more accurate sqrt function #129
Conversation
| // Complex64::new(2.4421097261308304e-162, 1.0115549693666347e-162) | ||
| // ); | ||
|
|
||
| if self.re.is_zero() && self.im.is_zero() { |
There was a problem hiding this comment.
Can you add a source for all these special cases? e.g.
https://en.cppreference.com/w/c/numeric/complex/csqrt
(and make sure all those are covered)
There was a problem hiding this comment.
I added more test to test_nan to make sure all of these are covered by theses and added a comment
src/lib.rs
Outdated
| } | ||
| if self.re.is_nan() { | ||
| // nan + nan i | ||
| return Self::new(self.re, (self.im - self.im) / (self.im - self.im)); |
There was a problem hiding this comment.
We have a direct NaN -- not sure if this should also copysign though.
| return Self::new(self.re, (self.im - self.im) / (self.im - self.im)); | |
| return Self::new(self.re, T::nan()); |
There was a problem hiding this comment.
I wasn't 100% sure about the sign of nan here and 100% matched the code just to be sure but it seems T::nan() is indeed sufficient and the sign doesn't change.
| // √(inf +/- x i) = inf +/- 0 i | ||
| // √(-inf +/- NaN i) = NaN +/- inf i | ||
| // √(-inf +/- x i) = 0 +/- inf i | ||
|
|
There was a problem hiding this comment.
Maybe add a variable to make this clearer:
| #[allow(clippy::eq_op)] | |
| let zero_or_nan = self.im - self.im; |
There was a problem hiding this comment.
that is indeed more readable, I also added a comments. good point
| if scale { | ||
| self = self / four; | ||
| } | ||
| if self.re.is_sign_negative() { |
There was a problem hiding this comment.
We could also use a citation and link in a comment for the algorithm you mentioned.
There was a problem hiding this comment.
I added a citation to the algorithm and the musl libc implementation as well as provide some additional background in a comement
|
Thanks for the fast review! Apologies for not getting back to this yet. I had to unexpectetly travel for work so I will not be able to get back to this before the weekend. |
|
This should be ready for another round of review, apologies again for the delay |
I noticed that the square root implementation in num-complex uses conversion trough polar coordinates to compute complex squre roots. Usually the algorithm from https://dl.acm.org/doi/abs/10.1145/363717.363780 is used to compute the complex square root instead.
The algorithm uses hypot/norm and square root to compute csqrt. This approach should be faster since less transcendental calls are needed. Hypot and sqrt also tend to be faster compared to exp/ln/atan/cos/sin. Both hypot and sqart also have much higher precision. Most implementations guarantee that these two functions return the correctly rounded infinite accuracy result.
For prior art you can look at the glibc and musl implementation (https://git.musl-libc.org/cgit/musl/tree/src/complex/csqrt.c). The glibc implementation is a lot more complicated/hard to read because they ensure that underflow floating point exceptions are triggered correctly. I don't think that is something num-complex needs to do. There is also some accuracy loss for subnormal numbers that I didn't handle yet. I left a comment about it, but it is very minor.