[FFmpeg-devel] [PATCH 1/8] swscale: fix sws_setColorspaceDetails after sws_init_context

[Michael Niedermayer]

On Sat, Oct 28, 2023 at 12:46:38PM +0200, Niklas Haas wrote:
> On Fri, 27 Oct 2023 23:42:41 +0200 Michael Niedermayer <michael at niedermayer.cc> wrote:
> > On Fri, Oct 27, 2023 at 07:04:39PM +0200, Niklas Haas wrote:
> > > From: Niklas Haas <git at haasn.dev>
> > > 
> > > More commonly, this fixes the case of sws_setColorspaceDetails after
> > > sws_getContext, since the latter implies sws_init_context.
> > > 
> > > The problem here is that sws_init_context sets up the range conversion
> > > and fast path tables based on the values of srcRange/dstRange at init
> > > time. This may result in locking in a "wrong" path (either using
> > > unscaled fast path when range conversion later required, or using
> > > scaled slow path when range conversion becomes no longer required).
> > > 
> > > There are two way outs:
> > > 
> > > 1. Always initialize range conversion and unscaled converters, even if
> > >    they will be unused, and extend the runtime check.
> > > 2. Re-do initialization if the values change after
> > >    sws_setColorspaceDetails.
> > > 
> > > I opted for approach 1 because it was simpler and easier to reason
> > > about.
> > > ---
> > >  libswscale/swscale.c | 2 +-
> > >  libswscale/utils.c   | 5 +----
> > >  2 files changed, 2 insertions(+), 5 deletions(-)
> > > 
> > > diff --git a/libswscale/swscale.c b/libswscale/swscale.c
> > > index 90e5b299ab..46ba68fe6a 100644
> > > --- a/libswscale/swscale.c
> > > +++ b/libswscale/swscale.c
> > > @@ -1016,7 +1016,7 @@ static int scale_internal(SwsContext *c,
> > >      reset_ptr(src2, c->srcFormat);
> > >      reset_ptr((void*)dst2, c->dstFormat);
> > >  
> > > -    if (c->convert_unscaled) {
> > > +    if (c->convert_unscaled && !c->lumConvertRange && !c->chrConvertRange) {
> > >          int offset  = srcSliceY_internal;
> > >          int slice_h = srcSliceH;
> > >  
> > > diff --git a/libswscale/utils.c b/libswscale/utils.c
> > > index e1ad685972..455955e907 100644
> > > --- a/libswscale/utils.c
> > > +++ b/libswscale/utils.c
> > > @@ -1728,9 +1728,7 @@ static av_cold int sws_init_single_context(SwsContext *c, SwsFilter *srcFilter,
> > >      }
> > >  
> > >      /* unscaled special cases */
> > > -    if (unscaled && !usesHFilter && !usesVFilter &&
> > > -        (c->srcRange == c->dstRange || isAnyRGB(dstFormat) ||
> > > -         isFloat(srcFormat) || isFloat(dstFormat))){
> > > +    if (unscaled && !usesHFilter && !usesVFilter) {
> > >          ff_get_unscaled_swscale(c);
> > >  
> > >          if (c->convert_unscaled) {
> > > @@ -1738,7 +1736,6 @@ static av_cold int sws_init_single_context(SwsContext *c, SwsFilter *srcFilter,
> > >                  av_log(c, AV_LOG_INFO,
> > >                         "using unscaled %s -> %s special converter\n",
> > >                         av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
> > > -            return 0;
> > >          }
> > >      }
> > 
> > the av_log() message will be wrong if this path is unused
> 
> Indeed. Perhaps the only way to fix that would be to instead try
> approach 2 and go through full reinit if any params change after
> setColorspaceDetails.
> 

> > also this ties convert_unscaled to never do range conversion, if thats
> > intended, i guess thats ok. Otherwise that maybe is a restriction
> > we dont want to add.
> 
> In the original code, c->convert_unscaled is set behind a `c->srcRange
> == c->dstRange` check, so how is that a change in behavior?

not "change in behavior" but it may be more complex if a
yuv->yuv unscaled convert is added that supports range convert
because before convert_unscaled is set only when it can be used
afterwards convert_unscaled is always set but sometimes it then
can be used with differing yuv ranges sometimes not.

what i meant is that the conditions related to convert_unscaled
could become more dispersed and duplciated _IF_ specific future
code is added
also its not just c->srcRange == c->dstRange, the condition is
more complex

thx

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Freedom in capitalist society always remains about the same as it was in
ancient Greek republics: Freedom for slave owners. -- Vladimir Lenin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20231028/b492273f/attachment.sig>