At Wed, 28 Oct 2009 12:31:52 -0700, Phil Nelson wrote: > > On Wednesday 28 October 2009 11:30:23 am Marko Schütz wrote: > > Is that setup using ccache? > > It does if "./build.sh release" does. (and I don't think > that build.sh uses ccache or cachecc1. I have experimented with using ccache on an _unmodified_ checkout. On my laptop ./build.sh build (basically) takes: Build started at: Wed Oct 28 17:23:48 AST 2009 Build finished at: Wed Oct 28 18:49:25 AST 2009 for a total of 1:25:37 whereas using ccache with a populated cache by shell-aliasing cc, c++, gcc, g++, /usr/bin/gcc, .../tooldir/bin/i386--netbsdelf-gcc, etc and then running ./build.sh build twice, the second run takes: Build started at: Wed Oct 28 19:04:53 AST 2009 Build finished at: Wed Oct 28 19:58:05 AST 2009 for a total of 0:53:12 That would be saving 38%, if it can be applied to the TNF build cluster (assuming it's not used there yet), that would take yield almost 5 builds per day. I _haven't_ yet tested the resulting executables, but even if the process needs some modifications, I believe it will work in principle and I don't expect the times to change much. I would suspect that the TNF build cluster has a HD speed to CPU speed ratio that would make the relative savings even more... Best regards, Marko
Attachment:
pgp4Yc9_pGhg0.pgp
Description: PGP signature