1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2004
5 * Out-of-line primitive operations
7 * This file contains the implementations of all the primitive
8 * operations ("primops") which are not expanded inline. See
9 * ghc/compiler/prelude/primops.txt.pp for a list of all the primops;
10 * this file contains code for most of those with the attribute
13 * Entry convention: the entry convention for a primop is that all the
14 * args are in Stg registers (R1, R2, etc.). This is to make writing
15 * the primops easier. (see compiler/codeGen/CgCallConv.hs).
17 * Return convention: results from a primop are generally returned
18 * using the ordinary unboxed tuple return convention. The C-- parser
19 * implements the RET_xxxx() macros to perform unboxed-tuple returns
20 * based on the prevailing return convention.
22 * This file is written in a subset of C--, extended with various
23 * features specific to GHC. It is compiled by GHC directly. For the
24 * syntax of .cmm files, see the parser in ghc/compiler/cmm/CmmParse.y.
26 * ---------------------------------------------------------------------------*/
30 /*-----------------------------------------------------------------------------
33 Basically just new*Array - the others are all inline macros.
35 The size arg is always passed in R1, and the result returned in R1.
37 The slow entry point is for returning from a heap check, the saved
38 size argument must be re-loaded from the stack.
39 -------------------------------------------------------------------------- */
41 /* for objects that are *less* than the size of a word, make sure we
42 * round up to the nearest word for the size of the array.
47 W_ words, payload_words, n, p;
48 MAYBE_GC(NO_PTRS,newByteArrayzh_fast);
50 payload_words = ROUNDUP_BYTES_TO_WDS(n);
51 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
52 "ptr" p = foreign "C" allocateLocal(MyCapability() "ptr",words) [];
53 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
54 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
55 StgArrWords_words(p) = payload_words;
59 newPinnedByteArrayzh_fast
61 W_ words, payload_words, n, p;
63 MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast);
65 payload_words = ROUNDUP_BYTES_TO_WDS(n);
67 // We want an 8-byte aligned array. allocatePinned() gives us
68 // 8-byte aligned memory by default, but we want to align the
69 // *goods* inside the ArrWords object, so we have to check the
70 // size of the ArrWords header and adjust our size accordingly.
71 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
72 if ((SIZEOF_StgArrWords & 7) != 0) {
76 "ptr" p = foreign "C" allocatePinned(words) [];
77 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
79 // Again, if the ArrWords header isn't a multiple of 8 bytes, we
80 // have to push the object forward one word so that the goods
81 // fall on an 8-byte boundary.
82 if ((SIZEOF_StgArrWords & 7) != 0) {
86 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
87 StgArrWords_words(p) = payload_words;
93 W_ words, n, init, arr, p;
94 /* Args: R1 = words, R2 = initialisation value */
97 MAYBE_GC(R2_PTR,newArrayzh_fast);
99 words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
100 "ptr" arr = foreign "C" allocateLocal(MyCapability() "ptr",words) [];
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
130 // to indicate that it is still on the mutable list.
132 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
133 // either it is on a mut_list, or it isn't. We adopt the convention that
134 // the closure type is MUT_ARR_PTRS_FROZEN0 if it is on the mutable list,
135 // and MUT_ARR_PTRS_FROZEN otherwise. In fact it wouldn't matter if
136 // we put it on the mutable list more than once, but it would get scavenged
137 // multiple times during GC, which would be unnecessarily slow.
139 if (StgHeader_info(R1) != stg_MUT_ARR_PTRS_FROZEN0_info) {
140 SET_INFO(R1,stg_MUT_ARR_PTRS_info);
141 foreign "C" recordMutableLock(R1 "ptr") [R1];
142 // must be done after SET_INFO, because it ASSERTs closure_MUTABLE()
145 SET_INFO(R1,stg_MUT_ARR_PTRS_info);
150 /* -----------------------------------------------------------------------------
152 -------------------------------------------------------------------------- */
157 /* Args: R1 = initialisation value */
159 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);
161 mv = Hp - SIZEOF_StgMutVar + WDS(1);
162 SET_HDR(mv,stg_MUT_VAR_info,W_[CCCS]);
163 StgMutVar_var(mv) = R1;
168 atomicModifyMutVarzh_fast
171 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
173 /* If x is the current contents of the MutVar#, then
174 We want to make the new contents point to
178 and the return value is
182 obviously we can share (f x).
184 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
185 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
186 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
190 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
191 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
193 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
194 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
198 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
199 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
201 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
202 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
205 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
207 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);
210 foreign "C" ACQUIRE_LOCK(sm_mutex "ptr");
213 x = StgMutVar_var(R1);
215 TICK_ALLOC_THUNK_2();
216 CCCS_ALLOC(THUNK_2_SIZE);
217 z = Hp - THUNK_2_SIZE + WDS(1);
218 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
219 LDV_RECORD_CREATE(z);
220 StgThunk_payload(z,0) = R2;
221 StgThunk_payload(z,1) = x;
223 TICK_ALLOC_THUNK_1();
224 CCCS_ALLOC(THUNK_1_SIZE);
225 y = z - THUNK_1_SIZE;
226 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
227 LDV_RECORD_CREATE(y);
228 StgThunk_payload(y,0) = z;
230 StgMutVar_var(R1) = y;
232 TICK_ALLOC_THUNK_1();
233 CCCS_ALLOC(THUNK_1_SIZE);
234 r = y - THUNK_1_SIZE;
235 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
236 LDV_RECORD_CREATE(r);
237 StgThunk_payload(r,0) = z;
240 foreign "C" RELEASE_LOCK(sm_mutex "ptr") [];
246 /* -----------------------------------------------------------------------------
247 Weak Pointer Primitives
248 -------------------------------------------------------------------------- */
250 STRING(stg_weak_msg,"New weak pointer at %p\n")
256 R3 = finalizer (or NULL)
261 R3 = stg_NO_FINALIZER_closure;
264 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
266 w = Hp - SIZEOF_StgWeak + WDS(1);
267 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
270 StgWeak_value(w) = R2;
271 StgWeak_finalizer(w) = R3;
273 StgWeak_link(w) = W_[weak_ptr_list];
274 W_[weak_ptr_list] = w;
276 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
291 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
292 RET_NP(0,stg_NO_FINALIZER_closure);
298 // A weak pointer is inherently used, so we do not need to call
299 // LDV_recordDead_FILL_SLOP_DYNAMIC():
300 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
301 // or, LDV_recordDead():
302 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
303 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
304 // large as weak pointers, so there is no need to fill the slop, either.
305 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
309 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
311 SET_INFO(w,stg_DEAD_WEAK_info);
312 LDV_RECORD_CREATE(w);
314 f = StgWeak_finalizer(w);
315 StgDeadWeak_link(w) = StgWeak_link(w);
317 /* return the finalizer */
318 if (f == stg_NO_FINALIZER_closure) {
319 RET_NP(0,stg_NO_FINALIZER_closure);
331 if (GET_INFO(w) == stg_WEAK_info) {
333 val = StgWeak_value(w);
341 /* -----------------------------------------------------------------------------
342 Arbitrary-precision Integer operations.
344 There are some assumptions in this code that mp_limb_t == W_. This is
345 the case for all the platforms that GHC supports, currently.
346 -------------------------------------------------------------------------- */
350 /* arguments: R1 = Int# */
352 W_ val, s, p; /* to avoid aliasing */
355 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
357 p = Hp - SIZEOF_StgArrWords;
358 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
359 StgArrWords_words(p) = 1;
361 /* mpz_set_si is inlined here, makes things simpler */
374 /* returns (# size :: Int#,
383 /* arguments: R1 = Word# */
385 W_ val, s, p; /* to avoid aliasing */
389 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
391 p = Hp - SIZEOF_StgArrWords;
392 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
393 StgArrWords_words(p) = 1;
402 /* returns (# size :: Int#,
403 data :: ByteArray# #)
410 * 'long long' primops for converting to/from Integers.
413 #ifdef SUPPORT_LONG_LONGS
415 int64ToIntegerzh_fast
417 /* arguments: L1 = Int64# */
420 W_ hi, s, neg, words_needed, p;
425 if ( %ge(val,0x100000000::L_) || %le(val,-0x100000000::L_) ) {
428 // minimum is one word
432 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
433 NO_PTRS, int64ToIntegerzh_fast );
435 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
436 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
437 StgArrWords_words(p) = words_needed;
439 if ( %lt(val,0::L_) ) {
444 hi = TO_W_(val >> 32);
446 if ( words_needed == 2 ) {
451 if ( val != 0::L_ ) {
454 } else /* val==0 */ {
462 /* returns (# size :: Int#,
463 data :: ByteArray# #)
468 word64ToIntegerzh_fast
470 /* arguments: L1 = Word64# */
473 W_ hi, s, words_needed, p;
476 if ( val >= 0x100000000::L_ ) {
482 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
483 NO_PTRS, word64ToIntegerzh_fast );
485 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
486 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
487 StgArrWords_words(p) = words_needed;
489 hi = TO_W_(val >> 32);
490 if ( val >= 0x100000000::L_ ) {
495 if ( val != 0::L_ ) {
498 } else /* val==0 */ {
503 /* returns (# size :: Int#,
504 data :: ByteArray# #)
510 #endif /* SUPPORT_LONG_LONGS */
512 /* ToDo: this is shockingly inefficient */
517 bits8 [SIZEOF_MP_INT];
522 bits8 [SIZEOF_MP_INT];
527 bits8 [SIZEOF_MP_INT];
532 bits8 [SIZEOF_MP_INT];
537 #define FETCH_MP_TEMP(X) \
539 X = BaseReg + (OFFSET_StgRegTable_r ## X);
541 #define FETCH_MP_TEMP(X) /* Nothing */
544 #define GMP_TAKE2_RET1(name,mp_fun) \
549 FETCH_MP_TEMP(mp_tmp1); \
550 FETCH_MP_TEMP(mp_tmp2); \
551 FETCH_MP_TEMP(mp_result1) \
552 FETCH_MP_TEMP(mp_result2); \
554 /* call doYouWantToGC() */ \
555 MAYBE_GC(R2_PTR & R4_PTR, name); \
562 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
563 MP_INT__mp_size(mp_tmp1) = (s1); \
564 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
565 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
566 MP_INT__mp_size(mp_tmp2) = (s2); \
567 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
569 foreign "C" mpz_init(mp_result1 "ptr") []; \
571 /* Perform the operation */ \
572 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
574 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
575 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
578 #define GMP_TAKE1_RET1(name,mp_fun) \
583 FETCH_MP_TEMP(mp_tmp1); \
584 FETCH_MP_TEMP(mp_result1) \
586 /* call doYouWantToGC() */ \
587 MAYBE_GC(R2_PTR, name); \
592 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
593 MP_INT__mp_size(mp_tmp1) = (s1); \
594 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
596 foreign "C" mpz_init(mp_result1 "ptr") []; \
598 /* Perform the operation */ \
599 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr") []; \
601 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
602 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
605 #define GMP_TAKE2_RET2(name,mp_fun) \
610 FETCH_MP_TEMP(mp_tmp1); \
611 FETCH_MP_TEMP(mp_tmp2); \
612 FETCH_MP_TEMP(mp_result1) \
613 FETCH_MP_TEMP(mp_result2) \
615 /* call doYouWantToGC() */ \
616 MAYBE_GC(R2_PTR & R4_PTR, name); \
623 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
624 MP_INT__mp_size(mp_tmp1) = (s1); \
625 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
626 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
627 MP_INT__mp_size(mp_tmp2) = (s2); \
628 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
630 foreign "C" mpz_init(mp_result1 "ptr") []; \
631 foreign "C" mpz_init(mp_result2 "ptr") []; \
633 /* Perform the operation */ \
634 foreign "C" mp_fun(mp_result1 "ptr",mp_result2 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
636 RET_NPNP(TO_W_(MP_INT__mp_size(mp_result1)), \
637 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords, \
638 TO_W_(MP_INT__mp_size(mp_result2)), \
639 MP_INT__mp_d(mp_result2) - SIZEOF_StgArrWords); \
642 GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add)
643 GMP_TAKE2_RET1(minusIntegerzh_fast, mpz_sub)
644 GMP_TAKE2_RET1(timesIntegerzh_fast, mpz_mul)
645 GMP_TAKE2_RET1(gcdIntegerzh_fast, mpz_gcd)
646 GMP_TAKE2_RET1(quotIntegerzh_fast, mpz_tdiv_q)
647 GMP_TAKE2_RET1(remIntegerzh_fast, mpz_tdiv_r)
648 GMP_TAKE2_RET1(divExactIntegerzh_fast, mpz_divexact)
649 GMP_TAKE2_RET1(andIntegerzh_fast, mpz_and)
650 GMP_TAKE2_RET1(orIntegerzh_fast, mpz_ior)
651 GMP_TAKE2_RET1(xorIntegerzh_fast, mpz_xor)
652 GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com)
654 GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr)
655 GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr)
659 mp_tmp_w: W_; // NB. mp_tmp_w is really an here mp_limb_t
665 /* R1 = the first Int#; R2 = the second Int# */
667 FETCH_MP_TEMP(mp_tmp_w);
670 r = foreign "C" mpn_gcd_1(mp_tmp_w "ptr", 1, R2) [];
673 /* Result parked in R1, return via info-pointer at TOS */
674 jump %ENTRY_CODE(Sp(0));
680 /* R1 = s1; R2 = d1; R3 = the int */
681 R1 = foreign "C" mpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3) [];
683 /* Result parked in R1, return via info-pointer at TOS */
684 jump %ENTRY_CODE(Sp(0));
690 /* R1 = s1; R2 = d1; R3 = the int */
691 W_ usize, vsize, v_digit, u_digit;
697 // paraphrased from mpz_cmp_si() in the GMP sources
698 if (%gt(v_digit,0)) {
701 if (%lt(v_digit,0)) {
707 if (usize != vsize) {
709 jump %ENTRY_CODE(Sp(0));
714 jump %ENTRY_CODE(Sp(0));
717 u_digit = W_[BYTE_ARR_CTS(R2)];
719 if (u_digit == v_digit) {
721 jump %ENTRY_CODE(Sp(0));
724 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
730 jump %ENTRY_CODE(Sp(0));
735 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
736 W_ usize, vsize, size, up, vp;
739 // paraphrased from mpz_cmp() in the GMP sources
743 if (usize != vsize) {
745 jump %ENTRY_CODE(Sp(0));
750 jump %ENTRY_CODE(Sp(0));
753 if (%lt(usize,0)) { // NB. not <, which is unsigned
759 up = BYTE_ARR_CTS(R2);
760 vp = BYTE_ARR_CTS(R4);
762 cmp = foreign "C" mpn_cmp(up "ptr", vp "ptr", size) [];
764 if (cmp == 0 :: CInt) {
766 jump %ENTRY_CODE(Sp(0));
769 if (%lt(cmp,0 :: CInt) == %lt(usize,0)) {
774 /* 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 r = W_[R2 + SIZEOF_StgArrWords];
811 /* Result parked in R1, return via info-pointer at TOS */
813 jump %ENTRY_CODE(Sp(0));
820 FETCH_MP_TEMP(mp_tmp1);
821 FETCH_MP_TEMP(mp_tmp_w);
823 /* arguments: F1 = Float# */
826 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
828 /* Be prepared to tell Lennart-coded __decodeFloat
829 where mantissa._mp_d can be put (it does not care about the rest) */
830 p = Hp - SIZEOF_StgArrWords;
831 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
832 StgArrWords_words(p) = 1;
833 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
835 /* Perform the operation */
836 foreign "C" __decodeFloat(mp_tmp1 "ptr",mp_tmp_w "ptr" ,arg) [];
838 /* returns: (Int# (expn), Int#, ByteArray#) */
839 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
842 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
843 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
849 FETCH_MP_TEMP(mp_tmp1);
850 FETCH_MP_TEMP(mp_tmp_w);
852 /* arguments: D1 = Double# */
855 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
857 /* Be prepared to tell Lennart-coded __decodeDouble
858 where mantissa.d can be put (it does not care about the rest) */
859 p = Hp - ARR_SIZE + WDS(1);
860 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
861 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
862 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
864 /* Perform the operation */
865 foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg) [];
867 /* returns: (Int# (expn), Int#, ByteArray#) */
868 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
871 /* -----------------------------------------------------------------------------
872 * Concurrency primitives
873 * -------------------------------------------------------------------------- */
877 /* args: R1 = closure to spark */
879 MAYBE_GC(R1_PTR, forkzh_fast);
881 // create it right now, return ThreadID in R1
882 "ptr" R1 = foreign "C" createIOThread( MyCapability() "ptr",
883 RtsFlags_GcFlags_initialStkSize(RtsFlags),
885 foreign "C" scheduleThread(MyCapability() "ptr", R1 "ptr");
887 // switch at the earliest opportunity
888 CInt[context_switch] = 1 :: CInt;
895 jump stg_yield_noregs;
910 foreign "C" labelThread(R1 "ptr", R2 "ptr");
912 jump %ENTRY_CODE(Sp(0));
915 isCurrentThreadBoundzh_fast
919 r = foreign "C" isThreadBound(CurrentTSO) [];
924 /* -----------------------------------------------------------------------------
926 * -------------------------------------------------------------------------- */
930 #define IF_NOT_REG_R1(x)
933 #define IF_NOT_REG_R1(x) x
936 // Catch retry frame ------------------------------------------------------------
938 #define CATCH_RETRY_FRAME_ERROR(label) \
939 label { foreign "C" barf("catch_retry_frame incorrectly entered!"); }
941 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_0_ret)
942 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_1_ret)
943 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_2_ret)
944 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_3_ret)
945 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_4_ret)
946 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_5_ret)
947 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_6_ret)
948 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_7_ret)
950 #if MAX_VECTORED_RTN > 8
951 #error MAX_VECTORED_RTN has changed: please modify stg_catch_retry_frame too.
954 #if defined(PROFILING)
955 #define CATCH_RETRY_FRAME_BITMAP 7
956 #define CATCH_RETRY_FRAME_WORDS 6
958 #define CATCH_RETRY_FRAME_BITMAP 1
959 #define CATCH_RETRY_FRAME_WORDS 4
962 INFO_TABLE_RET(stg_catch_retry_frame,
963 CATCH_RETRY_FRAME_WORDS, CATCH_RETRY_FRAME_BITMAP,
965 stg_catch_retry_frame_0_ret,
966 stg_catch_retry_frame_1_ret,
967 stg_catch_retry_frame_2_ret,
968 stg_catch_retry_frame_3_ret,
969 stg_catch_retry_frame_4_ret,
970 stg_catch_retry_frame_5_ret,
971 stg_catch_retry_frame_6_ret,
972 stg_catch_retry_frame_7_ret)
974 W_ r, frame, trec, outer;
975 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
978 trec = StgTSO_trec(CurrentTSO);
979 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
980 r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
982 /* Succeeded (either first branch or second branch) */
983 StgTSO_trec(CurrentTSO) = outer;
984 Sp = Sp + SIZEOF_StgCatchRetryFrame;
985 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
986 jump %ENTRY_CODE(Sp(SP_OFF));
988 /* Did not commit: retry */
990 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
991 StgTSO_trec(CurrentTSO) = new_trec;
992 if (StgCatchRetryFrame_running_alt_code(frame)) {
993 R1 = StgCatchRetryFrame_alt_code(frame);
995 R1 = StgCatchRetryFrame_first_code(frame);
996 StgCatchRetryFrame_first_code_trec(frame) = new_trec;
999 jump RET_LBL(stg_ap_v);
1004 // Atomically frame -------------------------------------------------------------
1007 #define ATOMICALLY_FRAME_ERROR(label) \
1008 label { foreign "C" barf("atomically_frame incorrectly entered!"); }
1010 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_0_ret)
1011 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_1_ret)
1012 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_2_ret)
1013 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_3_ret)
1014 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_4_ret)
1015 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_5_ret)
1016 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_6_ret)
1017 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_7_ret)
1019 #if MAX_VECTORED_RTN > 8
1020 #error MAX_VECTORED_RTN has changed: please modify stg_atomically_frame too.
1023 #if defined(PROFILING)
1024 #define ATOMICALLY_FRAME_BITMAP 3
1025 #define ATOMICALLY_FRAME_WORDS 3
1027 #define ATOMICALLY_FRAME_BITMAP 0
1028 #define ATOMICALLY_FRAME_WORDS 1
1032 INFO_TABLE_RET(stg_atomically_frame,
1033 ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP,
1035 stg_atomically_frame_0_ret,
1036 stg_atomically_frame_1_ret,
1037 stg_atomically_frame_2_ret,
1038 stg_atomically_frame_3_ret,
1039 stg_atomically_frame_4_ret,
1040 stg_atomically_frame_5_ret,
1041 stg_atomically_frame_6_ret,
1042 stg_atomically_frame_7_ret)
1044 W_ frame, trec, valid;
1045 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1048 trec = StgTSO_trec(CurrentTSO);
1050 /* The TSO is not currently waiting: try to commit the transaction */
1051 valid = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr");
1053 /* Transaction was valid: commit succeeded */
1054 StgTSO_trec(CurrentTSO) = NO_TREC;
1055 Sp = Sp + SIZEOF_StgAtomicallyFrame;
1056 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1057 jump %ENTRY_CODE(Sp(SP_OFF));
1059 /* Transaction was not valid: try again */
1060 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr");
1061 StgTSO_trec(CurrentTSO) = trec;
1062 R1 = StgAtomicallyFrame_code(frame);
1064 jump RET_LBL(stg_ap_v);
1068 INFO_TABLE_RET(stg_atomically_waiting_frame,
1069 ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP,
1071 stg_atomically_frame_0_ret,
1072 stg_atomically_frame_1_ret,
1073 stg_atomically_frame_2_ret,
1074 stg_atomically_frame_3_ret,
1075 stg_atomically_frame_4_ret,
1076 stg_atomically_frame_5_ret,
1077 stg_atomically_frame_6_ret,
1078 stg_atomically_frame_7_ret)
1080 W_ frame, trec, valid;
1081 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1085 /* The TSO is currently waiting: should we stop waiting? */
1086 valid = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr");
1088 /* Previous attempt is still valid: no point trying again yet */
1089 IF_NOT_REG_R1(Sp_adj(-2);
1090 Sp(1) = stg_NO_FINALIZER_closure;
1091 Sp(0) = stg_ut_1_0_unreg_info;)
1092 jump stg_block_noregs;
1094 /* Previous attempt is no longer valid: try again */
1095 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr");
1096 StgTSO_trec(CurrentTSO) = trec;
1097 StgHeader_info(frame) = stg_atomically_frame_info;
1098 R1 = StgAtomicallyFrame_code(frame);
1100 jump RET_LBL(stg_ap_v);
1104 // STM catch frame --------------------------------------------------------------
1106 #define CATCH_STM_FRAME_ENTRY_TEMPLATE(label,ret) \
1109 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \
1110 Sp = Sp + SIZEOF_StgCatchSTMFrame; \
1111 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \
1121 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_0_ret,%RET_VEC(Sp(SP_OFF),0))
1122 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_1_ret,%RET_VEC(Sp(SP_OFF),1))
1123 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_2_ret,%RET_VEC(Sp(SP_OFF),2))
1124 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_3_ret,%RET_VEC(Sp(SP_OFF),3))
1125 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_4_ret,%RET_VEC(Sp(SP_OFF),4))
1126 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_5_ret,%RET_VEC(Sp(SP_OFF),5))
1127 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_6_ret,%RET_VEC(Sp(SP_OFF),6))
1128 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_7_ret,%RET_VEC(Sp(SP_OFF),7))
1130 #if MAX_VECTORED_RTN > 8
1131 #error MAX_VECTORED_RTN has changed: please modify stg_catch_stm_frame too.
1134 #if defined(PROFILING)
1135 #define CATCH_STM_FRAME_BITMAP 3
1136 #define CATCH_STM_FRAME_WORDS 3
1138 #define CATCH_STM_FRAME_BITMAP 0
1139 #define CATCH_STM_FRAME_WORDS 1
1142 /* Catch frames are very similar to update frames, but when entering
1143 * one we just pop the frame off the stack and perform the correct
1144 * kind of return to the activation record underneath us on the stack.
1147 INFO_TABLE_RET(stg_catch_stm_frame,
1148 CATCH_STM_FRAME_WORDS, CATCH_STM_FRAME_BITMAP,
1150 stg_catch_stm_frame_0_ret,
1151 stg_catch_stm_frame_1_ret,
1152 stg_catch_stm_frame_2_ret,
1153 stg_catch_stm_frame_3_ret,
1154 stg_catch_stm_frame_4_ret,
1155 stg_catch_stm_frame_5_ret,
1156 stg_catch_stm_frame_6_ret,
1157 stg_catch_stm_frame_7_ret)
1158 CATCH_STM_FRAME_ENTRY_TEMPLATE(,%ENTRY_CODE(Sp(SP_OFF)))
1161 // Primop definition ------------------------------------------------------------
1169 // stmStartTransaction may allocate
1170 MAYBE_GC (R1_PTR, atomicallyzh_fast);
1172 /* Args: R1 = m :: STM a */
1173 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);
1175 /* Set up the atomically frame */
1176 Sp = Sp - SIZEOF_StgAtomicallyFrame;
1179 SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
1180 StgAtomicallyFrame_code(frame) = R1;
1182 /* Start the memory transcation */
1183 old_trec = StgTSO_trec(CurrentTSO);
1184 ASSERT(old_trec == NO_TREC);
1185 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr");
1186 StgTSO_trec(CurrentTSO) = new_trec;
1188 /* Apply R1 to the realworld token */
1190 jump RET_LBL(stg_ap_v);
1198 /* Args: R1 :: STM a */
1199 /* Args: R2 :: Exception -> STM a */
1200 STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast);
1202 /* Set up the catch frame */
1203 Sp = Sp - SIZEOF_StgCatchSTMFrame;
1206 SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
1207 StgCatchSTMFrame_handler(frame) = R2;
1209 /* Apply R1 to the realworld token */
1211 jump RET_LBL(stg_ap_v);
1221 // stmStartTransaction may allocate
1222 MAYBE_GC (R1_PTR & R2_PTR, catchRetryzh_fast);
1224 /* Args: R1 :: STM a */
1225 /* Args: R2 :: STM a */
1226 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);
1228 /* Start a nested transaction within which to run the first code */
1229 trec = StgTSO_trec(CurrentTSO);
1230 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr");
1231 StgTSO_trec(CurrentTSO) = new_trec;
1233 /* Set up the catch-retry frame */
1234 Sp = Sp - SIZEOF_StgCatchRetryFrame;
1237 SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
1238 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1239 StgCatchRetryFrame_first_code(frame) = R1;
1240 StgCatchRetryFrame_alt_code(frame) = R2;
1241 StgCatchRetryFrame_first_code_trec(frame) = new_trec;
1243 /* Apply R1 to the realworld token */
1245 jump RET_LBL(stg_ap_v);
1257 MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate
1259 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
1261 trec = StgTSO_trec(CurrentTSO);
1262 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr");
1263 StgTSO_sp(CurrentTSO) = Sp;
1264 frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr");
1265 Sp = StgTSO_sp(CurrentTSO);
1268 if (frame_type == CATCH_RETRY_FRAME) {
1269 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
1270 ASSERT(outer != NO_TREC);
1271 if (!StgCatchRetryFrame_running_alt_code(frame)) {
1272 // Retry in the first code: try the alternative
1273 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr");
1274 StgTSO_trec(CurrentTSO) = trec;
1275 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
1276 R1 = StgCatchRetryFrame_alt_code(frame);
1278 jump RET_LBL(stg_ap_v);
1280 // Retry in the alternative code: propagate
1282 other_trec = StgCatchRetryFrame_first_code_trec(frame);
1283 r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", other_trec "ptr");
1285 r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr");
1287 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr");
1290 // Merge between siblings succeeded: commit it back to enclosing transaction
1291 // and then propagate the retry
1292 StgTSO_trec(CurrentTSO) = outer;
1293 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1294 goto retry_pop_stack;
1296 // Merge failed: we musn't propagate the retry. Try both paths again.
1297 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr");
1298 StgCatchRetryFrame_first_code_trec(frame) = trec;
1299 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1300 StgTSO_trec(CurrentTSO) = trec;
1301 R1 = StgCatchRetryFrame_first_code(frame);
1303 jump RET_LBL(stg_ap_v);
1308 // We've reached the ATOMICALLY_FRAME: attempt to wait
1309 ASSERT(frame_type == ATOMICALLY_FRAME);
1310 ASSERT(outer == NO_TREC);
1311 r = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr");
1313 // Transaction was valid: stmWait put us on the TVars' queues, we now block
1314 StgHeader_info(frame) = stg_atomically_waiting_frame_info;
1316 // Fix up the stack in the unregisterised case: the return convention is different.
1317 IF_NOT_REG_R1(Sp_adj(-2);
1318 Sp(1) = stg_NO_FINALIZER_closure;
1319 Sp(0) = stg_ut_1_0_unreg_info;)
1320 R3 = trec; // passing to stmWaitUnblock()
1321 jump stg_block_stmwait;
1323 // Transaction was not valid: retry immediately
1324 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr");
1325 StgTSO_trec(CurrentTSO) = trec;
1326 R1 = StgAtomicallyFrame_code(frame);
1329 jump RET_LBL(stg_ap_v);
1339 /* Args: R1 = initialisation value */
1341 MAYBE_GC (R1_PTR, newTVarzh_fast);
1343 "ptr" tv = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr");
1354 /* Args: R1 = TVar closure */
1356 MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
1357 trec = StgTSO_trec(CurrentTSO);
1359 "ptr" result = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") [];
1371 /* Args: R1 = TVar closure */
1372 /* R2 = New value */
1374 MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
1375 trec = StgTSO_trec(CurrentTSO);
1378 foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") [];
1380 jump %ENTRY_CODE(Sp(0));
1384 /* -----------------------------------------------------------------------------
1387 * take & putMVar work as follows. Firstly, an important invariant:
1389 * If the MVar is full, then the blocking queue contains only
1390 * threads blocked on putMVar, and if the MVar is empty then the
1391 * blocking queue contains only threads blocked on takeMVar.
1394 * MVar empty : then add ourselves to the blocking queue
1395 * MVar full : remove the value from the MVar, and
1396 * blocking queue empty : return
1397 * blocking queue non-empty : perform the first blocked putMVar
1398 * from the queue, and wake up the
1399 * thread (MVar is now full again)
1401 * putMVar is just the dual of the above algorithm.
1403 * How do we "perform a putMVar"? Well, we have to fiddle around with
1404 * the stack of the thread waiting to do the putMVar. See
1405 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1406 * the stack layout, and the PerformPut and PerformTake macros below.
1408 * It is important that a blocked take or put is woken up with the
1409 * take/put already performed, because otherwise there would be a
1410 * small window of vulnerability where the thread could receive an
1411 * exception and never perform its take or put, and we'd end up with a
1414 * -------------------------------------------------------------------------- */
1418 /* args: R1 = MVar closure */
1420 if (GET_INFO(R1) == stg_EMPTY_MVAR_info) {
1432 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
1434 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1435 SET_HDR(mvar,stg_EMPTY_MVAR_info,W_[CCCS]);
1436 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1437 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1438 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1443 /* If R1 isn't available, pass it on the stack */
1445 #define PerformTake(tso, value) \
1446 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1447 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1449 #define PerformTake(tso, value) \
1450 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1451 W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info;
1454 #define PerformPut(tso,lval) \
1455 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1456 lval = W_[StgTSO_sp(tso) - WDS(1)];
1460 W_ mvar, val, info, tso;
1462 /* args: R1 = MVar closure */
1466 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1468 info = GET_INFO(mvar);
1471 /* If the MVar is empty, put ourselves on its blocking queue,
1472 * and wait until we're woken up.
1474 if (info == stg_EMPTY_MVAR_info) {
1475 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1476 StgMVar_head(mvar) = CurrentTSO;
1478 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1480 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1481 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1482 StgTSO_block_info(CurrentTSO) = mvar;
1483 StgMVar_tail(mvar) = CurrentTSO;
1485 jump stg_block_takemvar;
1488 /* we got the value... */
1489 val = StgMVar_value(mvar);
1491 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1493 /* There are putMVar(s) waiting...
1494 * wake up the first thread on the queue
1496 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1498 /* actually perform the putMVar for the thread that we just woke up */
1499 tso = StgMVar_head(mvar);
1500 PerformPut(tso,StgMVar_value(mvar));
1502 #if defined(GRAN) || defined(PAR)
1503 /* ToDo: check 2nd arg (mvar) is right */
1504 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar) [];
1505 StgMVar_head(mvar) = tso;
1507 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr",
1508 StgMVar_head(mvar) "ptr") [];
1509 StgMVar_head(mvar) = tso;
1512 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1513 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1517 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1523 /* No further putMVars, MVar is now empty */
1524 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1527 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1529 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1539 W_ mvar, val, info, tso;
1541 /* args: R1 = MVar closure */
1546 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1548 info = GET_INFO(mvar);
1551 if (info == stg_EMPTY_MVAR_info) {
1553 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1555 /* HACK: we need a pointer to pass back,
1556 * so we abuse NO_FINALIZER_closure
1558 RET_NP(0, stg_NO_FINALIZER_closure);
1561 /* we got the value... */
1562 val = StgMVar_value(mvar);
1564 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1566 /* There are putMVar(s) waiting...
1567 * wake up the first thread on the queue
1569 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1571 /* actually perform the putMVar for the thread that we just woke up */
1572 tso = StgMVar_head(mvar);
1573 PerformPut(tso,StgMVar_value(mvar));
1575 #if defined(GRAN) || defined(PAR)
1576 /* ToDo: check 2nd arg (mvar) is right */
1577 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr") [];
1578 StgMVar_head(mvar) = tso;
1580 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr",
1581 StgMVar_head(mvar) "ptr") [];
1582 StgMVar_head(mvar) = tso;
1585 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1586 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1589 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1594 /* No further putMVars, MVar is now empty */
1595 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1597 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1599 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1611 /* args: R1 = MVar, R2 = value */
1615 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1617 info = GET_INFO(mvar);
1620 if (info == stg_FULL_MVAR_info) {
1621 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1622 StgMVar_head(mvar) = CurrentTSO;
1624 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1626 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1627 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1628 StgTSO_block_info(CurrentTSO) = mvar;
1629 StgMVar_tail(mvar) = CurrentTSO;
1631 jump stg_block_putmvar;
1634 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1636 /* There are takeMVar(s) waiting: wake up the first one
1638 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1640 /* actually perform the takeMVar */
1641 tso = StgMVar_head(mvar);
1642 PerformTake(tso, R2);
1644 #if defined(GRAN) || defined(PAR)
1645 /* ToDo: check 2nd arg (mvar) is right */
1646 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") [];
1647 StgMVar_head(mvar) = tso;
1649 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") [];
1650 StgMVar_head(mvar) = tso;
1653 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1654 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1658 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1660 jump %ENTRY_CODE(Sp(0));
1664 /* No further takes, the MVar is now full. */
1665 StgMVar_value(mvar) = R2;
1668 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1670 SET_INFO(mvar,stg_FULL_MVAR_info);
1672 jump %ENTRY_CODE(Sp(0));
1675 /* ToDo: yield afterward for better communication performance? */
1683 /* args: R1 = MVar, R2 = value */
1687 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1689 info = GET_INFO(mvar);
1692 if (info == stg_FULL_MVAR_info) {
1694 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1699 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1701 /* There are takeMVar(s) waiting: wake up the first one
1703 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1705 /* actually perform the takeMVar */
1706 tso = StgMVar_head(mvar);
1707 PerformTake(tso, R2);
1709 #if defined(GRAN) || defined(PAR)
1710 /* ToDo: check 2nd arg (mvar) is right */
1711 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") [];
1712 StgMVar_head(mvar) = tso;
1714 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") [];
1715 StgMVar_head(mvar) = tso;
1718 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1719 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1723 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info);
1725 jump %ENTRY_CODE(Sp(0));
1729 /* No further takes, the MVar is now full. */
1730 StgMVar_value(mvar) = R2;
1733 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info);
1735 SET_INFO(mvar,stg_FULL_MVAR_info);
1737 jump %ENTRY_CODE(Sp(0));
1740 /* ToDo: yield afterward for better communication performance? */
1744 /* -----------------------------------------------------------------------------
1745 Stable pointer primitives
1746 ------------------------------------------------------------------------- */
1748 makeStableNamezh_fast
1752 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1754 index = foreign "C" lookupStableName(R1 "ptr") [];
1756 /* Is there already a StableName for this heap object?
1757 * stable_ptr_table is a pointer to an array of snEntry structs.
1759 if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) {
1760 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1761 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1762 StgStableName_sn(sn_obj) = index;
1763 snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj;
1765 sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry);
1772 makeStablePtrzh_fast
1776 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1777 "ptr" sp = foreign "C" getStablePtr(R1 "ptr") [];
1781 deRefStablePtrzh_fast
1783 /* Args: R1 = the stable ptr */
1786 r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry);
1790 /* -----------------------------------------------------------------------------
1791 Bytecode object primitives
1792 ------------------------------------------------------------------------- */
1803 W_ bco, bitmap_arr, bytes, words;
1806 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1809 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R4_PTR&R6_PTR, newBCOzh_fast );
1811 bco = Hp - bytes + WDS(1);
1812 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1814 StgBCO_instrs(bco) = R1;
1815 StgBCO_literals(bco) = R2;
1816 StgBCO_ptrs(bco) = R3;
1817 StgBCO_itbls(bco) = R4;
1818 StgBCO_arity(bco) = HALF_W_(R5);
1819 StgBCO_size(bco) = HALF_W_(words);
1821 // Copy the arity/bitmap info into the BCO
1825 if (i < StgArrWords_words(bitmap_arr)) {
1826 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1837 // R1 = the BCO# for the AP
1841 // This function is *only* used to wrap zero-arity BCOs in an
1842 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1843 // saturated and always points directly to a FUN or BCO.
1844 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
1845 StgBCO_arity(R1) == HALF_W_(0));
1847 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1848 TICK_ALLOC_UP_THK(0, 0);
1849 CCCS_ALLOC(SIZEOF_StgAP);
1851 ap = Hp - SIZEOF_StgAP + WDS(1);
1852 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1854 StgAP_n_args(ap) = HALF_W_(0);
1860 /* -----------------------------------------------------------------------------
1861 Thread I/O blocking primitives
1862 -------------------------------------------------------------------------- */
1864 /* Add a thread to the end of the blocked queue. (C-- version of the C
1865 * macro in Schedule.h).
1867 #define APPEND_TO_BLOCKED_QUEUE(tso) \
1868 ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \
1869 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
1870 W_[blocked_queue_hd] = tso; \
1872 StgTSO_link(W_[blocked_queue_tl]) = tso; \
1874 W_[blocked_queue_tl] = tso;
1880 foreign "C" barf("waitRead# on threaded RTS");
1883 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1884 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1885 StgTSO_block_info(CurrentTSO) = R1;
1886 // No locking - we're not going to use this interface in the
1887 // threaded RTS anyway.
1888 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1889 jump stg_block_noregs;
1897 foreign "C" barf("waitWrite# on threaded RTS");
1900 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1901 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1902 StgTSO_block_info(CurrentTSO) = R1;
1903 // No locking - we're not going to use this interface in the
1904 // threaded RTS anyway.
1905 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1906 jump stg_block_noregs;
1911 STRING(stg_delayzh_malloc_str, "delayzh_fast")
1914 #ifdef mingw32_HOST_OS
1922 foreign "C" barf("delay# on threaded RTS");
1925 /* args: R1 (microsecond delay amount) */
1926 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1927 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
1929 #ifdef mingw32_HOST_OS
1931 /* could probably allocate this on the heap instead */
1932 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1933 stg_delayzh_malloc_str);
1934 reqID = foreign "C" addDelayRequest(R1);
1935 StgAsyncIOResult_reqID(ares) = reqID;
1936 StgAsyncIOResult_len(ares) = 0;
1937 StgAsyncIOResult_errCode(ares) = 0;
1938 StgTSO_block_info(CurrentTSO) = ares;
1940 /* Having all async-blocked threads reside on the blocked_queue
1941 * simplifies matters, so change the status to OnDoProc put the
1942 * delayed thread on the blocked_queue.
1944 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1945 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1946 jump stg_block_async_void;
1951 time = foreign "C" getourtimeofday();
1952 target = (R1 / (TICK_MILLISECS*1000)) + time;
1953 StgTSO_block_info(CurrentTSO) = target;
1955 /* Insert the new thread in the sleeping queue. */
1957 t = W_[sleeping_queue];
1959 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
1965 StgTSO_link(CurrentTSO) = t;
1967 W_[sleeping_queue] = CurrentTSO;
1969 StgTSO_link(prev) = CurrentTSO;
1971 jump stg_block_noregs;
1973 #endif /* !THREADED_RTS */
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;
2005 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
2012 foreign "C" barf("asyncWrite# on threaded RTS");
2015 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2016 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2017 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2019 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2020 stg_asyncWritezh_malloc_str);
2021 reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr");
2023 StgAsyncIOResult_reqID(ares) = reqID;
2024 StgAsyncIOResult_len(ares) = 0;
2025 StgAsyncIOResult_errCode(ares) = 0;
2026 StgTSO_block_info(CurrentTSO) = ares;
2027 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2028 jump stg_block_async;
2032 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
2039 foreign "C" barf("asyncDoProc# on threaded RTS");
2042 /* args: R1 = proc, R2 = param */
2043 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2044 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2046 /* could probably allocate this on the heap instead */
2047 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2048 stg_asyncDoProczh_malloc_str);
2049 reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr");
2050 StgAsyncIOResult_reqID(ares) = reqID;
2051 StgAsyncIOResult_len(ares) = 0;
2052 StgAsyncIOResult_errCode(ares) = 0;
2053 StgTSO_block_info(CurrentTSO) = ares;
2054 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2055 jump stg_block_async;
2060 /* -----------------------------------------------------------------------------
2063 classes CCallable and CReturnable don't really exist, but the
2064 compiler insists on generating dictionaries containing references
2065 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
2066 for these. Some C compilers can't cope with zero-length static arrays,
2067 so we have to make these one element long.
2068 --------------------------------------------------------------------------- */
2071 GHC_ZCCCallable_static_info: W_ 0;
2075 GHC_ZCCReturnable_static_info: W_ 0;