A new "cache-aware" strategy for tail duplication. Differential Revision: https://reviews.llvm.org/D123050