NUMA support
[ghc.git] / includes / Cmm.h
1 /* -----------------------------------------------------------------------------
2 *
3 * (c) The University of Glasgow 2004-2013
4 *
5 * This file is included at the top of all .cmm source files (and
6 * *only* .cmm files). It defines a collection of useful macros for
7 * making .cmm code a bit less error-prone to write, and a bit easier
8 * on the eye for the reader.
9 *
10 * For the syntax of .cmm files, see the parser in ghc/compiler/cmm/CmmParse.y.
11 *
12 * Accessing fields of structures defined in the RTS header files is
13 * done via automatically-generated macros in DerivedConstants.h. For
14 * example, where previously we used
15 *
16 * CurrentTSO->what_next = x
17 *
18 * in C-- we now use
19 *
20 * StgTSO_what_next(CurrentTSO) = x
21 *
22 * where the StgTSO_what_next() macro is automatically generated by
23 * mkDerivedConstants.c. If you need to access a field that doesn't
24 * already have a macro, edit that file (it's pretty self-explanatory).
25 *
26 * -------------------------------------------------------------------------- */
27
28 #ifndef CMM_H
29 #define CMM_H
30
31 /*
32 * In files that are included into both C and C-- (and perhaps
33 * Haskell) sources, we sometimes need to conditionally compile bits
34 * depending on the language. CMINUSMINUS==1 in .cmm sources:
35 */
36 #define CMINUSMINUS 1
37
38 #include "ghcconfig.h"
39
40 /* -----------------------------------------------------------------------------
41 Types
42
43 The following synonyms for C-- types are declared here:
44
45 I8, I16, I32, I64 MachRep-style names for convenience
46
47 W_ is shorthand for the word type (== StgWord)
48 F_ shorthand for float (F_ == StgFloat == C's float)
49 D_ shorthand for double (D_ == StgDouble == C's double)
50
51 CInt has the same size as an int in C on this platform
52 CLong has the same size as a long in C on this platform
53
54 --------------------------------------------------------------------------- */
55
56 #define I8 bits8
57 #define I16 bits16
58 #define I32 bits32
59 #define I64 bits64
60 #define P_ gcptr
61
62 #if SIZEOF_VOID_P == 4
63 #define W_ bits32
64 /* Maybe it's better to include MachDeps.h */
65 #define TAG_BITS 2
66 #elif SIZEOF_VOID_P == 8
67 #define W_ bits64
68 /* Maybe it's better to include MachDeps.h */
69 #define TAG_BITS 3
70 #else
71 #error Unknown word size
72 #endif
73
74 /*
75 * The RTS must sometimes UNTAG a pointer before dereferencing it.
76 * See the wiki page Commentary/Rts/HaskellExecution/PointerTagging
77 */
78 #define TAG_MASK ((1 << TAG_BITS) - 1)
79 #define UNTAG(p) (p & ~TAG_MASK)
80 #define GETTAG(p) (p & TAG_MASK)
81
82 #if SIZEOF_INT == 4
83 #define CInt bits32
84 #elif SIZEOF_INT == 8
85 #define CInt bits64
86 #else
87 #error Unknown int size
88 #endif
89
90 #if SIZEOF_LONG == 4
91 #define CLong bits32
92 #elif SIZEOF_LONG == 8
93 #define CLong bits64
94 #else
95 #error Unknown long size
96 #endif
97
98 #define F_ float32
99 #define D_ float64
100 #define L_ bits64
101 #define V16_ bits128
102 #define V32_ bits256
103 #define V64_ bits512
104
105 #define SIZEOF_StgDouble 8
106 #define SIZEOF_StgWord64 8
107
108 /* -----------------------------------------------------------------------------
109 Misc useful stuff
110 -------------------------------------------------------------------------- */
111
112 #define ccall foreign "C"
113
114 #define NULL (0::W_)
115
116 #define STRING(name,str) \
117 section "rodata" { \
118 name : bits8[] str; \
119 } \
120
121 #ifdef TABLES_NEXT_TO_CODE
122 #define RET_LBL(f) f##_info
123 #else
124 #define RET_LBL(f) f##_ret
125 #endif
126
127 #ifdef TABLES_NEXT_TO_CODE
128 #define ENTRY_LBL(f) f##_info
129 #else
130 #define ENTRY_LBL(f) f##_entry
131 #endif
132
133 /* -----------------------------------------------------------------------------
134 Byte/word macros
135
136 Everything in C-- is in byte offsets (well, most things). We use
137 some macros to allow us to express offsets in words and to try to
138 avoid byte/word confusion.
139 -------------------------------------------------------------------------- */
140
141 #define SIZEOF_W SIZEOF_VOID_P
142 #define W_MASK (SIZEOF_W-1)
143
144 #if SIZEOF_W == 4
145 #define W_SHIFT 2
146 #elif SIZEOF_W == 8
147 #define W_SHIFT 3
148 #endif
149
150 /* Converting quantities of words to bytes */
151 #define WDS(n) ((n)*SIZEOF_W)
152
153 /*
154 * Converting quantities of bytes to words
155 * NB. these work on *unsigned* values only
156 */
157 #define BYTES_TO_WDS(n) ((n) / SIZEOF_W)
158 #define ROUNDUP_BYTES_TO_WDS(n) (((n) + SIZEOF_W - 1) / SIZEOF_W)
159
160 /* TO_W_(n) converts n to W_ type from a smaller type */
161 #if SIZEOF_W == 4
162 #define TO_W_(x) %sx32(x)
163 #define HALF_W_(x) %lobits16(x)
164 #elif SIZEOF_W == 8
165 #define TO_W_(x) %sx64(x)
166 #define HALF_W_(x) %lobits32(x)
167 #endif
168
169 #if SIZEOF_INT == 4 && SIZEOF_W == 8
170 #define W_TO_INT(x) %lobits32(x)
171 #elif SIZEOF_INT == SIZEOF_W
172 #define W_TO_INT(x) (x)
173 #endif
174
175 #if SIZEOF_LONG == 4 && SIZEOF_W == 8
176 #define W_TO_LONG(x) %lobits32(x)
177 #elif SIZEOF_LONG == SIZEOF_W
178 #define W_TO_LONG(x) (x)
179 #endif
180
181 /* -----------------------------------------------------------------------------
182 Heap/stack access, and adjusting the heap/stack pointers.
183 -------------------------------------------------------------------------- */
184
185 #define Sp(n) W_[Sp + WDS(n)]
186 #define Hp(n) W_[Hp + WDS(n)]
187
188 #define Sp_adj(n) Sp = Sp + WDS(n) /* pronounced "spadge" */
189 #define Hp_adj(n) Hp = Hp + WDS(n)
190
191 /* -----------------------------------------------------------------------------
192 Assertions and Debuggery
193 -------------------------------------------------------------------------- */
194
195 #ifdef DEBUG
196 #define ASSERT(predicate) \
197 if (predicate) { \
198 /*null*/; \
199 } else { \
200 foreign "C" _assertFail(NULL, __LINE__) never returns; \
201 }
202 #else
203 #define ASSERT(p) /* nothing */
204 #endif
205
206 #ifdef DEBUG
207 #define DEBUG_ONLY(s) s
208 #else
209 #define DEBUG_ONLY(s) /* nothing */
210 #endif
211
212 /*
213 * The IF_DEBUG macro is useful for debug messages that depend on one
214 * of the RTS debug options. For example:
215 *
216 * IF_DEBUG(RtsFlags_DebugFlags_apply,
217 * foreign "C" fprintf(stderr, stg_ap_0_ret_str));
218 *
219 * Note the syntax is slightly different to the C version of this macro.
220 */
221 #ifdef DEBUG
222 #define IF_DEBUG(c,s) if (RtsFlags_DebugFlags_##c(RtsFlags) != 0::I32) { s; }
223 #else
224 #define IF_DEBUG(c,s) /* nothing */
225 #endif
226
227 /* -----------------------------------------------------------------------------
228 Entering
229
230 It isn't safe to "enter" every closure. Functions in particular
231 have no entry code as such; their entry point contains the code to
232 apply the function.
233
234 ToDo: range should end in N_CLOSURE_TYPES-1, not N_CLOSURE_TYPES,
235 but switch doesn't allow us to use exprs there yet.
236
237 If R1 points to a tagged object it points either to
238 * A constructor.
239 * A function with arity <= TAG_MASK.
240 In both cases the right thing to do is to return.
241 Note: it is rather lucky that we can use the tag bits to do this
242 for both objects. Maybe it points to a brittle design?
243
244 Indirections can contain tagged pointers, so their tag is checked.
245 -------------------------------------------------------------------------- */
246
247 #ifdef PROFILING
248
249 // When profiling, we cannot shortcut ENTER() by checking the tag,
250 // because LDV profiling relies on entering closures to mark them as
251 // "used".
252
253 #define LOAD_INFO(ret,x) \
254 info = %INFO_PTR(UNTAG(x));
255
256 #define UNTAG_IF_PROF(x) UNTAG(x)
257
258 #else
259
260 #define LOAD_INFO(ret,x) \
261 if (GETTAG(x) != 0) { \
262 ret(x); \
263 } \
264 info = %INFO_PTR(x);
265
266 #define UNTAG_IF_PROF(x) (x) /* already untagged */
267
268 #endif
269
270 // We need two versions of ENTER():
271 // - ENTER(x) takes the closure as an argument and uses return(),
272 // for use in civilized code where the stack is handled by GHC
273 //
274 // - ENTER_NOSTACK() where the closure is in R1, and returns are
275 // explicit jumps, for use when we are doing the stack management
276 // ourselves.
277
278 #define ENTER(x) ENTER_(return,x)
279 #define ENTER_R1() ENTER_(RET_R1,R1)
280
281 #define RET_R1(x) jump %ENTRY_CODE(Sp(0)) [R1]
282
283 #define ENTER_(ret,x) \
284 again: \
285 W_ info; \
286 LOAD_INFO(ret,x) \
287 switch [INVALID_OBJECT .. N_CLOSURE_TYPES] \
288 (TO_W_( %INFO_TYPE(%STD_INFO(info)) )) { \
289 case \
290 IND, \
291 IND_STATIC: \
292 { \
293 x = StgInd_indirectee(x); \
294 goto again; \
295 } \
296 case \
297 FUN, \
298 FUN_1_0, \
299 FUN_0_1, \
300 FUN_2_0, \
301 FUN_1_1, \
302 FUN_0_2, \
303 FUN_STATIC, \
304 BCO, \
305 PAP: \
306 { \
307 ret(x); \
308 } \
309 default: \
310 { \
311 x = UNTAG_IF_PROF(x); \
312 jump %ENTRY_CODE(info) (x); \
313 } \
314 }
315
316 // The FUN cases almost never happen: a pointer to a non-static FUN
317 // should always be tagged. This unfortunately isn't true for the
318 // interpreter right now, which leaves untagged FUNs on the stack.
319
320 /* -----------------------------------------------------------------------------
321 Constants.
322 -------------------------------------------------------------------------- */
323
324 #include "rts/Constants.h"
325 #include "DerivedConstants.h"
326 #include "rts/storage/ClosureTypes.h"
327 #include "rts/storage/FunTypes.h"
328 #include "rts/OSThreads.h"
329
330 /*
331 * Need MachRegs, because some of the RTS code is conditionally
332 * compiled based on REG_R1, REG_R2, etc.
333 */
334 #include "stg/RtsMachRegs.h"
335
336 #include "rts/prof/LDV.h"
337
338 #undef BLOCK_SIZE
339 #undef MBLOCK_SIZE
340 #include "rts/storage/Block.h" /* For Bdescr() */
341
342
343 #define MyCapability() (BaseReg - OFFSET_Capability_r)
344
345 /* -------------------------------------------------------------------------
346 Info tables
347 ------------------------------------------------------------------------- */
348
349 #if defined(PROFILING)
350 #define PROF_HDR_FIELDS(w_,hdr1,hdr2) \
351 w_ hdr1, \
352 w_ hdr2,
353 #else
354 #define PROF_HDR_FIELDS(w_,hdr1,hdr2) /* nothing */
355 #endif
356
357 /* -------------------------------------------------------------------------
358 Allocation and garbage collection
359 ------------------------------------------------------------------------- */
360
361 /*
362 * ALLOC_PRIM is for allocating memory on the heap for a primitive
363 * object. It is used all over PrimOps.cmm.
364 *
365 * We make the simplifying assumption that the "admin" part of a
366 * primitive closure is just the header when calculating sizes for
367 * ticky-ticky. It's not clear whether eg. the size field of an array
368 * should be counted as "admin", or the various fields of a BCO.
369 */
370 #define ALLOC_PRIM(bytes) \
371 HP_CHK_GEN_TICKY(bytes); \
372 TICK_ALLOC_PRIM(SIZEOF_StgHeader,bytes-SIZEOF_StgHeader,0); \
373 CCCS_ALLOC(bytes);
374
375 #define HEAP_CHECK(bytes,failure) \
376 TICK_BUMP(HEAP_CHK_ctr); \
377 Hp = Hp + (bytes); \
378 if (Hp > HpLim) { HpAlloc = (bytes); failure; } \
379 TICK_ALLOC_HEAP_NOCTR(bytes);
380
381 #define ALLOC_PRIM_WITH_CUSTOM_FAILURE(bytes,failure) \
382 HEAP_CHECK(bytes,failure) \
383 TICK_ALLOC_PRIM(SIZEOF_StgHeader,bytes-SIZEOF_StgHeader,0); \
384 CCCS_ALLOC(bytes);
385
386 #define ALLOC_PRIM_(bytes,fun) \
387 ALLOC_PRIM_WITH_CUSTOM_FAILURE(bytes,GC_PRIM(fun));
388
389 #define ALLOC_PRIM_P(bytes,fun,arg) \
390 ALLOC_PRIM_WITH_CUSTOM_FAILURE(bytes,GC_PRIM_P(fun,arg));
391
392 #define ALLOC_PRIM_N(bytes,fun,arg) \
393 ALLOC_PRIM_WITH_CUSTOM_FAILURE(bytes,GC_PRIM_N(fun,arg));
394
395 /* CCS_ALLOC wants the size in words, because ccs->mem_alloc is in words */
396 #define CCCS_ALLOC(__alloc) CCS_ALLOC(BYTES_TO_WDS(__alloc), CCCS)
397
398 #define HP_CHK_GEN_TICKY(bytes) \
399 HP_CHK_GEN(bytes); \
400 TICK_ALLOC_HEAP_NOCTR(bytes);
401
402 #define HP_CHK_P(bytes, fun, arg) \
403 HEAP_CHECK(bytes, GC_PRIM_P(fun,arg))
404
405 // TODO I'm not seeing where ALLOC_P_TICKY is used; can it be removed?
406 // -NSF March 2013
407 #define ALLOC_P_TICKY(bytes, fun, arg) \
408 HP_CHK_P(bytes); \
409 TICK_ALLOC_HEAP_NOCTR(bytes);
410
411 #define CHECK_GC() \
412 (bdescr_link(CurrentNursery) == NULL || \
413 generation_n_new_large_words(W_[g0]) >= TO_W_(CLong[large_alloc_lim]))
414
415 // allocate() allocates from the nursery, so we check to see
416 // whether the nursery is nearly empty in any function that uses
417 // allocate() - this includes many of the primops.
418 //
419 // HACK alert: the __L__ stuff is here to coax the common-block
420 // eliminator into commoning up the call stg_gc_noregs() with the same
421 // code that gets generated by a STK_CHK_GEN() in the same proc. We
422 // also need an if (0) { goto __L__; } so that the __L__ label isn't
423 // optimised away by the control-flow optimiser prior to common-block
424 // elimination (it will be optimised away later).
425 //
426 // This saves some code in gmp-wrappers.cmm where we have lots of
427 // MAYBE_GC() in the same proc as STK_CHK_GEN().
428 //
429 #define MAYBE_GC(retry) \
430 if (CHECK_GC()) { \
431 HpAlloc = 0; \
432 goto __L__; \
433 __L__: \
434 call stg_gc_noregs(); \
435 goto retry; \
436 } \
437 if (0) { goto __L__; }
438
439 #define GC_PRIM(fun) \
440 jump stg_gc_prim(fun);
441
442 // Version of GC_PRIM for use in low-level Cmm. We can call
443 // stg_gc_prim, because it takes one argument and therefore has a
444 // platform-independent calling convention (Note [Syntax of .cmm
445 // files] in CmmParse.y).
446 #define GC_PRIM_LL(fun) \
447 R1 = fun; \
448 jump stg_gc_prim [R1];
449
450 // We pass the fun as the second argument, because the arg is
451 // usually already in the first argument position (R1), so this
452 // avoids moving it to a different register / stack slot.
453 #define GC_PRIM_N(fun,arg) \
454 jump stg_gc_prim_n(arg,fun);
455
456 #define GC_PRIM_P(fun,arg) \
457 jump stg_gc_prim_p(arg,fun);
458
459 #define GC_PRIM_P_LL(fun,arg) \
460 R1 = arg; \
461 R2 = fun; \
462 jump stg_gc_prim_p_ll [R1,R2];
463
464 #define GC_PRIM_PP(fun,arg1,arg2) \
465 jump stg_gc_prim_pp(arg1,arg2,fun);
466
467 #define MAYBE_GC_(fun) \
468 if (CHECK_GC()) { \
469 HpAlloc = 0; \
470 GC_PRIM(fun) \
471 }
472
473 #define MAYBE_GC_N(fun,arg) \
474 if (CHECK_GC()) { \
475 HpAlloc = 0; \
476 GC_PRIM_N(fun,arg) \
477 }
478
479 #define MAYBE_GC_P(fun,arg) \
480 if (CHECK_GC()) { \
481 HpAlloc = 0; \
482 GC_PRIM_P(fun,arg) \
483 }
484
485 #define MAYBE_GC_PP(fun,arg1,arg2) \
486 if (CHECK_GC()) { \
487 HpAlloc = 0; \
488 GC_PRIM_PP(fun,arg1,arg2) \
489 }
490
491 #define STK_CHK_LL(n, fun) \
492 TICK_BUMP(STK_CHK_ctr); \
493 if (Sp - (n) < SpLim) { \
494 GC_PRIM_LL(fun) \
495 }
496
497 #define STK_CHK_P_LL(n, fun, arg) \
498 TICK_BUMP(STK_CHK_ctr); \
499 if (Sp - (n) < SpLim) { \
500 GC_PRIM_P_LL(fun,arg) \
501 }
502
503 #define STK_CHK_PP(n, fun, arg1, arg2) \
504 TICK_BUMP(STK_CHK_ctr); \
505 if (Sp - (n) < SpLim) { \
506 GC_PRIM_PP(fun,arg1,arg2) \
507 }
508
509 #define STK_CHK_ENTER(n, closure) \
510 TICK_BUMP(STK_CHK_ctr); \
511 if (Sp - (n) < SpLim) { \
512 jump __stg_gc_enter_1(closure); \
513 }
514
515 // A funky heap check used by AutoApply.cmm
516
517 #define HP_CHK_NP_ASSIGN_SP0(size,f) \
518 HEAP_CHECK(size, Sp(0) = f; jump __stg_gc_enter_1 [R1];)
519
520 /* -----------------------------------------------------------------------------
521 Closure headers
522 -------------------------------------------------------------------------- */
523
524 /*
525 * This is really ugly, since we don't do the rest of StgHeader this
526 * way. The problem is that values from DerivedConstants.h cannot be
527 * dependent on the way (SMP, PROF etc.). For SIZEOF_StgHeader we get
528 * the value from GHC, but it seems like too much trouble to do that
529 * for StgThunkHeader.
530 */
531 #define SIZEOF_StgThunkHeader SIZEOF_StgHeader+SIZEOF_StgSMPThunkHeader
532
533 #define StgThunk_payload(__ptr__,__ix__) \
534 W_[__ptr__+SIZEOF_StgThunkHeader+ WDS(__ix__)]
535
536 /* -----------------------------------------------------------------------------
537 Closures
538 -------------------------------------------------------------------------- */
539
540 /* The offset of the payload of an array */
541 #define BYTE_ARR_CTS(arr) ((arr) + SIZEOF_StgArrBytes)
542
543 /* The number of words allocated in an array payload */
544 #define BYTE_ARR_WDS(arr) ROUNDUP_BYTES_TO_WDS(StgArrBytes_bytes(arr))
545
546 /* Getting/setting the info pointer of a closure */
547 #define SET_INFO(p,info) StgHeader_info(p) = info
548 #define GET_INFO(p) StgHeader_info(p)
549
550 /* Determine the size of an ordinary closure from its info table */
551 #define sizeW_fromITBL(itbl) \
552 SIZEOF_StgHeader + WDS(%INFO_PTRS(itbl)) + WDS(%INFO_NPTRS(itbl))
553
554 /* NB. duplicated from InfoTables.h! */
555 #define BITMAP_SIZE(bitmap) ((bitmap) & BITMAP_SIZE_MASK)
556 #define BITMAP_BITS(bitmap) ((bitmap) >> BITMAP_BITS_SHIFT)
557
558 /* Debugging macros */
559 #define LOOKS_LIKE_INFO_PTR(p) \
560 ((p) != NULL && \
561 LOOKS_LIKE_INFO_PTR_NOT_NULL(p))
562
563 #define LOOKS_LIKE_INFO_PTR_NOT_NULL(p) \
564 ( (TO_W_(%INFO_TYPE(%STD_INFO(p))) != INVALID_OBJECT) && \
565 (TO_W_(%INFO_TYPE(%STD_INFO(p))) < N_CLOSURE_TYPES))
566
567 #define LOOKS_LIKE_CLOSURE_PTR(p) (LOOKS_LIKE_INFO_PTR(GET_INFO(UNTAG(p))))
568
569 /*
570 * The layout of the StgFunInfoExtra part of an info table changes
571 * depending on TABLES_NEXT_TO_CODE. So we define field access
572 * macros which use the appropriate version here:
573 */
574 #ifdef TABLES_NEXT_TO_CODE
575 /*
576 * when TABLES_NEXT_TO_CODE, slow_apply is stored as an offset
577 * instead of the normal pointer.
578 */
579
580 #define StgFunInfoExtra_slow_apply(fun_info) \
581 (TO_W_(StgFunInfoExtraRev_slow_apply_offset(fun_info)) \
582 + (fun_info) + SIZEOF_StgFunInfoExtraRev + SIZEOF_StgInfoTable)
583
584 #define StgFunInfoExtra_fun_type(i) StgFunInfoExtraRev_fun_type(i)
585 #define StgFunInfoExtra_arity(i) StgFunInfoExtraRev_arity(i)
586 #define StgFunInfoExtra_bitmap(i) StgFunInfoExtraRev_bitmap(i)
587 #else
588 #define StgFunInfoExtra_slow_apply(i) StgFunInfoExtraFwd_slow_apply(i)
589 #define StgFunInfoExtra_fun_type(i) StgFunInfoExtraFwd_fun_type(i)
590 #define StgFunInfoExtra_arity(i) StgFunInfoExtraFwd_arity(i)
591 #define StgFunInfoExtra_bitmap(i) StgFunInfoExtraFwd_bitmap(i)
592 #endif
593
594 #define mutArrCardMask ((1 << MUT_ARR_PTRS_CARD_BITS) - 1)
595 #define mutArrPtrCardDown(i) ((i) >> MUT_ARR_PTRS_CARD_BITS)
596 #define mutArrPtrCardUp(i) (((i) + mutArrCardMask) >> MUT_ARR_PTRS_CARD_BITS)
597 #define mutArrPtrsCardWords(n) ROUNDUP_BYTES_TO_WDS(mutArrPtrCardUp(n))
598
599 #if defined(PROFILING) || (!defined(THREADED_RTS) && defined(DEBUG))
600 #define OVERWRITING_CLOSURE(c) foreign "C" overwritingClosure(c "ptr")
601 #define OVERWRITING_CLOSURE_OFS(c,n) \
602 foreign "C" overwritingClosureOfs(c "ptr", n)
603 #else
604 #define OVERWRITING_CLOSURE(c) /* nothing */
605 #define OVERWRITING_CLOSURE_OFS(c,n) /* nothing */
606 #endif
607
608 #ifdef THREADED_RTS
609 #define prim_write_barrier prim %write_barrier()
610 #else
611 #define prim_write_barrier /* nothing */
612 #endif
613
614 /* -----------------------------------------------------------------------------
615 Ticky macros
616 -------------------------------------------------------------------------- */
617
618 #ifdef TICKY_TICKY
619 #define TICK_BUMP_BY(ctr,n) CLong[ctr] = CLong[ctr] + n
620 #else
621 #define TICK_BUMP_BY(ctr,n) /* nothing */
622 #endif
623
624 #define TICK_BUMP(ctr) TICK_BUMP_BY(ctr,1)
625
626 #define TICK_ENT_DYN_IND() TICK_BUMP(ENT_DYN_IND_ctr)
627 #define TICK_ENT_DYN_THK() TICK_BUMP(ENT_DYN_THK_ctr)
628 #define TICK_ENT_VIA_NODE() TICK_BUMP(ENT_VIA_NODE_ctr)
629 #define TICK_ENT_STATIC_IND() TICK_BUMP(ENT_STATIC_IND_ctr)
630 #define TICK_ENT_PERM_IND() TICK_BUMP(ENT_PERM_IND_ctr)
631 #define TICK_ENT_PAP() TICK_BUMP(ENT_PAP_ctr)
632 #define TICK_ENT_AP() TICK_BUMP(ENT_AP_ctr)
633 #define TICK_ENT_AP_STACK() TICK_BUMP(ENT_AP_STACK_ctr)
634 #define TICK_ENT_BH() TICK_BUMP(ENT_BH_ctr)
635 #define TICK_ENT_LNE() TICK_BUMP(ENT_LNE_ctr)
636 #define TICK_UNKNOWN_CALL() TICK_BUMP(UNKNOWN_CALL_ctr)
637 #define TICK_UPDF_PUSHED() TICK_BUMP(UPDF_PUSHED_ctr)
638 #define TICK_CATCHF_PUSHED() TICK_BUMP(CATCHF_PUSHED_ctr)
639 #define TICK_UPDF_OMITTED() TICK_BUMP(UPDF_OMITTED_ctr)
640 #define TICK_UPD_NEW_IND() TICK_BUMP(UPD_NEW_IND_ctr)
641 #define TICK_UPD_NEW_PERM_IND() TICK_BUMP(UPD_NEW_PERM_IND_ctr)
642 #define TICK_UPD_OLD_IND() TICK_BUMP(UPD_OLD_IND_ctr)
643 #define TICK_UPD_OLD_PERM_IND() TICK_BUMP(UPD_OLD_PERM_IND_ctr)
644
645 #define TICK_SLOW_CALL_FUN_TOO_FEW() TICK_BUMP(SLOW_CALL_FUN_TOO_FEW_ctr)
646 #define TICK_SLOW_CALL_FUN_CORRECT() TICK_BUMP(SLOW_CALL_FUN_CORRECT_ctr)
647 #define TICK_SLOW_CALL_FUN_TOO_MANY() TICK_BUMP(SLOW_CALL_FUN_TOO_MANY_ctr)
648 #define TICK_SLOW_CALL_PAP_TOO_FEW() TICK_BUMP(SLOW_CALL_PAP_TOO_FEW_ctr)
649 #define TICK_SLOW_CALL_PAP_CORRECT() TICK_BUMP(SLOW_CALL_PAP_CORRECT_ctr)
650 #define TICK_SLOW_CALL_PAP_TOO_MANY() TICK_BUMP(SLOW_CALL_PAP_TOO_MANY_ctr)
651
652 #define TICK_SLOW_CALL_fast_v16() TICK_BUMP(SLOW_CALL_fast_v16_ctr)
653 #define TICK_SLOW_CALL_fast_v() TICK_BUMP(SLOW_CALL_fast_v_ctr)
654 #define TICK_SLOW_CALL_fast_p() TICK_BUMP(SLOW_CALL_fast_p_ctr)
655 #define TICK_SLOW_CALL_fast_pv() TICK_BUMP(SLOW_CALL_fast_pv_ctr)
656 #define TICK_SLOW_CALL_fast_pp() TICK_BUMP(SLOW_CALL_fast_pp_ctr)
657 #define TICK_SLOW_CALL_fast_ppv() TICK_BUMP(SLOW_CALL_fast_ppv_ctr)
658 #define TICK_SLOW_CALL_fast_ppp() TICK_BUMP(SLOW_CALL_fast_ppp_ctr)
659 #define TICK_SLOW_CALL_fast_pppv() TICK_BUMP(SLOW_CALL_fast_pppv_ctr)
660 #define TICK_SLOW_CALL_fast_pppp() TICK_BUMP(SLOW_CALL_fast_pppp_ctr)
661 #define TICK_SLOW_CALL_fast_ppppp() TICK_BUMP(SLOW_CALL_fast_ppppp_ctr)
662 #define TICK_SLOW_CALL_fast_pppppp() TICK_BUMP(SLOW_CALL_fast_pppppp_ctr)
663 #define TICK_VERY_SLOW_CALL() TICK_BUMP(VERY_SLOW_CALL_ctr)
664
665 /* NOTE: TICK_HISTO_BY and TICK_HISTO
666 currently have no effect.
667 The old code for it didn't typecheck and I
668 just commented it out to get ticky to work.
669 - krc 1/2007 */
670
671 #define TICK_HISTO_BY(histo,n,i) /* nothing */
672
673 #define TICK_HISTO(histo,n) TICK_HISTO_BY(histo,n,1)
674
675 /* An unboxed tuple with n components. */
676 #define TICK_RET_UNBOXED_TUP(n) \
677 TICK_BUMP(RET_UNBOXED_TUP_ctr++); \
678 TICK_HISTO(RET_UNBOXED_TUP,n)
679
680 /*
681 * A slow call with n arguments. In the unevald case, this call has
682 * already been counted once, so don't count it again.
683 */
684 #define TICK_SLOW_CALL(n) \
685 TICK_BUMP(SLOW_CALL_ctr); \
686 TICK_HISTO(SLOW_CALL,n)
687
688 /*
689 * This slow call was found to be to an unevaluated function; undo the
690 * ticks we did in TICK_SLOW_CALL.
691 */
692 #define TICK_SLOW_CALL_UNEVALD(n) \
693 TICK_BUMP(SLOW_CALL_UNEVALD_ctr); \
694 TICK_BUMP_BY(SLOW_CALL_ctr,-1); \
695 TICK_HISTO_BY(SLOW_CALL,n,-1);
696
697 /* Updating a closure with a new CON */
698 #define TICK_UPD_CON_IN_NEW(n) \
699 TICK_BUMP(UPD_CON_IN_NEW_ctr); \
700 TICK_HISTO(UPD_CON_IN_NEW,n)
701
702 #define TICK_ALLOC_HEAP_NOCTR(bytes) \
703 TICK_BUMP(ALLOC_RTS_ctr); \
704 TICK_BUMP_BY(ALLOC_RTS_tot,bytes)
705
706 /* -----------------------------------------------------------------------------
707 Saving and restoring STG registers
708
709 STG registers must be saved around a C call, just in case the STG
710 register is mapped to a caller-saves machine register. Normally we
711 don't need to worry about this the code generator has already
712 loaded any live STG registers into variables for us, but in
713 hand-written low-level Cmm code where we don't know which registers
714 are live, we might have to save them all.
715 -------------------------------------------------------------------------- */
716
717 #define SAVE_STGREGS \
718 W_ r1, r2, r3, r4, r5, r6, r7, r8; \
719 F_ f1, f2, f3, f4, f5, f6; \
720 D_ d1, d2, d3, d4, d5, d6; \
721 L_ l1; \
722 \
723 r1 = R1; \
724 r2 = R2; \
725 r3 = R3; \
726 r4 = R4; \
727 r5 = R5; \
728 r6 = R6; \
729 r7 = R7; \
730 r8 = R8; \
731 \
732 f1 = F1; \
733 f2 = F2; \
734 f3 = F3; \
735 f4 = F4; \
736 f5 = F5; \
737 f6 = F6; \
738 \
739 d1 = D1; \
740 d2 = D2; \
741 d3 = D3; \
742 d4 = D4; \
743 d5 = D5; \
744 d6 = D6; \
745 \
746 l1 = L1;
747
748
749 #define RESTORE_STGREGS \
750 R1 = r1; \
751 R2 = r2; \
752 R3 = r3; \
753 R4 = r4; \
754 R5 = r5; \
755 R6 = r6; \
756 R7 = r7; \
757 R8 = r8; \
758 \
759 F1 = f1; \
760 F2 = f2; \
761 F3 = f3; \
762 F4 = f4; \
763 F5 = f5; \
764 F6 = f6; \
765 \
766 D1 = d1; \
767 D2 = d2; \
768 D3 = d3; \
769 D4 = d4; \
770 D5 = d5; \
771 D6 = d6; \
772 \
773 L1 = l1;
774
775 /* -----------------------------------------------------------------------------
776 Misc junk
777 -------------------------------------------------------------------------- */
778
779 #define NO_TREC stg_NO_TREC_closure
780 #define END_TSO_QUEUE stg_END_TSO_QUEUE_closure
781 #define STM_AWOKEN stg_STM_AWOKEN_closure
782 #define END_INVARIANT_CHECK_QUEUE stg_END_INVARIANT_CHECK_QUEUE_closure
783
784 #define recordMutableCap(p, gen) \
785 W_ __bd; \
786 W_ mut_list; \
787 mut_list = Capability_mut_lists(MyCapability()) + WDS(gen); \
788 __bd = W_[mut_list]; \
789 if (bdescr_free(__bd) >= bdescr_start(__bd) + BLOCK_SIZE) { \
790 W_ __new_bd; \
791 ("ptr" __new_bd) = foreign "C" allocBlock_lock(); \
792 bdescr_link(__new_bd) = __bd; \
793 __bd = __new_bd; \
794 W_[mut_list] = __bd; \
795 } \
796 W_ free; \
797 free = bdescr_free(__bd); \
798 W_[free] = p; \
799 bdescr_free(__bd) = free + WDS(1);
800
801 #define recordMutable(p) \
802 P_ __p; \
803 W_ __bd; \
804 W_ __gen; \
805 __p = p; \
806 __bd = Bdescr(__p); \
807 __gen = TO_W_(bdescr_gen_no(__bd)); \
808 if (__gen > 0) { recordMutableCap(__p, __gen); }
809
810 /* -----------------------------------------------------------------------------
811 Arrays
812 -------------------------------------------------------------------------- */
813
814 /* Complete function body for the clone family of (mutable) array ops.
815 Defined as a macro to avoid function call overhead or code
816 duplication. */
817 #define cloneArray(info, src, offset, n) \
818 W_ words, size; \
819 gcptr dst, dst_p, src_p; \
820 \
821 again: MAYBE_GC(again); \
822 \
823 size = n + mutArrPtrsCardWords(n); \
824 words = BYTES_TO_WDS(SIZEOF_StgMutArrPtrs) + size; \
825 ("ptr" dst) = ccall allocate(MyCapability() "ptr", words); \
826 TICK_ALLOC_PRIM(SIZEOF_StgMutArrPtrs, WDS(size), 0); \
827 \
828 SET_HDR(dst, info, CCCS); \
829 StgMutArrPtrs_ptrs(dst) = n; \
830 StgMutArrPtrs_size(dst) = size; \
831 \
832 dst_p = dst + SIZEOF_StgMutArrPtrs; \
833 src_p = src + SIZEOF_StgMutArrPtrs + WDS(offset); \
834 while: \
835 if (n != 0) { \
836 n = n - 1; \
837 W_[dst_p] = W_[src_p]; \
838 dst_p = dst_p + WDS(1); \
839 src_p = src_p + WDS(1); \
840 goto while; \
841 } \
842 \
843 return (dst);
844
845 #define copyArray(src, src_off, dst, dst_off, n) \
846 W_ dst_elems_p, dst_p, src_p, dst_cards_p, bytes; \
847 \
848 if ((n) != 0) { \
849 SET_HDR(dst, stg_MUT_ARR_PTRS_DIRTY_info, CCCS); \
850 \
851 dst_elems_p = (dst) + SIZEOF_StgMutArrPtrs; \
852 dst_p = dst_elems_p + WDS(dst_off); \
853 src_p = (src) + SIZEOF_StgMutArrPtrs + WDS(src_off); \
854 bytes = WDS(n); \
855 \
856 prim %memcpy(dst_p, src_p, bytes, SIZEOF_W); \
857 \
858 dst_cards_p = dst_elems_p + WDS(StgMutArrPtrs_ptrs(dst)); \
859 setCards(dst_cards_p, dst_off, n); \
860 } \
861 \
862 return ();
863
864 #define copyMutableArray(src, src_off, dst, dst_off, n) \
865 W_ dst_elems_p, dst_p, src_p, dst_cards_p, bytes; \
866 \
867 if ((n) != 0) { \
868 SET_HDR(dst, stg_MUT_ARR_PTRS_DIRTY_info, CCCS); \
869 \
870 dst_elems_p = (dst) + SIZEOF_StgMutArrPtrs; \
871 dst_p = dst_elems_p + WDS(dst_off); \
872 src_p = (src) + SIZEOF_StgMutArrPtrs + WDS(src_off); \
873 bytes = WDS(n); \
874 \
875 if ((src) == (dst)) { \
876 prim %memmove(dst_p, src_p, bytes, SIZEOF_W); \
877 } else { \
878 prim %memcpy(dst_p, src_p, bytes, SIZEOF_W); \
879 } \
880 \
881 dst_cards_p = dst_elems_p + WDS(StgMutArrPtrs_ptrs(dst)); \
882 setCards(dst_cards_p, dst_off, n); \
883 } \
884 \
885 return ();
886
887 /*
888 * Set the cards in the cards table pointed to by dst_cards_p for an
889 * update to n elements, starting at element dst_off.
890 */
891 #define setCards(dst_cards_p, dst_off, n) \
892 W_ __start_card, __end_card, __cards; \
893 __start_card = mutArrPtrCardDown(dst_off); \
894 __end_card = mutArrPtrCardDown((dst_off) + (n) - 1); \
895 __cards = __end_card - __start_card + 1; \
896 prim %memset((dst_cards_p) + __start_card, 1, __cards, 1);
897
898 /* Complete function body for the clone family of small (mutable)
899 array ops. Defined as a macro to avoid function call overhead or
900 code duplication. */
901 #define cloneSmallArray(info, src, offset, n) \
902 W_ words, size; \
903 gcptr dst, dst_p, src_p; \
904 \
905 again: MAYBE_GC(again); \
906 \
907 words = BYTES_TO_WDS(SIZEOF_StgSmallMutArrPtrs) + n; \
908 ("ptr" dst) = ccall allocate(MyCapability() "ptr", words); \
909 TICK_ALLOC_PRIM(SIZEOF_StgSmallMutArrPtrs, WDS(n), 0); \
910 \
911 SET_HDR(dst, info, CCCS); \
912 StgSmallMutArrPtrs_ptrs(dst) = n; \
913 \
914 dst_p = dst + SIZEOF_StgSmallMutArrPtrs; \
915 src_p = src + SIZEOF_StgSmallMutArrPtrs + WDS(offset); \
916 while: \
917 if (n != 0) { \
918 n = n - 1; \
919 W_[dst_p] = W_[src_p]; \
920 dst_p = dst_p + WDS(1); \
921 src_p = src_p + WDS(1); \
922 goto while; \
923 } \
924 \
925 return (dst);
926
927 #endif /* CMM_H */