Update Integer32ToSmiField, SmiToInteger32, and SmiComprare to support 31-bit SMI for x32 port

Update Integer32ToSmiField, SmiToInteger32, and SmiComprare to support 31-bit SMI for x32 port

R=verwaest@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=20635

[haitao.feng, 2014-04-02 13:36:26]

PTAL.

[Toon Verwaest, 2014-04-03 11:47:48]

Mostly LG TM, just one comment.

https://codereview.chromium.org/222133003/diff/20001/src/x64/macro-assembler-...
File src/x64/macro-assembler-x64.cc (right):

https://codereview.chromium.org/222133003/diff/20001/src/x64/macro-assembler-...
src/x64/macro-assembler-x64.cc:1112: if (SmiValuesAre32Bits()) {
We shouldn't be able to call Integer32ToSmiField on x32. Int32 obviously doesn't
fit in a smi, so this can overflow. The optimization of requiring int32 for
storing smi is only available to SmiValuesAre32Bits from within Crankshaft
already, so we should ASSERT that we don't end up in this code-path in the other
cases.

[haitao.feng, 2014-04-08 14:25:26]

Toon, thanks for the review. Sorry for the late response as I was in the Chinese
holiday vacation.

The current usages of Integer32ToSmiField (3 in ic-x64.cc and 2 in
code-stubs-x64.cc) are all dealing with JSArray::Length property. From the
context, this property is less than FixedArray::kMaxLength (128MB) or
FixedDoubleArray::kMaxLength （64MB）when Integer32ToSmiField is used. There is no
overflow check for ia32 port either.

For the 32-bit SMI load/store optimization, I uploaded another CL at
https://codereview.chromium.org/228073004/.

On 2014/04/03 11:47:48, Toon Verwaest wrote:
> Mostly LG TM, just one comment.
> 
>
https://codereview.chromium.org/222133003/diff/20001/src/x64/macro-assembler-...
> File src/x64/macro-assembler-x64.cc (right):
> 
>
https://codereview.chromium.org/222133003/diff/20001/src/x64/macro-assembler-...
> src/x64/macro-assembler-x64.cc:1112: if (SmiValuesAre32Bits()) {
> We shouldn't be able to call Integer32ToSmiField on x32. Int32 obviously
doesn't
> fit in a smi, so this can overflow. The optimization of requiring int32 for
> storing smi is only available to SmiValuesAre32Bits from within Crankshaft
> already, so we should ASSERT that we don't end up in this code-path in the
other
> cases.

[Toon Verwaest, 2014-04-09 14:11:56]

lgtm

[haitao.feng, 2014-04-09 15:51:20]

The CQ bit was checked by haitao.feng@intel.com

[haitao.feng, 2014-04-09 15:51:20]

The CQ bit was unchecked by haitao.feng@intel.com

[haitao.feng, 2014-04-10 01:25:50]

Committed patchset #2 manually as r20635 (presubmit successful).