[MLton-devel] nucleic benchmark times

[Suresh Jagannathan]

I agree with Stephen that Matthew's numbers hold surprising
results that many people in the compiler community would find
quite counter-intuitive and valuable.  A really important contribution Mlton 
makes here is providing both a common optimizing front-end, and
radically different backend code-generators, allowing a more
systematic comparison of the two approaches, something most 
compilers don't provide. 

Re. the SSA optimization - Bodik, Gupta, and Sarkar's paper on
eliminating array bounds checks in pldi'00 uses simple symbolic
constraints to eliminate a large number of array checks in 
Java programs; the style of programs they mention in their paper
seems close in spirit to Matthew's example.

In looking over the numbers, it was interesting to see that nucleic did
significantly better simply by turning off overflow checks, which is what
the Gambit numbers Brad posted were based on.   I had a long talk with
Brad and Siskind last week, and both Gambit and Stalin's code generators
are structured to ensure that C enforces alias separation of various
memory classes.  E.g., Stalin uses a fixed set of regions, no stack, and
no heap which offers another ancedotal datapoint on the relative
importance of aliasing info.

-- Suresh


-------------------------------------------------------
This sf.net email is sponsored by: Are you worried about 
your web server security? Click here for a FREE Thawte 
Apache SSL Guide and answer your Apache SSL security 
needs: http://www.gothawte.com/rd523.html
_______________________________________________
MLton-devel mailing list
MLton-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mlton-devel