[Libav-user] getting CUdeviceptr formats and data

[Mathieu Prevot]

Le mar. 26 nov. 2024 à 19:26, Mathieu Prevot <mathieu.prevot at gmail.com> a
écrit :

>
>
> Le ven. 22 nov. 2024 à 13:55, Mathieu Prevot <mathieu.prevot at gmail.com> a
> écrit :
>
>>
>> Le mer. 20 nov. 2024 à 18:03, Mathieu Prevot <mathieu.prevot at gmail.com>
>> a écrit :
>>
>>>
>>>
>>> Le mar. 19 nov. 2024 à 00:39, Mathieu Prevot <mathieu.prevot at gmail.com>
>>> a écrit :
>>>
>>>>
>>>>
>>>> Le dim. 17 nov. 2024 à 20:57, Mathieu Prevot <mathieu.prevot at gmail.com>
>>>> a écrit :
>>>>
>>>>> Hello,
>>>>>
>>>>> I am using code, based on
>>>>> https://github.com/FFmpeg/FFmpeg/blob/master/doc/examples/hw_decode.c,
>>>>> but I am not transferring the decoded data to the host, I keep it on the
>>>>> device for later use with cuda code (my goal will be to use RGB 8-12 bits
>>>>> per channel data).
>>>>>
>>>>> I could not find a way to get the data format
>>>>> ((AVPixelFormat)frame->format translates to: "HW acceleration through CUDA.
>>>>> data[i] contain CUdeviceptr pointers exactly as for system memory frames.")
>>>>>
>>>>> I understand that I have two arrays: frame->data[0] and frame->data[1].
>>>>>
>>>>> I would like to translate the data into a cuda native structure like:
>>>>>
>>>>> cudaPitchedPtr CUdeviceptr{};
>>>>> CUdeviceptr.ptr = frame->data[0];
>>>>> CUdeviceptr.ysize = frame->height;
>>>>> CUdeviceptr.xsize = frame->width;
>>>>>
>>>>> Which is probably wrong, since width==1920 height==1080 and should
>>>>> have colors or YUV data.
>>>>>
>>>>> How to get to know the data format (RGB, BGR, YUV, 8/10/12bits per
>>>>> channel etc) and the size of the array ?
>>>>> If it is not RGB8 (or 10-12), can I get that through some ffmpeg
>>>>> implementation, maybe with an nvdec based function ?
>>>>> What is the correct cuda-native data ? cudaPitchedPtr or something
>>>>> else ?
>>>>> frame->data[0] is uint8_t, but the array is probably something else in
>>>>> the general case; so what is it (as redundancy or extension of "what is the
>>>>> data format") ?
>>>>>
>>>>
>>>> It seems that  frame->data[0] is the Y component of YUV, if not, one of
>>>> RGB channels, but I could not figure out what frame->data[1] is.
>>>> Can someone confirm ?
>>>>
>>>
>>> I got no answer from libav-user@, maybe is it a question for
>>> ffmpeg-devel@ (not subscribed, please cc) ?
>>>
>>
>> There seems to be no other data except that of `frame->data[0]`.
>>
>> I verified, there is nothing beyond `frame->data[0] +
>> frame->linesize[0]*frame->height`.
>>
>> frame->data[1] has another address but there is no data.
>>
>> frame->buf[0]->data and frame->extended_data  both point to the same
>> address as frame->data[0].
>>
>> no side_data or extended_buf,
>>
>> opaque_ref is NULL
>>
>> frame->ch_layout->order is AV_CHANNEL_ORDER_UNSPEC
>>
>> In the end, it comes down to this questions:
>>
>> 1. where are UV data of YUV when decoded by cuda ? Y is at frame->data[0].
>> 2. frame->linesize[1] is set too, what is this for ?
>>
>
For those interested.

<https://stackoverflow.com/posts/79242422/timeline>

For a given frame of a video that can be decoded with nvdec in ffmpeg,

(AVPixelFormat)frame->format will give:

AV_PIX_FMT_CUDA: HW acceleration through CUDA. data[i] contain CUdeviceptr
pointers exactly as for system memory frames.

which is not helpful, but
((AVHWFramesContext*)frame->hw_frames_ctx->data)->sw_format will give:

AV_PIX_FMT_NV12: planar YUV 4:2:0, 12bpp, 1 plane for Y and 1 plane for the
UV components, which are interleaved (first byte U and the following byte V)

which is everything that is needed.

Both are pitched arrays, which can be used directly as cudaPitchedPtr.ptr.
Their pitch is frame->linesize[0] and frame->linesize[1].
M
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://ffmpeg.org/pipermail/libav-user/attachments/20241201/f6dcd0cd/attachment.htm>