src/relay/qnn/op/subtract.cc

/*
 * Licensed to the Apache Software Foundation (ASF) under one
 * or more contributor license agreements.  See the NOTICE file
 * distributed with this work for additional information
 * regarding copyright ownership.  The ASF licenses this file
 * to you under the Apache License, Version 2.0 (the
 * "License"); you may not use this file except in compliance
 * with the License.  You may obtain a copy of the License at
 *
 *   http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing,
 * software distributed under the License is distributed on an
 * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
 * KIND, either express or implied.  See the License for the
 * specific language governing permissions and limitations
 * under the License.
 */

/*!
 * \file src/relay/qnn/op/subtract.cc
 * \brief QNN subtract operator.
 */
#include <tvm/relay/analysis.h>
#include <tvm/relay/op_attr_types.h>

#include "op_common.h"

namespace tvm {
namespace relay {
namespace qnn {

/*
 * \brief Canonicalizes the QNN subtract op.
 * \param attrs The empty attribute.
 * \param new_args The new mutated args to the call node.
 * \param arg_types The types of input and output.
 * \return The sequence of Relay ops for add op.
 */
Expr QnnSubtractCanonicalize(const Attrs& attrs, const Array<Expr>& new_args,
                             const Array<tvm::relay::Type>& arg_types) {
  // Get the args.
  QnnBinaryOpArguments args(new_args);

  // Get the input dtype and shape.
  QnnBinaryOpTensorType input_type(arg_types, 0);

  const auto* broadcast_attrs = attrs.as<BroadcastAttrs>();
  ICHECK(broadcast_attrs != nullptr);

  auto lhs_axis = broadcast_attrs->lhs_axis;
  auto rhs_axis = broadcast_attrs->rhs_axis;

  // TODO(shoubhik) - The lowering can be further optimized. Instead of inserting requantize in
  // the start, we can insert requantize at the end if both input tensors have same qnn params. In
  // that case, we can first subtract the tensors, add the zero point, and requantize at the end.
  // This can be done in future.

  // Since the input qnn params can be different than output qnn params, we first requantize the
  // input tensors to the output qnn params. Then we call relay.subtract on the requantized inputs.
  // This subtraction results in extra subtraction of the output zero point. We further add
  // the zero point. The whole process can be represented using following equations
  //
  //          scale_c * (Q_c - zp_c) = scale_a * (Q_a - zp_a) - scale_b * (Q_b - zp_b)
  //
  // After requantizing Q_a and Q_b, equation becomes,
  //          scale_c * (Q_c - zp_c) = scale_c * (Q_a' - zp_c) - scale_c * (Q_b' - zp_c)
  //          scale_c * (Q_c - zp_c) = scale_c * (Q_a' - Q_b')
  //
  // Comparing the LHS and RHS, it results in
  //          Q_c = Q_a' - Q_b' + zp_c
  // The subtract op is done in int32 precision.

  // Requantize LHS if necessary. Computes Q_a'
  auto requantized_lhs =
      RequantizeOrUpcast(args.lhs, args.lhs_scale, args.lhs_zero_point, args.output_scale,
                         args.output_zero_point, input_type.shape, lhs_axis);
  // Requantize RHS if necessary. Computes Q_b'
  auto requantized_rhs =
      RequantizeOrUpcast(args.rhs, args.rhs_scale, args.rhs_zero_point, args.output_scale,
                         args.output_zero_point, input_type.shape, rhs_axis);

  // Computes Q_a' - Q_b'
  auto output = Subtract(requantized_lhs, requantized_rhs);

  // Add zero point. Computes (Q_a' - Q_b') + zp_c
  auto zero_scalar = MakeConstantScalar(DataType::Int(32), 0);
  if (!IsEqualScalar(args.output_zero_point, zero_scalar)) {
    output = Add(output, args.output_zero_point);
  }

  // Go back to lower precision.
  return ConvertDtype(output, input_type.dtype);
}

// QNN Subtraction operator.
QNN_REGISTER_BINARY_OP("subtract")
    .describe("Elementwise subtract with broadcasting for quantized tensors.")
    .set_support_level(11)
    .set_attr<FTVMLegalize>("FTVMQnnCanonicalize", QnnSubtractCanonicalize);

}  // namespace qnn
}  // namespace relay
}  // namespace tvm