feature: detect MPI desynchronizations - #4091

Draft

kdrienCG wants to merge 22 commits into

feature/kdrienCG/detectMPIDesync

Commits on Jun 12, 2026

add build option to enable MPI desync detection
kdrienCG
committed
24f46c6
add desync dectection for barrier()
kdrienCG
committed
61701a3
move detectMpiDesync in the .hpp
kdrienCG
committed
71ac289
add desync detection to all collective operations
kdrienCG
committed
e3a406e
complete collective calls desync detection
kdrienCG
committed
72e6581
rename variables for readability
kdrienCG
committed
a19e09e
fix max Allreduce call
kdrienCG
committed
c881621
add MpiDesyncGuard RAII helper for readability
kdrienCG
committed
27ec2ee

Commits on Jun 17, 2026

fix global counter name
kdrienCG
committed
f7cf49f
replace standard MPI_* collective operations use with MpiWrapper ones
kdrienCG
committed
e4753b7

Commits on Jun 18, 2026

store stacktraces on desync detections
kdrienCG
committed
24f60b7

Commits on Jun 19, 2026

wip: use a timeout as desync detection method
kdrienCG
committed
b2574ef

Commits on Jun 22, 2026

wip: use a timeout as desync detection method
kdrienCG
committed
067d70f

Commits on Jul 3, 2026

Commits on Jul 8, 2026

resolve stacktrace on unhappy path only
kdrienCG
committed
0d5a684

Commits on Jul 15, 2026

revert to "match tag" implementation
kdrienCG
committed
f46cd74

Commits on Jul 16, 2026

fix: increment collective operation tag
kdrienCG
committed
b0f2a76

Commits on Jul 21, 2026

revert to "timeout" implementation
kdrienCG
committed
ab11209

Commits on Jul 22, 2026

refactor MpiDesyncGuard for testability
kdrienCG
committed
601be60