GPU support

Aperta
  • Qualcosa non va?
#93 4 commenti 0 reazioni 0 assegnatari Vedi su GitHub

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
30/100
Tipo di issue
Funzionalità
Chiarezza
Abbastanza chiara
Stato di attività
Ferma
Stack tecnologico
julia
Ambito
hpc

Direzione di ricerca

Inizia riproducendo i due casi segnalati: eseguire il broadcasting di una RuntimeGeneratedFunction su un CuVector e chiamare un kernel generato con @cuda. Traccia come RuntimeGeneratedFunction viene abbassata in ciascun percorso; il lavoro è completato quando il broadcasting GPU e i kernel vengono compilati senza argomenti non-isbits né un ritorno Any.

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

Descrizione

I've been using this project to circumvent the world age problems when generating functions and it has worked great so far.

Now I'm trying to also generate functions to run on the GPU, CUDA for now to be specific. There are two ways that I've tried to do that. The first is simply broadcasting the function on a CuVector, however, this does not work because RuntimeGeneratedFunctions are not isbits, which CUDA.jl does not like:

julia> func.(cu_inputs)
ERROR: GPU compilation of MethodInstance for (::GPUArrays.var"#34#36")(::CUDA.CuKernelContext, ::CuDeviceVector{ComplexF64, 1}, ::Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{…}, Tuple{…}, RuntimeGeneratedFunction{…}, Tuple{…}}, ::Int64) failed
KernelError: passing and using non-bitstype argument

Argument 4 to your kernel function is of type Base.Broadcast.Broadcasted{CUDA.CuArrayStyle{1, CUDA.Mem.DeviceBuffer}, Tuple{Base.OneTo{Int64}}, RuntimeGeneratedFunction{(:data_input,), ComputableDAGs.var"#_RGF_ModTag", var"#_RGF_ModTag", (0x3d86ea6b, 0x46940344, 0xf0c50769, 0x501ce261, 0x5c3fa2fa), Expr}, Tuple{Base.Broadcast.Extruded{CuDeviceVector{<LongArgumentType...>, 1}, Tuple{Bool}, Tuple{Int64}}}}, which is not isbits:
  .f is of type RuntimeGeneratedFunction{(:data_input,), ComputableDAGs.var"#_RGF_ModTag", var"#_RGF_ModTag", (0x3d86ea6b, 0x46940344, 0xf0c50769, 0x501ce261, 0x5c3fa2fa), Expr} which is not isbits.
    .body is of type Expr which is not isbits.
      .head is of type Symbol which is not isbits.
      .args is of type Vector{Any} which is not isbits.

The other option is to use @cuda and call a (generated) kernel. However, this also does not work as RuntimeGeneratedFunctions seem to always return an Any, and GPU kernels are not allowed to return anything:

julia> @cuda kernel(cu_inputs, cu_outputs, 1000)
ERROR: GPU compilation of MethodInstance for (::RuntimeGeneratedFunction{(:input_vector, :output_vector, :n), CUDAExt.var"#_RGF_ModTag", var"#_RGF_ModTag", (0x26ea8e24, 0xf73e2d2a, 0xb1e0f07e, 0x53860e81, 0xc0771049), Expr})(::CuDeviceVector{PhaseSpacePoint{…}, 1}, ::CuDeviceVector{Float64, 1}, ::Int64) failed
KernelError: kernel returns a value of type `Any`

Make sure your kernel function ends in `return`, `return nothing` or `nothing`.

It would be great if compatibility for GPU broadcasting and kernels could be added. Looking at the code in this package, I don't understand how it is doing what it is doing so I don't think I can add support myself. It seems that fixing the return nothing problem with kernels might not be too hard though.

Lingua principale
Julia
Stelle
114
Fork
19
Merge medio
2h 42m
PR unite (30g)
1

Preparare l'ambiente

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 SciML/RuntimeGeneratedFunctions.jl

Tutte le issue di SciML/RuntimeGeneratedFunctions.jl

Issue simili

Altre issue su Julia

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.