[llvm] [AArch64][SVE] Optimize svand_z/svorr_z with all-true predicates. (PR #160408)

[Paul Walker via llvm-commits]

================
@@ -1587,8 +1594,21 @@ static bool isAllActivePredicate(Value *Pred) {
     if (cast<ScalableVectorType>(Pred->getType())->getMinNumElements() <=
         cast<ScalableVectorType>(UncastedPred->getType())->getMinNumElements())
       Pred = UncastedPred;
-  auto *C = dyn_cast<Constant>(Pred);
-  return (C && C->isAllOnesValue());
+
+  // Also look through just convert.to.svbool if the input is an all-true splat
+  Value *ConvertArg;
+  if (match(Pred, m_Intrinsic<Intrinsic::aarch64_sve_convert_to_svbool>(
+                      m_Value(ConvertArg))))
+    Pred = ConvertArg;
+  // Check for splat(i1 true) pattern used by svptrue intrinsics
+  if (auto *C = dyn_cast<Constant>(Pred)) {
+    if (C->isAllOnesValue())
+      return true;
+    if (auto *SplatVal = C->getSplatValue())
+      if (auto *CI = dyn_cast<ConstantInt>(SplatVal))
+        return CI->isOne();
+  }
----------------
paulwalker-arm wrote:

This is not a valid transformation.  I'll elaborate more below (see test_sve_and_z_all_true_right).

https://github.com/llvm/llvm-project/pull/160408