MIPS: Fix [de]serialize problem of root objects.

MIPS: Fix [de]serialize problem of root objects.

BUG
TEST=mjsunit/deserialize-reference
R=yangguo@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=23487

[balazs.kilvady, 2014-08-28 09:43:13]

balazs.kilvady@imgtec.com changed reviewers:
+ akos.palfi@imgtec.com, danno@chromium.org, dusan.milosavljevic@imgtec.com,
gergely.kis@imgtec.com, machenbach@chromium.org, paul.lind@imgtec.com,
rmcilroy@chromium.org, yangguo@chromium.org

[balazs.kilvady, 2014-08-28 09:45:19]

PTAL. Could a v8 team member review?

[Yang, 2014-08-28 09:48:54]

On 2014/08/28 09:45:19, balazs.kilvady wrote:
> PTAL. Could a v8 team member review?

Why is it necessary for MIPS? Would the same also apply for ARM?

[balazs.kilvady, 2014-08-28 09:54:43]

On 2014/08/28 09:48:54, Yang wrote:
> On 2014/08/28 09:45:19, balazs.kilvady wrote:
> > PTAL. Could a v8 team member review?
> 
> Why is it necessary for MIPS? Would the same also apply for ARM?
Thank you for the quick review.

The same has done for arm/V8_OOL_CONSTANT_POOL arches in r r23242. Wee need this
change to fix failing mjsunit/deserialize-reference tests. To fix that test we
need serialize root_ objects by their root index and deserialize them with a
CASE_STATEMENT(kRootArray, kFromCode, kStartOfObject, 0) case branch.

[Yang, 2014-08-28 10:38:15]

On 2014/08/28 09:54:43, balazs.kilvady wrote:
> On 2014/08/28 09:48:54, Yang wrote:
> > On 2014/08/28 09:45:19, balazs.kilvady wrote:
> > > PTAL. Could a v8 team member review?
> > 
> > Why is it necessary for MIPS? Would the same also apply for ARM?
> Thank you for the quick review.
> 
> The same has done for arm/V8_OOL_CONSTANT_POOL arches in r r23242. Wee need
this
> change to fix failing mjsunit/deserialize-reference tests. To fix that test we
> need serialize root_ objects by their root index and deserialize them with a
> CASE_STATEMENT(kRootArray, kFromCode, kStartOfObject, 0) case branch.

I was just wondering whether it's better to just set V8_OOL_CONSTANT_POOL for
MIPS platforms.

[rmcilroy, 2014-08-28 10:47:08]

On 2014/08/28 10:38:15, Yang wrote:
> On 2014/08/28 09:54:43, balazs.kilvady wrote:
> > On 2014/08/28 09:48:54, Yang wrote:
> > > On 2014/08/28 09:45:19, balazs.kilvady wrote:
> > > > PTAL. Could a v8 team member review?
> > > 
> > > Why is it necessary for MIPS? Would the same also apply for ARM?
> > Thank you for the quick review.
> > 
> > The same has done for arm/V8_OOL_CONSTANT_POOL arches in r r23242. Wee need
> this
> > change to fix failing mjsunit/deserialize-reference tests. To fix that test
we
> > need serialize root_ objects by their root index and deserialize them with a
> > CASE_STATEMENT(kRootArray, kFromCode, kStartOfObject, 0) case branch.
> 
> I was just wondering whether it's better to just set V8_OOL_CONSTANT_POOL for
> MIPS platforms.

There is no ool constant pool support for MIPS yet AFAIK.

[Yang, 2014-08-28 10:50:17]

On 2014/08/28 10:47:08, rmcilroy wrote:
> On 2014/08/28 10:38:15, Yang wrote:
> > On 2014/08/28 09:54:43, balazs.kilvady wrote:
> > > On 2014/08/28 09:48:54, Yang wrote:
> > > > On 2014/08/28 09:45:19, balazs.kilvady wrote:
> > > > > PTAL. Could a v8 team member review?
> > > > 
> > > > Why is it necessary for MIPS? Would the same also apply for ARM?
> > > Thank you for the quick review.
> > > 
> > > The same has done for arm/V8_OOL_CONSTANT_POOL arches in r r23242. Wee
need
> > this
> > > change to fix failing mjsunit/deserialize-reference tests. To fix that
test
> we
> > > need serialize root_ objects by their root index and deserialize them with
a
> > > CASE_STATEMENT(kRootArray, kFromCode, kStartOfObject, 0) case branch.
> > 
> > I was just wondering whether it's better to just set V8_OOL_CONSTANT_POOL
for
> > MIPS platforms.
> 
> There is no ool constant pool support for MIPS yet AFAIK.

Hm... I don't think I understand the problem here.

The fix r23242 was for OOL constant pool on ARM. If MIPS doesn't have that, why
does it need thses case statements? Is there anything on MIPS regarding
referencing objects in the root array from code, that's different than all other
platforms?

[balazs.kilvady, 2014-08-28 11:14:04]

On 2014/08/28 10:50:17, Yang wrote:
> On 2014/08/28 10:47:08, rmcilroy wrote:
> > On 2014/08/28 10:38:15, Yang wrote:
> > > On 2014/08/28 09:54:43, balazs.kilvady wrote:
> > > > On 2014/08/28 09:48:54, Yang wrote:
> > > > > On 2014/08/28 09:45:19, balazs.kilvady wrote:
> > > > > > PTAL. Could a v8 team member review?
> > > > > 
> > > > > Why is it necessary for MIPS? Would the same also apply for ARM?
> > > > Thank you for the quick review.
> > > > 
> > > > The same has done for arm/V8_OOL_CONSTANT_POOL arches in r r23242. Wee
> need
> > > this
> > > > change to fix failing mjsunit/deserialize-reference tests. To fix that
> test
> > we
> > > > need serialize root_ objects by their root index and deserialize them
with
> a
> > > > CASE_STATEMENT(kRootArray, kFromCode, kStartOfObject, 0) case branch.
> > > 
> > > I was just wondering whether it's better to just set V8_OOL_CONSTANT_POOL
> for
> > > MIPS platforms.
> > 
> > There is no ool constant pool support for MIPS yet AFAIK.
> 
> Hm... I don't think I understand the problem here.
> 
> The fix r23242 was for OOL constant pool on ARM. If MIPS doesn't have that,
why
> does it need thses case statements? Is there anything on MIPS regarding
> referencing objects in the root array from code, that's different than all
other
> platforms?
We don't have a no ool constant pool support for MIPS. And
nothing special for MIPS. With this patch we wold like to remove the special
handling
for MIPS, the code block in serialize.cc at line 1305. If that block is removed
then
objects of root_ are serialized by their index and without removing that block
root
objects are serialized as copies of general objects.

But if we remove that block then we also need a
      CASE_STATEMENT(kRootArray, kFromCode, kStartOfObject, 0)
      CASE_BODY(kRootArray, kFromCode, kStartOfObject, 0)
case branch at deserialization (to deserialize root object references (from
code) by
their root index like on ARM).

[Yang, 2014-08-28 11:23:27]

On 2014/08/28 11:14:04, balazs.kilvady wrote:
> On 2014/08/28 10:50:17, Yang wrote:
> > On 2014/08/28 10:47:08, rmcilroy wrote:
> > > On 2014/08/28 10:38:15, Yang wrote:
> > > > On 2014/08/28 09:54:43, balazs.kilvady wrote:
> > > > > On 2014/08/28 09:48:54, Yang wrote:
> > > > > > On 2014/08/28 09:45:19, balazs.kilvady wrote:
> > > > > > > PTAL. Could a v8 team member review?
> > > > > > 
> > > > > > Why is it necessary for MIPS? Would the same also apply for ARM?
> > > > > Thank you for the quick review.
> > > > > 
> > > > > The same has done for arm/V8_OOL_CONSTANT_POOL arches in r r23242. Wee
> > need
> > > > this
> > > > > change to fix failing mjsunit/deserialize-reference tests. To fix that
> > test
> > > we
> > > > > need serialize root_ objects by their root index and deserialize them
> with
> > a
> > > > > CASE_STATEMENT(kRootArray, kFromCode, kStartOfObject, 0) case branch.
> > > > 
> > > > I was just wondering whether it's better to just set
V8_OOL_CONSTANT_POOL
> > for
> > > > MIPS platforms.
> > > 
> > > There is no ool constant pool support for MIPS yet AFAIK.
> > 
> > Hm... I don't think I understand the problem here.
> > 
> > The fix r23242 was for OOL constant pool on ARM. If MIPS doesn't have that,
> why
> > does it need thses case statements? Is there anything on MIPS regarding
> > referencing objects in the root array from code, that's different than all
> other
> > platforms?
> We don't have a no ool constant pool support for MIPS. And
> nothing special for MIPS. With this patch we wold like to remove the special
> handling
> for MIPS, the code block in serialize.cc at line 1305. If that block is
removed
> then
> objects of root_ are serialized by their index and without removing that block
> root
> objects are serialized as copies of general objects.
> 
> But if we remove that block then we also need a
>       CASE_STATEMENT(kRootArray, kFromCode, kStartOfObject, 0)
>       CASE_BODY(kRootArray, kFromCode, kStartOfObject, 0)
> case branch at deserialization (to deserialize root object references (from
> code) by
> their root index like on ARM).

alright. lgtm.

[balazs.kilvady, 2014-08-28 11:27:28]

Committed patchset #2 manually as r23487.