On Thu, Aug 09, 2012 at 08:23:10PM -0700, Ronald S. Bultje wrote:
> On Thu, Aug 9, 2012 at 4:49 PM, Diego Biurrun <[email protected]> wrote:
> > On Sun, Aug 05, 2012 at 07:20:40AM -0700, Ronald S. Bultje wrote:
> >> On Sun, Aug 5, 2012 at 2:44 AM, Diego Biurrun <[email protected]> wrote:
> >> > On Sat, Aug 04, 2012 at 06:19:38PM -0700, Ronald S. Bultje wrote:
> >> >> On Sat, Aug 4, 2012 at 2:28 PM, Loren Merritt <[email protected]> 
> >> >> wrote:
> >> >> > On Sat, 4 Aug 2012, Diego Biurrun wrote:
> >> >> >> On Sat, Aug 04, 2012 at 03:11:50PM -0400, Justin Ruggles wrote:
> >> >> >>> On 07/31/2012 06:17 PM, Diego Biurrun wrote:
> >> >> >>>> ---
> >> >> >>>>  30 files changed, 215 insertions(+), 211 deletions(-)
> >> >> >>>
> >> >> >>> Looks ok, but probably should get other opinions on this as well. I 
> >> >> >>> know
> >> >> >>> Ronald was trying to keep x86inc.asm sychronized with x264, and 
> >> >> >>> trying
> >> >> >>> to do so after this change would likely require similar extensive
> >> >> >>> cpuflag modifications in x264.
> >> >> >>
> >> >> >> I volunteer to patch x264 if such a change would be accepted on their
> >> >> >> side.
> >> >> >
> >> >> > Rejected. I like "mmx2" better.
> >> >> > However, I wouldn't be opposed to dropping mmx1 entirely and using
> >> >> > the name "mmx" to refer to mmx2. (x264 doesn't actually support mmx1
> >> >> > anyway; we use mmx2 inline asm that's actually inlined in places where
> >> >> > runtime cpu detection is impossible.) But that wouldn't help
> >> >> > synchronization if libav doesn't do so.
> >> >>
> >> >> Right - from my PoV, keeping us in sync with x264 is more important
> >> >> that having a slightly more accurate cpuflag for something that's
> >> >> never exposed to end users in any way.
> >> >
> >> > It is exposed through configure.
> >> >
> >> >> Diego, have ideas on how to fix this,
> >> >
> >> > Yes, why don't you convince Loren that mmxext is the more sensible name?
> >> >
> >> >> or can you live with changing everything to MMX2 instead?
> >> >
> >> > I believe mmxext is the better name and everybody except Loren seems
> >> > to agree.  There are very few uses of mmx2/mmxext in x86inc.asm and
> >> > they are unlikely to ever conflict with future changes to this file.
> >> > Besides, we only sync a few times per year.
> >> >
> >> > So I'm unconvinced that using mmx2 would be a net benefit.  I still
> >> > hope that Loren can change his mind, as I said, I volunteer to do
> >> > all the work.
> >>
> >> x86inc.asm is a x264 file that we sync, I'd like to keep it that way,
> >> especially given that I'm the guy who historically had to deal with
> >> the outfalls every time a sync doesn't work right (remember INIT_MMX
> >> not backing up XMM regs anymore on Win64? Remember the register
> >> reordering on Win64 recently?).
> >
> > Look at where in x86inc.asm the string mmx2/mmxext appears.  Merge
> > conflicts are very unlikely and trivial to fix.
> >
> >> Plus, I didn't say it was a good idea, I said I could live with it if
> >> others want it. Right now, it seems others (i.e. Loren) don't.
> >
> > So it's a stalemate?
> 
> OMG you're actually going to argue over something silly like this?
> 
> Please. I'm the one who has to fix these issues, and they have
> happened a lot over the past year or two. I'm asking you to not do
> this.

There's a simple solution if you are so afraid of conflicts, albeit they
are unlikely to happen and be trivial to fix.  I can just add

  %define cpuflags_mmxext cpuflags_mmx2

to the top of x86inc.asm.  Problem solved.

Diego
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to