Compilers cannot do this optimization for floating point [1] unless you're compiling with -ffast-math. In general, don't rely on compilers to optimize floating point sub-expressions.

[1]: https://godbolt.org/z/8bEjE9Wxx