bc7m16.c/.h is one of the strongest CPU BC7 encoders available at the moment for opaque textures:
In perceptual mode using luma PSNR on opaque textures it beats ispc_texcomp, without vectorization and only using 2 modes.
Using linear RGB metrics it's around 20% slower at same PSNR (compared to ispc_texcomp's fast profile). I don't suggest you use linear RGB metrics because the results visually look worse vs. perceptual metrics, as we learned years ago with BC1 encoding.
This encoder was basically the prototype for my new non-RDO ispc BC7 encoder I wrote for Basis, which supports all the modes and is even faster.