1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2004
5 * Out-of-line primitive operations
7 * This file contains the implementations of all the primitive
8 * operations ("primops") which are not expanded inline. See
9 * ghc/compiler/prelude/primops.txt.pp for a list of all the primops;
10 * this file contains code for most of those with the attribute
13 * Entry convention: the entry convention for a primop is that all the
14 * args are in Stg registers (R1, R2, etc.). This is to make writing
15 * the primops easier. (see compiler/codeGen/CgCallConv.hs).
17 * Return convention: results from a primop are generally returned
18 * using the ordinary unboxed tuple return convention. The C-- parser
19 * implements the RET_xxxx() macros to perform unboxed-tuple returns
20 * based on the prevailing return convention.
22 * This file is written in a subset of C--, extended with various
23 * features specific to GHC. It is compiled by GHC directly. For the
24 * syntax of .cmm files, see the parser in ghc/compiler/cmm/CmmParse.y.
26 * ---------------------------------------------------------------------------*/
30 /*-----------------------------------------------------------------------------
33 Basically just new*Array - the others are all inline macros.
35 The size arg is always passed in R1, and the result returned in R1.
37 The slow entry point is for returning from a heap check, the saved
38 size argument must be re-loaded from the stack.
39 -------------------------------------------------------------------------- */
41 /* for objects that are *less* than the size of a word, make sure we
42 * round up to the nearest word for the size of the array.
47 W_ words, payload_words, n, p;
48 MAYBE_GC(NO_PTRS,newByteArrayzh_fast);
50 payload_words = ROUNDUP_BYTES_TO_WDS(n);
51 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
52 "ptr" p = foreign "C" allocate(words);
53 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
54 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
55 StgArrWords_words(p) = payload_words;
59 newPinnedByteArrayzh_fast
61 W_ words, payload_words, n, p;
63 MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast);
65 payload_words = ROUNDUP_BYTES_TO_WDS(n);
67 // We want an 8-byte aligned array. allocatePinned() gives us
68 // 8-byte aligned memory by default, but we want to align the
69 // *goods* inside the ArrWords object, so we have to check the
70 // size of the ArrWords header and adjust our size accordingly.
71 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
72 if ((SIZEOF_StgArrWords & 7) != 0) {
76 "ptr" p = foreign "C" allocatePinned(words);
77 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
79 // Again, if the ArrWords header isn't a multiple of 8 bytes, we
80 // have to push the object forward one word so that the goods
81 // fall on an 8-byte boundary.
82 if ((SIZEOF_StgArrWords & 7) != 0) {
86 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
87 StgArrWords_words(p) = payload_words;
93 W_ words, n, init, arr, p;
94 /* Args: R1 = words, R2 = initialisation value */
97 MAYBE_GC(R2_PTR,newArrayzh_fast);
99 words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
100 "ptr" arr = foreign "C" allocate(words);
101 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
103 SET_HDR(arr, stg_MUT_ARR_PTRS_info, W_[CCCS]);
104 StgMutArrPtrs_ptrs(arr) = n;
106 // Initialise all elements of the the array with the value in R2
108 p = arr + SIZEOF_StgMutArrPtrs;
110 if (p < arr + WDS(words)) {
119 unsafeThawArrayzh_fast
121 SET_INFO(R1,stg_MUT_ARR_PTRS_info);
123 // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
125 // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
126 // normally doesn't. However, when we freeze a MUT_ARR_PTRS, we leave
127 // it on the mutable list for the GC to remove (removing something from
128 // the mutable list is not easy, because the mut_list is only singly-linked).
130 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
131 // either it is on a mut_list, or it isn't. We adopt the convention that
132 // the mut_link field is NULL if it isn't on a mut_list, and the GC
133 // maintains this invariant.
135 if (StgMutClosure_mut_link(R1) == NULL) {
136 foreign "C" recordMutable(R1 "ptr");
142 /* -----------------------------------------------------------------------------
144 -------------------------------------------------------------------------- */
149 /* Args: R1 = initialisation value */
151 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);
153 mv = Hp - SIZEOF_StgMutVar + WDS(1);
154 SET_HDR(mv,stg_MUT_VAR_info,W_[CCCS]);
155 StgMutVar_var(mv) = R1;
160 atomicModifyMutVarzh_fast
163 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
165 /* If x is the current contents of the MutVar#, then
166 We want to make the new contents point to
170 and the return value is
174 obviously we can share (f x).
176 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
177 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
178 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
182 #define THUNK_1_SIZE (SIZEOF_StgHeader + WDS(MIN_UPD_SIZE))
183 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
185 #define THUNK_1_SIZE (SIZEOF_StgHeader + WDS(1))
186 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
190 #define THUNK_2_SIZE (SIZEOF_StgHeader + WDS(MIN_UPD_SIZE))
191 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
193 #define THUNK_2_SIZE (SIZEOF_StgHeader + WDS(2))
194 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
197 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
199 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);
201 x = StgMutVar_var(R1);
203 TICK_ALLOC_THUNK_2();
204 CCCS_ALLOC(THUNK_2_SIZE);
205 z = Hp - THUNK_2_SIZE + WDS(1);
206 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
207 LDV_RECORD_CREATE(z);
208 StgClosure_payload(z,0) = R2;
209 StgClosure_payload(z,1) = x;
211 TICK_ALLOC_THUNK_1();
212 CCCS_ALLOC(THUNK_1_SIZE);
213 y = z - THUNK_1_SIZE;
214 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
215 LDV_RECORD_CREATE(y);
216 StgClosure_payload(y,0) = z;
218 StgMutVar_var(R1) = y;
220 TICK_ALLOC_THUNK_1();
221 CCCS_ALLOC(THUNK_1_SIZE);
222 r = y - THUNK_1_SIZE;
223 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
224 LDV_RECORD_CREATE(r);
225 StgClosure_payload(r,0) = z;
230 /* -----------------------------------------------------------------------------
231 Foreign Object Primitives
232 -------------------------------------------------------------------------- */
236 /* R1 = ptr to foreign object,
240 ALLOC_PRIM( SIZEOF_StgForeignObj, NO_PTRS, mkForeignObjzh_fast);
242 result = Hp - SIZEOF_StgForeignObj + WDS(1);
243 SET_HDR(result,stg_FOREIGN_info,W_[CCCS]);
244 StgForeignObj_data(result) = R1;
246 /* returns (# s#, ForeignObj# #) */
250 /* -----------------------------------------------------------------------------
251 Weak Pointer Primitives
252 -------------------------------------------------------------------------- */
254 STRING(stg_weak_msg,"New weak pointer at %p\n")
260 R3 = finalizer (or NULL)
265 R3 = stg_NO_FINALIZER_closure;
268 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
270 w = Hp - SIZEOF_StgWeak + WDS(1);
271 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
274 StgWeak_value(w) = R2;
275 StgWeak_finalizer(w) = R3;
277 StgWeak_link(w) = W_[weak_ptr_list];
278 W_[weak_ptr_list] = w;
280 IF_DEBUG(weak, foreign "C" fprintf(stderr,stg_weak_msg,w));
295 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
296 RET_NP(0,stg_NO_FINALIZER_closure);
302 // A weak pointer is inherently used, so we do not need to call
303 // LDV_recordDead_FILL_SLOP_DYNAMIC():
304 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
305 // or, LDV_recordDead():
306 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
307 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
308 // large as weak pointers, so there is no need to fill the slop, either.
309 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
313 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
315 SET_INFO(w,stg_DEAD_WEAK_info);
316 LDV_RECORD_CREATE(w);
318 f = StgWeak_finalizer(w);
319 StgDeadWeak_link(w) = StgWeak_link(w);
321 /* return the finalizer */
322 if (f == stg_NO_FINALIZER_closure) {
323 RET_NP(0,stg_NO_FINALIZER_closure);
335 if (GET_INFO(w) == stg_WEAK_info) {
337 val = StgWeak_value(w);
345 /* -----------------------------------------------------------------------------
346 Arbitrary-precision Integer operations.
348 There are some assumptions in this code that mp_limb_t == W_. This is
349 the case for all the platforms that GHC supports, currently.
350 -------------------------------------------------------------------------- */
354 /* arguments: R1 = Int# */
356 W_ val, s, p; /* to avoid aliasing */
359 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
361 p = Hp - SIZEOF_StgArrWords;
362 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
363 StgArrWords_words(p) = 1;
365 /* mpz_set_si is inlined here, makes things simpler */
378 /* returns (# size :: Int#,
387 /* arguments: R1 = Word# */
389 W_ val, s, p; /* to avoid aliasing */
393 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
395 p = Hp - SIZEOF_StgArrWords;
396 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
397 StgArrWords_words(p) = 1;
406 /* returns (# size :: Int#,
407 data :: ByteArray# #)
414 * 'long long' primops for converting to/from Integers.
417 #ifdef SUPPORT_LONG_LONGS
419 int64ToIntegerzh_fast
421 /* arguments: L1 = Int64# */
424 W_ hi, s, neg, words_needed, p;
429 if ( %ge(val,0x100000000::L_) || %le(val,-0x100000000::L_) ) {
432 // minimum is one word
436 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
437 NO_PTRS, int64ToIntegerzh_fast );
439 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
440 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
441 StgArrWords_words(p) = words_needed;
443 if ( %lt(val,0::L_) ) {
448 hi = TO_W_(val >> 32);
450 if ( words_needed == 2 ) {
455 if ( val != 0::L_ ) {
458 } else /* val==0 */ {
466 /* returns (# size :: Int#,
467 data :: ByteArray# #)
472 word64ToIntegerzh_fast
474 /* arguments: L1 = Word64# */
477 W_ hi, s, words_needed, p;
480 if ( val >= 0x100000000::L_ ) {
486 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
487 NO_PTRS, word64ToIntegerzh_fast );
489 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
490 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
491 StgArrWords_words(p) = words_needed;
493 hi = TO_W_(val >> 32);
494 if ( val >= 0x100000000::L_ ) {
499 if ( val != 0::L_ ) {
502 } else /* val==0 */ {
507 /* returns (# size :: Int#,
508 data :: ByteArray# #)
514 #endif /* SUPPORT_LONG_LONGS */
516 /* ToDo: this is shockingly inefficient */
520 bits8 [SIZEOF_MP_INT];
525 bits8 [SIZEOF_MP_INT];
530 bits8 [SIZEOF_MP_INT];
535 bits8 [SIZEOF_MP_INT];
538 #define GMP_TAKE2_RET1(name,mp_fun) \
544 /* call doYouWantToGC() */ \
545 MAYBE_GC(R2_PTR & R4_PTR, name); \
552 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
553 MP_INT__mp_size(mp_tmp1) = (s1); \
554 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
555 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
556 MP_INT__mp_size(mp_tmp2) = (s2); \
557 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
559 foreign "C" mpz_init(result1); \
561 /* Perform the operation */ \
562 foreign "C" mp_fun(result1,mp_tmp1,mp_tmp2); \
564 RET_NP(TO_W_(MP_INT__mp_size(result1)), \
565 MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \
568 #define GMP_TAKE1_RET1(name,mp_fun) \
574 /* call doYouWantToGC() */ \
575 MAYBE_GC(R2_PTR, name); \
580 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
581 MP_INT__mp_size(mp_tmp1) = (s1); \
582 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
584 foreign "C" mpz_init(result1); \
586 /* Perform the operation */ \
587 foreign "C" mp_fun(result1,mp_tmp1); \
589 RET_NP(TO_W_(MP_INT__mp_size(result1)), \
590 MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \
593 #define GMP_TAKE2_RET2(name,mp_fun) \
599 /* call doYouWantToGC() */ \
600 MAYBE_GC(R2_PTR & R4_PTR, name); \
607 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
608 MP_INT__mp_size(mp_tmp1) = (s1); \
609 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
610 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
611 MP_INT__mp_size(mp_tmp2) = (s2); \
612 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
614 foreign "C" mpz_init(result1); \
615 foreign "C" mpz_init(result2); \
617 /* Perform the operation */ \
618 foreign "C" mp_fun(result1,result2,mp_tmp1,mp_tmp2); \
620 RET_NPNP(TO_W_(MP_INT__mp_size(result1)), \
621 MP_INT__mp_d(result1) - SIZEOF_StgArrWords, \
622 TO_W_(MP_INT__mp_size(result2)), \
623 MP_INT__mp_d(result2) - SIZEOF_StgArrWords); \
626 GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add)
627 GMP_TAKE2_RET1(minusIntegerzh_fast, mpz_sub)
628 GMP_TAKE2_RET1(timesIntegerzh_fast, mpz_mul)
629 GMP_TAKE2_RET1(gcdIntegerzh_fast, mpz_gcd)
630 GMP_TAKE2_RET1(quotIntegerzh_fast, mpz_tdiv_q)
631 GMP_TAKE2_RET1(remIntegerzh_fast, mpz_tdiv_r)
632 GMP_TAKE2_RET1(divExactIntegerzh_fast, mpz_divexact)
633 GMP_TAKE2_RET1(andIntegerzh_fast, mpz_and)
634 GMP_TAKE2_RET1(orIntegerzh_fast, mpz_ior)
635 GMP_TAKE2_RET1(xorIntegerzh_fast, mpz_xor)
636 GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com)
638 GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr)
639 GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr)
642 aa: W_; // NB. aa is really an mp_limb_t
647 /* R1 = the first Int#; R2 = the second Int# */
651 r = foreign "C" mpn_gcd_1(aa, 1, R2);
654 /* Result parked in R1, return via info-pointer at TOS */
655 jump %ENTRY_CODE(Sp(0));
661 /* R1 = s1; R2 = d1; R3 = the int */
662 R1 = foreign "C" mpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3);
664 /* Result parked in R1, return via info-pointer at TOS */
665 jump %ENTRY_CODE(Sp(0));
671 /* R1 = s1; R2 = d1; R3 = the int */
672 W_ usize, vsize, v_digit, u_digit;
678 // paraphrased from mpz_cmp_si() in the GMP sources
679 if (%gt(v_digit,0)) {
682 if (%lt(v_digit,0)) {
688 if (usize != vsize) {
690 jump %ENTRY_CODE(Sp(0));
695 jump %ENTRY_CODE(Sp(0));
698 u_digit = W_[BYTE_ARR_CTS(R2)];
700 if (u_digit == v_digit) {
702 jump %ENTRY_CODE(Sp(0));
705 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
711 jump %ENTRY_CODE(Sp(0));
716 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
717 W_ usize, vsize, size, up, vp;
720 // paraphrased from mpz_cmp() in the GMP sources
724 if (usize != vsize) {
726 jump %ENTRY_CODE(Sp(0));
731 jump %ENTRY_CODE(Sp(0));
734 if (%lt(usize,0)) { // NB. not <, which is unsigned
740 up = BYTE_ARR_CTS(R2);
741 vp = BYTE_ARR_CTS(R4);
743 cmp = foreign "C" mpn_cmp(up "ptr", vp "ptr", size);
745 if (cmp == 0 :: CInt) {
747 jump %ENTRY_CODE(Sp(0));
750 if (%lt(cmp,0 :: CInt) == %lt(usize,0)) {
755 /* Result parked in R1, return via info-pointer at TOS */
756 jump %ENTRY_CODE(Sp(0));
768 r = W_[R2 + SIZEOF_StgArrWords];
773 /* Result parked in R1, return via info-pointer at TOS */
775 jump %ENTRY_CODE(Sp(0));
787 r = W_[R2 + SIZEOF_StgArrWords];
792 /* Result parked in R1, return via info-pointer at TOS */
794 jump %ENTRY_CODE(Sp(0));
806 /* arguments: F1 = Float# */
809 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
811 /* Be prepared to tell Lennart-coded __decodeFloat
812 where mantissa._mp_d can be put (it does not care about the rest) */
813 p = Hp - SIZEOF_StgArrWords;
814 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
815 StgArrWords_words(p) = 1;
816 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
818 /* Perform the operation */
819 foreign "C" __decodeFloat(mp_tmp1,exponent,arg);
821 /* returns: (Int# (expn), Int#, ByteArray#) */
822 RET_NNP(W_[exponent], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
825 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
826 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
833 /* arguments: D1 = Double# */
836 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
838 /* Be prepared to tell Lennart-coded __decodeDouble
839 where mantissa.d can be put (it does not care about the rest) */
840 p = Hp - ARR_SIZE + WDS(1);
841 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
842 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
843 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
845 /* Perform the operation */
846 foreign "C" __decodeDouble(mp_tmp1,exponent,arg);
848 /* returns: (Int# (expn), Int#, ByteArray#) */
849 RET_NNP(W_[exponent], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
852 /* -----------------------------------------------------------------------------
853 * Concurrency primitives
854 * -------------------------------------------------------------------------- */
858 /* args: R1 = closure to spark */
860 MAYBE_GC(R1_PTR, forkzh_fast);
862 // create it right now, return ThreadID in R1
863 "ptr" R1 = foreign "C" createIOThread( RtsFlags_GcFlags_initialStkSize(RtsFlags),
865 foreign "C" scheduleThread(R1 "ptr");
867 // switch at the earliest opportunity
868 CInt[context_switch] = 1 :: CInt;
875 jump stg_yield_noregs;
890 foreign "C" labelThread(R1 "ptr", R2 "ptr");
892 jump %ENTRY_CODE(Sp(0));
895 isCurrentThreadBoundzh_fast
899 r = foreign "C" isThreadBound(CurrentTSO);
904 /* -----------------------------------------------------------------------------
906 * -------------------------------------------------------------------------- */
910 #define IF_NOT_REG_R1(x)
913 #define IF_NOT_REG_R1(x) x
916 // Catch retry frame ------------------------------------------------------------
919 #define CATCH_RETRY_FRAME_ENTRY_TEMPLATE(label,ret) \
922 W_ r, frame, trec, outer; \
923 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \
926 trec = StgTSO_trec(CurrentTSO); \
927 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr"); \
928 r = foreign "C" stmCommitTransaction(trec "ptr"); \
930 /* Succeeded (either first branch or second branch) */ \
931 StgTSO_trec(CurrentTSO) = outer; \
932 Sp = Sp + SIZEOF_StgCatchRetryFrame; \
933 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \
936 /* Did not commit: retry */ \
938 "ptr" new_trec = foreign "C" stmStartTransaction(outer "ptr"); \
939 StgTSO_trec(CurrentTSO) = new_trec; \
940 if (StgCatchRetryFrame_running_alt_code(frame)) { \
941 R1 = StgCatchRetryFrame_alt_code(frame); \
943 R1 = StgCatchRetryFrame_first_code(frame); \
944 StgCatchRetryFrame_first_code_trec(frame) = new_trec; \
947 jump RET_LBL(stg_ap_v); \
951 CATCH_RETRY_FRAME_ENTRY_TEMPLATE(stg_catch_retry_frame_0_ret,%RET_VEC(Sp(SP_OFF),0))
952 CATCH_RETRY_FRAME_ENTRY_TEMPLATE(stg_catch_retry_frame_1_ret,%RET_VEC(Sp(SP_OFF),1))
953 CATCH_RETRY_FRAME_ENTRY_TEMPLATE(stg_catch_retry_frame_2_ret,%RET_VEC(Sp(SP_OFF),2))
954 CATCH_RETRY_FRAME_ENTRY_TEMPLATE(stg_catch_retry_frame_3_ret,%RET_VEC(Sp(SP_OFF),3))
955 CATCH_RETRY_FRAME_ENTRY_TEMPLATE(stg_catch_retry_frame_4_ret,%RET_VEC(Sp(SP_OFF),4))
956 CATCH_RETRY_FRAME_ENTRY_TEMPLATE(stg_catch_retry_frame_5_ret,%RET_VEC(Sp(SP_OFF),5))
957 CATCH_RETRY_FRAME_ENTRY_TEMPLATE(stg_catch_retry_frame_6_ret,%RET_VEC(Sp(SP_OFF),6))
958 CATCH_RETRY_FRAME_ENTRY_TEMPLATE(stg_catch_retry_frame_7_ret,%RET_VEC(Sp(SP_OFF),7))
960 #if MAX_VECTORED_RTN > 8
961 #error MAX_VECTORED_RTN has changed: please modify stg_catch_retry_frame too.
964 #if defined(PROFILING)
965 #define CATCH_RETRY_FRAME_BITMAP 7
966 #define CATCH_RETRY_FRAME_WORDS 6
968 #define CATCH_RETRY_FRAME_BITMAP 1
969 #define CATCH_RETRY_FRAME_WORDS 4
972 INFO_TABLE_RET(stg_catch_retry_frame,
973 CATCH_RETRY_FRAME_WORDS, CATCH_RETRY_FRAME_BITMAP,
975 stg_catch_retry_frame_0_ret,
976 stg_catch_retry_frame_1_ret,
977 stg_catch_retry_frame_2_ret,
978 stg_catch_retry_frame_3_ret,
979 stg_catch_retry_frame_4_ret,
980 stg_catch_retry_frame_5_ret,
981 stg_catch_retry_frame_6_ret,
982 stg_catch_retry_frame_7_ret)
983 CATCH_RETRY_FRAME_ENTRY_TEMPLATE(,%ENTRY_CODE(Sp(SP_OFF)))
987 // Atomically frame -------------------------------------------------------------
989 #define ATOMICALLY_FRAME_ENTRY_TEMPLATE(label,ret) \
992 W_ frame, trec, valid; \
993 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \
996 trec = StgTSO_trec(CurrentTSO); \
997 if (StgAtomicallyFrame_waiting(frame)) { \
998 /* The TSO is currently waiting: should we stop waiting? */ \
999 valid = foreign "C" stmReWait(trec "ptr"); \
1001 /* Previous attempt is still valid: no point trying again yet */ \
1002 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \
1003 jump stg_block_noregs; \
1005 /* Previous attempt is no longer valid: try again */ \
1006 "ptr" trec = foreign "C" stmStartTransaction(NO_TREC "ptr"); \
1007 StgTSO_trec(CurrentTSO) = trec; \
1008 StgAtomicallyFrame_waiting(frame) = 0 :: CInt; /* false; */ \
1009 R1 = StgAtomicallyFrame_code(frame); \
1011 jump RET_LBL(stg_ap_v); \
1014 /* The TSO is not currently waiting: try to commit the transaction */ \
1015 valid = foreign "C" stmCommitTransaction(trec "ptr"); \
1017 /* Transaction was valid: commit succeeded */ \
1018 StgTSO_trec(CurrentTSO) = NO_TREC; \
1019 Sp = Sp + SIZEOF_StgAtomicallyFrame; \
1020 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \
1023 /* Transaction was not valid: try again */ \
1024 "ptr" trec = foreign "C" stmStartTransaction(NO_TREC "ptr"); \
1025 StgTSO_trec(CurrentTSO) = trec; \
1026 R1 = StgAtomicallyFrame_code(frame); \
1028 jump RET_LBL(stg_ap_v); \
1033 ATOMICALLY_FRAME_ENTRY_TEMPLATE(stg_atomically_frame_0_ret,%RET_VEC(Sp(SP_OFF),0))
1034 ATOMICALLY_FRAME_ENTRY_TEMPLATE(stg_atomically_frame_1_ret,%RET_VEC(Sp(SP_OFF),1))
1035 ATOMICALLY_FRAME_ENTRY_TEMPLATE(stg_atomically_frame_2_ret,%RET_VEC(Sp(SP_OFF),2))
1036 ATOMICALLY_FRAME_ENTRY_TEMPLATE(stg_atomically_frame_3_ret,%RET_VEC(Sp(SP_OFF),3))
1037 ATOMICALLY_FRAME_ENTRY_TEMPLATE(stg_atomically_frame_4_ret,%RET_VEC(Sp(SP_OFF),4))
1038 ATOMICALLY_FRAME_ENTRY_TEMPLATE(stg_atomically_frame_5_ret,%RET_VEC(Sp(SP_OFF),5))
1039 ATOMICALLY_FRAME_ENTRY_TEMPLATE(stg_atomically_frame_6_ret,%RET_VEC(Sp(SP_OFF),6))
1040 ATOMICALLY_FRAME_ENTRY_TEMPLATE(stg_atomically_frame_7_ret,%RET_VEC(Sp(SP_OFF),7))
1042 #if MAX_VECTORED_RTN > 8
1043 #error MAX_VECTORED_RTN has changed: please modify stg_atomically_frame too.
1046 #if defined(PROFILING)
1047 #define ATOMICALLY_FRAME_BITMAP 7
1048 #define ATOMICALLY_FRAME_WORDS 4
1050 #define ATOMICALLY_FRAME_BITMAP 1
1051 #define ATOMICALLY_FRAME_WORDS 2
1054 INFO_TABLE_RET(stg_atomically_frame,
1055 ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP,
1057 stg_atomically_frame_0_ret,
1058 stg_atomically_frame_1_ret,
1059 stg_atomically_frame_2_ret,
1060 stg_atomically_frame_3_ret,
1061 stg_atomically_frame_4_ret,
1062 stg_atomically_frame_5_ret,
1063 stg_atomically_frame_6_ret,
1064 stg_atomically_frame_7_ret)
1065 ATOMICALLY_FRAME_ENTRY_TEMPLATE(,%ENTRY_CODE(Sp(SP_OFF)))
1068 // STM catch frame --------------------------------------------------------------
1070 #define CATCH_STM_FRAME_ENTRY_TEMPLATE(label,ret) \
1073 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \
1074 Sp = Sp + SIZEOF_StgCatchSTMFrame; \
1075 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \
1085 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_0_ret,%RET_VEC(Sp(SP_OFF),0))
1086 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_1_ret,%RET_VEC(Sp(SP_OFF),1))
1087 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_2_ret,%RET_VEC(Sp(SP_OFF),2))
1088 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_3_ret,%RET_VEC(Sp(SP_OFF),3))
1089 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_4_ret,%RET_VEC(Sp(SP_OFF),4))
1090 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_5_ret,%RET_VEC(Sp(SP_OFF),5))
1091 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_6_ret,%RET_VEC(Sp(SP_OFF),6))
1092 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_7_ret,%RET_VEC(Sp(SP_OFF),7))
1094 #if MAX_VECTORED_RTN > 8
1095 #error MAX_VECTORED_RTN has changed: please modify stg_catch_stm_frame too.
1098 #if defined(PROFILING)
1099 #define CATCH_STM_FRAME_BITMAP 3
1100 #define CATCH_STM_FRAME_WORDS 3
1102 #define CATCH_STM_FRAME_BITMAP 0
1103 #define CATCH_STM_FRAME_WORDS 1
1106 /* Catch frames are very similar to update frames, but when entering
1107 * one we just pop the frame off the stack and perform the correct
1108 * kind of return to the activation record underneath us on the stack.
1111 INFO_TABLE_RET(stg_catch_stm_frame,
1112 CATCH_STM_FRAME_WORDS, CATCH_STM_FRAME_BITMAP,
1114 stg_catch_stm_frame_0_ret,
1115 stg_catch_stm_frame_1_ret,
1116 stg_catch_stm_frame_2_ret,
1117 stg_catch_stm_frame_3_ret,
1118 stg_catch_stm_frame_4_ret,
1119 stg_catch_stm_frame_5_ret,
1120 stg_catch_stm_frame_6_ret,
1121 stg_catch_stm_frame_7_ret)
1122 CATCH_STM_FRAME_ENTRY_TEMPLATE(,%ENTRY_CODE(Sp(SP_OFF)))
1125 // Primop definition ------------------------------------------------------------
1133 /* Args: R1 = m :: STM a */
1134 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);
1136 /* Set up the atomically frame */
1137 Sp = Sp - SIZEOF_StgAtomicallyFrame;
1140 SET_HDR(frame,stg_atomically_frame_info,CCCS);
1141 StgAtomicallyFrame_waiting(frame) = 0 :: CInt; // False
1142 StgAtomicallyFrame_code(frame) = R1;
1144 /* Start the memory transcation */
1145 old_trec = StgTSO_trec(CurrentTSO);
1146 "ptr" new_trec = foreign "C" stmStartTransaction(old_trec "ptr");
1147 StgTSO_trec(CurrentTSO) = new_trec;
1149 /* Apply R1 to the realworld token */
1151 jump RET_LBL(stg_ap_v);
1159 /* Args: R1 :: STM a */
1160 /* Args: R2 :: Exception -> STM a */
1161 STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast);
1163 /* Set up the catch frame */
1164 Sp = Sp - SIZEOF_StgCatchSTMFrame;
1167 SET_HDR(frame, stg_catch_stm_frame_info, CCCS);
1168 StgCatchSTMFrame_handler(frame) = R2;
1170 /* Apply R1 to the realworld token */
1172 jump RET_LBL(stg_ap_v);
1182 /* Args: R1 :: STM a */
1183 /* Args: R2 :: STM a */
1184 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);
1186 /* Start a nested transaction within which to run the first code */
1187 trec = StgTSO_trec(CurrentTSO);
1188 "ptr" new_trec = foreign "C" stmStartTransaction(trec "ptr");
1189 StgTSO_trec(CurrentTSO) = new_trec;
1191 /* Set up the catch-retry frame */
1192 Sp = Sp - SIZEOF_StgCatchRetryFrame;
1195 SET_HDR(frame, stg_catch_retry_frame_info, CCCS);
1196 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1197 StgCatchRetryFrame_first_code(frame) = R1;
1198 StgCatchRetryFrame_alt_code(frame) = R2;
1199 StgCatchRetryFrame_first_code_trec(frame) = new_trec;
1201 /* Apply R1 to the realworld token */
1203 jump RET_LBL(stg_ap_v);
1215 MAYBE_GC (NO_PTRS, readTVarzh_fast); // STM operations may allocate
1217 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
1219 trec = StgTSO_trec(CurrentTSO);
1220 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr");
1221 StgTSO_sp(CurrentTSO) = Sp;
1222 frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr");
1223 Sp = StgTSO_sp(CurrentTSO);
1226 if (frame_type == CATCH_RETRY_FRAME) {
1227 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
1228 ASSERT(outer != NO_TREC);
1229 if (!StgCatchRetryFrame_running_alt_code(frame)) {
1230 // Retry in the first code: try the alternative
1231 "ptr" trec = foreign "C" stmStartTransaction(outer "ptr");
1232 StgTSO_trec(CurrentTSO) = trec;
1233 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
1234 R1 = StgCatchRetryFrame_alt_code(frame);
1236 jump RET_LBL(stg_ap_v);
1238 // Retry in the alternative code: propagate
1240 other_trec = StgCatchRetryFrame_first_code_trec(frame);
1241 r = foreign "C" stmMergeForWaiting(trec "ptr", other_trec "ptr");
1243 // Merge between siblings succeeded: commit it back to enclosing transaction
1244 // and then propagate the retry
1245 r = foreign "C" stmCommitTransaction(trec "ptr");
1246 StgTSO_trec(CurrentTSO) = outer;
1247 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1248 goto retry_pop_stack;
1250 // Merge failed: we musn't propagate the retry. Try both paths again.
1251 "ptr" trec = foreign "C" stmStartTransaction(outer "ptr");
1252 StgCatchRetryFrame_first_code_trec(frame) = trec;
1253 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1254 StgTSO_trec(CurrentTSO) = trec;
1255 R1 = StgCatchRetryFrame_first_code(frame);
1257 jump RET_LBL(stg_ap_v);
1262 // We've reached the ATOMICALLY_FRAME: attempt to wait
1263 ASSERT(frame_type == ATOMICALLY_FRAME);
1264 ASSERT(outer == NO_TREC);
1265 r = foreign "C" stmWait(CurrentTSO "ptr", trec "ptr");
1267 // Transaction was valid: stmWait put us on the TVars' queues, we now block
1268 StgAtomicallyFrame_waiting(frame) = 1 :: CInt; // true
1270 jump stg_block_noregs;
1272 // Transaction was not valid: retry immediately
1273 "ptr" trec = foreign "C" stmStartTransaction(outer "ptr");
1274 StgTSO_trec(CurrentTSO) = trec;
1275 R1 = StgAtomicallyFrame_code(frame);
1278 jump RET_LBL(stg_ap_v);
1287 /* Args: R1 = initialisation value */
1289 ALLOC_PRIM( SIZEOF_StgTVar, R1_PTR, newTVarzh_fast);
1290 tv = Hp - SIZEOF_StgTVar + WDS(1);
1291 SET_HDR(tv,stg_TVAR_info,W_[CCCS]);
1292 StgTVar_current_value(tv) = R1;
1293 StgTVar_first_wait_queue_entry(tv) = stg_END_STM_WAIT_QUEUE_closure;
1305 /* Args: R1 = TVar closure */
1307 MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
1308 trec = StgTSO_trec(CurrentTSO);
1310 "ptr" result = foreign "C" stmReadTVar(trec "ptr", tvar "ptr");
1322 /* Args: R1 = TVar closure */
1323 /* R2 = New value */
1325 MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
1326 trec = StgTSO_trec(CurrentTSO);
1329 foreign "C" stmWriteTVar(trec "ptr", tvar "ptr", new_value "ptr");
1331 jump %ENTRY_CODE(Sp(0));
1335 /* -----------------------------------------------------------------------------
1338 * take & putMVar work as follows. Firstly, an important invariant:
1340 * If the MVar is full, then the blocking queue contains only
1341 * threads blocked on putMVar, and if the MVar is empty then the
1342 * blocking queue contains only threads blocked on takeMVar.
1345 * MVar empty : then add ourselves to the blocking queue
1346 * MVar full : remove the value from the MVar, and
1347 * blocking queue empty : return
1348 * blocking queue non-empty : perform the first blocked putMVar
1349 * from the queue, and wake up the
1350 * thread (MVar is now full again)
1352 * putMVar is just the dual of the above algorithm.
1354 * How do we "perform a putMVar"? Well, we have to fiddle around with
1355 * the stack of the thread waiting to do the putMVar. See
1356 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1357 * the stack layout, and the PerformPut and PerformTake macros below.
1359 * It is important that a blocked take or put is woken up with the
1360 * take/put already performed, because otherwise there would be a
1361 * small window of vulnerability where the thread could receive an
1362 * exception and never perform its take or put, and we'd end up with a
1365 * -------------------------------------------------------------------------- */
1369 /* args: R1 = MVar closure */
1371 if (GET_INFO(R1) == stg_EMPTY_MVAR_info) {
1383 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
1385 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1386 SET_HDR(mvar,stg_EMPTY_MVAR_info,W_[CCCS]);
1387 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1388 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1389 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1394 /* If R1 isn't available, pass it on the stack */
1396 #define PerformTake(tso, value) \
1397 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1398 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1400 #define PerformTake(tso, value) \
1401 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1402 W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info;
1405 #define PerformPut(tso,lval) \
1406 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1407 lval = W_[StgTSO_sp(tso) - WDS(1)];
1412 W_ mvar, val, info, tso;
1414 /* args: R1 = MVar closure */
1417 info = GET_INFO(mvar);
1419 /* If the MVar is empty, put ourselves on its blocking queue,
1420 * and wait until we're woken up.
1422 if (info == stg_EMPTY_MVAR_info) {
1423 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1424 StgMVar_head(mvar) = CurrentTSO;
1426 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1428 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1429 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1430 StgTSO_block_info(CurrentTSO) = mvar;
1431 StgMVar_tail(mvar) = CurrentTSO;
1433 jump stg_block_takemvar;
1436 /* we got the value... */
1437 val = StgMVar_value(mvar);
1439 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1441 /* There are putMVar(s) waiting...
1442 * wake up the first thread on the queue
1444 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1446 /* actually perform the putMVar for the thread that we just woke up */
1447 tso = StgMVar_head(mvar);
1448 PerformPut(tso,StgMVar_value(mvar));
1450 #if defined(GRAN) || defined(PAR)
1451 /* ToDo: check 2nd arg (mvar) is right */
1452 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar);
1453 StgMVar_head(mvar) = tso;
1455 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1456 StgMVar_head(mvar) = tso;
1458 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1459 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1465 /* No further putMVars, MVar is now empty */
1467 /* do this last... we might have locked the MVar in the SMP case,
1468 * and writing the info pointer will unlock it.
1470 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1471 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1479 W_ mvar, val, info, tso;
1481 /* args: R1 = MVar closure */
1485 info = GET_INFO(mvar);
1487 if (info == stg_EMPTY_MVAR_info) {
1488 /* HACK: we need a pointer to pass back,
1489 * so we abuse NO_FINALIZER_closure
1491 RET_NP(0, stg_NO_FINALIZER_closure);
1494 /* we got the value... */
1495 val = StgMVar_value(mvar);
1497 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) "ptr", mvar "ptr");
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;
1522 /* No further putMVars, MVar is now empty */
1523 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1525 /* do this last... we might have locked the MVar in the SMP case,
1526 * and writing the info pointer will unlock it.
1528 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1539 /* args: R1 = MVar, R2 = value */
1542 info = GET_INFO(mvar);
1544 if (info == stg_FULL_MVAR_info) {
1545 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1546 StgMVar_head(mvar) = CurrentTSO;
1548 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1550 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1551 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1552 StgTSO_block_info(CurrentTSO) = mvar;
1553 StgMVar_tail(mvar) = CurrentTSO;
1555 jump stg_block_putmvar;
1558 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1559 /* There are takeMVar(s) waiting: wake up the first one
1561 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1563 /* actually perform the takeMVar */
1564 tso = StgMVar_head(mvar);
1565 PerformTake(tso, R2);
1567 #if defined(GRAN) || defined(PAR)
1568 /* ToDo: check 2nd arg (mvar) is right */
1569 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1570 StgMVar_head(mvar) = tso;
1572 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1573 StgMVar_head(mvar) = tso;
1576 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1577 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1580 jump %ENTRY_CODE(Sp(0));
1584 /* No further takes, the MVar is now full. */
1585 StgMVar_value(mvar) = R2;
1586 /* unlocks the MVar in the SMP case */
1587 SET_INFO(mvar,stg_FULL_MVAR_info);
1588 jump %ENTRY_CODE(Sp(0));
1591 /* ToDo: yield afterward for better communication performance? */
1599 /* args: R1 = MVar, R2 = value */
1602 info = GET_INFO(mvar);
1604 if (info == stg_FULL_MVAR_info) {
1608 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1609 /* There are takeMVar(s) waiting: wake up the first one
1611 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1613 /* actually perform the takeMVar */
1614 tso = StgMVar_head(mvar);
1615 PerformTake(tso, R2);
1617 #if defined(GRAN) || defined(PAR)
1618 /* ToDo: check 2nd arg (mvar) is right */
1619 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1620 StgMVar_head(mvar) = tso;
1622 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1623 StgMVar_head(mvar) = tso;
1626 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1627 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1630 jump %ENTRY_CODE(Sp(0));
1634 /* No further takes, the MVar is now full. */
1635 StgMVar_value(mvar) = R2;
1636 /* unlocks the MVar in the SMP case */
1637 SET_INFO(mvar,stg_FULL_MVAR_info);
1638 jump %ENTRY_CODE(Sp(0));
1641 /* ToDo: yield afterward for better communication performance? */
1645 /* -----------------------------------------------------------------------------
1646 Stable pointer primitives
1647 ------------------------------------------------------------------------- */
1649 makeStableNamezh_fast
1653 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1655 index = foreign "C" lookupStableName(R1 "ptr");
1657 /* Is there already a StableName for this heap object?
1658 * stable_ptr_table is an array of snEntry structs.
1660 if ( snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry) == NULL ) {
1661 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1662 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1663 StgStableName_sn(sn_obj) = index;
1664 snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry) = sn_obj;
1666 sn_obj = snEntry_sn_obj(stable_ptr_table + index*SIZEOF_snEntry);
1673 makeStablePtrzh_fast
1677 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1678 "ptr" sp = foreign "C" getStablePtr(R1 "ptr");
1682 deRefStablePtrzh_fast
1684 /* Args: R1 = the stable ptr */
1687 r = snEntry_addr(stable_ptr_table + sp*SIZEOF_snEntry);
1691 /* -----------------------------------------------------------------------------
1692 Bytecode object primitives
1693 ------------------------------------------------------------------------- */
1704 W_ bco, bitmap_arr, bytes, words;
1707 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1710 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R4_PTR&R6_PTR, newBCOzh_fast );
1712 bco = Hp - bytes + WDS(1);
1713 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1715 StgBCO_instrs(bco) = R1;
1716 StgBCO_literals(bco) = R2;
1717 StgBCO_ptrs(bco) = R3;
1718 StgBCO_itbls(bco) = R4;
1719 StgBCO_arity(bco) = HALF_W_(R5);
1720 StgBCO_size(bco) = HALF_W_(words);
1722 // Copy the arity/bitmap info into the BCO
1726 if (i < StgArrWords_words(bitmap_arr)) {
1727 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1738 // R1 = the BCO# for the AP
1742 // This function is *only* used to wrap zero-arity BCOs in an
1743 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1744 // saturated and always points directly to a FUN or BCO.
1745 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
1746 StgBCO_arity(R1) == HALF_W_(0));
1748 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1749 TICK_ALLOC_UP_THK(0, 0);
1750 CCCS_ALLOC(SIZEOF_StgAP);
1752 ap = Hp - SIZEOF_StgAP + WDS(1);
1753 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1755 StgAP_n_args(ap) = HALF_W_(0);
1761 /* -----------------------------------------------------------------------------
1762 Thread I/O blocking primitives
1763 -------------------------------------------------------------------------- */
1765 /* Add a thread to the end of the blocked queue. (C-- version of the C
1766 * macro in Schedule.h).
1768 #define APPEND_TO_BLOCKED_QUEUE(tso) \
1769 ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \
1770 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
1771 W_[blocked_queue_hd] = tso; \
1773 StgTSO_link(W_[blocked_queue_tl]) = tso; \
1775 W_[blocked_queue_tl] = tso;
1781 foreign "C" barf("waitRead# on threaded RTS");
1784 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1785 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1786 StgTSO_block_info(CurrentTSO) = R1;
1787 // No locking - we're not going to use this interface in the
1788 // threaded RTS anyway.
1789 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1790 jump stg_block_noregs;
1797 foreign "C" barf("waitWrite# on threaded RTS");
1800 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1801 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1802 StgTSO_block_info(CurrentTSO) = R1;
1803 // No locking - we're not going to use this interface in the
1804 // threaded RTS anyway.
1805 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1806 jump stg_block_noregs;
1810 STRING(stg_delayzh_malloc_str, "delayzh_fast")
1813 #ifdef mingw32_TARGET_OS
1821 foreign "C" barf("delay# on threaded RTS");
1824 /* args: R1 (microsecond delay amount) */
1825 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1826 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
1828 #ifdef mingw32_TARGET_OS
1830 /* could probably allocate this on the heap instead */
1831 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1832 stg_delayzh_malloc_str);
1833 reqID = foreign "C" addDelayRequest(R1);
1834 StgAsyncIOResult_reqID(ares) = reqID;
1835 StgAsyncIOResult_len(ares) = 0;
1836 StgAsyncIOResult_errCode(ares) = 0;
1837 StgTSO_block_info(CurrentTSO) = ares;
1839 /* Having all async-blocked threads reside on the blocked_queue
1840 * simplifies matters, so change the status to OnDoProc put the
1841 * delayed thread on the blocked_queue.
1843 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1844 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1845 jump stg_block_async_void;
1850 time = foreign "C" getourtimeofday();
1851 target = (R1 / (TICK_MILLISECS*1000)) + time;
1852 StgTSO_block_info(CurrentTSO) = target;
1854 /* Insert the new thread in the sleeping queue. */
1856 t = W_[sleeping_queue];
1858 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
1864 StgTSO_link(CurrentTSO) = t;
1866 W_[sleeping_queue] = CurrentTSO;
1868 StgTSO_link(prev) = CurrentTSO;
1870 jump stg_block_noregs;
1875 #ifdef mingw32_TARGET_OS
1876 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
1883 foreign "C" barf("asyncRead# on threaded RTS");
1886 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1887 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1888 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1890 /* could probably allocate this on the heap instead */
1891 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1892 stg_asyncReadzh_malloc_str);
1893 reqID = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr");
1894 StgAsyncIOResult_reqID(ares) = reqID;
1895 StgAsyncIOResult_len(ares) = 0;
1896 StgAsyncIOResult_errCode(ares) = 0;
1897 StgTSO_block_info(CurrentTSO) = ares;
1898 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1899 jump stg_block_async;
1902 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
1909 foreign "C" barf("asyncWrite# on threaded RTS");
1912 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1913 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1914 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1916 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1917 stg_asyncWritezh_malloc_str);
1918 reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr");
1920 StgAsyncIOResult_reqID(ares) = reqID;
1921 StgAsyncIOResult_len(ares) = 0;
1922 StgAsyncIOResult_errCode(ares) = 0;
1923 StgTSO_block_info(CurrentTSO) = ares;
1924 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1925 jump stg_block_async;
1928 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
1934 /* args: R1 = proc, R2 = param */
1935 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1936 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1938 /* could probably allocate this on the heap instead */
1939 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1940 stg_asyncDoProczh_malloc_str);
1941 reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr");
1942 StgAsyncIOResult_reqID(ares) = reqID;
1943 StgAsyncIOResult_len(ares) = 0;
1944 StgAsyncIOResult_errCode(ares) = 0;
1945 StgTSO_block_info(CurrentTSO) = ares;
1946 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1947 jump stg_block_async;
1951 /* -----------------------------------------------------------------------------
1954 classes CCallable and CReturnable don't really exist, but the
1955 compiler insists on generating dictionaries containing references
1956 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
1957 for these. Some C compilers can't cope with zero-length static arrays,
1958 so we have to make these one element long.
1959 --------------------------------------------------------------------------- */
1962 GHC_ZCCCallable_static_info: W_ 0;
1966 GHC_ZCCReturnable_static_info: W_ 0;