[XviD-devel] xvid_bench from HEAD cvs gave me 2 quantizer CRCerrors...is this expected?

Joe Millenbach jmillenbach at hotmail.com
Sun Jun 15 00:08:14 CEST 2003


The initial change I suggested was just performing the computation as .NET 
would (1000000/1000 = 1000) ahead of time.  The parenthesis is a better idea 
if what I'm guessing is the issue, and changes my output from 20 items being 
negative to 1...and weirdly enough the 1 isn't one of the original 20.

Multiply and divide have the same precedence, and it's left to right.  So 
possibly multiplying by 1,000,000 (if the item is already at greater than 4) 
would overload a 32 bit value (a good guess for the size of most items on 
x86 I would think)...and give the wrong answer at the end.

And this happens on both HEAD and the dev-api-4 versions.

>From: Edouard Gomez <ed.gomez at free.fr>
>Reply-To: xvid-devel at xvid.org
>To: xvid-devel at xvid.org
>Subject: Re: [XviD-devel]  xvid_bench from HEAD cvs gave me 2 quantizer 
>CRCerrors...is this expected?
>Date: Wed, 11 Jun 2003 22:56:11 +0200
>
>suxen_drol (suxen_drol at hotmail.com) wrote:
> > commit
>
>Well in fact the problem is that skal does a noop for loop to measure
>matrix filling overhead. The problem is the overhead seems not to be the
>same in the real for loops.
>
>Dunno if we can avoid that negative values. Anyway the millisecond
>measurement unit is the wrong way to fix that.
>
>--
>Edouard Gomez
>_______________________________________________
>XviD-devel mailing list
>XviD-devel at xvid.org
>http://list.xvid.org/mailman/listinfo/xvid-devel

_________________________________________________________________
Add photos to your messages with MSN 8. Get 2 months FREE*.  
http://join.msn.com/?page=features/featuredemail



More information about the XviD-devel mailing list