1 /* Copyright (c) 2010-2019. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
7 #ifndef SMPI_FILE_HPP_INCLUDED
8 #define SMPI_FILE_HPP_INCLUDED
9 #include "simgrid/plugins/file_system.h"
10 #include "smpi_comm.hpp"
11 #include "smpi_coll.hpp"
12 #include "smpi_datatype.hpp"
13 #include "smpi_errhandler.hpp"
14 #include "smpi_info.hpp"
17 XBT_LOG_EXTERNAL_CATEGORY(smpi_pmpi);
24 simgrid::s4u::File* file_;
26 MPI_Offset* shared_file_pointer_;
27 s4u::MutexPtr shared_mutex_;
30 MPI_Errhandler errhandler_;
33 File(MPI_Comm comm, const char *filename, int amode, MPI_Info info);
34 File(const File&) = delete;
35 File& operator=(const File&) = delete;
38 int get_position(MPI_Offset* offset);
39 int get_position_shared(MPI_Offset* offset);
43 int seek(MPI_Offset offset, int whence);
44 int seek_shared(MPI_Offset offset, int whence);
46 void set_info( MPI_Info info);
47 static int read(MPI_File fh, void *buf, int count,MPI_Datatype datatype, MPI_Status *status);
48 static int read_shared(MPI_File fh, void *buf, int count,MPI_Datatype datatype, MPI_Status *status);
49 static int read_ordered(MPI_File fh, void *buf, int count,MPI_Datatype datatype, MPI_Status *status);
50 static int write(MPI_File fh, void *buf, int count,MPI_Datatype datatype, MPI_Status *status);
51 static int write_shared(MPI_File fh, const void *buf, int count,MPI_Datatype datatype, MPI_Status *status);
52 static int write_ordered(MPI_File fh, const void *buf, int count,MPI_Datatype datatype, MPI_Status *status);
53 template <int (*T)(MPI_File, void *, int, MPI_Datatype, MPI_Status *)> int op_all(void *buf, int count,MPI_Datatype datatype, MPI_Status *status);
54 static int close(MPI_File *fh);
55 static int del(const char *filename, MPI_Info info);
56 MPI_Errhandler errhandler();
57 void set_errhandler( MPI_Errhandler errhandler);
60 /* Read_all, Write_all : loosely based on */
61 /* @article{Thakur:1996:ETM:245875.245879,*/
62 /* author = {Thakur, Rajeev and Choudhary, Alok},*/
63 /* title = {An Extended Two-phase Method for Accessing Sections of Out-of-core Arrays},*/
64 /* journal = {Sci. Program.},*/
65 /* issue_date = {Winter 1996},*/
66 /* pages = {301--317},*/
68 template <int (*T)(MPI_File, void *, int, MPI_Datatype, MPI_Status *)>
69 int File::op_all(void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
70 //get min and max offsets from everyone.
71 int size = comm_->size();
72 int rank = comm_-> rank();
73 MPI_Offset min_offset = file_->tell();
74 MPI_Offset max_offset = (min_offset + count * datatype->size());//cheating, as we don't care about exact data location, we can skip extent
75 MPI_Offset* min_offsets = new MPI_Offset[size];
76 MPI_Offset* max_offsets = new MPI_Offset[size];
77 simgrid::smpi::Colls::allgather(&min_offset, 1, MPI_OFFSET, min_offsets, 1, MPI_OFFSET, comm_);
78 simgrid::smpi::Colls::allgather(&max_offset, 1, MPI_OFFSET, max_offsets, 1, MPI_OFFSET, comm_);
79 MPI_Offset min=min_offset;
80 MPI_Offset max=max_offset;
83 for(int i=0;i<size;i++){
84 if(min_offsets[i]!=max_offsets[i])
86 tot+=(max_offsets[i]-min_offsets[i]);
87 if(min_offsets[i]<min)
89 if(max_offsets[i]>max)
93 XBT_CDEBUG(smpi_pmpi, "my offsets to read : %lld:%lld, global min and max %lld:%lld", min_offset, max_offset, min, max);
100 MPI_Offset total = max-min;
101 if(total==tot && (datatype->flags() & DT_FLAG_CONTIGUOUS)){
102 delete[] min_offsets;
103 delete[] max_offsets;
104 //contiguous. Just have each proc perform its read
105 status->count=count * datatype->size();
106 return T(this,buf,count,datatype, status);
109 //Interleaved case : How much do I need to read, and whom to send it ?
110 MPI_Offset my_chunk_start=(max-min+1)/size*rank;
111 MPI_Offset my_chunk_end=((max-min+1)/size*(rank+1));
112 XBT_CDEBUG(smpi_pmpi, "my chunks to read : %lld:%lld", my_chunk_start, my_chunk_end);
113 int* send_sizes = new int[size];
114 int* recv_sizes = new int[size];
115 int* send_disps = new int[size];
116 int* recv_disps = new int[size];
118 for(int i=0;i<size;i++){
120 send_disps[i]=0;//cheat to avoid issues when send>recv as we use recv buffer
121 if((my_chunk_start>=min_offsets[i] && my_chunk_start < max_offsets[i])||
122 ((my_chunk_end<=max_offsets[i]) && my_chunk_end> min_offsets[i])){
123 send_sizes[i]=(std::min(max_offsets[i]-1, my_chunk_end-1)-std::max(min_offsets[i], my_chunk_start));
124 //store min and max offest to actually read
125 min_offset=std::min(min_offset, min_offsets[i]);
126 total_sent+=send_sizes[i];
127 XBT_CDEBUG(smpi_pmpi, "will have to send %d bytes to %d", send_sizes[i], i);
130 min_offset=std::max(min_offset, my_chunk_start);
132 //merge the ranges of every process
133 std::vector<std::pair<MPI_Offset, MPI_Offset>> ranges;
134 for(int i=0; i<size; ++i)
135 ranges.push_back(std::make_pair(min_offsets[i],max_offsets[i]));
136 std::sort(ranges.begin(), ranges.end());
137 std::vector<std::pair<MPI_Offset, MPI_Offset>> chunks;
138 chunks.push_back(ranges[0]);
140 unsigned int nchunks=0;
142 while(i < ranges.size()){
143 if(ranges[i].second>chunks[nchunks].second){
144 // else range included - ignore
145 if(ranges[i].first>chunks[nchunks].second){
147 chunks.push_back(ranges[i]);
151 chunks[nchunks].second=ranges[i].second;
156 //what do I need to read ?
157 MPI_Offset totreads=0;
158 for(i=0; i<chunks.size();i++){
159 if(chunks[i].second < my_chunk_start)
161 else if (chunks[i].first > my_chunk_end)
164 totreads += (std::min(chunks[i].second, my_chunk_end-1)-std::max(chunks[i].first, my_chunk_start));
166 XBT_CDEBUG(smpi_pmpi, "will have to access %lld from my chunk", totreads);
168 unsigned char* sendbuf = smpi_get_tmp_sendbuffer(total_sent);
171 seek(min_offset, MPI_SEEK_SET);
172 T(this,sendbuf,totreads/datatype->size(),datatype, status);
174 simgrid::smpi::Colls::alltoall(send_sizes, 1, MPI_INT, recv_sizes, 1, MPI_INT, comm_);
176 for(int i=0;i<size;i++){
177 recv_disps[i]=total_recv;
178 total_recv+=recv_sizes[i];
180 //Set buf value to avoid copying dumb data
181 simgrid::smpi::Colls::alltoallv(sendbuf, send_sizes, send_disps, MPI_BYTE,
182 buf, recv_sizes, recv_disps, MPI_BYTE, comm_);
183 status->count=count * datatype->size();
184 smpi_free_tmp_buffer(sendbuf);
189 delete[] min_offsets;
190 delete[] max_offsets;