tmLQCD is a freely available software suite providing a set of tools to be used in lattice QCD simulations. This is mainly a HMC implementation (including PHMC and RHMC) for Wilson, Wilson Clover and Wilson twisted mass fermions and inverter for different versions of the Dirac operator. The code is fully parallelised and ships with optimisations for various modern architectures, such as commodity PC clusters and the Blue Gene family.
(quoting @kostrzewa from the AMD workshop's QUDA chat)
We should probably do two things:
finally remove the --enable-gpu flag and all the gpu stuff that Florian added about a decade ago. (we can do this later though, not now)
add the option to link against QUDA with different backends
if you look at configure.in line 964, this is where the -lcuda -lcublas linking flags come from. This section should be taken outside the --with-qudadir option and made stand-alone.
Then, one will be able to do something like --with-qudadir=[...] --with-cudadir=[...] for a CUDA build,
while doing something like --with-qudadir=[...] --with-hipdir=[...] for the HIP build.
(quoting @kostrzewa from the AMD workshop's QUDA chat)
We should probably do two things:
if you look at configure.in line 964, this is where the
-lcuda -lcublas
linking flags come from. This section should be taken outside the--with-qudadir
option and made stand-alone.Then, one will be able to do something like
--with-qudadir=[...] --with-cudadir=[...]
for a CUDA build, while doing something like--with-qudadir=[...] --with-hipdir=[...]
for the HIP build.