Hi,
My visionSDK version is 3.08.
My goal is realize 4 video in 1 video function. so I use VPE scale it then use edma copy and combine it from IPU2 to A15 with "Utils_dmaCopy2D". Everything work fine and Utils_dmaCopy2D execute time only 10ms
But VPE comsue 200MB/s DDR bandwidth, so I remove it and change Utils_dmaCopy2D to Utils_dmaCopy3D, which add cCnt and SRCCIDX to skip every two lines. Then Utils_dmaCopy3D API execute time raise to 70ms.
The total copy data size is same, but Utils_dmaCopy3D copy every line at once, don't like Utils_dmaCopy2D copy entire picture. So I think it's copy will be blocked(most is GPU) when DDR bandwidth is high.
So do you have any good suggestion about this problem? I'm thinking about use VIP to scale and combine it, but since it already occupied by camera orignal stream, I think is can not add another pipeline on same VIP.
Thanks!
