[davidwrighton]

• Currently, for best performance, one needs to know to update a switch statement in jitinterface.cpp whenever a helper is moved to managed
• Instead, with this change, if the helper has a specified managed method implementation, that check will just work
• Also, if there is a native implementation, then the native implementation will take precedence. (This is to allow for helpers to have alternative assembly based implementations)

This should help fix a regression #108203 introduced with my recent work around generics lookups.

[dotnet-policy-service]

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

[EgorBo]

@davidwrighton there is also this code that you might want to change

runtime/src/coreclr/vm/callcounting.cpp

Lines 663 to 668 in b3a113a

	// Let's tier up all cast helpers faster than other methods. This is because we want to import them as
	// direct calls in codegen and they need to be promoted earlier than their callers.
	if (methodDesc->GetMethodTable() == g_pCastHelpers)
	{
	callCountThreshold = max<CallCount>(1, (CallCount)(callCountThreshold / 2));
	}

so for this "indirect -> direct" optimization to kick in, especially in micro-benchmarks, it's better to promote such helpers to Tier1 earlier.

[davidwrighton]

@EgorBo ... thanks for the heads up. I'll look into that later.

[EgorBo]

I think this could be just instGen_Set_Reg_To_Imm(EA_PTR_DSP_RELOC, ...) without if-else here. It will either use addr+add or just emit a 64bit constant inside the function itself. See its impl:

runtime/src/coreclr/jit/codegenarm64.cpp

Lines 2239 to 2242 in c6e3eb7

	if (!compiler->opts.compReloc)
	{
	size = EA_SIZE(size); // Strip any Reloc flags from size if we aren't doing relocs
	}

[EgorBo]

Fine to merge as is, just a nit