Use pointer to async_send directly instead of a wrapper function

[vchuravy]

Fixes #1314

@ericphanson can you check that this fixes your issue?

[ericphanson]

Sure, I'll give it a try

[ericphanson]

it crashes at the same place it used to segfault, saying

ERROR: LoadError: could not load symbol "async_send":
julia: undefined symbol: async_send
Stacktrace:
  [1] launch(f::CUDA.var"#13#14"{Base.Event}; stream::CuStream)
    @ CUDA ~/.julia/packages/CUDA/ZRjZ5/lib/cudadrv/execution.jl:157
  [2] nonblocking_synchronize
    @ ~/.julia/packages/CUDA/ZRjZ5/lib/cudadrv/stream.jl:157 [inlined]
  [3] macro expansion
    @ ~/.julia/packages/CUDA/ZRjZ5/src/array.jl:409 [inlined]
  [4] macro expansion
    @ ~/.julia/packages/CUDA/ZRjZ5/src/state.jl:182 [inlined]
  [5] unsafe_copyto!(dest::Matrix{Float32}, doffs::Int64, src::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}, soffs::Int64, n::Int64)
    @ CUDA ~/.julia/packages/CUDA/ZRjZ5/src/array.jl:405
  [6] copyto!
    @ ~/.julia/packages/CUDA/ZRjZ5/src/array.jl:359 [inlined]
  [7] copyto!
    @ ~/.julia/packages/CUDA/ZRjZ5/src/array.jl:363 [inlined]
  [8] copyto_axcheck!(dest::Matrix{Float32}, src::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer})
    @ Base ./abstractarray.jl:1104
  [9] Array
    @ ./array.jl:563 [inlined]
 [10] Array
    @ ./boot.jl:481 [inlined]
 [11] convert
    @ ./array.jl:554 [inlined]
 [12] adapt_storage
    @ ~/.julia/packages/Adapt/wASZA/src/arrays.jl:6 [inlined]
 [13] adapt_structure
    @ ~/.julia/packages/Adapt/wASZA/src/Adapt.jl:42 [inlined]
 [14] adapt
    @ ~/.julia/packages/Adapt/wASZA/src/Adapt.jl:40 [inlined]
 [15] adapt_storage
    @ ~/.julia/packages/Flux/BPPNj/src/functor.jl:75 [inlined]
 [16] adapt_structure
    @ ~/.julia/packages/Adapt/wASZA/src/Adapt.jl:42 [inlined]
 [17] adapt
    @ ~/.julia/packages/Adapt/wASZA/src/Adapt.jl:40 [inlined]
 [18] #134
    @ ~/.julia/packages/Flux/BPPNj/src/functor.jl:115 [inlined]
 [19] fmap(f::Flux.var"#134#135", x::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer}; exclude::typeof(Functors.isleaf), walk::typeof(Functors._default_walk), cache::IdDict{Any, Any})
    @ Functors ~/.julia/packages/Functors/hIysk/src/functor.jl:121
 [20] fmap
    @ ~/.julia/packages/Functors/hIysk/src/functor.jl:120 [inlined]
 [21] cpu(x::CuArray{Float32, 2, CUDA.Mem.DeviceBuffer})
    @ Flux ~/.julia/packages/Flux/BPPNj/src/functor.jl:115
[truncated where it reaches my code]

[vchuravy]

it crashes at the same place it used to segfault, saying

I was a silly goose.

[ericphanson]

Success! No segfault or other issues

[codecov]

Codecov Report

Merging #1319 (0aea816) into master (51a03ba) will decrease coverage by 0.80%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1319      +/-   ##
==========================================
- Coverage   80.48%   79.67%   -0.81%     
==========================================
  Files         119      119              
  Lines        8393     8400       +7     
==========================================
- Hits         6755     6693      -62     
- Misses       1638     1707      +69     

Impacted Files	Coverage Δ
lib/cudadrv/execution.jl	89.65% <100.00%> (-6.90%)	⬇️
lib/cudadrv/occupancy.jl	75.86% <0.00%> (-15.81%)	⬇️
lib/cublas/util.jl	83.67% <0.00%> (-8.17%)	⬇️
src/compiler/exceptions.jl	88.23% <0.00%> (-5.89%)	⬇️
lib/cusolver/multigpu.jl	91.17% <0.00%> (-5.15%)	⬇️
lib/cusolver/linalg.jl	84.37% <0.00%> (-4.69%)	⬇️
lib/cudadrv/module/jit.jl	65.51% <0.00%> (-3.45%)	⬇️
src/compiler/gpucompiler.jl	80.64% <0.00%> (-3.23%)	⬇️
src/initialization.jl	43.18% <0.00%> (-3.16%)	⬇️
... and 24 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 51a03ba...0aea816. Read the comment docs.

[maleadt]

Great, thanks!

[ericphanson]

could we get a release with this fix?

[maleadt]

Yes, I'll work on a release today.

[ericphanson]

thanks!