Mentions légales du service

Skip to content

Fix matrix initialization in triangular timings

This PR replaces !112 (closed) as this last one was introducing an issue with the Cholesky checking. It tries to minimizes the matrix A initialization to a single triangular part if both sides of the matrix is not required. Note that for now, checking is using a GEMM operation and thus requires both sides of the original symmetric/hermitian matrix to be initialized.

Merge request reports