1 /* -----------------------------------------------------------------------------
2 * $Id: PrimOps.hc,v 1.91 2002/01/29 16:24:08 simonmar Exp $
4 * (c) The GHC Team, 1998-2000
6 * Primitive functions / data
8 * ---------------------------------------------------------------------------*/
14 #include "StgStartup.h"
19 #include "BlockAlloc.h" /* tmp */
20 #include "StablePriv.h"
27 classes CCallable and CReturnable don't really exist, but the
28 compiler insists on generating dictionaries containing references
29 to GHC_ZcCCallable_static_info etc., so we provide dummy symbols
30 for these. Some C compilers can't cope with zero-length static arrays,
31 so we have to make these one element long.
34 StgWord GHC_ZCCCallable_static_info[1];
35 StgWord GHC_ZCCReturnable_static_info[1];
37 /* -----------------------------------------------------------------------------
38 Macros for Hand-written primitives.
39 -------------------------------------------------------------------------- */
42 * Horrible macros for returning unboxed tuples.
44 * How an unboxed tuple is returned depends on two factors:
45 * - the number of real registers we have available
46 * - the boxedness of the returned fields.
48 * To return an unboxed tuple from a primitive operation, we have macros
49 * RET_<layout> where <layout> describes the boxedness of each field of the
50 * unboxed tuple: N indicates a non-pointer field, and P indicates a pointer.
52 * We only define the cases actually used, to avoid having too much
53 * garbage in this section. Warning: any bugs in here will be hard to
57 /*------ All Regs available */
59 # define RET_P(a) R1.w = (W_)(a); JMP_(ENTRY_CODE(Sp[0]));
60 # define RET_N(a) RET_P(a)
62 # define RET_PP(a,b) R1.w = (W_)(a); R2.w = (W_)(b); JMP_(ENTRY_CODE(Sp[0]));
63 # define RET_NN(a,b) RET_PP(a,b)
64 # define RET_NP(a,b) RET_PP(a,b)
66 # define RET_PPP(a,b,c) \
67 R1.w = (W_)(a); R2.w = (W_)(b); R3.w = (W_)(c); JMP_(ENTRY_CODE(Sp[0]));
68 # define RET_NNP(a,b,c) RET_PPP(a,b,c)
70 # define RET_NNNP(a,b,c,d) \
71 R1.w = (W_)(a); R2.w = (W_)(b); R3.w = (W_)(c); R4.w = (W_)d; \
72 JMP_(ENTRY_CODE(Sp[0]));
74 # define RET_NPNP(a,b,c,d) \
75 R1.w = (W_)(a); R2.w = (W_)(b); R3.w = (W_)(c); R4.w = (W_)(d); \
76 JMP_(ENTRY_CODE(Sp[0]));
78 # define RET_NNPNNP(a,b,c,d,e,f) \
79 R1.w = (W_)(a); R2.w = (W_)(b); R3.w = (W_)(c); \
80 R4.w = (W_)(d); R5.w = (W_)(e); R6.w = (W_)(f); \
81 JMP_(ENTRY_CODE(Sp[0]));
83 #elif defined(REG_R7) || defined(REG_R6) || defined(REG_R5) || \
84 defined(REG_R4) || defined(REG_R3)
85 # error RET_n macros not defined for this setup.
87 /*------ 2 Registers available */
90 # define RET_P(a) R1.w = (W_)(a); JMP_(ENTRY_CODE(Sp[0]));
91 # define RET_N(a) RET_P(a)
93 # define RET_PP(a,b) R1.w = (W_)(a); R2.w = (W_)(b); \
94 JMP_(ENTRY_CODE(Sp[0]));
95 # define RET_NN(a,b) RET_PP(a,b)
96 # define RET_NP(a,b) RET_PP(a,b)
98 # define RET_PPP(a,b,c) \
99 R1.w = (W_)(a); R2.w = (W_)(b); Sp[-1] = (W_)(c); Sp -= 1; \
100 JMP_(ENTRY_CODE(Sp[1]));
101 # define RET_NNP(a,b,c) \
102 R1.w = (W_)(a); R2.w = (W_)(b); Sp[-1] = (W_)(c); Sp -= 1; \
103 JMP_(ENTRY_CODE(Sp[1]));
105 # define RET_NNNP(a,b,c,d) \
108 /* Sp[-3] = ARGTAG(1); */ \
112 JMP_(ENTRY_CODE(Sp[3]));
114 # define RET_NPNP(a,b,c,d) \
117 /* Sp[-3] = ARGTAG(1); */ \
121 JMP_(ENTRY_CODE(Sp[3]));
123 # define RET_NNPNNP(a,b,c,d,e,f) \
127 /* Sp[-5] = ARGTAG(1); */ \
129 /* Sp[-3] = ARGTAG(1); */ \
133 JMP_(ENTRY_CODE(Sp[6]));
135 /*------ 1 Register available */
136 #elif defined(REG_R1)
137 # define RET_P(a) R1.w = (W_)(a); JMP_(ENTRY_CODE(Sp[0]));
138 # define RET_N(a) RET_P(a)
140 # define RET_PP(a,b) R1.w = (W_)(a); Sp[-1] = (W_)(b); Sp -= 1; \
141 JMP_(ENTRY_CODE(Sp[1]));
142 # define RET_NN(a,b) R1.w = (W_)(a); Sp[-1] = (W_)(b); Sp -= 2; \
143 JMP_(ENTRY_CODE(Sp[2]));
144 # define RET_NP(a,b) RET_PP(a,b)
146 # define RET_PPP(a,b,c) \
147 R1.w = (W_)(a); Sp[-2] = (W_)(b); Sp[-1] = (W_)(c); Sp -= 2; \
148 JMP_(ENTRY_CODE(Sp[2]));
149 # define RET_NNP(a,b,c) \
150 R1.w = (W_)(a); Sp[-2] = (W_)(b); Sp[-1] = (W_)(c); Sp -= 3; \
151 JMP_(ENTRY_CODE(Sp[3]));
153 # define RET_NNNP(a,b,c,d) \
155 /* Sp[-5] = ARGTAG(1); */ \
157 /* Sp[-3] = ARGTAG(1); */ \
161 JMP_(ENTRY_CODE(Sp[5]));
163 # define RET_NPNP(a,b,c,d) \
166 /* Sp[-3] = ARGTAG(1); */ \
170 JMP_(ENTRY_CODE(Sp[4]));
172 # define RET_NNPNNP(a,b,c,d,e,f) \
176 /* Sp[-3] = ARGTAG(1); */ \
178 /* Sp[-5] = ARGTAG(1); */ \
181 /* Sp[-8] = ARGTAG(1); */ \
183 JMP_(ENTRY_CODE(Sp[8]));
185 #else /* 0 Regs available */
187 #define PUSH_P(o,x) Sp[-o] = (W_)(x)
190 #define PUSH_N(o,x) Sp[1-o] = (W_)(x); Sp[-o] = ARG_TAG(1);
192 #define PUSH_N(o,x) Sp[1-o] = (W_)(x);
195 #define PUSHED(m) Sp -= (m); JMP_(ENTRY_CODE(Sp[m]));
197 /* Here's how to construct these macros:
199 * N = number of N's in the name;
200 * P = number of P's in the name;
202 * while (nonNull(name)) {
203 * if (nextChar == 'P') {
214 # define RET_P(a) PUSH_P(1,a); PUSHED(1)
215 # define RET_N(a) PUSH_N(2,a); PUSHED(2)
217 # define RET_PP(a,b) PUSH_P(2,a); PUSH_P(1,b); PUSHED(2)
218 # define RET_NN(a,b) PUSH_N(4,a); PUSH_N(2,b); PUSHED(4)
219 # define RET_NP(a,b) PUSH_N(3,a); PUSH_P(1,b); PUSHED(3)
221 # define RET_PPP(a,b,c) PUSH_P(3,a); PUSH_P(2,b); PUSH_P(1,c); PUSHED(3)
222 # define RET_NNP(a,b,c) PUSH_N(5,a); PUSH_N(3,b); PUSH_P(1,c); PUSHED(5)
224 # define RET_NNNP(a,b,c,d) PUSH_N(7,a); PUSH_N(5,b); PUSH_N(3,c); PUSH_P(1,d); PUSHED(7)
225 # define RET_NPNP(a,b,c,d) PUSH_N(6,a); PUSH_P(4,b); PUSH_N(3,c); PUSH_P(1,d); PUSHED(6)
226 # define RET_NNPNNP(a,b,c,d,e,f) PUSH_N(10,a); PUSH_N(8,b); PUSH_P(6,c); PUSH_N(5,d); PUSH_N(3,e); PUSH_P(1,f); PUSHED(10)
230 /*-----------------------------------------------------------------------------
233 Basically just new*Array - the others are all inline macros.
235 The size arg is always passed in R1, and the result returned in R1.
237 The slow entry point is for returning from a heap check, the saved
238 size argument must be re-loaded from the stack.
239 -------------------------------------------------------------------------- */
241 /* for objects that are *less* than the size of a word, make sure we
242 * round up to the nearest word for the size of the array.
245 #define BYTES_TO_STGWORDS(n) ((n) + sizeof(W_) - 1)/sizeof(W_)
247 FN_(newByteArrayzh_fast) \
249 W_ size, stuff_size, n; \
252 MAYBE_GC(NO_PTRS,newByteArrayzh_fast); \
254 stuff_size = BYTES_TO_STGWORDS(n); \
255 size = sizeofW(StgArrWords)+ stuff_size; \
256 p = (StgArrWords *)RET_STGCALL1(P_,allocate,size); \
257 TICK_ALLOC_PRIM(sizeofW(StgArrWords),stuff_size,0); \
258 SET_HDR(p, &stg_ARR_WORDS_info, CCCS); \
259 p->words = stuff_size; \
260 TICK_RET_UNBOXED_TUP(1) \
265 FN_(newPinnedByteArrayzh_fast) \
267 W_ size, stuff_size, n; \
270 MAYBE_GC(NO_PTRS,newPinnedByteArrayzh_fast); \
272 stuff_size = BYTES_TO_STGWORDS(n); \
273 size = sizeofW(StgArrWords)+ stuff_size; \
274 p = (StgArrWords *)RET_STGCALL1(P_,allocatePinned,size); \
275 TICK_ALLOC_PRIM(sizeofW(StgArrWords),stuff_size,0); \
276 SET_HDR(p, &stg_ARR_WORDS_info, CCCS); \
277 p->words = stuff_size; \
278 TICK_RET_UNBOXED_TUP(1) \
291 MAYBE_GC(R2_PTR,newArrayzh_fast);
293 size = sizeofW(StgMutArrPtrs) + n;
294 arr = (StgMutArrPtrs *)RET_STGCALL1(P_, allocate, size);
295 TICK_ALLOC_PRIM(sizeofW(StgMutArrPtrs), n, 0);
297 SET_HDR(arr,&stg_MUT_ARR_PTRS_info,CCCS);
301 for (p = (P_)arr + sizeofW(StgMutArrPtrs);
302 p < (P_)arr + size; p++) {
306 TICK_RET_UNBOXED_TUP(1);
311 FN_(newMutVarzh_fast)
314 /* Args: R1.p = initialisation value */
317 HP_CHK_GEN_TICKY(sizeofW(StgMutVar), R1_PTR, newMutVarzh_fast,);
318 TICK_ALLOC_PRIM(sizeofW(StgHeader)+1,1, 0); /* hack, dependent on rep. */
319 CCS_ALLOC(CCCS,sizeofW(StgMutVar));
321 mv = (StgMutVar *)(Hp-sizeofW(StgMutVar)+1);
322 SET_HDR(mv,&stg_MUT_VAR_info,CCCS);
325 TICK_RET_UNBOXED_TUP(1);
330 /* -----------------------------------------------------------------------------
331 Foreign Object Primitives
332 -------------------------------------------------------------------------- */
334 FN_(mkForeignObjzh_fast)
336 /* R1.p = ptr to foreign object,
338 StgForeignObj *result;
341 HP_CHK_GEN_TICKY(sizeofW(StgForeignObj), NO_PTRS, mkForeignObjzh_fast,);
342 TICK_ALLOC_PRIM(sizeofW(StgHeader),
343 sizeofW(StgForeignObj)-sizeofW(StgHeader), 0);
344 CCS_ALLOC(CCCS,sizeofW(StgForeignObj)); /* ccs prof */
346 result = (StgForeignObj *) (Hp + 1 - sizeofW(StgForeignObj));
347 SET_HDR(result,&stg_FOREIGN_info,CCCS);
350 /* returns (# s#, ForeignObj# #) */
351 TICK_RET_UNBOXED_TUP(1);
356 /* These two are out-of-line for the benefit of the NCG */
357 FN_(unsafeThawArrayzh_fast)
360 SET_INFO((StgClosure *)R1.cl,&stg_MUT_ARR_PTRS_info);
361 recordMutable((StgMutClosure*)R1.cl);
363 TICK_RET_UNBOXED_TUP(1);
368 /* -----------------------------------------------------------------------------
369 Weak Pointer Primitives
370 -------------------------------------------------------------------------- */
376 R3.p = finalizer (or NULL)
382 R3.cl = &stg_NO_FINALIZER_closure;
385 HP_CHK_GEN_TICKY(sizeofW(StgWeak),R1_PTR|R2_PTR|R3_PTR, mkWeakzh_fast,);
386 TICK_ALLOC_PRIM(sizeofW(StgHeader)+1, // +1 is for the link field
387 sizeofW(StgWeak)-sizeofW(StgHeader)-1, 0);
388 CCS_ALLOC(CCCS,sizeofW(StgWeak)); /* ccs prof */
390 w = (StgWeak *) (Hp + 1 - sizeofW(StgWeak));
391 SET_HDR(w, &stg_WEAK_info, CCCS);
395 w->finalizer = R3.cl;
397 w->link = weak_ptr_list;
399 IF_DEBUG(weak, fprintf(stderr,"New weak pointer at %p\n",w));
401 TICK_RET_UNBOXED_TUP(1);
406 FN_(finalizzeWeakzh_fast)
413 TICK_RET_UNBOXED_TUP(0);
414 w = (StgDeadWeak *)R1.p;
417 if (w->header.info == &stg_DEAD_WEAK_info) {
418 RET_NP(0,&stg_NO_FINALIZER_closure);
424 // A weak pointer is inherently used, so we do not need to call
425 // LDV_recordDead_FILL_SLOP_DYNAMIC():
426 // LDV_recordDead_FILL_SLOP_DYNAMIC((StgClosure *)w);
427 // or, LDV_recordDead():
428 // LDV_recordDead((StgClosure *)w, sizeofW(StgWeak) - sizeofW(StgProfHeader));
429 // Furthermore, when PROFILING is turned on, dead weak pointers are exactly as
430 // large as weak pointers, so there is no need to fill the slop, either.
431 // See stg_DEAD_WEAK_info in StgMiscClosures.hc.
434 // Todo: maybe use SET_HDR() and remove LDV_recordCreate()?
436 w->header.info = &stg_DEAD_WEAK_info;
439 LDV_recordCreate((StgClosure *)w);
441 f = ((StgWeak *)w)->finalizer;
442 w->link = ((StgWeak *)w)->link;
444 /* return the finalizer */
445 if (f == &stg_NO_FINALIZER_closure) {
446 RET_NP(0,&stg_NO_FINALIZER_closure);
453 FN_(deRefWeakzh_fast)
455 /* R1.p = weak ptr */
461 if (w->header.info == &stg_WEAK_info) {
463 val = (P_)((StgWeak *)w)->value;
472 /* -----------------------------------------------------------------------------
473 Arbitrary-precision Integer operations.
474 -------------------------------------------------------------------------- */
476 FN_(int2Integerzh_fast)
478 /* arguments: R1 = Int# */
480 I_ val, s; /* to avoid aliasing */
481 StgArrWords* p; /* address of array result */
485 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+1, NO_PTRS, int2Integerzh_fast,);
486 TICK_ALLOC_PRIM(sizeofW(StgArrWords),1,0);
487 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+1); /* ccs prof */
489 p = (StgArrWords *)Hp - 1;
490 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, 1);
492 /* mpz_set_si is inlined here, makes things simpler */
496 } else if (val > 0) {
503 /* returns (# size :: Int#,
507 TICK_RET_UNBOXED_TUP(2);
512 FN_(word2Integerzh_fast)
514 /* arguments: R1 = Word# */
516 W_ val; /* to avoid aliasing */
518 StgArrWords* p; /* address of array result */
522 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+1, NO_PTRS, word2Integerzh_fast,)
523 TICK_ALLOC_PRIM(sizeofW(StgArrWords),1,0);
524 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+1); /* ccs prof */
526 p = (StgArrWords *)Hp - 1;
527 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, 1);
536 /* returns (# size :: Int#,
540 TICK_RET_UNBOXED_TUP(2);
547 * 'long long' primops for converting to/from Integers.
550 #ifdef SUPPORT_LONG_LONGS
552 FN_(int64ToIntegerzh_fast)
554 /* arguments: L1 = Int64# */
556 StgInt64 val; /* to avoid aliasing */
558 I_ s, neg, words_needed;
559 StgArrWords* p; /* address of array result */
565 if ( val >= 0x100000000LL || val <= -0x100000000LL ) {
568 /* minimum is one word */
571 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+words_needed, NO_PTRS, int64ToIntegerzh_fast,)
572 TICK_ALLOC_PRIM(sizeofW(StgArrWords),words_needed,0);
573 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+words_needed); /* ccs prof */
575 p = (StgArrWords *)(Hp-words_needed+1) - 1;
576 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, words_needed);
583 hi = (W_)((LW_)val / 0x100000000ULL);
585 if ( words_needed == 2 ) {
589 } else if ( val != 0 ) {
592 } else /* val==0 */ {
595 s = ( neg ? -s : s );
597 /* returns (# size :: Int#,
601 TICK_RET_UNBOXED_TUP(2);
606 FN_(word64ToIntegerzh_fast)
608 /* arguments: L1 = Word64# */
610 StgWord64 val; /* to avoid aliasing */
613 StgArrWords* p; /* address of array result */
617 if ( val >= 0x100000000ULL ) {
622 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+words_needed, NO_PTRS, word64ToIntegerzh_fast,)
623 TICK_ALLOC_PRIM(sizeofW(StgArrWords),words_needed,0);
624 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+words_needed); /* ccs prof */
626 p = (StgArrWords *)(Hp-words_needed+1) - 1;
627 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, words_needed);
629 hi = (W_)((LW_)val / 0x100000000ULL);
630 if ( val >= 0x100000000ULL ) {
634 } else if ( val != 0 ) {
637 } else /* val==0 */ {
641 /* returns (# size :: Int#,
645 TICK_RET_UNBOXED_TUP(2);
651 #endif /* SUPPORT_LONG_LONGS */
653 /* ToDo: this is shockingly inefficient */
655 #define GMP_TAKE2_RET1(name,mp_fun) \
658 MP_INT arg1, arg2, result; \
664 /* call doYouWantToGC() */ \
665 MAYBE_GC(R2_PTR | R4_PTR, name); \
667 d1 = (StgArrWords *)R2.p; \
669 d2 = (StgArrWords *)R4.p; \
672 arg1._mp_alloc = d1->words; \
673 arg1._mp_size = (s1); \
674 arg1._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
675 arg2._mp_alloc = d2->words; \
676 arg2._mp_size = (s2); \
677 arg2._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d2)); \
679 STGCALL1(mpz_init,&result); \
681 /* Perform the operation */ \
682 STGCALL3(mp_fun,&result,&arg1,&arg2); \
684 TICK_RET_UNBOXED_TUP(2); \
685 RET_NP(result._mp_size, \
686 result._mp_d-sizeofW(StgArrWords)); \
690 #define GMP_TAKE1_RET1(name,mp_fun) \
693 MP_INT arg1, result; \
698 /* call doYouWantToGC() */ \
699 MAYBE_GC(R2_PTR, name); \
701 d1 = (StgArrWords *)R2.p; \
704 arg1._mp_alloc = d1->words; \
705 arg1._mp_size = (s1); \
706 arg1._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
708 STGCALL1(mpz_init,&result); \
710 /* Perform the operation */ \
711 STGCALL2(mp_fun,&result,&arg1); \
713 TICK_RET_UNBOXED_TUP(2); \
714 RET_NP(result._mp_size, \
715 result._mp_d-sizeofW(StgArrWords)); \
719 #define GMP_TAKE2_RET2(name,mp_fun) \
722 MP_INT arg1, arg2, result1, result2; \
728 /* call doYouWantToGC() */ \
729 MAYBE_GC(R2_PTR | R4_PTR, name); \
731 d1 = (StgArrWords *)R2.p; \
733 d2 = (StgArrWords *)R4.p; \
736 arg1._mp_alloc = d1->words; \
737 arg1._mp_size = (s1); \
738 arg1._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
739 arg2._mp_alloc = d2->words; \
740 arg2._mp_size = (s2); \
741 arg2._mp_d = (unsigned long int *) (BYTE_ARR_CTS(d2)); \
743 STGCALL1(mpz_init,&result1); \
744 STGCALL1(mpz_init,&result2); \
746 /* Perform the operation */ \
747 STGCALL4(mp_fun,&result1,&result2,&arg1,&arg2); \
749 TICK_RET_UNBOXED_TUP(4); \
750 RET_NPNP(result1._mp_size, \
751 result1._mp_d-sizeofW(StgArrWords), \
753 result2._mp_d-sizeofW(StgArrWords)); \
757 GMP_TAKE2_RET1(plusIntegerzh_fast, mpz_add);
758 GMP_TAKE2_RET1(minusIntegerzh_fast, mpz_sub);
759 GMP_TAKE2_RET1(timesIntegerzh_fast, mpz_mul);
760 GMP_TAKE2_RET1(gcdIntegerzh_fast, mpz_gcd);
761 GMP_TAKE2_RET1(quotIntegerzh_fast, mpz_tdiv_q);
762 GMP_TAKE2_RET1(remIntegerzh_fast, mpz_tdiv_r);
763 GMP_TAKE2_RET1(divExactIntegerzh_fast, mpz_divexact);
764 GMP_TAKE2_RET1(andIntegerzh_fast, mpz_and);
765 GMP_TAKE2_RET1(orIntegerzh_fast, mpz_ior);
766 GMP_TAKE2_RET1(xorIntegerzh_fast, mpz_xor);
767 GMP_TAKE1_RET1(complementIntegerzh_fast, mpz_com);
769 GMP_TAKE2_RET2(quotRemIntegerzh_fast, mpz_tdiv_qr);
770 GMP_TAKE2_RET2(divModIntegerzh_fast, mpz_fdiv_qr);
775 /* R1 = the first Int#; R2 = the second Int# */
779 aa = (mp_limb_t)(R1.i);
780 r = RET_STGCALL3(StgInt, mpn_gcd_1, (mp_limb_t *)(&aa), 1, (mp_limb_t)(R2.i));
785 FN_(gcdIntegerIntzh_fast)
787 /* R1 = s1; R2 = d1; R3 = the int */
790 r = RET_STGCALL3(I_,mpn_gcd_1,(mp_limb_t *)(BYTE_ARR_CTS(R2.p)), R1.i, R3.i);
795 FN_(cmpIntegerIntzh_fast)
797 /* R1 = s1; R2 = d1; R3 = the int */
808 // paraphrased from mpz_cmp_si() in the GMP sources
811 } else if (v_digit < 0) {
816 if (usize != vsize) {
817 RET_N(usize - vsize);
824 u_digit = *(mp_limb_t *)(BYTE_ARR_CTS(R2.p));
826 if (u_digit == (mp_limb_t) (unsigned long) v_digit) {
830 if (u_digit > (mp_limb_t) (unsigned long) v_digit) {
838 FN_(cmpIntegerzh_fast)
840 /* R1 = s1; R2 = d1; R3 = s2; R4 = d2 */
848 // paraphrased from mpz_cmp() in the GMP sources
852 if (usize != vsize) {
853 RET_N(usize - vsize);
862 up = BYTE_ARR_CTS(R2.p);
863 vp = BYTE_ARR_CTS(R4.p);
865 cmp = RET_STGCALL3(I_, mpn_cmp, (mp_limb_t *)up, (mp_limb_t *)vp, size);
871 if ((cmp < 0) == (usize < 0)) {
879 FN_(integer2Intzh_fast)
888 r = ((mp_limb_t *) (BYTE_ARR_CTS(R2.p)))[0];
895 FN_(integer2Wordzh_fast)
905 r = ((mp_limb_t *) (BYTE_ARR_CTS(R2.p)))[0];
913 FN_(decodeFloatzh_fast)
921 /* arguments: F1 = Float# */
924 HP_CHK_GEN_TICKY(sizeofW(StgArrWords)+1, NO_PTRS, decodeFloatzh_fast,);
925 TICK_ALLOC_PRIM(sizeofW(StgArrWords),1,0);
926 CCS_ALLOC(CCCS,sizeofW(StgArrWords)+1); /* ccs prof */
928 /* Be prepared to tell Lennart-coded __decodeFloat */
929 /* where mantissa._mp_d can be put (it does not care about the rest) */
930 p = (StgArrWords *)Hp - 1;
931 SET_ARR_HDR(p,&stg_ARR_WORDS_info,CCCS,1)
932 mantissa._mp_d = (void *)BYTE_ARR_CTS(p);
934 /* Perform the operation */
935 STGCALL3(__decodeFloat,&mantissa,&exponent,arg);
937 /* returns: (Int# (expn), Int#, ByteArray#) */
938 TICK_RET_UNBOXED_TUP(3);
939 RET_NNP(exponent,mantissa._mp_size,p);
943 #define DOUBLE_MANTISSA_SIZE (sizeofW(StgDouble))
944 #define ARR_SIZE (sizeofW(StgArrWords) + DOUBLE_MANTISSA_SIZE)
946 FN_(decodeDoublezh_fast)
953 /* arguments: D1 = Double# */
956 HP_CHK_GEN_TICKY(ARR_SIZE, NO_PTRS, decodeDoublezh_fast,);
957 TICK_ALLOC_PRIM(sizeofW(StgArrWords),DOUBLE_MANTISSA_SIZE,0);
958 CCS_ALLOC(CCCS,ARR_SIZE); /* ccs prof */
960 /* Be prepared to tell Lennart-coded __decodeDouble */
961 /* where mantissa.d can be put (it does not care about the rest) */
962 p = (StgArrWords *)(Hp-ARR_SIZE+1);
963 SET_ARR_HDR(p, &stg_ARR_WORDS_info, CCCS, DOUBLE_MANTISSA_SIZE);
964 mantissa._mp_d = (void *)BYTE_ARR_CTS(p);
966 /* Perform the operation */
967 STGCALL3(__decodeDouble,&mantissa,&exponent,arg);
969 /* returns: (Int# (expn), Int#, ByteArray#) */
970 TICK_RET_UNBOXED_TUP(3);
971 RET_NNP(exponent,mantissa._mp_size,p);
975 /* -----------------------------------------------------------------------------
976 * Concurrency primitives
977 * -------------------------------------------------------------------------- */
982 /* args: R1 = closure to spark */
984 MAYBE_GC(R1_PTR, forkzh_fast);
986 /* create it right now, return ThreadID in R1 */
987 R1.t = RET_STGCALL2(StgTSO *, createIOThread,
988 RtsFlags.GcFlags.initialStkSize, R1.cl);
989 STGCALL1(scheduleThread, R1.t);
991 /* switch at the earliest opportunity */
994 JMP_(ENTRY_CODE(Sp[0]));
1001 JMP_(stg_yield_noregs);
1005 /* -----------------------------------------------------------------------------
1008 * take & putMVar work as follows. Firstly, an important invariant:
1010 * If the MVar is full, then the blocking queue contains only
1011 * threads blocked on putMVar, and if the MVar is empty then the
1012 * blocking queue contains only threads blocked on takeMVar.
1015 * MVar empty : then add ourselves to the blocking queue
1016 * MVar full : remove the value from the MVar, and
1017 * blocking queue empty : return
1018 * blocking queue non-empty : perform the first blocked putMVar
1019 * from the queue, and wake up the
1020 * thread (MVar is now full again)
1022 * putMVar is just the dual of the above algorithm.
1024 * How do we "perform a putMVar"? Well, we have to fiddle around with
1025 * the stack of the thread waiting to do the putMVar. See
1026 * stg_block_putmvar and stg_block_takemvar in HeapStackCheck.c for
1027 * the stack layout, and the PerformPut and PerformTake macros below.
1029 * It is important that a blocked take or put is woken up with the
1030 * take/put already performed, because otherwise there would be a
1031 * small window of vulnerability where the thread could receive an
1032 * exception and never perform its take or put, and we'd end up with a
1035 * -------------------------------------------------------------------------- */
1037 FN_(isEmptyMVarzh_fast)
1039 /* args: R1 = MVar closure */
1042 r = (I_)((GET_INFO((StgMVar*)(R1.p))) == &stg_EMPTY_MVAR_info);
1055 HP_CHK_GEN_TICKY(sizeofW(StgMVar), NO_PTRS, newMVarzh_fast,);
1056 TICK_ALLOC_PRIM(sizeofW(StgMutVar)-1, // consider head,tail,link as admin wds
1058 CCS_ALLOC(CCCS,sizeofW(StgMVar)); /* ccs prof */
1060 mvar = (StgMVar *) (Hp - sizeofW(StgMVar) + 1);
1061 SET_HDR(mvar,&stg_EMPTY_MVAR_info,CCCS);
1062 mvar->head = mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1063 mvar->value = (StgClosure *)&stg_END_TSO_QUEUE_closure;
1065 TICK_RET_UNBOXED_TUP(1);
1070 #define PerformTake(tso, value) ({ \
1071 (tso)->sp[1] = (W_)value; \
1072 (tso)->sp[0] = (W_)&stg_gc_unpt_r1_info; \
1075 #define PerformPut(tso) ({ \
1076 StgClosure *val = (StgClosure *)(tso)->sp[2]; \
1077 (tso)->sp[2] = (W_)&stg_gc_noregs_info; \
1082 FN_(takeMVarzh_fast)
1086 const StgInfoTable *info;
1089 /* args: R1 = MVar closure */
1091 mvar = (StgMVar *)R1.p;
1094 info = LOCK_CLOSURE(mvar);
1096 info = GET_INFO(mvar);
1099 /* If the MVar is empty, put ourselves on its blocking queue,
1100 * and wait until we're woken up.
1102 if (info == &stg_EMPTY_MVAR_info) {
1103 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1104 mvar->head = CurrentTSO;
1106 mvar->tail->link = CurrentTSO;
1108 CurrentTSO->link = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1109 CurrentTSO->why_blocked = BlockedOnMVar;
1110 CurrentTSO->block_info.closure = (StgClosure *)mvar;
1111 mvar->tail = CurrentTSO;
1114 /* unlock the MVar */
1115 mvar->header.info = &stg_EMPTY_MVAR_info;
1117 JMP_(stg_block_takemvar);
1120 /* we got the value... */
1123 if (mvar->head != (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1124 /* There are putMVar(s) waiting...
1125 * wake up the first thread on the queue
1127 ASSERT(mvar->head->why_blocked == BlockedOnMVar);
1129 /* actually perform the putMVar for the thread that we just woke up */
1130 mvar->value = PerformPut(mvar->head);
1132 #if defined(GRAN) || defined(PAR)
1133 /* ToDo: check 2nd arg (mvar) is right */
1134 mvar->head = RET_STGCALL2(StgTSO *,unblockOne,mvar->head,mvar);
1136 mvar->head = RET_STGCALL1(StgTSO *,unblockOne,mvar->head);
1138 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1139 mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1142 /* unlock in the SMP case */
1143 SET_INFO(mvar,&stg_FULL_MVAR_info);
1145 TICK_RET_UNBOXED_TUP(1);
1148 /* No further putMVars, MVar is now empty */
1150 /* do this last... we might have locked the MVar in the SMP case,
1151 * and writing the info pointer will unlock it.
1153 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1154 mvar->value = (StgClosure *)&stg_END_TSO_QUEUE_closure;
1155 TICK_RET_UNBOXED_TUP(1);
1161 FN_(tryTakeMVarzh_fast)
1165 const StgInfoTable *info;
1168 /* args: R1 = MVar closure */
1170 mvar = (StgMVar *)R1.p;
1173 info = LOCK_CLOSURE(mvar);
1175 info = GET_INFO(mvar);
1178 if (info == &stg_EMPTY_MVAR_info) {
1181 /* unlock the MVar */
1182 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1185 /* HACK: we need a pointer to pass back,
1186 * so we abuse NO_FINALIZER_closure
1188 RET_NP(0, &stg_NO_FINALIZER_closure);
1191 /* we got the value... */
1194 if (mvar->head != (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1195 /* There are putMVar(s) waiting...
1196 * wake up the first thread on the queue
1198 ASSERT(mvar->head->why_blocked == BlockedOnMVar);
1200 /* actually perform the putMVar for the thread that we just woke up */
1201 mvar->value = PerformPut(mvar->head);
1203 #if defined(GRAN) || defined(PAR)
1204 /* ToDo: check 2nd arg (mvar) is right */
1205 mvar->head = RET_STGCALL2(StgTSO *,unblockOne,mvar->head,mvar);
1207 mvar->head = RET_STGCALL1(StgTSO *,unblockOne,mvar->head);
1209 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1210 mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1213 /* unlock in the SMP case */
1214 SET_INFO(mvar,&stg_FULL_MVAR_info);
1217 /* No further putMVars, MVar is now empty */
1218 mvar->value = (StgClosure *)&stg_END_TSO_QUEUE_closure;
1220 /* do this last... we might have locked the MVar in the SMP case,
1221 * and writing the info pointer will unlock it.
1223 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1226 TICK_RET_UNBOXED_TUP(1);
1234 const StgInfoTable *info;
1237 /* args: R1 = MVar, R2 = value */
1239 mvar = (StgMVar *)R1.p;
1242 info = LOCK_CLOSURE(mvar);
1244 info = GET_INFO(mvar);
1247 if (info == &stg_FULL_MVAR_info) {
1248 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1249 mvar->head = CurrentTSO;
1251 mvar->tail->link = CurrentTSO;
1253 CurrentTSO->link = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1254 CurrentTSO->why_blocked = BlockedOnMVar;
1255 CurrentTSO->block_info.closure = (StgClosure *)mvar;
1256 mvar->tail = CurrentTSO;
1259 /* unlock the MVar */
1260 SET_INFO(mvar,&stg_FULL_MVAR_info);
1262 JMP_(stg_block_putmvar);
1265 if (mvar->head != (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1266 /* There are takeMVar(s) waiting: wake up the first one
1268 ASSERT(mvar->head->why_blocked == BlockedOnMVar);
1270 /* actually perform the takeMVar */
1271 PerformTake(mvar->head, R2.cl);
1273 #if defined(GRAN) || defined(PAR)
1274 /* ToDo: check 2nd arg (mvar) is right */
1275 mvar->head = RET_STGCALL2(StgTSO *,unblockOne,mvar->head,mvar);
1277 mvar->head = RET_STGCALL1(StgTSO *,unblockOne,mvar->head);
1279 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1280 mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1283 /* unlocks the MVar in the SMP case */
1284 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1286 JMP_(ENTRY_CODE(Sp[0]));
1288 /* No further takes, the MVar is now full. */
1289 mvar->value = R2.cl;
1290 /* unlocks the MVar in the SMP case */
1291 SET_INFO(mvar,&stg_FULL_MVAR_info);
1292 JMP_(ENTRY_CODE(Sp[0]));
1295 /* ToDo: yield afterward for better communication performance? */
1299 FN_(tryPutMVarzh_fast)
1302 const StgInfoTable *info;
1305 /* args: R1 = MVar, R2 = value */
1307 mvar = (StgMVar *)R1.p;
1310 info = LOCK_CLOSURE(mvar);
1312 info = GET_INFO(mvar);
1315 if (info == &stg_FULL_MVAR_info) {
1318 /* unlock the MVar */
1319 mvar->header.info = &stg_FULL_MVAR_info;
1325 if (mvar->head != (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1326 /* There are takeMVar(s) waiting: wake up the first one
1328 ASSERT(mvar->head->why_blocked == BlockedOnMVar);
1330 /* actually perform the takeMVar */
1331 PerformTake(mvar->head, R2.cl);
1333 #if defined(GRAN) || defined(PAR)
1334 /* ToDo: check 2nd arg (mvar) is right */
1335 mvar->head = RET_STGCALL2(StgTSO *,unblockOne,mvar->head,mvar);
1337 mvar->head = RET_STGCALL1(StgTSO *,unblockOne,mvar->head);
1339 if (mvar->head == (StgTSO *)&stg_END_TSO_QUEUE_closure) {
1340 mvar->tail = (StgTSO *)&stg_END_TSO_QUEUE_closure;
1343 /* unlocks the MVar in the SMP case */
1344 SET_INFO(mvar,&stg_EMPTY_MVAR_info);
1346 JMP_(ENTRY_CODE(Sp[0]));
1348 /* No further takes, the MVar is now full. */
1349 mvar->value = R2.cl;
1350 /* unlocks the MVar in the SMP case */
1351 SET_INFO(mvar,&stg_FULL_MVAR_info);
1352 JMP_(ENTRY_CODE(Sp[0]));
1355 /* ToDo: yield afterward for better communication performance? */
1359 /* -----------------------------------------------------------------------------
1360 Stable pointer primitives
1361 ------------------------------------------------------------------------- */
1363 FN_(makeStableNamezh_fast)
1366 StgStableName *sn_obj;
1369 HP_CHK_GEN_TICKY(sizeofW(StgStableName), R1_PTR, makeStableNamezh_fast,);
1370 TICK_ALLOC_PRIM(sizeofW(StgHeader),
1371 sizeofW(StgStableName)-sizeofW(StgHeader), 0);
1372 CCS_ALLOC(CCCS,sizeofW(StgStableName)); /* ccs prof */
1374 index = RET_STGCALL1(StgWord,lookupStableName,R1.p);
1376 /* Is there already a StableName for this heap object? */
1377 if (stable_ptr_table[index].sn_obj == NULL) {
1378 sn_obj = (StgStableName *) (Hp - sizeofW(StgStableName) + 1);
1379 SET_HDR(sn_obj,&stg_STABLE_NAME_info,CCCS);
1381 stable_ptr_table[index].sn_obj = (StgClosure *)sn_obj;
1383 (StgClosure *)sn_obj = stable_ptr_table[index].sn_obj;
1386 TICK_RET_UNBOXED_TUP(1);
1391 FN_(makeStablePtrzh_fast)
1396 MAYBE_GC(R1_PTR, makeStablePtrzh_fast);
1397 sp = RET_STGCALL1(StgStablePtr,getStablePtr,R1.p);
1402 FN_(deRefStablePtrzh_fast)
1404 /* Args: R1 = the stable ptr */
1408 sp = (StgStablePtr)R1.w;
1409 r = stable_ptr_table[(StgWord)sp].addr;
1414 /* -----------------------------------------------------------------------------
1415 Bytecode object primitives
1416 ------------------------------------------------------------------------- */
1428 HP_CHK_GEN_TICKY(sizeofW(StgBCO),R1_PTR|R2_PTR|R3_PTR|R4_PTR, newBCOzh_fast,);
1429 TICK_ALLOC_PRIM(sizeofW(StgHeader), sizeofW(StgBCO)-sizeofW(StgHeader), 0);
1430 CCS_ALLOC(CCCS,sizeofW(StgBCO)); /* ccs prof */
1431 bco = (StgBCO *) (Hp + 1 - sizeofW(StgBCO));
1432 SET_HDR(bco, &stg_BCO_info, CCCS);
1434 bco->instrs = (StgArrWords*)R1.cl;
1435 bco->literals = (StgArrWords*)R2.cl;
1436 bco->ptrs = (StgMutArrPtrs*)R3.cl;
1437 bco->itbls = (StgArrWords*)R4.cl;
1439 TICK_RET_UNBOXED_TUP(1);
1444 FN_(mkApUpd0zh_fast)
1446 /* R1.p = the fn for the AP_UPD
1450 HP_CHK_GEN_TICKY(AP_sizeW(0), R1_PTR, mkApUpd0zh_fast,);
1451 TICK_ALLOC_PRIM(sizeofW(StgHeader), AP_sizeW(0)-sizeofW(StgHeader), 0);
1452 CCS_ALLOC(CCCS,AP_sizeW(0)); /* ccs prof */
1453 ap = (StgAP_UPD *) (Hp + 1 - AP_sizeW(0));
1454 SET_HDR(ap, &stg_AP_UPD_info, CCCS);
1459 TICK_RET_UNBOXED_TUP(1);
1464 /* -----------------------------------------------------------------------------
1465 Thread I/O blocking primitives
1466 -------------------------------------------------------------------------- */
1468 FN_(waitReadzh_fast)
1472 ASSERT(CurrentTSO->why_blocked == NotBlocked);
1473 CurrentTSO->why_blocked = BlockedOnRead;
1474 CurrentTSO->block_info.fd = R1.i;
1475 ACQUIRE_LOCK(&sched_mutex);
1476 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1477 RELEASE_LOCK(&sched_mutex);
1478 JMP_(stg_block_noregs);
1482 FN_(waitWritezh_fast)
1486 ASSERT(CurrentTSO->why_blocked == NotBlocked);
1487 CurrentTSO->why_blocked = BlockedOnWrite;
1488 CurrentTSO->block_info.fd = R1.i;
1489 ACQUIRE_LOCK(&sched_mutex);
1490 APPEND_TO_BLOCKED_QUEUE(CurrentTSO);
1491 RELEASE_LOCK(&sched_mutex);
1492 JMP_(stg_block_noregs);
1502 ASSERT(CurrentTSO->why_blocked == NotBlocked);
1503 CurrentTSO->why_blocked = BlockedOnDelay;
1505 ACQUIRE_LOCK(&sched_mutex);
1507 target = (R1.i / (TICK_MILLISECS*1000)) + getourtimeofday();
1508 CurrentTSO->block_info.target = target;
1510 /* Insert the new thread in the sleeping queue. */
1513 while (t != END_TSO_QUEUE && t->block_info.target < target) {
1518 CurrentTSO->link = t;
1520 sleeping_queue = CurrentTSO;
1522 prev->link = CurrentTSO;
1525 RELEASE_LOCK(&sched_mutex);
1526 JMP_(stg_block_noregs);