Add MathPowStub to x64 platform, and fix error in stub on ia32 platform.

Add MathPowStub to x64 platform, and fix error in stub on ia32 platform.

Committed: http://code.google.com/p/v8/source/detail?r=6973

[Lasse Reichstein, 2011-02-28 13:21:41]

LGTM

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc
File src/x64/code-stubs-x64.cc (right):

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2090: CpuFeatures::Scope use_sse2(SSE2);
No need to have SSE2 scopes in X64.

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2154: __ movl(rcx, Immediate(0x7FB00000));
Could you detect infinity by adding the value to itself and see that it doesn't
change? (Need to filter out zero at the beginning then, ofcourse).
Or subtract it from itself and not get zero (but NaN)?

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2160: __ movsd(xmm1, xmm3);
It's worth noticing that if xmm1 is very small, the division might give Infinity
- but that's the correct result for the operation.

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2163: // exponent (or both) is a heapnumber - no
matter what we should now work
Capitalize exponent.

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2189: __ j(greater_equal, &call_runtime);
If the value isn't -Infinity, won't the result be NaN/Infinity too. We shouldn't
need to go to runtime. 
It's only if it's -Infinity that we need to do something (and I have no idea
what :).

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2197: __ movl(rcx, Immediate(0xBF000000));
Is this smaller than loading the exact representation of 1/2 as 64-bit?

[William Hesse, 2011-02-28 14:36:11]

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc
File src/x64/code-stubs-x64.cc (right):

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2154: __ movl(rcx, Immediate(0x7FB00000));
We now divide first, then compare to 0.


On 2011/02/28 13:21:41, Lasse Reichstein wrote:
> Could you detect infinity by adding the value to itself and see that it
doesn't
> change? (Need to filter out zero at the beginning then, ofcourse).


> Or subtract it from itself and not get zero (but NaN)?

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2160: __ movsd(xmm1, xmm3);
On 2011/02/28 13:21:41, Lasse Reichstein wrote:
> It's worth noticing that if xmm1 is very small, the division might give
Infinity
> - but that's the correct result for the operation.

Done.

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2189: __ j(greater_equal, &call_runtime);
The value b^infinity depends on whether b is >, =, or < 1, and >, =, or < 0.  It
is complicated.

On 2011/02/28 13:21:41, Lasse Reichstein wrote:
> If the value isn't -Infinity, won't the result be NaN/Infinity too. We
shouldn't
> need to go to runtime. 
> It's only if it's -Infinity that we need to do something (and I have no idea
> what :).

http://codereview.chromium.org/6602007/diff/1/src/x64/code-stubs-x64.cc#newco...
src/x64/code-stubs-x64.cc:2197: __ movl(rcx, Immediate(0xBF000000));
Changed.

On 2011/02/28 13:21:41, Lasse Reichstein wrote:
> Is this smaller than loading the exact representation of 1/2 as 64-bit?