[Mlir-commits] [mlir] [MLIR][Bytecode] Enforce alignment requirements (PR #157004)

[Mehdi Amini]

================
@@ -117,6 +119,57 @@ TEST(Bytecode, MultiModuleWithResource) {
   checkResourceAttribute(*roundTripModule);
 }
 
+TEST(Bytecode, AlignmentFailure) {
+  MLIRContext context;
+  Builder builder(&context);
+  ParserConfig parseConfig(&context);
+  OwningOpRef<Operation *> module =
+      parseSourceString<Operation *>(irWithResources, parseConfig);
+  ASSERT_TRUE(module);
+
+  // Write the module to bytecode.
+  MockOstream ostream;
+  EXPECT_CALL(ostream, reserveExtraSpace).WillOnce([&](uint64_t space) {
+    ostream.buffer = std::make_unique<std::byte[]>(space);
+    ostream.size = space;
+  });
+  ASSERT_TRUE(succeeded(writeBytecodeToFile(module.get(), ostream)));
+
+  // Create copy of buffer which is not aligned to requested resource alignment.
+  std::string buffer((char *)ostream.buffer.get(),
+                     (char *)ostream.buffer.get() + ostream.size);
+  size_t bufferSize = buffer.size();
+
+  // Increment into the buffer until we get to a power of 2 alignment that is
+  // not 32 bit aligned.
----------------
joker-eph wrote:

32 bits or bytes? I would think that 32 bits would be `Align(4)`.

Also the comment is a bit strange: a buffer is always aligned to a power of two since 1 is a power of two. That probably deserves some rephrasing (and maybe code adjustment).

https://github.com/llvm/llvm-project/pull/157004