include/skepu/scan.h
Go to the documentation of this file.
00001 00005 #ifndef SCAN_H 00006 #define SCAN_H 00007 00008 #ifdef SKEPU_OPENCL 00009 #include <string> 00010 #include <vector> 00011 #include <CL/cl.h> 00012 #include "src/device_cl.h" 00013 #endif 00014 00015 #include "src/environment.h" 00016 #include "skepu/vector.h" 00017 #include "src/operator_macros.h" 00018 #include "src/exec_plan.h" 00019 00020 namespace skepu 00021 { 00022 00026 enum ScanType 00027 { 00028 INCLUSIVE, 00029 EXCLUSIVE 00030 }; 00031 00060 template <typename ScanFunc> 00061 class Scan 00062 { 00063 00064 public: 00065 00066 Scan(ScanFunc* scanFunc); 00067 00068 ~Scan(); 00069 00070 void finishAll() {m_environment->finishAll();} 00071 void setExecPlan(ExecPlan& plan) {m_execPlan = plan;} 00072 00073 private: 00074 Environment<int>* m_environment; 00075 ScanFunc* m_scanFunc; 00076 ExecPlan m_execPlan; 00077 00078 public: 00079 template <typename T> 00080 void operator()(Vector<T>& input, ScanType type, T init = T()); 00081 00082 template <typename InputIterator> 00083 void operator()(InputIterator inputBegin, InputIterator inputEnd, ScanType type, typename InputIterator::value_type init = typename InputIterator::value_type()); 00084 00085 template <typename T> 00086 void operator()(Vector<T>& input, Vector<T>& output, ScanType type, T init = T()); 00087 00088 template <typename InputIterator, typename OutputIterator> 00089 void operator()(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type, typename InputIterator::value_type init = typename InputIterator::value_type()); 00090 00091 public: 00092 template <typename T> 00093 void CPU(Vector<T>& input, ScanType type = INCLUSIVE, T init = T()); 00094 00095 template <typename InputIterator> 00096 void CPU(InputIterator inputBegin, InputIterator inputEnd, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type()); 00097 00098 template <typename T> 00099 void CPU(Vector<T>& input, Vector<T>& output, ScanType type = INCLUSIVE, T init = T()); 00100 00101 template <typename InputIterator, typename OutputIterator> 00102 void CPU(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type()); 00103 00104 #ifdef SKEPU_OPENMP 00105 public: 00106 template <typename T> 00107 void OMP(Vector<T>& input, ScanType type = INCLUSIVE, T init = T()); 00108 00109 template <typename InputIterator> 00110 void OMP(InputIterator inputBegin, InputIterator inputEnd, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type()); 00111 00112 template <typename T> 00113 void OMP(Vector<T>& input, Vector<T>& output, ScanType type = INCLUSIVE, T init = T()); 00114 00115 template <typename InputIterator, typename OutputIterator> 00116 void OMP(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type()); 00117 #endif 00118 00119 #ifdef SKEPU_CUDA 00120 public: 00121 template <typename T> 00122 void CU(Vector<T>& input, ScanType type = INCLUSIVE, T init = T(), int useNumGPU = 1); 00123 00124 template <typename InputIterator> 00125 void CU(InputIterator inputBegin, InputIterator inputEnd, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type(), int useNumGPU = 1); 00126 00127 template <typename T> 00128 void CU(Vector<T>& input, Vector<T>& output, ScanType type = INCLUSIVE, T init = T(), int useNumGPU = 1); 00129 00130 template <typename InputIterator, typename OutputIterator> 00131 void CU(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type(), int useNumGPU = 1); 00132 00133 private: 00134 template <typename InputIterator, typename OutputIterator> 00135 void scanSingleThread_CU(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type, typename InputIterator::value_type init, int deviceID); 00136 00137 template <typename T> 00138 T scanLargeVectorRecursively_CU(DeviceMemPointer_CU<T>* input, DeviceMemPointer_CU<T>* output, std::vector<DeviceMemPointer_CU<T>*>& blockSums, unsigned int numElements, int level, ScanType type, T init, int deviceID); 00139 00140 #endif 00141 00142 #ifdef SKEPU_OPENCL 00143 public: 00144 template <typename T> 00145 void CL(Vector<T>& input, ScanType type = INCLUSIVE, T init = T(), int useNumGPU = 1); 00146 00147 template <typename InputIterator> 00148 void CL(InputIterator inputBegin, InputIterator inputEnd, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type(), int useNumGPU = 1); 00149 00150 template <typename T> 00151 void CL(Vector<T>& input, Vector<T>& output, ScanType type = INCLUSIVE, T init = T(), int useNumGPU = 1); 00152 00153 template <typename InputIterator, typename OutputIterator> 00154 void CL(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type(), int useNumGPU = 1); 00155 00156 private: 00157 template <typename InputIterator, typename OutputIterator> 00158 void scanSingle_CL(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type, typename InputIterator::value_type init, int deviceID); 00159 00160 template <typename InputIterator, typename OutputIterator> 00161 void scanNumDevices_CL(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type, typename InputIterator::value_type init, int numDevices); 00162 00163 template <typename T> 00164 T scanLargeVectorRecursively_CL(DeviceMemPointer_CL<T>* input, DeviceMemPointer_CL<T>* output, std::vector<DeviceMemPointer_CL<T>*>& blockSums, unsigned int numElements, int level, ScanType type, T init, int deviceID); 00165 00166 private: 00167 std::vector<std::pair<cl_kernel, Device_CL*> > m_scanKernels_CL; 00168 std::vector<std::pair<cl_kernel, Device_CL*> > m_scanUpdateKernels_CL; 00169 std::vector<std::pair<cl_kernel, Device_CL*> > m_scanAddKernels_CL; 00170 00171 void replaceText(std::string& text, std::string find, std::string replace); 00172 void createOpenCLProgram(); 00173 #endif 00174 00175 }; 00176 00177 } 00178 00179 #include "src/scan.inl" 00180 00181 #include "src/scan_cpu.inl" 00182 00183 #ifdef SKEPU_OPENMP 00184 #include "src/scan_omp.inl" 00185 #endif 00186 00187 #ifdef SKEPU_OPENCL 00188 #include "src/scan_cl.inl" 00189 #endif 00190 00191 #ifdef SKEPU_CUDA 00192 #include "src/scan_cu.inl" 00193 #endif 00194 00195 #endif
