On Fri, Jan 10, 2020 at 11:23:34AM +0100, Richard Biener wrote:
>
> The following manages to avoid high EH indegree of landing pads
> during the sequence of cleaning up empty EH with a chain of those.
> By walking the landing pads in reverse order we mimic walking of
> the EH tree depth-first (which I am too lazy to write...). It
> looks like EH build assures that this is effectively the same.
>
> Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
>
> OK?
>
> With this the RedHat bugzilla testcase is down to a "flat" profile:
>
> parser function body : 14.60 ( 17%) 1.77 ( 31%) 16.37 (
> 18%) 720798 kB ( 23%)
> tree eh : 14.80 ( 17%) 0.06 ( 1%) 14.87 (
> 16%) 246315 kB ( 8%)
> integrated RA : 13.08 ( 15%) 0.30 ( 5%) 13.40 (
> 15%) 227799 kB ( 7%)
> TOTAL : 85.07 5.68 90.77
>
> 3183334 kB
>
> and memory usage is down as well to 4GB (it's actually this very patch
> that helps here for reasons I have not investigated - maybe we never
> shrink some EH data structures, who knows).
>
> Thanks,
> Richard.
>
> 2020-01-10 Richard Biener <[email protected]>
>
> PR middle-end/93199
> * tree-eh.c (redirect_eh_edge_1): Avoid some work if possible.
> (cleanup_all_empty_eh): Walk landing pads in reverse order to
> avoid quadraticness.
LGTM.
Jakub
