Fixed DoMulI to adopt correct operand size when the representation is Smi

Fixed DoMulI  to adopt correct operand size when the representation is Smi

BUG=

[Sven Panne, 2013-08-08 10:35:07]

Looking at the CL, this seems to be a correctness fix. Our current test suite
doesn't show any error, so please add a regression test exercising all code
paths in DoMulI you've changed. It's probably more work than the fix itself, but
it's very important to get better coverage.

[weiliang.lin2, 2013-08-08 15:05:29]

On 2013/08/08 10:35:07, Sven Panne wrote:
> Looking at the CL, this seems to be a correctness fix. Our current test suite
> doesn't show any error, so please add a regression test exercising all code
> paths in DoMulI you've changed. It's probably more work than the fix itself,
but
> it's very important to get better coverage.

There are two reasons why our current test suite doesn't show any error. The
first is as you said, maybe, current test cases don't cover all paths. The
second is that Smi representation is intentionally converted to Int 32 in HMul
instruction. I am not sure the reason. I submit a separate patch to try to
remove this intention code. Please see
https://codereview.chromium.org/22600005/. 

And I will work to add new regression test to exercise all code paths soon.

Thanks~

[weiliang.lin2, 2013-08-09 03:47:24]

On 2013/08/08 15:05:29, weiliang.lin2 wrote:
> On 2013/08/08 10:35:07, Sven Panne wrote:
> > Looking at the CL, this seems to be a correctness fix. Our current test
suite
> > doesn't show any error, so please add a regression test exercising all code
> > paths in DoMulI you've changed. It's probably more work than the fix itself,
> but
> > it's very important to get better coverage.
> 
> There are two reasons why our current test suite doesn't show any error. The
> first is as you said, maybe, current test cases don't cover all paths. The
> second is that Smi representation is intentionally converted to Int 32 in HMul
> instruction. I am not sure the reason. I submit a separate patch to try to
> remove this intention code. Please see
> https://codereview.chromium.org/22600005/. 
> 
> And I will work to add new regression test to exercise all code paths soon.
> 
> Thanks~

Add regression test. Pass of the test depends on
https://codereview.chromium.org/22600005/