[FFmpeg-devel] [PATCH 1/2] lavc/vvc: Fix NumEntryPoints derivation

[FFmpeg-devel] [PATCH 1/2] lavc/vvc: Fix NumEntryPoints derivation

[Frank Plowman]

If pps_single_slice_per_subpic_flag is 1,
slice_{width,height}_in_ctus are undefined and we must instead get the
dimensions of the slice by referring to the corresponding subpicture.

Signed-off-by: Frank Plowman <post@frankplowman.com>
---
 libavcodec/cbs_h266_syntax_template.c | 34 ++++++++++++++++++++++-----
 1 file changed, 28 insertions(+), 6 deletions(-)

diff --git a/libavcodec/cbs_h266_syntax_template.c b/libavcodec/cbs_h266_syntax_template.c
index 58211026bb..b84f3e29db 100644
--- a/libavcodec/cbs_h266_syntax_template.c
+++ b/libavcodec/cbs_h266_syntax_template.c
@@ -3440,13 +3440,35 @@ static int FUNC(slice_header) (CodedBitstreamContext *ctx, RWContext *rw,
             for (i = 0; i < current->curr_subpic_idx; i++) {
                 slice_idx += pps->num_slices_in_subpic[i];
             }
-            width_in_tiles =
-                pps->pps_slice_width_in_tiles_minus1[slice_idx] + 1;
 
-            if (entropy_sync)
-                height = pps->slice_height_in_ctus[slice_idx];
-            else
-                height = pps->pps_slice_height_in_tiles_minus1[slice_idx] + 1;
+            if (pps->pps_single_slice_per_subpic_flag) {
+                const int width_in_ctus = sps->sps_subpic_width_minus1[slice_idx] + 1;
+                const int subpic_l = sps->sps_subpic_ctu_top_left_x[slice_idx];
+                const int subpic_r = subpic_l + width_in_ctus;
+
+                int ctb_x = 0, tile_x = 0;
+                for (; ctb_x < subpic_l && tile_x < pps->num_tile_columns; tile_x++)
+                    ctb_x += pps->col_width_val[tile_x];
+
+                width_in_tiles = 0;
+                for (; ctb_x < subpic_r && tile_x < pps->num_tile_columns; tile_x++) {
+                    ctb_x += pps->col_width_val[tile_x];
+                    width_in_tiles++;
+                }
+
+                if (entropy_sync)
+                    height = sps->sps_subpic_height_minus1[slice_idx] + 1;
+                else
+                    height = pps->pps_slice_height_in_tiles_minus1[slice_idx] + 1;
+            } else {
+                width_in_tiles =
+                    pps->pps_slice_width_in_tiles_minus1[slice_idx] + 1;
+
+                if (entropy_sync)
+                    height = pps->slice_height_in_ctus[slice_idx];
+                else
+                    height = pps->pps_slice_height_in_tiles_minus1[slice_idx] + 1;
+            }
 
             current->num_entry_points = width_in_tiles * height;
         } else {
-- 
2.47.0

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

[FFmpeg-devel] [PATCH 2/2] lavc/vvc: Fix pps_single_slice_picture

[Frank Plowman]

Signed-off-by: Frank Plowman <post@frankplowman.com>
---
 libavcodec/vvc/ps.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/libavcodec/vvc/ps.c b/libavcodec/vvc/ps.c
index fae6655cc0..603a6d42d1 100644
--- a/libavcodec/vvc/ps.c
+++ b/libavcodec/vvc/ps.c
@@ -368,9 +368,10 @@ static int pps_add_ctus(VVCPPS *pps, int *off, const int rx, const int ry,
 
 static void pps_single_slice_picture(VVCPPS *pps, int *off)
 {
+    pps->num_ctus_in_slice[0] = 0;
     for (int j = 0; j < pps->r->num_tile_rows; j++) {
         for (int i = 0; i < pps->r->num_tile_columns; i++) {
-            pps->num_ctus_in_slice[0] = pps_add_ctus(pps, off,
+            pps->num_ctus_in_slice[0] += pps_add_ctus(pps, off,
                 pps->col_bd[i], pps->row_bd[j],
                 pps->r->col_width_val[i], pps->r->row_height_val[j]);
         }
-- 
2.47.0

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 2/2] lavc/vvc: Fix pps_single_slice_picture

[James Almer]

[-- Attachment #1.1.1: Type: text/plain, Size: 1077 bytes --]

On 2/23/2025 9:04 AM, Frank Plowman wrote:
> Signed-off-by: Frank Plowman <post@frankplowman.com>
> ---
>   libavcodec/vvc/ps.c | 3 ++-
>   1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/libavcodec/vvc/ps.c b/libavcodec/vvc/ps.c
> index fae6655cc0..603a6d42d1 100644
> --- a/libavcodec/vvc/ps.c
> +++ b/libavcodec/vvc/ps.c
> @@ -368,9 +368,10 @@ static int pps_add_ctus(VVCPPS *pps, int *off, const int rx, const int ry,
>   
>   static void pps_single_slice_picture(VVCPPS *pps, int *off)
>   {
> +    pps->num_ctus_in_slice[0] = 0;
>       for (int j = 0; j < pps->r->num_tile_rows; j++) {
>           for (int i = 0; i < pps->r->num_tile_columns; i++) {
> -            pps->num_ctus_in_slice[0] = pps_add_ctus(pps, off,
> +            pps->num_ctus_in_slice[0] += pps_add_ctus(pps, off,
>                   pps->col_bd[i], pps->row_bd[j],
>                   pps->r->col_width_val[i], pps->r->row_height_val[j]);
>           }

Is there a sample in the conformance suite this fixes? Or one we can 
upload to FATE to make a test.


[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 495 bytes --]

[-- Attachment #2: Type: text/plain, Size: 251 bytes --]

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 2/2] lavc/vvc: Fix pps_single_slice_picture

[Frank Plowman]

On 24/02/2025 02:23, James Almer wrote:
> On 2/23/2025 9:04 AM, Frank Plowman wrote:
>> Signed-off-by: Frank Plowman <post@frankplowman.com>
>> ---
>>   libavcodec/vvc/ps.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/libavcodec/vvc/ps.c b/libavcodec/vvc/ps.c
>> index fae6655cc0..603a6d42d1 100644
>> --- a/libavcodec/vvc/ps.c
>> +++ b/libavcodec/vvc/ps.c
>> @@ -368,9 +368,10 @@ static int pps_add_ctus(VVCPPS *pps, int *off, const int rx, const int ry,
>>   
>>   static void pps_single_slice_picture(VVCPPS *pps, int *off)
>>   {
>> +    pps->num_ctus_in_slice[0] = 0;
>>       for (int j = 0; j < pps->r->num_tile_rows; j++) {
>>           for (int i = 0; i < pps->r->num_tile_columns; i++) {
>> -            pps->num_ctus_in_slice[0] = pps_add_ctus(pps, off,
>> +            pps->num_ctus_in_slice[0] += pps_add_ctus(pps, off,
>>                   pps->col_bd[i], pps->row_bd[j],
>>                   pps->r->col_width_val[i], pps->r->row_height_val[j]);
>>           }
> 
> Is there a sample in the conformance suite this fixes? Or one we can 
> upload to FATE to make a test.

Hi James,

The conformance suite is pretty lacking here considering the number of
combinations of these different partitioning tools possible in VVC.
I think the lack of examples in the conformance suite is the main reason
we're still running into bugs in this area.

I crafted a sample which exhibits this bug though, and have sent it to
samples-request@ffmpeg.org.  I tried to also craft a sample which tests
93aae172ea909ec96b67077b8b3005886a83d0ae, but any attempt at this seemed
to segfault the reference software :)

--
All the best,
Frank

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 2/2] lavc/vvc: Fix pps_single_slice_picture

[Nuo Mi]

On Tue, Feb 25, 2025 at 3:16 AM Frank Plowman <post@frankplowman.com> wrote:

> On 24/02/2025 02:23, James Almer wrote:
> > On 2/23/2025 9:04 AM, Frank Plowman wrote:
> >> Signed-off-by: Frank Plowman <post@frankplowman.com>
> >> ---
> >>   libavcodec/vvc/ps.c | 3 ++-
> >>   1 file changed, 2 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/libavcodec/vvc/ps.c b/libavcodec/vvc/ps.c
> >> index fae6655cc0..603a6d42d1 100644
> >> --- a/libavcodec/vvc/ps.c
> >> +++ b/libavcodec/vvc/ps.c
> >> @@ -368,9 +368,10 @@ static int pps_add_ctus(VVCPPS *pps, int *off,
> const int rx, const int ry,
> >>
> >>   static void pps_single_slice_picture(VVCPPS *pps, int *off)
> >>   {
> >> +    pps->num_ctus_in_slice[0] = 0;
> >>       for (int j = 0; j < pps->r->num_tile_rows; j++) {
> >>           for (int i = 0; i < pps->r->num_tile_columns; i++) {
> >> -            pps->num_ctus_in_slice[0] = pps_add_ctus(pps, off,
> >> +            pps->num_ctus_in_slice[0] += pps_add_ctus(pps, off,
> >>                   pps->col_bd[i], pps->row_bd[j],
> >>                   pps->r->col_width_val[i], pps->r->row_height_val[j]);
> >>           }
> >
> > Is there a sample in the conformance suite this fixes? Or one we can
> > upload to FATE to make a test.
>
> Hi James,
>
> The conformance suite is pretty lacking here considering the number of
> combinations of these different partitioning tools possible in VVC.
> I think the lack of examples in the conformance suite is the main reason
> we're still running into bugs in this area.
>
> I crafted a sample which exhibits this bug though, and have sent it to
> samples-request@ffmpeg.org.  I tried to also craft a sample which tests
> 93aae172ea909ec96b67077b8b3005886a83d0ae, but any attempt at this seemed
> to segfault the reference software :)
>
Creating error-prone conformance streams is challenging; some companies
sell them at a high price

>
> --
> All the best,
> Frank
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 1/2] lavc/vvc: Fix NumEntryPoints derivation

[Nuo Mi]

On Sun, Feb 23, 2025 at 8:05 PM Frank Plowman <post@frankplowman.com> wrote:

> If pps_single_slice_per_subpic_flag is 1,
> slice_{width,height}_in_ctus are undefined and we must instead get the
> dimensions of the slice by referring to the corresponding subpicture.
>
> Signed-off-by: Frank Plowman <post@frankplowman.com>
> ---
>  libavcodec/cbs_h266_syntax_template.c | 34 ++++++++++++++++++++++-----
>  1 file changed, 28 insertions(+), 6 deletions(-)
>
> diff --git a/libavcodec/cbs_h266_syntax_template.c
> b/libavcodec/cbs_h266_syntax_template.c
> index 58211026bb..b84f3e29db 100644
> --- a/libavcodec/cbs_h266_syntax_template.c
> +++ b/libavcodec/cbs_h266_syntax_template.c
> @@ -3440,13 +3440,35 @@ static int FUNC(slice_header)
> (CodedBitstreamContext *ctx, RWContext *rw,
>              for (i = 0; i < current->curr_subpic_idx; i++) {
>                  slice_idx += pps->num_slices_in_subpic[i];
>              }
> -            width_in_tiles =
> -                pps->pps_slice_width_in_tiles_minus1[slice_idx] + 1;
>
> -            if (entropy_sync)
> -                height = pps->slice_height_in_ctus[slice_idx];
> -            else
> -                height = pps->pps_slice_height_in_tiles_minus1[slice_idx]
> + 1;
> +            if (pps->pps_single_slice_per_subpic_flag) {
> +                const int width_in_ctus =
> sps->sps_subpic_width_minus1[slice_idx] + 1;
> +                const int subpic_l =
> sps->sps_subpic_ctu_top_left_x[slice_idx];
> +                const int subpic_r = subpic_l + width_in_ctus;
> +
> +                int ctb_x = 0, tile_x = 0;
> +                for (; ctb_x < subpic_l && tile_x <
> pps->num_tile_columns; tile_x++)
> +                    ctb_x += pps->col_width_val[tile_x];
> +
> +                width_in_tiles = 0;
> +                for (; ctb_x < subpic_r && tile_x <
> pps->num_tile_columns; tile_x++) {
> +                    ctb_x += pps->col_width_val[tile_x];
> +                    width_in_tiles++;
> +                }
> +
> +                if (entropy_sync)
> +                    height = sps->sps_subpic_height_minus1[slice_idx] + 1;
> +                else
> +                    height =
> pps->pps_slice_height_in_tiles_minus1[slice_idx] + 1;
>
  pps_slice_height_in_tiles_minus1 is also undefined if
pps_single_slice_per_subpic_flag is true.

> +            } else {
> +                width_in_tiles =
> +                    pps->pps_slice_width_in_tiles_minus1[slice_idx] + 1;
> +
> +                if (entropy_sync)
> +                    height = pps->slice_height_in_ctus[slice_idx];
> +                else
> +                    height =
> pps->pps_slice_height_in_tiles_minus1[slice_idx] + 1;
> +            }
>
>              current->num_entry_points = width_in_tiles * height;
>          } else {
> --
> 2.47.0
>
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".

Re: [FFmpeg-devel] [PATCH 2/2] lavc/vvc: Fix pps_single_slice_picture

[Nuo Mi]

On Sun, Feb 23, 2025 at 8:05 PM Frank Plowman <post@frankplowman.com> wrote:

> Signed-off-by: Frank Plowman <post@frankplowman.com>
>
Applied this one.
Thank you, Frank

> ---
>  libavcodec/vvc/ps.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/libavcodec/vvc/ps.c b/libavcodec/vvc/ps.c
> index fae6655cc0..603a6d42d1 100644
> --- a/libavcodec/vvc/ps.c
> +++ b/libavcodec/vvc/ps.c
> @@ -368,9 +368,10 @@ static int pps_add_ctus(VVCPPS *pps, int *off, const
> int rx, const int ry,
>
>  static void pps_single_slice_picture(VVCPPS *pps, int *off)
>  {
> +    pps->num_ctus_in_slice[0] = 0;
>      for (int j = 0; j < pps->r->num_tile_rows; j++) {
>          for (int i = 0; i < pps->r->num_tile_columns; i++) {
> -            pps->num_ctus_in_slice[0] = pps_add_ctus(pps, off,
> +            pps->num_ctus_in_slice[0] += pps_add_ctus(pps, off,
>                  pps->col_bd[i], pps->row_bd[j],
>                  pps->r->col_width_val[i], pps->r->row_height_val[j]);
>          }
> --
> 2.47.0
>
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".