GStreamer Based Image Signal Processor/Performance/imx8
The following section describes some relevant performance measurements for the RGBA to NV12 color space conversion for a few standard resolutions. It begins with a summary of the results detailed on the following sections, as well as the pipelines used to capture such results.
Benchmarking
Pipeline
All the measurements below were made using the following minimal pipeline.
$ WIDTH=1920 $ HEIGHT=1080 $ gst-launch-1.0 videotestsrc is-live=true ! "video/x-raw,format=(string)RGBA,width=$WIDTH,height=$HEIGHT,framerate=(fraction)30/1" ! queue ! oclcsc ! fakesink -v
Summary
In this summary you can find measurements related with CPU usage, memory, execution time and maximum framerate supported by the CSC algorithm.
Resolution | CPU usage (%) * | Memory (kB) * | Execution Time (ms) |
---|---|---|---|
640x480@30fps | 4 | 18664 | 2.24 |
1280x720@30fps | 10 | 19484 | 6.19 |
1920x1080@30fps | 14 | 21424 | 13.17 |
Note: take into consideration that these values were taken using the videotestsrc element, and may decrease with a camera source capture.
Resolution | Maximum framerate (fps) |
---|---|
640x480 | 333 |
1280x720 | 120 |
1920x1080 | 55 |
CPU usage
These measurements were taken with top app to monitor the behavior of the pipeline while running. Notice, however, that videotestsrc was used to generate the buffers and it consumes a lot of CPU, so it is likely the results will differ if a camera is used for capture.
Memory usage
These measurements were taken with meminfo file to see the behavior of the pipeline while running, but it is important to consider that videotestsrc was used to generate the buffers.
Maximum Framerate
These measurements were taken with the gst-perf tool to monitor the behavior of the pipeline while running. When using videotestsrc we are able to process as many buffers as the element can handle, hence the maximum value reported. The image below shows the maximum framerate that the CSC element supports, which depends on the execution time of the CSC kernel algorithm.
Execution time
These measurements were taken using some timestamps before and after the processing to measure just the execution time for the CSC kernel algorithm on each frame and hence avoiding possible false measurements from the GStreamer plugin. The image below shows the CSC execution time by each frame for each resolution.