ADDQ	BX, AX; \
	MOVQ	AX, ((index)*8)(BP)

// Calculate T1 in AX - uses AX, CX and DX registers.
// h is also used as an accumulator. Wt is passed in AX.
//   T1 = h + BIGSIGMA1(e) + Ch(e, f, g) + Kt + Wt
//     BIGSIGMA1(x) = ROTR(14,x) XOR ROTR(18,x) XOR ROTR(41,x)
//     Ch(x, y, z) = (x AND y) XOR (NOT x AND z)
#define SHA512T1(const, e, f, g, h) \
	MOVQ	$const, DX; \
	ADDQ	AX, h; \
	MOVQ	e, AX; \

	ADDL	BX, AX; \
	MOVL	AX, ((index)*4)(BP)

// Calculate T1 in AX - uses AX, CX and DX registers.
// h is also used as an accumulator. Wt is passed in AX.
//   T1 = h + BIGSIGMA1(e) + Ch(e, f, g) + Kt + Wt
//     BIGSIGMA1(x) = ROTR(6,x) XOR ROTR(11,x) XOR ROTR(25,x)
//     Ch(x, y, z) = (x AND y) XOR (NOT x AND z)
#define SHA256T1(const, e, f, g, h) \
	ADDL	AX, h; \
	MOVL	e, AX; \
	ADDL	$const, h; \
	MOVL	e, CX; \

// Wt = SIGMA1(Wt-2) + SIGMA0(Wt-15) + Wt-16; for 16 <= t <= 63
//
// a = H0
// b = H1
// c = H2
// d = H3
// e = H4
// f = H5
// g = H6
// h = H7
//
// for t = 0 to 63 {
//    T1 = h + BIGSIGMA1(e) + Ch(e,f,g) + Kt + Wt
//    T2 = BIGSIGMA0(a) + Maj(a,b,c)
//    h = g
//    g = f
//    f = e
//    e = d + T1
//    d = c
//    c = b
//    b = a
//    a = T1 + T2
// }
//
// H0 = a + H0

// Wt = SIGMA1(Wt-2) + SIGMA0(Wt-15) + Wt-16; for 16 <= t <= 79
//
// a = H0
// b = H1
// c = H2
// d = H3
// e = H4
// f = H5
// g = H6
// h = H7
//
// for t = 0 to 79 {
//    T1 = h + BIGSIGMA1(e) + Ch(e,f,g) + Kt + Wt
//    T2 = BIGSIGMA0(a) + Maj(a,b,c)
//    h = g
//    g = f
//    f = e
//    e = d + T1
//    d = c
//    c = b
//    b = a
//    a = T1 + T2
// }
//
// H0 = a + H0