-4294967296, -1, 0, 1, 4294967296, 0x7FFFFFFFFFFFFFFE, 0x7FFFFFFFFFFFFFFF}},

	sizedTestData{name: "uint32", sn: "32", u: []uint64{0, 1, 4294967295}},
	sizedTestData{name: "int32", sn: "32", i: []int64{-0x80000000, -0x7FFFFFFF, -1, 0,
		1, 0x7FFFFFFF}},

	sizedTestData{name: "uint16", sn: "16", u: []uint64{0, 1, 65535}},
	sizedTestData{name: "int16", sn: "16", i: []int64{-32768, -32767, -1, 0, 1, 32766, 32767}},

		throw("cas1")
	}
	if z != 2 {
		throw("cas2")
	}

	z = 4
	if atomic.Cas(&z, 5, 6) {
		throw("cas3")
	}
	if z != 4 {
		throw("cas4")
	}

	z = 0xffffffff
	if !atomic.Cas(&z, 0xffffffff, 0xfffffffe) {
		throw("cas5")
	}
	if z != 0xfffffffe {
		throw("cas6")
	}

	m = [4]byte{1, 1, 1, 1}
	atomic.Or8(&m[1], 0xf0)

		}
		return (d * d) >> 2
	}
	testCases := []uint32{
		0,
		1,
		2,
		0x0fffd,
		0x0fffe,
		0x0ffff,
		0x10000,
		0x10001,
		0x10002,
		0xfffffffd,
		0xfffffffe,
		0xffffffff,
	}
	for _, x := range testCases {
		for _, y := range testCases {
			if got, want := sqDiff(x, y), orig(x, y); got != want {
				t.Fatalf("sqDiff(%#x, %#x): got %d, want %d", x, y, got, want)
			}
		}
	}

// specified in the ECX register into registers EDX:EAX.
// Currently, the only supported value for XCR is 0.
void
gccgoXgetbv(uint32_t *eax, uint32_t *edx)
{
	uint64_t v = _xgetbv(0);
	*eax = v & 0xffffffff;
	*edx = v >> 32;
}

#pragma clang attribute pop

	// 2,147,483,646 bytes: the maximum value for a 32-bit integer minus 1.
	// See https://docs.microsoft.com/en-us/windows/win32/api/mswsock/nf-mswsock-transmitfile
	const maxChunkSizePerCall = int64(0x7fffffff - 1)

	for n > 0 {
		chunkSize := maxChunkSizePerCall
		if chunkSize > n {
			chunkSize = n
		}

		o.qty = uint32(chunkSize)
		o.o.Offset = uint32(curpos)
		o.o.OffsetHigh = uint32(curpos >> 32)

	var x39 uint64
	x39, x38 = bits.Mul64(x30, 0xfffffffffffffffe)
	var x40 uint64
	var x41 uint64
	x41, x40 = bits.Mul64(x30, 0xffffffff00000000)
	var x42 uint64
	var x43 uint64
	x43, x42 = bits.Mul64(x30, 0xffffffff)
	var x44 uint64
	var x45 uint64
	x44, x45 = bits.Add64(x43, x40, uint64(0x0))
	var x46 uint64
	var x47 uint64
	x46, x47 = bits.Add64(x41, x38, uint64(p384Uint1(x45)))
	var x48 uint64
	var x49 uint64

                size += domain.length;
            }
            else {
                flags &= ( NTLMSSP_NEGOTIATE_OEM_DOMAIN_SUPPLIED ^ 0xffffffff );
            }

            byte[] workstation = new byte[0];
            String suppliedWorkstationString = getSuppliedWorkstation();

                domain = suppliedDomain.toUpperCase().getBytes(
                        getOEMEncoding());
            } else {
                flags &= (NTLMSSP_NEGOTIATE_OEM_DOMAIN_SUPPLIED ^ 0xffffffff);
            }
            byte[] workstation = new byte[0];
            if (suppliedWorkstation != null &&
                    suppliedWorkstation.length() != 0) {
                hostInfo = true;

	_EXCEPTION_FLT_OVERFLOW         = 0xc0000091
	_EXCEPTION_FLT_UNDERFLOW        = 0xc0000093
	_EXCEPTION_INT_DIVIDE_BY_ZERO   = 0xc0000094
	_EXCEPTION_INT_OVERFLOW         = 0xc0000095

	_INFINITE     = 0xffffffff
	_WAIT_TIMEOUT = 0x102

	_EXCEPTION_CONTINUE_EXECUTION  = -0x1
	_EXCEPTION_CONTINUE_SEARCH     = 0x0
	_EXCEPTION_CONTINUE_SEARCH_SEH = 0x1
)

type systeminfo struct {
	anon0                       [4]byte

big_loop_avx2:
	VMOVDQU	(SI), Y2
	VMOVDQU	(DI), Y3
	VMOVDQU	32(SI), Y4
	VMOVDQU	32(DI), Y5
	VPCMPEQB Y2, Y3, Y0
	VPMOVMSKB Y0, AX
	XORL	$0xffffffff, AX
	JNE	diff32_avx2
	VPCMPEQB Y4, Y5, Y6
	VPMOVMSKB Y6, AX
	XORL	$0xffffffff, AX
	JNE	diff64_avx2

	ADDQ	$64, SI
	ADDQ	$64, DI
	SUBQ	$64, R8
	CMPQ	R8, $64
	JB	big_loop_avx2_exit
	JMP	big_loop_avx2