Line data    Source code

       1              : // SPDX-License-Identifier: Apache-2.0
       2              : /**
       3              :  * Copyright (C) 2025 Sungsik Kong <ss.kong@samsung.com>
       4              :  *
       5              :  * @file   bs_threadpool_manager.cpp
       6              :  * @date   20 May 2025
       7              :  * @see    https://github.com/nnstreamer/nntrainer
       8              :  * @author Sungsik Kong <ss.kong@samsung.com>
       9              :  * @bug    No known bugs except for NYI items
      10              :  * @brief  BS threadpool manager class source file
      11              :  */
      12              : 
      13              : #ifndef THREAD_POOL_MANAGER_CPP
      14              : #define THREAD_POOL_MANAGER_CPP
      15              : 
      16              : #include "bs_thread_pool_manager.hpp"
      17              : #include <algorithm>
      18              : #include <cmath>
      19              : namespace nntrainer {
      20              : /**
      21              :  * @brief Instantiate thread pool with the number of hardware concurrency.
      22              :  *
      23              :  * @return BS::thread_pool<>
      24              :  */
      25              : 
      26            0 : std::size_t ThreadPoolManager::select_k_quant_thread_count(unsigned int M,
      27              :                                                            unsigned int N,
      28              :                                                            unsigned int K) {
      29            0 :   const std::size_t max_threads = std::thread::hardware_concurrency();
      30              : 
      31            0 :   const std::size_t work_size = static_cast<std::size_t>(M * N * K);
      32              :   std::size_t est_threads;
      33              : 
      34              :   //  Use log-scale thresholds to reduce threads on smaller work sizes
      35            0 :   if (work_size < 1536 * 1536)
      36              :     est_threads = 1;
      37            0 :   if (work_size < 1536 * 2048)
      38            0 :     est_threads = 2;
      39            0 :   if (work_size < 2048 * 2048)
      40            0 :     est_threads = 4;
      41              :   else {
      42            0 :     est_threads =
      43            0 :       static_cast<std::size_t>(std::log2(work_size / (1536 * 1536))) + 4;
      44              :   }
      45            0 :   return std::min(est_threads, max_threads);
      46              : }
      47              : 
      48              : } // namespace nntrainer
      49              : 
      50              : #endif // THREAD_POOL_MANAGER_CPP