Use float instead of doubles

Question

Use float instead of doubles

TerensTare opened this issue 4 years ago · 4 comments

Since this is supposed to be the "standard way" of doing 2d graphics in C++, I think we need it to be as fast as it can. From what I saw from the implementation, vector_2d and rectangle use double for members. C++ also offers a float type with smaller precision, but which is proven to be faster. I personally have been using float for my Vector2d class and I have seen no problems, so I think the implementation would gain some performance by using floats.

Answer 1 · 2020-08-09T12:46:34.000Z

afaik this is not true. https://stackoverflow.com/a/417591/347508

Answer 2 · 2020-08-09T15:43:56.000Z

Apparently, if we do computations on GPU, floats are faster (link here). But on Android, floats are even slower than ints (link here).

Answer 3 · 2020-08-09T16:46:28.000Z

before this gets religious ... i checked the source. can you provide links? i've found almost all double usage is in the examples and graphics backends, but not the main implementation:

https://github.com/cpp-io2d/P0267_RefImpl/search?q=double&unscoped_q=double

Answer 4 · 2020-08-09T18:23:06.000Z

I read the paper about this proposal (Revision 6). There I saw two classes (namely vector_2d and rectangle), which used double for members. I found a fork of this repo with the implementation of these classes. I decided to file an issue there, but since you can't open an issue on a fork, I went to this repo (the parent one). It appears that the forked repo is some commits (and a revision) behind this one, as the classes are now renamed to basic_point_2d and basic_bounding_box and use float for members. Guess it was just my bad at all😅😅.