1 /* -----------------------------------------------------------------------------
2 * $Id: PrimOps.hc,v 1.116 2004/01/08 15:26:44 simonmar Exp $
4 * (c) The GHC Team, 1998-2002
6 * Primitive functions / data
8 * ---------------------------------------------------------------------------*/
14 #include "StgStartup.h"
19 #include "BlockAlloc.h" /* tmp */
20 #include "StablePriv.h"
22 #include "Timer.h" /* TICK_MILLISECS */
24 #ifndef mingw32_TARGET_OS
25 #include "Itimer.h" /* getourtimeofday() */
28 #ifdef HAVE_SYS_TYPES_H
29 # include <sys/types.h>
34 #ifdef mingw32_TARGET_OS
36 #include "win32/AsyncIO.h"
41 classes CCallable and CReturnable don't really exist, but the
42 compiler insists on generating dictionaries containing references
43 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
44 for these. Some C compilers can't cope with zero-length static arrays,
45 so we have to make these one element long.
48 StgWord GHC_ZCCCallable_static_info[1];
49 StgWord GHC_ZCCReturnable_static_info[1];
51 /* -----------------------------------------------------------------------------
52 Macros for Hand-written primitives.
53 -------------------------------------------------------------------------- */
56 * Horrible macros for returning unboxed tuples.
58 * How an unboxed tuple is returned depends on two factors:
59 * - the number of real registers we have available
60 * - the boxedness of the returned fields.
62 * To return an unboxed tuple from a primitive operation, we have macros
63 * RET_<layout> where <layout> describes the boxedness of each field of the
64 * unboxed tuple: N indicates a non-pointer field, and P indicates a pointer.
66 * We only define the cases actually used, to avoid having too much
67 * garbage in this section. Warning: any bugs in here will be hard to
70 * The return convention for an unboxed tuple is as follows:
71 * - fit as many fields as possible in registers (as per the
72 * function fast-entry point calling convention).
73 * - sort the rest of the fields into pointers and non-pointers.
74 * push the pointers on the stack, followed by the non-pointers.
75 * (so the pointers have higher addresses).
78 /*------ All Regs available */
79 #if MAX_REAL_VANILLA_REG == 8
80 # define RET_P(a) R1.w = (W_)(a); JMP_(ENTRY_CODE(Sp[0]));
81 # define RET_N(a) RET_P(a)
83 # define RET_PP(a,b) R1.w = (W_)(a); R2.w = (W_)(b); JMP_(ENTRY_CODE(Sp[0]));
84 # define RET_NN(a,b) RET_PP(a,b)
85 # define RET_NP(a,b) RET_PP(a,b)
87 # define RET_PPP(a,b,c) \
88 R1.w = (W_)(a); R2.w = (W_)(b); R3.w = (W_)(c); JMP_(ENTRY_CODE(Sp[0]));
89 # define RET_NNP(a,b,c) RET_PPP(a,b,c)
91 # define RET_NNNP(a,b,c,d) \
92 R1.w = (W_)(a); R2.w = (W_)(b); R3.w = (W_)(c); R4.w = (W_)d; \
93 JMP_(ENTRY_CODE(Sp[0]));
95 # define RET_NPNP(a,b,c,d) \
96 R1.w = (W_)(a); R2.w = (W_)(b); R3.w = (W_)(c); R4.w = (W_)(d); \
97 JMP_(ENTRY_CODE(Sp[0]));
99 #elif MAX_REAL_VANILLA_REG > 2 && MAX_REAL_VANILLA_REG < 8
100 # error RET_n macros not defined for this setup.
102 /*------ 2 Registers available */
103 #elif MAX_REAL_VANILLA_REG == 2
105 # define RET_P(a) R1.w = (W_)(a); JMP_(ENTRY_CODE(Sp[0]));
106 # define RET_N(a) RET_P(a)
108 # define RET_PP(a,b) R1.w = (W_)(a); R2.w = (W_)(b); \
109 JMP_(ENTRY_CODE(Sp[0]));
110 # define RET_NN(a,b) RET_PP(a,b)
111 # define RET_NP(a,b) RET_PP(a,b)
113 # define RET_PPP(a,b,c) \
118 JMP_(ENTRY_CODE(Sp[1]));
120 # define RET_NNP(a,b,c) \
125 JMP_(ENTRY_CODE(Sp[1]));
127 # define RET_NNNP(a,b,c,d) \
133 JMP_(ENTRY_CODE(Sp[2]));
135 # define RET_NPNP(a,b,c,d) \
141 JMP_(ENTRY_CODE(Sp[2]));
143 /*------ 1 Register available */
144 #elif MAX_REAL_VANILLA_REG == 1
145 # define RET_P(a) R1.w = (W_)(a); JMP_(ENTRY_CODE(Sp[0]));
146 # define RET_N(a) RET_P(a)
148 # define RET_PP(a,b) R1.w = (W_)(a); Sp[-1] = (W_)(b); Sp -= 1; \
149 JMP_(ENTRY_CODE(Sp[1]));
150 # define RET_NN(a,b) R1.w = (W_)(a); Sp[-1] = (W_)(b); Sp -= 2; \
151 JMP_(ENTRY_CODE(Sp[2]));
152 # define RET_NP(a,b) RET_PP(a,b)
154 # define RET_PPP(a,b,c) \
159 JMP_(ENTRY_CODE(Sp[2]));
161 # define RET_NNP(a,b,c) \
166 JMP_(ENTRY_CODE(Sp[2]));
168 # define RET_NNNP(a,b,c,d) \
174 JMP_(ENTRY_CODE(Sp[3]));
176 # define RET_NPNP(a,b,c,d) \
182 JMP_(ENTRY_CODE(Sp[3]));
184 #else /* 0 Regs available */
186 #define PUSH(o,x) Sp[-o] = (W_)(x)
188 #define PUSHED(m) Sp -= (m); JMP_(ENTRY_CODE(Sp[m]));
190 # define RET_P(a) PUSH(1,a); PUSHED(1)
191 # define RET_N(a) PUSH(1,a); PUSHED(1)
193 # define RET_PP(a,b) PUSH(2,a); PUSH(1,b); PUSHED(2)
194 # define RET_NN(a,b) PUSH(2,a); PUSH(1,b); PUSHED(2)
195 # define RET_NP(a,b) PUSH(2,a); PUSH(1,b); PUSHED(2)
197 # define RET_PPP(a,b,c) PUSH(3,a); PUSH(2,b); PUSH(1,c); PUSHED(3)
198 # define RET_NNP(a,b,c) PUSH(3,a); PUSH(2,b); PUSH(1,c); PUSHED(3)
200 # define RET_NNNP(a,b,c,d) PUSH(4,a); PUSH(3,b); PUSH(2,c); PUSH(1,d); PUSHED(4)
201 # define RET_NPNP(a,b,c,d) PUSH(4,a); PUSH(3,c); PUSH(2,b); PUSH(1,d); PUSHED(4)
204 /*-----------------------------------------------------------------------------
207 Basically just new*Array - the others are all inline macros.
209 The size arg is always passed in R1, and the result returned in R1.
211 The slow entry point is for returning from a heap check, the saved
212 size argument must be re-loaded from the stack.
213 -------------------------------------------------------------------------- */
215 /* for objects that are *less* than the size of a word, make sure we
216 * round up to the nearest word for the size of the array.
219 #define BYTES_TO_STGWORDS(n) ((n) + sizeof(W_) - 1)/sizeof(W_)
221 FN_(newByteArrayzh_fast)
223 W_ size, stuff_size, n;
226 MAYBE_GC(NO_PTRS,newByteArrayzh_fast);
228 stuff_size = BYTES_TO_STGWORDS(n);
229 size = sizeofW(StgArrWords)+ stuff_size;
230 p = (StgArrWords *)RET_STGCALL1(P_,allocate,size);
231 TICK_ALLOC_PRIM(sizeofW(StgArrWords),stuff_size,0);
232 SET_HDR(p, &stg_ARR_WORDS_info, CCCS);
233 p->words = stuff_size;
234 TICK_RET_UNBOXED_TUP(1)
239 FN_(newPinnedByteArrayzh_fast)
241 W_ size, stuff_size, n;
244 MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast);
246 stuff_size = BYTES_TO_STGWORDS(n);
248 // We want an 8-byte aligned array. allocatePinned() gives us
249 // 8-byte aligned memory by default, but we want to align the
250 // *goods* inside the ArrWords object, so we have to check the
251 // size of the ArrWords header and adjust our size accordingly.
252 size = sizeofW(StgArrWords)+ stuff_size;
253 if ((sizeof(StgArrWords) & 7) != 0) {
257 p = (StgArrWords *)RET_STGCALL1(P_,allocatePinned,size);
258 TICK_ALLOC_PRIM(sizeofW(StgArrWords),stuff_size,0);
260 // Again, if the ArrWords header isn't a multiple of 8 bytes, we
261 // have to push the object forward one word so that the goods
262 // fall on an 8-byte boundary.
263 if ((sizeof(StgArrWords) & 7) != 0) {
267 SET_HDR(p, &stg_ARR_WORDS_info, CCCS);
268 p->words = stuff_size;
269 TICK_RET_UNBOXED_TUP(1)
282 MAYBE_GC(R2_PTR,newArrayzh_fast);
284 size = sizeofW(StgMutArrPtrs) + n;
285 arr = (StgMutArrPtrs *)RET_STGCALL1(P_, allocate, size);
286 TICK_ALLOC_PRIM(sizeofW(StgMutArrPtrs), n, 0);
288 SET_HDR(arr,&stg_MUT_ARR_PTRS_info,CCCS);
292 for (p = (P_)arr + sizeofW(StgMutArrPtrs);
293 p < (P_)arr + size; p++) {
297 TICK_RET_UNBOXED_TUP(1);
302 FN_(newMutVarzh_fast)
305 /* Args: R1.p = initialisation value */
308 HP_CHK_GEN_TICKY(sizeofW(StgMutVar), R1_PTR, newMutVarzh_fast);
309 TICK_ALLOC_PRIM(sizeofW(StgHeader)+1,1, 0); /* hack, dependent on rep. */
310 CCS_ALLOC(CCCS,sizeofW(StgMutVar));
312 mv = (StgMutVar *)(Hp-sizeofW(StgMutVar)+1);
313 SET_HDR(mv,&stg_MUT_VAR_info,CCCS);
316 TICK_RET_UNBOXED_TUP(1);
321 FN_(atomicModifyMutVarzh_fast)
324 StgClosure *z, *x, *y, *r;
326 /* Args: R1.p :: MutVar#, R2.p :: a -> (a,b) */
328 /* If x is the current contents of the MutVar#, then
329 We want to make the new contents point to
333 and the return value is
337 obviously we can share (f x).
339 z = [stg_ap_2 f x] (max (HS + 2) MIN_UPD_SIZE)
340 y = [stg_sel_0 z] (max (HS + 1) MIN_UPD_SIZE)
341 r = [stg_sel_1 z] (max (HS + 1) MIN_UPD_SIZE)
344 #define THUNK_SIZE(n) (sizeofW(StgHeader) + stg_max((n), MIN_UPD_SIZE))
345 #define SIZE (THUNK_SIZE(2) + THUNK_SIZE(1) + THUNK_SIZE(1))
347 HP_CHK_GEN_TICKY(SIZE, R1_PTR|R2_PTR, atomicModifyMutVarzh_fast);
348 CCS_ALLOC(CCCS,SIZE);
350 x = ((StgMutVar *)R1.cl)->var;
352 TICK_ALLOC_UP_THK(2,0); // XXX
353 z = (StgClosure *) Hp - THUNK_SIZE(2) + 1;
354 SET_HDR(z, (StgInfoTable *)&stg_ap_2_upd_info, CCCS);
355 z->payload[0] = R2.cl;
358 TICK_ALLOC_UP_THK(1,1); // XXX
359 y = (StgClosure *) (StgPtr)z - THUNK_SIZE(1);
360 SET_HDR(y, &stg_sel_0_upd_info, CCCS);
363 ((StgMutVar *)R1.cl)->var = y;
365 TICK_ALLOC_UP_THK(1,1); // XXX
366 r = (StgClosure *) (StgPtr)y - THUNK_SIZE(1);
367 SET_HDR(r, &stg_sel_1_upd_info, CCCS);
374 /* -----------------------------------------------------------------------------
375 Foreign Object Primitives
376 -------------------------------------------------------------------------- */
378 FN_(mkForeignObjzh_fast)
380 /* R1.p = ptr to foreign object,
382 StgForeignObj *result;
385 HP_CHK_GEN_TICKY(sizeofW(StgForeignObj), NO_PTRS, mkForeignObjzh_fast);
386 TICK_ALLOC_PRIM(sizeofW(StgHeader),
387 sizeofW(StgForeignObj)-sizeofW(StgHeader), 0);
388 CCS_ALLOC(CCCS,sizeofW(StgForeignObj)); /* ccs prof */
390 result = (StgForeignObj *) (Hp + 1 - sizeofW(StgForeignObj));
391 SET_HDR(result,&stg_FOREIGN_info,CCCS);
394 /* returns (# s#, ForeignObj# #) */
395 TICK_RET_UNBOXED_TUP(1);
400 /* These two are out-of-line for the benefit of the NCG */
401 FN_(unsafeThawArrayzh_fast)
404 SET_INFO((StgClosure *)R1.cl,&stg_MUT_ARR_PTRS_info);
406 // SUBTLETY TO DO WITH THE OLD GEN MUTABLE LIST
408 // A MUT_ARR_PTRS lives on the mutable list, but a MUT_ARR_PTRS_FROZEN
409 // normally doesn't. However, when we freeze a MUT_ARR_PTRS, we leave
410 // it on the mutable list for the GC to remove (removing something from
411 // the mutable list is not easy, because the mut_list is only singly-linked).
413 // So, when we thaw a MUT_ARR_PTRS_FROZEN, we must cope with two cases:
414 // either it is on a mut_list, or it isn't. We adopt the convention that
415 // the mut_link field is NULL if it isn't on a mut_list, and the GC
416 // maintains this invariant.
418 if (((StgMutArrPtrs *)R1.cl)->mut_link == NULL) {
419 recordMutable((StgMutClosure*)R1.cl);
422 TICK_RET_UNBOXED_TUP(1);
427 /* -----------------------------------------------------------------------------
428 Weak Pointer Primitives
429 -------------------------------------------------------------------------- */
435 R3.p = finalizer (or NULL)
441 R3.cl = &stg_NO_FINALIZER_closure;
444 HP_CHK_GEN_TICKY(sizeofW(StgWeak),R1_PTR|R2_PTR|R3_PTR, mkWeakzh_fast);
445 TICK_ALLOC_PRIM(sizeofW(StgHeader)+1, // +1 is for the link field
446 sizeofW(StgWeak)-sizeofW(StgHeader)-1, 0);
447 CCS_ALLOC(CCCS,sizeofW(StgWeak)); /* ccs prof */
449 w = (StgWeak *) (Hp + 1 - sizeofW(StgWeak));
450 SET_HDR(w, &stg_WEAK_info, CCCS);
454 w->finalizer = R3.cl;
456 w->link = weak_ptr_list;
458 IF_DEBUG(weak, fprintf(stderr,"New weak pointer at %p\n",w));
460 TICK_RET_UNBOXED_TUP(1);
465 FN_(finalizzeWeakzh_fast)
472 TICK_RET_UNBOXED_TUP(0);
473 w = (StgDeadWeak *)R1.p;
476 if (w->header.info == &stg_DEAD_WEAK_info) {
477 RET_NP(0,&stg_NO_FINALIZER_closure);
483 // A weak pointer is inherently used, so we do not need to call
484 // LDV_recordDead_FILL_SLOP_DYNAMIC():
485 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
486 // or, LDV_recordDead():
487 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
488 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
489 // large as weak pointers, so there is no need to fill the slop, either.
490 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
493 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
495 w->header.info = &stg_DEAD_WEAK_info;
498 LDV_recordCreate((StgClosure *)w);
500 f = ((StgWeak *)w)->finalizer;
501 w->link = ((StgWeak *)w)->link;
503 /* return the finalizer */
504 if (f == &stg_NO_FINALIZER_closure) {
505 RET_NP(0,&stg_NO_FINALIZER_closure);
512 FN_(deRefWeakzh_fast)
514 /* R1.p = weak ptr */
520 if (w->header.info == &stg_WEAK_info) {
522 val = (P_)((StgWeak *)w)->value;
531 /* -----------------------------------------------------------------------------
532 Arbitrary-precision Integer operations.
533 -------------------------------------------------------------------------- */
535 FN_(int2Integerzh_fast)
537 /* arguments: R1 = Int# */
539 I_ val, s; /* to avoid aliasing */
540 StgArrWords* p; /* address of array result */
544 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+1, NO_PTRS, int2Integerzh_fast);
545 TICK_ALLOC_PRIM(sizeofW(StgArrWords),1,0);
546 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+1); /* ccs prof */
548 p = (StgArrWords *)Hp - 1;
549 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, 1);
551 /* mpz_set_si is inlined here, makes things simpler */
555 } else if (val > 0) {
562 /* returns (# size :: Int#,
566 TICK_RET_UNBOXED_TUP(2);
571 FN_(word2Integerzh_fast)
573 /* arguments: R1 = Word# */
575 W_ val; /* to avoid aliasing */
577 StgArrWords* p; /* address of array result */
581 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+1, NO_PTRS, word2Integerzh_fast)
582 TICK_ALLOC_PRIM(sizeofW(StgArrWords),1,0);
583 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+1); /* ccs prof */
585 p = (StgArrWords *)Hp - 1;
586 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, 1);
595 /* returns (# size :: Int#,
599 TICK_RET_UNBOXED_TUP(2);
606 * 'long long' primops for converting to/from Integers.
609 #ifdef SUPPORT_LONG_LONGS
611 FN_(int64ToIntegerzh_fast)
613 /* arguments: L1 = Int64# */
615 StgInt64 val; /* to avoid aliasing */
617 I_ s, neg, words_needed;
618 StgArrWords* p; /* address of array result */
624 if ( val >= 0x100000000LL || val <= -0x100000000LL ) {
627 /* minimum is one word */
630 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+words_needed, NO_PTRS, int64ToIntegerzh_fast)
631 TICK_ALLOC_PRIM(sizeofW(StgArrWords),words_needed,0);
632 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+words_needed); /* ccs prof */
634 p = (StgArrWords *)(Hp-words_needed+1) - 1;
635 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, words_needed);
642 hi = (W_)((LW_)val / 0x100000000ULL);
644 if ( words_needed == 2 ) {
648 } else if ( val != 0 ) {
651 } else /* val==0 */ {
654 s = ( neg ? -s : s );
656 /* returns (# size :: Int#,
660 TICK_RET_UNBOXED_TUP(2);
665 FN_(word64ToIntegerzh_fast)
667 /* arguments: L1 = Word64# */
669 StgWord64 val; /* to avoid aliasing */
672 StgArrWords* p; /* address of array result */
676 if ( val >= 0x100000000ULL ) {
681 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+words_needed, NO_PTRS, word64ToIntegerzh_fast)
682 TICK_ALLOC_PRIM(sizeofW(StgArrWords),words_needed,0);
683 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+words_needed); /* ccs prof */
685 p = (StgArrWords *)(Hp-words_needed+1) - 1;
686 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, words_needed);
688 hi = (W_)((LW_)val / 0x100000000ULL);
689 if ( val >= 0x100000000ULL ) {
693 } else if ( val != 0 ) {
696 } else /* val==0 */ {
700 /* returns (# size :: Int#,
704 TICK_RET_UNBOXED_TUP(2);
710 #endif /* SUPPORT_LONG_LONGS */
712 /* ToDo: this is shockingly inefficient */
714 #define GMP_TAKE2_RET1(name,mp_fun) \
717 MP_INT arg1, arg2, result; \
723 /* call doYouWantToGC() */ \
724 MAYBE_GC(R2_PTR | R4_PTR, name); \
726 d1 = (StgArrWords *)R2.p; \
728 d2 = (StgArrWords *)R4.p; \
731 arg1._mp_alloc = d1->words; \
732 arg1._mp_size = (s1); \
733 arg1._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
734 arg2._mp_alloc = d2->words; \
735 arg2._mp_size = (s2); \
736 arg2._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d2)); \
738 STGCALL1(mpz_init,&result); \
740 /* Perform the operation */ \
741 STGCALL3(mp_fun,&result,&arg1,&arg2); \
743 TICK_RET_UNBOXED_TUP(2); \
744 RET_NP(result._mp_size, \
745 result._mp_d-sizeofW(StgArrWords)); \
749 #define GMP_TAKE1_RET1(name,mp_fun) \
752 MP_INT arg1, result; \
757 /* call doYouWantToGC() */ \
758 MAYBE_GC(R2_PTR, name); \
760 d1 = (StgArrWords *)R2.p; \
763 arg1._mp_alloc = d1->words; \
764 arg1._mp_size = (s1); \
765 arg1._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
767 STGCALL1(mpz_init,&result); \
769 /* Perform the operation */ \
770 STGCALL2(mp_fun,&result,&arg1); \
772 TICK_RET_UNBOXED_TUP(2); \
773 RET_NP(result._mp_size, \
774 result._mp_d-sizeofW(StgArrWords)); \
778 #define GMP_TAKE2_RET2(name,mp_fun) \
781 MP_INT arg1, arg2, result1, result2; \
787 /* call doYouWantToGC() */ \
788 MAYBE_GC(R2_PTR | R4_PTR, name); \
790 d1 = (StgArrWords *)R2.p; \
792 d2 = (StgArrWords *)R4.p; \
795 arg1._mp_alloc = d1->words; \
796 arg1._mp_size = (s1); \
797 arg1._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
798 arg2._mp_alloc = d2->words; \
799 arg2._mp_size = (s2); \
800 arg2._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d2)); \
802 STGCALL1(mpz_init,&result1); \
803 STGCALL1(mpz_init,&result2); \
805 /* Perform the operation */ \
806 STGCALL4(mp_fun,&result1,&result2,&arg1,&arg2); \
808 TICK_RET_UNBOXED_TUP(4); \
809 RET_NPNP(result1._mp_size, \
810 result1._mp_d-sizeofW(StgArrWords), \
812 result2._mp_d-sizeofW(StgArrWords)); \
816 GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add);
817 GMP_TAKE2_RET1(minusIntegerzh_fast, mpz_sub);
818 GMP_TAKE2_RET1(timesIntegerzh_fast, mpz_mul);
819 GMP_TAKE2_RET1(gcdIntegerzh_fast, mpz_gcd);
820 GMP_TAKE2_RET1(quotIntegerzh_fast, mpz_tdiv_q);
821 GMP_TAKE2_RET1(remIntegerzh_fast, mpz_tdiv_r);
822 GMP_TAKE2_RET1(divExactIntegerzh_fast, mpz_divexact);
823 GMP_TAKE2_RET1(andIntegerzh_fast, mpz_and);
824 GMP_TAKE2_RET1(orIntegerzh_fast, mpz_ior);
825 GMP_TAKE2_RET1(xorIntegerzh_fast, mpz_xor);
826 GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com);
828 GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr);
829 GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr);
834 /* R1 = the first Int#; R2 = the second Int# */
838 aa = (mp_limb_t)(R1.i);
839 r = RET_STGCALL3(StgInt, mpn_gcd_1, (mp_limb_t *)(&aa), 1, (mp_limb_t)(R2.i));
842 /* Result parked in R1, return via info-pointer at TOS */
843 JMP_(ENTRY_CODE(Sp[0]));
847 FN_(gcdIntegerIntzh_fast)
849 /* R1 = s1; R2 = d1; R3 = the int */
852 r = RET_STGCALL3(StgInt,mpn_gcd_1,(mp_limb_t *)(BYTE_ARR_CTS(R2.p)), R1.i, R3.i);
855 /* Result parked in R1, return via info-pointer at TOS */
856 JMP_(ENTRY_CODE(Sp[0]));
860 FN_(cmpIntegerIntzh_fast)
862 /* R1 = s1; R2 = d1; R3 = the int */
873 // paraphrased from mpz_cmp_si() in the GMP sources
876 } else if (v_digit < 0) {
881 if (usize != vsize) {
882 R1.i = usize - vsize; JMP_(ENTRY_CODE(Sp[0]));
886 R1.i = 0; JMP_(ENTRY_CODE(Sp[0]));
889 u_digit = *(mp_limb_t *)(BYTE_ARR_CTS(R2.p));
891 if (u_digit == (mp_limb_t) (unsigned long) v_digit) {
892 R1.i = 0; JMP_(ENTRY_CODE(Sp[0]));
895 if (u_digit > (mp_limb_t) (unsigned long) v_digit) {
901 JMP_(ENTRY_CODE(Sp[0]));
905 FN_(cmpIntegerzh_fast)
907 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
915 // paraphrased from mpz_cmp() in the GMP sources
919 if (usize != vsize) {
920 R1.i = usize - vsize; JMP_(ENTRY_CODE(Sp[0]));
924 R1.i = 0; JMP_(ENTRY_CODE(Sp[0]));
929 up = BYTE_ARR_CTS(R2.p);
930 vp = BYTE_ARR_CTS(R4.p);
932 cmp = RET_STGCALL3(I_, mpn_cmp, (mp_limb_t *)up, (mp_limb_t *)vp, size);
935 R1.i = 0; JMP_(ENTRY_CODE(Sp[0]));
938 if ((cmp < 0) == (usize < 0)) {
943 /* Result parked in R1, return via info-pointer at TOS */
944 JMP_(ENTRY_CODE(Sp[0]));
948 FN_(integer2Intzh_fast)
957 r = ((mp_limb_t *) (BYTE_ARR_CTS(R2.p)))[0];
960 /* Result parked in R1, return via info-pointer at TOS */
962 JMP_(ENTRY_CODE(Sp[0]));
966 FN_(integer2Wordzh_fast)
976 r = ((mp_limb_t *) (BYTE_ARR_CTS(R2.p)))[0];
979 /* Result parked in R1, return via info-pointer at TOS */
981 JMP_(ENTRY_CODE(Sp[0]));
986 FN_(decodeFloatzh_fast)
994 /* arguments: F1 = Float# */
997 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+1, NO_PTRS, decodeFloatzh_fast);
998 TICK_ALLOC_PRIM(sizeofW(StgArrWords),1,0);
999 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+1); /* ccs prof */
1001 /* Be prepared to tell Lennart-coded __decodeFloat */
1002 /* where mantissa._mp_d can be put (it does not care about the rest) */
1003 p = (StgArrWords *)Hp - 1;
1004 SET_ARR_HDR(p,&stg_ARR_WORDS_info,CCCS,1)
1005 mantissa._mp_d = (void *)BYTE_ARR_CTS(p);
1007 /* Perform the operation */
1008 STGCALL3(__decodeFloat,&mantissa,&exponent,arg);
1010 /* returns: (Int# (expn), Int#, ByteArray#) */
1011 TICK_RET_UNBOXED_TUP(3);
1012 RET_NNP(exponent,mantissa._mp_size,p);
1016 #define DOUBLE_MANTISSA_SIZE (sizeofW(StgDouble))
1017 #define ARR_SIZE (sizeofW(StgArrWords) + DOUBLE_MANTISSA_SIZE)
1019 FN_(decodeDoublezh_fast)
1026 /* arguments: D1 = Double# */
1029 HP_CHK_GEN_TICKY(ARR_SIZE, NO_PTRS, decodeDoublezh_fast);
1030 TICK_ALLOC_PRIM(sizeofW(StgArrWords),DOUBLE_MANTISSA_SIZE,0);
1031 CCS_ALLOC(CCCS,ARR_SIZE); /* ccs prof */
1033 /* Be prepared to tell Lennart-coded __decodeDouble */
1034 /* where mantissa.d can be put (it does not care about the rest) */
1035 p = (StgArrWords *)(Hp-ARR_SIZE+1);
1036 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, DOUBLE_MANTISSA_SIZE);
1037 mantissa._mp_d = (void *)BYTE_ARR_CTS(p);
1039 /* Perform the operation */
1040 STGCALL3(__decodeDouble,&mantissa,&exponent,arg);
1042 /* returns: (Int# (expn), Int#, ByteArray#) */
1043 TICK_RET_UNBOXED_TUP(3);
1044 RET_NNP(exponent,mantissa._mp_size,p);
1048 /* -----------------------------------------------------------------------------
1049 * Concurrency primitives
1050 * -------------------------------------------------------------------------- */
1055 /* args: R1 = closure to spark */
1057 MAYBE_GC(R1_PTR, forkzh_fast);
1059 /* create it right now, return ThreadID in R1 */
1060 R1.t = RET_STGCALL2(StgTSO *, createIOThread,
1061 RtsFlags.GcFlags.initialStkSize, R1.cl);
1062 STGCALL1(scheduleThread, R1.t);
1064 /* switch at the earliest opportunity */
1074 JMP_(stg_yield_noregs);
1078 FN_(myThreadIdzh_fast)
1082 RET_P((P_)CurrentTSO);
1086 FN_(labelThreadzh_fast)
1093 STGCALL2(labelThread,R1.p,(char *)R2.p);
1095 JMP_(ENTRY_CODE(Sp[0]));
1099 FN_(isCurrentThreadBoundzh_fast)
1104 r = (I_)(RET_STGCALL1(StgBool, isThreadBound, CurrentTSO));
1109 /* -----------------------------------------------------------------------------
1112 * take & putMVar work as follows. Firstly, an important invariant:
1114 * If the MVar is full, then the blocking queue contains only
1115 * threads blocked on putMVar, and if the MVar is empty then the
1116 * blocking queue contains only threads blocked on takeMVar.
1119 * MVar empty : then add ourselves to the blocking queue
1120 * MVar full : remove the value from the MVar, and
1121 * blocking queue empty : return
1122 * blocking queue non-empty : perform the first blocked putMVar
1123 * from the queue, and wake up the
1124 * thread (MVar is now full again)
1126 * putMVar is just the dual of the above algorithm.
1128 * How do we "perform a putMVar"? Well, we have to fiddle around with
1129 * the stack of the thread waiting to do the putMVar. See
1130 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1131 * the stack layout, and the PerformPut and PerformTake macros below.
1133 * It is important that a blocked take or put is woken up with the
1134 * take/put already performed, because otherwise there would be a
1135 * small window of vulnerability where the thread could receive an
1136 * exception and never perform its take or put, and we'd end up with a
1139 * -------------------------------------------------------------------------- */
1141 FN_(isEmptyMVarzh_fast)
1143 /* args: R1 = MVar closure */
1146 r = (I_)((GET_INFO((StgMVar*)(R1.p))) == &stg_EMPTY_MVAR_info);
1159 HP_CHK_GEN_TICKY(sizeofW(StgMVar), NO_PTRS, newMVarzh_fast);
1160 TICK_ALLOC_PRIM(sizeofW(StgMutVar)-1, // consider head,tail,link as admin wds
1162 CCS_ALLOC(CCCS,sizeofW(StgMVar)); /* ccs prof */
1164 mvar = (StgMVar *) (Hp - sizeofW(StgMVar) + 1);
1165 SET_HDR(mvar,&stg_EMPTY_MVAR_info,CCCS);
1166 mvar->head = mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1167 mvar->value = (StgClosure *)&stg_END_TSO_QUEUE_closure;
1169 TICK_RET_UNBOXED_TUP(1);
1174 /* If R1 isn't available, pass it on the stack */
1176 #define PerformTake(tso, value) ({ \
1177 (tso)->sp[1] = (W_)value; \
1178 (tso)->sp[0] = (W_)&stg_gc_unpt_r1_info; \
1181 #define PerformTake(tso, value) ({ \
1182 (tso)->sp[1] = (W_)value; \
1183 (tso)->sp[0] = (W_)&stg_ut_1_0_unreg_info; \
1188 #define PerformPut(tso) ({ \
1189 StgClosure *val = (StgClosure *)(tso)->sp[2]; \
1194 FN_(takeMVarzh_fast)
1198 const StgInfoTable *info;
1201 /* args: R1 = MVar closure */
1203 mvar = (StgMVar *)R1.p;
1206 info = LOCK_CLOSURE(mvar);
1208 info = GET_INFO(mvar);
1211 /* If the MVar is empty, put ourselves on its blocking queue,
1212 * and wait until we're woken up.
1214 if (info == &stg_EMPTY_MVAR_info) {
1215 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1216 mvar->head = CurrentTSO;
1218 mvar->tail->link = CurrentTSO;
1220 CurrentTSO->link = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1221 CurrentTSO->why_blocked = BlockedOnMVar;
1222 CurrentTSO->block_info.closure = (StgClosure *)mvar;
1223 mvar->tail = CurrentTSO;
1226 /* unlock the MVar */
1227 mvar->header.info = &stg_EMPTY_MVAR_info;
1229 JMP_(stg_block_takemvar);
1232 /* we got the value... */
1235 if (mvar->head != (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1236 /* There are putMVar(s) waiting...
1237 * wake up the first thread on the queue
1239 ASSERT(mvar->head->why_blocked == BlockedOnMVar);
1241 /* actually perform the putMVar for the thread that we just woke up */
1242 mvar->value = PerformPut(mvar->head);
1244 #if defined(GRAN) || defined(PAR)
1245 /* ToDo: check 2nd arg (mvar) is right */
1246 mvar->head = RET_STGCALL2(StgTSO *,unblockOne,mvar->head,mvar);
1248 mvar->head = RET_STGCALL1(StgTSO *,unblockOne,mvar->head);
1250 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1251 mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1254 /* unlock in the SMP case */
1255 SET_INFO(mvar,&stg_FULL_MVAR_info);
1257 TICK_RET_UNBOXED_TUP(1);
1260 /* No further putMVars, MVar is now empty */
1262 /* do this last... we might have locked the MVar in the SMP case,
1263 * and writing the info pointer will unlock it.
1265 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1266 mvar->value = (StgClosure *)&stg_END_TSO_QUEUE_closure;
1267 TICK_RET_UNBOXED_TUP(1);
1273 FN_(tryTakeMVarzh_fast)
1277 const StgInfoTable *info;
1280 /* args: R1 = MVar closure */
1282 mvar = (StgMVar *)R1.p;
1285 info = LOCK_CLOSURE(mvar);
1287 info = GET_INFO(mvar);
1290 if (info == &stg_EMPTY_MVAR_info) {
1293 /* unlock the MVar */
1294 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1297 /* HACK: we need a pointer to pass back,
1298 * so we abuse NO_FINALIZER_closure
1300 RET_NP(0, &stg_NO_FINALIZER_closure);
1303 /* we got the value... */
1306 if (mvar->head != (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1307 /* There are putMVar(s) waiting...
1308 * wake up the first thread on the queue
1310 ASSERT(mvar->head->why_blocked == BlockedOnMVar);
1312 /* actually perform the putMVar for the thread that we just woke up */
1313 mvar->value = PerformPut(mvar->head);
1315 #if defined(GRAN) || defined(PAR)
1316 /* ToDo: check 2nd arg (mvar) is right */
1317 mvar->head = RET_STGCALL2(StgTSO *,unblockOne,mvar->head,mvar);
1319 mvar->head = RET_STGCALL1(StgTSO *,unblockOne,mvar->head);
1321 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1322 mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1325 /* unlock in the SMP case */
1326 SET_INFO(mvar,&stg_FULL_MVAR_info);
1329 /* No further putMVars, MVar is now empty */
1330 mvar->value = (StgClosure *)&stg_END_TSO_QUEUE_closure;
1332 /* do this last... we might have locked the MVar in the SMP case,
1333 * and writing the info pointer will unlock it.
1335 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1338 TICK_RET_UNBOXED_TUP(1);
1346 const StgInfoTable *info;
1349 /* args: R1 = MVar, R2 = value */
1351 mvar = (StgMVar *)R1.p;
1354 info = LOCK_CLOSURE(mvar);
1356 info = GET_INFO(mvar);
1359 if (info == &stg_FULL_MVAR_info) {
1360 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1361 mvar->head = CurrentTSO;
1363 mvar->tail->link = CurrentTSO;
1365 CurrentTSO->link = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1366 CurrentTSO->why_blocked = BlockedOnMVar;
1367 CurrentTSO->block_info.closure = (StgClosure *)mvar;
1368 mvar->tail = CurrentTSO;
1371 /* unlock the MVar */
1372 SET_INFO(mvar,&stg_FULL_MVAR_info);
1374 JMP_(stg_block_putmvar);
1377 if (mvar->head != (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1378 /* There are takeMVar(s) waiting: wake up the first one
1380 ASSERT(mvar->head->why_blocked == BlockedOnMVar);
1382 /* actually perform the takeMVar */
1383 PerformTake(mvar->head, R2.cl);
1385 #if defined(GRAN) || defined(PAR)
1386 /* ToDo: check 2nd arg (mvar) is right */
1387 mvar->head = RET_STGCALL2(StgTSO *,unblockOne,mvar->head,mvar);
1389 mvar->head = RET_STGCALL1(StgTSO *,unblockOne,mvar->head);
1391 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1392 mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1395 /* unlocks the MVar in the SMP case */
1396 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1398 JMP_(ENTRY_CODE(Sp[0]));
1400 /* No further takes, the MVar is now full. */
1401 mvar->value = R2.cl;
1402 /* unlocks the MVar in the SMP case */
1403 SET_INFO(mvar,&stg_FULL_MVAR_info);
1404 JMP_(ENTRY_CODE(Sp[0]));
1407 /* ToDo: yield afterward for better communication performance? */
1411 FN_(tryPutMVarzh_fast)
1414 const StgInfoTable *info;
1417 /* args: R1 = MVar, R2 = value */
1419 mvar = (StgMVar *)R1.p;
1422 info = LOCK_CLOSURE(mvar);
1424 info = GET_INFO(mvar);
1427 if (info == &stg_FULL_MVAR_info) {
1430 /* unlock the MVar */
1431 mvar->header.info = &stg_FULL_MVAR_info;
1437 if (mvar->head != (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1438 /* There are takeMVar(s) waiting: wake up the first one
1440 ASSERT(mvar->head->why_blocked == BlockedOnMVar);
1442 /* actually perform the takeMVar */
1443 PerformTake(mvar->head, R2.cl);
1445 #if defined(GRAN) || defined(PAR)
1446 /* ToDo: check 2nd arg (mvar) is right */
1447 mvar->head = RET_STGCALL2(StgTSO *,unblockOne,mvar->head,mvar);
1449 mvar->head = RET_STGCALL1(StgTSO *,unblockOne,mvar->head);
1451 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1452 mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1455 /* unlocks the MVar in the SMP case */
1456 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1458 JMP_(ENTRY_CODE(Sp[0]));
1460 /* No further takes, the MVar is now full. */
1461 mvar->value = R2.cl;
1462 /* unlocks the MVar in the SMP case */
1463 SET_INFO(mvar,&stg_FULL_MVAR_info);
1464 JMP_(ENTRY_CODE(Sp[0]));
1467 /* ToDo: yield afterward for better communication performance? */
1471 /* -----------------------------------------------------------------------------
1472 Stable pointer primitives
1473 ------------------------------------------------------------------------- */
1475 FN_(makeStableNamezh_fast)
1478 StgStableName *sn_obj;
1481 HP_CHK_GEN_TICKY(sizeofW(StgStableName), R1_PTR, makeStableNamezh_fast);
1482 TICK_ALLOC_PRIM(sizeofW(StgHeader),
1483 sizeofW(StgStableName)-sizeofW(StgHeader), 0);
1484 CCS_ALLOC(CCCS,sizeofW(StgStableName)); /* ccs prof */
1486 index = RET_STGCALL1(StgWord,lookupStableName,R1.p);
1488 /* Is there already a StableName for this heap object? */
1489 if (stable_ptr_table[index].sn_obj == NULL) {
1490 sn_obj = (StgStableName *) (Hp - sizeofW(StgStableName) + 1);
1491 SET_HDR(sn_obj,&stg_STABLE_NAME_info,CCCS);
1493 stable_ptr_table[index].sn_obj = (StgClosure *)sn_obj;
1495 (StgClosure *)sn_obj = stable_ptr_table[index].sn_obj;
1498 TICK_RET_UNBOXED_TUP(1);
1503 FN_(makeStablePtrzh_fast)
1508 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1509 sp = RET_STGCALL1(StgStablePtr,getStablePtr,R1.p);
1514 FN_(deRefStablePtrzh_fast)
1516 /* Args: R1 = the stable ptr */
1520 sp = (StgStablePtr)R1.w;
1521 r = stable_ptr_table[(StgWord)sp].addr;
1526 /* -----------------------------------------------------------------------------
1527 Bytecode object primitives
1528 ------------------------------------------------------------------------- */
1541 StgArrWords *bitmap_arr;
1544 bitmap_arr = (StgArrWords *)R6.cl;
1545 size = sizeofW(StgBCO) + bitmap_arr->words;
1546 HP_CHK_GEN_TICKY(size,R1_PTR|R2_PTR|R3_PTR|R4_PTR|R6_PTR, newBCOzh_fast);
1547 TICK_ALLOC_PRIM(size, size-sizeofW(StgHeader), 0);
1548 CCS_ALLOC(CCCS,size); /* ccs prof */
1549 bco = (StgBCO *) (Hp + 1 - size);
1550 SET_HDR(bco, (const StgInfoTable *)&stg_BCO_info, CCCS);
1552 bco->instrs = (StgArrWords*)R1.cl;
1553 bco->literals = (StgArrWords*)R2.cl;
1554 bco->ptrs = (StgMutArrPtrs*)R3.cl;
1555 bco->itbls = (StgArrWords*)R4.cl;
1559 // Copy the arity/bitmap info into the BCO
1562 for (i = 0; i < bitmap_arr->words; i++) {
1563 bco->bitmap[i] = bitmap_arr->payload[i];
1567 TICK_RET_UNBOXED_TUP(1);
1572 FN_(mkApUpd0zh_fast)
1574 // R1.p = the BCO# for the AP
1579 // This function is *only* used to wrap zero-arity BCOs in an
1580 // updatable wrapper (see ByteCodeLink.lhs). An AP thunk is always
1581 // saturated and always points directly to a FUN or BCO.
1582 ASSERT(get_itbl(R1.cl)->type == BCO && ((StgBCO *)R1.p)->arity == 0);
1584 HP_CHK_GEN_TICKY(PAP_sizeW(0), R1_PTR, mkApUpd0zh_fast);
1585 TICK_ALLOC_PRIM(sizeofW(StgHeader), PAP_sizeW(0)-sizeofW(StgHeader), 0);
1586 CCS_ALLOC(CCCS,PAP_sizeW(0)); /* ccs prof */
1587 ap = (StgPAP *) (Hp + 1 - PAP_sizeW(0));
1588 SET_HDR(ap, &stg_AP_info, CCCS);
1593 TICK_RET_UNBOXED_TUP(1);
1598 /* -----------------------------------------------------------------------------
1599 Thread I/O blocking primitives
1600 -------------------------------------------------------------------------- */
1602 FN_(waitReadzh_fast)
1606 ASSERT(CurrentTSO->why_blocked == NotBlocked);
1607 CurrentTSO->why_blocked = BlockedOnRead;
1608 CurrentTSO->block_info.fd = R1.i;
1609 ACQUIRE_LOCK(&sched_mutex);
1610 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1611 RELEASE_LOCK(&sched_mutex);
1612 JMP_(stg_block_noregs);
1616 FN_(waitWritezh_fast)
1620 ASSERT(CurrentTSO->why_blocked == NotBlocked);
1621 CurrentTSO->why_blocked = BlockedOnWrite;
1622 CurrentTSO->block_info.fd = R1.i;
1623 ACQUIRE_LOCK(&sched_mutex);
1624 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1625 RELEASE_LOCK(&sched_mutex);
1626 JMP_(stg_block_noregs);
1632 #ifdef mingw32_TARGET_OS
1633 StgAsyncIOResult* ares;
1640 /* args: R1.i (microsecond delay amount) */
1641 ASSERT(CurrentTSO->why_blocked == NotBlocked);
1642 CurrentTSO->why_blocked = BlockedOnDelay;
1644 ACQUIRE_LOCK(&sched_mutex);
1645 #ifdef mingw32_TARGET_OS
1646 /* could probably allocate this on the heap instead */
1647 ares = (StgAsyncIOResult*)RET_STGCALL2(P_,stgMallocBytes,sizeof(StgAsyncIOResult), "delayzh_fast");
1648 reqID = RET_STGCALL1(W_,addDelayRequest,R1.i);
1649 ares->reqID = reqID;
1652 CurrentTSO->block_info.async_result = ares;
1653 /* Having all async-blocked threads reside on the blocked_queue simplifies matters, so
1654 * change the status to OnDoProc & put the delayed thread on the blocked_queue.
1656 CurrentTSO->why_blocked = BlockedOnDoProc;
1657 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1659 target = ((R1.i + TICK_MILLISECS*1000-1) / (TICK_MILLISECS*1000)) + getourtimeofday();
1660 CurrentTSO->block_info.target = target;
1662 /* Insert the new thread in the sleeping queue. */
1665 while (t != END_TSO_QUEUE && t->block_info.target < target) {
1670 CurrentTSO->link = t;
1672 sleeping_queue = CurrentTSO;
1674 prev->link = CurrentTSO;
1677 RELEASE_LOCK(&sched_mutex);
1678 JMP_(stg_block_noregs);
1682 #ifdef mingw32_TARGET_OS
1683 FN_(asyncReadzh_fast)
1685 StgAsyncIOResult* ares;
1688 /* args: R1.i = fd, R2.i = isSock, R3.i = len, R4.p = buf */
1689 ASSERT(CurrentTSO->why_blocked == NotBlocked);
1690 CurrentTSO->why_blocked = BlockedOnRead;
1691 ACQUIRE_LOCK(&sched_mutex);
1692 /* could probably allocate this on the heap instead */
1693 ares = (StgAsyncIOResult*)RET_STGCALL2(P_,stgMallocBytes,sizeof(StgAsyncIOResult), "asyncReadzh_fast");
1694 reqID = RET_STGCALL5(W_,addIORequest,R1.i,FALSE,R2.i,R3.i,(char*)R4.p);
1695 ares->reqID = reqID;
1698 CurrentTSO->block_info.async_result = ares;
1699 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1700 RELEASE_LOCK(&sched_mutex);
1701 JMP_(stg_block_async);
1705 FN_(asyncWritezh_fast)
1707 StgAsyncIOResult* ares;
1711 /* args: R1.i = fd, R2.i = isSock, R3.i = len, R4.p = buf */
1712 ASSERT(CurrentTSO->why_blocked == NotBlocked);
1713 CurrentTSO->why_blocked = BlockedOnWrite;
1714 ACQUIRE_LOCK(&sched_mutex);
1715 ares = (StgAsyncIOResult*)RET_STGCALL2(P_,stgMallocBytes,sizeof(StgAsyncIOResult), "asyncWritezh_fast");
1716 reqID = RET_STGCALL5(W_,addIORequest,R1.i,TRUE,R2.i,R3.i,(char*)R4.p);
1717 ares->reqID = reqID;
1720 CurrentTSO->block_info.async_result = ares;
1721 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1722 RELEASE_LOCK(&sched_mutex);
1723 JMP_(stg_block_async);
1727 FN_(asyncDoProczh_fast)
1729 StgAsyncIOResult* ares;
1732 /* args: R1.i = proc, R2.i = param */
1733 ASSERT(CurrentTSO->why_blocked == NotBlocked);
1734 CurrentTSO->why_blocked = BlockedOnDoProc;
1735 ACQUIRE_LOCK(&sched_mutex);
1736 /* could probably allocate this on the heap instead */
1737 ares = (StgAsyncIOResult*)RET_STGCALL2(P_,stgMallocBytes,sizeof(StgAsyncIOResult), "asyncDoProczh_fast");
1738 reqID = RET_STGCALL2(W_,addDoProcRequest,R1.p,R2.p);
1739 ares->reqID = reqID;
1742 CurrentTSO->block_info.async_result = ares;
1743 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1744 RELEASE_LOCK(&sched_mutex);
1745 JMP_(stg_block_async);