gp2flags1flags = regInfo{inputs: []regMask{gp, gp, 0}, outputs: []regMask{gp, 0}}

		gp2flags     = regInfo{inputs: []regMask{gpsp, gpsp}}
		gp1flags     = regInfo{inputs: []regMask{gpsp}}
		gp0flagsLoad = regInfo{inputs: []regMask{gpspsbg, 0}}
		gp1flagsLoad = regInfo{inputs: []regMask{gpspsbg, gpsp, 0}}
		gp2flagsLoad = regInfo{inputs: []regMask{gpspsbg, gpsp, gpsp, 0}}
		flagsgp      = regInfo{inputs: nil, outputs: gponly}

    }

    void assertHasImmediateCancel(
        ListenableFuture<String> a, ListenableFuture<String> b, Exception e) {
      ImmutableSet<ListenableFuture<String>> inputs = ImmutableSet.of(a, b);
      smartAssertTrue(inputs, e, intersect(inputs, ImmutableSet.of(doneCancelled)));
    }
  }

  /**

    }

    void assertHasImmediateCancel(
        ListenableFuture<String> a, ListenableFuture<String> b, Exception e) {
      ImmutableSet<ListenableFuture<String>> inputs = ImmutableSet.of(a, b);
      smartAssertTrue(inputs, e, intersect(inputs, ImmutableSet.of(doneCancelled)));
    }
  }

  /**

      outputs.push_back(tensor);
    }
  }

  if (!inputs.empty() || !outputs.empty()) {
    arg_nodes->resize(inputs.size());
    ret_nodes->resize(outputs.size());

    for (Node* n : GetOrderedNodes()) {
      // Handle inputs/arguments.
      auto input_it = inputs.find(n->name());
      if (input_it != inputs.end()) {
        (*arg_nodes)[input_it->second] = {n, 0};
      }

    private final boolean allMustSucceed;
    protected final ImmutableList<ClosingFuture<?>> inputs;

    private Combiner(boolean allMustSucceed, Iterable<? extends ClosingFuture<?>> inputs) {
      this.allMustSucceed = allMustSucceed;
      this.inputs = ImmutableList.copyOf(inputs);
      for (ClosingFuture<?> input : inputs) {
        input.becomeSubsumedInto(closeables);
      }
    }

    /**

    private final boolean allMustSucceed;
    protected final ImmutableList<ClosingFuture<?>> inputs;

    private Combiner(boolean allMustSucceed, Iterable<? extends ClosingFuture<?>> inputs) {
      this.allMustSucceed = allMustSucceed;
      this.inputs = ImmutableList.copyOf(inputs);
      for (ClosingFuture<?> input : inputs) {
        input.becomeSubsumedInto(closeables);
      }
    }

    /**

using ::mlir::tf_saved_model::kTfSavedModelIndexPathAttr;

// Helper struct that wraps inputs/outputs of a single SignatureDef.
struct SignatureDefData {
  // Note, we are using maps here to make order deterministic
  // for easily testing only.

  // Inputs defined in the signature def mapped to tensor names.
  std::map<std::string, std::string> inputs;
  // Outputs defined in the signature def mapped to tensor names.

}

void TF_AddInputList(TF_OperationDescription* desc, const TF_Output* inputs,
                     int num_inputs) {
  std::vector<NodeBuilder::NodeOut> input_list;
  input_list.reserve(num_inputs);
  for (int i = 0; i < num_inputs; ++i) {
    input_list.emplace_back(&inputs[i].oper->node, inputs[i].index);
  }
  desc->node_builder.Input(input_list);
}

}

def TPUAnnotateDynamicShapeInputsPass : Pass<"tf-tpu-annotate-dynamic-shape-inputs", "ModuleOp"> {
  let summary = "Annotate the inputs returned by TPUCopyWithDynamicShapeOp with dynamic shape";

  let description = [{
    This pass looks for the usage of the result of TPUCopyWithDynamicShapeOp
    and sets the shape of these inputs to be dynamic shaped. This will ensure

  let description = [{
    Performs transposed convolution operation on 3D inputs.
    Inputs:
      `inputs[0]`: required: the shape of output tensor
      `inputs[1]`: required: the filter weight tensor
      `inputs[2]`: required: the input activation tensor
      `inputs[3]`: optional: the bias tensor
  }];

  let arguments = (ins
    TFL_I32Tensor:$output_shape,