(no subject)

Hi all,

I’m investigating on integrating a gc to my llvm project, and when I read this document, one sentence confused me:

However, for a collector which wishes to relocate objects directly reachable from running code, a higher standard is required.

I don’t understand what the move “relocate objects directly reachable from running code” trying to do.

For my information, the concept “relocate” means the gc pointer refereed to a new location of an object, for example:

in Java:

Foo foo = new Foo();

foo = new Foo(); // ---> a relocation happens

So can anyone explain what the “relocate objects directly reachable from running code” trying to do?

I think it’s referring to a “moving” garbage collector (as opposed to a “non-moving” garbage collector that never moves/relocates objects.) The difference is that for a moving one, all pointers need to be tracked and potentially updated, whereas for a non-moving one, it’s sufficient that at least one pointer to a live object is seen (when there may be other pointers to the same object elsewhere) for correctness.

Exactly this. (As the person who wrote the line in question.)

Correct me if I’m wrong:
Relocation in this conversation “relocation” means GC trying to move objects in the heap for optimization (make the data more impact for bigger room for example), this move is invisible to a programmer, and if a compiler support to " relocate objects directly reachable from running code", a variable (Foo foo) may points to a different address after a GC happens, while a programmer could still use the name foo in the code as if nothing happened.

Correct, with a couple of nit picks.

Relocation isn’t an optimization the collector performs. It’s a key primitive the collector is built upon. Being unable to relocate is not an allowed state. (i.e. pinning can’t be required by the compiler)

When you talk about variables, that’s true for the source language and for the abstract machine before lowering. After lowering from the abstract machine, relocations are represented in the IR explicitly as an entirely new set of defs.


Thanks Philip, I’ll keep investigating.

Hi Philip, are there any C++ APIs to generate the llvm statepoint instructions? I didn’t found APIs similar to how to generate gcroot instructions.

The general assumption is that you use RS4GC to lower the abstract machine model.

I think there’s IRBuilder support. If not, patches welcome!