[llvm-commits] [PATCH] Add float abi CL option for lli

Alex Rosenberg alexr at leftfield.org
Thu Oct 11 10:02:19 PDT 2012


Doh. You're right, we split this when it gets past cc1. LGTM.

Sent from my iPad

On Oct 11, 2012, at 8:46 AM, Amara Emerson <Amara.Emerson at arm.com> wrote:

> This will just cause unnecessary confusion. lli and llc are, fundamentally, similar tools. llc doesn't accept -float-abi=softfp. Not to mention the fact that softfp isn't actually an ABI, despite its implications, so to make it an option for -float-abi would be a misnomer.
> 
> Anyone have any objections to this patch?
> 
> Amara
> 
> -----Original Message-----
> From: Alex Rosenberg [mailto:alexr at leftfield.org] 
> Sent: 11 October 2012 16:40
> To: Amara Emerson
> Cc: <llvm-commits at cs.uiuc.edu>
> Subject: Re: [llvm-commits] [PATCH] Add float abi CL option for lli
> 
> While I agree with this addition, I meant that -float-abi=softfp should be accepted.
> 
> Sent from my iPad
> 
> On Oct 11, 2012, at 4:28 AM, Amara Emerson <Amara.Emerson at arm.com> wrote:
> 
>> I see what you mean. To keep consistent with the rest of LLVM this does make sense.
>> 
>> New patch attached with the -soft-float option added, copied from llc.
>> 
>> Amara
>> 
>> -----Original Message-----
>> From: Alex Rosenberg [mailto:alexr at leftfield.org] 
>> Sent: 11 October 2012 05:36
>> To: Amara Emerson
>> Cc: <llvm-commits at cs.uiuc.edu>
>> Subject: Re: [llvm-commits] [PATCH] Add float abi CL option for lli
>> 
>> "softp" is a valid argument everywhere Clang and LLVM accept a "float ABI" option. I think we should be uniform in accepting it.
>> 
>> Alex
>> 
>> On Oct 10, 2012, at 9:28 AM, Amara Emerson wrote:
>> 
>>> Softfp isn't an ABI variant, it signals that software FP library calls should be used instead of hardware FP instructions. In llc, the -soft-float option *implies* the use of soft float ABI but it's still a distinct feature.
>>> 
>>> Are you asking why a soft-float lli option wasn't included in this patch? If so, our testing infrastructure is transitioning to hard float ABI so this is the minimal option needed in order to work, we don't need softfp at this moment in time.
>>> 
>>> Amara
>>> 
>>> -----Original Message-----
>>> From: Alex Rosenberg [mailto:alexr at leftfield.org] 
>>> Sent: 10 October 2012 17:09
>>> To: Amara Emerson
>>> Cc: <llvm-commits at cs.uiuc.edu>
>>> Subject: Re: [llvm-commits] [PATCH] Add float abi CL option for lli
>>> 
>>> What about the "softfp" variant?
>>> 
>>> Sent from my iPad
>>> 
>>> On Oct 8, 2012, at 8:32 AM, "Amara Emerson" <amara.emerson at arm.com> wrote:
>>> 
>>>> Hi,
>>>> 
>>>> lli currently uses the default target options for most things, including the
>>>> float ABI which defaults to soft. On hard float systems, this can cause
>>>> silent faults when the JITed code tries to call library functions built for
>>>> hard float ABI.
>>>> 
>>>> The attached patch allows specifying the float abi to lli in exactly the
>>>> same way as llc does it, via a -float-abi=[hard|soft|default] option.
>>>> 
>>>> Amara
>>>> <lli-specify-floatabi-2.patch>
>>>> _______________________________________________
>>>> llvm-commits mailing list
>>>> llvm-commits at cs.uiuc.edu
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> 
>> <lli-specify-floatabi-softfloat.patch>
> 
> 




More information about the llvm-commits mailing list