Good points. What I'm looking for would be primarily for use in 3D calculations, for 3D games. So just small vectors, small matrices, and quaternions. And operations like building translation/rotation/viewport matrices, generating camera projection/ortho matrices, quaternion interpolation, etc.
I know with C++ the math libraries can do operator overloading to make them feel natural, and I wouldn't have any of that in a C library, but that's fine. Even if the library was limited to integers and floats (and didn't mix them), since it doesn't have the template abstraction, it would be fine. When you're doing matrix operations with the goal of feeding it to the GPU, there isn't a lot of variety needed.
Simple enough set of requirements that I could probably write my own library if I had to, but I don't have the math optimization experience necessary to make the library as fast as it would need to be.