Using normal map for software bump mapping

I keep a copy of the 2D Bump Mapping article in my documents folder, and while there may be no practical need to do software bump mapping with today''s video hardware, I still like to dabble in software algorithms now and then, just for the fun of it. Anyway, I''ve taken an interest in using normal maps in place of bump maps because the bump data is already normalized, and in theory this should cut out about half the work of the bump mapping algorithm, making it even faster. Does anyone know of an algorithm, or have some tips for creating one, that utilizes normal maps in the place of bump maps? And while I''m on the subject of bump maps, does anyone know of a readable article that explains how to achieve self-shadowed bump mapping? I understand you can''t reasonably use normal maps to do this, but I just want to understand the technique. Thanks.