Summary

Increase the allowable error bounds of Math.hypot to 1.5 ulps.

Problem

The FDLIBM hypot implementation does not conform to its stated 1 ulp error bound.

Solution

Increase the error bound to cover the worst-case error known so far.

Specification

diff --git a/src/java.base/share/classes/java/lang/Math.java b/src/java.base/share/classes/java/lang/Math.java
index c851630b95a..11c636c8434 100644
--- a/src/java.base/share/classes/java/lang/Math.java
+++ b/src/java.base/share/classes/java/lang/Math.java
@@ -2755,7 +2755,7 @@ public static double tanh(double x) {
      * <li> If both arguments are zero, the result is positive zero.
      * </ul>
      *
-     * <p>The computed result must be within 1 ulp of the exact
+     * <p>The computed result must be within 1.5 ulps of the exact
      * result.  If one parameter is held constant, the results must be
      * semi-monotonic in the other parameter.
      *