Potential numerical and semantic issues with log-based primal-dual residual gap

[LucasBoTang]

Hi all,

I noticed the following line in solver.cu:

    double primal_dual_residual_gap = abs(
        log10(state->relative_dual_residual /
              state->relative_primal_residual));

Since log10(...) returns a double, I’m wondering why abs is used here instead of fabs. Is this intentional, or would fabs be safer?

Best,
Bo