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

test prefetch to L2 in stencil

Browse Source
This commit is contained in:
nmeyer-ur 2020-06-08 09:39:50 +02:00
parent 9872c76825
commit 93a37c8f68
1 changed files with 5 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
Grid/stencil/Stencil.h
View File

@ -75,20 +75,20 @@ void Gather_plane_simple_table (Vector<std::pair<int,int> >& table,const Lattice
// prefetching: // prefetching:
// +1% performance for Wilson on 32**4 // +1% performance for Wilson on 32**4
// -2% performance for DW on 24**4 x 12 // -2% performance for DW on 24**4 x 12
/*
const int dist = 2; const int dist = 7;
if (i+dist < num){ if (i+dist < num){
svbool_t pg1 = svptrue_b64(); svbool_t pg1 = svptrue_b64();
// prefetch input // prefetch input
auto in = rhs_v(so+table_v[i+dist].second); auto in = rhs_v(so+table_v[i+dist].second);
svprfd(pg1, (char*)&in, SV_PLDL1STRM); svprfd(pg1, (char*)&in, SV_PLDL2STRM);
// prefetch store buffer // prefetch store buffer
uint64_t o = table_v[i+dist].first; uint64_t o = table_v[i+dist].first;
svprfd(pg1, (char*)&buffer[off+o], SV_PSTL1STRM); svprfd(pg1, (char*)&buffer[off+o], SV_PSTL2STRM);
} }
*/
compressed_t tmp_c; compressed_t tmp_c;
uint64_t o = table_v[i].first; uint64_t o = table_v[i].first;
compress.Compress(&tmp_c,0,rhs_v(so+table_v[i].second)); compress.Compress(&tmp_c,0,rhs_v(so+table_v[i].second));