This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: mixing optimized and unoptimized code?
David Starner <dvdeug@x8b4e53cd.dhcp.okstate.edu> writes:
> > A related question: which optimizations are not especially memory hungry?
> > I was thinking I could selectively turn on *some* of them on those
> > problematic sources.
>
> Someone else will have to help you here. But do you know why are these
> are running out of memory? Are they large files, or template heavy, or
> large functions, or convulted code-wise? If the GCC people know where
> the hot spots are, they can sometimes do amazing improvements in the
> compiler.
They are large files (close to 1MB, preprocessed) containing large
functions. In fact, they are automatically generated files, which, I am
sure, aggravates the problem. (One of the files is a flex-generated
scanner for Java, which is basically one huge switch statement).
I've made the two problem files available here, if anyone is interested
in taking a look:
http://www.cs.berkeley.edu/~maratb/java-grammar.i.gz
and
http://www.cs.berkeley.edu/~maratb/java-lexer.i.gz
These files, compiled with
g++ -fPIC -fno-rtti -fno-exceptions -fno-gnu-keywords -fvtable-thunks -g -O1
give "out of virtual memory" errors at any optimization level. This is
with gcc 2.95.1 on Solaris (256MB Ultra 10); I've had similar result
with 2.95.2 on Linux (256MB PIII 450).
Marat.