Request for review (S): 7042122: JSR 292: adjust various inline thresholds for JSR 292 API methods and method handle adapters
christian.thalinger at oracle.com
Mon May 9 11:34:45 PDT 2011
On May 9, 2011, at 7:46 PM, Vladimir Kozlov wrote:
> Fine as point fix but one more thing. Should you have a check that MDO already exist?
Since we are spinning the bytecode during compilation there can't be an MDO. Anyway, I will the check. Maybe it's useful later.
> On 5/9/11 7:41 AM, Christian Thalinger wrote:
>> On May 7, 2011, at 12:29 AM, John Rose wrote:
>>> On May 6, 2011, at 2:42 PM, Vladimir Kozlov wrote:
>>>> OK, then it makes sense to have MDO. An other thing. You actually don't need real MDO. Do you? As I understand you need MDO only during compilation then you should use CURRENT_ENV->get_empty_methodData() in ciMethodHandle::get_adapter()
>>>> to set call counters and assign it to created ciMethod._method_data.
>>> It's probably safer to have the real MDO. The adapter is a real method, and it can (I think) reappear in the course of deoptimization and maybe even recompilation.
>>> (This raises the question of whether we should be throwing away the adapters generated during compilation. It seems non-parsimonious to do so, but it is simplest.)
>> Can we agree on this as a point-fix? I guess we have to adjust it later anyway as we get more customer feedback (or we are doing more optimizations).
>> -- Christian
More information about the hotspot-compiler-dev