[LLVMdev] GSoC 2009 application

Eli Friedman eli.friedman at gmail.com
Sun Mar 29 17:28:17 PDT 2009


On Sun, Mar 29, 2009 at 4:39 PM, Benoit Boissinot
<bboissin+llvm at gmail.com> wrote:
> You don't need to break any edges, the copies are "semantically" done on
> the edge. When you have a critical edge, you can simply fold the sigma into
> the following phi's.

The issue would be that the sigma output doesn't have a unique
identity... I'm not too familiar with algorithms using SSI, though, so
I don't know if that would be an issue.

> Where is the performance a concern? Using phi's instead of sigma's? Or all
> the new variables introduced by SSI? (I think that's a real concern,
> if I remember
> correctly the numbers from Ananian and Singer).

The performance concern is all the new PHI nodes getting inserted...

-Eli



More information about the llvm-dev mailing list