1
0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-14 01:35:36 +00:00
Grid/Grid_cshift_mpi.h

338 lines
11 KiB
C
Raw Normal View History

#ifndef _GRID_MPI_CSHIFT_H_
#define _GRID_MPI_CSHIFT_H_
#define MAX(x,y) ((x)>(y)?(x):(y))
#define MIN(x,y) ((x)>(y)?(y):(x))
//////////////////////////////////////////////////////////////////////////////////////////
// Must not lose sight that goal is to be able to construct really efficient
// gather to a point stencil code. CSHIFT is not the best way, so probably need
// additional stencil support.
//
// Could still do a templated syntax tree and make CSHIFT return lattice vector.
//
// Stencil based code could pre-exchange haloes and use a table lookup for neighbours
//
// Lattice <foo> could also allocate haloes which get used for stencil code.
//
// Grid could create a neighbour index table for a given stencil.
// Could also implement CovariantCshift.
//////////////////////////////////////////////////////////////////////////////////////////
/////////////////////////////////////////////////////////////
// Q. Further split this into separate sub functions?
/////////////////////////////////////////////////////////////
// CshiftCB_local
// CshiftCB_local_permute
// Cshift_comms_splice
// Cshift_comms
// Cshift_local
// Cshift_local_permute
friend Lattice<vobj> Cshift(Lattice<vobj> &rhs,int dimension,int shift)
{
typedef typename vobj::vector_type vector_type;
typedef typename vobj::scalar_type scalar_type;
Lattice<vobj> ret(rhs._grid);
int fd = rhs._grid->_fdimensions[dimension];
int rd = rhs._grid->_rdimensions[dimension];
// Map to always positive shift modulo global full dimension.
shift = (shift+fd)%fd;
ret.checkerboard = rhs._grid->CheckerBoardDestination(rhs.checkerboard,shift);
// the permute type
int simd_layout = rhs._grid->_simd_layout[dimension];
int comm_dim = rhs._grid->_processors[dimension] >1 ;
int splice_dim = rhs._grid->_simd_layout[dimension]>1 && (comm_dim);
if ( !comm_dim ) {
Cshift_local(ret,rhs,dimension,shift); // Handles checkerboarding
} else if ( splice_dim ) {
Cshift_comms_simd(ret,rhs,dimension,shift);
} else {
Cshift_comms(ret,rhs,dimension,shift);
}
return ret;
}
friend void Cshift_comms(Lattice<vobj>& ret,Lattice<vobj> &rhs,int dimension,int shift)
{
int sshift[2];
sshift[0] = rhs._grid->CheckerBoardShift(rhs.checkerboard,dimension,shift,0);
sshift[1] = rhs._grid->CheckerBoardShift(rhs.checkerboard,dimension,shift,1);
if ( sshift[0] == sshift[1] ) {
// printf("Cshift_comms : single pass\n");
Cshift_comms(ret,rhs,dimension,shift,0x3);
} else {
// printf("Cshift_comms : two pass\n");
// printf("call1\n");
Cshift_comms(ret,rhs,dimension,shift,0x1);// if checkerboard is unfavourable take two passes
// printf("call2\n");
Cshift_comms(ret,rhs,dimension,shift,0x2);// both with block stride loop iteration
// printf("done\n");
}
}
friend void Cshift_comms_simd(Lattice<vobj>& ret,Lattice<vobj> &rhs,int dimension,int shift)
{
int sshift[2];
sshift[0] = rhs._grid->CheckerBoardShift(rhs.checkerboard,dimension,shift,0);
sshift[1] = rhs._grid->CheckerBoardShift(rhs.checkerboard,dimension,shift,1);
if ( sshift[0] == sshift[1] ) {
Cshift_comms_simd(ret,rhs,dimension,shift,0x3);
} else {
// printf("call1 0x1 cb=even\n");
Cshift_comms_simd(ret,rhs,dimension,shift,0x1);// if checkerboard is unfavourable take two passes
// printf("call2 0x2 cb=odd\n");
Cshift_comms_simd(ret,rhs,dimension,shift,0x2);// both with block stride loop iteration
// printf("done\n");
}
}
friend void Cshift_comms(Lattice<vobj> &ret,Lattice<vobj> &rhs,int dimension,int shift,int cbmask)
{
typedef typename vobj::vector_type vector_type;
typedef typename vobj::scalar_type scalar_type;
SimdGrid *grid=rhs._grid;
Lattice<vobj> temp(rhs._grid);
int fd = rhs._grid->_fdimensions[dimension];
int rd = rhs._grid->_rdimensions[dimension];
int simd_layout = rhs._grid->_simd_layout[dimension];
int comm_dim = rhs._grid->_processors[dimension] >1 ;
assert(simd_layout==1);
assert(comm_dim==1);
assert(shift>=0);
assert(shift<fd);
// Packed gather sequence is clean
int buffer_size = rhs._grid->_slice_nblock[dimension]*rhs._grid->_slice_block[dimension];
std::vector<vobj,alignedAllocator<vobj> > send_buf(buffer_size);
std::vector<vobj,alignedAllocator<vobj> > recv_buf(buffer_size);
// This code could be simplified by multiple calls to single routine with extra params to
// encapsulate the difference in the code paths.
int cb= (cbmask==0x2)? 1 : 0;
int sshift= rhs._grid->CheckerBoardShift(rhs.checkerboard,dimension,shift,cb);
for(int x=0;x<rd;x++){
int offnode = ( x+sshift >= rd );
int sx = (x+sshift)%rd;
int comm_proc = (x+sshift)/rd;
if (!offnode) {
// printf("local x %d sshift %d offnode %d rd %d cb %d\n",x,sshift,offnode,rd,cb);
Copy_plane(ret,rhs,dimension,x,sx,cbmask);
} else {
int words = send_buf.size();
if (cbmask != 0x3) words=words>>1;
int bytes = words * sizeof(vobj);
// printf("nonlocal x %d sx %d sshift %d offnode %d rd %d cb %d cbmask %d rhscb %d comm_proc %d\n",
// x,sx,sshift,offnode,rd,cb,cbmask,rhs.checkerboard,comm_proc);
// Copy_plane(temp,rhs,dimension,x,sx,cbmask);
// Bug found; cbmask may differ between sx plan and rx plane.
Gather_plane_simple (rhs,send_buf,dimension,sx,cbmask);
// for(int i=0;i<MIN(words,8);i++){
// float *ptr = (float *)&send_buf[i];
// printf("send buf shift %d cbmask %d i %d %le\n",sshift,cbmask,i,*ptr);
// }
// Gather_plane_simple (rhs,send_buf,dimension,sx,cbmask^0x3);
// for(int i=0;i<MIN(words,8);i++){
// float *ptr = (float *)&send_buf[i];
// printf("send buf shift %d cbmask %d i %d %le\n",sshift,cbmask,i,*ptr);
// }
// recv_buf=send_buf;
int rank = grid->_processor;
int recv_from_rank;
int xmit_to_rank;
grid->ShiftedRanks(dimension,comm_proc,xmit_to_rank,recv_from_rank);
// printf("bytes %d node %d sending to %d receiving from %d\n",bytes,rank,xmit_to_rank,recv_from_rank );
grid->SendToRecvFrom((void *)&send_buf[0],
xmit_to_rank,
(void *)&recv_buf[0],
recv_from_rank,
bytes);
Scatter_plane_simple (ret,recv_buf,dimension,x,cbmask);
}
}
}
friend void Cshift_comms_simd(Lattice<vobj> &ret,Lattice<vobj> &rhs,int dimension,int shift,int cbmask)
{
const int Nsimd = vector_type::Nsimd();
SimdGrid *grid=rhs._grid;
typedef typename vobj::vector_type vector_type;
typedef typename vobj::scalar_type scalar_type;
int fd = grid->_fdimensions[dimension];
int rd = grid->_rdimensions[dimension];
int ld = grid->_ldimensions[dimension];
int simd_layout = grid->_simd_layout[dimension];
int comm_dim = grid->_processors[dimension] >1 ;
assert(comm_dim==1);
assert(simd_layout==2);
assert(shift>=0);
assert(shift<fd);
int permute_type=0;
for(int d=0;d<dimension;d++){
if (grid->_simd_layout[d]>1 ) permute_type++;
}
///////////////////////////////////////////////
// Simd direction uses an extract/merge pair
///////////////////////////////////////////////
int buffer_size = grid->_slice_nblock[dimension]*grid->_slice_block[dimension];
int words = sizeof(vobj)/sizeof(vector_type);
std::vector<std::vector<scalar_type> > send_buf_extract(Nsimd,std::vector<scalar_type>(buffer_size*words) );
std::vector<std::vector<scalar_type> > recv_buf_extract(Nsimd,std::vector<scalar_type>(buffer_size*words) );
int bytes = buffer_size*words*sizeof(scalar_type);
std::vector<scalar_type *> pointers(Nsimd); //
std::vector<scalar_type *> rpointers(Nsimd); // received pointers
///////////////////////////////////////////
// Work out what to send where
///////////////////////////////////////////
int cb = (cbmask==0x2)? 1 : 0;
int sshift= grid->CheckerBoardShift(rhs.checkerboard,dimension,shift,cb);
// printf("cshift-comms-simd: shift = %d ; sshift = %d ; cbmask %d ; simd_layout %d\n",shift,sshift,cbmask,simd_layout);
std::vector<int> comm_offnode(simd_layout);
std::vector<int> comm_proc (simd_layout); //relative processor coord in dim=dimension
// Strategy
//
//* Loop over source planes
//* if any communication needed extract and send
//* if communication needed extract and send
for(int x=0;x<rd;x++){
int comm_any = 0;
for(int s=0;s<simd_layout;s++) {
// does shift to "neighbour" takes us off node?
// coordinates (reduce plane, simd_lane) of neighbour?
// how many nodes away is this shift?
// where we should send to?
// where we should receive from?
int shifted_x = x+s*rd+sshift;
comm_offnode[s] = shifted_x >= ld;
comm_any = comm_any | comm_offnode[s];
comm_proc[s] = shifted_x/ld;
// printf("rd %d x %d shifted %d s=%d comm_any %d\n",rd, x,shifted_x,s,comm_any);
}
int o = 0;
int bo = x*grid->_ostride[dimension];
int sx = (x+sshift)%rd;
// Need Convenience function in _grid. Move this in
if ( comm_any ) {
for(int i=0;i<Nsimd;i++){
pointers[i] = (scalar_type *)&send_buf_extract[i][0];
}
Gather_plane_extract(rhs,pointers,dimension,sx,cbmask);
// for(int i=0;i<Nsimd;i++){
// printf("extracted %d %le\n",i,real(send_buf_extract[i][0]));
// }
for(int i=0;i<Nsimd;i++){
int s = grid->iCoordFromIsite(i,dimension);
if(comm_offnode[s]){
int rank = grid->_processor;
int recv_from_rank;
int xmit_to_rank;
grid->ShiftedRanks(dimension,comm_proc[s],xmit_to_rank,recv_from_rank);
grid->SendToRecvFrom((void *)&send_buf_extract[i][0],
xmit_to_rank,
(void *)&recv_buf_extract[i][0],
recv_from_rank,
bytes);
// printf("Cshift_simd comms %d %le %le\n",i,real(recv_buf_extract[i][0]),real(send_buf_extract[i][0]));
rpointers[i] = (scalar_type *)&recv_buf_extract[i][0];
} else {
rpointers[i] = (scalar_type *)&send_buf_extract[i][0];
// printf("Cshift_simd local %d %le \n",i,real(send_buf_extract[i][0]));
}
}
// Permute by swizzling pointers in merge
int permute_slice=0;
int lshift=sshift%ld;
int wrap =lshift/rd;
int num =lshift%rd;
if ( x< rd-num ) permute_slice=wrap;
else permute_slice = 1-wrap;
for(int i=0;i<vobj::vector_type::Nsimd();i++){
if ( permute_slice ) {
pointers[i] = rpointers[permute_map[permute_type][i]];
} else {
pointers[i] = rpointers[i];
}
// printf("Cshift_simd perm %d num %d wrap %d swiz %d %le unswiz %le\n",permute_slice,num,wrap,i,real(pointers[i][0]),real(rpointers[i][0]));
}
Scatter_plane_merge(ret,pointers,dimension,x,cbmask);
} else {
int permute_slice=0;
int wrap = sshift/rd;
int num = sshift%rd;
if ( x< rd-num ) permute_slice=wrap;
else permute_slice = 1-wrap;
if ( permute_slice ) Copy_plane_permute(ret,rhs,dimension,x,sx,cbmask,permute_type);
else Copy_plane(ret,rhs,dimension,x,sx,cbmask);
}
}
}
#endif