Hello,
The executing time of the universal_copy process is ~150us for a single frame of 1024bytes (measured from hlos side). However the total time that includes 8100 frames of file_read->memcpy->file_write reaches ~1764ms as it shown:
[t=0x000015fa] [tid=0x4003d000] xdc.runtime.Main: [+1] App-> Application starte.
[t=0x0000edd7] [tid=0x4003d000] xdc.runtime.Main: [+1] Alg version: 1.00.00.00
[t=0x001bda55] [tid=0x4003d000] xdc.runtime.Main: [+1] 8100 frames processed
[t=0x001bdac7] [tid=0x4003d000] xdc.runtime.Main: T2-T1: 1764
[t=0x00270657] [tid=0x4003d000] xdc.runtime.Main: [+1] app done.
Obviously it does not concern a real-time application and the question is:
It is possible to define a monolithic buffer to avoid multiple buffer transfers between ARM<->DSP?
Or there is an alternative way to speed up the execution time?
- dm816x
- EZSDK 5.04.00.11
Regards,
gaston