EXODUS/packages/seacas/doc-source/supes/memmgr.tex

\chapter{MEMORY MANAGER}
The purpose of the memory manager utilities is to allow an applications
programmer to write standard, readable FORTRAN-77 code while employing
dynamic memory management for REAL, INTEGER, LOGICAL and CHARACTER type
arrays.

Because the array sizes in most programs are problem dependent, a
program's memory requirements are not known until the program is running.
Since FORTRAN-77 does not provide for dynamic memory allocation, the
programmer has to either predict the maximum memory requirement or use
machine dependent requests for memory.  In addition, dynamic memory
allocation is an error prone exercise which tends to make the source code
difficult to read and maintain.

In SUPES, the memory manager utilities are written in standard FORTRAN-77
and provide an interface which encourages readable coding and efficient use
of memory resources.  Machine dependencies are isolated through the use of
the extension library (Chapter~\ref{sec:extlib}). All memory
requests are in terms of {\em numeric storage units} for numeric data
(integer, real, or logical) and {\em character storage units} for
character data~\cite{ansi}.

An important design feature of the memory manager is that the memory manager
can be supported even when the system-dependent dynamic memory request
routines are not implemented on a system.  In this case, the memory manager
will operate, allocating space from a user-supplied work array.  This mode
is described as dynamic allocation of static memory. Thus, modification of a
user's application program is minimal on systems where dynamic memory is not
implemented.

All user entry points to memory manager routines begin with either ``MD''
or ``MC.'' In most cases, the ``MD'' routines are used for numeric data,
while the ``MC'' routines are for character data. In some cases, however,
the routines are interchangeable. These routines are documented as synonyms.

In this document, the term ``Mx'' is used to refer simultaneously to
both ``MD'' and ``MC'' routines. Thus, MxRSRV is a reference to both MDRSRV
and MCRSRV subroutines.

The memory manager utility is divided into three categories; basic routines,
advanced routines, and development aids.  These categories will be discussed
in sections \ref{sec:mbas} through \ref{sec:mdev}.


\section{Indexing System}
In order to use the memory manager properly, the user must first understand
the concept of using a base array with indices for accessing memory
address locations.
At the core of this concept is FORTRAN's convention of passing
SUBROUTINE array references by address.
The memory manager references all memory addresses relative
to the addresses of user-supplied base
arrays---one each for numeric and character data. A reference
to memory is made in terms of a pointer to these base arrays.
Specifically,
the memory manager determines an indexing parameter by first determining the
offset of the appropriate memory location relative to the address
of the correct base
array.
The index is then computed in terms of the proper storage units
(either character or numeric).
Note that the resulting
indexes may take on a wide range of
values, including negative numbers.

The base arrays must comply with the following rules:
\begin{enumerate}
\item Numeric base arrays {\em must} be of type INTEGER, REAL, or LOGICAL.  Modified
word length storage arrays such as INTEGER$*$ 2 or REAL$*$ 8 will result
in invalid indexes with no error message.

\item Character base arrays {\em must} be declared CHARACTER$*$ 1.
\end{enumerate}

The following FORTRAN statements define valid base arrays:
\begin{verbatim}
DIMENSION NUMBAS(1)
CHARACTER*1 CHRBAS(1)
\end{verbatim}
Only one base array from each category (numeric and character) may
be used in a program.

In order to use memory allocated by the memory manager, the user merely
needs to pass the base array with the correct offsetting
index to a subprogram.
For example, for a base arrays NUMBAS and CHRBAS
and indexes IP1 and IP2, a subroutine call would be:

\verb+CALL SUBBIE ( NUMBAS(IP1), CHRBAS(IP2) )+

Although the programmer is not restricted to using the allocated memory in
subprograms only, the recommended usage for the memory manager is to
allocate dynamic arrays in the main program and then pass them to
subroutines.


\section{Basic Routines}\label{sec:mbas}
The basic memory manager routines are those which are most commonly used and
require little understanding of the internal workings of the utility.

\subsection{Initialize (MDINIT/MCINIT)}
The memory manager {\em must} be initialized with a calls to MDINIT and
MCINIT before any memory can be allocated.  The main purpose of the
initialization is to determine the location of the numeric and character
base arrays in memory. MDINIT must be called first, and MCINIT second. In
the case where character dynamic memory is not used, MCINIT need not be
called. When calling MxINIT, the user must pass (explicitly or
implicitly) subscript 1 of the base array.
\begin{verbatim}
CALL MDINIT (NUMBAS(1))
CALL MCINIT (CHRBAS(1))
\end{verbatim}

\begin{argy}{NUMBAS}{INTEGER, LOGICAL or REAL Array or Array Element}{Read
Only}
This array is used as a base reference to all dynamically allocated
numeric memory.
\end{argy}

\begin{argy}{CHRBAS}{CHARACTER$*$ 1 Array or Array Element}{Read Only}
This array is used as a base reference to all dynamically allocated
character memory.
\end{argy}

\subsection{Define Dynamic Array (MDRSRV/MCRSRV)}
MxRSRV declares new dynamic arrays.  The user specifies the space
required, and an index to the new space is returned.  Note that, by
default, the contents of the new storage are not initialized to any specific
value. MxFILL may be used for data initialization.
\begin{verbatim}
CALL MDRSRV (NAME, NEWIDX, NEWLEN)
CALL MCRSRV (NAME, NEWIDX, NEWLEN)
\end{verbatim}

\begin{argy}{NAME}{CHARACTER$*$($*$)}{Read Only}
This is the name of the new dynamic array. The memory manager will add this
name to its internal dictionary; each array must have a unique name.  The
first eight characters beginning with a nonblank are
used for comparison.
This comparison is case-insensitive and embedded blanks are
significant.
\end{argy}

\begin{argy}{NEWIDX}{INTEGER}{Write Only}
This is the index to storage allocated to this dynamic array relative to
the base array. The index for numeric data is to be used with the numeric
array supplied to MDINIT, and character data is to be used with the
character array supplied to MCINIT.
\end{argy}

\begin{argy}{NEWLEN}{INTEGER}{Read Only}
This is the length to be reserved for the new array. Any nonnegative number
is acceptable.  A zero length does not cause any storage to be allocated and
returns an index equal to one. The value of NEWLEN is in terms of numeric
storage units for numeric data and character storage units for character
data.
\end{argy}

\subsection{Delete Dynamic Array (MDDEL/MCDEL)}
MDDEL and MCDEL release the memory that is allocated to a dynamic array for
numeric and character storage, respectively.
\begin{verbatim}
CALL MDDEL (NAME)
CALL MCDEL (NAME)
\end{verbatim}

\begin{argy}{NAME}{CHARACTER$*$($*$)}{Read Only}
This is the name of the dynamic array which is to be deleted.  The array
name must match an existing name in the dictionary and be of the correct
type (numeric or character) for the operation. The first eight characters
beginning with a nonblank are used for comparison.
This comparison is case-insensitive and embedded blanks are
significant.
\end{argy}

\subsection{Reserve Memory Block (MDGET/MCGET)}
NOTE: This capability has been removed.  Calling \code{MDGET} or \code{MCGET} will have no affect.

MDGET and MCGET reserve a contiguous block of memory without associating the
block of memory with an array.  MxGET should be called prior to a
series of calls to MxRSRV to improve efficiency and to reduce memory
fragmentation.  Further discussion of the operation of MxGET is
found in section~\ref{sec:table}.
\begin{verbatim}
CALL MDGET (MNGET)
CALL MCGET (MNGET)
\end{verbatim}

\begin{argy}{MNGET}{INTEGER}{Read only}
This specifies the desired contiguous block size in numeric storage units
for MDGET or character storage units for MCGET.
\end{argy}

\subsection{Release Unallocated Memory (MDGIVE/MCGIVE)}
NOTE: This capability has been removed.  Calling \code{MDGIVE} or \code{MCGIVE} will have no affect.

MxGIVE
causes the memory manager to return unused storage to the operating
system, if possible. MDGIVE and MCGIVE are synonyms.
\begin{verbatim}
CALL MDGIVE ()
CALL MCGIVE ()
\end{verbatim}

\subsection{Obtain Statistics (MDSTAT/MCSTAT)}
MxSTAT returns memory manager statistics.  MxSTAT provides a method
for error checking, and thus should be used after other calls to the memory
manager to assure no errors have occurred. MDSTAT and MCSTAT are synonyms.
\begin{verbatim}
CALL MDSTAT (MNERRS, MNUSED)
CALL MCSTAT (MNERRS, MNUSED)
\end{verbatim}

\begin{argy}{MNERRS}{INTEGER}{Write Only}
This is the total number of errors detected by the memory manager during the
current execution.
\end{argy}

\begin{argy}{MNUSED}{INTEGER}{Write Only}
This is the total number of storage units that are currently allocated to
dynamic arrays. MDSTAT returns the numeric storage in numeric storage units,
and MCSTAT returns the character storage in character storage units. If any
storage has been requested in the deferred mode and not yet allocated by the
memory manager (Section~\ref{sec:wait}), this storage is counted as though
it were actually allocated.
\end{argy}

\subsection{Print Error Summary (MDEROR/MCEROR)}
MxEROR prints a summary of all errors detected by the memory
manager.  The return status of the last memory manager routine called is
also printed. MxEROR should be called any time an error is
detected by a call to MxSTAT.
Table~\ref{tab:ecode} lists the error codes.
MDEROR and MCEROR are synonyms.
\begin{table}[htbp]
\centering
\caption{Memory Manager Error Codes} \label{tab:ecode}
\begin{tabular}{|rcl|} \hline \hline
\multicolumn{3}{|c|}{ERROR CODES}\\ \hline
  1 &  &  SUCCESSFUL COMPLETION\\
  2 &  &  UNABLE TO GET REQUESTED SPACE FROM SYSTEM\\
  3 &  &  DATA MANAGER NOT INITIALIZED\\
  4 &  &  DATA MANAGER WAS PREVIOUSLY INITIALIZED\\
  5 &  &  NAME NOT FOUND IN DICTIONARY\\
  6 &  &  NAME ALREADY EXISTS IN DICTIONARY\\
  7 &  &  ILLEGAL LENGTH REQUEST\\
  8 &  &  UNKNOWN DATA TYPE\\
  9 & $*$&  DICTIONARY IS FULL\\
 10 & $*$&  VOID TABLE IS FULL\\
 11 & $*$&  MEMORY BLOCK TABLE IS FULL\\
 12 & $*$&  OVERLAPPING VOIDS - INTERNAL ERROR\\
 13 & $*$&  OVERLAPPING MEMORY BLOCKS - INTERNAL ERROR\\
 14 & $*$&  INVALID MEMORY BLOCK - EXTENSION LIBRARY ERROR\\
 15 &  &  INVALID ERROR CODE\\
 16 &  &  INVALID INPUT NAME\\
 17 &  &  ILLEGAL CALL WHILE IN DEFER MODE\\
 18 &  &  NAME IS OF WRONG TYPE FOR OPERATION\\ \hline \hline
\multicolumn{3}{c}{$*$ These are not user errors.}\\
\end{tabular}
\end{table}

Several of the error codes listed in Table~\ref{tab:ecode} are not a result
of a user error, but are used to signal an internal error, or that an
internal array is full. For example, the table which records the names of
the arrays allocated with MxRSRV may not be large enough for the
application. In this case, the memory manager subroutines must be modified
to accommodate the user. A local support person should perform this task.
\begin{verbatim}
CALL MDEROR (IUNIT)
CALL MCEROR (IUNIT)
\end{verbatim}

\begin{argy}{IUNIT}{INTEGER}{Read Only}
This is the FORTRAN unit number of the output device.
\end{argy}


\subsection{Enable data initialization (MDFILL/MCFILL)}
MxFILL defines a fill/initialization pattern that is to be used
for newly allocated storage. MDFILL and MCFILL are in effect until canceled
by MDFOFF and MCFOFF, respectively.  MDFILL and MCFILL operate
independently.
\begin{verbatim}
CALL MDFILL (NUMDAT)
CALL MCFILL (CHRDAT)
\end{verbatim}

\begin{argy}{NUMDAT}{INTEGER, REAL or LOGICAL}{Read Only}
This is the initialization datum for new storage al\-lo\-cat\-ed with MDRSRV
or extended with MDLONG. The memory manager makes no attempt to identify the
type (INTEGER, REAL, or LOGICAL) of either the initialization datum or of a
newly allocated array. Instead, the bit of the initialization datum is
stored without interpretation. This pattern is then used to initialize new
storage. Since the internal machine representation of REAL data is different
than INTEGER data (or LOGICAL data), the user may experience unexpected
results when dynamic memory is used as a numeric type which is different
from the type of the initialization datum.
\end{argy}

\begin{argy}{CHRDAT}{CHARACTER$*$($*$)}{Read Only}
This is the initialization data for new storage al\-lo\-cat\-ed with MCRSRV or
extended with MCLONG. Only the first character of CHRDAT is used.
\end{argy}

\subsection{Cancel Data Initialization (MDFOFF/MCFOFF)}
MDFOFF and MCFOFF cancel the data initialization for numeric and character
data, respectively.  MDFOFF and MCFOFF operate independently.
\begin{verbatim}
CALL MDFOFF ()
CALL MCFOFF ()
\end{verbatim}

\subsection{Basic Example}
\begin{verbatim}
    DIMENSION BASE(1)
    CHARACTER*1 CBASE(1)
    CALL MDINIT (BASE(1))
    CALL MCINIT (CBASE(1))
    CALL MDGET (20)
    CALL MDFILL (0.)
    CALL MCFILL ('Z')
    CALL MDRSRV ('FIRST',  I1, 10)
    CALL MDRSRV ('SECOND', I2, 10)
    CALL MCRSRV ('THIRD',  I3, 10)
    CALL MDDEL ('SECOND')
    CALL MDGIVE ()
    CALL MDSTAT (MNERRS, MNUSED)
    IF (MNERRS .NE. 0) THEN
       CALL MDEROR (6)
       STOP
    END IF
    CALL SUBPRG (BASE(I1), CBASE(I3))
\end{verbatim}

\section{Advanced Routines}
The advanced routines are supplied to give added capability to the user who
is interested in more sophisticated manipulation of memory.  These routines
are never necessary, but may be very desirable.

\subsection{Rename Dynamic Array (MDNAME/MCNAME)}
MxNAME renames a dynamic array from NAME1 to NAME2.  The location
of the array is not changed, nor is its length. MDNAME is used for numeric
arrays and MCNAME is used for character arrays.
\begin{verbatim}
CALL MDNAME (NAME1, NAME2)
CALL MCNAME (NAME1, NAME2)
\end{verbatim}

\begin{argy}{NAME1}{CHARACTER$*$($*$)}{Read Only}
This is the old name of the array.  The first eight characters after the
first nonblank are
used for comparison.
This comparison is case-insensitive and embedded blanks are
significant.
\end{argy}

\begin{argy}{NAME2}{CHARACTER$*$($*$)}{Read Only}
This is the new name of the array.  The first eight characters starting
from a nonblank are
used for the new name.
This comparison is case-insensitive and embedded blanks are
significant.
\end{argy}

\subsection{Adjust Dynamic Array Length (MDLONG/MCLONG)}
MxLONG changes the length of a dynamic array.  The memory manager
will relocate the array and move its data if storage cannot be extended at
the array's current location.  The user should assume that MxLONG
invalidates the previous index to this array if the array is extended.
MDLONG is used for numeric arrays and MCLONG is used for character arrays.
\begin{verbatim}
CALL MDLONG (NAME, NEWIDX, NEWLEN)
CALL MCLONG (NAME, NEWIDX, NEWLEN)
\end{verbatim}

\begin{argy}{NAME}{CHARACTER$*$($*$)}{Read Only}
This is the name of the dynamic array which the user wishes to extend or
shorten.
\end{argy}

\begin{argy}{NEWIDX}{INTEGER}{Write Only}
This is the new index to the dynamic array.
\end{argy}

\begin{argy}{NEWLEN}{INTEGER}{Read Only}
This is the new length for the dynamic array in numeric storage units for
MDLONG and in character storage units for MCLONG.
\end{argy}

\subsection{Locate Dynamic Array (MDFIND/MCFIND)}
MxFIND
returns the index and length of storage allocated to a dynamic
array.  This routine would be used if the index from an earlier call to
MxRSRV was not available in a particular program segment. MDFIND
is used for numeric arrays and MCFIND is used for character arrays.
\begin{verbatim}
CALL MDFIND (NAME, NEWIDX, NEWLEN)
CALL MCFIND (NAME, NEWIDX, NEWLEN)
\end{verbatim}

\begin{argy}{NAME}{CHARACTER$*$($*$)}{Read Only}
This is the name of the dynamic array to be located.
\end{argy}

\begin{argy}{NEWIDX}{INTEGER}{Write Only}
This is the index to the dynamic array relative to the user's reference
array. Because an index can take any value, the returned value cannot be
used as an indication of success or failure of MxFIND. MxSTAT should always
be used for error checking.
\end{argy}

\begin{argy}{NEWLEN}{INTEGER}{Write Only}
This is the length of the dynamic array in numeric or character storage
units for MDFIND and MCFIND, respectively.
\end{argy}

\subsection{Compress Storage (MDCOMP/MCCOMP)}
MxCOMP causes fragmented memory to be consolidated.  Note that
this may cause array storage locations to change.  It is important to
realize that all indexes must be recalculated by calling MxFIND
after a compress operation.  A call to MxCOMP prior to
MxGIVE will result in the return of the maximum memory to the system.
MDCOMP and MCCOMP are synonyms.
\begin{verbatim}
CALL MDCOMP ()
CALL MCCOMP ()
\end{verbatim}

\subsection{Error Flag Query (MDERPT/MCERPT)}
MxERPT requests the memory manager to report the number of errors
accumulated for a particular error flag. A programmer may use this to
determine more detailed information than what is available from
MxSTAT. MDERPT and MCERPT are synonyms.
\begin{verbatim}
CALL MDERPT (IFLAG, NERRS)
CALL MCERPT (IFLAG, NERRS)
\end{verbatim}

\begin{argy}{IFLAG}{INTEGER}{Read Only}
IFLAG specifies the flag number for which the user wishes an error count. A
list of the error flags can be printed by calling MxEROR.
\end{argy}

\begin{argy}{NERRS}{INTEGER}{Write Only}
NERSS will contain the error count.
\end{argy}

\subsection{Modify Error Count (MDEFIX/MCEFIX)}
MxEFIX allows the error count for a particular error flag to be
set to a specified value. MDEFIX and MCEFIX are synonyms.
\begin{verbatim}
CALL MDEFIX (IFLAG, NERRS)
CALL MCEFIX (IFLAG, NERRS)
\end{verbatim}

\begin{argy}{IFLAG}{INTEGER}{Read Only}
IFLAG specifies the number of the error flag which will be set.
See Table~\ref{tab:ecode} for a list and description of these flags.
\end{argy}

\begin{argy}{NERRS}{INTEGER}{Read Only}
NERRS is the new value for the error count.
\end{argy}

\subsection{Report Last Error (MDLAST/MCLAST)}
MxLAST requests the flag number of the last error.
MDLAST and MCLAST are synonyms.
\begin{verbatim}
CALL MDLAST (IFLAG)
CALL MCLAST (IFLAG)
\end{verbatim}

\begin{argy}{IFLAG}{INTEGER}{Write Only}
IFLAG is the number of the last error caused by a previous call to the
memory manager.
\end{argy}

\subsection{Enable Deferred Memory Mode (MDWAIT/MCWAIT)} \label{sec:wait}
NOTE: This capability has been removed.  Calling \code{MDWAIT} or \code{MCWAIT} will have no affect.

MxWAIT enables the deferred memory mode of the memory manager. In
this mode, any requests for additional memory with MxRSRV are
satisfied only if a system call is not required. If a system call is
required, the request for memory is deferred and will be satisfied when the
deferred mode is canceled by calling MxEXEC or a call to
MxLONG requires a system call for memory for an existing array.
MDWAIT and MCWAIT are synonyms.

Because the deferred mode may not actually provide array space at the time a
call to MxRSRV is made, the base array pointer re\-turn\-ed by
MxRSRV may not be valid. In fact, for a deferred request, an invalid
index is intentionally returned so that the requested array space cannot
be erroneously used. When the deferred memory requests are eventually
satisfied (by calling MxEXEC), the indexes are automatically,
asynchronously updated by the memory manager. Thus, upon return from
MxEXEC the indexes used in MxRSRV will have a valid value.

The deferred mode is provided as a means of reducing the sometimes
time-consuming calls to the operating system for new memory. A similar
efficiency could be realized by judicious use of MxGET, but the
deferred mode relieves the user of the burden of adding all memory requests
before calling MxRSRV. The deferred mode is a sophisticated
capability and should not be enabled if the user does not understand its
implications.

The deferred mode must be used as follows:
\begin{enumerate}
\item The deferred mode begins with a call to MxWAIT.
\item Except for MxPRNT, all memory manager calls are permissible
in the deferred mode.
\item Indexes returned by MxRSRV, MxFIND, and MxLONG
may not be assigned to other variables while the deferred mode is
in effect.
\item Dynamic memory may not be accessed while the deferred mode is in
effect.
\item The deferred mode is canceled by calling MxEXEC.
\end{enumerate}
\begin{verbatim}
CALL MDWAIT ()
CALL MCWAIT ()
\end{verbatim}

\subsection{Execute Deferred Memory Requests (MDEXEC/MCEXEC)}
NOTE: This capability has been removed.  Calling \code{MDEXEC} or \code{MCEXEC} will have no affect.

MxEXEC causes all deferred mode memory requests to be satisfied
with a single call to the operating system for the required memory. The
deferred mode is also canceled. MDEXEC and MCEXEC are synonyms.

After returning from MxEXEC, all indexes to array space which was
deferred are updated.
\begin{verbatim}
CALL MDEXEC ()
CALL MCEXEC ()
\end{verbatim}

\subsection{Report storage information (MDMEMS/MCMEMS)}
MxMEMS reports numeric or character storage information. This information
may be useful for planning storage strategies during code development
and for flow control during code execution.
\begin{verbatim}
CALL MDMEMS (NSUA, NSUD, NSUV, NSULV)
CALL MCMEMS (NSUA, NSUD, NSUV, NSULV)
\end{verbatim}

\begin{argy}{NSUA}{INTEGER}{Write Only}
NSUA is the number of numeric/character storage units currently allocated
and not deferred.
\end{argy}

\begin{argy}{NSUD}{INTEGER}{Write Only}
NSUD is the number of numeric/character storage units currently deferred.
\end{argy}

\begin{argy}{NSUV}{INTEGER}{Write Only}
NSUV is the amount of void space in numeric or character storage units. Note
that MDMEMS and MCMEMS may be reporting the same space for NSUV, but in
different units.
\end{argy}

\begin{argy}{NSULV}{INTEGER}{Write Only}
NSULV is the size of the largest void space in numeric or character storage
units. Note that MDMEMS and MCMEMS may be reporting the same space for
NSULV, but in different units.
\end{argy}

\section{Development Aids}\label{sec:mdev}
The routines in this section are designed to aid the programmer during
development of a program, and probably would not be used during execution of
a mature program, except as a response to a memory manager error.

\subsection{List Storage Tables (MDLIST/MCLIST)}
MxLIST prints the contents of the memory manager's internal
tables. Section~\ref{sec:table} describes these tables. MDLIST and MCLIST
are synonyms.
\begin{verbatim}
CALL MDLIST (IUNIT)
CALL MCLIST (IUNIT)
\end{verbatim}

\begin{argy}{IUNIT}{INTEGER}{Read Only}
This is the FORTRAN unit number of the output device.
\end{argy}

\subsection{Print Dynamic Array (MDPRNT/MCPRNT)}
MxPRNT prints the contents of an individual numeric and
character array, respectively.
\begin{verbatim}
CALL MDPRNT (NAME, IUNIT, TYPE)
CALL MCPRNT (NAME, IUNIT, NGRUP)
\end{verbatim}

\begin{argy}{NAME}{CHARACTER$*$($*$)}{Read Only}
This is the name of the array to be printed.
\end{argy}

\begin{argy}{IUNIT}{INTEGER}{Read Only}
This is the FORTRAN unit number of the output device.
\end{argy}

\begin{argy}{TYPE}{CHARACTER$*$($*$)}{Read Only}
TYPE indicates the data type of the data to be printed; "R" for REAL, or
"I" for INTEGER.  Note that this is not necessarily the declared type of the
base array.
\end{argy}

\begin{argy}{NGRUP}{INTEGER}{Read Only}
NGRUP controls the number of characters that are grouped together without
intervening spaces. Since the storage is declared as a CHARACTER$*$ 1
array, NGRUP allows the user to format the output consistent with longer
character strings.
\end{argy}

\subsection{Debug Printing (MDDEBG/MCDEBG)}
Debug printing causes error messages to be printed by the memory manager at
the time an error is detected. The default is no debug printing --- errors
are detected, but are only reported when re\-quest\-ed by MxSTAT and
MxERPT. MDDEBG and MCDEBG are synonyms.
\begin{verbatim}
CALL MDDEBG (IUNIT)
CALL MCDEBG (IUNIT)
\end{verbatim}

\begin{argy}{IUNIT}{INTEGER}{Read Only}
IUNIT controls debug printing. A negative or zero value
turns debug printing off. A positive value of IUNIT will cause any error
messages to be printed to FORTRAN unit number IUNIT.
\end{argy}