See if we can get perf wins with either clang/gcc through the different lto/pgo/bolt options. If we do get performance wins, this helps with both the inner loop, outer loop and costs.