Update ruby and mmtk-core repo rev #130
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The upstream has changed its way to handle the generic fields table. The generic_fields_tbl_ now maps each object to an imemo:fields, and the imemo:fields object will hold all the fields. We now treat the imemo:fields object as a child of its corresponding key in generic_fields_tbl_, which is just like CRuby's default GC. Like the default GC, we handle generic_fields_tbl_ like other weak tables during weak processing time. On the Rust side, we replaced the moved_gen_fields_tables hash map with a simple "backwarding table" that simply maps each moved object using generic_fields_tbl_ to its old address.
This is a reguler merging commit that synchronizes the MMTK's CRuby fork
with the upstream.
The upstream introduced the imemo:fields type, and is now using it for
the `generic_fields_tbl_`, i.e. it now holds instance variables for
objects other than `T_OBJECT`, `T_CLASS` and `T_MODULE`. When using
MMTk, we treat the key-value pair in the `generic_fields_tbl_` as a
strong edge, i.e. treating the imemo:fields of an object as if it were a
child. We now update the `generic_fields_tbl_` like other weak tables.
This simplified the handling of generic fields table in the MMTk
binding.
With imemo:fields added, we now have 17 imemo types including MMTk's
imemo:mmtk_strbuf and imemo:mmtk_objbuf. We increased the header bits
of the imemo type to 5 bits and changed the value of some imemo-specific
header offsets, such as `ISEQ_TRANSLATED`.
The upstream changed the API for acquiring/releasing the GVL. We make
changes accordingly.
YJIT assumes there is only one thread doing GC. It panics when two GC
worker threads try to mark two iseq objects simultaneously. We made
several changes to support parallel GC for YJIT-compiled iseq objects:
1. We replaced `Block::gc_obj_offsets` with `Block::gc_obj_addresses`
which are now absolute pointers instead of offsets. By doing so, we
no longer need to borrow the `Rc<RefCell<VirtualMem>>` during GC.
2. `rb_yjit_iseq_update_references` no longer uses `CodeBlock` to write
code, and no longer calls `mark_all_executable` after updating each
object. Instead we make the whole code memory writable before
updating any objects, and make it executable after the updating
phase finishes. This should both make it friendly to parallel GC
and improve performance.
The upstream now acquires the GVL when adding a freed fiber to the fiber
pool. Since MMTk worker threads cannot acquire the GVL, and MMTk only
calls `obj_free` in one GC worker thread, we skip the GVL when using
MMTk.
wks
force-pushed
the
update/merge-2025-06-25
branch
from
4dd4675 to
32e8dd0
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is a reguler merging commit that synchronizes the MMTK's CRuby fork
with the upstream.
The upstream introduced the imemo:fields type, and is now using it for
the
generic_fields_tbl_, i.e. it now holds instance variables forobjects other than
T_OBJECT,T_CLASSandT_MODULE. When usingMMTk, we treat the key-value pair in the
generic_fields_tbl_as astrong edge, i.e. treating the imemo:fields of an object as if it were a
child. We now update the
generic_fields_tbl_like other weak tables.This simplified the handling of generic fields table in the MMTk
binding.
With imemo:fields added, we now have 17 imemo types including MMTk's
imemo:mmtk_strbuf and imemo:mmtk_objbuf. We increased the header bits
of the imemo type to 5 bits and changed the value of some imemo-specific
header offsets, such as
ISEQ_TRANSLATED.The upstream changed the API for acquiring/releasing the GVL. We make
changes accordingly.
YJIT assumes there is only one thread doing GC. It panics when two GC
worker threads try to mark two iseq objects simultaneously. We made
several changes to support parallel GC for YJIT-compiled iseq objects:
Block::gc_obj_offsetswithBlock::gc_obj_addresseswhich are now absolute pointers instead of offsets. By doing so, we
no longer need to borrow the
Rc<RefCell<VirtualMem>>during GC.rb_yjit_iseq_update_referencesno longer usesCodeBlockto writecode, and no longer calls
mark_all_executableafter updating eachobject. Instead we make the whole code memory writable before
updating any objects, and make it executable after the updating
phase finishes. This should both make it friendly to parallel GC
and improve performance.