1 /* -----------------------------------------------------------------------------
3 * (c) The University of Glasgow 2004
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.
10 * For the syntax of .cmm files, see the parser in ghc/compiler/cmm/CmmParse.y.
12 * If you're used to the old HC file syntax, here's a quick cheat sheet
13 * for converting HC code:
16 * - Remove all type casts
18 * - STGFUN(foo) { ... } ==> foo { ... }
19 * - FN_(foo) { ... } ==> foo { ... }
20 * - JMP_(e) ==> jump e;
21 * - Remove EXTFUN(foo)
24 * - Sp += n ==> Sp_adj(n)
25 * - Hp += n ==> Hp_adj(n)
26 * - R1.i ==> R1 (similarly for R1.w, R1.cl etc.)
27 * - You need to explicitly dereference variables; eg.
28 * context_switch ==> CInt[context_switch]
29 * - convert all word offsets into byte offsets:
31 * - sizeofW(StgFoo) ==> SIZEOF_StgFoo
32 * - ENTRY_CODE(e) ==> %ENTRY_CODE(e)
33 * - get_itbl(c) ==> %GET_STD_INFO(c)
34 * - Change liveness masks in STK_CHK_GEN, HP_CHK_GEN:
35 * R1_PTR | R2_PTR ==> R1_PTR & R2_PTR
37 * - Declarations like 'StgPtr p;' become just 'W_ p;'
38 * - e->payload[n] ==> PAYLOAD(e,n)
39 * - Be very careful with comparisons: the infix versions (>, >=, etc.)
40 * are unsigned, so use %lt(a,b) to get signed less-than for example.
42 * Accessing fields of structures defined in the RTS header files is
43 * done via automatically-generated macros in DerivedConstants.h. For
44 * example, where previously we used
46 * CurrentTSO->what_next = x
50 * StgTSO_what_next(CurrentTSO) = x
52 * where the StgTSO_what_next() macro is automatically generated by
53 * mkDerivedConstnants.c. If you need to access a field that doesn't
54 * already have a macro, edit that file (it's pretty self-explanatory).
56 * -------------------------------------------------------------------------- */
62 * In files that are included into both C and C-- (and perhaps
63 * Haskell) sources, we sometimes need to conditionally compile bits
64 * depending on the language. CMINUSMINUS==1 in .cmm sources:
68 #include "ghcconfig.h"
69 #include "RtsConfig.h"
71 /* -----------------------------------------------------------------------------
74 The following synonyms for C-- types are declared here:
76 I8, I16, I32, I64 MachRep-style names for convenience
78 W_ is shorthand for the word type (== StgWord)
79 F_ shorthand for float (F_ == StgFloat == C's float)
80 D_ shorthand for double (D_ == StgDouble == C's double)
82 CInt has the same size as an int in C on this platform
83 CLong has the same size as a long in C on this platform
85 --------------------------------------------------------------------------- */
92 #if SIZEOF_VOID_P == 4
94 #elif SIZEOF_VOID_P == 8
97 #error Unknown word size
102 #elif SIZEOF_INT == 8
105 #error Unknown int size
110 #elif SIZEOF_LONG == 8
113 #error Unknown long size
120 #define SIZEOF_StgDouble 8
121 #define SIZEOF_StgWord64 8
123 /* -----------------------------------------------------------------------------
125 -------------------------------------------------------------------------- */
129 #define STRING(name,str) \
131 name : bits8[] str; \
134 /* -----------------------------------------------------------------------------
137 Everything in C-- is in byte offsets (well, most things). We use
138 some macros to allow us to express offsets in words and to try to
139 avoid byte/word confusion.
140 -------------------------------------------------------------------------- */
142 #define SIZEOF_W SIZEOF_VOID_P
143 #define W_MASK (SIZEOF_W-1)
151 /* Converting quantities of words to bytes */
152 #define WDS(n) ((n)*SIZEOF_W)
155 * Converting quantities of bytes to words
156 * NB. these work on *unsigned* values only
158 #define BYTES_TO_WDS(n) ((n) / SIZEOF_W)
159 #define ROUNDUP_BYTES_TO_WDS(n) (((n) + SIZEOF_W - 1) / SIZEOF_W)
161 /* TO_W_(n) converts n to W_ type from a smaller type */
163 #define TO_W_(x) %sx32(x)
164 #define HALF_W_(x) %lobits16(x)
166 #define TO_W_(x) %sx64(x)
167 #define HALF_W_(x) %lobits32(x)
170 #if SIZEOF_INT == 4 && SIZEOF_W == 8
171 #define W_TO_INT(x) %lobits32(x)
172 #elif SIZEOF_INT == SIZEOF_W
173 #define W_TO_INT(x) (x)
176 /* -----------------------------------------------------------------------------
177 Heap/stack access, and adjusting the heap/stack pointers.
178 -------------------------------------------------------------------------- */
180 #define Sp(n) W_[Sp + WDS(n)]
181 #define Hp(n) W_[Hp + WDS(n)]
183 #define Sp_adj(n) Sp = Sp + WDS(n)
184 #define Hp_adj(n) Hp = Hp + WDS(n)
186 /* -----------------------------------------------------------------------------
187 Assertions and Debuggery
188 -------------------------------------------------------------------------- */
191 #define ASSERT(predicate) \
195 foreign "C" _assertFail(NULL, __LINE__); \
198 #define ASSERT(p) /* nothing */
202 #define DEBUG_ONLY(s) s
204 #define DEBUG_ONLY(s) /* nothing */
208 * The IF_DEBUG macro is useful for debug messages that depend on one
209 * of the RTS debug options. For example:
211 * IF_DEBUG(RtsFlags_DebugFlags_apply,
212 * foreign "C" fprintf(stderr, stg_ap_0_ret_str));
214 * Note the syntax is slightly different to the C version of this macro.
217 #define IF_DEBUG(c,s) if (RtsFlags_DebugFlags_##c(RtsFlags) != 0::I32) { s; }
219 #define IF_DEBUG(c,s) /* nothing */
222 /* -----------------------------------------------------------------------------
225 It isn't safe to "enter" every closure. Functions in particular
226 have no entry code as such; their entry point contains the code to
229 ToDo: range should end in N_CLOSURE_TYPES-1, not N_CLOSURE_TYPES,
230 but switch doesn't allow us to use exprs there yet.
231 -------------------------------------------------------------------------- */
236 info = %INFO_PTR(R1); \
237 switch [INVALID_OBJECT .. N_CLOSURE_TYPES] \
238 (TO_W_( %INFO_TYPE(%STD_INFO(info)) )) { \
246 R1 = StgInd_indirectee(R1); \
260 jump %ENTRY_CODE(Sp(0)); \
264 jump %ENTRY_CODE(info); \
268 /* -----------------------------------------------------------------------------
270 -------------------------------------------------------------------------- */
272 #include "Constants.h"
273 #include "DerivedConstants.h"
274 #include "ClosureTypes.h"
276 #include "OSThreads.h"
280 * Need MachRegs, because some of the RTS code is conditionally
281 * compiled based on REG_R1, REG_R2, etc.
283 #define STOLEN_X86_REGS 4
284 #include "MachRegs.h"
286 #include "Liveness.h"
287 #include "StgLdvProf.h"
291 #include "Block.h" /* For Bdescr() */
294 /* Can't think of a better place to put this. */
295 #if SIZEOF_mp_limb_t != SIZEOF_VOID_P
296 #error mp_limb_t != StgWord: assumptions in PrimOps.cmm are now false
299 #define MyCapability() (BaseReg - OFFSET_Capability_r)
301 /* -------------------------------------------------------------------------
302 Allocation and garbage collection
303 ------------------------------------------------------------------------- */
306 * ALLOC_PRIM is for allocating memory on the heap for a primitive
307 * object. It is used all over PrimOps.cmm.
309 * We make the simplifying assumption that the "admin" part of a
310 * primitive closure is just the header when calculating sizes for
311 * ticky-ticky. It's not clear whether eg. the size field of an array
312 * should be counted as "admin", or the various fields of a BCO.
314 #define ALLOC_PRIM(bytes,liveness,reentry) \
315 HP_CHK_GEN_TICKY(bytes,liveness,reentry); \
316 TICK_ALLOC_PRIM(SIZEOF_StgHeader,bytes-SIZEOF_StgHeader,0); \
319 /* CCS_ALLOC wants the size in words, because ccs->mem_alloc is in words */
320 #define CCCS_ALLOC(__alloc) CCS_ALLOC(BYTES_TO_WDS(__alloc), W_[CCCS])
322 #define HP_CHK_GEN_TICKY(alloc,liveness,reentry) \
323 HP_CHK_GEN(alloc,liveness,reentry); \
324 TICK_ALLOC_HEAP_NOCTR(alloc);
326 // allocateLocal() allocates from the nursery, so we check to see
327 // whether the nursery is nearly empty in any function that uses
328 // allocateLocal() - this includes many of the primops.
329 #define MAYBE_GC(liveness,reentry) \
330 if (bdescr_link(CurrentNursery) == NULL || CInt[alloc_blocks] >= CInt[alloc_blocks_lim]) { \
334 jump stg_gc_gen_hp; \
337 /* -----------------------------------------------------------------------------
339 -------------------------------------------------------------------------- */
342 * This is really ugly, since we don't do the rest of StgHeader this
343 * way. The problem is that values from DerivedConstants.h cannot be
344 * dependent on the way (SMP, PROF etc.). For SIZEOF_StgHeader we get
345 * the value from GHC, but it seems like too much trouble to do that
346 * for StgThunkHeader.
348 #define SIZEOF_StgThunkHeader SIZEOF_StgHeader+SIZEOF_StgSMPThunkHeader
350 #define StgThunk_payload(__ptr__,__ix__) \
351 W_[__ptr__+SIZEOF_StgThunkHeader+ WDS(__ix__)]
353 /* -----------------------------------------------------------------------------
355 -------------------------------------------------------------------------- */
357 /* The offset of the payload of an array */
358 #define BYTE_ARR_CTS(arr) ((arr) + SIZEOF_StgArrWords)
360 /* Getting/setting the info pointer of a closure */
361 #define SET_INFO(p,info) StgHeader_info(p) = info
362 #define GET_INFO(p) StgHeader_info(p)
364 /* Determine the size of an ordinary closure from its info table */
365 #define sizeW_fromITBL(itbl) \
366 SIZEOF_StgHeader + WDS(%INFO_PTRS(itbl)) + WDS(%INFO_NPTRS(itbl))
368 /* NB. duplicated from InfoTables.h! */
369 #define BITMAP_SIZE(bitmap) ((bitmap) & BITMAP_SIZE_MASK)
370 #define BITMAP_BITS(bitmap) ((bitmap) >> BITMAP_BITS_SHIFT)
372 /* Debugging macros */
373 #define LOOKS_LIKE_INFO_PTR(p) \
375 (TO_W_(%INFO_TYPE(%STD_INFO(p))) != INVALID_OBJECT) && \
376 (TO_W_(%INFO_TYPE(%STD_INFO(p))) < N_CLOSURE_TYPES))
378 #define LOOKS_LIKE_CLOSURE_PTR(p) (LOOKS_LIKE_INFO_PTR(GET_INFO(p)))
381 * The layout of the StgFunInfoExtra part of an info table changes
382 * depending on TABLES_NEXT_TO_CODE. So we define field access
383 * macros which use the appropriate version here:
385 #ifdef TABLES_NEXT_TO_CODE
387 * when TABLES_NEXT_TO_CODE, slow_apply is stored as an offset
388 * instead of the normal pointer.
391 #define StgFunInfoExtra_slow_apply(fun_info) \
392 (TO_W_(StgFunInfoExtraRev_slow_apply_offset(fun_info)) \
393 + (fun_info) + SIZEOF_StgFunInfoExtraRev + SIZEOF_StgInfoTable)
395 #define StgFunInfoExtra_fun_type(i) StgFunInfoExtraRev_fun_type(i)
396 #define StgFunInfoExtra_arity(i) StgFunInfoExtraRev_arity(i)
397 #define StgFunInfoExtra_bitmap(i) StgFunInfoExtraRev_bitmap(i)
399 #define StgFunInfoExtra_slow_apply(i) StgFunInfoExtraFwd_slow_apply(i)
400 #define StgFunInfoExtra_fun_type(i) StgFunInfoExtraFwd_fun_type(i)
401 #define StgFunInfoExtra_arity(i) StgFunInfoExtraFwd_arity(i)
402 #define StgFunInfoExtra_bitmap(i) StgFunInfoExtraFwd_bitmap(i)
405 /* -----------------------------------------------------------------------------
406 Voluntary Yields/Blocks
408 We only have a generic version of this at the moment - if it turns
409 out to be slowing us down we can make specialised ones.
410 -------------------------------------------------------------------------- */
412 #define YIELD(liveness,reentry) \
417 #define BLOCK(liveness,reentry) \
422 /* -----------------------------------------------------------------------------
424 -------------------------------------------------------------------------- */
427 #define TICK_BUMP_BY(ctr,n) CLong[ctr] = CLong[ctr] + n
429 #define TICK_BUMP_BY(ctr,n) /* nothing */
432 #define TICK_BUMP(ctr) TICK_BUMP_BY(ctr,1)
434 #define TICK_ENT_DYN_IND() TICK_BUMP(ENT_DYN_IND_ctr)
435 #define TICK_ENT_DYN_THK() TICK_BUMP(ENT_DYN_THK_ctr)
436 #define TICK_ENT_VIA_NODE() TICK_BUMP(ENT_VIA_NODE_ctr)
437 #define TICK_ENT_STATIC_IND() TICK_BUMP(ENT_STATIC_IND_ctr)
438 #define TICK_ENT_PERM_IND() TICK_BUMP(ENT_PERM_IND_ctr)
439 #define TICK_ENT_PAP() TICK_BUMP(ENT_PAP_ctr)
440 #define TICK_ENT_AP() TICK_BUMP(ENT_AP_ctr)
441 #define TICK_ENT_AP_STACK() TICK_BUMP(ENT_AP_STACK_ctr)
442 #define TICK_ENT_BH() TICK_BUMP(ENT_BH_ctr)
443 #define TICK_UNKNOWN_CALL() TICK_BUMP(UNKNOWN_CALL_ctr)
444 #define TICK_UPDF_PUSHED() TICK_BUMP(UPDF_PUSHED_ctr)
445 #define TICK_CATCHF_PUSHED() TICK_BUMP(CATCHF_PUSHED_ctr)
446 #define TICK_UPDF_OMITTED() TICK_BUMP(UPDF_OMITTED_ctr)
447 #define TICK_UPD_NEW_IND() TICK_BUMP(UPD_NEW_IND_ctr)
448 #define TICK_UPD_NEW_PERM_IND() TICK_BUMP(UPD_NEW_PERM_IND_ctr)
449 #define TICK_UPD_OLD_IND() TICK_BUMP(UPD_OLD_IND_ctr)
450 #define TICK_UPD_OLD_PERM_IND() TICK_BUMP(UPD_OLD_PERM_IND_ctr)
452 #define TICK_SLOW_CALL_FUN_TOO_FEW() TICK_BUMP(SLOW_CALL_FUN_TOO_FEW_ctr)
453 #define TICK_SLOW_CALL_FUN_CORRECT() TICK_BUMP(SLOW_CALL_FUN_CORRECT_ctr)
454 #define TICK_SLOW_CALL_FUN_TOO_MANY() TICK_BUMP(SLOW_CALL_FUN_TOO_MANY_ctr)
455 #define TICK_SLOW_CALL_PAP_TOO_FEW() TICK_BUMP(SLOW_CALL_PAP_TOO_FEW_ctr)
456 #define TICK_SLOW_CALL_PAP_CORRECT() TICK_BUMP(SLOW_CALL_PAP_CORRECT_ctr)
457 #define TICK_SLOW_CALL_PAP_TOO_MANY() TICK_BUMP(SLOW_CALL_PAP_TOO_MANY_ctr)
459 #define TICK_SLOW_CALL_v() TICK_BUMP(SLOW_CALL_v_ctr)
460 #define TICK_SLOW_CALL_p() TICK_BUMP(SLOW_CALL_p_ctr)
461 #define TICK_SLOW_CALL_pv() TICK_BUMP(SLOW_CALL_pv_ctr)
462 #define TICK_SLOW_CALL_pp() TICK_BUMP(SLOW_CALL_pp_ctr)
463 #define TICK_SLOW_CALL_ppp() TICK_BUMP(SLOW_CALL_ppp_ctr)
464 #define TICK_SLOW_CALL_pppp() TICK_BUMP(SLOW_CALL_pppp_ctr)
465 #define TICK_SLOW_CALL_ppppp() TICK_BUMP(SLOW_CALL_ppppp_ctr)
466 #define TICK_SLOW_CALL_pppppp() TICK_BUMP(SLOW_CALL_pppppp_ctr)
468 /* NOTE: TICK_HISTO_BY and TICK_HISTO
469 currently have no effect.
470 The old code for it didn't typecheck and I
471 just commented it out to get ticky to work.
474 #define TICK_HISTO_BY(histo,n,i) /* nothing */
476 #define TICK_HISTO(histo,n) TICK_HISTO_BY(histo,n,1)
478 /* An unboxed tuple with n components. */
479 #define TICK_RET_UNBOXED_TUP(n) \
480 TICK_BUMP(RET_UNBOXED_TUP_ctr++); \
481 TICK_HISTO(RET_UNBOXED_TUP,n)
484 * A slow call with n arguments. In the unevald case, this call has
485 * already been counted once, so don't count it again.
487 #define TICK_SLOW_CALL(n) \
488 TICK_BUMP(SLOW_CALL_ctr); \
489 TICK_HISTO(SLOW_CALL,n)
492 * This slow call was found to be to an unevaluated function; undo the
493 * ticks we did in TICK_SLOW_CALL.
495 #define TICK_SLOW_CALL_UNEVALD(n) \
496 TICK_BUMP(SLOW_CALL_UNEVALD_ctr); \
497 TICK_BUMP_BY(SLOW_CALL_ctr,-1); \
498 TICK_HISTO_BY(SLOW_CALL,n,-1);
500 /* Updating a closure with a new CON */
501 #define TICK_UPD_CON_IN_NEW(n) \
502 TICK_BUMP(UPD_CON_IN_NEW_ctr); \
503 TICK_HISTO(UPD_CON_IN_NEW,n)
505 #define TICK_ALLOC_HEAP_NOCTR(n) \
506 TICK_BUMP(ALLOC_HEAP_ctr); \
507 TICK_BUMP_BY(ALLOC_HEAP_tot,n)
509 /* -----------------------------------------------------------------------------
511 -------------------------------------------------------------------------- */
513 #define NO_TREC stg_NO_TREC_closure
514 #define END_TSO_QUEUE stg_END_TSO_QUEUE_closure
515 #define END_INVARIANT_CHECK_QUEUE stg_END_INVARIANT_CHECK_QUEUE_closure
517 #define dirtyTSO(tso) \
518 StgTSO_flags(tso) = StgTSO_flags(tso) | TSO_DIRTY::I32;
520 #define recordMutableCap(p, gen, regs) \
523 mut_list = Capability_mut_lists(MyCapability()) + WDS(gen); \
524 __bd = W_[mut_list]; \
525 if (bdescr_free(__bd) >= bdescr_start(__bd) + BLOCK_SIZE) { \
527 "ptr" __new_bd = foreign "C" allocBlock_lock() [regs]; \
528 bdescr_link(__new_bd) = __bd; \
530 W_[mut_list] = __bd; \
533 free = bdescr_free(__bd); \
535 bdescr_free(__bd) = free + WDS(1);
537 #define recordMutable(p, regs) \
540 recordMutableCap(__p, TO_W_(bdescr_gen_no(Bdescr(__p))), regs)