Re: [AMBER] error while building Amber20 with Intel + CUDA

From: David A Case <>
Date: Tue, 2 Feb 2021 16:27:40 -0500

On Tue, Feb 02, 2021, Susan Chacko wrote:
>I compiled Amber 20 with gcc + CUDA + MPI on our Centos7 cluster, and
>everything worked fine.
>I also tried compiling with the Intel compilers, but am getting an
>error. The Amber config process reports
># Only ICC 15.0, ICC 16.0, ICC 17.0, ICC 18.0 allowed
># CUDA 7.5, 8.0, 9.0, 9.1, and 9.2 have been tested and are supported.
>so I tried Intel 2018.3.222 and CUDA 9.2.
>'configure -cuda intel' completed, but the following 'make install' died

>In file included from gpuContext.h(6),
>                  from gpu.cpp(23):
>gti_gpuContext.h(47): error: qualified name is not allowed

Thanks for your report. I'm cc-ing to people that are most familiar with
the gti_ code.

Here is a workaround: there is little to no point in compiling with CUDA and
Intel compilers. Almost all the time in pmemd.cuda is spent on the CPU, so
any optimizations a specialized compiler might perform on the CPU side are

Some GPU person should probably look into this, since if the problem shows
up with Intel, it may affect GNU compilers as well. But for now,
disable CUDA when you try to use Intel compilers. Then the system won't
even try to compile gti_gpuContext.h. I'm pretty sure Amber developers
haven't seen this because they have not felt a need to try this combination.

...hope this helps....dac

AMBER mailing list
Received on Tue Feb 02 2021 - 13:30:02 PST
Custom Search