al_io.c File Reference

Miscellaneous functions for IO operations. More...

#include "al_hidden.h"
#include <string.h>

Include dependency graph for al_io.c:

Go to the source code of this file.

Functions
int	AL_File_open (char *filename, int sz_ptr)
int	AL_File_close (int sz_ptr)
int	AL_Write_array (void *va, int sz_ptr, int istag)
int	AL_Read_array (void *va, int sz_ptr, int istag)
int	AL_Write_header (void *vbuffer, int nelem, AL_Datatype type, int sz_ptr)
long long	AL_Get_offset (int sz_ptr)
int	AL_Set_offset (int sz_ptr, long long offset)

Variables
SZ *	sz_stack [AL_MAX_ARRAYS]
int	stack_ptr [AL_MAX_ARRAYS]

Detailed Description

Miscellaneous functions for IO operations.

Author

A. Malagoli (University of Chicago)

A. Mignone (migno.nosp@m.ne@p.nosp@m.h.uni.nosp@m.to.i.nosp@m.t)

G. Muscianisi (g.mus.nosp@m.cian.nosp@m.isi@c.nosp@m.inec.nosp@m.a.it)

Date

Aug 26, 2012

Definition in file al_io.c.

Function Documentation

int AL_File_close

(

int

sz_ptr

)

Close a file associate with and array

Parameters

[in]

sz_ptr

integer pointer to array descriptor

Definition at line 83 of file al_io.c.

{
  int myrank;
  int errcode;
  SZ *s;

  s = sz_stack[sz_ptr];
  myrank = s->rank;

  errcode =  MPI_File_close(&(s->ifp));

  /* DIAGNOSTICS */
#ifdef DEBUG
  int myid, len;
  char es[128];
  if( errcode ){
     MPI_Comm_rank(MPI_COMM_WORLD, &myid);
     MPI_Error_string(errcode, es, &len);
     printf("Errcode from MPI_File_close: %d | %s\n", errcode, es);
  }
  printf("myid %d, AL_File_close: Closed file, errcode %d \n",myid, errcode);
#endif

  return (int) AL_SUCCESS;
}

Here is the caller graph for this function:

int AL_File_open	(	char *	filename,
		int	sz_ptr
	)

Open a file associated with a distributed array

Parameters

[in]	filename	name of the file
[in]	sz_ptr	integer pointer to the distributed array descriptor

Returns

ifp pointer to integer file pointer

Definition at line 27 of file al_io.c.

{
  int myrank, nproc;
  int errcode;
  MPI_File ifp;
  MPI_Comm comm;
  SZ *s;

  /* DIAGNOSTICS
    Check that sz_ptr points to an allocated SZ
  */
  if( stack_ptr[sz_ptr] == AL_STACK_FREE){
    printf("AL_Decompose: wrong SZ pointer\n");
  }

  s = sz_stack[sz_ptr];

  myrank = s->rank;
  nproc = s->size;

  comm = s->comm;

  MPI_Barrier(comm);

  errcode = MPI_File_open(comm, filename,
        MPI_MODE_CREATE | MPI_MODE_RDWR | MPI_MODE_UNIQUE_OPEN,
        MPI_INFO_NULL, &ifp);

  s->io_offset = 0;
  s->ifp = ifp;

  /* DIAGNOSTICS */
#ifdef DEBUG
  int myid, len;
  char es[128];
  MPI_Comm_rank(MPI_COMM_WORLD, &myid);
  if( errcode ){
     MPI_Error_string(errcode, es, &len);
     printf("Errcode from MPI_File_open: %d | %s\n", errcode, es);
  }
  printf("myid %d, AL_File_open: Opened file %s\n",myid,filename);
#endif

  return (int) AL_SUCCESS;
}

Here is the caller graph for this function:

long long AL_Get_offset

(

int

sz_ptr

)

Definition at line 327 of file al_io.c.

{
  SZ *s;

  if( stack_ptr[sz_ptr] == AL_STACK_FREE){
    printf("AL_Get_offset: wrong SZ pointer\n");
  }
  s  = sz_stack[sz_ptr];
  return s->io_offset;
}

Here is the caller graph for this function:

int AL_Read_array	(	void *	va,
		int	sz_ptr,
		int	istag
	)

Read a distributed array to a file in parallel using synchronous and collective IO operations

Parameters

[in]	buffer	pointer to the buffer to write
[in]	sz_ptr	integer pointer to a distributed array descriptor
[in]	istag	set it to -1 for centred variables, set to 0,1,2 for staggered field in the x,y,z direction

Bugs fixed on Aug 26, 2012:

• nelem has been declared long long in order to manage dbl and flt output 'single_file' in which each PLUTO variable is >= 4GB
• the updating of nelem after the writing of a variable has been modified from ''nelem *= (long long)(s->arrdim[i] +(istag == i));'' to ''nelem *= (long long)(s->arrdim[i]);'' because when a staggered variable is written, in the sz_ptr descriptor there is the information about the really number of element of this variable. This appens because the function AL_Set_staggered_dim is called for a staggered array.

Definition at line 209 of file al_io.c.

{
  char *a;
  register int i;
  int myrank;
  long long nelem;
  SZ *s;
  MPI_Status status;
  AL_Datatype gsub_arr, lsub_arr;


  MPI_File ifp;
  MPI_Offset offset;
  int size;

  a = (void *)va;

  s = sz_stack[sz_ptr];
  myrank = s->rank;

  ifp = s->ifp;

  offset   = s->io_offset;
  if (istag == -1){
    gsub_arr = s->gsubarr;
    lsub_arr = s->lsubarr;
  }else{
    gsub_arr = s->gsubarr_stag[istag];
    lsub_arr = s->lsubarr_stag[istag];
  }

  MPI_Barrier(s->comm);

  MPI_File_set_view(ifp, offset, MPI_BYTE, gsub_arr,
            "native", MPI_INFO_NULL);
  MPI_File_read_all(ifp, a, 1, lsub_arr, &status);
  
  MPI_Type_size( s->type, &size);

  nelem = 1;
  for(i = 0; i < s->ndim; i++) {
/* -------------------------------------------------------------------- */
/*!  Bugs fixed on Aug 26, 2012:
  \li nelem has been declared long long in order to manage dbl and flt 
     output 'single_file' in which each PLUTO variable is >= 4GB
  \li the updating of nelem after the writing of a variable has been 
     modified from  ''nelem *= (long long)(s->arrdim[i] +(istag == i));''
     to ''nelem *= (long long)(s->arrdim[i]);'' because when a staggered 
     variable is written, in the sz_ptr descriptor there is the 
     information about the really number of element of this variable.  
     This appens because the function AL_Set_staggered_dim is called 
     for a staggered array. */
/* -------------------------------------------------------------------- */
/*    nelem *= (long long)(s->arrdim[i] + (istag == i)); */
    nelem *= (long long)(s->arrdim[i]); 
  }
  s->io_offset += (long long)(size)*nelem;

  return (int) AL_SUCCESS;
}

Here is the caller graph for this function:

int AL_Set_offset	(	int	sz_ptr,
		long long	offset
	)

Definition at line 342 of file al_io.c.

{
  SZ *s;

  if( stack_ptr[sz_ptr] == AL_STACK_FREE){
    printf("AL_Get_offset: wrong SZ pointer\n");
  }
  s  = sz_stack[sz_ptr];
  s->io_offset = offset;
  return (int)AL_SUCCESS;
}

Here is the caller graph for this function:

int AL_Write_array	(	void *	va,
		int	sz_ptr,
		int	istag
	)

Write a distributed array to a file in parallel using synchronous and collective IO operations

Parameters

[in]	buffer	pointer to the buffer to write
[in]	sz_ptr	integer pointer to the distributed array descriptor
[in]	istag	set it to -1 for centred variables, set to 0,1,2 for staggered field in the x,y,z direction

Bugs fixed on Aug 26, 2012:

• nelem has been declared long long in order to manage dbl and flt output 'single_file' in which each PLUTO variable is >= 4GB
• the updating of nelem after the writing of a variable has been modified from ''nelem *= (long long)(s->arrdim[i] +(istag == i));'' to ''nelem *= (long long)(s->arrdim[i]);'' because when a staggered variable is written, in the sz_ptr descriptor there is the information about the really number of element of this variable. This appens because the function AL_Set_staggered_dim is called for a staggered array.

Definition at line 115 of file al_io.c.

{
  char *a;
  register int i;
  int myrank;
  long long nelem;
  int errcode;
  SZ *s;

  MPI_Status status;
  AL_Datatype gsub_arr, lsub_arr;

  MPI_File ifp;
  MPI_Offset offset;
  int  size;
  

  a = (char *) va;

  s = sz_stack[sz_ptr];
  myrank = s->rank;
  ifp = s->ifp;

  offset   = s->io_offset;
  if (istag == -1){
    gsub_arr = s->gsubarr;
    lsub_arr = s->lsubarr;
  }else{
    gsub_arr = s->gsubarr_stag[istag];
    lsub_arr = s->lsubarr_stag[istag];
  }

  MPI_Barrier(s->comm);

  errcode = MPI_File_set_view(ifp, offset, MPI_BYTE, gsub_arr,
                    "native", MPI_INFO_NULL);
 
#ifdef DEBUG
    int myid, len;
    char es[256];
    MPI_Comm_rank(MPI_COMM_WORLD, &myid);
    if( errcode ){
      MPI_Error_string(errcode, es, &len);
      printf("Errcode from MPI_File_set_view: %d | %s\n", errcode, es);
    }
    printf("myid %d, offset file_set_view %lld\n", myid,offset);
#endif

  errcode = MPI_File_write_all(ifp, a, 1, lsub_arr, &status);

#ifdef DEBUG
    if( errcode ){
      MPI_Error_string(errcode, es, &len);
      printf("Errcode from MPI_File_write_all: %d | %s\n", errcode, es);
    }
#endif
    
  MPI_Type_size( s->type, &size);

  nelem = 1;
  for(i = 0; i < s->ndim; i++) {
/* -------------------------------------------------------------------- */
/*!  Bugs fixed on Aug 26, 2012:
  \li nelem has been declared long long in order to manage dbl and flt 
     output 'single_file' in which each PLUTO variable is >= 4GB
  \li the updating of nelem after the writing of a variable has been 
     modified from  ''nelem *= (long long)(s->arrdim[i] +(istag == i));''
     to ''nelem *= (long long)(s->arrdim[i]);'' because when a staggered 
     variable is written, in the sz_ptr descriptor there is the 
     information about the really number of element of this variable. 
     This appens because the function AL_Set_staggered_dim is called 
     for a staggered array. */
/* -------------------------------------------------------------------- */
/*    nelem *= (long long)(s->arrdim[i] +(istag == i)); */
    nelem *= (long long)(s->arrdim[i]);
  } 

  s->io_offset += (long long)(size)*nelem;

  return (int) AL_SUCCESS;
}

Here is the caller graph for this function:

int AL_Write_header	(	void *	vbuffer,
		int	nelem,
		AL_Datatype	type,
		int	sz_ptr
	)

Definition at line 285 of file al_io.c.

{
  char *buffer;
  int myrank;
  int nbytes, size;
  SZ *s;
  MPI_Status status;
  MPI_File ifp;


  MPI_Offset offset;
  MPI_Type_size(type, &size);
  buffer = (char *) vbuffer;

  nbytes = nelem*size;
  s      = sz_stack[sz_ptr];
  myrank = s->rank;

  ifp    = s->ifp;
  offset = s->io_offset;
  MPI_Barrier(s->comm);
  MPI_File_set_view(ifp, offset, MPI_BYTE, MPI_CHAR,
                    "native", MPI_INFO_NULL);
  if( myrank == 0 ){
/*    MPI_File_write(ifp, buffer, strlen(buffer), MPI_CHAR, &status);  */
    MPI_File_write(ifp, buffer, nelem, type, &status); 
  }

  s->io_offset += nbytes;


  return (int) AL_SUCCESS;
}

Here is the caller graph for this function:

Variable Documentation

int stack_ptr[AL_MAX_ARRAYS]

Definition at line 26 of file al_szptr_.c.

SZ* sz_stack[AL_MAX_ARRAYS]

Definition at line 18 of file al_szptr_.c.