nir_constant_expressions.py conflicting float16_t definition
This script creates a bunch of fixed length named types. In particular it creates a float16_t type that aliases to 32-bit float. On platforms that have provide a float16_t type this results in compiler errors that are hard to resolve.
In particular arm_neon.h defines a float16_t type which conflicts with this.
Additional wrinkle, src/util/half_float.h also declares a float16_t type for C++ code which has 16-bit storage. This is in direct contrast to the nir type.
Not sure how best to resolve this short of renaming all the nir types to something prefixed with mesa_
and making everyone hate that.