Acovea doesn't link with gcc 4.0.0_beta20050305, complaining about a missing definition of BOGUS_RUN_TIME. The following patch solves that, and makes it possible to use acovea to tune your CFLAGS with gcc 4.0.0.
Created attachment 54062 [details, diff] gcc4 patch
Created attachment 54063 [details] patched ebuild ebuild that applies the patch, it works for me...
Created attachment 54373 [details] configuration file This is a sample configuration file that I made for my laptop (pentium-m). Some options that should be included in gcc-4.0 according to the docs are not recognised, so I commented them out. I hardcoded the path to the right compiler (4.0.0_beta20050305) so it can be used with the default compiler different from gcc4. All the benchmarks work, except for the treebench, but treebench never worked for me...
Created attachment 54422 [details, diff] Patch to make treebench work with the malloc-check in glibc This patch makes treebench work with glibc compiled with malloc-check activated. This is the only benchmark where I have seen succesfull tree-vectorisation, so it would be nice to be able to use this benchmark. This fix is non-gcc4 specific.
Bumping the package to the newest version would probably solve this problem. There are newer versions of libacovea, libcoyotl, and libevocosm on http://www.coyotegulch.com/products/acovea/index.html I didn't try it, but he tested GCC4 with it, so I'm assuming he fixed the compilation issues. Note, the location of the files has changed, so the ebuilds will need to be updated slightly. There is also a new package on the site as well.
the new version solves the problem of not building on gcc 4.0, but the treebench problem still exists, it won't run on either gcc 3.4 or 4.0. The patch works for the new version as well
Re-assign.
acovea-5.1.1 is now in the tree, let me know if there are any more issues. Thanks!