// the same because all strides are 1).
      low_padding[dim] =
          conv_grad_spatial_dim.pad_before - orig_padding[dim].first;
      high_padding[dim] =
          conv_grad_spatial_dim.pad_after - orig_padding[dim].second;

      // Update `out_grad_shape` to result shape of following `PadOp`.
      out_grad_shape[dim] = low_padding[dim] + high_padding[dim] +

      %1 = stablehlo.get_dimension_size %0, dim = 0 : (tensor<?x3xf32>) -> tensor<i32>
      %2 = stablehlo.reshape %1 : (tensor<i32>) -> tensor<1xi32>
      %3 = stablehlo.concatenate %2, %cst_0, dim = 0 : (tensor<1xi32>, tensor<1xi32>) -> tensor<2xi32>

    int64_t index = item.index();
    int64_t dim = item.value().getSExtValue();
    if (dim < 0 || dim > output_rank) {
      return emitOptionalError(location, "out of range broadcast dim");
    }
    if (is_broadcasted[dim]) {
      return emitOptionalError(location, "broadcast_dims has duplicates");
    }
    broadcast_shape[dim] = min_rank_ty.getDimSize(index);
    is_broadcasted[dim] = true;
  }

  IntegerAttr dim_i32;
  if (failed(ConvertToI32Attr(dim_int, &dim_i32))) return failure();

  int dim = dim_i32.getInt();
  if (dim < 0) dim += rank;

  const SmallVector<Value, 4>& slice_outputs = SliceOutputs(
      splitv_op, input, input_type, dim, splitv_op.getNumSplits(), &rewriter);
  rewriter.replaceOp(splitv_op, slice_outputs);

  return success();
}

  for (size_t i = 0; i < window_dimensions.size(); i++) {
    auto dim = window.add_dimensions();
    dim->set_size(window_dimensions[i]);
    if (!window_strides.empty()) {
      dim->set_stride(window_strides[i]);
    } else {
      dim->set_stride(1);
    }
    if (!padding.empty()) {
      dim->set_padding_low(padding[i].first);
      dim->set_padding_high(padding[i].second);
    } else {

foreach actFnPair = [[TFL_ReluOp, TFL_AF_Relu],
                     [TFL_Relu6Op, TFL_AF_Relu6],
                     [TFL_Relu1Op, TFL_AF_Relu1]] in {
  defm : FuseActFnIntoConvOpPat<!cast<Op>(actFnPair[0]), !cast<ConstantStrAttr>(actFnPair[1])>;
  defm : FuseActFnIntoPoolOpPat<!cast<Op>(actFnPair[0]), !cast<ConstantStrAttr>(actFnPair[1])>;
}

class CanFuseConvOrDepthwiseConv<string is_depthwise> : Constraint<

  if (h == nullptr) {
    status->status = tensorflow::errors::InvalidArgument("Invalid handle");
    return -1;
  }

  int64_t dim = -1;
  status->status = tensorflow::unwrap(h)->Dim(dim_index, &dim);
  return dim;
}

const char* TFE_TensorHandleDeviceName(TFE_TensorHandle* h, TF_Status* status) {
  if (h == nullptr) {

  bool ceil_mode;  // Rounding strategy (ceil or floor).
};

// Rounds the dimension based on the ceil mode.
int RoundDim(float dim, bool ceil_mode) {
  if (ceil_mode) {
    return std::ceil(dim);
  }
  return std::floor(dim);
}

// For H or W, calculate the output dimension for average pool.
int CalculateSpatialOutDim(int in, int k, int p, int s, bool ceil_mode) {

  TF_ASSERT_OK(output_proto.status());
  EXPECT_EQ(output_proto->dtype(), DT_FLOAT);
  EXPECT_EQ(output_proto->shape().dim_size(), 3);
  EXPECT_EQ(output_proto->shape().dim().at(0).size(), 1);
  EXPECT_EQ(output_proto->shape().dim().at(1).size(), 2);
  EXPECT_EQ(output_proto->shape().dim().at(2).size(), 3);
}

TEST(ConvertTypeToTensorSpecProtoTest, ScalarTensorType) {
  mlir::MLIRContext context;
  mlir::Builder b(&context);

bool HasStaticShapeAtDims(Value value, const ArrayRef<int> dims) {
  auto shaped_type = mlir::dyn_cast<ShapedType>(value.getType());
  if (!shaped_type || !shaped_type.hasRank()) return false;

  for (auto dim : dims) {
    if (shaped_type.isDynamicDim(dim)) return false;
  }
  return true;
}

Type CloneTypeWithNewElementType(Type old_type, Type element_type) {
  if (!mlir::isa<ShapedType>(old_type)) return {};