Andreas wrote:
Nice :)
I encountered a few days ago that the mixing code reduces quality.
Is this eliminated with the higher precision now?
Using fixed floats there is still a loss of quality.
To be more precise:
- S16 multiplication discards 5 bits if "accuracy" is not enabled in
configuration (./configure --enable-accuracy=yes)
- S24 always lose at least 2 bits
- S32 always lose at least 10 bits
I assume that dok's original code is still faster than your fixed point
solution?
More or less it's fast as the original code:
the differences are some additional shifts (from 1 to 3) during
buffer mixing, and an additional sum and shift before writing
the output to the device.
If that's true, it might be wise to include dok's code as
well. (For slow embedded devices)
The old code is no longer compatible because it was written
to always output Signed 16 samples.
With "accuracy" disabled, the new code should work fine on
embedded devices, too.
--
Regards,
Claudio Ciccani
[EMAIL PROTECTED]
http://directfb.org
http://sf.net/projects/php-directfb
_______________________________________________
directfb-dev mailing list
[email protected]
http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-dev
