[llvm] [AArch64] Prefer SVE2 for fixed-length i64 [S|U][MIN|MAX] reductions (PR #181161)

Paul Walker via llvm-commits llvm-commits at lists.llvm.org
Mon Feb 16 08:50:29 PST 2026


================
@@ -1936,10 +1943,6 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
       setOperationAction(ISD::UMAX, MVT::v2i64, Custom);
       setOperationAction(ISD::UMIN, MVT::v1i64, Custom);
       setOperationAction(ISD::UMIN, MVT::v2i64, Custom);
-      setOperationAction(ISD::VECREDUCE_SMAX, MVT::v2i64, Custom);
-      setOperationAction(ISD::VECREDUCE_SMIN, MVT::v2i64, Custom);
-      setOperationAction(ISD::VECREDUCE_UMAX, MVT::v2i64, Custom);
----------------
paulwalker-arm wrote:

I'd rather keep these here because this block "resets" all the lowering decisions, with these lines overriding the "new" default. I'm not sure why there are no test failures, so it's possible they are now redundant. This is another reason to keep them as removing them should be an isolated change.

FYI: I am going audit the structure now that the "without NEON" use case has settled down.

https://github.com/llvm/llvm-project/pull/181161


More information about the llvm-commits mailing list