si_high:
	// address ends in 111111xx. Load up to bytes we want, move to correct position.
	MOVL	-4(SI)(BX*1), SI
	SHRL	CX, SI
si_finish:

	// same for DI.
	MOVL	DI, DX
	CMPB	DX, $0xfc
	JA	di_high
	MOVL	(DI), DI
	JMP	di_finish
di_high:
	MOVL	-4(DI)(BX*1), DI
	SHRL	CX, DI
di_finish:

	SUBL	SI, DI
	SHLL	CX, DI
equal:
	SETEQ	(AX)

// func memclrNoHeapPointers(ptr unsafe.Pointer, n uintptr)
TEXT runtime·memclrNoHeapPointers(SB),NOSPLIT,$0-16
	MOVQ	ptr+0(FP), DI
	MOVQ	n+8(FP), CX
	MOVQ	CX, BX
	ANDQ	$7, BX
	SHRQ	$3, CX
	MOVQ	$0, AX
	CLD
	REP
	STOSQ
	MOVQ	BX, CX
	REP
	STOSB

fwdBy4:
	// Do 4 bytes at a time
	MOVL	BX, CX
	SHRL	$2, CX
	ANDL	$3, BX
	REP;	MOVSL
	JMP	tail

/*
 * check overlap
 */
back:
	MOVL	SI, CX
	ADDL	BX, CX
	CMPL	CX, DI
	JLS	forward
/*
 * whole thing backwards has
 * adjusted addresses
 */

	ADDL	BX, DI
	ADDL	BX, SI
	STD

/*
 * copy
 */
	MOVL	BX, CX
	SHRL	$2, CX
	ANDL	$3, BX

	SUBL	$4, DI

        String urlpath = getURLPath();
        String shr = getShare();
        if ( urlpath.length() > 1 ) {
            int i = urlpath.length() - 2;
            while ( urlpath.charAt(i) != '/' ) {
                i--;
            }
            return urlpath.substring(i + 1);
        }
        else if ( shr != null ) {
            return shr + '/';
        }

	hwcap2_SHA1  = 1 << 2
	hwcap2_SHA2  = 1 << 3
	hwcap2_CRC32 = 1 << 4
)

func initOptions() {
	options = []option{
		{Name: "pmull", Feature: &ARM.HasPMULL},
		{Name: "sha1", Feature: &ARM.HasSHA1},
		{Name: "sha2", Feature: &ARM.HasSHA2},
		{Name: "swp", Feature: &ARM.HasSWP},
		{Name: "thumb", Feature: &ARM.HasTHUMB},
		{Name: "thumbee", Feature: &ARM.HasTHUMBEE},
		{Name: "tls", Feature: &ARM.HasTLS},

//   SIGMA0(x) = ROTR(7,x) XOR ROTR(18,x) XOR SHR(3,x)
//   SIGMA1(x) = ROTR(17,x) XOR ROTR(19,x) XOR SHR(10,x)
#define MSGSCHEDULE1(index) \
	MOVL	((index-2)*4)(BP), AX; \
	MOVL	AX, CX; \
	RORL	$17, AX; \
	MOVL	CX, DX; \
	RORL	$19, CX; \
	SHRL	$10, DX; \
	MOVL	((index-15)*4)(BP), BX; \
	XORL	CX, AX; \
	MOVL	BX, CX; \
	XORL	DX, AX; \
	RORL	$7, BX; \
	MOVL	CX, DX; \
	SHRL	$3, DX; \
	RORL	$18, CX; \

	// generated code for
	//	func f(x uint64) (uint64, uint64) { return x/1000000000, x%1000000000 }
	// adapted to reduce duplication
	MOVQ	AX, CX
	MOVQ	$1360296554856532783, AX
	MULQ	CX
	ADDQ	CX, DX
	RCRQ	$1, DX
	SHRQ	$29, DX
	MOVQ	DX, sec+0(FP)
	IMULQ	$1000000000, DX
	SUBQ	DX, CX
	MOVL	CX, nsec+8(FP)

	XORL	AX, AX

tail:
	TESTL	BX, BX
	JEQ	_0
	CMPL	BX, $2
	JBE	_1or2
	CMPL	BX, $4
	JB	_3
	JE	_4
	CMPL	BX, $8
	JBE	_5through8
	CMPL	BX, $16
	JBE	_9through16
	MOVL	BX, CX
	SHRL	$2, CX
	REP
	STOSL
	ANDL	$3, BX
	JNE	tail
	RET

_1or2:
	MOVB	AX, (DI)
	MOVB	AX, -1(DI)(BX*1)
	RET
_0:
	RET
_3:
	MOVW	AX, (DI)
	MOVB	AX, 2(DI)
	RET
_4:

	case TYPE_BRANCH, TYPE_TEXTSIZE:
		if a.Reg != 0 || a.Index != 0 || a.Scale != 0 || a.Name != 0 {
			break
		}
		return

	case TYPE_MEM:
		return

	case TYPE_CONST:
		// TODO(rsc): After fixing SHRQ, check a.Index != 0 too.
		if a.Name != 0 || a.Sym != nil || a.Reg != 0 {
			ctxt.Diag("argument is TYPE_CONST, should be TYPE_ADDR, in %v", p)
			return
		}

    val sectionOffset = rangesBuffer.size.toInt() / 4

    // Section prefix.
    sectionIndexBuffer.writeByte(section and 0x1fc000 shr 14)
    sectionIndexBuffer.writeByte((section and 0x3f80) shr 7)

    // Section index.
    sectionIndexBuffer.writeByte((sectionOffset and 0x3f80) shr 7)
    sectionIndexBuffer.writeByte(sectionOffset and 0x7f)

    // Ranges.
    for (range in sectionMappedRanges) {