SIMD Math Test

A performance comparison between different implementations of an SIMD vector using SSE2. The code is a modified version of the source code to Designing Fast Cross-Platform SIMD Vector Libraries by Gustavo Oliveira.

VMath

An SIMD implementation where the __m128 data type is typedefed as a 4D vector.

XNAMath

Microsoft implementation included with DirectX.

VClass

An SIMD implementation where the __m128 data type is wrapped in a C++ class and used as a 4D vector.

VClassTypedef

An SIMD implementation where the __m128 data type is typedefed. A 4D vector is then built upon the __m128 typedef.

VClassSIMDType

An SIMD implementation where the __m128 data type is wrapped in a C++ class. A 4D vector is then built upon the SIMD class.

The goal of this comparison is to see if abstracting a low level SIMD interface and then using that interface to create math classes is a viable approach. This is the subject of a series of articles on #AltDevBlogADay.

Defining an SIMD Interface

Thanks

Thanks to Gustavo Oliveira for allowing the use of his source code.