This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: fix jump threading with liveness
On Tue, Mar 05, 2002 at 07:22:14PM +0100, Jan Hubicka wrote:
> This can be handled by anding with the liveness of edge destination.
Err, why isn't it _just_ the liveness of the destination?
> Other case is problem with different CC flags - basically different
> comparions have different CCmodes and the jump threading may fail just
> because of different modes. I believe this can be handled by trying
> to canonicalize the condtional if former does not match.
Maybe.
> By adding trivial conditional simplifier proving that condition A implies
> (negation of) condtiion B. Is there something similar present already? It
> would help loop unswitching as well.
Not as such, but perhaps you could put simplify-rtx to use.
I think most of the bits for this would have come in via
John Wehle's VRP pass.
> Last problem appears to be cost of jump threading. cfg_cleanup now
> takes about 8% of CPU time and it drops to 3% w/o threading and to
> about 1% w/o crossjumping. Do you have some ideas for speedup?
Not off the top of my head. I won't be able to ponder it for a while.
r~