Flemming Sondergaard wrote in a message to Tom Torfs:
 TT> AFAIK, REP MOVSW is still the fastest way to move blocks from memory to
 TT> memory (even faster than DMA). REP MOVSD is still faster, of course
 TT> (provided you have a 32-bit databus).
 FS> I once compared REP MOVSD to this method:
 FS> loop_000:
 FS>          mov       eax,ds:[si]
 FS>          add       si,4
 FS>          mov       es:[di],eax
 FS>          add       di,4
 FS>          dec       cx
 FS>          jnz       loop_000
 FS> And REP MOVSD _was_ the slowest of the two.
Sounds unlikely. Even by simply adding the separate instruction clock cycles 
you can calculate that the long method has to be a lot slower, and that's 
without taking into account the fact that the CPU has to fetch many more 
instructions. You were moving 4-byte-aligned addresses weren't you ?
greetz,
Tom
tomtorfs@mail.dma.be
--- timEd/2 1.10+
---------------