[Merged by Bors] - Split Component Ticks #6547
Closed
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
james7132
added
A-ECS
|
Quick answers:
Confirmed by checking the same test code's output used in #6461. It's identical save for the offset calculation on the This PR .LBB1_5:
movss (%rdi,%rbx,4), %xmm0
movl %r8d, (%rsi,%rbx,4)
addss (%r11,%rbx,4), %xmm0
movss %xmm0, (%r11,%rbx,4)
incq %rbx
.LBB1_1:
cmpq %rax, %rbx
jne .LBB1_5
.p2align 4, 0x90The output from #6461 .LBB5_5:
movss (%rbx,%rbp,4), %xmm0
movl %r14d, 4(%r8,%rbp,8)
addss (%rdi,%rbp,4), %xmm0
movss %xmm0, (%rdi,%rbp,4)
incq %rbp
.LBB5_1:
cmpq %rdx, %rbp
jne .LBB5_5
.p2align 4, 0x90 |
|
Finalized a set of microbenchmarks. Looks like there's a massive improvement in iteration times where there are mutable queries, slight improvements in overall change detection, and small regressions in commands. |
|
Sans the API surface, this is ready for review. |
| } | ||
|
|
||
| #[inline] | ||
| pub fn get_ticks(&self, row: usize) -> Option<ComponentTicks> { |
There was a problem hiding this comment.
Comments on these methods about how the row values works and why this might return None would be nice. Won't block on it though; this is no worse than before.
There was a problem hiding this comment.
I'll go ahead and make a separate PR for this.
|
Did some further investigation on why the perf difference grew bigger between The following is the same hot section of code in the original post, but in .LBB2_13:
movups (%rdi,%rax,4), %xmm1
movups 16(%rdi,%rax,4), %xmm2
movdqu %xmm0, (%rsi,%rax,4)
movdqu %xmm0, 16(%rsi,%rax,4)
movups (%rdx,%rax,4), %xmm3
addps %xmm1, %xmm3
movups 16(%rdx,%rax,4), %xmm1
addps %xmm2, %xmm1
movups %xmm3, (%rdx,%rax,4)
movups %xmm1, 16(%rdx,%rax,4)
movups 32(%rdi,%rax,4), %xmm1
movups 48(%rdi,%rax,4), %xmm2
movdqu %xmm0, 32(%rsi,%rax,4)
movdqu %xmm0, 48(%rsi,%rax,4)
movups 32(%rdx,%rax,4), %xmm3
addps %xmm1, %xmm3
movups 48(%rdx,%rax,4), %xmm1
addps %xmm2, %xmm1
movups %xmm3, 32(%rdx,%rax,4)
movups %xmm1, 48(%rdx,%rax,4)
addq $16, %rax
addq $-2, %rcx
jne .LBB2_13
testb $1, %r15b
je .LBB2_16 |
Diddykonga
approved these changes
Nov 14, 2022
| /// component_ticks.set_changed(world.read_change_tick()); | ||
| /// ``` | ||
| #[inline] | ||
| pub fn set_changed(&mut self, change_tick: u32) { |
There was a problem hiding this comment.
Is this needed, what if it is an Added Tick?
Also the Comments indirectly use the method they are documenting.
I would prefer direct access of the field.
Co-authored-by: PROMETHIA-27 <[email protected]>
james7132
added
the
S-Ready-For-Final-Review
|
bors try |
|
bors r+ |
bors bot
pushed a commit
that referenced
this pull request
Nov 21, 2022
# Objective Fixes #4884. `ComponentTicks` stores both added and changed ticks contiguously in the same 8 bytes. This is convenient when passing around both together, but causes half the bytes fetched from memory for the purposes of change detection to effectively go unused. This is inefficient when most queries (no filter, mutating *something*) only write out to the changed ticks. ## Solution Split the storage for change detection ticks into two separate `Vec`s inside `Column`. Fetch only what is needed during iteration. This also potentially also removes one blocker from autovectorization of dense queries. EDIT: This is confirmed to enable autovectorization of dense queries in `for_each` and `par_for_each` where possible. Unfortunately `iter` has other blockers that prevent it. ### TODO - [x] Microbenchmark - [x] Check if this allows query iteration to autovectorize simple loops. - [x] Clean up all of the spurious tuples now littered throughout the API ### Open Questions - ~~Is `Mut::is_added` absolutely necessary? Can we not just use `Added` or `ChangeTrackers`?~~ It's optimized out if unused. - ~~Does the fetch of the added ticks get optimized out if not used?~~ Yes it is. --- ## Changelog Added: `Tick`, a wrapper around a single change detection tick. Added: `Column::get_added_ticks` Added: `Column::get_column_ticks` Added: `SparseSet::get_added_ticks` Added: `SparseSet::get_column_ticks` Changed: `Column` now stores added and changed ticks separately internally. Changed: Most APIs returning `&UnsafeCell<ComponentTicks>` now returns `TickCells` instead, which contains two separate `&UnsafeCell<Tick>` for either component ticks. Changed: `Query::for_each(_mut)`, `Query::par_for_each(_mut)` will now leverage autovectorization to speed up query iteration where possible. ## Migration Guide TODO
|
Pull request successfully merged into main. Build succeeded:
|
bors
bot
changed the title
This was referenced Nov 23, 2022
bors bot
pushed a commit
that referenced
this pull request
Dec 10, 2022
# Objective #6547 accidentally broke change detection for SparseSet components by using `Ticks::from_tick_cells` with the wrong argument order. ## Solution Use the right argument order. Add a regression test.
taiyoungjang
pushed a commit
to taiyoungjang/bevy
that referenced
this pull request
Dec 15, 2022
# Objective Fixes bevyengine#4884. `ComponentTicks` stores both added and changed ticks contiguously in the same 8 bytes. This is convenient when passing around both together, but causes half the bytes fetched from memory for the purposes of change detection to effectively go unused. This is inefficient when most queries (no filter, mutating *something*) only write out to the changed ticks. ## Solution Split the storage for change detection ticks into two separate `Vec`s inside `Column`. Fetch only what is needed during iteration. This also potentially also removes one blocker from autovectorization of dense queries. EDIT: This is confirmed to enable autovectorization of dense queries in `for_each` and `par_for_each` where possible. Unfortunately `iter` has other blockers that prevent it. ### TODO - [x] Microbenchmark - [x] Check if this allows query iteration to autovectorize simple loops. - [x] Clean up all of the spurious tuples now littered throughout the API ### Open Questions - ~~Is `Mut::is_added` absolutely necessary? Can we not just use `Added` or `ChangeTrackers`?~~ It's optimized out if unused. - ~~Does the fetch of the added ticks get optimized out if not used?~~ Yes it is. --- ## Changelog Added: `Tick`, a wrapper around a single change detection tick. Added: `Column::get_added_ticks` Added: `Column::get_column_ticks` Added: `SparseSet::get_added_ticks` Added: `SparseSet::get_column_ticks` Changed: `Column` now stores added and changed ticks separately internally. Changed: Most APIs returning `&UnsafeCell<ComponentTicks>` now returns `TickCells` instead, which contains two separate `&UnsafeCell<Tick>` for either component ticks. Changed: `Query::for_each(_mut)`, `Query::par_for_each(_mut)` will now leverage autovectorization to speed up query iteration where possible. ## Migration Guide TODO
alradish
pushed a commit
to alradish/bevy
that referenced
this pull request
Jan 22, 2023
# Objective Fixes bevyengine#4884. `ComponentTicks` stores both added and changed ticks contiguously in the same 8 bytes. This is convenient when passing around both together, but causes half the bytes fetched from memory for the purposes of change detection to effectively go unused. This is inefficient when most queries (no filter, mutating *something*) only write out to the changed ticks. ## Solution Split the storage for change detection ticks into two separate `Vec`s inside `Column`. Fetch only what is needed during iteration. This also potentially also removes one blocker from autovectorization of dense queries. EDIT: This is confirmed to enable autovectorization of dense queries in `for_each` and `par_for_each` where possible. Unfortunately `iter` has other blockers that prevent it. ### TODO - [x] Microbenchmark - [x] Check if this allows query iteration to autovectorize simple loops. - [x] Clean up all of the spurious tuples now littered throughout the API ### Open Questions - ~~Is `Mut::is_added` absolutely necessary? Can we not just use `Added` or `ChangeTrackers`?~~ It's optimized out if unused. - ~~Does the fetch of the added ticks get optimized out if not used?~~ Yes it is. --- ## Changelog Added: `Tick`, a wrapper around a single change detection tick. Added: `Column::get_added_ticks` Added: `Column::get_column_ticks` Added: `SparseSet::get_added_ticks` Added: `SparseSet::get_column_ticks` Changed: `Column` now stores added and changed ticks separately internally. Changed: Most APIs returning `&UnsafeCell<ComponentTicks>` now returns `TickCells` instead, which contains two separate `&UnsafeCell<Tick>` for either component ticks. Changed: `Query::for_each(_mut)`, `Query::par_for_each(_mut)` will now leverage autovectorization to speed up query iteration where possible. ## Migration Guide TODO
alradish
pushed a commit
to alradish/bevy
that referenced
this pull request
Jan 22, 2023
# Objective bevyengine#6547 accidentally broke change detection for SparseSet components by using `Ticks::from_tick_cells` with the wrong argument order. ## Solution Use the right argument order. Add a regression test.
ItsDoot
pushed a commit
to ItsDoot/bevy
that referenced
this pull request
Feb 1, 2023
# Objective Fixes bevyengine#4884. `ComponentTicks` stores both added and changed ticks contiguously in the same 8 bytes. This is convenient when passing around both together, but causes half the bytes fetched from memory for the purposes of change detection to effectively go unused. This is inefficient when most queries (no filter, mutating *something*) only write out to the changed ticks. ## Solution Split the storage for change detection ticks into two separate `Vec`s inside `Column`. Fetch only what is needed during iteration. This also potentially also removes one blocker from autovectorization of dense queries. EDIT: This is confirmed to enable autovectorization of dense queries in `for_each` and `par_for_each` where possible. Unfortunately `iter` has other blockers that prevent it. ### TODO - [x] Microbenchmark - [x] Check if this allows query iteration to autovectorize simple loops. - [x] Clean up all of the spurious tuples now littered throughout the API ### Open Questions - ~~Is `Mut::is_added` absolutely necessary? Can we not just use `Added` or `ChangeTrackers`?~~ It's optimized out if unused. - ~~Does the fetch of the added ticks get optimized out if not used?~~ Yes it is. --- ## Changelog Added: `Tick`, a wrapper around a single change detection tick. Added: `Column::get_added_ticks` Added: `Column::get_column_ticks` Added: `SparseSet::get_added_ticks` Added: `SparseSet::get_column_ticks` Changed: `Column` now stores added and changed ticks separately internally. Changed: Most APIs returning `&UnsafeCell<ComponentTicks>` now returns `TickCells` instead, which contains two separate `&UnsafeCell<Tick>` for either component ticks. Changed: `Query::for_each(_mut)`, `Query::par_for_each(_mut)` will now leverage autovectorization to speed up query iteration where possible. ## Migration Guide TODO
ItsDoot
pushed a commit
to ItsDoot/bevy
that referenced
this pull request
Feb 1, 2023
# Objective bevyengine#6547 accidentally broke change detection for SparseSet components by using `Ticks::from_tick_cells` with the wrong argument order. ## Solution Use the right argument order. Add a regression test.
bors bot
pushed a commit
to bevyengine/bevy-website
that referenced
this pull request
Mar 6, 2023
## How This Works For the Bevy 0.10 release blog post (and for the first time ever), I'm publicly opening the doors to other people writing blog post sections. Specifically, if you worked on a feature in a substantial way and are interested in presenting it, you can now ask to claim a section by leaving a comment in this PR. If you claim a section, submit a pull request to the `release-0.10.0` branch in this repo. For the next week, we will be filling in sections (the release target is Saturday March 4th). Please don't claim a section if you don't plan on completing it within that timeline. Also don't claim a section if you weren't an active participant in the design and implementation of the change (unless you are a Maintainer or SME). I will claim any unclaimed sections. Try to match the style of previous release blog posts as much as possible. 1. Show, don't tell. Don't bombard people with information. Avoid large walls of text _and_ large walls of code. Prefer the pattern "byte sized description of one thing" -> "example code/picture/video contextualizing that one thing" -> repeat. Take readers on a journey step by simple step. 2. Don't use up reader's "mental bandwidth" without good reason. We can't afford page-long descriptions of minor bug fixes. If it isn't a "headliner change", keep the description short and sweet. If a change is self describing, let it do that (ex: We now support this new mesh shape primitive ... this is what it looks like). If it is a "headliner change", still try to keep it reasonable. We always have a lot to cover. 3. In slight competition with point (2), don't omit interesting technical information when it is truly fun and engaging. A good chunk of our users are highly technical and enjoy learning how the sausage is made. Try to strike a balance between "terse and simple" and "nerdy details". 4. When relevant, briefly describe the problem being solved first, then describe the solution we chose. This contextualizes the change and gives the feature value and purpose. 5. When possible, provide visuals. They create interest / keep people hooked / break up the monotony. 6. Record images and videos at the default bevy resolution (1280x720) 7. Provide an accurate listing of authors that meaningfully contributed to the feature. Try to sort in order of "contribution scale". This is hard to define, but try to be fair. When in doubt, ask other contributors, SMEs, and/or maintainers. 8. Provide numbers and graphs where possible. If something is faster, use numbers to back it up. We don't (yet) have automated graph generation in blog post style, so send data / info to me (@cart) if you want a graph made. ## Headliners Headliners are our "big ticket high importance / high profile" changes. They are listed briefly at the beginning of the blog post, their entries are roughly sorted "to the top", and they are given priority when it comes to "space in the blog post". If you think we missed something (or didn't prioritize something appropriately), let us know. * ECS Schedule v3 (previously known as "stageless") * Partial Android Support * Depth and Normal Prepass * Environment Map Lighting * Cascaded Shadow Maps * Distance and Atmospheric Fog * Smooth Skeletal Animation Transitions * Enable Parallel Pipelined Rendering * Windows as Entities * Renderer Optimizations * ECS Optimizations ## Sections These are the sections we will cover in the blog post. If a section has been claimed, it will have `(claimed by X)` in the title. If it is unclaimed it will have `(unclaimed)` in the title. Let us know if we missed a section. We don't cover every feature, but we should cover pretty much everything that would be interesting to users. Note that what is interesting or challenging to implement is not necessarily something that is relevant to our blog post readers. And sometimes the reverse is true! If you believe a section should be split up or reorganized, just bring it up here and we can discuss it. ### ~~Schedule V3 (claimed by @alice-i-cecile)~~ * [Migrate engine to Schedule v3][7267] * [Add `bevy_ecs::schedule_v3` module][6587] * [Stageless: fix unapplied systems][7446] * [Stageless: move final apply outside of spawned executor][7445] * Sets * Base Sets * [Base Sets][7466] * Reporting * [Report sets][7756] * [beter cycle reporting][7463] * Run Conditions * [Add condition negation][7559] * [And/Or][7605] * [Add more common run conditions][7579] * States * [States derive macro][7535] * System Piping Flexibility * [Support piping exclusive systems][7023] * [Allow piping run conditions][7547] ### ~~Depth and Normal Prepass (claimed by @IceSentry)~~ * [Add depth and normal prepass][6284] * [Move prepass functions to prepass_utils][7354] ### ~~Distance and Atmospheric Fog (claimed by @coreh)~~ * [Add Distance and Atmospheric Fog support][6412] ### ~~Cascaded Shadow Maps (claimed by @cart)~~ * [Cascaded shadow maps.][7064] * [Better cascades config defaults + builder, tweak example configs][7456] ### ~~Environment Map Lighting (claimed by @cart)~~ * [EnvironmentMapLight, BRDF Improvements][7051] * [Webgl2 support][7737] ### ~~Tonemapping options (claimed by @cart)~~ * [Initial tonemapping options][7594] ### ~~Android support + unification (claimed by @mockersf)~~ * [IOS, Android... same thing][7493] ### ~~Windows as Entities (claimed by @Aceeri)~~ * [Windows as Entities][5589] * [break feedback loop when moving cursor][7298] * [Fix `Window` feedback loop between the OS and Bevy][7517] ### ~~Enable Parallel Pipelined Rendering (claimed by @james7132)~~ * [Pipelined Rendering][6503] * [Stageless: add a method to scope to always run a task on the scope thread][7415] * [Separate Extract from Sub App Schedule][7046] ### ~~Smooth Skeletal Animation Transitions (claimed by @james7132)~~ * [Smooth Transition between Animations][6922] ### ~~Spatial Audio (claimed by @harudagondi)~~ * [Spatial Audio][6028] ### ~~Shader Processor Features (claimed by @cart)~~ * [Shader defs can now have a value][5900] * [Shaders can now have #else ifdef chains][7431] * [Define shader defs in shader][7518] ### ~~Shader Flexibility Improvements (claimed by @cart)~~ * [add ambient lighting hook][5428] * [Refactor Globals and View structs into separate shaders][7512] ### ~~Renderer Optimizations (claimed by @james7132)~~ * [bevy_pbr: Avoid copying structs and using registers in shaders][7069] * [Flatten render commands][6885] * [Replace UUID based IDs with a atomic-counted ones][6988] * [improve compile time by type-erasing wgpu structs][5950] * [Shrink DrawFunctionId][6944] * [Shrink ComputedVisibility][6305] * [Reduce branching in TrackedRenderPass][7053] * [Make PipelineCache internally mutable.][7205] * [Improve `Color::hex` performance][6940] * [Support recording multiple CommandBuffers in RenderContext][7248] * [Parallelized transform propagation][4775] * [Introduce detailed_trace macro, use in TrackedRenderPass][7639] * [Optimize color computation in prepare_uinodes][7311] * [Directly extract joints into SkinnedMeshJoints][6833] * [Parallelize forward kinematics animation systems][6785] * [Move system_commands spans into apply_buffers][6900] * [Reduce the use of atomics in the render phase][7084] ### ~~ECS Optimizations (claimed by @james7132 )~~ * [Remove redundant table and sparse set component IDs from Archetype][4927] * [Immutable sparse sets for metadata storage][4928] * [Replace BlobVec's swap_scratch with a swap_nonoverlapping][4853] * [Use T::Storage::STORAGE_TYPE to optimize out unused branches][6800] * [Remove unnecessary branching from bundle insertion][6902] * [Split Component Ticks][6547] * [use bevy_utils::HashMap for better performance. TypeId is predefined …][7642] * [Extend EntityLocation with TableId and TableRow][6681] * [Basic adaptive batching for parallel quer- [Speed up `CommandQueue` by storing commands more densely][6391]y iteration][4777] ### ~~Reflect Improvements (claimed by @cart)~~ * [bevy_reflect: Add `ReflectFromReflect` (v2)][6245] * [Add reflection support for VecDeque][6831] * [reflect: add `insert` and `remove` methods to `List`][7063] * [Add `remove` method to `Map` reflection trait.][6564] * [bevy_reflect: Fix binary deserialization not working for unit structs][6722] * [Add `TypeRegistrationDeserializer` and remove `BorrowedStr`][7094] * [bevy_reflect: Add simple enum support to reflection paths][6560] * [Enable deriving Reflect on structs with generic types][7364] * [bevy_reflect: Support tuple reflection paths][7324] * [bevy_reflect: Pre-parsed paths][7321] * [bevy_ecs: ReflectComponentFns without World][7206] ### ~~AsBindGroup Improvements (claimed by @cart)~~ * [Support storage buffers in derive `AsBindGroup`][6129] * [Support raw buffers in AsBindGroup][7701] ### ~~Cylinder Shape (claimed by @cart)~~ * [Add cylinder shape][6809] ### ~~Subdividable Plane Shape (claimed by @cart)~~ * [added subdivisions to shape::Plane][7546] ### ~~StandardMaterial Blend Modes (claimed by @coreh)~~ * [Standard Material Blend Modes][6644] ### ~~Configurable Visibility Component (claimed by @cart)~~ * [enum `Visibility` component][6320] ### Task Improvements (claimed by @cart) * [Fix panicking on another scope][6524] * [Add thread create/destroy callbacks to TaskPool][6561] * [Thread executor for running tasks on specific threads.][7087] * [await tasks to cancel][6696] * [Stageless: move MainThreadExecutor to schedule_v3][7444] * [Stageless: close the finish channel so executor doesn't deadlock][7448] ### ~~Upgrade to wgpu 0.15 (claimed by @cart)~~ * [Wgpu 0.15][7356] ### ~~Expose Bindless / Non-uniform Indexing Support (claimed by @cart)~~ * [Request WGPU Capabilities for Non-uniform Indexing][6995] ### ~~Cubic Spline (claimed by @aevyrie)~~ * [Bezier][7653] ### ~~Revamp Bloom (claimed by @JMS55)~~ * [Revamp bloom](bevyengine/bevy#6677) ### ~~Use Prepass Shaders for Shadows (claimed by @superdump)~~ * [use prepass shaders for shadows](bevyengine/bevy#7784) ### ~~AccessKit (claimed by @alice-i-cecile)~~ * [accesskit](bevyengine/bevy#6874) ### ~~Camera Output Modes (claimed by @cart)~~ * [camera output modes](bevyengine/bevy#7671) ### ~~SystemParam Improvements (claimed by @JoJoJet)~~ * [Make the `SystemParam` derive macro more flexible][6694] * [Add a `SystemParam` primitive for deferred mutations; allow `#[derive]`ing more types of SystemParam][6817] ### ~~Gamepad Improvements (claimed by @cart)~~ * [Gamepad events refactor][6965] * [add `Axis::devices` to get all the input devices][5400] ### ~~Input Methods (claimed by @cart)~~ * [add Input Method Editor support][7325] ### ~~Color Improvements (claimed by @cart)~~ * [Add LCH(ab) color space to `bevy_render::color::Color`][7483] * [Add a more familiar hex color entry][7060] ### ~~Split Up CorePlugin (claimed by @cart)~~ * [Break `CorePlugin` into `TaskPoolPlugin`, `TypeRegistrationPlugin`, `FrameCountPlugin`.][7083] ### ~~ExtractComponent Derive (claimed by @cart)~~ * [Extract component derive][7399] ### ~~Added OpenGL and DX11 Backends By Default (claimed by @cart)~~ * [add OpenGL and DX11 backends][7481] ### ~~UnsafeWorldCell (claimed by @BoxyUwU)~~ * [Move all logic to `UnsafeWorldCell`][7381] * [Rename `UnsafeWorldCellEntityRef` to `UnsafeEntityCell`][7568] ### ~~Entity Commands (claimed by @cart)~~ * [Add a trait for commands that run for a given `Entity`][7015] * [Add an extension trait to `EntityCommands` to update hierarchy while preserving `GlobalTransform`][7024] * [Add ReplaceChildren and ClearChildren EntityCommands][6035] ### ~~Iterate EntityRef (claimed by @james7132)~~ * [Allow iterating over with EntityRef over the entire World][6843] ### ~~Ref Queries (@JoJoJet)~~ * [Added Ref to allow immutable access with change detection][7097] ### ~~Taffy Upgrade (claimed by @cart)~~ * [Upgrade to Taffy 0.2][6743] ### ~~Relative Cursor Position (claimed by @cart)~~ * [Relative cursor position][7199] ### ~~Const UI Config (claimed by @cart)~~ * [Add const to methods and const defaults to bevy_ui][5542] ### ~~Examples (claimed by @cart)~~ * [Add pixelated Bevy to assets and an example][6408] * [Organized scene_viewer into plugins for reuse and organization][6936] ### ~~CI Improvements (claimed by @cart)~~ * [add rust-version for MSRV and CI job to check][6852] * [msrv: only send a message on failure during the actual msrv part][7532] * [Make CI friendlier][7398] * [Fix CI welcome message][7428] * [add an action to ask for a migration guide when one is missing][7507] ### ~~SMEs (@cart)~~ This was already covered in another blog post. Just briefly call out what they are and that this is the first release that used them. Link to the other blog post. * [Subject Matter Experts and new Bevy Org docs][7185] [4775]: bevyengine/bevy#4775 [4777]: bevyengine/bevy#4777 [4853]: bevyengine/bevy#4853 [4927]: bevyengine/bevy#4927 [4928]: bevyengine/bevy#4928 [5400]: bevyengine/bevy#5400 [5428]: bevyengine/bevy#5428 [5542]: bevyengine/bevy#5542 [5589]: bevyengine/bevy#5589 [5900]: bevyengine/bevy#5900 [5950]: bevyengine/bevy#5950 [6028]: bevyengine/bevy#6028 [6035]: bevyengine/bevy#6035 [6129]: bevyengine/bevy#6129 [6179]: bevyengine/bevy#6179 [6245]: bevyengine/bevy#6245 [6284]: bevyengine/bevy#6284 [6305]: bevyengine/bevy#6305 [6320]: bevyengine/bevy#6320 [6391]: bevyengine/bevy#6391 [6408]: bevyengine/bevy#6408 [6412]: bevyengine/bevy#6412 [6503]: bevyengine/bevy#6503 [6524]: bevyengine/bevy#6524 [6547]: bevyengine/bevy#6547 [6557]: bevyengine/bevy#6557 [6560]: bevyengine/bevy#6560 [6561]: bevyengine/bevy#6561 [6564]: bevyengine/bevy#6564 [6587]: bevyengine/bevy#6587 [6644]: bevyengine/bevy#6644 [6649]: bevyengine/bevy#6649 [6681]: bevyengine/bevy#6681 [6694]: bevyengine/bevy#6694 [6696]: bevyengine/bevy#6696 [6722]: bevyengine/bevy#6722 [6743]: bevyengine/bevy#6743 [6785]: bevyengine/bevy#6785 [6800]: bevyengine/bevy#6800 [6802]: bevyengine/bevy#6802 [6809]: bevyengine/bevy#6809 [6817]: bevyengine/bevy#6817 [6831]: bevyengine/bevy#6831 [6833]: bevyengine/bevy#6833 [6843]: bevyengine/bevy#6843 [6852]: bevyengine/bevy#6852 [6885]: bevyengine/bevy#6885 [6900]: bevyengine/bevy#6900 [6902]: bevyengine/bevy#6902 [6922]: bevyengine/bevy#6922 [6926]: bevyengine/bevy#6926 [6936]: bevyengine/bevy#6936 [6940]: bevyengine/bevy#6940 [6944]: bevyengine/bevy#6944 [6965]: bevyengine/bevy#6965 [6988]: bevyengine/bevy#6988 [6995]: bevyengine/bevy#6995 [7015]: bevyengine/bevy#7015 [7023]: bevyengine/bevy#7023 [7024]: bevyengine/bevy#7024 [7046]: bevyengine/bevy#7046 [7051]: bevyengine/bevy#7051 [7053]: bevyengine/bevy#7053 [7060]: bevyengine/bevy#7060 [7063]: bevyengine/bevy#7063 [7064]: bevyengine/bevy#7064 [7069]: bevyengine/bevy#7069 [7083]: bevyengine/bevy#7083 [7084]: bevyengine/bevy#7084 [7087]: bevyengine/bevy#7087 [7094]: bevyengine/bevy#7094 [7097]: bevyengine/bevy#7097 [7185]: bevyengine/bevy#7185 [7199]: bevyengine/bevy#7199 [7205]: bevyengine/bevy#7205 [7206]: bevyengine/bevy#7206 [7248]: bevyengine/bevy#7248 [7267]: bevyengine/bevy#7267 [7298]: bevyengine/bevy#7298 [7311]: bevyengine/bevy#7311 [7321]: bevyengine/bevy#7321 [7324]: bevyengine/bevy#7324 [7325]: bevyengine/bevy#7325 [7354]: bevyengine/bevy#7354 [7356]: bevyengine/bevy#7356 [7364]: bevyengine/bevy#7364 [7381]: bevyengine/bevy#7381 [7398]: bevyengine/bevy#7398 [7399]: bevyengine/bevy#7399 [7415]: bevyengine/bevy#7415 [7428]: bevyengine/bevy#7428 [7431]: bevyengine/bevy#7431 [7444]: bevyengine/bevy#7444 [7445]: bevyengine/bevy#7445 [7446]: bevyengine/bevy#7446 [7448]: bevyengine/bevy#7448 [7456]: bevyengine/bevy#7456 [7463]: bevyengine/bevy#7463 [7466]: bevyengine/bevy#7466 [7481]: bevyengine/bevy#7481 [7483]: bevyengine/bevy#7483 [7493]: bevyengine/bevy#7493 [7507]: bevyengine/bevy#7507 [7510]: bevyengine/bevy#7510 [7512]: bevyengine/bevy#7512 [7517]: bevyengine/bevy#7517 [7518]: bevyengine/bevy#7518 [7532]: bevyengine/bevy#7532 [7535]: bevyengine/bevy#7535 [7546]: bevyengine/bevy#7546 [7547]: bevyengine/bevy#7547 [7559]: bevyengine/bevy#7559 [7568]: bevyengine/bevy#7568 [7579]: bevyengine/bevy#7579 [7594]: bevyengine/bevy#7594 [7605]: bevyengine/bevy#7605 [7639]: bevyengine/bevy#7639 [7642]: bevyengine/bevy#7642 [7653]: bevyengine/bevy#7653 [7701]: bevyengine/bevy#7701 [7737]: bevyengine/bevy#7737 [7756]: bevyengine/bevy#7756 Co-authored-by: François <[email protected]> Co-authored-by: Alice Cecile <[email protected]> Co-authored-by: Mike <[email protected]> Co-authored-by: Boxy <[email protected]> Co-authored-by: IceSentry <[email protected]> Co-authored-by: JoJoJet <[email protected]> Co-authored-by: Aevyrie <[email protected]> Co-authored-by: James Liu <[email protected]> Co-authored-by: Marco Buono <[email protected]> Co-authored-by: Aceeri <[email protected]>
2 tasks
github-merge-queue bot
pushed a commit
that referenced
this pull request
Dec 1, 2023
… Iterator combinators (#6773) # Objective After #6547, `Query::for_each` has been capable of automatic vectorization on certain queries, which is seeing a notable (>50% CPU time improvements) for iteration. However, `Query::for_each` isn't idiomatic Rust, and lacks the flexibility of iterator combinators. Ideally, `Query::iter` and friends should be able to achieve the same results. However, this does seem to blocked upstream (rust-lang/rust#104914) by Rust's loop optimizations. ## Solution This is an intermediate solution and refactor. This moves the `Query::for_each` implementation onto the `Iterator::fold` implementation for `QueryIter` instead. This should result in the same automatic vectorization optimization on all `Iterator` functions that internally use fold, including `Iterator::for_each`, `Iterator::count`, etc. With this, it should close the gap between the two completely. Internally, this PR changes `Query::for_each` to use `query.iter().for_each(..)` instead of the duplicated implementation. Separately, the duplicate implementations of internal iteration (i.e. `Query::par_for_each`) now use portions of the current `Query::for_each` implementation factored out into their own functions. This also massively cleans up our internal fragmentation of internal iteration options, deduplicating the iteration code used in `for_each` and `par_iter().for_each()`. --- ## Changelog Changed: `Query::for_each`, `Query::for_each_mut`, `Query::for_each`, and `Query::for_each_mut` have been moved to `QueryIter`'s `Iterator::for_each` implementation, and still retains their performance improvements over normal iteration. These APIs are deprecated in 0.13 and will be removed in 0.14. --------- Co-authored-by: JoJoJet <[email protected]> Co-authored-by: Alice Cecile <[email protected]>
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.
Objective
Fixes #4884.
ComponentTicksstores both added and changed ticks contiguously in the same 8 bytes. This is convenient when passing around both together, but causes half the bytes fetched from memory for the purposes of change detection to effectively go unused. This is inefficient when most queries (no filter, mutating something) only write out to the changed ticks.Solution
Split the storage for change detection ticks into two separate
Vecs insideColumn. Fetch only what is needed during iteration.This also potentially also removes one blocker from autovectorization of dense queries.
EDIT: This is confirmed to enable autovectorization of dense queries in
for_eachandpar_for_eachwhere possible. Unfortunatelyiterhas other blockers that prevent it.TODO
Open Questions
IsIt's optimized out if unused.Mut::is_addedabsolutely necessary? Can we not just useAddedorChangeTrackers?Does the fetch of the added ticks get optimized out if not used?Yes it is.Changelog
Added:
Tick, a wrapper around a single change detection tick.Added:
Column::get_added_ticksAdded:
Column::get_column_ticksAdded:
SparseSet::get_added_ticksAdded:
SparseSet::get_column_ticksChanged:
Columnnow stores added and changed ticks separately internally.Changed: Most APIs returning
&UnsafeCell<ComponentTicks>now returnsTickCellsinstead, which contains two separate&UnsafeCell<Tick>for either component ticks.Changed:
Query::for_each(_mut),Query::par_for_each(_mut)will now leverage autovectorization to speed up query iteration where possible.Migration Guide
Various low level APIs interacting with the change detection ticks no longer return
&UnsafeCell<ComponentTicks>, instead returningTickCellswhich contains two separate&UnsafeCell<Tick>s instead.