tomtor wrote: loop-reduce does not preserve predecrement operations in loops which it cannot optimize. This is a non AVR specific bug/issue, but on architectures with rich addressing modes it is hidden (a scaled index is used on Intel/ARM) and RISCV has no post/pre addressing.
In functions which have an `incdec` ptr I disable the loop-reduce. That is a bit of overkill, this could/should be done on an individual loop instead of the whole function, but in practice the difference will be small, because `incdec` ptrs are rare in generated AVR code. I see no easy way to do this on individual loops. @Patryk27 suggestions? https://github.com/llvm/llvm-project/pull/152028 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
