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) [R2];
101 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
103 SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_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_DIRTY_info);
141 recordMutable(R1, R1);
142 // must be done after SET_INFO, because it ASSERTs closure_MUTABLE()
145 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_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_DIRTY_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);
209 #if defined(THREADED_RTS)
210 foreign "C" ACQUIRE_LOCK(atomic_modify_mutvar_mutex "ptr") [R1,R2];
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;
231 foreign "C" dirty_MUT_VAR(BaseReg "ptr", R1 "ptr") [R1];
233 TICK_ALLOC_THUNK_1();
234 CCCS_ALLOC(THUNK_1_SIZE);
235 r = y - THUNK_1_SIZE;
236 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
237 LDV_RECORD_CREATE(r);
238 StgThunk_payload(r,0) = z;
240 #if defined(THREADED_RTS)
241 foreign "C" RELEASE_LOCK(atomic_modify_mutvar_mutex "ptr") [];
247 /* -----------------------------------------------------------------------------
248 Weak Pointer Primitives
249 -------------------------------------------------------------------------- */
251 STRING(stg_weak_msg,"New weak pointer at %p\n")
257 R3 = finalizer (or NULL)
262 R3 = stg_NO_FINALIZER_closure;
265 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
267 w = Hp - SIZEOF_StgWeak + WDS(1);
268 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
271 StgWeak_value(w) = R2;
272 StgWeak_finalizer(w) = R3;
274 StgWeak_link(w) = W_[weak_ptr_list];
275 W_[weak_ptr_list] = w;
277 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
292 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
293 RET_NP(0,stg_NO_FINALIZER_closure);
299 // A weak pointer is inherently used, so we do not need to call
300 // LDV_recordDead_FILL_SLOP_DYNAMIC():
301 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
302 // or, LDV_recordDead():
303 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
304 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
305 // large as weak pointers, so there is no need to fill the slop, either.
306 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
310 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
312 SET_INFO(w,stg_DEAD_WEAK_info);
313 LDV_RECORD_CREATE(w);
315 f = StgWeak_finalizer(w);
316 StgDeadWeak_link(w) = StgWeak_link(w);
318 /* return the finalizer */
319 if (f == stg_NO_FINALIZER_closure) {
320 RET_NP(0,stg_NO_FINALIZER_closure);
332 if (GET_INFO(w) == stg_WEAK_info) {
334 val = StgWeak_value(w);
342 /* -----------------------------------------------------------------------------
343 Arbitrary-precision Integer operations.
345 There are some assumptions in this code that mp_limb_t == W_. This is
346 the case for all the platforms that GHC supports, currently.
347 -------------------------------------------------------------------------- */
351 /* arguments: R1 = Int# */
353 W_ val, s, p; /* to avoid aliasing */
356 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
358 p = Hp - SIZEOF_StgArrWords;
359 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
360 StgArrWords_words(p) = 1;
362 /* mpz_set_si is inlined here, makes things simpler */
375 /* returns (# size :: Int#,
384 /* arguments: R1 = Word# */
386 W_ val, s, p; /* to avoid aliasing */
390 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
392 p = Hp - SIZEOF_StgArrWords;
393 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
394 StgArrWords_words(p) = 1;
403 /* returns (# size :: Int#,
404 data :: ByteArray# #)
411 * 'long long' primops for converting to/from Integers.
414 #ifdef SUPPORT_LONG_LONGS
416 int64ToIntegerzh_fast
418 /* arguments: L1 = Int64# */
421 W_ hi, lo, s, neg, words_needed, p;
426 hi = TO_W_(val >> 32);
429 if ( hi != 0 && hi != 0xFFFFFFFF ) {
432 // minimum is one word
436 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
437 NO_PTRS, int64ToIntegerzh_fast );
439 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
440 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
441 StgArrWords_words(p) = words_needed;
453 if ( words_needed == 2 ) {
461 } else /* val==0 */ {
469 /* returns (# size :: Int#,
470 data :: ByteArray# #)
474 word64ToIntegerzh_fast
476 /* arguments: L1 = Word64# */
479 W_ hi, lo, s, words_needed, p;
482 hi = TO_W_(val >> 32);
491 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
492 NO_PTRS, word64ToIntegerzh_fast );
494 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
495 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
496 StgArrWords_words(p) = words_needed;
506 } else /* val==0 */ {
511 /* returns (# size :: Int#,
512 data :: ByteArray# #)
519 #endif /* SUPPORT_LONG_LONGS */
521 /* ToDo: this is shockingly inefficient */
526 bits8 [SIZEOF_MP_INT];
531 bits8 [SIZEOF_MP_INT];
536 bits8 [SIZEOF_MP_INT];
541 bits8 [SIZEOF_MP_INT];
546 #define FETCH_MP_TEMP(X) \
548 X = BaseReg + (OFFSET_StgRegTable_r ## X);
550 #define FETCH_MP_TEMP(X) /* Nothing */
553 #define GMP_TAKE2_RET1(name,mp_fun) \
558 FETCH_MP_TEMP(mp_tmp1); \
559 FETCH_MP_TEMP(mp_tmp2); \
560 FETCH_MP_TEMP(mp_result1) \
561 FETCH_MP_TEMP(mp_result2); \
563 /* call doYouWantToGC() */ \
564 MAYBE_GC(R2_PTR & R4_PTR, name); \
571 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
572 MP_INT__mp_size(mp_tmp1) = (s1); \
573 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
574 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
575 MP_INT__mp_size(mp_tmp2) = (s2); \
576 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
578 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
580 /* Perform the operation */ \
581 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
583 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
584 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
587 #define GMP_TAKE1_RET1(name,mp_fun) \
592 FETCH_MP_TEMP(mp_tmp1); \
593 FETCH_MP_TEMP(mp_result1) \
595 /* call doYouWantToGC() */ \
596 MAYBE_GC(R2_PTR, name); \
601 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
602 MP_INT__mp_size(mp_tmp1) = (s1); \
603 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
605 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
607 /* Perform the operation */ \
608 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr") []; \
610 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
611 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
614 #define GMP_TAKE2_RET2(name,mp_fun) \
619 FETCH_MP_TEMP(mp_tmp1); \
620 FETCH_MP_TEMP(mp_tmp2); \
621 FETCH_MP_TEMP(mp_result1) \
622 FETCH_MP_TEMP(mp_result2) \
624 /* call doYouWantToGC() */ \
625 MAYBE_GC(R2_PTR & R4_PTR, name); \
632 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
633 MP_INT__mp_size(mp_tmp1) = (s1); \
634 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
635 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
636 MP_INT__mp_size(mp_tmp2) = (s2); \
637 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
639 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
640 foreign "C" __gmpz_init(mp_result2 "ptr") []; \
642 /* Perform the operation */ \
643 foreign "C" mp_fun(mp_result1 "ptr",mp_result2 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
645 RET_NPNP(TO_W_(MP_INT__mp_size(mp_result1)), \
646 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords, \
647 TO_W_(MP_INT__mp_size(mp_result2)), \
648 MP_INT__mp_d(mp_result2) - SIZEOF_StgArrWords); \
651 GMP_TAKE2_RET1(plusIntegerzh_fast, __gmpz_add)
652 GMP_TAKE2_RET1(minusIntegerzh_fast, __gmpz_sub)
653 GMP_TAKE2_RET1(timesIntegerzh_fast, __gmpz_mul)
654 GMP_TAKE2_RET1(gcdIntegerzh_fast, __gmpz_gcd)
655 GMP_TAKE2_RET1(quotIntegerzh_fast, __gmpz_tdiv_q)
656 GMP_TAKE2_RET1(remIntegerzh_fast, __gmpz_tdiv_r)
657 GMP_TAKE2_RET1(divExactIntegerzh_fast, __gmpz_divexact)
658 GMP_TAKE2_RET1(andIntegerzh_fast, __gmpz_and)
659 GMP_TAKE2_RET1(orIntegerzh_fast, __gmpz_ior)
660 GMP_TAKE2_RET1(xorIntegerzh_fast, __gmpz_xor)
661 GMP_TAKE1_RET1(complementIntegerzh_fast, __gmpz_com)
663 GMP_TAKE2_RET2(quotRemIntegerzh_fast, __gmpz_tdiv_qr)
664 GMP_TAKE2_RET2(divModIntegerzh_fast, __gmpz_fdiv_qr)
668 mp_tmp_w: W_; // NB. mp_tmp_w is really an here mp_limb_t
674 /* R1 = the first Int#; R2 = the second Int# */
676 FETCH_MP_TEMP(mp_tmp_w);
679 r = foreign "C" __gmpn_gcd_1(mp_tmp_w "ptr", 1, R2) [];
682 /* Result parked in R1, return via info-pointer at TOS */
683 jump %ENTRY_CODE(Sp(0));
689 /* R1 = s1; R2 = d1; R3 = the int */
690 R1 = foreign "C" __gmpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3) [];
692 /* Result parked in R1, return via info-pointer at TOS */
693 jump %ENTRY_CODE(Sp(0));
699 /* R1 = s1; R2 = d1; R3 = the int */
700 W_ usize, vsize, v_digit, u_digit;
706 // paraphrased from __gmpz_cmp_si() in the GMP sources
707 if (%gt(v_digit,0)) {
710 if (%lt(v_digit,0)) {
716 if (usize != vsize) {
718 jump %ENTRY_CODE(Sp(0));
723 jump %ENTRY_CODE(Sp(0));
726 u_digit = W_[BYTE_ARR_CTS(R2)];
728 if (u_digit == v_digit) {
730 jump %ENTRY_CODE(Sp(0));
733 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
739 jump %ENTRY_CODE(Sp(0));
744 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
745 W_ usize, vsize, size, up, vp;
748 // paraphrased from __gmpz_cmp() in the GMP sources
752 if (usize != vsize) {
754 jump %ENTRY_CODE(Sp(0));
759 jump %ENTRY_CODE(Sp(0));
762 if (%lt(usize,0)) { // NB. not <, which is unsigned
768 up = BYTE_ARR_CTS(R2);
769 vp = BYTE_ARR_CTS(R4);
771 cmp = foreign "C" __gmpn_cmp(up "ptr", vp "ptr", size) [];
773 if (cmp == 0 :: CInt) {
775 jump %ENTRY_CODE(Sp(0));
778 if (%lt(cmp,0 :: CInt) == %lt(usize,0)) {
783 /* Result parked in R1, return via info-pointer at TOS */
784 jump %ENTRY_CODE(Sp(0));
796 r = W_[R2 + SIZEOF_StgArrWords];
801 /* Result parked in R1, return via info-pointer at TOS */
803 jump %ENTRY_CODE(Sp(0));
815 r = W_[R2 + SIZEOF_StgArrWords];
820 /* Result parked in R1, return via info-pointer at TOS */
822 jump %ENTRY_CODE(Sp(0));
829 FETCH_MP_TEMP(mp_tmp1);
830 FETCH_MP_TEMP(mp_tmp_w);
832 /* arguments: F1 = Float# */
835 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, decodeFloatzh_fast );
837 /* Be prepared to tell Lennart-coded __decodeFloat
838 where mantissa._mp_d can be put (it does not care about the rest) */
839 p = Hp - SIZEOF_StgArrWords;
840 SET_HDR(p,stg_ARR_WORDS_info,W_[CCCS]);
841 StgArrWords_words(p) = 1;
842 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
844 /* Perform the operation */
845 foreign "C" __decodeFloat(mp_tmp1 "ptr",mp_tmp_w "ptr" ,arg) [];
847 /* returns: (Int# (expn), Int#, ByteArray#) */
848 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
851 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
852 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
858 FETCH_MP_TEMP(mp_tmp1);
859 FETCH_MP_TEMP(mp_tmp_w);
861 /* arguments: D1 = Double# */
864 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
866 /* Be prepared to tell Lennart-coded __decodeDouble
867 where mantissa.d can be put (it does not care about the rest) */
868 p = Hp - ARR_SIZE + WDS(1);
869 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
870 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
871 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
873 /* Perform the operation */
874 foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg) [];
876 /* returns: (Int# (expn), Int#, ByteArray#) */
877 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
880 /* -----------------------------------------------------------------------------
881 * Concurrency primitives
882 * -------------------------------------------------------------------------- */
886 /* args: R1 = closure to spark */
888 MAYBE_GC(R1_PTR, forkzh_fast);
894 "ptr" threadid = foreign "C" createIOThread( MyCapability() "ptr",
895 RtsFlags_GcFlags_initialStkSize(RtsFlags),
897 foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") [];
899 // switch at the earliest opportunity
900 CInt[context_switch] = 1 :: CInt;
907 /* args: R1 = cpu, R2 = closure to spark */
909 MAYBE_GC(R2_PTR, forkOnzh_fast);
917 "ptr" threadid = foreign "C" createIOThread( MyCapability() "ptr",
918 RtsFlags_GcFlags_initialStkSize(RtsFlags),
920 foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") [];
922 // switch at the earliest opportunity
923 CInt[context_switch] = 1 :: CInt;
930 jump stg_yield_noregs;
945 foreign "C" labelThread(R1 "ptr", R2 "ptr") [];
947 jump %ENTRY_CODE(Sp(0));
950 isCurrentThreadBoundzh_fast
954 r = foreign "C" isThreadBound(CurrentTSO) [];
959 /* -----------------------------------------------------------------------------
961 * -------------------------------------------------------------------------- */
965 #define IF_NOT_REG_R1(x)
968 #define IF_NOT_REG_R1(x) x
971 // Catch retry frame ------------------------------------------------------------
973 #define CATCH_RETRY_FRAME_ERROR(label) \
974 label { foreign "C" barf("catch_retry_frame incorrectly entered!"); }
976 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_0_ret)
977 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_1_ret)
978 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_2_ret)
979 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_3_ret)
980 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_4_ret)
981 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_5_ret)
982 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_6_ret)
983 CATCH_RETRY_FRAME_ERROR(stg_catch_retry_frame_7_ret)
985 #if MAX_VECTORED_RTN > 8
986 #error MAX_VECTORED_RTN has changed: please modify stg_catch_retry_frame too.
989 #if defined(PROFILING)
990 #define CATCH_RETRY_FRAME_BITMAP 7
991 #define CATCH_RETRY_FRAME_WORDS 5
993 #define CATCH_RETRY_FRAME_BITMAP 1
994 #define CATCH_RETRY_FRAME_WORDS 3
997 INFO_TABLE_RET(stg_catch_retry_frame,
998 CATCH_RETRY_FRAME_WORDS, CATCH_RETRY_FRAME_BITMAP,
1000 stg_catch_retry_frame_0_ret,
1001 stg_catch_retry_frame_1_ret,
1002 stg_catch_retry_frame_2_ret,
1003 stg_catch_retry_frame_3_ret,
1004 stg_catch_retry_frame_4_ret,
1005 stg_catch_retry_frame_5_ret,
1006 stg_catch_retry_frame_6_ret,
1007 stg_catch_retry_frame_7_ret)
1009 W_ r, frame, trec, outer;
1010 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1013 trec = StgTSO_trec(CurrentTSO);
1014 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1015 r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
1017 /* Succeeded (either first branch or second branch) */
1018 StgTSO_trec(CurrentTSO) = outer;
1019 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1020 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1021 jump %ENTRY_CODE(Sp(SP_OFF));
1023 /* Did not commit: re-execute */
1025 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1026 StgTSO_trec(CurrentTSO) = new_trec;
1027 if (StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
1028 R1 = StgCatchRetryFrame_alt_code(frame);
1030 R1 = StgCatchRetryFrame_first_code(frame);
1037 // Atomically frame -------------------------------------------------------------
1040 #define ATOMICALLY_FRAME_ERROR(label) \
1041 label { foreign "C" barf("atomically_frame incorrectly entered!"); }
1043 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_0_ret)
1044 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_1_ret)
1045 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_2_ret)
1046 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_3_ret)
1047 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_4_ret)
1048 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_5_ret)
1049 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_6_ret)
1050 ATOMICALLY_FRAME_ERROR(stg_atomically_frame_7_ret)
1052 #if MAX_VECTORED_RTN > 8
1053 #error MAX_VECTORED_RTN has changed: please modify stg_atomically_frame too.
1056 #if defined(PROFILING)
1057 #define ATOMICALLY_FRAME_BITMAP 3
1058 #define ATOMICALLY_FRAME_WORDS 4
1060 #define ATOMICALLY_FRAME_BITMAP 0
1061 #define ATOMICALLY_FRAME_WORDS 2
1065 INFO_TABLE_RET(stg_atomically_frame,
1066 ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP,
1068 stg_atomically_frame_0_ret,
1069 stg_atomically_frame_1_ret,
1070 stg_atomically_frame_2_ret,
1071 stg_atomically_frame_3_ret,
1072 stg_atomically_frame_4_ret,
1073 stg_atomically_frame_5_ret,
1074 stg_atomically_frame_6_ret,
1075 stg_atomically_frame_7_ret)
1077 W_ frame, trec, valid, next_invariant, q, outer;
1078 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1081 trec = StgTSO_trec(CurrentTSO);
1082 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1084 if (outer == NO_TREC) {
1085 /* First time back at the atomically frame -- pick up invariants */
1086 "ptr" q = foreign "C" stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr") [];
1087 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
1090 /* Second/subsequent time back at the atomically frame -- abort the
1091 * tx that's checking the invariant and move on to the next one */
1092 StgTSO_trec(CurrentTSO) = outer;
1093 q = StgAtomicallyFrame_next_invariant_to_check(frame);
1094 StgInvariantCheckQueue_my_execution(q) = trec;
1095 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1096 /* Don't free trec -- it's linked from q and will be stashed in the
1097 * invariant if we eventually commit. */
1098 q = StgInvariantCheckQueue_next_queue_entry(q);
1099 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
1103 q = StgAtomicallyFrame_next_invariant_to_check(frame);
1105 if (q != END_INVARIANT_CHECK_QUEUE) {
1106 /* We can't commit yet: another invariant to check */
1107 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [];
1108 StgTSO_trec(CurrentTSO) = trec;
1110 next_invariant = StgInvariantCheckQueue_invariant(q);
1111 R1 = StgAtomicInvariant_code(next_invariant);
1116 /* We've got no more invariants to check, try to commit */
1117 valid = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") [];
1119 /* Transaction was valid: commit succeeded */
1120 StgTSO_trec(CurrentTSO) = NO_TREC;
1121 Sp = Sp + SIZEOF_StgAtomicallyFrame;
1122 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;)
1123 jump %ENTRY_CODE(Sp(SP_OFF));
1125 /* Transaction was not valid: try again */
1126 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1127 StgTSO_trec(CurrentTSO) = trec;
1128 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
1129 R1 = StgAtomicallyFrame_code(frame);
1135 INFO_TABLE_RET(stg_atomically_waiting_frame,
1136 ATOMICALLY_FRAME_WORDS, ATOMICALLY_FRAME_BITMAP,
1138 stg_atomically_frame_0_ret,
1139 stg_atomically_frame_1_ret,
1140 stg_atomically_frame_2_ret,
1141 stg_atomically_frame_3_ret,
1142 stg_atomically_frame_4_ret,
1143 stg_atomically_frame_5_ret,
1144 stg_atomically_frame_6_ret,
1145 stg_atomically_frame_7_ret)
1147 W_ frame, trec, valid;
1148 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); )
1152 /* The TSO is currently waiting: should we stop waiting? */
1153 valid = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") [];
1155 /* Previous attempt is still valid: no point trying again yet */
1156 IF_NOT_REG_R1(Sp_adj(-2);
1157 Sp(1) = stg_NO_FINALIZER_closure;
1158 Sp(0) = stg_ut_1_0_unreg_info;)
1159 jump stg_block_noregs;
1161 /* Previous attempt is no longer valid: try again */
1162 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1163 StgTSO_trec(CurrentTSO) = trec;
1164 StgHeader_info(frame) = stg_atomically_frame_info;
1165 R1 = StgAtomicallyFrame_code(frame);
1170 // STM catch frame --------------------------------------------------------------
1172 #define CATCH_STM_FRAME_ENTRY_TEMPLATE(label,ret) \
1175 IF_NOT_REG_R1(W_ rval; rval = Sp(0); Sp_adj(1); ) \
1176 W_ r, frame, trec, outer; \
1178 trec = StgTSO_trec(CurrentTSO); \
1179 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") []; \
1180 r = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") []; \
1182 /* Commit succeeded */ \
1183 StgTSO_trec(CurrentTSO) = outer; \
1184 Sp = Sp + SIZEOF_StgCatchSTMFrame; \
1185 IF_NOT_REG_R1(Sp_adj(-1); Sp(0) = rval;) \
1188 /* Commit failed */ \
1190 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") []; \
1191 StgTSO_trec(CurrentTSO) = new_trec; \
1192 R1 = StgCatchSTMFrame_code(frame); \
1193 jump stg_ap_v_fast; \
1203 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_0_ret,%RET_VEC(Sp(SP_OFF),0))
1204 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_1_ret,%RET_VEC(Sp(SP_OFF),1))
1205 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_2_ret,%RET_VEC(Sp(SP_OFF),2))
1206 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_3_ret,%RET_VEC(Sp(SP_OFF),3))
1207 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_4_ret,%RET_VEC(Sp(SP_OFF),4))
1208 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_5_ret,%RET_VEC(Sp(SP_OFF),5))
1209 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_6_ret,%RET_VEC(Sp(SP_OFF),6))
1210 CATCH_STM_FRAME_ENTRY_TEMPLATE(stg_catch_stm_frame_7_ret,%RET_VEC(Sp(SP_OFF),7))
1212 #if MAX_VECTORED_RTN > 8
1213 #error MAX_VECTORED_RTN has changed: please modify stg_catch_stm_frame too.
1216 #if defined(PROFILING)
1217 #define CATCH_STM_FRAME_BITMAP 3
1218 #define CATCH_STM_FRAME_WORDS 4
1220 #define CATCH_STM_FRAME_BITMAP 0
1221 #define CATCH_STM_FRAME_WORDS 2
1224 /* Catch frames are very similar to update frames, but when entering
1225 * one we just pop the frame off the stack and perform the correct
1226 * kind of return to the activation record underneath us on the stack.
1229 INFO_TABLE_RET(stg_catch_stm_frame,
1230 CATCH_STM_FRAME_WORDS, CATCH_STM_FRAME_BITMAP,
1232 stg_catch_stm_frame_0_ret,
1233 stg_catch_stm_frame_1_ret,
1234 stg_catch_stm_frame_2_ret,
1235 stg_catch_stm_frame_3_ret,
1236 stg_catch_stm_frame_4_ret,
1237 stg_catch_stm_frame_5_ret,
1238 stg_catch_stm_frame_6_ret,
1239 stg_catch_stm_frame_7_ret)
1240 CATCH_STM_FRAME_ENTRY_TEMPLATE(,%ENTRY_CODE(Sp(SP_OFF)))
1243 // Primop definition ------------------------------------------------------------
1251 // stmStartTransaction may allocate
1252 MAYBE_GC (R1_PTR, atomicallyzh_fast);
1254 /* Args: R1 = m :: STM a */
1255 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);
1257 old_trec = StgTSO_trec(CurrentTSO);
1259 /* Nested transactions are not allowed; raise an exception */
1260 if (old_trec != NO_TREC) {
1261 R1 = base_GHCziIOBase_NestedAtomically_closure;
1265 /* Set up the atomically frame */
1266 Sp = Sp - SIZEOF_StgAtomicallyFrame;
1269 SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
1270 StgAtomicallyFrame_code(frame) = R1;
1271 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
1273 /* Start the memory transcation */
1274 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1];
1275 StgTSO_trec(CurrentTSO) = new_trec;
1277 /* Apply R1 to the realworld token */
1286 /* Args: R1 :: STM a */
1287 /* Args: R2 :: Exception -> STM a */
1288 STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast);
1290 /* Set up the catch frame */
1291 Sp = Sp - SIZEOF_StgCatchSTMFrame;
1294 SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
1295 StgCatchSTMFrame_handler(frame) = R2;
1296 StgCatchSTMFrame_code(frame) = R1;
1298 /* Start a nested transaction to run the body of the try block in */
1301 cur_trec = StgTSO_trec(CurrentTSO);
1302 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", cur_trec "ptr");
1303 StgTSO_trec(CurrentTSO) = new_trec;
1305 /* Apply R1 to the realworld token */
1316 // stmStartTransaction may allocate
1317 MAYBE_GC (R1_PTR & R2_PTR, catchRetryzh_fast);
1319 /* Args: R1 :: STM a */
1320 /* Args: R2 :: STM a */
1321 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);
1323 /* Start a nested transaction within which to run the first code */
1324 trec = StgTSO_trec(CurrentTSO);
1325 "ptr" new_trec = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2];
1326 StgTSO_trec(CurrentTSO) = new_trec;
1328 /* Set up the catch-retry frame */
1329 Sp = Sp - SIZEOF_StgCatchRetryFrame;
1332 SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
1333 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1334 StgCatchRetryFrame_first_code(frame) = R1;
1335 StgCatchRetryFrame_alt_code(frame) = R2;
1337 /* Apply R1 to the realworld token */
1350 MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate
1352 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
1354 StgTSO_sp(CurrentTSO) = Sp;
1355 frame_type = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") [];
1356 Sp = StgTSO_sp(CurrentTSO);
1358 trec = StgTSO_trec(CurrentTSO);
1359 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1361 if (frame_type == CATCH_RETRY_FRAME) {
1362 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
1363 ASSERT(outer != NO_TREC);
1364 // Abort the transaction attempting the current branch
1365 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1366 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
1367 if (!StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
1368 // Retry in the first branch: try the alternative
1369 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1370 StgTSO_trec(CurrentTSO) = trec;
1371 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
1372 R1 = StgCatchRetryFrame_alt_code(frame);
1375 // Retry in the alternative code: propagate the retry
1376 StgTSO_trec(CurrentTSO) = outer;
1377 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1378 goto retry_pop_stack;
1382 // We've reached the ATOMICALLY_FRAME: attempt to wait
1383 ASSERT(frame_type == ATOMICALLY_FRAME);
1384 if (outer != NO_TREC) {
1385 // We called retry while checking invariants, so abort the current
1386 // invariant check (merging its TVar accesses into the parents read
1387 // set so we'll wait on them)
1388 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1389 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
1391 StgTSO_trec(CurrentTSO) = trec;
1392 "ptr" outer = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1394 ASSERT(outer == NO_TREC);
1396 r = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") [];
1398 // Transaction was valid: stmWait put us on the TVars' queues, we now block
1399 StgHeader_info(frame) = stg_atomically_waiting_frame_info;
1401 // Fix up the stack in the unregisterised case: the return convention is different.
1402 IF_NOT_REG_R1(Sp_adj(-2);
1403 Sp(1) = stg_NO_FINALIZER_closure;
1404 Sp(0) = stg_ut_1_0_unreg_info;)
1405 R3 = trec; // passing to stmWaitUnblock()
1406 jump stg_block_stmwait;
1408 // Transaction was not valid: retry immediately
1409 "ptr" trec = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1410 StgTSO_trec(CurrentTSO) = trec;
1411 R1 = StgAtomicallyFrame_code(frame);
1422 /* Args: R1 = invariant closure */
1423 MAYBE_GC (R1_PTR, checkzh_fast);
1425 trec = StgTSO_trec(CurrentTSO);
1427 foreign "C" stmAddInvariantToCheck(MyCapability() "ptr",
1431 jump %ENTRY_CODE(Sp(0));
1440 /* Args: R1 = initialisation value */
1442 MAYBE_GC (R1_PTR, newTVarzh_fast);
1444 "ptr" tv = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") [];
1455 /* Args: R1 = TVar closure */
1457 MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
1458 trec = StgTSO_trec(CurrentTSO);
1460 "ptr" result = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") [];
1472 /* Args: R1 = TVar closure */
1473 /* R2 = New value */
1475 MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
1476 trec = StgTSO_trec(CurrentTSO);
1479 foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") [];
1481 jump %ENTRY_CODE(Sp(0));
1485 /* -----------------------------------------------------------------------------
1488 * take & putMVar work as follows. Firstly, an important invariant:
1490 * If the MVar is full, then the blocking queue contains only
1491 * threads blocked on putMVar, and if the MVar is empty then the
1492 * blocking queue contains only threads blocked on takeMVar.
1495 * MVar empty : then add ourselves to the blocking queue
1496 * MVar full : remove the value from the MVar, and
1497 * blocking queue empty : return
1498 * blocking queue non-empty : perform the first blocked putMVar
1499 * from the queue, and wake up the
1500 * thread (MVar is now full again)
1502 * putMVar is just the dual of the above algorithm.
1504 * How do we "perform a putMVar"? Well, we have to fiddle around with
1505 * the stack of the thread waiting to do the putMVar. See
1506 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1507 * the stack layout, and the PerformPut and PerformTake macros below.
1509 * It is important that a blocked take or put is woken up with the
1510 * take/put already performed, because otherwise there would be a
1511 * small window of vulnerability where the thread could receive an
1512 * exception and never perform its take or put, and we'd end up with a
1515 * -------------------------------------------------------------------------- */
1519 /* args: R1 = MVar closure */
1521 if (GET_INFO(R1) == stg_EMPTY_MVAR_info) {
1533 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
1535 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1536 SET_HDR(mvar,stg_EMPTY_MVAR_info,W_[CCCS]);
1537 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1538 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1539 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1544 /* If R1 isn't available, pass it on the stack */
1546 #define PerformTake(tso, value) \
1547 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1548 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1550 #define PerformTake(tso, value) \
1551 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1552 W_[StgTSO_sp(tso) + WDS(0)] = stg_ut_1_0_unreg_info;
1555 #define PerformPut(tso,lval) \
1556 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1557 lval = W_[StgTSO_sp(tso) - WDS(1)];
1561 W_ mvar, val, info, tso;
1563 /* args: R1 = MVar closure */
1566 #if defined(THREADED_RTS)
1567 "ptr" info = foreign "C" lockClosure(mvar "ptr") [];
1569 info = GET_INFO(mvar);
1572 /* If the MVar is empty, put ourselves on its blocking queue,
1573 * and wait until we're woken up.
1575 if (info == stg_EMPTY_MVAR_info) {
1576 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1577 StgMVar_head(mvar) = CurrentTSO;
1579 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1581 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1582 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1583 StgTSO_block_info(CurrentTSO) = mvar;
1584 StgMVar_tail(mvar) = CurrentTSO;
1586 jump stg_block_takemvar;
1589 /* we got the value... */
1590 val = StgMVar_value(mvar);
1592 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1594 /* There are putMVar(s) waiting...
1595 * wake up the first thread on the queue
1597 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1599 /* actually perform the putMVar for the thread that we just woke up */
1600 tso = StgMVar_head(mvar);
1601 PerformPut(tso,StgMVar_value(mvar));
1604 #if defined(GRAN) || defined(PAR)
1605 /* ToDo: check 2nd arg (mvar) is right */
1606 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar),mvar) [];
1607 StgMVar_head(mvar) = tso;
1609 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr",
1610 StgMVar_head(mvar) "ptr") [];
1611 StgMVar_head(mvar) = tso;
1614 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1615 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1618 #if defined(THREADED_RTS)
1619 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1625 /* No further putMVars, MVar is now empty */
1626 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1628 #if defined(THREADED_RTS)
1629 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1631 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1641 W_ mvar, val, info, tso;
1643 /* args: R1 = MVar closure */
1647 #if defined(THREADED_RTS)
1648 "ptr" info = foreign "C" lockClosure(mvar "ptr") [];
1650 info = GET_INFO(mvar);
1653 if (info == stg_EMPTY_MVAR_info) {
1654 #if defined(THREADED_RTS)
1655 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1657 /* HACK: we need a pointer to pass back,
1658 * so we abuse NO_FINALIZER_closure
1660 RET_NP(0, stg_NO_FINALIZER_closure);
1663 /* we got the value... */
1664 val = StgMVar_value(mvar);
1666 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1668 /* There are putMVar(s) waiting...
1669 * wake up the first thread on the queue
1671 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1673 /* actually perform the putMVar for the thread that we just woke up */
1674 tso = StgMVar_head(mvar);
1675 PerformPut(tso,StgMVar_value(mvar));
1678 #if defined(GRAN) || defined(PAR)
1679 /* ToDo: check 2nd arg (mvar) is right */
1680 "ptr" tso = foreign "C" unblockOne(StgMVar_head(mvar) "ptr", mvar "ptr") [];
1681 StgMVar_head(mvar) = tso;
1683 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr",
1684 StgMVar_head(mvar) "ptr") [];
1685 StgMVar_head(mvar) = tso;
1688 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1689 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1691 #if defined(THREADED_RTS)
1692 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1697 /* No further putMVars, MVar is now empty */
1698 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1699 #if defined(THREADED_RTS)
1700 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1702 SET_INFO(mvar,stg_EMPTY_MVAR_info);
1714 /* args: R1 = MVar, R2 = value */
1717 #if defined(THREADED_RTS)
1718 "ptr" info = foreign "C" lockClosure(mvar "ptr") [R2];
1720 info = GET_INFO(mvar);
1723 if (info == stg_FULL_MVAR_info) {
1724 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1725 StgMVar_head(mvar) = CurrentTSO;
1727 StgTSO_link(StgMVar_tail(mvar)) = CurrentTSO;
1729 StgTSO_link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1730 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1731 StgTSO_block_info(CurrentTSO) = mvar;
1732 StgMVar_tail(mvar) = CurrentTSO;
1734 jump stg_block_putmvar;
1737 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1739 /* There are takeMVar(s) waiting: wake up the first one
1741 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1743 /* actually perform the takeMVar */
1744 tso = StgMVar_head(mvar);
1745 PerformTake(tso, R2);
1748 #if defined(GRAN) || defined(PAR)
1749 /* ToDo: check 2nd arg (mvar) is right */
1750 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") [];
1751 StgMVar_head(mvar) = tso;
1753 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") [];
1754 StgMVar_head(mvar) = tso;
1757 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1758 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1761 #if defined(THREADED_RTS)
1762 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1764 jump %ENTRY_CODE(Sp(0));
1768 /* No further takes, the MVar is now full. */
1769 StgMVar_value(mvar) = R2;
1771 #if defined(THREADED_RTS)
1772 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1774 SET_INFO(mvar,stg_FULL_MVAR_info);
1776 jump %ENTRY_CODE(Sp(0));
1779 /* ToDo: yield afterward for better communication performance? */
1787 /* args: R1 = MVar, R2 = value */
1790 #if defined(THREADED_RTS)
1791 "ptr" info = foreign "C" lockClosure(mvar "ptr") [R2];
1793 info = GET_INFO(mvar);
1796 if (info == stg_FULL_MVAR_info) {
1797 #if defined(THREADED_RTS)
1798 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1803 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1805 /* There are takeMVar(s) waiting: wake up the first one
1807 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1809 /* actually perform the takeMVar */
1810 tso = StgMVar_head(mvar);
1811 PerformTake(tso, R2);
1814 #if defined(GRAN) || defined(PAR)
1815 /* ToDo: check 2nd arg (mvar) is right */
1816 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr",mvar "ptr") [];
1817 StgMVar_head(mvar) = tso;
1819 "ptr" tso = foreign "C" unblockOne(MyCapability() "ptr", StgMVar_head(mvar) "ptr") [];
1820 StgMVar_head(mvar) = tso;
1823 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1824 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1827 #if defined(THREADED_RTS)
1828 foreign "C" unlockClosure(mvar "ptr", stg_EMPTY_MVAR_info) [];
1833 /* No further takes, the MVar is now full. */
1834 StgMVar_value(mvar) = R2;
1836 #if defined(THREADED_RTS)
1837 foreign "C" unlockClosure(mvar "ptr", stg_FULL_MVAR_info) [];
1839 SET_INFO(mvar,stg_FULL_MVAR_info);
1844 /* ToDo: yield afterward for better communication performance? */
1848 /* -----------------------------------------------------------------------------
1849 Stable pointer primitives
1850 ------------------------------------------------------------------------- */
1852 makeStableNamezh_fast
1856 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1858 index = foreign "C" lookupStableName(R1 "ptr") [];
1860 /* Is there already a StableName for this heap object?
1861 * stable_ptr_table is a pointer to an array of snEntry structs.
1863 if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) {
1864 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1865 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1866 StgStableName_sn(sn_obj) = index;
1867 snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj;
1869 sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry);
1876 makeStablePtrzh_fast
1880 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1881 "ptr" sp = foreign "C" getStablePtr(R1 "ptr") [];
1885 deRefStablePtrzh_fast
1887 /* Args: R1 = the stable ptr */
1890 r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry);
1894 /* -----------------------------------------------------------------------------
1895 Bytecode object primitives
1896 ------------------------------------------------------------------------- */
1907 W_ bco, bitmap_arr, bytes, words;
1910 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
1913 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R4_PTR&R6_PTR, newBCOzh_fast );
1915 bco = Hp - bytes + WDS(1);
1916 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
1918 StgBCO_instrs(bco) = R1;
1919 StgBCO_literals(bco) = R2;
1920 StgBCO_ptrs(bco) = R3;
1921 StgBCO_itbls(bco) = R4;
1922 StgBCO_arity(bco) = HALF_W_(R5);
1923 StgBCO_size(bco) = HALF_W_(words);
1925 // Copy the arity/bitmap info into the BCO
1929 if (i < StgArrWords_words(bitmap_arr)) {
1930 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
1941 // R1 = the BCO# for the AP
1945 // This function is *only* used to wrap zero-arity BCOs in an
1946 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1947 // saturated and always points directly to a FUN or BCO.
1948 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
1949 StgBCO_arity(R1) == HALF_W_(0));
1951 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
1952 TICK_ALLOC_UP_THK(0, 0);
1953 CCCS_ALLOC(SIZEOF_StgAP);
1955 ap = Hp - SIZEOF_StgAP + WDS(1);
1956 SET_HDR(ap, stg_AP_info, W_[CCCS]);
1958 StgAP_n_args(ap) = HALF_W_(0);
1966 /* args: R1 = closure to analyze */
1968 MAYBE_GC(R1_PTR, infoPtrzh_fast);
1971 info = %GET_STD_INFO(R1);
1975 closurePayloadzh_fast
1977 /* args: R1 = closure to analyze */
1978 // TODO: Consider the absence of ptrs or nonptrs as a special case ?
1980 MAYBE_GC(R1_PTR, closurePayloadzh_fast);
1982 W_ info, ptrs, nptrs, p, ptrs_arr, nptrs_arr;
1983 info = %GET_STD_INFO(R1);
1984 ptrs = TO_W_(%INFO_PTRS(info));
1985 nptrs = TO_W_(%INFO_NPTRS(info));
1988 ALLOC_PRIM (SIZEOF_StgMutArrPtrs + WDS(ptrs), R1_PTR, closurePayloadzh_fast);
1989 ptrs_arr = Hp - SIZEOF_StgMutArrPtrs - WDS(ptrs) + WDS(1);
1990 SET_HDR(ptrs_arr, stg_MUT_ARR_PTRS_FROZEN_info, W_[CCCS]);
1991 StgMutArrPtrs_ptrs(ptrs_arr) = ptrs;
1994 W_[ptrs_arr + SIZEOF_StgMutArrPtrs + WDS(p)] = StgClosure_payload(R1,p);
1999 ALLOC_PRIM (SIZEOF_StgArrWords + WDS(nptrs), R1_PTR, closurePayloadzh_fast);
2000 nptrs_arr = Hp - SIZEOF_StgArrWords - WDS(nptrs) + WDS(1);
2001 SET_HDR(nptrs_arr, stg_ARR_WORDS_info, W_[CCCS]);
2002 StgArrWords_words(nptrs_arr) = nptrs;
2006 W_[BYTE_ARR_CTS(nptrs_arr) + WDS(p)] = StgClosure_payload(R1, p+ptrs);
2010 RET_PP(ptrs_arr, nptrs_arr);
2013 /* -----------------------------------------------------------------------------
2014 Thread I/O blocking primitives
2015 -------------------------------------------------------------------------- */
2017 /* Add a thread to the end of the blocked queue. (C-- version of the C
2018 * macro in Schedule.h).
2020 #define APPEND_TO_BLOCKED_QUEUE(tso) \
2021 ASSERT(StgTSO_link(tso) == END_TSO_QUEUE); \
2022 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
2023 W_[blocked_queue_hd] = tso; \
2025 StgTSO_link(W_[blocked_queue_tl]) = tso; \
2027 W_[blocked_queue_tl] = tso;
2033 foreign "C" barf("waitRead# on threaded RTS");
2036 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2037 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
2038 StgTSO_block_info(CurrentTSO) = R1;
2039 // No locking - we're not going to use this interface in the
2040 // threaded RTS anyway.
2041 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2042 jump stg_block_noregs;
2050 foreign "C" barf("waitWrite# on threaded RTS");
2053 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2054 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2055 StgTSO_block_info(CurrentTSO) = R1;
2056 // No locking - we're not going to use this interface in the
2057 // threaded RTS anyway.
2058 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2059 jump stg_block_noregs;
2064 STRING(stg_delayzh_malloc_str, "delayzh_fast")
2067 #ifdef mingw32_HOST_OS
2075 foreign "C" barf("delay# on threaded RTS");
2078 /* args: R1 (microsecond delay amount) */
2079 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2080 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
2082 #ifdef mingw32_HOST_OS
2084 /* could probably allocate this on the heap instead */
2085 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2086 stg_delayzh_malloc_str);
2087 reqID = foreign "C" addDelayRequest(R1);
2088 StgAsyncIOResult_reqID(ares) = reqID;
2089 StgAsyncIOResult_len(ares) = 0;
2090 StgAsyncIOResult_errCode(ares) = 0;
2091 StgTSO_block_info(CurrentTSO) = ares;
2093 /* Having all async-blocked threads reside on the blocked_queue
2094 * simplifies matters, so change the status to OnDoProc put the
2095 * delayed thread on the blocked_queue.
2097 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2098 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2099 jump stg_block_async_void;
2105 time = foreign "C" getourtimeofday() [R1];
2106 divisor = TO_W_(RtsFlags_MiscFlags_tickInterval(RtsFlags))*1000;
2107 target = ((R1 + divisor - 1) / divisor) /* divide rounding up */
2108 + time + 1; /* Add 1 as getourtimeofday rounds down */
2109 StgTSO_block_info(CurrentTSO) = target;
2111 /* Insert the new thread in the sleeping queue. */
2113 t = W_[sleeping_queue];
2115 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
2121 StgTSO_link(CurrentTSO) = t;
2123 W_[sleeping_queue] = CurrentTSO;
2125 StgTSO_link(prev) = CurrentTSO;
2127 jump stg_block_noregs;
2129 #endif /* !THREADED_RTS */
2133 #ifdef mingw32_HOST_OS
2134 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
2141 foreign "C" barf("asyncRead# on threaded RTS");
2144 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2145 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2146 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
2148 /* could probably allocate this on the heap instead */
2149 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2150 stg_asyncReadzh_malloc_str)
2152 reqID = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr") [];
2153 StgAsyncIOResult_reqID(ares) = reqID;
2154 StgAsyncIOResult_len(ares) = 0;
2155 StgAsyncIOResult_errCode(ares) = 0;
2156 StgTSO_block_info(CurrentTSO) = ares;
2157 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2158 jump stg_block_async;
2162 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
2169 foreign "C" barf("asyncWrite# on threaded RTS");
2172 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2173 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2174 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2176 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2177 stg_asyncWritezh_malloc_str)
2179 reqID = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr") [];
2181 StgAsyncIOResult_reqID(ares) = reqID;
2182 StgAsyncIOResult_len(ares) = 0;
2183 StgAsyncIOResult_errCode(ares) = 0;
2184 StgTSO_block_info(CurrentTSO) = ares;
2185 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2186 jump stg_block_async;
2190 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
2197 foreign "C" barf("asyncDoProc# on threaded RTS");
2200 /* args: R1 = proc, R2 = param */
2201 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2202 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2204 /* could probably allocate this on the heap instead */
2205 "ptr" ares = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2206 stg_asyncDoProczh_malloc_str)
2208 reqID = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr") [];
2209 StgAsyncIOResult_reqID(ares) = reqID;
2210 StgAsyncIOResult_len(ares) = 0;
2211 StgAsyncIOResult_errCode(ares) = 0;
2212 StgTSO_block_info(CurrentTSO) = ares;
2213 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2214 jump stg_block_async;
2219 /* -----------------------------------------------------------------------------
2222 classes CCallable and CReturnable don't really exist, but the
2223 compiler insists on generating dictionaries containing references
2224 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
2225 for these. Some C compilers can't cope with zero-length static arrays,
2226 so we have to make these one element long.
2227 --------------------------------------------------------------------------- */
2230 GHC_ZCCCallable_static_info: W_ 0;
2234 GHC_ZCCReturnable_static_info: W_ 0;