[LLVMdev] llvm2cpp attributes handling

[David Blaikie]

+Bill who worked on the attribute change

On Mon, Mar 25, 2013 at 9:15 AM, Harris BAKIRAS <h.bakiras at gmail.com> wrote:
> Hi everyone,
>
> We are using the llvm2cpp feature of LLVM in the VMKit project and there are
> some issues that I would like to point out.
>
> I made a tiny reproducible example here, to be clear:
>
> echo "int main() { return 0; };" > test.c && clang test.c -emit-llvm -c -o -
> | llc -march=cpp -cppgen=function -cppfor=main -o -
>
> This command is supposed to generate the C++ code to construct the main
> function in LLVM (see main_llvm2cpp.cc joined file).
> But an error occurs while generating the code (see stack trace below).
>
> It appears that an assertion is broken while treating attributes in the file
> CPPBackend.cpp.
> So I dumped the main function's attributes before treating them, here is the
> result:
>
> PAL[
>   { ~0U => nounwind uwtable "less-precise-fpmad"="false"
> "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="true"
> "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false"
> "use-soft-float"="false" }
> ]
>
> The problem is that llvm2cpp code generation supports only the attributes
> present in the AttrKind enum of the Attribute class.
> It means that the following attributes are not supported:
>     - "less-precise-fpmad"
>     - "no-frame-pointer-elim"
>     - "no-frame-pointer-elim-non-leaf"
>     - "no-infs-fp-math"
>     - "no-nans-fp-math"
>     - "unsafe-fp-math"
>     - "use-soft-float"
>
> What is extremely strange, is that those attributes are set by clang with no
> particular option.
>
> So, is the assert in CPPBackend.cpp deprecated or should those attributes be
> handled somehow in C++ code generation (which means that the previous
> attribute list should be added to the AttrKind enum of the Attribute class)?

I believe that the C++ code generation should be updated (patches
welcome - if you're relying on this functionality you may wish to
contribute to it as I believe it's not much more than a learning tool
for most other users/developers & is liable to rot fairly easily (if
it's important to you, a bot/test cases to keep it working might be
useful)) but probably not by "handling" the new attributes
specifically. I think with the new attributes API there should be a
general solution that doesn't require handling specific attributes
explicitly, though I may be wrong.

- David

>
> Cheers,
>
> --
>
> Harris Bakiras
>
>    llc: CPPBackend.cpp:521: void
> {anonymous}::CppWriter::printAttributes(const llvm::AttributeSet&, const
> string&): Assertion `!attrs.hasAttributes() && "Unhandled attribute!"'
> failed.
> 0  llc             0x00000000010edc22 llvm::sys::PrintStackTrace(_IO_FILE*)
> + 34
> 1  llc             0x00000000010ed049
> 2  libpthread.so.0 0x00007f9701addcb0
> 3  libc.so.6       0x00007f9700d2f425 gsignal + 53
> 4  libc.so.6       0x00007f9700d32b8b abort + 379
> 5  libc.so.6       0x00007f9700d280ee
> 6  libc.so.6       0x00007f9700d28192
> 7  llc             0x000000000064fed3
> 8  llc             0x000000000065335b
> 9  llc             0x0000000000659fed
> 10 llc             0x000000000065b0dc
> 11 llc             0x0000000001081230
> llvm::MPPassManager::runOnModule(llvm::Module&) + 688
> 12 llc             0x0000000001081405
> llvm::PassManagerImpl::run(llvm::Module&) + 245
> 13 llc             0x0000000000565d27
> 14 llc             0x0000000000551281 main + 353
> 15 libc.so.6       0x00007f9700d1a76d __libc_start_main + 237
> 16 llc             0x0000000000561fd5
> Stack dump:
> 0.    Program arguments: llc -march=cpp -cppgen=function -cppfor=main -o -
> 1.    Running pass 'C++ backend' on module '<stdin>'.
> Aborted (core dumped)
>
>
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
>