#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
-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
\r
- File::File(MPI_Comm comm, char *filename, int amode, MPI_Info info): comm_(comm), flags_(amode), info_(info), shared_file_pointer_(0) {\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
File::~File(){\r
}\r
\r
int File::get_position_shared(MPI_Offset* offset){\r
- lock();\r
- *offset=shared_file_pointer_;\r
- unlock();\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_DEBUG("Seeking in MPI_File %s, setting offset %lld", file_->get_path(), offset);\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_DEBUG("Seeking in MPI_File %s, current offset + %lld", file_->get_path(), offset);\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_DEBUG("Seeking in MPI_File %s, end offset + %lld", file_->get_path(), offset);\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
}\r
\r
int File::seek_shared(MPI_Offset offset, int whence){\r
- lock();\r
+ shared_mutex_->lock();\r
seek(offset,whence);\r
- shared_file_pointer_=file_->tell();\r
- unlock();\r
+ *shared_file_pointer_=offset;\r
+ shared_mutex_->unlock();\r
return MPI_SUCCESS;\r
}\r
\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_DEBUG("Read in MPI_File %s, %lld bytes read, readsize %lld bytes, movesize %lld", fh->file_->get_path(), read, readsize, movesize);\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_DEBUG("Position after read in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());\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->lock();\r
- fh->seek(fh->shared_file_pointer_,MPI_SEEK_SET);\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->unlock();\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
- fh->lock();\r
- fh->unlock();\r
+ val=*(fh->shared_file_pointer_);\r
}else{\r
- fh->shared_file_pointer_=count*datatype->size();\r
+ val=count*datatype->size();\r
}\r
+\r
MPI_Offset result;\r
- simgrid::smpi::Colls::scan(&(fh->shared_file_pointer_), &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);\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->lock();\r
- fh->unlock();\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
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_DEBUG("Write in MPI_File %s, %lld bytes read, readsize %lld bytes, movesize %lld", fh->file_->get_path(), write, writesize, movesize);\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_DEBUG("Position after write in MPI_File %s : %llu",fh->file_->get_path(), fh->file_->tell());\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->lock();\r
- fh->seek(fh->shared_file_pointer_,MPI_SEEK_SET);\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->unlock();\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
- fh->lock();\r
- fh->unlock();\r
+ val=*(fh->shared_file_pointer_);\r
}else{\r
- fh->shared_file_pointer_=count*datatype->size();\r
+ val=count*datatype->size();\r
}\r
MPI_Offset result;\r
- simgrid::smpi::Colls::scan(&(fh->shared_file_pointer_), &result, 1, MPI_OFFSET, MPI_SUM, fh->comm_);\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->lock();\r
- fh->unlock();\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 simgrid::smpi::Colls::barrier(comm_);\r
}\r
\r
- int File::lock()\r
-{\r
- int rank = comm_->rank();\r
- int size = comm_->size();\r
- char waitlist[size];\r
- char lock = 1;\r
- int tag=444;\r
- int i;\r
- win_->lock(MPI_LOCK_EXCLUSIVE, 0, 0);\r
- win_->put(&lock, 1, MPI_CHAR, 0, FP_SIZE+rank, 1, MPI_CHAR);\r
- win_->get(waitlist, size, MPI_CHAR, 0, FP_SIZE, size, MPI_CHAR);\r
- win_->get(&shared_file_pointer_ , 1 , MPI_OFFSET , 0 , 0, 1, MPI_OFFSET);\r
- win_->unlock(0);\r
- for (i = 0; i < size; i++) {\r
- if (waitlist[i] == 1 && i != rank) {\r
- // wait for the lock\r
- MPI_Recv(&lock, 1, MPI_CHAR, MPI_ANY_SOURCE, tag, comm_, MPI_STATUS_IGNORE);\r
- break;\r
- }\r
- }\r
- return 0;\r
-}\r
-\r
-int File::unlock()\r
-{\r
- int rank = comm_->rank();\r
- int size = comm_->size();\r
- char waitlist[size];\r
- char lock = 0;\r
- int tag=444;\r
- int i, next;\r
- win_->lock(MPI_LOCK_EXCLUSIVE, 0, 0);\r
- win_->put(&lock, 1, MPI_CHAR, 0, FP_SIZE+rank, 1, MPI_CHAR);\r
- win_->get(waitlist, size, MPI_CHAR, 0, FP_SIZE, size, MPI_CHAR);\r
- shared_file_pointer_=file_->tell();\r
- win_->put(&shared_file_pointer_, 1 , MPI_OFFSET , 0 , 0, 1, MPI_OFFSET);\r
-\r
- win_->unlock(0);\r
- next = (rank + 1 + size) % size;\r
- for (i = 0; i < size; i++, next = (next + 1) % size) {\r
- if (waitlist[next] == 1) {\r
- MPI_Send(&lock, 1, MPI_CHAR, next, tag, comm_);\r
- break;\r
- }\r
- }\r
- return 0;\r
-}\r
-\r
MPI_Info File::info(){\r
if(info_== MPI_INFO_NULL)\r
info_ = new Info();\r