Modifications to add use_smeared_gauge to InvertParam #1522
Conversation
There was a problem hiding this comment.
@cjmorningstar10 do these changes fix the issues that were being seen?
include/quda.h
Outdated
| @@ -507,7 +511,7 @@ extern "C" { | |||
|
|
|||
| /** Whether to use the smeared gauge field for the Dirac operator | |||
| for whose eigenvalues are are computing. */ | |||
| bool use_smeared_gauge; | |||
| QudaBoolean use_smeared_gauge; | |||
There was a problem hiding this comment.
Why the change to QudaBoolean from bool? I intentionally used bool with this variable as we intend to deprecate QudaBoolean.
There was a problem hiding this comment.
Also I think with this addition of the use_smeared_gauge parameter to QudaInvertParm the QudaEigParam variant is never used now? Can we just delete it?
There was a problem hiding this comment.
I used QudaBoolean because all the other boolean members did...it looked weird to see a difference. But bool is fine. The use_smeared_gauge is needed in the InvertParam because only that gets sent to the setDiracParam functions, where it is needed. It is ok to keep it in the EigParam too. I think I put in a check that these two bools have to be the same.
|
cscs-ci run |
No. The main reason for this PR is to enable computing the LapH eigenvectors if you read the smeared gauge and NOT the original gauge. Once the perambulators are computed, the original gauge field is never needed any more. When we start evaluating correlator functions, we will just read the smeared gauge, compute the Laph evs on the fly, then start computing baryon triplets/meson doublets/ then the correlators. Even displacements in the operators need only the smeared gauge field. No need to read the original gauge, smear it, then .... Just cuts out one step. No need to waste device mem on the original gauge. This is a minor issue, but .... why not support this choice? |
Added use_smeared_gauge to QudaInvertParam to get rid of awkward extra bool in some routines. In setDiracEigParam(...), added code to allow this to run without needing gaugePrecise having already been set. Some first tests have been done, but would like feedback on how this was done! These changes facilitate computation of the LapH eigenvectors using the smeared gauge field.