This is the mail archive of the
mailing list for the GCC project.
Re: [patch] tree-ssa-ccp.c: Propagate ADDR_EXPRs further.
On Fri, 2005-05-27 at 11:47 -0400, Kazu Hirata wrote:
> Hi Jeff,
> > I'd like to see at least some measurement of compile-time impacts.
> Here is a timing in seconds for five runs of cc1 -O2 for several cc1-i
> files. (Actually each is run 6 times with the timing from the first
> run discarded to discount hot cache effects.)
> original patched diff%
> c-common.i 22.403 22.338 -0.290%
> combine.i 18.668 18.676 +0.042%
> fold-const.i 42.545 42.531 -0.032%
> reload1.i 14.505 14.509 +0.027%
> reload.i 13.394 13.379 -0.111%
> insn-attrtab.i 200.815 200.856 +0.020%
Looks like noise to me. I can speculate that we're much better
at systematically avoiding moving through the lattice -- ie
we're much better at identifying those objects which are VARYING
and thus avoiding a lot of transitions and re-evaluations.
I'm OK with the patch. If you wouldn't mind, can you investigate
removing the fold_XXX_to_const routines?