gcj performance: benchmarks

[Bryce McKinlay]

Anthony Green wrote:

> Was your libgcj compiled with optimization?  The .java files aren't
> currently compiled with optimization by default.

No, I am using a default libgcj build. It would be interesting to see what
difference, if any, compiling it with optimization makes. Unfortunatly libgcj
seems to be a bit broken at the moment so I can't run the benchmarks using the
current tree.

> As Per mentioned, there are many opportunities for us to improve
> performance.  Per also made an important point about compiling from
> source code.  Most of these benchmarks involve loops.  Gcj doesn't
> currently recover loop information from bytecode and so loses out on a
> handful of important loop optimizations.

Again, it would be nice to see what sort of improvement we would get by
compiling "jBYTEmark" from source rather than bytecode. gcj segfaults on the
source, and so far I havn't been able to figure out why. There have been a
couple of requests for the benchmark programs that I used, so I'll put the code
up somewhere and maybe Alexandre can take a look.

regards

  [ bryce ]