[FFmpeg-devel] Different behavior of ff_float_to_int16_c() and float_to_int16_sse()

[Michael Niedermayer]

On Mon, Mar 03, 2008 at 08:14:15PM +0800, Zuxy Meng wrote:
> Hi,
> 
> Obviously, these two functions produce quite different output given
> the same input. So IIRC whenever trying to use dsp.float_to_int16 we
> have to check if it's the C version or the SIMD version, but in

There should be a float_to_int16_offset & float_to_int16_scale in the
dsp context, so that the "user" can provide proper input. Or one of
the 2 functions should be changed to behave like the other (the multiply
is easier to merge into the transform/dequant than the add ...) so the
add could be merged into float_to_int16().

Benchmarks for the merge case would be required for it to be accepted.


[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

While the State exists there can be no freedom; when there is freedom there
will be no State. -- Vladimir Lenin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080303/9712efec/attachment.pgp>