Fix bug in propagation of type information into registers....

Fix bug in propagation of type information into registers.

The number type information of results has to be also copied when
calling ToRegister with a fixed register as destination.

Also fix an unbound label and a missing CpuFeatures scope.


Committed: http://code.google.com/p/v8/source/detail?r=4155

[fschneider, 2010-03-16 12:45:46]

Small code review.

[Lasse Reichstein, 2010-03-17 07:10:39]

LGTM

http://codereview.chromium.org/987003/diff/5001/6002
File src/ia32/codegen-ia32.cc (right):

http://codereview.chromium.org/987003/diff/5001/6002#newcode6525
src/ia32/codegen-ia32.cc:6525: __ and_(operand.reg(), ~kSmiTagMask);  // Remove
inverted smi-tag.
ASSERT_EQ(0, kSmiTag).
Seems wasteful to use a 32-bit immediate to clear one bit.
How about:
  lea(operand.reg(), Operand(operand.reg(), kSmiTagMask));
  not(operand.reg());
(i.e., set it before not'ing and use lea to reduce size).
That, or use and_b (but that's a partial register write, I'm not sure that's
good for efficiency).

http://codereview.chromium.org/987003/diff/5001/6002#newcode6543
src/ia32/codegen-ia32.cc:6543: __ and_(answer.reg(), ~kSmiTagMask);  // Remove
inverted smi-tag.
Ditto here.

http://codereview.chromium.org/987003/diff/5001/6002#newcode6571
src/ia32/codegen-ia32.cc:6571: } else {
If operand.number_info().IsInteger32, could that also be propagated?

[fschneider, 2010-03-17 09:19:50]

http://codereview.chromium.org/987003/diff/5001/6002
File src/ia32/codegen-ia32.cc (right):

http://codereview.chromium.org/987003/diff/5001/6002#newcode6525
src/ia32/codegen-ia32.cc:6525: __ and_(operand.reg(), ~kSmiTagMask);  // Remove
inverted smi-tag.
On 2010/03/17 07:10:39, Lasse Reichstein wrote:
> ASSERT_EQ(0, kSmiTag).
> Seems wasteful to use a 32-bit immediate to clear one bit.
> How about:
>   lea(operand.reg(), Operand(operand.reg(), kSmiTagMask));
>   not(operand.reg());
> (i.e., set it before not'ing and use lea to reduce size).
> That, or use and_b (but that's a partial register write, I'm not sure that's
> good for efficiency).

Good point. I'll go for your first suggestion using lea.

http://codereview.chromium.org/987003/diff/5001/6002#newcode6543
src/ia32/codegen-ia32.cc:6543: __ and_(answer.reg(), ~kSmiTagMask);  // Remove
inverted smi-tag.
On 2010/03/17 07:10:39, Lasse Reichstein wrote:
> Ditto here.

Done.

http://codereview.chromium.org/987003/diff/5001/6002#newcode6571
src/ia32/codegen-ia32.cc:6571: } else {
On 2010/03/17 07:10:39, Lasse Reichstein wrote:
> If operand.number_info().IsInteger32, could that also be propagated?

Done.