1 /* Copyright (c) 2007-2023. 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 = _smpi_cfg_default_errhandler_is_error ? MPI_ERRORS_ARE_FATAL : MPI_ERRORS_RETURN;;
26 namespace simgrid::smpi {
28 File::File(MPI_Comm comm, const char* filename, int amode, MPI_Info info) : comm_(comm), flags_(amode), info_(info)
30 if (info_ != MPI_INFO_NULL)
32 std::string fullname = filename;
33 xbt_assert(not simgrid::s4u::Host::current()->get_disks().empty(),
34 "SMPI/IO : Trying to open file on a diskless host ! Add one to your platform file");
36 // in case no fullpath is provided ... just pick the first mountpoint.
37 if (size_t found = fullname.find('/'); found == std::string::npos || fullname.rfind("./", 1) != std::string::npos) {
38 const auto* disk = simgrid::s4u::Host::current()->get_disks().front();
40 if (disk->get_host() != simgrid::s4u::Host::current())
41 mount = disk->extension<simgrid::s4u::FileSystemDiskExt>()->get_mount_point(disk->get_host());
43 mount = disk->extension<simgrid::s4u::FileSystemDiskExt>()->get_mount_point();
44 XBT_DEBUG("No absolute path given for file opening, use '%s'", mount.c_str());
45 if (fullname.rfind("./", 1) != std::string::npos)
46 fullname.replace(fullname.begin(), fullname.begin() + 1, mount);
49 fullname.insert(0, mount);
52 XBT_DEBUG("Opening %s", fullname.c_str());
53 file_ = simgrid::s4u::File::open(fullname, nullptr);
58 if (comm_->rank() == 0) {
59 int size = comm_->size() + FP_SIZE;
60 list_ = new char[size];
61 errhandler_ = SMPI_default_File_Errhandler;
63 memset(list_, 0, size);
64 shared_file_pointer_ = new MPI_Offset();
65 shared_mutex_ = s4u::Mutex::create();
66 *shared_file_pointer_ = 0;
67 win_ = new Win(list_, size, 1, MPI_INFO_NULL, comm_);
69 errhandler_ = MPI_ERRHANDLER_NULL;
70 win_ = new Win(list_, 0, 1, MPI_INFO_NULL, comm_);
72 simgrid::smpi::colls::bcast(&shared_file_pointer_, 1, MPI_AINT, 0, comm);
73 simgrid::smpi::colls::bcast(&shared_mutex_, 1, MPI_AINT, 0, comm);
74 if (comm_->rank() != 0)
75 intrusive_ptr_add_ref(&*shared_mutex_);
81 if (comm_->rank() == 0) {
82 delete shared_file_pointer_;
85 simgrid::smpi::Win::del(win_);
87 F2C::free_f(this->f2c_id());
88 if (info_ != MPI_INFO_NULL)
89 simgrid::smpi::Info::unref(info_);
90 if (errhandler_ != MPI_ERRHANDLER_NULL)
91 simgrid::smpi::Errhandler::unref(errhandler_);
94 int File::close(MPI_File* fh)
96 XBT_DEBUG("Closing MPI_File %s", (*fh)->file_->get_path());
98 if ((*fh)->flags() & MPI_MODE_DELETE_ON_CLOSE)
99 (*fh)->file_->unlink();
104 int File::del(const char* filename, const Info*)
106 // get the file with MPI_MODE_DELETE_ON_CLOSE and then close it
107 auto* f = new File(MPI_COMM_SELF, filename, MPI_MODE_DELETE_ON_CLOSE | MPI_MODE_RDWR, nullptr);
112 int File::get_position(MPI_Offset* offset) const
114 *offset = file_->tell()/etype_->get_extent();
118 int File::get_position_shared(MPI_Offset* offset) const
120 shared_mutex_->lock();
121 *offset = *shared_file_pointer_/etype_->get_extent();
122 shared_mutex_->unlock();
126 int File::seek(MPI_Offset offset, int whence)
130 XBT_VERB("Seeking in MPI_File %s, setting offset %lld", file_->get_path(), offset);
131 file_->seek(offset, SEEK_SET);
134 XBT_VERB("Seeking in MPI_File %s, current offset + %lld", file_->get_path(), offset);
135 file_->seek(offset, SEEK_CUR);
138 XBT_VERB("Seeking in MPI_File %s, end offset + %lld", file_->get_path(), offset);
139 file_->seek(offset, SEEK_END);
147 int File::seek_shared(MPI_Offset offset, int whence)
149 shared_mutex_->lock();
150 seek(offset, whence);
151 *shared_file_pointer_ = file_->tell();
152 shared_mutex_->unlock();
156 int File::read(MPI_File fh, void* /*buf*/, int count, const Datatype* datatype, MPI_Status* status)
158 // get position first as we may be doing non contiguous reads and it will probably be updated badly
159 MPI_Offset position = fh->file_->tell();
160 MPI_Offset movesize = datatype->get_extent() * count;
161 MPI_Offset readsize = datatype->size() * count;
162 XBT_DEBUG("Position before read in MPI_File %s : %llu, size %llu", fh->file_->get_path(), fh->file_->tell(), fh->file_->size());
163 MPI_Offset read = fh->file_->read(readsize);
164 XBT_VERB("Read in MPI_File %s, %lld bytes read, count %d, readsize %lld bytes, movesize %lld", fh->file_->get_path(), read, count,
166 if (readsize != movesize) {
167 fh->file_->seek(position + movesize, SEEK_SET);
169 XBT_VERB("Position after read in MPI_File %s : %llu", fh->file_->get_path(), fh->file_->tell());
170 if (status != MPI_STATUS_IGNORE)
171 status->count = count * datatype->size();
175 /*Ordered and Shared Versions, with RMA-based locks : Based on the model described in :*/
176 /* @InProceedings{10.1007/11557265_15,*/
177 /* author="Latham, Robert and Ross, Robert and Thakur, Rajeev and Toonen, Brian",*/
178 /* title="Implementing MPI-IO Shared File Pointers Without File System Support",*/
179 /* booktitle="Recent Advances in Parallel Virtual Machine and Message Passing Interface",*/
181 /* publisher="Springer Berlin Heidelberg",*/
182 /* address="Berlin, Heidelberg",*/
185 int File::read_shared(MPI_File fh, void* buf, int count, const Datatype* datatype, MPI_Status* status)
187 fh->shared_mutex_->lock();
188 fh->seek(*(fh->shared_file_pointer_), MPI_SEEK_SET);
189 read(fh, buf, count, datatype, status);
190 *(fh->shared_file_pointer_) = fh->file_->tell();
191 fh->shared_mutex_->unlock();
192 fh->seek(*(fh->shared_file_pointer_), MPI_SEEK_SET);
196 int File::read_ordered(MPI_File fh, void* buf, int count, const Datatype* datatype, MPI_Status* status)
198 // 0 needs to get the shared pointer value
200 if (fh->comm_->rank() == 0) {
201 val = *(fh->shared_file_pointer_);
203 val = count * datatype->size();
207 simgrid::smpi::colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);
209 fh->get_position(&prev);
210 fh->seek(result, MPI_SEEK_SET);
211 int ret = fh->op_all<simgrid::smpi::File::read>(buf, count, datatype, status);
212 if (fh->comm_->rank() == fh->comm_->size() - 1) {
213 fh->shared_mutex_->lock();
214 *(fh->shared_file_pointer_)=fh->file_->tell();
215 fh->shared_mutex_->unlock();
218 simgrid::smpi::colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size() - 1, fh->comm_);
219 fh->seek(prev, MPI_SEEK_SET);
223 int File::write(MPI_File fh, void* /*buf*/, int count, const Datatype* datatype, MPI_Status* status)
225 // get position first as we may be doing non contiguous reads and it will probably be updated badly
226 MPI_Offset position = fh->file_->tell();
227 MPI_Offset movesize = datatype->get_extent() * count;
228 MPI_Offset writesize = datatype->size() * count;
229 XBT_DEBUG("Position before write in MPI_File %s : %llu, size %llu", fh->file_->get_path(), fh->file_->tell(), fh->file_->size());
230 MPI_Offset write = fh->file_->write(writesize, true);
231 XBT_VERB("Write in MPI_File %s, %lld bytes written, count %d, writesize %lld bytes, movesize %lld", fh->file_->get_path(), write,
232 count, writesize, movesize);
233 if (writesize != movesize) {
234 fh->file_->seek(position + movesize, SEEK_SET);
236 XBT_VERB("Position after write in MPI_File %s : %llu", fh->file_->get_path(), fh->file_->tell());
237 if (status != MPI_STATUS_IGNORE)
238 status->count = count * datatype->size();
242 int File::write_shared(MPI_File fh, const void* buf, int count, const Datatype* datatype, MPI_Status* status)
244 fh->shared_mutex_->lock();
245 XBT_DEBUG("Write shared on %s - Shared ptr before : %lld", fh->file_->get_path(), *(fh->shared_file_pointer_));
246 fh->seek(*(fh->shared_file_pointer_), MPI_SEEK_SET);
247 write(fh, const_cast<void*>(buf), count, datatype, status);
248 *(fh->shared_file_pointer_) = fh->file_->tell();
249 XBT_DEBUG("Write shared on %s - Shared ptr after : %lld", fh->file_->get_path(), *(fh->shared_file_pointer_));
250 fh->seek(*(fh->shared_file_pointer_), MPI_SEEK_SET);
251 fh->shared_mutex_->unlock();
255 int File::write_ordered(MPI_File fh, const void* buf, int count, const Datatype* datatype, MPI_Status* status)
257 // 0 needs to get the shared pointer value
259 if (fh->comm_->rank() == 0) {
260 val = *(fh->shared_file_pointer_);
262 val = count * datatype->size();
265 simgrid::smpi::colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);
267 fh->get_position(&prev);
268 fh->seek(result, MPI_SEEK_SET);
269 int ret = fh->op_all<simgrid::smpi::File::write>(const_cast<void*>(buf), count, datatype, status);
270 if (fh->comm_->rank() == fh->comm_->size() - 1) {
271 fh->shared_mutex_->lock();
272 *(fh->shared_file_pointer_)=fh->file_->tell();
273 fh->shared_mutex_->unlock();
276 simgrid::smpi::colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size() - 1, fh->comm_);
277 fh->seek(prev, MPI_SEEK_SET);
281 int File::set_view(MPI_Offset disp, MPI_Datatype etype, MPI_Datatype filetype, const char* datarep, const Info*)
284 filetype_ = filetype;
287 if (comm_->rank() == 0){
288 if(disp != MPI_DISPLACEMENT_CURRENT)
289 seek_shared(disp, MPI_SEEK_SET);
291 seek_shared(0, MPI_SEEK_CUR);
297 int File::get_view(MPI_Offset* disp, MPI_Datatype* etype, MPI_Datatype* filetype, char* datarep) const
301 *filetype = filetype_;
302 snprintf(datarep, MPI_MAX_NAME_STRING + 1, "%s", datarep_.c_str());
306 int File::size() const
308 return file_->size();
311 void File::set_size(int size)
313 file_->write(size, true);
316 int File::flags() const
321 MPI_Datatype File::etype() const
329 return simgrid::smpi::colls::barrier(comm_);
332 MPI_Info File::info()
337 void File::set_info(MPI_Info info)
339 if (info_ != MPI_INFO_NULL)
340 simgrid::smpi::Info::unref(info_);
342 if (info_ != MPI_INFO_NULL)
346 MPI_Comm File::comm() const
351 MPI_Errhandler File::errhandler()
353 if (errhandler_ != MPI_ERRHANDLER_NULL)
358 void File::set_errhandler(MPI_Errhandler errhandler)
360 if (errhandler_ != MPI_ERRHANDLER_NULL)
361 simgrid::smpi::Errhandler::unref(errhandler_);
362 errhandler_ = errhandler;
363 if (errhandler_ != MPI_ERRHANDLER_NULL)
367 File* File::f2c(int id)
369 return static_cast<File*>(F2C::f2c(id));
372 void File::set_atomicity(bool a){
376 bool File::get_atomicity() const
381 } // namespace simgrid::smpi