X87: fix the bug which is introduced by R24169.

X87: fix the bug which is introduced by R24169.

    Two more DeoptimizeIf(...) are added so the branch distance changed.
    we need to use Label::kFar.

BUG=

[chunyang.dai, 2014-09-25 08:29:09]

chunyang.dai@intel.com changed reviewers:
+ weiliang.lin@intel.com

[chunyang.dai, 2014-09-25 08:29:20]

WL.
PTAL.
thanks

[Weiliang, 2014-09-25 12:56:55]

lgtm

https://codereview.chromium.org/598353002/diff/1/src/x87/lithium-codegen-x87.cc
File src/x87/lithium-codegen-x87.cc (right):

https://codereview.chromium.org/598353002/diff/1/src/x87/lithium-codegen-x87....
src/x87/lithium-codegen-x87.cc:5308: __ jmp(&done, Label::kFar);
The default jmp is far. So you don't need to do it explicitly.
I will do it for you.

[Weiliang, 2014-09-25 12:57:58]

Committed as r24226

[Sven Panne, 2014-09-29 07:15:58]

Hmmm, I think this is still fragile: On x87 deopts emit much more code than on
other platforms, so is it really safe to do *any* near jump in DoDoubleToI?

[Weiliang, 2014-09-29 08:14:09]

On 2014/09/29 07:15:58, Sven Panne wrote:
> Hmmm, I think this is still fragile: On x87 deopts emit much more code than on
> other platforms, so is it really safe to do *any* near jump in DoDoubleToI?

Agree. It is still fragile here. We'd better to use the far jump in DoDoubleToI.
Thanks a lot.

Thanks
-Weiliang