Open sandreim opened 3 weeks ago
We profile cache misses using valgrind. Could add more reports using other instruments from valgrind. But what exactly we need and for which cases?
Recently, @alexggh has played with profiling the amount of memory is allocated in different calls which lead to discovering some improvement opportunities. jemalloc
has a profiling feature that allows it, but Alex can share more details about how we could automate as much as possible what he did.
These are the steps that I've been playing with: https://www.magiroux.com/rust-jemalloc-profiling/, I'll follow with a PR to integrate it in subsystem-bench.
Most likely the same thing would have been achieved with valgrind instrumentation.
It would be great to have this working along with CPU profiling. We could use it to investigate performance as well as memory leaks.