快速平方根计算.pdfVIP

FAST INVERSE SQUARE ROOT CHRIS LOMONT Abstract. Computing reciprocal square roots is necessary in many applications, such as vector normalization in video games. Often, some loss of precision is acceptable for a large increase in speed. This note examines and improves a fast method found in source- code for several online libraries, and provides the ideas to derive similar methods for other functions.1 1. Introduction Reading the math programming forum on [1], I ran across an interesting method to compute an inverse square root. The C code was essentially (my comments): float InvSqrt(float x) { float xhalf = 0.5f*x; int i = *(int*)x; // get bits for floating value i = 0x5f3759df - (i1); // gives initial guess y0 x = *(float*)i; // convert bits back to float x = x*(1.5f-xhalf*x*x); // Newton step, repeating increases accuracy return x; } The interesting part was the constant 0x5f3759df: where did it come from and why does the code work? Some quick testing in Visual C++.NET [2] showed the code above to be roughly 4 times faster than the naive (float)(1.0/sqrt(x)), and the maximum relative error over all floating point numbers was 0 so the method seems very useful. Three immediate questions were: 1) how does it work, 2) can it be improved, and 3) what bit master designed such an incredible hack? A quick search on Google for 0x5f3759df returned several hits, the most relevant being a reference to a thread about this code on comp.graphics.algorithms from Jan 9, 2002 [3], and an (incorrect, 12000 Mathematics Subject Classification. Primary 65G99, 65Y99; Secondary 68W40.