So, technically, if I want to implement this on a downstream level, what would I do?
- I built Python with
./configure—-enable-experimental-jit=yes-offand have the proper LLVM version installed. The stencils are generated injit_stencils.h– I save them – for each architecture (and for each build type debug/optimized?). - Elsewhere. I don’t have LLVM, I copy the
jit_stencils.hfrom 1 to the build directory. I build Python with./configure—-enable-experimental-jit=yes-off.
That’s it?