Fix matrix binary ops for large case where int4 isn't enough (#222)

* Fix matrix binary ops for large case where int4 isn't enough

* fix offset too

* lint

Source/Fortran/PSMatrixModule.F90

@@ -651,11 +651,12 @@ RECURSIVE SUBROUTINE ConstructMatrixFromBinary_ps(this, file_name, &
 
  !! Compute Offset
  local_triplets = total_values / this%process_grid%total_processors
- local_offset = local_triplets * this%process_grid%global_rank
+ local_offset = INT(local_triplets, KIND=NTLONG) * &
+ & this%process_grid%global_rank
  header_size = 3 * bytes_per_int + bytes_per_long
  IF (this%process_grid%global_rank .EQ. &
  & this%process_grid%total_processors - 1) THEN
- local_triplets = INT(total_values) - INT(local_offset)
+ local_triplets = INT(total_values - local_offset)
  END IF
  local_offset = local_offset*(bytes_per_int*2 + bytes_per_data) + &
  & header_size

Source/Fortran/distributed_includes/WriteMatrixToBinary.f90

@@ -25,6 +25,7 @@
  CALL MPI_Allgather(SIZE(merged_local_data%values), 1, MPINTINTEGER, &
  & local_values_buffer, 1, MPINTINTEGER, &
  & this%process_grid%within_slice_comm, ierr)
+ total_values = GetMatrixSize(this)
 
  write_offset = 0
  write_offset = write_offset + header_size
@@ -53,7 +54,6 @@
  END IF
  CALL MPI_File_write_at(mpi_file_handler, zero_offset, header_buffer, &
  & 3, MPINTINTEGER, message_status, ierr)
- total_values = SUM(local_values_buffer)
  CALL MPI_File_write_at(mpi_file_handler, &
  & zero_offset + bytes_per_int * 3, total_values, &
  & 1, MPINTLONG, message_status, ierr)