Hotfix/mg - Githubissues

lattice / quda

QUDA is a library for performing calculations in lattice QCD on GPUs.

https://lattice.github.io/quda

Other

279 stars 94 forks source link

Hotfix/mg #1439

Closed maddyscientist closed 5 months ago

maddyscientist commented 5 months ago

This is a bugfix PR, and close #1438. This bug arose because of the temporary removal in #1434, which took care of precision conversion.

Adds support to the fine-grid restrictor to directly read from half-precision fields
blas::caxpyNorm is now a wrapper to the more general function blas::caxpbyNorm function (which support mixed precision)
blas::xmyNorm and blas::max_deviation now support mixed precision

maddyscientist commented 5 months ago

Further testing revealed that there was a bug in the half-precision staggered colorspinor::AccessorCB::load function. With this fixed, we can also enable restriction from half precision fields in the staggered case, which I have now enabled.