inline void copy( In const& a, Out& b , HDI const& , HDO const&
, cudaStream_t stream = 0)
{
using T = typename Out::value_type;
//TODO
CUDA_ERROR(cudaMemcpyAsync( (T*)b.data()
, a.data()
, a.size()* sizeof(T)
, copy_<HDI,HDO>::mode()
, stream
));
}
inline void transfer_htd( In & in, int blockid, Stream & stream ,std::size_t streamid
, std::size_t leftover , nt2::pinned_ & )
{
std::size_t sizeb = blocksize;
if(leftover !=0) sizeb = leftover ;
if( block_stream_htd[blockid] == false )
{
block_stream_htd[blockid] = true;
CUDA_ERROR(cudaMemcpyAsync( buffers.get_device(streamid)
, in.data()
, sizeb* sizeof(T)
, cudaMemcpyHostToDevice
, stream
));
cudaStreamSynchronize(stream);
}
}
