Current Event callback uses undefined behaviour

[vchuravy]

I learned today [1] that the solution I implemented for user callbacks in [2] is using undefined/illegal behaviour.
I am quite amazed that it worked as long as it did...

code_llvm(OpenCL.event_notify, (OpenCL.CL_event, OpenCL.CL_int, Ptr{Void}))

%7 = load %jl_value_t*** @jl_pgcstack, align 8

I am not quite sure how to solve that problem without reimplementing Base.SingleAsyncWork since we can't pass along evt_id::CL_event, status::CL_int.

@jakebolewski Any ideas?

[1] https://groups.google.com/forum/#!topic/julia-dev/9J1GYfCyVpE
[2]

OpenCL.jl/src/event.jl

Lines 97 to 106 in f529546

	function event_notify(evt_id::CL_event, status::CL_int, julia_func::Ptr{Void})
	# Obtain the Function object from the opaque pointer
	callback = unsafe_pointer_to_objref(julia_func)::Function
	# In order to callback into the Julia thread create an AsyncWork package.
	cb_packaged = Base.SingleAsyncWork(data -> callback(evt_id, status))
	# Use uv_async_send to notify the main thread
	ccall(:uv_async_send, Void, (Ptr{Void},), cb_packaged.handle)
	end

[dfdx]

Can it be the reason for segfault in tests? (note that I commented out possibly erroneous tests for arrays).

[vchuravy]

Yes that is entirely possible and probably my fault. I am going to look
into this. How often do you see it occur?

On Thu, 12 Nov 2015 07:13 Andrei Zhabinski [email protected] wrote:

Can it be the reason for segfault in tests
https://travis-ci.org/JuliaGPU/OpenCL.jl/jobs/90612668? (note that I
commented out possibly erroneous tests for arrays).

—
Reply to this email directly or view it on GitHub
#86 (comment).

[dfdx]

It is pretty reproducible in Travis (40% of time, in other 60% error is not printed, but build fails, so it may also be caused by segfault). Though, currently I don't have access to a GPU-enabled computer, so cannot test it locally.

Note 2 things:

• segfault-test branch is a copy of master with array tests commented out
• segfault occurs only for Julia v0.4 on Linux, nightly build and Mac are not affected (according to Travis)

[vchuravy]

@dfdx Sorry I completely missed the point you were trying to make. I though #87 was already merged and was about to investigate the seqfault you mentionend and wondered why it is still the old code. I think it is time to merge #87 now and that should hopefully solve your issue.

[dfdx]

Thanks, that error has gone now. However, I encounter another (minor) issue which might be related to even callbacks - see #92 .