As per our conversation on IRC: Dalibor claimed that Kaffe would compile in both native GMP-based math and some slow java-based math, and then only use the native if -Xnative-big-math is enabled. This means it's even slower than Sun's impl, by default. Please make it use native big math if it possibly can. Personally I don't see any reason why not to just link it in if it's available at configure time. And then fail with a dynamic linker error if it's not available. Or link it in statically if the user asked for that. But if you want to load it on the fly, that's fine and won't be much performance hit.
My interest in all this: Freenet has just received a significant reduction in its CPU usage by enclosing native GMP libs for various x86 platforms. Sun's math is _really_ slow, simply because they don't open source it and therefore can't use libgmp. We've had a factor of 8 or 10 gain in BigInteger modPow, and a factor of 3 in BigInteger.doubleValue(), from using libgmp. But it means we have to deal with platform specific native code in the extra jar, which is a pain in the ass. Hopefully we can get Freenet working on Kaffe soon. -- Matthew J Toseland - [EMAIL PROTECTED] Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so.
signature.asc
Description: Digital signature
_______________________________________________ kaffe mailing list [EMAIL PROTECTED] http://kaffe.org/cgi-bin/mailman/listinfo/kaffe
