This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Speed up DOM -- pr15524
On Wed, 2004-09-29 at 05:40, Jan Hubicka wrote:
> How did you measured this?
"this" referring to the ggc_free cost. via gprof.
On the smaller testcase we got > 1 million calls to ggc_free, 97% of
them coming from remove_edge. This accounted for roughly 1.5% of the
total compilation time.
> I oprofiled this testcase about month ago
> and didn't see ggc_free on the top (only ggc_alloc). Perhaps something
> changed with the edge vector changes, I will try again on the yesterday
> code.
Did you profile before or after the addition of tree-ssa-threadupdate?
As I mentioned in the message, tree-ssa-threadupdate had the rather
nasty behavior of creating an insane number of useless edges for
cases where we thread a large number of incoming edges to a large
number of unique outgoing edges.
> Also can your patch be resoponsible for 0.6% memory consumption increase
> on combine.c tester reported tonight? It is not big deal, but would be
> nice to keep track of things...
Highly unlikely. It would have to be some highly obscure subtle
change since the majority of the benefit of this patch comes from
*reducing* the amount of useless tree nodes and edges we create.
> Also what is making tree CFG cleanup that slow?
I haven't looked at it.
jeff