Function device

devices/cpu/cpu_engine.cpp:20–52

#include "cpu_image_copy.h"

OIDN_NAMESPACE_BEGIN

  CPUEngine::CPUEngine(CPUDevice* device, int numThreads)
    : device(device)
  {
    // Get the thread affinities for one thread per core on non-hybrid CPUs with SMT
  #if !(defined(__APPLE__) && defined(OIDN_ARCH_ARM64))
    if (device->setAffinity
      #if TBB_INTERFACE_VERSION >= 12020 // oneTBB 2021.2 or later
        && tbb::info::core_types().size() <= 1 // non-hybrid cores
      #endif
       )
    {
      affinity = std::make_shared<ThreadAffinity>(1, device->verbose);
      if (affinity->getNumThreads() == 0 ||                                           // detection failed
          tbb::this_task_arena::max_concurrency() == affinity->getNumThreads() ||     // no SMT
          (tbb::this_task_arena::max_concurrency() % affinity->getNumThreads()) != 0) // hybrid SMT
        affinity.reset(); // disable affinitization
    }
  #endif

    // Create the task arena
    const int maxNumThreads = affinity ? affinity->getNumThreads() : tbb::this_task_arena::max_concurrency();
    numThreads = (numThreads > 0) ? min(numThreads, maxNumThreads) : maxNumThreads;
    arena = std::make_shared<tbb::task_arena>(numThreads);

    // Automatically set the thread affinities
    if (affinity)
      observer = std::make_shared<PinningObserver>(affinity, *arena);

    // Start the queue processing thread
    queueThread = std::thread([&]() { processQueue(); });
  }

  CPUEngine::~CPUEngine()
  {