Fix gc on MIPS (live registers were spilled in reversed order).

Fix gc on MIPS (live registers were spilled in reversed order).
Add comments on all platforms.

R=koda@google.com

Committed: https://code.google.com/p/dart/source/detail?r=42125

[regis, 2014-12-05 00:33:15]

regis@google.com changed reviewers:
+ koda@google.com, zra@google.com

[koda, 2014-12-05 00:46:43]

Great find!

Would it be possible to write a regression test that reliably triggers the bug
by using the --gc_at_instance_allocation flag?

https://codereview.chromium.org/785443002/diff/1/runtime/vm/flow_graph_compil...
File runtime/vm/flow_graph_compiler_mips.cc (right):

https://codereview.chromium.org/785443002/diff/1/runtime/vm/flow_graph_compil...
runtime/vm/flow_graph_compiler_mips.cc:1508: }
ASSERT(offset == 0)?

https://codereview.chromium.org/785443002/diff/1/runtime/vm/flow_graph_compil...
runtime/vm/flow_graph_compiler_mips.cc:1529: __ addiu(SP, SP,
Immediate(register_count * kWordSize));
ASSERT(offset == 0)?

[regis, 2014-12-05 00:56:11]

Regression test? Possible, maybe, but not easy. It would be difficult to
convince the optimizer to allocate registers for both objects and non-objects in
a pattern that would reliably trigger this issue. Considering how long this bug
survived undetected, I'd rather not try. Let's hope that the added comments will
prevent a regression.

Thanks,
Regis

https://codereview.chromium.org/785443002/diff/1/runtime/vm/flow_graph_compil...
File runtime/vm/flow_graph_compiler_mips.cc (right):

https://codereview.chromium.org/785443002/diff/1/runtime/vm/flow_graph_compil...
runtime/vm/flow_graph_compiler_mips.cc:1508: }
On 2014/12/05 00:46:43, koda wrote:
> ASSERT(offset == 0)?

Done.

https://codereview.chromium.org/785443002/diff/1/runtime/vm/flow_graph_compil...
runtime/vm/flow_graph_compiler_mips.cc:1529: __ addiu(SP, SP,
Immediate(register_count * kWordSize));
On 2014/12/05 00:46:43, koda wrote:
> ASSERT(offset == 0)?

Done.

[koda, 2014-12-05 01:04:43]

lgtm

[regis, 2014-12-05 01:07:13]

Committed patchset #2 (id:20001) manually as r42125 (presubmit successful).