Page 1 of 1

p2y segmentation fault with yambo-3.4.1-rev61 and QE-5.0.3

Posted: Tue Mar 03, 2015 10:47 am
by Philippe
Dear all,

I compiled Yambo on JUROPA, linking with Intel MKL libraries. When running p2y I obtain a segmentation fault on exiting the function 'pickup_a_random' (mod_logo.F). This error disappears when I configure without lapack and scalapack. Any ideas?

I've attached the config.log as well as the output given by valgrind.

Thanks and best regards,
Philippe

Re: p2y segmentation fault with yambo-3.4.1-rev61 and QE-5.0

Posted: Tue Mar 03, 2015 3:12 pm
by Davide Sangalli
Dear Philippe,
it is strange. We wil try to investigate the problem. Could you provide your input files and pseudo-potentials of your DFT run.

Meanwhile, you could fix the problem replacing the line "serial_number=pickup_a_random(10000._SP)" with "serial_number=1325" or some other number you like.
Keep in mind that the serial number is used by yambo to check that a user is not mixing databases from different runs.
So if you keep a fix number this check will not exist any-more in practice, but you should not have any other problem.

Regards,
D.

Re: p2y segmentation fault with yambo-3.4.1-rev61 and QE-5.0

Posted: Wed Mar 04, 2015 12:23 pm
by Conor Hogan
Dear Philippe,
My guess is that, since the error comes from lapack, etc, the problem is in the call to the dlaran random number generator inside pickup_a_random (src/modules/mod_logo.F). Could you try to replace

pickup_a_random=int(dlaran(iseed(4:))*rand_range)

with

pickup_a_random=int(0.5*rand_range)

and compile with full linking to MKL?
Dlaran is used elsewhere in the code so it's important to isolate this.

Re: p2y segmentation fault with yambo-3.4.1-rev61 and QE-5.0

Posted: Wed Mar 04, 2015 1:07 pm
by Philippe
Dear Davide,

thank you, setting the serial number manually and removing 'write_the_logo' fixed the problem.

However, now p2y stops with
[ERROR] STOP signal received while in :
[ERROR] WF databases too old. Please regenerate them with a newer Yambo release

Doing 'ncdump ns.db1 ' returns
netcdf ns {
dimensions:
D_0000000017 = 17 ;
variables:
float DIMENSIONS(D_0000000017) ;
data:

DIMENSIONS = 1325, 0, 0, 0, 180, 260, 443527, 55580, 1, 2, 1, 1, 0, 264, 3,
1009, _ ;
}

I tried with different versions of NetCDF and also without, but no success.

As requested, I've attached the PW input and pseudo potentials (as .txt files, otherwise they wouldn't be accepted).

Best regards,
Philippe

Re: p2y segmentation fault with yambo-3.4.1-rev61 and QE-5.0

Posted: Wed Mar 04, 2015 1:32 pm
by Conor Hogan
Another workaround, for now, is to compile p2y without the libraries, and compile yambo with them. The databases will not change, and at least you should have no problems with serial numbers. p2y is just a simple convertor to be run once.

Re: p2y segmentation fault with yambo-3.4.1-rev61 and QE-5.0

Posted: Wed Mar 04, 2015 2:55 pm
by Philippe
Dear Conor,

you're right, the problem does originate from dlaran. In fact, in the config.log it says

/tmp/ifortRcOuF5.o: In function `MAIN__':
conftest.f90:(.text+0x10): undefined reference to `dlaran__'
configure:8893: $? = 1
configure: failed program was:
| program main
| call dlaran_
| end

I somehow missed that error message before, sorry.

Best regards,
Philippe