Status: Accepted
Owner: [email protected]
CC: [email protected]
Labels: Type-Bug Priority-Medium ES5
New issue 1975 by [email protected]: Precision of trignometric
functions is low
http://code.google.com/p/v8/issues/detail?id=1975
The precision of some builtin trignometric functions (i.e. Math.pow(),
Math.sin() and Math.tan()) is very low on some corner cases when compared
to precomputed values.
1) We should verify that the precomputed values are actually a good measure
of what a good approximation of IEEE 754 arithmetic should return. The spec
recommends (see section 15.8.2) using "fdlibm" as a reference
implementation.
2) We should verify that the test cases mentioned below actually perform
correct precision measurement. The computation of the deviation itself can
be tricky when it comes to floating point arithmetic.
3) We should find out whether (and how) to improve our precision.
Affected test cases are:
* test262/S15.8.2.16_A7
* test262/S15.8.2.18_A7
* test262/S15.8.2.13_A23
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
