[FFmpeg-devel] [Issue 664] [PATCH] Fix AAC PNS Scaling

[Guillaume POIRIER]

Hello,

On Tue, Oct 7, 2008 at 11:47 PM, M?ns Rullg?rd <mans at mansr.com> wrote:
> Uoti Urpala <uoti.urpala at pp1.inet.fi> writes:
>> Plain C says so little about floating point that it is impossible to
>> write a useful program relying only on "guaranteed to work by the
>> standard" assumptions. There's no widely used standard between the
>> precision levels of C and IEEE-754, so in practice a program must choose
>> between  1) require IEEE-754  2) list a custom set of assumptions  3)
>> decide "how likely is this to work on practical platforms?" on a
>> case-by-case basis.
>
> Add the fact that many CPUs have a fast, not-quite-IEEE floating-point
> mode.

So what it's worth the Cell's SPU doesn't strictly follow IEEE 754:

"The SPU supports both single-precision and double-precision
floating-point operations. Single-
precision instructions are performed in 4-way SIMD fashion. The data
formats for single-preci-
sion and double-precision instructions are those defined by IEEE
Standard 754, but the results
calculated by single-precision instructions depart from the IEEE
Standard 754 by placing
emphasis on real-time graphics requirements that are typical of
multimedia processing. "



Guillaume
-- 
One should not give up hope on imbeciles. With a little training, you
can make them into soldiers.
 -- Pierre Desproges