AMPI: MPI_Waitall over an MPI_Ialltoall request hangs; alltoall messages unordered; four more request-completion items

Aperta
  • Qualcosa non va?
#4,097 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
35/100
Tipo di issue
Bug
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
cpp
Ambito
distributed-systems, hpc

Direzione di ricerca

Start with the AMPI request completion and collective matching code, especially ATAReq, RednReq::receive, ScanReq, freeOp, and putOutOfOrder; compare the related fixes in #4084, #4085, and #4093. Reproduce the MPI_Ialltoall plus MPI_Waitall hang and inspect existing tests that use separate MPI_Wait calls. Done means resolving the five listed completion, ordering, operation lifetime, and sequence-wrap issues with regression coverage.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Split out of #4072 (found 2026-10-23/24 while fixing items 7 and the scan family; all pre-existing on recharm, none fixed yet).

  1. MPI_Waitall over an MPI_Ialltoall request hangs, even alone (direct reproducer 2026-10-24, code unchanged from origin/recharm). ATAReq sets its own blocked flag, but its sub-requests never decrement numBlockedReqs, so the waiting rank is never woken. Tests use separate MPI_Wait calls meanwhile. This is the most user-visible item here: Ialltoall followed by Waitall is a common pattern.
  2. Alltoall-family messages carry sequence number 0 (unordered), so two outstanding Ialltoalls on one communicator have the ordering exposure the scans had before #4085 gave scan messages per-sender sequence numbers.
  3. Two outstanding Ibcasts with the same root are matched on (MPI_BCAST_TAG, root) in posting order. Since #4093 array broadcasts are delivered in order on reconverse, so this is a robustness item; a sequence number per broadcast, as #4084 did for reductions, would close it.
  4. MPI_Op_free while a non-blocking reduction is outstanding: RednReq::receive resolves the op index per round and freeOp pops the entry, so a later round asserts or calls a reused slot. MPI-3.1 says pending operations complete normally. #4085 captures the function in ScanReq; RednReq (non-commutative Ireduce/Iallreduce) still looks it up.
  5. putOutOfOrder's "received late out-of-order message" check does not handle 16-bit sequence wraparound, so an error-checking build can abort after 65536 messages from one sender.

Related: #4084 (collective matching), #4085 (scan family), #4093 (broadcast ordering and storage).

Lingua principale
C++
Stelle
241
Fork
62
Merge medio
1g 1h
PR unite (30g)
121

Preparare l'ambiente

Questo progetto non fornisce container di sviluppo, Dockerfile né guida per i contributori, quindi l'ambiente è a tuo carico: parti dal suo README e consulta la nostra guida al primo contributo per i passaggi generali.

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di charmplusplus/charm

Tutte le issue di charmplusplus/charm

Issue simili

Altre issue su C++

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.