quda
quda copied to clipboard
lattice
QUDA is a library for performing calculations in lattice QCD on GPUs.
There appears to be a suspicious generator statement in `lib/CMakeLists.txt` in [feature/generic_kernel] which looks like so: ```bash # set up QUDA compile options target_compile_definitions( quda PRIVATE $ $ $ $...
With the fixes to UCX regarding its inability to deal with reused CUDA IPC memory handles (https://github.com/openucx/ucx/issues/3192) we should be able to remove the default behaviour `QUDA_ENABLE_P2P=7` which essentially forbids...
With JITIFY on, the generic_kernels branch will fail due to a vestigial instance of `constant_param_t` at ``` /scratch/CPviolator/work/QUDA_GENERIC_KERNELS/quda/lib/targets/cuda/jitify_helper.cpp:76 ```
Right now it saves a `double` precision gauge field regardless of the input precision---I need to fix that.
See https://github.com/Neargye/magic_enum This allows things like ```cpp enum class Flags { A = 1
NVSHMEM enabled Dslashes (#1062) add an additional `in_pack` to the Arg which is basically identical to the input `in` field but needed to reset the ghosts differently.
Reminder to myself. Since it was made generic, some kernels based on `transform_reduce` are seeing significant regressions. This is because it maps to the multi-reduction kernel, which limits the number...
To complete the suite of gauge tools we need to add n-HEX and n-HYP smearing for measurements, and a recursive smearing framework for HMC. n-HEX and n-HYP will be straightforward...
Improvements to split grid in the future: - Add support for split grid + multi-shift. It should be straight forward. - Add support for split grid when the number of...
Described in title