Jonathan R. Madsen

> Use cases: > > * often there are significant run-to-run variation of an application due to the inherent randomness, e.g., for Monte-Carlo simulations. Well, realistically, a Monte Carlo application...

Why do we need to even rely on rocprof to do application replay? Doing a whole application replay is trivial to implement without forking. LD_PRELOAD library with wrapper around `__libc_start_main`...

Basically, you'd just build a library with something like [main.c in omnitrace]( https://github.com/AMDResearch/omnitrace/blob/main/source/lib/omnitrace-dl/main.c) and implement that logic after the call to [main_real](https://github.com/AMDResearch/omnitrace/blob/2449e7cd462672e3ff9f040c710dcac236af28be/source/lib/omnitrace-dl/main.c#L102)

[execve](https://man7.org/linux/man-pages/man2/execve.2.html) basically replaces the current program with a new program: > execve() executes the program referred to by pathname. This causes the program that is currently being run by the...

> as it's the one who's actually cycling through various sets of counters. This doesn't sound particularly complicated to me once you figure out the number of HW counter slots...

Just chiming in out of curiousity. > - Go through each of the vendor connectors and ensure the corresponding functionality is the same What does this mean? I think the...

@daboehme I think you'd have to use the C APIs of nvidia-smi/rocm-smi and poll the current memory usage (potentially in a background thread), keeping track of the max. I am...

Hi @rbberger, sorry for the delay, I was on vacation last week. It was always my intention to enable this but it got sort of lost in all the changes...

Interesting, and no worries, I enjoy a semi-related rambling. Luckily, we won't run into this issue though. The include paths of `` vs. `` are accounted for in the [Findroctracer.cmake](https://github.com/AMDResearch/omnitrace/blob/main/cmake/Modules/Findroctracer.cmake)...

Add `OMNITRACE_PAPI_EVENTS = amd64_rapl::RAPL_ENERGY_PKG` to a config file and you should see them in the trace timeline... assuming your machine has the privileges to read them (which it sounds like...