Commit 8a1f77e3 authored by Philip Carns's avatar Philip Carns

carry over file ref comments to mpiio mod

parent 905e0ac4
......@@ -28,12 +28,62 @@
#include "darshan.h"
#include "darshan-mpiio-log-format.h"
/* The mpiio_file_runtime structure maintains necessary runtime metadata
* for the MPIIO file record (darshan_mpiio_file structure, defined in
* darshan-mpiio-log-format.h) pointed to by 'file_record'. This metadata
* assists with the instrumenting of specific statistics in the file record.
* 'hlink' is a hash table link structure used to add/remove this record
* from the hash table of MPIIO file records for this process.
*
* RATIONALE: the MPIIO module needs to track some stateful, volatile
* information about each open file (like the current file offset, most recent
* access time, etc.) to aid in instrumentation, but this information can't be
* stored in the darshan_mpiio_file struct because we don't want it to appear in
* the final darshan log file. We therefore associate a mpiio_file_runtime
* struct with each darshan_mpiio_file struct in order to track this information.
*
* NOTE: There is a one-to-one mapping of mpiio_file_runtime structs to
* darshan_mpiio_file structs.
*
* NOTE: The mpiio_file_runtime struct contains a pointer to a darshan_mpiio_file
* struct (see the *file_record member) rather than simply embedding an entire
* darshan_mpiio_file struct. This is done so that all of the darshan_mpiio_file
* structs can be kept contiguous in memory as a single array to simplify
* reduction, compression, and storage.
*/
struct mpiio_runtime_file
{
struct darshan_mpiio_file* file_record;
/* TODO: any stateful (but not intended for persistent storage in the log)
* information about MPI-IO access. If we don't have any then this struct
* could be eliminated.
*/
UT_hash_handle hlink;
};
/* The mpiio_file_runtime_ref structure is used to associate a MPIIO
* file handle with an already existing MPIIO file record. This is
* necessary as many MPIIO I/O functions take only a file handle as input,
* but MPIIO file records are indexed by their full file paths (i.e., darshan
* record identifiers for MPIIO files are created by hashing the file path).
* In other words, this structure is necessary as it allows us to look up a
* file record either by a pathname (mpiio_file_runtime) or by MPIIO file
* descriptor (mpiio_file_runtime_ref), depending on which parameters are
* available. This structure includes another hash table link, since separate
* hashes are maintained for mpiio_file_runtime structures and mpiio_file_runtime_ref
* structures.
*
* RATIONALE: In theory the file handle information could be included in the
* mpiio_file_runtime struct rather than in a separate structure here. The
* reason we don't do that is to handle the potential for an MPI implementation
* to produce a new file handle instance each time MPI_File_open() is called on a
* file. Thus there might be multiple file handles referring to the same
* underlying record.
*
* NOTE: there are potentially multiple mpiio_file_runtime_ref structures
* referring to a single mpiio_file_runtime structure. Most of the time there is
* only one, however.
*/
struct mpiio_runtime_file_ref
{
struct mpiio_runtime_file* file;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment