chelini wrote: Could you pls clarify “lower through appropriate arith ops” in the commit message? The connection is indirect: this patch maps the new GPU reduction kinds to the existing vector combining kinds. https://github.com/llvm/llvm-project/pull/226903