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(BaseReg "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(BaseReg "ptr",words);
101 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
103 SET_HDR(arr, stg_MUT_ARR_PTRS_info, W_[CCCS]);
104 StgMutArrPtrs_ptrs(arr) = n;
106 // Initialise all elements of the the array with the value in R2
108 p = arr + SIZEOF_StgMutArrPtrs;
110 if (p < arr + WDS(words)) {
119 unsafeThawArrayzh_fast
121 // 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 to indicate
130 // 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 mut_link field is NULL if it isn't on a mut_list, and the GC
135 // maintains this invariant.
137 if (%INFO_TYPE(%GET_STD_INFO(R1)) != HALF_W_(MUT_ARR_PTRS_FROZEN0)) {
138 foreign "C" recordMutableLock(R1 "ptr");
141 SET_INFO(R1,stg_MUT_ARR_PTRS_info);
146 /* -----------------------------------------------------------------------------
148 -------------------------------------------------------------------------- */
153 /* Args: R1 = initialisation value */
155 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);
157 mv = Hp - SIZEOF_StgMutVar + WDS(1);
158 SET_HDR(mv,stg_MUT_VAR_info,W_[CCCS]);
159 StgMutVar_var(mv) = R1;
164 atomicModifyMutVarzh_fast
167 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
169 /* If x is the current contents of the MutVar#, then
170 We want to make the new contents point to
174 and the return value is
178 obviously we can share (f x).
180 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
181 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
182 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
186 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
187 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
189 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
190 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
194 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
195 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
197 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
198 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
201 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
203 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);
206 foreign "C" ACQUIRE_LOCK(sm_mutex "ptr");
209 x = StgMutVar_var(R1);
211 TICK_ALLOC_THUNK_2();
212 CCCS_ALLOC(THUNK_2_SIZE);
213 z = Hp - THUNK_2_SIZE + WDS(1);
214 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
215 LDV_RECORD_CREATE(z);
216 StgThunk_payload(z,0) = R2;
217 StgThunk_payload(z,1) = x;
219 TICK_ALLOC_THUNK_1();
220 CCCS_ALLOC(THUNK_1_SIZE);
221 y = z - THUNK_1_SIZE;
222 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
223 LDV_RECORD_CREATE(y);
224 StgThunk_payload(y,0) = z;
226 StgMutVar_var(R1) = y;
228 TICK_ALLOC_THUNK_1();
229 CCCS_ALLOC(THUNK_1_SIZE);
230 r = y - THUNK_1_SIZE;
231 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
232 LDV_RECORD_CREATE(r);
233 StgThunk_payload(r,0) = z;
236 foreign "C" RELEASE_LOCK(sm_mutex "ptr");
242 /* -----------------------------------------------------------------------------
243 Foreign Object Primitives
244 -------------------------------------------------------------------------- */
248 /* R1 = ptr to foreign object,
252 ALLOC_PRIM( SIZEOF_StgForeignObj, NO_PTRS, mkForeignObjzh_fast);
254 result = Hp - SIZEOF_StgForeignObj + WDS(1);
255 SET_HDR(result,stg_FOREIGN_info,W_[CCCS]);
256 StgForeignObj_data(result) = R1;
258 /* returns (# s#, ForeignObj# #) */
262 /* -----------------------------------------------------------------------------
263 Weak Pointer Primitives
264 -------------------------------------------------------------------------- */
266 STRING(stg_weak_msg,"New weak pointer at %p\n")
272 R3 = finalizer (or NULL)
277 R3 = stg_NO_FINALIZER_closure;
280 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
282 w = Hp - SIZEOF_StgWeak + WDS(1);
283 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
286 StgWeak_value(w) = R2;
287 StgWeak_finalizer(w) = R3;
289 StgWeak_link(w) = W_[weak_ptr_list];
290 W_[weak_ptr_list] = w;
292 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w));
307 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
308 RET_NP(0,stg_NO_FINALIZER_closure);
314 // A weak pointer is inherently used, so we do not need to call
315 // LDV_recordDead_FILL_SLOP_DYNAMIC():
316 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
317 // or, LDV_recordDead():
318 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
319 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
320 // large as weak pointers, so there is no need to fill the slop, either.
321 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
325 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
327 SET_INFO(w,stg_DEAD_WEAK_info);
328 LDV_RECORD_CREATE(w);
330 f = StgWeak_finalizer(w);
331 StgDeadWeak_link(w) = StgWeak_link(w);
333 /* return the finalizer */
334 if (f == stg_NO_FINALIZER_closure) {
335 RET_NP(0,stg_NO_FINALIZER_closure);
347 if (GET_INFO(w) == stg_WEAK_info) {
349 val = StgWeak_value(w);
357 /* -----------------------------------------------------------------------------
358 Arbitrary-precision Integer operations.
360 There are some assumptions in this code that mp_limb_t == W_. This is
361 the case for all the platforms that GHC supports, currently.
362 -------------------------------------------------------------------------- */
366 /* arguments: R1 = Int# */
368 W_ val, s, p; /* to avoid aliasing */
371 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
373 p = Hp - SIZEOF_StgArrWords;
374 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
375 StgArrWords_words(p) = 1;
377 /* mpz_set_si is inlined here, makes things simpler */
390 /* returns (# size :: Int#,
399 /* arguments: R1 = Word# */
401 W_ val, s, p; /* to avoid aliasing */
405 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
407 p = Hp - SIZEOF_StgArrWords;
408 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
409 StgArrWords_words(p) = 1;
418 /* returns (# size :: Int#,
419 data :: ByteArray# #)
426 * 'long long' primops for converting to/from Integers.
429 #ifdef SUPPORT_LONG_LONGS
431 int64ToIntegerzh_fast
433 /* arguments: L1 = Int64# */
436 W_ hi, s, neg, words_needed, p;
441 if ( %ge(val,0x100000000::L_) || %le(val,-0x100000000::L_) ) {
444 // minimum is one word
448 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
449 NO_PTRS, int64ToIntegerzh_fast );
451 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
452 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
453 StgArrWords_words(p) = words_needed;
455 if ( %lt(val,0::L_) ) {
460 hi = TO_W_(val >> 32);
462 if ( words_needed == 2 ) {
467 if ( val != 0::L_ ) {
470 } else /* val==0 */ {
478 /* returns (# size :: Int#,
479 data :: ByteArray# #)
484 word64ToIntegerzh_fast
486 /* arguments: L1 = Word64# */
489 W_ hi, s, words_needed, p;
492 if ( val >= 0x100000000::L_ ) {
498 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
499 NO_PTRS, word64ToIntegerzh_fast );
501 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
502 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
503 StgArrWords_words(p) = words_needed;
505 hi = TO_W_(val >> 32);
506 if ( val >= 0x100000000::L_ ) {
511 if ( val != 0::L_ ) {
514 } else /* val==0 */ {
519 /* returns (# size :: Int#,
520 data :: ByteArray# #)
526 #endif /* SUPPORT_LONG_LONGS */
528 /* ToDo: this is shockingly inefficient */
533 bits8 [SIZEOF_MP_INT];
538 bits8 [SIZEOF_MP_INT];
543 bits8 [SIZEOF_MP_INT];
548 bits8 [SIZEOF_MP_INT];
553 #define FETCH_MP_TEMP(X) \
555 X = BaseReg + (OFFSET_StgRegTable_r ## X);
557 #define FETCH_MP_TEMP(X) /* Nothing */
560 #define GMP_TAKE2_RET1(name,mp_fun) \
565 FETCH_MP_TEMP(mp_tmp1); \
566 FETCH_MP_TEMP(mp_tmp2); \
567 FETCH_MP_TEMP(mp_result1) \
568 FETCH_MP_TEMP(mp_result2); \
570 /* call doYouWantToGC() */ \
571 MAYBE_GC(R2_PTR & R4_PTR, name); \
578 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
579 MP_INT__mp_size(mp_tmp1) = (s1); \
580 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
581 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
582 MP_INT__mp_size(mp_tmp2) = (s2); \
583 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
585 foreign "C" mpz_init(mp_result1 "ptr"); \
587 /* Perform the operation */ \
588 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr"); \
590 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
591 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
594 #define GMP_TAKE1_RET1(name,mp_fun) \
599 FETCH_MP_TEMP(mp_tmp1); \
600 FETCH_MP_TEMP(mp_result1) \
602 /* call doYouWantToGC() */ \
603 MAYBE_GC(R2_PTR, name); \
608 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
609 MP_INT__mp_size(mp_tmp1) = (s1); \
610 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
612 foreign "C" mpz_init(mp_result1 "ptr"); \
614 /* Perform the operation */ \
615 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr"); \
617 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
618 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
621 #define GMP_TAKE2_RET2(name,mp_fun) \
626 FETCH_MP_TEMP(mp_tmp1); \
627 FETCH_MP_TEMP(mp_tmp2); \
628 FETCH_MP_TEMP(mp_result1) \
629 FETCH_MP_TEMP(mp_result2) \
631 /* call doYouWantToGC() */ \
632 MAYBE_GC(R2_PTR & R4_PTR, name); \
639 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
640 MP_INT__mp_size(mp_tmp1) = (s1); \
641 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
642 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
643 MP_INT__mp_size(mp_tmp2) = (s2); \
644 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
646 foreign "C" mpz_init(mp_result1 "ptr"); \
647 foreign "C" mpz_init(mp_result2 "ptr"); \
649 /* Perform the operation */ \
650 foreign "C" mp_fun(mp_result1 "ptr",mp_result2 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr"); \
652 RET_NPNP(TO_W_(MP_INT__mp_size(mp_result1)), \
653 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords, \
654 TO_W_(MP_INT__mp_size(mp_result2)), \
655 MP_INT__mp_d(mp_result2) - SIZEOF_StgArrWords); \
658 GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add)
659 GMP_TAKE2_RET1(minusIntegerzh_fast, mpz_sub)
660 GMP_TAKE2_RET1(timesIntegerzh_fast, mpz_mul)
661 GMP_TAKE2_RET1(gcdIntegerzh_fast, mpz_gcd)
662 GMP_TAKE2_RET1(quotIntegerzh_fast, mpz_tdiv_q)
663 GMP_TAKE2_RET1(remIntegerzh_fast, mpz_tdiv_r)
664 GMP_TAKE2_RET1(divExactIntegerzh_fast, mpz_divexact)
665 GMP_TAKE2_RET1(andIntegerzh_fast, mpz_and)
666 GMP_TAKE2_RET1(orIntegerzh_fast, mpz_ior)
667 GMP_TAKE2_RET1(xorIntegerzh_fast, mpz_xor)
668 GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com)
670 GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr)
671 GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr)
675 mp_tmp_w: W_; // NB. mp_tmp_w is really an here mp_limb_t
681 /* R1 = the first Int#; R2 = the second Int# */
683 FETCH_MP_TEMP(mp_tmp_w);
686 r = foreign "C" mpn_gcd_1(mp_tmp_w "ptr", 1, R2);
689 /* Result parked in R1, return via info-pointer at TOS */
690 jump %ENTRY_CODE(Sp(0));
696 /* R1 = s1; R2 = d1; R3 = the int */
697 R1 = foreign "C" mpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3);
699 /* Result parked in R1, return via info-pointer at TOS */
700 jump %ENTRY_CODE(Sp(0));
706 /* R1 = s1; R2 = d1; R3 = the int */
707 W_ usize, vsize, v_digit, u_digit;
713 // paraphrased from mpz_cmp_si() in the GMP sources
714 if (%gt(v_digit,0)) {
717 if (%lt(v_digit,0)) {
723 if (usize != vsize) {
725 jump %ENTRY_CODE(Sp(0));
730 jump %ENTRY_CODE(Sp(0));
733 u_digit = W_[BYTE_ARR_CTS(R2)];
735 if (u_digit == v_digit) {
737 jump %ENTRY_CODE(Sp(0));
740 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
746 jump %ENTRY_CODE(Sp(0));
751 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
752 W_ usize, vsize, size, up, vp;
755 // paraphrased from mpz_cmp() in the GMP sources
759 if (usize != vsize) {
761 jump %ENTRY_CODE(Sp(0));
766 jump %ENTRY_CODE(Sp(0));
769 if (%lt(usize,0)) { // NB. not <, which is unsigned
775 up = BYTE_ARR_CTS(R2);
776 vp = BYTE_ARR_CTS(R4);
778 cmp = foreign "C" mpn_cmp(up "ptr", vp "ptr", size);
780 if (cmp == 0 :: CInt) {
782 jump %ENTRY_CODE(Sp(0));
785 if (%lt(cmp,0 :: CInt) == %lt(usize,0)) {
790 /* Result parked in R1, return via info-pointer at TOS */
791 jump %ENTRY_CODE(Sp(0));
803 r = W_[R2 + SIZEOF_StgArrWords];
808 /* Result parked in R1, return via info-pointer at TOS */
810 jump %ENTRY_CODE(Sp(0));
822 r = W_[R2 + SIZEOF_StgArrWords];
827 /* Result parked in R1, return via info-pointer at TOS */
829 jump %ENTRY_CODE(Sp(0));
836 FETCH_MP_TEMP(mp_tmp1);
837 FETCH_MP_TEMP(mp_tmp_w);
839 /* arguments: F1 = Float# */
842 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
844 /* Be prepared to tell Lennart-coded __decodeFloat
845 where mantissa._mp_d can be put (it does not care about the rest) */
846 p = Hp - SIZEOF_StgArrWords;
847 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
848 StgArrWords_words(p) = 1;
849 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
851 /* Perform the operation */
852 foreign "C" __decodeFloat(mp_tmp1 "ptr",mp_tmp_w "ptr" ,arg);
854 /* returns: (Int# (expn), Int#, ByteArray#) */
855 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
858 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
859 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
865 FETCH_MP_TEMP(mp_tmp1);
866 FETCH_MP_TEMP(mp_tmp_w);
868 /* arguments: D1 = Double# */
871 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
873 /* Be prepared to tell Lennart-coded __decodeDouble
874 where mantissa.d can be put (it does not care about the rest) */
875 p = Hp - ARR_SIZE + WDS(1);
876 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
877 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
878 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
880 /* Perform the operation */
881 foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg);
883 /* returns: (Int# (expn), Int#, ByteArray#) */
884 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
887 /* -----------------------------------------------------------------------------
888 * Concurrency primitives
889 * -------------------------------------------------------------------------- */
893 /* args: R1 = closure to spark */
895 MAYBE_GC(R1_PTR, forkzh_fast);
897 // create it right now, return ThreadID in R1
898 "ptr" R1 = foreign "C" createIOThread( RtsFlags_GcFlags_initialStkSize(RtsFlags),
900 foreign "C" scheduleThread(R1 "ptr");
902 // switch at the earliest opportunity
903 CInt[context_switch] = 1 :: CInt;
910 jump stg_yield_noregs;
925 foreign "C" labelThread(R1 "ptr", R2 "ptr");
927 jump %ENTRY_CODE(Sp(0));
930 isCurrentThreadBoundzh_fast
934 r = foreign "C" isThreadBound(CurrentTSO);
939 /* -----------------------------------------------------------------------------
941 * -------------------------------------------------------------------------- */
945 #define IF_NOT_REG_R1(x)
948 #define IF_NOT_REG_R1(x) x
951 // Catch retry frame ------------------------------------------------------------
953 #define CATCH_RETRY_FRAME_ERROR(label) \
954 label { foreign "C" barf("catch_retry_frame incorrectly entered!"); }
956 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_0_ret)
957 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_1_ret)
958 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_2_ret)
959 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_3_ret)
960 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_4_ret)
961 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_5_ret)
962 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_6_ret)
963 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_7_ret)
965 #if MAX_VECTORED_RTN > 8
966 #error MAX_VECTORED_RTN has changed: please modify stg_catch_retry_frame too.
969 #if defined(PROFILING)
970 #define CATCH_RETRY_FRAME_BITMAP 7
971 #define CATCH_RETRY_FRAME_WORDS 6
973 #define CATCH_RETRY_FRAME_BITMAP 1
974 #define CATCH_RETRY_FRAME_WORDS 4
977 INFO_TABLE_RET(stg_catch_retry_frame,
978 CATCH_RETRY_FRAME_WORDS, CATCH_RETRY_FRAME_BITMAP,
980 stg_catch_retry_frame_0_ret,
981 stg_catch_retry_frame_1_ret,
982 stg_catch_retry_frame_2_ret,
983 stg_catch_retry_frame_3_ret,
984 stg_catch_retry_frame_4_ret,
985 stg_catch_retry_frame_5_ret,
986 stg_catch_retry_frame_6_ret,
987 stg_catch_retry_frame_7_ret)
989 W_ r, frame, trec, outer;
990 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
993 trec = StgTSO_trec(CurrentTSO);
994 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr");
995 r = foreign "C" stmCommitNestedTransaction(BaseReg "ptr", trec "ptr");
997 /* Succeeded (either first branch or second branch) */
998 StgTSO_trec(CurrentTSO) = outer;
999 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1000 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1001 jump %ENTRY_CODE(Sp(SP_OFF));
1003 /* Did not commit: retry */
1005 "ptr" new_trec = foreign "C" stmStartTransaction(BaseReg "ptr", outer "ptr");
1006 StgTSO_trec(CurrentTSO) = new_trec;
1007 if (StgCatchRetryFrame_running_alt_code(frame)) {
1008 R1 = StgCatchRetryFrame_alt_code(frame);
1010 R1 = StgCatchRetryFrame_first_code(frame);
1011 StgCatchRetryFrame_first_code_trec(frame) = new_trec;
1014 jump RET_LBL(stg_ap_v);
1019 // Atomically frame -------------------------------------------------------------
1022 #define ATOMICALLY_FRAME_ERROR(label) \
1023 label { foreign "C" barf("atomically_frame incorrectly entered!"); }
1025 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_0_ret)
1026 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_1_ret)
1027 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_2_ret)
1028 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_3_ret)
1029 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_4_ret)
1030 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_5_ret)
1031 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_6_ret)
1032 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_7_ret)
1034 #if MAX_VECTORED_RTN > 8
1035 #error MAX_VECTORED_RTN has changed: please modify stg_atomically_frame too.
1038 #if defined(PROFILING)
1039 #define ATOMICALLY_FRAME_BITMAP 7
1040 #define ATOMICALLY_FRAME_WORDS 4
1042 #define ATOMICALLY_FRAME_BITMAP 1
1043 #define ATOMICALLY_FRAME_WORDS 2
1047 INFO_TABLE_RET(stg_atomically_frame,
1048 ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP,
1050 stg_atomically_frame_0_ret,
1051 stg_atomically_frame_1_ret,
1052 stg_atomically_frame_2_ret,
1053 stg_atomically_frame_3_ret,
1054 stg_atomically_frame_4_ret,
1055 stg_atomically_frame_5_ret,
1056 stg_atomically_frame_6_ret,
1057 stg_atomically_frame_7_ret)
1059 W_ frame, trec, valid;
1060 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1063 trec = StgTSO_trec(CurrentTSO);
1064 if (StgAtomicallyFrame_waiting(frame)) {
1065 /* The TSO is currently waiting: should we stop waiting? */
1066 valid = foreign "C" stmReWait(CurrentTSO "ptr");
1068 /* Previous attempt is still valid: no point trying again yet */
1069 IF_NOT_REG_R1(Sp_adj(-2);
1070 Sp(1) = stg_NO_FINALIZER_closure;
1071 Sp(0) = stg_ut_1_0_unreg_info;)
1072 jump stg_block_noregs;
1074 /* Previous attempt is no longer valid: try again */
1075 "ptr" trec = foreign "C" stmStartTransaction(BaseReg "ptr", NO_TREC "ptr");
1076 StgTSO_trec(CurrentTSO) = trec;
1077 StgAtomicallyFrame_waiting(frame) = 0 :: CInt; /* false; */
1078 R1 = StgAtomicallyFrame_code(frame);
1080 jump RET_LBL(stg_ap_v);
1083 /* The TSO is not currently waiting: try to commit the transaction */
1084 valid = foreign "C" stmCommitTransaction(BaseReg "ptr", trec "ptr");
1086 /* Transaction was valid: commit succeeded */
1087 StgTSO_trec(CurrentTSO) = NO_TREC;
1088 Sp = Sp + SIZEOF_StgAtomicallyFrame;
1089 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1090 jump %ENTRY_CODE(Sp(SP_OFF));
1092 /* Transaction was not valid: try again */
1093 "ptr" trec = foreign "C" stmStartTransaction(BaseReg "ptr", NO_TREC "ptr");
1094 StgTSO_trec(CurrentTSO) = trec;
1095 R1 = StgAtomicallyFrame_code(frame);
1097 jump RET_LBL(stg_ap_v);
1103 // STM catch frame --------------------------------------------------------------
1105 #define CATCH_STM_FRAME_ENTRY_TEMPLATE(label,ret) \
1108 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \
1109 Sp = Sp + SIZEOF_StgCatchSTMFrame; \
1110 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \
1120 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_0_ret,%RET_VEC(Sp(SP_OFF),0))
1121 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_1_ret,%RET_VEC(Sp(SP_OFF),1))
1122 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_2_ret,%RET_VEC(Sp(SP_OFF),2))
1123 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_3_ret,%RET_VEC(Sp(SP_OFF),3))
1124 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_4_ret,%RET_VEC(Sp(SP_OFF),4))
1125 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_5_ret,%RET_VEC(Sp(SP_OFF),5))
1126 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_6_ret,%RET_VEC(Sp(SP_OFF),6))
1127 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_7_ret,%RET_VEC(Sp(SP_OFF),7))
1129 #if MAX_VECTORED_RTN > 8
1130 #error MAX_VECTORED_RTN has changed: please modify stg_catch_stm_frame too.
1133 #if defined(PROFILING)
1134 #define CATCH_STM_FRAME_BITMAP 3
1135 #define CATCH_STM_FRAME_WORDS 3
1137 #define CATCH_STM_FRAME_BITMAP 0
1138 #define CATCH_STM_FRAME_WORDS 1
1141 /* Catch frames are very similar to update frames, but when entering
1142 * one we just pop the frame off the stack and perform the correct
1143 * kind of return to the activation record underneath us on the stack.
1146 INFO_TABLE_RET(stg_catch_stm_frame,
1147 CATCH_STM_FRAME_WORDS, CATCH_STM_FRAME_BITMAP,
1149 stg_catch_stm_frame_0_ret,
1150 stg_catch_stm_frame_1_ret,
1151 stg_catch_stm_frame_2_ret,
1152 stg_catch_stm_frame_3_ret,
1153 stg_catch_stm_frame_4_ret,
1154 stg_catch_stm_frame_5_ret,
1155 stg_catch_stm_frame_6_ret,
1156 stg_catch_stm_frame_7_ret)
1157 CATCH_STM_FRAME_ENTRY_TEMPLATE(,%ENTRY_CODE(Sp(SP_OFF)))
1160 // Primop definition ------------------------------------------------------------
1168 // stmStartTransaction may allocate
1169 MAYBE_GC (R1_PTR, atomicallyzh_fast);
1171 /* Args: R1 = m :: STM a */
1172 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);
1174 /* Set up the atomically frame */
1175 Sp = Sp - SIZEOF_StgAtomicallyFrame;
1178 SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
1179 StgAtomicallyFrame_waiting(frame) = 0 :: CInt; // False
1180 StgAtomicallyFrame_code(frame) = R1;
1182 /* Start the memory transcation */
1183 old_trec = StgTSO_trec(CurrentTSO);
1184 "ptr" new_trec = foreign "C" stmStartTransaction(BaseReg "ptr", old_trec "ptr");
1185 StgTSO_trec(CurrentTSO) = new_trec;
1187 /* Apply R1 to the realworld token */
1189 jump RET_LBL(stg_ap_v);
1197 /* Args: R1 :: STM a */
1198 /* Args: R2 :: Exception -> STM a */
1199 STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast);
1201 /* Set up the catch frame */
1202 Sp = Sp - SIZEOF_StgCatchSTMFrame;
1205 SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
1206 StgCatchSTMFrame_handler(frame) = R2;
1208 /* Apply R1 to the realworld token */
1210 jump RET_LBL(stg_ap_v);
1220 // stmStartTransaction may allocate
1221 MAYBE_GC (R1_PTR & R2_PTR, catchRetryzh_fast);
1223 /* Args: R1 :: STM a */
1224 /* Args: R2 :: STM a */
1225 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);
1227 /* Start a nested transaction within which to run the first code */
1228 trec = StgTSO_trec(CurrentTSO);
1229 "ptr" new_trec = foreign "C" stmStartTransaction(BaseReg "ptr", trec "ptr");
1230 StgTSO_trec(CurrentTSO) = new_trec;
1232 /* Set up the catch-retry frame */
1233 Sp = Sp - SIZEOF_StgCatchRetryFrame;
1236 SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
1237 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1238 StgCatchRetryFrame_first_code(frame) = R1;
1239 StgCatchRetryFrame_alt_code(frame) = R2;
1240 StgCatchRetryFrame_first_code_trec(frame) = new_trec;
1242 /* Apply R1 to the realworld token */
1244 jump RET_LBL(stg_ap_v);
1256 MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate
1258 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
1260 trec = StgTSO_trec(CurrentTSO);
1261 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr");
1262 StgTSO_sp(CurrentTSO) = Sp;
1263 frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr");
1264 Sp = StgTSO_sp(CurrentTSO);
1267 if (frame_type == CATCH_RETRY_FRAME) {
1268 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
1269 ASSERT(outer != NO_TREC);
1270 if (!StgCatchRetryFrame_running_alt_code(frame)) {
1271 // Retry in the first code: try the alternative
1272 "ptr" trec = foreign "C" stmStartTransaction(BaseReg "ptr", outer "ptr");
1273 StgTSO_trec(CurrentTSO) = trec;
1274 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
1275 R1 = StgCatchRetryFrame_alt_code(frame);
1277 jump RET_LBL(stg_ap_v);
1279 // Retry in the alternative code: propagate
1281 other_trec = StgCatchRetryFrame_first_code_trec(frame);
1282 r = foreign "C" stmCommitNestedTransaction(BaseReg "ptr", other_trec "ptr");
1284 r = foreign "C" stmCommitNestedTransaction(BaseReg "ptr", trec "ptr");
1287 // Merge between siblings succeeded: commit it back to enclosing transaction
1288 // and then propagate the retry
1289 StgTSO_trec(CurrentTSO) = outer;
1290 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1291 goto retry_pop_stack;
1293 // Merge failed: we musn't propagate the retry. Try both paths again.
1294 "ptr" trec = foreign "C" stmStartTransaction(BaseReg "ptr", outer "ptr");
1295 StgCatchRetryFrame_first_code_trec(frame) = trec;
1296 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1297 StgTSO_trec(CurrentTSO) = trec;
1298 R1 = StgCatchRetryFrame_first_code(frame);
1300 jump RET_LBL(stg_ap_v);
1305 // We've reached the ATOMICALLY_FRAME: attempt to wait
1306 ASSERT(frame_type == ATOMICALLY_FRAME);
1307 ASSERT(outer == NO_TREC);
1308 r = foreign "C" stmWait(BaseReg "ptr", CurrentTSO "ptr", trec "ptr");
1310 // Transaction was valid: stmWait put us on the TVars' queues, we now block
1311 StgAtomicallyFrame_waiting(frame) = 1 :: CInt; // true
1313 // Fix up the stack in the unregisterised case: the return convention is different.
1314 IF_NOT_REG_R1(Sp_adj(-2);
1315 Sp(1) = stg_NO_FINALIZER_closure;
1316 Sp(0) = stg_ut_1_0_unreg_info;)
1317 jump stg_block_noregs;
1319 // Transaction was not valid: retry immediately
1320 "ptr" trec = foreign "C" stmStartTransaction(BaseReg "ptr", outer "ptr");
1321 StgTSO_trec(CurrentTSO) = trec;
1322 R1 = StgAtomicallyFrame_code(frame);
1325 jump RET_LBL(stg_ap_v);
1335 /* Args: R1 = initialisation value */
1337 ALLOC_PRIM( SIZEOF_StgTVar, R1_PTR, newTVarzh_fast);
1338 tv = Hp - SIZEOF_StgTVar + WDS(1);
1339 SET_HDR(tv,stg_TVAR_info,W_[CCCS]);
1340 StgTVar_current_value(tv) = R1;
1341 StgTVar_first_wait_queue_entry(tv) = stg_END_STM_WAIT_QUEUE_closure;
1343 trec = StgTSO_trec(CurrentTSO);
1344 StgTVar_last_update_by(tv) = trec;
1346 StgTVar_last_update_by(tv) = NO_TREC;
1359 /* Args: R1 = TVar closure */
1361 MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
1362 trec = StgTSO_trec(CurrentTSO);
1364 "ptr" result = foreign "C" stmReadTVar(BaseReg "ptr", trec "ptr", tvar "ptr");
1376 /* Args: R1 = TVar closure */
1377 /* R2 = New value */
1379 MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
1380 trec = StgTSO_trec(CurrentTSO);
1383 foreign "C" stmWriteTVar(BaseReg "ptr", trec "ptr", tvar "ptr", new_value "ptr");
1385 jump %ENTRY_CODE(Sp(0));
1389 /* -----------------------------------------------------------------------------
1392 * take & putMVar work as follows. Firstly, an important invariant:
1394 * If the MVar is full, then the blocking queue contains only
1395 * threads blocked on putMVar, and if the MVar is empty then the
1396 * blocking queue contains only threads blocked on takeMVar.
1399 * MVar empty : then add ourselves to the blocking queue
1400 * MVar full : remove the value from the MVar, and
1401 * blocking queue empty : return
1402 * blocking queue non-empty : perform the first blocked putMVar
1403 * from the queue, and wake up the
1404 * thread (MVar is now full again)
1406 * putMVar is just the dual of the above algorithm.
1408 * How do we "perform a putMVar"? Well, we have to fiddle around with
1409 * the stack of the thread waiting to do the putMVar. See
1410 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1411 * the stack layout, and the PerformPut and PerformTake macros below.
1413 * It is important that a blocked take or put is woken up with the
1414 * take/put already performed, because otherwise there would be a
1415 * small window of vulnerability where the thread could receive an
1416 * exception and never perform its take or put, and we'd end up with a
1419 * -------------------------------------------------------------------------- */
1423 /* args: R1 = MVar closure */
1425 if (GET_INFO(R1) == stg_EMPTY_MVAR_info) {
1437 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
1439 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1440 SET_HDR(mvar,stg_EMPTY_MVAR_info,W_[CCCS]);
1441 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1442 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1443 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1448 /* If R1 isn't available, pass it on the stack */
1450 #define PerformTake(tso, value) \
1451 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1452 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1454 #define PerformTake(tso, value) \
1455 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1456 W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info;
1459 #define PerformPut(tso,lval) \
1460 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1461 lval = W_[StgTSO_sp(tso) - WDS(1)];
1465 W_ mvar, val, info, tso;
1467 /* args: R1 = MVar closure */
1471 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1473 info = GET_INFO(mvar);
1476 /* If the MVar is empty, put ourselves on its blocking queue,
1477 * and wait until we're woken up.
1479 if (info == stg_EMPTY_MVAR_info) {
1480 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1481 StgMVar_head(mvar) = CurrentTSO;
1483 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1485 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1486 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1487 StgTSO_block_info(CurrentTSO) = mvar;
1488 StgMVar_tail(mvar) = CurrentTSO;
1490 jump stg_block_takemvar;
1493 /* we got the value... */
1494 val = StgMVar_value(mvar);
1496 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1498 /* There are putMVar(s) waiting...
1499 * wake up the first thread on the queue
1501 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1503 /* actually perform the putMVar for the thread that we just woke up */
1504 tso = StgMVar_head(mvar);
1505 PerformPut(tso,StgMVar_value(mvar));
1507 #if defined(GRAN) || defined(PAR)
1508 /* ToDo: check 2nd arg (mvar) is right */
1509 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar);
1510 StgMVar_head(mvar) = tso;
1512 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1513 StgMVar_head(mvar) = tso;
1516 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1517 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1521 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1527 /* No further putMVars, MVar is now empty */
1528 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1531 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1533 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1543 W_ mvar, val, info, tso;
1545 /* args: R1 = MVar closure */
1550 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1552 info = GET_INFO(mvar);
1555 if (info == stg_EMPTY_MVAR_info) {
1557 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1559 /* HACK: we need a pointer to pass back,
1560 * so we abuse NO_FINALIZER_closure
1562 RET_NP(0, stg_NO_FINALIZER_closure);
1565 /* we got the value... */
1566 val = StgMVar_value(mvar);
1568 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1570 /* There are putMVar(s) waiting...
1571 * wake up the first thread on the queue
1573 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1575 /* actually perform the putMVar for the thread that we just woke up */
1576 tso = StgMVar_head(mvar);
1577 PerformPut(tso,StgMVar_value(mvar));
1579 #if defined(GRAN) || defined(PAR)
1580 /* ToDo: check 2nd arg (mvar) is right */
1581 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr");
1582 StgMVar_head(mvar) = tso;
1584 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1585 StgMVar_head(mvar) = tso;
1588 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1589 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1592 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1597 /* No further putMVars, MVar is now empty */
1598 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1600 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1602 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1614 /* args: R1 = MVar, R2 = value */
1618 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1620 info = GET_INFO(mvar);
1623 if (info == stg_FULL_MVAR_info) {
1624 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1625 StgMVar_head(mvar) = CurrentTSO;
1627 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1629 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1630 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1631 StgTSO_block_info(CurrentTSO) = mvar;
1632 StgMVar_tail(mvar) = CurrentTSO;
1634 jump stg_block_putmvar;
1637 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1639 /* There are takeMVar(s) waiting: wake up the first one
1641 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1643 /* actually perform the takeMVar */
1644 tso = StgMVar_head(mvar);
1645 PerformTake(tso, R2);
1647 #if defined(GRAN) || defined(PAR)
1648 /* ToDo: check 2nd arg (mvar) is right */
1649 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1650 StgMVar_head(mvar) = tso;
1652 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1653 StgMVar_head(mvar) = tso;
1656 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1657 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1661 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1663 jump %ENTRY_CODE(Sp(0));
1667 /* No further takes, the MVar is now full. */
1668 StgMVar_value(mvar) = R2;
1671 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1673 SET_INFO(mvar,stg_FULL_MVAR_info);
1675 jump %ENTRY_CODE(Sp(0));
1678 /* ToDo: yield afterward for better communication performance? */
1686 /* args: R1 = MVar, R2 = value */
1690 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1692 info = GET_INFO(mvar);
1695 if (info == stg_FULL_MVAR_info) {
1697 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1702 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1704 /* There are takeMVar(s) waiting: wake up the first one
1706 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1708 /* actually perform the takeMVar */
1709 tso = StgMVar_head(mvar);
1710 PerformTake(tso, R2);
1712 #if defined(GRAN) || defined(PAR)
1713 /* ToDo: check 2nd arg (mvar) is right */
1714 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1715 StgMVar_head(mvar) = tso;
1717 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1718 StgMVar_head(mvar) = tso;
1721 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1722 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1726 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1728 jump %ENTRY_CODE(Sp(0));
1732 /* No further takes, the MVar is now full. */
1733 StgMVar_value(mvar) = R2;
1736 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1738 SET_INFO(mvar,stg_FULL_MVAR_info);
1740 jump %ENTRY_CODE(Sp(0));
1743 /* ToDo: yield afterward for better communication performance? */
1747 /* -----------------------------------------------------------------------------
1748 Stable pointer primitives
1749 ------------------------------------------------------------------------- */
1751 makeStableNamezh_fast
1755 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1757 index = foreign "C" lookupStableName(R1 "ptr");
1759 /* Is there already a StableName for this heap object?
1760 * stable_ptr_table is a pointer to an array of snEntry structs.
1762 if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) {
1763 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1764 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1765 StgStableName_sn(sn_obj) = index;
1766 snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj;
1768 sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry);
1775 makeStablePtrzh_fast
1779 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1780 "ptr" sp = foreign "C" getStablePtr(R1 "ptr");
1784 deRefStablePtrzh_fast
1786 /* Args: R1 = the stable ptr */
1789 r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry);
1793 /* -----------------------------------------------------------------------------
1794 Bytecode object primitives
1795 ------------------------------------------------------------------------- */
1806 W_ bco, bitmap_arr, bytes, words;
1809 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1812 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R4_PTR&R6_PTR, newBCOzh_fast );
1814 bco = Hp - bytes + WDS(1);
1815 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1817 StgBCO_instrs(bco) = R1;
1818 StgBCO_literals(bco) = R2;
1819 StgBCO_ptrs(bco) = R3;
1820 StgBCO_itbls(bco) = R4;
1821 StgBCO_arity(bco) = HALF_W_(R5);
1822 StgBCO_size(bco) = HALF_W_(words);
1824 // Copy the arity/bitmap info into the BCO
1828 if (i < StgArrWords_words(bitmap_arr)) {
1829 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1840 // R1 = the BCO# for the AP
1844 // This function is *only* used to wrap zero-arity BCOs in an
1845 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1846 // saturated and always points directly to a FUN or BCO.
1847 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
1848 StgBCO_arity(R1) == HALF_W_(0));
1850 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1851 TICK_ALLOC_UP_THK(0, 0);
1852 CCCS_ALLOC(SIZEOF_StgAP);
1854 ap = Hp - SIZEOF_StgAP + WDS(1);
1855 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1857 StgAP_n_args(ap) = HALF_W_(0);
1863 /* -----------------------------------------------------------------------------
1864 Thread I/O blocking primitives
1865 -------------------------------------------------------------------------- */
1867 /* Add a thread to the end of the blocked queue. (C-- version of the C
1868 * macro in Schedule.h).
1870 #define APPEND_TO_BLOCKED_QUEUE(tso) \
1871 ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \
1872 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
1873 W_[blocked_queue_hd] = tso; \
1875 StgTSO_link(W_[blocked_queue_tl]) = tso; \
1877 W_[blocked_queue_tl] = tso;
1883 foreign "C" barf("waitRead# on threaded RTS");
1886 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1887 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1888 StgTSO_block_info(CurrentTSO) = R1;
1889 // No locking - we're not going to use this interface in the
1890 // threaded RTS anyway.
1891 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1892 jump stg_block_noregs;
1899 foreign "C" barf("waitWrite# on threaded RTS");
1902 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1903 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1904 StgTSO_block_info(CurrentTSO) = R1;
1905 // No locking - we're not going to use this interface in the
1906 // threaded RTS anyway.
1907 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1908 jump stg_block_noregs;
1912 STRING(stg_delayzh_malloc_str, "delayzh_fast")
1915 #ifdef mingw32_HOST_OS
1923 foreign "C" barf("delay# on threaded RTS");
1926 /* args: R1 (microsecond delay amount) */
1927 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1928 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
1930 #ifdef mingw32_HOST_OS
1932 /* could probably allocate this on the heap instead */
1933 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1934 stg_delayzh_malloc_str);
1935 reqID = foreign "C" addDelayRequest(R1);
1936 StgAsyncIOResult_reqID(ares) = reqID;
1937 StgAsyncIOResult_len(ares) = 0;
1938 StgAsyncIOResult_errCode(ares) = 0;
1939 StgTSO_block_info(CurrentTSO) = ares;
1941 /* Having all async-blocked threads reside on the blocked_queue
1942 * simplifies matters, so change the status to OnDoProc put the
1943 * delayed thread on the blocked_queue.
1945 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1946 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1947 jump stg_block_async_void;
1952 time = foreign "C" getourtimeofday();
1953 target = (R1 / (TICK_MILLISECS*1000)) + time;
1954 StgTSO_block_info(CurrentTSO) = target;
1956 /* Insert the new thread in the sleeping queue. */
1958 t = W_[sleeping_queue];
1960 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
1966 StgTSO_link(CurrentTSO) = t;
1968 W_[sleeping_queue] = CurrentTSO;
1970 StgTSO_link(prev) = CurrentTSO;
1972 jump stg_block_noregs;
1977 #ifdef mingw32_HOST_OS
1978 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
1985 foreign "C" barf("asyncRead# on threaded RTS");
1988 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1989 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1990 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1992 /* could probably allocate this on the heap instead */
1993 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1994 stg_asyncReadzh_malloc_str);
1995 reqID = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr");
1996 StgAsyncIOResult_reqID(ares) = reqID;
1997 StgAsyncIOResult_len(ares) = 0;
1998 StgAsyncIOResult_errCode(ares) = 0;
1999 StgTSO_block_info(CurrentTSO) = ares;
2000 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2001 jump stg_block_async;
2004 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
2011 foreign "C" barf("asyncWrite# on threaded RTS");
2014 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2015 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2016 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2018 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2019 stg_asyncWritezh_malloc_str);
2020 reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr");
2022 StgAsyncIOResult_reqID(ares) = reqID;
2023 StgAsyncIOResult_len(ares) = 0;
2024 StgAsyncIOResult_errCode(ares) = 0;
2025 StgTSO_block_info(CurrentTSO) = ares;
2026 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2027 jump stg_block_async;
2030 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
2036 /* args: R1 = proc, R2 = param */
2037 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2038 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2040 /* could probably allocate this on the heap instead */
2041 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2042 stg_asyncDoProczh_malloc_str);
2043 reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr");
2044 StgAsyncIOResult_reqID(ares) = reqID;
2045 StgAsyncIOResult_len(ares) = 0;
2046 StgAsyncIOResult_errCode(ares) = 0;
2047 StgTSO_block_info(CurrentTSO) = ares;
2048 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2049 jump stg_block_async;
2053 /* -----------------------------------------------------------------------------
2056 classes CCallable and CReturnable don't really exist, but the
2057 compiler insists on generating dictionaries containing references
2058 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
2059 for these. Some C compilers can't cope with zero-length static arrays,
2060 so we have to make these one element long.
2061 --------------------------------------------------------------------------- */
2064 GHC_ZCCCallable_static_info: W_ 0;
2068 GHC_ZCCReturnable_static_info: W_ 0;