VASP_6.1.2_patched Intel Compiler Error in Test Suite

[robert_molt]

Good evening,

I wish to resolve the following error observed when the test suite is executed:

SUMMARY:

==================================================================

The following tests failed, please check the output file manually:

bulk_GaAs_G0W0_sym bulk_GaAs_G0W0_sym_RPR bulk_GaAs_scGW0_ALGO=D_sym bulk_GaAs_scGW0_ALGO=D_sym_RPR bulk_GaAs_scGW0_sym bulk_GaAs_scGW0_sym_RPR bulk_GaAs_scGW_ALGO=D_sym bulk_GaAs_scGW_ALGO=D_sym_RPR bulk_GaAs_scGW_sym bulk_GaAs_scGW_sym_RPR bulk_InP_SOC_G0W0_sym bulk_InP_SOC_G0W0_sym_RPR bulk_SiO2_LOPTICS bulk_SiO2_LOPTICS_RPR bulk_SiO2_LPEAD bulk_SiO2_LPEAD_RPR SiC8_GW0R Tl_x Tl_y Tl_z

The common variable seemingly being GW calculations. This is the result of running the test suite on VASP 6.1.2 patched on intel compilers. There was an error in compilation, but I would think it should have nothing to do with this. The error I received was:

fortran.c(68): catastrophic error: cannot open source file "cuda_runtime.h"

#include <cuda_runtime.h>

^



compilation aborted for fortran.c (code 4)

make[3]: *** [obj/x86_64/release/fortran.c.o] Error 4

make[3]: Leaving directory `/ensco/mb-nss-unix/molt.robert/VASP_Compile_2/src/vasp.6.1.2_patched/build/gpu/CUDA'

make[2]: *** [release] Error 2

make[2]: Leaving directory `/ensco/mb-nss-unix/molt.robert/VASP_Compile_2/src/vasp.6.1.2_patched/build/gpu/CUDA'

make[1]: *** [CUDA] Error 2

make[1]: Leaving directory `/ensco/mb-nss-unix/molt.robert/VASP_Compile_2/src/vasp.6.1.2_patched/build/gpu'

make: *** [gpu] Error 2

I was surprised to get this error at all; I only specified the makefile for intel compilers, and did not think it would assume to use cuda. I therefore dismissed this error as the makefile being overzealous in trying out the GPU that was not there. Am I wrong, perhaps, and a subsequent part of the compilation post-GPU was neglected? Specifically, I copied makefile.include.linux_intel as my makefile.include . I also used intel64, if that matters.

I have tried to add the OUTCAR as an attachment, either with no file extension or as a .txt; the forum will not let me upload either. Each time I get a message of Invalid File Extension: OUTCAR.txt. If someone can advise, I would be happy to post the OUTCAR of the first failed test calculation, as well as the testsuite.log file.

[henrique_miranda]

Regarding your point as to why the test suite fails, could you maybe post the log file + the OUTCAR of one of the failing tests?

As for the CUDA error, you are probably compiling the code with:

Code: Select all

make all

This will try to build all the versions of VASP including the GPU port.
If you don't need the GPU port I recommend that you compile only the CPU version of the code:

Code: Select all

make gam std ncl

The two issues are unlikely to be related.
Once you post the files I suggested above I will try to help you to track down the reason why the tests are failing on your system.

[robert_molt]

Pursuant to the above:

"I have tried to add the OUTCAR as an attachment, either with no file extension or as a .txt; the forum will not let me upload either. Each time I get a message of Invalid File Extension: OUTCAR.txt. If someone can advise, I would be happy to post the OUTCAR of the first failed test calculation, as well as the testsuite.log file."

What is the proper file extension that this forum will allow me to attach the files?

[henrique_miranda]

Only zip files are allowed
As written in the forum posting guidelines:
https://www.vasp.at/forum/viewtopic.php?f=4&t=17928

[robert_molt]

My mistake; I apologize for failing to follow the rules. I have now successfully attached the test.log record and a sample output file. It does not have an obvious error in electronic structure, but I am a neophyte to periodic electronic structure.

[henrique_miranda]

It is a bit difficult to know exactly what is going wrong from the backtraces you posted:

Code: Select all

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image              PC                Routine            Line        Source
vasp_ncl           0000000001C5A9D5  Unknown               Unknown  Unknown
vasp_ncl           0000000001C585F7  Unknown               Unknown  Unknown
vasp_ncl           0000000001BC0414  Unknown               Unknown  Unknown
vasp_ncl           0000000001BC0226  Unknown               Unknown  Unknown
vasp_ncl           0000000001B5BDD6  Unknown               Unknown  Unknown
vasp_ncl           0000000001B639F0  Unknown               Unknown  Unknown
libpthread.so.0    00007FDA41D5B630  Unknown               Unknown  Unknown
vasp_ncl           00000000006705A0  Unknown               Unknown  Unknown
vasp_ncl           00000000006CB504  Unknown               Unknown  Unknown
vasp_ncl           0000000000ECCD21  Unknown               Unknown  Unknown
vasp_ncl           0000000000F4C4C9  Unknown               Unknown  Unknown
vasp_ncl           000000000198D4A3  Unknown               Unknown  Unknown
vasp_ncl           00000000019686ED  Unknown               Unknown  Unknown
vasp_ncl           000000000040AF9E  Unknown               Unknown  Unknown
libc.so.6          00007FDA4169E555  Unknown               Unknown  Unknown
vasp_ncl           000000000040AEA9  Unknown               Unknown  Unknown

Could you post here the makefile.include you used? I will try to make a suggestion based on that.
Could you maybe also re-compile with the '-traceback' or '-g' options for the intel compiler and re-run for a couple of the tests that failed?

[robert_molt]

The makefile is attached. I will also do as you suggest and begin compiling with the suggested flags; I will report the result as soon as it is available. I thank you for your help in this matter!

[robert_molt]

I apologize for my ignorance, but I do not understand exactly what is meant by -traceback or -g, after all. I assumed this was intended to be
make all -traceback
or
make all -g
but this is clearly not what you intended, since those flags have no meaning when I go to compile. Did you want me to do the traceback when I execute the test suite?

[henrique_miranda]

Don't worry, maybe I should have explained better.
What I meant was to replace FFLAGS and OFLAG in your makefile.include:

Code: Select all

FFLAGS     = -traceback -debug -g -assume byterecl -w
OFLAG      = -O0

Note that these are debugging flags just to help us to understand where the problem is coming from.
For production code, you should not use these as the code will run slower.

I had a look at the makefile you sent and it looks correct to me.
Can you indicate what version of intel and mkl you are using?

Code: Select all

 mpiifort -v

[robert_molt]

Gotcha, now I understand the compiler flags. I just started the recompilation; I will post the results when it finishes (will take a few hours).

Code: Select all

[molt.robert@beluga0 vasp.6.1.2_patched]$ mpiifort -v
mpiifort for the Intel(R) MPI Library 5.1.3 for Linux*
Copyright(C) 2003-2015, Intel Corporation.  All rights reserved.
ifort version 16.0.3

is the compiler version

[henrique_miranda]

Have you tried using a more recent version of the intel compiler?
Alternatively, you could try to build with the GNU toolchain (as an additional check if the error persists).
https://www.vasp.at/wiki/wiki/index.php/Toolchains