Make more use of the NumberInfo data....

Make more use of the NumberInfo data.
Makes NumberInfo into a real class.
Fix bug where NumberInfo was lost in ToRegister.
Allow 17 bits in safe Smis instead of 16.

Committed: http://code.google.com/p/v8/source/detail?r=4046

[fschneider, 2010-03-05 15:01:06]

LGTM.

I'm looking forward to see the overall performance impact!

http://codereview.chromium.org/668151/diff/1/11
File src/number-info.h (right):

http://codereview.chromium.org/668151/diff/1/11#newcode60
src/number-info.h:60: // Return compact representation.  Very sensitive to enum
values above!
Comment: above --> below

http://codereview.chromium.org/668151/diff/1/11#newcode107
src/number-info.h:107: return ((type_ & kNumberType) != 0);
for consistency with the other functions:

return ((type_ & kNumberType) == kNumberType);

http://codereview.chromium.org/668151/diff/37/43
File src/ia32/codegen-ia32.cc (right):

http://codereview.chromium.org/668151/diff/37/43#newcode1047
src/ia32/codegen-ia32.cc:1047: // Result is always a number. Smi property of
inputs is preserved.
// Result is always a integer32.

http://codereview.chromium.org/668151/diff/37/43#newcode1053
src/ia32/codegen-ia32.cc:1053: // Result is a smi if we shift by a constant >=
1, otherwise a number.
number --> integer32

http://codereview.chromium.org/668151/diff/37/43#newcode1059
src/ia32/codegen-ia32.cc:1059: // Result is a smi if we shift by a constant >=
2, otherwise a number.
number --> integer32

http://codereview.chromium.org/668151/diff/37/43#newcode1077
src/ia32/codegen-ia32.cc:1077: // The Integer32 range is big enough to take the
difference of any two Smis.
long line

http://codereview.chromium.org/668151/diff/37/43#newcode1875
src/ia32/codegen-ia32.cc:1875: operand->number_info(),
long line?

http://codereview.chromium.org/668151/diff/37/43#newcode1913
src/ia32/codegen-ia32.cc:1913: operand->number_info(),
You could avoid creating the DeferredInlineSmiOperation object in case we have a
smi input here.

http://codereview.chromium.org/668151/diff/37/43#newcode2021
src/ia32/codegen-ia32.cc:2021: if (!operand->number_info().IsSmi()) {
Also here only instantiate DeferredInlineSmiOperation if we don't have a smi.

http://codereview.chromium.org/668151/diff/37/43#newcode2069
src/ia32/codegen-ia32.cc:2069: operand->number_info(),
long line.

http://codereview.chromium.org/668151/diff/37/43#newcode2079
src/ia32/codegen-ia32.cc:2079: if (!operand->number_info().IsSmi()) {
Also here only instantiate DeferredInlineSmiOperation if we don't have a smi.

http://codereview.chromium.org/668151/diff/37/43#newcode2144
src/ia32/codegen-ia32.cc:2144: operand->number_info(),
Long line.

http://codereview.chromium.org/668151/diff/37/43#newcode9072
src/ia32/codegen-ia32.cc:9072: if (number_info.IsHeapNumber()) {
This function is getting quite messy with all the different cases for
number_info. I wonder if it would be better (for readability) to do at the
beginning

if (!number_info.IsNumber()) {
  ...
} else {
  ...
}

and then have the different cases for smi, heap number in the else part.

http://codereview.chromium.org/668151/diff/37/43#newcode9073
src/ia32/codegen-ia32.cc:9073: __ jmp (&arg1_is_object);
Remove extra space.

http://codereview.chromium.org/668151/diff/37/47
File src/number-info.h (right):

http://codereview.chromium.org/668151/diff/37/47#newcode60
src/number-info.h:60: // Return compact representation.  Very sensitive to enum
values above!
above -> below

http://codereview.chromium.org/668151/diff/37/51
File src/register-allocator.cc (right):

http://codereview.chromium.org/668151/diff/37/51#newcode56
src/register-allocator.cc:56: NumberInfo Result::number_info() const {
Move this to register-allocator-inl.h.

http://codereview.chromium.org/668151/diff/37/51#newcode58
src/register-allocator.cc:58: if (!is_constant()) return
NumberInfo::FromInt(NumberInfoField::decode(value_));
Long line.

http://codereview.chromium.org/668151/diff/37/51#newcode66
src/register-allocator.cc:66: void Result::set_number_info(NumberInfo info) {
Move this to register-allocator-inl.h.

http://codereview.chromium.org/668151/diff/37/38
File src/register-allocator.h (right):

http://codereview.chromium.org/668151/diff/37/38#newcode104
src/register-allocator.h:104: NumberInfo number_info() const;
Maybe make this an inline function?

http://codereview.chromium.org/668151/diff/37/38#newcode105
src/register-allocator.h:105: void set_number_info(NumberInfo info);
Maybe make this an inline function?

[Erik Corry, 2010-03-05 20:20:15]

Fixed everything else.  Unfortunately I can't land it now because x64 doesn't
compile.

http://codereview.chromium.org/668151/diff/37/43
File src/ia32/codegen-ia32.cc (right):

http://codereview.chromium.org/668151/diff/37/43#newcode2079
src/ia32/codegen-ia32.cc:2079: if (!operand->number_info().IsSmi()) {
On 2010/03/05 15:01:07, fschneider wrote:
> Also here only instantiate DeferredInlineSmiOperation if we don't have a smi.

Actually instead I will remove the 'if' since shifting a known Smi by constant 0
seems like a stupid thing to do.