Hi all:
I do alg optimize on dm8168 use dvrrdk4.00.
And i want to do gauss 4x4 pyramid,
gaussDownCoef[16] = { 6, 8, 8, 6,
8, 10, 10, 8,
8, 10, 10, 8,
6, 8, 8, 6
};
and the downsample image is to do sobel, use IMG_sobel_asm in imglib from ti,
i used ping-pong buffer dma and internal mem,
and i found a strange thing is that when i split gauss downsample and sobel in two fun and do alg optimize by each fun,
the time is less than i do in one fun, which is IMG_sobel_asm is done after gauss downsample,
i think in one fun ping-pong and dam transfer will less, so the time delay will be less,
but the i print the time delay, in one fun it cost more time,
why?
best regards
why
