Commit 57ac3b1a authored by Giovanni La Mura's avatar Giovanni La Mura
Browse files

Make TFRFME::vec_wsum directly accessible

parent 0becb9e3

src/include/tfrfme.h

+1 −3
Original line number Diff line number Diff line
@@ -342,8 +342,6 @@ protected: 
  double *yv;

  //! Vector of computed z positions

  double *zv;

  //! QUESTION: definition?

  dcomplex *vec_wsum;



  /*! \brief Load a configuration instance from a HDF5 binary file.

   *
@@ -407,7 +405,7 @@ public: 
  //! QUESTION: definition?

  const double& exril = _exril;

  //! QUESTION: definition?

  dcomplex **wsum;

  dcomplex *vec_wsum;

  

  /*! \brief Trapping configuration instance constructor.

   *

src/libnptm/tfrfme.cpp

+8 −10
Original line number Diff line number Diff line
@@ -584,8 +584,6 @@ TFRFME::TFRFME(int lmode, int lm, int nkv, int nxv, int nyv, int nzv) { 
  _nlmmt = _lm * (_lm + 2) * 2;

  _nrvc = _nxv * _nyv * _nzv;

  vec_wsum = new dcomplex[nrvc * nlmmt]();

  wsum = new dcomplex*[nlmmt];

  for (int wi = 0; wi < nlmmt; wi++) wsum[wi] = vec_wsum + wi * nrvc;

}



TFRFME::~TFRFME() {
@@ -593,7 +591,6 @@ TFRFME::~TFRFME() { 
  delete[] yv;

  delete[] zv;

  delete[] vec_wsum;

  delete[] wsum;

}



TFRFME* TFRFME::from_binary(const std::string& file_name, const std::string& mode) {
@@ -660,7 +657,7 @@ TFRFME* TFRFME::from_hdf5(const std::string& file_name) { 
    for (int wj = 0; wj < nrvc; wj++) {

      for (int wi = 0; wi < nlmmt; wi++) {

	value = elements[index] + elements[index + 1] * I;

	instance->wsum[wi][wj] = value;

	instance->vec_wsum[nrvc * wi + wj] = value;

	index += 2;

      } // wi loop

    } // wj loop
@@ -725,7 +722,7 @@ TFRFME* TFRFME::from_legacy(const std::string& file_name) { 
	input.read(reinterpret_cast<char *>(&rval), sizeof(double));

	input.read(reinterpret_cast<char *>(&ival), sizeof(double));

	dcomplex value = rval + ival * I;

	instance->wsum[wi][wj] = value;

	instance->vec_wsum[nrvc * wi + wj] = value;

      } // wi loop

    } // wj loop

    input.close();
@@ -840,8 +837,8 @@ void TFRFME::write_hdf5(const std::string& file_name) { 
  int index = 0;

  for (int wj = 0; wj < nrvc; wj++) {

    for (int wi = 0; wi < nlmmt; wi++) {

      ptr_elements[index++] = real(wsum[wi][wj]);

      ptr_elements[index++] = imag(wsum[wi][wj]);

      ptr_elements[index++] = real(vec_wsum[nrvc * wi + wj]);

      ptr_elements[index++] = imag(vec_wsum[nrvc * wi + wj]);

    } // wi loop

  } // wj loop

  rec_ptr_list.append(ptr_elements);
@@ -889,8 +886,8 @@ void TFRFME::write_legacy(const std::string& file_name) { 
      output.write(reinterpret_cast<char *>(&(zv[zi])), sizeof(double));

    for (int wj = 0; wj < _nrvc; wj++) {

      for (int wi = 0; wi < _nlmmt; wi++) {

	double rval = real(wsum[wi][wj]);

	double ival = imag(wsum[wi][wj]);

	double rval = real(vec_wsum[nrvc * wi + wj]);

	double ival = imag(vec_wsum[nrvc * wi + wj]);

	output.write(reinterpret_cast<char *>(&rval), sizeof(double));

	output.write(reinterpret_cast<char *>(&ival), sizeof(double));

      } // wi loop
@@ -960,8 +957,9 @@ bool TFRFME::operator ==(const TFRFME& other) { 
    }

  }

  for (int wi = 0; wi < _nlmmt; wi++) {

    int i = _nrvc * wi;

    for (int wj = 0; wj < _nrvc; wj++) {

      if (wsum[wi][wj] != other.wsum[wi][wj]) {

      if (vec_wsum[i + wj] != other.vec_wsum[i + wj]) {

	return false;

      }

    } // wj loop

src/trapping/cfrfme.cpp

+9 −2
Original line number Diff line number Diff line
@@ -66,6 +66,8 @@ 
#endif



#ifdef USE_TARGET_OFFLOAD

#include <cstdlib>



/*! \brief Specialized function to perform GPU-offloaded trapping loop.

 *

 * The offload of GPU operations through interface layers, such as OpenMP,
@@ -461,9 +463,13 @@ void frfme(string data_file, string output_path) { 
	  int size_global_vec_w = nkvs * (jlml - jlmf + 1);

	  int size_vec_tt1_wk = nkvs * nlmmt;

	  const dcomplex *vec_tt1_wk = tt1->wk;

	  dcomplex *vec_wsum = tfrfme->wsum[0];

	  dcomplex *vec_wsum = tfrfme->vec_wsum;

	  double *vec_vkzm = tt2->vec_vkzm;

#ifdef USE_TARGET_OFFLOAD

	  dcomplex *global_vec_w = (dcomplex *)aligned_alloc(64, size_global_vec_w * sizeof(dcomplex));

#else

	  dcomplex *global_vec_w = new dcomplex[size_global_vec_w]();

#endif // USE_TARGET_OFFLOAD

	  message = "INFO: looping over " + to_string(jlml - jlmf + 1) + " J iterations.\n";

	  logger.log(message);

#ifdef USE_TARGET_OFFLOAD
@@ -490,6 +496,7 @@ void frfme(string data_file, string output_path) { 
	  sprintf(buffer, "INFO: loop calculation took %lfs.\n", elapsed.count());

	  message = string(buffer);

	  logger.log(message);

	  free(global_vec_w);

#else

#pragma omp parallel for

	  for (int j80 = jlmf - 1; j80 < jlml; j80++) {
@@ -550,8 +557,8 @@ void frfme(string data_file, string output_path) { 
	      vec_wsum[(j80 * nrvc) + ixyz] = sumy * delks;

	    } // ixyz loop

	  } // j80 loop

#endif // USE_TARGET_OFFLOAD

	  delete[] global_vec_w;

#endif // USE_TARGET_OFFLOAD

#ifdef USE_NVTX

	  nvtxRangePop();

#endif

src/trapping/clffft.cpp

+2 −1
Original line number Diff line number Diff line
@@ -303,6 +303,7 @@ void lffft(string data_file, string output_path) { 
	  // label 160

	  const int nlmm = lm * (lm + 2);

	  const int nlmmt = nlmm + nlmm;

	  const int nrvc = nxv * nyv * nzv;

	  ws = new dcomplex[nlmmt]();

	  if (lm > le) wsl = new dcomplex[nlmmt]();

	  // FORTRAN writes two output formatted files without opening them
@@ -320,7 +321,7 @@ void lffft(string data_file, string output_path) { 
		  //binary_input.read(reinterpret_cast<char *>(&vimag), sizeof(double));

		  int row = i;

		  int col = (nyv * nxv * iz475) + (nxv * iy475) + ix475;

		  dcomplex value = tfrfme->wsum[row][col];

		  dcomplex value = tfrfme->vec_wsum[nrvc * row + col];

		  if (lm <= le) {

		    ws[i] = value;

		  } else { // label 170