[PATCH] media: verisilicon: HEVC: Initialize start_bit field

[PATCH] media: verisilicon: HEVC: Initialize start_bit field

[Benjamin Gaignard]

Always set start_bit field to 0, if not it could lead to corrupted frames
specially when decoding VP9 bitstreams at the same time since VP9 driver
set it for it own purpose.

Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
---
 drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
index 85a44143b378..0e212198dd65 100644
--- a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
+++ b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
@@ -518,6 +518,7 @@ static void set_buffers(struct hantro_ctx *ctx)
 	hantro_reg_write(vpu, &g2_stream_len, src_len);
 	hantro_reg_write(vpu, &g2_strm_buffer_len, src_buf_len);
 	hantro_reg_write(vpu, &g2_strm_start_offset, 0);
+	hantro_reg_write(vpu, &g2_start_bit, 0);
 	hantro_reg_write(vpu, &g2_write_mvs_e, 1);
 
 	hantro_write_addr(vpu, G2_TILE_SIZES_ADDR, ctx->hevc_dec.tile_sizes.dma);
-- 
2.43.0

Re: [PATCH] media: verisilicon: HEVC: Initialize start_bit field

[Benjamin Gaignard]

Le 20/01/2025 à 09:10, Benjamin Gaignard a écrit :
> Always set start_bit field to 0, if not it could lead to corrupted frames
> specially when decoding VP9 bitstreams at the same time since VP9 driver
> set it for it own purpose.
>
> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>

I could add this tag:

Fixes: cb5dd5a0fa518 ("media: hantro: Introduce G2/HEVC decoder")

> ---
>   drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c | 1 +
>   1 file changed, 1 insertion(+)
>
> diff --git a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> index 85a44143b378..0e212198dd65 100644
> --- a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> +++ b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> @@ -518,6 +518,7 @@ static void set_buffers(struct hantro_ctx *ctx)
>   	hantro_reg_write(vpu, &g2_stream_len, src_len);
>   	hantro_reg_write(vpu, &g2_strm_buffer_len, src_buf_len);
>   	hantro_reg_write(vpu, &g2_strm_start_offset, 0);
> +	hantro_reg_write(vpu, &g2_start_bit, 0);
>   	hantro_reg_write(vpu, &g2_write_mvs_e, 1);
>   
>   	hantro_write_addr(vpu, G2_TILE_SIZES_ADDR, ctx->hevc_dec.tile_sizes.dma);

Re: [PATCH] media: verisilicon: HEVC: Initialize start_bit field

[Nicolas Dufresne]

Le lundi 20 janvier 2025 à 11:00 +0100, Benjamin Gaignard a écrit :
> Le 20/01/2025 à 09:10, Benjamin Gaignard a écrit :
> > Always set start_bit field to 0, if not it could lead to corrupted frames
> > specially when decoding VP9 bitstreams at the same time since VP9 driver
> > set it for it own purpose.

               its

> > 
> > Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
> 
> I could add this tag:
> 
> Fixes: cb5dd5a0fa518 ("media: hantro: Introduce G2/HEVC decoder")

I have tested this on IMX8MQ board using the following GStreamer pipeline.
Before the change, the HEVC window was entirely corrupted. The streams don't
matter as long as they use both HEVC and VP9 codec.

gst-launch-1.0 \
  filesrc location=hevc.mp4 ! parsebin ! v4l2slh265dec ! fakevideosink \
  filesrc location=vp9.mkv ! parsebin ! v4l2slvp9dec ! fakevideosink

Tested-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>


> 
> > ---
> >   drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c | 1 +
> >   1 file changed, 1 insertion(+)
> > 
> > diff --git a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> > index 85a44143b378..0e212198dd65 100644
> > --- a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> > +++ b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> > @@ -518,6 +518,7 @@ static void set_buffers(struct hantro_ctx *ctx)
> >   	hantro_reg_write(vpu, &g2_stream_len, src_len);
> >   	hantro_reg_write(vpu, &g2_strm_buffer_len, src_buf_len);
> >   	hantro_reg_write(vpu, &g2_strm_start_offset, 0);
> > +	hantro_reg_write(vpu, &g2_start_bit, 0);
> >   	hantro_reg_write(vpu, &g2_write_mvs_e, 1);

I've also crossed against "decoder_swreg_map_g2.xlsx" documentation, if you are
lucky to have access to that, and within swreg5 there is only g2_tempor_mvp_e
that is also shared, and its already being set in both drivers.

Reviewed-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>


> >   
> >   	hantro_write_addr(vpu, G2_TILE_SIZES_ADDR, ctx->hevc_dec.tile_sizes.dma);
> 

Re: [PATCH] media: verisilicon: HEVC: Initialize start_bit field

[Adam Ford]

On Mon, Jan 20, 2025 at 8:10 AM Nicolas Dufresne
<nicolas.dufresne@collabora.com> wrote:
>
> Le lundi 20 janvier 2025 à 11:00 +0100, Benjamin Gaignard a écrit :
> > Le 20/01/2025 à 09:10, Benjamin Gaignard a écrit :
> > > Always set start_bit field to 0, if not it could lead to corrupted frames
> > > specially when decoding VP9 bitstreams at the same time since VP9 driver
> > > set it for it own purpose.
>
>                its
>

Does this impact the Fluster score?

> > >
> > > Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
> >
> > I could add this tag:
> >
> > Fixes: cb5dd5a0fa518 ("media: hantro: Introduce G2/HEVC decoder")
>
> I have tested this on IMX8MQ board using the following GStreamer pipeline.
> Before the change, the HEVC window was entirely corrupted. The streams don't
> matter as long as they use both HEVC and VP9 codec.
>
> gst-launch-1.0 \
>   filesrc location=hevc.mp4 ! parsebin ! v4l2slh265dec ! fakevideosink \
>   filesrc location=vp9.mkv ! parsebin ! v4l2slvp9dec ! fakevideosink
>
> Tested-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
>
>
> >
> > > ---
> > >   drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c | 1 +
> > >   1 file changed, 1 insertion(+)
> > >
> > > diff --git a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> > > index 85a44143b378..0e212198dd65 100644
> > > --- a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> > > +++ b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> > > @@ -518,6 +518,7 @@ static void set_buffers(struct hantro_ctx *ctx)
> > >     hantro_reg_write(vpu, &g2_stream_len, src_len);
> > >     hantro_reg_write(vpu, &g2_strm_buffer_len, src_buf_len);
> > >     hantro_reg_write(vpu, &g2_strm_start_offset, 0);
> > > +   hantro_reg_write(vpu, &g2_start_bit, 0);
> > >     hantro_reg_write(vpu, &g2_write_mvs_e, 1);
>
> I've also crossed against "decoder_swreg_map_g2.xlsx" documentation, if you are
> lucky to have access to that, and within swreg5 there is only g2_tempor_mvp_e
> that is also shared, and its already being set in both drivers.
>
> Reviewed-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
>
>
> > >
> > >     hantro_write_addr(vpu, G2_TILE_SIZES_ADDR, ctx->hevc_dec.tile_sizes.dma);
> >
>
>

Re: [PATCH] media: verisilicon: HEVC: Initialize start_bit field

[Benjamin Gaignard]

Le 20/01/2025 à 15:15, Adam Ford a écrit :
> On Mon, Jan 20, 2025 at 8:10 AM Nicolas Dufresne
> <nicolas.dufresne@collabora.com> wrote:
>> Le lundi 20 janvier 2025 à 11:00 +0100, Benjamin Gaignard a écrit :
>>> Le 20/01/2025 à 09:10, Benjamin Gaignard a écrit :
>>>> Always set start_bit field to 0, if not it could lead to corrupted frames
>>>> specially when decoding VP9 bitstreams at the same time since VP9 driver
>>>> set it for it own purpose.
>>                 its
>>
> Does this impact the Fluster score?

No because that happens only when decoding VP9 and HEVC bitstreams at the same time.

>
>>>> Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
>>> I could add this tag:
>>>
>>> Fixes: cb5dd5a0fa518 ("media: hantro: Introduce G2/HEVC decoder")
>> I have tested this on IMX8MQ board using the following GStreamer pipeline.
>> Before the change, the HEVC window was entirely corrupted. The streams don't
>> matter as long as they use both HEVC and VP9 codec.
>>
>> gst-launch-1.0 \
>>    filesrc location=hevc.mp4 ! parsebin ! v4l2slh265dec ! fakevideosink \
>>    filesrc location=vp9.mkv ! parsebin ! v4l2slvp9dec ! fakevideosink
>>
>> Tested-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
>>
>>
>>>> ---
>>>>    drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c | 1 +
>>>>    1 file changed, 1 insertion(+)
>>>>
>>>> diff --git a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
>>>> index 85a44143b378..0e212198dd65 100644
>>>> --- a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
>>>> +++ b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
>>>> @@ -518,6 +518,7 @@ static void set_buffers(struct hantro_ctx *ctx)
>>>>      hantro_reg_write(vpu, &g2_stream_len, src_len);
>>>>      hantro_reg_write(vpu, &g2_strm_buffer_len, src_buf_len);
>>>>      hantro_reg_write(vpu, &g2_strm_start_offset, 0);
>>>> +   hantro_reg_write(vpu, &g2_start_bit, 0);
>>>>      hantro_reg_write(vpu, &g2_write_mvs_e, 1);
>> I've also crossed against "decoder_swreg_map_g2.xlsx" documentation, if you are
>> lucky to have access to that, and within swreg5 there is only g2_tempor_mvp_e
>> that is also shared, and its already being set in both drivers.
>>
>> Reviewed-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
>>
>>
>>>>      hantro_write_addr(vpu, G2_TILE_SIZES_ADDR, ctx->hevc_dec.tile_sizes.dma);
>>

Re: [PATCH] media: verisilicon: HEVC: Initialize start_bit field

[Nicolas Dufresne]

Le lundi 20 janvier 2025 à 08:15 -0600, Adam Ford a écrit :
> On Mon, Jan 20, 2025 at 8:10 AM Nicolas Dufresne
> <nicolas.dufresne@collabora.com> wrote:
> > 
> > Le lundi 20 janvier 2025 à 11:00 +0100, Benjamin Gaignard a écrit :
> > > Le 20/01/2025 à 09:10, Benjamin Gaignard a écrit :
> > > > Always set start_bit field to 0, if not it could lead to corrupted frames
> > > > specially when decoding VP9 bitstreams at the same time since VP9 driver
> > > > set it for it own purpose.
> > 
> >                its
> > 
> 
> Does this impact the Fluster score?

Only if you concurrently test VP9 and HEVC at the same time. Fluster does not
parallelized across codecs, could be a nice addition to catch more issues.

Nicolas

> 
> > > > 
> > > > Signed-off-by: Benjamin Gaignard <benjamin.gaignard@collabora.com>
> > > 
> > > I could add this tag:
> > > 
> > > Fixes: cb5dd5a0fa518 ("media: hantro: Introduce G2/HEVC decoder")
> > 
> > I have tested this on IMX8MQ board using the following GStreamer pipeline.
> > Before the change, the HEVC window was entirely corrupted. The streams don't
> > matter as long as they use both HEVC and VP9 codec.
> > 
> > gst-launch-1.0 \
> >   filesrc location=hevc.mp4 ! parsebin ! v4l2slh265dec ! fakevideosink \
> >   filesrc location=vp9.mkv ! parsebin ! v4l2slvp9dec ! fakevideosink
> > 
> > Tested-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
> > 
> > 
> > > 
> > > > ---
> > > >   drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c | 1 +
> > > >   1 file changed, 1 insertion(+)
> > > > 
> > > > diff --git a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> > > > index 85a44143b378..0e212198dd65 100644
> > > > --- a/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> > > > +++ b/drivers/media/platform/verisilicon/hantro_g2_hevc_dec.c
> > > > @@ -518,6 +518,7 @@ static void set_buffers(struct hantro_ctx *ctx)
> > > >     hantro_reg_write(vpu, &g2_stream_len, src_len);
> > > >     hantro_reg_write(vpu, &g2_strm_buffer_len, src_buf_len);
> > > >     hantro_reg_write(vpu, &g2_strm_start_offset, 0);
> > > > +   hantro_reg_write(vpu, &g2_start_bit, 0);
> > > >     hantro_reg_write(vpu, &g2_write_mvs_e, 1);
> > 
> > I've also crossed against "decoder_swreg_map_g2.xlsx" documentation, if you are
> > lucky to have access to that, and within swreg5 there is only g2_tempor_mvp_e
> > that is also shared, and its already being set in both drivers.
> > 
> > Reviewed-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
> > 
> > 
> > > > 
> > > >     hantro_write_addr(vpu, G2_TILE_SIZES_ADDR, ctx->hevc_dec.tile_sizes.dma);
> > > 
> > 
> >