[svn-r10558] use MPI_Count in MPI_Status

This commit actually does a few things to the status object:

(1) `int count` becomes `MPI_Count count`;
(2) `count` & `cancelled` have been reordered after the "public"
fields (e.g., MPI_TAG) to ensure that direct access from Fortran is
not somehow broken by the (potentially) non-INTEGER count field;
(3) and a new future-proofing field, `abi_slush_fund`, has been added to
attempt to allow us to add fields to the status object in the future
without having to break ABI compatibility.

There's still too much duplication of the status object's contents/layout
in multiple pieces of the code, but I don't see a reasonable way to fix
this in the near term.

Warning: large count breaks the experimental Fortran binding code
generated for handling the "sizeof(int)!=sizeof(MPI_Fint)" case.

Reviewed by balaji@.

configure.ac

@@ -3748,61 +3748,6 @@ dnl     len_doublecplx=$hexlen
 	fi
     fi
 
-    #
-    # Set size of MPI_Status.  We may need sizeof_int here as well
-    #
-
-    #
-    # The size of MPI_Status is needed for the Fortran interface. 
-    # This is not quite right unless the device setup script, executed above,
-    # sets the EXTRA_STATUS_DECL that will be used in defining a status.
-    # The default size assumes that there are 5 ints
-    AC_CACHE_CHECK([for size of MPI_Status],pac_cv_sizeof_mpi_status,[
-        rm -f pac_mpi_status.h
-        cat > pac_mpi_status.h <<_EOF
-typedef struct {
-    int count;
-    int cancelled;
-    int MPI_SOURCE;
-    int MPI_TAG;
-    int MPI_ERROR;
-    $EXTRA_STATUS_DECL
-} MPI_Status;
-_EOF
-        dnl another alternative is to use AC_CHECK_SIZEOF as follows:
-        dnl AC_CHECK_SIZEOF([MPI_Status],[], [#include "pac_status.h"])
-        dnl But AC_CHECK_SIZEOF has its own AC_MSG_CHECKING, so can't use
-        dnl AC_CACHE_CHECK, so use AC_COMPUTE_INT here.
-        default_status_size=`expr 5 \* $ac_cv_sizeof_int`
-        AC_COMPUTE_INT([pac_cv_sizeof_mpi_status],
-                       [sizeof(MPI_Status)],
-                       [#include "pac_mpi_status.h"],
-                       [pac_cv_sizeof_mpi_status=$default_status_size])
-        rm -f pac_mpi_status.h
-    ])
-
-    SIZEOF_MPI_STATUS=$pac_cv_sizeof_mpi_status
-    export SIZEOF_MPI_STATUS
-    AC_SUBST(SIZEOF_MPI_STATUS)
-
-    if test -z "$MPI_STATUS_SIZE" ; then
-        if test -n "$SIZEOF_MPI_STATUS" ; then
-            # compute from the C sizeof
-            MPI_STATUS_SIZE=`expr $SIZEOF_MPI_STATUS / $ac_cv_sizeof_int`
-            if test "$MPI_STATUS_SIZE" = "0" ; then
-                AC_MSG_WARN([Could not compute the size of MPI_Status])
-                MPI_STATUS_SIZE=5
-            fi
-        else
-            AC_MSG_WARN([MPI_STATUS_SIZE was not defined!])
-            # The default size is 5.  See mpi.h.in
-            MPI_STATUS_SIZE=5
-        fi
-    fi
-    AC_SUBST(MPI_STATUS_SIZE)
-    MPIF_STATUS_SIZE=$MPI_STATUS_SIZE
-    AC_DEFINE_UNQUOTED(MPIF_STATUS_SIZE,$MPIF_STATUS_SIZE,[Size of an MPI_STATUS, in Fortran, in Fortran integers])
-
     # We must convert all hex values to decimal (!).
     # This is for the C types so they are also available in Fortran
     for var in CHAR SIGNED_CHAR UNSIGNED_CHAR WCHAR SHORT \
@@ -5563,6 +5508,76 @@ AC_DEFINE_UNQUOTED([MPIR_COUNT_MAX],[$MPIR_COUNT_MAX],[limits.h _MAX constant fo
 
 # ----------------------------------------------------------------------------
 
+#
+# Set size of MPI_Status.  Must come after MPI_Count determination.
+#
+
+#
+# The size of MPI_Status is needed for the Fortran interface. 
+# This is not quite right unless the device prereq macro, expanded above,
+# sets the EXTRA_STATUS_DECL that will be used in defining a status.
+#
+# WARNING!!! this is a spot where we duplicate code from mpi.h.in and it *must*
+# be kept in sync in order to make a proper computation
+AC_CACHE_CHECK([for size of MPI_Status],[pac_cv_sizeof_mpi_status],[
+dnl "double-quote" (in m4), otherwise you end up with a heinous bug because of
+dnl the array subscripting below
+[
+    rm -f pac_mpi_status.h
+    cat > pac_mpi_status.h <<_EOF
+typedef struct {
+    int MPI_SOURCE;
+    int MPI_TAG;
+    int MPI_ERROR;
+    $MPI_COUNT count;
+    int cancelled;
+    int abi_slush_fund[2];
+    $EXTRA_STATUS_DECL
+} MPI_Status;
+_EOF
+]
+
+dnl just compute it, since a 1s-complement or sign-and-magnitude machine is
+dnl *highly* unlikely.  Users will report the error if it is ever dnl
+dnl encountered, which will be safer than attempting some never-tested
+dnl default fallback.
+    AC_COMPUTE_INT([pac_cv_sizeof_mpi_status],
+                   [sizeof(MPI_Status)],
+                   [#include "pac_mpi_status.h"],
+                   [AC_MSG_ERROR([unable to compute status size, are you compiling on a non-2s-complement host?])])
+    rm -f pac_mpi_status.h
+])
+
+SIZEOF_MPI_STATUS=$pac_cv_sizeof_mpi_status
+export SIZEOF_MPI_STATUS
+AC_SUBST([SIZEOF_MPI_STATUS])
+
+if test -z "$MPI_STATUS_SIZE" ; then
+    if test -n "$SIZEOF_MPI_STATUS" ; then
+        # compute from the C sizeof
+        # FIXME shouldn't these calculations be based on the size of a Fortran
+        # 'INTEGER' rather than a C 'int'?
+        AS_VAR_ARITH([MPI_STATUS_SIZE],[$SIZEOF_MPI_STATUS / $ac_cv_sizeof_int])
+        if test "$MPI_STATUS_SIZE" = "0" ; then
+            AC_MSG_ERROR([Could not compute the size of MPI_Status])
+        fi
+
+        # sanity: otherwise we have a problem in Fortran where we must implement
+        # MPI_STATUS as an array of INTEGERs with MPI_STATUS_SIZE elements
+        AS_VAR_ARITH([status_int_rem],[$SIZEOF_MPI_STATUS '%' $ac_cv_sizeof_int])
+        AS_IF([test "$status_int_rem" -gt 0],
+              [AC_MSG_ERROR([status size does not divide evenly by the integer size])])
+        AS_UNSET([status_int_rem])
+    else
+        AC_MSG_ERROR([MPI_STATUS_SIZE was not defined!])
+    fi
+fi
+AC_SUBST([MPI_STATUS_SIZE])
+MPIF_STATUS_SIZE=$MPI_STATUS_SIZE
+AC_DEFINE_UNQUOTED([MPIF_STATUS_SIZE],[$MPIF_STATUS_SIZE],[Size of an MPI_STATUS, in Fortran, in Fortran integers])
+
+# ----------------------------------------------------------------------------
+
 dnl Extra status information, from device subsystem, to be included in the
 dnl declaration of MPI_Status in mpi.h.in
 AC_SUBST(EXTRA_STATUS_DECL)

src/binding/f77/buildiface

@@ -3591,7 +3591,7 @@ if ($write_mpif) {
     # FIXME: The offsets for the status elements are hardwired.  If they
     # change in mpi.h.in, they need to change here as well.
     print MPIFFD "       INTEGER MPI_SOURCE, MPI_TAG, MPI_ERROR\n";
-    print MPIFFD "       PARAMETER (MPI_SOURCE=3,MPI_TAG=4,MPI_ERROR=5)\n";
+    print MPIFFD "       PARAMETER (MPI_SOURCE=1,MPI_TAG=2,MPI_ERROR=3)\n";
     print MPIFFD "       INTEGER MPI_STATUS_SIZE\n";
     print MPIFFD "       PARAMETER (MPI_STATUS_SIZE=\@MPI_STATUS_SIZE\@)\n";
     # Temporary until configure handles these.  Define as arrays to keep
@@ -4378,11 +4378,13 @@ print $OUTFD "\
 #ifdef HAVE_FINT_IS_INT
     *c_status = *(MPI_Status *)	f_status;
 #else
-    c_status->count      = f_status\[0\];
-    c_status->cancelled  = f_status\[1\];
-    c_status->MPI_SOURCE = f_status\[2\];
-    c_status->MPI_TAG    = f_status\[3\];
-    c_status->MPI_ERROR  = f_status\[4\];
+#error this code is broken because of the new MPI_Count type
+    c_status->MPI_SOURCE = f_status\[0\];
+    c_status->MPI_TAG    = f_status\[1\];
+    c_status->MPI_ERROR  = f_status\[2\];
+    c_status->count      = f_status\[3\];
+    c_status->cancelled  = f_status\[4\];
+    /* no need to copy abi_slush_fund field */
 #endif\n";
     }
     else {
@@ -4451,13 +4453,13 @@ int MPI_Status_c2f( const MPI_Status *c_status, MPI_Fint *f_status )
 #ifdef HAVE_FINT_IS_INT
     *(MPI_Status *)f_status = *c_status;
 #else
-    f_status\[0\]   = c_status->count;
-    f_status\[1\]   = c_status->cancelled;
-    f_status\[2\] = c_status->MPI_SOURCE;
-    f_status\[3\] = c_status->MPI_TAG;
-    f_status\[4\] = c_status->MPI_ERROR;
+#error this code is broken because of the new MPI_Count type
+    f_status\[0\] = c_status->MPI_SOURCE;
+    f_status\[1\] = c_status->MPI_TAG;
+    f_status\[2\] = c_status->MPI_ERROR;
+    f_status\[3\] = c_status->count;
+    f_status\[4\] = c_status->cancelled;
 #endif\n";
-    #    printf( \"c_status = %d, %d, %d\\n\", c_status->MPI_SOURCE, c_status->MPI_TAG, c_status->MPI_ERROR );
     }
     else {
         print $OUTFD "    *(MPI_Status *)f_status = *c_status;\n";

src/binding/f90/buildiface

@@ -642,11 +642,15 @@ $PRIVATEVAR = "";
 #
 #$BINDACCESS = ", BIND(C)";
 #$BINDDEF ="";
-print MPIFD "        TYPE$BINDACCESS :: MPI_Status
+print MPIFD <<EOT;
+        TYPE$BINDACCESS :: MPI_Status
            $BINDDEF
-           INTEGER$PRIVATEVAR :: count, cancelled
            INTEGER$PUBLICVAR :: MPI_SOURCE, MPI_TAG, MPI_ERROR
-        END TYPE MPI_Status\n";
+           INTEGER$PRIVATEVAR(KIND=MPI_COUNT_KIND) :: count
+           INTEGER$PRIVATEVAR :: cancelled
+           INTEGER$PRIVATEVAR(2) :: abi_slush_fund
+        END TYPE MPI_Status
+EOT
 
 %handles = ( 'comm' => 'Comm', 'datatype' => 'Datatype', 
 	     'group' => 'Group', 'win' => 'Win', 

src/include/mpi.h.in

@@ -523,11 +523,12 @@ static const MPI_Datatype mpich_mpi_offset MPICH_ATTR_TYPE_TAG(MPI_Offset) = MPI
 
 /* The order of these elements must match that in mpif.h */
 typedef struct MPI_Status {
-    int count;
-    int cancelled;
     int MPI_SOURCE;
     int MPI_TAG;
     int MPI_ERROR;
+    MPI_Count count;
+    int cancelled;
+    int abi_slush_fund[2];
     @EXTRA_STATUS_DECL@
 } MPI_Status;
 

src/mpi/init/initthread.c

@@ -429,6 +429,9 @@ int MPIR_Init_thread(int * argc, char ***argv, int required, int * provided)
     /* Assert: tag_ub is at least the minimum asked for in the MPI spec */
     MPIU_Assert( MPIR_Process.attrs.tag_ub >= 32767 );
 
+    /* very nasty bugs will occur if this does not hold */
+    MPIU_Assert(sizeof(MPI_Status) == MPIF_STATUS_SIZE*sizeof(MPI_Fint));
+
     /* Capture the level of thread support provided */
     MPIR_ThreadInfo.thread_provided = thread_provided;
     if (provided) *provided = thread_provided;