JIT: Don't give up on runtime lookup in inliner too early #81432
Description
static bool Test<T>() => Callee<T>();
static bool Callee<T>() => typeof(T) == typeof(int);It is expected that Test<string> will be compiled to just return false. Instead, it's compiled down to:
; Assembly listing for method Program:<<Main>$>g__Test|0_0[System.__Canon]():bool
G_M24604_IG01:
sub rsp, 40
mov qword ptr [rsp+20H], rcx
G_M24604_IG02:
mov rdx, qword ptr [rcx+38H]
mov rdx, qword ptr [rdx+10H]
test rdx, rdx
je SHORT G_M24604_IG04
G_M24604_IG03:
jmp SHORT G_M24604_IG05
G_M24604_IG04:
mov rdx, 0xD1FFAB1E ; global ptr
call CORINFO_HELP_RUNTIMEHANDLE_METHOD
mov rdx, rax
G_M24604_IG05:
mov rcx, rdx
call [Program:<<Main>$>g__Callee|0_1[System.__Canon]():bool]
nop
G_M24604_IG06:
add rsp, 40
ret
; Total bytes of code 57It seems that JIT's inliner could be a bit smarter here when currently it gives up when it sees ldtoken T! as a potential runtime lookup. We either can teach it to recognize typeof() == typeof() idiom upfront (and fold) or we can do a sort of "I've just hit a runtime lookup, let me import up to 3 more opcodes to see if I will be able to fold it"
cc @AndyAyersMS @jkotas