INFO: Using _declspec(dllimport) & _declspec(dllexport) in Code

This article supplements the information covered in the following article in the Microsoft Knowledge Base: This article discusses the advantages and mechanics of using _declspec(dllimport) and _declspec(dllexport) in your application.

The 32-bit edition of Visual C++ uses _declspec(dllimport) and _declspec(dllexport) to replace the __export keyword previously used in 16-bit versions of Visual C++.

You do not need to use _declspec(dllimport) for your code to compile correctly, but doing so allows the compiler to generate better code. The compiler is able to generate better code because it knows for sure whether a function exists in a DLL or not, so the compiler can produce codes that skip a level of indirection that would normally be present in a function call that crossed a DLL boundary.

With the proper .DEF file EXPORTS section, _declspec(dllexport) is not required. _declspec(dllexport) was added to provide an easy way to export functions from an .EXE or .DLL without using a .DEF file.

The remainder of this article provides a fairly low-level, thorough discussion of these issues.

The Win32 Portable Executable format is designed to minimize the number of pages that must be touched to fix imports. To do this, it places all the import addresses for any program in one place called the Import Address Table. This allows the loader to modify only one or two pages when accessing these imports.

Using _declspec(dllimport) for Function Calls

In the following code example, assume func1 is a function that resides in a DLL separate from the .EXE file that contains the main() function.

Without _declspec(dllimport), given this code: the compiler generates code that looks like this: and the linker translates the call into something like this: If 'func1' exists in another DLL, the linker can't resolve this directly because it has no way of knowing what the address of 'func1' is. In 16-bit environments, the linker adds this code address to a list in the .EXE that the loader would patch at run-time with the correct address. In 32-bit environments, the linker generates a thunk for which it does know the address. The thunk looks like this: Here __imp_func1 is the address for func1's slot in the import address table of the .EXE file. All the addresses are thus known to the linker. The loader only has to update the .EXE file's import address table at load time for everything to work correctly.

Therefore, using _declspec(dllimport) is better because it is better if the linker does not generate a thunk if it does not have to. Thunks make the code larger (on RISC systems, it can be several instructions) and can degrade your cache performance. If you tell the compiler the function is in a DLL, it can generate an indirect call for you.

So now this code: generates this instruction: There is no thunk and no jmp instruction, so the code is smaller and faster.

On the other hand, for function calls inside a DLL, you don't want to have to use an indirect call. You already know a function's address. Time and space are required to load and store the address of the function before an indirect call, so a direct call is always faster and smaller. You only want to use __declspec(dllimport) when calling DLL functions from the outside the DLL itself. Don't use __declspec(dllimport) on functions inside a DLL when building that DLL.

Using _declspec(dllexport)

Microsoft introduced __export in the 16-bit compiler version to allow the compiler to generate the export names automatically and place them in a .LIB file. This .LIB file could then be used just like a static .LIB to link with a DLL.

Microsoft added __declspec(dllexport) to continue this convenience. Its purpose is to add the export directive to the object file so you don't need a .DEF file.

This convenience is most apparent when trying to export decorated C++ function names. There is no standard specification for name decoration, so the name of an exported function may change between compiler versions. If you use _declspec(dllexport), recompiling the DLL and dependent .EXE files is necessary only to account for any naming convention changes.

Many export directives such as ordinals, NONAME, or PRIVATE, can be made only in a .DEF file, and there is no way to specify these attributes without a .DEF file. However, using _declspec(dllexport) in addition to using a .DEF file does not cause build errors.

As a reference, search through the Win32 WINBASE.H header file. It contains examples of preferred __declspec(dllexport) and __declspec(dllimport) usage.

Using _declspec(dllexport) and _declspec(dllimport) on Data

In the case of data, using _declspec(dllimport) is a convenience item that removes a layer of indirection. When you import data from a DLL, you still have to go through the import address table. In the Win32 days before _declspec(dllimport), this meant you had to remember to do an extra level of indirection when accessing data exported from the DLL: You would then export the data in your .DEF file: and access it outside the DLL: When you mark the data as __declspec(dllimport), the compiler automatically generates the indirection code for you. You no longer have to worry about the steps above. As stated previously, do not use _declspec(dllimport) declaration on the data when building the DLL. Functions within the DLL will not use the Import Address Table to access the data object. Therefore, you will not have the extra level of indirection present.

To export the data automatically from the DLL, use this declaration:

Using a .DEF File

If you choose to use __declspec(dllimport) along with a .DEF file, you should change the .DEF file to use DATA in place of CONSTANT to reduce the likelihood that incorrect coding will cause a problem: The following chart shows why: Using _declspec(dllimport)and CONSTANT lists both the __imp_ version and the undecorated name in the .LIB DLL import library that is created to allow explicit linking. Using _declspec(dllimport) and DATA lists just the __imp_ version of the name.

If you use CONSTANT, either of the following code constructs could be used to access the ulDataInDll: However, if you use DATA in your .DEF file, only code compiled with the following definition can access the variable ulDataInDll: Using CONSTANT is more risky because if you forget to use the extra level of indirection, you could potentially access the Import Address Table's pointer to the variable -- not the variable itself. This type of problem can often manifest as an Access Violation because the Import Address Table is currently made read-only by the Microsoft compiler and linkers.

The Current Visual C++ linker issues a warning if it sees CONSTANT in the .DEF file to account for this case. The only real reason to use CONSTANT is if you can't recompile some object file where the header file didn't list dllimport on the prototype.

The Visual C++ Books Online provide a substantial amount of documentation on the dllexport and dllimport storage-class attributes. This includes "The dllexport and dllimport Attributes" and the "Using dllimport and dllexport in C++" topics in the "Microsoft-Specific Modifiers" chapter of the C++ Language Reference, and the "Exporting Symbols" topics in the "Creating DLLs for Win32" chapter of the Programming Techniques reference. For a thorough listing related topics, search the Books Online for "dllimport" or "dllexport".

For more information, please see the following articles in the Microsoft Knowledge Base: