eic-opticks/sysrap/SCU_.h

0001 #pragma once
0002 /**
0003 SCU_.h
0004 =========
0005
0006 TODO : merge this with SCU.h
0007
0008 **/
0009
0010 #include <cuda_runtime.h>
0011 #include <string>
0012 #include <sstream>
0013
0014
0015 struct SCU_Exception : public std::runtime_error
0016 {
0017      SCU_Exception( const char* msg ) : std::runtime_error( msg ) {}
0018 };
0019
0020 #define SCU_CHECK( call )                                                     \
0021     do                                                                         \
0022     {                                                                          \
0023         cudaError_t error = call;                                              \
0024         if( error != cudaSuccess )                                             \
0025         {                                                                      \
0026             std::stringstream ss;                                              \
0027             ss << "CUDA call (" << #call << " ) failed with error: '"          \
0028                << cudaGetErrorString( error )                                  \
0029                << "' (" __FILE__ << ":" << __LINE__ << ")\n";                  \
0030             throw SCU_Exception( ss.str().c_str() );                        \
0031         }                                                                      \
0032     } while( 0 )
0033
0034
0035
0036 struct SCU_
0037 {
0038     using ULL = unsigned long long ;
0039     static void _cudaMalloc( void** p2p, size_t size, const char* label );
0040
0041     template <typename T>
0042     static T*   device_alloc( unsigned num_items, const char* label ) ;
0043
0044     template<typename T>
0045     static void copy_host_to_device( T* d, const T* h, unsigned num_items);
0046
0047 };
0048
0049
0050 inline void SCU_::_cudaMalloc( void** p2p, size_t size, const char* label )
0051 {
0052     cudaError_t err = cudaMalloc(p2p, size ) ;
0053     if( err != cudaSuccess )
0054     {
0055         std::stringstream ss;
0056         ss << "CUDA call (" << label << " ) failed with error: '"
0057            << cudaGetErrorString( err )
0058            << "' (" __FILE__ << ":" << __LINE__ << ")\n";
0059
0060         throw SCU_Exception( ss.str().c_str() );
0061     }
0062 }
0063
0064
0065 template<typename T>
0066 inline T* SCU_::device_alloc( unsigned num_items, const char* label )
0067 {
0068     size_t size = num_items*sizeof(T) ;
0069
0070     T* d ;
0071     _cudaMalloc( reinterpret_cast<void**>( &d ), size, label );
0072
0073     return d ;
0074 }
0075
0076
0077 template<typename T>
0078 inline void SCU_::copy_host_to_device( T* d, const T* h, unsigned num_items)
0079 {
0080     size_t size = num_items*sizeof(T) ;
0081     SCU_CHECK( cudaMemcpy(reinterpret_cast<void*>( d ), h , size, cudaMemcpyHostToDevice ));
0082 }
0083
0084