You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently PatchEnergyJIT is the parent of PatchEnergyJITGPU, but since PatchEnergyJIT maintains CPU compiled code among other things leads this to not be a good paradigm. A PatchEnergyJIT class (distinct from the one above) should be created with the things in common, and PatchEnergyJITCPU and PatchEnergyJITGPU both should inherit from the new PatchEnergyJIT class.
Motivation and context
The text was updated successfully, but these errors were encountered:
We will release 3.0.0 with the JIT as is and labelled experimental. We will delay this refactoring to a future 3.x release.
I do have a question about this though. Currently, PatchEnergyGPUJIT defers some of its computations, such as the total energy to the CPU via base class methods. Do you propose implementing all those methods on the GPU? This pattern is used throughout HOOMD where the base class implements the method on the CPU and the *GPU derived class overrides just a few methods to put the slow parts of the computation on the GPU.
Description
Currently
PatchEnergyJIT
is the parent ofPatchEnergyJITGPU
, but sincePatchEnergyJIT
maintains CPU compiled code among other things leads this to not be a good paradigm. APatchEnergyJIT
class (distinct from the one above) should be created with the things in common, andPatchEnergyJITCPU
andPatchEnergyJITGPU
both should inherit from the newPatchEnergyJIT
class.Motivation and context
The text was updated successfully, but these errors were encountered: