Make the placement-new buffer in LazyInstance<Type> aligned.

Make the placement-new buffer in LazyInstance<Type> aligned.
Before, the LazyInstance::buf_ was 4-byte aligned on x64, which is wrong.

WHY?! I thought buf_ is the first member of LazyInstance?!
NO! LazyInstance inherits LazyInstanceHelper, sizeof(LIH) = 4.

Then, buf_ is given to placement new.

As a result, the LazyInstance<Type> instances are all 4-byte aligned on x64.
This may break some stuff like SSE-based optimizations assuming the instance is
8-bytes aligned (fair assumption).
Also, if Type contains a bunch of std::vector/hash_map's, their pointers occupy
two half-words and Valgrind doesn't traverse to their data, reporting a false leak.

BUG=64930
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=106763

[Timur Iskhodzhanov, 2011-10-21 16:04:08]

Hi Nico, Will,

Mind reviewing this tiny CL?
(+see description)

Re: DCHECK_ALIGNED - feel free to suggest better places to add it and/or
different names.

Thanks,
Timur Iskhodzhanov

[willchan no longer on Chromium, 2011-10-21 16:10:41]

http://codereview.chromium.org/8366041/diff/2001/base/logging.h
File base/logging.h (right):

http://codereview.chromium.org/8366041/diff/2001/base/logging.h#newcode719
base/logging.h:719: #define DCHECK_ALIGNED(ptr) \
I don't think we need to add this into logging.h. I don't think many places will
use it.

[willchan no longer on Chromium, 2011-10-21 16:11:13]

Other than the previous comment about logging.h, LGTM.

On Fri, Oct 21, 2011 at 9:10 AM, <willchan@chromium.org> wrote:

>
>
http://codereview.chromium.**org/8366041/diff/2001/base/**logging.h<http://co...
> File base/logging.h (right):
>
> http://codereview.chromium.**org/8366041/diff/2001/base/**
>
logging.h#newcode719<http://codereview.chromium.org/8366041/diff/2001/base/logging.h#newcode719>
> base/logging.h:719: #define DCHECK_ALIGNED(ptr) \
> I don't think we need to add this into logging.h. I don't think many
> places will use it.
>
>
http://codereview.chromium.**org/8366041/<http://codereview.chromium.org/8366...
>

[Timur Iskhodzhanov, 2011-10-21 16:23:49]

PTAL

[Nico, 2011-10-21 16:26:22]

lgtm

http://codereview.chromium.org/8366041/diff/2004/base/lazy_instance.h
File base/lazy_instance.h (right):

http://codereview.chromium.org/8366041/diff/2004/base/lazy_instance.h#newcode55
base/lazy_instance.h:55: DCHECK_EQ(reinterpret_cast<uintptr_t>(instance) %
sizeof(instance), 0u)
do you need this cast?

http://codereview.chromium.org/8366041/diff/2004/base/lazy_instance.h#newcode56
base/lazy_instance.h:56: << ": Bad boy, the buffer is not aligned!";
Maybe explain why this is important and what to do when this dcheck hits

[Timur Iskhodzhanov, 2011-10-21 16:40:28]

PTAL

http://codereview.chromium.org/8366041/diff/2004/base/lazy_instance.h
File base/lazy_instance.h (right):

http://codereview.chromium.org/8366041/diff/2004/base/lazy_instance.h#newcode55
base/lazy_instance.h:55: DCHECK_EQ(reinterpret_cast<uintptr_t>(instance) %
sizeof(instance), 0u)
void * ptr;
ptr % const -> doesn't compile.

Suggestions are welcome, I don't like reinterpret_cast myself

http://codereview.chromium.org/8366041/diff/2004/base/lazy_instance.h#newcode56
base/lazy_instance.h:56: << ": Bad boy, the buffer is not aligned!";
Why - Done.
What to do -> it shouldn't happen unless someone changes the LazyInstance impl.
In this case I expect one to be experienced enough to read stacktraces :)

On 2011/10/21 16:26:22, Nico wrote:
> Maybe explain why this is important and what to do when this dcheck hits

[Nico, 2011-10-21 16:41:56]

lgtm

[commit-bot: I haz the power, 2011-10-21 18:47:18]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/timurrrr@chromium.org/8366041/8002

[commit-bot: I haz the power, 2011-10-21 19:46:01]

Change committed as 106763