Re: [rvm-research] Compiler in-lining limits for Magic methods

Hi,

On 03.03.2016 18:27, Kulig, Joe wrote:
> Do the compilers, baseline and adaptive, still limit the amount of inline code generated by Magic methods as stated in the comments in the OutOfLineMachineCode class?
AFAIK the JavaDoc comment for OutOfLineMachineCode is wrong. The current compilers "inline" magic methods by replacing the call with the appropriate machine code instructions (baseline compilers) or appropriate IR (opt compiler). This process isn't subject to any size limits.

Magics that are implemented via OutOfLineMachineCode are different from the others in that they directly access multiple hardware registers. AFAIK this can't easily be mapped to operators in the current high-level IR (which operates on an unlimited number of symbolic registers).


Kind regards,

Erik Brangs

------------------------------------------------------------------------------