memcpy() function

[matic]

Posted on November 06, 2015 at 19:58

Hi there.

I read an article about copying data using CPU (''for'' loop) and DMA (Mem2Mem).http://www.embedds.com/using-direct-memory-access-dma-in-stm23-projects/Below in the comments of that article is described also an approach, which uses memcpy() function. Interestingly, this is also (much) faster than DMA. I never used memcpy() function, so I would like to ask, how could it be faster than DMA? As long as it is a function, I suppose it occupy CPU? But how it works? Where can I see its definition?Thanks #memcpy

[Tesla DeLorean]

Posted on November 06, 2015 at 21:09

You'd get yourself the library source, or disassemble it.

The memcpy doesn't have any setup overhead, the memory is tightly coupled and the processor has write-buffers. You'd have to dig into the gate-level implementation, but I suspect the DMA state-machine has more states, and has to arbitrate/request access from the core. One of the typical memcpy() optimizations is reading multiple words, and then writing multiple words. DMA would have a single holding register

[matic]

Posted on November 06, 2015 at 22:32

Do you have maybe an example code to see how it is used?