/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck_tile/host/reference/reference_batched_dropout.hpp Source File

/home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck_tile/host/reference/reference_batched_dropout.hpp Source File#

Composable Kernel: /home/docs/checkouts/readthedocs.org/user_builds/advanced-micro-devices-composable-kernel/checkouts/develop/include/ck_tile/host/reference/reference_batched_dropout.hpp Source File

Go to the documentation of this file.

 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
  
 #pragma once
  
 #include "ck_tile/core.hpp"
 #include "ck_tile/host/host_tensor.hpp"
 #include <thread>
  
 namespace ck_tile {
  
 template <typename DataType, typename RandValOutputDataType>
 CK_TILE_HOST void reference_batched_dropout(HostTensor<DataType>& in_out_b_m_n,
                                             const HostTensor<RandValOutputDataType>& randval_b_m_n,
                                             const uint8_t& p_undrop_in_uint8_t,
                                             const float scale)
 {
     const int N = in_out_b_m_n.mDesc.get_lengths()[2];
     auto f      = [&](auto batch, auto m) {
         for(int n = 0; n < N; ++n)
         {
             float tmp = ck_tile::type_convert<float>(in_out_b_m_n(batch, m, n)) * scale;
             in_out_b_m_n(batch, m, n) = randval_b_m_n(batch, m, n) <= p_undrop_in_uint8_t
                                                  ? ck_tile::type_convert<DataType>(tmp)
                                                  : DataType(0);
         }
     };
  
     make_ParallelTensorFunctor(
         f, randval_b_m_n.mDesc.get_lengths()[0], randval_b_m_n.mDesc.get_lengths()[1])(
         std::thread::hardware_concurrency());
 }
 } // namespace ck_tile