max frame size

Stephen Weeks sweeks@intertrust.com
Mon, 10 May 1999 15:21:41 -0700 (PDT)


> I  ran  through  a bunch of the benchmark examples, comparing my coalescer to
> the old one.  Here are the results:
> 
>         life.sml                no change
>         count-graphs.sml        no change
>         knuth-bendix.sml        10 chunks -> 3 chunks
>         ratio-regions.sml       7 chunks -> 4 chunks
>         simple.sml              360 chunks -> 283 chunks
>         barnes-hut.sml          no change
>         lexgen.sml              25 chunks -> 6 chunks
>         vliw.sml                1455 chunks -> 376 chunks
>         nucleic.sml             no change
>         mlyacc.sml              471 chunks -> 29 chunks

Cool.  This should give some decent performance improvments.  Can
y'all run the benchmarks to see?

> In NONE of these examples did the maximum frame size change.
> 
> I remain confused.

My new guess is that there is no bug and that, in fact, you changed
the command line switches passed to MLton.  In particular, perhaps you
forgot the "-i 15 -f 1 -np" when you tried the improved coalescer.
This could definitely cause increased stack frame sizes.