quda
quda copied to clipboard
lattice
QUDA is a library for performing calculations in lattice QCD on GPUs.
There's currently an all-to-all of to, from precisions for the MG `copy_color_spinor_mg.*` files. Historically this split was done for compile time reasons. We should investigate if this is still a...
The routines to: * Add the identity to the coarse clover * Add a staggered mass to the coarse clover * Add a twist to the coarse clover Are all...
We use a version based on 1.8 with some QUDA specific additions. Maybe they are no longer necessary.
This is something to do post GK merge. Some possible options include ``` quda_parallel quda_offload QUDA_PARALLEL QUDA_OFFLOAD ``` For aesthetic reasons I prefer lower case, but understand the reasoning for...
This is a task to be done after generic kernels is merged, before HIP support is merged in. CUB is the outstanding CUDA device library that has not been abstracted...
https://github.com/lattice/quda/blob/ff2a4068eb4e571b8e357f73d1e67cc827131ef2/lib/solver.cpp#L41-L47 Should this include a test for `QUDA_CA_GCR_INVERTER`?
Revert #1152 (or re-write the files as appropriate) once the multi-rhs work is done.
Currently `struct Aggregates` in `restrictor.cu` and `struct OrthoAggregates` in `block_orthogonalize.cu` are essentially identical. They could be unified using a Tag template. E.g. in `include/powers_of_two_array.h` ``` template struct TaggedAggregate { /*...
This would allow us more flexibility in algorithm exploration, and enable more efficient compilation.
Enabling `GPU_MULTIGRID_DOUBLE` in the generic_kernels branch cause a build fail with the following error: ``` /scratch/CPviolator/work/QUDA_COMPRESSED_TRLM/quda/lib/../include/kernels/coarse_op_preconditioned_mma.cuh(122): error: argument of type "double *" is incompatible with parameter of type "float *"...
