mirror/hdf5
2
0
Fork 0
mirror of https://github.com/HDFGroup/hdf5.git synced 2025-02-17 16:10:24 +08:00
Code Issues Packages Projects Releases Wiki Activity
b91e17b6cf
hdf5/src/H5Dmpio.c
Raymond Lu 7f2a3a97c3 [svn-r13997] Moved H5D_istore_lock and H5D_istore_unlock from H5D_istore_writevv and H5D_istore_readvv to 
H5D_chunk_write and H5D_chunk_read to avoid frequent lock and unlock and to improve some
performance.

Tested with h5committest on THG machines.
2007-07-21 14:50:43 -05:00

1984 lines
75 KiB
C
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* Copyright by The HDF Group. *
* Copyright by the Board of Trustees of the University of Illinois. *
* All rights reserved. *
* *
* This file is part of HDF5. The full HDF5 copyright notice, including *
* terms governing use, modification, and redistribution, is contained in *
* the files COPYING and Copyright.html. COPYING can be found at the root *
* of the source code distribution tree; Copyright.html can be found at the *
* root level of an installed copy of the electronic HDF5 document set and *
* is linked from the top-level documents page. It can also be found at *
* http://hdfgroup.org/HDF5/doc/Copyright.html. If you do not have *
* access to either file, you may request a copy from help@hdfgroup.org. *
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
/*
* Programmer: rky 980813
* KY 2005 revised the code and made the change to support and optimize
* collective IO support.
* Purpose: Functions to read/write directly between app buffer and file.
*
* Beware of the ifdef'ed print statements.
* I didn't make them portable.
*/
/****************/
/* Module Setup */
/****************/
#define H5D_PACKAGE /*suppress error about including H5Dpkg */
/***********/
/* Headers */
/***********/
#include "H5private.h" /* Generic Functions */
#include "H5Iprivate.h"
#include "H5Dpkg.h" /* Datasets */
#include "H5Eprivate.h" /* Error handling */
#include "H5Fprivate.h" /* File access */
#include "H5FDprivate.h" /* File drivers */
#include "H5MMprivate.h"
#include "H5Oprivate.h" /* Object headers */
#include "H5Pprivate.h" /* Property lists */
#include "H5Sprivate.h" /* Dataspaces */
#include "H5Vprivate.h" /* Vector */
#ifdef H5_HAVE_PARALLEL
/****************/
/* Local Macros */
/****************/
/* Macros to represent different IO options */
#define H5D_ONE_LINK_CHUNK_IO 0
#define H5D_MULTI_CHUNK_IO 1
#define H5D_ONE_LINK_CHUNK_IO_MORE_OPT 2
#define H5D_MULTI_CHUNK_IO_MORE_OPT 3
/***** Macros for One linked collective IO case. *****/
/* The default value to do one linked collective IO for all chunks.
If the average number of chunks per process is greater than this value,
the library will create an MPI derived datatype to link all chunks to do collective IO.
The user can set this value through an API. */
/* Macros to represent options on how to obtain chunk address for one linked-chunk IO case */
#define H5D_OBTAIN_ONE_CHUNK_ADDR_IND 0
#define H5D_OBTAIN_ALL_CHUNK_ADDR_COL 2
/* Macros to define the default ratio of obtaining all chunk addresses for one linked-chunk IO case */
#define H5D_ALL_CHUNK_ADDR_THRES_COL 30
#define H5D_ALL_CHUNK_ADDR_THRES_COL_NUM 10000
/***** Macros for multi-chunk collective IO case. *****/
/* The default value of the threshold to do collective IO for this chunk.
If the average number of processes per chunk is greater than the default value,
collective IO is done for this chunk.
*/
/* Macros to represent different IO modes(NONE, Independent or collective)for multiple chunk IO case */
#define H5D_CHUNK_IO_MODE_IND 0
#define H5D_CHUNK_IO_MODE_COL 1
/* Macros to represent the regularity of the selection for multiple chunk IO case. */
#define H5D_CHUNK_SELECT_REG 1
#define H5D_CHUNK_SELECT_IRREG 2
#define H5D_CHUNK_SELECT_NONE 0
/******************/
/* Local Typedefs */
/******************/
/* Combine chunk address and chunk info into a struct for better performance. */
typedef struct H5D_chunk_addr_info_t {
haddr_t chunk_addr;
H5D_chunk_info_t chunk_info;
} H5D_chunk_addr_info_t;
/* Combine all information that needs to know for collective MPI-IO of this selection. */
typedef struct H5D_common_coll_info_t {
hbool_t mbt_is_derived;
hbool_t mft_is_derived;
size_t mpi_buf_count;
haddr_t chunk_addr;
} H5D_common_coll_info_t;
/********************/
/* Local Prototypes */
/********************/
static herr_t
H5D_multi_chunk_collective_io(H5D_io_info_t *io_info,fm_map *fm,const void *buf,
hbool_t do_write);
static herr_t
H5D_multi_chunk_collective_io_no_opt(H5D_io_info_t *io_info,fm_map *fm,const void *buf,
hbool_t do_write);
static herr_t
H5D_link_chunk_collective_io(H5D_io_info_t *io_info,fm_map *fm,const void *buf,
hbool_t do_write,int sum_chunk);
static herr_t
H5D_inter_collective_io(H5D_io_info_t *io_info,const H5S_t *file_space,
const H5S_t *mem_space,haddr_t addr,
const void *buf, hbool_t do_write );
static herr_t
H5D_final_collective_io(H5D_io_info_t *io_info,MPI_Datatype*mpi_file_type,
MPI_Datatype *mpi_buf_type,
H5D_common_coll_info_t* coll_info,
const void *buf, hbool_t do_write);
static herr_t
H5D_sort_chunk(H5D_io_info_t * io_info,
fm_map *fm,
H5D_chunk_addr_info_t chunk_addr_info_array[],
int many_chunk_opt);
static herr_t
H5D_obtain_mpio_mode(H5D_io_info_t* io_info,
fm_map *fm,
uint8_t assign_io_mode[],
haddr_t chunk_addr[]);
static herr_t H5D_ioinfo_make_ind(H5D_io_info_t *io_info);
static herr_t H5D_ioinfo_make_coll_opt(H5D_io_info_t *io_info);
static herr_t H5D_ioinfo_make_coll(H5D_io_info_t *io_info);
static herr_t H5D_mpio_get_min_chunk(const H5D_io_info_t *io_info,
const fm_map *fm, int *min_chunkf);
static int H5D_cmp_chunk_addr(const void *addr1, const void *addr2);
static herr_t H5D_mpio_get_sum_chunk(const H5D_io_info_t *io_info,
const fm_map *fm, int *sum_chunkf);
/*********************/
/* Package Variables */
/*********************/
/*******************/
/* Local Variables */
/*******************/
/*-------------------------------------------------------------------------
* Function: H5D_mpio_opt_possible
*
* Purpose: Checks if an direct I/O transfer is possible between memory and
* the file.
*
* Return: Success: Non-negative: TRUE or FALSE
* Failure: Negative
*
* Programmer: Quincey Koziol
* Wednesday, April 3, 2002
*
*-------------------------------------------------------------------------
*/
htri_t
H5D_mpio_opt_possible( const H5D_io_info_t *io_info,
const H5S_t *mem_space, const H5S_t *file_space, const H5T_path_t *tpath)
{
int local_opinion = TRUE; /* This process's idea of whether to perform collective I/O or not */
int consensus; /* Consensus opinion of all processes */
int mpi_code; /* MPI error code */
htri_t ret_value=TRUE;
FUNC_ENTER_NOAPI(H5D_mpio_opt_possible, FAIL);
/* Check args */
assert(io_info);
assert(mem_space);
assert(file_space);
/* For independent I/O, get out quickly and don't try to form consensus */
if (io_info->dxpl_cache->xfer_mode==H5FD_MPIO_INDEPENDENT)
HGOTO_DONE(FALSE);
/* Optimized MPI types flag must be set and it must be collective IO */
/* (Don't allow parallel I/O for the MPI-posix driver, since it doesn't do real collective I/O) */
if (!(H5S_mpi_opt_types_g && io_info->dxpl_cache->xfer_mode==H5FD_MPIO_COLLECTIVE && !IS_H5FD_MPIPOSIX(io_info->dset->oloc.file))) {
local_opinion = FALSE;
goto broadcast;
} /* end if */
/* Check whether these are both simple or scalar dataspaces */
if (!((H5S_SIMPLE==H5S_GET_EXTENT_TYPE(mem_space) || H5S_SCALAR==H5S_GET_EXTENT_TYPE(mem_space))
&& (H5S_SIMPLE==H5S_GET_EXTENT_TYPE(file_space) || H5S_SCALAR==H5S_GET_EXTENT_TYPE(file_space)))) {
local_opinion = FALSE;
goto broadcast;
} /* end if */
/* Can't currently handle point selections */
if (H5S_SEL_POINTS==H5S_GET_SELECT_TYPE(mem_space) || H5S_SEL_POINTS==H5S_GET_SELECT_TYPE(file_space)) {
local_opinion = FALSE;
goto broadcast;
} /* end if */
/* Dataset storage must be contiguous or chunked */
if (!(io_info->dset->shared->layout.type == H5D_CONTIGUOUS ||
io_info->dset->shared->layout.type == H5D_CHUNKED)) {
local_opinion = FALSE;
goto broadcast;
} /* end if */
/* The handling of memory space is different for chunking
and contiguous storage,
For contigous storage, mem_space and file_space won't
change when it it is doing disk IO.
For chunking storage, mem_space will change for different
chunks. So for chunking storage, whether we can use
collective IO will defer until each chunk IO is reached.
For contiguous storage, if we find MPI-IO cannot
support complicated MPI derived data type and the shape
of data space is not regular, we will
set use_par_opt_io = FALSE.
*/
#ifndef H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS
if(io_info->dset->shared->layout.type == H5D_CONTIGUOUS)
if((H5S_SELECT_IS_REGULAR(file_space) != TRUE) ||
(H5S_SELECT_IS_REGULAR(mem_space) != TRUE)) {
local_opinion = FALSE;
goto broadcast;
} /* end if */
#endif
/* Don't allow collective operations if filters need to be applied */
if(io_info->dset->shared->layout.type == H5D_CHUNKED)
if(io_info->dset->shared->dcpl_cache.pline.nused>0) {
local_opinion = FALSE;
goto broadcast;
} /* end if */
/* Don't allow collective operations if datatype conversions need to happen */
if(!H5T_path_noop(tpath)) {
local_opinion = FALSE;
goto broadcast;
} /* end if */
/* Don't allow collective operations if data transform operations should occur */
if(!H5Z_xform_noop(io_info->dxpl_cache->data_xform_prop)) {
local_opinion = FALSE;
goto broadcast;
} /* end if */
broadcast:
/* Form consensus opinion among all processes about whether to perform
* collective I/O */
if (MPI_SUCCESS != (mpi_code = MPI_Allreduce(&local_opinion, &consensus, 1, MPI_INT, MPI_LAND, io_info->comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Allreduce failed", mpi_code)
ret_value = consensus > 0 ? TRUE : FALSE;
done:
FUNC_LEAVE_NOAPI(ret_value);
} /* H5D_mpio_opt_possible() */
/*-------------------------------------------------------------------------
* Function: H5D_mpio_chunk_adjust_iomode
*
* Decription: If H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS and
H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS are defined,
the HDF5 library will do collective IO if the application asks for it.
If H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS is not defined
and one or more processes are not participating in the IO,
then collective IO is not assured. The library will check
each process for the
number of chunks it involves. If any process involves zero chunks,
the library will use independent IO mode instead.
This function is only used for linked chunk IO.
* Purpose: Checks if it is possible to do collective IO
*
* Return: Success: Non-negative: TRUE or FALSE
* Failure: Negative
*
* Programmer: Muqun Yang
* Monday, Feb. 13th, 2006
*
*-------------------------------------------------------------------------
*/
#ifndef H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS
herr_t
H5D_mpio_chunk_adjust_iomode(H5D_io_info_t *io_info, const fm_map *fm)
{
int min_chunk;
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI_NOINIT(H5D_mpio_chunk_adjust_iomode)
if(H5D_mpio_get_min_chunk(io_info,fm,&min_chunk) < 0)
HGOTO_ERROR(H5E_DATASPACE, H5E_CANTSWAP, FAIL, "unable to obtain the min chunk number of all processes");
if(min_chunk == 0) {
/* Switch to independent I/O */
if(H5D_ioinfo_make_ind(io_info) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't switch to independent I/O")
} /* end if */
done:
FUNC_LEAVE_NOAPI(ret_value)
}
#endif
/*-------------------------------------------------------------------------
* Function: H5D_mpio_select_read
*
* Purpose: MPI-IO function to read directly from app buffer to file.
*
* Return: non-negative on success, negative on failure.
*
* Programmer:
*
*-------------------------------------------------------------------------
*/
herr_t
H5D_mpio_select_read(H5D_io_info_t *io_info,
size_t mpi_buf_count,
const size_t UNUSED elmt_size,
const H5S_t UNUSED *file_space,
const H5S_t UNUSED *mem_space,
haddr_t addr,
void UNUSED *pointer,
void *buf/*out*/)
{
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI(H5D_mpio_select_read,FAIL);
if(H5F_block_read (io_info->dset->oloc.file, H5FD_MEM_DRAW, addr, mpi_buf_count, io_info->dxpl_id, buf) < 0)
HGOTO_ERROR(H5E_IO,H5E_READERROR,FAIL,"can't finish collective parallel read");
done:
FUNC_LEAVE_NOAPI(ret_value);
} /* end H5D_mpio_select_read() */
/*-------------------------------------------------------------------------
* Function: H5D_mpio_select_write
*
* Purpose: MPI-IO function to write directly from app buffer to file.
*
* Return: non-negative on success, negative on failure.
*
* Programmer:
*
*-------------------------------------------------------------------------
*/
herr_t
H5D_mpio_select_write(H5D_io_info_t *io_info,
size_t mpi_buf_count,
const size_t UNUSED elmt_size,
const H5S_t UNUSED *file_space,
const H5S_t UNUSED *mem_space,
haddr_t addr,
void UNUSED *pointer,
const void *buf)
{
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI(H5D_mpio_select_write,FAIL);
/*OKAY: CAST DISCARDS CONST QUALIFIER*/
if(H5F_block_write (io_info->dset->oloc.file, H5FD_MEM_DRAW, addr, mpi_buf_count, io_info->dxpl_id, buf)<0)
HGOTO_ERROR(H5E_IO,H5E_WRITEERROR,FAIL,"can't finish collective parallel write");
done:
FUNC_LEAVE_NOAPI(ret_value);
} /* end H5D_mpio_select_write() */
/*-------------------------------------------------------------------------
* Function: H5D_ioinfo_make_ind
*
* Purpose: Switch to MPI independent I/O
*
* Return: Non-negative on success/Negative on failure
*
* Programmer: Quincey Koziol
* Friday, August 12, 2005
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_ioinfo_make_ind(H5D_io_info_t *io_info)
{
H5P_genplist_t *dx_plist; /* Data transer property list */
herr_t ret_value = SUCCEED; /*return value */
FUNC_ENTER_NOAPI_NOINIT(H5D_ioinfo_make_ind)
/* Get the dataset transfer property list */
if (NULL == (dx_plist = H5I_object(io_info->dxpl_id)))
HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a dataset transfer property list")
/* Change the xfer_mode to independent, handle the request,
* then set xfer_mode before return.
*/
io_info->dxpl_cache->xfer_mode = H5FD_MPIO_INDEPENDENT;
if(H5P_set (dx_plist, H5D_XFER_IO_XFER_MODE_NAME, &io_info->dxpl_cache->xfer_mode) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "can't set transfer mode")
/* Set the pointers to the non-MPI-specific routines */
io_info->ops.read = H5D_select_read;
io_info->ops.write = H5D_select_write;
/* Indicate that the transfer mode should be restored before returning
* to user.
*/
io_info->xfer_mode_changed=TRUE;
done:
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_ioinfo_make_ind() */
/*-------------------------------------------------------------------------
* Function: H5D_ioinfo_make_coll_opt
*
* Purpose: Switch to MPI independent I/O with file set view
*
* Return: Non-negative on success/Negative on failure
*
* Programmer: MuQun Yang
* Oct. 5th, 2006
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_ioinfo_make_coll_opt(H5D_io_info_t *io_info)
{
H5P_genplist_t *dx_plist; /* Data transer property list */
herr_t ret_value = SUCCEED; /*return value */
FUNC_ENTER_NOAPI_NOINIT(H5D_ioinfo_make_coll_opt)
/* Get the dataset transfer property list */
if (NULL == (dx_plist = H5I_object(io_info->dxpl_id)))
HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a dataset transfer property list")
/* Change the optimial xfer_mode to independent, handle the request,
* then set xfer_mode before return.
*/
io_info->dxpl_cache->xfer_opt_mode = H5FD_MPIO_INDIVIDUAL_IO;
if(H5P_set (dx_plist, H5D_XFER_IO_XFER_OPT_MODE_NAME, &io_info->dxpl_cache->xfer_opt_mode) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "can't set transfer mode")
/* Set the pointers to the non-MPI-specific routines */
io_info->ops.read = H5D_mpio_select_read;
io_info->ops.write = H5D_mpio_select_write;
/* Indicate that the transfer mode should be restored before returning
* to user.
*/
io_info->xfer_opt_mode_changed = TRUE;
done:
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_ioinfo_make_coll_opt() */
/*-------------------------------------------------------------------------
* Function: H5D_ioinfo_make_coll
*
* Purpose: Switch to MPI collective I/O
*
* Return: Non-negative on success/Negative on failure
*
* Programmer: Quincey Koziol
* Friday, August 12, 2005
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_ioinfo_make_coll(H5D_io_info_t *io_info)
{
H5P_genplist_t *dx_plist; /* Data transer property list */
herr_t ret_value = SUCCEED; /*return value */
FUNC_ENTER_NOAPI_NOINIT(H5D_ioinfo_make_coll)
/* Get the dataset transfer property list */
if (NULL == (dx_plist = H5I_object(io_info->dxpl_id)))
HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a dataset transfer property list")
/* Change the xfer_mode to independent, handle the request,
* then set xfer_mode before return.
*/
io_info->dxpl_cache->xfer_mode = H5FD_MPIO_COLLECTIVE;
if(H5P_set (dx_plist, H5D_XFER_IO_XFER_MODE_NAME, &io_info->dxpl_cache->xfer_mode) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "can't set transfer mode")
io_info->dxpl_cache->xfer_opt_mode = H5FD_MPIO_COLLECTIVE_IO;
if(H5P_set (dx_plist, H5D_XFER_IO_XFER_OPT_MODE_NAME, &io_info->dxpl_cache->xfer_opt_mode) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "can't set transfer mode")
/* Set the pointers to the MPI-specific routines */
io_info->ops.read = H5D_mpio_select_read;
io_info->ops.write = H5D_mpio_select_write;
/* Indicate that the transfer mode should _NOT_ be restored before returning
* to user.
*/
io_info->xfer_mode_changed=FALSE;
io_info->xfer_opt_mode_changed=FALSE;
done:
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_ioinfo_make_coll() */
/*-------------------------------------------------------------------------
* Function: H5D_mpio_get_min_chunk
*
* Purpose: Routine for obtaining minimum number of chunks to cover
* hyperslab selection selected by all processors.
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_mpio_get_min_chunk(const H5D_io_info_t *io_info,
const fm_map *fm, int *min_chunkf)
{
int num_chunkf; /* Number of chunks to iterate over */
int mpi_code; /* MPI return code */
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI_NOINIT(H5D_mpio_get_min_chunk);
/* Get the number of chunks to perform I/O on */
num_chunkf = H5SL_count(fm->fsel);
/* Determine the minimum # of chunks for all processes */
if (MPI_SUCCESS != (mpi_code = MPI_Allreduce(&num_chunkf, min_chunkf, 1, MPI_INT, MPI_MIN, io_info->comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Allreduce failed", mpi_code)
done:
FUNC_LEAVE_NOAPI(ret_value);
} /* end H5D_mpio_get_min_chunk() */
/*-------------------------------------------------------------------------
* Function: H5D_mpio_get_sum_chunk
*
* Purpose: Routine for obtaining total number of chunks to cover
* hyperslab selection selected by all processors.
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_mpio_get_sum_chunk(const H5D_io_info_t *io_info,
const fm_map *fm, int *sum_chunkf)
{
int num_chunkf; /* Number of chunks to iterate over */
size_t ori_num_chunkf;
int mpi_code; /* MPI return code */
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI_NOINIT(H5D_mpio_get_sum_chunk);
/* Get the number of chunks to perform I/O on */
num_chunkf = 0;
ori_num_chunkf = H5SL_count(fm->fsel);
H5_ASSIGN_OVERFLOW(num_chunkf,ori_num_chunkf,size_t,int);
/* Determine the summation of number of chunks for all processes */
if (MPI_SUCCESS != (mpi_code = MPI_Allreduce(&num_chunkf, sum_chunkf, 1, MPI_INT, MPI_SUM, io_info->comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Allreduce failed", mpi_code)
done:
FUNC_LEAVE_NOAPI(ret_value);
} /* end H5D_mpio_get_sum_chunk() */
/*-------------------------------------------------------------------------
* Function: H5D_contig_collective_io
*
* Purpose: Wrapper Routine for H5D_inter_collective_io
The starting file address of contiguous layout
will be calculated and passed to H5D_inter_collective_io routine.
*
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5D_contig_collective_io(H5D_io_info_t *io_info,
const H5S_t *file_space,
const H5S_t *mem_space,
const void *buf,
hbool_t do_write)
{
haddr_t addr = HADDR_UNDEF; /* Address of dataset (or selection) within file */
herr_t ret_value = SUCCEED; /* return value */
FUNC_ENTER_NOAPI_NOINIT(H5D_contig_collective_io)
assert (IS_H5FD_MPIO(io_info->dset->oloc.file));
/* Make certain we have the correct type of property list */
assert(TRUE==H5P_isa_class(io_info->dxpl_id,H5P_DATASET_XFER));
/* Get the base address of the contiguous dataset */
if(io_info->dset->shared->layout.type == H5D_CONTIGUOUS)
addr = H5D_contig_get_addr(io_info->dset);
if(H5D_inter_collective_io(io_info,file_space,mem_space,addr,buf,do_write)<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish shared collective MPI-IO");
done:
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_contig_collective_io */
/*-------------------------------------------------------------------------
* Function: H5D_chunk_collective_io
*
* Purpose: Routine for
1) choose an IO option:
a) One collective IO defined by one MPI derived datatype to link through all chunks
or b) multiple chunk IOs,to do MPI-IO for each chunk, the IO mode may be adjusted
due to the selection pattern for each chunk.
* For option a)
1. Sort the chunk address, obtain chunk info according to the sorted chunk address
2. Build up MPI derived datatype for each chunk
3. Build up the final MPI derived datatype
4. Set up collective IO property list
5. Do IO
* For option b)
1. Use MPI_gather and MPI_Bcast to obtain information of *collective/independent/none*
IO mode for each chunk of the selection
2. Depending on whether the IO mode is collective or independent or none,
Create either MPI derived datatype for each chunk to do collective IO or
just do independent IO or independent IO with file set view
3. Set up collective IO property list for collective mode
4. DO IO
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5D_chunk_collective_io(H5D_io_info_t *io_info,fm_map *fm,const void *buf, hbool_t do_write)
{
int io_option = H5D_MULTI_CHUNK_IO_MORE_OPT;
int sum_chunk = 0,mpi_size;
unsigned one_link_chunk_io_threshold;
H5P_genplist_t *plist;
H5FD_mpio_chunk_opt_t chunk_opt_mode;
#ifdef H5_HAVE_INSTRUMENTED_LIBRARY
htri_t check_prop,temp_not_link_io = FALSE;
int new_value;
#endif
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI_NOINIT(H5D_chunk_collective_io)
assert (IS_H5FD_MPIO(io_info->dset->oloc.file));
/* Obtain the data transfer properties */
if(NULL == (plist = H5I_object(io_info->dxpl_id)))
HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a file access property list")
/* Check the optional property list on what to do with collective chunk IO. */
chunk_opt_mode=(H5FD_mpio_chunk_opt_t)H5P_peek_unsigned(plist,H5D_XFER_MPIO_CHUNK_OPT_HARD_NAME);
if(chunk_opt_mode == H5FD_MPIO_CHUNK_ONE_IO)
io_option = H5D_ONE_LINK_CHUNK_IO;/*no opt*/
else if(chunk_opt_mode == H5FD_MPIO_CHUNK_MULTI_IO)
io_option = H5D_MULTI_CHUNK_IO;/*no opt */
else {
if(H5D_mpio_get_sum_chunk(io_info,fm,&sum_chunk)<0)
HGOTO_ERROR(H5E_DATASPACE, H5E_CANTSWAP, FAIL, "unable to obtain the total chunk number of all processes");
if((mpi_size = H5F_mpi_get_size(io_info->dset->oloc.file))<0)
HGOTO_ERROR (H5E_IO, H5E_MPI, FAIL, "unable to obtain mpi size");
if(NULL == (plist = H5I_object(io_info->dxpl_id)))
HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a file access property list")
one_link_chunk_io_threshold =H5P_peek_unsigned(plist,H5D_XFER_MPIO_CHUNK_OPT_NUM_NAME);
/* step 1: choose an IO option */
/* If the average number of chunk per process is greater than a threshold, we will do one link chunked IO. */
if((unsigned)sum_chunk/mpi_size >= one_link_chunk_io_threshold)
io_option = H5D_ONE_LINK_CHUNK_IO_MORE_OPT;
#ifdef H5_HAVE_INSTRUMENTED_LIBRARY
else
temp_not_link_io = TRUE;
#endif
}
#ifdef H5_HAVE_INSTRUMENTED_LIBRARY
/*** Test collective chunk user-input optimization APIs. ***/
check_prop = H5Pexist(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_LINK_HARD_NAME);
if(check_prop > 0) {
if(io_option == H5D_ONE_LINK_CHUNK_IO) {
new_value = 0;
if(H5Pset(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_LINK_HARD_NAME,&new_value)<0)
HGOTO_ERROR(H5E_PLIST, H5E_UNSUPPORTED, FAIL, "unable to get property value");
}
}
check_prop = H5Pexist(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_MULTI_HARD_NAME);
if(check_prop > 0) {
if(io_option == H5D_MULTI_CHUNK_IO) {
new_value = 0;
if(H5Pset(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_MULTI_HARD_NAME,&new_value)<0)
HGOTO_ERROR(H5E_PLIST, H5E_UNSUPPORTED, FAIL, "unable to get property value");
}
}
check_prop = H5Pexist(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_LINK_NUM_TRUE_NAME);
if(check_prop > 0) {
if(io_option == H5D_ONE_LINK_CHUNK_IO_MORE_OPT) {
new_value = 0;
if(H5Pset(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_LINK_NUM_TRUE_NAME,&new_value)<0)
HGOTO_ERROR(H5E_PLIST, H5E_UNSUPPORTED, FAIL, "unable to get property value");
}
}
check_prop = H5Pexist(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_LINK_NUM_FALSE_NAME);
if(check_prop > 0) {
if(temp_not_link_io){
new_value = 0;
if(H5Pset(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_LINK_NUM_FALSE_NAME,&new_value)<0)
HGOTO_ERROR(H5E_PLIST, H5E_UNSUPPORTED, FAIL, "unable to get property value");
}
}
#endif
#ifndef H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS
if(io_option == H5D_ONE_LINK_CHUNK_IO )
io_option = H5D_MULTI_CHUNK_IO ;/* We can not do this with one chunk IO. */
if(io_option == H5D_ONE_LINK_CHUNK_IO_MORE_OPT)
io_option = H5D_MULTI_CHUNK_IO_MORE_OPT;
#endif
/* step 2: Go ahead to do IO.*/
if(io_option == H5D_ONE_LINK_CHUNK_IO || io_option == H5D_ONE_LINK_CHUNK_IO_MORE_OPT) {
if(H5D_link_chunk_collective_io(io_info,fm,buf,do_write,sum_chunk)<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish linked chunk MPI-IO");
}
else if(io_option == H5D_MULTI_CHUNK_IO) {
if(H5D_multi_chunk_collective_io_no_opt(io_info,fm,buf,do_write)<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish multiple chunk MPI-IO");
}
else { /*multiple chunk IOs with opt */
if(H5D_multi_chunk_collective_io(io_info,fm,buf,do_write)<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish multiple chunk MPI-IO");
}
done:
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_chunk_collective_io */
/*-------------------------------------------------------------------------
* Function: H5D_link_chunk_collective_io
*
* Purpose: Routine for one collective IO with one MPI derived datatype to link with all chunks
1. Sort the chunk address and chunk info
2. Build up MPI derived datatype for each chunk
3. Build up the final MPI derived datatype
4. Use common collective IO routine to do MPI-IO
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_link_chunk_collective_io(H5D_io_info_t *io_info,fm_map *fm,const void *buf, hbool_t do_write,int sum_chunk)
{
size_t src_type_size; /*size of source type */
size_t dst_type_size; /*size of destination type*/
hsize_t mpi_buf_extra_offset;
hsize_t mpi_file_extra_offset;
size_t mpi_buf_count;
size_t mpi_file_count;
hbool_t mbt_is_derived=0, /* Whether the buffer (memory) type is derived and needs to be free'd */
mft_is_derived=0; /* Whether the file type is derived and needs to be free'd */
int mpi_size,mpi_code; /* MPI return code */
int i,num_chunk=0,total_chunks;
size_t ori_num_chunk;
hsize_t ori_total_chunks;
haddr_t chunk_base_addr;
haddr_t* total_chunk_addr_array=NULL;
MPI_Datatype *chunk_mtype=NULL;
MPI_Datatype *chunk_ftype=NULL;
MPI_Datatype chunk_final_mtype;
MPI_Datatype chunk_final_ftype;
MPI_Aint *chunk_disp_array=NULL;
MPI_Aint *chunk_mem_disp_array=NULL;
int *blocklen=NULL;
int blocklen_value;
int actual_bsearch_coll_chunk_threshold;
int bsearch_coll_chunk_threshold;
int many_chunk_opt = 0;
H5D_common_coll_info_t coll_info;
H5D_chunk_addr_info_t* chunk_addr_info_array=NULL;
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI_NOINIT(H5D_link_chunk_collective_io)
ori_total_chunks = fm->total_chunks;
H5_ASSIGN_OVERFLOW(total_chunks,ori_total_chunks,hsize_t,int);
/* Handle with a special case when only one chunk is covered by all processes */
if(total_chunks == 1){
H5SL_node_t *chunk_node;
H5D_chunk_info_t *chunk_info;
H5D_storage_t store;
chunk_node = H5SL_first(fm->fsel);
if(chunk_node == NULL) {
if(H5D_istore_chunkmap(io_info, &chunk_base_addr, fm->down_chunks) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk address");
if(H5D_inter_collective_io(io_info,NULL,NULL,chunk_base_addr,buf,do_write)<0)
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't finish shared collective MPI-IO");
}
else {
if(NULL ==(chunk_info = H5SL_item(chunk_node)))
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't get chunk info from skipped list");
io_info->store = &store;
store.chunk.offset = chunk_info->coords;
store.chunk.index = chunk_info->index;
if(HADDR_UNDEF==(chunk_base_addr = H5D_istore_get_addr(io_info,NULL)))
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't get chunk info from skipped list");
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"before inter_collective_io for total chunk = 1 \n");
#endif
if(H5D_inter_collective_io(io_info,chunk_info->fspace,chunk_info->mspace,chunk_base_addr,buf,do_write)<0)
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't finish shared collective MPI-IO");
}
goto done;
}
/* Allocate chunking information */
ori_num_chunk = H5SL_count(fm->fsel);
H5_ASSIGN_OVERFLOW(num_chunk,ori_num_chunk,size_t,int);
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"total_chunks = %d\n",(int)total_chunks);
#endif
if(num_chunk == 0)
total_chunk_addr_array = H5MM_malloc(sizeof(haddr_t)*total_chunks);
else
{
chunk_addr_info_array= H5MM_malloc(num_chunk*sizeof(H5D_chunk_addr_info_t));
chunk_mtype = H5MM_malloc(num_chunk*sizeof(MPI_Datatype));
chunk_ftype = H5MM_malloc(num_chunk*sizeof(MPI_Datatype));
chunk_disp_array = H5MM_malloc(num_chunk*sizeof(MPI_Aint));
chunk_mem_disp_array = H5MM_calloc(num_chunk*sizeof(MPI_Aint));
blocklen = H5MM_malloc(num_chunk*sizeof(int));
}
/* Obtain information to do collective IO,
in order to do collective IO, no datatype conversion should happen. */
if((src_type_size = H5T_get_size(io_info->dset->shared->type))==0)
HGOTO_ERROR(H5E_DATATYPE, H5E_BADSIZE, FAIL, "datatype size invalid");
dst_type_size = src_type_size;
bsearch_coll_chunk_threshold = H5D_ALL_CHUNK_ADDR_THRES_COL;
if((mpi_size = H5F_mpi_get_size(io_info->dset->oloc.file))<0)
HGOTO_ERROR (H5E_IO, H5E_MPI, FAIL, "unable to obtain mpi size");
/* Calculate the actual threshold to obtain all chunk addresses collectively
The bigger this number is, the more possible the use of obtaining chunk address collectively. */
/* For non-optimization one-link IO,
actual bsearch threshold is always 0,
we would always want to obtain the chunk addresses individually
for each process. */
actual_bsearch_coll_chunk_threshold = sum_chunk*100/(total_chunks*mpi_size);
if((actual_bsearch_coll_chunk_threshold > bsearch_coll_chunk_threshold)
&&(sum_chunk/mpi_size >= H5D_ALL_CHUNK_ADDR_THRES_COL_NUM))
many_chunk_opt = H5D_OBTAIN_ALL_CHUNK_ADDR_COL;
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"before sorting the chunk address \n");
#endif
/* Sort the chunk address
when chunk optimization selection is either H5D_OBTAIN_*/
if(num_chunk == 0){ /* special case: this process doesn't select anything */
if(H5D_istore_chunkmap(io_info, total_chunk_addr_array, fm->down_chunks)<0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk address");
chunk_base_addr = total_chunk_addr_array[0];
}
else {
if(H5D_sort_chunk(io_info,fm,chunk_addr_info_array,many_chunk_opt)<0)
HGOTO_ERROR (H5E_DATASPACE, H5E_CANTSWAP, FAIL, "unable to sort chunk address");
chunk_base_addr = chunk_addr_info_array[0].chunk_addr;
}
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"after sorting the chunk address \n");
#endif
/* Obtain MPI derived datatype from all individual chunks */
for ( i = 0; i < num_chunk; i++) {
/* Disk MPI derived datatype */
if(H5S_mpio_space_type(chunk_addr_info_array[i].chunk_info.fspace,src_type_size,&chunk_ftype[i],
&mpi_file_count,&mpi_file_extra_offset,&mft_is_derived)<0)
HGOTO_ERROR(H5E_DATASPACE, H5E_BADTYPE, FAIL,"couldn't create MPI file type");
/* Buffer MPI derived datatype */
if(H5S_mpio_space_type(chunk_addr_info_array[i].chunk_info.mspace,dst_type_size,&chunk_mtype[i],
&mpi_buf_count,&mpi_buf_extra_offset,&mbt_is_derived)<0)
HGOTO_ERROR(H5E_DATASPACE, H5E_BADTYPE, FAIL,"couldn't create MPI buf type");
/* Chunk address relative to the first chunk */
chunk_addr_info_array[i].chunk_addr -= chunk_base_addr;
H5_ASSIGN_OVERFLOW(chunk_disp_array[i],chunk_addr_info_array[i].chunk_addr,haddr_t,MPI_Aint);
}
blocklen_value = 1;
if(num_chunk){
/* initialize the buffer with the constant value 1; this algo. is very fast. */
H5V_array_fill(blocklen,&blocklen_value,sizeof(int),(size_t)num_chunk);
/* Create final MPI derived datatype */
if(MPI_SUCCESS != (mpi_code = MPI_Type_struct(num_chunk,blocklen,chunk_disp_array,chunk_ftype,&chunk_final_ftype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_struct failed", mpi_code);
if(MPI_SUCCESS != (mpi_code = MPI_Type_commit(&chunk_final_ftype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_commit failed", mpi_code);
if(MPI_SUCCESS != (mpi_code = MPI_Type_struct(num_chunk,blocklen,chunk_mem_disp_array,chunk_mtype,&chunk_final_mtype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_struct failed", mpi_code);
if(MPI_SUCCESS != (mpi_code = MPI_Type_commit(&chunk_final_mtype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_commit failed", mpi_code);
for ( i = 0; i< num_chunk;i++){
if (MPI_SUCCESS != (mpi_code= MPI_Type_free( chunk_mtype+i )))
HMPI_DONE_ERROR(FAIL, "MPI_Type_free failed", mpi_code);
if (MPI_SUCCESS != (mpi_code= MPI_Type_free( chunk_ftype+i )))
HMPI_DONE_ERROR(FAIL, "MPI_Type_free failed", mpi_code);
}
/* buffer, file derived datatypes should be true */
coll_info.mbt_is_derived = 1;
coll_info.mft_is_derived = 1;
coll_info.mpi_buf_count = 1;
coll_info.chunk_addr = chunk_base_addr;
}
else {/* no selection at all for this process */
chunk_final_ftype = MPI_BYTE;
chunk_final_mtype = MPI_BYTE;
/* buffer, file derived datatypes should be true */
coll_info.mbt_is_derived = 0;
coll_info.mft_is_derived = 0;
coll_info.mpi_buf_count = 0;
coll_info.chunk_addr = chunk_base_addr;
}
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"before coming to final collective IO\n");
#endif
if(H5D_final_collective_io(io_info,&chunk_final_ftype,&chunk_final_mtype,&coll_info,buf,do_write)<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish MPI-IO");
done:
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"before freeing memory inside H5D_link_collective_io ret_value = %d\n",ret_value);
#endif
if (fm->total_chunks != 1) {
if(num_chunk == 0) HDfree(total_chunk_addr_array);
else {
HDfree(chunk_addr_info_array);
HDfree(chunk_mtype);
HDfree(chunk_ftype);
HDfree(chunk_disp_array);
HDfree(chunk_mem_disp_array);
HDfree(blocklen);
}
}
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_link_chunk_collective_io */
/*-------------------------------------------------------------------------
* Function: H5D_multi_chunk_collective_io
*
* Purpose: To do IO per chunk according to IO mode(collective/independent/none)
1. Use MPI_gather and MPI_Bcast to obtain IO mode in each chunk(collective/independent/none)
2. Depending on whether the IO mode is collective or independent or none,
Create either MPI derived datatype for each chunk or just do independent IO
3. Use common collective IO routine to do MPI-IO
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_multi_chunk_collective_io(H5D_io_info_t *io_info,fm_map *fm,const void *buf, hbool_t do_write)
{
unsigned i, total_chunk;
hsize_t ori_total_chunk;
uint8_t *chunk_io_option;
H5SL_node_t *chunk_node; /* Current node in chunk skip list */
H5D_chunk_info_t *chunk_info=NULL;
haddr_t *chunk_addr;
H5D_storage_t store; /* union of EFL and chunk pointer in file space */
hbool_t select_chunk;
hbool_t last_io_mode_coll = TRUE;
herr_t ret_value = SUCCEED;
#ifdef H5Dmpio_DEBUG
int mpi_rank;
#endif
FUNC_ENTER_NOAPI_NOINIT(H5D_multi_chunk_collective_io)
#ifdef H5Dmpio_DEBUG
mpi_rank = H5F_mpi_get_rank(io_info->dset->oloc.file);
#endif
/* Allocate memories */
ori_total_chunk = fm->total_chunks;
H5_ASSIGN_OVERFLOW(total_chunk,ori_total_chunk,hsize_t,unsigned);
HDassert(total_chunk!=0);
chunk_io_option = (uint8_t *)H5MM_calloc(total_chunk*sizeof(MPI_BYTE));
chunk_addr = (haddr_t *)H5MM_calloc(total_chunk*sizeof(haddr_t));
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"total_chunk %u\n",total_chunk);
#endif
/* obtain IO option for each chunk */
if(H5D_obtain_mpio_mode(io_info,fm,chunk_io_option,chunk_addr)<0)
HGOTO_ERROR (H5E_DATASET, H5E_CANTRECV, FAIL, "unable to obtain MPIO mode");
for(i = 0; i < total_chunk; i++) {
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"mpi_rank = %d, chunk index = %u\n",mpi_rank,i);
#endif
select_chunk = fm->select_chunk[i];
if(select_chunk == 1){/* Have selection elements in this chunk. Find the chunk info. */
if(NULL ==(chunk_node = H5SL_first(fm->fsel)))
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't get chunk node from skipped list");
while(chunk_node){
if(NULL ==(chunk_info = H5SL_item(chunk_node)))
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't get chunk info from skipped list");
if(chunk_info->index == i) {
/* Set dataset storage for I/O info */
io_info->store=&store;
/* Pass in chunk's coordinates in a union. */
store.chunk.offset = chunk_info->coords;
store.chunk.index = chunk_info->index;
break;
}
chunk_node = H5SL_next(chunk_node);
}
}
if(chunk_io_option[i] == 1){ /*collective IO for this chunk,
note: even there is no selection for this process,
the process still needs to contribute MPI NONE TYPE.*/
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"inside collective chunk IO mpi_rank = %d, chunk index = %u\n",mpi_rank,i);
#endif
if(!last_io_mode_coll)
/* Switch back to collective I/O */
if(H5D_ioinfo_make_coll(io_info) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't switch to collective I/O")
if(select_chunk){
if(H5D_inter_collective_io(io_info,chunk_info->fspace,chunk_info->mspace,
chunk_addr[i],buf,do_write )<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish shared collective MPI-IO");
}
else{
if(H5D_inter_collective_io(io_info,NULL,NULL,
chunk_addr[i],buf,do_write )<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish shared collective MPI-IO");
}
last_io_mode_coll = TRUE;
}
else {/*possible independent IO for this chunk*/
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"inside independent IO mpi_rank = %d, chunk index = %u\n",mpi_rank,i);
#endif
HDassert(chunk_io_option[i] == 0);
#if !defined(H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS) || !defined(H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS)
if(!select_chunk)
continue; /* this process has nothing to do with this chunk, continue! */
if(last_io_mode_coll)
/* Switch to independent I/O */
if(H5D_ioinfo_make_ind(io_info) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't switch to independent I/O")
if(do_write) {
if((io_info->ops.write)(io_info,
chunk_info->chunk_points,H5T_get_size(io_info->dset->shared->type),
chunk_info->fspace,chunk_info->mspace,0, NULL, buf) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "optimized write failed")
}
else {
if((io_info->ops.read)(io_info,
chunk_info->chunk_points,H5T_get_size(io_info->dset->shared->type),
chunk_info->fspace,chunk_info->mspace,0, NULL, buf) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "optimized read failed")
}
#else
if(!last_io_mode_coll)
/* using independent I/O with file setview.*/
if(H5D_ioinfo_make_coll_opt(io_info) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't switch to collective I/O")
if(select_chunk){
if(H5D_inter_collective_io(io_info,chunk_info->fspace,chunk_info->mspace,
chunk_addr[i],buf,do_write )<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish shared collective MPI-IO");
}
else {
if(H5D_inter_collective_io(io_info,NULL,NULL,
chunk_addr[i],buf,do_write )<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish shared collective MPI-IO");
}
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"after inter collective IO\n");
#endif
#endif
last_io_mode_coll = FALSE;
}
}
if(!last_io_mode_coll)
/* Switch back to collective I/O */
if(H5D_ioinfo_make_coll(io_info) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't switch to collective I/O")
done:
HDfree(chunk_io_option);
HDfree(chunk_addr);
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_multi_chunk_collective_io */
/*-------------------------------------------------------------------------
* Function: H5D_multi_chunk_collective_io_no_opt
*
* Purpose: To do collective IO without any optimization per chunk base
* The internal independent IO inside HDF5 cannot handle
* non-contiguous(or with holes) storage efficiently.
* Under this case, the one independent IO call may consist of
* many small disk IOs. So we may use independent IO with derived datatype
to replace the independent IO when we find this chunk is not good to
do collective IO. However, according to our performance study,
this approach may not overcome the overhead caused by MPI gather/scatter.
So we decide to leave the original collective IO per chunk approach as
an option for users. NO MPI gather/scatter calls are used.
HDF5 will try to collective IO if possible.
If users choose to use
H5Pset_dxpl_mpio_chunk_opt(dxpl_id,H5FD_MPIO_OPT_MULTI_IO),
this function will be called.
The HDF5 library won't do any IO management but leave it to MPI-IO to figure
out.
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_multi_chunk_collective_io_no_opt(H5D_io_info_t *io_info,fm_map *fm,const void *buf, hbool_t do_write)
{
int count_chunk,min_num_chunk;
haddr_t chunk_addr;
H5SL_node_t *chunk_node; /* Current node in chunk skip list */
H5D_storage_t store; /* union of EFL and chunk pointer in file space */
herr_t ret_value = SUCCEED;
#ifdef H5Dmpio_DEBUG
int mpi_rank;
#endif
FUNC_ENTER_NOAPI_NOINIT(H5D_multi_chunk_collective_io_no_opt)
#ifdef H5D_DEBUG
if(H5DEBUG(D)){
mpi_rank = H5F_mpi_get_rank(io_info->dset->oloc.file);
HDfprintf(H5DEBUG(D),"coming to multi_chunk_collective_io_no_opt\n");
}
#endif
if(H5D_mpio_get_min_chunk(io_info,fm,&min_num_chunk)<0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get minimum number of chunk");
count_chunk = 0;
/* Get first node in chunk skip list */
chunk_node=H5SL_first(fm->fsel);
/* Iterate through chunks to be operated on */
while(chunk_node) {
H5D_chunk_info_t *chunk_info; /* chunk information */
hbool_t make_ind, make_coll; /* Flags to indicate that the MPI mode should change */
/* Get the actual chunk information from the skip list node */
chunk_info=H5SL_item(chunk_node);
/* Set dataset storage for I/O info */
io_info->store=&store;
/* Pass in chunk's coordinates in a union. */
store.chunk.offset = chunk_info->coords;
store.chunk.index = chunk_info->index;
/* Reset flags for changing parallel I/O mode */
make_ind = make_coll = FALSE;
count_chunk++;
/* If the number of chunk is greater than minimum number of chunk,
Do independent read */
if(count_chunk > min_num_chunk) {
/* Switch to independent I/O (permanently) */
make_ind = TRUE;
}
#ifndef H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS
/* This case needs to be improved to check if the selected space
is regular. If all selections are regular, collective IO can still be done.
However, since we find an MPI-IO bug at a DOE machine(mcr) that cannot
handle collective I/O selection for this case correctly,
we turn off this optimization but leave the following code
for future optimization. Otherwise, the following else {} doesn't make sense.
KY 2006/8/4/ */
else {
/* Switch to independent I/O (temporarily) */
make_ind = TRUE;
make_coll = TRUE;
} /* end else */
#endif /* H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS */
/* Switch to independent I/O */
if(make_ind)
if(H5D_ioinfo_make_ind(io_info) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't switch to independent I/O")
if(make_ind) {/*independent I/O */
if(do_write) {
if((io_info->ops.write)(io_info,
chunk_info->chunk_points,H5T_get_size(io_info->dset->shared->type),
chunk_info->fspace,chunk_info->mspace, (hsize_t)0, NULL, buf) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "optimized write failed")
}
else {
if((io_info->ops.read)(io_info,
chunk_info->chunk_points,H5T_get_size(io_info->dset->shared->type),
chunk_info->fspace,chunk_info->mspace, (hsize_t)0, NULL, buf) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "optimized read failed")
}
}
else { /*collective I/O */
if(HADDR_UNDEF==(chunk_addr = H5D_istore_get_addr(io_info,NULL)))
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't get chunk info from skipped list");
if(H5D_inter_collective_io(io_info,chunk_info->fspace,chunk_info->mspace,
chunk_addr,buf,do_write ) < 0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish shared collective MPI-IO");
}
if(make_coll)
if(H5D_ioinfo_make_coll(io_info) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't switch to independent I/O")
/* Get the next chunk node in the skip list */
chunk_node=H5SL_next(chunk_node);
} /* end while */
done:
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_multi_chunk_collective_io_no_opt */
/*-------------------------------------------------------------------------
* Function: H5D_inter_collective_io
*
* Purpose: Routine for the shared part of collective IO between multiple chunk
collective IO and contiguous collective IO
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_inter_collective_io(H5D_io_info_t *io_info,const H5S_t *file_space,const H5S_t *mem_space,
haddr_t addr, const void *buf, hbool_t do_write )
{
size_t mpi_buf_count, mpi_file_count; /* Number of "objects" to transfer */
MPI_Datatype mpi_file_type,mpi_buf_type;
hsize_t mpi_buf_offset, mpi_file_offset; /* Offset within dataset where selection (ie. MPI type) begins */
hbool_t mbt_is_derived=0, /* Whether the buffer (memory) type is derived and needs to be free'd */
mft_is_derived=0; /* Whether the file type is derived and needs to be free'd */
H5D_common_coll_info_t coll_info;
herr_t ret_value = SUCCEED; /* return value */
FUNC_ENTER_NOAPI_NOINIT(H5D_inter_collective_io)
if((file_space!=NULL) && (mem_space != NULL)) {
/*Obtain disk and memory MPI derived datatype */
if(H5S_mpio_space_type(file_space,H5T_get_size(io_info->dset->shared->type),
&mpi_file_type,&mpi_file_count,&mpi_file_offset,&mft_is_derived)<0)
HGOTO_ERROR(H5E_DATASPACE, H5E_BADTYPE, FAIL,"couldn't create MPI file type");
if(H5S_mpio_space_type(mem_space,H5T_get_size(io_info->dset->shared->type),
&mpi_buf_type,&mpi_buf_count,&mpi_buf_offset,&mbt_is_derived)<0)
HGOTO_ERROR(H5E_DATASPACE, H5E_BADTYPE, FAIL,"couldn't create MPI buffer type");
}
else {
/* For non-selection, participate with a none MPI derived datatype, the count is 0. */
mpi_buf_type = MPI_BYTE;
mpi_file_type = MPI_BYTE;
mpi_file_count = 0;
mpi_buf_count = 0;
}
coll_info.mbt_is_derived = mbt_is_derived;
coll_info.mft_is_derived = mft_is_derived;
coll_info.mpi_buf_count = mpi_buf_count;
coll_info.chunk_addr = addr;
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"before final collective IO \n");
#endif
if(H5D_final_collective_io(io_info,&mpi_file_type,&mpi_buf_type,&coll_info,buf,do_write)<0)
HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish collective MPI-IO");
done:
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"before leaving inter_collective_io ret_value = %d\n",ret_value);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_inter_collective_io */
/*-------------------------------------------------------------------------
* Function: H5D_final_collective_io
*
* Purpose: Routine for the common part of collective IO with different storages.
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_final_collective_io(H5D_io_info_t *io_info,MPI_Datatype*mpi_file_type,MPI_Datatype *mpi_buf_type,
H5D_common_coll_info_t* coll_info, const void *buf, hbool_t do_write)
{
int mpi_code; /* MPI return code */
hbool_t plist_is_setup=0; /* Whether the dxpl has been customized */
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI_NOINIT(H5D_final_collective_io)
/*
* Pass buf type, file type to the file driver.
*/
if(H5FD_mpi_setup_collective(io_info->dxpl_id, *mpi_buf_type, *mpi_file_type)<0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "can't set MPI-I/O properties");
plist_is_setup=1;
#ifdef H5D_DEBUG
if(H5DEBUG(D)){
HDfprintf(H5DEBUG(D),"chunk addr %Hu\n",coll_info->chunk_addr);
HDfprintf(H5DEBUG(D),"mpi_buf_count %d\n",coll_info->mpi_buf_count);
}
#endif
if(do_write) {
if((io_info->ops.write)(io_info,
coll_info->mpi_buf_count,0,NULL,NULL,coll_info->chunk_addr,
NULL, buf) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_WRITEERROR, FAIL, "optimized write failed")
}
else {
if((io_info->ops.read)(io_info,
coll_info->mpi_buf_count,0,NULL,NULL,coll_info->chunk_addr,
NULL, buf) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_READERROR, FAIL, "optimized read failed")
}
done:
/* Reset the dxpl settings */
if(plist_is_setup) {
if(H5FD_mpi_teardown_collective(io_info->dxpl_id)<0)
HDONE_ERROR(H5E_DATASPACE, H5E_CANTFREE, FAIL, "unable to reset dxpl values");
} /* end if */
/* free the MPI buf and file types */
if (coll_info->mbt_is_derived) {
if (MPI_SUCCESS != (mpi_code= MPI_Type_free( mpi_buf_type )))
HMPI_DONE_ERROR(FAIL, "MPI_Type_free failed", mpi_code);
}
if (coll_info->mft_is_derived) {
if (MPI_SUCCESS != (mpi_code= MPI_Type_free( mpi_file_type )))
HMPI_DONE_ERROR(FAIL, "MPI_Type_free failed", mpi_code);
}
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"ret_value before leaving final_collective_io=%d\n",ret_value);
#endif
FUNC_LEAVE_NOAPI(ret_value)
}/* end H5D_final_collective_io */
/*-------------------------------------------------------------------------
* Function: H5D_sort_chunk
*
* Purpose: Routine to sort chunks in increasing order of chunk address
Each chunk address is also obtained.
Description:
For most cases, the chunk address has already been sorted in increasing order.
The special sorting flag is used to optimize this common case.
quick sort is used for necessary sorting.
Parameters:
Input: H5D_io_info_t* io_info,
fm_map *fm(global chunk map struct)
Input/Output: H5D_chunk_addr_info_t chunk_addr_info_array[] : array to store chunk address and information
many_chunk_opt : flag to optimize the way to obtain chunk addresses
for many chunks
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_sort_chunk(H5D_io_info_t * io_info,
fm_map *fm,
H5D_chunk_addr_info_t chunk_addr_info_array[],
int many_chunk_opt)
{
H5SL_node_t *chunk_node; /* Current node in chunk skip list */
H5D_chunk_info_t *chunk_info; /* Current chunking info. of this node. */
haddr_t chunk_addr; /* Current chunking address of this node */
haddr_t *total_chunk_addr_array=NULL; /* The array of chunk address for the total number of chunk */
int i,mpi_code;
int total_chunks;
size_t num_chunks;
int mpi_type_cleanup = 0;
int tchunk_addr_cleanup = 0;
MPI_Datatype chunk_addrtype;
H5D_storage_t store; /*union of EFL and chunk pointer in file space */
hbool_t do_sort = FALSE;
herr_t ret_value = SUCCEED; /*return value */
FUNC_ENTER_NOAPI_NOINIT(H5D_sort_chunk)
num_chunks = H5SL_count(fm->fsel);
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"many_chunk_opt= %d\n",many_chunk_opt);
#endif
/* If we need to optimize the way to obtain the chunk address */
if(many_chunk_opt != H5D_OBTAIN_ONE_CHUNK_ADDR_IND){
int mpi_rank, root;
total_chunks = (int)fm->total_chunks;
total_chunk_addr_array = H5MM_malloc(sizeof(haddr_t)*total_chunks);
tchunk_addr_cleanup = 1;
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"Coming inside H5D_OBTAIN_ALL_CHUNK_ADDR_COL\n");
#endif
root = 0;
if((mpi_rank = H5F_mpi_get_rank(io_info->dset->oloc.file))<0)
HGOTO_ERROR (H5E_IO, H5E_MPI, FAIL, "unable to obtain mpi rank");
/*Create received MPI derived datatype */
if(MPI_SUCCESS !=(mpi_code = MPI_Type_contiguous((int)(sizeof(haddr_t)*total_chunks), MPI_BYTE, &chunk_addrtype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_contiguous failed", mpi_code);
if(MPI_SUCCESS !=(mpi_code = MPI_Type_commit(&chunk_addrtype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_commit failed", mpi_code);
mpi_type_cleanup = 1;
if(mpi_rank == root) {
if(H5D_istore_chunkmap(io_info, total_chunk_addr_array, fm->down_chunks)<0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk address");
}
/* Broadcasting the MPI_IO option info. and chunk address info. */
if(MPI_SUCCESS !=(mpi_code = MPI_Bcast(total_chunk_addr_array,1,chunk_addrtype,root,io_info->comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_BCast failed", mpi_code);
} /* end if */
/* Get first node in chunk skip list */
if(NULL ==(chunk_node = H5SL_first(fm->fsel)))
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't get chunk node from skipped list");
/* Set dataset storage for I/O info */
io_info->store = &store;
if(NULL ==(chunk_info = H5SL_item(chunk_node)))
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't get chunk info from skipped list");
store.chunk.offset = chunk_info->coords;
store.chunk.index = chunk_info->index;
i = 0;
if(many_chunk_opt == H5D_OBTAIN_ONE_CHUNK_ADDR_IND){
if(HADDR_UNDEF==(chunk_addr = H5D_istore_get_addr(io_info,NULL)))
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't get chunk info from skipped list");
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"coming to obtain each chunk address individually \n");
#endif
}
else
chunk_addr = total_chunk_addr_array[chunk_info->index];
chunk_addr_info_array[i].chunk_addr = chunk_addr;
chunk_addr_info_array[i].chunk_info = *chunk_info;
chunk_node = H5SL_next(chunk_node);
while(chunk_node) {
chunk_info = H5SL_item(chunk_node);
store.chunk.offset = chunk_info->coords;
store.chunk.index = chunk_info->index;
if(many_chunk_opt == H5D_OBTAIN_ONE_CHUNK_ADDR_IND){
if(HADDR_UNDEF==(chunk_addr = H5D_istore_get_addr(io_info,NULL)))
HGOTO_ERROR(H5E_STORAGE, H5E_CANTGET, FAIL,"couldn't get chunk info from skipped list");
}
else
chunk_addr = total_chunk_addr_array[chunk_info->index];
if(chunk_addr < chunk_addr_info_array[i].chunk_addr) do_sort = TRUE;
chunk_addr_info_array[i+1].chunk_addr = chunk_addr;
chunk_addr_info_array[i+1].chunk_info =*chunk_info;
i++;
chunk_node = H5SL_next(chunk_node);
}
#ifdef H5D_DEBUG
if(H5DEBUG(D))
HDfprintf(H5DEBUG(D),"before Qsort\n");
#endif
if(do_sort)
HDqsort(chunk_addr_info_array,num_chunks,sizeof(chunk_addr_info_array),H5D_cmp_chunk_addr);
done:
if(tchunk_addr_cleanup)
HDfree(total_chunk_addr_array);
if(mpi_type_cleanup) {
if (MPI_SUCCESS != (mpi_code= MPI_Type_free( &chunk_addrtype )))
HMPI_DONE_ERROR(FAIL, "MPI_Type_free failed", mpi_code);
}
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5D_sort_chunk() */
/*-------------------------------------------------------------------------
* Function: H5D_obtain_mpio_mode
*
* Purpose: Routine to obtain each io mode(collective,independent or none) for each chunk;
Each chunk address is also obtained.
Description:
1) Each process provides two piece of information for all chunks having selection
a) chunk index
b) wheather this chunk is regular(for MPI derived datatype not working case)
2) Gather all the information to the root process
3) Root process will do the following:
a) Obtain chunk addresses for all chunks in this data space
b) With the consideration of the user option, calculate IO mode for each chunk
c) Build MPI derived datatype to combine "chunk address" and "assign_io" information
in order to do MPI Bcast only once
d) MPI Bcast the IO mode and chunk address information for each chunk.
4) Each process then retrieves IO mode and chunk address information to assign_io_mode and chunk_addr.
Parameters:
Input: H5D_io_info_t* io_info,
fm_map *fm,(global chunk map struct)
Output: uint8_t assign_io_mode[], : IO mode, collective, independent or none
haddr_t chunk_addr[], : chunk address array for each chunk
*
* Return: Non-negative on success/Negative on failure
*
* Programmer:
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
static herr_t
H5D_obtain_mpio_mode(H5D_io_info_t* io_info,
fm_map *fm,
uint8_t assign_io_mode[],
haddr_t chunk_addr[])
{
int total_chunks;
hsize_t ori_total_chunks;
unsigned percent_nproc_per_chunk,threshold_nproc_per_chunk;
H5FD_mpio_chunk_opt_t chunk_opt_mode;
uint8_t* io_mode_info=NULL;
uint8_t* recv_io_mode_info=NULL;
uint8_t* mergebuf=NULL;
uint8_t* tempbuf;
H5SL_node_t* chunk_node;
H5D_chunk_info_t* chunk_info;
MPI_Datatype bastype[2];
MPI_Datatype chunk_addrtype;
int bascount;
int basblock[2];
MPI_Aint basdisp[2];
MPI_Datatype rtype;
MPI_Datatype stype;
int mpi_size,mpi_rank;
MPI_Comm comm;
int ic,root;
int mpi_code;
H5P_genplist_t *plist;
int mem_cleanup = 0,
mpi_type_cleanup = 0;
#ifdef H5_HAVE_INSTRUMENTED_LIBRARY
int new_value;
htri_t check_prop;
#endif
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI_NOINIT(H5D_obtain_mpio_mode)
/* Assign the rank 0 to the root */
root = 0;
comm = io_info->comm;
/* Obtain the number of process and the current rank of the process */
if((mpi_rank = H5F_mpi_get_rank(io_info->dset->oloc.file))<0)
HGOTO_ERROR (H5E_IO, H5E_MPI, FAIL, "unable to obtain mpi rank");
if((mpi_size = H5F_mpi_get_size(io_info->dset->oloc.file))<0)
HGOTO_ERROR (H5E_IO, H5E_MPI, FAIL, "unable to obtain mpi size");
/* Allocate memory */
ori_total_chunks = fm->total_chunks;
H5_ASSIGN_OVERFLOW(total_chunks,ori_total_chunks,hsize_t,int);
/* Obtain the data transfer properties */
if(NULL == (plist = H5I_object(io_info->dxpl_id)))
HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a file access property list")
percent_nproc_per_chunk=H5P_peek_unsigned(plist,H5D_XFER_MPIO_CHUNK_OPT_RATIO_NAME);
#if defined(H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS) && defined(H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS)
chunk_opt_mode=(H5FD_mpio_chunk_opt_t)H5P_peek_unsigned(plist,H5D_XFER_MPIO_CHUNK_OPT_HARD_NAME);
if((chunk_opt_mode == H5FD_MPIO_CHUNK_MULTI_IO) || (percent_nproc_per_chunk == 0)){
if(H5D_istore_chunkmap(io_info, chunk_addr, fm->down_chunks) < 0)
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk address");
for(ic = 0; ic<total_chunks;ic++)
assign_io_mode[ic] = H5D_CHUNK_IO_MODE_COL;
goto done;
}
#endif
threshold_nproc_per_chunk = mpi_size * percent_nproc_per_chunk/100;
io_mode_info = (uint8_t *)H5MM_calloc(total_chunks*sizeof(MPI_BYTE));
mergebuf = H5MM_malloc((sizeof(haddr_t)+sizeof(MPI_BYTE))*total_chunks);
tempbuf = mergebuf + sizeof(MPI_BYTE)*total_chunks;
if(mpi_rank == root)
recv_io_mode_info = (uint8_t *)H5MM_malloc(total_chunks*sizeof(MPI_BYTE)*mpi_size);
mem_cleanup = 1;
chunk_node = H5SL_first(fm->fsel);
/*Obtain the regularity and selection information for all chunks in this process. */
while(chunk_node){
chunk_info = H5SL_item(chunk_node);
#ifndef H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS
/* regularity information: 1, selection information: 2 */
if(H5S_SELECT_IS_REGULAR(chunk_info->fspace) == TRUE &&
H5S_SELECT_IS_REGULAR(chunk_info->mspace) == TRUE)
#endif
io_mode_info[chunk_info->index] = H5D_CHUNK_SELECT_REG; /* this chunk is selected and is "regular" without defining H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS. */
#ifndef H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS
else
io_mode_info[chunk_info->index] = H5D_CHUNK_SELECT_IRREG; /* this chunk is selected and is irregular*/
#endif
chunk_node = H5SL_next(chunk_node);
}
/*Create sent MPI derived datatype */
if(MPI_SUCCESS !=(mpi_code = MPI_Type_contiguous(total_chunks,MPI_BYTE,&stype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Comm_rank failed", mpi_code);
if(MPI_SUCCESS != (mpi_code = MPI_Type_commit(&stype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_commit failed", mpi_code);
/*Create received basic MPI derived datatype */
bascount = 2;
basblock[0] = total_chunks;
basblock[1] = total_chunks;
basdisp[0] = 0;
basdisp[1] = (MPI_Aint)(sizeof(MPI_BYTE)*total_chunks);/* may need to check overflow */
bastype[0] = MPI_BYTE;
if(MPI_SUCCESS !=(mpi_code = MPI_Type_contiguous(sizeof(haddr_t),MPI_BYTE,&chunk_addrtype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_contiguous failed", mpi_code);
if(MPI_SUCCESS != (mpi_code = MPI_Type_commit(&chunk_addrtype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_commit failed", mpi_code);
bastype[1] = chunk_addrtype;
if(MPI_SUCCESS !=(mpi_code = MPI_Type_struct(bascount,basblock,basdisp,bastype,&rtype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_struct failed", mpi_code);
if(MPI_SUCCESS !=(mpi_code = MPI_Type_commit(&rtype)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_commit failed", mpi_code);
/* Set up a flag to clean up the MPI derived datatype later */
mpi_type_cleanup = 1;
/*Gather all the information */
if(MPI_SUCCESS !=(mpi_code = MPI_Gather(io_mode_info,1,stype,recv_io_mode_info,1,stype,root,comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Gather failed", mpi_code);
/* Calculate the mode for IO(collective, independent or none) at root process */
if(mpi_rank == root) {
int nproc;
int* nproc_per_chunk;
#if !defined(H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS) || !defined(H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS)
int* ind_this_chunk;
#endif
/* pre-computing: calculate number of processes and
regularity of the selection occupied in each chunk */
nproc_per_chunk = (int*)H5MM_calloc(total_chunks*sizeof(int));
#if !defined(H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS) || !defined(H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS)
ind_this_chunk = (int*)H5MM_calloc(total_chunks*sizeof(int));
#endif
/* calculating the chunk address */
if(H5D_istore_chunkmap(io_info, chunk_addr, fm->down_chunks)<0){
HDfree(nproc_per_chunk);
#if !defined(H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS) || !defined(H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS)
HDfree(ind_this_chunk);
#endif
HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get chunk address");
}
/* checking for number of process per chunk and regularity of the selection*/
for (nproc = 0;nproc <mpi_size;nproc++){
uint8_t *tmp_recv_io_mode_info = recv_io_mode_info + nproc*total_chunks;
/* calculate the number of process per chunk and adding irregular selection option */
for(ic = 0; ic < total_chunks; ic++, tmp_recv_io_mode_info++){
if(*tmp_recv_io_mode_info != 0) {
nproc_per_chunk[ic]++;
#ifndef H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS
if(*tmp_recv_io_mode_info == H5D_CHUNK_SELECT_IRREG)
ind_this_chunk[ic] = 1;
#endif
}
#ifndef H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS
else {
/*checking whether we have a selection in this chunk */
ind_this_chunk[ic] = 1;
}
#endif
}
}
/* Calculating MPIO mode for each chunk (collective, independent, none) */
for(ic = 0; ic < total_chunks; ic++){
if(nproc_per_chunk[ic]>MAX(1,threshold_nproc_per_chunk)){
#if !defined(H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS) || !defined(H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS)
if(!ind_this_chunk[ic]) assign_io_mode[ic] = H5D_CHUNK_IO_MODE_COL;
#else
assign_io_mode[ic] = H5D_CHUNK_IO_MODE_COL;
#endif
}
}
/* merge buffer io_mode info and chunk addr into one */
HDmemcpy(mergebuf,assign_io_mode,sizeof(MPI_BYTE)*total_chunks);
HDmemcpy(tempbuf,chunk_addr,sizeof(haddr_t)*total_chunks);
HDfree(nproc_per_chunk);
#if !defined(H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS) || !defined(H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS)
HDfree(ind_this_chunk);
#endif
}
/* Broadcasting the MPI_IO option info. and chunk address info. */
if(MPI_SUCCESS !=(mpi_code = MPI_Bcast(mergebuf,1,rtype,root,comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_BCast failed", mpi_code);
HDmemcpy(assign_io_mode,mergebuf,sizeof(MPI_BYTE)*total_chunks);
HDmemcpy(chunk_addr,tempbuf,sizeof(haddr_t)*total_chunks);
#ifdef H5_HAVE_INSTRUMENTED_LIBRARY
check_prop = H5Pexist(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_MULTI_RATIO_COLL_NAME);
if(check_prop > 0) {
#if !defined(H5_MPI_COMPLEX_DERIVED_DATATYPE_WORKS) || !defined(H5_MPI_SPECIAL_COLLECTIVE_IO_WORKS)
new_value = 0;
if(H5Pset(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_MULTI_RATIO_COLL_NAME,&new_value)<0)
HGOTO_ERROR(H5E_PLIST, H5E_UNSUPPORTED, FAIL, "unable to set property value");
#else
for(ic = 0; ic < total_chunks; ic++){
if(assign_io_mode[ic] == H5D_CHUNK_IO_MODE_COL) {
new_value = 0;
if(H5Pset(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_MULTI_RATIO_COLL_NAME,&new_value)<0)
HGOTO_ERROR(H5E_PLIST, H5E_UNSUPPORTED, FAIL, "unable to set property value");
break;
}
}
#endif
}
check_prop = H5Pexist(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_MULTI_RATIO_IND_NAME);
if(check_prop > 0) {
int temp_count = 0;
for(ic = 0; ic < total_chunks; ic++){
if(assign_io_mode[ic] == H5D_CHUNK_IO_MODE_COL) {
temp_count++;
break;
}
}
if(temp_count==0){
new_value = 0;
if(H5Pset(io_info->dxpl_id,H5D_XFER_COLL_CHUNK_MULTI_RATIO_IND_NAME,&new_value)<0)
HGOTO_ERROR(H5E_PLIST, H5E_UNSUPPORTED, FAIL, "unable to set property value");
}
}
#endif
done:
if(mpi_type_cleanup) {
if (MPI_SUCCESS != (mpi_code= MPI_Type_free( &chunk_addrtype )))
HMPI_DONE_ERROR(FAIL, "MPI_Type_free failed", mpi_code);
if (MPI_SUCCESS != (mpi_code= MPI_Type_free( &stype )))
HMPI_DONE_ERROR(FAIL, "MPI_Type_free failed", mpi_code);
if (MPI_SUCCESS != (mpi_code= MPI_Type_free( &rtype )))
HMPI_DONE_ERROR(FAIL, "MPI_Type_free failed", mpi_code);
}
if(mem_cleanup){
HDfree(io_mode_info);
HDfree(mergebuf);
if(mpi_rank == root)
HDfree(recv_io_mode_info);
}
FUNC_LEAVE_NOAPI(ret_value)
}/* end H5D_obtain_mpio_mode*/
static int
H5D_cmp_chunk_addr(const void *chunk_addr_info1, const void *chunk_addr_info2)
{
haddr_t addr1, addr2;
FUNC_ENTER_NOAPI_NOINIT_NOFUNC(H5D_cmp_chunk_addr)
addr1 = ((const H5D_chunk_addr_info_t *)chunk_addr_info1)->chunk_addr;
addr2 = ((const H5D_chunk_addr_info_t *)chunk_addr_info2)->chunk_addr;
FUNC_LEAVE_NOAPI(H5F_addr_cmp(addr1, addr2))
}
#endif /* H5_HAVE_PARALLEL */
Reference in New Issue View Git Blame Copy Permalink