Fix JITEEInterface::IsValueClass to return false for pointers.

[sandreenko]

Fixes #44730

In the past, we did not see REF types from ldobj handle, we had:

    [ 0]   4 (0x004) ldloca.s 0
    [ 1]   6 (0x006) ldobj pointer 
    [ 1]  11 (0x00b) ret

and jit treated it like a value type with POINTER_SIZE(in this case 8-byte, so used long):

               [000010] *--XG-------              *  IND       long  
               [000009] ------------              \--*  ADDR      byref 
               [000008] -------N----                 \--*  LCL_VAR   long   V01 tmp1 

after the VM change we see the real REF type (a pointer to an object on a heap):

               [000010] *--XG-------              *  IND       ref   
               [000009] ------------              \--*  ADDR      long  
               [000008] -------N----                 \--*  LCL_VAR   long   V01 tmp1 

Note that we allow REF->IMPL conversion, but not IMPL->REF, cc @AndyAyersMS .

I guess I need to revert part of #44465 but not sure what exactly.

[sandreenko]

PTAL @jkotas, @AndyAyersMS , cc @MichalStrehovsky @dotnet/jit-contrib

[sandreenko]

There are no diffs crossgen/pmi on --f. I have manually collected dasm for base and diff using different VM versions and checked with jit-analyze.

[jkotas]

Could you please also make a matching change in src\coreclr\tools\Common\JitInterface\CorInfoImpl.cs:

Change isValueClass to just:

        private bool isValueClass(CORINFO_CLASS_STRUCT_* cls)
        {
            return HandleToObject(cls).IsValueType;
        }

And delete this from getClassAttribsInternal:

            // CoreCLR uses UIntPtr in place of pointers here
            if (type.IsPointer || type.IsFunctionPointer)
                type = _compilation.TypeSystemContext.GetWellKnownType(WellKnownType.UIntPtr);

The VM side of the change looks good to me otherwise.

[sandreenko]

ping @dotnet/jit-contrib

[AndyAyersMS]

This will need a new jit GUID.

[AndyAyersMS]

LGTM.