\r
#include "smpi_comm.hpp"\r
#include "smpi_coll.hpp"\r
+#include "smpi_datatype.hpp"\r
#include "smpi_info.hpp"\r
+#include "smpi_win.hpp"\r
+#include "smpi_request.hpp"\r
+\r
+//setup here, because we have templates in smpi_file we want to log\r
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_io, smpi, "Logging specific to SMPI (RMA operations)");\r
+\r
#include "smpi_file.hpp"\r
+#include "smpi_status.hpp"\r
#include "simgrid/plugins/file_system.h"\r
\r
-\r
-XBT_LOG_NEW_DEFAULT_SUBCATEGORY(smpi_io, smpi, "Logging specific to SMPI (RMA operations)");\r
+#define FP_SIZE sizeof(MPI_Offset)\r
\r
\r
namespace simgrid{\r
namespace smpi{\r
- File::File(MPI_Comm comm, char *filename, int amode, MPI_Info info): comm_(comm), flags_(amode), info_(info){\r
+\r
+ File::File(MPI_Comm comm, char *filename, int amode, MPI_Info info): comm_(comm), flags_(amode), info_(info) {\r
file_= new simgrid::s4u::File(filename, nullptr);\r
+ list_=nullptr;\r
+ if (comm_->rank() == 0) {\r
+ int size= comm_->size() + FP_SIZE;\r
+ list_ = new char[size];\r
+ memset(list_, 0, size);\r
+ shared_file_pointer_ = new MPI_Offset[1];\r
+ shared_mutex_ = s4u::Mutex::create();\r
+ *shared_file_pointer_ = 0;\r
+ win_=new Win(list_, size, 1, MPI_INFO_NULL, comm_);\r
+ }else{\r
+ win_=new Win(list_, 0, 1, MPI_INFO_NULL, comm_);\r
+ }\r
+ simgrid::smpi::Colls::bcast(&shared_file_pointer_, 1, MPI_AINT, 0, comm);\r
+ simgrid::smpi::Colls::bcast(&shared_mutex_, 1, MPI_AINT, 0, comm);\r
+ if(comm_->rank() != 0)\r
+ intrusive_ptr_add_ref(&*shared_mutex_);\r
}\r
- \r
+\r
File::~File(){\r
delete file_;\r
}\r
- \r
+\r
int File::close(MPI_File *fh){\r
+ XBT_DEBUG("Closing MPI_File %s", (*fh)->file_->get_path());\r
(*fh)->sync();\r
if((*fh)->flags() & MPI_MODE_DELETE_ON_CLOSE)\r
(*fh)->file_->unlink();\r
- delete fh;\r
+ delete (*fh);\r
return MPI_SUCCESS;\r
}\r
- \r
+\r
int File::del(char *filename, MPI_Info info){\r
+ //get the file with MPI_MODE_DELETE_ON_CLOSE and then close it\r
File* f = new File(MPI_COMM_SELF,filename,MPI_MODE_DELETE_ON_CLOSE|MPI_MODE_RDWR, nullptr);\r
close(&f);\r
return MPI_SUCCESS;\r
}\r
- \r
+\r
+ int File::get_position(MPI_Offset* offset){\r
+ *offset=file_->tell();\r
+ return MPI_SUCCESS;\r
+ }\r
+\r
+ int File::get_position_shared(MPI_Offset* offset){\r
+ shared_mutex_->lock();\r
+ *offset=*shared_file_pointer_;\r
+ shared_mutex_->unlock();\r
+ return MPI_SUCCESS;\r
+ }\r
+\r
+ int File::seek(MPI_Offset offset, int whence){\r
+ switch(whence){\r
+ case(MPI_SEEK_SET):\r
+ XBT_VERB("Seeking in MPI_File %s, setting offset %lld", file_->get_path(), offset);\r
+ file_->seek(offset,SEEK_SET);\r
+ break;\r
+ case(MPI_SEEK_CUR):\r
+ XBT_VERB("Seeking in MPI_File %s, current offset + %lld", file_->get_path(), offset);\r
+ file_->seek(offset,SEEK_CUR);\r
+ break;\r
+ case(MPI_SEEK_END):\r
+ XBT_VERB("Seeking in MPI_File %s, end offset + %lld", file_->get_path(), offset);\r
+ file_->seek(offset,SEEK_END);\r
+ break;\r
+ default:\r
+ return MPI_ERR_FILE;\r
+ }\r
+ return MPI_SUCCESS;\r
+ }\r
+\r
+ int File::seek_shared(MPI_Offset offset, int whence){\r
+ shared_mutex_->lock();\r
+ seek(offset,whence);\r
+ *shared_file_pointer_=offset;\r
+ shared_mutex_->unlock();\r
+ return MPI_SUCCESS;\r
+ }\r
+\r
+ int File::read(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){\r
+ //get position first as we may be doing non contiguous reads and it will probably be updated badly\r
+ MPI_Offset position = fh->file_->tell();\r
+ MPI_Offset movesize = datatype->get_extent()*count;\r
+ MPI_Offset readsize = datatype->size()*count;\r
+ XBT_DEBUG("Position before read in MPI_File %s : %llu",fh->file_->get_path(),fh->file_->tell());\r
+ MPI_Offset read = fh->file_->read(readsize);\r
+ XBT_VERB("Read in MPI_File %s, %lld bytes read, readsize %lld bytes, movesize %lld", fh->file_->get_path(), read, readsize, movesize);\r
+ if(readsize!=movesize){\r
+ fh->file_->seek(position+movesize, SEEK_SET);\r
+ }\r
+ XBT_VERB("Position after read in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());\r
+ status->count=count*datatype->size();\r
+ return MPI_SUCCESS;\r
+ }\r
+\r
+ /*Ordered and Shared Versions, with RMA-based locks : Based on the model described in :*/\r
+ /* @InProceedings{10.1007/11557265_15,*/\r
+ /* author="Latham, Robert and Ross, Robert and Thakur, Rajeev and Toonen, Brian",*/ \r
+ /* title="Implementing MPI-IO Shared File Pointers Without File System Support",*/\r
+ /* booktitle="Recent Advances in Parallel Virtual Machine and Message Passing Interface",*/\r
+ /* year="2005",*/\r
+ /* publisher="Springer Berlin Heidelberg",*/\r
+ /* address="Berlin, Heidelberg",*/\r
+ /* pages="84--93"*/\r
+ /* }*/\r
+ int File::read_shared(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){\r
+ fh->shared_mutex_->lock();\r
+ fh->seek(*(fh->shared_file_pointer_),MPI_SEEK_SET);\r
+ read(fh, buf, count, datatype, status);\r
+ *(fh->shared_file_pointer_)=fh->file_->tell();\r
+ fh->shared_mutex_->unlock();\r
+ return MPI_SUCCESS;\r
+ }\r
+\r
+ int File::read_ordered(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){\r
+ //0 needs to get the shared pointer value\r
+ MPI_Offset val;\r
+ if(fh->comm_->rank()==0){\r
+ val=*(fh->shared_file_pointer_);\r
+ }else{\r
+ val=count*datatype->size();\r
+ }\r
+\r
+ MPI_Offset result;\r
+ simgrid::smpi::Colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);\r
+ fh->seek(result, MPI_SEEK_SET);\r
+ int ret = fh->op_all<simgrid::smpi::File::read>(buf, count, datatype, status);\r
+ if(fh->comm_->rank()==fh->comm_->size()-1){\r
+ fh->shared_mutex_->lock();\r
+ *(fh->shared_file_pointer_)=fh->file_->tell();\r
+ fh->shared_mutex_->unlock();\r
+ }\r
+ char c;\r
+ simgrid::smpi::Colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size()-1, fh->comm_);\r
+ return ret;\r
+ }\r
+\r
+ int File::write(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){\r
+ //get position first as we may be doing non contiguous reads and it will probably be updated badly\r
+ MPI_Offset position = fh->file_->tell();\r
+ MPI_Offset movesize = datatype->get_extent()*count;\r
+ MPI_Offset writesize = datatype->size()*count;\r
+ XBT_DEBUG("Position before write in MPI_File %s : %llu",fh->file_->get_path(),fh->file_->tell());\r
+ MPI_Offset write = fh->file_->write(writesize);\r
+ XBT_VERB("Write in MPI_File %s, %lld bytes written, readsize %lld bytes, movesize %lld", fh->file_->get_path(), write, writesize, movesize);\r
+ if(writesize!=movesize){\r
+ fh->file_->seek(position+movesize, SEEK_SET);\r
+ }\r
+ XBT_VERB("Position after write in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());\r
+ status->count=count*datatype->size();\r
+ return MPI_SUCCESS;\r
+ }\r
+\r
+ int File::write_shared(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){\r
+ fh->shared_mutex_->lock();\r
+ fh->seek(*(fh->shared_file_pointer_),MPI_SEEK_SET);\r
+ write(fh, buf, count, datatype, status);\r
+ *(fh->shared_file_pointer_)=fh->file_->tell();\r
+ fh->shared_mutex_->unlock();\r
+ return MPI_SUCCESS;\r
+ }\r
+\r
+ int File::write_ordered(MPI_File fh, void *buf, int count, MPI_Datatype datatype, MPI_Status *status){\r
+ //0 needs to get the shared pointer value\r
+ MPI_Offset val;\r
+ if(fh->comm_->rank()==0){\r
+ val=*(fh->shared_file_pointer_);\r
+ }else{\r
+ val=count*datatype->size();\r
+ }\r
+ MPI_Offset result;\r
+ simgrid::smpi::Colls::scan(&val, &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);\r
+ fh->seek(result, MPI_SEEK_SET);\r
+ int ret = fh->op_all<simgrid::smpi::File::write>(buf, count, datatype, status);\r
+ if(fh->comm_->rank()==fh->comm_->size()-1){\r
+ fh->shared_mutex_->lock();\r
+ *(fh->shared_file_pointer_)=fh->file_->tell();\r
+ fh->shared_mutex_->unlock();\r
+ }\r
+ char c;\r
+ simgrid::smpi::Colls::bcast(&c, 1, MPI_BYTE, fh->comm_->size()-1, fh->comm_);\r
+ return ret;\r
+ }\r
+\r
int File::size(){\r
return file_->size();\r
}\r
- \r
+\r
int File::flags(){\r
return flags_;\r
}\r
+\r
int File::sync(){\r
//no idea\r
return simgrid::smpi::Colls::barrier(comm_);\r
}\r
+\r
+MPI_Info File::info(){\r
+ if(info_== MPI_INFO_NULL)\r
+ info_ = new Info();\r
+ info_->ref();\r
+ return info_;\r
+}\r
+\r
+void File::set_info(MPI_Info info){\r
+ if(info_!= MPI_INFO_NULL)\r
+ info->ref();\r
+ info_=info;\r
+}\r
+\r
+}\r
}\r
-}
\ No newline at end of file
