[MLton-devel] Basis-2002 benchmarks

[Matthew Fluet]

> > BTW, here are the results from benchmarks I ran yesterday.  Runtimes
> > aren't bad with the exception of wc-{input1,scanStream},
> 
> Yeah, ratios of 58 and 33 are pretty bad.  It would at least be nice
> to get the ratios down under 10, which would make us competitive with
> other SML compilers (but not blowing them away like we used to :-().

I think we can speed up StreamIO and doing the imperative/fall back to
stream approach should speed up wc-input1.

> BTW, any idea what is wrong with md5 and tensor?
> Is this a MLton bug or a basis problem?

They have type-checking errors under the new basis.  md5 uses the
Word8Vector.extract function which no longer exists
(use Word8VectorSlice.vector o Word8VectorSlice.slice).
tensor uses the MONO_VECTOR signature constraint on a "handmade" mono
vector, so it's missing functions defined in the new basis.  And some of
the types don't match.

There are no known MLton bugs with the new basis.  I'm able to get through
all the regressions (modulo some type errors) and do a self-compile, so
it's stable (just slow in places).

If anyone with a large body of ML code (besides the MLton sources ;) wants
to check out the new basis, I could try to make rpms/tgzs available.
Again, this was branched back in mid-summer, so it's missing a lot of the
gc stuff that was added on the main branch, but it might be useful to get
an idea of what changes need to be made to an existing code base.




-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
MLton-devel mailing list
MLton-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mlton-devel