[Patch] Mangle C __stdcall and __fastcall in clang

[Rafael Espíndola]

A new patch is attached. I think I implemented all of Reid's suggestions.

On 9 October 2013 16:01, Reid Kleckner <rnk at google.com> wrote:
> This sounds like the right direction to me, if Anton's OK with removing the
> mangling from LLVM.
>
> -  virtual bool shouldMangleDeclName(const NamedDecl *D) = 0;
> +  bool shouldMangleDeclName(const NamedDecl *D);
> +  virtual bool shouldMangleDeclNameImpl(const NamedDecl *D) = 0;
>
>    // FIXME: consider replacing raw_ostream & with something like
> SmallString &.
> -  virtual void mangleName(const NamedDecl *D, raw_ostream &) = 0;
> +  void mangleName(const NamedDecl *D, raw_ostream &);
> +  virtual void mangleNameImpl(const NamedDecl *D, raw_ostream &) = 0;
>
> Rather than mangleNameImpl, how about mangleCXXName?  Basically, the shared
> code is C / asm mangling only, and the ABI-specific code is all C++.
>
> ....
>
> +static bool isStdOrFastCallFunc(const ASTContext &Context,
> +                                const NamedDecl *ND) {
> +  llvm::Triple Triple = Context.getTargetInfo().getTriple();
> +  if (!Triple.isOSWindows() || Triple.getArch() != llvm::Triple::x86 ||
> +      Context.getLangOpts().CPlusPlus)
> +    return false;
> +
> +  const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND);
> +  if (!FD)
> +    return false;
> +  QualType T = FD->getType();
> +  const AttributedType *AT = dyn_cast<AttributedType>(T);
> +  if (!AT)
> +    return false;
> +  AttributedType::Kind Kind = AT->getAttrKind();
> +  return Kind == AttributedType::attr_stdcall ||
> +         Kind == AttributedType::attr_fastcall;
> +}
>
> Don't look for the attributed type, it won't be there with -mrtd, for
> example.  That'd be a good test, btw.  You can pull the true CC directly
> from FunctionType::getCallConv().
>
> ...
>
> +    unsigned ArgWords = 0;
> +    for (FunctionProtoType::arg_type_iterator Arg =
> Proto->arg_type_begin(),
> +                                              ArgEnd =
> Proto->arg_type_end();
> +         Arg != ArgEnd; ++Arg) {
> +      QualType AT = *Arg;
> +      // Size should be aligned to DWORD boundary
> +      ArgWords += (ASTContext.getTypeSize(AT) + 31) / 32;
>
> Grumble grumble RoundUpToAlignment().
>
> +    }
> +    Out << 4 * ArgWords;
>
> ...
>
> diff --git a/test/CodeGen/mangle-ms.c b/test/CodeGen/mangle-ms.c
> new file mode 100644
> index 0000000..6706492
> --- /dev/null
> +++ b/test/CodeGen/mangle-ms.c
>
> mangle-windows.c seems more accurate.
>
>
>
> On Wed, Oct 9, 2013 at 12:45 PM, Rafael Espíndola
> <rafael.espindola at gmail.com> wrote:
>>
>> On 9 October 2013 14:48, Anton Korobeynikov <anton at korobeynikov.info>
>> wrote:
>> >> Doesn't mingw need these manglings to match the Windows C ABI?
>> > Correct, it does need it.
>>
>> Good catch. The attached patch handles mingw too.
>>
>> > Another problem, which is much more severe, is that moving mangling in
>> > clang will force non-clang users to handle the mangling by themselves.
>>
>> Judging from
>> http://lists.cs.uiuc.edu/pipermail/llvmdev/2013-February/059626.html
>>
>> that would be an welcome change. The mangling consists of adding how
>> many bytes are used for arguments. If a frontend can produce a call at
>> all it has that information.
>>
>> In any case. This patch changes only clang. Even if we decide to not
>> change llvm, I think this patch is an improvement as it makes it
>> possible for us to test that clang produces the desired name for C
>> code (as we do for C++ already).
>>
>> > Also, keep in mind, that gcc on linux does support stdcall / fastcall
>> > calling conventions, but does not mangle the stuff.
>>
>> That is also handled correctly.
>>
>> Cheers,
>> Rafael
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: t.patch
Type: application/octet-stream
Size: 13429 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20131015/2e9612b2/attachment.obj>