From: Bruno Haible <haible@il...> - 2002-05-28 18:55:32
OK, I'm now able to build with gcc-3.1 on i386. I made two changes
- For the gcc bug that affected record.d and other files at -O2, I
submitted a test case (even a runnable one with a main() that
returns 0 when compiled correctly and 1 when miscompiled). It is
registered as c/6833. The workaround is to use -fno-gcse (done in
- The other one also bites with no optimization at all. It was
a problem in clisp: we were assuming that the stack can be
accessed in a traditional way, like on m68k. But in gcc-3.1
the i386 backend has largely been rewritten, and the stack is
now accessed the same way as in RISC CPUs. It probably results
in better performance. But it killed the FAST_SP feature of the
bytecode interpreter in eval.d.
Get latest updates about Open Source Projects, Conferences and News.