Page 1 of 1

Testing EPW-4.1 installation

Posted: Sat Aug 12, 2017 9:22 pm
by Fedor
Dear EPW developers and users,

Since a few months we have a simple graphical interface for QE6.0 for mac, windows & linux and provide (openMP-parallelized) binaries with the ADF Modeling Suite.
Because there was interest in EPW, I've asked our developers if they could try to compile with ph and epw include on those platforms.
That all seemed to work OK, when making the following changes

---For Windows---
(for main PW.x - see changes here: https://www.scm.com/support/downloads/g ... -espresso/)
additionally:
cd wannier90-2.0.1
cp config/make.sys.gfort make.sys
vim make.sys # change the LIBDIR at the bottom to point to /c/64/software/qe-6.0/LAPACK
make all
cd ..
make epw
---MacOS---
vi install/make_wannier90.inc # , replace LIBS with LIBS = /opt/intel/compilers_and_libraries_2017.1.126/mac/mkl/lib/libmkl_blas95_ilp64.a /opt/intel/compilers_and_libraries_2017.1.126/mac/mkl/lib/libmkl_lapack95_ilp64.a -L/opt/intel/compilers_and_libraries_2017.1.126/mac/mkl/lib -Wl,-rpath,/opt/intel/compilers_and_libraries_2017.1.126/mac/mkl/lib -lmkl_intel_ilp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -lm -ldl
make epw
--Linux--
vi install/make_wannier90.inc # change blas to blas95, and lapack to lapack95
make epw

I tested a little bit on Windows a few weeks back with the epw_base in the test-suite (http://qe-forge.org/gf/download/frsrele ... ite.tar.gz) but couldn't work my way to why the pp.py script would not work (I understand it corresponds to the B-doped diamond tutorial).

I'm now trying on a macbook by just running from the cmd line
pw.x <scf.in > scf.out
ph.x <ph.in > ph.out
that all works fine
then I copy the diam.dyn? and _ph0/diam.dvscf and _ph0/diam.phsave and diam.q_? to /save/
and run scf_epw.in and nscf_epw.in.
Those also run fine.

Contents of save and diam.save:
Fedors-MacBook-Pro:epw_base fedor$ ls -l save
total 18448
-rw-r--r-- 1 fedor staff 5308416 Aug 12 22:25 diam.dvscf
-rw-r--r-- 1 fedor staff 310 Aug 12 22:23 diam.dyn0
-rw-r--r-- 1 fedor staff 2855 Aug 12 22:23 diam.dyn1
-rw-r--r-- 1 fedor staff 10338 Aug 12 22:23 diam.dyn2
-rw-r--r-- 1 fedor staff 8200 Aug 12 22:23 diam.dyn3
-rw-r--r-- 1 fedor staff 14614 Aug 12 22:23 diam.dyn4
-rw-r--r-- 1 fedor staff 8207 Aug 12 22:25 diam.nnkp
drwxr-xr-x 28 fedor staff 952 Aug 12 22:25 diam.phsave
drwxr-xr-x 5 fedor staff 170 Aug 12 22:25 diam.q_2
drwxr-xr-x 5 fedor staff 170 Aug 12 22:25 diam.q_3
drwxr-xr-x 5 fedor staff 170 Aug 12 22:25 diam.q_4
-rw-r--r-- 1 fedor staff 4071168 Aug 12 22:25 diam.wfc
-rw-r--r-- 1 fedor staff 179 Aug 12 22:25 diam.win

Fedors-MacBook-Pro:epw_base fedor$ ls -l diam.save/
total 624
-rw-r--r-- 1 fedor staff 78562 Aug 12 22:17 C_3.98148.UPF
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00001
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00002
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00003
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00004
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00005
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00006
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00007
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00008
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00009
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00010
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00011
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00012
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00013
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00014
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00015
drwxr-xr-x 5 fedor staff 170 Aug 12 22:05 K00016
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00017
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00018
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00019
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00020
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00021
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00022
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00023
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00024
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00025
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00026
drwxr-xr-x 5 fedor staff 170 Aug 12 22:17 K00027
-rw-r--r-- 1 fedor staff 113796 Aug 12 22:17 charge-density.dat
-rw-r--r-- 1 fedor staff 65389 Aug 12 22:17 data-file.xml
-rw-r--r-- 1 fedor staff 56251 Aug 12 22:17 gvectors.dat

now I run
epw.x < epw.in > epw.out

However, this ends with an error:
Running Wannier90
Error allocating fermi_energy_read in param_read
Error: examine the output/error file for details

Could that have to do with replacing the BLAS/LAPACK Libs in Wannier90? Or did I just not put the right files in the right place for the EPW calculations?

Thanks in advance for your insight and help!
-Fedor

Re: Testing EPW-4.1 installation

Posted: Sun Aug 13, 2017 10:27 am
by sponce
Dear Fedor,

Thank you for your interest in EPW.

I have never tried EPW on Windows (we do not support it). However, I'm interested if you manage to make it work.

The python pp.py script is only copying and renaming files generated by the ph.x code and place then in the "save" folder. Your save folder seems fine to me.

May I suggest you to try the EPW test from the test-suite. Those tests are indeed tested nightly and should be working on different compilers/lib (on linux) [1].

You can find them in the quantum-espresso folder under /test-suite. If you download from the webpage, you need to download the test-suite package:
qe-6.1-test-suite.tar.gz and just un-tar it into the main qe folder.

from

http://www.qe-forge.org/gf/project/q-e/ ... kage_id=18

You can then run automatically the EPW tests by running:
make run-tests-epw-serial

It will take care of copying everything and will also verify the numerical accuracy.

Regarding your error in Wannier. Are you using the latest wannier 2.1.0 version? This should be downloaded automatically when you do "make epw" inside QE.

In principle you should not have to replace blas by blas95 in linux. You are using GCC ? If you do not specify libraries, it should use internal blas/lapack.
Let me know if you have any further issue.

[1] Fair warning. There was an issue when releasing the QE version 6.1 (along with version 4.1 of EPW). This was patched (see http://epw.org.uk/Main/DownloadAndInstall).
As a result, the reference files from the test-suite will have expected numerical inaccuracies (the code will run but the results will be off with respect to the reference benchmark).
This will soon be resolved as there will be a new release of QE August/Sept 2017.
In the meantime you can download the development version of QE on Github https://github.com/QEF/q-e
The test-farm there should be correct.


Best,
Samuel

Re: Testing EPW-4.1 installation

Posted: Sun Aug 13, 2017 8:12 pm
by Fedor
Dear Samuel,

Thanks for the quick reply. Note this is about EPW 4.1 (with qe-6.0).
We haven't yet updated to qe-6.1 as it was released just after we had a major new release ourselves. When we have next major new release (Mar/Apr 2018?) we'll look at the latest major QE release (including EPW) shortly before (qe-6.2 or even 6.3 ?).

I didn't do the compilation myself so I don't if the latest Wannier is picked up (for Windows I see 2.0.1 was used). We do tend to compile as much as possible with ifort, icc and mkl to make faster code.
The Windows QE (6.0) binaries seem to work fine for most regular calculations, and now also phonons.

I'll ask our developers to run the (6.0) test suite, and see if it chokes on EPW / Wannier on all platforms.
If you have a Windows machine at hand, you are more than welcome to try out the QE 6.0 binaries, including the ones with EPW 4.1 (which are not yet public).

Best wishes,
Fedor

Re: Testing EPW-4.1 installation

Posted: Mon Aug 14, 2017 8:37 am
by sponce
Dear Fedor,

There was bugs in the library version of Wannier 2.0.1. Those were corrected by copying some files from EPW/src to Wannier. This was taken care by the Makefile during the installation but required to run "make" twice. This has been solved in qe 6.1/EPW 4.2 and Wannier 2.1.0.

Best,
Samuel

Re: Testing EPW-4.1 installation

Posted: Mon Aug 14, 2017 8:39 am
by Fedor
Hi Samuel,

OK that sounds like a potential culprit. I'll get our developers to have a look at dropping in a more recent Wannier.f90
-Fedor