On Mon, Jun 10, 2013 at 12:07 PM, <[email protected]> wrote:
> > With these instructions, we can collect results from a variety of
> > machines, and perhaps narrow down the factors that affect errors in
> > the bit exactness of the codec and modem.
>
> Small differences in the bit-exactness of floating point calculation are
> not errors. That's the problem. These happen due to instruction order and
> register use changes due to the compiler and optimizer. Floating point
> numbers are not counting numbers, and do not provide bit-exact results. We
> don't have a test suite for the codec or modem that is tolerant of small
> deltas of the type that would result from optimization. This kind of
> problem is going to keep coming up until we do.
Perhaps in this case the difference is between 32bit and 64bit arches
instead of compiler flags?
Richard
------------------------------------------------------------------------------
This SF.net email is sponsored by Windows:
Build for Windows Store.
http://p.sf.net/sfu/windows-dev2dev
_______________________________________________
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
