[cfe-commits] r134697 - /cfe/trunk/lib/Analysis/UninitializedValues.cpp

[Chandler Carruth]

On Fri, Jul 8, 2011 at 9:16 AM, Chris Lattner <clattner at apple.com> wrote:

> Wow, this is a great win.  With dataflow, the standard traversal order is a
> reverse postorder traversal (e.g. as vended by
> llvm/include/llvm/ADT/PostOrderIterator.h).  Did you consider using a RPOT?
>

Nope, I was going for minimum change. I'll look into that today if I can get
some time. It should be easy to evaluate especially with the synthetic CFG
test inputs I checked in.

I'm also planning on doing some more fine grained profiling of this warning
with those test cases. I have several ideas for further optimizations, but
not sure they'll pan out. I prioritized testing the daylights out of this to
make sure it didn't regress anything else significantly.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20110708/842dd7a4/attachment.html>