RFR (S) : 8014362 : Need to expose some processor features via Unsafe interface
john.r.rose at oracle.com
Wed May 15 16:24:17 PDT 2013
On May 15, 2013, at 3:19 PM, Vladimir Kozlov <vladimir.kozlov at oracle.com> wrote:
> I think, sun.zip.clmulSupported approach is only temporary solution. As Christian told we would need more CPU information for Graal.
> We should start discussing (new RFE) about new JVM - JAVA API to pass information about hybrid hw environment (GPU+CPU+etc) JVM is running on.
Yes, I also view this as a temporary approach, to be evaluated after we understand better how to do these aggressive platform-sensitive optimizations.
(I would hate to see a series of follow-ons like sun.arraycopy.bisSupported, for example.)
David is breaking a lot of ice here, and we are learning from his experience. Besides the benefit of better CRC32 performance, we will also have better hindsight to do similar processor-sensitive optimizations in a cleaner way.
Given that we may be running on heterogeneous processors in the future, I think a sun.misc.Processor class, designed with similar trust and packaging parameters as sun.misc.Unsafe, will be a useful aid to processor-sensitive optimizations and Java-based compilation mechanisms. Let's work up a simple initial design, tailored like Unsafe to the mission of the compiler group, and see where it goes.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the hotspot-compiler-dev