Vyas Ramasubramani
Vyas Ramasubramani
Well then... looks like we've got to work our way all the way up the stack for this. For the purpose of something like clang-tidy we might be able to...
Compile times are an ever-present problem for us. This issue as currently framed isn't clearly actionable, so let's lay out some concrete points. > We should make sure that best...
The async resource will use the pool managed by the CUDA driver, which we do not own and would probably be fine. Ideally everyone would use that and then all...
My mistake, I didn't realize that we were allocating from a specific pool that we created. The failure mode should still be relatively graceful if two processes both use the...
I think that in order to answer this question effectively we will need to rerun a lot of benchmarks. The original allocator design was many GPU generations ago, and I...
I never looked into what the `replay` benchmark was supposed to do, that does sound very helpful. Having rmm devs sit down and take stock of what we want to...
Does @wence- have time to finish this or do we want someone else to take over and get this past the finish line? I think the remaining changes are all...
Yes, I agree with that approach. Regarding with de-templatizing, we will have to work through the various parts of rmm piecemeal, but definitely starting with the memory resources makes sense....
> Is there an issue for the equivalent feature for `rmm::device_scalar`? Nope, please feel free to create one.
Great thanks for the note!