[llvm] [BasicAA] Check for Overflow using vscale_range (PR #81144)

[Nikita Popov via llvm-commits]

================
@@ -1183,12 +1183,22 @@ AliasResult BasicAAResult::aliasGEP(
         ScalableVar.IsNegated ? -ScalableVar.Scale : ScalableVar.Scale;
     LocationSize VLeftSize = Scale.isNegative() ? V1Size : V2Size;
 
-    // Note that we do not check that the typesize is scalable, as vscale >= 1
-    // so noalias still holds so long as the dependency distance is at least as
-    // big as the typesize.
-    if (VLeftSize.hasValue() &&
-        Scale.uge(VLeftSize.getValue().getKnownMinValue()))
-      return AliasResult::NoAlias;
+    // Check if the offset is known to not overflow, if it does then attempt to
+    // prove it with the known values of vscale_range.
+    bool Overflows = !DecompGEP1.VarIndices[0].IsNSW;
+    if (Overflows) {
+      ConstantRange CR = getVScaleRange(&F, Scale.getBitWidth());
+      (void)CR.getSignedMax().smul_ov(Scale, Overflows);
+    }
+
+    if (!Overflows) {
+      // Note that we do not check that the typesize is scalable, as vscale >= 1
+      // so noalias still holds so long as the dependency distance is at least
+      // as big as the typesize.
+      if (VLeftSize.hasValue() &&
+          Scale.uge(VLeftSize.getValue().getKnownMinValue()))
----------------
nikic wrote:

I missed this when reviewing the original patch: Isn't this Scale supposed to be abs(Scale)? I don't think this does the right thing for negative scales right now.

https://github.com/llvm/llvm-project/pull/81144