defective MPI_Pack/Unpack/Pack_size interface

MPI_Pack, MPI_Unpack, and MPI_Pack_size use integers to describe counts
of bytes and posistions in a bytestream.  In a world of 2 GiB+
datatypes, this is not correct.  It will take an army to change the
standard, but we can at least do the right thing internally.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

src/include/mpiimpl.h

@@ -4375,9 +4375,9 @@ int MPIR_Type_indexed_impl(int count, const int blocklens[], const int indices[]
 void MPIR_Type_free_impl(MPI_Datatype *datatype);
 int MPIR_Type_vector_impl(int count, int blocklength, int stride, MPI_Datatype old_type, MPI_Datatype *newtype_p);
 int MPIR_Type_struct_impl(int count, const int blocklens[], const MPI_Aint indices[], const MPI_Datatype old_types[], MPI_Datatype *newtype);
-int MPIR_Pack_impl(const void *inbuf, int incount, MPI_Datatype datatype, void *outbuf, int outcount, int *position);
-void MPIR_Pack_size_impl(int incount, MPI_Datatype datatype, int *size);
-int MPIR_Unpack_impl(const void *inbuf, int insize, int *position,
+int MPIR_Pack_impl(const void *inbuf, int incount, MPI_Datatype datatype, void *outbuf, MPI_Aint outcount, MPI_Aint *position);
+void MPIR_Pack_size_impl(int incount, MPI_Datatype datatype, MPI_Aint *size);
+int MPIR_Unpack_impl(const void *inbuf, MPI_Aint insize, MPI_Aint *position,
                      void *outbuf, int outcount, MPI_Datatype datatype);
 void MPIR_Type_lb_impl(MPI_Datatype datatype, MPI_Aint *displacement);
 int MPIR_Ibsend_impl(const void *buf, int count, MPI_Datatype datatype, int dest, int tag,

src/mpi/coll/alltoall.c

@@ -91,7 +91,8 @@ int MPIR_Alltoall_intra(
     int mpi_errno=MPI_SUCCESS, src, dst, rank, nbytes;
     int mpi_errno_ret = MPI_SUCCESS;
     MPI_Status status;
-    int sendtype_size, pack_size, block, position, *displs, count;
+    int sendtype_size, block, *displs, count;
+    MPI_Aint pack_size, position;
     MPI_Datatype newtype = MPI_DATATYPE_NULL;
     void *tmp_buf;
     MPI_Comm comm;

src/mpi/coll/bcast.c

@@ -47,7 +47,7 @@ static int MPIR_Bcast_binomial(
     MPI_Status status;
     int is_contig, is_homogeneous;
     MPI_Aint type_size;
-    int position;
+    MPI_Aint position;
     void *tmp_buf=NULL;
     MPI_Comm comm;
     MPID_Datatype *dtp;
@@ -404,7 +404,8 @@ static int MPIR_Bcast_scatter_doubling_allgather(
     int j, k, i, tmp_mask, is_contig, is_homogeneous;
     MPI_Aint type_size, nbytes = 0;
     int relative_dst, dst_tree_root, my_tree_root, send_offset;
-    int recv_offset, tree_root, nprocs_completed, offset, position;
+    int recv_offset, tree_root, nprocs_completed, offset;
+    MPI_Aint position;
     MPIU_CHKLMEM_DECL(1);
     MPI_Comm comm;
     MPID_Datatype *dtp;
@@ -707,10 +708,9 @@ static int MPIR_Bcast_scatter_ring_allgather(
     int rank, comm_size;
     int mpi_errno = MPI_SUCCESS;
     int mpi_errno_ret = MPI_SUCCESS;
-    int scatter_size, nbytes;
+    int scatter_size;
     int j, i, is_contig, is_homogeneous;
-    MPI_Aint type_size;
-    int position;
+    MPI_Aint nbytes, type_size, position;
     int left, right, jnext;
     int curr_size = 0;
     void *tmp_buf;
@@ -1143,7 +1143,8 @@ int MPIR_Bcast_intra (
     int mpi_errno_ret = MPI_SUCCESS;
     int comm_size;
     int nbytes=0;
-    int type_size, is_homogeneous;
+    int is_homogeneous;
+    MPI_Aint type_size;
     MPID_MPI_STATE_DECL(MPID_STATE_MPIR_BCAST);
 
     MPID_MPI_FUNC_ENTER(MPID_STATE_MPIR_BCAST);

src/mpi/coll/helper_fns.c

@@ -493,8 +493,8 @@ int MPIC_Sendrecv_replace(void *buf, int count, MPI_Datatype datatype,
     MPID_Request *sreq;
     MPID_Request *rreq;
     void *tmpbuf = NULL;
-    int tmpbuf_size = 0;
-    int tmpbuf_count = 0;
+    MPI_Aint tmpbuf_size = 0;
+    MPI_Aint tmpbuf_count = 0;
     MPID_Comm *comm_ptr;
     MPIU_CHKLMEM_DECL(1);
     MPIDI_STATE_DECL(MPID_STATE_MPIC_SENDRECV_REPLACE_FT);

src/mpi/coll/ibcast.c

@@ -40,8 +40,8 @@ int MPIR_Ibcast_binomial(void *buffer, int count, MPI_Datatype datatype, int roo
     int mpi_errno = MPI_SUCCESS;
     int mask;
     int comm_size, rank;
-    int type_size, is_contig, is_homogeneous;
-    int nbytes;
+    int is_contig, is_homogeneous;
+    MPI_Aint nbytes, type_size;
     int relative_rank;
     int src, dst;
     void *tmp_buf = NULL;
@@ -669,7 +669,8 @@ fn_fail:
 int MPIR_Ibcast_SMP(void *buffer, int count, MPI_Datatype datatype, int root, MPID_Comm *comm_ptr, MPID_Sched_t s)
 {
     int mpi_errno = MPI_SUCCESS;
-    int type_size, is_homogeneous;
+    int is_homogeneous;
+    MPI_Aint type_size;
 
     if (!MPIR_CVAR_ENABLE_SMP_COLLECTIVES || !MPIR_CVAR_ENABLE_SMP_BCAST)
         MPID_Abort(comm_ptr, MPI_ERR_OTHER, 1, "SMP collectives are disabled!");
@@ -754,7 +755,8 @@ fn_fail:
 int MPIR_Ibcast_intra(void *buffer, int count, MPI_Datatype datatype, int root, MPID_Comm *comm_ptr, MPID_Sched_t s)
 {
     int mpi_errno = MPI_SUCCESS;
-    int comm_size, is_homogeneous, type_size, nbytes;
+    int comm_size, is_homogeneous;
+    MPI_Aint type_size, nbytes;
 
     MPIU_Assert(comm_ptr->comm_kind == MPID_INTRACOMM);
 

src/mpi/datatype/pack.c

@@ -31,8 +31,8 @@ int MPIR_Pack_impl(const void *inbuf,
                    int incount,
                    MPI_Datatype datatype,
                    void *outbuf,
-                   int outsize,
-                   int *position)
+                   MPI_Aint outsize,
+                   MPI_Aint *position)
 {
     int mpi_errno = MPI_SUCCESS;
     MPI_Aint first, last;
@@ -151,6 +151,7 @@ int MPI_Pack(const void *inbuf,
 	     MPI_Comm comm)
 {
     int mpi_errno = MPI_SUCCESS;
+    MPI_Aint position_x;
     MPID_Comm *comm_ptr = NULL;
     
     MPID_MPI_STATE_DECL(MPID_STATE_MPI_PACK);
@@ -240,7 +241,9 @@ int MPI_Pack(const void *inbuf,
 
     /* ... body of routine ... */
 
-    mpi_errno = MPIR_Pack_impl(inbuf, incount, datatype, outbuf, outsize, position);
+    position_x = *position;
+    mpi_errno = MPIR_Pack_impl(inbuf, incount, datatype, outbuf, outsize, &position_x);
+    MPIU_Assign_trunc(*position, position_x, int);
     if (mpi_errno) goto fn_fail;
     
    /* ... end of body of routine ... */

src/mpi/datatype/pack_size.c

@@ -27,9 +27,9 @@
 #define FUNCNAME MPIR_Pack_size_impl
 #undef FCNAME
 #define FCNAME MPIU_QUOTE(FUNCNAME)
-void MPIR_Pack_size_impl(int incount, MPI_Datatype datatype, int *size)
+void MPIR_Pack_size_impl(int incount, MPI_Datatype datatype, MPI_Aint *size)
 {
-    int typesize;
+    MPI_Aint typesize;
     MPID_Datatype_get_size_macro(datatype, typesize);
     *size = incount * typesize;
 }
@@ -76,6 +76,7 @@ int MPI_Pack_size(int incount,
 {
     MPID_Comm *comm_ptr = NULL;
     int mpi_errno = MPI_SUCCESS;
+    MPI_Aint size_x = MPI_UNDEFINED;
     MPID_MPI_STATE_DECL(MPID_STATE_MPI_PACK_SIZE);
 
     MPIR_ERRTEST_INITIALIZED_ORDIE();
@@ -124,8 +125,9 @@ int MPI_Pack_size(int incount,
 
     /* ... body of routine ... */
 
-    MPIR_Pack_size_impl(incount, datatype, size);
-    
+    MPIR_Pack_size_impl(incount, datatype, &size_x);
+    MPIU_Assign_trunc(*size, size_x, int);
+
     /* ... end of body of routine ... */
 
 #ifdef HAVE_ERROR_CHECKING

src/mpi/datatype/unpack.c

@@ -27,7 +27,7 @@
 #define FUNCNAME MPIR_Unpack_impl
 #undef FCNAME
 #define FCNAME MPIU_QUOTE(FUNCNAME)
-int MPIR_Unpack_impl(const void *inbuf, int insize, int *position,
+int MPIR_Unpack_impl(const void *inbuf, MPI_Aint insize, MPI_Aint *position,
                      void *outbuf, int outcount, MPI_Datatype datatype)
 {
     int mpi_errno = MPI_SUCCESS;
@@ -137,6 +137,7 @@ int MPI_Unpack(const void *inbuf, int insize, int *position,
 	       MPI_Comm comm)
 {
     int mpi_errno = MPI_SUCCESS;
+    MPI_Aint position_x;
     MPID_Comm *comm_ptr = NULL;
     MPID_MPI_STATE_DECL(MPID_STATE_MPI_UNPACK);
 
@@ -193,8 +194,10 @@ int MPI_Unpack(const void *inbuf, int insize, int *position,
 
     /* ... body of routine ...  */
     
-    mpi_errno = MPIR_Unpack_impl(inbuf, insize, position, outbuf, outcount, datatype);
+    position_x = *position;
+    mpi_errno = MPIR_Unpack_impl(inbuf, insize, &position_x, outbuf, outcount, datatype);
     if (mpi_errno) goto fn_fail;
+    MPIU_Assign_trunc(*position, position_x, int);
     
     /* ... end of body of routine ... */
 

src/mpi/pt2pt/bsendutil.c

@@ -213,7 +213,8 @@ int MPIR_Bsend_isend(const void *buf, int count, MPI_Datatype dtype,
     int mpi_errno = MPI_SUCCESS;
     MPIR_Bsend_data_t *p;
     MPIR_Bsend_msg_t *msg;
-    int packsize, pass;
+    MPI_Aint packsize;
+    int pass;
 
     /* Find a free segment and copy the data into it.  If we could 
        have, we would already have used tBsend to send the message with
@@ -234,7 +235,7 @@ int MPIR_Bsend_isend(const void *buf, int count, MPI_Datatype dtype,
     else
         packsize = count;
 
-    MPIU_DBG_MSG_D(BSEND,TYPICAL,"looking for buffer of size %d", packsize);
+    MPIU_DBG_MSG_D(BSEND,TYPICAL,"looking for buffer of size %ld", packsize);
     /*
      * Use two passes.  Each pass is the same; between the two passes,
      * attempt to complete any active requests, and start any pending
@@ -246,7 +247,7 @@ int MPIR_Bsend_isend(const void *buf, int count, MPI_Datatype dtype,
 	p = MPIR_Bsend_find_buffer( packsize );
 	if (p) {
 	    MPIU_DBG_MSG_FMT(BSEND,TYPICAL,(MPIU_DBG_FDEST,
-                     "found buffer of size %d with address %p",packsize,p));
+                     "found buffer of size %ld with address %p",packsize,p));
 	    /* Found a segment */
 
 	    msg = &p->msg;

src/mpi/pt2pt/sendrecv_rep.c

@@ -140,8 +140,8 @@ int MPI_Sendrecv_replace(void *buf, int count, MPI_Datatype datatype,
 	MPID_Request * sreq;
 	MPID_Request * rreq;
 	void * tmpbuf = NULL;
-	int tmpbuf_size = 0;
-	int tmpbuf_count = 0;
+	MPI_Aint tmpbuf_size = 0;
+	MPI_Aint tmpbuf_count = 0;
 
 	if (count > 0 && dest != MPI_PROC_NULL)
 	{