Re:Approximating sin(x) to 5 ULP with Chebyshev polynomials
by raiph on Saturday May 13, 2017 @09:35AM
(#14330881)
I hear your basic points, and this is a tangent on a tangent, but:
Rakudo, the existing implementation of standard Perl 6, is a compiler.
(I wouldn't be surprised if the code currently generated by Rakudo for this algorithm is far slower than the code generated by rustc but that isn't because it isn't a compiler.)
It won't - and I don't have to measure to know.
* OP uses standard optimization techniques. Their use has been justified over decades of actual research.
* OP uses a compiled language. OP can look at disassembly and see what's happening on the CPU.
* Your use of +reduce+ does not make sense at all - it's just syntactic sugar (you of course can use HOFs to handle constants, but that's completely missing the idea of function composition, which arguably is the only reason +reduce+ exists).