include FFT in DKSOPAL and DKSBaseMuSR

auto-tuning/CMakeLists.txt

@@ -28,4 +28,7 @@ IF (ENABLE_OPAL)
   TARGET_LINK_LIBRARIES(testPushKick dks ${Boost_LIBRARIES} ${CLFFT_LIBRARIES})
 ENDIF(ENABLE_OPAL)
 
+ADD_EXECUTABLE(testFFT testFFT.cpp)
+TARGET_LINK_LIBRARIES(testFFT dks ${Boost_LIBRARIES} ${CLFFT_LIBRARIES})
+
 

auto-tuning/testFFT.cpp

@@ -0,0 +1,216 @@
+#include <iostream>
+#include <cstdlib>
+#include <complex>
+
+#include "Utility/TimeStamp.h"
+#include "DKSFFT.h"
+#include "DKSOPAL.h"
+#include "DKSBaseMuSR.h"
+
+using namespace std;
+
+void compareData(complex<double>* data1, complex<double>* data2, int N, int dim);
+void compareData(double* data1, double *data2, int N, int dim);
+
+void initData(complex<double> *data, int dimsize[3], int dim);
+void initData(double *data, int dimsize[3], int dim);
+
+bool readParams(int argc, char *argv[], int &N1, int &N2, int &N3, int &dim, 
+		char *api_name, char *device_name);
+
+void printHelp();
+
+int main(int argc, char *argv[]) {
+
+  int ierr;
+  int N1 = 8;
+  int N2 = 8;
+  int N3 = 8;
+  int dim = 3;
+  char *api_name = new char[10];
+  char *device_name = new char[10];
+
+  if ( readParams(argc, argv, N1, N2, N3, dim, api_name, device_name) )
+    return 0;
+
+  cout << "Use api: " << api_name << ", " << device_name << endl;
+
+  int dimsize[3] = {N1, N2, N3};
+  int sizereal = dimsize[0] * dimsize[1] * dimsize[2];
+  int sizecomp = (dimsize[0]/2+1) * dimsize[1] *dimsize[2];
+
+  double *rdata = new double[sizereal];
+  double *ordata = new double[sizereal];
+  complex<double> *cdata = new complex<double>[sizereal];
+  complex<double> *codata = new complex<double>[sizereal];
+
+  initData(rdata, dimsize, 3);
+  initData(cdata, dimsize, 3);
+
+  /* init DKSBase */
+  cout << "Init device and set function" << endl;
+  DKSBaseMuSR base;
+  base.setAPI(api_name, strlen(api_name));
+  base.setDevice(device_name, strlen(device_name));
+  cout << "init device" << endl;
+  base.initDevice();
+  cout << "setup fft" << endl;
+  base.setupFFT(dim, dimsize);
+
+  //Test RC FFT -> CR FFT
+  void *real_ptr, *comp_ptr, *res_ptr;
+  cout << "allocate memory" << endl;
+  real_ptr = base.allocateMemory<double>(sizereal, ierr);
+  res_ptr = base.allocateMemory<double>(sizereal, ierr);
+  comp_ptr = base.allocateMemory< complex<double> >(sizecomp, ierr);
+
+  cout << "write data" << endl;
+  base.writeData<double>(real_ptr, rdata, sizereal);
+
+  cout << "perform fft" << endl;
+  base.callR2CFFT(real_ptr, comp_ptr, dim, dimsize);
+  base.callC2RFFT(res_ptr, comp_ptr, dim, dimsize);
+  base.callNormalizeC2RFFT(res_ptr, dim, dimsize);
+
+  cout << "read data" << endl;
+  base.readData<double>(res_ptr, ordata, sizereal);
+
+  compareData(rdata, ordata, N1, 3);
+
+  base.freeMemory<double>(real_ptr, sizereal);
+  base.freeMemory<double>(res_ptr, sizereal);
+  base.freeMemory< complex<double> >(comp_ptr, sizecomp);
+
+  //Test CC FFT
+  void *mem_ptr;
+  mem_ptr = base.allocateMemory< complex<double> >(sizereal, ierr);
+  base.writeData< complex<double> >(mem_ptr, cdata, sizereal);
+  base.callFFT(mem_ptr, 3, dimsize);
+  base.callIFFT(mem_ptr, 3, dimsize);
+  base.callNormalizeFFT(mem_ptr, 3, dimsize);
+  base.readData< complex<double> >(mem_ptr, codata, sizereal);
+
+  compareData(cdata, codata, N1, 3);
+
+  base.freeMemory< complex<double> > (mem_ptr, sizereal);
+
+  delete[] rdata;
+  delete[] ordata;
+  delete[] cdata;
+  delete[] codata;
+
+}
+
+void compareData(complex<double>* data1, complex<double>* data2, int N, int dim) {
+  int ni, nj, nk, id;
+  ni = (dim > 2) ? N : 1;
+  nj = (dim > 1) ? N : 1;
+  nk = N;
+  double sum = 0;
+  for (int i = 0; i < ni; i++) {
+    for (int j = 0; j < nj; j++) {
+      for (int k = 0; k < nk; k++) {
+	id = i*ni*ni + j*nj + k;
+	sum += fabs(data1[id].real() - data2[id].real());
+	sum += fabs(data1[id].imag() - data2[id].imag());
+      }
+    }
+  }
+  cout << "Size " << N << " CC <--> CC diff: " << sum << endl;
+}
+
+void compareData(double* data1, double* data2, int N, int dim) {
+  int ni, nj, nk, id;
+  ni = (dim > 2) ? N : 1;
+  nj = (dim > 1) ? N : 1;
+  nk = N;
+  double sum = 0;
+  for (int i = 0; i < ni; i++) {
+    for (int j = 0; j < nj; j++) {
+      for (int k = 0; k < nk; k++) {
+	id = i*ni*ni + j*nj + k;
+	sum += fabs(data1[id] - data2[id]);
+      }
+    }
+  }
+  cout << "Size " << N << " RC <--> CR diff: " << sum << endl;
+}
+
+void initData(complex<double> *data, int dimsize[3], int dim) {
+  if (dim == 3) {
+    for (int i = 0; i < dimsize[2]; i++)
+      for (int j = 0; j < dimsize[1]; j++) 
+	for (int k = 0; k < dimsize[0]; k++) 
+	  data[i*dimsize[1]*dimsize[0] + j*dimsize[0] + k] = complex<double>(sin(k), 0.0);
+  } else if (dim == 2) {
+    for (int j = 0; j < dimsize[1]; j++) {
+      for (int k = 0; k < dimsize[0]; k++) {
+	data[j*dimsize[0] + k] = complex<double>(sin(k), 0.0);
+      }
+    }
+  } else {
+    for (int k = 0; k < dimsize[0]; k++) 
+      data[k] = complex<double>(sin(k), 0.0);
+  }
+}
+
+void initData(double *data, int dimsize[3], int dim) {
+  if (dim == 3) {
+    for (int i = 0; i < dimsize[2]; i++)
+      for (int j = 0; j < dimsize[1]; j++) 
+	for (int k = 0; k < dimsize[0]; k++) 
+	  data[i*dimsize[1]*dimsize[0] + j*dimsize[0] + k] = sin(k);
+  } else if (dim == 2) {
+    for (int j = 0; j < dimsize[1]; j++) {
+      for (int k = 0; k < dimsize[0]; k++) {
+	data[j*dimsize[0] + k] = sin(k);
+      }
+    }
+  } else {
+    for (int k = 0; k < dimsize[0]; k++) 
+      data[k] = sin(k);
+  }
+}
+
+bool readParams(int argc, char *argv[], int &N1, int &N2, int &N3, int &dim,
+		char *api_name, char *device_name) 
+{
+
+  for (int i = 1; i < argc; i++) {
+
+    if ( argv[i] == std::string("-dim")) {
+      dim = atoi(argv[i + 1]);
+      i++;
+    }
+
+    if ( argv[i] == std::string("-grid") ) {
+      N1 = atoi(argv[i + 1]);
+      N2 = atoi(argv[i + 2]);
+      N3 = atoi(argv[i + 3]);
+      i += 3;
+    }
+
+    if (argv[i] == string("-cuda")) {
+      strcpy(api_name, "Cuda");
+      strcpy(device_name, "-gpu");
+    } 
+
+    if (argv[i] == string("-opencl")) {
+      strcpy(api_name, "OpenCL");
+      strcpy(device_name, "-gpu");
+    } 
+
+    if (argv[i] == string("-mic")) {
+      strcpy(api_name, "OpenMP");
+      strcpy(device_name, "-mic");
+    } 
+
+    if (argv[i] == string("-cpu")) {
+      strcpy(api_name, "OpenCL");
+      strcpy(device_name, "-cpu");
+    }
+  }
+
+  return false;
+}
+

src/Algorithms/FFT.h

@@ -6,7 +6,7 @@
 
 #include "../DKSDefinitions.h"
 
-class DKSFFT {
+class FFT {
 
 protected:
   int defaultN[3];
@@ -22,7 +22,7 @@ protected:
 
 public:
 
-  virtual ~DKSFFT() { }
+  virtual ~FFT() { }
 
   virtual int setupFFT(int ndim, int N[3]) = 0;
   virtual int setupFFTRC(int ndim, int N[3], double scale = 1.0) = 0;

src/CMakeLists.txt

@@ -35,12 +35,12 @@ ENDMACRO ()
 SET (DKS_BASEDIR_HDRS
   DKSBase.h
   DKSDefinitions.h
-  DKSOPAL.h
+  DKSFFT.h
   )
 
 SET (DKS_BASEDIR_SRCS
   DKSBase.cpp
-  DKSOPAL.cpp
+  DKSFFT.cpp
   )
 
 #add opal to DKS if enable_opal is set

src/CUDA/CMakeLists.txt

@@ -1,9 +1,9 @@
-SET (_HDRS CudaBase.cuh)
+SET (_HDRS CudaBase.cuh CudaFFT.cuh)
-SET (_SRCS CudaBase.cu)
+SET (_SRCS CudaBase.cu CudaFFT.cu)
 
 IF (ENABLE_OPAL)
-  SET (_HDRS ${_HDRS} CudaFFT.cuh CudaGreensFunction.cuh CudaCollimatorPhysics.cuh)
+  SET (_HDRS ${_HDRS} CudaGreensFunction.cuh CudaCollimatorPhysics.cuh)
-  SET (_SRCS ${_SRCS} CudaFFT.cu CudaGreensFunction.cu CudaCollimatorPhysics.cu)
+  SET (_SRCS ${_SRCS} CudaGreensFunction.cu CudaCollimatorPhysics.cu)
 ENDIF (ENABLE_OPAL)
 
 IF (ENABLE_MUSR)

src/CUDA/CudaFFT.cuh

@@ -10,7 +10,7 @@
 #include "../Algorithms/FFT.h"
 #include "CudaBase.cuh"
 
-class CudaFFT : public DKSFFT{
+class CudaFFT : public FFT {
 
 private:
 

src/DKSBaseMuSR.h

@@ -8,6 +8,7 @@
 #include "AutoTuning/DKSAutoTuningTester.h"
 
 #include "DKSBase.h"
+#include "DKSFFT.h"
 
 #include "Algorithms/ChiSquareRuntime.h"
 
@@ -19,7 +20,7 @@
 #include "OpenCL/OpenCLChiSquareRuntime.h"
 #endif
 
-class DKSBaseMuSR : public DKSBase {
+class DKSBaseMuSR : public DKSFFT {
 
 private:
 

src/DKSFFT.cpp

@@ -4,7 +4,7 @@ DKSFFT::DKSFFT() {
   dksfft = nullptr;
 }
 
-~DKSFFT::DKSFFT() {
+DKSFFT::~DKSFFT() {
   delete dksfft;
 }
 
@@ -12,8 +12,10 @@ DKSFFT::DKSFFT() {
 int DKSFFT::setupFFT(int ndim, int N[3]) {
 
   if (apiCuda()) {
+    dksfft = CUDA_SAFEINIT( new CudaFFT(getCudaBase()) );
     return dksfft->setupFFT(ndim, N);
   } else if (apiOpenCL()) {
+    dksfft = OPENCL_SAFEINIT_AMD( new OpenCLFFT(getOpenCLBase()) );
     int ierr1 = dksfft->setupFFT(ndim, N);
     int ierr2 = dksfft->setupFFTRC(ndim, N);
     int ierr3 = dksfft->setupFFTCR(ndim, N);
@@ -24,6 +26,7 @@ int DKSFFT::setupFFT(int ndim, int N[3]) {
   } else if (apiOpenMP()) {
     //micbase.mic_setupFFT(ndim, N);
     //BENI: setting up RC and CR transformations on MIC
+    dksfft = MIC_SAFEINIT( new MICFFT(getMICBase()) );
     int ierr1 = dksfft->setupFFTRC(ndim, N, 1.);
     int ierr2 = dksfft->setupFFTCR(ndim, N, 1./(N[0]*N[1]*N[2]));
     if (ierr1 != DKS_SUCCESS)

src/DKSFFT.h

@@ -1,5 +1,5 @@
-#ifndef H_DKS_FFT
+#ifndef H_DKSBASE_FFT
-#define H_DKS_FFT
+#define H_DKSBASE_FFT
 
 #include <iostream>
 #include "AutoTuning/DKSAutoTuning.h"
@@ -18,7 +18,6 @@
 
 #ifdef DKS_CUDA
 #include "CUDA/CudaFFT.cuh"
-
 #endif
 
 #ifdef DKS_MIC
@@ -29,7 +28,7 @@ class DKSFFT : public DKSBase {
 
 private:
 
-  DKSFFT *dksfft;
+  FFT *dksfft;
 
   int initFFT();
   
@@ -105,3 +104,5 @@ public:
 
 
 };
+
+#endif

src/DKSOPAL.cpp

@@ -11,7 +11,6 @@ DKSOPAL::DKSOPAL(const char* api_name, const char* device_name) {
 }
 
 DKSOPAL::~DKSOPAL() {
-  delete dksfft;
   delete dkscol;
   delete dksgreens;
 }
@@ -21,17 +20,14 @@ int DKSOPAL::setupOPAL() {
   if (apiOpenCL()) {
     ierr = OPENCL_SAFECALL( DKS_SUCCESS );
     //TODO: only enable if AMD libraries are available
-    dksfft = OPENCL_SAFEINIT_AMD( new OpenCLFFT(getOpenCLBase()) );
     dkscol = OPENCL_SAFEINIT_AMD( new OpenCLCollimatorPhysics(getOpenCLBase()) );
     dksgreens = OPENCL_SAFEINIT_AMD( new OpenCLGreensFunction(getOpenCLBase()) );
   } else if (apiCuda()) {
     ierr = CUDA_SAFECALL( DKS_SUCCESS );
-    dksfft = CUDA_SAFEINIT( new CudaFFT(getCudaBase()) );
     dkscol = CUDA_SAFEINIT( new CudaCollimatorPhysics(getCudaBase()) );
     dksgreens = CUDA_SAFEINIT( new CudaGreensFunction(getCudaBase()) );
   } else if (apiOpenMP()) {
     ierr = MIC_SAFECALL( DKS_SUCCESS );
-    dksfft = MIC_SAFEINIT( new MICFFT(getMICBase()) );
     dkscol = MIC_SAFEINIT( new MICCollimatorPhysics(getMICBase()) );
     dksgreens = MIC_SAFEINIT( new MICGreensFunction(getMICBase()) );
   } else {

src/DKSOPAL.h

@@ -5,6 +5,7 @@
 #include "AutoTuning/DKSAutoTuning.h"
 
 #include "DKSBase.h"
+#include "DKSFFT.h"
 
 #include "DKSDefinitions.h"
 
@@ -32,11 +33,10 @@
 #include "MIC/MICCollimatorPhysics.h"
 #endif
 
-class DKSOPAL : public DKSBase {
+class DKSOPAL : public DKSFFT {
 
 private: 
 
-  DKSFFT *dksfft;
   DKSCollimatorPhysics *dkscol;
   GreensFunction *dksgreens;
 

src/MIC/CMakeLists.txt

@@ -1,11 +1,10 @@
-SET (_SRCS MICBase.cpp)
+SET (_SRCS MICBase.cpp MICFFT.cpp)
-SET (_HDRS MICBase.h)
+SET (_HDRS MICBase.h MICFFT.h)
 
 IF (ENABLE_OPAL)
   SET (_SRCS
     ${_SRCS}
     MICChiSquare.cpp
-    MICFFT.cpp
     MICGreensFunction.cpp  
     MICCollimatorPhysics.cpp
     )
@@ -13,7 +12,6 @@ IF (ENABLE_OPAL)
   SET (_HDRS
     ${_HDRS}
     MICChiSquare.h
-    MICFFT.h
     MICCollimatorPhysics.h
     MICGreensFunction.hpp    
     MICMergeSort.h

src/MIC/MICFFT.h

@@ -10,7 +10,7 @@
 #include "../Algorithms/FFT.h"
 #include "MICBase.h"
 
-class MICFFT : public DKSFFT {
+class MICFFT : public FFT {
 
 private:
 

src/OpenCL/CMakeLists.txt

@@ -4,6 +4,25 @@ SET (_HDRS OpenCLBase.h)
 SET (_SRCS OpenCLBase.cpp)
 SET (_KERNELS "")
 
+IF (ENABLE_AMD)
+  SET (_SRCS
+    ${_SRCS}
+    OpenCLFFT.cpp
+    )
+
+  SET (_HDRS
+    ${_HDRS}
+    OpenCLFFT.h
+    )
+
+  SET (_KERNELS
+    ${_KERNELS}
+    OpenCLKernels/OpenCLFFT.cl
+    OpenCLKernels/OpenCLFFTStockham.cl
+    OpenCLKernels/OpenCLTranspose.cl
+  )
+ENDIF (ENABLE_AMD)
+
 IF (ENABLE_MUSR)
   SET (_HDRS ${_HDRS} OpenCLChiSquareRuntime.h)
   SET (_SRCS ${_SRCS} OpenCLChiSquareRuntime.cpp)
@@ -13,23 +32,18 @@ ENDIF (ENABLE_MUSR)
 IF (ENABLE_AMD AND ENABLE_OPAL)
   SET (_SRCS
     ${_SRCS}
-    OpenCLFFT.cpp
     OpenCLCollimatorPhysics.cpp
     OpenCLGreensFunction.cpp
     )
 
   SET (_HDRS
     ${_HDRS}
-    OpenCLFFT.h
     OpenCLCollimatorPhysics.h
     OpenCLGreensFunction.h
     )
 
   SET (_KERNELS
     ${_KERNELS}
-    OpenCLKernels/OpenCLFFT.cl
-    OpenCLKernels/OpenCLFFTStockham.cl
-    OpenCLKernels/OpenCLTranspose.cl
     OpenCLKernels/OpenCLCollimatorPhysics.cl
     OpenCLKernels/OpenCLGreensFunction.cl
   )

src/OpenCL/OpenCLFFT.h

@@ -22,7 +22,7 @@
 
 #include "clFFT.h"
 
-class OpenCLFFT : public DKSFFT {
+class OpenCLFFT : public FFT {
 
 private:
 
@@ -112,10 +112,9 @@ public:
 		   int streamId = -1);
   int executeCRFFT(void * real_ptr, void * comp_ptr, int ndim, int N[3], 
 		   int streamId = -1);
-  int normalizeCRFFT(void *real_ptr, int ndim, int N[3], int streamId = -1)
+  int normalizeCRFFT(void *real_ptr, int ndim, int N[3], int streamId = -1) {
-    {
+    return DKS_ERROR;
-      return DKS_ERROR;
+  }
-    }
 
   //void printData3DN4(cl_double2* &data, int N);