From: Milton Aupperle <milton@outcastsoft.com>

Date: April 27, 2010 11:57:13 AM MDT

To: Astro_IIDC@yahoogroups.com

Subject: Re: [Astro_IIDC] Encoder  self-correction question....

Hi Alberto;

I had a look at the code base and to do what you want is a tonne of work. I have to basically re-write the stacking code for this. And I have to do it twice as I have completely separate code paths for 8 bit and 16 bit movies too.

So I'll add it to the list, but I doubt if it will get implemented until I have to completely re-write all the code and even then I question how many users this benefits.

On an Astro IIDC 4.06.00 Alpha  note, I hope to release A2 later this week which fixes two bugs and should offer some stacking speed improvements. I'm rewriting some of my pixel comparison Altivec code for SSE 3 and there should be a fairly hefty improvement in alignment speed over the previous Universal versions. The main issue is trying to find workaround functions as Altivec has a very rich API, where as SSE does not. And your really hampered with the limited number of registers for SSE has (Altivec has 32 registers, SSE has 8) which means you often have register spillage and stuff has to be loaded / unloaded form the CPU cache, but that has always been the problem with PowerPC versus x86. My bayer decode actually runs within the registers of the PowerPC CPU (I'm using up 24 registers for that), where as it has to hit the cache in x86 multiple times. That means a G4 1.25 gigahz single core PowerPC CPU actually decodes bayer 30% faster decodes than a 1.66 gigahz dual core x86 does. On a G5 with the fatser bus and CPU, it runs 100% faster than x86.

TTYL..

Milton Aupperle

On 26-Apr-10, at 4:17 PM, albe albo wrote:

ok... got the messages about encoding pec etc...

back to Astro IIDC as is.

About batch align with manually selected points:

I already did all my 177 reference movies choosing a certain confidence.

This way the refence movies are already sorted with the 1st frame as the sharpest one. (isn't it?)

Using such reference movies could avoid to preprocess for sharpness estimation and it would be possible to batch select the alignment points  on the first (sharpest) frame.

Perhaps my thought is wrong but perhaps it isn't...

Cheers

TTYL