[llvm] [SLP]Prefer segmented/deinterleaved loads to strided and fix codegen (PR #135058)

[Gaëtan Bossu via llvm-commits]

================
@@ -5787,6 +5789,18 @@ BoUpSLP::canVectorizeLoads(ArrayRef<Value *> VL, const Value *VL0,
     // Check that the sorted loads are consecutive.
     if (static_cast<unsigned>(*Diff) == Sz - 1)
       return LoadsState::Vectorize;
+    bool IsMasked;
+    unsigned InterleaveFactor;
+    SmallVector<int> CompressMask;
+    VectorType *LoadVecTy;
----------------
gbossu wrote:

Nit: I think all those new local variables aren't used, and they pollute an already very long function. Maybe create an overloaded `isMaskedLoadCompress` which does not take those "output" parameters, or turn them into pointers that can be `nullptr`.

https://github.com/llvm/llvm-project/pull/135058