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/s4u/Disk.hpp"
17 #include "simgrid/s4u/Host.hpp"
18 #include "simgrid/plugins/file_system.h"
20 #define FP_SIZE sizeof(MPI_Offset)
22 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_io, smpi, "Logging specific to SMPI (RMA operations)");
24 MPI_Errhandler SMPI_default_File_Errhandler = MPI_ERRORS_RETURN;
29 File::File(MPI_Comm comm, const char *filename, int amode, MPI_Info info): comm_(comm), flags_(amode), info_(info) {
30 std::string fullname=filename;
31 if (simgrid::s4u::Host::current()->get_disks().empty())
32 xbt_die("SMPI/IO : Trying to open file on a diskless host ! Add one to your platform file");
34 size_t found=fullname.find('/');
35 //in case no fullpath is provided ... just pick the first mountpoint.
36 if(found==std::string::npos){
37 auto disk = simgrid::s4u::Host::current()->get_disks().front();
39 if (disk->get_host() != simgrid::s4u::Host::current())
40 mount = disk->extension<simgrid::s4u::FileSystemDiskExt>()->get_mount_point(disk->get_host());
42 mount = disk->extension<simgrid::s4u::FileSystemDiskExt>()->get_mount_point();
43 XBT_DEBUG("No absolute path given for file opening, use '%s'", mount.c_str());
44 fullname.insert(0, mount);
47 file_= new simgrid::s4u::File(fullname, nullptr);
49 if (comm_->rank() == 0) {
50 int size= comm_->size() + FP_SIZE;
51 list_ = new char[size];
52 errhandler_= SMPI_default_File_Errhandler;
53 memset(list_, 0, size);
54 shared_file_pointer_ = new MPI_Offset();
55 shared_mutex_ = s4u::Mutex::create();
56 *shared_file_pointer_ = 0;
57 win_=new Win(list_, size, 1, MPI_INFO_NULL, comm_);
59 win_=new Win(list_, 0, 1, MPI_INFO_NULL, comm_);
61 simgrid::smpi::colls::bcast(&shared_file_pointer_, 1, MPI_AINT, 0, comm);
62 simgrid::smpi::colls::bcast(&shared_mutex_, 1, MPI_AINT, 0, comm);
63 if(comm_->rank() != 0)
64 intrusive_ptr_add_ref(&*shared_mutex_);
68 if(comm_->rank() == 0){
69 delete shared_file_pointer_;
76 int File::close(MPI_File *fh){
77 XBT_DEBUG("Closing MPI_File %s", (*fh)->file_->get_path());
79 if((*fh)->flags() & MPI_MODE_DELETE_ON_CLOSE)
80 (*fh)->file_->unlink();
85 int File::del(const char* filename, MPI_Info)
87 //get the file with MPI_MODE_DELETE_ON_CLOSE and then close it
88 File* f = new File(MPI_COMM_SELF,filename,MPI_MODE_DELETE_ON_CLOSE|MPI_MODE_RDWR, nullptr);
93 int File::get_position(MPI_Offset* offset){
94 *offset=file_->tell();
98 int File::get_position_shared(MPI_Offset* offset){
99 shared_mutex_->lock();
100 *offset=*shared_file_pointer_;
101 shared_mutex_->unlock();
105 int File::seek(MPI_Offset offset, int whence){
108 XBT_VERB("Seeking in MPI_File %s, setting offset %lld", file_->get_path(), offset);
109 file_->seek(offset,SEEK_SET);
112 XBT_VERB("Seeking in MPI_File %s, current offset + %lld", file_->get_path(), offset);
113 file_->seek(offset,SEEK_CUR);
116 XBT_VERB("Seeking in MPI_File %s, end offset + %lld", file_->get_path(), offset);
117 file_->seek(offset,SEEK_END);
125 int File::seek_shared(MPI_Offset offset, int whence){
126 shared_mutex_->lock();
128 *shared_file_pointer_=offset;
129 shared_mutex_->unlock();
133 int File::read(MPI_File fh, void* /*buf*/, int count, MPI_Datatype datatype, MPI_Status* status)
135 //get position first as we may be doing non contiguous reads and it will probably be updated badly
136 MPI_Offset position = fh->file_->tell();
137 MPI_Offset movesize = datatype->get_extent()*count;
138 MPI_Offset readsize = datatype->size()*count;
139 XBT_DEBUG("Position before read in MPI_File %s : %llu",fh->file_->get_path(),fh->file_->tell());
140 MPI_Offset read = fh->file_->read(readsize);
141 XBT_VERB("Read in MPI_File %s, %lld bytes read, readsize %lld bytes, movesize %lld", fh->file_->get_path(), read, readsize, movesize);
142 if(readsize!=movesize){
143 fh->file_->seek(position+movesize, SEEK_SET);
145 XBT_VERB("Position after read in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());
146 if(status != MPI_STATUS_IGNORE)
147 status->count=count*datatype->size();
151 /*Ordered and Shared Versions, with RMA-based locks : Based on the model described in :*/
152 /* @InProceedings{10.1007/11557265_15,*/
153 /* author="Latham, Robert and Ross, Robert and Thakur, Rajeev and Toonen, Brian",*/
154 /* title="Implementing MPI-IO Shared File Pointers Without File System Support",*/
155 /* booktitle="Recent Advances in Parallel Virtual Machine and Message Passing Interface",*/
157 /* publisher="Springer Berlin Heidelberg",*/
158 /* address="Berlin, Heidelberg",*/
161 int File::read_shared(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
162 fh->shared_mutex_->lock();
163 fh->seek(*(fh->shared_file_pointer_),MPI_SEEK_SET);
164 read(fh, buf, count, datatype, status);
165 *(fh->shared_file_pointer_)=fh->file_->tell();
166 fh->shared_mutex_->unlock();
170 int File::read_ordered(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
171 //0 needs to get the shared pointer value
173 if(fh->comm_->rank()==0){
174 val=*(fh->shared_file_pointer_);
176 val=count*datatype->size();
180 simgrid::smpi::colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);
181 fh->seek(result, MPI_SEEK_SET);
182 int ret = fh->op_all<simgrid::smpi::File::read>(buf, count, datatype, status);
183 if(fh->comm_->rank()==fh->comm_->size()-1){
184 fh->shared_mutex_->lock();
185 *(fh->shared_file_pointer_)=fh->file_->tell();
186 fh->shared_mutex_->unlock();
189 simgrid::smpi::colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size() - 1, fh->comm_);
193 int File::write(MPI_File fh, void* /*buf*/, int count, MPI_Datatype datatype, MPI_Status* status)
195 //get position first as we may be doing non contiguous reads and it will probably be updated badly
196 MPI_Offset position = fh->file_->tell();
197 MPI_Offset movesize = datatype->get_extent()*count;
198 MPI_Offset writesize = datatype->size()*count;
199 XBT_DEBUG("Position before write in MPI_File %s : %llu",fh->file_->get_path(),fh->file_->tell());
200 MPI_Offset write = fh->file_->write(writesize, 1);
201 XBT_VERB("Write in MPI_File %s, %lld bytes written, readsize %lld bytes, movesize %lld", fh->file_->get_path(), write, writesize, movesize);
202 if(writesize!=movesize){
203 fh->file_->seek(position+movesize, SEEK_SET);
205 XBT_VERB("Position after write in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());
206 if(status != MPI_STATUS_IGNORE)
207 status->count=count*datatype->size();
211 int File::write_shared(MPI_File fh, const void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
212 fh->shared_mutex_->lock();
213 fh->seek(*(fh->shared_file_pointer_),MPI_SEEK_SET);
214 write(fh, const_cast<void*>(buf), count, datatype, status);
215 *(fh->shared_file_pointer_)=fh->file_->tell();
216 fh->shared_mutex_->unlock();
220 int File::write_ordered(MPI_File fh, const void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
221 //0 needs to get the shared pointer value
223 if(fh->comm_->rank()==0){
224 val=*(fh->shared_file_pointer_);
226 val=count*datatype->size();
229 simgrid::smpi::colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);
230 fh->seek(result, MPI_SEEK_SET);
231 int ret = fh->op_all<simgrid::smpi::File::write>(const_cast<void*>(buf), count, datatype, status);
232 if(fh->comm_->rank()==fh->comm_->size()-1){
233 fh->shared_mutex_->lock();
234 *(fh->shared_file_pointer_)=fh->file_->tell();
235 fh->shared_mutex_->unlock();
238 simgrid::smpi::colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size() - 1, fh->comm_);
243 return file_->size();
252 return simgrid::smpi::colls::barrier(comm_);
255 MPI_Info File::info(){
256 if(info_== MPI_INFO_NULL)
262 void File::set_info(MPI_Info info){
263 if(info_!= MPI_INFO_NULL)
268 MPI_Comm File::comm(){
272 MPI_Errhandler File::errhandler(){
276 void File::set_errhandler(MPI_Errhandler errhandler){
277 errhandler_=errhandler;
278 if(errhandler_!= MPI_ERRHANDLER_NULL)