As the conjugate gradient method uses some matrix-vector products, I would expect some performance improvements if OpenMP is activated.
The documentation mentions that OpenMP activation allows parallelization on general matrix products and (as a special case?) on PartialPivLU.
Trying OpenMP on ConjugateGradient shows that internal matrix-vector products are actually not parallelized. How is OpenMP disabled here ?
Is it on purpose ? How can I enable OpenMP within GradientConjugate without recoding my own conjugate gradient method ?

Thank you,