Commit f8d864ea authored by Claudio Gheller's avatar Claudio Gheller
Browse files

added some comments

parent 4601caf6

Makefile

+3 −3
Original line number Diff line number Diff line
@@ -42,10 +42,10 @@ FFTWLIBS = 
OPT += -DUSE_FFTW



# use omp-ized version of fftw routines

#OPT += -DHYBRID_FFTW

OPT += -DHYBRID_FFTW



# switch on the OpenMP parallelization

#OPT += -DUSE_OMP

OPT += -DUSE_OMP



# ========================================================
@@ -92,7 +92,7 @@ OPT += -DGAUSS_HI_PRECISION 
#OPT += -DCUFFTMP



# FULL NVIDIA GPU SUPPORT - Recommended for full NVIDIA GPU code execution

OPT += -DFULL_NVIDIA

#OPT += -DFULL_NVIDIA

ifeq (FULL_NVIDIA,$(findstring FULL_NVIDIA,$(OPT)))

OPT += -DCUDACC -DNCCL_REDUCE -DCUFFTMP

endif

cd

deleted100644 → 0
+0 −0

Empty file deleted.

gridding_cpu.c

+8 −7
Original line number Diff line number Diff line
@@ -67,12 +67,12 @@ void gridding_data() 
      

    }   // closes reduce_method == REDUCE_RING



  

  timing_wt.kernel     = 0.0;

  timing_wt.reduce     = 0.0;

  timing_wt.reduce_mpi = 0.0;

  timing_wt.reduce_sh  = 0.0;

  timing_wt.compose    = 0.0;

  //CLAAAA

  //timing_wt.kernel     = 0.0;

  //timing_wt.reduce     = 0.0;

  //timing_wt.reduce_mpi = 0.0;

  //timing_wt.reduce_sh  = 0.0;

  //timing_wt.compose    = 0.0;



  // calculate the resolution in radians

  resolution = 1.0/MAX(fabs(metaData.uvmin),fabs(metaData.uvmax));
@@ -142,6 +142,7 @@ void gridding_data() 
      double uumax = -1e20;

      double vvmax = -1e20;



     /*

     #pragma omp parallel reduction( min: uumin, vvmin) reduction( max: uumax, vvmax) num_threads(param.num_threads)

      {

	double my_uumin = 1e20;
@@ -165,7 +166,7 @@ void gridding_data() 
      }



      //printf("UU, VV, min, max = %f %f %f %f\n", uumin, uumax, vvmin, vvmax);

      

      */



      timing_wt.compose += CPU_TIME_wt - start;

gridding_nccl.cu

+6 −5
Original line number Diff line number Diff line
@@ -52,11 +52,12 @@ void gridding_data(){ 


  double shift = (double)(dx*yaxis);



  timing_wt.kernel     = 0.0;

  timing_wt.reduce     = 0.0;

  timing_wt.reduce_mpi = 0.0;

  timing_wt.reduce_sh  = 0.0;

  timing_wt.compose    = 0.0;

  // CLAAAA

  //timing_wt.kernel     = 0.0;

  //timing_wt.reduce     = 0.0;

  //timing_wt.reduce_mpi = 0.0;

  //timing_wt.reduce_sh  = 0.0;

  //timing_wt.compose    = 0.0;

  

  // calculate the resolution in radians

  resolution = 1.0/MAX(fabs(metaData.uvmin),fabs(metaData.uvmax));

main.c

+7 −0
Original line number Diff line number Diff line
@@ -33,6 +33,13 @@ int main(int argc, char * argv[]) 
{





//CLAAAA

  timing_wt.kernel     = 0.0;

  timing_wt.reduce     = 0.0;

  timing_wt.reduce_mpi = 0.0;

  timing_wt.reduce_sh  = 0.0;

  timing_wt.compose    = 0.0;



  if(argc > 1)

    {

      strcpy(in.paramfile, argv[1]);