Inliner fails to inline a simple RVA pattern

[EgorBo]

class C
{
    static bool IsWhiteSpace(char c) => IsLatin1(c);

    static bool IsLatin1(char c) => c < RVA.Length;

    static ReadOnlySpan<byte> RVA => new byte[]{1,2,3};
}

codegen for IsWhiteSpace:

; Method C:IsWhiteSpace(ushort):bool
       movzx    rcx, cx
       tail.jmp [C:IsLatin1(ushort):bool]
; Total bytes of code: 9

weight= 10 : state   3 [ ldarg.0 ]
weight= 79 : state  40 [ call ]
weight=  6 : state  11 [ stloc.0 ]
Named Intrinsic System.ReadOnlySpan`1.get_Length: Recognized
weight= 61 : state  19 [ ldloca.s ]
weight= 79 : state  40 [ call ]
weight= 26 : state 171 [ clt ]
weight= 19 : state  42 [ ret ]

Inline has 1 intrinsics.  Multiplier increased to 1.3.
Inline candidate callsite is boring.  Multiplier increased to 2.6.
calleeNativeSizeEstimate=280
callsiteNativeSizeEstimate=85
benefit multiplier=2.6
threshold=221
Native estimate for function size exceeds threshold for inlining 28 > 22.1 (multiplier = 2.6)

The IL for RVA spans is a bit verbose, inliner sees two calls (one of them is recognized as Span.Length though) and is not able to predict that the whole thing is lowered to a few instructions

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

Issue Details

---

class C
{
    static bool IsWhiteSpace(char c) => IsLatin1(c);

    static bool IsLatin1(char c) => c < RVA.Length;

    static ReadOnlySpan<byte> RVA => new byte[]{1,2,3};
}

codegen for IsWhiteSpace:

; Method C:IsWhiteSpace(ushort):bool
       movzx    rcx, cx
       tail.jmp [C:IsLatin1(ushort):bool]
; Total bytes of code: 9

weight= 10 : state   3 [ ldarg.0 ]
weight= 79 : state  40 [ call ]
weight=  6 : state  11 [ stloc.0 ]
Named Intrinsic System.ReadOnlySpan`1.get_Length: Recognized
weight= 61 : state  19 [ ldloca.s ]
weight= 79 : state  40 [ call ]
weight= 26 : state 171 [ clt ]
weight= 19 : state  42 [ ret ]

Inline has 1 intrinsics.  Multiplier increased to 1.3.
Inline candidate callsite is boring.  Multiplier increased to 2.6.
calleeNativeSizeEstimate=280
callsiteNativeSizeEstimate=85
benefit multiplier=2.6
threshold=221
Native estimate for function size exceeds threshold for inlining 28 > 22.1 (multiplier = 2.6)

The IL for RVA spans is a bit verbose, inliner sees two calls (one of them is recognized as Span.Length though) and is not able to predict that the whole thing is lowered to a few instructions

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

[EgorBo]

Generic fix: R2R should leave a flag in MethodDesc that method itself is compiled into a small piece of assembly code - JIT will rely on that flag in cases like this.