[llvm-dev] RFC: Dynamic dominators

[Jakub (Kuba) Kuderski via llvm-dev]

>
> Do you mean the algorithm is slower in debug builds of LLVM, or when
> LLVM is used to do debug builds of things?
>
> It sounds like you say the algorithm can be 8% slower for debug
> builds, but also likely to speed up debug builds? I'm confused :-)

The algorithm can be slower when LLVM is built in Debug mode instead of
Release.
DomTree is only an analysis, so as long as it is correct it doesn't affect
performance of compiled programs.

On Tue, Jun 13, 2017 at 11:44 AM, Hans Wennborg <hans at chromium.org> wrote:

> On Mon, Jun 12, 2017 at 5:12 PM, Jakub (Kuba) Kuderski via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
> > 4. Performance
> >
> > The real life performance of full dynamic DBS-based DomTree
> recalculation is
> > between 20% and 2% better on a machine with two Xeon E5-2680v2 CPUs than
> the
> > existing SLT algorithm, which is consistent with the findings from this
> > thesis [3]. The advantage is not that visible on debug builds, where the
> > DBS-algorithm can be up to 8% slower. It is most like possible to speed
> up
> > debug builds, but I haven’t looked into that yet.
>
> Do you mean the algorithm is slower in debug builds of LLVM, or when
> LLVM is used to do debug builds of things?
>
> It sounds like you say the algorithm can be 8% slower for debug
> builds, but also likely to speed up debug builds? I'm confused :-)
>



-- 
Jakub Kuderski
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170613/43704ce6/attachment.html>