Notes: Supports left and right preconditioning
See KSPBCGSL for additional stabilization
Unlike the Bi-CG-stab algorithm, this requires one multiplication be the transpose of the operator before the iteration starts.
The paper has two errors in the algorithm presented, they are fixed in the code in KSPSolve_IBCGS()
For maximum reduction in the number of global reduction operations, this solver should be used with KSPSetLagNorm().
This is not supported for complex numbers.
Reference: The Improved BiCGStab Method for Large and Sparse Unsymmetric Linear Systems on Parallel Distributed Memory Architectures. L. T. Yang and R. Brent, Proceedings of the Fifth International Conference on Algorithms and Architectures for Parallel Processing, 2002, IEEE.