#define ROUND3FIRST(a, b, c, d, index, const, shift) \
	MOVL	d,		R9; \
	XORL	c,		R9; \
	XORL	b,		R9; \
	ADDL	$const,	a; \
	ADDL	R8,		a; \
	MOVL	(index*4)(SI),R8; \
	ADDL	R9,		a; \
	ROLL	$shift,		a; \
	ADDL	b,		a

#define ROUND3(a, b, c, d, index, const, shift) \
	XORL	a,		R9; \
	XORL	b,		R9; \
	ADDL	$const,	a; \
	ADDL	R8,		a; \

}

// Expand 48-bit input to 64-bit, with each 6-bit block padded by extra two bits at the top.
// By doing so, we can have the input blocks (four bits each), and the key blocks (six bits each) well-aligned without
// extra shifts/rotations for alignments.
func unpack(x uint64) uint64 {
	return ((x>>(6*1))&0xff)<<(8*0) |
		((x>>(6*3))&0xff)<<(8*1) |
		((x>>(6*5))&0xff)<<(8*2) |
		((x>>(6*7))&0xff)<<(8*3) |
		((x>>(6*0))&0xff)<<(8*4) |

    --node-image)
      NODE_IMAGE=$2
      shift 2
    ;;
    # Config for enabling different Kubernetes features in KinD (see prow/config{endpointslice.yaml,trustworthy-jwt.yaml}).
    --kind-config)
    KIND_CONFIG=$2
    shift 2
    ;;
    --skip-setup)
      SKIP_SETUP=true
      shift
    ;;
    --skip-cleanup)
      SKIP_CLEANUP=true
      shift
    ;;
    --skip-build)
      SKIP_BUILD=true

	CMPBEQ	R6, R7, end

loop:
	STMY	R2, R5, tmp-16(SP)

	MOVWBR	0(R6), R8
	MOVWZ	R5, R9

#define ROUND1(a, b, c, d, index, const, shift) \
	XOR	c, R9; \
	ADD	$const, a; \
	ADD	R8, a; \
	MOVWBR	(index*4)(R6), R8; \
	AND	b, R9; \
	XOR	d, R9; \
	ADD	R9, a; \
	RLL	$shift, a; \
	MOVWZ	c, R9; \
	ADD	b, a

	ROUND1(R2,R3,R4,R5, 1,0xd76aa478, 7);
	ROUND1(R5,R2,R3,R4, 2,0xe8c7b756,12);

	// badCharSkip[b] is len(pattern).
	//
	// Whenever a mismatch is found with byte b in the text, we can safely
	// shift the matching frame at least badCharSkip[b] until the next time
	// the matching char could be in alignment.
	badCharSkip [256]int

	// goodSuffixSkip[i] defines how far we can shift the matching frame given
	// that the suffix pattern[i+1:] matches, but the byte pattern[i] does
	// not. There are two cases to consider:

	_ = 0xabci // ERROR "hexadecimal floating-point"
	_ = 0x0p1i // ERROR "hexadecimal floating-point"
)

// signed shift counts
var (
	s int
	_ = 1 << s // ERROR "invalid operation: 1 << s \(signed shift count type int\) requires go1.13 or later|signed shift count"
	_ = 1 >> s // ERROR "signed shift count"

	_ = 0xabci // ERROR "hexadecimal floating-point"
	_ = 0x0p1i // ERROR "hexadecimal floating-point"
)

// signed shift counts
var (
	s int
	_ = 1 << s // ERROR "invalid operation: signed shift count s (variable of type int) requires go1.13 or later"
	_ = 1 >> s // ERROR "signed shift count"

	MOVM.IA (Rc0), [Ra,Rb,Rc,Rd]

// a += (((c^d)&b)^d) + X[index] + const
// a = a<<shift | a>>(32-shift) + b
#define ROUND1(Ra, Rb, Rc, Rd, index, shift, Rconst) \
	EOR	Rc, Rd, Rt0		; \
	AND	Rb, Rt0			; \
	EOR	Rd, Rt0			; \
	MOVW	(index<<2)(Rdata), Rt1	; \
	ADD	Rt1, Rt0			; \
	ADD	Rconst, Rt0			; \
	ADD	Rt0, Ra			; \
	ADD	Ra@>(32-shift), Rb, Ra	;

	MOVM.IA.W (Rtable), [Rc0,Rc1,Rc2,Rc3]
	ROUND1(Ra, Rb, Rc, Rd,  0,	7, Rc0)

    val sink = sink()
    val bitCount = 64 - java.lang.Long.numberOfLeadingZeros(v)
    val byteCount = (bitCount + 6) / 7
    for (shift in (byteCount - 1) * 7 downTo 0 step 7) {
      val lastBit = if (shift == 0) 0 else 0b1000_0000
      sink.writeByte(((v shr shift) and 0b0111_1111).toInt() or lastBit)
    }
  }

  override fun toString(): String = path.joinToString(separator = " / ")

	ORL	DI,		BP; \
	MOVL	(index*4)(SI),DI; \
	ADDL	BP,		a; \
	ROLL	$shift,	a; \
	ADDL	b,		a

#define ROUND3(a, b, c, d, index, const, shift) \
	LEAL	const(a)(DI*1),a; \
	MOVL	(index*4)(SI),DI; \
	XORL	d,		BP; \
	XORL	b,		BP; \
	ADDL	BP,		a; \
	ROLL	$shift,		a; \
	MOVL	b,		BP; \
	ADDL	b,		a

#define ROUND4(a, b, c, d, index, const, shift) \
	LEAL	const(a)(DI*1),a; \
	ORL	b,		BP; \
	XORL	c,		BP; \