VCNTMBB V1, $1, R3                      // 10790e42
	VCNTMBD V1, $1, R3                      // 107f0e42
	VCNTMBH V1, $1, R3                      // 107b0e42
	VCNTMBW V1, $1, R3                      // 107d0e42
	VCTZDM V1, V2, V3                       // 106117c4
	VDIVESD V1, V2, V3                      // 106113cb
	VDIVESQ V1, V2, V3                      // 1061130b
	VDIVESW V1, V2, V3                      // 1061138b

	"VDIVSD",
	"VDIVEUW",
	"VDIVEUQ",
	"VDIVEUD",
	"VDIVESW",
	"VDIVESQ",
	"VDIVESD",
	"VCTZDM",
	"VCNTMBW",
	"VCNTMBH",
	"VCNTMBD",
	"VCNTMBB",
	"VCMPUQ",
	"VCMPSQ",
	"VCMPGTUQCC",
	"VCMPGTUQ",
	"VCMPGTSQCC",
	"VCMPGTSQ",
	"VCMPEQUQCC",
	"VCMPEQUQ",
	"VCLZDM",
	"VCLRRB",
	"VCLRLB",
	"VCFUGED",
	"STXVRWX",
	"STXVRHX",
	"STXVRDX",
	"STXVRBX",

// Separators with lengths of 2, 3 or 4 are handled
// specially.

// This works on power8 and above. The loads and
// compares are done in big endian order
// since that allows the used of VCLZD, and allows
// the same implementation to work on big and little
// endian platforms with minimal conditional changes.

// NOTE: There is a power9 implementation that
// improves performance by 10-15% on little