/***************************************************************************** * * * Copyright (c) 2003-2006 Intel Corporation. * * All rights reserved. * * * ***************************************************************************** This code is covered by the Community Source License (CPL), version 1.0 as published by IBM and reproduced in the file "license.txt" in the "license" subdirectory. Redistribution in source and binary form, with or without modification, is permitted ONLY within the regulations contained in above mentioned license. Use of the name and trademark "Intel(R) MPI Benchmarks" is allowed ONLY within the regulations of the "License for Use of "Intel(R) MPI Benchmarks" Name and Trademark" as reproduced in the file "use-of-trademark-license.txt" in the "license" subdirectory. THE PROGRAM IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OR CONDITIONS OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Each Recipient is solely responsible for determining the appropriateness of using and distributing the Program and assumes all risks associated with its exercise of rights under this Agreement, including but not limited to the risks and costs of program errors, compliance with applicable laws, damage to or loss of data, programs or equipment, and unavailability or interruption of operations. EXCEPT AS EXPRESSLY SET FORTH IN THIS AGREEMENT, NEITHER RECIPIENT NOR ANY CONTRIBUTORS SHALL HAVE ANY LIABILITY FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING WITHOUT LIMITATION LOST PROFITS), HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OR DISTRIBUTION OF THE PROGRAM OR THE EXERCISE OF ANY RIGHTS GRANTED HEREUNDER, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. EXPORT LAWS: THIS LICENSE ADDS NO RESTRICTIONS TO THE EXPORT LAWS OF YOUR JURISDICTION. It is licensee's responsibility to comply with any export regulations applicable in licensee's jurisdiction. Under CURRENT U.S. export regulations this software is eligible for export from the U.S. and can be downloaded by or otherwise exported or reexported worldwide EXCEPT to U.S. embargoed destinations which include Cuba, Iraq, Libya, North Korea, Iran, Syria, Sudan, Afghanistan and any other country to which the U.S. has embargoed goods and services. *************************************************************************** For more documentation than found here, see [1] doc/ReadMe_IMB.txt [2] Intel (R) MPI Benchmarks Users Guide and Methodology Description In doc/IMB_ug.pdf File: IMB_read.c Implemented functions: IMB_read_shared; IMB_read_indv; IMB_read_expl; IMB_read_ij; IMB_iread_ij; ***************************************************************************/ #include "IMB_declare.h" #include "IMB_benchmark.h" #include "IMB_prototypes.h" /*************************************************************************/ void IMB_read_shared(struct comm_info* c_info, int size, int n_sample, MODES RUN_MODE, double* time) /* MPI-IO benchmark kernel Driver for read benchmarks with shared file pointers Input variables: -c_info (type struct comm_info*) Collection of all base data for MPI; see [1] for more information -size (type int) Basic message size in bytes -n_sample (type int) Number of repetitions (for timing accuracy) -RUN_MODE (type MODES) Mode (aggregate/non aggregate; blocking/nonblocking); see "IMB_benchmark.h" for definition Output variables: -time (type double*) Timing result per sample */ { if( RUN_MODE->AGGREGATE ) IMB_read_ij(c_info, size, shared, RUN_MODE->type, 1, n_sample, 1, time); else IMB_read_ij(c_info, size, shared, RUN_MODE->type, n_sample, 1, 0, time); if( RUN_MODE->NONBLOCKING ) { MPI_File_close(&c_info->fh); IMB_open_file(c_info); if( RUN_MODE->AGGREGATE ) IMB_iread_ij(c_info, size, shared, RUN_MODE->type, 1, n_sample, 1, 1, time+1); else IMB_iread_ij(c_info, size, shared, RUN_MODE->type, n_sample, 1, 0, 1, time+1); } } /*************************************************************************/ void IMB_read_indv(struct comm_info* c_info, int size, int n_sample, MODES RUN_MODE, double* time) /* MPI-IO benchmark kernel Driver for read benchmarks with individual file pointers Input variables: -c_info (type struct comm_info*) Collection of all base data for MPI; see [1] for more information -size (type int) Basic message size in bytes -n_sample (type int) Number of repetitions (for timing accuracy) -RUN_MODE (type MODES) Mode (aggregate/non aggregate; blocking/nonblocking); see "IMB_benchmark.h" for definition Output variables: -time (type double*) Timing result per sample */ { if( RUN_MODE->AGGREGATE ) IMB_read_ij(c_info, size, indv_block, RUN_MODE->type, 1, n_sample, 1, time); else IMB_read_ij(c_info, size, indv_block, RUN_MODE->type, n_sample, 1, 0, time); if( RUN_MODE->NONBLOCKING ) { MPI_File_close(&c_info->fh); IMB_open_file(c_info); if( RUN_MODE->AGGREGATE ) IMB_iread_ij(c_info, size, indv_block, RUN_MODE->type, 1, n_sample, 1, 1, time+1); else IMB_iread_ij(c_info, size, indv_block, RUN_MODE->type, n_sample, 1, 0, 1, time+1); } } void IMB_read_expl(struct comm_info* c_info, int size, int n_sample, MODES RUN_MODE, double* time) /* MPI-IO benchmark kernel Driver for read benchmarks with explicit offsets Input variables: -c_info (type struct comm_info*) Collection of all base data for MPI; see [1] for more information -size (type int) Basic message size in bytes -n_sample (type int) Number of repetitions (for timing accuracy) -RUN_MODE (type MODES) Mode (aggregate/non aggregate; blocking/nonblocking); see "IMB_benchmark.h" for definition Output variables: -time (type double*) Timing result per sample */ { if( RUN_MODE->AGGREGATE ) IMB_read_ij(c_info, size, explicit, RUN_MODE->type, 1, n_sample, 1, time); else IMB_read_ij(c_info, size, explicit, RUN_MODE->type, n_sample, 1, 0, time); if( RUN_MODE->NONBLOCKING ) { MPI_File_close(&c_info->fh); IMB_open_file(c_info); if( RUN_MODE->AGGREGATE ) IMB_iread_ij(c_info, size, explicit, RUN_MODE->type, 1, n_sample, 1, 1, time+1); else IMB_iread_ij(c_info, size, explicit, RUN_MODE->type, n_sample, 1, 0, 1, time+1); } } void IMB_read_ij(struct comm_info* c_info, int size, POSITIONING pos, BTYPES type, int i_sample, int j_sample, int time_inner, double* time) /* Calls the proper read functions, blocking case Input variables: -c_info (type struct comm_info*) Collection of all base data for MPI; see [1] for more information -size (type int) portion size in bytes -pos (type POSITIONING) (see IMB_benchmark.h for definition of this enum type) descriptor for the file positioning -type (type BTYPES) (see IMB_benchmark.h for definition of this enum type) descriptor for the file access synchronism -i_sample,j_sample (type int) aggregate case: i_sample=1, j_sample=n_sample (set by driving function) non aggregate case: i_sample=n_sample, j_sample=1 (set by driving function) Benchmark logistics then: for( i=0 .. i_sample-1 ) for( j=0 .. j_sample-1 ) input ... Synchronize (!) -time_inner (type int) logical flag: should timing be issued for the innermost loop (and then averaged by outermost count), or for outermost loop Output variables: -time (type double*) Timing result per sample */ { int i, j; int Locsize, Totalsize; MPI_Status stat; MPI_Offset Offset; int (* GEN_File_read)(MPI_File fh, void* buf, int count, MPI_Datatype datatype, MPI_Status *status); int (* GEN_File_read_shared) (MPI_File fh, void* buf, int count, MPI_Datatype datatype, MPI_Status *status); int (* GEN_File_read_at) (MPI_File fh, MPI_Offset offset, void* buf, int count, MPI_Datatype datatype, MPI_Status *status); #ifdef CHECK defect = 0.; #endif ierr = 0; *time = 0.; if( c_info->File_rank >= 0 ) { if (type == Collective ) { GEN_File_read = MPI_File_read_all; GEN_File_read_shared = MPI_File_read_ordered; GEN_File_read_at = MPI_File_read_at_all; #ifdef DEBUG fprintf(unit,"Collective input\n"); #endif } else { GEN_File_read = MPI_File_read; GEN_File_read_shared = MPI_File_read_shared; GEN_File_read_at = MPI_File_read_at; #ifdef DEBUG fprintf(unit,"Non collective input\n"); #endif } Locsize = c_info->split.Locsize; Totalsize = c_info->split.Totalsize; /* !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! BLOCKING COLLECTIVE/NON COLLECTIVE INPUT CASES COMBINED (function pointer GEN_File_read_XXX either standard or collective MPI_File_read_XXX !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! */ if( !time_inner ) *time = MPI_Wtime(); for ( i=0; iFile_comm); *time = MPI_Wtime(); } if( pos == indv_block ) { for( j=0; jfh,c_info->r_buffer,Locsize,c_info->etype,&stat); MPI_ERRHAND(ierr); DIAGNOSTICS("Read standard ",c_info,c_info->r_buffer,Locsize,Totalsize,j,pos); CHK_DIFF("Read_indv",c_info, c_info->r_buffer, 0, Locsize, Totalsize, asize, get, pos, i_sample*j_sample, j, -2, &defect); } } else if( pos == explicit ) { for( j=0; jsplit.Offset+(MPI_Offset)(j*Totalsize); ierr=GEN_File_read_at (c_info->fh, Offset, c_info->r_buffer,Locsize,c_info->etype,&stat); MPI_ERRHAND(ierr); DIAGNOSTICS("Read explicit ",c_info,c_info->r_buffer,Locsize,Totalsize,j,pos); CHK_DIFF("Read_expl",c_info, c_info->r_buffer, 0, Locsize, Totalsize, asize, get, pos, i_sample*j_sample, j, -2, &defect); } } else if( pos == shared ) { for( j=0; jfh,c_info->r_buffer,Locsize,c_info->etype,&stat); MPI_ERRHAND(ierr); DIAGNOSTICS("Read shared ",c_info,c_info->r_buffer,Locsize,Totalsize,j,pos); #ifdef CHECK IMB_chk_diff("Read_shared",c_info, c_info->r_buffer, 0, Locsize, Totalsize, asize, get, pos, i_sample*j_sample, j, -3, &defect); MPI_Barrier(c_info->File_comm); #endif } } if( time_inner ) *time = (MPI_Wtime()-*time)/(i_sample*j_sample); } if( !time_inner ) *time = (MPI_Wtime() - *time)/(i_sample*j_sample); } /* end if FIle_rank >= 0 */ } void IMB_iread_ij(struct comm_info* c_info, int size, POSITIONING pos, BTYPES type, int i_sample, int j_sample, int time_inner, int do_ovrlp, double* time) /* Calls the proper read functions, non blocking case (See IMB_read_ij for documentation of calling sequence) */ { int i, j; int Locsize, Totalsize; MPI_Status *STAT, stat; MPI_Request *REQUESTS; MPI_Offset Offset; #ifdef CHECK defect = 0.; #endif ierr = 0; *time = 0.; if( c_info->File_rank >= 0 ) { Locsize = c_info->split.Locsize; Totalsize = c_info->split.Totalsize; if( type == Collective ) /* !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! NON BLOCKING COLLECTIVE INPUT CASES !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! */ { for(i=0; iFile_comm); *time = MPI_Wtime(); if( pos == indv_block ) for ( j=0; jfh,c_info->r_buffer,Locsize,c_info->etype); MPI_ERRHAND(ierr); if( do_ovrlp ) IMB_cpu_exploit(TARGET_CPU_SECS,0); ierr=MPI_File_read_all_end (c_info->fh,c_info->r_buffer,&stat); MPI_ERRHAND(ierr); DIAGNOSTICS("IRead coll. ",c_info,c_info->r_buffer,Locsize,Totalsize,j,pos); CHK_DIFF("Coll. IRead_indv",c_info, c_info->r_buffer, 0, Locsize, Totalsize, asize, get, pos, i_sample*j_sample, j, -2, &defect); } else if ( pos == explicit ) for ( j=0; jsplit.Offset+(MPI_Offset)(j*Totalsize); ierr=MPI_File_read_at_all_begin (c_info->fh,Offset,c_info->r_buffer,Locsize,c_info->etype); MPI_ERRHAND(ierr); if( do_ovrlp ) IMB_cpu_exploit(TARGET_CPU_SECS,0); ierr=MPI_File_read_at_all_end (c_info->fh,c_info->r_buffer,&stat); MPI_ERRHAND(ierr); DIAGNOSTICS("IRead expl coll. ",c_info,c_info->r_buffer,Locsize,Totalsize,j,pos); CHK_DIFF("Coll. IRead_expl",c_info, c_info->r_buffer, 0, Locsize, Totalsize, asize, get, pos, i_sample*j_sample, j, -2, &defect); } else if ( pos == shared ) for ( j=0; jfh,c_info->r_buffer,Locsize,c_info->etype); MPI_ERRHAND(ierr); if( do_ovrlp ) IMB_cpu_exploit(TARGET_CPU_SECS,0); ierr=MPI_File_read_ordered_end (c_info->fh,c_info->r_buffer,&stat); MPI_ERRHAND(ierr); DIAGNOSTICS("IRead shared coll. ",c_info,c_info->r_buffer,Locsize,Totalsize,j,pos); CHK_DIFF("Coll. IRead_shared",c_info, c_info->r_buffer, 0, Locsize, Totalsize, asize, get, pos, i_sample*j_sample, j, -3, &defect); } *time = (MPI_Wtime() - *time)/(i_sample*j_sample); } else /* type non-Collective */ /* !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! NON BLOCKING NON COLLECTIVE INPUT CASES !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! */ { REQUESTS = (MPI_Request*)IMB_v_alloc(j_sample*sizeof(MPI_Request), "IRead_ij"); STAT = (MPI_Status *)IMB_v_alloc(j_sample*sizeof(MPI_Status ), "IRead_ij"); for( j=0; jFile_comm); if( !time_inner ) *time = MPI_Wtime(); for ( i=0; iFile_comm); *time = MPI_Wtime(); } if( pos == indv_block ) { for( j=0; jfh,c_info->r_buffer,Locsize,c_info->etype,&REQUESTS[j]); MPI_ERRHAND(ierr); #if (defined CHECK || defined DEBUG) MPI_Wait(REQUESTS+j,STAT); DIAGNOSTICS("IRead standard ",c_info,c_info->r_buffer,Locsize,Totalsize,j,pos); CHK_DIFF("IRead_indv",c_info, c_info->r_buffer, 0, Locsize, Totalsize, asize, get, pos, i_sample*j_sample, j, -2, &defect); #endif } } else if( pos == explicit ) { for( j=0; jsplit.Offset+(MPI_Offset)(j*Totalsize); ierr=MPI_File_iread_at (c_info->fh,Offset,c_info->r_buffer,Locsize,c_info->etype,&REQUESTS[j]); MPI_ERRHAND(ierr); #if (defined CHECK || defined DEBUG) MPI_Wait(REQUESTS+j,STAT); DIAGNOSTICS("IRead expl ",c_info,c_info->r_buffer,Locsize,Totalsize,j,pos); CHK_DIFF("IRead_expl",c_info, c_info->r_buffer, 0, Locsize, Totalsize, asize, get, pos, i_sample*j_sample, j, -2, &defect); #endif } } else if( pos == shared ) { for( j=0; jfh,c_info->r_buffer,Locsize,c_info->etype,&REQUESTS[j]); MPI_ERRHAND(ierr); #if (defined CHECK || defined DEBUG) MPI_Wait(REQUESTS+j,STAT); DIAGNOSTICS("IRead shared ",c_info,c_info->r_buffer,Locsize,Totalsize,j,pos); CHK_DIFF("IRead_shared",c_info, c_info->r_buffer, 0, Locsize, Totalsize, asize, get, pos, i_sample*j_sample, j, -3, &defect); MPI_Barrier(c_info->File_comm); #endif } } if( do_ovrlp ) for ( j=0; j= 0 */ }