Giters

microsoft / DirectXMath

DirectXMath is an all inline SIMD C++ linear algebra library for use in games and graphics apps

Home Page:https://walbourn.github.io/introducing-directxmath/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Optimise 'XMVectorSin/Cos/SinEst/CosEst/SinCos/SinCosEst' with FMA

john-h-k opened this issue · comments

commented

None of them currently use it for the polynomial approximation (appears tan does), but with slight rearrangement of the constant FMA can be used to accelerate the x64 pathway. I saw a speedup of roughly 6.5ns->4.8ns when using FMA with Sin

commented

Thanks. I'll take a look at this for AVX2 improvements in a future release.

XMVectorTan is implemented in terms of other DirectXMath functions which is why it ends up using FMA3 with /arch:AVX2. The Sin/Cos are implemented directly in platform intrinsics so I need to review those.