// 8(FP) is p.len
//12(FP) is p.cap
//
// Stack frame
#define p_end	end-4(SP)	// pointer to the end of data
#define p_data	data-8(SP)	// current data pointer
#define buf	buffer-(8+4*16)(SP)	//16 words temporary buffer
		// 3 words at 4..12(R13) for called routine parameters

TEXT	·block(SB), NOSPLIT, $84-16
	MOVW	p+4(FP), Rdata	// pointer to the data
	MOVW	p_len+8(FP), Rt0	// number of bytes
	ADD	Rdata, Rt0

	MOVQ		p_len+16(FP), numBytes		// get number of input bytes to hash
	SHRQ		$6, numBytes			// force modulo 64 input buffer length
	SHLQ		$6, numBytes
	CMPQ		numBytes, $0			// exit early for zero-length input buffer
	JEQ		done
	ADDQ		dataPtr, numBytes		// point numBytes to end of input buffer
	VMOVDQU		(0*16)(digestPtr), state0	// load initial hash values and reorder
	VMOVDQU		(1*16)(digestPtr), state1	// DCBA, HGFE -> ABEF, CDGH

	VMOVDQU Y0, (OFFSET*2)(R14)


// Message scheduling pre-compute for rounds 0-15
// R13 is a pointer to even 64-byte block
// R10 is a pointer to odd 64-byte block
// R14 is a pointer to temp buffer
// X0 is used as temp register
// YREG is clobbered as part of computation
// OFFSET chooses 16 byte chunk within a block
// R8 is a pointer to constants block
// K_OFFSET chooses K constants relevant to this round

	 *     USHORT Length;
	 *     USHORT MaximumLength;
	 *     [size_is(MaximumLength / 2), length_is((Length) / 2) ] USHORT *  Buffer;
	 * } UNICODE_STRING;
	 */

	typedef struct {
		uint16_t length;
		uint16_t maximum_length;
		[length_is(length / 2),size_is(maximum_length / 2)] uint16_t *buffer;
	} unicode_string;

	/* 
	 * typedef struct _SID_IDENTIFIER_AUTHORITY {
	 *     UCHAR Value[6];

	 *     USHORT Length;
	 *     USHORT MaximumLength;
	 *     [size_is(MaximumLength / 2), length_is((Length) / 2) ] USHORT *  Buffer;
	 * } UNICODE_STRING;
	 */

	typedef struct {
		uint16_t length;
		uint16_t maximum_length;
		[length_is(length / 2),size_is(maximum_length / 2)] uint16_t *buffer;
	} unicode_string;

	/* 
	 * typedef struct _SID_IDENTIFIER_AUTHORITY {
	 *     UCHAR Value[6];

#define Rctr	R12	// loop counter
#define Rw	R14		// point to w buffer

// func block(dig *digest, p []byte)
// 0(FP) is *digest
// 4(FP) is p.array (struct Slice)
// 8(FP) is p.len
//12(FP) is p.cap
//
// Stack frame
#define p_end	end-4(SP)		// pointer to the end of data
#define p_data	data-8(SP)	// current data pointer (unused?)
#define w_buf	buf-(8+4*80)(SP)	//80 words temporary buffer w uint32[80]

TEXT ·cryptBlocksGCM(SB),NOSPLIT,$0-112
	MOVD	src_len+64(FP), R0
	MOVD	buf_base+80(FP), R1
	MOVD	cnt+104(FP), R12
	LMG	(R12), R2, R3

	// Check that the src size is less than or equal to the buffer size.
	MOVD	buf_len+88(FP), R4
	CMP	R0, R4
	BGT	crash

	// Check that the src size is a multiple of 16-bytes.
	MOVD	R0, R4
	AND	$0xf, R4
	BLT	crash // non-zero

github.com/libp2p/go-addr-util,v0.0.1,h1:TpTQm9cXVRVSKsYbgQ7GKc3KbbHVTnbostgGaDEP+88=,d49a37e15540c8b95f845dde6cdf802e7af490bc13fd88fec3da318d08464f7b
github.com/libp2p/go-buffer-pool,v0.0.2,h1:QNK2iAFa8gjAe1SPz6mHSMuCcjs+X1wlHzeOSqcmlfs=,fef932705b72198df3d50befd9d2aa157aea1b5f3d23712b09d627d02cfe841e