The optimized routine

AN2367

4/13

2

The optimized routine

2.1

Interface

void MEMCOPY_L2S_BY4(void *pDEST, void *pSRC, ushort NbrByte)

Note about pointers:

In this description above, the pointers are pointing to a void.

A way of looking at it is that the pointer is pointing to a buffer of long with the first 2 bytes

filled with zeros.

Another way of looking at it is that the pointer is pointing to a buffer of chars.

While it is beyond this application note to explain the problem of mapping bytes, short and

long in the same memory space in C, it is worth noting that the routine could be declared

differently.

In fact, when running the program to test the routine on its own, it was declared as

void MEMCOPY_L2S_BY4(short *pDEST, long *pSRC, long NbrByte)

See Appendix B on page 11 for an example of the C call.

2.2

Algorithm description

The algorithm works as follows:

●

First level - granularity 16: The CPU transfer 16 bytes per iteration until the number of

bytes is less than 16. Go to second level.

●

Second level: Granularity 4: Depending on the remaining number of bytes, the CPU

transfer either 4, 8 or 12 bytes. Go to third level.

●

Third level: Granularity1: Depending on the remaining number of bytes, the CPU

transfer either 1, 2 or 3 bytes.