portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2026-08-31 14:59:35 +01:00
Code Releases Activity

Both changes to improve the buffer persistence in teh dense distributed inverse and to add verbosity to polynomial smoothers

Browse Source
This commit is contained in:
Peter Boyle
2026-08-26 17:48:03 -04:00
parent 5f8f396e9b
commit e15099eb7b
10 changed files with 266 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/debug/Test_blockcyclic.cc
+6 -6
View File
@@ -38,7 +38,7 @@ Author: Peter Boyle <pboyle@bnl.gov>
// equal mloc*nloc; LocalOffset is a bijection onto [0,mloc*nloc).
// T5 : block contiguity: within any owned global block, consecutive
// global rows are consecutive local rows (what SUMMA panels rely on).
// T6 : ChooseProcessGrid: exact factorisation, Pr<=Pc, most-square.
// T6 : ChooseProcessGrid: exact factorisation, Pr>=Pc, most-square.
//////////////////////////////////////////////////////////////////////////////
#include <Grid/Grid.h>
@@ -227,15 +227,15 @@ int main(int argc, char **argv)
int Pr,Pc;
BlockCyclicLayout::ChooseProcessGrid(P,Pr,Pc);
if ( Pr*Pc != P ) ok = false;
if ( Pr > Pc ) ok = false;
// most-square: no divisor r with Pr < r <= sqrt(P)
for(int r=Pr+1; (int64_t)r*r <= (int64_t)P; r++)
if ( Pr < Pc ) ok = false; // Pr >= Pc (SUMMA B panels unpadded)
// most-square: no divisor r with Pc < r <= sqrt(P)
for(int r=Pc+1; (int64_t)r*r <= (int64_t)P; r++)
if ( P % r == 0 ) ok = false;
}
int Pr,Pc;
BlockCyclicLayout::ChooseProcessGrid(288,Pr,Pc);
if ( !(Pr==16 && Pc==18) ) ok = false;
Report("T6 ChooseProcessGrid exact, Pr<=Pc, most-square (288 -> 16x18)", ok);
if ( !(Pr==18 && Pc==16) ) ok = false; // Pr >= Pc: SUMMA B panels unpadded (S = ceil(Pc/Pr) = 1)
Report("T6 ChooseProcessGrid exact, Pr>=Pc, most-square (288 -> 18x16)", ok);
}
////////////////////////////////////////////////////////////////////////