Krystof Beyls has recently uploaded a patch which contributes the Arm optimized math routines (AOR) to LLVM libc: https://reviews.llvm.org/D75355.
To take this forward, I would like to propose the following sequence of steps:
Step 1: Submit this patch as is .
Step 2: Add a step on the libc builders to run tests from the AOR directory .
Step 3: Move the implementations from the AOR directory in parts over to their right place in the LLVM-libc tree. This will have to be done in many small patches which go through the normal code review process.
Step 4: Setup benchmarking and fuzzing infrastructure for the math functions.
 - This will give up proper diffs to look at during code reviews et al.
 - The math functions in the AOR directory are built on a common infrastructure residing within the AOR directory. As we move parts of AOR to their right place, we will have to update/move this common infrastructure. Hence, running the tests on the bots helps us ensure that our changes do not break the parts which have not yet been moved to their right place in the LLVM-libc tree.