Add an IsFuncletCache to the EECodeInfo

[davidwrighton]

• Do a cheap computation in EECodeInfo::Init time to fill it in if there aren't any funclets in the method
• This avoids the fairly complex IsFunclet logic on the JitManager, as well as the virtual call to get there in some cases.

This should improve the performance of any EH stack trace operations that repeatedly call IsFunclet, especially if many of the methods on the stack are simple and do not have any EH constructs in them.

In a simple benchmark of EH throw performance, when running in the Windows X86 Funclets mode, the performance improves by about 15%.

[Copilot]

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

[dotnet-policy-service]

Tagging subscribers to this area: @mangod9
See info in area-owners.md if you want to be subscribed.

[jkotas]

Suggested change

	IsFuncletCache ComputeInitialIsFuncletCacheValue(PCODE codePtr)
	{
	if (GetNumberOfUnwindInfos() == 1)
	return IsFuncletCache::IsNotFunclet; // Optimize the scenario where there are not any funclets in a method
	return IsFuncletCache::NotSet;
	}
	bool MayHaveFunclets()
	{
	return GetNumberOfUnwindInfos() > 1;
	}

It would be nice if IsFuncletCache is kept as a private detail of EECodeInfo.

[jkotas]

This looks very similar to the lazy initialization used for GetFunctionEntry() that is also initialized eagerly in some cases and lazily in other cases. It would be nice if the two look exactly the same:

• Move the implementation of IsFunclet to .cpp file next to GetFunctionEntry
• Rename method on the code manager interface to LazyIsFunclet

[jkotas]

Suggested change

	pCodeInfo->m_isFuncletCache = pCHdr->ComputeInitialIsFuncletCacheValue(currentPC);
	if (pCHdr->MayHaveFunclets())
	{
	// Computed lazily by code:EEJitManager::LazyGetFunctionEntry
	pCodeInfo->m_pFunctionEntry = NULL;
	pCodeInfo->m_isFuncletCache = IsFuncletCache::NotSet;
	}
	else
	{
	pCodeInfo->m_pFunctionEntry = pCHdr->GetUnwindInfo(0);
	pCodeInfo->m_isFuncletCache = IsFuncletCache::IsNotFunclet;
	}

We may be able to do even more cheap caching for the common case here.

[filipnavara]

Nit: Typo - s/(0)(/(0).

I already had this optimization locally and it's useful.

[jkotas]

LGTM!