[MLton] Benchmark question
Lukasz S Ziarek
Wed, 25 Feb 2004 10:12:55 -0500 (EST)
On Tue, 24 Feb 2004, Matthew Fluet wrote:
> > After running the benchmarks on while using flattener, the benchmark
> > script reported that about half of the benchmarks failed. All the
> > benchmarks compile when I run the compiler with my flag set to true.
> > Obviously there is a bug, and to help with debugging I was wondering if
> > there is a way to see what errors occured during the benchmark suite.
> There isn't an easy way of getting the compile errors.
> One simple thing to check: note that all the .sml files in the
> benchmarks/tests directory are "incomplete" in some sense. The benchmark
> program appends
> val _ = Main.doit 10
> (where 10 is some number specific to each benchmark; essentially, a repeat
> count so the benchmark runs long enough) to each benchmark and compiles
> that program. Without the call to Main.doit, the programs essentially do
> nothing, and the first couple rounds of the SSA optimizer will usually
> simplify the program down to pretty much nothing; by the time your
> flattener runs, there may not be anything interesting for it to do.
Yes that is what was happening, I could compile the .sml files fine,
because they would be simplified, though they still were decently large,
just not interesting for the sake of debugging. What I did, is to quickly
code up something incorrect so that all the tests would fail, but the
suite would generate the test.batch.sml files. Now, these files actually do not
compile for whatever reason and I can proceed with my debugging, by
running my flattener on them.
> MLton mailing list