I don’t know what OS you are working on though I would assume it is some linux distro. These often ship with a version of gcc pre-installed. It should also be possible to install a “second version” of gcc, not making this the default.
There might be a gcc4 version hidden somewhere already, which you should be able to check using something like
$ find /usr/ -name “gcc"
or with an appropriate search path, you could potentially solve the issue through symlinking. Though I also assume that this might cause problems if there are other software depending on cuda.
$ sudo ln -s /path/to/bin/gcc-4.X /path/to/cuda/bin/gcc and sudo ln -s /path/to/bin/g++-4.X /path/to/cuda/bin/g++
I think the default cuda binary directory is "/usr/local/cuda/bin”. Though I would suspect that this would also mean that you should build the rest of amber with the same version of gcc to avoid issues? (Someone with more insight might be able to chime in here).
I would guess that you could export/alias another version of gcc for the particular install and the repeating this manually for a session when amber is being used, not making any changes persistent (CXX=g++-4.4 CC=gcc-4.4)
However, it seems that the problem might be something else, if you have Cuda 10 installed, it should support gnu gcc 7. So the best way to solve this issue seems to be to get the correct version of Cuda included in the configuration of the amber install, then the gnu gcc problem should resolve itself. Does amber support Cuda 10?
Best regards
// Gustaf
> On 2 Jul 2020, at 07:57, Suchetana Gupta <tutulg.gmail.com> wrote:
>
> Thanks for the response. Actually we have other programs running on this
> machine which will be disrupted if we shift to gcc6. That is why we were
> thinking of there is some other alternative.
> Thanks
> Suchetana Gupta
> IACS, Kolkata
>
> On Thu, 2 Jul 2020, 11:25 Gustaf Olsson, <gustaf.olsson.lnu.se> wrote:
>
>> The error log seems pretty clear:
>>
>> #error -- unsupported GNU version! gcc versions later than 6 are not
>> supported!
>>
>> I am uncertain if this is a restriction from Amber or the installed CUDA
>> version though this matters less. What happens if you install gcc6 and try
>> to build?
>>
>> It seems that compiling amber using the gnu compilers (pretty much been my
>> default since 2010-2018) is being met with more complications as the build
>> seems picky regarding versions and particularly in combination with
>> openmpi/cuda.
>>
>> Anyway, I would initially start by reverting to gcc6 and see if that
>> produces a working build.
>> Best regards
>> // Gustaf
>>
>> On 1 Jul 2020, at 20:13, Suchetana Gupta <tutulg.gmail.com<mailto:
>> tutulg.gmail.com>> wrote:
>>
>> _______________________________________________
>>
>> _______________________________________________
>> 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 Thu Jul 02 2020 - 00:00:03 PDT
