Stefanos Kornilios Mitsis Poiitidis
Stefanos Kornilios Mitsis Poiitidis
- Make sure SS / SD opcodes don't to merges - Optimize suffles - Make sure SIMD generates good enough stuff
We currently handle only 0x2, 0x3, 0x6, 0x7, 0xA, 0xB, everything else is forced to do full flags calculation
There's quite a bit of cruft in core.cpp and other files, that is best tackled as a follow up because of interdependencies with other open PRs. - Better code reuse...
Follow up from #1926 `set_target_properties(MyTarget PROPERTIES EXCLUDE_FROM_ALL TRUE)` Should also be done to other targets, like the IDE ones for thunks. #1923
Relocation pools can be marked MAP_PRIVATE instead of MAP_SHARED, and as such the object code caches can be used as is for execution instead of having to be imported to...
Follow up from #1842. Also make sure caching passes the smc tests and handles anonymous fds, etc.
... and also add FLTs for that. Follow up from #1842
.. to avoid a race condition around stale caches. Possibly best implemented via #1961 Follow up from #1842
... should reduce stutter contention Follow up from #1842
This will enable us to - Avoid rehashing pages for Code Caches - Better track SMC - Open a way for PROT_EXEC page deduplication in the future Follow up from...