Commit a42e9056 authored by Giovanni Lacopo's avatar Giovanni Lacopo
Browse files

AMD GPU bug fix

parent 404e0296

gridding_rccl.cpp

+9 −9
Original line number Diff line number Diff line 
#include "allvars_rccl.h"

#include "proto.h"

#include <hip/hip_runtime.h>

#include <rccl/rccl.h>

#include <rccl.h>



/* 

 * Implements the gridding of data via GPU
@@ -104,18 +104,18 @@ void gridding_data(){ 
      uint Nsec            = histo_send[isector];

      uint Nweightss       = Nsec*metaData.polarisations;

      uint Nvissec         = Nweightss*metaData.freq_per_chan;

      double_t *memory     = (double*) malloc ( (Nsec*3)*sizeof(double_t) +

						(Nvissec*2+Nweightss)*sizeof(float_t) );

      double_ty *memory     = (double*) malloc ( (Nsec*3)*sizeof(double_ty) +

						(Nvissec*2+Nweightss)*sizeof(float_ty) );



      if ( memory == NULL )

	shutdown_wstacking(NOT_ENOUGH_MEM_STACKING, "Not enough memory for stacking", __FILE__, __LINE__);

  

      double_t *uus        = (double_t*) memory;

      double_t *vvs        = (double_t*) uus+Nsec;

      double_t *wws        = (double_t*) vvs+Nsec;

      float_t  *weightss   = (float_t*)((double_t*)wws+Nsec);

      float_t  *visreals   = (float_t*)weightss + Nweightss;

      float_t  *visimgs    = (float_t*)visreals + Nvissec;

      double_ty *uus        = (double_ty*) memory;

      double_ty *vvs        = (double_ty*) uus+Nsec;

      double_ty *wws        = (double_ty*) vvs+Nsec;

      float_ty  *weightss   = (float_ty*)((double_t*)wws+Nsec);

      float_ty  *visreals   = (float_ty*)weightss + Nweightss;

      float_ty  *visimgs    = (float_ty*)visreals + Nvissec;