WorkerPool::WorkerPool(int numWorkers)
: m_readyWorkers(), m_ready(), m_readyMutex()
{
for (int i = 0; i < numWorkers; ++i)
m_workers.push_back(WorkerPtr(new Worker(this)));
}
bool OctreeConstructor::compute()
{
const Global& global = IV::getGlobal();
level_t readLevel = 0;
#ifdef IV_USE_CUDA
if( global.useCuda() )
readLevel = global.getBrickLevel();
else
#endif
readLevel = global.getCubeLevel();
index_node_t idStart =
_cube <<
3 * ( readLevel - _constructorLevel );
index_node_t idFinish =
( ( _cube + 1 ) <<
3 * ( readLevel - _constructorLevel ) ) - 1;
index_node_t min =
_cube <<
3 * ( global.getOctreeLevel() - _constructorLevel );
index_node_t max =
( ( _cube + 1 ) <<
3 * ( global.getOctreeLevel() - _constructorLevel ) ) - 1;
// Create DataWarehouse
_dataWarehouse = DataWarehousePtr( new DataWarehouse( min, max ) );
// Create Cache
CacheAttrPtr cacheAttr( new iv::DataHandler::CacheAttr() );
CachePtr cache( new iv::DataHandler::Cache( ) );
if( ! cache )
return false;
class Data
{
public:
Data( DataWarehousePtr& dPtr,
CachePtr& cPtr,
CacheAttrPtr& cAttr,
OctreeConstructorStats& stats )
: _dPtr( dPtr )
, _cPtr( cPtr )
, _stats( stats )
{
wasFine = _dPtr->start() && _cPtr->init( cAttr );
}
~Data()
{
auto startC = std::chrono::high_resolution_clock::now();
_dPtr->wait();
auto endC = std::chrono::high_resolution_clock::now();
_stats.incSortingTime( std::chrono::duration_cast<
std::chrono::milliseconds>(
endC - startC ) );
_cPtr->stop();
}
bool wasFine;
private:
DataWarehousePtr& _dPtr;
CachePtr& _cPtr;
OctreeConstructorStats& _stats;
};
Data data( _dataWarehouse, cache, cacheAttr, _stats );
if( !data.wasFine )
return false;
bool wasFine = true;
uint32_t numCubes = 0;
for( index_node_t id = idStart; id <= idFinish; id++ )
{
// Check cube is outside of volume
vec3int32_t coordCubeStart = getMinBoxIndex2( id,
readLevel,
global.getnLevels() );
if( coordCubeStart.x() < cache->getRealDimension().x() &&
coordCubeStart.y() < cache->getRealDimension().y() &&
coordCubeStart.z() < cache->getRealDimension().z() )
{
numCubes++;
}
}
uint32_t eachThread = numCubes / global.getMaxNumThreads();
index_node_t id = idStart;
uint32_t worker = 0;
while( id <= idFinish )
{
#ifdef IV_USE_CUDA
if( global.useCuda() )
_workers.push_back( WorkerPtr( new WorkerCPU( _dataWarehouse,
cache,
readLevel,
_stats ) ) );
#endif
else
_workers.push_back( WorkerPtr( new WorkerCPU( _dataWarehouse,
cache,
readLevel,
_stats ) ) );
uint32_t cubes = 0;
while( cubes <= eachThread && id <= idFinish )
{
vec3int32_t coordCubeStart = getMinBoxIndex2( id,
readLevel,
global.getnLevels() );
if( coordCubeStart.x() < cache->getRealDimension().x() &&
coordCubeStart.y() < cache->getRealDimension().y() &&
coordCubeStart.z() < cache->getRealDimension().z() )
{
cubes++;
_workers[ worker ]->addCube( id );
}
id++;
}
worker++;
}
for( uint32_t w = 0; w < _workers.size() && wasFine; w++ )
wasFine &= _workers[w]->start();
if( !wasFine )
return false;
for( uint32_t w = 0; w < _workers.size() && wasFine; w++ )
_workers[w]->wait();
return true;
}