A note on JIT code
gback at cs.utah.edu
Fri Feb 4 09:27:39 PST 2000
may provide for interesting reading.
> I have been looking into the kaffe sources for quite sometime,
> especially the implementation of JIT. I am not sure what all were the primary
> considerations that went into, before the actual implementation, of the
> JIT code, especially to note that we dont seem to be doing coloring of
> any sort. Personally, I am not very much obsessed with graph coloring,
> though I am curious to know the extent to which redundant spills and
> restores can be mimized over the existing schema of the least recently
> used register, by "working a bit more harder" at compile time by
> generating graphs at the "right places" and trying to use as many machine
> registers to the non-overlapping slots as possible. Has any worked on
> getting statistics on the number of spills and restores with the
> existing approach against the compile time overhead of doing these using
> coloring approach or probably a "better approach"? If so what were the
> statistics based upon? If not, just out of curiosity, is any one having plans
> on improving the same :) ?
> I do appreciate when one would respond with comments like coloring
> may not be very much justifiable for Dynamic Code genaration as against
> "proper compilers", but would certainly feel a lot more convinced after if these
> arguments are backed with some interesting statistics on platforms like
> Intel for JIT code.
> I did go through the news groups archive, but did not find any
> discussion threads on the same.
> Also please let me know if I am missing something in the overall
> Thanks & Regards,
More information about the kaffe