[FFmpeg-devel] [PATCH 1/2] hwcontext_vulkan: dynamically load functions

[Lynne]

Jun 10, 2021, 12:27 by dev at lynne.ee:

> Jun 10, 2021, 03:38 by wenbin.chen at intel.com:
>
>>> Jun 8, 2021, 07:38 by wenbin.chen at intel.com:
>>>
>>> >> Apr 29, 2021, 03:52 by dev at lynne.ee:
>>> >>
>>> >> > This patch allows for alternative loader implementations.
>>> >> >
>>> >> > Patch attached.
>>> >> >
>>> >>
>>> >> Forgot to fix a flag, v2 attached.
>>> >>
>>> >
>>> > Hi Lynne:
>>> > I tried the following command:
>>> > "ffmpeg -init_hw_device vulkan=vul:0 -filter_hw_device vul -i 1080p.264 -vf
>>> "hwupload=extra_hw_frames=16,scale_vulkan=w=1920:h1080" -f null"
>>> > It report a segmentation fault at
>>> https://github.com/FFmpeg/FFmpeg/blob/282b9f4cba7ca361c43ac9f94031a
>>> 43109df0a4f/libavutil/hwcontext_vulkan.c#L1018
>>> > The function pointer vk->CreateCommandPool is NULL.
>>> > I build on ffmpeg master.
>>> >
>>>
>>> Weird. What Vulkan drivers and what OS was used?
>>> If CreateCommandPool is NULL then initialization probably
>>> wouldn't even happen.
>>>
>>
>> I use linux ubuntu 20.04 and I get the vulkan driver from: https://www.lunarg.com/vulkan-sdk/.
>> I see the load_functions() is called three times, but none of them set "has_dev" argument to 1.
>> I tried to change the code in https://github.com/FFmpeg/FFmpeg/blob/591b88e6787c4e678237f02a50421d101abd25c2/libavutil/hwcontext_vulkan.c#L1352  to "load_functions(ctx, 1, 1)". The problem is unseen.
>>
>
> Thanks for testing, fix pushed.
>

Out of curiosity, are you planning on contributing to the Vulkan code?