}

  // Also fold if `input` has a known rank.
  auto input_type = getInput().getType().cast<ShapedType>();
  // Do not fold if rank is zero because the TFLite converter doesn't
  // distinguish between unranked input and scalar input due to b/138865275.
  // TODO(b/138865275): Remove `input_type.getRank() != 0` in the following
  // predicate and fold the op when rank is zero.

                          << range_begin << ", " << range_end
                          << "); actual value: " << axis;
  }

  return success();
}

OpFoldResult PackOp::fold(FoldAdaptor) {
  // Fold pack operation if it computes the input tensor shape:
  //
  //   %shape  = tf.Shape(%arg)                    // [? x ...]
  //   %dim0   = tf.StridedSlice(%shape, 0, 1, 1)  // get unknown dim0 value

// AddNOp
//===----------------------------------------------------------------------===//

OpFoldResult AddNOp::fold(FoldAdaptor adaptor) {
  auto operands = adaptor.getOperands();
  if (operands.size() == 1) return *getInputs().begin();

  // Fold if there is only one single non-zero operand or all operands are zero.
  int non_zero_index = -1;
  auto IsKnownZero = [](Attribute attr) {

		t.Second.goString(indent+2, "Second: "),
		t.Third.goString(indent+2, "Third: "))
}

// Fold is a C++17 fold-expression.  Arg2 is nil for a unary operator.
type Fold struct {
	Left bool
	Op   AST
	Arg1 AST
	Arg2 AST
}

func (f *Fold) print(ps *printState) {
	op, _ := f.Op.(*Operator)
	printOp := func() {
		if op != nil {
			if ps.llvmStyle {

  let summary = "Move transposes pass.";
  let constructor = "TF::CreateMoveTransposesPass()";
  let options = [
    Option<"fold_transpose_in_ops_", "fold-transpose-in-ops", "bool", /*default=*/"true",
           "Whether to fold transposes in ops which can support folding.">,
    Option<"direction_", "direction", "enum MoveTransposeDirection",

      Map<? extends K, ? extends V> map) {
    return new HashMap<>(map);
  }

  /**
   * Creates a {@code HashMap} instance, with a high enough "initial capacity" that it <i>should</i>
   * hold {@code expectedSize} elements without growth. This behavior cannot be broadly guaranteed,
   * but it is observed to be true for OpenJDK 1.7. It also can't be guaranteed that the method

      Map<? extends K, ? extends V> map) {
    return new HashMap<>(map);
  }

  /**
   * Creates a {@code HashMap} instance, with a high enough "initial capacity" that it <i>should</i>
   * hold {@code expectedSize} elements without growth. This behavior cannot be broadly guaranteed,
   * but it is observed to be true for OpenJDK 1.7. It also can't be guaranteed that the method

    // TF::MulOp is used to fold the constant.
    // TODO(b/139192933): switch to the TFL constant folding
    auto filter_type = mlir::cast<ShapedType>(filter.getType());
    if (filter_type.hasStaticShape()) {
      auto size =
          filter_type.getNumElements() * filter_type.getElementTypeBitWidth();
      // Don't constant fold if the filter is too large for TF to fold.

	}
}

// casGToWaiting transitions gp from old to _Gwaiting, and sets the wait reason.
//
// Use this over casgstatus when possible to ensure that a waitreason is set.
func casGToWaiting(gp *g, old uint32, reason waitReason) {
	// Set the wait reason before calling casgstatus, because casgstatus will use it.
	gp.waitreason = reason
	casgstatus(gp, old, _Gwaiting)
}

  for (int i = 0; i < noutputs; ++i) {
    c_outputs[i] = nullptr;
  }
}

// TF_TensorToTensorV1 decodes a string serialization to DT_RESOURCE.
// In the TFv1 convention, TF_Tensor can hold a string serialization of
// DT_RESOURCE. The string serialization is converted back to a
// ResourceHandle during Session run where the TF_Tensor is converted to a
// Tensor.