portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-09 21:50:45 +01:00
Code Releases Activity

Hypercube optimisation

Browse Source
This commit is contained in:
Peter Boyle 2018-05-02 14:10:21 +01:00
parent 172f412102
commit 12982a4455
1 changed files with 5 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
lib/communicator/SharedMemoryMPI.cc
View File

@ -173,15 +173,9 @@ void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,
HyperCubeCoords[d] = (hypercoor>>d)&0x1; HyperCubeCoords[d] = (hypercoor>>d)&0x1;
} }
std::cerr << " Hcoor (" ;
for(int d=0;d<maxhdim;d++){
std::cerr << " "<< HyperCubeCoords[d] ;
}
std::cerr << " )"<<std::endl;
std::string hname(name); std::string hname(name);
std::cerr << "hostname "<<hname<<std::endl; std::cout << "hostname "<<hname<<std::endl;
std::cerr << "R " << R << " I " << I << " N "<< N<<" nhi "<<nhi<<" nlo "<<nlo std::cout << "R " << R << " I " << I << " N "<< N<<
<< " hypercoor 0x"<<std::hex<<hypercoor<<std::dec<<std::endl; << " hypercoor 0x"<<std::hex<<hypercoor<<std::dec<<std::endl;
////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////
@ -191,7 +185,7 @@ void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,
hypercoor=hypercoor-rootcoor; hypercoor=hypercoor-rootcoor;
assert(hypercoor<WorldSize); assert(hypercoor<WorldSize);
assert(hypercoor>=0); assert(hypercoor>=0);
std::cerr << " WorldRank "<<WorldRank << " relative hypercoor " << std::hex << hypercoor <<std::dec<<std::endl;
////////////////////////////////////// //////////////////////////////////////
// Printing // Printing
////////////////////////////////////// //////////////////////////////////////
@ -199,12 +193,6 @@ void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,
HyperCubeCoords[d] = (hypercoor>>d)&0x1; HyperCubeCoords[d] = (hypercoor>>d)&0x1;
} }
std::cerr << " rel Hcoor (";
for(int d=0;d<maxhdim;d++){
std::cerr << " "<< HyperCubeCoords[d] ;
}
std::cerr << " )"<<std::endl;
//////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////
// Identify subblock of ranks on node spreading across dims // Identify subblock of ranks on node spreading across dims
// in a maximally symmetrical way // in a maximally symmetrical way
@ -254,12 +242,6 @@ void GlobalSharedMemory::OptimalCommunicator(const std::vector<int> &processors,
int rank; int rank;
Lexicographic::CoorFromIndexReversed(NodeCoor,WorldNode ,NodeDims); Lexicographic::CoorFromIndexReversed(NodeCoor,WorldNode ,NodeDims);
std::cerr << "NodeCoor ";
for(int d=0;d<ndimension;d++) std::cerr << NodeCoor[d]<<" ";
std::cerr << std::endl;
std::cerr << "HyperCoor ";
for(int d=0;d<ndimension;d++) std::cerr << HyperCoor[d]<<" ";
std::cerr << std::endl;
for(int d=0;d<ndimension;d++) NodeCoor[d]=HyperCoor[d]; for(int d=0;d<ndimension;d++) NodeCoor[d]=HyperCoor[d];
@ -395,7 +377,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
#ifdef GRID_MPI3_SHMMMAP #ifdef GRID_MPI3_SHMMMAP
void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags) void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
{ {
std::cout << "SharedMemoryAllocate "<< bytes<< " MMAP implementation "<<std::endl; std::cout << "SharedMemoryAllocate "<< bytes<< " MMAP implementation "<< GRID_SHM_PATH <<std::endl;
assert(_ShmSetup==1); assert(_ShmSetup==1);
assert(_ShmAlloc==0); assert(_ShmAlloc==0);
////////////////////////////////////////////////////////////////////////////////////////////////////////// //////////////////////////////////////////////////////////////////////////////////////////////////////////
@ -405,7 +387,7 @@ void GlobalSharedMemory::SharedMemoryAllocate(uint64_t bytes, int flags)
WorldShmCommBufs.resize(WorldShmSize); WorldShmCommBufs.resize(WorldShmSize);
//////////////////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////////////////
// Hugetlbf and others map filesystems as mappable huge pages // Hugetlbfs and others map filesystems as mappable huge pages
//////////////////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////////////////
char shm_name [NAME_MAX]; char shm_name [NAME_MAX];
for(int r=0;r<WorldShmSize;r++){ for(int r=0;r<WorldShmSize;r++){