phynexis-deps/eigen
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Perf: Avoid heap allocation for fixed-max decomposition temporaries

Browse Source 
libeigen/eigen!2593

Closes #1477
This commit is contained in:
Florian Maurin
2026-06-01 01:40:21 +00:00
committed by Charles Schlosser
parent 662ba79d79
commit ed45bac320
3 changed files with 12 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/nomalloc.cpp
+10
View File
@@ -26,6 +26,7 @@ void nomalloc(const MatrixType& m) {
/* this test check no dynamic memory allocation are issued with fixed-size matrices
*/
typedef typename MatrixType::Scalar Scalar;
using RealScalar = typename NumTraits<Scalar>::Real;
Index rows = m.rows();
Index cols = m.cols();
@@ -78,6 +79,12 @@ void nomalloc(const MatrixType& m) {
m2.template selfadjointView<Upper>().rankUpdate(m1.row(0), -1);
m2.template selfadjointView<Lower>().rankUpdate(m1.col(0), m1.col(0)); // rank-2
MatrixType spd = m1 * m1.adjoint();
spd.diagonal().array() += RealScalar(rows);
LLT<MatrixType> llt(spd);
VERIFY_IS_EQUAL(llt.info(), Success);
llt.rankUpdate(m1.col(0), RealScalar(1));
// The following fancy matrix-matrix products are not safe yet regarding static allocation
m2.template selfadjointView<Lower>().rankUpdate(m1);
m2 += m2.template triangularView<Upper>() * m1;
@@ -109,6 +116,9 @@ void ctms_decompositions() {
LLT.compute(A);
X = LLT.solve(B);
x = LLT.solve(b);
LLT.compute(saA);
VERIFY_IS_EQUAL(LLT.info(), Eigen::Success);
LLT.rankUpdate(b, Scalar(1));
Eigen::LDLT<Matrix> LDLT;
LDLT.compute(A);
X = LDLT.solve(B);