Fix/unify root handling in code generation.

Fix/unify root handling in code generation.

R=danno@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=16132

[danno, 2013-08-09 13:36:03]

What about x64?

https://codereview.chromium.org/22670004/diff/1/src/ia32/macro-assembler-ia32.cc
File src/ia32/macro-assembler-ia32.cc (right):

https://codereview.chromium.org/22670004/diff/1/src/ia32/macro-assembler-ia32...
src/ia32/macro-assembler-ia32.cc:68: roots_array_start));
The two moves all be done in one instruction, you might have to add other
infrastructure for that but I would hate to make this code sequence so long if
you are cleaning . Comment holds for instances below, too.

If you do that, you probably don't need a scratch register, either.

[danno, 2013-08-09 13:39:45]

lgtm after discussion offline

[Benedikt Meurer, 2013-08-09 13:41:48]

On 2013/08/09 13:36:03, danno wrote:
> What about x64?

Michi suggested to use the kRootRegister on x64, which generates code that is
both shorter and faster than the 64-bit immediate movq.

[Benedikt Meurer, 2013-08-09 13:41:55]

https://codereview.chromium.org/22670004/diff/1/src/ia32/macro-assembler-ia32.cc
File src/ia32/macro-assembler-ia32.cc (right):

https://codereview.chromium.org/22670004/diff/1/src/ia32/macro-assembler-ia32...
src/ia32/macro-assembler-ia32.cc:68: roots_array_start));
On 2013/08/09 13:36:03, danno wrote:
> The two moves all be done in one instruction, you might have to add other
> infrastructure for that but I would hate to make this code sequence so long if
> you are cleaning . Comment holds for instances below, too.
> 
> If you do that, you probably don't need a scratch register, either.

There's no way to do this in one instruction, cause it would break
serialization.

[Benedikt Meurer, 2013-08-09 13:43:53]

Committed patchset #1 manually as r16132 (presubmit successful).