portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-09 23:45:36 +00:00
Code Releases Activity

Better padded cell exchange

Browse Source
This commit is contained in:
Peter Boyle 2023-12-22 18:05:41 -05:00
parent dd13937bb6
commit 78ab955fec
1 changed files with 4 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
Grid/lattice/PaddedCell.h
View File

@ -126,10 +126,8 @@ template<class vobj> inline void ScatterSlice(const cshiftVector<vobj> &buf,
/////////////////////////////////////////// ///////////////////////////////////////////
// Transfer into lattice - will coalesce // Transfer into lattice - will coalesce
/////////////////////////////////////////// ///////////////////////////////////////////
#if 0 // sobj obj = extractLane(blane,buf_p[ss+offset]);
sobj obj = extractLane(blane,buf_p[ss+offset]); // insertLane(lane,lat_v[osite],obj);
insertLane(lane,lat_v[osite],obj);
#else
const int words=sizeof(vobj)/sizeof(vector_type); const int words=sizeof(vobj)/sizeof(vector_type);
vector_type * from = (vector_type *)&buf_p[ss+offset]; vector_type * from = (vector_type *)&buf_p[ss+offset];
vector_type * to = (vector_type *)&lat_v[osite]; vector_type * to = (vector_type *)&lat_v[osite];
@ -138,7 +136,6 @@ template<class vobj> inline void ScatterSlice(const cshiftVector<vobj> &buf,
stmp = getlane(from[w], blane); stmp = getlane(from[w], blane);
putlane(to[w], stmp, lane); putlane(to[w], stmp, lane);
} }
#endif
} }
}); });
} }
@ -215,10 +212,8 @@ template<class vobj> inline void GatherSlice(cshiftVector<vobj> &buf,
/////////////////////////////////////////// ///////////////////////////////////////////
// Take out of lattice // Take out of lattice
/////////////////////////////////////////// ///////////////////////////////////////////
#if 0 // sobj obj = extractLane(lane,lat_v[osite]);
sobj obj = extractLane(lane,lat_v[osite]); // insertLane(blane,buf_p[ss+offset],obj);
insertLane(blane,buf_p[ss+offset],obj);
#else
const int words=sizeof(vobj)/sizeof(vector_type); const int words=sizeof(vobj)/sizeof(vector_type);
vector_type * to = (vector_type *)&buf_p[ss+offset]; vector_type * to = (vector_type *)&buf_p[ss+offset];
vector_type * from = (vector_type *)&lat_v[osite]; vector_type * from = (vector_type *)&lat_v[osite];
@ -227,22 +222,8 @@ template<class vobj> inline void GatherSlice(cshiftVector<vobj> &buf,
stmp = getlane(from[w], lane); stmp = getlane(from[w], lane);
putlane(to[w], stmp, blane); putlane(to[w], stmp, blane);
} }
#endif
} }
}); });
/*
int words =block*nblock/simd[dim];
std::vector<vobj> tbuf(words);
acceleratorCopyFromDevice((void *)&buf[offset],(void *)&tbuf[0],words*sizeof(vobj));
typedef typename vobj::scalar_type scalar;
scalar *sbuf = (scalar *)&tbuf[0];
scalar tmp=0.0;
for(int w=0;w<words*sizeof(vobj)/sizeof(scalar);w++){
tmp=tmp+conjugate(sbuf[w])*sbuf[w];
}
std::cout << " Gathered buffer norm "<<tmp<<std::endl;
*/
} }