Made the blocking computation aware of the l3 cache

Also optimized the blocking parameters to take into account the number of threads used for a computation
This commit is contained in:
Benoit Steiner
2014-10-15 15:32:59 -07:00
parent dba55041ab
commit bfdd9f3ac9
13 changed files with 117 additions and 79 deletions

View File

@@ -96,7 +96,7 @@ struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,Dynamic>
typedef internal::gemm_blocking_space<(Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor,Scalar,Scalar,
Rhs::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime, Lhs::MaxRowsAtCompileTime,4> BlockingType;
BlockingType blocking(rhs.rows(), rhs.cols(), size);
BlockingType blocking(rhs.rows(), rhs.cols(), size, 1, false);
triangular_solve_matrix<Scalar,Index,Side,Mode,LhsProductTraits::NeedToConjugate,(int(Lhs::Flags) & RowMajorBit) ? RowMajor : ColMajor,
(Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor>