[clang] [llvm] [AArch64][SVE] Improve fixed-length addressing modes. (PR #129732)

[Paul Walker via cfe-commits]

================
@@ -7380,12 +7380,26 @@ bool AArch64DAGToDAGISel::SelectAddrModeIndexedSVE(SDNode *Root, SDValue N,
     return false;
 
   SDValue VScale = N.getOperand(1);
-  if (VScale.getOpcode() != ISD::VSCALE)
+  int64_t MulImm = std::numeric_limits<int64_t>::max();
+  if (VScale.getOpcode() == ISD::VSCALE) {
+    MulImm = cast<ConstantSDNode>(VScale.getOperand(0))->getSExtValue();
+  } else if (auto C = dyn_cast<ConstantSDNode>(VScale)) {
+    int64_t ByteOffset = C->getSExtValue();
+    const auto KnownVScale =
+        Subtarget->getSVEVectorSizeInBits() / AArch64::SVEBitsPerBlock;
+
+    if (!KnownVScale || ByteOffset % KnownVScale != 0)
+      return false;
+
+    MulImm = ByteOffset / KnownVScale;
+  } else
     return false;
 
+  assert(MulImm != std::numeric_limits<int64_t>::max() &&
+         "Uninitialized MulImm.");
+
----------------
paulwalker-arm wrote:

This is not strictly true given you are initialising `MulImm` and whilst unlikely there's nothing stopping `ISD::VSCALE`'s operand from being `std::numeric_limits<int64_t>::max()`.

Given the following code will bail if `MulImm` is not changed from its initial value, you could just remove the assert?

https://github.com/llvm/llvm-project/pull/129732