PGO: Profile isinst/castclass

[EgorBo]

This PR adds class probes for some castclass/isinst operations in order to then optimize them in Tier1.
These probes are only supposed to be run for static PGO since they add some overhead for the dynamic one and we want to avoid it, while for the static one we can afford some overhead for a training session.

I didn't push the change to utilize the PGO for cast/isinst because I want this to be merged first, then flow into our static pgo infrastructure (where I'll enable this flag) and collect isinst/castclass for StandardOptimizationData.mibc. Then, I'll merge the second change and it will allow me to catch improvements properly for dotnet/performance.

Simple benchmark with the 2nd change (PGO is enabled):

usingSystem.Runtime.CompilerServices;usingBenchmarkDotNet.Attributes;usingBenchmarkDotNet.Running;publicclassProgram2:Program{}publicclassProgram{publicstaticvoidMain(string[]args){BenchmarkSwitcher.FromAssembly(typeof(Program).Assembly).Run(args);}publicobject[]testData;[GlobalSetup]publicvoidSetup(){testData=newobject[1024];for(inti=0;i<testData.Length;i++)testData[i]=newProgram2();}[Benchmark]publicintCountPrograms(){vararray=testData;intcount=0;for(inti=0;i<array.Length;i++){if(array[i]isProgram)count++;}returncount;}[Benchmark]publicvoidCastPrograms(){vararray=testData;for(inti=0;i<array.Length;i++)Consume((Program)array[i]);}[MethodImpl(MethodImplOptions.NoInlining)]staticvoidConsume(Programp){}}

Method	Job	Toolchain	Mean	Error	StdDev	Ratio
CountPrograms	Job-AGRJVL	\Core_Root\corerun.exe	895.6 ns	2.03 ns	1.69 ns	1.00
CountPrograms	Job-UTLWER	\Core_Root_base\corerun.exe	2,014.3 ns	5.81 ns	4.53 ns	2.25
CastPrograms	Job-AGRJVL	\Core_Root\corerun.exe	1,577.2 ns	6.27 ns	5.24 ns	1.00
CastPrograms	Job-UTLWER	\Core_Root_base\corerun.exe	2,922.8 ns	18.81 ns	16.67 ns	1.85

[ghost]

Tagging subscribers to this area: @JulieLeeMSFT
See info in area-owners.md if you want to be subscribed.

Issue Details

---

This PR adds class probes for some castclass/isinst operations in order to then optimize them in Tier1.
These probes are only supposed to be run for static PGO since they add some overhead for the dynamic one and we want to avoid it, while for the static one we can afford some overhead for a training session.

I didn't push the change to utilize the PGO for cast/isinst because I want this to be merged first, then flow into our static pgo infrastructure (where I'll enable this flag) and collect isinst/castclass for StandardOptimizationData.mibc. Then, I'll merge the second change and it will allow me to catch improvements properly for dotnet/performance.

Simple benchmark with the 2nd change (PGO is enabled):

usingSystem.Runtime.CompilerServices;usingBenchmarkDotNet.Attributes;usingBenchmarkDotNet.Running;publicclassProgram2:Program{}publicclassProgram{publicstaticvoidMain(string[]args){BenchmarkSwitcher.FromAssembly(typeof(Program).Assembly).Run(args);}publicobject[]testData;[GlobalSetup]publicvoidSetup(){testData=newobject[1024];for(inti=0;i<testData.Length;i++)testData[i]=newProgram2();}[Benchmark]publicintCountPrograms(){vararray=testData;intcount=0;for(inti=0;i<array.Length;i++){if(array[i]isProgram)count++;}returncount;}[Benchmark]publicvoidCastPrograms(){vararray=testData;for(inti=0;i<array.Length;i++)Consume((Program)array[i]);}[MethodImpl(MethodImplOptions.NoInlining)]staticvoidConsume(Programp){}}

Method	Job	Toolchain	Mean	Error	StdDev	Ratio
CountPrograms	Job-AGRJVL	\Core_Root\corerun.exe	895.6 ns	2.03 ns	1.69 ns	1.00
CountPrograms	Job-UTLWER	\Core_Root_base\corerun.exe	2,014.3 ns	5.81 ns	4.53 ns	2.25
CastPrograms	Job-AGRJVL	\Core_Root\corerun.exe	1,577.2 ns	6.27 ns	5.24 ns	1.00
CastPrograms	Job-UTLWER	\Core_Root_base\corerun.exe	2,922.8 ns	18.81 ns	16.67 ns	1.85

Author:	EgorBo
Assignees:	-
Labels:	area-CodeGen-coreclr
Milestone:	-

[EgorBo]

/azp list

[EgorBo]

/azp run runtime-coreclr pgo, runtime-coreclr libraries-pgo

[azure-pipelines]

Azure Pipelines successfully started running 2 pipeline(s).

[EgorBo]

@AndyAyersMS@jakobbotsch PTAL

[AndyAyersMS]

LGTM.

One minor suggestion.