Re: Pros and cons of high-level intermediate languages

graham@maths.su.oz.au (Graham Matthews) said:
>[to handle cases where GC might move objects, generate C code like:]


> for( i = 0; i < 100; i++ )
> object_access(p)[i] = object_access(q)[i];


> where object_access() is a function returning a pointer to the storage
> space for the array. The C compiler will not optimise this code (you
> may have problems with global inter-procedural optimisations here but
> you can usually turn them off).


> You take a performance hit here but in the presence of GC surely you
> have to take that hit no matter what IL you use?


Looks like a *big* performance hit for every array access, since you're
talking about doing a real procedure call (otherwise, it may boil down to what
I wrote and you get back into the problems previously discussed)!


> Now if you are really sure that 'q' and 'p' above will not move (eg:
> you only preform GC on allocation or deletion and you are sure that
> there will be no collections inside the loop) you can code the loop as
> you suggest above and the performance will be better. If you cannot
> guarantee this condition then you cannot code the loop as above.


See our paper in the last SIGPLAN conference which describes a *working
system* that can handle the kind of code I originally presented. It works
by generating tabular information for the collector to use. This incurs a
space cost, but impacts run-time performance minimally. There are cases
where additional values have to be kept around, which might increase
register pressure a little, but it appears to be a minor effect.


As for changes/extensions to C to make it easier to gc, Hans Boehm has
worked up a proposal for making C safe for gc (for conservative gc at
least; I can't recall if it would allow other kinds of gc). I certainly
think it's *possible*; whether people would accept the necessary language
definition changes is another matter.
--


J. Eliot B. Moss, Associate Professor
Department of Computer Science
Lederle Graduate Research Center
University of Massachusetts
Amherst, MA 01003
(413) 545-4206, 545-1249 (fax); Moss@cs.umass.edu
--

Return to the comp.compilers page.
Search the comp.compilers archives again.