diff options
author | MuQun Yang <ymuqun@hdfgroup.org> | 2006-03-18 02:59:32 (GMT) |
---|---|---|
committer | MuQun Yang <ymuqun@hdfgroup.org> | 2006-03-18 02:59:32 (GMT) |
commit | 68385c1746a926f2ea876794f08622e919816e6b (patch) | |
tree | ab2e3bab6e55d4acf1f08650fd4ccccff8ccfc8e /src/H5Dmpio.c | |
parent | e2a8d4e04433220abc398e8a91511dc8733731d6 (diff) | |
download | hdf5-68385c1746a926f2ea876794f08622e919816e6b.zip hdf5-68385c1746a926f2ea876794f08622e919816e6b.tar.gz hdf5-68385c1746a926f2ea876794f08622e919816e6b.tar.bz2 |
[svn-r12117] Purpose:
Enhancing the optimiziation of collective IO per chunk
Description:
When the user does one of the following two things:
1. to do collective IO per chunk without using our optimization code
2. or the user passes the percent of number of process per chunk to be 0 when choosing
to do collective IO per chunk
It is not necessary that the library uses MPI-IO collective calls to do any optimization.
Solution:
Modify the code so that no MPI communication-involved analyses will be done for the above cases. Chunk addresses are obtained globally and IO modes are assigned to collective always.
Platforms tested:
h5committest
Misc. update:
Diffstat (limited to 'src/H5Dmpio.c')
-rw-r--r-- | src/H5Dmpio.c | 63 |
1 files changed, 25 insertions, 38 deletions
diff --git a/src/H5Dmpio.c b/src/H5Dmpio.c index 2d03283..d6c135f 100644 --- a/src/H5Dmpio.c +++ b/src/H5Dmpio.c @@ -111,9 +111,6 @@ typedef struct H5D_common_coll_info_t { 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, @@ -685,13 +682,9 @@ H5D_chunk_collective_io(H5D_io_info_t *io_info,fm_map *fm,const void *buf, hbool HGOTO_ERROR(H5E_IO, H5E_CANTGET, FAIL,"couldn't finish linked chunk MPI-IO"); } - else if(io_option == H5D_MULTI_CHUNK_IO_MORE_OPT) { /* 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"); - } else { /*multiple chunk IOs without opt */ - if(H5D_multi_chunk_collective_io_no_opt(io_info,fm,buf,do_write)<0) + 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"); } @@ -1041,6 +1034,8 @@ H5D_multi_chunk_collective_io(H5D_io_info_t *io_info,fm_map *fm,const void *buf, printf("total_chunk %d\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"); @@ -1161,29 +1156,6 @@ printf("inside independent IO mpi_rank = %d, chunk index = %d\n",mpi_rank,i); FUNC_LEAVE_NOAPI(ret_value) } /* end H5D_multi_chunk_collective_io */ -/*------------------------------------------------------------------------- - * Function: H5D_multi_chunk_collective_io_no_opt - * - * 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_no_opt(H5D_io_info_t *io_info,fm_map *fm,const void *buf, hbool_t do_write) -{ -return SUCCEED; -} /*------------------------------------------------------------------------- @@ -1594,6 +1566,7 @@ H5D_obtain_mpio_mode(H5D_io_info_t* io_info, 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; uint8_t* recv_io_mode_info; uint8_t* mergebuf; @@ -1611,7 +1584,7 @@ H5D_obtain_mpio_mode(H5D_io_info_t* io_info, MPI_Datatype stype; int mpi_size,mpi_rank; MPI_Comm comm; - int root; + int ic,root; int mpi_code; H5P_genplist_t *plist; int mem_cleanup = 0, @@ -1630,16 +1603,30 @@ H5D_obtain_mpio_mode(H5D_io_info_t* io_info, 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"); - /* Obtain the data transfer properties */ + + /* 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_OPT_MULTI_IO) || (percent_nproc_per_chunk == 0)){ + if(H5D_istore_chunkmap(io_info,total_chunks,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; - /* Allocate memory */ - ori_total_chunks = fm->total_chunks; - H5_ASSIGN_OVERFLOW(total_chunks,ori_total_chunks,hsize_t,int); io_mode_info = (uint8_t *)H5MM_calloc(total_chunks*sizeof(MPI_BYTE)); mergebuf = H5MM_malloc((sizeof(haddr_t)+sizeof(MPI_BYTE))*total_chunks); @@ -1706,7 +1693,7 @@ H5D_obtain_mpio_mode(H5D_io_info_t* io_info, /* Calculate the mode for IO(collective, independent or none) at root process */ if(mpi_rank == root) { - int ic,nproc; + 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; |