add 'tm_rho' parameter for Hasenbusch mass preconditioning for twiste… #15
Conversation
…d clover HMC using tmLQCD
|
The |
| @@ -342,6 +342,8 @@ void printQudaInvertParam(QudaInvertParam *param) { | |||
| P(m5, INVALID_DOUBLE); | |||
| P(Ls, INVALID_INT); | |||
| P(mu, INVALID_DOUBLE); | |||
| P(epsilon, INVALID_DOUBLE); | |||
There was a problem hiding this comment.
It was somehow never added. None of these are really needed, but it makes a cleaner parameter struct if it can be properly initialised with default values and the checks will catch incorrect entries where possible.
|
We have a problem now though, as the QUDA-MG in the generic_kernel framework is not done yet, which means that we either get the 1+1 twisted clover operator (which I based on the generic_kernel branch) or the MG, but not both right now...
|
2-level MG works, however. |
…d clover HMC using tmLQCD
@Marcogarofalo @sunpho84 After some more thought I realised that as far as I understand, we won't be able to rely on the
rhoparameter in the clover accessor for our purposes.I've added a
tm_rhoparameter to QUDA which acts like a twisted mass in the diagonal of the asymetrically preconditioned twisted clover operator (but not in the inverse of the diagonal).I think this doesn't break anything (unless one sets
tm_rhonon-zero). If it does break anything, the PR acceptance tests in the QUDA repo should uncover any breakage.