[NativeAOT] Convert operand of newarr to int when producing reverse P/Invoke array marshalling

[filipnavara]

Fixes the following error on linux-arm (#98022):

ILC: /home/navara/runtime/src/coreclr/jit/importer.cpp:9382
ILC: Assertion failed 'genActualTypeIsIntOrI(op2) : Possibly bad IL with CEE_newarr at offset 002Bh (op1=NULL op2=long stkDepth=0)' in 'SizeParamIndex.PInvoke.PassingByOutTest:MarshalCStyleArrayLong_AsByOut_AsSizeParamIndex(byref,byref):ubyte' during 'Importation' (IL size 147; hash 0x8bc3ebdf; FullOpts)

ILC: /home/navara/runtime/src/coreclr/jit/importer.cpp:9382
ILC: Assertion failed 'genActualTypeIsIntOrI(op2) : Possibly bad IL with CEE_newarr at offset 000Ch (op1=NULL op2=long stkDepth=0)' in 'Internal.CompilerGenerated.<Module>:ReverseDelegateStub__DelUlongArrByRefAsCdeclCaller(uint,uint):int' during 'Importation' (IL size 204; hash 0x41204394; FullOpts)

[MichalStrehovsky]

Most uses of EmitElementCount seem to assume int32 though. I think the bug is that we have a couple (two that I saw) places that store it into an nint local. But then seem to still use it as int32.

Weird that this shows up as a problem on 32bit platform. I would expect it to show up on 64bit due to size differences.

[filipnavara]

The IL opcode newarr is documented to take native int from the stack (https://learn.microsoft.com/en-us/dotnet/api/system.reflection.emit.opcodes.newarr?view=net-8.0):

The number of elements in the new array should be specified as a native int.

On 64-bit platforms it happens to work with any size of int but on 32-bit ones this gets hit for ulong out size parameter.

I can change it to conv.i4 or conv.ovf.i4 if you prefer. The array size cannot be bigger anyway.

[MichalStrehovsky]

The IL opcode newarr is documented to take native int from the stack

Huh, that's weird. ECMA-335 spec says int32 or nint and normally one would see int32.

Here we store the result into int32:

runtime/src/coreclr/tools/Common/TypeSystem/Interop/IL/Marshaller.cs

Lines 1142 to 1144 in 0842930

	// loads the number of elements
	EmitElementCount(codeStream, MarshalDirection.Forward);
	codeStream.EmitStLoc(vLength);

If I'm reading it right, coreclr will convert to i4 with overflow checking:

runtime/src/coreclr/vm/ilmarshalers.cpp

Line 4160 in a715719

pslILEmit->EmitCONV_OVF_I4();

[filipnavara]

If I'm reading it right, coreclr will convert to i4 with overflow checking:

Let's match CoreCLR then.

[am11]

On 64-bit platforms it happens to work with any size of int but on 32-bit ones this gets hit for ulong out size parameter.

Nice quote to have as a code comment! 😉