DirectXMath 3.15 includes the following changes:

• Added XMMatrixVectorTensorProduct for creating a matrix from two vectors
• Use of __m256 registers and FMA3 with /arch:AVX2 for stream and some matrix functions
• Optimized load/stores for SSE2 float2 & float3 functions
• Optimized some instruction choices for better AMD CPU support
• Improved conformance for clang/LLVM, GCC, and MinGW compilers
• Code review (constexpr / noexcept usage)
• Retired VS 2015 support

See this post.

This version is also available on NuGet.