logic in Math.nextAfter for handling -0.0 input
Tom Rodriguez
tom.rodriguez at oracle.com
Sat Jun 28 00:13:26 UTC 2014
That’s supposed to work correctly. I suppose it could be because of inline everything though I’m not sure how that would trigger the problem. Can try running your problematic test case against regular graal and see if it reproduces?
I’ve seen cases like this with recursive deopts. Maybe that’s what’s happening here?
tom
On Jun 27, 2014, at 4:40 PM, Deneau, Tom <tom.deneau at amd.com> wrote:
> That's what I was doing below. Is the problem that we are "inlining everything" on the hsail backend?
>
> -- Tom
>
> -----Original Message-----
> From: Tom Rodriguez [mailto:tom.rodriguez at oracle.com]
> Sent: Friday, June 27, 2014 6:39 PM
> To: Deneau, Tom
> Cc: graal-dev at openjdk.java.net
> Subject: Re: logic in Math.nextAfter for handling -0.0 input
>
> Just call it. A call to the original method stays a real call to that method and doesn't get substituted.
>
> tom
>
> On Jun 27, 2014, at 4:35 PM, Deneau, Tom <tom.deneau at amd.com> wrote:
>
>> Pinging, since I never saw an answer to this.
>>
>> I thought there was some way to substitute for a method and still call
>> the original method without getting into a recursive call...
>>
>> (The original problem below is resolved but need this kind of solution for a different reason).
>>
>> -- Tom
>>
>>
>> -----Original Message-----
>> From: Deneau, Tom
>> Sent: Monday, June 23, 2014 3:43 PM
>> To: Deneau, Tom; 'graal-dev at openjdk.java.net'
>> Subject: RE: logic in Math.nextAfter for handling -0.0 input
>>
>> Related to this...
>> Should I be able to work around this with the following hsail MethodSubstitution?
>> When I try this, I get a StackOverflowError...
>>
>> -- Tom
>>
>> @ClassSubstitution(java.lang.Math.class)
>> public class HSAILMathSubstitutions {
>> ...
>> @MethodSubstitution
>> public static double nextAfter(double x, double direction) {
>> // work around special case of -0.0
>> double xx = (x == -0.0 ? 0.0 : x);
>> return Math.nextAfter(xx, direction);
>> }
>> }
>>
>>
>>
>>
>> -----Original Message-----
>> From: Deneau, Tom
>> Sent: Monday, June 23, 2014 3:34 PM
>> To: 'graal-dev at openjdk.java.net'
>> Subject: logic in Math.nextAfter for handling -0.0 input
>>
>> The JDK method Math.nextAfter contains the logic shown below to handle
>> an input of -0.0
>>
>> When the graal compiler compiles this for the hsail backend, it makes the reasonable assumption that "start + 0.0d" can be reduced to "start" which is a problem for this algorithm.
>>
>> From what I could tell, c2 or graal for amd64 backend do not do this optimization and so get the right answer for -0.0 input. How do they know not to do this optimization?
>>
>>
>> } else { // start > direction or start < direction
>> // Add +0.0 to get rid of a -0.0 (+0.0 + -0.0 => +0.0)
>> // then bitwise convert start to integer.
>> long transducer = Double.doubleToRawLongBits(start + 0.0d); <==============
>>
>> if (direction > start) { // Calculate next greater value
>> transducer = transducer + (transducer >= 0L ? 1L:-1L);
>> } else { // Calculate next lesser value
>> ....
>>
>> -- Tom
>
More information about the graal-dev
mailing list