[windows] is there a way to tell llvm to generate symbol with a leading _ ?

Clang Frontend
1

Hello,

here is the context for my question.
On Windows, mingw generates symbols with a leading _ in their name.

I have a project that combines regular C libraries compiled by mingw (runtime.a) and IR code that I generated and compile via llvm libraries (generated.a).
In my case generated.a uses variables exported by runtime.a.
And here lies the run, while those variables have the same name in the c files that make runtime.a and in the IR code that I feed llvm, once they get compiled, the former has a leading _ and the latter not.

To complicate things a little further, that project already work on macosx and linux.
So, I can’t want to rename the said variables in the IR code without recreating ripples on the platforms.
Which I do want to avoid.

What I am looking for is the switch to tell llvm assembler “hey, we’re on windows, please add the dreaded _”.
Does such thing exist?

Where can I look for it?

Cheers,
Carl.

PS: I’m posting in the correct mailing list?

2

bcc cfe-dev

llvmdev would probably be more appropriate.

We used to choose whether to use a _ prefix based on the OS, which would be mingw32 or win32. Now I believe we ask the user to put something in the IR data layout to tell us what mangling they want:
http://llvm.org/docs/LangRef.html#data-layout

You probably need to add “m:w” to your data layout string.

3

Hello,

I was recommended to post my question here.

Here is the context for my question.
On Windows, mingw generates symbols with a leading _ in their name.

I have a project that combines regular C libraries compiled by mingw (runtime.a) and IR code that I generated and compile via llvm libraries (generated.a).
In my case generated.a uses variables exported by runtime.a.
And here lies the run, while those variables have the same name in the c files that make runtime.a and in the IR code that I feed llvm, once they get compiled, the former has a leading _ and the latter not.

To complicate things a little further, that project already work on macosx and linux.
So, I can’t want to rename the said variables in the IR code without recreating ripples on the platforms.
Which I do want to avoid.

What I am looking for is the switch to tell llvm assembler “hey, we’re on windows, please add the dreaded _”.
Does such thing exist?

Where can I look for it?

I was told that I should add “m:w” to your data layout string. How so?
By using the feature parameter of Target?

createTargetMachine ( StringRef Triple,
StringRef CPU,
StringRef Features,
const TargetOptions & Options,
Reloc::Model RM = Reloc::Default,
CodeModel::Model CM = CodeModel::Default,
CodeGenOpt::Level OL = CodeGenOpt::Default
) const [inline]

Cheers,
Carl.

PS: I’m posting in the correct mailing list?

4

Hello,

I was recommended to post my question here.

...

I was told that I should add "m:w" to your data layout string. How so?
By using the feature parameter of Target?

No, if this is codegen, it should just work. Check computeDataLayout
in X86TargetMachine.cpp. The call to getManglingComponent should be
adding the "m:w".

If codegen is not involved, you need to set the datalayout in the
Module by calling setDataLayout.

Cheers,
Rafael