path: root/src/RecoverGH.c

 /*@@
   @file      RecoverGH.c
   @date      Fri Jun 19 09:14:22 1998
   @author    Tom Goodale
   @desc 
   Contains the routines to do the internal checkpoint recovery.

   Currently can recover from:
     (1) One file containing recombined data
     (2) Multiple unrecombined files, where the current
	     number of processors and outputing processors
		 match those used to write the data.
   @enddesc 
   @history
   @hauthor Gabrielle Allen @hdate 19 Oct 1998      
   @hdesc Changed names ready for thorn_IO
   @endhistory
   @version $Id$
 @@*/

static char *rcsid = "$Id$";

#include <stdio.h>

#include "cctk.h"
#include "flesh.h"
#include "declare_parameters.h"
#include "GHExtensions.h"
#include "WarnLevel.h"
#include "Comm.h"
#ifdef CACTUSBASE_PUGH
#include "CactusBase/pugh/src/include/pugh.h"
#endif
#include "CactusBase/IOUtil/src/ioGH.h"
#include "ioFlexGH.h"


/* this one comes from RestoreFile.c */
int IOFlexIO_RestoreIEEEIOfile (cGH *GH, IOFile ifp,
                              int IOrecover_ioproc,
                              int IOrecover_ioproc_every,
                              int IOrecover_unchunked);


 /*@@
   @routine    IOFlexIO_RecoverGH
   @date       Fri Jun 19 09:22:52 1998
   @author     Tom Goodale
   @desc 
   Recovers a GH.
   @enddesc 
   @calls IOUtil_PrepareFilename IOFlexIO_RestoreIEEEIOfile
   @calledby   
   @history 
   @hauthor Gabrielle Allen
   @hdate Thu Jul  2 18:17:59 1998
   @hdesc Restore the physical time and iteration count, pass myproc to
          IEEEIOparamRestore
	  Derives the filename from IOUtil_PrepareFilename (in thorn IOUtil)
   @hauthor Gabrielle Allen @hdate Oct 17 1998
   @hdesc Added input of (some) GH structure variables
   @endhistory 

@@*/

int IOFlexIO_RecoverGH (cGH *GH, const char *basename, int called_from)
{
#ifdef CACTUSBASE_PUGH

  DECLARE_PARAMETERS
  IOFile ifp;
  char ftmp [1024], fname [1024];
  int proc, nprocs, myproc;
  int index;
  int is_IEEEIO_file;
  Long nels_stored;
  int nt_stored;
  int iteration_stored;
  int file_ioproc, file_ioproc_every;
  int file_nprocs, file_unchunked;
  CCTK_INT4 tmpInt;
  char msg [512];
  pGH *pughGH;
  ioGH *ioUtilGH;
  cTimer total_time, dataset_time, param_time;
#ifdef MPI
  CCTK_INT info [3];
#endif


  /* Get the handles for PUGH and IOUtil extensions */
  pughGH = (pGH *) GH->extensions [CCTK_GetGHExtensionHandle ("PUGH")];
  ioUtilGH = (ioGH *) GH->extensions [CCTK_GetGHExtensionHandle ("IO")];

  /* identify myself */
  nprocs = CCTK_GetnProcs (GH);
  myproc = CCTK_GetMyProc (GH);

  /* initialize timers */
  CactusResetTimer (&total_time);
  CactusResetTimer (&dataset_time);
  CactusResetTimer (&param_time);

  CactusStartTimer (&total_time);

  /* Examine base file to find whether recovering from
   * one (recombined) file or from multiple files
   */

  if (myproc == 0) {
    
    /* Determine name of base file
       NOTE: As we don't know whether the file is chunked or not
             we need to try both file names. */
    /* at first try with current chunking mode */
    file_unchunked = ioUtilGH->unchunked;
    IOUtil_PrepareFilename(GH, basename, fname, called_from, 0, file_unchunked);
    if (called_from == CP_RECOVER_DATA)
      strcat (fname, ".chkpt");
    strcat (fname, ".ieee");

    if (IO_verbose)
      printf ("Opening file %s\n", fname);

    /* Open file, make sure the file is valid */
    ifp = IEEEopen (fname, "r");
    if (IOisValid (ifp))
      is_IEEEIO_file = 1;
    else {
      if (IO_verbose)
        printf ("Cannot open file '%s'\n", fname);

      /* now try with the other chunking mode */
      file_unchunked = ! ioUtilGH->unchunked;
      IOUtil_PrepareFilename (GH, basename, fname, called_from,
                          0, file_unchunked);
      if (called_from == CP_RECOVER_DATA)
        strcat (fname, ".chkpt");
      strcat (fname, ".ieee");
      if (IO_verbose)
        printf ("Trying now file '%s'...\n", fname);

      /* Open file, make sure the file is valid */
      ifp = IEEEopen (fname, "r");
      is_IEEEIO_file = IOisValid (ifp);
    }
  }
    
  if (myproc == 0 && is_IEEEIO_file) {
    /* Now determine how the data was written */

    /* Read nioprocs used to write data */
    index = IOreadAttributeInfo (ifp, "GH$ioproc_every", &nt_stored, &nels_stored);
    if (index >= 0 && nt_stored == FLEXIO_INT4 && nels_stored == 1) {
      IOreadAttribute (ifp, index, &tmpInt);
      file_ioproc_every = tmpInt;
    } else {
      CCTK_WARN (1, "Unable to restore GH$ioproc_every. "
                 "Assuming it is nprocs and continuing");
      file_ioproc_every = nprocs;
    }

    /* Read nprocs used to write data */
    index = IOreadAttributeInfo (ifp, "GH$nprocs", &nt_stored, &nels_stored);
    if (index >= 0 && nt_stored == FLEXIO_INT4 && nels_stored == 1) {
      IOreadAttribute (ifp, index, &tmpInt);
      file_nprocs = tmpInt;
    } else {
      CCTK_WARN (1, "Unable to restore GH$nprocs. "
                 "Assuming it is 1 and continuing");
      file_nprocs = 1;
    }
    
    /* Determine whether data is chunked or unchunked
       We could derive this from the filename itself but just to be sure ... */
    index = IOreadAttributeInfo (ifp, "unchunked", &nt_stored, &nels_stored);
    if (index >= 0 && nt_stored == FLEXIO_INT4 && nels_stored == 1) {
      IOreadAttribute (ifp, index, &tmpInt);
      file_unchunked = tmpInt;
    } else {
      sprintf (msg, "Unable to restore 'unchunked' attribute. "
               "Assuming it is %s and continuing",
               file_unchunked ? "true" : "false");
      CCTK_WARN (1, msg);
    }

    /* If we restore from multiple files
     * the number of processors must match.
     */
 
    if (file_ioproc_every == nprocs || file_unchunked) {
      if (IO_verbose)
        printf ("Recovering from one %s file\n",
                file_unchunked ? "unchunked" : "chunked");
    } else {
      if (file_nprocs != nprocs) {
        sprintf (msg, "Must restart on %d processors with multiple files "
                 "or recombine them", file_nprocs);
        CCTK_WARN (0, msg);
      }
      if (IO_verbose)
        printf ("Recovering from %d chunked files\n",
                nprocs / file_ioproc_every + (nprocs % file_ioproc_every?1:0));
    }
  }
  
#ifdef MPI
  /* Broadcast chunking mode to all processors from processor zero */
  info [0] = is_IEEEIO_file;
  info [1] = unchunked;
  info [2] = file_ioproc_every;
  CACTUS_MPI_ERROR (MPI_Bcast (info, 3, PUGH_MPI_INT, 0,
                    pughGH->PUGH_COMM_WORLD));
  is_IEEEIO_file = info [0];
  unchunked = info [1];
  file_ioproc_every = info [2];
#endif

  /* return here to IOUtil if no valid file could be found */
  if (! is_IEEEIO_file) {
    if (myproc == 0) {
      sprintf (msg, "No valid IEEEIO file '%s' found !", fname);
      CCTK_WARN (2, msg);
    }
    return (-1);
  }

  /* Determine the IO processors for each node and the corresponding
     checkpoint file */
  file_ioproc = myproc - (myproc % file_ioproc_every);
  IOUtil_PrepareFilename (GH, basename, fname, called_from,
                      file_ioproc/file_ioproc_every, file_unchunked);
  if (called_from == CP_RECOVER_DATA)
    strcat (fname, ".chkpt");
  strcat (fname, ".ieee");

  /* Open chunked files on other IO processors */
  if (myproc != 0 && myproc == file_ioproc) {
      
    if (IO_verbose)
      printf ("Opening chunked file '%s' on processor %d.\n",
              fname, myproc);

    /* Open file, make sure the file is valid */
    ifp = IEEEopen (fname, "r");
    if (! IOisValid (ifp)) {
      sprintf (msg, "Cannot open file '%s' on processor %d",
               fname, myproc);
      CCTK_WARN (0, msg);
    }
  }
 
  /* Restore the data */
  if (IO_verbose && myproc == 0)
    printf ("Recovering %schunked data with ioproc %d, ioproc_every %d.\n",
            file_unchunked ? "un" : "", file_ioproc, file_ioproc_every);

  CactusStartTimer (&dataset_time);
  IOFlexIO_RestoreIEEEIOfile (GH, ifp, file_ioproc, file_ioproc_every,
                            file_unchunked);
  CactusStopTimer (&dataset_time);

  /* Close the file. */
  if (myproc == file_ioproc) {
    if (IO_verbose)
      printf ("Closing file '%s' after recovery.\n", fname);
    IOclose (ifp);
  }
 
  if (called_from == CP_RECOVER_DATA) {
    /* Must read in parameters and scalars on all processors. */
    CactusStartTimer (&param_time);
    for (proc = file_ioproc;
         proc < file_ioproc+file_ioproc_every && proc < nprocs;
         proc++) {
    
      /* Only have the file open by one proc at any time. */
      if (proc == myproc) {
      
        /* Open file, make sure the file is valid */
        ifp = IEEEopen (fname, "r");
        if (! IOisValid (ifp)) {
          sprintf (msg, "Cannot open checkpoint file '%s' on processor %d",
                   fname, myproc);
          CCTK_WARN (0, msg);
        }

        /* Restore the parameters. */
        if (IO_verbose)
          printf ("Recovering parameters on processor %d.\n", myproc);
/*** FIXME ***/
#if	0
        IO_IEEEIOparamRestore (ifp, myproc);
      
        /* Restore the structure variables. */
        if (IO_verbose)
          printf ("Recovering GH variables.\n");
        IO_IEEEIOStructRestore (GH, ifp);
#endif

        /* Restore global variables */
      
        /* Get the iteration number. */
        if (IO_verbose)
          printf ("Recovering iteration number.\n");
        index = IOreadAttributeInfo (ifp, "iteration", &nt_stored, &nels_stored);
	
        if (index >= 0 && nt_stored == FLEXIO_INT4 && nels_stored == 1) {
          IOreadAttribute (ifp, index, &iteration_stored);
          if (IO_verbose)
            printf ("Iteration number is %d\n", (int) iteration_stored);
          GH->cctk_iteration = iteration_stored;
        } else
          printf ("*Warning* Unable to restore iteration number\n");
      
        /* Close the file. */
        if (IO_verbose)
          printf ("Closing '%s' after recovery.\n", fname);
        IOclose (ifp);
      }
    
      /* Synchronise all processors */
      CCTK_Barrier (GH);
    }
    CactusStopTimer (&param_time);
  }

  /* print timing output */
  if (IO_verbose && called_from == CP_RECOVER_DATA && myproc == 0) {
    printf (
      "----------------------------------------------------------------\n");
/*** FIXME: choose right component of basic[] ***/
    printf ("Time to restore data:            %10.3lf sec\n",
            dataset_time.total.basic [0]);
    printf ("Time to restore parameters:      %10.3lf sec\n",
            param_time.total.basic [0]);
    printf ("Time to recover from checkpoint: %10.3lf sec\n",
            total_time.total.basic [0]);
  }

  return (0);
#endif /* CACTUSBASE_PUGH */
}