On Tue, Aug 28, 2012 at 9:53 PM, Michael Barker <[address removed]> wrote:
>> And no... no JIT magic/optimization can be applied.
> For the overflow case, I thought it would be fairly straight forward
> for the JVM to implement an intrinsic to provide machine specific
> assembly. Is there are reason why this technique can be applied?
Did you mean "Are there reasons why this technique can't be applied"?
Richard - are you aware of how HotSpot can utilize highly
machine-specific code using techniques like intrinsics? If not, this
could well be an interesting topic for us to discuss at the next
OpenJDK hack day.
In general, I think that JIT-based techniques for unsigned arithmetic
can be problematic precisely because they don't sit within the overall
language or platform specification.
However, expecting HPC techniques to be defined by a standard which
only specifies the interpreted form of the language seems also to be a
problematic way of approaching the issue. I'm not sure there are any
easy answers here. :(