[llvm] [mlir] [mlir] fix overflow in LLVM dialect inlining (PR #72878)

[Tobias Gysi via llvm-commits]

================
@@ -597,16 +597,16 @@ static Value handleByValArgument(OpBuilder &builder, Operation *callable,
                     memoryEffects.getArgMem() != LLVM::ModRefInfo::Mod;
   // Check if there's an alignment mismatch requiring us to copy.
   DataLayout dataLayout = DataLayout::closest(callable);
-  unsigned minimumAlignment = dataLayout.getTypeABIAlignment(elementType);
+  uint64_t minimumAlignment = dataLayout.getTypeABIAlignment(elementType);
   if (isReadOnly) {
     if (requestedAlignment <= minimumAlignment)
       return argument;
-    unsigned currentAlignment =
+    uint64_t currentAlignment =
         tryToEnforceAlignment(argument, requestedAlignment, dataLayout);
----------------
gysit wrote:

`tryToEnforceAlignment` and `tryToEnforceAllocaAlignment` should probably also use uint64_t?

https://github.com/llvm/llvm-project/pull/72878