-
Did anyone test qio interface in the current master? I got a segfault while tried to read a gauge config via qio. As a bypass I replaced qio stuff with that one from multigrid branch. It seems working…
-
Mobius and 4-d domain-wall fermions need some final testing. In particular some possible issues have been noticed when running with Chroma. Moreover, we need to check that the interface is consisten…
-
Running MILC RHMC on Fermi cards gives unexpected differences when using QUDA HISQ Fermion Force
Delta S from the MILC test example (su3_rhmc_hisq.1.sample_in)
GPU w/ CPU FF + GF: ACCEPT: delta S =…
-
I built `su3_rhmc_hisq` from MILC with quad 0.7 for single precision and tried to run it with the `test/su3_rhmc_hisq.1.sample-in` input.
While it does start it hangs for a while.
If I disable the FF …
-
The twisted-clover dslash kernels are presently generated by hand. This is an issue to create modify the current python scripts so that they can be created automatically.
-
For qudaMultishiftInvert there is a line in milc_interface.cpp
`invertParam.residual_type = (target_fermilab_residual[0] != 0) ? QUDA_HEAVY_QUARK_RESIDUAL : QUDA_L2_RELATIVE_RESIDUAL;`
while for qud…
-
Probably it's a known issue but to keep track of it:
Some of the CUB headers (which are used by ./lib/clover_invert.cu and ./lib/gauge_plaq.cu) make direct use of cudaTextureObject_t which requires co…
-
The extended gauge copy and ghost extraction routines do not presently handle the time boundary conditions correctly. When extended, the boundary condition lies on the the R[3]-1 on the first time sl…
-
non-degenerate dslash operator is currently broken, also a single GPU version . In particular, the single GPU dslash test fails for single precision and reconstruction 12 (while all other options loo…
-
The latest OpenMPI and MVAPICH libraries support GPU pointer directly. This is something we should support in QUDA as soon as possible as it had a huge potential for better scaling:
1. Inter-GPU memo…