Optimize by introducing stages #33

torkleyy · 2017-06-04T07:07:31Z

This optimization is big! 🎉 Benchmark running time decreased by 74%.

Fixes #6

Object905 · 2017-06-04T07:16:04Z

src/lib.rs

@@ -58,18 +58,18 @@ extern crate fnv;
 extern crate mopa;
 #[cfg(not(target_os = "emscripten"))]
 extern crate pulse;
+extern crate rayon;
 #[cfg(not(target_os = "emscripten"))]
 extern crate rayon_core;


rayon_core probably not needed anymore?

Given that rayon_core is stable and rayon is not, we can just use rayon_core where possible I guess.

Well rayon itself uses rayon_core, and reexports all parts of it, so using this parts from rayon_core is all the same.

Object905 · 2017-06-04T07:28:20Z

src/dispatch/builder.rs

+    {
+        let id = self.id_gen.next();
+        let mut reads: Vec<_> = unsafe { T::SystemData::reads(0) };
+        let writes = unsafe { T::SystemData::writes(0) };


Why are these calls unsafe? Yes if returned ids are invalid it will cause a undefined behavior, but i doesn't mean that it must be marked as unsafe.

You're right, this should be changed. My initial version of shred didn't check borrow rules in release mode, but it does now..let's remove the unsafe.

It won't be undefined behaviour. If it was, unsafe would be just right.

Actually data race is undefined behavior.

Yes, it is, but we don't have any data races, it panics.

This optimization is big! 🎉 Benchmark running time decreased by 74%

WaDelma

Nice to see so good speed up!

WaDelma · 2017-06-04T10:24:40Z

src/dispatch/builder.rs

+    }
+
+    /// Adds a new system with a given name and a list of dependencies.
+    /// Please not that the dependency should be added before


WaDelma · 2017-06-04T10:26:29Z

src/dispatch/builder.rs

+        reads.sort();
+        reads.dedup();
+
+        let mut reads = SmallVec::from_vec(reads);


Is there a reason that SystemData::reads and SystemData::writes don't return SmallVecs?

Two reasons:

Didn't want to make it a public dependency

The system data may read from a resource several times. Therefore, the stack storage would often be too small at first, but not necessary anymore later. That's why I do the conversion after removing duplicates

I didn't yet check if it makes sense to use SmallVec for reads / writes at all. It's only used for building the stages, after all.

WaDelma · 2017-06-04T10:27:24Z

src/dispatch/builder.rs

+    {
+        let id = self.id_gen.next();
+        let mut reads: Vec<_> = T::SystemData::reads(0);
+        let writes = T::SystemData::writes(0);


Shouldn't these 0s be id instead? Or am I missing something?

Thanks for reviewing, @WaDelma!

My code is correct, however it's not obvious why I did it like that, because I never wrote my design decision down.

First of all, in case you don't know, system data most likely consists of other structs implementing SystemData, so for example:

#[derive(SystemData)] struct Data<'a> { a: ReadStorage<'a, A>, b: WriteStorage<'a, B>, }

Now, I'd like resources to have additional ids (in Specs these were called component ids). So fetch / reads / writes should be parameterized over this resource id. However, the system data directly associated with a system, doesn't need such an id. It just gets passed 0 for that reason.

Shouldn't these 0s be id instead?

The id you're seeing here is the system id, which is totally unrelated to the resource id.

WaDelma · 2017-06-04T10:33:01Z

src/dispatch/builder.rs

+        let dependencies: SystemDependencies = dep.iter()
+            .map(|x| {
+                     *self.map
+                          .get(x.to_owned())


Because how HashMap::get is defined &str can be used for HashMap<String, _>::get and to_owned is unnessary here.

torkleyy · 2017-06-04T11:33:47Z

I got another idea how to optimize this even more, hold on.

torkleyy · 2017-06-05T15:25:11Z

Yeah! I finally got it finished.

kvark · 2017-06-05T17:42:22Z

src/dispatch/stage.rs

+
+        (self.barrier..self.stages.len())
+            .map(|stage| {
+                let conflict = Self::find_conflict(&*self.ids,


any reason this is not just self.find_conflict(stage, new_reads.clone(), new_writes.clone(), new_dep)?

Yes, easier to test.

kvark · 2017-06-05T17:43:13Z

src/dispatch/stage.rs

+            .find(|&(stage, conflict)| match conflict {
+                      Conflict::None => true,
+                      Conflict::Single(group) => {
+                          self.stages[stage].groups[group].len() < MAX_SYSTEMS_PER_GROUP - 1 &&


why MAX_SYSTEMS_PER_GROUP - 1 ?

The ArrayVec only supports up to MAX_SYSTEMS_PER_GROUP, so any additional system would just be discarded. So I check if there's enough space left.

kvark · 2017-06-05T17:43:35Z

src/dispatch/stage.rs

+            .map(|(stage, conflict)| match conflict {
+                     Conflict::None => InsertionTarget::Stage(stage),
+                     Conflict::Single(group) => InsertionTarget::Group(stage, group),
+                     _ => unreachable!(),


let's have Conflict::Multiple here explicitly instead of _

kvark · 2017-06-05T17:50:47Z

src/dispatch/stage.rs

+    fn remove_ids(&self, stage: usize, new_dep: &mut SmallVec<[SystemId; 4]>) {
+        if !new_dep.is_empty() {
+            for id in self.ids[stage].iter().flat_map(|id_group| id_group) {
+                if let Some(index) = new_dep.iter().position(|x| *x == *id) {


is there SmallVec::retain()?

Wanted to use that, but no, unfortunately there isn't.

Add rayon and arrayvec dependencies Removes rayon_core Add RunningTime and System::running_time This allows for further optimizations. Do not rustfmt macro invocations in system module Reexport RunningTime Removed debug printlns Remove typ annotation Make match exhaustive

torkleyy · 2017-06-05T18:03:52Z

bors r+

Thanks @Object905 @WaDelma @kvark for reviewing!

33: Optimize by introducing stages r=torkleyy This optimization is big! 🎉 Benchmark running time decreased by 74%. Fixes #6

bors · 2017-06-05T18:06:46Z

Build succeeded

continuous-integration/travis-ci/push

166: Update to shred 0.4 r=torkleyy Now specs can make use of [a major optimization](amethyst/shred#33) in `shred` and [a bugfix](amethyst/shred#35) in `shred-derive`. Also improves bench by specifying running time.

126: Update hashbrown requirement from 0.1.7 to 0.2.0 r=torkleyy a=dependabot[bot] Updates the requirements on [hashbrown](https://github.com/Amanieu/hashbrown) to permit the latest version. <details> <summary>Changelog</summary> *Sourced from [hashbrown's changelog](https://github.com/Amanieu/hashbrown/blob/master/CHANGELOG.md).* > ## [v0.2.0] - 2019-03-31 > > ### Changed > - The code has been updated to Rust 2018 edition. This means that the minimum > Rust version has been bumped to 1.31 (2018 edition). > > ### Added > - Added `insert_with_hasher` to the raw_entry API to allow `K: !(Hash + Eq)`. ([#54](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/54)) > - Added support for using hashbrown as the hash table implementation in libstd. ([#46](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/46)) > > ### Fixed > - Fixed cargo build with minimal-versions. ([#45](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/45)) > - Fixed `#[may_dangle]` attributes to match the libstd `HashMap`. ([#46](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/46)) > - ZST keys and values are now handled properly. ([#46](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/46)) > > ## [v0.1.8] - 2019-01-14 > > ### Added > - Rayon parallel iterator support ([#37](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/37)) > - `raw_entry` support ([#31](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/31)) > - `#[may_dangle]` on nightly ([#31](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/31)) > - `try_reserve` support ([#31](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/31)) > > ### Fixed > - Fixed variance on `IterMut`. ([#31](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/31)) > > ## [v0.1.7] - 2018-12-05 > > ### Fixed > - Fixed non-SSE version of convert_special_to_empty_and_full_to_deleted. ([#32](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/32)) > - Fixed overflow in rehash_in_place. ([#33](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/33)) > > ## [v0.1.6] - 2018-11-17 > > ### Fixed > - Fixed compile error on nightly. ([#29](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/29)) > > ## [v0.1.5] - 2018-11-08 > > ### Fixed > - Fixed subtraction overflow in generic::Group::match_byte. ([#28](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/28)) > > ## [v0.1.4] - 2018-11-04 > > ### Fixed > - Fixed a bug in the `erase_no_drop` implementation. ([#26](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/26)) > > ## [v0.1.3] - 2018-11-01 > > ### Added ></tr></table> ... (truncated) </details> <details> <summary>Commits</summary> - [`af90bc9`](Amanieu/hashbrown@af90bc9) Version 0.2.0 - [`83ce3d5`](Amanieu/hashbrown@83ce3d5) Merge [#46](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/46) - [`f9bcc42`](Amanieu/hashbrown@f9bcc42) Add #[may_dangle] to RawIntoIter - [`b11f8cd`](Amanieu/hashbrown@b11f8cd) Upgrade to 2018 edition - [`2bd15ed`](Amanieu/hashbrown@2bd15ed) Apply feedback from https://github-redirect.dependabot.com/rust-lang/rust/pull/56241 - [`91d3e4a`](Amanieu/hashbrown@91d3e4a) Merge [#55](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/55) - [`49a9d6c`](Amanieu/hashbrown@49a9d6c) use --exclude-should-panic - [`c36d49a`](Amanieu/hashbrown@c36d49a) install Miri via rustup - [`332db4d`](Amanieu/hashbrown@332db4d) Prepare hashbrown for inclusion in the standard library - [`edf79d0`](Amanieu/hashbrown@edf79d0) Merge [#54](https://github-redirect.dependabot.com/Amanieu/hashbrown/issues/54) - Additional commits viewable in [compare view](Amanieu/hashbrown@v0.1.7...v0.2.0) </details> <br /> Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting `@dependabot rebase`. [//]: # (dependabot-automerge-start) [//]: # (dependabot-automerge-end) --- <details> <summary>Dependabot commands and options</summary> <br /> You can trigger Dependabot actions by commenting on this PR: - `@dependabot rebase` will rebase this PR - `@dependabot recreate` will recreate this PR, overwriting any edits that have been made to it - `@dependabot merge` will merge this PR after your CI passes on it - `@dependabot squash and merge` will squash and merge this PR after your CI passes on it - `@dependabot cancel merge` will cancel a previously requested merge and block automerging - `@dependabot reopen` will reopen this PR if it is closed - `@dependabot ignore this [patch|minor|major] version` will close this PR and stop Dependabot creating any more for this minor/major version (unless you reopen the PR or upgrade to it yourself) - `@dependabot ignore this dependency` will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself) - `@dependabot use these labels` will set the current labels as the default for future PRs for this repo and language - `@dependabot use these reviewers` will set the current reviewers as the default for future PRs for this repo and language - `@dependabot use these assignees` will set the current assignees as the default for future PRs for this repo and language - `@dependabot use this milestone` will set the current milestone as the default for future PRs for this repo and language - `@dependabot badge me` will comment on this PR with code to add a "Dependabot enabled" badge to your readme Additionally, you can set the following in your Dependabot [dashboard](https://app.dependabot.com): - Update frequency (including time of day and day of week) - Automerge options (never/patch/minor, and dev/runtime dependencies) - Pull request limits (per update run and/or open at any time) - Out-of-range updates (receive only lockfile updates, if desired) - Security updates (receive only security updates, if desired) Finally, you can contact us by mentioning @dependabot. </details> Co-authored-by: dependabot[bot] <[email protected]>

torkleyy requested a review from kvark June 4, 2017 07:09

Object905 reviewed Jun 4, 2017

View reviewed changes

Optimize by introducing stages

2f04b69

This optimization is big! 🎉 Benchmark running time decreased by 74%

torkleyy force-pushed the opt branch from cac5bba to 2f04b69 Compare June 4, 2017 07:53

WaDelma reviewed Jun 4, 2017

View reviewed changes

torkleyy added the in progress label Jun 4, 2017

torkleyy force-pushed the opt branch from acf8292 to ec2e01b Compare June 5, 2017 15:21

torkleyy removed the in progress label Jun 5, 2017

torkleyy force-pushed the opt branch 2 times, most recently from 457f86c to f8ae782 Compare June 5, 2017 16:55

kvark approved these changes Jun 5, 2017

View reviewed changes

torkleyy force-pushed the opt branch from f8ae782 to e9acf68 Compare June 5, 2017 17:59

bors bot added a commit that referenced this pull request Jun 5, 2017

Merge #33

832d2f2

33: Optimize by introducing stages r=torkleyy This optimization is big! 🎉 Benchmark running time decreased by 74%. Fixes #6

bors bot merged commit e9acf68 into amethyst:master Jun 5, 2017

torkleyy deleted the opt branch June 5, 2017 18:15

torkleyy mentioned this pull request Jun 6, 2017

Update to shred 0.4 amethyst/specs#166

Merged

torkleyy mentioned this pull request Jun 7, 2017

Registering systems with the same name should panic #38

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Optimize by introducing stages #33

Optimize by introducing stages #33

torkleyy commented Jun 4, 2017 •

edited

Loading

Object905 Jun 4, 2017

torkleyy Jun 4, 2017

Object905 Jun 4, 2017

Object905 Jun 4, 2017

torkleyy Jun 4, 2017 •

edited

Loading

torkleyy Jun 4, 2017

Object905 Jun 4, 2017

torkleyy Jun 4, 2017

WaDelma left a comment

WaDelma Jun 4, 2017

WaDelma Jun 4, 2017

torkleyy Jun 4, 2017

WaDelma Jun 4, 2017

torkleyy Jun 4, 2017

WaDelma Jun 4, 2017

torkleyy commented Jun 4, 2017

torkleyy commented Jun 5, 2017

kvark Jun 5, 2017

torkleyy Jun 5, 2017

kvark Jun 5, 2017

torkleyy Jun 5, 2017

kvark Jun 5, 2017

kvark Jun 5, 2017

torkleyy Jun 5, 2017

torkleyy commented Jun 5, 2017

bors bot commented Jun 5, 2017

Optimize by introducing stages #33

Optimize by introducing stages #33

Conversation

torkleyy commented Jun 4, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

torkleyy Jun 4, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

WaDelma left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

torkleyy commented Jun 4, 2017

torkleyy commented Jun 5, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

torkleyy commented Jun 5, 2017

bors bot commented Jun 5, 2017

Build succeeded

torkleyy commented Jun 4, 2017 •

edited

Loading

torkleyy Jun 4, 2017 •

edited

Loading