Gcc optimization when copying an array

I need to profile an application that runs multiple copies of arrays, so I ended up profiling this very simple function:

typedef unsigned char UChar;
void copy_mem(UChar *src, UChar *dst, unsigned int len) {
        UChar *end = src + len;
        while (src < end)
                *dst++ = *src++;
}

I use Intel VTune to do the actual profiling, and from there I saw that when compiling with gcc -O3 and "plain" gcc (4.4) there are significant differences.

To understand why and how, I have an assembly of both compilations.

Non-optimized version:

.L3:
        movl    8(%ebp), %eax
        movzbl  (%eax), %edx
        movl    12(%ebp), %eax
        movb    %dl, (%eax)
        addl    $1, 12(%ebp)
        addl    $1, 8(%ebp)
.L2:
        movl    8(%ebp), %eax
        cmpl    -4(%ebp), %eax
        jb      .L3
        leave

So, I see that it first loads the dword from * src and places the low byte in edx, then it saves it in * dst and updates the pointers: simple enough.

Then I saw an optimized version, and I did not understand anything.

EDIT : There is an optimized build here.

: gcc ?