Minutes 2015 04 23 - lattice/quda GitHub Wiki
Minutes 2015-04-23
Participants
- Steve, Alexei, Carleton, Nuno, Alex, Thorsten, Mathias
Issues discussed
MILC interface changes
tracked in #220
- aim at next release
- not pressing right now but requires changes to MILC or
- defining a version 2 interface and keeping the old one for compatibility ?
- QDP interface should server as example
MILC clover issues
tracked in #224
- BiCG seems to converge but residual reported after inversion is bad (on the order of the estimate when missing clover term)
- inversions run with reliable updates in full double precision, 25 iterations w/ 15 reliable updates
- It is possible to pass clover coefficient and / or clover field
- if clover field is NULL QUDA calculates it using the coefficient (and print that to its output)
- different convention ?
- Alexei will test inversion from qudainvert
- we need to check BiCG with fermilab residual, probably completely untested
- Wilson-Colver tests
- Alex tested Clover with CG
- Thorsten uses Clover from Chroma
Numa issues
tracked in #223
- Don will look into this
CUDA memcheck with MPI
tracked in #222
- seems like issues with getting pointer attributes within MPI
- Nuno tested and will close that issue w/ comments on details
Pull requests
* postponed to next call
Other news
- TML QCD interface from Mario Schroeck
- probably still untested
Follow up in next call (May 7th)
- clover issues
- eigcg mixed precision with HISQ (Alexei reported and will create an issue to track this)
- numa issue status
- regression tests