//go:noescape
func gcmAesData(productTable *[256]byte, data []byte, T *[16]byte)

//go:noescape
func gcmAesEnc(productTable *[256]byte, dst, src []byte, ctr, T *[16]byte, ks []uint32)

//go:noescape
func gcmAesDec(productTable *[256]byte, dst, src []byte, ctr, T *[16]byte, ks []uint32)

//go:noescape
func gcmAesFinish(productTable *[256]byte, tagMask, T *[16]byte, pLen, dLen uint64)

const (
	gcmBlockSize         = 16

	PXOR T0, ACC0

	reduceRound(ACC0)
	reduceRound(ACC0)
	PXOR ACC1, ACC0

gcmAesEncDone:
	MOVOU ACC0, (tPtr)
	RET
#undef increment

// func gcmAesDec(productTable *[256]byte, dst, src []byte, ctr, T *[16]byte, ks []uint32)
TEXT ·gcmAesDec(SB),0,$128-96
#define increment(i) ADDL $1, aluCTR; MOVL aluCTR, aluTMP; XORL aluK, aluTMP; BSWAPL aluTMP; MOVL aluTMP, (3*4 + i*16)(SP)
#define combinedDecRound(i) \

tailLast:
	VEOR	T0.B16, B0.B16, B0.B16
	VAND	T3.B16, B0.B16, B0.B16
	B	encReduce

done:
	VST1	[ACC0.B16], (tPtr)
	RET

// func gcmAesDec(productTable *[256]byte, dst, src []byte, ctr, T *[16]byte, ks []uint32)
TEXT ·gcmAesDec(SB),NOSPLIT,$0
	MOVD	productTable+0(FP), pTbl
	MOVD	dst+8(FP), dstPtr
	MOVD	src_base+32(FP), srcPtr
	MOVD	src_len+40(FP), srcPtrLen
	MOVD	ctr+56(FP), ctrPtr