MIPS: Porting r9605 to arm (elements kind conversion in generated code).

MIPS: Porting r9605 to arm (elements kind conversion in generated code).

Port r9690 (857eacf)

BUG=
TEST=

Committed: http://code.google.com/p/v8/source/detail?r=9745

[Paul Lind, 2011-10-21 05:29:02]

Yang, Danno - This submission covers the following commits from the past couple
days:

9690     Porting r9605 to arm (elements kind conversion in generated code).
9696     Bugfix for r9690.
9702     Introduce HTransitionElementsKind instruction. (non-crankshaft portion)
9704     Optimize fast element conversion in arm using batch store/loads.
9717     Fix x64 and ARM builds.
9724     Refactor elements kind conversion.

These commits are overlapping/inter-dependent. I did not think it made sense to
send them all out separately. I did commit them as individual patch-sets within
this single issue, so that you can look at the code for each independently. I
hope that helps.

There are 3 other related commits I am submitting tonight, and with all of them
landed, we will be current to r 9730 and pass all tests:

    http://codereview.chromium.org/8362024
    http://codereview.chromium.org/8363025
    http://codereview.chromium.org/8365026

thanks,

paul

[Yang, 2011-10-21 07:43:33]

LGTM, but got one question.

http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc
File src/mips/codegen-mips.cc (right):

http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc#new...
src/mips/codegen-mips.cc:79: __ mov(v0, a0);
Why is a0 moved to v0?
KeyedStoreStubCompiler::GenerateStoreFastElement that follows expects value in
a0 as well.
In the case that the generator is called from crankshaft, no value or key exist
in the first place.

http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc#new...
src/mips/codegen-mips.cc:200: __ mov(v0, a0);
Ditto.

http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc#new...
src/mips/codegen-mips.cc:311: __ mov(v0, a0);
Ditto.

[Paul Lind, 2011-10-21 16:39:49]

Hi Yang, thanks for catching those extra reg-moves. They are gone now. 

thanks,

paul

http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc
File src/mips/codegen-mips.cc (right):

http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc#new...
src/mips/codegen-mips.cc:79: __ mov(v0, a0);
On 2011/10/21 07:43:34, Yang wrote:
> Why is a0 moved to v0?
> KeyedStoreStubCompiler::GenerateStoreFastElement that follows expects value in
> a0 as well.
> In the case that the generator is called from crankshaft, no value or key
exist
> in the first place.

Yep, you are correct, it is not needed here. Removed.

This is a common pattern for mips, needed in lots of places, and we just
found/fixed a bug where this was needed in porting FastElementsConversionStub
earlier in this commit. So here I guess we errored in the other direction.

http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc#new...
src/mips/codegen-mips.cc:200: __ mov(v0, a0);
On 2011/10/21 07:43:34, Yang wrote:
> Ditto.

Removed.

http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc#new...
src/mips/codegen-mips.cc:311: __ mov(v0, a0);
On 2011/10/21 07:43:34, Yang wrote:
> Ditto.

Removed.

[Yang, 2011-10-21 18:36:39]

On 2011/10/21 16:39:49, Paul Lind wrote:
> Hi Yang, thanks for catching those extra reg-moves. They are gone now. 
> 
> thanks,
> 
> paul
> 
> http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc
> File src/mips/codegen-mips.cc (right):
> 
>
http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc#new...
> src/mips/codegen-mips.cc:79: __ mov(v0, a0);
> On 2011/10/21 07:43:34, Yang wrote:
> > Why is a0 moved to v0?
> > KeyedStoreStubCompiler::GenerateStoreFastElement that follows expects value
in
> > a0 as well.
> > In the case that the generator is called from crankshaft, no value or key
> exist
> > in the first place.
> 
> Yep, you are correct, it is not needed here. Removed.
> 
> This is a common pattern for mips, needed in lots of places, and we just
> found/fixed a bug where this was needed in porting FastElementsConversionStub
> earlier in this commit. So here I guess we errored in the other direction.
> 
>
http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc#new...
> src/mips/codegen-mips.cc:200: __ mov(v0, a0);
> On 2011/10/21 07:43:34, Yang wrote:
> > Ditto.
> 
> Removed.
> 
>
http://codereview.chromium.org/8366031/diff/1006/src/mips/codegen-mips.cc#new...
> src/mips/codegen-mips.cc:311: __ mov(v0, a0);
> On 2011/10/21 07:43:34, Yang wrote:
> > Ditto.
> 
> Removed.

LGTM, landing.