MOVL	d,		R9; \
	XORL	c,		R9; \
	XORL	b,		R9; \
	ADDL	$const,	a; \
	ADDL	R8,		a; \
	MOVL	(index*4)(SI),R8; \
	ADDL	R9,		a; \
	ROLL	$shift,		a; \
	ADDL	b,		a

#define ROUND3(a, b, c, d, index, const, shift) \
	XORL	a,		R9; \
	XORL	b,		R9; \
	ADDL	$const,	a; \
	ADDL	R8,		a; \
	MOVL	(index*4)(SI),R8; \
	ADDL	R9,		a; \
	ROLL	$shift,		a; \
	ADDL	b,		a

	ROUND3FIRST(AX,BX,CX,DX, 8,0xfffa3942, 4);

	ADDL	(4*4)(DI), AX
	MOVL	AX, (4*4)(DI)
	MOVL	AX, (4*4)(BP)
	MOVL	(5*4)(BP), BX		// H5 = f + H5
	ADDL	(5*4)(DI), BX
	MOVL	BX, (5*4)(DI)
	MOVL	BX, (5*4)(BP)
	MOVL	(6*4)(BP), CX		// H6 = g + H6
	ADDL	(6*4)(DI), CX
	MOVL	CX, (6*4)(DI)
	MOVL	CX, (6*4)(BP)
	MOVL	(7*4)(BP), DX		// H7 = h + H7
	ADDL	(7*4)(DI), DX
	MOVL	DX, (7*4)(DI)
	MOVL	DX, (7*4)(BP)

	ADDL	$64, SI
	CMPL	SI, 288(SP)

	MOVQ dig+0(FP), BP
	ADDL (0*4)(BP), R8  // H0 = a + H0
	MOVL R8, (0*4)(BP)
	ADDL (1*4)(BP), R9  // H1 = b + H1
	MOVL R9, (1*4)(BP)
	ADDL (2*4)(BP), R10 // H2 = c + H2
	MOVL R10, (2*4)(BP)
	ADDL (3*4)(BP), R11 // H3 = d + H3
	MOVL R11, (3*4)(BP)
	ADDL (4*4)(BP), R12 // H4 = e + H4
	MOVL R12, (4*4)(BP)
	ADDL (5*4)(BP), R13 // H5 = f + H5
	MOVL R13, (5*4)(BP)

	ROUND4(AX,BX,CX,DX,11,0xf7537e82, 6);
	ROUND4(DX,AX,BX,CX, 2,0xbd3af235,10);
	ROUND4(CX,DX,AX,BX, 9,0x2ad7d2bb,15);
	ROUND4(BX,CX,DX,AX, 0,0xeb86d391,21);

	ADDL	0(SP),	AX
	ADDL	4(SP),	BX
	ADDL	8(SP),	CX
	ADDL	12(SP),	DX

	ADDL	$64,		SI
	CMPL	SI,		16(SP)
	JB	loop

end:
	MOVL	dig+0(FP),	BP
	MOVL	AX,		(0*4)(BP)
	MOVL	BX,		(1*4)(BP)
	MOVL	CX,		(2*4)(BP)
	MOVL	DX,		(3*4)(BP)

	ROUND4(BP, AX, BX, CX, DX, 76)
	ROUND4(DX, BP, AX, BX, CX, 77)
	ROUND4(CX, DX, BP, AX, BX, 78)
	ROUND4(BX, CX, DX, BP, AX, 79)

	ADDL	64(SP), AX
	ADDL	68(SP), BX
	ADDL	72(SP), CX
	ADDL	76(SP), DX
	ADDL	80(SP), BP

	MOVL	88(SP), SI
	ADDL	$64, SI
	CMPL	SI, 84(SP)
	JB	loop

end:
	MOVL	dig+0(FP), DI
	MOVL	AX, (0*4)(DI)
	MOVL	BX, (1*4)(DI)
	MOVL	CX, (2*4)(DI)

	ROUND4(AX, BX, CX, DX, BP, 75)
	ROUND4(BP, AX, BX, CX, DX, 76)
	ROUND4(DX, BP, AX, BX, CX, 77)
	ROUND4(CX, DX, BP, AX, BX, 78)
	ROUND4(BX, CX, DX, BP, AX, 79)

	ADDL	R11, AX
	ADDL	R12, BX
	ADDL	R13, CX
	ADDL	R14, DX
	ADDL	R15, BP

	ADDQ	$64, SI
	CMPQ	SI, DI
	JB	loop

end:
	MOVQ	dig+0(FP), DI
	MOVL	AX, (0*4)(DI)
	MOVL	BX, (1*4)(DI)
	MOVL	CX, (2*4)(DI)
	MOVL	DX, (3*4)(DI)

	// amd64:`SUBQ\s[A-Z]+,\s24\([A-Z]+\)`
	arr[3] -= b
	// 386:`DECL\s16\([A-Z]+\)`
	arr[4]--
	// 386:`ADDL\s[$]-20,\s20\([A-Z]+\)`
	arr[5] -= 20
	// 386:`SUBL\s\([A-Z]+\)\([A-Z]+\*4\),\s[A-Z]+`
	ef -= arr[b]
	// 386:`SUBL\s[A-Z]+,\s\([A-Z]+\)\([A-Z]+\*4\)`
	arr[c] -= b
	// 386:`ADDL\s[$]-15,\s\([A-Z]+\)\([A-Z]+\*4\)`
	arr[d] -= 15
	// 386:`DECL\s\([A-Z]+\)\([A-Z]+\*4\)`
	arr[b]--
	// amd64:`DECQ\s64\([A-Z]+\)`

// Atomically:
//	*val += delta;
//	return *val;
TEXT ·Xadd(SB), NOSPLIT, $0-12
	MOVL	ptr+0(FP), BX
	MOVL	delta+4(FP), AX
	MOVL	AX, CX
	LOCK
	XADDL	AX, 0(BX)
	ADDL	CX, AX
	MOVL	AX, ret+8(FP)
	RET

TEXT ·Xadd64(SB), NOSPLIT, $0-20
	NO_LOCAL_POINTERS
	// no XADDQ so use CMPXCHG8B loop
	MOVL	ptr+0(FP), BP
	TESTL	$7, BP
	JZ	2(PC)
	CALL	·panicUnaligned(SB)

	/* copy return values back */		\
	MOVL	stackArgsType+0(FP), DX;		\
	MOVL	stackArgs+8(FP), DI;		\
	MOVL	stackArgsSize+12(FP), CX;		\
	MOVL	stackRetOffset+16(FP), BX;		\
	MOVL	SP, SI;				\
	ADDL	BX, DI;				\
	ADDL	BX, SI;				\
	SUBL	BX, CX;				\
	CALL	callRet<>(SB);			\
	RET

// callRet copies return values back at the end of call*. This is a
// separate function so it can allocate stack space for the arguments

// uint32 Xadd(uint32 volatile *val, int32 delta)
// Atomically:
//	*val += delta;
//	return *val;
TEXT ·Xadd(SB), NOSPLIT, $0-20
	MOVQ	ptr+0(FP), BX
	MOVL	delta+8(FP), AX
	MOVL	AX, CX
	LOCK
	XADDL	AX, 0(BX)
	ADDL	CX, AX
	MOVL	AX, ret+16(FP)
	RET

// uint64 Xadd64(uint64 volatile *val, int64 delta)
// Atomically:
//	*val += delta;
//	return *val;
TEXT ·Xadd64(SB), NOSPLIT, $0-24
	MOVQ	ptr+0(FP), BX