[llvm] [IA][RISCV] Add support for vp.load/vp.store with shufflevector (PR #135445)

[Alexey Bataev via llvm-commits]

================
@@ -339,25 +371,61 @@ bool InterleavedAccessImpl::lowerInterleavedLoad(
     return false;
 
   bool BinOpShuffleChanged =
-      replaceBinOpShuffles(BinOpShuffles.getArrayRef(), Shuffles, LI);
+      replaceBinOpShuffles(BinOpShuffles.getArrayRef(), Shuffles, LoadOp);
+
+  // Check if we extract only the unmasked elements.
+  if (MaskedIndices.any()) {
+    if (any_of(Shuffles, [&](const auto *Shuffle) {
+          ArrayRef<int> ShuffleMask = Shuffle->getShuffleMask();
+          for (int Idx : ShuffleMask) {
+            if (Idx < 0)
+              continue;
+            if (MaskedIndices.test(unsigned(Idx)))
+              return true;
+          }
+          return false;
+        })) {
+      LLVM_DEBUG(dbgs() << "IA: trying to extract a masked element through "
+                        << "shufflevector\n");
+      return false;
+    }
+  }
+  // Check if we extract only the elements within evl.
+  if (auto *VPLoad = dyn_cast<VPIntrinsic>(LoadOp)) {
+    uint64_t EVL = cast<ConstantInt>(VPLoad->getArgOperand(2))->getZExtValue();
----------------
alexey-bataev wrote:

`int64_t EVL = `

https://github.com/llvm/llvm-project/pull/135445