CUDA 9.1

Tomasz Rybak tomasz.rybak at post.pl
Sun Feb 11 22:35:24 UTC 2018


On Thu, 2018-02-08 at 11:34 +0200, Graham Inggs wrote:
> Hi
> 
> On 05/02/2018 14:05, Andreas Beckmann wrote:
> > Please test it and test whether all your packages compile against
> > it,
> > s.t. we can do a transition at some point ...
> 
> FWIW, the following all rebuilt locally against nvidia-cuda-toolkit
> 9.1 
> in Ubuntu without modification:
> caffe-contrib
> eztrace-contrib
> hwloc-contrib
> pycuda

Thanks.
I compiled and tested PyCUDA on AWS p2.xlarge instance with CUDA 9.1.
(used AWS because my last machine with NVIDIA GPU died some time ago).

It mostaly works, there are only 2 tests failing:
/usr/lib/python3/dist-packages/pycuda/compiler.py:432:
FileNotFoundError
____________________________________ TestDriver.test_jit_link_module
____________________________________

args = (<test_driver.TestDriver object at 0x7fd774148278>,), kwargs =
{}
pycuda = <module 'pycuda' from '/usr/lib/python3/dist-
packages/pycuda/__init__.py'>
ctx = <pycuda._driver.Context object at 0x7fd7742e8660>
clear_context_caches = <function clear_context_caches at
0x7fd776826620>
collect = <built-in function collect>

    def f(*args, **kwargs):
        import pycuda.driver
        # appears to be idempotent, i.e. no harm in calling it more
than once
        pycuda.driver.init()
    
        ctx = make_default_context()
        try:
            assert isinstance(ctx.get_device().name(), str)
            assert isinstance(ctx.get_device().compute_capability(),
tuple)
            assert isinstance(ctx.get_device().get_attributes(), dict)
>           inner_f(*args, **kwargs)

/usr/lib/python3/dist-packages/pycuda/tools.py:460: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
test_driver.py:941: in test_jit_link_module
    mod.add_stdlib('cudadevrt')
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

self = <pycuda.compiler.DynamicModule object at 0x7fd7741489b0>,
libname = 'cudadevrt'

    def add_stdlib(self, libname):
        if self.libdir is None:
            self.libdir, self.libptn = self._locate_cuda_libdir()
        from os.path import isfile, join
        libpath = join(self.libdir, self.libptn % libname)
        if not isfile(libpath):
>           raise FileNotFoundError('CUDA SDK library file "%s" not
found' % libpath)
E           FileNotFoundError: CUDA SDK library file
"/usr/bin/../lib64/libcudadevrt.a" not found

/usr/lib/python3/dist-packages/pycuda/compiler.py:432:
FileNotFoundError
================================== 2 failed, 27 passed in 8.00 seconds

I'll investigate it during the week, but except for that CUDA 9.1 looks
OK.


Best regards.

-- 
Tomasz Rybak, Debian Maintainer
GPG: A565 CE64 F866 A258 4DDC F9C7 ECB7 3E37 E887 AA8C
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part
URL: <http://lists.alioth.debian.org/pipermail/pkg-nvidia-devel/attachments/20180211/c0a80c13/attachment.sig>


More information about the pkg-nvidia-devel mailing list