1 /* -----------------------------------------------------------------------------
3 * (c) The GHC Team, 1998-2004
5 * Out-of-line primitive operations
7 * This file contains the implementations of all the primitive
8 * operations ("primops") which are not expanded inline. See
9 * ghc/compiler/prelude/primops.txt.pp for a list of all the primops;
10 * this file contains code for most of those with the attribute
13 * Entry convention: the entry convention for a primop is that all the
14 * args are in Stg registers (R1, R2, etc.). This is to make writing
15 * the primops easier. (see compiler/codeGen/CgCallConv.hs).
17 * Return convention: results from a primop are generally returned
18 * using the ordinary unboxed tuple return convention. The C-- parser
19 * implements the RET_xxxx() macros to perform unboxed-tuple returns
20 * based on the prevailing return convention.
22 * This file is written in a subset of C--, extended with various
23 * features specific to GHC. It is compiled by GHC directly. For the
24 * syntax of .cmm files, see the parser in ghc/compiler/cmm/CmmParse.y.
26 * ---------------------------------------------------------------------------*/
30 /*-----------------------------------------------------------------------------
33 Basically just new*Array - the others are all inline macros.
35 The size arg is always passed in R1, and the result returned in R1.
37 The slow entry point is for returning from a heap check, the saved
38 size argument must be re-loaded from the stack.
39 -------------------------------------------------------------------------- */
41 /* for objects that are *less* than the size of a word, make sure we
42 * round up to the nearest word for the size of the array.
47 W_ words, payload_words, n, p;
48 MAYBE_GC(NO_PTRS,newByteArrayzh_fast);
50 payload_words = ROUNDUP_BYTES_TO_WDS(n);
51 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
52 "ptr" p = foreign "C" allocateLocal(BaseReg "ptr",words);
53 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
54 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
55 StgArrWords_words(p) = payload_words;
59 newPinnedByteArrayzh_fast
61 W_ words, payload_words, n, p;
63 MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast);
65 payload_words = ROUNDUP_BYTES_TO_WDS(n);
67 // We want an 8-byte aligned array. allocatePinned() gives us
68 // 8-byte aligned memory by default, but we want to align the
69 // *goods* inside the ArrWords object, so we have to check the
70 // size of the ArrWords header and adjust our size accordingly.
71 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
72 if ((SIZEOF_StgArrWords & 7) != 0) {
76 "ptr" p = foreign "C" allocatePinned(words);
77 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
79 // Again, if the ArrWords header isn't a multiple of 8 bytes, we
80 // have to push the object forward one word so that the goods
81 // fall on an 8-byte boundary.
82 if ((SIZEOF_StgArrWords & 7) != 0) {
86 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
87 StgArrWords_words(p) = payload_words;
93 W_ words, n, init, arr, p;
94 /* Args: R1 = words, R2 = initialisation value */
97 MAYBE_GC(R2_PTR,newArrayzh_fast);
99 words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
100 "ptr" arr = foreign "C" allocateLocal(BaseReg "ptr",words);
101 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
103 SET_HDR(arr, stg_MUT_ARR_PTRS_info, W_[CCCS]);
104 StgMutArrPtrs_ptrs(arr) = n;
106 // Initialise all elements of the the array with the value in R2
108 p = arr + SIZEOF_StgMutArrPtrs;
110 if (p < arr + WDS(words)) {
119 unsafeThawArrayzh_fast
121 // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
123 // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
124 // normally doesn't. However, when we freeze a MUT_ARR_PTRS, we leave
125 // it on the mutable list for the GC to remove (removing something from
126 // the mutable list is not easy, because the mut_list is only singly-linked).
128 // So that we can tell whether a MUT_ARR_PTRS_FROZEN is on the mutable list,
129 // when we freeze it we set the info ptr to be MUT_ARR_PTRS_FROZEN0 to indicate
130 // that it is still on the mutable list.
132 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
133 // either it is on a mut_list, or it isn't. We adopt the convention that
134 // the mut_link field is NULL if it isn't on a mut_list, and the GC
135 // maintains this invariant.
137 if (%INFO_TYPE(%GET_STD_INFO(R1)) != HALF_W_(MUT_ARR_PTRS_FROZEN0)) {
138 foreign "C" recordMutableLock(R1 "ptr");
141 SET_INFO(R1,stg_MUT_ARR_PTRS_info);
146 /* -----------------------------------------------------------------------------
148 -------------------------------------------------------------------------- */
153 /* Args: R1 = initialisation value */
155 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);
157 mv = Hp - SIZEOF_StgMutVar + WDS(1);
158 SET_HDR(mv,stg_MUT_VAR_info,W_[CCCS]);
159 StgMutVar_var(mv) = R1;
164 atomicModifyMutVarzh_fast
167 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
169 /* If x is the current contents of the MutVar#, then
170 We want to make the new contents point to
174 and the return value is
178 obviously we can share (f x).
180 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
181 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
182 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
186 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
187 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
189 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
190 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
194 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
195 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
197 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
198 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
201 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
203 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);
206 foreign "C" ACQUIRE_LOCK(sm_mutex "ptr");
209 x = StgMutVar_var(R1);
211 TICK_ALLOC_THUNK_2();
212 CCCS_ALLOC(THUNK_2_SIZE);
213 z = Hp - THUNK_2_SIZE + WDS(1);
214 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
215 LDV_RECORD_CREATE(z);
216 StgThunk_payload(z,0) = R2;
217 StgThunk_payload(z,1) = x;
219 TICK_ALLOC_THUNK_1();
220 CCCS_ALLOC(THUNK_1_SIZE);
221 y = z - THUNK_1_SIZE;
222 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
223 LDV_RECORD_CREATE(y);
224 StgThunk_payload(y,0) = z;
226 StgMutVar_var(R1) = y;
228 TICK_ALLOC_THUNK_1();
229 CCCS_ALLOC(THUNK_1_SIZE);
230 r = y - THUNK_1_SIZE;
231 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
232 LDV_RECORD_CREATE(r);
233 StgThunk_payload(r,0) = z;
236 foreign "C" RELEASE_LOCK(sm_mutex "ptr");
242 /* -----------------------------------------------------------------------------
243 Foreign Object Primitives
244 -------------------------------------------------------------------------- */
248 /* R1 = ptr to foreign object,
252 ALLOC_PRIM( SIZEOF_StgForeignObj, NO_PTRS, mkForeignObjzh_fast);
254 result = Hp - SIZEOF_StgForeignObj + WDS(1);
255 SET_HDR(result,stg_FOREIGN_info,W_[CCCS]);
256 StgForeignObj_data(result) = R1;
258 /* returns (# s#, ForeignObj# #) */
262 /* -----------------------------------------------------------------------------
263 Weak Pointer Primitives
264 -------------------------------------------------------------------------- */
266 STRING(stg_weak_msg,"New weak pointer at %p\n")
272 R3 = finalizer (or NULL)
277 R3 = stg_NO_FINALIZER_closure;
280 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
282 w = Hp - SIZEOF_StgWeak + WDS(1);
283 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
286 StgWeak_value(w) = R2;
287 StgWeak_finalizer(w) = R3;
289 StgWeak_link(w) = W_[weak_ptr_list];
290 W_[weak_ptr_list] = w;
292 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w));
307 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
308 RET_NP(0,stg_NO_FINALIZER_closure);
314 // A weak pointer is inherently used, so we do not need to call
315 // LDV_recordDead_FILL_SLOP_DYNAMIC():
316 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
317 // or, LDV_recordDead():
318 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
319 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
320 // large as weak pointers, so there is no need to fill the slop, either.
321 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
325 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
327 SET_INFO(w,stg_DEAD_WEAK_info);
328 LDV_RECORD_CREATE(w);
330 f = StgWeak_finalizer(w);
331 StgDeadWeak_link(w) = StgWeak_link(w);
333 /* return the finalizer */
334 if (f == stg_NO_FINALIZER_closure) {
335 RET_NP(0,stg_NO_FINALIZER_closure);
347 if (GET_INFO(w) == stg_WEAK_info) {
349 val = StgWeak_value(w);
357 /* -----------------------------------------------------------------------------
358 Arbitrary-precision Integer operations.
360 There are some assumptions in this code that mp_limb_t == W_. This is
361 the case for all the platforms that GHC supports, currently.
362 -------------------------------------------------------------------------- */
366 /* arguments: R1 = Int# */
368 W_ val, s, p; /* to avoid aliasing */
371 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
373 p = Hp - SIZEOF_StgArrWords;
374 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
375 StgArrWords_words(p) = 1;
377 /* mpz_set_si is inlined here, makes things simpler */
390 /* returns (# size :: Int#,
399 /* arguments: R1 = Word# */
401 W_ val, s, p; /* to avoid aliasing */
405 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
407 p = Hp - SIZEOF_StgArrWords;
408 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
409 StgArrWords_words(p) = 1;
418 /* returns (# size :: Int#,
419 data :: ByteArray# #)
426 * 'long long' primops for converting to/from Integers.
429 #ifdef SUPPORT_LONG_LONGS
431 int64ToIntegerzh_fast
433 /* arguments: L1 = Int64# */
436 W_ hi, s, neg, words_needed, p;
441 if ( %ge(val,0x100000000::L_) || %le(val,-0x100000000::L_) ) {
444 // minimum is one word
448 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
449 NO_PTRS, int64ToIntegerzh_fast );
451 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
452 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
453 StgArrWords_words(p) = words_needed;
455 if ( %lt(val,0::L_) ) {
460 hi = TO_W_(val >> 32);
462 if ( words_needed == 2 ) {
467 if ( val != 0::L_ ) {
470 } else /* val==0 */ {
478 /* returns (# size :: Int#,
479 data :: ByteArray# #)
484 word64ToIntegerzh_fast
486 /* arguments: L1 = Word64# */
489 W_ hi, s, words_needed, p;
492 if ( val >= 0x100000000::L_ ) {
498 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
499 NO_PTRS, word64ToIntegerzh_fast );
501 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
502 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
503 StgArrWords_words(p) = words_needed;
505 hi = TO_W_(val >> 32);
506 if ( val >= 0x100000000::L_ ) {
511 if ( val != 0::L_ ) {
514 } else /* val==0 */ {
519 /* returns (# size :: Int#,
520 data :: ByteArray# #)
526 #endif /* SUPPORT_LONG_LONGS */
528 /* ToDo: this is shockingly inefficient */
532 bits8 [SIZEOF_MP_INT];
537 bits8 [SIZEOF_MP_INT];
542 bits8 [SIZEOF_MP_INT];
547 bits8 [SIZEOF_MP_INT];
550 #define GMP_TAKE2_RET1(name,mp_fun) \
556 /* call doYouWantToGC() */ \
557 MAYBE_GC(R2_PTR & R4_PTR, name); \
564 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
565 MP_INT__mp_size(mp_tmp1) = (s1); \
566 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
567 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
568 MP_INT__mp_size(mp_tmp2) = (s2); \
569 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
571 foreign "C" mpz_init(result1); \
573 /* Perform the operation */ \
574 foreign "C" mp_fun(result1,mp_tmp1,mp_tmp2); \
576 RET_NP(TO_W_(MP_INT__mp_size(result1)), \
577 MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \
580 #define GMP_TAKE1_RET1(name,mp_fun) \
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(result1); \
598 /* Perform the operation */ \
599 foreign "C" mp_fun(result1,mp_tmp1); \
601 RET_NP(TO_W_(MP_INT__mp_size(result1)), \
602 MP_INT__mp_d(result1) - SIZEOF_StgArrWords); \
605 #define GMP_TAKE2_RET2(name,mp_fun) \
611 /* call doYouWantToGC() */ \
612 MAYBE_GC(R2_PTR & R4_PTR, name); \
619 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
620 MP_INT__mp_size(mp_tmp1) = (s1); \
621 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
622 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
623 MP_INT__mp_size(mp_tmp2) = (s2); \
624 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
626 foreign "C" mpz_init(result1); \
627 foreign "C" mpz_init(result2); \
629 /* Perform the operation */ \
630 foreign "C" mp_fun(result1,result2,mp_tmp1,mp_tmp2); \
632 RET_NPNP(TO_W_(MP_INT__mp_size(result1)), \
633 MP_INT__mp_d(result1) - SIZEOF_StgArrWords, \
634 TO_W_(MP_INT__mp_size(result2)), \
635 MP_INT__mp_d(result2) - SIZEOF_StgArrWords); \
638 GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add)
639 GMP_TAKE2_RET1(minusIntegerzh_fast, mpz_sub)
640 GMP_TAKE2_RET1(timesIntegerzh_fast, mpz_mul)
641 GMP_TAKE2_RET1(gcdIntegerzh_fast, mpz_gcd)
642 GMP_TAKE2_RET1(quotIntegerzh_fast, mpz_tdiv_q)
643 GMP_TAKE2_RET1(remIntegerzh_fast, mpz_tdiv_r)
644 GMP_TAKE2_RET1(divExactIntegerzh_fast, mpz_divexact)
645 GMP_TAKE2_RET1(andIntegerzh_fast, mpz_and)
646 GMP_TAKE2_RET1(orIntegerzh_fast, mpz_ior)
647 GMP_TAKE2_RET1(xorIntegerzh_fast, mpz_xor)
648 GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com)
650 GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr)
651 GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr)
654 aa: W_; // NB. aa is really an mp_limb_t
659 /* R1 = the first Int#; R2 = the second Int# */
663 r = foreign "C" mpn_gcd_1(aa, 1, R2);
666 /* Result parked in R1, return via info-pointer at TOS */
667 jump %ENTRY_CODE(Sp(0));
673 /* R1 = s1; R2 = d1; R3 = the int */
674 R1 = foreign "C" mpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3);
676 /* Result parked in R1, return via info-pointer at TOS */
677 jump %ENTRY_CODE(Sp(0));
683 /* R1 = s1; R2 = d1; R3 = the int */
684 W_ usize, vsize, v_digit, u_digit;
690 // paraphrased from mpz_cmp_si() in the GMP sources
691 if (%gt(v_digit,0)) {
694 if (%lt(v_digit,0)) {
700 if (usize != vsize) {
702 jump %ENTRY_CODE(Sp(0));
707 jump %ENTRY_CODE(Sp(0));
710 u_digit = W_[BYTE_ARR_CTS(R2)];
712 if (u_digit == v_digit) {
714 jump %ENTRY_CODE(Sp(0));
717 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
723 jump %ENTRY_CODE(Sp(0));
728 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
729 W_ usize, vsize, size, up, vp;
732 // paraphrased from mpz_cmp() in the GMP sources
736 if (usize != vsize) {
738 jump %ENTRY_CODE(Sp(0));
743 jump %ENTRY_CODE(Sp(0));
746 if (%lt(usize,0)) { // NB. not <, which is unsigned
752 up = BYTE_ARR_CTS(R2);
753 vp = BYTE_ARR_CTS(R4);
755 cmp = foreign "C" mpn_cmp(up "ptr", vp "ptr", size);
757 if (cmp == 0 :: CInt) {
759 jump %ENTRY_CODE(Sp(0));
762 if (%lt(cmp,0 :: CInt) == %lt(usize,0)) {
767 /* Result parked in R1, return via info-pointer at TOS */
768 jump %ENTRY_CODE(Sp(0));
780 r = W_[R2 + SIZEOF_StgArrWords];
785 /* Result parked in R1, return via info-pointer at TOS */
787 jump %ENTRY_CODE(Sp(0));
799 r = W_[R2 + SIZEOF_StgArrWords];
804 /* Result parked in R1, return via info-pointer at TOS */
806 jump %ENTRY_CODE(Sp(0));
818 /* arguments: F1 = Float# */
821 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
823 /* Be prepared to tell Lennart-coded __decodeFloat
824 where mantissa._mp_d can be put (it does not care about the rest) */
825 p = Hp - SIZEOF_StgArrWords;
826 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
827 StgArrWords_words(p) = 1;
828 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
830 /* Perform the operation */
831 foreign "C" __decodeFloat(mp_tmp1,exponent,arg);
833 /* returns: (Int# (expn), Int#, ByteArray#) */
834 RET_NNP(W_[exponent], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
837 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
838 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
845 /* arguments: D1 = Double# */
848 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
850 /* Be prepared to tell Lennart-coded __decodeDouble
851 where mantissa.d can be put (it does not care about the rest) */
852 p = Hp - ARR_SIZE + WDS(1);
853 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
854 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
855 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
857 /* Perform the operation */
858 foreign "C" __decodeDouble(mp_tmp1,exponent,arg);
860 /* returns: (Int# (expn), Int#, ByteArray#) */
861 RET_NNP(W_[exponent], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
864 /* -----------------------------------------------------------------------------
865 * Concurrency primitives
866 * -------------------------------------------------------------------------- */
870 /* args: R1 = closure to spark */
872 MAYBE_GC(R1_PTR, forkzh_fast);
874 // create it right now, return ThreadID in R1
875 "ptr" R1 = foreign "C" createIOThread( RtsFlags_GcFlags_initialStkSize(RtsFlags),
877 foreign "C" scheduleThread(R1 "ptr");
879 // switch at the earliest opportunity
880 CInt[context_switch] = 1 :: CInt;
887 jump stg_yield_noregs;
902 foreign "C" labelThread(R1 "ptr", R2 "ptr");
904 jump %ENTRY_CODE(Sp(0));
907 isCurrentThreadBoundzh_fast
911 r = foreign "C" isThreadBound(CurrentTSO);
916 /* -----------------------------------------------------------------------------
918 * -------------------------------------------------------------------------- */
922 #define IF_NOT_REG_R1(x)
925 #define IF_NOT_REG_R1(x) x
928 // Catch retry frame ------------------------------------------------------------
930 #define CATCH_RETRY_FRAME_ERROR(label) \
931 label { foreign "C" barf("catch_retry_frame incorrectly entered!"); }
933 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_0_ret)
934 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_1_ret)
935 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_2_ret)
936 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_3_ret)
937 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_4_ret)
938 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_5_ret)
939 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_6_ret)
940 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_7_ret)
942 #if MAX_VECTORED_RTN > 8
943 #error MAX_VECTORED_RTN has changed: please modify stg_catch_retry_frame too.
946 #if defined(PROFILING)
947 #define CATCH_RETRY_FRAME_BITMAP 7
948 #define CATCH_RETRY_FRAME_WORDS 6
950 #define CATCH_RETRY_FRAME_BITMAP 1
951 #define CATCH_RETRY_FRAME_WORDS 4
954 INFO_TABLE_RET(stg_catch_retry_frame,
955 CATCH_RETRY_FRAME_WORDS, CATCH_RETRY_FRAME_BITMAP,
957 stg_catch_retry_frame_0_ret,
958 stg_catch_retry_frame_1_ret,
959 stg_catch_retry_frame_2_ret,
960 stg_catch_retry_frame_3_ret,
961 stg_catch_retry_frame_4_ret,
962 stg_catch_retry_frame_5_ret,
963 stg_catch_retry_frame_6_ret,
964 stg_catch_retry_frame_7_ret)
966 W_ r, frame, trec, outer;
967 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
970 trec = StgTSO_trec(CurrentTSO);
971 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr");
972 r = foreign "C" stmCommitTransaction(trec "ptr");
974 /* Succeeded (either first branch or second branch) */
975 StgTSO_trec(CurrentTSO) = outer;
976 Sp = Sp + SIZEOF_StgCatchRetryFrame;
977 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
978 jump %ENTRY_CODE(Sp(SP_OFF));
980 /* Did not commit: retry */
982 "ptr" new_trec = foreign "C" stmStartTransaction(outer "ptr");
983 StgTSO_trec(CurrentTSO) = new_trec;
984 if (StgCatchRetryFrame_running_alt_code(frame)) {
985 R1 = StgCatchRetryFrame_alt_code(frame);
987 R1 = StgCatchRetryFrame_first_code(frame);
988 StgCatchRetryFrame_first_code_trec(frame) = new_trec;
991 jump RET_LBL(stg_ap_v);
996 // Atomically frame -------------------------------------------------------------
999 #define ATOMICALLY_FRAME_ERROR(label) \
1000 label { foreign "C" barf("atomically_frame incorrectly entered!"); }
1002 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_0_ret)
1003 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_1_ret)
1004 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_2_ret)
1005 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_3_ret)
1006 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_4_ret)
1007 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_5_ret)
1008 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_6_ret)
1009 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_7_ret)
1011 #if MAX_VECTORED_RTN > 8
1012 #error MAX_VECTORED_RTN has changed: please modify stg_atomically_frame too.
1015 #if defined(PROFILING)
1016 #define ATOMICALLY_FRAME_BITMAP 7
1017 #define ATOMICALLY_FRAME_WORDS 4
1019 #define ATOMICALLY_FRAME_BITMAP 1
1020 #define ATOMICALLY_FRAME_WORDS 2
1024 INFO_TABLE_RET(stg_atomically_frame,
1025 ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP,
1027 stg_atomically_frame_0_ret,
1028 stg_atomically_frame_1_ret,
1029 stg_atomically_frame_2_ret,
1030 stg_atomically_frame_3_ret,
1031 stg_atomically_frame_4_ret,
1032 stg_atomically_frame_5_ret,
1033 stg_atomically_frame_6_ret,
1034 stg_atomically_frame_7_ret)
1036 W_ frame, trec, valid;
1037 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1040 trec = StgTSO_trec(CurrentTSO);
1041 if (StgAtomicallyFrame_waiting(frame)) {
1042 /* The TSO is currently waiting: should we stop waiting? */
1043 valid = foreign "C" stmReWait(CurrentTSO "ptr");
1045 /* Previous attempt is still valid: no point trying again yet */
1046 IF_NOT_REG_R1(Sp_adj(-2);
1047 Sp(1) = stg_NO_FINALIZER_closure;
1048 Sp(0) = stg_ut_1_0_unreg_info;)
1049 jump stg_block_noregs;
1051 /* Previous attempt is no longer valid: try again */
1052 "ptr" trec = foreign "C" stmStartTransaction(NO_TREC "ptr");
1053 StgTSO_trec(CurrentTSO) = trec;
1054 StgAtomicallyFrame_waiting(frame) = 0 :: CInt; /* false; */
1055 R1 = StgAtomicallyFrame_code(frame);
1057 jump RET_LBL(stg_ap_v);
1060 /* The TSO is not currently waiting: try to commit the transaction */
1061 valid = foreign "C" stmCommitTransaction(trec "ptr");
1063 /* Transaction was valid: commit succeeded */
1064 StgTSO_trec(CurrentTSO) = NO_TREC;
1065 Sp = Sp + SIZEOF_StgAtomicallyFrame;
1066 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1067 jump %ENTRY_CODE(Sp(SP_OFF));
1069 /* Transaction was not valid: try again */
1070 "ptr" trec = foreign "C" stmStartTransaction(NO_TREC "ptr");
1071 StgTSO_trec(CurrentTSO) = trec;
1072 R1 = StgAtomicallyFrame_code(frame);
1074 jump RET_LBL(stg_ap_v);
1080 // STM catch frame --------------------------------------------------------------
1082 #define CATCH_STM_FRAME_ENTRY_TEMPLATE(label,ret) \
1085 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \
1086 Sp = Sp + SIZEOF_StgCatchSTMFrame; \
1087 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \
1097 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_0_ret,%RET_VEC(Sp(SP_OFF),0))
1098 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_1_ret,%RET_VEC(Sp(SP_OFF),1))
1099 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_2_ret,%RET_VEC(Sp(SP_OFF),2))
1100 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_3_ret,%RET_VEC(Sp(SP_OFF),3))
1101 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_4_ret,%RET_VEC(Sp(SP_OFF),4))
1102 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_5_ret,%RET_VEC(Sp(SP_OFF),5))
1103 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_6_ret,%RET_VEC(Sp(SP_OFF),6))
1104 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_7_ret,%RET_VEC(Sp(SP_OFF),7))
1106 #if MAX_VECTORED_RTN > 8
1107 #error MAX_VECTORED_RTN has changed: please modify stg_catch_stm_frame too.
1110 #if defined(PROFILING)
1111 #define CATCH_STM_FRAME_BITMAP 3
1112 #define CATCH_STM_FRAME_WORDS 3
1114 #define CATCH_STM_FRAME_BITMAP 0
1115 #define CATCH_STM_FRAME_WORDS 1
1118 /* Catch frames are very similar to update frames, but when entering
1119 * one we just pop the frame off the stack and perform the correct
1120 * kind of return to the activation record underneath us on the stack.
1123 INFO_TABLE_RET(stg_catch_stm_frame,
1124 CATCH_STM_FRAME_WORDS, CATCH_STM_FRAME_BITMAP,
1126 stg_catch_stm_frame_0_ret,
1127 stg_catch_stm_frame_1_ret,
1128 stg_catch_stm_frame_2_ret,
1129 stg_catch_stm_frame_3_ret,
1130 stg_catch_stm_frame_4_ret,
1131 stg_catch_stm_frame_5_ret,
1132 stg_catch_stm_frame_6_ret,
1133 stg_catch_stm_frame_7_ret)
1134 CATCH_STM_FRAME_ENTRY_TEMPLATE(,%ENTRY_CODE(Sp(SP_OFF)))
1137 // Primop definition ------------------------------------------------------------
1145 /* Args: R1 = m :: STM a */
1146 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);
1148 /* Set up the atomically frame */
1149 Sp = Sp - SIZEOF_StgAtomicallyFrame;
1152 SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
1153 StgAtomicallyFrame_waiting(frame) = 0 :: CInt; // False
1154 StgAtomicallyFrame_code(frame) = R1;
1156 /* Start the memory transcation */
1157 old_trec = StgTSO_trec(CurrentTSO);
1158 "ptr" new_trec = foreign "C" stmStartTransaction(old_trec "ptr");
1159 StgTSO_trec(CurrentTSO) = new_trec;
1161 /* Apply R1 to the realworld token */
1163 jump RET_LBL(stg_ap_v);
1171 /* Args: R1 :: STM a */
1172 /* Args: R2 :: Exception -> STM a */
1173 STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast);
1175 /* Set up the catch frame */
1176 Sp = Sp - SIZEOF_StgCatchSTMFrame;
1179 SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
1180 StgCatchSTMFrame_handler(frame) = R2;
1182 /* Apply R1 to the realworld token */
1184 jump RET_LBL(stg_ap_v);
1194 /* Args: R1 :: STM a */
1195 /* Args: R2 :: STM a */
1196 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);
1198 /* Start a nested transaction within which to run the first code */
1199 trec = StgTSO_trec(CurrentTSO);
1200 "ptr" new_trec = foreign "C" stmStartTransaction(trec "ptr");
1201 StgTSO_trec(CurrentTSO) = new_trec;
1203 /* Set up the catch-retry frame */
1204 Sp = Sp - SIZEOF_StgCatchRetryFrame;
1207 SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
1208 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1209 StgCatchRetryFrame_first_code(frame) = R1;
1210 StgCatchRetryFrame_alt_code(frame) = R2;
1211 StgCatchRetryFrame_first_code_trec(frame) = new_trec;
1213 /* Apply R1 to the realworld token */
1215 jump RET_LBL(stg_ap_v);
1227 MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate
1229 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
1231 trec = StgTSO_trec(CurrentTSO);
1232 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr");
1233 StgTSO_sp(CurrentTSO) = Sp;
1234 frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr");
1235 Sp = StgTSO_sp(CurrentTSO);
1238 if (frame_type == CATCH_RETRY_FRAME) {
1239 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
1240 ASSERT(outer != NO_TREC);
1241 if (!StgCatchRetryFrame_running_alt_code(frame)) {
1242 // Retry in the first code: try the alternative
1243 "ptr" trec = foreign "C" stmStartTransaction(outer "ptr");
1244 StgTSO_trec(CurrentTSO) = trec;
1245 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
1246 R1 = StgCatchRetryFrame_alt_code(frame);
1248 jump RET_LBL(stg_ap_v);
1250 // Retry in the alternative code: propagate
1252 other_trec = StgCatchRetryFrame_first_code_trec(frame);
1253 r = foreign "C" stmMergeForWaiting(trec "ptr", other_trec "ptr");
1255 r = foreign "C" stmCommitTransaction(trec "ptr");
1258 // Merge between siblings succeeded: commit it back to enclosing transaction
1259 // and then propagate the retry
1260 StgTSO_trec(CurrentTSO) = outer;
1261 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1262 goto retry_pop_stack;
1264 // Merge failed: we musn't propagate the retry. Try both paths again.
1265 "ptr" trec = foreign "C" stmStartTransaction(outer "ptr");
1266 StgCatchRetryFrame_first_code_trec(frame) = trec;
1267 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1268 StgTSO_trec(CurrentTSO) = trec;
1269 R1 = StgCatchRetryFrame_first_code(frame);
1271 jump RET_LBL(stg_ap_v);
1276 // We've reached the ATOMICALLY_FRAME: attempt to wait
1277 ASSERT(frame_type == ATOMICALLY_FRAME);
1278 ASSERT(outer == NO_TREC);
1279 r = foreign "C" stmWait(CurrentTSO "ptr", trec "ptr");
1281 // Transaction was valid: stmWait put us on the TVars' queues, we now block
1282 StgAtomicallyFrame_waiting(frame) = 1 :: CInt; // true
1284 // Fix up the stack in the unregisterised case: the return convention is different.
1285 IF_NOT_REG_R1(Sp_adj(-2);
1286 Sp(1) = stg_NO_FINALIZER_closure;
1287 Sp(0) = stg_ut_1_0_unreg_info;)
1288 jump stg_block_noregs;
1290 // Transaction was not valid: retry immediately
1291 "ptr" trec = foreign "C" stmStartTransaction(outer "ptr");
1292 StgTSO_trec(CurrentTSO) = trec;
1293 R1 = StgAtomicallyFrame_code(frame);
1296 jump RET_LBL(stg_ap_v);
1305 /* Args: R1 = initialisation value */
1307 ALLOC_PRIM( SIZEOF_StgTVar, R1_PTR, newTVarzh_fast);
1308 tv = Hp - SIZEOF_StgTVar + WDS(1);
1309 SET_HDR(tv,stg_TVAR_info,W_[CCCS]);
1310 StgTVar_current_value(tv) = R1;
1311 StgTVar_first_wait_queue_entry(tv) = stg_END_STM_WAIT_QUEUE_closure;
1323 /* Args: R1 = TVar closure */
1325 MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
1326 trec = StgTSO_trec(CurrentTSO);
1328 "ptr" result = foreign "C" stmReadTVar(trec "ptr", tvar "ptr");
1340 /* Args: R1 = TVar closure */
1341 /* R2 = New value */
1343 MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
1344 trec = StgTSO_trec(CurrentTSO);
1347 foreign "C" stmWriteTVar(trec "ptr", tvar "ptr", new_value "ptr");
1349 jump %ENTRY_CODE(Sp(0));
1353 /* -----------------------------------------------------------------------------
1356 * take & putMVar work as follows. Firstly, an important invariant:
1358 * If the MVar is full, then the blocking queue contains only
1359 * threads blocked on putMVar, and if the MVar is empty then the
1360 * blocking queue contains only threads blocked on takeMVar.
1363 * MVar empty : then add ourselves to the blocking queue
1364 * MVar full : remove the value from the MVar, and
1365 * blocking queue empty : return
1366 * blocking queue non-empty : perform the first blocked putMVar
1367 * from the queue, and wake up the
1368 * thread (MVar is now full again)
1370 * putMVar is just the dual of the above algorithm.
1372 * How do we "perform a putMVar"? Well, we have to fiddle around with
1373 * the stack of the thread waiting to do the putMVar. See
1374 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1375 * the stack layout, and the PerformPut and PerformTake macros below.
1377 * It is important that a blocked take or put is woken up with the
1378 * take/put already performed, because otherwise there would be a
1379 * small window of vulnerability where the thread could receive an
1380 * exception and never perform its take or put, and we'd end up with a
1383 * -------------------------------------------------------------------------- */
1387 /* args: R1 = MVar closure */
1389 if (GET_INFO(R1) == stg_EMPTY_MVAR_info) {
1401 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
1403 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1404 SET_HDR(mvar,stg_EMPTY_MVAR_info,W_[CCCS]);
1405 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1406 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1407 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1412 /* If R1 isn't available, pass it on the stack */
1414 #define PerformTake(tso, value) \
1415 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1416 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1418 #define PerformTake(tso, value) \
1419 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1420 W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info;
1423 #define PerformPut(tso,lval) \
1424 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1425 lval = W_[StgTSO_sp(tso) - WDS(1)];
1430 W_ mvar, val, info, tso;
1432 /* args: R1 = MVar closure */
1436 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1438 info = GET_INFO(mvar);
1441 /* If the MVar is empty, put ourselves on its blocking queue,
1442 * and wait until we're woken up.
1444 if (info == stg_EMPTY_MVAR_info) {
1445 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1446 StgMVar_head(mvar) = CurrentTSO;
1448 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1450 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1451 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1452 StgTSO_block_info(CurrentTSO) = mvar;
1453 StgMVar_tail(mvar) = CurrentTSO;
1456 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1459 jump stg_block_takemvar;
1462 /* we got the value... */
1463 val = StgMVar_value(mvar);
1465 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1467 /* There are putMVar(s) waiting...
1468 * wake up the first thread on the queue
1470 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1472 /* actually perform the putMVar for the thread that we just woke up */
1473 tso = StgMVar_head(mvar);
1474 PerformPut(tso,StgMVar_value(mvar));
1476 #if defined(GRAN) || defined(PAR)
1477 /* ToDo: check 2nd arg (mvar) is right */
1478 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar);
1479 StgMVar_head(mvar) = tso;
1481 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1482 StgMVar_head(mvar) = tso;
1484 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1485 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1489 SET_INFO(mvar,stg_FULL_MVAR_info);
1496 /* No further putMVars, MVar is now empty */
1497 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1499 /* unlocks the closure in the SMP case */
1500 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1509 W_ mvar, val, info, tso;
1511 /* args: R1 = MVar closure */
1516 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1518 info = GET_INFO(mvar);
1521 if (info == stg_EMPTY_MVAR_info) {
1523 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1525 /* HACK: we need a pointer to pass back,
1526 * so we abuse NO_FINALIZER_closure
1528 RET_NP(0, stg_NO_FINALIZER_closure);
1531 /* we got the value... */
1532 val = StgMVar_value(mvar);
1534 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1535 /* There are putMVar(s) waiting...
1536 * wake up the first thread on the queue
1538 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1540 /* actually perform the putMVar for the thread that we just woke up */
1541 tso = StgMVar_head(mvar);
1542 PerformPut(tso,StgMVar_value(mvar));
1544 #if defined(GRAN) || defined(PAR)
1545 /* ToDo: check 2nd arg (mvar) is right */
1546 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr");
1547 StgMVar_head(mvar) = tso;
1549 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1550 StgMVar_head(mvar) = tso;
1553 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1554 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1557 SET_INFO(mvar,stg_FULL_MVAR_info);
1562 /* No further putMVars, MVar is now empty */
1563 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1564 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1575 /* args: R1 = MVar, R2 = value */
1579 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1581 info = GET_INFO(mvar);
1584 if (info == stg_FULL_MVAR_info) {
1585 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1586 StgMVar_head(mvar) = CurrentTSO;
1588 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1590 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1591 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1592 StgTSO_block_info(CurrentTSO) = mvar;
1593 StgMVar_tail(mvar) = CurrentTSO;
1596 SET_INFO(mvar,stg_FULL_MVAR_info);
1598 jump stg_block_putmvar;
1601 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1602 /* There are takeMVar(s) waiting: wake up the first one
1604 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1606 /* actually perform the takeMVar */
1607 tso = StgMVar_head(mvar);
1608 PerformTake(tso, R2);
1610 #if defined(GRAN) || defined(PAR)
1611 /* ToDo: check 2nd arg (mvar) is right */
1612 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1613 StgMVar_head(mvar) = tso;
1615 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1616 StgMVar_head(mvar) = tso;
1619 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1620 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1624 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1626 jump %ENTRY_CODE(Sp(0));
1630 /* No further takes, the MVar is now full. */
1631 StgMVar_value(mvar) = R2;
1632 /* unlocks the MVar in the SMP case */
1633 SET_INFO(mvar,stg_FULL_MVAR_info);
1635 jump %ENTRY_CODE(Sp(0));
1638 /* ToDo: yield afterward for better communication performance? */
1646 /* args: R1 = MVar, R2 = value */
1650 "ptr" info = foreign "C" lockClosure(mvar "ptr");
1652 info = GET_INFO(mvar);
1655 if (info == stg_FULL_MVAR_info) {
1657 SET_INFO(mvar,stg_FULL_MVAR_info);
1662 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1663 /* There are takeMVar(s) waiting: wake up the first one
1665 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1667 /* actually perform the takeMVar */
1668 tso = StgMVar_head(mvar);
1669 PerformTake(tso, R2);
1671 #if defined(GRAN) || defined(PAR)
1672 /* ToDo: check 2nd arg (mvar) is right */
1673 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr",mvar "ptr");
1674 StgMVar_head(mvar) = tso;
1676 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr");
1677 StgMVar_head(mvar) = tso;
1680 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1681 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1685 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1687 jump %ENTRY_CODE(Sp(0));
1691 /* No further takes, the MVar is now full. */
1692 StgMVar_value(mvar) = R2;
1693 /* unlocks the MVar in the SMP case */
1694 SET_INFO(mvar,stg_FULL_MVAR_info);
1696 jump %ENTRY_CODE(Sp(0));
1699 /* ToDo: yield afterward for better communication performance? */
1703 /* -----------------------------------------------------------------------------
1704 Stable pointer primitives
1705 ------------------------------------------------------------------------- */
1707 makeStableNamezh_fast
1711 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1713 index = foreign "C" lookupStableName(R1 "ptr");
1715 /* Is there already a StableName for this heap object?
1716 * stable_ptr_table is a pointer to an array of snEntry structs.
1718 if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) {
1719 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1720 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1721 StgStableName_sn(sn_obj) = index;
1722 snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj;
1724 sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry);
1731 makeStablePtrzh_fast
1735 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1736 "ptr" sp = foreign "C" getStablePtr(R1 "ptr");
1740 deRefStablePtrzh_fast
1742 /* Args: R1 = the stable ptr */
1745 r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry);
1749 /* -----------------------------------------------------------------------------
1750 Bytecode object primitives
1751 ------------------------------------------------------------------------- */
1762 W_ bco, bitmap_arr, bytes, words;
1765 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1768 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R4_PTR&R6_PTR, newBCOzh_fast );
1770 bco = Hp - bytes + WDS(1);
1771 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1773 StgBCO_instrs(bco) = R1;
1774 StgBCO_literals(bco) = R2;
1775 StgBCO_ptrs(bco) = R3;
1776 StgBCO_itbls(bco) = R4;
1777 StgBCO_arity(bco) = HALF_W_(R5);
1778 StgBCO_size(bco) = HALF_W_(words);
1780 // Copy the arity/bitmap info into the BCO
1784 if (i < StgArrWords_words(bitmap_arr)) {
1785 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1796 // R1 = the BCO# for the AP
1800 // This function is *only* used to wrap zero-arity BCOs in an
1801 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1802 // saturated and always points directly to a FUN or BCO.
1803 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
1804 StgBCO_arity(R1) == HALF_W_(0));
1806 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1807 TICK_ALLOC_UP_THK(0, 0);
1808 CCCS_ALLOC(SIZEOF_StgAP);
1810 ap = Hp - SIZEOF_StgAP + WDS(1);
1811 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1813 StgAP_n_args(ap) = HALF_W_(0);
1819 /* -----------------------------------------------------------------------------
1820 Thread I/O blocking primitives
1821 -------------------------------------------------------------------------- */
1823 /* Add a thread to the end of the blocked queue. (C-- version of the C
1824 * macro in Schedule.h).
1826 #define APPEND_TO_BLOCKED_QUEUE(tso) \
1827 ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \
1828 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
1829 W_[blocked_queue_hd] = tso; \
1831 StgTSO_link(W_[blocked_queue_tl]) = tso; \
1833 W_[blocked_queue_tl] = tso;
1839 foreign "C" barf("waitRead# on threaded RTS");
1842 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1843 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1844 StgTSO_block_info(CurrentTSO) = R1;
1845 // No locking - we're not going to use this interface in the
1846 // threaded RTS anyway.
1847 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1848 jump stg_block_noregs;
1855 foreign "C" barf("waitWrite# on threaded RTS");
1858 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1859 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1860 StgTSO_block_info(CurrentTSO) = R1;
1861 // No locking - we're not going to use this interface in the
1862 // threaded RTS anyway.
1863 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1864 jump stg_block_noregs;
1868 STRING(stg_delayzh_malloc_str, "delayzh_fast")
1871 #ifdef mingw32_HOST_OS
1879 foreign "C" barf("delay# on threaded RTS");
1882 /* args: R1 (microsecond delay amount) */
1883 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1884 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
1886 #ifdef mingw32_HOST_OS
1888 /* could probably allocate this on the heap instead */
1889 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1890 stg_delayzh_malloc_str);
1891 reqID = foreign "C" addDelayRequest(R1);
1892 StgAsyncIOResult_reqID(ares) = reqID;
1893 StgAsyncIOResult_len(ares) = 0;
1894 StgAsyncIOResult_errCode(ares) = 0;
1895 StgTSO_block_info(CurrentTSO) = ares;
1897 /* Having all async-blocked threads reside on the blocked_queue
1898 * simplifies matters, so change the status to OnDoProc put the
1899 * delayed thread on the blocked_queue.
1901 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1902 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1903 jump stg_block_async_void;
1908 time = foreign "C" getourtimeofday();
1909 target = (R1 / (TICK_MILLISECS*1000)) + time;
1910 StgTSO_block_info(CurrentTSO) = target;
1912 /* Insert the new thread in the sleeping queue. */
1914 t = W_[sleeping_queue];
1916 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
1922 StgTSO_link(CurrentTSO) = t;
1924 W_[sleeping_queue] = CurrentTSO;
1926 StgTSO_link(prev) = CurrentTSO;
1928 jump stg_block_noregs;
1933 #ifdef mingw32_HOST_OS
1934 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
1941 foreign "C" barf("asyncRead# on threaded RTS");
1944 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1945 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1946 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
1948 /* could probably allocate this on the heap instead */
1949 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1950 stg_asyncReadzh_malloc_str);
1951 reqID = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr");
1952 StgAsyncIOResult_reqID(ares) = reqID;
1953 StgAsyncIOResult_len(ares) = 0;
1954 StgAsyncIOResult_errCode(ares) = 0;
1955 StgTSO_block_info(CurrentTSO) = ares;
1956 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1957 jump stg_block_async;
1960 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
1967 foreign "C" barf("asyncWrite# on threaded RTS");
1970 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
1971 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1972 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
1974 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1975 stg_asyncWritezh_malloc_str);
1976 reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr");
1978 StgAsyncIOResult_reqID(ares) = reqID;
1979 StgAsyncIOResult_len(ares) = 0;
1980 StgAsyncIOResult_errCode(ares) = 0;
1981 StgTSO_block_info(CurrentTSO) = ares;
1982 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1983 jump stg_block_async;
1986 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
1992 /* args: R1 = proc, R2 = param */
1993 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
1994 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
1996 /* could probably allocate this on the heap instead */
1997 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
1998 stg_asyncDoProczh_malloc_str);
1999 reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr");
2000 StgAsyncIOResult_reqID(ares) = reqID;
2001 StgAsyncIOResult_len(ares) = 0;
2002 StgAsyncIOResult_errCode(ares) = 0;
2003 StgTSO_block_info(CurrentTSO) = ares;
2004 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2005 jump stg_block_async;
2009 /* -----------------------------------------------------------------------------
2012 classes CCallable and CReturnable don't really exist, but the
2013 compiler insists on generating dictionaries containing references
2014 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
2015 for these. Some C compilers can't cope with zero-length static arrays,
2016 so we have to make these one element long.
2017 --------------------------------------------------------------------------- */
2020 GHC_ZCCCallable_static_info: W_ 0;
2024 GHC_ZCCReturnable_static_info: W_ 0;