On Mon, 28 Oct 2024 23:48:48 GMT, Quan Anh Mai <qa...@openjdk.org> wrote:
>> Generic vector operand resolution cocretizes generic operands based on type >> agnostic node size, its a post matcher pass, and its job is to replace >> generic MachOper operand nodes with cocrete ones (vec[SDXYZ]) which holds >> precise register mask needed by register allocator. > > @jaskarth I notice we have `process_late_inline_calls_no_inline` below, > please put lowering after it because `process_late_inline_calls_no_inline` > does do igvn. Invariants in the lowered graph will get scheduled out of the loop during GCM, but lowering may still impact the unrolling policy which is based on the [loop body size. ](https://github.com/openjdk/jdk/blob/master/src/hotspot/share/opto/loopTransform.cpp#L1024) ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/21599#discussion_r1820624424
