/***********************
 * Computes:
 *	a[] = b[] + c[]
 */

T[] _arraySliceSliceAddSliceAssign_d(T[] a, T[] c, T[] b)
in
{
	assert(a.length == b.length && b.length == c.length);
	assert(disjoint(a, b));
	assert(disjoint(a, c));
	assert(disjoint(b, c));
}
body
{
    auto aptr = a.ptr;
    auto aend = aptr + a.length;
    auto bptr = b.ptr;
    auto cptr = c.ptr;

    version (D_InlineAsm_X86)
    {
	// SSE2 version is 333% faster 
	if (sse2() && b.length >= 16)
	{
	    auto n = aptr + (b.length & ~15);

        /*
	    // Unaligned case
	    asm 
	    {
		mov EAX, bptr; // left operand
		mov ECX, cptr; // right operand
		mov ESI, aptr; // destination operand
		mov EDI, n;    // end comparison

		align 8;
	    startsseloopb:
		movupd XMM0, [EAX]; 
		movupd XMM1, [EAX+16];
		movupd XMM2, [EAX+32];
		movupd XMM3, [EAX+48];
		add EAX, 64;
		movupd XMM4, [ECX]; 
		movupd XMM5, [ECX+16];
		movupd XMM6, [ECX+32];
		movupd XMM7, [ECX+48];
		add ESI, 64;
		addpd XMM0, XMM4;
		addpd XMM1, XMM5;
		addpd XMM2, XMM6;
		addpd XMM3, XMM7;
		add ECX, 64;
		movupd [ESI+ 0-64], XMM0;
		movupd [ESI+16-64], XMM1;
		movupd [ESI+32-64], XMM2;
		movupd [ESI+48-64], XMM3;
		cmp ESI, EDI; 
		jb startsseloopb;

		mov aptr, ESI;
		mov bptr, EAX;
		mov cptr, ECX;
	    }
        */
	    if (((cast(size_t) aptr | cast(size_t) bptr | cast(size_t) cptr) & 15) == 0)
        {
	    // Aligned case
	    asm 
	    {
		mov EAX, bptr; // left operand
		mov ECX, cptr; // right operand
		mov ESI, aptr; // destination operand
		mov EDI, n;    // end comparison

		align 8;
	    startsseloopa:
		movapd XMM0, [EAX]; 
		movapd XMM1, [EAX+16];
		movapd XMM2, [EAX+32];
		movapd XMM3, [EAX+48];
		add EAX, 64;
		movapd XMM4, [ECX]; 
		movapd XMM5, [ECX+16];
		movapd XMM6, [ECX+32];
		movapd XMM7, [ECX+48];
		add ESI, 64;
		addpd XMM0, XMM4;
		addpd XMM1, XMM5;
		addpd XMM2, XMM6;
		addpd XMM3, XMM7;
		add ECX, 64;
		movapd [ESI+ 0-64], XMM0;
		movapd [ESI+16-64], XMM1;
		movapd [ESI+32-64], XMM2;
		movapd [ESI+48-64], XMM3;
		cmp ESI, EDI; 
		jb startsseloopa;

		mov aptr, ESI;
		mov bptr, EAX;
		mov cptr, ECX;
	    }
        }
        else
        {
	    // Unaligned case
	    asm 
	    {
		mov EAX, bptr; // left operand
		mov ECX, cptr; // right operand
		mov ESI, aptr; // destination operand
		mov EDI, n;    // end comparison

		align 8;
	    startsseloopu:
		movupd XMM0, [EAX]; 
		movupd XMM1, [EAX+16];
		movupd XMM2, [EAX+32];
		movupd XMM3, [EAX+48];
		add EAX, 64;
		movupd XMM4, [ECX]; 
		movupd XMM5, [ECX+16];
		movupd XMM6, [ECX+32];
		movupd XMM7, [ECX+48];
		add ESI, 64;
		addpd XMM0, XMM4;
		addpd XMM1, XMM5;
		addpd XMM2, XMM6;
		addpd XMM3, XMM7;
		add ECX, 64;
		movupd [ESI+ 0-64], XMM0;
		movupd [ESI+16-64], XMM1;
		movupd [ESI+32-64], XMM2;
		movupd [ESI+48-64], XMM3;
		cmp ESI, EDI; 
		jb startsseloopu;

		mov aptr, ESI;
		mov bptr, EAX;
		mov cptr, ECX;
	    }
        }
	}
    }

    // Handle remainder
    while (aptr < aend)
	*aptr++ = *bptr++ + *cptr++;

    return a;
}