Answering msg from Paul Edwards to Frank Malcolm,
on Monday January 08 1996 at 23:48

Hi Paul, a couple of optimizations and suggestions.

 PE> public memcpy
 PE> memcpy proc
 PE> push ebp
 PE> mov ebp, esp
 PE> pushf
 PE> push edi
 PE> push esi
 PE> push ecx
 PE> mov edi, [ebp+8]
     mov eax,edi        ; used for return value
 PE> mov esi, [ebp+12]
 PE> mov ecx, [ebp+16]
 PE> cld
 PE> shr ecx, 2
 PE> rep movsd

----
 PE> mov ecx, [ebp+16]
 PE> and ecx, 00000003H
----
; Try replacing the above two lines with this.
; It may or may not be faster, time it and see. (Would be a small
; difference only)

     mov cl,byte ptr [ebp+16]   ; ECX=0 so we can safely manipulate
     and cl,3                   ; the bottom byte only

 PE> rep movsb

 PE> mov eax, [ebp+8]
; remove above line, eax is set above

 PE> pop ecx
 PE> pop esi
 PE> pop edi
 PE> popf
 PE> pop ebp
 PE> ret
 PE> memcpy endp

In certain cases, it can be faster to move memory with
"primitives" rather than the MOVSx commands, such as:

mov eax,[esi]    ; staggered to avoid AGI
mov ebx,[esi+4]
mov ecx,[esi+8]
mov [edi],eax
mov [edi+4],ebx
mov [edi+8],ecx
add si,12
loop ...

This is entirely off the top of my head, but you get the idea. This will be
most efficient for 486+ processors.

---