Replacement String Library Strcmp Using SSE4.2

// http://www.intel.com/content/dam/doc/manual/64-ia-32-architectures-optimization-manual.pdf
// return 0 if strings are equal, 1 if greater, -1if less
int strcmp_sse4_2(const char *src1, const char *src2) {
    int val;
    __asm{
        mov esi, src1 ;
        mov edi, src2
        mov edx, -16 ; common index relative to base of either string pointer xor eax, eax
    topofloop:
        add edx, 16 ; prevent loop carry dependency
    next:
        lea ecx, [esi+edx] ; address of fragment that we want to load
        and ecx, 0x0fff ; check least significant12 bits of addr for page boundary cmp ecx, 0x0ff0
        jg too_close_pgb ; branch to byte-granular if within 16 bytes of boundary lea ecx, [edi+edx] ; do the same check for each fragment of 2nd string
        and ecx, 0x0fff
        cmp ecx, 0x0ff0
        jg too_close_pgb
        movdqu xmm2, BYTE PTR[esi+edx]
        movdqu xmm1, BYTE PTR[edi+edx]
        pcmpistri xmm2, xmm1, 0x18 ; equal each
        ja topofloop
        jnc ret_tag
        add edx, ecx ; ecx points to the byte offset that differ
    not_equal:
        movzx eax, BYTE PTR[esi+edx] movzx edx, BYTE PTR[edi+edx] cmp eax, edx
        cmova eax, ONE
        cmovb eax, NEG_ONE
        jmp ret_tag
    too_close_pgb:
        add edx, 1 ; do byte granular compare movzx ecx, BYTE PTR[esi+edx-1]
        movzx ebx, BYTE PTR[edi+edx-1]
        cmp ecx, ebx
        jne inequality
        add ebx, ecx
        jnz next
        jmp ret_tag
    inequality:
        cmovb eax, NEG_ONE
        cmova eax, ONE ret_tag:
        mov [val], eax
    }
    return(val);
}