etmc
Refurbish correlator online measurement
we need to refurbish this, because right now it will always use the pure TMWilson operator for the correlator computation. This is not what we need.
we could use the operator struct we anyhow use for invert also in hmc_tm for the online measurements. One would therefore need to specify the operators one would like to have inverted and measured. This would then have effect only for the correlator online measurement, of course.
Maybe we could also make it write into a single file with a header for each write and a double newline between them? I had to modify the code because the lustre file system in Zeuthen couldn't deal with the many files generated during my high statistics run.
in principle we could do that, however, thats always a mess if you restart jobs and redo trajectories. Maybe we could implement the choice?
I agree with you, although in the current situation the default would be to overwrite data with the same trajectory number, no?
If we're happy with your changes I will close this, I guess we will know tomorrow.
Hi, Thank you for solving this issue. As we said in the phone conference I can test it in Fermi (online versus offline correlators). But it was not clear to me what do I have to do in order to get the code. My working copy of the code is cloned from Bartek's: $ git branch
-
interleaving_hs_mergeNDTwistedClover master $ git log -n 1 commit fde90345e53366b0b1ee20796c002a4027040908 Merge: 8b0505e 1896db1 Author: Bartosz Kostrzewa [email protected] Date: Wed Dec 5 17:45:55 2012 +0100
Merge remote-tracking branch 'urbach/NDTwistedClover' into
interleaving_hs_mergeNDTwistedClover
Does the following command bring me the update? $ git pull origin interleaving_hs_mergeNDTwistedClover
Thanks in advance and best regards,
David
On Wed, Dec 12, 2012 at 4:39 PM, Bartosz Kostrzewa <[email protected]
wrote:
If we're happy with your changes I will close this, I guess we will know tomorrow.
— Reply to this email directly or view it on GitHubhttps://github.com/etmc/tmLQCD/issues/169#issuecomment-11293803.
On 13/12/12 16:39, palao wrote:
Hi,
Hi David,
Does the following command bring me the update? $ git pull origin interleaving_hs_mergeNDTwistedClover
yes, that should bring you up to date!
All the best, Bartek
Oh, please note that the debug levels have been adjusted. At level 1 you now have only basic output such as total iteration count. At level 2 you have timings for many common function calls (derivatives and such) and at level 3 there is output for every CG iteration. Effectively what used to be level 2 has become level 3.
Also note that on BG/Q using level 2 or 3 seems to make the code quite a bit slower, probably because the master process is already quite busy and having it produce output just causes everyone else to wait excessively.
the bug is fixed, however, maybe we should also allow to measure the kaon mass online!?
the bug is fixed, however, maybe we should also allow to measure the kaon mass online!?
You mean in the unitary case for 2+1(+1) runs?