fix(pc-sampling): read samples into a buffer of their own - #164
Merged
Merged
Conversation
In KERNEL_SERIALIZED mode CUPTI moves every finished kernel's PC records into the SAMPLING_DATA_BUFFER given at configure time. gpufl passed that same buffer to cuptiPCSamplingGetData and zeroed its count first, which threw away everything held there. Only the per-PC overflow CUPTI creates once the 65,536-record buffer is full reached the output, so a session returned nothing until about 65,536 / (PC records per launch) launches had run, and then only later launches, under one correlation id. Read into a separate buffer, as NVIDIA's samples do, and drain until CUPTI reports nothing pending. Emit one row per (launch, PC, stall reason) through Monitor::PushProfileSamples so per-launch attribution survives and large collects do not overrun the monitor ring. Correct the comments, zero-sample warning and CLI help that blamed launch count or armed GetData, and fix the example the same way.
This was referenced Sep 30, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
In KERNEL_SERIALIZED mode CUPTI moves every finished kernel's PC records into the SAMPLING_DATA_BUFFER given at configure time. gpufl passed that same buffer to cuptiPCSamplingGetData and zeroed its count first, which threw away everything held there. Only the per-PC overflow CUPTI creates once the 65,536-record buffer is full reached the output, so a session returned nothing until about 65,536 / (PC records per launch) launches had run, and then only later launches, under one correlation id.
Read into a separate buffer, as NVIDIA's samples do, and drain until CUPTI reports nothing pending. Emit one row per (launch, PC, stall reason) through Monitor::PushProfileSamples so per-launch attribution survives and large collects do not overrun the monitor ring.
Correct the comments, zero-sample warning and CLI help that blamed launch count or armed GetData, and fix the example the same way.
Type of Change
Testing
Checklist