Message-ID: <HE1PR06MB40112032B360DE217C939FB3AC9A0@HE1PR06MB4011.eurprd06.prod.outlook.com>
Date:   Tue, 8 Oct 2019 13:57:31 +0000
From:   Jonas Karlman <jonas@...boo.se>
To:     Tomasz Figa <tfiga@...omium.org>
CC:     Ezequiel Garcia <ezequiel@...labora.com>,
        "linux-media@...r.kernel.org" <linux-media@...r.kernel.org>,
        "kernel@...labora.com" <kernel@...labora.com>,
        Nicolas Dufresne <nicolas.dufresne@...labora.com>,
        "linux-rockchip@...ts.infradead.org" 
        <linux-rockchip@...ts.infradead.org>,
        Heiko Stuebner <heiko@...ech.de>,
        Philipp Zabel <p.zabel@...gutronix.de>,
        Boris Brezillon <boris.brezillon@...labora.com>,
        Alexandre Courbot <acourbot@...omium.org>,
        "fbuergisser@...omium.org" <fbuergisser@...omium.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        Douglas Anderson <dianders@...omium.org>
Subject: Re: [PATCH v2 for 5.4 3/4] media: hantro: Fix motion vectors usage
 condition

On 2019-10-08 12:26, Tomasz Figa wrote:
> On Tue, Oct 8, 2019 at 3:23 PM Jonas Karlman <jonas@...boo.se> wrote:
>> On 2019-10-08 05:29, Tomasz Figa wrote:
>>> Hi Jonas,
>>>
>>> On Tue, Oct 8, 2019 at 3:33 AM Jonas Karlman <jonas@...boo.se> wrote:
>>>> On 2019-10-07 19:45, Ezequiel Garcia wrote:
>>>>> From: Francois Buergisser <fbuergisser@...omium.org>
>>>>>
>>>>> The setting of the motion vectors usage and the setting of motion
>>>>> vectors address are currently done under different conditions.
>>>>>
>>>>> When decoding pre-recorded videos, this results of leaving the motion
>>>>> vectors address unset, resulting in faulty memory accesses. Fix it
>>>>> by using the same condition everywhere, which matches the profiles
>>>>> that support motion vectors.
>>>> This does not fully match hantro sdk:
>>>>
>>>>   enable direct MV writing and POC tables for high/main streams.
>>>>   enable it also for any "baseline" stream which have main/high tools enabled.
>>>>
>>>>   (sps->profile_idc > 66 && sps->constrained_set0_flag == 0) ||
>>>>   sps->frame_mbs_only_flag != 1 ||
>>>>   sps->chroma_format_idc != 1 ||
>>>>   sps->scaling_matrix_present_flag != 0 ||
>>>>   pps->entropy_coding_mode_flag != 0 ||
>>>>   pps->weighted_pred_flag != 0 ||
>>>>   pps->weighted_bi_pred_idc != 0 ||
>>>>   pps->transform8x8_flag != 0 ||
>>>>   pps->scaling_matrix_present_flag != 0
>>> Thanks for double checking this. I can confirm that it's what Hantro
>>> SDK does indeed.
>>>
>>> However, would a stream with sps->profile_idc <= 66 and those other
>>> conditions met be still a compliant stream?
>> You are correct, if a non-compliant video is having decoding problems it should probably be handled
>> on userspace side (by not reporting baseline profile) and not in kernel.
>> All my video samples that was having the issue fixed in this patch are now decoded correctly.
>>
>>>> Above check is used when DIR_MV_BASE should be written.
>>>> And WRITE_MVS_E is set to nal_ref_idc != 0 when above is true.
>>>>
>>>> I think it may be safer to always set DIR_MV_BASE and keep the existing nal_ref_idc check for WRITE_MVS_E.
>>> That might have a performance penalty or some other side effects,
>>> though. Otherwise Hantro SDK wouldn't have enable it conditionally.
>>>
>>>> (That is what I did in my "media: hantro: H264 fixes and improvements" series, v2 is incoming)
>>>> Or have you found any video that is having issues in such case?
>>> We've been running the code with sps->profile_idc > 66 in production
>>> for 4 years and haven't seen any reports of a stream that wasn't
>>> decoded correctly.
>>>
>>> If we decide to go with a different behavior, I'd suggest thoroughly
>>> verifying the behavior on a big number of streams, including some
>>> performance measurements.
>> I agree, I would however suggest to change the if statement to the following (or similar)
>> as that should be the optimal for performance reasons and match the hantro sdk.
>>
>> if (sps->profile_idc > 66 && dec_param->nal_ref_idc)
> Sorry for my ignorance, but could you elaborate on this? What's the
> meaning of nal_ref_idc? I don't see it being checked in the Hantro SDK
> condition you mentioned earlier.

My somewhat limited understanding of h264 spec is that nal_ref_idc should be 0 for non-reference field/frame/pictures
and because of this there should not be any need to write motion vector data as the field/frame should not be referenced.

My base for the hantro sdk code is the imx8 imx-vpu-hantro package and it uses (simplified):
  SetDecRegister(h264_regs, HWIF_WRITE_MVS_E, nal_ref_idc != 0)
for MVC there is an additional condition.

Regards,
Jonas

>
>> Regards,
>> Jonas
>>
>>> Best regards,
>>> Tomasz
>>>
>>>> Regards,
>>>> Jonas
>>>>
>>>>> Fixes: dea0a82f3d22 ("media: hantro: Add support for H264 decoding on G1")
>>>>> Signed-off-by: Francois Buergisser <fbuergisser@...omium.org>
>>>>> Signed-off-by: Ezequiel Garcia <ezequiel@...labora.com>
>>>>> ---
>>>>> v2:
>>>>> * New patch.
>>>>>
>>>>>  drivers/staging/media/hantro/hantro_g1_h264_dec.c | 2 +-
>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/staging/media/hantro/hantro_g1_h264_dec.c b/drivers/staging/media/hantro/hantro_g1_h264_dec.c
>>>>> index 7ab534936843..c92460407613 100644
>>>>> --- a/drivers/staging/media/hantro/hantro_g1_h264_dec.c
>>>>> +++ b/drivers/staging/media/hantro/hantro_g1_h264_dec.c
>>>>> @@ -35,7 +35,7 @@ static void set_params(struct hantro_ctx *ctx)
>>>>>       if (sps->flags & V4L2_H264_SPS_FLAG_MB_ADAPTIVE_FRAME_FIELD)
>>>>>               reg |= G1_REG_DEC_CTRL0_SEQ_MBAFF_E;
>>>>>       reg |= G1_REG_DEC_CTRL0_PICORD_COUNT_E;
>>>>> -     if (dec_param->nal_ref_idc)
>>>>> +     if (sps->profile_idc > 66)
>>>>>               reg |= G1_REG_DEC_CTRL0_WRITE_MVS_E;
>>>>>
>>>>>       if (!(sps->flags & V4L2_H264_SPS_FLAG_FRAME_MBS_ONLY) &&

Powered by blists - more mailing lists