
2034 /* Add a thread to the end of the blocked queue. (C-- version of the C
2035 * macro in Schedule.h).
2037 #define APPEND_TO_BLOCKED_QUEUE(tso) \
2038 ASSERT(StgTSO__link(tso) == END_TSO_QUEUE); \
2039 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
2040 W_[blocked_queue_hd] = tso; \
2042 foreign "C" setTSOLink(MyCapability() "ptr", W_[blocked_queue_tl] "ptr", tso) []; \
2044 W_[blocked_queue_tl] = tso;
2050 foreign "C" barf("waitRead# on threaded RTS") never returns;
2053 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2054 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
2055 StgTSO_block_info(CurrentTSO) = R1;
2056 // No locking - we're not going to use this interface in the
2057 // threaded RTS anyway.
2058 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2059 jump stg_block_noregs;
2067 foreign "C" barf("waitWrite# on threaded RTS") never returns;
2070 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2071 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2072 StgTSO_block_info(CurrentTSO) = R1;
2073 // No locking - we're not going to use this interface in the
2074 // threaded RTS anyway.
2075 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2076 jump stg_block_noregs;
2081 STRING(stg_delayzh_malloc_str, "delayzh_fast")
2084 #ifdef mingw32_HOST_OS
2092 foreign "C" barf("delay# on threaded RTS") never returns;
2095 /* args: R1 (microsecond delay amount) */
2096 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2097 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
2099 #ifdef mingw32_HOST_OS
2101 /* could probably allocate this on the heap instead */
2102 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2103 stg_delayzh_malloc_str);
2104 (reqID) = foreign "C" addDelayRequest(R1);
2105 StgAsyncIOResult_reqID(ares) = reqID;
2106 StgAsyncIOResult_len(ares) = 0;
2107 StgAsyncIOResult_errCode(ares) = 0;
2108 StgTSO_block_info(CurrentTSO) = ares;
2110 /* Having all async-blocked threads reside on the blocked_queue
2111 * simplifies matters, so change the status to OnDoProc put the
2112 * delayed thread on the blocked_queue.
2114 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2115 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2116 jump stg_block_async_void;
2122 (time) = foreign "C" getourtimeofday() [R1];
2123 divisor = TO_W_(RtsFlags_MiscFlags_tickInterval(RtsFlags));
2127 divisor = divisor * 1000;
2128 target = ((R1 + divisor - 1) / divisor) /* divide rounding up */
2129 + time + 1; /* Add 1 as getourtimeofday rounds down */
2130 StgTSO_block_info(CurrentTSO) = target;
2132 /* Insert the new thread in the sleeping queue. */
2134 t = W_[sleeping_queue];
2136 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
2138 t = StgTSO__link(t);
2142 StgTSO__link(CurrentTSO) = t;
2144 W_[sleeping_queue] = CurrentTSO;
2146 foreign "C" setTSOLink(MyCapability() "ptr", prev "ptr", CurrentTSO) [];
2148 jump stg_block_noregs;
2150 #endif /* !THREADED_RTS */
2154 #ifdef mingw32_HOST_OS
2155 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
2162 foreign "C" barf("asyncRead# on threaded RTS") never returns;
2165 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2166 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2167 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
2169 /* could probably allocate this on the heap instead */
2170 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2171 stg_asyncReadzh_malloc_str)
2173 (reqID) = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr") [];
2174 StgAsyncIOResult_reqID(ares) = reqID;
2175 StgAsyncIOResult_len(ares) = 0;
2176 StgAsyncIOResult_errCode(ares) = 0;
2177 StgTSO_block_info(CurrentTSO) = ares;
2178 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2179 jump stg_block_async;
2183 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
2190 foreign "C" barf("asyncWrite# on threaded RTS") never returns;
2193 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2194 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2195 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2197 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2198 stg_asyncWritezh_malloc_str)
2200 (reqID) = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr") [];
2202 StgAsyncIOResult_reqID(ares) = reqID;
2203 StgAsyncIOResult_len(ares) = 0;
2204 StgAsyncIOResult_errCode(ares) = 0;
2205 StgTSO_block_info(CurrentTSO) = ares;
2206 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2207 jump stg_block_async;
2211 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
2218 foreign "C" barf("asyncDoProc# on threaded RTS") never returns;
2221 /* args: R1 = proc, R2 = param */
2222 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2223 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2225 /* could probably allocate this on the heap instead */
2226 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2227 stg_asyncDoProczh_malloc_str)
2229 (reqID) = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr") [];
2230 StgAsyncIOResult_reqID(ares) = reqID;
2231 StgAsyncIOResult_len(ares) = 0;
2232 StgAsyncIOResult_errCode(ares) = 0;
2233 StgTSO_block_info(CurrentTSO) = ares;
2234 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2235 jump stg_block_async;
2240 // noDuplicate# tries to ensure that none of the thunks under
2241 // evaluation by the current thread are also under evaluation by
2242 // another thread. It relies on *both* threads doing noDuplicate#;
2243 // the second one will get blocked if they are duplicating some work.
2246 SAVE_THREAD_STATE();
2247 ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16);
2248 foreign "C" threadPaused (MyCapability() "ptr", CurrentTSO "ptr") [];
2250 if (StgTSO_what_next(CurrentTSO) == ThreadKilled::I16) {
2251 jump stg_threadFinished;
2253 LOAD_THREAD_STATE();
2254 ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16);
2255 jump %ENTRY_CODE(Sp(0));
2259 getApStackValzh_fast
2261 W_ ap_stack, offset, val, ok;
2263 /* args: R1 = AP_STACK, R2 = offset */
2267 if (%INFO_PTR(ap_stack) == stg_AP_STACK_info) {
2269 val = StgAP_STACK_payload(ap_stack,offset);