need elegant way to handle byte-level pointer math used extensively in algos

[raffenet]

CIVL does not like when you do pointer math at byte-level, which MPICH does due to type information being external to the buffer supplied by the user.

Violation 0 encountered at depth 472:
CIVL execution violation in p1 (property: OUT_OF_BOUNDS, certainty: PROVEABLE) at
(char *) recvbuf
civl_allred_ring.c:185:15-30 | (char *) recvbuf

Pointer addition &<d7>my_allreduce_result[0] + 5*SIZEOF_REAL results in an index out of bound error.
Object: Variable my_allreduce_result
Object type :real[10]
Pointer value: &<d7>my_allreduce_result[0]
Offset value: 5*SIZEOF_REAL
Violated constraint: 0 < 0 + 5*SIZEOF_REAL <= 10

Input:
  _civl_argc=X__civl_argc
  _civl_argv=X__civl_argv
  N=10
  x_val=X_x_val
  _mpi_nprocs=2
  _mpi_nprocs_lo=1
  _mpi_nprocs_hi=10
Context:
  0==(32768 div SIZEOF_REAL)
  forall _t : dynamicType . (0 <= CIVL_SIZEOF(_t) - 1)
  0<=(SIZEOF_INT-1)
  0<=(SIZEOF_REAL-6554)
  0<=(X__civl_argc-1)
Call stacks:
process 0:
  main@23
process 1:
  MPIR_Allgatherv_intra_ring@573 civl_allred_ring.c:185:8 rbuf called from
  MPIR_Allreduce_intra_ring@372 civl_allred_ring.c:102:16 MPIR_Allgatherv_intra_ring called from
  _civl_main@400 ring_allreduce_driver.cvl:45:4 MPIR_Allreduce_intra_ring called from
  _mpi_process@230  called from
  _par_proc0@816
process 2:
  $mpi_sendrecv@685 civl-mpi-blocking.cvl:275:6 $comm_dequeue called from
  MPI_Sendrecv@257 mpi.cvl:225:2 $mpi_sendrecv called from
  MPIR_Allreduce_intra_ring@368 mpiimpl_cvl.h:8:47 MPI_Sendrecv called from
  _civl_main@400 ring_allreduce_driver.cvl:45:4 MPIR_Allreduce_intra_ring called from
  _mpi_process@230  called from
  _par_proc0@816