cuda 0.6.5.0 fails to install

Hi,

I get the following error when trying to install the cuda package via cabal.

Preprocessing library cuda-0.6.5.0...
<command-line>: error: "1" may not appear in macro parameter list
<command-line>: warning: missing whitespace after the macro name
c2hs: Error during preprocessing custom header file

environment:

-bash-4.1$ ghc --version
The Glorious Glasgow Haskell Compilation System, version 7.8.3
-bash-4.1$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2014 NVIDIA Corporation
Built on Thu_Jul_17_21:41:27_CDT_2014
Cuda compilation tools, release 6.5, V6.5.12

Cheers

[tmcdonell]

I have not encountered this before. What version of c2hs are you using?

Right, I knew I forgot something! :)

-bash-4.1$ c2hs --version
C->Haskell Compiler, version 0.18.2 The shapeless maps, 31 Oct 2014
build platform is "x86_64-linux" <1, True, True, 1>

[tmcdonell]

Okay, I was using c2hs-0.17.2. It appears something changed with the newer c2hs. I'll see if I can get a workaround/fix.

[tmcdonell]

By the way what platform are you running on?

[tmcdonell]

I would suggest just using the older 1.17.2 version of c2hs so that you can continue working until we sort this problem out.

Thanks for bringing the issue to my attention!

Thanks Trevor, that fixed the problem.
I'm not really running on platform... I partly installed the newest one and just used cabal to install whatever I needed.

I'll shoot you an email by the early next week with some results from our benchmarks and thoughts on Accelerate.
Keep up the good work :)

[tmcdonell]

Thanks! Yes, please do let us know how we can improve Accelerate in any way (and sorry the installation was not as smooth as it should have been).

[tmcdonell]

Submitted as NV Bug ID #1557230

[vjeranc]

I'll just say that I had similar problems installing accelerate-cuda, where cuda-0.5.1.1 couldn't finish installing. Depending on the c2hs version I would get the OP's error or

c2hs: Errors during expansion of binding hooks:

./Foreign/CUDA/Driver/Marshal.chs:151: (column 15) [ERROR] >>> Unknown identifier!
Cannot find a definition for `cuMemHostRegister' in the header file.

After seeing the nvcc --version line up there, by installing nvidia-cuda-toolkit everything started working fine.

[tmcdonell]

@vjeranc Do you happen to know which version of nvcc you had installed when you got the error you posted? I try to keep the bindings mostly backwards compatible with old CUDA releases.

[vjeranc]

@tmcdonell 331.38 was the version of the binary driver I was using, I'm not sure which version of nvcc was present then, currently it's 340.29. nvcc --> 5.5.
I'll try to recreate the issue again.
I hit a wall somewhere and didn't manage to resolve the dependency issues surrounding the accelerate package, so, I'll try again soon.

Although, if it is possible that these errors show up if there's no presence of the nvcc (be it missing installation or missing path) that also might be the reason.

[tmcdonell]

The oldest version of the CUDA toolkit I have on hand is 4.2, which includes cuMemHostRegister. This is already pretty old, so I wouldn't worry too much about installing very old versions in order to get this information, unless you actually need to use them.

Certainly we should fix the accelerate issues though.