/// quantized and the operand type is quantizable.

LogicalResult QuantizedConstRewrite::matchAndRewrite(
    QuantizeCastOp qbarrier, PatternRewriter &rewriter) const {
  Attribute value;

  // Is the operand a constant?
  if (!matchPattern(qbarrier.getArg(), m_Constant(&value))) {
    return failure();
  }

  // Does the qbarrier convert to a quantized type. This will not be true

  bool SetConstantResultParams(Operation* op);

  // Inserts the Quantize and Dequantize ops after `op`'s `index`-th result. The
  // quantized element type for the result is `quantized_type`.
  void QuantizeOpResult(Operation* op, int result_index,
                        QuantizedType quantized_type);

  // Inserts the Quantize and Dequantize ops after `arg`. The quantized element
  // type for `arg` is `quantized_type`.

// (e.g. matmul) has both quantized and unquantized inputs by dequantizing
// the quantized inputs, performing the operation in the expressed type, then
// requantizing if a quantized output is required.
//
// The motivation behind these changes is for Dialects that assume only float
// or quantized computation, and do not support a mixture of these types on

      llvm::SmallVector<Value, 4> quantized;
      for (auto user : returned.getUsers()) {
        if (auto q = Quantized(user)) {
          quantized.push_back(q);
        }
      }
      if (quantized.size() == 1) {
        ret.setOperand(i, quantized.front());
      }
      i++;
    }
  });

  // Check for  (Quant (Dequant $in), $qA) "qdq" pairs that couldn't be

    std::optional<const absl::string_view> mlir_dump_file_prefix =
        std::nullopt);

// Converts dequantize-(quantizable) call-quantize pattern to a single call op
// that has quantized input and output types. It is expected for this pass to
// emit illegal IR with unsupported quantized input and output types. The
// pass following immediately after this one will be responsible for legalizing

// and NumericVerify ops to compare output values from the quantized and float
// ops.
//
// When `legacy_float_scale` is true, the quantizer will use float scale instead
// of double, and call TOCO's quantization routines to maintain bit-exactness of
// the values with the TOCO quantizer.
TfLiteStatus QuantizeModel(
    absl::string_view model_buffer, const tflite::TensorType &input_type,

    // This op has been quantized, so we should not consider it again.
    if (quantized_.contains(op)) continue;
    quantized_.insert(op);

    if (auto constant_op = dyn_cast<arith::ConstantOp>(op); constant_op) {
      // If the workflow requires inferring ranges from the content
      // (post-training quantization) and it is weight (filter) and hasn't

    // For consistency, we require all quantized composite function to have
    // the "tf_quant.quantized_ops" attribute.
    if (!new_func.getSymName().starts_with("quantized_")) continue;
    if (!new_func->hasAttrOfType<ArrayAttr>("tf_quant.quantized_ops")) {
      new_func->emitError() << "Missing \"tf_quant.quantized_ops\" "
                               "attribute in the quantized composite function.";
      signalPassFailure();

                         << " is expected to be quantized with "
                         << tensor_property.number_of_bits << " bits, but got "
                         << num_storage_bits << " bits instead.";
          return failure();
        }
        continue;  // skip if it is already quantized.
      }
      quant::UniformQuantizedType qtype;

    return %0 : tensor<1x3xf32>
  }
// Checks that the entry function is quantized for dot_general. Quantized
// dot_general outputs an i32 quantized tensor, followed by requantization to
// i8 quantized tensor.