GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-12 03:39:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

Avoid undeflow when blocking size are tuned manually.

Browse Source
This commit is contained in:
Gael Guennebaud 2015-03-06 21:51:09 +01:00
parent 14a5f135a3
commit 0ee391863e
1 changed files with 9 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
Eigen/src/Core/products/GeneralBlockPanelKernel.h
View File

@ -975,7 +975,11 @@ void gebp_kernel<LhsScalar,RhsScalar,Index,DataMapper,mr,nr,ConjugateLhs,Conjuga
// However, if depth is too small, we can extend the number of rows of these horizontal panels. // However, if depth is too small, we can extend the number of rows of these horizontal panels.
// This actual number of rows is computed as follow: // This actual number of rows is computed as follow:
const Index l1 = 32*1024; // in Bytes, TODO, l1 should be passed to this function. const Index l1 = 32*1024; // in Bytes, TODO, l1 should be passed to this function.
const Index actual_panel_rows = (3*LhsProgress) * ( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 3*LhsProgress) ); #ifdef EIGEN_TEST_SPECIFIC_BLOCKING_SIZES
const Index actual_panel_rows = (3*LhsProgress) * std::max<Index>(1,( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 3*LhsProgress) ));
#else
const Index actual_panel_rows = (3*LhsProgress) * ( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 3*LhsProgress) );
#endif
for(Index i1=0; i1<peeled_mc3; i1+=actual_panel_rows) for(Index i1=0; i1<peeled_mc3; i1+=actual_panel_rows)
{ {
const Index actual_panel_end = (std::min)(i1+actual_panel_rows, peeled_mc3); const Index actual_panel_end = (std::min)(i1+actual_panel_rows, peeled_mc3);
@ -1208,7 +1212,11 @@ void gebp_kernel<LhsScalar,RhsScalar,Index,DataMapper,mr,nr,ConjugateLhs,Conjuga
if(mr>=2*Traits::LhsProgress) if(mr>=2*Traits::LhsProgress)
{ {
const Index l1 = 32*1024; // in Bytes, TODO, l1 should be passed to this function. const Index l1 = 32*1024; // in Bytes, TODO, l1 should be passed to this function.
#ifdef EIGEN_TEST_SPECIFIC_BLOCKING_SIZES
Index actual_panel_rows = (2*LhsProgress) * std::max<Index>(1,( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 2*LhsProgress) ));
#else
Index actual_panel_rows = (2*LhsProgress) * ( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 2*LhsProgress) ); Index actual_panel_rows = (2*LhsProgress) * ( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 2*LhsProgress) );
#endif
for(Index i1=peeled_mc3; i1<peeled_mc2; i1+=actual_panel_rows) for(Index i1=peeled_mc3; i1<peeled_mc2; i1+=actual_panel_rows)
{ {
Index actual_panel_end = (std::min)(i1+actual_panel_rows, peeled_mc2); Index actual_panel_end = (std::min)(i1+actual_panel_rows, peeled_mc2);