Add support for sub-architecture when building for iOS

Add support for sub-architecture when building for iOS

Since iOS 7, the iPhone 5s, the iPad Air and the iPad Mini Retina
supports 64-bits binaries. This requires building a fat binary with
support for both armv7 and arm64 (and optionally armv7s).

In an effort to prepare building such a fat binary for Chrome on
iOS, introduce a "target_subarch" variable that control the variant
we want to build (32-bit only, 64-bit only -- mainly for debugging,
fat binary).

This require introducing way to use the clang compiler shipped with
Xcode, as the clang compiler shipped with Chromium code does not
support "arm64" architecture yet (Apple has not yet opensource the
implementation).

BUG=341453, 314720
R=mark@chromium.org

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=251083

[sdefresne, 2014-02-10 14:42:10]

Please take a look.

[Mark Mentovai, 2014-02-10 14:50:02]

https://codereview.chromium.org/156473005/diff/1/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/156473005/diff/1/build/common.gypi#newcode4102
build/common.gypi:4102: '-Wno-deprecated-register',
Is this in here because the clang that ships with Xcode doesn’t care about this
flag?

Does the clang that ships with Xcode complain when it sees this flag?

https://codereview.chromium.org/156473005/diff/1/build/common.gypi#newcode4830
build/common.gypi:4830: ['clang==1 and clang_xcode==0 and OS!="win"', {
Seeing clang_xcode here makes it seem like that’s relevant, but this block is
effective on any non-Windows ninja platform. Since clang_xcode is only effective
on OS=="mac" and OS=="ios", I think the condition should be revised to reflect
that:

clang==1 and ((OS!="mac" and OS!="ios") or clang_xcode==0) and OS!="win"

https://codereview.chromium.org/156473005/diff/1/build/common.gypi#newcode4866
build/common.gypi:4866: ['use_goma==1 and clang_xcode==0 and '
Same.

[sdefresne, 2014-02-10 15:04:09]

Please take another look.

https://codereview.chromium.org/156473005/diff/1/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/156473005/diff/1/build/common.gypi#newcode4102
build/common.gypi:4102: '-Wno-deprecated-register',
On 2014/02/10 14:50:02, Mark Mentovai wrote:
> Is this in here because the clang that ships with Xcode doesn’t care about
this
> flag?
> 
> Does the clang that ships with Xcode complain when it sees this flag?

Yes, it does complain:

$ clang++ -Wno-deprecated-register empty.mm 
warning: unknown warning option '-Wno-deprecated-register'; did you mean
'-Wno-deprecated'? [-Wunknown-warning-option]

And since we configure warnings to be errors for Chromium code, this cause the
compilation to fail. But the clang in Xcode does not yet consider the register
keyword as deprecated, so the flag is not required.

https://codereview.chromium.org/156473005/diff/1/build/common.gypi#newcode4830
build/common.gypi:4830: ['clang==1 and clang_xcode==0 and OS!="win"', {
On 2014/02/10 14:50:02, Mark Mentovai wrote:
> Seeing clang_xcode here makes it seem like that’s relevant, but this block is
> effective on any non-Windows ninja platform. Since clang_xcode is only
effective
> on OS=="mac" and OS=="ios", I think the condition should be revised to reflect
> that:
> 
> clang==1 and ((OS!="mac" and OS!="ios") or clang_xcode==0) and OS!="win"

Done.

https://codereview.chromium.org/156473005/diff/1/build/common.gypi#newcode4866
build/common.gypi:4866: ['use_goma==1 and clang_xcode==0 and '
On 2014/02/10 14:50:02, Mark Mentovai wrote:
> Same.

Done.

[Mark Mentovai, 2014-02-10 15:05:17]

https://codereview.chromium.org/156473005/diff/1/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/156473005/diff/1/build/common.gypi#newcode4102
build/common.gypi:4102: '-Wno-deprecated-register',
On 2014/02/10 15:04:09, sdefresne wrote:
> On 2014/02/10 14:50:02, Mark Mentovai wrote:
> > Is this in here because the clang that ships with Xcode doesn’t care about
> this
> > flag?
> > 
> > Does the clang that ships with Xcode complain when it sees this flag?
> 
> Yes, it does complain:
> 
> $ clang++ -Wno-deprecated-register empty.mm 
> warning: unknown warning option '-Wno-deprecated-register'; did you mean
> '-Wno-deprecated'? [-Wunknown-warning-option]
> 
> And since we configure warnings to be errors for Chromium code, this cause the
> compilation to fail. But the clang in Xcode does not yet consider the register
> keyword as deprecated, so the flag is not required.

Then you should add a quick note to the comment saying why you’re disabling the
warning for trunk clang but not Xcode clang.

[sdefresne, 2014-02-10 15:10:54]

On 2014/02/10 15:05:17, Mark Mentovai wrote:
> https://codereview.chromium.org/156473005/diff/1/build/common.gypi
> File build/common.gypi (right):
> 
> https://codereview.chromium.org/156473005/diff/1/build/common.gypi#newcode4102
> build/common.gypi:4102: '-Wno-deprecated-register',
> On 2014/02/10 15:04:09, sdefresne wrote:
> > On 2014/02/10 14:50:02, Mark Mentovai wrote:
> > > Is this in here because the clang that ships with Xcode doesn’t care about
> > this
> > > flag?
> > > 
> > > Does the clang that ships with Xcode complain when it sees this flag?
> > 
> > Yes, it does complain:
> > 
> > $ clang++ -Wno-deprecated-register empty.mm 
> > warning: unknown warning option '-Wno-deprecated-register'; did you mean
> > '-Wno-deprecated'? [-Wunknown-warning-option]
> > 
> > And since we configure warnings to be errors for Chromium code, this cause
the
> > compilation to fail. But the clang in Xcode does not yet consider the
register
> > keyword as deprecated, so the flag is not required.
> 
> Then you should add a quick note to the comment saying why you’re disabling
the
> warning for trunk clang but not Xcode clang.

I did add a comment with my second patch set. Did you miss it?

[Mark Mentovai, 2014-02-10 15:20:23]

Yeah, missed it.

The iOS bots on patch set 1 had build problems. Looks like something’s not quite
right with the NEON extensions in Skia.

https://codereview.chromium.org/156473005/diff/80001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/156473005/diff/80001/build/common.gypi#newcod...
build/common.gypi:4105: # this flag, so don't enable it of "clang_xcode" is "1".
of->if

[sdefresne, 2014-02-10 18:06:33]

I've fixed the compilation on simulator by not setting the "target_arch"
to "arm" (and thus neon is not enabled when building simulator). I've
created a new bug to try to enable neon when building skia for device
(would probably involve adding a new macro and some auto-detection
in code when building for iOS).

Please take another look (and thank you for the amazingly fast reviews
that you already did).

https://codereview.chromium.org/156473005/diff/80001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/156473005/diff/80001/build/common.gypi#newcod...
build/common.gypi:4105: # this flag, so don't enable it of "clang_xcode" is "1".
On 2014/02/10 15:20:23, Mark Mentovai wrote:
> of->if

Done.

[Mark Mentovai, 2014-02-10 18:42:41]

LGTM. You’re welcome.

[sdefresne, 2014-02-13 14:53:15]

The CQ bit was checked by sdefresne@chromium.org

[commit-bot: I haz the power, 2014-02-13 14:53:35]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/sdefresne@chromium.org/156473005/200001

[commit-bot: I haz the power, 2014-02-13 18:17:36]

Change committed as 251083