ffmpeg's v360 hardware filters
Hey, I've asked Paul about the `v360` hardware-accelerated FFmpeg filters, thread below. He referred me to Lynne @ #ffmpeg-devel on Libera, to ask about the possible Vulkan implementation. I'll get in touch soon and update ============ Forwarded message ============ From: Paul B Mahol < mailto:onemda@gmail.com > To: "Milos"< mailto:milos@whitebox.aero > Cc: "xavier"< mailto:xavier@whitebox.aero > Date: Sat, 24 Jan 2026 13:51:52 +0100 Subject: Re: ffmpeg's v360 hardware filters ============ Forwarded message ============ Hi, I have had only half-finished but working OpenCL filter version of v360 for ffmpeg/libavfilter. I need to recheck if I still have source code of vf_v360_opencl.c and related files/changes still somewhere. The Vulkan version never was really functional like OpenCL version is, and if you want Vulkan version better contact Vulkan expert(s) at FFmpeg. At least over IRC channel, #ffmpeg-devel on libera IRC servers, by IRC nick name Lynne. I'm not working on Vulkan stack so I do not have all necessary skills for very optimal filter implementation, moved to do other stuff instead. The old Vulkan WIP patch was here, which was then tested on old integrated Intel GPU: https://ffmpeg.org/pipermail/ffmpeg-devel/2020-June/265057.html But Vulkan API is changing and FFmpeg related support code too, so it needs more work, and more other transforms needs to be implemented for other projections in v360 vulkan filter... Regards On Sat, Jan 24, 2026 at 12:07 PM Milos < mailto:milos@whitebox.aero > wrote: Hello Paul, I'm working on a project utilizing the ffmpeg's v360 filter and we're looking into ways to accelerate it. We found your messages on ffmpeg-devel's mailing list: https://ffmpeg.org/pipermail/ffmpeg-devel/2024-June/330229.html , and I was wondering if you'd be willing to share the patches for us to try them out. You can check the project out here: https://docs.whitebox.aero If that's alright, we'd like to try them out, and, considering your ffmpeg expertise, potentially contract you for some help getting it running / merged upstream if you'd be willing to help. Looking forward to hearing from you. Thanks! Milos
Hey Paul, Hope you're doing well. Just wanted to ask if you had time to check if you still have the source code of `vf_v360_opencl.c`. Thanks!
No attempt was made to compile/test it with recent libavfilter. It does not have EAC or cubemap support, among many other stuff. On todays GPUs with GBs of GPU RAM whole remap data could be stored in memory like what CPU v360 filter does. Then speed could be even faster?! or not... On Fri, Feb 13, 2026 at 12:45 AM <milos@whitebox.aero> wrote:
Hey Paul,
Hope you're doing well. Just wanted to ask if you had time to check if you still have the source code of `vf_v360_opencl.c`.
Thanks!
Hey Paul, To provide some more context, we're trying to optimize one specific use case for now, converting dual-fisheye to equirectangular format in real-time. We're using it like this: v360=dfisheye:e:ih_fov=193:iv_fov=193. As our project is running on a battery powered SBC (Orange Pi 5 Plus, running RK3588), we're quite constrained on resources. We've already optimized most of our FFmpeg pipeline, but the v360 filter is the biggest remaining resource hog. Currently, we're able to run a single pipeline in real-time for a single video source, but in the future we'd like to extend that further to multiple real-time video sources, for which the CPU-based filter is going to be too heavy for on our environment. If I understand it right, I think that neither EAC nor cubemap are used here, but I don't have experience with FFmpeg internals so I might be wrong. Is this use case something that the code you mentioned has previously supported? I understand that the code would need more work to support current libavfilter, so we're wondering if you'd be open to sponsorship to help us bring v360_opencl to support our use case? Thanks! On 14. 2. 2026. 12:46, Paul B Mahol wrote:
No attempt was made to compile/test it with recent libavfilter.
It does not have EAC or cubemap support, among many other stuff. On todays GPUs with GBs of GPU RAM whole remap data could be stored in memory like what CPU v360 filter does. Then speed could be even faster?! or not...
On Fri, Feb 13, 2026 at 12:45 AM <milos@whitebox.aero> wrote:
Hey Paul,
Hope you're doing well. Just wanted to ask if you had time to check if you still have the source code of `vf_v360_opencl.c`.
Thanks!
participants (3)
-
Milos -
milos＠whitebox.aero
-
Paul B Mahol