GpuSurfDetectorInternal::GpuSurfDetectorInternal(GpuSurfConfiguration config) :
m_initialized(false),
m_config(config)
{
int deviceCount;
int device;
cudaError_t err;
cudaGetDeviceCount(&deviceCount);
ASRL_ASSERT_GT(deviceCount,0,"There are no CUDA capable devices present");
err = cudaGetDevice(&device);
ASRL_ASSERT_EQ(err,cudaSuccess, "Unable to get the CUDA device: " << cudaGetErrorString(err));
//std::cout << "Found device " << device << std::endl;
err = cudaGetDeviceProperties(&m_deviceProp,device);
ASRL_ASSERT_EQ(err,cudaSuccess, "Unable to get the CUDA device properties: " << cudaGetErrorString(err));
// Some more checking...
ASRL_ASSERT_GE(m_deviceProp.major,1,"Minimum compute capability 1.1 is necessary");
ASRL_ASSERT_GE(m_deviceProp.minor,1,"Minimum compute capability 1.1 is necessary");
m_maxmin.init(ASRL_SURF_MAX_CANDIDATES,false);
m_maxmin.memset(0);
}
void CudaSynchronizedMemory<T>::pullFromDeviceAsync(cudaStream_t stream, size_t nElements)
{
ASRL_ASSERT_GT(m_size,0, "The array is empty");
ASRL_ASSERT(m_pageLocked, "Asynchronous transfer is only valid for page-locked host memory");
if(nElements > m_size)
nElements = m_size;
cudaError_t err = (cudaMemcpyAsync((void*) m_host, (void *)m_device, nElements*sizeof(T), cudaMemcpyDeviceToHost, stream));
ASRL_ASSERT_EQ(err,cudaSuccess, "Unable to copy " << typeid(T).name() << " array of size " << m_size << " from device. Stream " << stream << ": (" << err << "): " << cudaGetErrorString(err));
}
