GCC 8 and BLAS/LAPACK

[Anton Kudelin]

Dear Alfio,

The problem has been fixed with the new release of OpenBLAS 0.3.0. Thank 
you!

Anton

вторник, 22 мая 2018 г., 10:49:44 UTC пользователь Alfio Lazzaro написал:
>
> Dear Anton,
> I did a test with the toolchain and GCC 8.1. Note that we use Netlib 
> BLAS/LAPACK in the toolchain, therefore I should reproduce your problem.
> Then I compiled a PSMP version of CP2K. I got some warnings related to 
> implicit reallocation and uninitialized values, which are not supposed to 
> be there (at least they were not recognized by GCC 7.3). I ignored them and 
> run regression tests. It works.
>
> This is the way to run CP2K with toolchain and GCC 8.1.
> 1) Download CP2K trunk (svn checkout 
> http://svn.code.sf.net/p/cp2k/code/trunk/cp2k cp2k)
> 2) open the file cp2k/tools/toolchain/scripts/package_versions.sh and 
> change the GCC version:
>
> -gcc_ver=${gcc_ver:-7.3.0}
> +gcc_ver=${gcc_ver:-8.1.0}
>
> 3) Go inside the directory: tools/toolchain and run the command:
>
> ./install_cp2k_toolchain.sh --install-all --with-quip=no
>
> It will install all packages, including GCC and BLAS.
>
> 4) Compile CP2K as usual and then with your input. As I mentioned, I get 
> some warnings, that's why I have removed the flags:
>
> -Werror=uninitialized -Werror=realloc-lhs-all
>
> Let me know if it works. 
>
> Alfio
>
>
>
>
>
> Il giorno sabato 19 maggio 2018 13:36:38 UTC+2, Anton Kudelin ha scritto:
>>
>> Hi all,
>>
>> I run Arch Linux, and recently it changed the major version of the system 
>> compiler. So I tried to compile CP2K with GCC 8.1.0 after the environment 
>> being rebuilt and found something strange. The problem may lay far beyond 
>> CP2K, but the program is a reliable indicator of the quality of compilers 
>> and libraries, by the way it has very experienced community.
>>
>> In short, BLAS/LAPACK routines don't working when compiled with GCC 
>> 8.1.0. I tried NetLib BLAS/LAPACK, ATLAS, and OpenBLAS and obtained the 
>> same results: SCF cannot get converged. On the other hand, Intel MKL does 
>> things as earlier.
>>
>> Could anybody check my results? If I get them proved, it could mean a 
>> terrible issue with GCC 8.1.0 or even with BLAS/LAPACK that has to be 
>> reported.
>>
>> Anton
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20180604/536bbe56/attachment.htm>