Jonathan R. Madsen

But in the meantime, I’ll just let you know that you’ll probably want to try to play with LD_LIBRARY_PATH to get Omnitrace to use the same ROCm libraries as PyTorch,...

Well actually, I’ve probably got enough time now. The fundamental problem I’ve seen with some PyTorch apps in the past is that PyTorch has an RPATH to the ROCm libraries...

Could you do me a favor and run your app normally (without Omnitrace) and before it exits, print out `/proc/self/maps`? If there aren’t any ROCm libs loaded, the last comment...

And for the record, the way we are addressing this issue in the new rocprofiler (which combines the capabilities of roctracer and rocprofiler) is that rocprofiler doesn’t link to the...

Sorry for the delay. I believe, based on the backtrace, you need to make the code importable with “main” support (i.e. wrap the code in a function and only have...

On the plus side, the approach usually makes the traces much cleaner because you can easily put the `@profile` decorator above `def run():` and use the `-b` option to omnitrace-python...

Example: ```python import torch @profile def run(): n_warmup = 1 m = 64 n = 64 k = 64 dtype = torch.float32 A = torch.rand(m, k, device="cuda", dtype=dtype) B =...

Try prefixing the command with `omnitrace-run -- python3 -m omnitrace -b -- `. I suspect the later initialization of omnitrace due to the `@profile` is causing in omnitrace getting initialized...

There are a couple things going on here. I believe the default view of the timelines is the accumulation of the counters, so you will not see them fluctuate but...

FYI, this is probably a rocprofiler bug, not an omnitrace bug. There have been a lot of issues with its stability between minor releases recently.