[llvm] [LV] Use wide lane masks as the canonical form when tail-folding & interleaving (PR #209484)
Benjamin Maxwell via llvm-commits
llvm-commits at lists.llvm.org
Wed Aug 5 01:46:20 PDT 2026
================
@@ -785,20 +788,25 @@ Value *VPInstruction::generate(VPTransformState &State) {
return Builder.CreateSelectFMF(Cond, Op1, Op2, getFastMathFlagsOrNone(),
Name);
}
- case VPInstruction::ActiveLaneMask: {
+ case VPInstruction::ActiveLaneMask:
+ case VPInstruction::WideActiveLaneMask: {
// Get first lane of vector induction variable.
Value *VIVElem0 = State.get(getOperand(0), VPLane(0));
// Get the original loop tripcount.
Value *ScalarTC = State.get(getOperand(1), VPLane(0));
+ unsigned Multiplier =
+ getOpcode() == VPInstruction::ActiveLaneMask
----------------
MacDue wrote:
very nit: I'd maybe swap this check to `== WideActiveLaneMask` (since that's the case with the extra operand).
https://github.com/llvm/llvm-project/pull/209484
More information about the llvm-commits
mailing list