>>105545585
>Also compilers are terrible at estimating call depths and will rather generate a CALL/RET pair than a LEA return + JMP target + JMP return combination even though that would improve out-of-order execution (because CALL/RET use the SP register implicitly)
You mean like
f:
jmp rax
start:
lea rax, [rip + 2]
jmp f
would be better than
f:
ret
start:
call f
for out of order execution?