This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: getting rid of ggc_push_context
On Sat, 2004-10-02 at 09:37, Daniel Jacobowitz wrote:
> On Fri, Oct 01, 2004 at 01:16:40PM -0700, Richard Henderson wrote:
> > Until we get rid of cse-follow-jumps, I don't see that much has
> > changed in how CSE operates.
>
> Yes, but a lot has changed in its input. Like Zack said: "with the
> three optimizers in place, it probably isn't generating nearly as much
> garbage as it used to".
Possibly because we are likely generating fewer statements and fewer
pseudos which need to be tracked in CSE.
However, there may be cases where we generate more pseudos (think
about SRA, then look at the Plum Hall tests and ponder what would
happen if all those structure elements are scalarized).
> I instrumented cc1-i-files to measure CSE memory use, and the most cse
> ever allocated for any function was 800K. I didn't try the referenced
> test from Plum Hall, but that would be easy to do next week...
There's a few tests which are notorious. Due to the way the Plum Hall
license works, I probably can't name them in a public forum. But
you can identify them as the ones that are the slowest to compile
and which spend 20-30% of their time in CSE.
Jeff