Thou shalt have no other gods before the ANSI C standard 1412

In response to Trevor's challenge, I provided a code example for a time-critical cryptographic computation: basically, evaluating an input polynomial modulo 2^130-5. The code needs 64-bit IEEE floating-point ``double,'' among other things.

According to the 36-bit-la-la-land religion, this code is heresy, because it relies on guarantees that aren't provided by the ANSI C standard. I'm supposed to instead write much slower code that works with 13-bit bytes, gaps in the integer representation, blah blah blah.

Trevor L. Jackson, III

outline of the 64-bit implementation.

This code was actually translated from another language where the casts the lines labelled ``input.''

Your suggestion to remove the casts is poor engineering: nonzero cost, zero benefit.

The subtractions all happen in parallel on typical computers. Copying from one register to another would require waiting for that first zero register---a pointless slowdown. Loading from memory would compete with other loads happening at the same time---again, a pointless slowdown.

Sounds like you need better fonts. Anyway, in the input language, none of the uses of that variable are in positions where a number is allowed.

---D. J. Bernstein, buttociate Professor, Department of Mathematics, Statistics, and Computer Science, University of Illinois at Chicago