solve_bicgstab: avoid iter==1 check

[eebasso]

Summary

This PR restructures solve_bicgstab such that the iter==1 check in the iter loop is avoided.

Additional background

This is one of the split off changes to the CG and BiCGSTAB implementations in MLCGSolver.H

[WeiqunZhang]

Could you please explain why the change is made? To prepare for more changes or just to avoid the iter == 1 test? But now rho = dotxy(rh,r); appears twice.

[eebasso]

The main reason is to avoid the iter==1 test, but it will also help prepare for more changes related to the larger PR suggesting improvements to the BiCGSTAB implementation. It's true that now rho = dotxy(rh,r) appears twice, but I currently can't think of a way to eliminate that while also avoiding the iter==1 test. One way to look at it is that everything above the iter loop is essentially the iter=0 initialization of the variables, including sol, r, rh, p, and rho.

[WeiqunZhang]

I think if (iter == 1) is much less an evil than duplicating rho = dotxy(rh,r);. Even if the latter were slightly better, that would not be a good reason to make the change, because any changes have the potential of introducing bugs.

[eebasso]

Okay, it was been rewritten so that rho = dotxy(rh,r) is not duplicated. The unnecessary initializations of alpha, omega, and rho_1 have also been eliminated. As far as bugs are concerned, what do the tests show?