I created an IPU task using IPU API as below:
task.input.width = 1920;
task.input.height = 1080;
task.input.format = v4l2_fourcc('U', 'Y', 'V', 'Y');
// Overlay image size and format
task.overlay.width = 1920;
task.overlay.height = 1080;
task.overlay.format = v4l2_fourcc('R', 'G', 'B', 'A');
task.overlay_en = 1; // ENABLE COMPOSITION
task.overlay.alpha.mode = 1;
task.overlay.alpha.gvalue = 0;
task.overlay.colorkey.enable = 0;
task.overlay.colorkey.value = 0;
// Output image size and format
task.output.width = 1920;
task.output.height = 1080;
task.output.format = v4l2_fourcc('I', '4', '2', '0');
But the output image seems wrong. (Describing it may help: shapes are distinguishable but it is distorted and pink). Is there a limitation about color-space conversion and composition in IPU or/and IPU API?
I have tried performing same task with other formats (for example UYVY-RGBA->RGBA) and there was no problem. Also I have tried to split it into two tasks, again the result was good but total operation time is almost doubled.
As a side question, Where can I find information about the standard operation times for IPU? I couldn't find it on IPU documents.
Thanks in advance.