[PATCH] D28129: NewGVN: Sort Dominator Tree in RPO order, and use that for generating order.

Davide Italiano via llvm-commits llvm-commits at lists.llvm.org
Fri Dec 30 10:52:42 PST 2016


On Tue, Dec 27, 2016 at 12:47 PM, Daniel Berlin via Phabricator via
llvm-commits <llvm-commits at lists.llvm.org> wrote:
> dberlin created this revision.
> dberlin added reviewers: chandlerc, davide.
> dberlin added a subscriber: llvm-commits.
>
> The optimal iteration order for this problem is RPO order. We want to
> process as many preds of a backedge as we can before we process the
> backedge.
>
> At the same time, as we add predicate handling, we want to be able to
[...]

Replying to the original mail, something I forgot to mention earlier.
As you're speaking about predicate handling, are you planning to stick
with what the paper describes?
The main reason why I'm asking is because in our early discussions you
mentioned the possibility of using Bodik's e-SSA to potentially reduce
the worst case complexity of predicate handling by a factor of N. At
that point, we could kill one bird with one stone (kinda) using e-SSA
for CVP as explained in the CGO paper you propose here
http://lists.llvm.org/pipermail/llvm-dev/2016-January/093999.html (the
code on github has actually an e-SSA implementation, uses 3.4 but
porting shouldn't be a lot of work).

For those watching/interested, the reference is:
http://www.sdsc.edu/~allans/cs231/ABCD.ps

-- 
Davide

"There are no solved problems; there are only problems that are more
or less solved" -- Henri Poincare


More information about the llvm-commits mailing list