Konstantin devel Wilson+Clover N_f=2+1 implementation#673
Conversation
…ed explicit "single_flavor" flag in monomials to e.g. fix issue with double counting of trlog term and make the code more expressive in several places.
|
@chaoos @mtaillefumier any thoughts on the best order of merging this or #664 ? |
|
@kostrzewa If this is a quick and small thing, I'd merge it. Yes, it gives conflicts on the cmake branch, but else we have to suspend all merges until cmake is merged (which might still take a while?). |
@chaoos compared to the cmake_support branch the number of changes here is small |
|
Please merge it. I will resolve the merge conflicts afterwards. |
…QUDA_MATPC_ODD_ODD_ASYMMETRIC in quda_interface.h. This does not seem to affect the simulation but more correctly reflects the theoretical setup.
monomial/rat_monomial.c
Outdated
| // if(mnl->type != CLOVERRAT) phmc_compute_ev(hf->traj_counter-1, id, &Qtm_pm_ndbipsi); | ||
| // else phmc_compute_ev(hf->traj_counter-1, id, &Qsw_pm_ndbipsi); | ||
| if(mnl->type != CLOVERRAT) { | ||
| phmc_compute_ev(hf->traj_counter-1, id, &Qtm_pm_ndbipsi); |
There was a problem hiding this comment.
@kottnad could you please add a comment here specifying that this call is technically not correct but that when running with QUDA the EV computation will work nonetheless?
|
Thanks for this. Could you please apply clang-format to the files that you modified and add the comment that I mention above? Down the line (not in this PR) we will need to review the non-HMC inversion pathway but we need to design a test for that. Already have something in mind. |
|
Thanks! |
4 participants
Provides a working implementation of N_f=2+1 (plain) Wilson+Clover simulations.
Still lacks a GPU implementation of the force for the single flavor part though.