fallthrough
			case ssStarter:
				if outp > 0 {
					copy(i.buf[outp:], d)
					return i.buf[:p]
				}
				return d
			}
			copy(i.buf[outp:], d)
			outp = p
			inCopyStart, outCopyStart = i.p, outp
			if i.info.ccc < prevCC {
				goto doNorm
			}
			continue
		} else if r := i.rb.src.hangul(i.p); r != 0 {
			outp = decomposeHangul(i.buf[:], r)
			i.p += hangulUTF8Size

}

func (c *CodeDirectory) put(out []byte) []byte {
	out = put32be(out, c.magic)
	out = put32be(out, c.length)
	out = put32be(out, c.version)
	out = put32be(out, c.flags)
	out = put32be(out, c.hashOffset)
	out = put32be(out, c.identOffset)
	out = put32be(out, c.nSpecialSlots)
	out = put32be(out, c.nCodeSlots)
	out = put32be(out, c.codeLimit)
	out = put8(out, c.hashSize)
	out = put8(out, c.hashType)

				fallthrough
			case ssStarter:
				if outp > 0 {
					copy(i.buf[outp:], d)
					return i.buf[:p]
				}
				return d
			}
			copy(i.buf[outp:], d)
			outp = p
			inCopyStart, outCopyStart = i.p, outp
			if i.info.ccc < prevCC {
				goto doNorm
			}
			continue
		} else if r := i.rb.src.hangul(i.p); r != 0 {
			outp = decomposeHangul(i.buf[:], r)
			i.p += hangulUTF8Size

	ADD	$16, INP
	VXOR	INOUT, V6, INOUT
	CIPHER_BLOCK(INOUT, IVEC, INOUT, VCIPHER, VCIPHERLAST, Lcbc_enc10, Lcbc_enc12)
	VOR	INOUT, INOUT, IVEC // ciphertext (INOUT) is IVEC for next block.
	P8_STXVB16X(INOUT, OUTP, R0)
	ADD	$16, OUTP
	BDNZ	Lcbc_enc

	P8_STXVB16X(INOUT, IVP, R0)
	CLEAR_KEYS()
	RET

	PCALIGN $16
Lcbc_dec:
	P8_LXVB16X(INP, R0, TMP)
	ADD	$16, INP

    add_attrs(g.node());

    auto out0 = ops::XlaClusterOutput(scope.WithOpName("Out0"), b_identity);
    auto out1 = ops::XlaClusterOutput(scope.WithOpName("Out1"), e);
    auto out2 = ops::XlaClusterOutput(scope.WithOpName("Out2"), g);
    auto out3 = ops::XlaClusterOutput(scope.WithOpName("Out3"), read_u);

    auto consumer0_a = ops::Identity(scope.WithOpName("consumer0_a"), out0);

			if err != nil {
				return nil, err
			}
			outbuf[out4] = byte(t4 >> 8)
			out4++
			rbr4.cnt -= uint32(t4 & 0xff)
		}

		outbuf[out1] = byte(t1 >> 8)
		out1++
		rbr1.cnt -= uint32(t1 & 0xff)

		outbuf[out2] = byte(t2 >> 8)
		out2++
		rbr2.cnt -= uint32(t2 & 0xff)

		outbuf[out3] = byte(t3 >> 8)
		out3++
		rbr3.cnt -= uint32(t3 & 0xff)
	}

	return outbuf, nil

// We will only calculate for the "necessary" tensor transferred.
// from_graph & to_graph are used to compute the "necessary" tensors.
//     from_graph
//    /    \   \
//  out1   out2  out3
//           \   /
//           to_graph
// So only out2 & out3 are counted.
float GetTransferCost(const std::string& from_hardware_str,
                      const std::string& to_hardware_str,

	for i, test := range testSplitPreMasterSecretTests {
		in, _ := hex.DecodeString(test.in)
		out1, out2 := splitPreMasterSecret(in)
		s1 := hex.EncodeToString(out1)
		s2 := hex.EncodeToString(out2)
		if s1 != test.out1 || s2 != test.out2 {
			t.Errorf("#%d: got: (%s, %s) want: (%s, %s)", i, s1, s2, test.out1, test.out2)
		}
	}
}

type testKeysFromTest struct {
	version                                        uint16

    // Add StopGradients according to 'stop_outputs'.
    auto out0 = stop_outputs[0]
                    ? StopGradient(scope_, (Identity(scope_, z))).output
                    : Identity(scope_, z).output;
    auto out1 = stop_outputs[1]
                    ? StopGradient(scope_, (Identity(scope_, z))).output
                    : Identity(scope_, z).output;
    auto out2 = stop_outputs[2]

				}()
			}

			// Apply fix: should get tt.Out.
			out, fixed, ok := parseFixPrint(t, tt.Fn, tt.Name, tt.In, true)
			if !ok {
				return
			}

			// reformat to get printing right
			out, _, ok = parseFixPrint(t, fnop, tt.Name, out, false)
			if !ok {
				return
			}

			if tt.Out == "" {
				tt.Out = tt.In
			}
			if out != tt.Out {
				t.Errorf("incorrect output.\n")