Add a DroplessArena and utilize it as a more efficient arena when possible #38653
Conversation
| } | ||
| } | ||
| } | ||
|
|
||
| pub struct CtxtInterners<'tcx> { | ||
| /// The arenas that types etc are allocated from. |
There was a problem hiding this comment.
No need to remove the comment, just make it singular.
| @@ -387,6 +369,7 @@ impl<'a, 'gcx, 'tcx> Deref for TyCtxt<'a, 'gcx, 'tcx> { | |||
| } | |||
|
|
|||
| pub struct GlobalCtxt<'tcx> { | |||
| global_arenas: &'tcx CtxtArenas<'tcx>, | |||
There was a problem hiding this comment.
A rename to GlobalArenas might make more sense. In fact, a GlobalInterners type would help too, I think layout isn't the only one, and stability also is global-only.
| @@ -4560,9 +4561,9 @@ impl<'a, 'gcx, 'tcx> FnCtxt<'a, 'gcx, 'tcx> { | |||
| /// Report errors if the provided parameters are too few or too many. | |||
| fn check_path_parameter_count(&self, | |||
| span: Span, | |||
| segment: &mut Option<(&hir::PathSegment, &ty::Generics)>) { | |||
| segment: &mut Option<(&hir::PathSegment, Rc<ty::Generics>)>) { | |||
There was a problem hiding this comment.
Maybe you shouldn't change anything about ty::Generics, if you can avoid it (since we're keeping at least some of the typed drop-ful arenas for now).
It relies on the underlying types not having special Drop behavior.
Mark-Simulacrum
force-pushed
the
dropless-arena
branch
from
c6f8cfb to
ace23f1
Compare
CtxtInterners contains a single DroplessArena, while GlobalArenas contains the TypedArenas still required for the remaining Drop-containing types.
Mark-Simulacrum
force-pushed
the
dropless-arena
branch
from
ace23f1 to
1676bec
Compare
|
|
||
| stability_interner: RefCell<FxHashSet<&'tcx attr::Stability>>, | ||
|
|
||
| layout_interner: RefCell<FxHashSet<&'tcx Layout>>, |
There was a problem hiding this comment.
My main concern now might be grouping these in a GlobalInterners type.
|
@bors r+ |
|
📌 Commit 1676bec has been approved by |
|
Leaving this here in case anyone has any ideas: https://gist.githubusercontent.com/Mark-Simulacrum/6a088a61f8eff2c2ded205432b4b5de9/raw/982f7329e4ac5c2c8aec57fbccf20257b8a350de/gistfile1.txt. This is a list of the However, no meaningful optimizations based on this have been thought of quite yet. |
|
⌛ Testing commit 1676bec with merge ca4a0b1... |
|
💔 Test failed - status-travis |
|
@bors: retry
* osx linker segfaulted
…On Sat, Dec 31, 2016 at 2:08 PM, bors ***@***.***> wrote:
💔 Test failed - status-travis
<https://travis-ci.org/rust-lang/rust/builds/187951270>
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#38653 (comment)>, or mute
the thread
<https://github.com/notifications/unsubscribe-auth/AAD95PGVU473c94j8FKuu0DWGHbz_IBfks5rNtJNgaJpZM4LW-Oo>
.
|
Add a DroplessArena and utilize it as a more efficient arena when possible I will collect performance (probably just `-Ztime-passes`, and more if that shows significant differences, perhaps). 6feba98 also fixes a potential infinite loop if inplace reallocation failed for `TypedArena` (and `DroplessArena` via copied code). r? @eddyb
|
☀️ Test successful - status-appveyor, status-travis |
I will collect performance (probably just
-Ztime-passes, and more if that shows significant differences, perhaps).6feba98 also fixes a potential infinite loop if inplace reallocation failed for
TypedArena(andDroplessArenavia copied code).r? @eddyb