On Fri, Jun 24, 2011 at 07:43:20AM -0700, Ronald S. Bultje wrote:
> This means that precision is retained when scaling between sample
> formats with >8 bits per component (48bit RGB, 16bit grayscale,
> 9/10/16bit YUV). Also document hscale functions.
>
> Known problems:
> - 1:1 scaling path (unscaled) uses a 4x hscaling path, there should be
> a special 1x (point/noscale) and 2x bilin hscale filter path like in
^^^ bilinear
> vscale/output. Also, 1x/2x vscale uses 4095 as 1-x path somewhere,
> should be 4096.
Does anybody understand the last sentence?
> - in hScale(), xInc and srcW are unused - remove them.
> - can we not align hScale() filterSize (as a value) by 4? Allows faster
> asm in some cases.
These two lines should not go to Git.
> - no dither yet.
> - DSP'ize the 8->16 and 19->15 bit routines in hy/cscale().
Somewhere Diego is crying (and not because he doesn't work on FSV2 decoder).
> ---
> libswscale/ppc/swscale_altivec.c | 2 +
> libswscale/swscale.c | 649
> +++++++++++++++++++++++--------------
> libswscale/swscale_internal.h | 51 +++
> libswscale/utils.c | 9 +-
> libswscale/x86/swscale_template.c | 73 +----
> 5 files changed, 476 insertions(+), 308 deletions(-)
In general looks ok but please edit the log message before committing.
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
