bdenhollander
bdenhollander
`integrateLangevinMiddlePart3` has this little gem: ```C++ void CommonIntegrateLangevinMiddleStepKernel::execute(ContextImpl& context, const LangevinMiddleIntegrator& integrator) { ... // Reduce UI lag. #ifdef WIN32 cc.flushQueue(); #endif } ``` Commenting out `cc.flushQueue();` eliminates the `integrateLangevinMiddlePart3`...
I know I broke the simulation...but in the process I think I confirmed `enqueueReadBuffer` and `downloadCountEvent.wait()` introduce pauses.
> Original: 320 ns/day > Download the interaction count, but don't wait on the event: 340 ns/day > Don't download it: 406 ns/day My results are similar: Original: 300 ns/day...
> The download causes about a 25% slowdown, even though it's supposed to happen asynchronously and nothing ever depends on it having completed. The C++ call is non-blocking but the...
AMD definitely supports multiple streams and several HIP optimization talks mention it as a good idea. In the profiler on Windows I can see multiple kernels executing simultaneously when `usePmeStream`...
RX 6600 is not listed as supported but HIP SDK 5.5.1 works just fine for me on Windows 10. Did you try it and confirm it's not working?
> Are you doing the development with rocm or just running programs that uses rocm? If the SDK works well, I'd like to see if all the libraries are supported...
@jcoffland I returned [P13426 R5990 C18 G4](https://apps.foldingathome.org/wu#project=13426&run=5990&clone=18&gen=4) and noticed that Gen 5 was already done before I was assigned Gen 4. Earlier Gens were completed several times. Gen 0 ...
Received [P13426 R5308 C24 G4](https://apps.foldingathome.org/wu#project=13426&run=5308&clone=24&gen=4), which already has 2 successful returns and Gen 5 has already been returned twice. Does the work server check for the presence of a result...
> @jcoffland I returned [P13426 R5990 C18 G4](https://apps.foldingathome.org/wu#project=13426&run=5990&clone=18&gen=4) and noticed that Gen 5 was already done before I was assigned Gen 4. Earlier Gens were completed several times. There's another...