Performance of memcpy (+memset)

[Bas van den Berg]

When testing memcpy, it seems to handle about 4 MB/sec, running on the 60 MHz app-core.

This seems to be way too slow. Creating our own version of memcpy() that just does a simple

byte-for-byte copy, this also has the same performance, so my guess is that the default implemention

is similar. What kind of performance could be expected? 32 MB/sec? The RAM we use is the first

RAM bank (so fastest for app-core)

[Turbo J]

The memcpy() implementation depends on the LibC used. Newlib-nano uses a simple (size optimized) one byte at a time loop.

You can optimize the speed by about 4x using word (32-bit) RAM access, but this also requires source, destination addresses and buffer size to be dividable by 4.