On 5/29/23 16:51, Martin Jambor wrote:
Hi,
On Mon, May 22 2023, Aldy Hernandez via Gcc-patches wrote:
Implement hashing for ipa_vr. When all is said and done, all these
patches incurr a 7.64% slowdown for ipa-cp, with is entirely covered by
the similar 7% increase in this area last week. So we get type agnostic
ranges with "infinite" range precision close to free.
Do you know why/where this slow-down happens? Do we perhaps want to
limit the "infiniteness" a little somehow?
It happens in ipcp_vr_lattice::meet_with_1, because we have a lot more
sub-ranges to union. The latest numbers are 6.6%, and as I said I've
already improved ipa-cp by the an equal amount, so we're good :).
The infiniteness is already capped. We start at 3 sub-ranges, and grow
up to 255. I suppose if this is a problem, we could write a custom
Value_Range temporary for IPA to fit specific needs, but I'd really like
to avoid such special casing.
Also, jump functions live for a long time, have you looked at how memory
hungry they become? I hope that the hashing would be good at preventing
any issues.
See my previous mail on memory usage.
Aldy
