Hi,
What decision by the glibc / Ubuntu developers was so "bleeding edge?"
Trying to optimize memcpy performance seems a quite reasonable thing to do.
Do we need to contribute more CUDA-based code to repo test sets to save
ourselves pain?
Mark
On Sat, 4 Jun 2016 18:37 Ross Walker <ross.rosswalker.co.uk> wrote:
> Hi Manuel,
>
> This is not an AMBER unique problem. It's a problem of Ubuntu being too
> far past the bleeding edge.
>
> CUDA 7.5 does NOT support Ubuntu 16.04 yet - might not ever if the changes
> are too aggressive:
>
>
> CUDA 8.0 RC 1 purports to support Ubuntu 16.04 so you could try this -
> you'll probably need AMBER 16 though - we never tested AMBER 14 with CUDA 8
> (it may or may not work).
>
>
> So your options are register as a developer and download CUDA 8.0 RC -
> noting that it is a release candidate and thus not extensively tested yet
> with AMBER. Or rewind to Ubuntu 14.04.
>
> All the best
> Ross
>
> > On Jun 3, 2016, at 22:41, mmaestre <mmaestre.gate.sinica.edu.tw> wrote:
> >
> > Dear Amber users:
> >
> > Recently, I updated my laptop to Ubuntu 16.04 (I jump from LTS to LTS,
> meaning that I'm coming from ubuntu 14.04). Since I do most of my
> computational stuff on two servers which I maintain myself, I tend to you
> my laptop as a bit of a test-ground for amber and co.
> > When installing amber, both serial and parallel worked wonderfully, but
> when it comes to enabling hardware acceleration, I've had a bit of trouble.
> > When compiling, I get the following error:
> >
> > /usr/include/string.h: In function ‘void* __mempcpy_inline(void*, const
> void*, size_t)’:
> > /usr/include/string.h:652:42: error: ‘memcpy’ was not declared in this
> scope
> > return (char *) memcpy (__dest, __src, __n) + __n;
> > ^
> > Makefile:38: recipe for target 'kForcesUpdate.o' failed
> > make[5]: *** [kForcesUpdate.o] Error 1
> > make[5]: Leaving directory '/usr/local/xtal/amber16/src/pmemd/src/cuda'
> > Makefile:169: recipe for target 'cuda_spfp_libs' failed
> > make[4]: *** [cuda_spfp_libs] Error 2
> > make[4]: Leaving directory '/usr/local/xtal/amber16/src/pmemd/src'
> > Makefile:64: recipe for target 'cuda_SPFP' failed
> > make[3]: *** [cuda_SPFP] Error 2
> > make[3]: Leaving directory '/usr/local/xtal/amber16/src/pmemd/src'
> > Makefile:27: recipe for target 'cuda_serial' failed
> > make[2]: *** [cuda_serial] Error 2
> > make[2]: Leaving directory '/usr/local/xtal/amber16/src/pmemd'
> > Makefile:43: recipe for target 'cuda_serial' failed
> > make[1]: *** [cuda_serial] Error 2
> > make[1]: Leaving directory '/usr/local/xtal/amber16/src'
> > Makefile:7: recipe for target 'install' failed
> > make: *** [install] Error 2
> >
> > I've been googling a bit, and found that, back in 2010, Jason had a
> similar message regarding memcpy, which was related to the gcc compiler on
> ubuntu being too new. He had some success by rolling back to a precious
> version. I tried the same, going from 5.3.1 to 4.9.3, and to 4.8.5 (the
> configure script says that amber compilation was tested for 4.8.4 or
> higher). Neither of the three worked, and I still get the exact same error
> message.
> >
> > I've also read that there are some bugs between Cuda-7.5 and Ubuntu
> 16.04, so I'm thinking perhaps that's the problem. I'm getting the error
> message both with amber 14 and amber 16.
> >
> > Any help would be greatly appreciated! Should I perhaps try to apply for
> Cuda-8.0 RC?
> > Thank you very much!
> > Manuel
> >
> > P.S. Under Ubuntu 14.04, everything worked fine... but due to other
> pieces of software, I'd like to maintain 16.04, if possible.
> >
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Jun 04 2016 - 11:00:02 PDT
