Dagger.jl Files

Dagger v0.22.7

Diff since v0.22.6

Merged pull requests:

• Fix GPU IPC and collect (#758) (@jpsamaroo)
• Reduce CI pressure on the JuliaGPU cluster (#761) (@maleadt)
• Sch: Propagate a failure to dependents before deleting the failed task (#763) (@aalquwayfili)
• DArray: Add diag (#766) (@aalquwayfili)
• DArray: Infer the result type of reductions through f (#767) (@aalquwayfili)
• DArray: Do not pass Base._InitialValue to partition reductions (#769) (@aalquwayfili)
• MPI: spread plain @spawn tasks across ranks and threads (#771) (@jpsamaroo)
• Fix GPU CI startup and move CPU jobs to GitHub Actions (#772) (@jpsamaroo)
• DArray: Add accumulate, cumsum and cumprod (#775) (@aalquwayfili)

Closed issues:

• using mean with distributed Dagger arrays (#279)
• Spawn macro does not compose with logging macros (#514)
• Wait does not work for MPI chunks (#591)
• Logging is slow (#621)
• MPI: plain Dagger.@spawn tasks all run on rank 0, thread 1 (#760)
• A failed task with dependents can take down the eager scheduler (#762)
• Reductions on a DArray return the wrong type when f changes the element type (#764)
• diag and tr of a DMatrix throw an ArgumentError (#765)
• sum, prod, maximum and mapreduce fail on a GPU backed DArray (#768)
• cumsum(DA; dims) throws a ConcurrencyViolationError (#770)