[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: floats ??????

From: Eric Blake
Subject: Re: floats ??????
Date: Mon, 16 Sep 2002 07:17:32 -0600
User-agent: Mozilla/5.0 (Windows; U; Win98; en-US; rv:1.0.1) Gecko/20020823 Netscape/7.0

In the .class file, any version of NaN is legal (both fff8000000000000 and 7ff8000000000000 are NaNs, and which one is prefered by the architecture is machine dependent). This may be a bug in doubleToLongBits for not flattening to the Java canonical NaN properly. Also, since jikes 1.13, jikes emulates all floating point ops when calculating compiled constants, rather than using hardware floating point, so the particular problem you mention may be solved in a newer version of jikes.

I'm still trying to convince the jikes release manager that it is time for 1.17; as 1.16 incorrectly compiles a few of the Classpath files, while exiting with a success status. The current CVS of jikes does not have these problems with Classpath, as far as I can tell.

Chris Gray wrote:
On 16 Sep 2002, John Leuner wrote:

It's almost certainly a kissme bug.

Mind you some versions at least of Jikes generate a bad bit pattern for NaN:

FAIL $$java.lang.Double$$doubleToLongBits(double)long$$Value Double.NaN$$: gnu.testlet.wonka.lang.Double.DoubleTest: doubleToLongBits(double)long (number 9) got -2251799813685248 but expected 9221120237041090560 NaN = fff8000000000000, but should be 0x7ff8000000000000
That's jikes 1.12, which was the last version to be able to compile Wonka
without generating bad code or dumping core ...

This signature intentionally left boring.

Eric Blake             address@hidden
  BYU student, free software programmer

reply via email to

[Prev in Thread] Current Thread [Next in Thread]