Actually, for your case, since it will be a little while before the bugfix
can go through the usual channels and since I still don't have CUDA8 on
hand, would you mind going into AmberTools/src/mdgx/KernelMacros.h and line
12 from:
# define __ballot_sync(a, b) __ballot(a, b)
to:
# define __ballot_sync(a, b) __ballot(b)
That should solve your problem, and if it does not I'll at least know there
is something else.
Dave
On Thu, Jul 9, 2020 at 12:30 PM David Cerutti <dscerutti.gmail.com> wrote:
> The module in question was written for CUDA9 and above; I made some
> provisions in the hope that it wouldn't break on CUDA7 or 8, but those are
> very old CUDA compilers at this point and I didn't have access to machines
> with them. I was kind of flying blind, and it looks like I didn't properly
> fix the __ballot_sync(unsigned int, bool) function (which is just
> __ballot(bool) in CUDA8 and below). That's why you're getting these
> errors. Best solution, for performance in pmemd and going forward, is to
> upgrade to CUDA10. The mdgx CUDA modules are separate from pmemd, and are
> designed to run massive numbers of small GB simulations on the same GPU
> (aggregate throughput tops a millisecond of TrpCage per week on a V100).
> So, if all you need is pmemd.cuda, you can skip (fornow) the AmberTools
> CUDA installation and just go straight to pmemd. Nevertheless, we claim to
> support CUDA8 and a promise is a promise, so I will patch the problem.
>
> Dave
>
>
> On Thu, Jul 9, 2020 at 10:11 AM Camila Clemente <
> camilamaraclemente.gmail.com> wrote:
>
>> Hi
>> I can't install pmemd.cuda (amber20) on Centos7.
>> The following modules were loaded:
>> -gcc/5
>> -cuda/8.0
>> -cmake/3.16.2
>> -export HOME_CUDA=/opt/cuda/8.0
>> -export LD_LIBRARY_PATH="/opt/cuda/8.0/lib:${LD_LIBRARY_PATH
>> - add to run_cmake -DCUDA_TOOLKIT_ROOT_DIR=/opt/cuda/8.0
>>
>> I did the following:
>>
>> cd $AMBERHOME
>> make clean
>> ./configure -cuda gnu
>> make install
>>
>> and the error was:
>>
>> Makefile:226: atención: se imponen comandos para el objetivo
>> `ArraySimulator.o'
>> Makefile:218: atención: se ignoran las instrucciones viejas para el
>> objetivo `ArraySimulator.o'
>> [MDGX] NVCC ArraySimulator.cu
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> kDynamics.h(311): error: too many arguments in function call
>>
>> kDynamics.h(341): error: too many arguments in function call
>>
>> kDynamics.h(1565): error: too many arguments in function call
>>
>> kDynamics.h(1599): error: too many arguments in function call
>>
>> 48 errors detected in the compilation of
>> "/tmp/tmpxft_00005702_00000000-21_ArraySimulator.compute_61.cpp1.ii".
>> make[2]: *** [ArraySimulator.o] Error 2
>> make[2]: se sale del directorio
>> `/home/cclemente/amber20_src/AmberTools/src/mdgx'
>> make[1]: *** [cuda_serial] Error 2
>> make[1]: se sale del directorio
>> `/home/cclemente/amber20_src/AmberTools/src'
>> make: *** [install] Error 2
>>
>> Any help is appreciated
>> Thanks
>>
>> --
>> *Camila*
>> _______________________________________________
>> 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 09 2020 - 10:00:03 PDT
