don't require the JitDiagnosers (TailCall & Inlining) to run benchmarks once again just to gather JIT info, the overhead is very small

[adamsitnik]

While working on dotnet/runtime#2191 I got surprised by how long it takes to get the benchmark results.

It turned out that so far we were always running the benchmarks one more time if Inlining or TailCall diagnosers were enabled. We simply don't need to do that, the overhead of these two diagnosers is very, very small (we just sign up for 3 events).

git clone https://github.com/dotnet/BenchmarkDotNet/
cd BenchmarkDotNet
cd samples
cd BenchmarkDotNet.Samples.FSharp
dotnet run -c Release -f net461 --filter TailCallDetector

Before:

Method	facRank	Mean	Error	StdDev
test	7	9.460 ns	0.2158 ns	0.2310 ns
test1	7	4.758 ns	0.0824 ns	0.0731 ns
test2	7	51.820 ns	0.9814 ns	0.9180 ns

Run time: 00:02:29 (149.11 sec), executed benchmarks: 3

Global total time: 00:02:31 (151.73 sec), executed benchmarks: 3

After:

Method	facRank	Mean	Error	StdDev
test	7	9.383 ns	0.0837 ns	0.0783 ns
test1	7	4.688 ns	0.0655 ns	0.0613 ns
test2	7	49.748 ns	0.2633 ns	0.2199 ns

Run time: 00:01:23 (83.31 sec), executed benchmarks: 3

Global total time: 00:01:26 (86.03 sec), executed benchmarks: 3

[0xGeorgii]

@adamsitnik make sense, I believe 👍

[adamsitnik]

@GeorgePlotnikov thanks for a very fast response!

@AndreyAkinshin I need your approval to be able to merge

[adamsitnik]

@AndreyAkinshin ping

[AndreyAkinshin]

@adamsitnik thanks!