You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Will arrayfire-r be updated soon to allow Windows usage? #11
I see arrayfire-r has been dormant for a few years. In the ArrayFire page it says there is R support. In arrayfire-r, it says Linux is supported, but Windows will be supported "soon".
Is it expected a new arrayfire-r version, which will allow Windows usage?
@DespertaFerro Unfortunately, there isn't any active development going on the R wrapper support for quite some time. We haven't had that many requests for R wrapper, so updating/maintaining this language binding is currently not at the top of the priority list.
@pavanky Any chance that you are working on latest version but haven't published it yet ?
You have my vote for an R wrapper. I was actually looking to contribute to a ViennaCL R wrapper when I came across ArrayFire, whose core appears to be much more actively maintained.
@jeffkeller87 At the moment, we don't have any active development going on for R bindings. You are welcome to contribute to R wrapper for ArrayFire. We are available on Slack for quick questions. We are also on Google groups for much detailed discussions.
Getting R and AF to work together on Windows is complicated. If you use the pre-build binaries you have to use MSVC for compiling the R package's library, while normally mingw (from Rtools) is used. Alternatively you could compile AF with the mingw tool chain, but that rules out CUDA. See daqana/rcpparrayfire#3 for a similar discussion also without solution as of now.
@rstub Can you please explain what do you mean "mingw rules out CUDA".
As long as C-API from our website binaries is used, you wouldn't face any issues even if our binaries are built using MSVC and your target toolchain is mingw.
@rstub Can you please explain what do you mean "mingw rules out CUDA".
I meant "you cannot build the CUDA backend if you build AF using the mingw tool chain".
As long as C-API from our website binaries is used, you wouldn't face any issues even if our binaries are built using MSVC and your target toolchain is mingw.
I do not remember the details, but I think I had linking problems using the AF's C-API together with mingw. I have not investigated this in detail, though, since I need the C++-API for RcppArrayFire anyway. And that cannot work with mingw due to different C++ name mangling schema.
That is right, C++ will have the problem you mentioned. But C-API shouldn't be a problem. I believe I have once tried using ArrayFire binaries(website download) with rust mingw toolchain, if I remember correctly, it did work.
To compile AF, you need a C++ compiler. Now, I haven't tried CUDA backend with mingw toolchain, is it a well known issue ? or something you noticed only with ArrayFire ?
Activity
9prady9 commentedon Sep 23, 2018
@DespertaFerro Unfortunately, there isn't any active development going on the R wrapper support for quite some time. We haven't had that many requests for R wrapper, so updating/maintaining this language binding is currently not at the top of the priority list.
@pavanky Any chance that you are working on latest version but haven't published it yet ?
jeffkeller87 commentedon May 22, 2019
You have my vote for an R wrapper. I was actually looking to contribute to a ViennaCL R wrapper when I came across ArrayFire, whose core appears to be much more actively maintained.
9prady9 commentedon May 22, 2019
@jeffkeller87 At the moment, we don't have any active development going on for R bindings. You are welcome to contribute to R wrapper for ArrayFire. We are available on Slack for quick questions. We are also on Google groups for much detailed discussions.
rstub commentedon Jun 7, 2019
Getting R and AF to work together on Windows is complicated. If you use the pre-build binaries you have to use MSVC for compiling the R package's library, while normally mingw (from Rtools) is used. Alternatively you could compile AF with the mingw tool chain, but that rules out CUDA. See daqana/rcpparrayfire#3 for a similar discussion also without solution as of now.
9prady9 commentedon Jun 7, 2019
@rstub Can you please explain what do you mean "mingw rules out CUDA".
As long as C-API from our website binaries is used, you wouldn't face any issues even if our binaries are built using MSVC and your target toolchain is mingw.
rstub commentedon Jun 7, 2019
I meant "you cannot build the CUDA backend if you build AF using the mingw tool chain".
I do not remember the details, but I think I had linking problems using the AF's C-API together with mingw. I have not investigated this in detail, though, since I need the C++-API for RcppArrayFire anyway. And that cannot work with mingw due to different C++ name mangling schema.
9prady9 commentedon Jun 7, 2019
That is right, C++ will have the problem you mentioned. But C-API shouldn't be a problem. I believe I have once tried using ArrayFire binaries(website download) with rust mingw toolchain, if I remember correctly, it did work.
To compile AF, you need a C++ compiler. Now, I haven't tried CUDA backend with mingw toolchain, is it a well known issue ? or something you noticed only with ArrayFire ?
rstub commentedon Jun 7, 2019
Interesting. I might give it another shot once I have a Windows dev machine running.
AFAIK CUDA on windows requires MSVC, c.f. https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html. I am not sure if it possible to compile the CUDA parts with nvcc, which uses MSVC, and the CPU parts in mingw.