TEXT ·gcmAesFinish(SB),NOSPLIT,$0
#define pTbl DI
#define tMsk SI
#define tPtr DX
#define plen AX
#define dlen CX

	MOVQ productTable+0(FP), pTbl
	MOVQ tagMask+8(FP), tMsk
	MOVQ T+16(FP), tPtr
	MOVQ pLen+24(FP), plen
	MOVQ dLen+32(FP), dlen

	MOVOU (tPtr), ACC0
	MOVOU (tMsk), T2

	MOVOU bswapMask<>(SB), BSWAP
	MOVOU gcmPoly<>(SB), POLY

	SHLQ $3, plen

#define pTbl R0
#define tMsk R1
#define tPtr R2
#define plen R3
#define dlen R4

	MOVD	$0xC2, R1
	LSL	$56, R1
	MOVD	$1, R0
	VMOV	R1, POLY.D[0]
	VMOV	R0, POLY.D[1]
	VEOR	ZERO.B16, ZERO.B16, ZERO.B16

	MOVD	productTable+0(FP), pTbl
	MOVD	tagMask+8(FP), tMsk
	MOVD	T+16(FP), tPtr
	MOVD	pLen+24(FP), plen
	MOVD	dLen+32(FP), dlen

	VLD1	(tPtr), [ACC0.B16]
	VLD1	(tMsk), [B1.B16]

	LSL	$3, plen

	SYS_ADJTIME              = 421 // { int|sys|50|adjtime(const struct timeval *delta, struct timeval *olddelta); }
	SYS_FUTIMES              = 423 // { int|sys|50|futimes(int fd, const struct timeval *tptr); }
	SYS_LUTIMES              = 424 // { int|sys|50|lutimes(const char *path, const struct timeval *tptr); }

	SYS_ADJTIME              = 421 // { int|sys|50|adjtime(const struct timeval *delta, struct timeval *olddelta); }
	SYS_FUTIMES              = 423 // { int|sys|50|futimes(int fd, const struct timeval *tptr); }
	SYS_LUTIMES              = 424 // { int|sys|50|lutimes(const char *path, const struct timeval *tptr); }

	SYS_ADJTIME              = 421 // { int|sys|50|adjtime(const struct timeval *delta, struct timeval *olddelta); }
	SYS_FUTIMES              = 423 // { int|sys|50|futimes(int fd, const struct timeval *tptr); }
	SYS_LUTIMES              = 424 // { int|sys|50|lutimes(const char *path, const struct timeval *tptr); }

	SYS_ADJTIME              = 421 // { int|sys|50|adjtime(const struct timeval *delta, struct timeval *olddelta); }
	SYS_FUTIMES              = 423 // { int|sys|50|futimes(int fd, const struct timeval *tptr); }
	SYS_LUTIMES              = 424 // { int|sys|50|lutimes(const char *path, const struct timeval *tptr); }

	SYS_ADJTIME              = 421 // { int|sys|50|adjtime(const struct timeval *delta, struct timeval *olddelta); }
	SYS_FUTIMES              = 423 // { int|sys|50|futimes(int fd, const struct timeval *tptr); }
	SYS_LUTIMES              = 424 // { int|sys|50|lutimes(const char *path, const struct timeval *tptr); }

	SYS_ADJTIME              = 421 // { int|sys|50|adjtime(const struct timeval *delta, struct timeval *olddelta); }
	SYS_FUTIMES              = 423 // { int|sys|50|futimes(int fd, const struct timeval *tptr); }
	SYS_LUTIMES              = 424 // { int|sys|50|lutimes(const char *path, const struct timeval *tptr); }

	SYS_ADJTIME              = 421 // { int|sys|50|adjtime(const struct timeval *delta, struct timeval *olddelta); }
	SYS_FUTIMES              = 423 // { int|sys|50|futimes(int fd, const struct timeval *tptr); }
	SYS_LUTIMES              = 424 // { int|sys|50|lutimes(const char *path, const struct timeval *tptr); }

	SYS_ADJTIME              = 421 // { int|sys|50|adjtime(const struct timeval *delta, struct timeval *olddelta); }
	SYS_FUTIMES              = 423 // { int|sys|50|futimes(int fd, const struct timeval *tptr); }
	SYS_LUTIMES              = 424 // { int|sys|50|lutimes(const char *path, const struct timeval *tptr); }