[PATCH] [AArch64] Implement Clang CLI interface proposal about "-march".

Eric Christopher echristo at gmail.com
Tue Jul 15 14:49:31 PDT 2014

On Mon, Jul 14, 2014 at 8:05 AM, Kristof Beyls <kristof.beyls at arm.com> wrote:
> Thanks for the feedback, Eric.
> I've thought further about the potential deprecation of -mcpu.
> TBH, I haven't seen any good reason for deprecating -mcpu.  I've
> looked through the discussion threads on this topic on the LLVM
> mailing lists and tried to find discussions on the gcc side about
> deprecating -mcpu. I haven't found any strong arguments to
> deprecate -mcpu for the AArch64 architecture.

Probably not though it shouldn't have been added in the first place :\

> Therefore, I don't see a good reason for implementing
> -march/-mcpu/-mtune option behaviour for AArch64 any other way
> than gcc does it.

I can live with this logic. I was hoping you'd change both places and
get rid of -mcpu. :)


> Thanks,
> Kristof
>> -----Original Message-----
>> From: Eric Christopher [mailto:echristo at gmail.com]
>> Sent: 09 July 2014 19:49
>> To: Kristof Beyls
>> Cc: Kevin Qin; Clang Commits; kanheim at a-bix.com; LLVM Commits;
>> reviews+D4346+public+5fab7236f933ff44 at reviews.llvm.org
>> Subject: Re: [PATCH] [AArch64] Implement Clang CLI interface proposal
>> about "-march".
>> > * On deprecating -mcpu: I think that should be left to a follow-on
>> >   patch and probably needs some further discussing, as I'm not
>> >   fully convinced it's the right thing to do. If we did deprecate
>> >   -mcpu, I do think that we would have to start accepting cpu
>> >   names to the -march flag, as a shorthand for "the architecture
>> >   variant as implemented by that CPU". For example, -march cortex-a57
>> >   would be equivalent to specifying -march armv8-a+fp+simd+crypto+crc.
>> >   We would also lose command line compatibility with gcc, which
>> >   would make it harder for people using gcc to start trying out clang.
>> Let's just have -march=cyclone/cortex-a57 work right from the start and
>> we won't have to worry about it.
>> Thanks.
>> -eric

More information about the llvm-commits mailing list