Roll Clang r186332:r188423.

Roll Clang r186332:r188423.

BUG=none

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=219657

[hans, 2013-08-23 16:24:29]

Here's a new clang roll. Let's see what the bots say.

When this was tested internally, they did find a codegen bug in the SLP
vectorizer. It's very unlikely that we would hit it, but if we want to be
completely safe, we could add -fno-slp-vectorize to our build flags, or we could
wait for the next release.

[Nico, 2013-08-23 17:06:38]

On 2013/08/23 16:24:29, hans wrote:
> Here's a new clang roll. Let's see what the bots say.
> 
> When this was tested internally, they did find a codegen bug in the SLP
> vectorizer. It's very unlikely that we would hit it, but if we want to be
> completely safe, we could add -fno-slp-vectorize to our build flags, or we
could
> wait for the next release.

Do you have a link to the bug? Why do you think it's unlikely we'll hit it?

[hans, 2013-08-23 17:17:15]

On 2013/08/23 17:06:38, Nico (vacation Aug 2 - Aug 22) wrote:
> On 2013/08/23 16:24:29, hans wrote:
> > Here's a new clang roll. Let's see what the bots say.
> > 
> > When this was tested internally, they did find a codegen bug in the SLP
> > vectorizer. It's very unlikely that we would hit it, but if we want to be
> > completely safe, we could add -fno-slp-vectorize to our build flags, or we
> could
> > wait for the next release.
> 
> Do you have a link to the bug? Why do you think it's unlikely we'll hit it?

I don't think a bug was ever filed, Chandler just fixed it upstream:
http://llvm.org/viewvc/llvm-project?view=revision&revision=189007

Chandler said the bug can only hit for structs where the non-first member is a
type that's smaller than its alignment, such as long double (what tickled the
bug internally), or maybe packed bit-fields. I haven't seen any long doubles in
Chromium, and I don't think we use bit-fields either.

[Nico, 2013-08-23 17:29:29]

On 2013/08/23 17:17:15, hans wrote:
> On 2013/08/23 17:06:38, Nico (vacation Aug 2 - Aug 22) wrote:
> > On 2013/08/23 16:24:29, hans wrote:
> > > Here's a new clang roll. Let's see what the bots say.
> > > 
> > > When this was tested internally, they did find a codegen bug in the SLP
> > > vectorizer. It's very unlikely that we would hit it, but if we want to be
> > > completely safe, we could add -fno-slp-vectorize to our build flags, or we
> > could
> > > wait for the next release.
> > 
> > Do you have a link to the bug? Why do you think it's unlikely we'll hit it?
> 
> I don't think a bug was ever filed, Chandler just fixed it upstream:
> http://llvm.org/viewvc/llvm-project?view=revision&revision=189007
> 
> Chandler said the bug can only hit for structs where the non-first member is a
> type that's smaller than its alignment, such as long double (what tickled the
> bug internally), or maybe packed bit-fields. I haven't seen any long doubles
in
> Chromium, and I don't think we use bit-fields either.

I think there are a few packed bitfields in ANGLE.

But since it's early in the release cycle, I guess this is a good opportunity to
check how our test coverage is, should some code run into this bug. lgtm.

[hans, 2013-08-26 20:47:04]

The mac_asan failures are the same as I get when I try a whitespace change:
https://codereview.chromium.org/22815035

Let's take this to the CQ.

[commit-bot: I haz the power, 2013-08-26 21:08:36]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/hans@chromium.org/23172018/1

[commit-bot: I haz the power, 2013-08-27 01:09:28]

Change committed as 219657