ARM: Correctness fix to Math.pow optimization...

ARM: Correctness fix to Math.pow optimization

The change in r4990 contained a bug in Math.pow when then exponent was a large negative smi. In that case calculating 1/Math.pow(x,-y) did not provide the correct result as Math.pow(x,-y) would overflow ti infinity. This was caught by Sputnik test S8.5_A13_T1.

Committed: http://code.google.com/p/v8/source/detail?r=4994

[Erik Corry, 2010-06-30 15:10:53]

LGTM.  We should have this test in our own test suite too.

[Søren Thygesen Gjesse, 2010-06-30 15:20:17]

On 2010/06/30 15:10:53, Erik Corry wrote:
> LGTM.  We should have this test in our own test suite too.

The following tests from Sputnik S8.5_A13_T1 are added to mjsunit test
math-pow.js.

assertTrue((1*((Math.pow(2,53))-1)*(Math.pow(2,-1074))) ===
4.4501477170144023e-308);
assertTrue((1*(Math.pow(2,52))*(Math.pow(2,-1074))) ===
2.2250738585072014e-308);
assertTrue((-1*(Math.pow(2,52))*(Math.pow(2,-1074))) ===
-2.2250738585072014e-308);