f30505b836141030f50cdd941f9cbc8f4cda0995,scipy/sparse/linalg/isolve/lgmres.py,,lgmres,#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#,14

Before Change

                v_new = axpy(v, v_new, v.shape[0], -alpha)  // v_new -= alpha*v
            hcur.append(nrm2(v_new))

            if hcur[-1] == 0:
                // Exact solution.
                break
            else:
                v_new = scal(1.0/hcur[-1], v_new)

            vs.append(v_new)
            ws.append(z)

            hcur = np.asarray(hcur, dtype=Q.dtype)

After Change

                // Careful with denormals
                alpha = 1/hcur[-1]

            if np.isfinite(alpha):
                v_new = scal(alpha, v_new)
            else:
                // v_new either zero (solution in span of previous
                // vectors) or we have nans.  If we already have
                // previous vectors in R, we can discard the current
                // vector and bail out.
                if j > 1:
                    j -= 1
                    break

            vs.append(v_new)
            ws.append(z)

            // -- GMRES optimization problem

Instances