GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-13 20:26:03 +08:00
Code Issues Packages Projects Releases Wiki Activity

Make sure that micro-panel-size is smaller than blocking sizes (otherwise we might get a buffer overflow)

Browse Source
This commit is contained in:
Gael Guennebaud 2016-01-26 23:31:48 +01:00
parent 5b0a9ee003
commit aa8c6a251e
1 changed files with 6 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
Eigen/src/Core/products/TriangularMatrixMatrix.h
View File

@ -126,6 +126,10 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
Index kc = blocking.kc(); // cache block size along the K direction Index kc = blocking.kc(); // cache block size along the K direction
Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction
// The small panel size must not be larger than blocking size.
// Usually this should never be the case because SmallPanelWidth^2 is very small
// compared to L2 cache size, but let's be safe:
Index panelWidth = (std::min)(Index(SmallPanelWidth),(std::min)(kc,mc));
std::size_t sizeA = kc*mc; std::size_t sizeA = kc*mc;
std::size_t sizeB = kc*cols; std::size_t sizeB = kc*cols;
@ -169,9 +173,9 @@ EIGEN_DONT_INLINE void product_triangular_matrix_matrix<Scalar,Index,Mode,true,
if(IsLower || actual_k2<rows) if(IsLower || actual_k2<rows)
{ {
// for each small vertical panels of lhs // for each small vertical panels of lhs
for (Index k1=0; k1<actual_kc; k1+=SmallPanelWidth) for (Index k1=0; k1<actual_kc; k1+=panelWidth)
{ {
Index actualPanelWidth = std::min<Index>(actual_kc-k1, SmallPanelWidth); Index actualPanelWidth = std::min<Index>(actual_kc-k1, panelWidth);
Index lengthTarget = IsLower ? actual_kc-k1-actualPanelWidth : k1; Index lengthTarget = IsLower ? actual_kc-k1-actualPanelWidth : k1;
Index startBlock = actual_k2+k1; Index startBlock = actual_k2+k1;
Index blockBOffset = k1; Index blockBOffset = k1;