i've seen this multiple times in the past few years, each time i've seen it i've understood just a tiny bit more... i would love to see an interview with the original author of this...
my impression (from reading the clang/llvm posts a while back) was that they tend to assume that anything with undefined behaviour in the spec returns a value that makes the code faster (eg. by simply ignoring the statement altogether).
I am wondering if in a lot of code taking this inverse square root is not necessary like when calculating length = sqrt(x^2+y^2) in a lot of code, you just want to know a minimum of maximum value so getting the min or max of x^2+y^2 is what you want and calculating the sqrt is not necessary.
Comments
i've seen this multiple times in the past few years, each time i've seen it i've understood just a tiny bit more... i would love to see an interview with the original author of this...
my impression (from reading the clang/llvm posts a while back) was that they tend to assume that anything with undefined behaviour in the spec returns a value that makes the code faster (eg. by simply ignoring the statement altogether).
edit: http://blog.llvm.org/2011/05/what-every-c-programmer-should-...
I am wondering if in a lot of code taking this inverse square root is not necessary like when calculating length = sqrt(x^2+y^2) in a lot of code, you just want to know a minimum of maximum value so getting the min or max of x^2+y^2 is what you want and calculating the sqrt is not necessary.