Hi Hal,
Thanks for the very detailed email. I followed the ongoing discussion about moving ISL and Polly from a subproject of LLVM into the LLVM library. I was not convinced by the arguments in the threads. I believe that the potential benefits of the change that you are trying to make are not proportional to the high cost for the rest of the users of the compiler library. Traditional compiler optimizations do a much better job in almost every dimension: complexity, compile time, binary size, and performance. Polly has been developed for a decade and I have not seen the major adoption or benchmarks that justify the cost associated with adding an ILP solver into LLVM. Is there a reason why you don’t want to develop polly as an optimization that loads as a plugin into the clang optimization pipeline?
Thanks,
Nadav