1 /* Copyright (c) 2007-2019. The SimGrid Team. All rights reserved. */
3 /* This program is free software; you can redistribute it and/or modify it
4 * under the terms of the license (GNU LGPL) which comes with this package. */
7 #include "smpi_comm.hpp"
8 #include "smpi_coll.hpp"
9 #include "smpi_datatype.hpp"
10 #include "smpi_info.hpp"
11 #include "smpi_win.hpp"
12 #include "smpi_request.hpp"
14 #include "smpi_file.hpp"
15 #include "smpi_status.hpp"
16 #include "simgrid/plugins/file_system.h"
18 #define FP_SIZE sizeof(MPI_Offset)
20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_io, smpi, "Logging specific to SMPI (RMA operations)");
25 File::File(MPI_Comm comm, char *filename, int amode, MPI_Info info): comm_(comm), flags_(amode), info_(info) {
26 file_= new simgrid::s4u::File(filename, nullptr);
28 if (comm_->rank() == 0) {
29 int size= comm_->size() + FP_SIZE;
30 list_ = new char[size];
31 memset(list_, 0, size);
32 shared_file_pointer_ = new MPI_Offset();
33 shared_mutex_ = s4u::Mutex::create();
34 *shared_file_pointer_ = 0;
35 win_=new Win(list_, size, 1, MPI_INFO_NULL, comm_);
37 win_=new Win(list_, 0, 1, MPI_INFO_NULL, comm_);
39 simgrid::smpi::Colls::bcast(&shared_file_pointer_, 1, MPI_AINT, 0, comm);
40 simgrid::smpi::Colls::bcast(&shared_mutex_, 1, MPI_AINT, 0, comm);
41 if(comm_->rank() != 0)
42 intrusive_ptr_add_ref(&*shared_mutex_);
46 if(comm_->rank() == 0){
47 delete shared_file_pointer_;
54 int File::close(MPI_File *fh){
55 XBT_DEBUG("Closing MPI_File %s", (*fh)->file_->get_path());
57 if((*fh)->flags() & MPI_MODE_DELETE_ON_CLOSE)
58 (*fh)->file_->unlink();
63 int File::del(char *filename, MPI_Info info){
64 //get the file with MPI_MODE_DELETE_ON_CLOSE and then close it
65 File* f = new File(MPI_COMM_SELF,filename,MPI_MODE_DELETE_ON_CLOSE|MPI_MODE_RDWR, nullptr);
70 int File::get_position(MPI_Offset* offset){
71 *offset=file_->tell();
75 int File::get_position_shared(MPI_Offset* offset){
76 shared_mutex_->lock();
77 *offset=*shared_file_pointer_;
78 shared_mutex_->unlock();
82 int File::seek(MPI_Offset offset, int whence){
85 XBT_VERB("Seeking in MPI_File %s, setting offset %lld", file_->get_path(), offset);
86 file_->seek(offset,SEEK_SET);
89 XBT_VERB("Seeking in MPI_File %s, current offset + %lld", file_->get_path(), offset);
90 file_->seek(offset,SEEK_CUR);
93 XBT_VERB("Seeking in MPI_File %s, end offset + %lld", file_->get_path(), offset);
94 file_->seek(offset,SEEK_END);
102 int File::seek_shared(MPI_Offset offset, int whence){
103 shared_mutex_->lock();
105 *shared_file_pointer_=offset;
106 shared_mutex_->unlock();
110 int File::read(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
111 //get position first as we may be doing non contiguous reads and it will probably be updated badly
112 MPI_Offset position = fh->file_->tell();
113 MPI_Offset movesize = datatype->get_extent()*count;
114 MPI_Offset readsize = datatype->size()*count;
115 XBT_DEBUG("Position before read in MPI_File %s : %llu",fh->file_->get_path(),fh->file_->tell());
116 MPI_Offset read = fh->file_->read(readsize);
117 XBT_VERB("Read in MPI_File %s, %lld bytes read, readsize %lld bytes, movesize %lld", fh->file_->get_path(), read, readsize, movesize);
118 if(readsize!=movesize){
119 fh->file_->seek(position+movesize, SEEK_SET);
121 XBT_VERB("Position after read in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());
122 status->count=count*datatype->size();
126 /*Ordered and Shared Versions, with RMA-based locks : Based on the model described in :*/
127 /* @InProceedings{10.1007/11557265_15,*/
128 /* author="Latham, Robert and Ross, Robert and Thakur, Rajeev and Toonen, Brian",*/
129 /* title="Implementing MPI-IO Shared File Pointers Without File System Support",*/
130 /* booktitle="Recent Advances in Parallel Virtual Machine and Message Passing Interface",*/
132 /* publisher="Springer Berlin Heidelberg",*/
133 /* address="Berlin, Heidelberg",*/
136 int File::read_shared(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
137 fh->shared_mutex_->lock();
138 fh->seek(*(fh->shared_file_pointer_),MPI_SEEK_SET);
139 read(fh, buf, count, datatype, status);
140 *(fh->shared_file_pointer_)=fh->file_->tell();
141 fh->shared_mutex_->unlock();
145 int File::read_ordered(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
146 //0 needs to get the shared pointer value
148 if(fh->comm_->rank()==0){
149 val=*(fh->shared_file_pointer_);
151 val=count*datatype->size();
155 simgrid::smpi::Colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);
156 fh->seek(result, MPI_SEEK_SET);
157 int ret = fh->op_all<simgrid::smpi::File::read>(buf, count, datatype, status);
158 if(fh->comm_->rank()==fh->comm_->size()-1){
159 fh->shared_mutex_->lock();
160 *(fh->shared_file_pointer_)=fh->file_->tell();
161 fh->shared_mutex_->unlock();
164 simgrid::smpi::Colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size()-1, fh->comm_);
168 int File::write(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
169 //get position first as we may be doing non contiguous reads and it will probably be updated badly
170 MPI_Offset position = fh->file_->tell();
171 MPI_Offset movesize = datatype->get_extent()*count;
172 MPI_Offset writesize = datatype->size()*count;
173 XBT_DEBUG("Position before write in MPI_File %s : %llu",fh->file_->get_path(),fh->file_->tell());
174 MPI_Offset write = fh->file_->write(writesize);
175 XBT_VERB("Write in MPI_File %s, %lld bytes written, readsize %lld bytes, movesize %lld", fh->file_->get_path(), write, writesize, movesize);
176 if(writesize!=movesize){
177 fh->file_->seek(position+movesize, SEEK_SET);
179 XBT_VERB("Position after write in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());
180 status->count=count*datatype->size();
184 int File::write_shared(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
185 fh->shared_mutex_->lock();
186 fh->seek(*(fh->shared_file_pointer_),MPI_SEEK_SET);
187 write(fh, buf, count, datatype, status);
188 *(fh->shared_file_pointer_)=fh->file_->tell();
189 fh->shared_mutex_->unlock();
193 int File::write_ordered(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
194 //0 needs to get the shared pointer value
196 if(fh->comm_->rank()==0){
197 val=*(fh->shared_file_pointer_);
199 val=count*datatype->size();
202 simgrid::smpi::Colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);
203 fh->seek(result, MPI_SEEK_SET);
204 int ret = fh->op_all<simgrid::smpi::File::write>(buf, count, datatype, status);
205 if(fh->comm_->rank()==fh->comm_->size()-1){
206 fh->shared_mutex_->lock();
207 *(fh->shared_file_pointer_)=fh->file_->tell();
208 fh->shared_mutex_->unlock();
211 simgrid::smpi::Colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size()-1, fh->comm_);
216 return file_->size();
225 return simgrid::smpi::Colls::barrier(comm_);
228 MPI_Info File::info(){
229 if(info_== MPI_INFO_NULL)
235 void File::set_info(MPI_Info info){
236 if(info_!= MPI_INFO_NULL)
241 MPI_Comm File::comm(){