var tHdr, tMeta []byte // Zero copy bytes
	for i := 0; i < versions; i++ {
		tHdr, buf, e = msgp.ReadBytesZC(buf)
		if e != nil {
			return e
		}
		tMeta, buf, e = msgp.ReadBytesZC(buf)
		if e != nil {
			return e
		}
		if e = fn(i, tHdr, tMeta); e != nil {
			return e
		}
	}
	return nil
}

	var tHdr, tMeta []byte // Zero copy bytes
	for i := range versions {
		tHdr, buf, err = msgp.ReadBytesZC(buf)
		if err != nil {
			return err
		}
		tMeta, buf, err = msgp.ReadBytesZC(buf)
		if err != nil {
			return err
		}
		if err = fn(i, tHdr, tMeta); err != nil {
			return err
		}
	}
	return nil
}

	var tHdr, tMeta []byte // Zero copy bytes
	for i := range versions {
		tHdr, buf, err = msgp.ReadBytesZC(buf)
		if err != nil {
			return err
		}
		tMeta, buf, err = msgp.ReadBytesZC(buf)
		if err != nil {
			return err
		}
		if err = fn(i, tHdr, tMeta); err != nil {
			if err == errDoneForNow {
				err = nil
			}
			return err

    // Take Difference of both estimates: (f(theta + eps) - f(theta - eps)).
    TF_RETURN_IF_ERROR(
        ops::Sub(ctx, fPlus.get(), fMinus.get(), f_outputs, "sub_top"));
    AbstractTensorHandlePtr fDiff(f_outputs[0]);

    // Calculate using the difference quotient definition:
    // (f(theta + eps) - f(theta - eps)) / (2 * eps).
    TF_RETURN_IF_ERROR(

[V10.D1, V11.D1] VLD1.P 16(R0), [V12.D1, V13.D1] VLD1.P 16(R0), [V14.D1, V15.D1] VLD1.P 16(R0), [V16.D1, V17.D1] VLD1.P 16(R0), [V18.D1, V19.D1] VLD1.P 16(R0), [V20.D1, V21.D1] VLD1.P 16(R0), [V22.D1, V23.D1] VLD1 (R0), [V24.D1] SUB $192, R0, R0 loop: // theta VEOR3 V20.B16, V15.B16, V10.B16, V25.B16 VEOR3 V21.B16, V16.B16, V11.B16, V26.B16 VEOR3 V22.B16, V17.B16, V12.B16, V27.B16 VEOR3 V23.B16, V18.B16, V13.B16, V28.B16 VEOR3 V24.B16, V19.B16, V14.B16, V29.B16 VEOR3 V25.B16, V5.B16, V0.B16, V25.B16 VEOR3...