Commit 90e669b8 authored by Giovanni Lacopo's avatar Giovanni Lacopo
Browse files

Last modification

parent 783e2134

Makefile

+4 −5
Original line number Diff line number Diff line
@@ -19,8 +19,8 @@ endif 


LINKER=$(MPICC)



FFTW_MPI_INC = -I/home/giacopo/Library_fftw/include

FFTW_MPI_LIB = -L/home/giacopo/Library_fftw/lib

FFTW_MPI_INC = 

FFTW_MPI_LIB = 



CFLAGS += -I./

FFTWLIBS =
@@ -51,7 +51,7 @@ OPT += -DPHASE_ON 
#OPT += -DNVIDIA



#use cuda for GPUs

#OPT += -D__CUDACC__

#OPT += -DCUDACC



# use GPU acceleration via OMP

#OPT += -DACCOMP
@@ -153,7 +153,6 @@ endif 
w-stacking: $(COBJ) $(DEPS) Makefile

	@$(LINKER) $(FLAGS) $(OPT) $(FFTWLIBS) $(LIBS) -lmpi -o $(EXEC)$(EXEC_EXT)





$(COBJ): $(DEPS) Makefile



%.o: %.c $(DEPS)

fourier_transform.c

+204 −217
Original line number Diff line number Diff line
@@ -7,8 +7,9 @@ void fftw_data(){ 
 #ifdef USE_FFTW

  // FFT transform the data (using distributed FFTW)

  if(rank == 0)printf("PERFORMING FFT\n");

        clock_gettime(CLOCK_MONOTONIC, &begin);

        start = clock();



  double start = CPU_TIME_wt;

	

  fftw_plan plan;

  fftw_complex *fftwgrid;

  ptrdiff_t alloc_local, local_n0, local_0_start;
@@ -16,7 +17,7 @@ void fftw_data(){ 


  //Use the hybrid MPI-OpenMP FFTW

 #ifdef HYBRID_FFTW

	if (threads_ok) fftw_plan_with_nthreads(param.num_threads);

  fftw_plan_with_nthreads(param.num_threads);

 #endif

  // map the 1D array of complex visibilities to a 2D array required by FFTW (complex[*][2])

  // x is the direction of contiguous data and maps to the second parameter
@@ -75,20 +76,10 @@ void fftw_data(){ 
  fftw_free(fftwgrid);



		

        #ifdef ONE_SIDE

        MPI_Win_fence(0,slabwin);

  MPI_Barrier(MPI_COMM_WORLD);

	#else

	MPI_Barrier(MPI_COMM_WORLD);

        #endif

        

        end = clock();

        clock_gettime(CLOCK_MONOTONIC, &finish);

	timing.fftw_time = ((double) (end - start)) / CLOCKS_PER_SEC;



	timing.fftw_time1 = (finish.tv_sec - begin.tv_sec);

        timing.fftw_time1 += (finish.tv_nsec - begin.tv_nsec) / 1000000000.0;

        clock_gettime(CLOCK_MONOTONIC, &begin);

  timing_wt.fftw += CPU_TIME_wt - start;

	

 #endif
@@ -98,7 +89,7 @@ void write_fftw_data(){ 


 #ifdef USE_FFTW

 #ifdef WRITE_DATA

        // Write results

  // Write results let's skip this part for the moment

 #ifdef USE_MPI

  MPI_Win writewin;

  MPI_Win_create(gridss, size_of_grid*sizeof(double), sizeof(double), MPI_INFO_NULL, MPI_COMM_WORLD, &writewin);
@@ -175,19 +166,15 @@ void write_fftw_data(){ 


  // Phase correction  



     clock_gettime(CLOCK_MONOTONIC, &begin);

     start = clock();

  double start = CPU_TIME_wt;

	       

  if(rank == 0)printf("PHASE CORRECTION\n");

  double* image_real = (double*) calloc(xaxis*yaxis,sizeof(double));

  double* image_imag = (double*) calloc(xaxis*yaxis,sizeof(double));



  phase_correction(gridss,image_real,image_imag,xaxis,yaxis,param.num_w_planes,param.grid_size_x,param.grid_size_y,resolution,metaData.wmin,metaData.wmax,param.num_threads,rank);



     end = clock();

     clock_gettime(CLOCK_MONOTONIC, &finish);

     timing.phase_time = ((double) (end - start)) / CLOCKS_PER_SEC;

     timing.phase_time1 = (finish.tv_sec - begin.tv_sec);

     timing.phase_time1 += (finish.tv_nsec - begin.tv_nsec) / 1000000000.0;

  timing_wt.phase += CPU_TIME_wt - start;

  

 #ifdef WRITE_IMAGE
@@ -198,15 +185,15 @@ void write_fftw_data(){ 
      fclose(file.pFilereal);

      fclose(file.pFileimg);

    }

        #ifdef USE_MPI



  MPI_Barrier(MPI_COMM_WORLD);

        #endif



  if(rank == 0)printf("WRITING IMAGE\n");

  for (int isector=0; isector<size; isector++)

    {

            #ifdef USE_MPI



      MPI_Barrier(MPI_COMM_WORLD);

            #endif

      

      if(isector == rank)

	{

	  printf("%d writing\n",isector);
@@ -224,9 +211,9 @@ void write_fftw_data(){ 
	  fclose(file.pFileimg);

	}

    }

        #ifdef USE_MPI

 

  MPI_Barrier(MPI_COMM_WORLD);

        #endif

 



 #endif //WRITE_IMAGE

gridding.c

+0 −1
Original line number Diff line number Diff line
@@ -8,7 +8,6 @@ 
void free_array       ( uint *, uint **, int );

void initialize_array ( void );

void gridding_data    ( void );

int  reduce_ring      ( int );

gridding_std.c

+1 −0
Original line number Diff line number Diff line
@@ -12,6 +12,7 @@ 


#if !defined( NCCL_REDUCE )



int reduce_ring (int);



//   .....................................................................

//

init.c

+3 −1
Original line number Diff line number Diff line
@@ -5,6 +5,8 @@ 
void init(int index)

{



  double start_tot;

  start_tot = CPU_TIME_wt;

  double begin = CPU_TIME_pr;

  

  // DAV: the corresponding KernelLen is calculated within the wstack function. It can be anyway hardcoded for optimization