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());
45 fullname.insert(0, mount);
48 file_= new simgrid::s4u::File(fullname, nullptr);
50 if (comm_->rank() == 0) {
51 int size= comm_->size() + FP_SIZE;
52 list_ = new char[size];
53 errhandler_= SMPI_default_File_Errhandler;
54 memset(list_, 0, size);
55 shared_file_pointer_ = new MPI_Offset();
56 shared_mutex_ = s4u::Mutex::create();
57 *shared_file_pointer_ = 0;
58 win_=new Win(list_, size, 1, MPI_INFO_NULL, comm_);
60 win_=new Win(list_, 0, 1, MPI_INFO_NULL, comm_);
62 simgrid::smpi::colls::bcast(&shared_file_pointer_, 1, MPI_AINT, 0, comm);
63 simgrid::smpi::colls::bcast(&shared_mutex_, 1, MPI_AINT, 0, comm);
64 if(comm_->rank() != 0)
65 intrusive_ptr_add_ref(&*shared_mutex_);
69 if(comm_->rank() == 0){
70 delete shared_file_pointer_;
77 int File::close(MPI_File *fh){
78 XBT_DEBUG("Closing MPI_File %s", (*fh)->file_->get_path());
80 if((*fh)->flags() & MPI_MODE_DELETE_ON_CLOSE)
81 (*fh)->file_->unlink();
86 int File::del(const char* filename, MPI_Info)
88 //get the file with MPI_MODE_DELETE_ON_CLOSE and then close it
89 File* f = new File(MPI_COMM_SELF,filename,MPI_MODE_DELETE_ON_CLOSE|MPI_MODE_RDWR, nullptr);
94 int File::get_position(MPI_Offset* offset){
95 *offset=file_->tell();
99 int File::get_position_shared(MPI_Offset* offset){
100 shared_mutex_->lock();
101 *offset=*shared_file_pointer_;
102 shared_mutex_->unlock();
106 int File::seek(MPI_Offset offset, int whence){
109 XBT_VERB("Seeking in MPI_File %s, setting offset %lld", file_->get_path(), offset);
110 file_->seek(offset,SEEK_SET);
113 XBT_VERB("Seeking in MPI_File %s, current offset + %lld", file_->get_path(), offset);
114 file_->seek(offset,SEEK_CUR);
117 XBT_VERB("Seeking in MPI_File %s, end offset + %lld", file_->get_path(), offset);
118 file_->seek(offset,SEEK_END);
126 int File::seek_shared(MPI_Offset offset, int whence){
127 shared_mutex_->lock();
129 *shared_file_pointer_=offset;
130 shared_mutex_->unlock();
134 int File::read(MPI_File fh, void* /*buf*/, int count, MPI_Datatype datatype, MPI_Status* status)
136 //get position first as we may be doing non contiguous reads and it will probably be updated badly
137 MPI_Offset position = fh->file_->tell();
138 MPI_Offset movesize = datatype->get_extent()*count;
139 MPI_Offset readsize = datatype->size()*count;
140 XBT_DEBUG("Position before read in MPI_File %s : %llu",fh->file_->get_path(),fh->file_->tell());
141 MPI_Offset read = fh->file_->read(readsize);
142 XBT_VERB("Read in MPI_File %s, %lld bytes read, readsize %lld bytes, movesize %lld", fh->file_->get_path(), read, readsize, movesize);
143 if(readsize!=movesize){
144 fh->file_->seek(position+movesize, SEEK_SET);
146 XBT_VERB("Position after read in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());
147 if(status != MPI_STATUS_IGNORE)
148 status->count=count*datatype->size();
152 /*Ordered and Shared Versions, with RMA-based locks : Based on the model described in :*/
153 /* @InProceedings{10.1007/11557265_15,*/
154 /* author="Latham, Robert and Ross, Robert and Thakur, Rajeev and Toonen, Brian",*/
155 /* title="Implementing MPI-IO Shared File Pointers Without File System Support",*/
156 /* booktitle="Recent Advances in Parallel Virtual Machine and Message Passing Interface",*/
158 /* publisher="Springer Berlin Heidelberg",*/
159 /* address="Berlin, Heidelberg",*/
162 int File::read_shared(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
163 fh->shared_mutex_->lock();
164 fh->seek(*(fh->shared_file_pointer_),MPI_SEEK_SET);
165 read(fh, buf, count, datatype, status);
166 *(fh->shared_file_pointer_)=fh->file_->tell();
167 fh->shared_mutex_->unlock();
171 int File::read_ordered(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
172 //0 needs to get the shared pointer value
174 if(fh->comm_->rank()==0){
175 val=*(fh->shared_file_pointer_);
177 val=count*datatype->size();
181 simgrid::smpi::colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);
182 fh->seek(result, MPI_SEEK_SET);
183 int ret = fh->op_all<simgrid::smpi::File::read>(buf, count, datatype, status);
184 if(fh->comm_->rank()==fh->comm_->size()-1){
185 fh->shared_mutex_->lock();
186 *(fh->shared_file_pointer_)=fh->file_->tell();
187 fh->shared_mutex_->unlock();
190 simgrid::smpi::colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size() - 1, fh->comm_);
194 int File::write(MPI_File fh, void* /*buf*/, int count, MPI_Datatype datatype, MPI_Status* status)
196 //get position first as we may be doing non contiguous reads and it will probably be updated badly
197 MPI_Offset position = fh->file_->tell();
198 MPI_Offset movesize = datatype->get_extent()*count;
199 MPI_Offset writesize = datatype->size()*count;
200 XBT_DEBUG("Position before write in MPI_File %s : %llu",fh->file_->get_path(),fh->file_->tell());
201 MPI_Offset write = fh->file_->write(writesize, 1);
202 XBT_VERB("Write in MPI_File %s, %lld bytes written, readsize %lld bytes, movesize %lld", fh->file_->get_path(), write, writesize, movesize);
203 if(writesize!=movesize){
204 fh->file_->seek(position+movesize, SEEK_SET);
206 XBT_VERB("Position after write in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());
207 if(status != MPI_STATUS_IGNORE)
208 status->count=count*datatype->size();
212 int File::write_shared(MPI_File fh, const void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
213 fh->shared_mutex_->lock();
214 fh->seek(*(fh->shared_file_pointer_),MPI_SEEK_SET);
215 write(fh, const_cast<void*>(buf), count, datatype, status);
216 *(fh->shared_file_pointer_)=fh->file_->tell();
217 fh->shared_mutex_->unlock();
221 int File::write_ordered(MPI_File fh, const void *buf, int count, MPI_Datatype datatype, MPI_Status *status){
222 //0 needs to get the shared pointer value
224 if(fh->comm_->rank()==0){
225 val=*(fh->shared_file_pointer_);
227 val=count*datatype->size();
230 simgrid::smpi::colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);
231 fh->seek(result, MPI_SEEK_SET);
232 int ret = fh->op_all<simgrid::smpi::File::write>(const_cast<void*>(buf), count, datatype, status);
233 if(fh->comm_->rank()==fh->comm_->size()-1){
234 fh->shared_mutex_->lock();
235 *(fh->shared_file_pointer_)=fh->file_->tell();
236 fh->shared_mutex_->unlock();
239 simgrid::smpi::colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size() - 1, fh->comm_);
243 int File::set_view(MPI_Offset disp, MPI_Datatype etype, MPI_Datatype filetype, const char *datarep, MPI_Info info){
246 datarep_=std::string(datarep);
247 seek_shared(0,MPI_SEEK_SET);
251 int File::get_view(MPI_Offset *disp, MPI_Datatype *etype, MPI_Datatype *filetype, char *datarep){
254 snprintf(datarep, MPI_MAX_NAME_STRING+1, "%s", datarep_.c_str());
259 return file_->size();
268 return simgrid::smpi::colls::barrier(comm_);
271 MPI_Info File::info(){
272 if(info_== MPI_INFO_NULL)
278 void File::set_info(MPI_Info info){
279 if(info_!= MPI_INFO_NULL)
284 MPI_Comm File::comm(){
288 MPI_Errhandler File::errhandler(){
292 void File::set_errhandler(MPI_Errhandler errhandler){
293 errhandler_=errhandler;
294 if(errhandler_!= MPI_ERRHANDLER_NULL)