//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

	print("r28  ", hex(c.r28()), "\t")
	print("r29  ", hex(c.r29()), "\n")
	print("r30  ", hex(c.r30()), "\t")
	print("r31  ", hex(c.r31()), "\n")
	print("pc   ", hex(c.pc()), "\t")
	print("ctr  ", hex(c.ctr()), "\n")
	print("link ", hex(c.link()), "\t")
	print("xer  ", hex(c.xer()), "\n")
	print("ccr  ", hex(c.ccr()), "\t")
	print("trap ", hex(c.trap()), "\n")
}

//go:nosplit
//go:nowritebarrierrec

			x.outUsed = 0
		}

		if x.decrypt {
			// We can precompute a larger segment of the
			// keystream on decryption. This will allow
			// larger batches for xor, and we should be
			// able to match CTR/OFB performance.
			copy(x.next[x.outUsed:], src)
		}
		n := subtle.XORBytes(dst, src, x.out[x.outUsed:])
		if !x.decrypt {
			copy(x.next[x.outUsed:], dst)
		}
		dst = dst[n:]
		src = src[n:]
		x.outUsed += n

	ADD     $-64, LEN
	STXVW4X VS62, (OUT)(R10)
	ADD     $64, OUT

	MOVD $10, R14
	MOVD R14, CTR
	BNE  loop_outer_vsx

done_vsx:
	// Increment counter by number of 64 byte blocks
	MOVD (CNT), R14
	ADD  BLOCKS, R14
	MOVD R14, (CNT)
	RET

tail_vsx:
	ADD  $32, R1, R11
	MOVD LEN, CTR

	// Save values on stack to copy from
	STXVW4X VS32, (R11)(R0)
	STXVW4X VS36, (R11)(R8)

	__extctx_magic int32
	__pad          int32
}

type context64 struct {
	gpr        [32]uint64
	msr        uint64
	iar        uint64
	lr         uint64
	ctr        uint64
	cr         uint32
	xer        uint32
	fpscr      uint32
	fpscrx     uint32
	except     [1]uint64
	fpr        [32]float64
	fpeu       uint8
	fpinfo     uint8
	fpscr24_31 uint8

	_O_TRUNC     = 0x200
	_O_NONBLOCK  = 0x800
	_O_CLOEXEC   = 0x80000
	_SA_RESTORER = 0
)

type ptregs struct {
	gpr       [32]uint64
	nip       uint64
	msr       uint64
	orig_gpr3 uint64
	ctr       uint64
	link      uint64
	xer       uint64
	ccr       uint64
	softe     uint64
	trap      uint64
	dar       uint64
	dsisr     uint64
	result    uint64
}

type vreg struct {

	}
	if r == REG_CR {
		return "CR"
	}
	if REG_SPR0 <= r && r <= REG_SPR0+1023 {
		switch r {
		case REG_XER:
			return "XER"

		case REG_LR:
			return "LR"

		case REG_CTR:
			return "CTR"
		}

		return fmt.Sprintf("SPR(%d)", r-REG_SPR0)
	}

	if r == REG_FPSCR {
		return "FPSCR"
	}
	if r == REG_MSR {
		return "MSR"
	}

	return fmt.Sprintf("Rgok(%d)", r-obj.RBasePPC64)
}

// If other callers are added this function might
// need to change.
TEXT addMulVVWx<>(SB), NOSPLIT, $0
	MOVD	z+0(FP), R3
	MOVD	x+8(FP), R4
	MOVD	y+16(FP), R5

	MOVD	$0, R9		// R9 = c = 0
	MOVD	R6, CTR		// Initialize loop counter
	PCALIGN	$16

loop:
	MOVD	0(R4), R14	// x[i]
	MOVD	8(R4), R16	// x[i+1]
	MOVD	16(R4), R18	// x[i+2]
	MOVD	24(R4), R20	// x[i+3]
	MOVD	0(R3), R15	// z[i]
	MOVD	8(R3), R17	// z[i+1]

	_O_TRUNC     = 0x200
	_O_NONBLOCK  = 0x800
	_O_CLOEXEC   = 0x80000
	_SA_RESTORER = 0
)

type ptregs struct {
	gpr       [32]uint64
	nip       uint64
	msr       uint64
	orig_gpr3 uint64
	ctr       uint64
	link      uint64
	xer       uint64
	ccr       uint64
	softe     uint64
	trap      uint64
	dar       uint64
	dsisr     uint64
	result    uint64
}

type vreg struct {

	CMPU	R6, $8, CR6	// Check if 8 ≤ n < 64 bytes
	BLE	CR6, small	// <= 8
	BLT	CR7, xor32	// Case for 32 ≤ n < 64 bytes

	// Case for n ≥ 64 bytes
preloop64:
	SRD	$6, R6, R7	// Set up loop counter
	MOVD	R7, CTR
	MOVD	$16, R10
	MOVD	$32, R14
	MOVD	$48, R15
	ANDCC	$63, R6, R9	// Check for tailing bytes for later
	PCALIGN $16
	// Case for >= 64 bytes
	// Process 64 bytes per iteration
	// Load 4 vectors of a and b