2 % (c) The GRASP/AQUA Project, Glasgow University, 1993-1994
4 \section[StgMacros]{C macros used in GHC-generated \tr{.hc} files}
11 %************************************************************************
13 \subsection[StgMacros-abbrev]{Abbreviatory(?) and general macros}
15 %************************************************************************
19 /* for function declarations */
20 #define STGFUN(f) F_ f(STG_NO_ARGS)
21 #define STATICFUN(f) static F_ f(STG_NO_ARGS)
23 /* for functions/data that are really external to this module */
24 #define EXTFUN(f) extern F_ f(STG_NO_ARGS)
25 #define EXTDATA(d) extern W_ d[]
26 #define EXTDATA_RO(d) extern const W_ d[] /* read-only */
28 /* for fwd decls to functions/data somewhere else in this module */
29 /* (identical for the mo') */
30 #define INTFUN(f) static F_ f(STG_NO_ARGS)
31 #define INTDATA(d) extern W_ d[]
32 #define INTDATA_RO(d) extern const W_ d[] /* read-only */
34 /* short forms of most of the above */
36 #define FN_(f) F_ f(STG_NO_ARGS)
37 #define IFN_(f) static F_ f(STG_NO_ARGS)
38 #define EF_(f) extern F_ f(STG_NO_ARGS)
39 #define ED_(d) extern W_ d[]
40 #define ED_RO_(d) extern const W_ d[] /* read-only */
41 #define IF_(f) static F_ f(STG_NO_ARGS)
43 /* GCC is uncooperative about the next one: */
44 /* But, the "extern" prevents initialisation... ADR */
46 #define ID_(d) extern W_ d[]
47 #define ID_RO_(d) extern const W_ d[] /* read-only */
49 #define ID_(d) static W_ d[]
50 #define ID_RO_(d) static const W_ d[] /* read-only */
54 General things; note: general-but-``machine-dependent'' macros are
55 given in \tr{StgMachDeps.lh}.
57 I_ STG_MAX PROTO((I_, I_)); /* GCC -Wall loves prototypes */
61 STG_MAX(I_ a, I_ b) { return((a >= b) ? a : b); }
62 /* NB: the naive #define macro version of STG_MAX
63 can lead to exponential CPP explosion, if you
64 have very-nested STG_MAXes.
68 Macros to combine two short words into a single
69 word and split such a word back into two.
71 Dependent on machine word size :-)
74 #define COMBINE_WORDS(word,short1,short2) \
76 ((packed_shorts *)&(word))->wu.s1 = short1; \
77 ((packed_shorts *)&(word))->wu.s2 = short2; \
80 #define SPLIT_WORD(word,short1,short2) \
82 short1 = ((packed_shorts *)&(word))->wu.s1; \
83 short2 = ((packed_shorts *)&(word))->wu.s2; \
88 %************************************************************************
90 \subsection[StgMacros-gen-stg]{General STGish macros}
92 %************************************************************************
94 Common sizes of vector-return tables.
96 Claim: don't need fwd decls for return pts in \tr{VECTBL*}, because
97 the AbsC flattener ensures that things come out sufficiently
101 #ifdef __STG_REV_TBLS__
102 #define UNVECTBL(staticp,label,a) /* nothing */
104 #define UNVECTBL(staticp,label,a) \
106 staticp const W_ label[] = { \
113 #if defined(USE_SPLIT_MARKERS)
114 #define __STG_SPLIT_MARKER(n) FN_(CAT2(__stg_split_marker,n)){ }
116 #define __STG_SPLIT_MARKER(n) /* nothing */
120 %************************************************************************
122 \subsection[StgMacros-exceptions]{Exception-checking macros}
124 %************************************************************************
126 Argument-satisfaction check, stack(s) overflow check, heap overflow
129 The @SUBTRACT(upper, lower)@ macros return a positive result in words
130 indicating the amount by which upper is above lower on the stack.
133 #define SUBTRACT_A_STK( upper, lower ) AREL( (lower) - (upper) )
134 #define SUBTRACT_B_STK( upper, lower ) BREL( (lower) - (upper) )
137 %************************************************************************
139 \subsubsection[StgMacros-arg-satis]{Argument-satisfaction checks}
141 %************************************************************************
143 @ARGS_CHK(n)@ sees of there are @n@ words of args on the A/B stack.
144 If not, it jumps to @UpdatePAP@.
146 @ARGS_CHK@ args are pre-directionified.
147 Notice that we do the comparisons in the form (x < a+n), for
148 some constant n. This generates more efficient code (with GCC at least)
152 #define ARGS_CHK_A(n) \
153 if (SuA /*SUBTRACT_A_STK( SpA, SuA )*/ < (SpA+(n))) { \
157 #define ARGS_CHK_A_LOAD_NODE(n, closure_addr) \
158 if (SuA /*SUBTRACT_A_STK( SpA, SuA )*/ < (SpA+(n))) { \
159 Node = (P_) closure_addr; \
163 #define ARGS_CHK_B(n) \
164 if (SpB /*SUBTRACT_B_STK( SpB, SuB )*/ < (SuB-(n))) { \
169 #define ARGS_CHK_B_LOAD_NODE(n, closure_addr) \
170 if (SpB /*SUBTRACT_B_STK( SpB, SuB )*/ < (SuB-(n))) { \
171 Node = (P_) closure_addr; \
176 %************************************************************************
178 \subsubsection[StgMacros-stk-chks]{Stack-overflow check}
180 %************************************************************************
182 @STK_CHK(a,b)@ [misc args omitted...] checks that we can allocate @a@
183 words of A stack and @b@ words of B stack. If not, it calls
184 @StackOverflow@ (which dies).
186 (It will be different in the parallel case.)
188 NB: args @a@ and @b@ are pre-direction-ified!
190 I_ SqueezeUpdateFrames PROTO((P_, P_, P_));
191 int sanityChk_StkO (P_ stko); /* ToDo: move to a sane place */
193 #if ! defined(CONCURRENT)
195 extern void StackOverflow(STG_NO_ARGS) STG_NORETURN;
197 #if STACK_CHECK_BY_PAGE_FAULT
199 #define STACK_OVERFLOW(liveness,hda,hdb,spa,spb,rtype,reenter) \
200 /* use memory protection instead; still need ticky-ness */
204 #define STACK_OVERFLOW(liveness,hda,hdb,spa,spb,rtype,reenter) \
205 ULTRASAFESTGCALL0(void,(void *),StackOverflow)
207 #endif /* not using page-faulting */
211 I_ StackOverflow PROTO((W_, W_));
214 * On a uniprocessor, we do *NOT* context switch on a stack overflow
215 * (though we may GC). Therefore, we never have to reenter node.
218 #define STACK_OVERFLOW(liveness,hda,hdb,spa,spb,rtype,reenter) \
219 DO_STACKOVERFLOW((hda+hdb)<<2|((rtype)<<1)|(reenter),((spa)<<20)|((spb)<<8)|(liveness))
221 #define STACK_OVERFLOW_HEADROOM(args,y) ((args) >> 2)
222 #define STACK_OVERFLOW_PRIM_RETURN(args,y) ((args) & 2)
223 #define STACK_OVERFLOW_REENTER(args,y) ((args) & 1)
225 #define STACK_OVERFLOW_AWORDS(x,args) (((args) >> 20) & 0x0fff)
226 #define STACK_OVERFLOW_BWORDS(x,args) (((args) >> 8) & 0x0fff)
227 #define STACK_OVERFLOW_LIVENESS(x,args) ((args) & 0xff)
229 #endif /* CONCURRENT */
231 #define STK_CHK(liveness_mask,a_headroom,b_headroom,spa,spb,ret_type,reenter)\
233 DO_ASTK_HWM(); /* ticky-ticky profiling */ \
235 if (STKS_OVERFLOW_OP((a_headroom) + 1, (b_headroom) + 1)) { \
236 STACK_OVERFLOW(liveness_mask,a_headroom,b_headroom,spa,spb,ret_type,reenter);\
241 %************************************************************************
243 \subsubsection[StgMacros-heap-chks]{Heap-overflow checks}
245 %************************************************************************
247 Please see the general discussion/commentary about ``what really
248 happens in a GC,'' in \tr{SMinterface.lh}.
251 void PerformGC PROTO((W_));
252 void RealPerformGC PROTO((W_ liveness, W_ reqsize, W_ always_reenter_node, rtsBool do_full_collection));
253 void checkInCCallGC(STG_NO_ARGS);
256 void StgPerformGarbageCollection(STG_NO_ARGS);
261 #define OR_MSG_PENDING /* never */
263 #define HEAP_OVERFLOW(liveness,n,reenter) \
265 DO_GC((((W_)n)<<8)|(liveness)); \
268 #define REQSIZE_BITMASK ((1L << ((BITS_IN(W_) - 8 + 1))) - 1)
269 #define HEAP_OVERFLOW_REQSIZE(args) (((args) >> 8) & REQSIZE_BITMASK)
270 #define HEAP_OVERFLOW_REENTER(args) 0
271 #define HEAP_OVERFLOW_LIVENESS(args) ((args) & 0xff)
273 #else /* CONCURRENT */
275 void ReallyPerformThreadGC PROTO((W_, rtsBool));
277 #define HEAP_OVERFLOW(liveness,n,reenter) \
279 DO_GC((((W_)(n))<<9)|((reenter)<<8)|(liveness)); \
282 #define REQSIZE_BITMASK ((1L << ((BITS_IN(W_) - 9 + 1))) - 1)
283 #define HEAP_OVERFLOW_REQSIZE(args) (((args) >> 9) & REQSIZE_BITMASK)
284 #define HEAP_OVERFLOW_REENTER(args) (((args) >> 8) & 0x1)
285 #define HEAP_OVERFLOW_LIVENESS(args) ((args) & 0xff)
289 #define OR_MSG_PENDING /* never */
293 extern int PacketsWaiting; /*Probes for incoming messages*/
294 extern int heapChkCounter; /*Not currently used! We check for messages when*/
295 /*a thread is resheduled PWT*/
296 /* #define OR_MSG_PENDING || (--heapChkCounter == 0 && PacketsWaiting())*/
297 #define OR_MSG_PENDING /* never */
300 #endif /* CONCURRENT */
302 #if 0 /* alpha_TARGET_ARCH */
303 #define CACHE_LINE 4 /* words */
304 #define LINES_AHEAD 3
305 #define PRE_FETCH(n) \
308 j = ((STG_VOLATILE StgInt *) Hp)[LINES_AHEAD * CACHE_LINE]; \
310 #define EXTRA_HEAP_WORDS (CACHE_LINE * LINES_AHEAD)
312 #define PRE_FETCH(reg)
313 #define EXTRA_HEAP_WORDS 0
317 #define HEAP_CHK(liveness_mask,n,reenter) \
319 /* TICKY_PARANOIA(__FILE__, __LINE__); */ \
320 /* THREAD_CONTEXT_SWITCH(liveness_mask,reenter); */ \
321 ALLOC_HEAP(n); /* ticky profiling */ \
322 GRAN_ALLOC_HEAP(n,liveness_mask); /* Granularity Simulation */ \
323 if (((Hp = Hp + (n)) > HpLim)) { \
324 /* Old: STGCALL3_GC(PerformGC,liveness_mask,n,StgFalse); */\
325 HEAP_OVERFLOW(liveness_mask,n,StgFalse); \
330 #define HEAP_CHK(liveness_mask,n,reenter) \
332 /* TICKY_PARANOIA(__FILE__, __LINE__); */ \
334 ALLOC_HEAP(n); /* ticky profiling */ \
335 if (((Hp = Hp + (n)) > HpLim) OR_INTERVAL_EXPIRED OR_CONTEXT_SWITCH OR_MSG_PENDING) { \
336 HEAP_OVERFLOW(liveness_mask,n,reenter); \
344 #define HEAP_CHK_AND_RESTORE_N(liveness_mask,n,reenter) \
346 /* TICKY_PARANOIA(__FILE__, __LINE__); */ \
348 ALLOC_HEAP(n); /* ticky profiling */ \
349 if (((Hp = Hp + (n)) > HpLim) OR_INTERVAL_EXPIRED OR_CONTEXT_SWITCH OR_MSG_PENDING) { \
350 HEAP_OVERFLOW(liveness_mask,n,reenter); \
351 n = TSO_ARG1(CurrentTSO); \
356 #define HEAP_CHK_AND_RESTORE_N(liveness_mask,n,reenter) \
357 HEAP_CHK(liveness_mask,n,reenter)
364 %************************************************************************
366 \subsection[StgMacros-prim-ops]{Primitive operations}
368 %************************************************************************
370 One thing to be {\em very careful about} with these macros that assign
371 to results is that the assignment must come {\em last}. Some of the
372 other arguments may be in terms of addressing modes that get clobbered
373 by the assignment. (Dirty imperative programming RULES!)
375 The order here is roughly that in \tr{compiler/prelude/PrimOps.lhs}.
377 %************************************************************************
379 \subsubsection[StgMacros-compare-primops]{Primitive comparison ops on basic types}
381 %************************************************************************
383 We cast the chars in case one of them is a literal (so C things work right
384 even for 8-bit chars).
386 #define gtCharZh(r,a,b) r=(I_)((a)> (b))
387 #define geCharZh(r,a,b) r=(I_)((a)>=(b))
388 #define eqCharZh(r,a,b) r=(I_)((a)==(b))
389 #define neCharZh(r,a,b) r=(I_)((a)!=(b))
390 #define ltCharZh(r,a,b) r=(I_)((a)< (b))
391 #define leCharZh(r,a,b) r=(I_)((a)<=(b))
393 /* Int comparisons: >#, >=# etc */
394 #define ZgZh(r,a,b) r=(I_)((a) >(b))
395 #define ZgZeZh(r,a,b) r=(I_)((a)>=(b))
396 #define ZeZeZh(r,a,b) r=(I_)((a)==(b))
397 #define ZdZeZh(r,a,b) r=(I_)((a)!=(b))
398 #define ZlZh(r,a,b) r=(I_)((a) <(b))
399 #define ZlZeZh(r,a,b) r=(I_)((a)<=(b))
401 #define gtWordZh(r,a,b) r=(I_)((a) >(b))
402 #define geWordZh(r,a,b) r=(I_)((a)>=(b))
403 #define eqWordZh(r,a,b) r=(I_)((a)==(b))
404 #define neWordZh(r,a,b) r=(I_)((a)!=(b))
405 #define ltWordZh(r,a,b) r=(I_)((a) <(b))
406 #define leWordZh(r,a,b) r=(I_)((a)<=(b))
408 #define gtAddrZh(r,a,b) r=(I_)((a) >(b))
409 #define geAddrZh(r,a,b) r=(I_)((a)>=(b))
410 #define eqAddrZh(r,a,b) r=(I_)((a)==(b))
411 #define neAddrZh(r,a,b) r=(I_)((a)!=(b))
412 #define ltAddrZh(r,a,b) r=(I_)((a) <(b))
413 #define leAddrZh(r,a,b) r=(I_)((a)<=(b))
415 #define gtFloatZh(r,a,b) r=(I_)((a)> (b))
416 #define geFloatZh(r,a,b) r=(I_)((a)>=(b))
417 #define eqFloatZh(r,a,b) r=(I_)((a)==(b))
418 #define neFloatZh(r,a,b) r=(I_)((a)!=(b))
419 #define ltFloatZh(r,a,b) r=(I_)((a)< (b))
420 #define leFloatZh(r,a,b) r=(I_)((a)<=(b))
422 /* Double comparisons: >##, >=#@ etc */
423 #define ZgZhZh(r,a,b) r=(I_)((a) >(b))
424 #define ZgZeZhZh(r,a,b) r=(I_)((a)>=(b))
425 #define ZeZeZhZh(r,a,b) r=(I_)((a)==(b))
426 #define ZdZeZhZh(r,a,b) r=(I_)((a)!=(b))
427 #define ZlZhZh(r,a,b) r=(I_)((a) <(b))
428 #define ZlZeZhZh(r,a,b) r=(I_)((a)<=(b))
431 %************************************************************************
433 \subsubsection[StgMacros-char-primops]{Primitive @Char#@ ops (and @LitString#@ish things, too)}
435 %************************************************************************
437 We cast the chars in case one of them is a literal (so C things work right
438 even for 8-bit chars).
440 #define ordZh(r,a) r=(I_)((W_) (a))
441 #define chrZh(r,a) r=(StgChar)((W_)(a))
444 %************************************************************************
446 \subsubsection[StgMacros-int-primops]{Primitive @Int#@ ops}
448 %************************************************************************
451 I_ stg_div PROTO((I_ a, I_ b));
453 #define ZpZh(r,a,b) r=(a)+(b)
454 #define ZmZh(r,a,b) r=(a)-(b)
455 #define ZtZh(r,a,b) r=(a)*(b)
456 #define quotIntZh(r,a,b) r=(a)/(b)
457 #define ZdZh(r,a,b) r=ULTRASAFESTGCALL2(I_,(void *, I_, I_),stg_div,(a),(b))
458 #define remIntZh(r,a,b) r=(a)%(b)
459 #define negateIntZh(r,a) r=-(a)
460 /* Ever used ? -- SOF */
461 #define absIntZh(a) r=(( (a) >= 0 ) ? (a) : (-(a)))
464 %************************************************************************
466 \subsubsection[StgMacros-word-primops]{Primitive @Word#@ ops}
468 %************************************************************************
471 #define andZh(r,a,b) r=(a)&(b)
472 #define orZh(r,a,b) r=(a)|(b)
473 #define notZh(r,a) r=~(a)
475 #define shiftLZh(r,a,b) r=(a)<<(b)
476 #define shiftRAZh(r,a,b) r=(a)>>(b)
477 #define shiftRLZh(r,a,b) r=(a)>>(b)
478 #define iShiftLZh(r,a,b) r=(a)<<(b)
479 #define iShiftRAZh(r,a,b) r=(a)>>(b)
480 #define iShiftRLZh(r,a,b) r=(a)>>(b)
482 #define int2WordZh(r,a) r=(W_)(a)
483 #define word2IntZh(r,a) r=(I_)(a)
486 %************************************************************************
488 \subsubsection[StgMacros-addr-primops]{Primitive @Addr#@ ops}
490 %************************************************************************
493 #define int2AddrZh(r,a) r=(A_)(a)
494 #define addr2IntZh(r,a) r=(I_)(a)
497 %************************************************************************
499 \subsubsection[StgMacros-float-primops]{Primitive @Float#@ ops}
501 %************************************************************************
504 #define plusFloatZh(r,a,b) r=(a)+(b)
505 #define minusFloatZh(r,a,b) r=(a)-(b)
506 #define timesFloatZh(r,a,b) r=(a)*(b)
507 #define divideFloatZh(r,a,b) r=(a)/(b)
508 #define negateFloatZh(r,a) r=-(a)
510 #define int2FloatZh(r,a) r=(StgFloat)(a)
511 #define float2IntZh(r,a) r=(I_)(a)
513 #define expFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),exp,a)
514 #define logFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),log,a)
515 #define sqrtFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),sqrt,a)
516 #define sinFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),sin,a)
517 #define cosFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),cos,a)
518 #define tanFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),tan,a)
519 #define asinFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),asin,a)
520 #define acosFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),acos,a)
521 #define atanFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),atan,a)
522 #define sinhFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),sinh,a)
523 #define coshFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),cosh,a)
524 #define tanhFloatZh(r,a) r=(StgFloat) SAFESTGCALL1(StgDouble,(void *, StgDouble),tanh,a)
525 #define powerFloatZh(r,a,b) r=(StgFloat) SAFESTGCALL2(StgDouble,(void *, StgDouble,StgDouble),pow,a,b)
527 /* encoding/decoding given w/ Integer stuff */
530 %************************************************************************
532 \subsubsection[StgMacros-double-primops]{Primitive @Double#@ ops}
534 %************************************************************************
537 #define ZpZhZh(r,a,b) r=(a)+(b)
538 #define ZmZhZh(r,a,b) r=(a)-(b)
539 #define ZtZhZh(r,a,b) r=(a)*(b)
540 #define ZdZhZh(r,a,b) r=(a)/(b)
541 #define negateDoubleZh(r,a) r=-(a)
543 #define int2DoubleZh(r,a) r=(StgDouble)(a)
544 #define double2IntZh(r,a) r=(I_)(a)
546 #define float2DoubleZh(r,a) r=(StgDouble)(a)
547 #define double2FloatZh(r,a) r=(StgFloat)(a)
549 #define expDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),exp,a)
550 #define logDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),log,a)
551 #define sqrtDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),sqrt,a)
552 #define sinDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),sin,a)
553 #define cosDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),cos,a)
554 #define tanDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),tan,a)
555 #define asinDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),asin,a)
556 #define acosDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),acos,a)
557 #define atanDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),atan,a)
558 #define sinhDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),sinh,a)
559 #define coshDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),cosh,a)
560 #define tanhDoubleZh(r,a) r=(StgDouble) SAFESTGCALL1(StgDouble,(void *, StgDouble),tanh,a)
562 #define ZtZtZhZh(r,a,b) r=(StgDouble) SAFESTGCALL2(StgDouble,(void *, StgDouble,StgDouble),pow,a,b)
565 %************************************************************************
567 \subsubsection[StgMacros-integer-primops]{Primitive @Integer@-related ops (GMP stuff)}
569 %************************************************************************
571 Dirty macros we use for the real business.
573 INVARIANT: When one of these macros is called, the only live data is
574 tidily on the STG stacks or in the STG registers (the code generator
575 ensures this). If there are any pointer-arguments, they will be in
576 the first \tr{Ret*} registers (e.g., \tr{da} arg of \tr{gmpTake1Return1}).
578 OK, here are the real macros:
580 #define gmpTake1Return1(size_chk_macro, liveness, mpz_op, ar,sr,dr, aa,sa,da) \
583 I_ space = size_chk_macro(sa); \
585 /* Check that there will be enough heap & make Hp visible to GMP allocator */ \
586 GMP_HEAP_LOOKAHEAD(liveness,space); \
588 /* Now we can initialise (post possible GC) */ \
591 arg.d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
593 SAFESTGCALL1(void,(void *, MP_INT *),mpz_init,&result); \
595 /* Perform the operation */ \
596 SAFESTGCALL2(void,(void *, MP_INT *, MP_INT *),mpz_op,&result,&arg); \
598 GMP_HEAP_HANDBACK(); /* restore Hp */ \
599 (ar) = result.alloc; \
600 (sr) = result.size; \
601 (dr) = (B_) (result.d - DATA_HS); \
602 /* pt to *beginning* of object (GMP has been monkeying around in the middle) */ \
606 #define gmpTake2Return1(size_chk_macro, liveness, mpz_op, ar,sr,dr, a1,s1,d1, a2,s2,d2)\
610 I_ space = size_chk_macro(s1,s2); \
612 /* Check that there will be enough heap & make Hp visible to GMP allocator */ \
613 GMP_HEAP_LOOKAHEAD(liveness,space); \
615 /* Now we can initialise (post possible GC) */ \
618 arg1.d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
621 arg2.d = (unsigned long int *) (BYTE_ARR_CTS(d2)); \
623 SAFESTGCALL1(void,(void *, MP_INT *),mpz_init,&result); \
625 /* Perform the operation */ \
626 SAFESTGCALL3(void,(void *, MP_INT *, MP_INT *, MP_INT *),mpz_op,&result,&arg1,&arg2); \
628 GMP_HEAP_HANDBACK(); /* restore Hp */ \
629 (ar) = result.alloc; \
630 (sr) = result.size; \
631 (dr) = (B_) (result.d - DATA_HS); \
632 /* pt to *beginning* of object (GMP has been monkeying around in the middle) */ \
635 #define gmpTake2Return2(size_chk_macro, liveness, mpz_op, ar1,sr1,dr1, ar2,sr2,dr2, a1,s1,d1, a2,s2,d2) \
640 I_ space = size_chk_macro(s1,s2); \
642 /* Check that there will be enough heap & make Hp visible to GMP allocator */ \
643 GMP_HEAP_LOOKAHEAD(liveness,space); \
645 /* Now we can initialise (post possible GC) */ \
648 arg1.d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
651 arg2.d = (unsigned long int *) (BYTE_ARR_CTS(d2)); \
653 SAFESTGCALL1(void,(void *, MP_INT *),mpz_init,&result1); \
654 SAFESTGCALL1(void,(void *, MP_INT *),mpz_init,&result2); \
656 /* Perform the operation */ \
657 SAFESTGCALL4(void,(void *, MP_INT *, MP_INT *, MP_INT *, MP_INT *),mpz_op,&result1,&result2,&arg1,&arg2); \
659 GMP_HEAP_HANDBACK(); /* restore Hp */ \
660 (ar1) = result1.alloc; \
661 (sr1) = result1.size; \
662 (dr1) = (B_) (result1.d - DATA_HS); \
663 (ar2) = result2.alloc; \
664 (sr2) = result2.size; \
665 (dr2) = (B_) (result2.d - DATA_HS); \
669 Some handy size-munging macros: sometimes gratuitously {\em conservative}.
670 The \tr{+16} is to allow for the initial allocation of \tr{MP_INT} results.
671 The \tr{__abs} stuff is because negative-ness of GMP things is encoded
674 #define __abs(a) (( (a) >= 0 ) ? (a) : (-(a)))
675 #define GMP_SIZE_ONE() (2 + DATA_HS + 16)
676 #define GMP_SAME_SIZE(a) (__abs(a) + DATA_HS + 16)
677 #define GMP_MAX_SIZE(a,b) ((__abs(a) > __abs(b) ? __abs(a) : __abs(b)) + 1 + DATA_HS + 16)
678 /* NB: the +1 is for the carry (or whatever) */
679 #define GMP_2MAX_SIZE(a,b) (2 * GMP_MAX_SIZE(a,b))
680 #define GMP_ADD_SIZES(a,b) (__abs(a) + __abs(b) + 1 + DATA_HS + 16)
681 /* the +1 may just be paranoia */
684 For the Integer/GMP stuff, we have macros that {\em look ahead} for
685 some space, but don't actually grab it.
687 If there are live pointers at the time of the lookahead, the caller
688 must make sure they are in \tr{Ret1}, \tr{Ret2}, ..., so they can be
689 handled normally. We achieve this by having the code generator {\em
690 always} pass args to may-invoke-GC primitives in registers, using the
691 normal pointers-first policy. This means that, if we do go to garbage
692 collection, everything is already in the Right Place.
694 Saving and restoring Hp register so the MP allocator can see them. If we are
695 performing liftime profiling need to save and restore HpLim as well so that
696 it can be bumped if allocation occurs.
698 The second argument to @GMP_HEAP_LOOKAHEAD@ must be an lvalue so that
699 it can be restored from @TSO_ARG1@ after a failed @HEAP_CHK@ in
703 #define GMP_HEAP_LOOKAHEAD(liveness,n) \
705 HEAP_CHK_AND_RESTORE_N(liveness,n,0); \
707 UN_ALLOC_HEAP(n); /* Undo ticky-ticky */ \
708 SAVE_Hp = Hp; /* Hand over the hp */ \
709 DEBUG_SetGMPAllocBudget(n) \
712 #define GMP_HEAP_HANDBACK() \
714 DEBUG_ResetGMPAllocBudget()
718 void *stgAllocForGMP PROTO((size_t size_in_bytes));
719 void *stgReallocForGMP PROTO((void *ptr, size_t old_size, size_t new_size));
720 void stgDeallocForGMP PROTO((void *ptr, size_t size));
723 extern StgInt DEBUG_GMPAllocBudget;
724 #define DEBUG_SetGMPAllocBudget(n) DEBUG_GMPAllocBudget = (n);
725 #define DEBUG_ResetGMPAllocBudget() DEBUG_GMPAllocBudget = 0;
727 #define DEBUG_SetGMPAllocBudget(n) /*nothing*/
728 #define DEBUG_ResetGMPAllocBudget() /*nothing*/
732 The real business (defining Integer primops):
734 #define negateIntegerZh(ar,sr,dr, liveness, aa,sa,da) \
735 gmpTake1Return1(GMP_SAME_SIZE, liveness, mpz_neg, ar,sr,dr, aa,sa,da)
737 #define plusIntegerZh(ar,sr,dr, liveness, a1,s1,d1, a2,s2,d2) \
738 gmpTake2Return1(GMP_MAX_SIZE, liveness, mpz_add, ar,sr,dr, a1,s1,d1, a2,s2,d2)
739 #define minusIntegerZh(ar,sr,dr, liveness, a1,s1,d1, a2,s2,d2) \
740 gmpTake2Return1(GMP_MAX_SIZE, liveness, mpz_sub, ar,sr,dr, a1,s1,d1, a2,s2,d2)
741 #define timesIntegerZh(ar,sr,dr, liveness, a1,s1,d1, a2,s2,d2) \
742 gmpTake2Return1(GMP_ADD_SIZES, liveness, mpz_mul, ar,sr,dr, a1,s1,d1, a2,s2,d2)
744 /* div, mod, quot, rem are defined w/ quotRem & divMod */
746 #define quotRemIntegerZh(ar1,sr1,dr1, ar2,sr2,dr2, liveness, a1,s1,d1, a2,s2,d2) \
747 gmpTake2Return2(GMP_2MAX_SIZE, liveness, mpz_divmod, ar1,sr1,dr1, ar2,sr2,dr2, a1,s1,d1, a2,s2,d2)
748 #define divModIntegerZh(ar1,sr1,dr1, ar2,sr2,dr2, liveness, a1,s1,d1, a2,s2,d2) \
749 gmpTake2Return2(GMP_2MAX_SIZE, liveness, mpz_mdivmod, ar1,sr1,dr1, ar2,sr2,dr2, a1,s1,d1, a2,s2,d2)
752 Comparison ops (@<@, @>=@, etc.) are defined in terms of the cmp
753 fellow (returns -ve, 0, or +ve).
755 #define cmpIntegerZh(r, hp, a1,s1,d1, a2,s2,d2) /* calls mpz_cmp */ \
758 /* Does not allocate memory */ \
762 arg1.d = (unsigned long int *) (BYTE_ARR_CTS(d1)); \
765 arg2.d = (unsigned long int *) (BYTE_ARR_CTS(d2)); \
767 (r) = SAFESTGCALL2(I_,(void *, MP_INT *, MP_INT *),mpz_cmp,&arg1,&arg2); \
774 #define integer2IntZh(r, hp, aa,sa,da) \
776 /* Does not allocate memory */ \
780 arg.d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
782 (r) = SAFESTGCALL1(I_,(void *, MP_INT *),mpz_get_si,&arg); \
785 /* Since we're forced to know a little bit about MP_INT layout to do this with
786 pre-allocated heap, we just inline the whole of mpz_init_set_si here.
787 ** DIRE WARNING. if mpz_init_set_si changes, so does this! ***
790 #define int2IntegerZh(ar,sr,dr, hp, i) \
791 { StgInt val; /* to snaffle arg to avoid aliasing */ \
793 val = (i); /* snaffle... */ \
795 SET_DATA_HDR((hp),ArrayOfData_info,CCC,DATA_VHS+MIN_MP_INT_SIZE,0); \
797 if ((val) < 0) { (sr) = -1; (hp)[DATA_HS] = -(val); } \
798 else if ((val) > 0) { (sr) = 1; (hp)[DATA_HS] = (val); } \
799 else /* val==0 */ { (sr) = 0; } \
801 (dr) = (B_)(hp); /* dr is an StgByteArray */ \
804 #define word2IntegerZh(ar,sr,dr, hp, i) \
805 { StgWord val; /* to snaffle arg to avoid aliasing */ \
807 val = (i); /* snaffle... */ \
809 SET_DATA_HDR((hp),ArrayOfData_info,CCC,DATA_VHS+MIN_MP_INT_SIZE,0); \
811 if ((val) != 0) { (sr) = 1; (hp)[DATA_HS] = (val); } \
812 else /* val==0 */ { (sr) = 0; } \
814 (dr) = (B_)(hp); /* dr is an StgByteArray */ \
819 Then there are a few oddments to make life easier:
823 The "str" argument must be a literal C string.
825 addr2Integer( ..., "foo") OK!
828 addr2Integer( ..., x) NO! NO!
831 #define addr2IntegerZh(ar,sr,dr, liveness, str) \
833 /* taking the number of bytes/8 as the number of words of lookahead \
834 is plenty conservative */ \
835 I_ space = GMP_SAME_SIZE(sizeof(str) / 8 + 1); \
837 GMP_HEAP_LOOKAHEAD(liveness, space); \
839 /* Perform the operation */ \
840 if (SAFESTGCALL3(I_,(void *, MP_INT *, char *, int), mpz_init_set_str,&result,(str),/*base*/10)) \
843 GMP_HEAP_HANDBACK(); /* restore Hp */ \
844 (ar) = result.alloc; \
845 (sr) = result.size; \
846 (dr) = (B_) (result.d - DATA_HS); \
847 /* pt to *beginning* of object (GMP has been monkeying around in the middle) */ \
851 Encoding and decoding float-ish things is pretty Integer-ish. We use
852 these pretty magical support functions, essentially stolen from Lennart:
854 StgFloat __encodeFloat PROTO((MP_INT *, I_));
855 void __decodeFloat PROTO((MP_INT * /*result1*/,
859 StgDouble __encodeDouble PROTO((MP_INT *, I_));
860 void __decodeDouble PROTO((MP_INT * /*result1*/,
865 Some floating-point format info, made with the \tr{enquire} program
866 (version~4.3) [comes with gcc].
868 /* this should be done by CPU architecture, insofar as possible [WDP] */
870 #if sparc_TARGET_ARCH \
871 || alpha_TARGET_ARCH \
872 || hppa1_1_TARGET_ARCH \
873 || i386_TARGET_ARCH \
874 || m68k_TARGET_ARCH \
875 || mipsel_TARGET_ARCH \
876 || mipseb_TARGET_ARCH \
877 || powerpc_TARGET_ARCH \
878 || rs6000_TARGET_ARCH
880 /* yes, it is IEEE floating point */
881 #include "ieee-flpt.h"
883 #if alpha_dec_osf1_TARGET \
884 || i386_TARGET_ARCH \
885 || mipsel_TARGET_ARCH
887 #undef BIGENDIAN /* little-endian weirdos... */
892 #else /* unknown floating-point format */
894 ******* ERROR *********** Any ideas about floating-point format?
896 #endif /* unknown floating-point */
900 #if alpha_dec_osf1_TARGET
901 #define encodeFloatZh(r, hp, aa,sa,da, expon) encodeDoubleZh(r, hp, aa,sa,da, expon)
903 #define encodeFloatZh(r, hp, aa,sa,da, expon) \
905 /* Does not allocate memory */ \
909 arg.d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
911 r = SAFESTGCALL2(StgFloat,(void *, MP_INT *, I_), __encodeFloat,&arg,(expon)); \
915 #define encodeDoubleZh(r, hp, aa,sa,da, expon) \
917 /* Does not allocate memory */ \
921 arg.d = (unsigned long int *) (BYTE_ARR_CTS(da)); \
923 r = SAFESTGCALL2(StgDouble,(void *, MP_INT *, I_), __encodeDouble,&arg,(expon));\
926 #if alpha_dec_osf1_TARGET
927 #define decodeFloatZh(exponr, ar,sr,dr, hp, f) decodeDoubleZh(exponr, ar,sr,dr, hp, f)
929 #define decodeFloatZh(exponr, ar,sr,dr, hp, f) \
932 StgFloat arg = (f); \
934 /* Be prepared to tell Lennart-coded __decodeFloat */ \
935 /* where mantissa.d can be put (it does not care about the rest) */ \
936 SET_DATA_HDR(hp,ArrayOfData_info,CCC,DATA_VHS+MIN_MP_INT_SIZE,0); \
937 mantissa.d = (hp) + DATA_HS; \
939 /* Perform the operation */ \
940 SAFESTGCALL3(void,(void *, MP_INT *, I_ *, StgFloat),__decodeFloat,&mantissa,&exponent,arg); \
942 ar = mantissa.alloc; \
943 sr = mantissa.size; \
948 #define decodeDoubleZh(exponr, ar,sr,dr, hp, f) \
951 StgDouble arg = (f); \
953 /* Be prepared to tell Lennart-coded __decodeDouble */ \
954 /* where mantissa.d can be put (it does not care about the rest) */ \
955 SET_DATA_HDR(hp,ArrayOfData_info,CCC,DATA_VHS+MIN_MP_INT_SIZE,0); \
956 mantissa.d = (hp) + DATA_HS; \
958 /* Perform the operation */ \
959 SAFESTGCALL3(void,(void *, MP_INT *, I_ *, StgDouble),__decodeDouble,&mantissa,&exponent,arg); \
961 ar = mantissa.alloc; \
962 sr = mantissa.size; \
967 %************************************************************************
969 \subsubsection[StgMacros-mv-floats]{Moving floats and doubles around (e.g., to/from stacks)}
971 %************************************************************************
973 With GCC, we use magic non-standard inlining; for other compilers, we
974 just use functions (see also \tr{runtime/prims/PrimArith.lc}).
976 (The @OMIT_...@ is only used in compiling some of the RTS, none of
977 which uses these anyway.)
980 #if alpha_TARGET_ARCH \
981 || i386_TARGET_ARCH \
984 #define ASSIGN_FLT(dst, src) *(StgFloat *)(dst) = (src);
985 #define PK_FLT(src) (*(StgFloat *)(src))
987 #define ASSIGN_DBL(dst, src) *(StgDouble *)(dst) = (src);
988 #define PK_DBL(src) (*(StgDouble *)(src))
990 #else /* not m68k || alpha || i[34]86 */
992 /* Special handling for machines with troublesome alignment constraints */
994 #define FLOAT_ALIGNMENT_TROUBLES TRUE
996 #if ! defined(__GNUC__) || ! defined(__STG_GCC_REGS__)
998 void ASSIGN_DBL PROTO((W_ [], StgDouble));
999 StgDouble PK_DBL PROTO((W_ []));
1000 void ASSIGN_FLT PROTO((W_ [], StgFloat));
1001 StgFloat PK_FLT PROTO((W_ []));
1003 #else /* yes, its __GNUC__ && we really want them */
1005 #if sparc_TARGET_ARCH
1007 #define ASSIGN_FLT(dst, src) *(StgFloat *)(dst) = (src);
1008 #define PK_FLT(src) (*(StgFloat *)(src))
1010 #define ASSIGN_DBL(dst,src) \
1011 __asm__("st %2,%0\n\tst %R2,%1" : "=m" (((P_)(dst))[0]), \
1012 "=m" (((P_)(dst))[1]) : "f" (src));
1014 #define PK_DBL(src) \
1015 ( { register double d; \
1016 __asm__("ld %1,%0\n\tld %2,%R0" : "=f" (d) : \
1017 "m" (((P_)(src))[0]), "m" (((P_)(src))[1])); d; \
1022 /* (not very) forward prototype declarations */
1023 void ASSIGN_DBL PROTO((W_ [], StgDouble));
1024 StgDouble PK_DBL PROTO((W_ []));
1025 void ASSIGN_FLT PROTO((W_ [], StgFloat));
1026 StgFloat PK_FLT PROTO((W_ []));
1030 ASSIGN_DBL(W_ p_dest[], StgDouble src)
1034 p_dest[0] = y.du.dhi;
1035 p_dest[1] = y.du.dlo;
1038 /* GCC also works with this version, but it generates
1039 the same code as the previous one, and is not ANSI
1041 #define ASSIGN_DBL( p_dest, src ) \
1042 *p_dest = ((double_thing) src).du.dhi; \
1043 *(p_dest+1) = ((double_thing) src).du.dlo \
1051 y.du.dhi = p_src[0];
1052 y.du.dlo = p_src[1];
1058 ASSIGN_FLT(W_ p_dest[], StgFloat src)
1074 #endif /* ! sparc */
1076 #endif /* __GNUC__ */
1078 #endif /* not __m68k__ */
1081 %************************************************************************
1083 \subsubsection[StgMacros-array-primops]{Primitive arrays}
1085 %************************************************************************
1087 We regularly use this macro to fish the ``contents'' part
1088 out of a DATA or TUPLE closure, which is what is used for
1089 non-ptr and ptr arrays (respectively).
1091 BYTE_ARR_CTS returns a @C_ *@!
1093 We {\em ASSUME} we can use the same macro for both!!
1097 #define BYTE_ARR_CTS(a) \
1098 ({ ASSERT(INFO_PTR(a) == (W_) ArrayOfData_info); \
1099 ((C_ *) (((StgPtr) (a))+DATA_HS)); })
1100 #define PTRS_ARR_CTS(a) \
1101 ({ ASSERT((INFO_PTR(a) == (W_) ArrayOfPtrs_info) \
1102 || (INFO_PTR(a) == (W_) ImMutArrayOfPtrs_info));\
1103 ((a)+MUTUPLE_HS);} )
1105 #define BYTE_ARR_CTS(a) ((char *) (((StgPtr) (a))+DATA_HS))
1106 #define PTRS_ARR_CTS(a) ((a)+MUTUPLE_HS)
1110 extern I_ genSymZh(STG_NO_ARGS);
1111 extern I_ resetGenSymZh(STG_NO_ARGS);
1112 extern I_ incSeqWorldZh(STG_NO_ARGS);
1114 extern I_ byteArrayHasNUL__ PROTO((const char *, I_));
1117 OK, the easy ops first: (all except \tr{newArr*}:
1119 (OLD:) VERY IMPORTANT! The read/write/index primitive ops
1120 on @ByteArray#@s index the array using a {\em BYTE} offset, even
1121 if the thing begin gotten out is a multi-byte @Int#@, @Float#@ etc.
1122 This is because you might be trying to take apart a C struct, where
1123 the offset from the start of the struct isn't a multiple of the
1124 size of the thing you're getting. Hence the @(char *)@ casts.
1126 EVEN MORE IMPORTANT! The above is a lie. The offsets for BlahArrays
1127 are in Blahs. WDP 95/08
1129 In the case of messing with @StgAddrs@ (@A_@), which are really \tr{void *},
1130 we cast to @P_@, because you can't index off an uncast \tr{void *}.
1132 In the case of @Array#@ (which contain pointers), the offset is in units
1133 of one ptr (not bytes).
1136 #define sameMutableArrayZh(r,a,b) r=(I_)((a)==(b))
1137 #define sameMutableByteArrayZh(r,a,b) r=(I_)((B_)(a)==(B_)(b))
1139 #define readArrayZh(r,a,i) r=((PP_) PTRS_ARR_CTS(a))[(i)]
1141 #define readCharArrayZh(r,a,i) indexCharOffAddrZh(r,BYTE_ARR_CTS(a),i)
1142 #define readIntArrayZh(r,a,i) indexIntOffAddrZh(r,BYTE_ARR_CTS(a),i)
1143 #define readAddrArrayZh(r,a,i) indexAddrOffAddrZh(r,BYTE_ARR_CTS(a),i)
1144 #define readFloatArrayZh(r,a,i) indexFloatOffAddrZh(r,BYTE_ARR_CTS(a),i)
1145 #define readDoubleArrayZh(r,a,i) indexDoubleOffAddrZh(r,BYTE_ARR_CTS(a),i)
1147 /* result ("r") arg ignored in write macros! */
1148 #define writeArrayZh(a,i,v) ((PP_) PTRS_ARR_CTS(a))[(i)]=(v)
1150 #define writeCharArrayZh(a,i,v) ((C_ *)(BYTE_ARR_CTS(a)))[i] = (v)
1151 #define writeIntArrayZh(a,i,v) ((I_ *)(BYTE_ARR_CTS(a)))[i] = (v)
1152 #define writeAddrArrayZh(a,i,v) ((PP_)(BYTE_ARR_CTS(a)))[i] = (v)
1153 #define writeFloatArrayZh(a,i,v) \
1154 ASSIGN_FLT((P_) (((StgFloat *)(BYTE_ARR_CTS(a))) + i),v)
1155 #define writeDoubleArrayZh(a,i,v) \
1156 ASSIGN_DBL((P_) (((StgDouble *)(BYTE_ARR_CTS(a))) + i),v)
1158 #define indexArrayZh(r,a,i) r=((PP_) PTRS_ARR_CTS(a))[(i)]
1160 #define indexCharArrayZh(r,a,i) indexCharOffAddrZh(r,BYTE_ARR_CTS(a),i)
1161 #define indexIntArrayZh(r,a,i) indexIntOffAddrZh(r,BYTE_ARR_CTS(a),i)
1162 #define indexAddrArrayZh(r,a,i) indexAddrOffAddrZh(r,BYTE_ARR_CTS(a),i)
1163 #define indexFloatArrayZh(r,a,i) indexFloatOffAddrZh(r,BYTE_ARR_CTS(a),i)
1164 #define indexDoubleArrayZh(r,a,i) indexDoubleOffAddrZh(r,BYTE_ARR_CTS(a),i)
1166 #define indexCharOffForeignObjZh(r,fo,i) indexCharOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
1167 #define indexIntOffForeignObjZh(r,fo,i) indexIntOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
1168 #define indexAddrOffForeignObjZh(r,fo,i) indexAddrOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
1169 #define indexFloatOffForeignObjZh(r,fo,i) indexFloatOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
1170 #define indexDoubleOffForeignObjZh(r,fo,i) indexDoubleOffAddrZh(r,ForeignObj_CLOSURE_DATA(fo),i)
1172 #define indexCharOffAddrZh(r,a,i) r= ((C_ *)(a))[i]
1173 #define indexIntOffAddrZh(r,a,i) r= ((I_ *)(a))[i]
1174 #define indexAddrOffAddrZh(r,a,i) r= ((PP_)(a))[i]
1175 #define indexFloatOffAddrZh(r,a,i) r= PK_FLT((P_) (((StgFloat *)(a)) + i))
1176 #define indexDoubleOffAddrZh(r,a,i) r= PK_DBL((P_) (((StgDouble *)(a)) + i))
1178 /* Freezing arrays-of-ptrs requires changing an info table, for the
1179 benefit of the generational collector. It needs to scavenge mutable
1180 objects, even if they are in old space. When they become immutable,
1181 they can be removed from this scavenge list. */
1182 #define unsafeFreezeArrayZh(r,a) \
1186 FREEZE_MUT_HDR(result,ImMutArrayOfPtrs_info); \
1190 #define unsafeFreezeByteArrayZh(r,a) r=(B_)(a)
1193 Now the \tr{newArr*} ops:
1197 --------------------
1198 Will: ToDo: we need to find suitable places to put this comment, and the
1199 "in-general" one which follows.
1201 ************ Nota Bene. The "n" in this macro is guaranteed to
1202 be a register, *not* (say) Node[1]. That means that it is guaranteed
1203 to survive GC, provided only that the register is kept unaltered.
1204 This is important, because "n" is used after the HEAP_CHK.
1206 In general, *all* parameters to these primitive-op macros are always
1207 registers. (Will: For exactly *which* primitive-op macros is this guaranteed?
1208 Exactly those which can trigger GC?)
1209 ------------------------
1211 NOTE: the above may now be OLD (WDP 94/02/10)
1215 For char arrays, the size is in {\em BYTES}.
1218 #define newCharArrayZh(r,liveness,n) newByteArray(r,liveness,(n) * sizeof(C_))
1219 #define newIntArrayZh(r,liveness,n) newByteArray(r,liveness,(n) * sizeof(I_))
1220 #define newAddrArrayZh(r,liveness,n) newByteArray(r,liveness,(n) * sizeof(P_))
1221 #define newFloatArrayZh(r,liveness,n) newByteArray(r,liveness,(n) * sizeof(StgFloat))
1222 #define newDoubleArrayZh(r,liveness,n) newByteArray(r,liveness,(n) * sizeof(StgDouble))
1224 #define newByteArray(r,liveness,n) \
1229 HEAP_CHK(liveness,DATA_HS+BYTES_TO_STGWORDS(n),0); \
1230 size = BYTES_TO_STGWORDS(n); \
1231 ALLOC_PRIM(DATA_HS,size,0,DATA_HS+size) /* ticky ticky */; \
1232 CC_ALLOC(CCC,DATA_HS+size,ARR_K); \
1234 result = Hp-(DATA_HS+size)+1; \
1235 SET_DATA_HDR(result,ArrayOfData_info,CCC,DATA_VHS+size,0); \
1240 Arrays of pointers need to be initialised; uses \tr{TUPLES}!
1241 The initialisation value is guaranteed to be in a register,
1242 and will be indicated by the liveness mask, so it's ok to do
1243 a \tr{HEAP_CHK}, which may trigger GC.
1246 /* The new array initialization routine for the NCG */
1247 void newArrZh_init PROTO((P_ result, I_ n, P_ init));
1249 #define newArrayZh(r,liveness,n,init) \
1254 HEAP_CHK(liveness, MUTUPLE_HS+(n),0); \
1255 ALLOC_PRIM(MUTUPLE_HS,(n),0,MUTUPLE_HS+(n)) /* ticky ticky */; \
1256 CC_ALLOC(CCC,MUTUPLE_HS+(n),ARR_K); /* cc prof */ \
1258 result = Hp + 1 - (MUTUPLE_HS+(n)); \
1259 SET_MUTUPLE_HDR(result,ArrayOfPtrs_info,CCC,MUTUPLE_VHS+(n),0) \
1260 for (p = result+MUTUPLE_HS; p < (result+MUTUPLE_HS+(n)); p++) { \
1268 %************************************************************************
1270 \subsubsection[StgMacros-SynchVar-primops]{Synchronizing Variables PrimOps}
1272 %************************************************************************
1275 ED_(PrelBase_Z91Z93_closure);
1277 #define newSynchVarZh(r, hp) \
1279 ALLOC_PRIM(MUTUPLE_HS,3,0,MUTUPLE_HS+3) /* ticky ticky */; \
1280 CC_ALLOC(CCC,MUTUPLE_HS+3,ARR_K); /* cc prof */ \
1281 SET_SVAR_HDR(hp,EmptySVar_info,CCC); \
1282 SVAR_HEAD(hp) = SVAR_TAIL(hp) = SVAR_VALUE(hp) = PrelBase_Z91Z93_closure; \
1290 void Yield PROTO((W_));
1292 #define takeMVarZh(r, liveness, node) \
1294 while (INFO_PTR(node) != (W_) FullSVar_info) { \
1295 if (SVAR_HEAD(node) == PrelBase_Z91Z93_closure) \
1296 SVAR_HEAD(node) = CurrentTSO; \
1298 TSO_LINK(SVAR_TAIL(node)) = CurrentTSO; \
1299 TSO_LINK(CurrentTSO) = (P_) PrelBase_Z91Z93_closure; \
1300 SVAR_TAIL(node) = CurrentTSO; \
1301 DO_YIELD(liveness << 1); \
1303 SET_INFO_PTR(node, EmptySVar_info); \
1304 r = SVAR_VALUE(node); \
1305 SVAR_VALUE(node) = PrelBase_Z91Z93_closure; \
1310 #define takeMVarZh(r, liveness, node) \
1312 if (INFO_PTR(node) != (W_) FullSVar_info) { \
1313 /* Don't wrap the calls; we're done with STG land */\
1315 fprintf(stderr, "takeMVar#: MVar is empty.\n"); \
1316 EXIT(EXIT_FAILURE); \
1318 SET_INFO_PTR(node, EmptySVar_info); \
1319 r = SVAR_VALUE(node); \
1320 SVAR_VALUE(node) = PrelBase_Z91Z93_closure; \
1331 /* Only difference to the !GRAN def: RunnableThreadsHd has been replaced by */
1332 /* ThreadQueueHd i.e. the tso is added at the end of the thread queue on */
1333 /* the CurrentProc. This means we have an implicit context switch after */
1334 /* putMVar even if unfair scheduling is used in GranSim (default)! -- HWL */
1336 #define putMVarZh(node, value) \
1339 if (INFO_PTR(node) == (W_) FullSVar_info) { \
1340 /* Don't wrap the calls; we're done with STG land */\
1342 fprintf(stderr, "putMVar#: MVar already full.\n"); \
1343 EXIT(EXIT_FAILURE); \
1345 SET_INFO_PTR(node, FullSVar_info); \
1346 SVAR_VALUE(node) = value; \
1347 tso = SVAR_HEAD(node); \
1348 if (tso != (P_) PrelBase_Z91Z93_closure) { \
1350 STGCALL3(void,(void *, char *, P_, P_),QP_Event2,do_qp_prof > 1 ? "RA" : "RG",tso,CurrentTSO); \
1351 if (ThreadQueueHd == PrelBase_Z91Z93_closure) \
1352 ThreadQueueHd = tso; \
1354 TSO_LINK(ThreadQueueTl) = tso; \
1355 ThreadQueueTl = tso; \
1356 SVAR_HEAD(node) = TSO_LINK(tso); \
1357 TSO_LINK(tso) = (P_) PrelBase_Z91Z93_closure; \
1358 if(SVAR_HEAD(node) == (P_) PrelBase_Z91Z93_closure) \
1359 SVAR_TAIL(node) = (P_) PrelBase_Z91Z93_closure; \
1365 #define putMVarZh(node, value) \
1368 if (INFO_PTR(node) == (W_) FullSVar_info) { \
1369 /* Don't wrap the calls; we're done with STG land */\
1371 fprintf(stderr, "putMVar#: MVar already full.\n"); \
1372 EXIT(EXIT_FAILURE); \
1374 SET_INFO_PTR(node, FullSVar_info); \
1375 SVAR_VALUE(node) = value; \
1376 tso = SVAR_HEAD(node); \
1377 if (tso != (P_) PrelBase_Z91Z93_closure) { \
1379 STGCALL3(void,(void *, char *, P_, P_),QP_Event2,do_qp_prof > 1 ? "RA" : "RG",tso,CurrentTSO); \
1380 if (RunnableThreadsHd == PrelBase_Z91Z93_closure) \
1381 RunnableThreadsHd = tso; \
1383 TSO_LINK(RunnableThreadsTl) = tso; \
1384 RunnableThreadsTl = tso; \
1385 SVAR_HEAD(node) = TSO_LINK(tso); \
1386 TSO_LINK(tso) = (P_) PrelBase_Z91Z93_closure; \
1387 if(SVAR_HEAD(node) == (P_) PrelBase_Z91Z93_closure) \
1388 SVAR_TAIL(node) = (P_) PrelBase_Z91Z93_closure; \
1396 #define putMVarZh(node, value) \
1399 if (INFO_PTR(node) == (W_) FullSVar_info) { \
1400 /* Don't wrap the calls; we're done with STG land */\
1402 fprintf(stderr, "putMVar#: MVar already full.\n"); \
1403 EXIT(EXIT_FAILURE); \
1405 SET_INFO_PTR(node, FullSVar_info); \
1406 SVAR_VALUE(node) = value; \
1415 #define readIVarZh(r, liveness, node) \
1417 if (INFO_PTR(node) != (W_) ImMutArrayOfPtrs_info) { \
1418 if (SVAR_HEAD(node) == PrelBase_Z91Z93_closure) \
1419 SVAR_HEAD(node) = CurrentTSO; \
1421 TSO_LINK(SVAR_TAIL(node)) = CurrentTSO; \
1422 TSO_LINK(CurrentTSO) = (P_) PrelBase_Z91Z93_closure; \
1423 SVAR_TAIL(node) = CurrentTSO; \
1424 DO_YIELD(liveness << 1); \
1426 r = SVAR_VALUE(node); \
1431 #define readIVarZh(r, liveness, node) \
1433 if (INFO_PTR(node) != (W_) ImMutArrayOfPtrs_info) { \
1434 /* Don't wrap the calls; we're done with STG land */\
1436 fprintf(stderr, "readIVar#: IVar is empty.\n"); \
1437 EXIT(EXIT_FAILURE); \
1439 r = SVAR_VALUE(node); \
1450 /* Only difference to the !GRAN def: RunnableThreadsHd has been replaced by */
1451 /* ThreadQueueHd i.e. the tso is added at the end of the thread queue on */
1452 /* the CurrentProc. This means we have an implicit context switch after */
1453 /* writeIVar even if unfair scheduling is used in GranSim (default)! -- HWL */
1455 #define writeIVarZh(node, value) \
1458 if (INFO_PTR(node) == (W_) ImMutArrayOfPtrs_info) { \
1459 /* Don't wrap the calls; we're done with STG land */\
1461 fprintf(stderr, "writeIVar#: IVar already full.\n");\
1462 EXIT(EXIT_FAILURE); \
1464 tso = SVAR_HEAD(node); \
1465 if (tso != (P_) PrelBase_Z91Z93_closure) { \
1466 if (ThreadQueueHd == PrelBase_Z91Z93_closure) \
1467 ThreadQueueHd = tso; \
1469 TSO_LINK(ThreadQueueTl) = tso; \
1470 while(TSO_LINK(tso) != PrelBase_Z91Z93_closure) { \
1472 STGCALL3(void,(void *, char *, P_, P_),QP_Event2,do_qp_prof > 1 ? "RA" : "RG",tso,CurrentTSO); \
1473 tso = TSO_LINK(tso); \
1476 STGCALL3(void,(void *, char *, P_, P_),QP_Event2,do_qp_prof > 1 ? "RA" : "RG",tso,CurrentTSO); \
1477 ThreadQueueTl = tso; \
1479 /* Don't use freeze, since it's conditional on GC */ \
1480 SET_INFO_PTR(node, ImMutArrayOfPtrs_info); \
1481 MUTUPLE_CLOSURE_SIZE(node) = (MUTUPLE_VHS+1); \
1482 SVAR_VALUE(node) = value; \
1487 #define writeIVarZh(node, value) \
1490 if (INFO_PTR(node) == (W_) ImMutArrayOfPtrs_info) { \
1491 /* Don't wrap the calls; we're done with STG land */\
1493 fprintf(stderr, "writeIVar#: IVar already full.\n");\
1494 EXIT(EXIT_FAILURE); \
1496 tso = SVAR_HEAD(node); \
1497 if (tso != (P_) PrelBase_Z91Z93_closure) { \
1498 if (RunnableThreadsHd == PrelBase_Z91Z93_closure) \
1499 RunnableThreadsHd = tso; \
1501 TSO_LINK(RunnableThreadsTl) = tso; \
1502 while(TSO_LINK(tso) != PrelBase_Z91Z93_closure) { \
1504 STGCALL3(void,(void *, char *, P_, P_),QP_Event2,do_qp_prof > 1 ? "RA" : "RG",tso,CurrentTSO); \
1505 tso = TSO_LINK(tso); \
1508 STGCALL3(void,(void *, char *, P_, P_),QP_Event2,do_qp_prof > 1 ? "RA" : "RG",tso,CurrentTSO); \
1509 RunnableThreadsTl = tso; \
1511 /* Don't use freeze, since it's conditional on GC */ \
1512 SET_INFO_PTR(node, ImMutArrayOfPtrs_info); \
1513 MUTUPLE_CLOSURE_SIZE(node) = (MUTUPLE_VHS+1); \
1514 SVAR_VALUE(node) = value; \
1521 #define writeIVarZh(node, value) \
1524 if (INFO_PTR(node) == (W_) ImMutArrayOfPtrs_info) { \
1525 /* Don't wrap the calls; we're done with STG land */\
1527 fprintf(stderr, "writeIVar#: IVar already full.\n");\
1528 EXIT(EXIT_FAILURE); \
1530 /* Don't use freeze, since it's conditional on GC */ \
1531 SET_INFO_PTR(node, ImMutArrayOfPtrs_info); \
1532 MUTUPLE_CLOSURE_SIZE(node) = (MUTUPLE_VHS+1); \
1533 SVAR_VALUE(node) = value; \
1539 %************************************************************************
1541 \subsubsection[StgMacros-Wait-primops]{Delay/Wait PrimOps}
1543 %************************************************************************
1548 /* ToDo: for GRAN */
1550 #define delayZh(liveness, us) \
1552 if (WaitingThreadsTl == PrelBase_Z91Z93_closure) \
1553 WaitingThreadsHd = CurrentTSO; \
1555 TSO_LINK(WaitingThreadsTl) = CurrentTSO; \
1556 WaitingThreadsTl = CurrentTSO; \
1557 TSO_LINK(CurrentTSO) = PrelBase_Z91Z93_closure; \
1558 TSO_EVENT(CurrentTSO) = (W_) ((us) < 1 ? 1 : (us)); \
1559 DO_YIELD(liveness << 1); \
1564 #define delayZh(liveness, us) \
1567 fprintf(stderr, "delay#: unthreaded build.\n"); \
1568 EXIT(EXIT_FAILURE); \
1575 /* ToDo: something for GRAN */
1577 #define waitReadZh(liveness, fd) \
1579 if (WaitingThreadsTl == PrelBase_Z91Z93_closure) \
1580 WaitingThreadsHd = CurrentTSO; \
1582 TSO_LINK(WaitingThreadsTl) = CurrentTSO; \
1583 WaitingThreadsTl = CurrentTSO; \
1584 TSO_LINK(CurrentTSO) = PrelBase_Z91Z93_closure; \
1585 TSO_EVENT(CurrentTSO) = (W_) (-(fd)); \
1586 DO_YIELD(liveness << 1); \
1591 #define waitReadZh(liveness, fd) \
1594 fprintf(stderr, "waitRead#: unthreaded build.\n"); \
1595 EXIT(EXIT_FAILURE); \
1602 /* ToDo: something for GRAN */
1604 #ifdef HAVE_SYS_TYPES_H
1605 #include <sys/types.h>
1606 #endif HAVE_SYS_TYPES_H */
1608 #define waitWriteZh(liveness, fd) \
1610 if (WaitingThreadsTl == PrelBase_Z91Z93_closure) \
1611 WaitingThreadsHd = CurrentTSO; \
1613 TSO_LINK(WaitingThreadsTl) = CurrentTSO; \
1614 WaitingThreadsTl = CurrentTSO; \
1615 TSO_LINK(CurrentTSO) = PrelBase_Z91Z93_closure; \
1616 TSO_EVENT(CurrentTSO) = (W_) (-(fd+FD_SETSIZE)); \
1617 DO_YIELD(liveness << 1); \
1622 #define waitWriteZh(liveness, fd) \
1625 fprintf(stderr, "waitWrite#: unthreaded build.\n"); \
1626 EXIT(EXIT_FAILURE); \
1633 %************************************************************************
1635 \subsubsection[StgMacros-IO-primops]{Primitive I/O, error-handling primops}
1637 %************************************************************************
1640 extern P_ TopClosure;
1641 EXTFUN(ErrorIO_innards);
1642 EXTFUN(__std_entry_error__);
1644 #define errorIOZh(a) \
1645 do { TopClosure=(a); \
1646 (void) SAFESTGCALL1(I_,(void *, FILE *),fflush,stdout); \
1647 (void) SAFESTGCALL1(I_,(void *, FILE *),fflush,stderr); \
1648 JMP_(ErrorIO_innards); \
1651 #if !defined(CALLER_SAVES_SYSTEM)
1652 /* can use the macros */
1653 #define stg_getc(stream) getc((FILE *) (stream))
1654 #define stg_putc(c,stream) putc((c),((FILE *) (stream)))
1656 /* must not use the macros (they contain embedded calls to _filbuf/whatnot) */
1657 #define stg_getc(stream) SAFESTGCALL1(I_,(void *, FILE *),fgetc,(FILE *) (stream))
1658 #define stg_putc(c,stream) SAFESTGCALL2(I_,(void *, char, FILE *),fputc,(c),((FILE *) (stream)))
1661 int initialize_virtual_timer(int us);
1662 int install_segv_handler(STG_NO_ARGS);
1663 int install_vtalrm_handler(STG_NO_ARGS);
1664 void initUserSignals(STG_NO_ARGS);
1665 void blockUserSignals(STG_NO_ARGS);
1666 void unblockUserSignals(STG_NO_ARGS);
1667 IF_RTS(void blockVtAlrmSignal(STG_NO_ARGS);)
1668 IF_RTS(void unblockVtAlrmSignal(STG_NO_ARGS);)
1669 IF_RTS(void AwaitEvent(I_ delta);)
1671 #if defined(_POSIX_SOURCE) && !defined(nextstep3_TARGET_OS)
1672 /* For nextstep3_TARGET_OS comment see stgdefs.h. CaS */
1673 extern I_ sig_install PROTO((I_, I_, sigset_t *));
1674 #define stg_sig_ignore(s,m) SAFESTGCALL3(I_,(void *, I_, I_),sig_install,s,STG_SIG_IGN,(sigset_t *)m)
1675 #define stg_sig_default(s,m) SAFESTGCALL3(I_,(void *, I_, I_),sig_install,s,STG_SIG_DFL,(sigset_t *)m)
1676 #define stg_sig_catch(s,sp,m) SAFESTGCALL3(I_,(void *, I_, I_),sig_install,s,sp,(sigset_t *)m)
1678 extern I_ sig_install PROTO((I_, I_));
1679 #define stg_sig_ignore(s,m) SAFESTGCALL2(I_,(void *, I_, I_),sig_install,s,STG_SIG_IGN)
1680 #define stg_sig_default(s,m) SAFESTGCALL2(I_,(void *, I_, I_),sig_install,s,STG_SIG_DFL)
1681 #define stg_sig_catch(s,sp,m) SAFESTGCALL2(I_,(void *, I_, I_),sig_install,s,sp)
1684 #define STG_SIG_DFL (-1)
1685 #define STG_SIG_IGN (-2)
1686 #define STG_SIG_ERR (-3)
1688 StgInt getErrorHandler(STG_NO_ARGS);
1690 void raiseError PROTO((StgStablePtr handler));
1691 StgInt catchError PROTO((StgStablePtr newErrorHandler));
1693 void decrementErrorCount(STG_NO_ARGS);
1695 #define stg_catchError(sp) SAFESTGCALL1(I_,(void *, StgStablePtr),catchError,sp)
1696 #define stg_decrementErrorCount() SAFESTGCALL0(void,(void *),decrementErrorCount)
1699 %************************************************************************
1701 \subsubsection[StgMacros-stable-ptr]{Primitive ops for manipulating stable pointers}
1703 %************************************************************************
1706 The type of these should be:
1709 makeStablePointer# :: a -> State# _RealWorld -> StateAndStablePtr# _RealWorld a
1710 deRefStablePointer# :: StablePtr# a -> State# _RealWorld -> StateAndPtr _RealWorld a
1713 Since world-tokens are no longer explicitly passed around, the
1714 implementations have a few less arguments/results.
1716 The simpler one is @deRefStablePointer#@ (which is only a primop
1717 because it is more polymorphic than is allowed of a ccall).
1722 #define deRefStablePtrZh(ri,sp) \
1725 fprintf(stderr, "deRefStablePtr#: no stable pointer support.\n");\
1726 EXIT(EXIT_FAILURE); \
1731 extern StgPtr _deRefStablePointer PROTO((StgInt, StgPtr));
1733 #define deRefStablePtrZh(ri,sp) \
1734 ri = SAFESTGCALL2(I_,(void *, I_, P_),_deRefStablePointer,sp,StorageMgrInfo.StablePointerTable);
1737 Declarations for other stable pointer operations.
1740 void freeStablePointer PROTO((I_ stablePtr));
1742 void enterStablePtr PROTO((StgStablePtr, StgFunPtr));
1743 void performIO PROTO((StgStablePtr));
1744 I_ enterInt PROTO((StgStablePtr));
1745 I_ enterFloat PROTO((StgStablePtr));
1746 P_ deRefStablePointer PROTO((StgStablePtr));
1747 IF_RTS(I_ catchSoftHeapOverflow PROTO((StgStablePtr, I_));)
1748 IF_RTS(I_ getSoftHeapOverflowHandler(STG_NO_ARGS);)
1749 IF_RTS(extern StgStablePtr softHeapOverflowHandler;)
1750 IF_RTS(void shutdownHaskell(STG_NO_ARGS);)
1752 EXTFUN(stopPerformIODirectReturn);
1753 EXTFUN(startPerformIO);
1754 EXTFUN(stopEnterIntDirectReturn);
1755 EXTFUN(startEnterInt);
1756 EXTFUN(stopEnterFloatDirectReturn);
1757 EXTFUN(startEnterFloat);
1759 void enterStablePtr PROTO((StgStablePtr stableIndex, StgFunPtr startCode));
1763 IF_RTS(extern I_ ErrorIO_call_count;)
1766 Somewhat harder is @makeStablePointer#@ --- it is usually simple but
1767 if we're unlucky, it will have to allocate a new table and copy the
1768 old bit over. Since we might, very occasionally, have to call the
1769 garbage collector, this has to be a macro... sigh!
1771 NB @newSP@ is required because it is entirely possible that
1772 @stablePtr@ and @unstablePtr@ are aliases and so we can't do the
1773 assignment to @stablePtr@ until we've finished with @unstablePtr@.
1775 Another obscure piece of coding is the recalculation of the size of
1776 the table. We do this just in case Jim's threads decide they want to
1777 context switch---in which case any stack-allocated variables may get
1778 trashed. (If only there was a special heap check which didn't
1779 consider context switching...)
1784 /* Calculate SP Table size from number of pointers */
1785 #define SPTSizeFromNoPtrs( newNP ) (DYN_VHS + 1 + 2 * (newNP))
1787 /* Calculate number of pointers in new table from number in old table:
1788 any strictly increasing expression will do here */
1789 #define CalcNewNoSPtrs( i ) ((i)*2 + 100)
1791 void enlargeSPTable PROTO((P_, P_));
1793 #define makeStablePtrZh(stablePtr,liveness,unstablePtr) \
1795 EXTDATA_RO(StablePointerTable_info); \
1796 EXTDATA(UnusedSP); \
1797 StgStablePtr newSP; \
1799 if (SPT_EMPTY(StorageMgrInfo.StablePointerTable)) { /* free stack is empty */ \
1800 { /* Variables used before the heap check */ \
1801 I_ OldNoPtrs = SPT_NoPTRS( StorageMgrInfo.StablePointerTable ); \
1802 I_ NewNoPtrs = CalcNewNoSPtrs( OldNoPtrs ); \
1803 I_ NewSize = SPTSizeFromNoPtrs( NewNoPtrs ); \
1804 HEAP_CHK(liveness, _FHS+NewSize, 0); \
1806 { /* Variables used after the heap check - same values */ \
1807 I_ OldNoPtrs = SPT_NoPTRS( StorageMgrInfo.StablePointerTable ); \
1808 I_ NewNoPtrs = CalcNewNoSPtrs( OldNoPtrs ); \
1809 I_ NewSize = SPTSizeFromNoPtrs( NewNoPtrs ); \
1810 P_ SPTable = Hp + 1 - (_FHS + NewSize); \
1812 CC_ALLOC(CCC, _FHS+NewSize, SPT_K); /* cc prof */ \
1813 SET_DYN_HDR(SPTable,StablePointerTable_info,CCC,NewSize,NewNoPtrs);\
1814 SAFESTGCALL2(void, (void *, P_, P_), enlargeSPTable, SPTable, StorageMgrInfo.StablePointerTable); \
1815 StorageMgrInfo.StablePointerTable = SPTable; \
1819 newSP = SPT_POP(StorageMgrInfo.StablePointerTable); \
1820 SPT_SPTR(StorageMgrInfo.StablePointerTable, newSP) = unstablePtr; \
1821 CHECK_SPT_CLOSURE( StorageMgrInfo.StablePointerTable ); \
1822 stablePtr = newSP; \
1827 #define makeStablePtrZh(stablePtr,liveness,unstablePtr) \
1830 fprintf(stderr, "makeStablePtr#: no stable pointer support.\n");\
1831 EXIT(EXIT_FAILURE); \
1837 %************************************************************************
1839 \subsubsection[StgMacros-unsafePointerEquality]{Primitive `op' for breaking referential transparency}
1841 %************************************************************************
1843 The type of this is @reallyUnsafePtrEquality :: a -> a -> Int#@ so we
1844 can expect three parameters: the two arguments and a "register" to put
1847 Message to Will: This primop breaks referential transparency so badly
1848 you might want to leave it out. On the other hand, if you hide it
1849 away in an appropriate monad, it's perfectly safe. [ADR]
1851 Note that this primop is non-deterministic: different results can be
1852 obtained depending on just what the garbage collector (and code
1853 optimiser??) has done. However, we can guarantee that if two objects
1854 are pointer-equal, they have the same denotation --- the converse most
1855 certainly doesn't hold.
1857 ToDo ADR: The degree of non-determinism could be greatly reduced by
1858 following indirections.
1861 #define reallyUnsafePtrEqualityZh(r,a,b) r=((StgPtr)(a) == (StgPtr)(b))
1864 %************************************************************************
1866 \subsubsection[StgMacros-parallel-primop]{Primitive `op' for sparking (etc)}
1868 %************************************************************************
1870 Assuming local sparking in some form, we can now inline the spark request.
1872 We build a doubly-linked list in the heap, so that we can handle FIFO
1873 or LIFO scheduling as we please.
1875 Anything with tag >= 0 is in WHNF, so we discard it.
1880 ED_(PrelBase_Z91Z93_closure);
1884 #define parZh(r,node) \
1885 PARZh(r,node,1,0,0,0,0,0)
1887 #define parAtZh(r,node,where,identifier,gran_info,size_info,par_info,rest) \
1888 parATZh(r,node,where,identifier,gran_info,size_info,par_info,rest,1)
1890 #define parAtAbsZh(r,node,proc,identifier,gran_info,size_info,par_info,rest) \
1891 parATZh(r,node,proc,identifier,gran_info,size_info,par_info,rest,2)
1893 #define parAtRelZh(r,node,proc,identifier,gran_info,size_info,par_info,rest) \
1894 parATZh(r,node,proc,identifier,gran_info,size_info,par_info,rest,3)
1896 #define parAtForNowZh(r,node,where,identifier,gran_info,size_info,par_info,rest) \
1897 parATZh(r,node,where,identifier,gran_info,size_info,par_info,rest,0)
1899 #define parATZh(r,node,where,identifier,gran_info,size_info,par_info,rest,local) \
1902 if (SHOULD_SPARK(node)) { \
1905 result = NewSpark((P_)node,identifier,gran_info,size_info,par_info,local); \
1906 if (local==2) { /* special case for parAtAbs */ \
1907 GranSimSparkAtAbs(result,(I_)where,identifier);\
1908 } else if (local==3) { /* special case for parAtRel */ \
1909 GranSimSparkAtAbs(result,(I_)(CurrentProc+where),identifier); \
1911 GranSimSparkAt(result,where,identifier); \
1913 context_switch = 1; \
1915 RestoreAllStgRegs(); \
1916 } else if (do_qp_prof) { \
1917 I_ tid = threadId++; \
1918 SAFESTGCALL2(void,(I_, P_),QP_Event0,tid,node); \
1920 r = 1; /* return code for successful spark -- HWL */ \
1923 #define parLocalZh(r,node,identifier,gran_info,size_info,par_info,rest) \
1924 PARZh(r,node,rest,identifier,gran_info,size_info,par_info,1)
1926 #define parGlobalZh(r,node,identifier,gran_info,size_info,par_info,rest) \
1927 PARZh(r,node,rest,identifier,gran_info,size_info,par_info,0)
1931 #define PARZh(r,node,rest,identifier,gran_info,size_info,par_info,local) \
1933 if (SHOULD_SPARK(node)) { \
1936 result = NewSpark((P_)node,identifier,gran_info,size_info,par_info,local);\
1937 add_to_spark_queue(result); \
1938 GranSimSpark(local,(P_)node); \
1939 context_switch = 1; \
1941 RestoreAllStgRegs(); \
1942 } else if (do_qp_prof) { \
1943 I_ tid = threadId++; \
1944 SAFESTGCALL2(void,(I_, P_),QP_Event0,tid,node); \
1946 r = 1; /* return code for successful spark -- HWL */ \
1951 #define PARZh(r,node,rest,identifier,gran_info,size_info,par_info,local) \
1954 if (SHOULD_SPARK(node)) { \
1955 result = NewSpark((P_)node,identifier,gran_info,size_info,par_info,local);\
1956 ADD_TO_SPARK_QUEUE(result); \
1957 SAFESTGCALL2(void,(W_),GranSimSpark,local,(P_)node); \
1958 /* context_switch = 1; not needed any more -- HWL */ \
1959 } else if (do_qp_prof) { \
1960 I_ tid = threadId++; \
1961 SAFESTGCALL2(void,(I_, P_),QP_Event0,tid,node); \
1963 r = 1; /* return code for successful spark -- HWL */ \
1968 #define copyableZh(r,node) \
1969 /* copyable not yet implemented!! */
1971 #define noFollowZh(r,node) \
1972 /* noFollow not yet implemented!! */
1976 extern I_ required_thread_count;
1979 #define COUNT_SPARK TSO_GLOBALSPARKS(CurrentTSO)++
1985 Note that we must bump the required thread count NOW, rather
1986 than when the thread is actually created.
1989 #define forkZh(r,liveness,node) \
1991 while (PendingSparksTl[REQUIRED_POOL] == PendingSparksLim[REQUIRED_POOL]) \
1992 DO_YIELD((liveness << 1) | 1); \
1994 if (SHOULD_SPARK(node)) { \
1995 *PendingSparksTl[REQUIRED_POOL]++ = (P_)(node); \
1996 } else if (DO_QP_PROF) { \
1997 I_ tid = threadId++; \
1998 SAFESTGCALL2(void,(I_, P_),QP_Event0,tid,node); \
2000 required_thread_count++; \
2001 context_switch = 1; \
2002 r = 1; /* Should not be necessary */ \
2005 #define parZh(r,node) \
2008 if (SHOULD_SPARK(node) && \
2009 PendingSparksTl[ADVISORY_POOL] < PendingSparksLim[ADVISORY_POOL]) { \
2010 *PendingSparksTl[ADVISORY_POOL]++ = (P_)(node); \
2014 I_ tid = threadId++; \
2015 SAFESTGCALL2(void,(I_, P_),QP_Event0,tid,node); \
2018 r = 1; /* Should not be necessary */ \
2024 The following seq# code should only be used in unoptimized code.
2025 Be warned: it's a potential bug-farm.
2027 First we push two words on the B stack: the current value of RetReg
2028 (which may or may not be live), and a continuation snatched largely out
2029 of thin air (it's a point within this code block). Then we set RetReg
2030 to the special polymorphic return code for seq, load up Node with the
2031 closure to be evaluated, and we're off. When the eval returns to the
2032 polymorphic seq return point, the two words are popped off the B stack,
2033 RetReg is restored, and we jump to the continuation, completing the
2034 primop and going on our merry way.
2040 #define seqZh(r,liveness,node) \
2043 /* STK_CHK(liveness,0,2,0,0,0,0); */ \
2044 /* SpB -= BREL(2); */ \
2045 SpB[BREL(0)] = (W_) RetReg; \
2046 SpB[BREL(1)] = (W_) &&cont; \
2047 RetReg = (StgRetAddr) vtbl_seq; \
2050 InfoPtr = (D_)(INFO_PTR(Node)); \
2051 JMP_(ENTRY_CODE(InfoPtr)); \
2053 r = 1; /* Should be unnecessary */ \
2056 #endif /* CONCURRENT */
2059 %************************************************************************
2061 \subsubsection[StgMacros-foreign-objects]{Foreign Objects}
2063 %************************************************************************
2065 [Based on previous MallocPtr comments -- SOF]
2067 This macro is used to construct a ForeignObj on the heap.
2069 What this does is plug the pointer (which will be in a local
2070 variable) together with its finalising/free routine, into a fresh heap
2071 object and then sets a result (which will be a register) to point
2072 to the fresh heap object.
2074 To accommodate per-object finalisation, augment the macro with a
2075 finalisation routine argument. Nothing spectacular, just plug the
2076 pointer to the routine into the ForeignObj -- SOF 4/96
2078 Question: what's this "SET_ACTIVITY" stuff - should I be doing this
2079 too? (It's if you want to use the SPAT profiling tools to
2080 characterize program behavior by ``activity'' -- tail-calling,
2081 heap-checking, etc. -- see Ticky.lh. It is quite specialized.
2084 (Swapped first two arguments to make it come into line with what appears
2085 to be `standard' format, return register then liveness mask. -- SOF 4/96)
2090 StgInt eqForeignObj PROTO((StgForeignObj p1, StgForeignObj p2));
2092 #define makeForeignObjZh(r, liveness, mptr, finalise) \
2096 HEAP_CHK((W_)liveness, _FHS + ForeignObj_SIZE,0); \
2097 CC_ALLOC(CCC,_FHS + ForeignObj_SIZE,ForeignObj_K); /* cc prof */ \
2099 result = Hp + 1 - (_FHS + ForeignObj_SIZE); \
2100 SET_ForeignObj_HDR(result,ForeignObj_info,CCC,_FHS + ForeignObj_SIZE,0); \
2101 ForeignObj_CLOSURE_DATA(result) = (P_)mptr; \
2102 ForeignObj_CLOSURE_FINALISER(result) = (P_)finalise; \
2103 ForeignObj_CLOSURE_LINK(result) = StorageMgrInfo.ForeignObjList; \
2104 StorageMgrInfo.ForeignObjList = result; \
2107 /*fprintf(stderr,"DEBUG: ForeignObj(0x%x) = <0x%x, 0x%x, 0x%x, 0x%x>\n", \
2109 result[0],result[1], \
2110 result[2],result[3]);*/ \
2112 CHECK_ForeignObj_CLOSURE( result ); \
2113 VALIDATE_ForeignObjList( StorageMgrInfo.ForeignObjList ); \
2115 (r) = (P_) result; \
2118 #define writeForeignObjZh(res,datum) ((PP_) ForeignObj_CLOSURE_DATA(res)) = ((P_)datum)
2121 #define makeForeignObjZh(r, liveness, mptr, finalise) \
2124 fprintf(stderr, "makeForeignObj#: no foreign object support.\n");\
2125 EXIT(EXIT_FAILURE); \
2128 #define writeForeignObjZh(res,datum) \
2131 fprintf(stderr, "writeForeignObj#: no foreign object support.\n");\
2132 EXIT(EXIT_FAILURE); \
2139 End-of-file's multi-slurp protection:
2141 #endif /* ! STGMACROS_H */