5.2 multicore markdelay #3029
Merged
5.2 multicore markdelay #3029
Conversation
This file contains 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
stedolan
force-pushed
the
5.2-multicore-markdelay
branch
from
e3fd82b to
5e85fe9
Compare
stedolan
force-pushed
the
5.2-multicore-markdelay
branch
from
5e85fe9 to
69cd4ed
Compare
|
(This was automatically closed by Github when |
NickBarnes
reviewed
Oct 11, 2024
- Domains now begin marking simultaneously at some minor GC during the major cycle
- Before marking, the write barrier is disabled and fresh allocations are UNMARKED
- Change to ephemeron logic: per-cycle ephemeron data structure reset is now done
at marking start rather than at the start of the cycle. (In particular, this makes
orphaned work simpler to deal with)
(cherry picked from commit f376410)
Includes several systhreads tests that rely on the interaction between systhreads, domains and backup threads. (cherry picked from commit c31eebd)
…tion (cherry picked from commit 2368930)
stedolan
force-pushed
the
5.2-multicore-markdelay
branch
from
69cd4ed to
5d2b2dc
Compare
|
I rebased to current |
NickBarnes
reviewed
Oct 14, 2024
mshinwell
approved these changes
Oct 16, 2024
ncik-roberts
pushed a commit
that referenced
this pull request
Oct 21, 2024
* Markdelay support for multi-domain programs
- Domains now begin marking simultaneously at some minor GC during the major cycle
- Before marking, the write barrier is disabled and fresh allocations are UNMARKED
- Change to ephemeron logic: per-cycle ephemeron data structure reset is now done
at marking start rather than at the start of the cycle. (In particular, this makes
orphaned work simpler to deal with)
* Re-enable parallel tests
Includes several systhreads tests that rely on the interaction between
systhreads, domains and backup threads.
* Add a new test (tests/parallel/churn.ml) stressing cross-domain promotion
lukemaurer
pushed a commit
that referenced
this pull request
Oct 23, 2024
* Markdelay support for multi-domain programs
- Domains now begin marking simultaneously at some minor GC during the major cycle
- Before marking, the write barrier is disabled and fresh allocations are UNMARKED
- Change to ephemeron logic: per-cycle ephemeron data structure reset is now done
at marking start rather than at the start of the cycle. (In particular, this makes
orphaned work simpler to deal with)
* Re-enable parallel tests
Includes several systhreads tests that rely on the interaction between
systhreads, domains and backup threads.
* Add a new test (tests/parallel/churn.ml) stressing cross-domain promotion
This was referenced Dec 2, 2024
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 patch extends the existing single-domain-only markdelay patch to work with multiple domains, and (in separate commits) re-enables all the parallel tests and adds a new parallel test.
The idea is to add a new phase,
Phase_sweep_main, prior toPhase_sweep_and_mark_main, in which only sweeping and not marking can be done. Since the roots have not yet been marked, it is safe to give new allocations theUNMARKEDstatus during this phase, allowing them to be collected a cycle earlier than otherwise. Also, caml_modify need not do darkening during this phase. This was implemented for single-domain programs in #2348, and this PR extends it to multi-domain programs.Since domains freely share new allocations, all must agree on phase transitions. So, when any domain finishes sweeping, it requests a transition to Phase_sweep_and_mark_main. Instead of being acted upon directly, this sets a flag to trigger the phase transition at the next minor GC. (The phase transition involves marking roots, which must be done with the minor heap empty, and the simplest way to achieve this is to wait until the next point when the minor heap naturally empties)