[llvm-commits] [llvm] r77397 - in /llvm/trunk: include/llvm/Support/FormattedStream.h lib/Support/FormattedStream.cpp

[David Greene]

On Wednesday 29 July 2009 11:43, Daniel Dunbar wrote:
> On Wed, Jul 29, 2009 at 9:03 AM, David Greene<dag at cray.com> wrote:
> > On Tuesday 28 July 2009 23:48, Evan Cheng wrote:
> >> Ok. David, I'm going to revert AsmWriterEmitter.cpp back to 74742 for
> >> now. Please re-commit as you have fixed the efficiency problem.
> >
> > As you know, I can't SEE the efficiency problem.  I'm flying blind.
>
> I don't understand this, what times do you see for the example I gave
> (or something similar)? I can't see why they wouldn't reproduce
> elsewhere.

We knew the N^2 algorithm would be a problem but Chris asked me to commit
anyway.  I reverted your revert and applied a fix to make the algorithm
linear again.  But due to Evan's revert this doesn't get called yet.  That's
fine.

I am not able to see the original compile time regression caused by
excessive flushing.  That's what I was referring to.

I'm trying to get us back to where we were.  I think I have it and am
running tests now.  I will be reverting Evan's revert after making sure
I don't see a problem.  But as I said, that's no guarantee there won't
be some kind of problem I can't see.

                                    -Dave