Use an intercepter interface to allow GPU trace matching AFTER symbolization

[gnurizen]

Symbolize GPU traces before kernel timing fixup

GPU samples can sit as raw traces for awhile waiting for the fixer to
match them with GPU timing information, during this time pointers in
the raw traces could grow stale due to functional program GC'ing
activation records. Avoid this by doing trace symbolizing before
parking traces in the fixer maps.

This has the nice side affect of removing some channel indirection
and now traces go straight into the fixer maps and when matched they
go straight to ReportTraceEvent.

Move CUDA symbolization earlier in the pipeline: ConvertTrace now
handles CUDA frames directly, and parcagpu.Start returns a
TraceInterceptor instead of a filtered channel. The interceptor
diverts symbolized CUDA traces into the GPU fixer post-ConvertTrace,
and completed traces (with timing and kernel name) are reported
directly. This eliminates the Symbolize method on the CUDA
interpreter in favor of demangling in prepTrace.

[Copilot]

Pull request overview

This PR restructures the CUDA/GPU trace pipeline so traces are symbolized before being parked for GPU timing correlation, using a new interceptor hook after ConvertTrace to divert CUDA traces into the GPU “fixer” and report completed traces directly.

Changes:

• Add TraceInterceptor support to tracehandler and update call sites/tests.
• Move CUDA frame handling into ProcessManager.ConvertTrace (no longer relying on CUDA interpreter Symbolize for demangling).
• Refactor parcagpu + CUDA fixer to accept symbolized traces, attach timing/kernel-name later, recompute hashes, and emit completed traces.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
tracehandler/tracehandler.go	Adds interceptor hook and threads it through handler construction/start.
tracehandler/tracehandler_test.go	Adds interceptor behavior tests and updates Start call signature.
processmanager/manager.go	Handles CUDA frames directly during ConvertTrace (preserving correlation ID encoding).
parcagpu/parcagpu.go	Reworks timing reader and returns an interceptor that diverts CUDA traces post-ConvertTrace.
interpreter/gpu/cuda.go	Refactors CUDA fixer to store symbolized traces, attach kernel timing/name, recompute hash, and return completed outputs.
internal/controller/controller.go	Updates tracehandler.Start signature usage (currently still passes nil interceptor).

Comments suppressed due to low confidence (1)

parcagpu/parcagpu.go:79

• This select loop calls eventReader.ReadInto() in the default branch. ReadInto blocks, so the goroutine won’t service logTicker/clearTicker ticks (or ctx cancellation) while it’s blocked waiting for events. Consider using Reader.SetDeadline / timed reads, or reading perf events in a dedicated goroutine and sending them over a channel so the outer loop can select on ctx/tickers reliably.

			case <-ctx.Done():
				return
			default:
				if err := eventReader.ReadInto(&data); err != nil {
					readErrorCount.Add(1)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[gnurizen]

Converting back to draft as we're looking at doing the demangling server side