Use sort_by_cached_key when the key function is not trivial/free #55821

ljedrz · 2018-11-09T14:16:24Z

I'm not 100% sure about def_path_hash (everything it does is inlined) but it seems like a good idea at least for the rest, as they are cloning.

rust-highfive · 2018-11-09T14:16:27Z

r? @cramertj

(rust_highfive has picked a reviewer for you, use r? to override)

Mark-Simulacrum · 2018-11-15T18:44:01Z

@bors try for perf

r? @eddyb

bors · 2018-11-15T18:44:13Z

⌛ Trying commit 1649c2e with merge a22afbf...

Use sort_by_cached_key when the key function is not trivial/free I'm not 100% sure about `def_path_hash` (everything it does is inlined) but it seems like a good idea at least for the rest, as they are cloning.

eddyb · 2018-11-15T19:12:12Z

r? @nikomatsakis

bors · 2018-11-15T21:03:11Z

☀️ Test successful - status-travis
State: approved= try=True

sinkuu · 2018-11-16T00:10:58Z

src/librustc_mir/monomorphize/partitioning.rs

@@ -585,7 +585,7 @@ fn merge_codegen_units<'tcx>(tcx: TyCtxt<'_, 'tcx, 'tcx>,
    // smallest into each other) we're sure to start off with a deterministic
    // order (sorted by name). This'll mean that if two cgus have the same size
    // the stable sort below will keep everything nice and deterministic.
-    codegen_units.sort_by_key(|cgu| cgu.name().clone());
+    codegen_units.sort_by_cached_key(|cgu| cgu.name().clone());


~~Can't it be sort_by(|a, b| a.name().cmp(b.name())) to avoid heap allocations completely?~~
name() returns a ref to InternedString which is Copy, so this clone is free.

nikomatsakis · 2018-11-16T20:03:45Z

@rust-timer build a22afbf

rust-timer · 2018-11-16T20:03:47Z

Success: Queued a22afbf with parent 9649c1f, comparison URL.

nikomatsakis

Let's see what perf says. I think that the primary_span and def_path_hash calls are likely not a win. Not sure about the others.

nikomatsakis · 2018-11-16T20:08:31Z

src/librustc/middle/resolve_lifetime.rs

@@ -1573,7 +1573,7 @@ impl<'a, 'tcx> LifetimeContext<'a, 'tcx> {
            .collect();

        // ensure that we issue lints in a repeatable order
-        def_ids.sort_by_key(|&def_id| self.tcx.def_path_hash(def_id));
+        def_ids.sort_by_cached_key(|&def_id| self.tcx.def_path_hash(def_id));


As you noted, I think this is already cached -- it's something we do a lot in incremental. You agree, @michaelwoerister ?

It's cached, yes, but still not entirely trivial. I.e. it has to do the local/upstream crate dispatch and for upstream DefIds it has to go through a number of function calls:

rust/src/librustc/ty/context.rs

Line 1325 in 780658a

pub fn def_path_hash(self, def_id: DefId) -> hir_map::DefPathHash {

So, I can imagine this being a win for sorting where each key might be compared against multiple times.

nikomatsakis · 2018-11-16T20:08:41Z

src/librustc/traits/object_safety.rs

@@ -408,7 +408,7 @@ impl<'a, 'tcx> TyCtxt<'a, 'tcx, 'tcx> {
            .collect::<Vec<_>>();

        // existential predicates need to be in a specific order
-        associated_types.sort_by_key(|item| self.def_path_hash(item.def_id));
+        associated_types.sort_by_cached_key(|item| self.def_path_hash(item.def_id));


Also here, of course

nikomatsakis · 2018-11-16T20:09:50Z

src/librustc_mir/borrow_check/mod.rs

@@ -341,7 +341,7 @@ fn do_mir_borrowck<'a, 'gcx, 'tcx>(
    }

    if !mbcx.errors_buffer.is_empty() {
-        mbcx.errors_buffer.sort_by_key(|diag| diag.span.primary_span());
+        mbcx.errors_buffer.sort_by_cached_key(|diag| diag.span.primary_span());


This is the 'error reporting' path, so it really doesn't matter... also primary_span is a very simple operation... doesn't strike me as worthwhile.

nikomatsakis · 2018-11-16T20:10:28Z

src/librustc_mir/monomorphize/partitioning.rs

@@ -985,7 +985,7 @@ fn collect_and_partition_mono_items<'a, 'tcx>(
                output.push_str(" @@");
                let mut empty = Vec::new();
                let cgus = item_to_cgus.get_mut(i).unwrap_or(&mut empty);
-                cgus.as_mut_slice().sort_by_key(|&(ref name, _)| name.clone());
+                cgus.as_mut_slice().sort_by_cached_key(|&(ref name, _)| name.clone());


This might make sense. Not sure what type name has here... :)

rust-timer · 2018-11-16T21:55:41Z

Finished benchmarking try commit a22afbf

ljedrz · 2018-11-17T08:58:47Z

Looks like a win:

inflate-check
	avg: -2.4%	min: -3.3%	max: 0.3%
keccak-check
	avg: -2.1%	min: -3.1%	max: 0.5%
keccak-opt
	avg: -1.6%	min: -2.7%	max: 0.4%
keccak-debug
	avg: -1.5%	min: -2.5%	max: 0.3%
inflate-debug
	avg: -1.6%	min: -2.3%	max: 0.3%

I can update the commit taking into account the comments - theoretically it could make the wins even nicer.

ljedrz · 2018-11-17T09:07:32Z

Not much was left after including the remarks ^^. We might want to run another round of perf to make sure we didn't lose the gains reported by the first run.

nikomatsakis · 2018-11-20T20:31:25Z

Yeah, we should try again. I'm a bit surprised to see a 3% swing, I have to admit.

nikomatsakis · 2018-11-20T20:31:30Z

@bors try

bors · 2018-11-20T20:31:42Z

⌛ Trying commit df9526808c94b90bec3599bcd25ecb089ab934c5 with merge c992107af31574b1105b4058931d9e371cfaf8ad...

bors · 2018-11-20T23:02:45Z

☀️ Test successful - status-travis
State: approved= try=True

michaelwoerister · 2018-11-21T10:02:31Z

The only change left at this point is in debugging code that is only executed by the test suite -- so I expect no wins there. The original changes (except for the cgu.name().clone()) looked fine to me though.

nikomatsakis · 2018-11-26T20:29:15Z

@rust-timer build c992107af31574b1105b4058931d9e371cfaf8ad

rust-timer · 2018-11-26T20:29:16Z

Success: Queued c992107af31574b1105b4058931d9e371cfaf8ad with parent 3991bfb, comparison URL.

rust-timer · 2018-11-26T22:22:10Z

Finished benchmarking try commit c992107af31574b1105b4058931d9e371cfaf8ad

ljedrz · 2018-11-27T08:28:36Z

Looks like @michaelwoerister was right - the original gains were lost.

I've restored the earlier changes without the Copy (that's free) and primary_span (in error reporting and pretty trivial). This should bring back the greens.

michaelwoerister · 2018-11-27T10:17:10Z

lgtm: @bors r+

bors · 2018-11-27T10:17:11Z

📌 Commit d4a6e73 has been approved by michaelwoerister

…oerister Use sort_by_cached_key when the key function is not trivial/free I'm not 100% sure about `def_path_hash` (everything it does is inlined) but it seems like a good idea at least for the rest, as they are cloning.

Rollup of 19 pull requests Successful merges: - #55011 (Add libstd Cargo feature "panic_immediate_abort") - #55821 (Use sort_by_cached_key when the key function is not trivial/free) - #56014 (add test for issue #21335) - #56131 (Assorted tweaks) - #56214 (Implement chalk unification routines) - #56216 (Add TryFrom<&[T]> for [T; $N] where T: Copy) - #56268 (Reuse the `P` in `InvocationCollector::fold_{,opt_}expr`.) - #56324 (Use raw_entry for more efficient interning) - #56336 (Clean up and streamline the pretty-printer) - #56337 (Fix const_fn ICE with non-const function pointer) - #56339 (Remove not used option) - #56341 (Rename conversion util; remove duplicate util in librustc_codegen_llvm.) - #56349 (rustc 1.30.0's linker flavor inference is a non-backwards compat change to -Clinker) - #56355 (Add inline attributes and add unit to CommonTypes) - #56360 (Optimize local linkchecker program) - #56364 (Fix panic with outlives in existential type) - #56365 (Stabilize self_struct_ctor feature.) - #56367 (Moved some feature gate tests to correct location) - #56373 (Update books)

rust-highfive assigned cramertj Nov 9, 2018

rust-highfive added the S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. label Nov 9, 2018

rust-highfive assigned eddyb and unassigned cramertj Nov 15, 2018

rust-highfive assigned nikomatsakis and unassigned eddyb Nov 15, 2018

sinkuu reviewed Nov 16, 2018

View reviewed changes

nikomatsakis requested changes Nov 16, 2018

View reviewed changes

ljedrz force-pushed the cached_key_sorts branch from 1649c2e to df95268 Compare November 17, 2018 09:05

Use sort_by_cached_key when key the function is not trivial/free

d4a6e73

ljedrz force-pushed the cached_key_sorts branch from df95268 to d4a6e73 Compare November 27, 2018 08:28

bors added S-waiting-on-bors Status: Waiting on bors to run and complete tests. Bors will change the label on completion. and removed S-waiting-on-review Status: Awaiting review from the assignee but also interested parties. labels Nov 27, 2018

kennytm mentioned this pull request Nov 27, 2018

Rollup of 9 pull requests #56276

Closed

pietroalbini mentioned this pull request Nov 28, 2018

Rollup of 29 pull requests #56325

Closed

pietroalbini mentioned this pull request Nov 29, 2018

Rollup of 11 pull requests #56353

Closed

kennytm mentioned this pull request Nov 30, 2018

Rollup of 16 pull requests #56376

Closed

kennytm mentioned this pull request Nov 30, 2018

Rollup of 19 pull requests #56381

Merged

bors merged commit d4a6e73 into rust-lang:master Nov 30, 2018

ljedrz deleted the cached_key_sorts branch November 30, 2018 22:03

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use sort_by_cached_key when the key function is not trivial/free #55821

Use sort_by_cached_key when the key function is not trivial/free #55821

ljedrz commented Nov 9, 2018

rust-highfive commented Nov 9, 2018

Mark-Simulacrum commented Nov 15, 2018

bors commented Nov 15, 2018

eddyb commented Nov 15, 2018

bors commented Nov 15, 2018

sinkuu Nov 16, 2018 •

edited

Loading

nikomatsakis commented Nov 16, 2018

rust-timer commented Nov 16, 2018

nikomatsakis left a comment

nikomatsakis Nov 16, 2018

michaelwoerister Nov 21, 2018

nikomatsakis Nov 16, 2018

nikomatsakis Nov 16, 2018

nikomatsakis Nov 16, 2018

rust-timer commented Nov 16, 2018

ljedrz commented Nov 17, 2018

ljedrz commented Nov 17, 2018

nikomatsakis commented Nov 20, 2018

nikomatsakis commented Nov 20, 2018

bors commented Nov 20, 2018

bors commented Nov 20, 2018

michaelwoerister commented Nov 21, 2018

nikomatsakis commented Nov 26, 2018

rust-timer commented Nov 26, 2018

rust-timer commented Nov 26, 2018

ljedrz commented Nov 27, 2018

michaelwoerister commented Nov 27, 2018

bors commented Nov 27, 2018

Use sort_by_cached_key when the key function is not trivial/free #55821

Use sort_by_cached_key when the key function is not trivial/free #55821

Conversation

ljedrz commented Nov 9, 2018

rust-highfive commented Nov 9, 2018

Mark-Simulacrum commented Nov 15, 2018

bors commented Nov 15, 2018

eddyb commented Nov 15, 2018

bors commented Nov 15, 2018

sinkuu Nov 16, 2018 • edited Loading

Choose a reason for hiding this comment

nikomatsakis commented Nov 16, 2018

rust-timer commented Nov 16, 2018

nikomatsakis left a comment

Choose a reason for hiding this comment

nikomatsakis Nov 16, 2018

Choose a reason for hiding this comment

michaelwoerister Nov 21, 2018

Choose a reason for hiding this comment

nikomatsakis Nov 16, 2018

Choose a reason for hiding this comment

nikomatsakis Nov 16, 2018

Choose a reason for hiding this comment

nikomatsakis Nov 16, 2018

Choose a reason for hiding this comment

rust-timer commented Nov 16, 2018

ljedrz commented Nov 17, 2018

ljedrz commented Nov 17, 2018

nikomatsakis commented Nov 20, 2018

nikomatsakis commented Nov 20, 2018

bors commented Nov 20, 2018

bors commented Nov 20, 2018

michaelwoerister commented Nov 21, 2018

nikomatsakis commented Nov 26, 2018

rust-timer commented Nov 26, 2018

rust-timer commented Nov 26, 2018

ljedrz commented Nov 27, 2018

michaelwoerister commented Nov 27, 2018

bors commented Nov 27, 2018

sinkuu Nov 16, 2018 •

edited

Loading