memcpy

[N1ghtm4r3]

C++ memcpy function...

Example:

Code Select Expand

include \masm32\include\masm32rt.inc

include _memcpy.asm

Main PROTO

.data
szSrc db 20h  dup(1)
szDst db 20h  dup(0)

.code

Start:
  Invoke Main
  Invoke ExitProcess,0

Main proc

push 10
push offset szSrc
push offset szDst
call _memcpy

ret
Main endp

end Start

[dedndave]

 :bg

Code Select Expand

        mov     esi,offset szSrc
        mov     edi,offset szDst
        mov     ecx,sizeof szSrc
        rep     movsb

[N1ghtm4r3]

sure!  :bg
but the code I posted might be useful in some especial situations. as it was for me today  :wink

[brethren]

you can use c++ functions.................or you can just roll your own

Code Select Expand

.code
MemCopy PROC USES esi edi, Source:DWORD, Dest:DWORD, ln:DWORD

cld
mov esi, Source
mov edi, Dest
mov ecx, ln
shr ecx, 2
rep movsd

mov ecx, ln
and ecx, 3
rep movsb

ret
MemCopy ENDP
END

[N1ghtm4r3]

About using C++ functions you are right, but as I said it's not just matter of copying memory! I had to use exactly this code for a special case and I just shared it here maybe someday, someone find it useful. that's all ;)

[qWord]

if speed is need, the CRT (dynamical linked)  is probably the best choice. Using this, you application will also take speed gain of future instruction sets.

[hutch--]

I have a reasonably simple view on memory copy, if its a terrabyte it matter, if its a megabyte it doesn't. If you have the registers to spare a simple REP MOVSB does the job in most instances.

Code Select Expand


mov esi, src
mov edi, dst
mov ecx, cnt
rep movsb


[dedndave]

that's right
the real problem with trying to speed it up by using REP MOVSD is not dealing with the mod 3 count at the end
it is the fact that strings are not usually both dword aligned - that negates the speed advantage
you have a 1 in 16 chance of going really fast   :bg

[jj2007]

that's right
the real problem with trying to speed it up by using REP MOVSD is not dealing with the mod 3 count at the end
it is the fact that strings are not usually both dword aligned - that negates the speed advantage
you have a 1 in 16 chance of going really fast   :bg

Dave,
In real life big buffers are 8- or 16-bit aligned - and then rep movsd is almost unbeatable. Remember the Code location sensitivity of timings thread...? Look for MemCo1.

[drizz]

It looks exactly like if you have disassembled ms crt library memcpy function with IDA.
And invoked it as stdcall while the function is clearly cdecl.

[N1ghtm4r3]

Yes, that's what I did :p