[llvm] [LV] Support interleaving in order to reduce stalls (PR #197959)

Florian Hahn via llvm-commits llvm-commits at lists.llvm.org
Tue Jul 21 07:53:18 PDT 2026


================
@@ -1504,6 +1504,10 @@ class TargetTransformInfo {
   LLVM_ABI unsigned getMaxInterleaveFactor(ElementCount VF,
                                            bool HasUnorderedReductions) const;
 
+  /// \return True if the loop vectorizer should interleave in order to reduce
+  /// the number of stall cycles due to long latency instructions.
+  LLVM_ABI bool shouldInterleaveToReduceStalls() const;
----------------
fhahn wrote:

Do we already have a more general way to check if the CPU is in-order? Also, it looks like all implementations currently return false in the patch. can this be added separately?

https://github.com/llvm/llvm-project/pull/197959


More information about the llvm-commits mailing list