MOVL	(5*4)(SI),	DI
	MOVL	CX,		BP

	ROUND3(AX,BX,CX,DX, 8,0xfffa3942, 4);
	ROUND3(DX,AX,BX,CX,11,0x8771f681,11);
	ROUND3(CX,DX,AX,BX,14,0x6d9d6122,16);
	ROUND3(BX,CX,DX,AX, 1,0xfde5380c,23);
	ROUND3(AX,BX,CX,DX, 4,0xa4beea44, 4);
	ROUND3(DX,AX,BX,CX, 7,0x4bdecfa9,11);
	ROUND3(CX,DX,AX,BX,10,0xf6bb4b60,16);
	ROUND3(BX,CX,DX,AX,13,0xbebfbc70,23);
	ROUND3(AX,BX,CX,DX, 0,0x289b7ec6, 4);

	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
	ROUND3(Ra, Rb, Rc, Rd,  1,	4, Rc0)
	ROUND3(Rd, Ra, Rb, Rc,  4, 11, Rc1)
	ROUND3(Rc, Rd, Ra, Rb,  7, 16, Rc2)
	ROUND3(Rb, Rc, Rd, Ra, 10, 23, Rc3)

	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
	ROUND3(Ra, Rb, Rc, Rd, 13,	4, Rc0)
	ROUND3(Rd, Ra, Rb, Rc,  0, 11, Rc1)
	ROUND3(Rc, Rd, Ra, Rb,  3, 16, Rc2)
	ROUND3(Rb, Rc, Rd, Ra,  6, 23, Rc3)

	ROUND2(Rc, Rd, Re, Ra, Rb)
	ROUND2(Rb, Rc, Rd, Re, Ra)
	SUB.S	$1, Rctr
	BNE	loop2

	MOVW	$0x8F1BBCDC, Rconst
	MOVW	$4, Rctr
loop3:	ROUND3(Ra, Rb, Rc, Rd, Re)
	ROUND3(Re, Ra, Rb, Rc, Rd)
	ROUND3(Rd, Re, Ra, Rb, Rc)
	ROUND3(Rc, Rd, Re, Ra, Rb)
	ROUND3(Rb, Rc, Rd, Re, Ra)
	SUB.S	$1, Rctr
	BNE	loop3

	MOVW	$0xCA62C1D6, Rconst
	MOVW	$4, Rctr
loop4:	ROUND4(Ra, Rb, Rc, Rd, Re)

	ROUND3(R7,R4,R5,R6, 7,0x4bdecfa9,11);
	ROUND3(R6,R7,R4,R5,10,0xf6bb4b60,16);
	ROUND3(R5,R6,R7,R4,13,0xbebfbc70,23);
	ROUND3(R4,R5,R6,R7, 0,0x289b7ec6, 4);
	ROUND3(R7,R4,R5,R6, 3,0xeaa127fa,11);
	ROUND3(R6,R7,R4,R5, 6,0xd4ef3085,16);
	ROUND3(R5,R6,R7,R4, 9, 0x4881d05,23);
	ROUND3(R4,R5,R6,R7,12,0xd9d4d039, 4);
	ROUND3(R7,R4,R5,R6,15,0xe6db99e5,11);
	ROUND3(R6,R7,R4,R5, 2,0x1fa27cf8,16);

	ROUND3(R5,R2,R3,R4, 7,0x4bdecfa9,11);
	ROUND3(R4,R5,R2,R3,10,0xf6bb4b60,16);
	ROUND3(R3,R4,R5,R2,13,0xbebfbc70,23);
	ROUND3(R2,R3,R4,R5, 0,0x289b7ec6, 4);
	ROUND3(R5,R2,R3,R4, 3,0xeaa127fa,11);
	ROUND3(R4,R5,R2,R3, 6,0xd4ef3085,16);
	ROUND3(R3,R4,R5,R2, 9, 0x4881d05,23);
	ROUND3(R2,R3,R4,R5,12,0xd9d4d039, 4);
	ROUND3(R5,R2,R3,R4,15,0xe6db99e5,11);
	ROUND3(R4,R5,R2,R3, 2,0x1fa27cf8,16);

	ROUND2(R3,R4,R5,R22,M12,0x8d2a4c8a,20);

	ROUND3(R22,R3,R4,R5,M05,0xfffa3942, 4);
	ROUND3(R5,R22,R3,R4,M08,0x8771f681,11);
	ROUND3(R4,R5,R22,R3,M11,0x6d9d6122,16);
	ROUND3(R3,R4,R5,R22,M14,0xfde5380c,23);
	ROUND3(R22,R3,R4,R5,M01,0xa4beea44, 4);
	ROUND3(R5,R22,R3,R4,M04,0x4bdecfa9,11);
	ROUND3(R4,R5,R22,R3,M07,0xf6bb4b60,16);
	ROUND3(R3,R4,R5,R22,M10,0xbebfbc70,23);
	ROUND3(R22,R3,R4,R5,M13,0x289b7ec6, 4);

	ROUND2(BX, CX, DX, BP, AX, 39)

	ROUND3(AX, BX, CX, DX, BP, 40)
	ROUND3(BP, AX, BX, CX, DX, 41)
	ROUND3(DX, BP, AX, BX, CX, 42)
	ROUND3(CX, DX, BP, AX, BX, 43)
	ROUND3(BX, CX, DX, BP, AX, 44)
	ROUND3(AX, BX, CX, DX, BP, 45)
	ROUND3(BP, AX, BX, CX, DX, 46)
	ROUND3(DX, BP, AX, BX, CX, 47)
	ROUND3(CX, DX, BP, AX, BX, 48)
	ROUND3(BX, CX, DX, BP, AX, 49)
	ROUND3(AX, BX, CX, DX, BP, 50)
	ROUND3(BP, AX, BX, CX, DX, 51)

	ROUND3(DX,AX,BX,CX, 7,0x4bdecfa9,11);
	ROUND3(CX,DX,AX,BX,10,0xf6bb4b60,16);
	ROUND3(BX,CX,DX,AX,13,0xbebfbc70,23);
	ROUND3(AX,BX,CX,DX, 0,0x289b7ec6, 4);
	ROUND3(DX,AX,BX,CX, 3,0xeaa127fa,11);
	ROUND3(CX,DX,AX,BX, 6,0xd4ef3085,16);
	ROUND3(BX,CX,DX,AX, 9, 0x4881d05,23);
	ROUND3(AX,BX,CX,DX,12,0xd9d4d039, 4);
	ROUND3(DX,AX,BX,CX,15,0xe6db99e5,11);
	ROUND3(CX,DX,AX,BX, 2,0x1fa27cf8,16);
	ROUND3(BX,CX,DX,AX, 0,0xc4ac5665,23);

	ROUND2(BX, CX, DX, BP, AX, 39)

	ROUND3(AX, BX, CX, DX, BP, 40)
	ROUND3(BP, AX, BX, CX, DX, 41)
	ROUND3(DX, BP, AX, BX, CX, 42)
	ROUND3(CX, DX, BP, AX, BX, 43)
	ROUND3(BX, CX, DX, BP, AX, 44)
	ROUND3(AX, BX, CX, DX, BP, 45)
	ROUND3(BP, AX, BX, CX, DX, 46)
	ROUND3(DX, BP, AX, BX, CX, 47)
	ROUND3(CX, DX, BP, AX, BX, 48)
	ROUND3(BX, CX, DX, BP, AX, 49)
	ROUND3(AX, BX, CX, DX, BP, 50)
	ROUND3(BP, AX, BX, CX, DX, 51)

		0x2441453,
		0xd8a1e681,
		0xe7d3fbc8,
		0x21e1cde6,
		0xc33707d6,
		0xf4d50d87,
		0x455a14ed,
		0xa9e3e905,
		0xfcefa3f8,
		0x676f02d9,
		0x8d2a4c8a,
	},
	Table3: []uint32{
		// round3
		0xfffa3942,
		0x8771f681,
		0x6d9d6122,
		0xfde5380c,
		0xa4beea44,
		0x4bdecfa9,
		0xf6bb4b60,
		0xbebfbc70,
		0x289b7ec6,
		0xeaa127fa,
		0xd4ef3085,
		0x4881d05,
		0xd9d4d039,