RFR: 8342826: Improve performance of oopDesc::klass() after JDK-8305895 [v2]
Roman Kennke
rkennke at openjdk.org
Thu Nov 14 11:54:56 UTC 2024
On Thu, 14 Nov 2024 11:44:33 GMT, Coleen Phillimore <coleenp at openjdk.org> wrote:
> I have sort of a different idea. Instead of these switch statements, can you templatize these functions based on ObjLayoutMode?
There are a few places where these methods (or callers of these methods, like oopDesc::size()) are used in hot loops. In those cases I guess we could do that. But templatizing the methods always needs 'driver methods', e.g. some place that decides, based on the runtime value of the flags, which temlatized loop to call. It's ugly.
AFAIK, @tstuefe has done something similar and slightly better than that in his KLUTE prototype, but it's also more complex and nothing that helps us right now with the problems that we have (e.g. in stack-trace building, which is what affected us in dacapo:pmd).
-------------
PR Comment: https://git.openjdk.org/jdk/pull/22020#issuecomment-2476149932
More information about the hotspot-runtime-dev
mailing list