cuda 0.6.5.0 fails to install about cuda

[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.

[pnutus]

When I use c2hs 0.18.2 I get the error

Preprocessing library cuda-0.6.5.0...
In file included from <built-in>:170:
<command line>:5:29: error: expected comma in macro parameter list
#define C2HS_MIN_VERSION(mj 1
                            ^
<command line>:7:11: warning: ISO C99 requires whitespace after the macro name [-Wc99-extensions]
#define rv) (mj<=0&&mn<=18&&rv<=2)
          ^
1 warning and 1 error generated.
c2hs: Error during preprocessing custom header file

When I use c2hs 0.17.2 I get the error

Preprocessing library cuda-0.6.5.0...
c2hs: C header contains errors:

/usr/include/stdlib.h:177: (column 46) [ERROR]  >>> Syntax error !
  The symbol `__OSX_AVAILABLE_BUT_DEPRECATED_MSG' does not fit here.

cabal: Error: some packages failed to install:
cuda-0.6.5.0 failed during the building phase. The exception was:
ExitFailure 1

I am running OS X 10.10 and CUDA 6.5.14.

[tmcdonell]

I haven't upgraded to Yosemite yet.
If you can find the definition of the __OSX_AVAILABLE_BUT_DEPRECATED_MSG macro that would help. Presumably it has changed since 10.9 and we need to update the dummy implementation here: https://github.com/tmcdonell/cuda/blob/master/cbits/stubs.h#L21

[pnutus]

Can't really find a definition, the closest is this (look at the bottom of the file): http://www.opensource.apple.com/source/xnu/xnu-2422.1.72/EXTERNAL_HEADERS/Availability.h

There is a newer Availability.h on my local computer, but I can't find it online. It looks the same in terms of __OSX_AVAILABLE_BUT_DEPRECATED_MSG, though.

It seems like what's needed is just a message argument. I am not confident enough with compiler macros to make a pull request, though.

[tmcdonell]

Okay, thanks for chasing that! I should get a change to upgrade to yosemite in the next few days and try it out (hopefully).

[aesadde]

Hi,

any progress on this issue yet?
I'm having the same problem as @pnutus.
/usr/include/stdlib.h:177: (column 46) [ERROR] >>> Syntax error ! The symbol __OSX_AVAILABLE_BUT_DEPRECATED_MSG' does not fit here.`

I'm running osx 10.10, nvcc 6.5, and c2hs 0.17.2.

Thanks,

[tmcdonell]

Hi all,

Sorry I haven't gotten around to this yet, lots of deadlines recently. I'm downloading the Yosemite installer as we speak though, so hopefully will have a fix soon.

[tmcdonell]

Version 0.6.5.1 has been uploaded that fixes compilation on Yosemite.

c2hs has been fixed so that this package will compile c2hs HEAD; that version is currently unreleased, but I can edit the comment on Hackage later. For the time being, the documentation header still advises to use version 0.17.