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 * ---------------------------------------------------------------------------*/
31 #ifndef mingw32_HOST_OS
41 import __gmpz_tdiv_qr;
42 import __gmpz_fdiv_qr;
43 import __gmpz_divexact;
49 import pthread_mutex_lock;
50 import pthread_mutex_unlock;
52 import base_ControlziExceptionziBase_nestedAtomically_closure;
53 import EnterCriticalSection;
54 import LeaveCriticalSection;
55 import ghczmprim_GHCziBool_False_closure;
57 /*-----------------------------------------------------------------------------
60 Basically just new*Array - the others are all inline macros.
62 The size arg is always passed in R1, and the result returned in R1.
64 The slow entry point is for returning from a heap check, the saved
65 size argument must be re-loaded from the stack.
66 -------------------------------------------------------------------------- */
68 /* for objects that are *less* than the size of a word, make sure we
69 * round up to the nearest word for the size of the array.
74 W_ words, payload_words, n, p;
75 MAYBE_GC(NO_PTRS,newByteArrayzh_fast);
77 payload_words = ROUNDUP_BYTES_TO_WDS(n);
78 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
79 ("ptr" p) = foreign "C" allocateLocal(MyCapability() "ptr",words) [];
80 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
81 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
82 StgArrWords_words(p) = payload_words;
87 #define BA_MASK (BA_ALIGN-1)
89 newPinnedByteArrayzh_fast
91 W_ words, bytes, payload_words, p;
93 MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast);
95 /* payload_words is what we will tell the profiler we had to allocate */
96 payload_words = ROUNDUP_BYTES_TO_WDS(bytes);
97 /* When we actually allocate memory, we need to allow space for the
99 bytes = bytes + SIZEOF_StgArrWords;
100 /* And we want to align to BA_ALIGN bytes, so we need to allow space
101 to shift up to BA_ALIGN - 1 bytes: */
102 bytes = bytes + BA_ALIGN - 1;
103 /* Now we convert to a number of words: */
104 words = ROUNDUP_BYTES_TO_WDS(bytes);
106 ("ptr" p) = foreign "C" allocatePinned(words) [];
107 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
109 /* Now we need to move p forward so that the payload is aligned
110 to BA_ALIGN bytes: */
111 p = p + ((-p - SIZEOF_StgArrWords) & BA_MASK);
113 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
114 StgArrWords_words(p) = payload_words;
118 newAlignedPinnedByteArrayzh_fast
120 W_ words, bytes, payload_words, p, alignment;
122 MAYBE_GC(NO_PTRS,newAlignedPinnedByteArrayzh_fast);
126 /* payload_words is what we will tell the profiler we had to allocate */
127 payload_words = ROUNDUP_BYTES_TO_WDS(bytes);
129 /* When we actually allocate memory, we need to allow space for the
131 bytes = bytes + SIZEOF_StgArrWords;
132 /* And we want to align to <alignment> bytes, so we need to allow space
133 to shift up to <alignment - 1> bytes: */
134 bytes = bytes + alignment - 1;
135 /* Now we convert to a number of words: */
136 words = ROUNDUP_BYTES_TO_WDS(bytes);
138 ("ptr" p) = foreign "C" allocatePinned(words) [];
139 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
141 /* Now we need to move p forward so that the payload is aligned
142 to <alignment> bytes. Note that we are assuming that
143 <alignment> is a power of 2, which is technically not guaranteed */
144 p = p + ((-p - SIZEOF_StgArrWords) & (alignment - 1));
146 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
147 StgArrWords_words(p) = payload_words;
153 W_ words, n, init, arr, p;
154 /* Args: R1 = words, R2 = initialisation value */
157 MAYBE_GC(R2_PTR,newArrayzh_fast);
159 words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + n;
160 ("ptr" arr) = foreign "C" allocateLocal(MyCapability() "ptr",words) [R2];
161 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(n), 0);
163 SET_HDR(arr, stg_MUT_ARR_PTRS_DIRTY_info, W_[CCCS]);
164 StgMutArrPtrs_ptrs(arr) = n;
166 // Initialise all elements of the the array with the value in R2
168 p = arr + SIZEOF_StgMutArrPtrs;
170 if (p < arr + WDS(words)) {
179 unsafeThawArrayzh_fast
181 // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
183 // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
184 // normally doesn't. However, when we freeze a MUT_ARR_PTRS, we leave
185 // it on the mutable list for the GC to remove (removing something from
186 // the mutable list is not easy, because the mut_list is only singly-linked).
188 // So that we can tell whether a MUT_ARR_PTRS_FROZEN is on the mutable list,
189 // when we freeze it we set the info ptr to be MUT_ARR_PTRS_FROZEN0
190 // to indicate that it is still on the mutable list.
192 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
193 // either it is on a mut_list, or it isn't. We adopt the convention that
194 // the closure type is MUT_ARR_PTRS_FROZEN0 if it is on the mutable list,
195 // and MUT_ARR_PTRS_FROZEN otherwise. In fact it wouldn't matter if
196 // we put it on the mutable list more than once, but it would get scavenged
197 // multiple times during GC, which would be unnecessarily slow.
199 if (StgHeader_info(R1) != stg_MUT_ARR_PTRS_FROZEN0_info) {
200 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
201 recordMutable(R1, R1);
202 // must be done after SET_INFO, because it ASSERTs closure_MUTABLE()
205 SET_INFO(R1,stg_MUT_ARR_PTRS_DIRTY_info);
210 /* -----------------------------------------------------------------------------
212 -------------------------------------------------------------------------- */
217 /* Args: R1 = initialisation value */
219 ALLOC_PRIM( SIZEOF_StgMutVar, R1_PTR, newMutVarzh_fast);
221 mv = Hp - SIZEOF_StgMutVar + WDS(1);
222 SET_HDR(mv,stg_MUT_VAR_DIRTY_info,W_[CCCS]);
223 StgMutVar_var(mv) = R1;
228 atomicModifyMutVarzh_fast
230 W_ mv, f, z, x, y, r, h;
231 /* Args: R1 :: MutVar#, R2 :: a -> (a,b) */
233 /* If x is the current contents of the MutVar#, then
234 We want to make the new contents point to
238 and the return value is
242 obviously we can share (f x).
244 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
245 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
246 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
250 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
251 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),WDS(MIN_UPD_SIZE-1))
253 #define THUNK_1_SIZE (SIZEOF_StgThunkHeader + WDS(1))
254 #define TICK_ALLOC_THUNK_1() TICK_ALLOC_UP_THK(WDS(1),0)
258 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(MIN_UPD_SIZE))
259 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),WDS(MIN_UPD_SIZE-2))
261 #define THUNK_2_SIZE (SIZEOF_StgThunkHeader + WDS(2))
262 #define TICK_ALLOC_THUNK_2() TICK_ALLOC_UP_THK(WDS(2),0)
265 #define SIZE (THUNK_2_SIZE + THUNK_1_SIZE + THUNK_1_SIZE)
267 HP_CHK_GEN_TICKY(SIZE, R1_PTR & R2_PTR, atomicModifyMutVarzh_fast);
272 TICK_ALLOC_THUNK_2();
273 CCCS_ALLOC(THUNK_2_SIZE);
274 z = Hp - THUNK_2_SIZE + WDS(1);
275 SET_HDR(z, stg_ap_2_upd_info, W_[CCCS]);
276 LDV_RECORD_CREATE(z);
277 StgThunk_payload(z,0) = f;
279 TICK_ALLOC_THUNK_1();
280 CCCS_ALLOC(THUNK_1_SIZE);
281 y = z - THUNK_1_SIZE;
282 SET_HDR(y, stg_sel_0_upd_info, W_[CCCS]);
283 LDV_RECORD_CREATE(y);
284 StgThunk_payload(y,0) = z;
286 TICK_ALLOC_THUNK_1();
287 CCCS_ALLOC(THUNK_1_SIZE);
288 r = y - THUNK_1_SIZE;
289 SET_HDR(r, stg_sel_1_upd_info, W_[CCCS]);
290 LDV_RECORD_CREATE(r);
291 StgThunk_payload(r,0) = z;
294 x = StgMutVar_var(mv);
295 StgThunk_payload(z,1) = x;
297 (h) = foreign "C" cas(mv + SIZEOF_StgHeader + OFFSET_StgMutVar_var, x, y) [];
298 if (h != x) { goto retry; }
300 StgMutVar_var(mv) = y;
303 if (GET_INFO(mv) == stg_MUT_VAR_CLEAN_info) {
304 foreign "C" dirty_MUT_VAR(BaseReg "ptr", mv "ptr") [];
310 /* -----------------------------------------------------------------------------
311 Weak Pointer Primitives
312 -------------------------------------------------------------------------- */
314 STRING(stg_weak_msg,"New weak pointer at %p\n")
320 R3 = finalizer (or NULL)
325 R3 = stg_NO_FINALIZER_closure;
328 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR & R3_PTR, mkWeakzh_fast );
330 w = Hp - SIZEOF_StgWeak + WDS(1);
331 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
333 // We don't care about cfinalizer here.
334 // Should StgWeak_cfinalizer(w) be stg_NO_FINALIZER_closure or
338 StgWeak_value(w) = R2;
339 StgWeak_finalizer(w) = R3;
340 StgWeak_cfinalizer(w) = stg_NO_FINALIZER_closure;
342 StgWeak_link(w) = W_[weak_ptr_list];
343 W_[weak_ptr_list] = w;
345 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
350 mkWeakForeignEnvzh_fast
356 R5 = has environment (0 or 1)
359 W_ w, payload_words, words, p;
361 W_ key, val, fptr, ptr, flag, eptr;
370 ALLOC_PRIM( SIZEOF_StgWeak, R1_PTR & R2_PTR, mkWeakForeignEnvzh_fast );
372 w = Hp - SIZEOF_StgWeak + WDS(1);
373 SET_HDR(w, stg_WEAK_info, W_[CCCS]);
376 words = BYTES_TO_WDS(SIZEOF_StgArrWords) + payload_words;
377 ("ptr" p) = foreign "C" allocateLocal(MyCapability() "ptr", words) [];
379 TICK_ALLOC_PRIM(SIZEOF_StgArrWords,WDS(payload_words),0);
380 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
382 StgArrWords_words(p) = payload_words;
383 StgArrWords_payload(p,0) = fptr;
384 StgArrWords_payload(p,1) = ptr;
385 StgArrWords_payload(p,2) = eptr;
386 StgArrWords_payload(p,3) = flag;
388 // We don't care about the value here.
389 // Should StgWeak_value(w) be stg_NO_FINALIZER_closure or something else?
391 StgWeak_key(w) = key;
392 StgWeak_value(w) = val;
393 StgWeak_finalizer(w) = stg_NO_FINALIZER_closure;
394 StgWeak_cfinalizer(w) = p;
396 StgWeak_link(w) = W_[weak_ptr_list];
397 W_[weak_ptr_list] = w;
399 IF_DEBUG(weak, foreign "C" debugBelch(stg_weak_msg,w) []);
413 if (GET_INFO(w) == stg_DEAD_WEAK_info) {
414 RET_NP(0,stg_NO_FINALIZER_closure);
420 // A weak pointer is inherently used, so we do not need to call
421 // LDV_recordDead_FILL_SLOP_DYNAMIC():
422 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
423 // or, LDV_recordDead():
424 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
425 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
426 // large as weak pointers, so there is no need to fill the slop, either.
427 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
431 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
433 SET_INFO(w,stg_DEAD_WEAK_info);
434 LDV_RECORD_CREATE(w);
436 f = StgWeak_finalizer(w);
437 arr = StgWeak_cfinalizer(w);
439 StgDeadWeak_link(w) = StgWeak_link(w);
441 if (arr != stg_NO_FINALIZER_closure) {
442 foreign "C" runCFinalizer(StgArrWords_payload(arr,0),
443 StgArrWords_payload(arr,1),
444 StgArrWords_payload(arr,2),
445 StgArrWords_payload(arr,3)) [];
448 /* return the finalizer */
449 if (f == stg_NO_FINALIZER_closure) {
450 RET_NP(0,stg_NO_FINALIZER_closure);
462 if (GET_INFO(w) == stg_WEAK_info) {
464 val = StgWeak_value(w);
472 /* -----------------------------------------------------------------------------
473 Arbitrary-precision Integer operations.
475 There are some assumptions in this code that mp_limb_t == W_. This is
476 the case for all the platforms that GHC supports, currently.
477 -------------------------------------------------------------------------- */
481 /* arguments: R1 = Int# */
483 W_ val, s, p; /* to avoid aliasing */
486 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, int2Integerzh_fast );
488 p = Hp - SIZEOF_StgArrWords;
489 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
490 StgArrWords_words(p) = 1;
492 /* mpz_set_si is inlined here, makes things simpler */
505 /* returns (# size :: Int#,
514 /* arguments: R1 = Word# */
516 W_ val, s, p; /* to avoid aliasing */
520 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(1), NO_PTRS, word2Integerzh_fast);
522 p = Hp - SIZEOF_StgArrWords;
523 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
524 StgArrWords_words(p) = 1;
533 /* returns (# size :: Int#,
534 data :: ByteArray# #)
541 * 'long long' primops for converting to/from Integers.
544 #ifdef SUPPORT_LONG_LONGS
546 int64ToIntegerzh_fast
548 /* arguments: L1 = Int64# */
551 W_ hi, lo, s, neg, words_needed, p;
556 hi = TO_W_(val >> 32);
559 if ( hi == 0 || (hi == 0xFFFFFFFF && lo != 0) ) {
560 // minimum is one word
566 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
567 NO_PTRS, int64ToIntegerzh_fast );
569 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
570 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
571 StgArrWords_words(p) = words_needed;
583 if ( words_needed == 2 ) {
591 } else /* val==0 */ {
599 /* returns (# size :: Int#,
600 data :: ByteArray# #)
604 word64ToIntegerzh_fast
606 /* arguments: L1 = Word64# */
609 W_ hi, lo, s, words_needed, p;
612 hi = TO_W_(val >> 32);
621 ALLOC_PRIM( SIZEOF_StgArrWords + WDS(words_needed),
622 NO_PTRS, word64ToIntegerzh_fast );
624 p = Hp - SIZEOF_StgArrWords - WDS(words_needed) + WDS(1);
625 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
626 StgArrWords_words(p) = words_needed;
636 } else /* val==0 */ {
641 /* returns (# size :: Int#,
642 data :: ByteArray# #)
649 #endif /* SUPPORT_LONG_LONGS */
651 /* ToDo: this is shockingly inefficient */
656 bits8 [SIZEOF_MP_INT];
661 bits8 [SIZEOF_MP_INT];
666 bits8 [SIZEOF_MP_INT];
671 bits8 [SIZEOF_MP_INT];
676 #define FETCH_MP_TEMP(X) \
678 X = BaseReg + (OFFSET_StgRegTable_r ## X);
680 #define FETCH_MP_TEMP(X) /* Nothing */
683 #define GMP_TAKE2_RET1(name,mp_fun) \
688 FETCH_MP_TEMP(mp_tmp1); \
689 FETCH_MP_TEMP(mp_tmp2); \
690 FETCH_MP_TEMP(mp_result1) \
691 FETCH_MP_TEMP(mp_result2); \
693 /* call doYouWantToGC() */ \
694 MAYBE_GC(R2_PTR & R4_PTR, name); \
701 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
702 MP_INT__mp_size(mp_tmp1) = (s1); \
703 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
704 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
705 MP_INT__mp_size(mp_tmp2) = (s2); \
706 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
708 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
710 /* Perform the operation */ \
711 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
713 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
714 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
717 #define GMP_TAKE1_RET1(name,mp_fun) \
722 FETCH_MP_TEMP(mp_tmp1); \
723 FETCH_MP_TEMP(mp_result1) \
725 /* call doYouWantToGC() */ \
726 MAYBE_GC(R2_PTR, name); \
731 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
732 MP_INT__mp_size(mp_tmp1) = (s1); \
733 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
735 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
737 /* Perform the operation */ \
738 foreign "C" mp_fun(mp_result1 "ptr",mp_tmp1 "ptr") []; \
740 RET_NP(TO_W_(MP_INT__mp_size(mp_result1)), \
741 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords); \
744 #define GMP_TAKE2_RET2(name,mp_fun) \
749 FETCH_MP_TEMP(mp_tmp1); \
750 FETCH_MP_TEMP(mp_tmp2); \
751 FETCH_MP_TEMP(mp_result1) \
752 FETCH_MP_TEMP(mp_result2) \
754 /* call doYouWantToGC() */ \
755 MAYBE_GC(R2_PTR & R4_PTR, name); \
762 MP_INT__mp_alloc(mp_tmp1) = W_TO_INT(StgArrWords_words(d1)); \
763 MP_INT__mp_size(mp_tmp1) = (s1); \
764 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(d1); \
765 MP_INT__mp_alloc(mp_tmp2) = W_TO_INT(StgArrWords_words(d2)); \
766 MP_INT__mp_size(mp_tmp2) = (s2); \
767 MP_INT__mp_d(mp_tmp2) = BYTE_ARR_CTS(d2); \
769 foreign "C" __gmpz_init(mp_result1 "ptr") []; \
770 foreign "C" __gmpz_init(mp_result2 "ptr") []; \
772 /* Perform the operation */ \
773 foreign "C" mp_fun(mp_result1 "ptr",mp_result2 "ptr",mp_tmp1 "ptr",mp_tmp2 "ptr") []; \
775 RET_NPNP(TO_W_(MP_INT__mp_size(mp_result1)), \
776 MP_INT__mp_d(mp_result1) - SIZEOF_StgArrWords, \
777 TO_W_(MP_INT__mp_size(mp_result2)), \
778 MP_INT__mp_d(mp_result2) - SIZEOF_StgArrWords); \
781 GMP_TAKE2_RET1(plusIntegerzh_fast, __gmpz_add)
782 GMP_TAKE2_RET1(minusIntegerzh_fast, __gmpz_sub)
783 GMP_TAKE2_RET1(timesIntegerzh_fast, __gmpz_mul)
784 GMP_TAKE2_RET1(gcdIntegerzh_fast, __gmpz_gcd)
785 GMP_TAKE2_RET1(quotIntegerzh_fast, __gmpz_tdiv_q)
786 GMP_TAKE2_RET1(remIntegerzh_fast, __gmpz_tdiv_r)
787 GMP_TAKE2_RET1(divExactIntegerzh_fast, __gmpz_divexact)
788 GMP_TAKE2_RET1(andIntegerzh_fast, __gmpz_and)
789 GMP_TAKE2_RET1(orIntegerzh_fast, __gmpz_ior)
790 GMP_TAKE2_RET1(xorIntegerzh_fast, __gmpz_xor)
791 GMP_TAKE1_RET1(complementIntegerzh_fast, __gmpz_com)
793 GMP_TAKE2_RET2(quotRemIntegerzh_fast, __gmpz_tdiv_qr)
794 GMP_TAKE2_RET2(divModIntegerzh_fast, __gmpz_fdiv_qr)
798 mp_tmp_w: W_; // NB. mp_tmp_w is really an here mp_limb_t
804 /* R1 = the first Int#; R2 = the second Int# */
806 FETCH_MP_TEMP(mp_tmp_w);
809 (r) = foreign "C" __gmpn_gcd_1(mp_tmp_w "ptr", 1, R2) [];
812 /* Result parked in R1, return via info-pointer at TOS */
813 jump %ENTRY_CODE(Sp(0));
819 /* R1 = s1; R2 = d1; R3 = the int */
821 (s1) = foreign "C" __gmpn_gcd_1( BYTE_ARR_CTS(R2) "ptr", R1, R3) [];
824 /* Result parked in R1, return via info-pointer at TOS */
825 jump %ENTRY_CODE(Sp(0));
831 /* R1 = s1; R2 = d1; R3 = the int */
832 W_ usize, vsize, v_digit, u_digit;
838 // paraphrased from __gmpz_cmp_si() in the GMP sources
839 if (%gt(v_digit,0)) {
842 if (%lt(v_digit,0)) {
848 if (usize != vsize) {
850 jump %ENTRY_CODE(Sp(0));
855 jump %ENTRY_CODE(Sp(0));
858 u_digit = W_[BYTE_ARR_CTS(R2)];
860 if (u_digit == v_digit) {
862 jump %ENTRY_CODE(Sp(0));
865 if (%gtu(u_digit,v_digit)) { // NB. unsigned: these are mp_limb_t's
871 jump %ENTRY_CODE(Sp(0));
876 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
877 W_ usize, vsize, size, up, vp;
880 // paraphrased from __gmpz_cmp() in the GMP sources
884 if (usize != vsize) {
886 jump %ENTRY_CODE(Sp(0));
891 jump %ENTRY_CODE(Sp(0));
894 if (%lt(usize,0)) { // NB. not <, which is unsigned
900 up = BYTE_ARR_CTS(R2);
901 vp = BYTE_ARR_CTS(R4);
903 (cmp) = foreign "C" __gmpn_cmp(up "ptr", vp "ptr", size) [];
905 if (cmp == 0 :: CInt) {
907 jump %ENTRY_CODE(Sp(0));
910 if (%lt(cmp,0 :: CInt) == %lt(usize,0)) {
915 /* Result parked in R1, return via info-pointer at TOS */
916 jump %ENTRY_CODE(Sp(0));
928 r = W_[R2 + SIZEOF_StgArrWords];
933 /* Result parked in R1, return via info-pointer at TOS */
935 jump %ENTRY_CODE(Sp(0));
947 r = W_[R2 + SIZEOF_StgArrWords];
952 /* Result parked in R1, return via info-pointer at TOS */
954 jump %ENTRY_CODE(Sp(0));
957 decodeFloatzuIntzh_fast
961 FETCH_MP_TEMP(mp_tmp1);
962 FETCH_MP_TEMP(mp_tmp_w);
964 /* arguments: F1 = Float# */
967 /* Perform the operation */
968 foreign "C" __decodeFloat_Int(mp_tmp1 "ptr", mp_tmp_w "ptr", arg) [];
970 /* returns: (Int# (mantissa), Int# (exponent)) */
971 RET_NN(W_[mp_tmp1], W_[mp_tmp_w]);
974 #define DOUBLE_MANTISSA_SIZE SIZEOF_DOUBLE
975 #define ARR_SIZE (SIZEOF_StgArrWords + DOUBLE_MANTISSA_SIZE)
981 FETCH_MP_TEMP(mp_tmp1);
982 FETCH_MP_TEMP(mp_tmp_w);
984 /* arguments: D1 = Double# */
987 ALLOC_PRIM( ARR_SIZE, NO_PTRS, decodeDoublezh_fast );
989 /* Be prepared to tell Lennart-coded __decodeDouble
990 where mantissa.d can be put (it does not care about the rest) */
991 p = Hp - ARR_SIZE + WDS(1);
992 SET_HDR(p, stg_ARR_WORDS_info, W_[CCCS]);
993 StgArrWords_words(p) = BYTES_TO_WDS(DOUBLE_MANTISSA_SIZE);
994 MP_INT__mp_d(mp_tmp1) = BYTE_ARR_CTS(p);
996 /* Perform the operation */
997 foreign "C" __decodeDouble(mp_tmp1 "ptr", mp_tmp_w "ptr",arg) [];
999 /* returns: (Int# (expn), Int#, ByteArray#) */
1000 RET_NNP(W_[mp_tmp_w], TO_W_(MP_INT__mp_size(mp_tmp1)), p);
1003 decodeDoublezu2Intzh_fast
1007 FETCH_MP_TEMP(mp_tmp1);
1008 FETCH_MP_TEMP(mp_tmp2);
1009 FETCH_MP_TEMP(mp_result1);
1010 FETCH_MP_TEMP(mp_result2);
1012 /* arguments: D1 = Double# */
1015 /* Perform the operation */
1016 foreign "C" __decodeDouble_2Int(mp_tmp1 "ptr", mp_tmp2 "ptr",
1017 mp_result1 "ptr", mp_result2 "ptr",
1021 (Int# (mant sign), Word# (mant high), Word# (mant low), Int# (expn)) */
1022 RET_NNNN(W_[mp_tmp1], W_[mp_tmp2], W_[mp_result1], W_[mp_result2]);
1025 /* -----------------------------------------------------------------------------
1026 * Concurrency primitives
1027 * -------------------------------------------------------------------------- */
1031 /* args: R1 = closure to spark */
1033 MAYBE_GC(R1_PTR, forkzh_fast);
1039 ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr",
1040 RtsFlags_GcFlags_initialStkSize(RtsFlags),
1043 /* start blocked if the current thread is blocked */
1044 StgTSO_flags(threadid) =
1045 StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) &
1046 (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));
1048 foreign "C" scheduleThread(MyCapability() "ptr", threadid "ptr") [];
1050 // context switch soon, but not immediately: we don't want every
1051 // forkIO to force a context-switch.
1052 Capability_context_switch(MyCapability()) = 1 :: CInt;
1059 /* args: R1 = cpu, R2 = closure to spark */
1061 MAYBE_GC(R2_PTR, forkOnzh_fast);
1069 ("ptr" threadid) = foreign "C" createIOThread( MyCapability() "ptr",
1070 RtsFlags_GcFlags_initialStkSize(RtsFlags),
1073 /* start blocked if the current thread is blocked */
1074 StgTSO_flags(threadid) =
1075 StgTSO_flags(threadid) | (StgTSO_flags(CurrentTSO) &
1076 (TSO_BLOCKEX::I32 | TSO_INTERRUPTIBLE::I32));
1078 foreign "C" scheduleThreadOn(MyCapability() "ptr", cpu, threadid "ptr") [];
1080 // context switch soon, but not immediately: we don't want every
1081 // forkIO to force a context-switch.
1082 Capability_context_switch(MyCapability()) = 1 :: CInt;
1089 jump stg_yield_noregs;
1104 foreign "C" labelThread(R1 "ptr", R2 "ptr") [];
1106 jump %ENTRY_CODE(Sp(0));
1109 isCurrentThreadBoundzh_fast
1113 (r) = foreign "C" isThreadBound(CurrentTSO) [];
1119 /* args: R1 :: ThreadId# */
1127 if (TO_W_(StgTSO_what_next(tso)) == ThreadRelocated) {
1128 tso = StgTSO__link(tso);
1132 what_next = TO_W_(StgTSO_what_next(tso));
1133 why_blocked = TO_W_(StgTSO_why_blocked(tso));
1134 // Note: these two reads are not atomic, so they might end up
1135 // being inconsistent. It doesn't matter, since we
1136 // only return one or the other. If we wanted to return the
1137 // contents of block_info too, then we'd have to do some synchronisation.
1139 if (what_next == ThreadComplete) {
1140 ret = 16; // NB. magic, matches up with GHC.Conc.threadStatus
1142 if (what_next == ThreadKilled) {
1151 /* -----------------------------------------------------------------------------
1153 * -------------------------------------------------------------------------- */
1157 // Catch retry frame ------------------------------------------------------------
1159 INFO_TABLE_RET(stg_catch_retry_frame, CATCH_RETRY_FRAME,
1160 #if defined(PROFILING)
1161 W_ unused1, W_ unused2,
1163 W_ unused3, P_ unused4, P_ unused5)
1165 W_ r, frame, trec, outer;
1168 trec = StgTSO_trec(CurrentTSO);
1169 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1170 (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
1172 /* Succeeded (either first branch or second branch) */
1173 StgTSO_trec(CurrentTSO) = outer;
1174 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1175 jump %ENTRY_CODE(Sp(SP_OFF));
1177 /* Did not commit: re-execute */
1179 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1180 StgTSO_trec(CurrentTSO) = new_trec;
1181 if (StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
1182 R1 = StgCatchRetryFrame_alt_code(frame);
1184 R1 = StgCatchRetryFrame_first_code(frame);
1191 // Atomically frame ------------------------------------------------------------
1193 INFO_TABLE_RET(stg_atomically_frame, ATOMICALLY_FRAME,
1194 #if defined(PROFILING)
1195 W_ unused1, W_ unused2,
1197 P_ unused3, P_ unused4)
1199 W_ frame, trec, valid, next_invariant, q, outer;
1202 trec = StgTSO_trec(CurrentTSO);
1203 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1205 if (outer == NO_TREC) {
1206 /* First time back at the atomically frame -- pick up invariants */
1207 ("ptr" q) = foreign "C" stmGetInvariantsToCheck(MyCapability() "ptr", trec "ptr") [];
1208 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
1211 /* Second/subsequent time back at the atomically frame -- abort the
1212 * tx that's checking the invariant and move on to the next one */
1213 StgTSO_trec(CurrentTSO) = outer;
1214 q = StgAtomicallyFrame_next_invariant_to_check(frame);
1215 StgInvariantCheckQueue_my_execution(q) = trec;
1216 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1217 /* Don't free trec -- it's linked from q and will be stashed in the
1218 * invariant if we eventually commit. */
1219 q = StgInvariantCheckQueue_next_queue_entry(q);
1220 StgAtomicallyFrame_next_invariant_to_check(frame) = q;
1224 q = StgAtomicallyFrame_next_invariant_to_check(frame);
1226 if (q != END_INVARIANT_CHECK_QUEUE) {
1227 /* We can't commit yet: another invariant to check */
1228 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [];
1229 StgTSO_trec(CurrentTSO) = trec;
1231 next_invariant = StgInvariantCheckQueue_invariant(q);
1232 R1 = StgAtomicInvariant_code(next_invariant);
1237 /* We've got no more invariants to check, try to commit */
1238 (valid) = foreign "C" stmCommitTransaction(MyCapability() "ptr", trec "ptr") [];
1240 /* Transaction was valid: commit succeeded */
1241 StgTSO_trec(CurrentTSO) = NO_TREC;
1242 Sp = Sp + SIZEOF_StgAtomicallyFrame;
1243 jump %ENTRY_CODE(Sp(SP_OFF));
1245 /* Transaction was not valid: try again */
1246 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1247 StgTSO_trec(CurrentTSO) = trec;
1248 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
1249 R1 = StgAtomicallyFrame_code(frame);
1255 INFO_TABLE_RET(stg_atomically_waiting_frame, ATOMICALLY_FRAME,
1256 #if defined(PROFILING)
1257 W_ unused1, W_ unused2,
1259 P_ unused3, P_ unused4)
1261 W_ frame, trec, valid;
1265 /* The TSO is currently waiting: should we stop waiting? */
1266 (valid) = foreign "C" stmReWait(MyCapability() "ptr", CurrentTSO "ptr") [];
1268 /* Previous attempt is still valid: no point trying again yet */
1269 jump stg_block_noregs;
1271 /* Previous attempt is no longer valid: try again */
1272 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", NO_TREC "ptr") [];
1273 StgTSO_trec(CurrentTSO) = trec;
1274 StgHeader_info(frame) = stg_atomically_frame_info;
1275 R1 = StgAtomicallyFrame_code(frame);
1280 // STM catch frame --------------------------------------------------------------
1284 /* Catch frames are very similar to update frames, but when entering
1285 * one we just pop the frame off the stack and perform the correct
1286 * kind of return to the activation record underneath us on the stack.
1289 INFO_TABLE_RET(stg_catch_stm_frame, CATCH_STM_FRAME,
1290 #if defined(PROFILING)
1291 W_ unused1, W_ unused2,
1293 P_ unused3, P_ unused4)
1295 W_ r, frame, trec, outer;
1297 trec = StgTSO_trec(CurrentTSO);
1298 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1299 (r) = foreign "C" stmCommitNestedTransaction(MyCapability() "ptr", trec "ptr") [];
1301 /* Commit succeeded */
1302 StgTSO_trec(CurrentTSO) = outer;
1303 Sp = Sp + SIZEOF_StgCatchSTMFrame;
1308 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1309 StgTSO_trec(CurrentTSO) = new_trec;
1310 R1 = StgCatchSTMFrame_code(frame);
1316 // Primop definition ------------------------------------------------------------
1324 // stmStartTransaction may allocate
1325 MAYBE_GC (R1_PTR, atomicallyzh_fast);
1327 /* Args: R1 = m :: STM a */
1328 STK_CHK_GEN(SIZEOF_StgAtomicallyFrame + WDS(1), R1_PTR, atomicallyzh_fast);
1330 old_trec = StgTSO_trec(CurrentTSO);
1332 /* Nested transactions are not allowed; raise an exception */
1333 if (old_trec != NO_TREC) {
1334 R1 = base_ControlziExceptionziBase_nestedAtomically_closure;
1338 /* Set up the atomically frame */
1339 Sp = Sp - SIZEOF_StgAtomicallyFrame;
1342 SET_HDR(frame,stg_atomically_frame_info, W_[CCCS]);
1343 StgAtomicallyFrame_code(frame) = R1;
1344 StgAtomicallyFrame_next_invariant_to_check(frame) = END_INVARIANT_CHECK_QUEUE;
1346 /* Start the memory transcation */
1347 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", old_trec "ptr") [R1];
1348 StgTSO_trec(CurrentTSO) = new_trec;
1350 /* Apply R1 to the realworld token */
1359 /* Args: R1 :: STM a */
1360 /* Args: R2 :: Exception -> STM a */
1361 STK_CHK_GEN(SIZEOF_StgCatchSTMFrame + WDS(1), R1_PTR & R2_PTR, catchSTMzh_fast);
1363 /* Set up the catch frame */
1364 Sp = Sp - SIZEOF_StgCatchSTMFrame;
1367 SET_HDR(frame, stg_catch_stm_frame_info, W_[CCCS]);
1368 StgCatchSTMFrame_handler(frame) = R2;
1369 StgCatchSTMFrame_code(frame) = R1;
1371 /* Start a nested transaction to run the body of the try block in */
1374 cur_trec = StgTSO_trec(CurrentTSO);
1375 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", cur_trec "ptr");
1376 StgTSO_trec(CurrentTSO) = new_trec;
1378 /* Apply R1 to the realworld token */
1389 // stmStartTransaction may allocate
1390 MAYBE_GC (R1_PTR & R2_PTR, catchRetryzh_fast);
1392 /* Args: R1 :: STM a */
1393 /* Args: R2 :: STM a */
1394 STK_CHK_GEN(SIZEOF_StgCatchRetryFrame + WDS(1), R1_PTR & R2_PTR, catchRetryzh_fast);
1396 /* Start a nested transaction within which to run the first code */
1397 trec = StgTSO_trec(CurrentTSO);
1398 ("ptr" new_trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", trec "ptr") [R1,R2];
1399 StgTSO_trec(CurrentTSO) = new_trec;
1401 /* Set up the catch-retry frame */
1402 Sp = Sp - SIZEOF_StgCatchRetryFrame;
1405 SET_HDR(frame, stg_catch_retry_frame_info, W_[CCCS]);
1406 StgCatchRetryFrame_running_alt_code(frame) = 0 :: CInt; // false;
1407 StgCatchRetryFrame_first_code(frame) = R1;
1408 StgCatchRetryFrame_alt_code(frame) = R2;
1410 /* Apply R1 to the realworld token */
1423 MAYBE_GC (NO_PTRS, retryzh_fast); // STM operations may allocate
1425 // Find the enclosing ATOMICALLY_FRAME or CATCH_RETRY_FRAME
1427 StgTSO_sp(CurrentTSO) = Sp;
1428 (frame_type) = foreign "C" findRetryFrameHelper(CurrentTSO "ptr") [];
1429 Sp = StgTSO_sp(CurrentTSO);
1431 trec = StgTSO_trec(CurrentTSO);
1432 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1434 if (frame_type == CATCH_RETRY_FRAME) {
1435 // The retry reaches a CATCH_RETRY_FRAME before the atomic frame
1436 ASSERT(outer != NO_TREC);
1437 // Abort the transaction attempting the current branch
1438 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1439 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
1440 if (!StgCatchRetryFrame_running_alt_code(frame) != 0::I32) {
1441 // Retry in the first branch: try the alternative
1442 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1443 StgTSO_trec(CurrentTSO) = trec;
1444 StgCatchRetryFrame_running_alt_code(frame) = 1 :: CInt; // true;
1445 R1 = StgCatchRetryFrame_alt_code(frame);
1448 // Retry in the alternative code: propagate the retry
1449 StgTSO_trec(CurrentTSO) = outer;
1450 Sp = Sp + SIZEOF_StgCatchRetryFrame;
1451 goto retry_pop_stack;
1455 // We've reached the ATOMICALLY_FRAME: attempt to wait
1456 ASSERT(frame_type == ATOMICALLY_FRAME);
1457 if (outer != NO_TREC) {
1458 // We called retry while checking invariants, so abort the current
1459 // invariant check (merging its TVar accesses into the parents read
1460 // set so we'll wait on them)
1461 foreign "C" stmAbortTransaction(MyCapability() "ptr", trec "ptr") [];
1462 foreign "C" stmFreeAbortedTRec(MyCapability() "ptr", trec "ptr") [];
1464 StgTSO_trec(CurrentTSO) = trec;
1465 ("ptr" outer) = foreign "C" stmGetEnclosingTRec(trec "ptr") [];
1467 ASSERT(outer == NO_TREC);
1469 (r) = foreign "C" stmWait(MyCapability() "ptr", CurrentTSO "ptr", trec "ptr") [];
1471 // Transaction was valid: stmWait put us on the TVars' queues, we now block
1472 StgHeader_info(frame) = stg_atomically_waiting_frame_info;
1474 // Fix up the stack in the unregisterised case: the return convention is different.
1475 R3 = trec; // passing to stmWaitUnblock()
1476 jump stg_block_stmwait;
1478 // Transaction was not valid: retry immediately
1479 ("ptr" trec) = foreign "C" stmStartTransaction(MyCapability() "ptr", outer "ptr") [];
1480 StgTSO_trec(CurrentTSO) = trec;
1481 R1 = StgAtomicallyFrame_code(frame);
1492 /* Args: R1 = invariant closure */
1493 MAYBE_GC (R1_PTR, checkzh_fast);
1495 trec = StgTSO_trec(CurrentTSO);
1497 foreign "C" stmAddInvariantToCheck(MyCapability() "ptr",
1501 jump %ENTRY_CODE(Sp(0));
1510 /* Args: R1 = initialisation value */
1512 MAYBE_GC (R1_PTR, newTVarzh_fast);
1514 ("ptr" tv) = foreign "C" stmNewTVar(MyCapability() "ptr", new_value "ptr") [];
1525 /* Args: R1 = TVar closure */
1527 MAYBE_GC (R1_PTR, readTVarzh_fast); // Call to stmReadTVar may allocate
1528 trec = StgTSO_trec(CurrentTSO);
1530 ("ptr" result) = foreign "C" stmReadTVar(MyCapability() "ptr", trec "ptr", tvar "ptr") [];
1540 result = StgTVar_current_value(R1);
1541 if (%INFO_PTR(result) == stg_TREC_HEADER_info) {
1553 /* Args: R1 = TVar closure */
1554 /* R2 = New value */
1556 MAYBE_GC (R1_PTR & R2_PTR, writeTVarzh_fast); // Call to stmWriteTVar may allocate
1557 trec = StgTSO_trec(CurrentTSO);
1560 foreign "C" stmWriteTVar(MyCapability() "ptr", trec "ptr", tvar "ptr", new_value "ptr") [];
1562 jump %ENTRY_CODE(Sp(0));
1566 /* -----------------------------------------------------------------------------
1569 * take & putMVar work as follows. Firstly, an important invariant:
1571 * If the MVar is full, then the blocking queue contains only
1572 * threads blocked on putMVar, and if the MVar is empty then the
1573 * blocking queue contains only threads blocked on takeMVar.
1576 * MVar empty : then add ourselves to the blocking queue
1577 * MVar full : remove the value from the MVar, and
1578 * blocking queue empty : return
1579 * blocking queue non-empty : perform the first blocked putMVar
1580 * from the queue, and wake up the
1581 * thread (MVar is now full again)
1583 * putMVar is just the dual of the above algorithm.
1585 * How do we "perform a putMVar"? Well, we have to fiddle around with
1586 * the stack of the thread waiting to do the putMVar. See
1587 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1588 * the stack layout, and the PerformPut and PerformTake macros below.
1590 * It is important that a blocked take or put is woken up with the
1591 * take/put already performed, because otherwise there would be a
1592 * small window of vulnerability where the thread could receive an
1593 * exception and never perform its take or put, and we'd end up with a
1596 * -------------------------------------------------------------------------- */
1600 /* args: R1 = MVar closure */
1602 if (StgMVar_value(R1) == stg_END_TSO_QUEUE_closure) {
1614 ALLOC_PRIM ( SIZEOF_StgMVar, NO_PTRS, newMVarzh_fast );
1616 mvar = Hp - SIZEOF_StgMVar + WDS(1);
1617 SET_HDR(mvar,stg_MVAR_DIRTY_info,W_[CCCS]);
1618 // MVARs start dirty: generation 0 has no mutable list
1619 StgMVar_head(mvar) = stg_END_TSO_QUEUE_closure;
1620 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1621 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1626 #define PerformTake(tso, value) \
1627 W_[StgTSO_sp(tso) + WDS(1)] = value; \
1628 W_[StgTSO_sp(tso) + WDS(0)] = stg_gc_unpt_r1_info;
1630 #define PerformPut(tso,lval) \
1631 StgTSO_sp(tso) = StgTSO_sp(tso) + WDS(3); \
1632 lval = W_[StgTSO_sp(tso) - WDS(1)];
1636 W_ mvar, val, info, tso;
1638 /* args: R1 = MVar closure */
1641 #if defined(THREADED_RTS)
1642 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1644 info = GET_INFO(mvar);
1647 if (info == stg_MVAR_CLEAN_info) {
1648 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr") [];
1651 /* If the MVar is empty, put ourselves on its blocking queue,
1652 * and wait until we're woken up.
1654 if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1655 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1656 StgMVar_head(mvar) = CurrentTSO;
1658 foreign "C" setTSOLink(MyCapability() "ptr",
1659 StgMVar_tail(mvar) "ptr",
1662 StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1663 StgTSO_block_info(CurrentTSO) = mvar;
1664 // write barrier for throwTo(), which looks at block_info
1665 // if why_blocked==BlockedOnMVar.
1666 prim %write_barrier() [];
1667 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1668 StgMVar_tail(mvar) = CurrentTSO;
1671 jump stg_block_takemvar;
1674 /* we got the value... */
1675 val = StgMVar_value(mvar);
1677 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure)
1679 /* There are putMVar(s) waiting...
1680 * wake up the first thread on the queue
1682 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1684 /* actually perform the putMVar for the thread that we just woke up */
1685 tso = StgMVar_head(mvar);
1686 PerformPut(tso,StgMVar_value(mvar));
1688 if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1689 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1692 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1693 StgMVar_head(mvar) "ptr", 1) [];
1694 StgMVar_head(mvar) = tso;
1696 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1697 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1700 #if defined(THREADED_RTS)
1701 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1703 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1709 /* No further putMVars, MVar is now empty */
1710 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1712 #if defined(THREADED_RTS)
1713 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1715 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1725 W_ mvar, val, info, tso;
1727 /* args: R1 = MVar closure */
1731 #if defined(THREADED_RTS)
1732 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1734 info = GET_INFO(mvar);
1737 if (StgMVar_value(mvar) == stg_END_TSO_QUEUE_closure) {
1738 #if defined(THREADED_RTS)
1739 unlockClosure(mvar, info);
1741 /* HACK: we need a pointer to pass back,
1742 * so we abuse NO_FINALIZER_closure
1744 RET_NP(0, stg_NO_FINALIZER_closure);
1747 if (info == stg_MVAR_CLEAN_info) {
1748 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1751 /* we got the value... */
1752 val = StgMVar_value(mvar);
1754 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1756 /* There are putMVar(s) waiting...
1757 * wake up the first thread on the queue
1759 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1761 /* actually perform the putMVar for the thread that we just woke up */
1762 tso = StgMVar_head(mvar);
1763 PerformPut(tso,StgMVar_value(mvar));
1764 if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1765 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1768 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1769 StgMVar_head(mvar) "ptr", 1) [];
1770 StgMVar_head(mvar) = tso;
1772 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1773 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1775 #if defined(THREADED_RTS)
1776 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1778 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1783 /* No further putMVars, MVar is now empty */
1784 StgMVar_value(mvar) = stg_END_TSO_QUEUE_closure;
1785 #if defined(THREADED_RTS)
1786 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1788 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1798 W_ mvar, val, info, tso;
1800 /* args: R1 = MVar, R2 = value */
1804 #if defined(THREADED_RTS)
1805 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [];
1807 info = GET_INFO(mvar);
1810 if (info == stg_MVAR_CLEAN_info) {
1811 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1814 if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) {
1815 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1816 StgMVar_head(mvar) = CurrentTSO;
1818 foreign "C" setTSOLink(MyCapability() "ptr",
1819 StgMVar_tail(mvar) "ptr",
1822 StgTSO__link(CurrentTSO) = stg_END_TSO_QUEUE_closure;
1823 StgTSO_block_info(CurrentTSO) = mvar;
1824 // write barrier for throwTo(), which looks at block_info
1825 // if why_blocked==BlockedOnMVar.
1826 prim %write_barrier() [];
1827 StgTSO_why_blocked(CurrentTSO) = BlockedOnMVar::I16;
1828 StgMVar_tail(mvar) = CurrentTSO;
1832 jump stg_block_putmvar;
1835 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1837 /* There are takeMVar(s) waiting: wake up the first one
1839 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1841 /* actually perform the takeMVar */
1842 tso = StgMVar_head(mvar);
1843 PerformTake(tso, val);
1844 if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1845 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1848 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1849 StgMVar_head(mvar) "ptr", 1) [];
1850 StgMVar_head(mvar) = tso;
1852 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1853 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1856 #if defined(THREADED_RTS)
1857 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1859 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1861 jump %ENTRY_CODE(Sp(0));
1865 /* No further takes, the MVar is now full. */
1866 StgMVar_value(mvar) = val;
1868 #if defined(THREADED_RTS)
1869 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1871 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1873 jump %ENTRY_CODE(Sp(0));
1876 /* ToDo: yield afterward for better communication performance? */
1884 /* args: R1 = MVar, R2 = value */
1887 #if defined(THREADED_RTS)
1888 ("ptr" info) = foreign "C" lockClosure(mvar "ptr") [R2];
1890 info = GET_INFO(mvar);
1893 if (StgMVar_value(mvar) != stg_END_TSO_QUEUE_closure) {
1894 #if defined(THREADED_RTS)
1895 unlockClosure(mvar, info);
1900 if (info == stg_MVAR_CLEAN_info) {
1901 foreign "C" dirty_MVAR(BaseReg "ptr", mvar "ptr");
1904 if (StgMVar_head(mvar) != stg_END_TSO_QUEUE_closure) {
1906 /* There are takeMVar(s) waiting: wake up the first one
1908 ASSERT(StgTSO_why_blocked(StgMVar_head(mvar)) == BlockedOnMVar::I16);
1910 /* actually perform the takeMVar */
1911 tso = StgMVar_head(mvar);
1912 PerformTake(tso, R2);
1913 if (TO_W_(StgTSO_flags(tso)) & TSO_DIRTY == 0) {
1914 foreign "C" dirty_TSO(MyCapability() "ptr", tso "ptr") [];
1917 ("ptr" tso) = foreign "C" unblockOne_(MyCapability() "ptr",
1918 StgMVar_head(mvar) "ptr", 1) [];
1919 StgMVar_head(mvar) = tso;
1921 if (StgMVar_head(mvar) == stg_END_TSO_QUEUE_closure) {
1922 StgMVar_tail(mvar) = stg_END_TSO_QUEUE_closure;
1925 #if defined(THREADED_RTS)
1926 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1928 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1933 /* No further takes, the MVar is now full. */
1934 StgMVar_value(mvar) = R2;
1936 #if defined(THREADED_RTS)
1937 unlockClosure(mvar, stg_MVAR_DIRTY_info);
1939 SET_INFO(mvar,stg_MVAR_DIRTY_info);
1944 /* ToDo: yield afterward for better communication performance? */
1948 /* -----------------------------------------------------------------------------
1949 Stable pointer primitives
1950 ------------------------------------------------------------------------- */
1952 makeStableNamezh_fast
1956 ALLOC_PRIM( SIZEOF_StgStableName, R1_PTR, makeStableNamezh_fast );
1958 (index) = foreign "C" lookupStableName(R1 "ptr") [];
1960 /* Is there already a StableName for this heap object?
1961 * stable_ptr_table is a pointer to an array of snEntry structs.
1963 if ( snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) == NULL ) {
1964 sn_obj = Hp - SIZEOF_StgStableName + WDS(1);
1965 SET_HDR(sn_obj, stg_STABLE_NAME_info, W_[CCCS]);
1966 StgStableName_sn(sn_obj) = index;
1967 snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry) = sn_obj;
1969 sn_obj = snEntry_sn_obj(W_[stable_ptr_table] + index*SIZEOF_snEntry);
1976 makeStablePtrzh_fast
1980 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1981 ("ptr" sp) = foreign "C" getStablePtr(R1 "ptr") [];
1985 deRefStablePtrzh_fast
1987 /* Args: R1 = the stable ptr */
1990 r = snEntry_addr(W_[stable_ptr_table] + sp*SIZEOF_snEntry);
1994 /* -----------------------------------------------------------------------------
1995 Bytecode object primitives
1996 ------------------------------------------------------------------------- */
2006 W_ bco, bitmap_arr, bytes, words;
2010 words = BYTES_TO_WDS(SIZEOF_StgBCO) + StgArrWords_words(bitmap_arr);
2013 ALLOC_PRIM( bytes, R1_PTR&R2_PTR&R3_PTR&R5_PTR, newBCOzh_fast );
2015 bco = Hp - bytes + WDS(1);
2016 SET_HDR(bco, stg_BCO_info, W_[CCCS]);
2018 StgBCO_instrs(bco) = R1;
2019 StgBCO_literals(bco) = R2;
2020 StgBCO_ptrs(bco) = R3;
2021 StgBCO_arity(bco) = HALF_W_(R4);
2022 StgBCO_size(bco) = HALF_W_(words);
2024 // Copy the arity/bitmap info into the BCO
2028 if (i < StgArrWords_words(bitmap_arr)) {
2029 StgBCO_bitmap(bco,i) = StgArrWords_payload(bitmap_arr,i);
2040 // R1 = the BCO# for the AP
2044 // This function is *only* used to wrap zero-arity BCOs in an
2045 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
2046 // saturated and always points directly to a FUN or BCO.
2047 ASSERT(%INFO_TYPE(%GET_STD_INFO(R1)) == HALF_W_(BCO) &&
2048 StgBCO_arity(R1) == HALF_W_(0));
2050 HP_CHK_GEN_TICKY(SIZEOF_StgAP, R1_PTR, mkApUpd0zh_fast);
2051 TICK_ALLOC_UP_THK(0, 0);
2052 CCCS_ALLOC(SIZEOF_StgAP);
2054 ap = Hp - SIZEOF_StgAP + WDS(1);
2055 SET_HDR(ap, stg_AP_info, W_[CCCS]);
2057 StgAP_n_args(ap) = HALF_W_(0);
2063 unpackClosurezh_fast
2065 /* args: R1 = closure to analyze */
2066 // TODO: Consider the absence of ptrs or nonptrs as a special case ?
2068 W_ info, ptrs, nptrs, p, ptrs_arr, nptrs_arr;
2069 info = %GET_STD_INFO(UNTAG(R1));
2071 // Some closures have non-standard layout, so we omit those here.
2073 type = TO_W_(%INFO_TYPE(info));
2074 switch [0 .. N_CLOSURE_TYPES] type {
2075 case THUNK_SELECTOR : {
2080 case THUNK, THUNK_1_0, THUNK_0_1, THUNK_2_0, THUNK_1_1,
2081 THUNK_0_2, THUNK_STATIC, AP, PAP, AP_STACK, BCO : {
2087 ptrs = TO_W_(%INFO_PTRS(info));
2088 nptrs = TO_W_(%INFO_NPTRS(info));
2093 W_ ptrs_arr_sz, nptrs_arr_sz;
2094 nptrs_arr_sz = SIZEOF_StgArrWords + WDS(nptrs);
2095 ptrs_arr_sz = SIZEOF_StgMutArrPtrs + WDS(ptrs);
2097 ALLOC_PRIM (ptrs_arr_sz + nptrs_arr_sz, R1_PTR, unpackClosurezh_fast);
2102 ptrs_arr = Hp - nptrs_arr_sz - ptrs_arr_sz + WDS(1);
2103 nptrs_arr = Hp - nptrs_arr_sz + WDS(1);
2105 SET_HDR(ptrs_arr, stg_MUT_ARR_PTRS_FROZEN_info, W_[CCCS]);
2106 StgMutArrPtrs_ptrs(ptrs_arr) = ptrs;
2110 W_[ptrs_arr + SIZEOF_StgMutArrPtrs + WDS(p)] = StgClosure_payload(clos,p);
2115 SET_HDR(nptrs_arr, stg_ARR_WORDS_info, W_[CCCS]);
2116 StgArrWords_words(nptrs_arr) = nptrs;
2120 W_[BYTE_ARR_CTS(nptrs_arr) + WDS(p)] = StgClosure_payload(clos, p+ptrs);
2124 RET_NPP(info, ptrs_arr, nptrs_arr);
2127 /* -----------------------------------------------------------------------------
2128 Thread I/O blocking primitives
2129 -------------------------------------------------------------------------- */
2131 /* Add a thread to the end of the blocked queue. (C-- version of the C
2132 * macro in Schedule.h).
2134 #define APPEND_TO_BLOCKED_QUEUE(tso) \
2135 ASSERT(StgTSO__link(tso) == END_TSO_QUEUE); \
2136 if (W_[blocked_queue_hd] == END_TSO_QUEUE) { \
2137 W_[blocked_queue_hd] = tso; \
2139 foreign "C" setTSOLink(MyCapability() "ptr", W_[blocked_queue_tl] "ptr", tso) []; \
2141 W_[blocked_queue_tl] = tso;
2147 foreign "C" barf("waitRead# on threaded RTS") never returns;
2150 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2151 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
2152 StgTSO_block_info(CurrentTSO) = R1;
2153 // No locking - we're not going to use this interface in the
2154 // threaded RTS anyway.
2155 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2156 jump stg_block_noregs;
2164 foreign "C" barf("waitWrite# on threaded RTS") never returns;
2167 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2168 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2169 StgTSO_block_info(CurrentTSO) = R1;
2170 // No locking - we're not going to use this interface in the
2171 // threaded RTS anyway.
2172 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2173 jump stg_block_noregs;
2178 STRING(stg_delayzh_malloc_str, "delayzh_fast")
2181 #ifdef mingw32_HOST_OS
2189 foreign "C" barf("delay# on threaded RTS") never returns;
2192 /* args: R1 (microsecond delay amount) */
2193 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2194 StgTSO_why_blocked(CurrentTSO) = BlockedOnDelay::I16;
2196 #ifdef mingw32_HOST_OS
2198 /* could probably allocate this on the heap instead */
2199 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2200 stg_delayzh_malloc_str);
2201 (reqID) = foreign "C" addDelayRequest(R1);
2202 StgAsyncIOResult_reqID(ares) = reqID;
2203 StgAsyncIOResult_len(ares) = 0;
2204 StgAsyncIOResult_errCode(ares) = 0;
2205 StgTSO_block_info(CurrentTSO) = ares;
2207 /* Having all async-blocked threads reside on the blocked_queue
2208 * simplifies matters, so change the status to OnDoProc put the
2209 * delayed thread on the blocked_queue.
2211 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2212 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2213 jump stg_block_async_void;
2219 (time) = foreign "C" getourtimeofday() [R1];
2220 divisor = TO_W_(RtsFlags_MiscFlags_tickInterval(RtsFlags));
2224 divisor = divisor * 1000;
2225 target = ((R1 + divisor - 1) / divisor) /* divide rounding up */
2226 + time + 1; /* Add 1 as getourtimeofday rounds down */
2227 StgTSO_block_info(CurrentTSO) = target;
2229 /* Insert the new thread in the sleeping queue. */
2231 t = W_[sleeping_queue];
2233 if (t != END_TSO_QUEUE && StgTSO_block_info(t) < target) {
2235 t = StgTSO__link(t);
2239 StgTSO__link(CurrentTSO) = t;
2241 W_[sleeping_queue] = CurrentTSO;
2243 foreign "C" setTSOLink(MyCapability() "ptr", prev "ptr", CurrentTSO) [];
2245 jump stg_block_noregs;
2247 #endif /* !THREADED_RTS */
2251 #ifdef mingw32_HOST_OS
2252 STRING(stg_asyncReadzh_malloc_str, "asyncReadzh_fast")
2259 foreign "C" barf("asyncRead# on threaded RTS") never returns;
2262 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2263 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2264 StgTSO_why_blocked(CurrentTSO) = BlockedOnRead::I16;
2266 /* could probably allocate this on the heap instead */
2267 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2268 stg_asyncReadzh_malloc_str)
2270 (reqID) = foreign "C" addIORequest(R1, 0/*FALSE*/,R2,R3,R4 "ptr") [];
2271 StgAsyncIOResult_reqID(ares) = reqID;
2272 StgAsyncIOResult_len(ares) = 0;
2273 StgAsyncIOResult_errCode(ares) = 0;
2274 StgTSO_block_info(CurrentTSO) = ares;
2275 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2276 jump stg_block_async;
2280 STRING(stg_asyncWritezh_malloc_str, "asyncWritezh_fast")
2287 foreign "C" barf("asyncWrite# on threaded RTS") never returns;
2290 /* args: R1 = fd, R2 = isSock, R3 = len, R4 = buf */
2291 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2292 StgTSO_why_blocked(CurrentTSO) = BlockedOnWrite::I16;
2294 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2295 stg_asyncWritezh_malloc_str)
2297 (reqID) = foreign "C" addIORequest(R1, 1/*TRUE*/,R2,R3,R4 "ptr") [];
2299 StgAsyncIOResult_reqID(ares) = reqID;
2300 StgAsyncIOResult_len(ares) = 0;
2301 StgAsyncIOResult_errCode(ares) = 0;
2302 StgTSO_block_info(CurrentTSO) = ares;
2303 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2304 jump stg_block_async;
2308 STRING(stg_asyncDoProczh_malloc_str, "asyncDoProczh_fast")
2315 foreign "C" barf("asyncDoProc# on threaded RTS") never returns;
2318 /* args: R1 = proc, R2 = param */
2319 ASSERT(StgTSO_why_blocked(CurrentTSO) == NotBlocked::I16);
2320 StgTSO_why_blocked(CurrentTSO) = BlockedOnDoProc::I16;
2322 /* could probably allocate this on the heap instead */
2323 ("ptr" ares) = foreign "C" stgMallocBytes(SIZEOF_StgAsyncIOResult,
2324 stg_asyncDoProczh_malloc_str)
2326 (reqID) = foreign "C" addDoProcRequest(R1 "ptr",R2 "ptr") [];
2327 StgAsyncIOResult_reqID(ares) = reqID;
2328 StgAsyncIOResult_len(ares) = 0;
2329 StgAsyncIOResult_errCode(ares) = 0;
2330 StgTSO_block_info(CurrentTSO) = ares;
2331 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
2332 jump stg_block_async;
2337 // noDuplicate# tries to ensure that none of the thunks under
2338 // evaluation by the current thread are also under evaluation by
2339 // another thread. It relies on *both* threads doing noDuplicate#;
2340 // the second one will get blocked if they are duplicating some work.
2343 SAVE_THREAD_STATE();
2344 ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16);
2345 foreign "C" threadPaused (MyCapability() "ptr", CurrentTSO "ptr") [];
2347 if (StgTSO_what_next(CurrentTSO) == ThreadKilled::I16) {
2348 jump stg_threadFinished;
2350 LOAD_THREAD_STATE();
2351 ASSERT(StgTSO_what_next(CurrentTSO) == ThreadRunGHC::I16);
2352 jump %ENTRY_CODE(Sp(0));
2356 getApStackValzh_fast
2358 W_ ap_stack, offset, val, ok;
2360 /* args: R1 = AP_STACK, R2 = offset */
2364 if (%INFO_PTR(ap_stack) == stg_AP_STACK_info) {
2366 val = StgAP_STACK_payload(ap_stack,offset);
2374 /* -----------------------------------------------------------------------------
2376 -------------------------------------------------------------------------- */
2378 // Write the cost center stack of the first argument on stderr; return
2379 // the second. Possibly only makes sense for already evaluated
2386 ccs = StgHeader_ccs(UNTAG(R1));
2387 foreign "C" fprintCCS_stderr(ccs "ptr") [R2];
2398 #ifndef THREADED_RTS
2399 RET_NP(0,ghczmprim_GHCziBool_False_closure);
2401 (spark) = foreign "C" findSpark(MyCapability());
2405 RET_NP(0,ghczmprim_GHCziBool_False_closure);