Port fdlibm implementation for Math.cosh.

Port fdlibm implementation for Math.cosh.

R=rtoy@chromium.org
BUG=v8:3494
LOG=N

Committed: https://code.google.com/p/v8/source/detail?r=23548

[Raymond Toy, 2014-08-29 16:23:32]

lgtm with some nits.

https://codereview.chromium.org/522723002/diff/1/test/mjsunit/es6/math-hyperb...
File test/mjsunit/es6/math-hyperbolic.js (right):

https://codereview.chromium.org/522723002/diff/1/test/mjsunit/es6/math-hyperb...
test/mjsunit/es6/math-hyperbolic.js:111: assertEqualsDelta(74.209948524787,
Math.cosh(-5), 1E-12);
Is there any reason to keep the tests in 108-111? If yes, then you should
probably use the correct values and test for equality: cosh(0.5) =
1.1276259652063807d0, and cosh(5) = 74.20994852478785d0

The same goes for the sinh tests in lines 105-106: sinh(5) = 74.20321057778875d0

If you want some further tests, you could use the properties cosh(x) + sinh(x) =
exp(x) and cosh(x)^2-sinh(x)^2 = 1.

https://codereview.chromium.org/522723002/diff/1/test/mjsunit/es6/math-hyperb...
test/mjsunit/es6/math-hyperbolic.js:173: assertEquals(1, Math.cosh(-Math.pow(2,
-55)));
Oops. This is a mistake in original test. If we want to test |x| < 2^-55, we
can't use 2^-55.  Replace with 2^-56.

https://codereview.chromium.org/522723002/diff/1/third_party/fdlibm/fdlibm.js
File third_party/fdlibm/fdlibm.js (right):

https://codereview.chromium.org/522723002/diff/1/third_party/fdlibm/fdlibm.js...
third_party/fdlibm/fdlibm.js:785: const KCOSH_OVERFLOW = kMath[52];
You might want to add a comment in fdlibm.cc that kMath[22] is the overflow
threshold for both sinh and cosh. The comment (from the sinh commit) only says
sinh.

[Yang, 2014-09-01 09:35:42]

https://codereview.chromium.org/522723002/diff/1/test/mjsunit/es6/math-hyperb...
File test/mjsunit/es6/math-hyperbolic.js (right):

https://codereview.chromium.org/522723002/diff/1/test/mjsunit/es6/math-hyperb...
test/mjsunit/es6/math-hyperbolic.js:111: assertEqualsDelta(74.209948524787,
Math.cosh(-5), 1E-12);
On 2014/08/29 16:23:31, Raymond Toy wrote:
> Is there any reason to keep the tests in 108-111? If yes, then you should
> probably use the correct values and test for equality: cosh(0.5) =
> 1.1276259652063807d0, and cosh(5) = 74.20994852478785d0
> 
> The same goes for the sinh tests in lines 105-106: sinh(5) =
74.20321057778875d0
> 
> If you want some further tests, you could use the properties cosh(x) + sinh(x)
=
> exp(x) and cosh(x)^2-sinh(x)^2 = 1.

I corrected those test expectations. I think we are fine regarding tests so far.
Actually I was hoping that we would eventually get some coverage in Test262 once
we have formal specification on accuracy.

https://codereview.chromium.org/522723002/diff/1/test/mjsunit/es6/math-hyperb...
test/mjsunit/es6/math-hyperbolic.js:173: assertEquals(1, Math.cosh(-Math.pow(2,
-55)));
On 2014/08/29 16:23:31, Raymond Toy wrote:
> Oops. This is a mistake in original test. If we want to test |x| < 2^-55, we
> can't use 2^-55.  Replace with 2^-56.

Done.

https://codereview.chromium.org/522723002/diff/1/third_party/fdlibm/fdlibm.js
File third_party/fdlibm/fdlibm.js (right):

https://codereview.chromium.org/522723002/diff/1/third_party/fdlibm/fdlibm.js...
third_party/fdlibm/fdlibm.js:785: const KCOSH_OVERFLOW = kMath[52];
On 2014/08/29 16:23:31, Raymond Toy wrote:
> You might want to add a comment in fdlibm.cc that kMath[22] is the overflow
> threshold for both sinh and cosh. The comment (from the sinh commit) only says
> sinh.

Done.

[Yang, 2014-09-01 09:36:11]

Committed patchset #3 (id:40001) manually as 23548 (presubmit successful).

[Raymond Toy, 2014-09-02 22:33:27]

https://codereview.chromium.org/522723002/diff/1/test/mjsunit/es6/math-hyperb...
File test/mjsunit/es6/math-hyperbolic.js (right):

https://codereview.chromium.org/522723002/diff/1/test/mjsunit/es6/math-hyperb...
test/mjsunit/es6/math-hyperbolic.js:111: assertEqualsDelta(74.209948524787,
Math.cosh(-5), 1E-12);
On 2014/09/01 09:35:42, Yang wrote:
> On 2014/08/29 16:23:31, Raymond Toy wrote:
> > Is there any reason to keep the tests in 108-111? If yes, then you should
> > probably use the correct values and test for equality: cosh(0.5) =
> > 1.1276259652063807d0, and cosh(5) = 74.20994852478785d0
> > 
> > The same goes for the sinh tests in lines 105-106: sinh(5) =
> 74.20321057778875d0
> > 
> > If you want some further tests, you could use the properties cosh(x) +
sinh(x)
> =
> > exp(x) and cosh(x)^2-sinh(x)^2 = 1.
> 
> I corrected those test expectations. I think we are fine regarding tests so
far.
> Actually I was hoping that we would eventually get some coverage in Test262
once
> we have formal specification on accuracy.

I was assuming you'd change the assertEqualsDelta to assertEquals, since there's
no delta required now. Oh well.