Improve the Correctness & Performance of Time Formatting #576
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
This improves the correctness and the performance of the time formatting by making use of the fact that the underlying times are represented as a pair of integers (seconds and the subsecond nanoseconds). This way instead of formatting the fractional part as a floating point number we already have the fractional part in form of the nanoseconds. On top of that we can improve the performance a little bit by removing the data dependency we introduced by calculating the hours from the total minutes (and so on) when we can just calculate the hours directly from the total seconds. This allows for a little bit of instruction level parallelism where the days, hours, minutes and seconds can all be calculated in parallel.
CryZe
added
enhancement
CryZe
added a commit
to CryZe/livesplit-core
that referenced
this pull request
Oct 6, 2022
Going through the standard library's formatter is usually quite slow. And usually we want to show either minutes or seconds that are in the range from 0 to 59, or even tenths from 0 to 99. For those we can just define a lookup table where we look up the formatted strings and directly call `write_str` on the formatter instead of going through the `write!` macro, which would go through a lot more setup. Additionally we already have `itoa` as an indirect dependency which we now directly use for formatting integers that are not bound that way. `itoa` generally does the same algorithm as `std` but does so without going through any formatting machinery and is thus a lot faster, but also much less customizable. Overall this and LiveSplit#576 together result in a `~3.86x` performance improvement when formatting a time. | When | Time | |-------------|------------:| | Both PRs | `59.205 ns` | | Previous PR | `154.63 ns` | | Before | `228.47 ns` |
CryZe
added a commit
to CryZe/livesplit-core
that referenced
this pull request
Oct 6, 2022
Going through the standard library's formatter is usually quite slow. And usually we want to show either minutes or seconds that are in the range from 0 to 59, or even tenths from 0 to 99. For those we can just define a lookup table where we look up the formatted strings and directly call `write_str` on the formatter instead of going through the `write!` macro, which would go through a lot more setup. Additionally we already have `itoa` as an indirect dependency which we now directly use for formatting integers that are not bound that way. `itoa` generally does the same algorithm as `std` but does so without going through any formatting machinery and is thus a lot faster, but also much less customizable. Overall this and LiveSplit#576 together result in a `~3.86x` performance improvement when formatting a time. | When | Time | |-------------|------------:| | Both PRs | `59.205 ns` | | Previous PR | `154.63 ns` | | Before | `228.47 ns` |
CryZe
added a commit
that referenced
this pull request
Oct 6, 2022
Going through the standard library's formatter is usually quite slow. And usually we want to show either minutes or seconds that are in the range from 0 to 59, or even tenths from 0 to 99. For those we can just define a lookup table where we look up the formatted strings and directly call `write_str` on the formatter instead of going through the `write!` macro, which would go through a lot more setup. Additionally we already have `itoa` as an indirect dependency which we now directly use for formatting integers that are not bound that way. `itoa` generally does the same algorithm as `std` but does so without going through any formatting machinery and is thus a lot faster, but also much less customizable. Overall this and #576 together result in a `~3.86x` performance improvement when formatting a time. | When | Time | |:------------|------------:| | Both PRs | `59.205 ns` | | Previous PR | `154.63 ns` | | Before | `228.47 ns` |
CryZe
added a commit
that referenced
this pull request
Dec 29, 2022
- The `livesplit-hotkey` crate is now documented. (@CryZe) [#479](#479) - Not every key press emits a scan code on Windows. For those the virtual key code is now translated to a scan code. (@CryZe) [#480](#480) - Time parsing is now a lot more robust, handles more edge cases, and is also a lot more accurate. (@CryZe) [#483](#483) and [#578](#578) - When parsing a GDI based font name, platforms other than Windows now don't attempt to parse "normal" as part of the font name anymore as it is too ambigious. It could either refer to a font weight or stretch. (@kadiwa4) [#487](#487) - The text engine can now be customized. You can either provide your own text engine or use the one provided by `livesplit-core`. The one provided is now behind the `path-based-text-engine` and converts all glyphs to paths that can easily be drawn. (@CryZe) [#495](#495) - The path based text engine now caches the width of digits for tabular numbers, as well as the ellipsis glyph and its width, so that they can be layed out faster. (@kadiwa4) [#490](#490) and [#499](#499) - On Windows GDI is now used to resolve GDI based font names. (@CryZe) [#500](#500) - (Total) Possible Time Save now properly indicates that it's updating frequently. This results in faster rendering times. (@kadiwa4) [#501](#501) - Initial support for auto splitting has landed in `livesplit-core`. Auto splitters are provided as WebAssembly modules. Support can be activated via the `auto-splitting` feature. (@P1n3appl3) [#477](#477) - Auto splitting is also supported via the C API when activating its `auto-splitting` feature. (@DarkRTA) [#503](#503) - A watchdog for the Auto Splitting Runtime was added which unloads scripts that aren't responsive. (@CryZe) [#528](#528) - Splits and layouts can now be parsed and saved on `no_std` platforms. (@CryZe) [#532](#532) - The splits component column labels can now be queried via the C API. (@MichaelJBerk) [#526](#526) - The Software Renderer is now supported on `no_std` platforms. (@CryZe) [#536](#536) - The parsers are now faster because they don't allocate as much memory anymore. (@CryZe) [#546](#546) - The auto splitters have unstable support the `WebAssembly System Interface` via the `unstable-auto-splitting` feature. (@CryZe) [#547](#547) - The Timer component can now use the color of the delta for its background. (@Hurricane996) [#539](#539) - The splits component now takes the font into account when calculating the width of the columns. (@Hurricane996) [#550](#550) - The `Resource Allocator` now decodes the images, allowing the underlying renderer to do the encoding by itself. (@CryZe) [#562](#562) - Cargo's `--crate-type` parameter is now used to build the C API. (@CryZe) [#565](#565) - The columns of the splits component can now show the custom variables. (@CryZe) [#566](#566) - On the web, the `keydown` event may not always pass a `KeyboardEvent` despite the specification saying that this should be the case. This is now properly handled. (@CryZe) [#567](#567) - An integer overflow in the `FuzzyList` used for searching game and category names has been fixed. (@CryZe) [#569](#569) - The way the background is handled in the Detailed Timer component has been fixed. (@CryZe) [#572](#572) - The times are now formatted as strings without going through floating point numbers which increases both the correctness and the performance. (@CryZe) [#576](#576) - Instead of using `core::fmt` formatting machinery to format the times as strings, we now use a custom implementation that's much faster. (@CryZe) [#577](#577) and [#580](#580) - Holding down a hotkey on Windows now doesn't cause it to be triggered over and over again. Other platforms already behaved this way. (@CryZe) [#584](#584) - The `base64` crate is now replaced with `base64-simd` which uses SIMD to speed up the decoding of the images. (@CryZe) [#585](#585) - Splits from `SpeedRunIGT`, which is a Minecraft speedruning mod, can now be parsed. (@CryZe) [#591](#591) - It turns out using `evdev` for the hotkeys on Linux requires the user to be in the `input` group, which is not always the case. Therefore we now fall back to `X11` if `evdev` is not usable. (@CryZe) [#592](#592) - When an auto splitter wants to attach to a Process by name, the start time and process id are now used to prioritize duplicate processes. (@Eein) [#589](#589) - It is now possible to resolve the key codes to the particular name of the key based on the current keyboard layout on Linux and the web. This was already the case on Windows and macOS. (@CryZe) [#594](#594) and [#595](#595) - It is now possible to trust the user of the C API to always pass valid UTF-8 strings to the C API via the optional `assume-str-parameters-are-utf8` feature. This is also always the case when using WebAssembly on the web. This improves the performance because no validation of the strings is necessary. (@CryZe) [#597](#597) - There is now a new `max-opt` cargo profile that can be used to maximally optimize the resulting executable. The release profile is now using its default configuration again. (@CryZe) [#598](#598) - When encountering images `livesplit-core` checks their dimensions to potentially automatically shrink them if they are larger than necessary. It turns out that checking the dimensions of PNG images was a lot less efficient than it could have been. This even improves parsing speed of entire splits files by up to 30%. (@CryZe) [#600](#600) - The documentation now uses links to types mentioned. (@Eein) [#596](#596) - Auto splitters can now query size of the modules of a process. (@CryZe) [#602](#602) - The log messages emitted by auto splitters can now be consumed directly instead of always being emitted via the `log` crate. (@CryZe) [#603](#603) - The auto splitters can provide settings that can be configured. For now the auto splitters need to be reloaded when the settings change. (@CryZe) [#606](#606) - The file path used to be tracked in the `Run`, but no frontend even used this. So it has been removed. (@CryZe) [#616](#616) - The documentation states that the title component's lines store the unabbreviated line as their last element. This was not actually the case and has been fixed. (@DarkRTA) [#615](#615)
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 improves the correctness and the performance of the time formatting by making use of the fact that the underlying times are represented as a pair of integers (seconds and the subsecond nanoseconds). This way instead of formatting the fractional part as a floating point number we already have the fractional part in form of the nanoseconds. On top of that we can improve the performance a little bit by removing the data dependency we introduced by calculating the hours from the total minutes (and so on) when we can just calculate the hours directly from the total seconds. This allows for a little bit of instruction level parallelism where the days, hours, minutes and seconds can all be calculated in parallel.