README

CPPC@DOC@v0.8.1

A. INSTALLATION
________________________________________________________________________________

To install CPPC on your computer, do the following:

1/ Untar the archive and move into the CPPC directory.
2/ Type "./configure" to check for system components needed to compile
   different CPPC plugins. The following options are available:

  --prefix=<path>           	Install CPPC files under <path> (see step 4).
                            	Defaults to `pwd`/Build.
  --disable-fortran         	Disable support for Fortran 77 applications.
  --with-mpi=yes/no/<path>  	<path> to mpicc for MPI configuration. MPI is 
			    	enabled by default. Note that compiling CPPC with 
			    	MPI support will prevent the library from
			    	checkpointing sequential applications (an error
			    	complaining about MPI not being initialized will be
			    	raised during runtime).
  --with-hdf5=yes/no/<path> 	<path> to h5cc for HDF5 configuration. Disabling
			    	the HDF5 writer currently renders the library
			    	unusable and causes runtime errors.
  --with-xerces=yes/no/<path>	Search for Xerces-C in <path>. Disabled by default
			    	since CPPC 0.8.
  --with-xerces-inc=<path>  	Search for Xerces include files under <path>.
  --with-xerces-lib=<path>  	Search for Xerces library files under <path>.

3/ Type "make" to build CPPC.

4/ Typing "make install" will copy CPPC essential files to the destination
   directory specified using the --prefix option when running ./configure. This
   will create three subdirectories:

   bin       Containing the compiler scripts.
   include   Header files needed by other programs to use CPPC. These
             include the C and Fortran interfaces (cppc.h and cppcf.h
             respectively), various datatype definitions
             (data/cppc_types.h), and the semantic files to direct
             compiler operation.
   lib       The shared libraries for CPPC, both C and Fortran 77
             versions (libcppc-c.so and libcppc-f.so, respectively).


B. USING CPPC
________________________________________________________________________________

In order to introduce fault-tolerance through CPPC, the code of your application
needs to be changed so that it communicates with the runtime library, passing
information about variables that need to be dumped in the next checkpoint, where
to create the state files, etc. Also, flow-control structures are placed to
control the re-execution of certain critical portions of code at restart. This
will enable the recovery of certain non-portable parts of data, such as MPI
communicators or open files, that cannot be just stored as binary data in a
state file. As the insertion of these function calls and flow-control structures
would imply significant effort by the end user, the CPPC runtime library is
distributed along with a compiler that helps the user by automatically
performing necessary transformations to the original application code. 

The automatic communication analysis and checkpoint insertion sported from
v0.7.x of the compiler are fairly stable, and a huge improvement over v0.6.x,
but might still not work well with all applications. You can deactivate both and
rely on manual directive insertion if you experience trouble with the analyses
or the output code.

The compiler-provided directives, in case the user wants to manually guide the
compiler operation, are:

  * cppc execute/end execute: These mark a block of code that needs to be
    re-executed upon application restart. This directive should be
    inserted when you want to recover state by re-execution, instead of
    saving/reading it to/from disk.

  * cppc checkpoint: This directive may be used for manually marking
    points where the state is dumped to a state file. If so, it must be
    inserted at safe points in the application: locations where there are
    neither in-transit, nor orphan messages between processes. In a typical
    example, a checkpoint should not be placed in between an MPI_Send()
    and its matching MPI_Recv(). If this happened, the message would not
    be resent upon application restart, but the destination process would
    still expect to receive it.

  * cppc checkpoint loop: This directive is the same as the previous one,
    except that you mark a loop in whose body you want the checkpoint
    inserted. The compiler will take into account communications between
    processes and insert a checkpoint in the first safe point it can find
    inside the loop body.

Checkpoints are usually inserted inside loops. You usually won't expect a state
file to be dumped at each loop iteration. Checkpoint frequency can be controlled
by using the CPPC/Controller/Frequency parameter (see the EXAMPLE APPLICATION
section).

If the automatic checkpoint/communication analyses work for your application,
then there is nothing you will need to manually alter in its code. Else, the
steps that you want to take in order to integrate your application with the
CPPC framework are:

1/ Decide where you want to dump the state. Place checkpoint or checkpoint loop
   directives in those spots. If the communication analysis is not used, be sure
   to check that those spots are safe points as defined above. Bear in mind that
   the code after a checkpoint and up to the end of the application will be the
   code being executed upon application restart.

2/ Compile the application linking with the appropriate CPPC dynamic library.

C. UNDERSTANDING SEMANTIC DESCRIPTIONS OF FUNCTION FILES
--------------------------------------------------------------------------------

Some of the transforms implemented by the CPPC compiler are semantic-directed.
These require certain semantic information about which function calls implement
certain semantics and how. Since source code in imperative languages is not
intrinsically semantic, it is necessary to provide semantic information to the
compiler. The mechanism for supplying this information is extensible and not
tied to specific implementations of a semantic, fulfilling the portability
target present in CPPC.

The compiler reads this semantic information from XML files having the following
format:

<cppc-semantic-module>

  <module name="...">

    <function name="...">

      <!-- Block used for feeding the compiler with flow control
           data, which improves the efficiency of the analysis for
          variable registration. -->
      <input parameters="..."/>
      <output parameters="..."/>
      <input-output parameters="..."/>

      <!-- Block containing semantic information about the roles
           this function fulfills. -->
      <semantics>

        <semantic role="...">
          <attribute name="..." value="..."/>
          <attribute .../>
        </semantic>

        <semantic ...   />

      </semantics>

    </function>

    <function ... />

  </module>

</cppc-semantic-module>

Take a look at the "linux-module.xml" in the "compiler/semantic-meta"
subdirectory. Each of this files is called a "semantic module", and contains
semantic information about a collection of related procedures. Each procedure
can have any number of semantic tags, each one of them having a "role"
identifier and, optionally, a number of attributes that give further information
on how the procedure implements the given semantic.

Look, for example, at the UNIX fopen and open functions. Both accept two
parameters: a path string and modifiers on how to open the file, and both return
a reference to the opened file. This information is passed to the compiler in
the semantic description of the functions. This reference is different for each
one of them, a FILE * for the fopen function and an integer for open. The given
"DescriptorType" attribute tells the compiler which type of file descriptor does
the function return.

The current release of CPPC includes semantic modules for common POSIX functions,
MPI, and common Fortran intrinsics. If you intend to provide support for a specific
family of functions not included with CPPC, follow these steps:

1/ Create a new XML file, and insert the necessary <cppc-semantic-module> and
   <module> tags. Give the module a name that fits the function family which
   documents.

2/ Add the functions in that family (or simply the ones you need). For each
   function, add a classification of its parameters in terms of data flow:
   input/output/input-output.

3/ Add the semantic roles to the relevant functions. Current recognized roles
   are:

   - CPPC/Nonportable: Identifies a procedure as having a non portable outcome,
     which has to be recovered by code re-execution rather than by storing it
     into the save file.

   - CPPC/IO/Open and CPPC/IO/Close: For marking functions that open/close
     sequential access streams. It is required if you want CPPC to automatically
     track the state of such media. Attributes for the semantic role are the
     "Path" to the file being opened, the "FileDescriptor" being returned and
     its "DescriptorType". You can see an example in the semantic modules
     bundled with this distribution.

   - CPPC/Comm/Initializer: Used for identifying functions that initialize the
     parallel system. Since CPPC might perform interprocess communications at
     restart time, it needs to locate the point in the code where the comm
     system is started.

   - CPPC/Comm/Ranker: Used for identifying functions that return the rank of a
     parallel process. It is necessary to correctly perform the communications
     analysis.

   - CPPC/Comm/Sizer: Used for identifying functions that return the size of
     a communicator in a parallel execution. It is necessary to correctly
     perform the communications analysis.

4/ If your semantic module documents a widely used family of functions, please
   send it to gabriel.rodriguez@udc.es so it can be added to the next version of
   the CPPC compiler.


D. EXAMPLE APPLICATION
________________________________________________________________________________

In the 'Example' directory there's an example application which solves an
equation system using the Gauss-Jordan method. There are two different versions
of this application. The system will build the appropriate one depending on
whether you have chosen to build the MPI or the sequential version of CPPC.

1/ Enter the 'Example' directory.

2/ Execute 'make' to build all necessary targets.

3/ The syntax accepted by the "gauss" executable:

   gauss <dimA> <dimB> <file1> <file2> <file3>

   Where:

   dimA     Rows in the grid of processors.
   dimB     Columns in the grid of processors.
   file1    File containing the system matrix.
   file2    File containing the independent vector.
   file3    Output file.

   Hence, if the equation system to be solved is A.b=x, then <file1>=A,
   <file2>=b and x=<file3>.

4/ Execution of the 'gauss' file automatically uses the CPPC library. To achieve
   a suitable configuration for CPPC, a number of parameters must be defined. In
   the 'Example' directory you will find a file called 'cppc_config.example'.
   This is an XML configuration file. Edit it and change it at will. Every CPPC
   plugin defines its own configuration parameters, and to allow for this the
   XML is structured in modules and parameters. The list of currently accepted
   options is:

   Module: Controller
     RootDir                    Directory where CPPC will store and search for
                                checkpoint files.
     ApplicationName            Subdirectory of "RootDir" where checkpoint files
                                will be effectively stored.
     Restart                    Controls whether the application must be
                                restarted. It is highly recommended to set this
                                to false, and enforce restart using command-line
                                parameters.
     Frequency                  Number of calls to CPPC_Do_checkpoint() between
                                effective state dumping. Set to -1 to disable
				checkpointing.
     FullFrequency		Number of effective checkpoint dumps in between two
				full checkpoints. Set to 1 to disable incremental
				checkpointing.
     CheckpointOnFirstTouch     If this is set to "true", the state dumping will
                                be performed every first time that a particular
                                call to CPPC_Do_checkpoint() is reached.
     Suffix                     Suffix for the files being generated.
     StoredRecoverySets		Maximum number of recovery sets stored (per node)
				before beginning deletion of the older ones as new
				ones are being created. A recovery set is composed
				of a full checkpoint plus a number of incremental to
				be applied on top of it. Set to 1 to store only the
				latest recovery set (not recommended if using MPI).
				Set to -1 to disable deletion of created checkpoints.
     DeleteCheckpoints          If set to "true", stored checkpoints will
                                be removed upon succesful execution.

   Module: Writer
     Type                       Type of writer used. See 4.1/.
   Submodule: HDF-5
     Compression                Type of compression used by the HDF-5 writer.
                                See 4.2/.
     CompressionLevel           Level of compression used by the ZLib
                                compressor. It is an integer in the 0-9 range,
                                9 being the maximum compression level.
     MinimumSize                An integer defining the minimum data size for
                                enabling compression. E.g., if set to 100, any
                                array with less than a hundred elements will
                                not be compressed, but stored directly.

4.1/ Currently theres just one valid value for this parameter:
      3                         HDF-5 writer.

4.2/ Two valid values:
      None                      No compression.
      ZLib                      ZLib compression. When this is enabled, it is
                                mandatory to also include the
                                HDF-5/CompressionLevel and HDF-5/MinimumSize
                                parameters.


** NOTE: In order for CPPC to find the configuration file, copy it to
   $HOME/.cppc_config, or use the CPPC/Configuration/FilePath command-line
   parameter (see 5). A configuration file example can be found under
   Example/cppc-config.example. 

   For compatibility with older versions, CPPC also accepts XML configuration files.
   An example of the equivalent XML configuration file can be found under
   Example/cppc-config.xml. The DTD for the configuration file can be found under
   include/util/xml/cppc-config.dtd. XML configuration files are deprecated and will
   go away at some point.

5/ CPPC accepts command-line parameters that overwrite those set in the configuration
   file. Used syntaxis, supposing we are running the 'gauss' application, is:

     gauss <gauss-params> -CPPC,<CPPC-Param1>,<CPPC-Param2>,...

   Where every CPPC-Param has the following structure:

     CPPC/<Module>/<Submodule>/.../<Parameter>=<value>

   For instance, if we want to overwrite the configuration file placement to be
   /tmp/cppc-config and command CPPC to restart the application, we will use:

     gauss <gauss-params> -CPPC,CPPC/Configuration/FilePath=/tmp/cppc-config,\
       CPPC/Controller/Restart=true

   Any parameter seen in the configuration file can be overwritten using a
   command-line parameter.

   Please note that the Fortran 77 version of the compiler does not accept
   command-line parameters, due to the Fortran 77 lack of a standard way for
   accessing them. It will be implemented in future versions of CPPC, shall we
   find a portable way to do it.

E. COMPILER
________________________________________________________________________________

The compiler used by CPPC has been ported to the Cetus framework so that it is
easier to use and more portable than the previous SUIF version in CPPC v0.1.x.
Also, the Fortran 77 compiler is now working (however, the Fortran 77 parser is
somehow experimental and won't parse any Fortan 77 code out there).

Two script files, bin/cppc_cc and bin/cppc_fc will be copied to the installation
directory. They run the C and Fortran 77 compilers, respectively. Executing
them without parameters will give a brief description of its functionalities.

Please note that the compiler, specially the Fortran 77 version, is highly
experimental code. It is based on a custom Cetus distribution. Extensions have
been made to the IR so that it supports Fortran 77 structures. Any bug report
concerning the Fortran 77 compiler will be specially appreciated.

F. KNOWN ISSUES
_________________________________________________________________________________

1/ Fortran code not yet accepted by the Fortran compiler:

   - Octal numbers such as "0777". This number will get translated as "777"
     instead of "511", resulting in code malfunction. If you are using any
     octal integer representation, be sure to check the resulting code after
     running the precompiler to correct this issue.

   - "CONTINUE" statements. Ending a labeled loop with "<label> CONTINUE" is
     ok, but executable "CONTINUE" statements are not yet supported. Note that
     ending multiple loops with the same label with just one CONTINUE statement
     is not yet supported.

   - "ELSE IF" constructs. If you want to use an ELSEIF construct, write it
     as "ELSEIF". Otherwise, the new if has to be properly closed with an
     END IF (ENDIF) statement.

2/ Compiler analyses might fail if a nonportable function is inserted into
   a non-do loop (Fortran) or a non standard for loop (C) (being standard the
   usual "for( var = value; <condition>; <step> )" construct.

3/ Currently the CPPC framework does only support the Unix standard filesystem,
   described in headers "fcntl.h", "sys/stat.h", "sys/types.h" and "unistd.ht".
   If you are interested in an adapter for other filesystems, please send
   a petition.

4/ CPPC needs the HDF-5 libraries to be installed in order to build a checkpoint
   writer. If it cannot find HDF-5 execution will fail when trying to create
   checkpoint files.

5/ The Fortran compiler used to build the Fortran 77 interface needs to
   implement the FSEEK GNU intrinsic, in order to be able to reposition files
   upon application restart. For this reason, the Fortran interface in CPPC is
   not compatible with GCC4 below v4.3 (it works with GCC3 and g77, though).

G. COPYRIGHT NOTICES
_________________________________________________________________________________

- This distribution includes a modified copy of the Cetus 0.5 Standard
  Version, by Troy A. Johnson and Sang-Ik Lee. Modifications were done to
  include Fortran support and adapt it to the CPPC compiler. A copy of the
  current Cetus Standard Version can be obtained at:

    http://cetus.ecn.purdue.edu/

  A copy of the Cetus License can be found under compiler/lib/cetus-license.

- This distribution includes a copy of the Antlr software by Terence Parr. It
  also uses this software, and software generated by the Antlr framework, for
  source-to-source transformations. A copy of the current Antlr distribution
  can be obtained at:

    http://www.antlr.org

H. LICENSE
_________________________________________________________________________________

This program is free software; you can redistribute it and/or modify it under the
terms of the GNU General Public License as published by the Free Software Foundation;
either version 2 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE. See the GNU General Public License for more details.

A copy of the GNU General Public License should be located in the 'License' file;
if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330,
Boston, MA 02111-1307 USA.

I. CONTACTING AUTHORS
_________________________________________________________________________________

This implementation of the CPPC framework has been done by Gabriel Rodríguez.
Send feedback to 'gabriel.rodriguez@udc.es'.