Can CUDArt be loaded with VS2015? about cudart.jl

[musm]

I can confirm this works if add VS140COMNTOOLS
https://github.com/JuliaGPU/CUDArt.jl/blob/master/deps/build.jl#L3
yayy

Sans the slew of deprecation warnings

[musm]

@liusihaoshiwo Did you get this to work? I think the problem is that CUDA 7.5 doesn't even properly work with VS 2015, thus I don't see why you would expect this to build too

julia> Pkg.build("CUDArt")
INFO: Building CUDArt

Microsoft (R) Program Maintenance Utility Version 14.00.23918.0
Copyright (C) Microsoft Corporation.  All rights reserved.

        nvcc --shared --compiler-options="/wd4819" --linker-options= wrapcuda.c -o libwrapcuda.dll
nvcc fatal   : nvcc cannot find a supported version of Microsoft Visual Studio. Only the versions 2010, 2012, and 2013 are supported
NMAKE : fatal error U1077: '"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\bin\nvcc.EXE"' : return code '0x1'
Stop.
========================================================[ ERROR: CUDArt ]========================================================


LoadError: failed process: Process(`cmd /C 'C:\Program Files (x86)\Microsoft Visual Studio 14.0\Common7\Tools\..\..\VC\vcvarsall.bat' amd64 & nmake -f Windows.mk`, ProcessExited(2)) [2]
while loading C:\Users\Mustafa\.julia\v0.5\CUDArt\deps\build.jl, in expression starting on line 23

=================================================================================================================================


========================================================[ BUILD ERRORS ]=========================================================


WARNING: CUDArt had build errors.

 - packages with build errors remain installed in C:\Users\Mustafa\.julia\v0.5
 - build the package(s) and all dependencies with `Pkg.build("CUDArt")`
 - build a single package by running its `deps/build.jl` script

==================================================================================================

We will have to wait till CUDA 8 for VS2015 support...

[musm]

can be now closed

[timholy]

I don't know much about Windows, so others may have to help you if you can't figure this out. But check out the build.jl script as listed in the backtrace; presumably it's just a matter of adding a check for a newer version of VS.

Assuming you do this in a way that doesn't break things for other users, please submit your fix as a pull request!

@timholy yeah.. I guess so. But I don't know what check should I add in my VS. I will continue trying.

[timholy]

I don't think you need to change your VS in any way, this seems to be an environment variable and/or path issue.

Before trying to load CUDArt you can do this:

julia> ENV["foo"] = "bar"
"bar"

julia> haskey(ENV, "foo")
true

So you could spoof having VS 2012 (VS120COMNTOOLS) until you figure out the magic incantation.

[ChrisRackauckas]

Yes, VS2015 has issues with CUDA 7.5 that would be hard to get around. CUDA 8 solves those issues, so CUDArt.jl should work pretty much out of the box with VS2015 once CUDA 8 is out.

[kshyatt]

CUDA 8 is out. Does VS2015 work with this package now? Can we add that line to build.jl? What about VS14?

[musm]

@kshyatt finally VS2015 support for cuda: https://devblogs.nvidia.com/parallelforall/cuda-8-features-revealed/ older versions of VS will not work.

[kshyatt]

Yes, but has anyone actually built this package with it?

[ChrisRackauckas]

I've been using it for awhile (since the beta release, using it with a GTX 1080). It works. Never tried it with VS though. Can't say I really have a plan to either...

[kshyatt]

I can certainly believe CUDA 8 works and this package works with gcc and CUDA 8. But the issue is about VS :). Can someone try it out?

[kshyatt]

Thanks for checking! Do you mind opening a PR (and maybe fixing some depwarns?)

[musm]

Also tests don't pass but that's another thing.

[musm]

Troublesome test:

ERROR: MethodError: no method matching copy!(::CUDArt.CudaArray{Float64,2}, ::Array{Float32,2})
Closest candidates are:
copy!{T}(::Array{T,N}, ::Array{T,N}) at array.jl:68
copy!(::BitArray{N}, ::Array{T,N}) at bitarray.jl:457
copy!(::Base.PermutedDimsArrays.PermutedDimsArray{T,N,perm,iperm,AA<:AbstractArray}, ::AbstractArray{T,N}) at permuteddimsarray.jl:62

[musm]

hopefully Tim can update the package with a pr for this fix

[timholy]

Not sure which PR you mean?

With all the work on Images and the zoo of associated changes to arrays/colors/numbers/io, GPUs are just beyond me to think about right now. I'm happy to allow others to make decisions here. It's not like I'm a huge expert on the subject, anyway (there are many who know much more). Perhaps the pool of contributors should be enlarged if we're missing people who know stuff who can merge things.

[musm]

ah sorry I meant other TIm @maleadt : )

[maleadt]

@musm I'm not a Windows user, and CUDArt's tests aren't failing on my setup (Linux x64, Julia nightly, CUDA 8) -- except occasionally for the gc() ones which are just fragile. Maybe make a separate bug report with some more information and a full backtrace (the one you posted doesn't show which test failed)?

[musm]

#59 , the test failure above was on the tagged version..

[kshyatt]

Thanks for all your work on this @musm!