Capture critical PMCs for perf and render flamegraph using these

[ambud]

Currently we are only capturing default perf for perf based profiling, it doesn't capture any other types of events. This issue is created to request we that we allow ability for dynamic profiler to invoke perf with specific events and build a flamegraph for these to aid in solving and tracing performance issues.

To start with we want:

• cache-misses
• stalled-frontend / backend

[mlim19]

@ambud, can you elaborate this request? I think the perfspect tool can provides such metrics already or can extend if not there. Do you want to build a different flamegraph or attribute them to the existing flamegraph?

[prashantbytesyntax]

@mlim19 to clarify this requirement
support additional metrics for perf profiling rather than just default ones . These metrics includes cache-misses, stalled-cycles-frontend/backend etc as listed. So the output visualization is still flamegraph here but based on these metrics. To improve UX, we can tag these flamegraphs or add it as metrics selection.
Just some example use cases where this could be helpful - detect Java app running slow due to cold JIT compiler, branch-misses flamegraphs pinpoint unpredictable code paths causing CPU pipeline stalls, cache-misses + stalled-cycles distinguish between "need faster algorithm" vs "need better data layout, etc.