03-12-2021, 06:58 PM
I've fixed some bugs in my test script that now show vainfo correctly. The updated script is vputests.sh, as in this example.
|----------------------------------------------------
++++ /dev/video0:
|----------------------------------------------------
LIBVA_DRIVER_NAME=v4l2_request vainfo --display drm --device /dev/video0
libva info: VA-API version 1.11.0
libva info: User environment variable requested driver 'v4l2_request'
libva info: Trying to open /usr/lib/dri/v4l2_request_drv_video.so
libva info: Found init function __vaDriverInit_1_11
libva info: va_openDriver() returns 0
vainfo: VA-API version: 1.11 (libva 2.11.0.pre1)
vainfo: Driver version: v4l2-request
vainfo: Supported profile and entrypoints
VAProfileH264Main : VAEntrypointVLD
VAProfileH264High : VAEntrypointVLD
VAProfileH264ConstrainedBaseline: VAEntrypointVLD
VAProfileH264MultiviewHigh : VAEntrypointVLD
VAProfileH264StereoHigh : VAEntrypointVLD
VAProfileHEVCMain : VAEntrypointVLD
I also ported the kernel patch to 5.10.17. Before and after the patch I have only seen what appears to be HW accelleration with mpeg2video files. H264 doesn't not appear to be getting accelerated based on CPU uage. The patch has been posted to the pbp site.
I've also created my own sample files using OpenShot to save in H264, mpeg2video and mpeg4 formats (I'm not trying VP9 because its not one of our use cases) as 720p 25fps 1280x720 videos in MP4 and AVI containers. The videos are longer than most of the samples I've been using to give me a chance to watch CPU usage over time. That's how I'm testing if HW accel is apparently working. With mpeg2video I'm seeing about 7% CPU usage. With the others it varies from about 50% to 80% cpu usage.
I tried kwiboo's ffmpeg but there was no difference in the tests.
At this point I'm convinced that only mpeg2video are HW accelerated with the upstreamed hantro/rkvdec drivers plus the patches available here to both the kernel and user space utilities for VAAPI support.
|----------------------------------------------------
++++ /dev/video0:
|----------------------------------------------------
LIBVA_DRIVER_NAME=v4l2_request vainfo --display drm --device /dev/video0
libva info: VA-API version 1.11.0
libva info: User environment variable requested driver 'v4l2_request'
libva info: Trying to open /usr/lib/dri/v4l2_request_drv_video.so
libva info: Found init function __vaDriverInit_1_11
libva info: va_openDriver() returns 0
vainfo: VA-API version: 1.11 (libva 2.11.0.pre1)
vainfo: Driver version: v4l2-request
vainfo: Supported profile and entrypoints
VAProfileH264Main : VAEntrypointVLD
VAProfileH264High : VAEntrypointVLD
VAProfileH264ConstrainedBaseline: VAEntrypointVLD
VAProfileH264MultiviewHigh : VAEntrypointVLD
VAProfileH264StereoHigh : VAEntrypointVLD
VAProfileHEVCMain : VAEntrypointVLD
I also ported the kernel patch to 5.10.17. Before and after the patch I have only seen what appears to be HW accelleration with mpeg2video files. H264 doesn't not appear to be getting accelerated based on CPU uage. The patch has been posted to the pbp site.
I've also created my own sample files using OpenShot to save in H264, mpeg2video and mpeg4 formats (I'm not trying VP9 because its not one of our use cases) as 720p 25fps 1280x720 videos in MP4 and AVI containers. The videos are longer than most of the samples I've been using to give me a chance to watch CPU usage over time. That's how I'm testing if HW accel is apparently working. With mpeg2video I'm seeing about 7% CPU usage. With the others it varies from about 50% to 80% cpu usage.
I tried kwiboo's ffmpeg but there was no difference in the tests.
At this point I'm convinced that only mpeg2video are HW accelerated with the upstreamed hantro/rkvdec drivers plus the patches available here to both the kernel and user space utilities for VAAPI support.
Michael J. Hammel
mjhammel@graphics-muse.org
michaelhammel@acm.org
mjhammel@graphics-muse.org
michaelhammel@acm.org