Currently glm::fma implementation just returns a * b + c, which defeats the purpose of fma.
It should instead use the std::fma function, which invokes an optimized machine instruction if available. This gives higher precision in this calculation, which is the whole reason why fma exists at all.