Andy Ayers

@amanasifkhalid can you take another look? I removed the Next block and just wire up the flow directly. TP diffs good, PerfScore diffs good. Code size increases, but mainly from...

Failure is a timeout spmi replay for linux arm32.

Calls via function pointers would not get inlined currently.

Another thought is to not optimize these methods, that way the return value can be unconsumed but presumably would always still be produced. But it would make overhead higher, which...

Now have the ability to get method entry counts for jitted methods. So by measuring before/after a given inline, we can see the drop in counts for the inlined method...

Have a [PR for this](https://github.com/dotnet/coreclr/pull/5581) over in coreclr.

Using offset helps disambiguate some cases, but not enough. Here's the same run as above, with offset used to try and identify call sites: ``` $$$ --- legacy: inlines in...

Have plumbed through improved IL offsets and how the results look more plausible: ``` $$$ --- legacy: inlines in order of impact --- $$$ --- [08] Bench -> G -1.38%...

I have some ad-hoc data that shows the number is indeed pretty stable for a wide variety of the jit perf benchmarks. Will gather this more systematically soon. One interesting...

Wondering now why we can't just invoke the xunit entry point. Seems like this ought to work. So basically Main emulates the (single iteration) activity of xunit-perf. At any rate...