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

[Linaro Golin]

On 1 July 2014 04:05, Kevin Qin <kevinqindev at gmail.com> wrote:
> I only got it removed from AArch64 target. Gcc has deprecated this option
> for a while, and I don't know any tools or build systems reply on this
> option. If there's demand of this option, I can add it back and provide a
> warning on it, just like "-mcpu".

Deprecated or removed? I'd be surprised if the latter.

I'm not expecting new software to use deprecated options, but old
build systems that can be made to build on AArch64 by just changing
the triple will still use all the legacy options in hidden places
which will extend the usage of -mcpu well beyond AArch64.


> I didn't explain it clearly. Your point is totally what I did in this patch.
> I emphasize " ONLY get micro-architecture level feature enabled" is want to
> say ISA won't be changed by this option. This option is to select target CPU
> to optimize for, including enabling micro-architecture level feature,
> choosing MI scheduler and triggering any optimizations specific for target.

Oh, I see. My bad.


> In my patch, the difference between "-mcpu" and "-mtune" is that, "-mcpu"
> will enable all ISAs which target CPU supports, while "-mtune" won't do
> this.

This is correct.


> And "-mcpu" can accept extra feature modifiers to make a change, but
> "-mtune" accepts CPU name only.

This is new, and since we're trying to discourage the usage of -mcpu,
a little in the wrong direction. We shouldn't make it *more*
convenient for people to use -mcpu...


> So "-mcpu" is an shortcut of "-march" and "-tune".

Which seems to be the case in GCC, too.


> Keeping this option alive in clang is because it's still alive in
> gcc, and may still be used in many projects.  An warning is added to
> discourage use of this option.

I agree with this, but if GCC doesn't have the features in -mcpu, I
don't think we should add it, since the only reason we do is for
compatibility with legacy GCC options.


> They don't need to complain
> typing too much as splitting "-mcpu" into "-march" and "-mtune" because they
> can use "-mtune" only.

Maybe they do... :)

The reason why you need -mcpu is (AFAIK) *only* because you haven't
specified your arch correctly. So, if you specify "-march arm", you'll
need to specify your CPU to generate anything other than ARMv4 code,
since -mtune won't help. If you specify "armv7", you'll need "-mcpu
cortex-m4" to get M-class, and so on. But if you choose "armv7a", you
only need -mtune to make it faster on A9, or "armv7ave" to mean
virt+hdiv, than it's obvious A15-class, which you'll need -mtune to
make it faster on A15 and not A7.

All in all, -mcpu is redundant if the arch is correct. I believe GCC
understands quite well all the ARM arch strings, and Clang/LLVM should
if it doesn't.

And with the features in -march (armv7a+noneon), it'd be even less
important to have a "-mcpu tegra3" option.

Makes sense?

cheers,
--renato