[llvm] Add Addend Checks for MOVT and MOVW instructions. (PR #111970)

[David Spickett via llvm-commits]

================
@@ -446,6 +447,16 @@ unsigned ARMAsmBackend::adjustFixupValue(const MCAssembler &Asm,
                                          const MCSubtargetInfo* STI) const {
   unsigned Kind = Fixup.getKind();
 
+  // For MOVW/MOVT Instructions, the Fixup Value needs to be 16 bit aligned.
+  // If this is not the case, we should reject compilation.
+  if((Kind == ARM::fixup_arm_movw_lo16 || Kind == ARM::fixup_arm_movt_hi16 ||
+      Kind == ARM::fixup_t2_movw_lo16 || Kind == ARM::fixup_t2_movt_hi16) &&
+      (!(minIntN(16) <= static_cast<int64_t>(Value) &&
+    static_cast<int64_t>(Value) <= maxIntN(16)))) {
----------------
DavidSpickett wrote:

One day we'll be able to do Python's `x < y < z` :unicorn: 

https://github.com/llvm/llvm-project/pull/111970