Linux: Floating point expressions aren't rounded correctly

Name: asR10013 Date: 05/29/2000


    Test vm/misc/FloatingPointInline.java from regression testsuite fails
on linux jdk1.3b07. The reason is the same described in bug 4105954.
First and second testcases report incorrect floating point round scheme.
Below is testcases descriptions and output.
/**
* First test:
* The value tiny is near the minimum representable double.
* Dividing tiny by small should get us a value
* that is not representable.
* Page 35 says the result should be "the representable value
* nearest to the the infinitely precise result".
* The nearest representable value is 0.0, being 4 times
* nearer than Double.MIN_VALUE.
* Multiply that by big and you should still get 0.0.
* What you should not get is the extended-precision
* value of (tiny/small) multiplied by big.
* Note that the computation of tiny/small should
* yield 0.0, whether it is an intermediate result or a
* stored value.
*/
------------------------------ output --------------------------------
$ java -version
java version "1.3.0beta"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.3.0beta-b07)
Java HotSpot(TM) Client VM (build 1.3.0beta-b04, mixed mode)
$ java FloatingPointInline
Sigh: double range too big near 0
Oops: arg double range too big near 0
Oops: result double range too big near 0
        tiny: 1.0E-323
        tiny/small: 0.0
        (tiny/small)*big: 3.16E-322
        MIN_VALUE*big: 1.265E-321
        (tiny*big)/small: 3.16E-322
<<the rest is skipped>>
----------------------------------------------------------------------
/**
* Second test:
* Huge is near the maximum representable double.
* Multiplying huge by small should get us a value that
* is not representable, so should be Double.POSITIVE_INFINITY.
* Divide that by big and you should still get Infinity.
* What you should not get is the extended-precision
* value of (huge*small) divided by big.
*/
------------------------------ output --------------------------------
$ java -version
java version "1.3.0beta"
Java(TM) 2 Runtime Environment, Standard Edition (build 1.3.0beta-b07)
Java HotSpot(TM) Client VM (build 1.3.0beta-b04, mixed mode)
$ java FloatingPointInline
Sigh: double range too big near infinity
Oops: arg double range too big near infinity
Oops: result double range too big near infinity
        huge: 8.988465674311579E307
        huge*small: Infinity
        (huge*small)/big: 2.8088955232223683E306
<<the rest is skipped>>
======================================================================

======================================================================