Skip to content

Using shards inside spawn_datadeps leads to generic memory error #665

@cneverett

Description

@cneverett

Using a shard inside a spawn_datadeps environment leads to the error ERROR: Type GenericMemory does not have a definite size.

A simple example of this is

using Distributed; addprocs(2)
nworkers = length(workers())
@everywhere using Dagger
@everywhere using LinearAlgebra

A = rand(Blocks(8, 4), 4*4, 4) 
B = rand(Blocks(2), 4) 
C = Dagger.@shard rand(4*4) 

Dagger.spawn_datadeps() do 

  for (idx, w) in enumerate(workers())
    Dagger.@spawn scope=Dagger.scope(worker=w) mul!(Out(C), A.chunks[idx], B.chunks[idx])
  end

  # further operations that use C 
end

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions