Lightning-AI / Lightning-AI/lightning-thunder
Enable NvtxProfileTransform by default
@tfogal is already working on this.
Since Sep 27, 2024.
- Dominant language
- Python
- Stars
- 1.5k
- Forks
- 121
- PR merge metrics
- No merged PRs in 30d
Description
🚀 Feature
Add NvtxProfileTransform to the default set of transforms.
Motivation
NvtxProfileTransform adds nvtx around each thunder op, which is great for understanding what thunder is spending its time doing.
Pitch
Anyone that wants to look at performance wants this transform on. If you're not looking at perf, the overhead should be negligible.
Alternatives
- detect if we're running in the profiler (is this possible?) and turn it on automagically
- automagically enable it when the
THUNDER_ANNOTATE_TRACESenvironment variable is set, which has precedence as doing something similar - enable it but automagically disable it when an environment variable is set
Additional context
It was pointed out that seeing a bunch of 'nvtx' nodes in a trace might be annoying for people developing their own transforms--it may feel like noise that obscures what's happening in the trace.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.