Mercurial > cgi-bin > hgwebdir.cgi > PR > Applications > SSR > SSR__Blocked_Matrix_Mult__Bench

changeset 19:233fe8a5208f

find changesets by author, revision, files, or words in the commit message
better load balance
author Nina Engelhardt <nengel@mailbox.tu-berlin.de>
date Mon, 16 Apr 2012 18:27:12 +0200
parents 5671f96e0b5f
children bf6a4474f0e0 8972c00c00dd
files SSR_Matrix_Mult/Divide_Pr.c
diffstat 1 files changed, 5 insertions(+), 5 deletions(-) [+]
line diff
     1.1 --- a/SSR_Matrix_Mult/Divide_Pr.c	Mon Mar 19 10:29:43 2012 -0700
     1.2 +++ b/SSR_Matrix_Mult/Divide_Pr.c	Mon Apr 16 18:27:12 2012 +0200
     1.3 @@ -196,7 +196,7 @@
     1.4  
     1.5              DEBUG__printf(dbgAppFlow,"**create result Pr**")
     1.6        resultPr =
     1.7 -         SSR__create_procr_with( &gatherResults, resultsParams, animPr);
     1.8 +         SSR__create_procr_with_affinity( &gatherResults, resultsParams, animPr,0);
     1.9  
    1.10           //Make the sub-matrices, and pair them up, and make processor to
    1.11           // calc product of each pair.
    1.12 @@ -261,7 +261,7 @@
    1.13     idealNumWorkUnits = SSR__giveIdealNumWorkUnits();
    1.14     
    1.15     idealSizeOfSide2 = leftMatrix->numRows / rint(cbrt( idealNumWorkUnits ));
    1.16 -   idealSizeOfSide2 *= 0.6; //finer granularity to help load balance
    1.17 +   idealSizeOfSide2 *= 0.4; //finer granularity to help load balance
    1.18  
    1.19     if( idealSizeOfSide1 > idealSizeOfSide2 )
    1.20        idealSizeOfSide = idealSizeOfSide1;
    1.21 @@ -367,10 +367,10 @@
    1.22  
    1.23     numCores = SSR__give_number_of_cores_to_schedule_onto();
    1.24  
    1.25 -   numToPutOntoEachCore = numRowIdxs*numColIdxs/numCores;
    1.26 +   numToPutOntoEachCore = numRowIdxs*numColIdxs/(numCores-1);
    1.27     leftOverFraction = 0;
    1.28     numVecOnCurrCore = 0;
    1.29 -   coreToAssignOnto = 0;
    1.30 +   coreToAssignOnto = 1;
    1.31  
    1.32     for( resRowIdx = 0; resRowIdx < numRowIdxs; resRowIdx++ )
    1.33      {
    1.34 @@ -412,7 +412,7 @@
    1.35  
    1.36                    //Move to next core, max core-value to incr to is numCores -1
    1.37                 coreToAssignOnto += 1;
    1.38 -               if( coreToAssignOnto >= numCores ) coreToAssignOnto = 0;
    1.39 +               if( coreToAssignOnto >= numCores ) coreToAssignOnto = 1;
    1.40               } //if
    1.41            } //for( vecIdx
    1.42         } //for( resColIdx