// CHECK: %[[lstm:.*]] = "tfl.unidirectional_sequence_lstm"(%[[input_0]], %[[input_1]], %[[input_2]], %[[input_3]], %[[input_4]], %[[input_5]], %[[input_6]], %[[input_7]], %[[input_8]],
// CHECK-SAME: %[[input_9]], %[[input_10]], %[[input_11]], %[[input_12]], %[[input_13]], %[[input_14]], %[[input_15]], %[[input_16]], %[[input_17]], %[[input_18]], %[[input_19]]

  EXPECT_THAT(input2->type, Eq(TensorType_INT8));

  // Check if the quantization params of the minimum/maximum inputs match
  // after requantization
  EXPECT_THAT(input1->quantization->scale, Eq(input2->quantization->scale));
  EXPECT_THAT(input1->quantization->zero_point,
              Eq(input2->quantization->zero_point));

  // Check the input quantization params match the output ones.

[[sec:task_inputs_outputs]]
== Task inputs and outputs

In the most common case, a task takes some inputs and generates some outputs.
We can consider the process of Java compilation as an example of a task.
The Java source files act as inputs of the task, while the generated class files, i.e. the result of the compilation, are the outputs of the task.

.Example task inputs and outputs
image::taskInputsOutputs.png[]

		gp21           = regInfo{inputs: []regMask{gpg, gpg}, outputs: []regMask{gp}}
		gp21nog        = regInfo{inputs: []regMask{gp, gp}, outputs: []regMask{gp}}
		gp21flags      = regInfo{inputs: []regMask{gp, gp}, outputs: []regMask{gp, 0}}
		gp2flags       = regInfo{inputs: []regMask{gpg, gpg}}
		gp2flags1      = regInfo{inputs: []regMask{gp, gp}, outputs: []regMask{gp}}

    Supports generating random value input tensors mapped by the `input_key`.

    Args:
      input_key: The string key that identifies the created tensor as an input.
      shape: Shape of the tensor data.
      minval: The lower bound of the generated input
      maxval: The upper bound of the generated input
      dtype: The type of the generated input - usually dtypes.float32 for float

    full_output_types.push_back(output.getType());
  }

  // Convert split sharded inputs to MANUAL sharded inputs.
  // common_split_sharding is the split sharding that is common to all inputs
  // and outputs.
  llvm::SmallVector<Value, 4> manual_inputs;
  manual_inputs.reserve(inputs.size());
  for (Value in : inputs) {
    Type shard_type;
    if (failed(GetShardShapedType(original_op, num_cores_per_replica,

It is possible to temporarily opt out of configuration input detection in the following cases:

* Since Gradle 8.1, using many APIs related to the file system is correctly tracked as configuration inputs, including the file system checks, such as `File.exists()` or `File.isFile()`.
+

        }
        if (TF::TPUReplicatedInputOp input =
                llvm::dyn_cast<TF::TPUReplicatedInputOp>(user)) {
          if (!input.getIsPacked()) {
            input.emitOpError() << "unexpected variable input, not packed";
            return LogicalResult::failure();
          }

          if (is_variable) {
            input.emitOpError() << "unexpected multiple TPUReplicatedInputOp "

      TF_AddGradientsWithPrefix(graph_, prefix, outputs, noutputs, inputs,
                                ninputs, grad_inputs, s_, grad_outputs);
    } else {
      TF_AddGradientsWithPrefix(graph_, prefix, outputs, noutputs, inputs,
                                ninputs, nullptr, s_, grad_outputs);
    }
  }

  void BuildErrorGraph(TF_Output* inputs, TF_Output* outputs) {

  } else if (sparse_index_vector.type ==
             tflite::SparseIndexVector_Uint16Vector) {
    const auto& inputs = sparse_index_vector.AsUint16Vector()->values;
    std::vector<int32_t> outputs(inputs.size());
    std::transform(inputs.begin(), inputs.end(), outputs.begin(),
                   [](auto x) { return static_cast<int32_t>(x); });
    return outputs;
  } else if (sparse_index_vector.type ==