ARM64: Access has_pending_message_ correctly.

ARM64: Access has_pending_message_ correctly.

This fixes accesses in ARM and ARM64; the field is a bool, with size 1,
but we were accessing it with pointer-sized loads and stores.

BUG=
R=ulan@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=20771

[jbramley, 2014-04-09 13:41:12]

The other back-ends have the same problem, but I don't know the proper way to
fix them. I found an appropriate instruction for the access in x64, but I
couldn't find a MacroAssembler helper for using it with an ExternalReference. If
it's accepted, I'll raise an issue to get this change ported.

https://codereview.chromium.org/230553004/diff/1/src/arm64/full-codegen-arm64.cc
File src/arm64/full-codegen-arm64.cc (right):

https://codereview.chromium.org/230553004/diff/1/src/arm64/full-codegen-arm64...
src/arm64/full-codegen-arm64.cc:4853: STATIC_ASSERT(sizeof(bool) == 1);   //
NOLINT(runtime/sizeof)
This is defined in the ABI for both ARM and AArch64. It's also true for the
simulator host environments I tested (gcc 4.8.1 on Ubuntu 13.10). If it causes
trouble, we'll have to review how we handle accesses to fields where the size is
determined by the host environment. We might need templated Ldr/Str variants to
load host data types, for example.

[ulan, 2014-04-09 13:51:47]

https://codereview.chromium.org/230553004/diff/1/src/arm64/full-codegen-arm64.cc
File src/arm64/full-codegen-arm64.cc (right):

https://codereview.chromium.org/230553004/diff/1/src/arm64/full-codegen-arm64...
src/arm64/full-codegen-arm64.cc:4853: STATIC_ASSERT(sizeof(bool) == 1);   //
NOLINT(runtime/sizeof)
On 2014/04/09 13:41:12, jbramley wrote:
> This is defined in the ABI for both ARM and AArch64. It's also true for the
> simulator host environments I tested (gcc 4.8.1 on Ubuntu 13.10). If it causes
> trouble, we'll have to review how we handle accesses to fields where the size
is
> determined by the host environment. We might need templated Ldr/Str variants
to
> load host data types, for example.

Another solution would be to change the type of "has_pending_message" to
something platform-independent, like int8_t or int32_t. Jochen, wdyt?

[jbramley, 2014-04-09 13:53:13]

On 2014/04/09 13:51:47, ulan wrote:
>
https://codereview.chromium.org/230553004/diff/1/src/arm64/full-codegen-arm64.cc
> File src/arm64/full-codegen-arm64.cc (right):
> 
>
https://codereview.chromium.org/230553004/diff/1/src/arm64/full-codegen-arm64...
> src/arm64/full-codegen-arm64.cc:4853: STATIC_ASSERT(sizeof(bool) == 1);   //
> NOLINT(runtime/sizeof)
> On 2014/04/09 13:41:12, jbramley wrote:
> > This is defined in the ABI for both ARM and AArch64. It's also true for the
> > simulator host environments I tested (gcc 4.8.1 on Ubuntu 13.10). If it
causes
> > trouble, we'll have to review how we handle accesses to fields where the
size
> is
> > determined by the host environment. We might need templated Ldr/Str variants
> to
> > load host data types, for example.
> 
> Another solution would be to change the type of "has_pending_message" to
> something platform-independent, like int8_t or int32_t. Jochen, wdyt?

This probably isn't the only one, though. It would be nice if ExternalReferences
could have an implicit access type, but due to the way they're used, that would
be really hard to enforce.

[ulan, 2014-04-15 11:29:34]

lgtm

I talked with Jochen. Both approaches are brittle. As you  already mentioned
making compiler check is hard. Let's land this.

[jbramley, 2014-04-15 14:03:34]

Committed patchset #1 manually as r20771 (presubmit successful).