// This implementation is 2x unrolled, and interleaves vector instructions,
// used to precompute W, with scalar computation of current round
// for optimal scheduling.

// Trivial helper macros.
#define UPDATE_HASH(A,TB,C,D,E) \
	ADDL	(R9), A \
	MOVL	A, (R9) \
	ADDL	4(R9), TB \
	MOVL	TB, 4(R9) \
	ADDL	8(R9), C \
	MOVL	C, 8(R9) \
	ADDL	12(R9), D \
	MOVL	D, 12(R9) \
	ADDL	16(R9), E \
	MOVL	E, 16(R9)