Re: [scipion-users] [EXT] Re: Problem compiling xmipp in Scipion3 on Centos 7 cluster

[Lugmayr, W. <w.l...@uk...>]

dear all, 

just to finalize - it worked also for me in a fresh scipion install when i did: 

conda activate scipion3 
conda install -y hdf5 
# load my gcc10.3, cuda 11.5, and openmpi with cuda11 enabled 
scipion3 installp -p scipion-em-xmipp -j 12 | tee -a install.log 

all went fine except the -mtune/march parameters do not fit to our environment. 
we have intel and amd nodes in the cluster. 
so i did in the same environment as above: 
cd to_xmipp_location 
./xmipp cleanBin 
vi ./scripts/config.py 
changed: #self.configDict["CXXFLAGS"] += " -mtune=native -march=native -flto" 
to: self.configDict["CXXFLAGS"] += " -mavx2" 
./xmipp config noAsk | tee config.log 
./xmipp compile | tee compile.log 
./xmipp install | tee install.log 

cheers, 
wolfgang 



From: "Alberto Garcia Mena" <alb...@cn...> 
To: "Thu Nguyen" <t.n...@un...>, "Mailing list for Scipion users" <sci...@li...> 
Cc: "Irene" <isa...@cn...> 
Sent: Thursday, 17 March, 2022 11:51:35 
Subject: Re: [scipion-users] [EXT] Re: Problem compiling xmipp in Scipion3 on Centos 7 cluster 

Glad to hear it. 

I'm sure hdf5 creates some problems, also the compiler version. 

Thanks for your patience and for your installation report. [ https://github.com/I2PC/xmipp/wiki/Installing-Xmipp-on-CentOS-7-9.2009 | I
      have updated the wiki page ] for installing Xmipp on Centos7, reporting the problems with hdf5. 

Thanks again for your patience and I hope you enjoy Scipion&Xmipp. 

If you have any other problems related to Scipion / xmipp, please write to us. 




BR 






On 17/3/22 11:18, Thu Nguyen wrote: 



Dear All, 


BQ_BEGIN

Thanks to All that have responded and helped. For a whole week, I have tried without success with countless combinations of gcc, cuda and hdf5. Now I can report that I have "sort-of" solved my own problem. I hope this will help someone who might have the same issue. 



    * Asumming the centos 7 hdf5-devel installed with yum (1.8.12) was not doing its job, I uninstalled it. No trace of hdf5* in /usr/include of /usr/lib64. 
    * I used gcc 8.5.0 to compile and install hdf 1.10.3 from source, installed at /usr/local/hdf5. Please note, I have used this version before, but as rpms from a 3 rd party (Springdale Computational, Princeton); this time I built my own. 
    * I ran ./xmipp cleanAll. This erase averything including contents of the src folder I think. 
    * then ./xmipp config. This time the config said it could not find hdf5 and asked me to put it in, so I was able to specify the include and lib paths to the new hdf5 (usr/local/hdf5/include and lib) 
    * ./xmipp all will redownload everything, compile and install 
I am so glad to finally see the line that said xmipp installed sucessfully. 



In the process I learned to clean up after a failed build; how to rebuild the xmipp.config file and do other things with the ./xmipp. Also when trying with gcc5.3, I got a message complaining about "inline variables", which on my search I read that it is a C++17 feature that gcc only fully supports in their version 8's. That confirms the need for gcc8; I am not sure if my previous use of gcc 7.2.0 contributed to the failure or not (would not have the energy to go back and try !) 

In saying I sort-of solve the problem, the cluster needs packages to be in rpm form to be auto-deployed to compute nodes, so I will have to find a way to get my hdf5 to them, but this is outside the focus of this forum 

Again thank for all your help, specially those who have email me personally with plenty of hints. 







___________________________________________________ 

Thu D. Nguyen 

Bio21 Institute 


The University of Melbourne 












BQ_END

-- 
- Alberto GM - 


_______________________________________________ 
scipion-users mailing list 
sci...@li... 
https://lists.sourceforge.net/lists/listinfo/scipion-users