OSDN
> Finden Software
> Scientific/Engineering
> Physics
> Superconducting Toolkit
> Foren
> SCTK_forum
> ** Some questions about parallel **

Dear developers, Recently, I tried to calculate the tutorial in SCTK/examples/MgB2/. And I have a problem. When calculating the screened Coulomb interaction, if I use the command " mpiexec -np 32 PATH/sctk.x -nk 32 -in sctk.in > kel.out" It will work. But if I change the number of np and nk. It will have some problems.

For example, I'd like to use three nodes in my real work, and every node has 48 tasks. So I usually make np=144 and nk=6 in every step. But when I use the parameter to create kel.out. The kel.out will stop under the line: "##### Compute K_el #####

q-point : 1

Number of PWs for W : 1

Total RAM for Vscr : 0.26E-06 GB"

And say

" Intel MKL ERROR: Parameter 10 was incorrect on entry to ZGEMM ."

I think the problem is because I change the number of np and nk. So, I did some tests. First, I think if there is a rule that np=nk, so I make np=144,nk=144. But it still has the problem.

Then I try to use one node, make np=nk=48. It works, but in my other work, even I use one node and make np=nk=48, it also does not work. I am really confused about that.

And then, I try np=48 nk=6. There is another kind of error that I also meet in my real work. The output file kel.out will stop in : " ##### Compute K_el #####

q-point : 1

Number of PWs for W : 1499

Total RAM for Vscr : 0.58E+00 GB"

And say
" *** Error in `/work/software/SCTK/sctk/bin/sctk.x': malloc(): memory corruption: 0x0000000004732550 ***"
`

So, I want to know why this error will happen. If it is because I chose the wrong number of np and nk? And if so, if there is a rule to choose it?

I look forward to hearing from you .

Sincerely, Wenbo Zhao

Reply To Message #89006 Dear developers,

I have solved this problem. It happened in "SUBROUTINE fft_wfc()." In this subroutine, when "cnt=0", it will make some error in "igmin" and "igmax" matrix. " igmin = MINVAL(igv(1:3, 1:npwxtot, 1:cnt, 1:2))

igmax = MAXVAL(igv(1:3, 1:npwxtot, 1:cnt, 1:2)) "

I have another question if SCTK can output the averaged phonon frequencies. I didn't find it in the standard output.

Look forward to your reply. Sincerely, Wenbo Zhao

Reply to #89006