On Tue, 13 Jul 2021 23:29:38 GMT, Joe Darcy <da...@openjdk.org> wrote:
>> Brian Burkhalter has updated the pull request incrementally with one
>> additional commit since the last revision:
>>
>> 8211002: Change so that Tests.testUlpDiff() handles the NaNs and accepts a
>> ulp tolerance of 2
>
> test/jdk/java/lang/Math/PowTests.java line 65:
>
>> 63: int failures = 0;
>> 64: if (Double.isNaN(smResult)) {
>> 65: if (!Double.isNaN(mResult)) failures = 1;
>
> Pretty sure the Tests.testUlpDiff call will handle two NaNs as well as
> NaN/non-NaN argument.
Looks like that is true.
> test/jdk/java/lang/Math/PowTests.java line 69:
>
>> 67: failures += Tests.testUlpDiff(
>> 68: "StrictMath.pow(double, double) vs Math.pow(double,
>> double)",
>> 69: input1, input2, mResult, smResult, 1.0
>
> In theory at least, one pow method could round an ulp up and the other could
> round an ulp down so the difference should be 2 ulps (ignoring the special
> case of values straddling a power of two).
Good point.
-------------
PR: https://git.openjdk.java.net/jdk/pull/4758
