This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH]: Speed up dominance frontiers computation
> Daniel Berlin wrote:
>
> >+/* Compute dominance frontiers, ala Harvey.
> >
> >
Diego also wanted more details on the numbers.
On two *very* large cases i have, dominance frontier calculation took
~10 seconds on each one.
It now takes ~6 seconds on each one. This was roughly a 1% speedup for
these testcases.
On a large C++ testcase i have, where we were processing thousands of
functions, dominance frontiers took a total of ~30 seconds.
It now takes ~15.
This is roughly a 1% speedup for these testcases.
On bootstraps, we save roughly 5 seconds (this is with checking, i
haven't done a disable-checking bootstrap), which is not significant.
I've only submitted this for consideration for 4.0 because it is a small
speedup.
I was going to submit it to 4.1 anyway because the algorithm is a
cleanup and simplification of the old one.
I honestly have no problem pushing it off till 4.1 if people would
rather me do that.
--Dan