Wed Feb 22 13:35:00 GMT 2006
Norman Hendrich writes:
> Hello all,
> after some absence I tried my luck with gcj again (2006.02.18 snapshot).
> However, it seems that I did something wrong:
> 1) building gcc 4.2.0 20060218 (C,C++,Java) took over _three hours/
> on my Athlon 2600+ machine. For comparison, I last built gcc/gcj
> before the classpath merge, which completed in about 35 minutes (iirc)
> on the same machine. This is on gnu-linux-x86 with gcc-4.0.0 as my
> system compiler.
We now do a full bootstrap by default; you can configure with
--disable-bootstrap if you wish.
> Is this normal?
> 2) I "only" have 512 MBytes RAM on this machine. Building gcj (both
> "make" and crazily also "make install") required a smidgen more than
> 512 megs with corresponding swapping and an unusable machine during
> several minutes.
This is probably the real cause of the slowdown.
> Is this intentional? Do you get provision from DRAM vendors? :-)
> 3) The resulting compiler seems to work, but AWT/Swing performance is
> not what I expected. Actually, jamvm+classpath runs some of my test
> apps faster than gcj -O3 ... see next post for an example.
> Any insights as how to speed up build-times or why AWT/Swing might
> end up crawling under gcj are really appreciated.
The solution for speeding up build times is pretty obvious: you just
don't have enough RAM. It might be that we could reorganize things a
bit, but at some point we have to link the library, and that takes
More information about the Java