MIPS: [builtins] Fix MathMaxMin.

MIPS: [builtins] Fix MathMaxMin.

Port b86ac0e05a1e4a34944bdc3c0c6702945387b431

Original commit message:
Both of these were broken in different ways:
* On arm, the loop counter was passed as argc on the stack.
* On arm64, we passed argc + 1 instead of argc.

The result in both cases was an incorrect receiver for the builtin frame
when generating stack traces.

BUG=v8:4815
Committed: https://crrev.com/e65035f2572144d317e1559ea69a054317649e36
Cr-Commit-Position: refs/heads/master@{#37481}

[balazs.kilvady, 2016-07-01 11:50:59]

balazs.kilvady@imgtec.com changed reviewers:
+ akos.palfi@imgtec.com, bmeurer@chromium.org, ivica.bogosavljevic@imgtec.com,
jgruber@chromium.org

[ivica.bogosavljevic, 2016-07-01 12:40:51]

https://codereview.chromium.org/2120463002/diff/1/src/mips/builtins-mips.cc
File src/mips/builtins-mips.cc (right):

https://codereview.chromium.org/2120463002/diff/1/src/mips/builtins-mips.cc#n...
src/mips/builtins-mips.cc:153: __ Branch(&done_loop, lt, a3, Operand(zero_reg));
You could fill delay slot here, the next instruction writes to at, so if jump is
taken, it doesn't matter, otherwise you save one nop :)

[balazs.kilvady, 2016-07-01 13:17:26]

https://codereview.chromium.org/2120463002/diff/1/src/mips/builtins-mips.cc
File src/mips/builtins-mips.cc (right):

https://codereview.chromium.org/2120463002/diff/1/src/mips/builtins-mips.cc#n...
src/mips/builtins-mips.cc:153: __ Branch(&done_loop, lt, a3, Operand(zero_reg));
On 2016/07/01 12:40:50, ivica.bogosavljevic wrote:
> You could fill delay slot here, the next instruction writes to at, so if jump
is
> taken, it doesn't matter, otherwise you save one nop :)

Since we support r6 I try to avoid delay slot usage in conditional branches as
with them we lose the chance for an r6 compact branch optimization. Especially
when the result of the previous op is the input of the branch condition.

[akos.palfi.imgtec, 2016-07-01 14:38:51]

LGTM. I think we should not fill the delay slot, since it breaks the compact
branch optimization, but it is up to you Balazs. :)

[balazs.kilvady, 2016-07-01 14:40:04]

The CQ bit was checked by balazs.kilvady@imgtec.com

[commit-bot: I haz the power, 2016-07-01 14:40:08]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-07-01 15:11:26]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2016-07-01 15:13:07]

Description was changed from

==========
MIPS: [builtins] Fix MathMaxMin.

Port b86ac0e05a1e4a34944bdc3c0c6702945387b431

Original commit message:
Both of these were broken in different ways:
* On arm, the loop counter was passed as argc on the stack.
* On arm64, we passed argc + 1 instead of argc.

The result in both cases was an incorrect receiver for the builtin frame
when generating stack traces.

BUG=v8:4815
==========

to

==========
MIPS: [builtins] Fix MathMaxMin.

Port b86ac0e05a1e4a34944bdc3c0c6702945387b431

Original commit message:
Both of these were broken in different ways:
* On arm, the loop counter was passed as argc on the stack.
* On arm64, we passed argc + 1 instead of argc.

The result in both cases was an incorrect receiver for the builtin frame
when generating stack traces.

BUG=v8:4815

Committed: https://crrev.com/e65035f2572144d317e1559ea69a054317649e36
Cr-Commit-Position: refs/heads/master@{#37481}
==========

[commit-bot: I haz the power, 2016-07-01 15:13:09]

Patchset 1 (id:??) landed as
https://crrev.com/e65035f2572144d317e1559ea69a054317649e36
Cr-Commit-Position: refs/heads/master@{#37481}