SkePU: include/skepu/scan.h Source File

SkePU 0.7
00001 
00005 #ifndef SCAN_H
00006 #define SCAN_H
00007 
00008 #ifdef SKEPU_OPENCL
00009 #include <string>
00010 #include <vector>
00011 #ifdef USE_MAC_OPENCL
00012 #include <OpenCL/opencl.h>
00013 #else
00014 #include <CL/cl.h>
00015 #endif
00016 #include "src/device_cl.h"
00017 #endif
00018 
00019 #include "src/environment.h"
00020 #include "skepu/vector.h"
00021 #include "src/operator_macros.h"
00022 #include "src/exec_plan.h"
00023 
00024 namespace skepu
00025 {
00026 
00030 enum ScanType
00031 {
00032     INCLUSIVE,
00033     EXCLUSIVE
00034 };
00035 
00062 template <typename ScanFunc>
00063 class Scan
00064 {
00065 
00066 public:
00067 
00068     Scan(ScanFunc* scanFunc);
00069 
00070     ~Scan();
00071 
00072     void finishAll() {m_environment->finishAll();}
00073     void setExecPlan(ExecPlan& plan) {m_execPlan = plan;}
00074 
00075 private:
00076     Environment<int>* m_environment;
00077     ScanFunc* m_scanFunc;
00078     ExecPlan m_execPlan;
00079 
00080 public:
00081     template <typename T>
00082     void operator()(Vector<T>& input, ScanType type, T init = T());
00083 
00084     template <typename InputIterator>
00085     void operator()(InputIterator inputBegin, InputIterator inputEnd, ScanType type, typename InputIterator::value_type init = typename InputIterator::value_type());
00086 
00087     template <typename T>
00088     void operator()(Vector<T>& input, Vector<T>& output, ScanType type, T init = T());
00089 
00090     template <typename InputIterator, typename OutputIterator>
00091     void operator()(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type, typename InputIterator::value_type init = typename InputIterator::value_type());
00092 
00093 public:
00094     template <typename T>
00095     void CPU(Vector<T>& input, ScanType type = INCLUSIVE, T init = T());
00096 
00097     template <typename InputIterator>
00098     void CPU(InputIterator inputBegin, InputIterator inputEnd, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type());
00099 
00100     template <typename T>
00101     void CPU(Vector<T>& input, Vector<T>& output, ScanType type = INCLUSIVE, T init = T());
00102 
00103     template <typename InputIterator, typename OutputIterator>
00104     void CPU(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type());
00105 
00106 #ifdef SKEPU_OPENMP
00107 public:
00108     template <typename T>
00109     void OMP(Vector<T>& input, ScanType type = INCLUSIVE, T init = T());
00110 
00111     template <typename InputIterator>
00112     void OMP(InputIterator inputBegin, InputIterator inputEnd, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type());
00113 
00114     template <typename T>
00115     void OMP(Vector<T>& input, Vector<T>& output, ScanType type = INCLUSIVE, T init = T());
00116 
00117     template <typename InputIterator, typename OutputIterator>
00118     void OMP(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type());
00119 #endif
00120 
00121 #ifdef SKEPU_CUDA
00122 public:
00123     template <typename T>
00124     void CU(Vector<T>& input, ScanType type = INCLUSIVE, T init = T(), int useNumGPU = 1);
00125 
00126     template <typename InputIterator>
00127     void CU(InputIterator inputBegin, InputIterator inputEnd, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type(), int useNumGPU = 1);
00128 
00129     template <typename T>
00130     void CU(Vector<T>& input, Vector<T>& output, ScanType type = INCLUSIVE, T init = T(), int useNumGPU = 1);
00131 
00132     template <typename InputIterator, typename OutputIterator>
00133     void CU(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type(), int useNumGPU = 1);
00134 
00135 private:
00136     template <typename InputIterator, typename OutputIterator>
00137     void scanSingleThread_CU(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type, typename InputIterator::value_type init, int deviceID);
00138 
00139     template <typename T>
00140     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);
00141 
00142 #endif
00143 
00144 #ifdef SKEPU_OPENCL
00145 public:
00146     template <typename T>
00147     void CL(Vector<T>& input, ScanType type = INCLUSIVE, T init = T(), int useNumGPU = 1);
00148 
00149     template <typename InputIterator>
00150     void CL(InputIterator inputBegin, InputIterator inputEnd, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type(), int useNumGPU = 1);
00151 
00152     template <typename T>
00153     void CL(Vector<T>& input, Vector<T>& output, ScanType type = INCLUSIVE, T init = T(), int useNumGPU = 1);
00154 
00155     template <typename InputIterator, typename OutputIterator>
00156     void CL(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type = INCLUSIVE, typename InputIterator::value_type init = typename InputIterator::value_type(), int useNumGPU = 1);
00157 
00158 private:
00159     template <typename InputIterator, typename OutputIterator>
00160     void scanSingle_CL(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type, typename InputIterator::value_type init, int deviceID);
00161 
00162     template <typename InputIterator, typename OutputIterator>
00163     void scanNumDevices_CL(InputIterator inputBegin, InputIterator inputEnd, OutputIterator outputBegin, ScanType type, typename InputIterator::value_type init, int numDevices);
00164 
00165     template <typename T>
00166     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);
00167 
00168 private:
00169     std::vector<std::pair<cl_kernel, Device_CL*> > m_scanKernels_CL;
00170     std::vector<std::pair<cl_kernel, Device_CL*> > m_scanUpdateKernels_CL;
00171     std::vector<std::pair<cl_kernel, Device_CL*> > m_scanAddKernels_CL;
00172 
00173     void replaceText(std::string& text, std::string find, std::string replace);
00174     void createOpenCLProgram();
00175 #endif
00176 
00177 };
00178 
00179 }
00180 
00181 #include "src/scan.inl"
00182 
00183 #include "src/scan_cpu.inl"
00184 
00185 #ifdef SKEPU_OPENMP
00186 #include "src/scan_omp.inl"
00187 #endif
00188 
00189 #ifdef SKEPU_OPENCL
00190 #include "src/scan_cl.inl"
00191 #endif
00192 
00193 #ifdef SKEPU_CUDA
00194 #include "src/scan_cu.inl"
00195 #endif
00196 
00197 #endif
00198