amd64 arch and sse optimization

[Jaromír Mikeš]

2014-02-03 Reinhard Tartler <siretart at gmail.com>:

> On Mon, Feb 3, 2014 at 2:01 PM, Jaromír Mikeš <mira.mikes at gmail.com>
> wrote:
>
> > Please go ahead ... It looks like debian-devel would be right place.
> > It sound strange that sse is redundant in amd64.
>
> Please elaborate, why is this strange?
>
> > Some time ago I played with sse optimization flags and packages build
> with
> > and without them
> > got 30-40% difference in performance in my cases.
> >
> > Would be great to make it clear.
>
> Unfortunately not. Please clarify what exact packages did you compile
> with what additional see flags on what architecture to get the 30-40%
> difference.
>

I did test with jconvover ... I had to dig in old emails exchanged with
Free Ekanayaka.
I was 4 years ago! ;)
Used flags -mmmx -msse -mfpmath=sse
Improvement in performance wasn't so huge ... it was about 15% ... sorry
for my bad memory.
Testing also haven't been very scientific just running as much as possible
instances of jconvolver until first xruns in Jack.


> Yes, this can make a huge difference, but using those flags is like
> firing a shotgun: it may solve some symptoms, but is likely to cause but
> is likely to cause
> collateral damage.


Sure, that's why I would like to know more.
Not make big disaster instead of little improvement.

mira
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-multimedia-maintainers/attachments/20140203/9318dd6a/attachment-0001.html>