Tianle Qiu

> Yes, markcompact moves objects through a linear scan

The weird thing is in jetbrains' IDEs, when ligature is not enabled, it seems to be fine. However, in other editors like vs code and sublime text 4, disabling ligature...

> But I also remember that when running benchmarks, we also disable `System.gc()` and completely rely on `harness_begin` to trigger GC. See [the `ignore_system_gc` option](https://github.com/mmtk/mmtk-core/blob/170630fc84bc82070ca6e06ea4fed76894aa703f/src/util/options.rs#L800) and [the `ignoreSystemGC` option in...

> Results for all benchmarks are here. DaCapo Chopin MR2, vole.moma, 2.4x min heap size w.r.t semispace on vole.moma, 40 invocations, 5 iterations, using ConcurrentImmix, Immix and GenImmix. > >...

A side note, in C1, code patching is done on instruction instead of LIR. When code needs patching, the value of `slot` is incorrect. To support field barrier, we need...

> I fixed one obvious bug which is a typo that removes the barrier when the new field value is null. [940c691](https://github.com/mmtk/mmtk-openjdk/commit/940c6915a8d39661cb78112afc97ef4b80cb1305) > > But the SIGSEGV bug remains. It...

Well, I suggest we should add "old"/"previous" object to the barrier interface. In a SATB barrier for example, fast-path needs to load the object to check if slow-path is needed...