2 % (c) Kevin Hammond, Parade/AQUA Projects, Glasgow University, February 15th. 1995
4 % This is for GUM only.
6 %************************************************************************
8 \section[Parallel.lh]{Definitions for parallel machines}
10 %************************************************************************
12 Multi-slurp protection:
18 This section contains definitions applicable only to programs compiled
19 to run on a parallel machine. Some of these things can probably be
20 ripped out now that we don't store GAs in closures, but beware that
21 this {\em might} break GranSim, so check first! KH
23 These basic definitions need to be around, one way or the other:
25 #define PAR_FIXED_HDR (GA_HDR_SIZE)
26 #define PAR_HDR_POSN AFTER_INFO_HDR
27 #define AFTER_PAR_HDR (PAR_HDR_POSN+PAR_FIXED_HDR)
29 #define SET_PAR_HDR(closure,ga) /* nothing */
30 #define SET_STATIC_PAR_HDR(closure) /* nothing */
35 # define MAX_PES 256 /* Maximum number of processors */
36 /* MAX_PES is enforced by SysMan, which does not
37 allow more than this many "processors".
38 This is important because PackGA [GlobAddr.lc]
39 **assumes** that a PE# can fit in 8+ bits.
42 extern I_ do_sp_profile;
44 extern P_ PendingFetches;
45 extern GLOBAL_TASK_ID *PEs;
47 extern rtsBool IAmMainThread, GlobalStopPending;
48 extern rtsBool fishing;
49 extern GLOBAL_TASK_ID SysManTask;
50 extern int seed; /*pseudo-random-number generator seed:*/
51 /*Initialised in ParInit*/
52 extern I_ threadId; /*Number of Threads that have existed on a PE*/
53 extern GLOBAL_TASK_ID mytid;
57 extern rtsBool InGlobalGC; /* Are we in the midst of performing global GC */
59 extern HashTable *pGAtoGALAtable;
60 extern HashTable *LAtoGALAtable;
61 extern GALA *freeIndirections;
62 extern GALA *liveIndirections;
63 extern GALA *freeGALAList;
64 extern GALA *liveRemoteGAs;
67 void RunParallelSystem PROTO((StgPtr program_closure));
68 void initParallelSystem(STG_NO_ARGS);
69 void SynchroniseSystem(STG_NO_ARGS);
71 void registerTask PROTO((GLOBAL_TASK_ID gtid));
72 globalAddr *LAGAlookup PROTO((P_ addr));
73 P_ GALAlookup PROTO((globalAddr *ga));
74 globalAddr *MakeGlobal PROTO((P_ addr, rtsBool preferred));
75 globalAddr *setRemoteGA PROTO((P_ addr, globalAddr *ga, rtsBool preferred));
76 void splitWeight PROTO((globalAddr *to, globalAddr *from));
77 globalAddr *addWeight PROTO((globalAddr *ga));
78 void initGAtables(STG_NO_ARGS);
79 W_ taskIDtoPE PROTO((GLOBAL_TASK_ID gtid));
80 void RebuildLAGAtable(STG_NO_ARGS);
82 void *lookupHashTable PROTO((HashTable *table, StgWord key));
83 void insertHashTable PROTO((HashTable *table, StgWord key, void *data));
84 void freeHashTable PROTO((HashTable *table, void (*freeDataFun) PROTO((void *data))));
85 HashTable *allocHashTable(STG_NO_ARGS);
86 void *removeHashTable PROTO((HashTable *table, StgWord key, void *data));
88 extern void myexit PROTO((I_));
90 # define exit : error : Wrong exit!
96 %************************************************************************
98 \subsection[anti-parallel-SM]{But if we're {\em not} compiling for a parallel system...}
100 %************************************************************************
102 Get this out of the way. These are all null definitions.
105 # if !(defined(GRAN) || defined(PAR))
107 # define GA_HDR_SIZE 0
108 # define GA(closure) /*nothing*/
110 # define SET_GA(closure,ga) /* nothing */
111 # define SET_STATIC_GA(closure) /* nothing */
112 # define SET_GRAN_HDR(closure,pe) /* nothing */
113 # define SET_STATIC_PROCS(closure) /* nothing */
115 # define SET_TASK_ACTIVITY(act) /* nothing */
121 # define GA_HDR_SIZE 1
123 # define PROCS_HDR_POSN PAR_HDR_POSN
124 # define PROCS_HDR_SIZE 1
126 /* Accessing components of the field */
127 # define PROCS(closure) (*((P_)(closure)+PROCS_HDR_POSN))
129 # define SET_PROCS(closure, procs) \
130 PROCS(closure) = (W_)(procs) /* Set closure's location */
131 # define SET_GRAN_HDR(closure,pe) SET_PROCS(closure,pe)
133 # define SET_STATIC_PROCS(closure) , (W_) (Everywhere)
135 # define SET_TASK_ACTIVITY(act) /* nothing */
139 %************************************************************************
141 \subsection[parallel-GAs]{Parallel-only part of fixed headers (global addresses)}
143 %************************************************************************
145 Definitions relating to the entire parallel-only fixed-header field.
147 On GUM, the global addresses for each local closure are stored in a separate
148 hash table, rather then with the closure in the heap. We call @getGA@ to
149 look up the global address associated with a local closure (0 is returned
150 for local closures that have no global address), and @setGA@ to store a new
151 global address for a local closure which did not previously have one.
156 # define GA_HDR_SIZE 0
158 # define GA(closure) getGA(closure)
160 # define SET_GA(closure, ga) setGA(closure,ga)
161 # define SET_STATIC_GA(closure)
162 # define SET_GRAN_HDR(closure,pe)
163 # define SET_STATIC_PROCS(closure)
165 # define MAX_GA_WEIGHT 0 /* Treat as 2^n */
167 W_ PackGA PROTO((W_, int));
168 /* There was a PACK_GA macro here; but we turned it into the PackGA
169 routine [GlobAddr.lc] (because it needs to do quite a bit of
170 paranoia checking. Phil & Will (95/08)
174 At the moment, there is no activity profiling for GUM. This may change.
178 # define SET_TASK_ACTIVITY(act)
182 %************************************************************************
184 \subsection[parallel-heap-objs]{Special parallel-only heap objects (`closures')}
186 %************************************************************************
188 % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
189 % NB: The following definitons are BOTH for GUM and GrAnSim -- HWL
190 % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
192 The rest of this file contains definitions for {\it GUM and GrAnSim}.
193 Although we don't create FetchMe nodes in GrAnSim (we simulate it by
194 bitmask twiddling) we use FetchMe_info when converting nodes into RBHs
195 (mainly to keep the code as close to GUM as possible). So, we define all
196 the FetchMe related stuff in GrAnSim, too. % -- HWL
198 %************************************************************************
200 \subsubsection[FETCHME-closures]{@FETCHME@ heap objects (`closures')}
202 %************************************************************************
204 FetchMes are pointers into the global heap. When evaluated, the value
205 they point to is read from the global heap.
207 A FetchMe closure has the form:
212 \begin{tabular}{||l|l||}\hline
213 \tr{FETCHME_info} & junk \\ \hline
219 The argument word is a pointer (outside of the heap) to a globalAddr structure...
220 in particular, the one corresponding to the object to be fetched. Note that
221 we can't just used the LAGA table, because weight-splitting may force us to
222 reassign a local GA to the @FetchMe@ so that we can give out new references.
224 A @FetchMe@ must have a valid @MUT_LINK@ field, because it may act as
225 a transition between an RBH on the OldMutables list and a BQ still on
226 the OldMutables list.
230 # define FETCHME_VHS IND_VHS
231 # define FETCHME_HS IND_HS
233 # define FETCHME_GA_LOCN FETCHME_HS
235 # define FETCHME_CLOSURE_SIZE(closure) IND_CLOSURE_SIZE(closure)
236 # define FETCHME_CLOSURE_NoPTRS(closure) 0L
237 # define FETCHME_CLOSURE_NoNONPTRS(closure) (IND_CLOSURE_SIZE(closure)-IND_VHS)
239 # define SET_FETCHME_HDR(closure,infolbl,cc,size,ptrs) \
240 { SET_FIXED_HDR(closure,FetchMe_info,<bogus CC>); \
241 SET_MUT_RESERVED_WORDS(closure); \
244 # define FETCHME_GA(closure) (((globalAddr **)(closure))[FETCHME_GA_LOCN])
246 EXTFUN(FetchMe_entry);
247 EXTDATA_RO(FetchMe_info);
251 %************************************************************************
253 \subsubsection[BlockedFetch-closures]{@BlockedFetch@ heap objects (`closures')}
255 %************************************************************************
257 @BlockedFetch@s are inbound fetch messages blocked on local closures.
258 They arise as entries in a local blocking queue when a fetch has been
259 received for a local black hole. When awakened, we look at their
260 contents to figure out where to send a resume.
262 A @BlockedFetch@ closure has the form:
267 \begin{tabular}{||l|l|l||}\hline
268 \tr{BF_info} & link & node & gtid & slot & weight \\ \hline
275 # define BF_VHS (GC_MUT_RESERVED_WORDS)
276 # define BF_HS (FIXED_HS + BF_VHS)
278 # define BF_LINK_LOCN (BF_HS)
279 # define BF_NODE_LOCN (BF_HS + 1)
280 # define BF_GTID_LOCN (BF_HS + 2)
281 # define BF_SLOT_LOCN (BF_HS + 3)
282 # define BF_WEIGHT_LOCN (BF_HS + 4)
284 # define BF_CLOSURE_NoPTRS(closure) 2
285 # define BF_CLOSURE_NoNONPTRS(closure) 3
287 # define BF_CLOSURE_SIZE(closure) (BF_VHS + 5)
289 # define BF_LINK(closure) (((PP_)closure)[BF_LINK_LOCN])
290 # define BF_NODE(closure) (((PP_)closure)[BF_NODE_LOCN])
291 # define BF_GTID(closure) (((P_)closure)[BF_GTID_LOCN])
292 # define BF_SLOT(closure) (((P_)closure)[BF_SLOT_LOCN])
293 # define BF_WEIGHT(closure) (((P_)closure)[BF_WEIGHT_LOCN])
295 # define BF_CLOSURE_PTR(closure, no) (((P_)(closure))[BF_HS + (no) - 1])
297 /* std start-filling-in macro: */
298 # define SET_BF_HDR(closure,infolbl,cc) \
299 { SET_FIXED_HDR(closure,infolbl,cc); \
300 SET_MUT_RESERVED_WORDS(closure); \
308 %************************************************************************
310 \subsubsection[FMBQ-closures]{@FMBQ@ (FetchMe with blocking queue) heap objects (`closures')}
312 %************************************************************************
314 FetchMe's with blocking queues are @Fetchme@ nodes which have been entered
315 (and therefore a fetch has been sent), but for which we have not yet received
316 a @Resume@ message. They look just like normal blocking queues, but have
317 a distinguished info pointer.
320 # define FMBQ_VHS BQ_VHS
321 # define FMBQ_HS BQ_HS
323 # define FMBQ_CLOSURE_SIZE(closure) BQ_CLOSURE_SIZE(closure)
324 # define FMBQ_CLOSURE_NoPTRS(closure) BQ_CLOSURE_NoPTRS(closure)
325 # define FMBQ_CLOSURE_NoNONPTRS(closure) BQ_CLOSURE_NoNONPTRS(closure)
326 # define FMBQ_CLOSURE_PTR(closure, no) BQ_CLOSURE_PTR(closure, no)
328 # define FMBQ_ENTRIES(closure) BQ_ENTRIES(closure)
329 # define FMBQ_LINK(closure) BQ_LINK(closure)
332 EXTDATA_RO(FMBQ_info);
335 %************************************************************************
337 \subsection[parallel-info-tables]{Special parallel-only info-table stuff}
339 %************************************************************************
341 %************************************************************************
343 \subsubsection[FETCHME_ITBL]{@FETCHME_ITBL@}
345 %************************************************************************
347 ToDo: delete FetchMe CAT (because we don't profile and parallelize at
348 the same time...) Even better...set things up so that we can profile
349 and parallelize at the same time!
353 # define FETCHME_ITBL(itbl_name,entry_code) \
354 CAT_DECLARE(FetchMe,INTERNAL_KIND,"FetchMe","<FetchMe>") \
355 EXTFUN(entry_code); \
356 EXTDATA_RO(MK_REP_LBL(FetchMe,,)); \
357 const W_ itbl_name[] = { \
359 ,(W_) INFO_OTHER_TAG \
360 ,(W_) MK_REP_REF(FetchMe,,) \
361 INCLUDE_PROFILING_INFO(FetchMe) \
364 # define FETCHME_RTBL() \
365 const W_ MK_REP_LBL(FetchMe,,)[] = { \
366 INCLUDE_TYPE_INFO(FETCHME) \
367 INCLUDE_SIZE_INFO(MIN_UPD_SIZE, 0L) \
369 INCLUDE_COPYING_INFO(_Evacuate_FetchMe,_Scavenge_FetchMe) \
370 INCLUDE_COMPACTING_INFO(_ScanLink_FetchMe,_PRStart_FetchMe,_ScanMove_FetchMe,_PRIn_Error) \
375 %************************************************************************
377 \subsubsection[BF_ITBL]{@BF_ITBL@}
379 %************************************************************************
381 The special info table used for thread state objects (BlockedFetchs).
386 CAT_DECLARE(BF,INTERNAL_KIND,"BlockedFetch","<BlockedFetch>") \
388 EXTDATA_RO(MK_REP_LBL(BF,,)); \
389 const W_ BF_info[] = { \
391 ,(W_) INFO_OTHER_TAG \
392 ,(W_) MK_REP_REF(BF,,) \
393 INCLUDE_PROFILING_INFO(BF) \
397 const W_ MK_REP_LBL(BF,,)[] = { \
398 INCLUDE_TYPE_INFO(BF) \
399 INCLUDE_SIZE_INFO(BF_CLOSURE_SIZE(dummy),BF_CLOSURE_NoPTRS(dummy)) \
401 INCLUDE_COPYING_INFO(_Evacuate_BF,_Scavenge_BF) \
402 INCLUDE_COMPACTING_INFO(_ScanLink_BF,_PRStart_BF,_ScanMove_BF,_PRIn_BF) \
407 %************************************************************************
409 \subsubsection[FMBQ_ITBL]{@FMBQ_ITBL@}
411 %************************************************************************
413 Special info-table for local blocking queues.
416 # define FMBQ_ITBL() \
417 CAT_DECLARE(FMBQ,INTERNAL_KIND,"FMBQ","<FMBQ>") \
418 EXTFUN(FMBQ_entry); \
419 EXTDATA_RO(MK_REP_LBL(FMBQ,,)); \
420 const W_ FMBQ_info[] = { \
422 ,(W_) INFO_OTHER_TAG \
423 ,(W_) MK_REP_REF(FMBQ,,) \
424 INCLUDE_PROFILING_INFO(FMBQ) \
427 # define FMBQ_RTBL() \
428 const W_ MK_REP_LBL(FMBQ,,)[] = { \
429 INCLUDE_TYPE_INFO(FMBQ) \
430 INCLUDE_SIZE_INFO(MIN_UPD_SIZE,INFO_UNUSED) \
432 INCLUDE_COPYING_INFO(_Evacuate_BQ,_Scavenge_BQ) \
433 SPEC_COMPACTING_INFO(_ScanLink_BQ,_PRStart_BQ,_ScanMove_BQ,_PRIn_BQ) \
438 %************************************************************************
440 \subsection[parallel-spark-pool-defs]{Parallel-only Spark pool definitions}
442 %************************************************************************
445 # define HAVE_SPARK ((PendingSparksHd[REQUIRED_POOL] < PendingSparksTl[REQUIRED_POOL]) \
446 || (PendingSparksHd[ADVISORY_POOL] < PendingSparksTl[ADVISORY_POOL]))
448 # define HAVE_WORK (RUNNING_THREAD || HAVE_SPARK)
449 # define RUNNING_THREAD (CurrentTSO != NULL)
452 %************************************************************************
454 \subsection[parallel-pack-defs]{Parallel-only Packing definitions}
456 %************************************************************************
459 Symbolic constants for the packing code.
461 This constant defines how many words of data we can pack into a single
462 packet in the parallel (GUM) system.
465 void InitPackBuffer(STG_NO_ARGS);
466 P_ PackTSO PROTO((P_ tso, W_ *size));
467 P_ PackStkO PROTO((P_ stko, W_ *size));
468 P_ AllocateHeap PROTO((W_ size)); /* Doesn't belong */
470 void InitClosureQueue (STG_NO_ARGS);
471 P_ DeQueueClosure(STG_NO_ARGS);
472 void QueueClosure PROTO((P_ closure));
473 rtsBool QueueEmpty(STG_NO_ARGS);
474 void PrintPacket PROTO((P_ buffer));
476 P_ get_closure_info PROTO((P_ closure, W_ *size, W_ *ptrs, W_ *nonptrs, W_ *vhs, char *type));
478 rtsBool isOffset PROTO((globalAddr *ga)),
479 isFixed PROTO((globalAddr *ga));
481 void doGlobalGC(STG_NO_ARGS);
483 P_ PackNearbyGraph PROTO((P_ closure,W_ *size));
484 P_ UnpackGraph PROTO((W_ *buffer, globalAddr **gamap, W_ *nGAs));
488 # define PACK_HEAP_REQUIRED \
489 ((RTSflags.ParFlags.packBufferSize - PACK_HDR_SIZE) / (PACK_GA_SIZE + _FHS) * (SPEC_HS + 2))
491 extern W_ *PackBuffer; /* size: can be set via option */
492 extern long *buffer; /* HWL_ */
493 extern W_ *freeBuffer; /* HWL_ */
494 extern W_ *packBuffer; /* HWL_ */
496 extern void InitPackBuffer(STG_NO_ARGS);
497 extern void InitMoreBuffers(STG_NO_ARGS);
498 extern void InitPendingGABuffer(W_ size);
499 extern void AllocClosureQueue(W_ size);
501 # define MAX_GAS (RTSflags.ParFlags.packBufferSize / PACK_GA_SIZE)
504 # define PACK_GA_SIZE 3 /* Size of a packed GA in words */
505 /* Size of a packed fetch-me in words */
506 # define PACK_FETCHME_SIZE (PACK_GA_SIZE + FIXED_HS)
508 # define PACK_HDR_SIZE 1 /* Words of header in a packet */
510 # define PACK_PLC_SIZE 2 /* Size of a packed PLC in words */
517 /* ToDo: Check which of the PAR routines are needed in GranSim -- HWL */
518 void InitPackBuffer(STG_NO_ARGS);
519 P_ AllocateHeap PROTO((W_ size)); /* Doesn't belong */
520 P_ PackNearbyGraph PROTO((P_ closure, P_ tso, W_ *packbuffersize));
521 P_ PackOneNode PROTO((P_ closure, P_ tso, W_ *packbuffersize));
522 P_ UnpackGraph PROTO((P_ buffer));
524 void InitClosureQueue (STG_NO_ARGS);
525 P_ DeQueueClosure(STG_NO_ARGS);
526 void QueueClosure PROTO((P_ closure));
527 rtsBool QueueEmpty(STG_NO_ARGS);
528 void PrintPacket PROTO((P_ buffer));
530 P_ get_closure_info PROTO((P_ closure, W_ *size, W_ *ptrs, W_ *nonptrs, W_ *vhs, char *type));
532 /* These are needed in the packing code to get the size of the packet
533 right. The closures itself are never built in GrAnSim. */
534 # define FETCHME_VHS IND_VHS
535 # define FETCHME_HS IND_HS
537 # define FETCHME_GA_LOCN FETCHME_HS
539 # define FETCHME_CLOSURE_SIZE(closure) IND_CLOSURE_SIZE(closure)
540 # define FETCHME_CLOSURE_NoPTRS(closure) 0L
541 # define FETCHME_CLOSURE_NoNONPTRS(closure) (IND_CLOSURE_SIZE(closure)-IND_VHS)
543 # define MAX_GAS (RTSflags.GranFlags.packBufferSize / PACK_GA_SIZE)
544 # define PACK_GA_SIZE 3 /* Size of a packed GA in words */
545 /* Size of a packed fetch-me in words */
546 # define PACK_FETCHME_SIZE (PACK_GA_SIZE + FIXED_HS)
547 # define PACK_HDR_SIZE 4 /* Words of header in a packet */
549 # define PACK_HEAP_REQUIRED \
550 ((RTSflags.GranFlags.packBufferSize - PACK_HDR_SIZE) / (PACK_GA_SIZE + _FHS) * (SPEC_HS + 2))
552 # define PACK_FLAG_LOCN 0
553 # define PACK_TSO_LOCN 1
554 # define PACK_UNPACKED_SIZE_LOCN 2
555 # define PACK_SIZE_LOCN 3
556 # define MAGIC_PACK_FLAG 0xfabc
559 #endif /* Parallel_H */