//
// func castagnoliSSE42(crc uint32, p []byte) uint32
TEXT ·castagnoliSSE42(SB),NOSPLIT,$0
	MOVL crc+0(FP), AX  // CRC value
	MOVQ p+8(FP), SI  // data pointer
	MOVQ p_len+16(FP), CX  // len(p)

	// If there are fewer than 8 bytes to process, skip alignment.
	CMPQ CX, $8
	JL less_than_8

	MOVQ SI, BX
	ANDQ $7, BX
	JZ aligned

	MOVD    p+8(FP), R3       // data pointer
	MOVD    p_len+16(FP), R4  // len(p)

	MOVD    $·crclecons(SB), R5
	BR      vectorizedBody<>(SB)

// func vectorizedCastagnoli(crc uint32, p []byte) uint32
TEXT ·vectorizedCastagnoli(SB),NOSPLIT,$0
	MOVWZ   crc+0(FP), R2     // R2 stores the CRC value
	MOVD    p+8(FP), R3       // data pointer
	MOVD    p_len+16(FP), R4  // len(p)

		x.mode = value
		x.typ = S
		if check.recordTypes() {
			check.recordBuiltinType(call.Fun, sig)
		}

	case _Cap, _Len:
		// cap(x)
		// len(x)
		mode := invalid
		var val constant.Value
		switch t := arrayPtrDeref(under(x.typ)).(type) {
		case *Basic:
			if isString(t) && id == _Len {
				if x.mode == constant_ {
					mode = constant_
					val = constant.MakeInt64(int64(len(constant.StringVal(x.val))))

		x.mode = value
		x.typ = S
		if check.recordTypes() {
			check.recordBuiltinType(call.Fun, sig)
		}

	case _Cap, _Len:
		// cap(x)
		// len(x)
		mode := invalid
		var val constant.Value
		switch t := arrayPtrDeref(under(x.typ)).(type) {
		case *Basic:
			if isString(t) && id == _Len {
				if x.mode == constant_ {
					mode = constant_
					val = constant.MakeInt64(int64(len(constant.StringVal(x.val))))

        int clen = s.length();
        int blen = 2 * clen;
        byte[] b = new byte[blen];
        char[] c = new char[clen];
        s.getChars(0, clen, c, 0);
        for ( int i = 0, j = 0; i < clen; i++ ) {
            b[ j++ ] = (byte) ( c[ i ] >>> 8 );
            b[ j++ ] = (byte) ( c[ i ] >>> 0 );
        }
        write(b, 0, blen);
    }


    @Override

// A builtinId is the id of a builtin function.
type builtinId int

const (
	// universe scope
	_Append builtinId = iota
	_Cap
	_Clear
	_Close
	_Complex
	_Copy
	_Delete
	_Imag
	_Len
	_Make
	_Max
	_Min
	_New
	_Panic
	_Print
	_Println
	_Real
	_Recover

	// package unsafe
	_Add
	_Alignof
	_Offsetof
	_Sizeof
	_Slice
	_SliceData
	_String

// A builtinId is the id of a builtin function.
type builtinId int

const (
	// universe scope
	_Append builtinId = iota
	_Cap
	_Clear
	_Close
	_Complex
	_Copy
	_Delete
	_Imag
	_Len
	_Make
	_Max
	_Min
	_New
	_Panic
	_Print
	_Println
	_Real
	_Recover

	// package unsafe
	_Add
	_Alignof
	_Offsetof
	_Sizeof
	_Slice
	_SliceData
	_String

//
// ARM64 version of md5block.go
// derived from crypto/md5/md5block_amd64.s

//go:build !purego

#include "textflag.h"

TEXT	·block(SB),NOSPLIT,$0-32
	MOVD	dig+0(FP), R0
	MOVD	p+8(FP), R1
	MOVD	p_len+16(FP), R2
	AND	$~63, R2
	CBZ	R2, zero

	ADD	R1, R2, R21
	LDPW	(0*8)(R0), (R4, R5)
	LDPW	(1*8)(R0), (R6, R7)

loop:
	MOVW	R4, R12
	MOVW	R5, R13
	MOVW	R6, R14
	MOVW	R7, R15

	MOVD	p_base+24(FP), R1                            // message first address
	MOVD	k_base+48(FP), R2                            // k constants first address
	MOVD	p_len+32(FP), R3                             // message length
	VLD1.P	16(R0), [V0.S4]
	FMOVS	(R0), F20
	SUB	$16, R0, R0

blockloop:

	VLD1.P	16(R1), [V4.B16]                             // load message

	ADDL	BP,		a; \
	MOVL	(index*4)(SI),DI; \
	MOVL	$0xffffffff,	BP; \
	ROLL	$shift,		a; \
	XORL	c,		BP; \
	ADDL	b,		a

TEXT	·block(SB),NOSPLIT,$24-16
	MOVL	dig+0(FP),	BP
	MOVL	p+4(FP),	SI
	MOVL	p_len+8(FP), DX
	SHRL	$6,		DX
	SHLL	$6,		DX

	LEAL	(SI)(DX*1),	DI
	MOVL	(0*4)(BP),	AX
	MOVL	(1*4)(BP),	BX
	MOVL	(2*4)(BP),	CX
	MOVL	(3*4)(BP),	DX

	CMPL	SI,		DI
	JEQ	end

	MOVL	DI,		16(SP)