A 1.3 to 5.2 times faster memcpy, optimizing depends on data blocks alignment on Cortex-M4.
test code: memcpy_fast(dest + a, src + b, TEST_BUFF_SIZE + c) and memcpy(dest + a, src + b, TEST_BUFF_SIZE + c)
dest and src are 8 bytes aligned array variables.
TEST_BUFF_SIZE = 10240
