[FFmpeg-devel] [PATCH 4/4] mpegaudio_tablegen: Don't use llrint

[Michael Niedermayer]

On Tue, Jan 15, 2013 at 08:09:52PM -0500, Derek Buitenhuis wrote:
> On 15/01/2013 7:38 PM, Michael Niedermayer wrote:
> > On Tue, Jan 15, 2013 at 07:07:27PM -0500, Derek Buitenhuis wrote:
> >> On 15/01/2013 6:46 PM, Michael Niedermayer wrote:
> >>> ok if it produces the same values
> >>> otherwise might be worth looking at the differences
> >>
> >> Same values. I ran FATE on both Linux and Windows.
> > 
> > just to make sure i dont misunderstand
> > you checked that the table values match or just that fate passes
> > 
> > i do not think fate tests all values from the tables ...
> 
> The llrint and lrint replacements produce exact tables, but the cbrtf replacement
> based off libm.h's has some differences such as:
> 
> -    7.958060377802014720e+17, 8.854350645805711360e+17, 9.773938878007214080e+17, 1.071570082376712192e+18
> +    7.958060377802014720e+17, 8.854351333000478720e+17, 9.773939565201981440e+17, 1.071570082376712192e+18
> 
> Mind you, these would be present on any system lacking cbrtf at runtime as well.
> 
> Comments / Ideas?

do these occur with pow() too or just powf() ?
in powf the 1.0/3  is at a much lower precission compared to a ideal
cbrtf()

but i dont think small differences here a a major concern


[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Those who are best at talking, realize last or never when they are wrong.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20130116/1a1710a5/attachment.asc>