On Fri, Dec 10, 2021, jinfeng liu wrote:
>
>I am trying to install the GPU version of Amber20 on my centos7 server with
>A100 GPU card. Both of the GNU (gcc version 8.3.1) and INTEL (intel oneapi,
>icc version 2021.1) compilers have been tried, and all failed during the
>make install procudure.
>
>When I used the GNU compiler, the error message is like the following,
>
>*[ 65%] Linking CXX executable
I'll leave this one to Dan: full error message is in the original post; last
few lines are here:
>*CMakeFiles/cpptraj_common_obj.dir/DataIO_Cpout.cpp.o：在函数‘DataIO_Cpout::ReadCpin(FileName
>const&)’中：DataIO_Cpout.cpp:(.text+0x3775)：对‘std::__throw_out_of_range_fmt(char
>const*,
>...)’未定义的引用CMakeFiles/cpptraj_common_obj.dir/DataIO_Cpout.cpp.o:DataIO_Cpout.cpp:(.text+0x38d1):
>跟着更多未定义的参考到 std::__throw_out_of_range_fmt(char const*, ...)collect2: error:
>ld returned 1 exit statusmake[2]: *** [AmberTools/src/cpptraj/src/cpptraj]
>错误 1make[1]: *** [AmberTools/src/cpptraj/src/CMakeFiles/cpptraj.dir/all] 错误
>2make: *** [all] 错误 2*
>
>When I used the INTEL compiler, the error message is like the following,
>*/usr/local/cuda-11.1/include/crt/host_config.h(110): error: #error
>directive: -- unsupported ICC configuration! Only ICC 15.0, ICC 16.0, ICC
>17.0, ICC 18.0 and ICC 19.x on Linux x86_64 are supported!
This message comes from NVIDIA, not Amber. There is (generally) no reason
to use Intel compilers for GPU builds, since optimization at the CPU level
is irrelevant.
>What should I do to solve this problem ? Does Amber20 support A100 card ?
Yes.
>May I skip the cpptraj installation to avoid this problem ?
That is certainly worth a try. Add this line to your run_cmake script:
-DDISABLE_TOOLS="cpptraj"
and re-run run_cmake. More info is here:
https://ambermd.org/pmwiki/pmwiki.php/Main/CMake-Common-Options
...good luck....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Dec 10 2021 - 07:00:02 PST
