This short book contains a terrific explanation of the IEEE floating point standard- a document that is of indescribable importance to the field of numerical analysis as well as modern computing. All of the numerical analysis books I've read (or at least paged through) mention the standard in passing, and if you are lucky, a few fundamental concepts. Overton's book goes into much more detail, explaining not only how floating point arithmetic is implemented, but why certain things happen (exceptions, over/underflow) and the historical basis for the decisions that were made in developing the standard. Several example programs (in C) are provided which demonstrate the key concepts, and nearly every section of the book provides questions and exercises that are thought-provoking and well beyond the realm of simple comprehension checks.

That would have been fine, but interjected in this otherwise excellent book is a lot of Overton's opinions. I think he overstepped the scope of the book a bit... big-endian, MATLAB-loving, subnormal-floating-point-numbers-supporting soap boxes are better left for post-conference evenings at the pub. It's easy to poke holes in the standards- look no further than ASCII- but extremely widely-adopted standards are hard to change, especially on the whims of a single professor. Despite the occasional diatribe, this is still a wonderful little book that both numerical analysts and scientists that write code should read.

Скачать книгу бесплатно (pdf, 6.31 Mb)

---

Читать «Numerical Computing with IEEE Floating Point Arithmetic»