Re: [PATCH v2 0/7] MR62: vkd3d-shader/hlsl: sin(), cos(), sqrt() and step() intrinsics.

[Zebediah Figura (＠zfigura)]

On Tue Jan 17 01:04:43 2023 +0000, Francisco Casas wrote:
> I see, and the difference between `8.94069672e-007` and plain old `0.0`
> in terms of _ulps_ seems to be impractically high.
> How about using the `round()` intrinsic, as in the update?
> Worth noting: The [d3d11 functional
> specification](https://microsoft.github.io/DirectX-Specs/d3d/archive/D3D11_3_FunctionalSpec...)
> says:
> > `The maximum absolute error is 0.0008 in the interval from -100*Pi to +100*Pi.`

Seems reasonable; also reasonable would be to just pick some different constants (since it seems to be the ones close to zero which are broken).