On Mon, Oct 24, 2011 at 08:30:21PM +0100, Davide Anastasia wrote:
> On 24/10/2011 19:18, Bruce Guenter wrote:
> > There should be nothing in particular wrong with the fattal02 and
> > mantiuk06 commits. In fact, I've been using them for some time. I was
> > holding them back to double check that they were correct and to test how
> > much of a benefit they provided.
> Ok, then let's leave them: I don't see any reason to revert the entire
> However, if you prefer, I can branch your last commits and bring back
> the master branch to the previous state.
At this point, given that others have fetched it, I'd say not to. More
hassle than it's worth.
> How much confident are you about those changes?
> I've seen you have removed a bunch of lines from fattal02: are you sure
> they are really unnecessary?
Quite confident. The wholesale function removals should be obvious --
nothing references them, and if something did the compile would fail.
The parameter removals were done carefully reviewing all callers
(usually just one) to make sure the parameter was in fact constant.
> Other options are here: http://cmake.org/Wiki/CMake_Useful_Variables, in
> particular there is the option "Minimum size", which should bring the
> optimization to -O3.
I doubt it. -O3 introduces things like loop unrolling which are
virtually guaranteed to expand the code size. More likely it would use
-Os for gcc at least, which tells it to optimize for size.
> However, from my experience, you won't see any
> noticable effect between -O2 and -O3 with GCC. Correct me if I'm wrong.
It could be for Luminance. For some code bases there is enough
difference to justify using it. I haven't benchmarked that yet.
> Let's use the mailing list from now on.
Bruce Guenter <bruce@...> http://untroubled.org/