On 2024/06/25 22:27, Paul B Mahol wrote:
On Tue, Jun 25, 2024 at 3:17 PM Dennis Mungai <dmngaie@gmail.com> wrote:
On Tue, 25 Jun 2024 at 16:07, Paul B Mahol <onemda@gmail.com> wrote:
On Tue, Jun 25, 2024 at 2:47 PM Dennis Mungai <dmngaie@gmail.com> wrote:
On Sun, 28 Jun 2020 at 16:59, James Almer <jamrial@gmail.com> wrote:
On 6/27/2020 7:54 AM, Paul B Mahol wrote:
On 6/26/20, James Almer <jamrial@gmail.com> wrote: > On 6/26/2020 4:16 PM, Lynne wrote: >> Jun 26, 2020, 19:05 by jamrial@gmail.com: >> >>> On 6/26/2020 1:58 PM, Paul B Mahol wrote: >>> >>>> On 6/26/20, James Almer <jamrial@gmail.com> wrote: >>>> >>>>> On 6/26/2020 1:13 PM, Paul B Mahol wrote: >>>>> >>>>>> On 6/26/20, Paul B Mahol <onemda@gmail.com> wrote: >>>>>> >>>>>>> On 6/24/20, Paul B Mahol <onemda@gmail.com> wrote: >>>>>>> >>>>>>>> Signed-off-by: Paul B Mahol <onemda@gmail.com> >>>>>>>> --- >>>>>>>> Please test it and reports benchmark results on non toyish GPUs. >>>>>>>> >>>>>>> >>>>>>> Please, I really need it! >>>>>>> >>>>>> >>>>>> >>>>>> Why nobody wants to help? >>>>>> >>>>>> This makes me really really sad. >>>>>> >>>>> >>>>> I can try. Can you give an example command line with a publicly >>>>> available sample? >>>>> >>>> >>>> You can use any sample, samples are irrelevant. just report resolution >>>> used. >>>> Also compare with CPU solution with just v360 filter. >>>> >>>> For yuv420p inputs: >>>> >>>> ffmpeg -init_hw_device vulkan=vulkan -i INPUT -filter_hw_device vulkan >>>> -vf hwupload,v360_vulkan=fisheye:e:ih_fov=180:iv_fov=180:w=2048:h=1024,hwdownload,format=yuv420p -f null - >>>> >>>> ffmpeg -i INPUT -vf v360=fisheye:e:ih_fov=180:iv_fov=180:w=2048:h=1024 -f null - >>>> >>>> Just make sure that you get same output resolution in both cases. If >>>> not, adjust w/h accordingly. >>>> >>> >>> I'm getting a segfault on vkGetInstanceProcAddr, and i don't know if >>> it's your filter or some issue in the driver. gdb is not being helpful. >>> _______________________________________________ >>> ffmpeg-devel mailing list >>> ffmpeg-devel@ffmpeg.org >>> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel >>> >>> To unsubscribe, visit link above, or email >>> ffmpeg-devel-request@ffmpeg.org> with subject "unsubscribe". >>> >> >> Can you test again but replace EXT_EXTERNAL_HOST_MEMORY in >> optional_device_exts >> with EXT_NO_FLAG? > > Applying > >> diff --git a/libavutil/hwcontext_vulkan.c b/libavutil/hwcontext_vulkan.c >> index 5e51d0390f..49fb7827ba 100644 >> --- a/libavutil/hwcontext_vulkan.c >> +++ b/libavutil/hwcontext_vulkan.c >> @@ -228,7 +228,7 @@ static const VulkanOptExtension optional_device_exts[] >> = { >> { VK_EXT_EXTERNAL_MEMORY_DMA_BUF_EXTENSION_NAME, >> EXT_EXTERNAL_DMABUF_MEMORY, }, >> { VK_EXT_IMAGE_DRM_FORMAT_MODIFIER_EXTENSION_NAME, >> EXT_DRM_MODIFIER_FLAGS, }, >> { VK_KHR_EXTERNAL_SEMAPHORE_FD_EXTENSION_NAME, >> EXT_EXTERNAL_FD_SEM, }, >> - { VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME, >> EXT_EXTERNAL_HOST_MEMORY, }, >> + { VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME, >> EXT_NO_FLAG, }, >> }; >> >> /* Converts return values to strings */ > > I still get a segfault: > >> Thread 1 received signal SIGSEGV, Segmentation fault. >> 0x00007ff8ed9274c0 in vkGetInstanceProcAddr () >> from >>
C:\Windows\System32\DriverStore\FileRepository\nv_dispi.inf_amd64_ccad5caddc3a3d35\nvoglv64.dll
> > Debug output shows > >> [AVHWDeviceContext @ 000001532a762800] GPU listing: >> [AVHWDeviceContext @ 000001532a762800] 0: GeForce GTX 1660 (discrete) >> (0x2184) >> [AVHWDeviceContext @ 000001532a762800] Using queue family 0 (total queues: >> 16) for graphics >> [AVHWDeviceContext @ 000001532a762800] QF 0 flags: (graphics) >> (compute) (transfers) (sparse) >> [AVHWDeviceContext @ 000001532a762800] Using queue family 2 (total queues: >> 8) for compute >> [AVHWDeviceContext @ 000001532a762800] QF 2 flags: (compute) >> (transfers) (sparse) >> [AVHWDeviceContext @ 000001532a762800] Using queue family 1 (total queues: >> 2) for transfers >> [AVHWDeviceContext @ 000001532a762800] QF 1 flags: (transfers) >> (sparse) >> [AVHWDeviceContext @ 000001532a762800] Using device extension >> "VK_EXT_external_memory_host" >> [AVHWDeviceContext @ 000001532a762800] Using device extension >> VK_EXT_external_memory_host >> [AVHWDeviceContext @ 000001532a762800] Using device: GeForce GTX 1660 >> [AVHWDeviceContext @ 000001532a762800] Alignments: >> [AVHWDeviceContext @ 000001532a762800] >> optimalBufferCopyRowPitchAlignment: 1 >> [AVHWDeviceContext @ 000001532a762800] minMemoryMapAlignment: 64 > > And crashes right after printing: > >> [Parsed_v360_vulkan_1 @ 000001532abd5c80] Shader v360_compute linked! >> Size: 5508 bytes > > I'm using the latest driver, 451.48, on Windows 10 2004.
So other vulkan filters, like vulkan_chromaber works for you?
All fail the same way in vkGetInstanceProcAddr, so the issue is in hwcontext.
Lynne should be able to test/bench this filter.
Hello Paul, et al,
Any updates on this filter?
I'm no longer working on FFmpeg.
The v360_opencl filter got in more useful state but due complete lack of interest in it from users, I forgot about its existence. The v360_vulkan filter is probably in worse shape than v360_opencl filter and 0/negative interest, and I forgot about its existence too.
Is there an FFmpeg tree with v360_opencl (or a similar filter) out there?
I can give you patches/files for both v360 vulkan and opencl filter if you want to hack on it. But I will not adapt them for free to current library versions, nor provide any kind of free support for it.
I tested v360_opencl filter with slight modification to your patch. It works, but it generates different output from v360 filter. For example, I tested a sample GoPro Max 360 video on: https://drive.google.com/drive/folders/1DlP6FVw-O6AIm6O0nSsC-mIoCrEBtmdG ffmpeg -i GS010010.360 -init_hw_device opencl:0.0 -vf "hwupload,v360_opencl=yaw=90,hwdownload,format=yuvj420p" output-v360cl.mkv ffmpeg -i GS010010.360 -vf v360=yaw=90 output-v360.mkv I'm not sure which is correct, but I feel the output of v360_opencl filter is better.
A GPU-accelerated v360 filter is definitely useful to test and implement. Definitely more useful than the DNN stack in FFmpeg (by leaps and bounds) as far as applicability is concerned. _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
_______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".