MIPS: Fix bug in allocation stub's slow path.

MIPS: Fix bug in allocation stub's slow path.

The CL r32697 that changed the argument count of the AllocateObject
runtime function from 3 to 2 did not correctly adjust SP before
the runtime call.

BUG=dartbug.com/16873
R=regis@google.com

Committed: https://code.google.com/p/dart/source/detail?r=32735

[regis, 2014-02-18 03:27:42]

LGTM

https://codereview.chromium.org/167893006/diff/80001/tests/language/vm/regres...
File tests/language/vm/regress_16873_test.dart (right):

https://codereview.chromium.org/167893006/diff/80001/tests/language/vm/regres...
tests/language/vm/regress_16873_test.dart:6: // Regression test for slow-path
allocation in the allocation stub.
Technically speaking, --slow_path is not synonym with --no_inline_alloc.
We discussed last week the benefits of keeping the --slow_path flag. They are
not huge. We may remove the flag at some time. Any opinion?

[Florian Schneider, 2014-02-18 09:58:10]

https://codereview.chromium.org/167893006/diff/80001/tests/language/vm/regres...
File tests/language/vm/regress_16873_test.dart (right):

https://codereview.chromium.org/167893006/diff/80001/tests/language/vm/regres...
tests/language/vm/regress_16873_test.dart:6: // Regression test for slow-path
allocation in the allocation stub.
On 2014/02/18 03:27:43, regis wrote:
> Technically speaking, --slow_path is not synonym with --no_inline_alloc.
> We discussed last week the benefits of keeping the --slow_path flag. They are
> not huge. We may remove the flag at some time. Any opinion?

I'm using --no-inline-alloc here. We should have at least one flag for testing
purposes. A generic slow-path - I don't know where it should be used, so I would
be fine with removing it.

[Florian Schneider, 2014-02-18 10:10:12]

Committed patchset #3 manually as r32735 (presubmit successful).