diff options
| author |  Rasmus Munk Larsen <rmlarsen@google.com> | 2016-10-04 15:08:41 -0700 |
|---|---|---|
| committer | Rasmus Munk Larsen <rmlarsen@google.com> | 2016-10-04 15:08:41 -0700 |
| commit | 765615609d40f485c718d60b76b35a277c36d5ce (patch) | |
| tree | 1657e741a5f02111eb1d2a48f4d07d8052470fea /Eigen/src/Core/arch/SSE/MathFunctions.h | |
| parent | 3ed67cb0bb4af65fbf243df598604a8c7630bf7d (diff) | |
Update comment for fast sqrt.
Diffstat (limited to 'Eigen/src/Core/arch/SSE/MathFunctions.h')
| -rw-r--r-- | Eigen/src/Core/arch/SSE/MathFunctions.h | 9 |
1 files changed, 7 insertions, 2 deletions
diff --git a/Eigen/src/Core/arch/SSE/MathFunctions.h b/Eigen/src/Core/arch/SSE/MathFunctions.h index 2c34a869d..7b5f948e1 100644 --- a/Eigen/src/Core/arch/SSE/MathFunctions.h +++ b/Eigen/src/Core/arch/SSE/MathFunctions.h @@ -444,9 +444,14 @@ Packet4f pcos<Packet4f>(const Packet4f& _x) #if EIGEN_FAST_MATH -// This is based on Quake3's fast inverse square root. +// Functions for sqrt. +// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step +// of Newton's method, at a cost of 1-2 bits of precision as opposed to the +// exact solution. It does not handle +inf, or denormalized numbers correctly. +// The main advantage of this approach is not just speed, but also the fact that +// it can be inlined and pipelined with other computations, further reducing its +// effective latency. This is similar to Quake3's fast inverse square root. // For detail see here: http://www.beyond3d.com/content/articles/8/ -// It lacks 1 (or 2 bits in some rare cases) of precision, and does not handle negative, +inf, or denormalized numbers correctly. template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f psqrt<Packet4f>(const Packet4f& _x) { |