[cfe-dev] Clang problems with Solaris and SPARC

[James Molloy]

Hi,

[Forgot to reply-all]

Looking at Tools.cpp:AddSPARCTargetArgs(), it looks like the softfloat
behaviour has been copy-pasted from another architecture's TargetArgs
(notice the switch only has a default case?).

I don't know anything about SPARC so would not be able to say what the
default behaviour should be - perhaps someone else could weigh in on that
point?

Cheers,

James

On 25 May 2011 16:28, Witold Sowa <witold.sowa at gmail.com> wrote:

> Hi James,
>
> Your fix is working for me. Thanks a lot. The new issue is that now clang
> displays following message:
> $ ./clang++ ~/q.cpp -o ~/q
> clang-2: warning: unknown platform, assuming -mfloat-abi=soft
> '+soft-float' is not a recognized feature for this target (ignoring
> feature)
>
> I would assume that if soft-float is not recognized for this target (ie.
> sparc) a hardware floating point operations should be used in first place
> and we shouldn't even try to use soft-float. Do you know what is the status
> of sparc target support? Does it support floating point operations or not?
>
> Cheers!
> Witek
>
>
>
> 2011/5/25 James Molloy <james at jamesmolloy.co.uk>
>
>> Hi Witold,
>>
>> This appears to be a bug in Clang. The offending code is in
>> lib/Driver/Tools.cpp:
>>
>>   if (FloatABI == "soft") {
>>     // Floating point operations and argument passing are soft.
>>     //
>>     // FIXME: This changes CPP defines, we need -target-soft-float.
>>     CmdArgs.push_back("-msoft-float");
>>     CmdArgs.push_back("soft");
>>     CmdArgs.push_back("-target-feature");
>>     CmdArgs.push_back("+soft-float");
>>   } else {
>>
>>
>> but –msoft-float is defined in CC1Options.td  as:
>>
>> def msoft_float : Flag<"-msoft-float">,
>>   HelpText<"Use software floating point">;
>>
>>
>> i.e. it doesn’t take an argument. So Clang (as cc1) is interpreting the
>> extra ‘soft’ argument as an input file.
>>
>> Fix is to remove the line “CmdArgs.push_back("soft");” as far as I can
>> tell?
>>
>> Cheers,
>>
>> James
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20110525/10a6d5e3/attachment.html>