Re: d3dx9_36 [try 2]: Save multiplications and divisions to improve performance

[Matteo Bruni]

2015-06-25 23:45 GMT+02:00 David Adam <david.adam.cnrs(a)gmail.com>:
> Hi,
>
> I added a test with 65000 sides. All the test pass with a native d3dx9_36
> dll.

You should have mentioned that the new test doesn't pass with the
previous implementation, which is what makes this patch really
convincing. So this patch is okay to me.

Also notice that this in practice doesn't save any multiplication in
the loop, at least with my version of GCC: the multiplication 2.0f *
D3DX_PI was precalculated at compile time. A better patch title
mentioning D3DXCreatePolygon would be nice.

The tests generally have some room for improvement. As a followup it
would be nice to try and improve them, maybe by replacing the
comparison function with one which allows for a relative error and
makes it possible to choose a potentially different error value for
each test (to be able to lower those as much as possible).

I would like to point you to some other place in Wine where the float
comparison is done right but, on the spot, I only found relative_error
+ expect_color macros from d3dx9_36/tests/math.c which are really
ugly. FWIW https://randomascii.wordpress.com/2012/02/25/comparing-floating-point-number...
seems to be a good read on the subject.