The method Math.tan has a 1 ulp error bound.
Recent work (https://members.loria.fr/PZimmermann/papers/accuracy.pdf) has found a case where the reference implementation of Math.tan has a 1.02 ulp error, which is (slightly) larger than the stated bound. The problematic input is 0x1.3f9605aaeb51bp+21.
The specification should be amended to accommodate the larger observed error bound.
Recent work (https://members.loria.fr/PZimmermann/papers/accuracy.pdf) has found a case where the reference implementation of Math.tan has a 1.02 ulp error, which is (slightly) larger than the stated bound. The problematic input is 0x1.3f9605aaeb51bp+21.
The specification should be amended to accommodate the larger observed error bound.
- csr for
-
JDK-8326537 Widen allowable error bound of Math.tan
- Closed
- relates to
-
JDK-8316708 Augment WorstCaseTests with more cases
- Resolved
-
JDK-8316688 Widen allowable error bound of Math.hypot
- Resolved
- links to
-
Commit openjdk/jdk/63f6a563
-
Review(master) openjdk/jdk/17973