diff options
Diffstat (limited to 'vendor/memchr')
53 files changed, 16099 insertions, 0 deletions
diff --git a/vendor/memchr/.cargo-checksum.json b/vendor/memchr/.cargo-checksum.json new file mode 100644 index 0000000..4b915f3 --- /dev/null +++ b/vendor/memchr/.cargo-checksum.json @@ -0,0 +1 @@ +{"files":{"COPYING":"01c266bced4a434da0051174d6bee16a4c82cf634e2679b6155d40d75012390f","Cargo.toml":"8f521805349c8df78da2ccaeec8b48f2f9fa1667887aaf89950ae555cbf6f8b2","LICENSE-MIT":"0f96a83840e146e43c0ec96a22ec1f392e0680e6c1226e6f3ba87e0740af850f","README.md":"92a74aaffe011bdaa06fbc34a01686a6eba58ca1322e976759417a547fddf734","UNLICENSE":"7e12e5df4bae12cb21581ba157ced20e1986a0508dd10d0e8a4ab9a4cf94e85c","rustfmt.toml":"1ca600239a27401c4a43f363cf3f38183a212affc1f31bff3ae93234bbaec228","src/arch/aarch64/memchr.rs":"5bb70f915084e629d940dbc322f5b9096b2e658cf63fea8a2f6e7550412e73a0","src/arch/aarch64/mod.rs":"44cd1a614bd66f1e66fc86c541d3c3b8d3a14a644c13e8bf816df3f555eac2d4","src/arch/aarch64/neon/memchr.rs":"e8c00b8fb2c7e2711832ae3cedefe59f32ebedd7dfa4d0ec6de2a566c979daea","src/arch/aarch64/neon/mod.rs":"eab6d56c2b2354db4ee395f40282cd49f97e2ab853547be5de6e65fbe1b2f634","src/arch/aarch64/neon/packedpair.rs":"fbdfdbfaf7b76b234db261fbe55a55c4479d32cdc65a654d60417c2d1c237849","src/arch/all/memchr.rs":"f6c51e50309f80e749887572297aa5efc19a3c7e5434b4404a101b3e6b152be1","src/arch/all/mod.rs":"05f3fc2b069682eb1545fc6366d167bb620a454365dac8b8dd6cde6cd64de18a","src/arch/all/packedpair/default_rank.rs":"abffd1b5b8b7a3be95c03dd1105b905c246a379854dc56f1e846ea7c4408f2c7","src/arch/all/packedpair/mod.rs":"292b66042c5b5c78bba33db6526aeae6904db803d601fcdd29032b87b3eb3754","src/arch/all/rabinkarp.rs":"236f69c04b90c14c253ae6c8d9b78150b4a56df75bb50af6d63b15145668b7cc","src/arch/all/shiftor.rs":"0d79117f52a1e4795843603a3bb0b45397df4ad5e4184bbc923658dab9dc3b5f","src/arch/all/twoway.rs":"47c97a265bfbafde90a618946643d3e97dfd9a85f01aa4ac758cd4c1573a450d","src/arch/generic/memchr.rs":"88290761bab740878401e914d71866da6501cdcef53d1249ec6fda4c7f9c12ae","src/arch/generic/mod.rs":"1dd75f61e0ea2563b8205a08aaa7b55500130aa331d18b9e9f995724b66c7a39","src/arch/generic/packedpair.rs":"a4a6efb29877ced9cf4c4e5ae9f36a79f019a16b831f2b9424899a1513d458ad","src/arch/mod.rs":"6dbd9e0b1b89fecb9faac5df6edfc87e24607e9099136aa831f3f056b14e22db","src/arch/wasm32/memchr.rs":"bfaaeca702cc32e605a06d5078d26ac59263d3c4eb04f9756e6be5e2850c3d0d","src/arch/wasm32/mod.rs":"a20377aa8fe07d68594879101dc73061e4f51d9c8d812b593b1f376e3c8add79","src/arch/wasm32/simd128/memchr.rs":"bac2c4c43fe710c83a6f2b1118fede043be89dd821d4b532907f129f09fdb5cf","src/arch/wasm32/simd128/mod.rs":"c157b373faedbfd65323be432e25bc411d97aa1b7bc58e76048614c7b2bf3bf6","src/arch/wasm32/simd128/packedpair.rs":"47e7875f1a0b502f3f30ddfd9257ed7ad4568fb7d968b5e6c01ba9e2aab2a459","src/arch/x86_64/avx2/memchr.rs":"576ec0c30f49874f7fd9f6caeb490d56132c0fbbaa4d877b1aa532cafce19323","src/arch/x86_64/avx2/mod.rs":"0033d1b712d0b10f0f273ef9aa8caa53e05e49f4c56a64f39af0b9df97eec584","src/arch/x86_64/avx2/packedpair.rs":"87b69cb4301815906127db4f6370f572c7c5d5dad35c0946c00ad888dbcaec8c","src/arch/x86_64/memchr.rs":"99a1dbe4156d498e6f910d06d3d3b31e7f6d06dff7d13a4c51b33a02b7e2fba9","src/arch/x86_64/mod.rs":"61b2aa876942fd3e78714c2ae21e356c8634545c06995020f443fa50218df027","src/arch/x86_64/sse2/memchr.rs":"68fc3b8f9eddf82192979c3aa11e5141f085cbb993c49c340558719a904679dc","src/arch/x86_64/sse2/mod.rs":"38b70ae52a64ec974dbb91d04d6ca8013d9e06d1fe4af852206bbc2faf1c59aa","src/arch/x86_64/sse2/packedpair.rs":"241ea981d8eea6024769f1c9375f726a9bb9700160c5857781d4befd9f5ef55d","src/cow.rs":"34eddd02cb82cc2d5a2c640891d64efe332dabcc1eea5115764200d8f46b66f7","src/ext.rs":"c472bcc41a7ef48980d976e954e87ef9fdfdfd30ac0199b959cc7e5b9d563ab3","src/lib.rs":"22509be5f0c201773ee26bb21bf5c0491e287c4414c02b7faa6ea5177ce0f32f","src/macros.rs":"3e4b39252bfa471fad384160a43f113ebfec7bec46a85d16f006622881dd2081","src/memchr.rs":"36f1c03304261877cd7f75c7ed8f7daff7a5c570cedce375e38e9b1ca44467f7","src/memmem/mod.rs":"1b0a9d6a681fd0887c677c4fc8d4c8f9719ddde250bdd5ea545365c1a7fb9094","src/memmem/searcher.rs":"69c38fb33d8f1a2a26769a81e514428240c8f8f15cea5302873d90b80391dd89","src/tests/memchr/mod.rs":"269f8e4b4f7f5ea458f27a3c174eb1020ffb2484eeba9464170beb51747df69b","src/tests/memchr/naive.rs":"6a0bee033e5edfb5b1d5769a5fa1c78388f7e9ff7bb91cb67f0ad029289e00e7","src/tests/memchr/prop.rs":"7bf7435087fbf08c5014c216b76575349735590d6b1d0e448921a1dc17bc0ea7","src/tests/mod.rs":"7cec8f809e279310a465c6a7725087970f219a676cc76c83de30c695bb490740","src/tests/packedpair.rs":"b02ec4fbb61a8653cb5f2268c31bc9168b8043347f2abdcc74081acf83b98e15","src/tests/substring/mod.rs":"c7660d10749363ac4687e7da2b5fda60768230425df8ba416c0c28b8d56a5c74","src/tests/substring/naive.rs":"df6f55d165382b8a53762ba4c324926cac13ebc62cde1805f4ce08740b326483","src/tests/substring/prop.rs":"38c15992609b5681a95d838ae6f2933e00a1219f2c971bfba245f96e0729fcdc","src/tests/x86_64-soft_float.json":"c0e416487fe9b4809534edb7db2a9eff3453dc40d9f1e23362c37f45a77ec717","src/vector.rs":"ef823ae8c54053780a0e7aeaee14b6c6ac2aea4567bf701ae8be137806c6d293"},"package":"523dc4f511e55ab87b694dc30d0f820d60906ef06413f93d4d7a1385599cc149"}
\ No newline at end of file diff --git a/vendor/memchr/COPYING b/vendor/memchr/COPYING new file mode 100644 index 0000000..bb9c20a --- /dev/null +++ b/vendor/memchr/COPYING @@ -0,0 +1,3 @@ +This project is dual-licensed under the Unlicense and MIT licenses. + +You may use this code under the terms of either license. diff --git a/vendor/memchr/Cargo.toml b/vendor/memchr/Cargo.toml new file mode 100644 index 0000000..9e95051 --- /dev/null +++ b/vendor/memchr/Cargo.toml @@ -0,0 +1,89 @@ +# THIS FILE IS AUTOMATICALLY GENERATED BY CARGO +# +# When uploading crates to the registry Cargo will automatically +# "normalize" Cargo.toml files for maximal compatibility +# with all versions of Cargo and also rewrite `path` dependencies +# to registry (e.g., crates.io) dependencies. +# +# If you are reading this file be aware that the original Cargo.toml +# will likely look very different (and much more reasonable). +# See Cargo.toml.orig for the original contents. + +[package] +edition = "2021" +rust-version = "1.61" +name = "memchr" +version = "2.7.1" +authors = [ + "Andrew Gallant <jamslam@gmail.com>", + "bluss", +] +exclude = [ + "/.github", + "/benchmarks", + "/fuzz", + "/scripts", + "/tmp", +] +description = """ +Provides extremely fast (uses SIMD on x86_64, aarch64 and wasm32) routines for +1, 2 or 3 byte search and single substring search. +""" +homepage = "https://github.com/BurntSushi/memchr" +documentation = "https://docs.rs/memchr/" +readme = "README.md" +keywords = [ + "memchr", + "memmem", + "substring", + "find", + "search", +] +license = "Unlicense OR MIT" +repository = "https://github.com/BurntSushi/memchr" + +[package.metadata.docs.rs] +rustdoc-args = ["--generate-link-to-definition"] + +[profile.bench] +debug = 2 + +[profile.release] +debug = 2 + +[profile.test] +opt-level = 3 +debug = 2 + +[lib] +name = "memchr" +bench = false + +[dependencies.compiler_builtins] +version = "0.1.2" +optional = true + +[dependencies.core] +version = "1.0.0" +optional = true +package = "rustc-std-workspace-core" + +[dependencies.log] +version = "0.4.20" +optional = true + +[dev-dependencies.quickcheck] +version = "1.0.3" +default-features = false + +[features] +alloc = [] +default = ["std"] +libc = [] +logging = ["dep:log"] +rustc-dep-of-std = [ + "core", + "compiler_builtins", +] +std = ["alloc"] +use_std = ["std"] diff --git a/vendor/memchr/LICENSE-MIT b/vendor/memchr/LICENSE-MIT new file mode 100644 index 0000000..3b0a5dc --- /dev/null +++ b/vendor/memchr/LICENSE-MIT @@ -0,0 +1,21 @@ +The MIT License (MIT) + +Copyright (c) 2015 Andrew Gallant + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in +all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN +THE SOFTWARE. diff --git a/vendor/memchr/README.md b/vendor/memchr/README.md new file mode 100644 index 0000000..db00ebb --- /dev/null +++ b/vendor/memchr/README.md @@ -0,0 +1,196 @@ +memchr +====== +This library provides heavily optimized routines for string search primitives. + +[](https://github.com/BurntSushi/memchr/actions) +[](https://crates.io/crates/memchr) + +Dual-licensed under MIT or the [UNLICENSE](https://unlicense.org/). + + +### Documentation + +[https://docs.rs/memchr](https://docs.rs/memchr) + + +### Overview + +* The top-level module provides routines for searching for 1, 2 or 3 bytes + in the forward or reverse direction. When searching for more than one byte, + positions are considered a match if the byte at that position matches any + of the bytes. +* The `memmem` sub-module provides forward and reverse substring search + routines. + +In all such cases, routines operate on `&[u8]` without regard to encoding. This +is exactly what you want when searching either UTF-8 or arbitrary bytes. + +### Compiling without the standard library + +memchr links to the standard library by default, but you can disable the +`std` feature if you want to use it in a `#![no_std]` crate: + +```toml +[dependencies] +memchr = { version = "2", default-features = false } +``` + +On `x86_64` platforms, when the `std` feature is disabled, the SSE2 accelerated +implementations will be used. When `std` is enabled, AVX2 accelerated +implementations will be used if the CPU is determined to support it at runtime. + +SIMD accelerated routines are also available on the `wasm32` and `aarch64` +targets. The `std` feature is not required to use them. + +When a SIMD version is not available, then this crate falls back to +[SWAR](https://en.wikipedia.org/wiki/SWAR) techniques. + +### Minimum Rust version policy + +This crate's minimum supported `rustc` version is `1.61.0`. + +The current policy is that the minimum Rust version required to use this crate +can be increased in minor version updates. For example, if `crate 1.0` requires +Rust 1.20.0, then `crate 1.0.z` for all values of `z` will also require Rust +1.20.0 or newer. However, `crate 1.y` for `y > 0` may require a newer minimum +version of Rust. + +In general, this crate will be conservative with respect to the minimum +supported version of Rust. + + +### Testing strategy + +Given the complexity of the code in this crate, along with the pervasive use +of `unsafe`, this crate has an extensive testing strategy. It combines multiple +approaches: + +* Hand-written tests. +* Exhaustive-style testing meant to exercise all possible branching and offset + calculations. +* Property based testing through [`quickcheck`](https://github.com/BurntSushi/quickcheck). +* Fuzz testing through [`cargo fuzz`](https://github.com/rust-fuzz/cargo-fuzz). +* A huge suite of benchmarks that are also run as tests. Benchmarks always + confirm that the expected result occurs. + +Improvements to the testing infrastructure are very welcome. + + +### Algorithms used + +At time of writing, this crate's implementation of substring search actually +has a few different algorithms to choose from depending on the situation. + +* For very small haystacks, + [Rabin-Karp](https://en.wikipedia.org/wiki/Rabin%E2%80%93Karp_algorithm) + is used to reduce latency. Rabin-Karp has very small overhead and can often + complete before other searchers have even been constructed. +* For small needles, a variant of the + ["Generic SIMD"](http://0x80.pl/articles/simd-strfind.html#algorithm-1-generic-simd) + algorithm is used. Instead of using the first and last bytes, a heuristic is + used to select bytes based on a background distribution of byte frequencies. +* In all other cases, + [Two-Way](https://en.wikipedia.org/wiki/Two-way_string-matching_algorithm) + is used. If possible, a prefilter based on the "Generic SIMD" algorithm + linked above is used to find candidates quickly. A dynamic heuristic is used + to detect if the prefilter is ineffective, and if so, disables it. + + +### Why is the standard library's substring search so much slower? + +We'll start by establishing what the difference in performance actually +is. There are two relevant benchmark classes to consider: `prebuilt` and +`oneshot`. The `prebuilt` benchmarks are designed to measure---to the extent +possible---search time only. That is, the benchmark first starts by building a +searcher and then only tracking the time for _using_ the searcher: + +``` +$ rebar rank benchmarks/record/x86_64/2023-08-26.csv --intersection -e memchr/memmem/prebuilt -e std/memmem/prebuilt +Engine Version Geometric mean of speed ratios Benchmark count +------ ------- ------------------------------ --------------- +rust/memchr/memmem/prebuilt 2.5.0 1.03 53 +rust/std/memmem/prebuilt 1.73.0-nightly 180dffba1 6.50 53 +``` + +Conversely, the `oneshot` benchmark class measures the time it takes to both +build the searcher _and_ use it: + +``` +$ rebar rank benchmarks/record/x86_64/2023-08-26.csv --intersection -e memchr/memmem/oneshot -e std/memmem/oneshot +Engine Version Geometric mean of speed ratios Benchmark count +------ ------- ------------------------------ --------------- +rust/memchr/memmem/oneshot 2.5.0 1.04 53 +rust/std/memmem/oneshot 1.73.0-nightly 180dffba1 5.26 53 +``` + +**NOTE:** Replace `rebar rank` with `rebar cmp` in the above commands to +explore the specific benchmarks and their differences. + +So in both cases, this crate is quite a bit faster over a broad sampling of +benchmarks regardless of whether you measure only search time or search time +plus construction time. The difference is a little smaller when you include +construction time in your measurements. + +These two different types of benchmark classes make for a nice segue into +one reason why the standard library's substring search can be slower: API +design. In the standard library, the only APIs available to you require +one to re-construct the searcher for every search. While you can benefit +from building a searcher once and iterating over all matches in a single +string, you cannot reuse that searcher to search other strings. This might +come up when, for example, searching a file one line at a time. You'll need +to re-build the searcher for every line searched, and this can [really +matter][burntsushi-bstr-blog]. + +**NOTE:** The `prebuilt` benchmark for the standard library can't actually +avoid measuring searcher construction at some level, because there is no API +for it. Instead, the benchmark consists of building the searcher once and then +finding all matches in a single string via an iterator. This tends to +approximate a benchmark where searcher construction isn't measured, but it +isn't perfect. While this means the comparison is not strictly +apples-to-apples, it does reflect what is maximally possible with the standard +library, and thus reflects the best that one could do in a real world scenario. + +While there is more to the story than just API design here, it's important to +point out that even if the standard library's substring search were a precise +clone of this crate internally, it would still be at a disadvantage in some +workloads because of its API. (The same also applies to C's standard library +`memmem` function. There is no way to amortize construction of the searcher. +You need to pay for it on every call.) + +The other reason for the difference in performance is that +the standard library has trouble using SIMD. In particular, substring search +is implemented in the `core` library, where platform specific code generally +can't exist. That's an issue because in order to utilize SIMD beyond SSE2 +while maintaining portable binaries, one needs to use [dynamic CPU feature +detection][dynamic-cpu], and that in turn requires platform specific code. +While there is [an RFC for enabling target feature detection in +`core`][core-feature], it doesn't yet exist. + +The bottom line here is that `core`'s substring search implementation is +limited to making use of SSE2, but not AVX. + +Still though, this crate does accelerate substring search even when only SSE2 +is available. The standard library could therefore adopt the techniques in this +crate just for SSE2. The reason why that hasn't happened yet isn't totally +clear to me. It likely needs a champion to push it through. The standard +library tends to be more conservative in these things. With that said, the +standard library does use some [SSE2 acceleration on `x86-64`][std-sse2] added +in [this PR][std-sse2-pr]. However, at the time of writing, it is only used +for short needles and doesn't use the frequency based heuristics found in this +crate. + +**NOTE:** Another thing worth mentioning is that the standard library's +substring search routine requires that both the needle and haystack have type +`&str`. Unless you can assume that your data is valid UTF-8, building a `&str` +will come with the overhead of UTF-8 validation. This may in turn result in +overall slower searching depending on your workload. In contrast, the `memchr` +crate permits both the needle and the haystack to have type `&[u8]`, where +`&[u8]` can be created from a `&str` with zero cost. Therefore, the substring +search in this crate is strictly more flexible than what the standard library +provides. + +[burntsushi-bstr-blog]: https://blog.burntsushi.net/bstr/#motivation-based-on-performance +[dynamic-cpu]: https://doc.rust-lang.org/std/arch/index.html#dynamic-cpu-feature-detection +[core-feature]: https://github.com/rust-lang/rfcs/pull/3469 +[std-sse2]: https://github.com/rust-lang/rust/blob/bf9229a2e366b4c311f059014a4aa08af16de5d8/library/core/src/str/pattern.rs#L1719-L1857 +[std-sse2-pr]: https://github.com/rust-lang/rust/pull/103779 diff --git a/vendor/memchr/UNLICENSE b/vendor/memchr/UNLICENSE new file mode 100644 index 0000000..68a49da --- /dev/null +++ b/vendor/memchr/UNLICENSE @@ -0,0 +1,24 @@ +This is free and unencumbered software released into the public domain. + +Anyone is free to copy, modify, publish, use, compile, sell, or +distribute this software, either in source code form or as a compiled +binary, for any purpose, commercial or non-commercial, and by any +means. + +In jurisdictions that recognize copyright laws, the author or authors +of this software dedicate any and all copyright interest in the +software to the public domain. We make this dedication for the benefit +of the public at large and to the detriment of our heirs and +successors. We intend this dedication to be an overt act of +relinquishment in perpetuity of all present and future rights to this +software under copyright law. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, +EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF +MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. +IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR +OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, +ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR +OTHER DEALINGS IN THE SOFTWARE. + +For more information, please refer to <http://unlicense.org/> diff --git a/vendor/memchr/rustfmt.toml b/vendor/memchr/rustfmt.toml new file mode 100644 index 0000000..aa37a21 --- /dev/null +++ b/vendor/memchr/rustfmt.toml @@ -0,0 +1,2 @@ +max_width = 79 +use_small_heuristics = "max" diff --git a/vendor/memchr/src/arch/aarch64/memchr.rs b/vendor/memchr/src/arch/aarch64/memchr.rs new file mode 100644 index 0000000..e0053b2 --- /dev/null +++ b/vendor/memchr/src/arch/aarch64/memchr.rs @@ -0,0 +1,137 @@ +/*! +Wrapper routines for `memchr` and friends. + +These routines choose the best implementation at compile time. (This is +different from `x86_64` because it is expected that `neon` is almost always +available for `aarch64` targets.) +*/ + +macro_rules! defraw { + ($ty:ident, $find:ident, $start:ident, $end:ident, $($needles:ident),+) => {{ + #[cfg(target_feature = "neon")] + { + use crate::arch::aarch64::neon::memchr::$ty; + + debug!("chose neon for {}", stringify!($ty)); + debug_assert!($ty::is_available()); + // SAFETY: We know that wasm memchr is always available whenever + // code is compiled for `aarch64` with the `neon` target feature + // enabled. + $ty::new_unchecked($($needles),+).$find($start, $end) + } + #[cfg(not(target_feature = "neon"))] + { + use crate::arch::all::memchr::$ty; + + debug!( + "no neon feature available, using fallback for {}", + stringify!($ty), + ); + $ty::new($($needles),+).$find($start, $end) + } + }} +} + +/// memchr, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::find_raw`. +#[inline(always)] +pub(crate) unsafe fn memchr_raw( + n1: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(One, find_raw, start, end, n1) +} + +/// memrchr, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::rfind_raw`. +#[inline(always)] +pub(crate) unsafe fn memrchr_raw( + n1: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(One, rfind_raw, start, end, n1) +} + +/// memchr2, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Two::find_raw`. +#[inline(always)] +pub(crate) unsafe fn memchr2_raw( + n1: u8, + n2: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(Two, find_raw, start, end, n1, n2) +} + +/// memrchr2, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Two::rfind_raw`. +#[inline(always)] +pub(crate) unsafe fn memrchr2_raw( + n1: u8, + n2: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(Two, rfind_raw, start, end, n1, n2) +} + +/// memchr3, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Three::find_raw`. +#[inline(always)] +pub(crate) unsafe fn memchr3_raw( + n1: u8, + n2: u8, + n3: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(Three, find_raw, start, end, n1, n2, n3) +} + +/// memrchr3, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Three::rfind_raw`. +#[inline(always)] +pub(crate) unsafe fn memrchr3_raw( + n1: u8, + n2: u8, + n3: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(Three, rfind_raw, start, end, n1, n2, n3) +} + +/// Count all matching bytes, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::count_raw`. +#[inline(always)] +pub(crate) unsafe fn count_raw( + n1: u8, + start: *const u8, + end: *const u8, +) -> usize { + defraw!(One, count_raw, start, end, n1) +} diff --git a/vendor/memchr/src/arch/aarch64/mod.rs b/vendor/memchr/src/arch/aarch64/mod.rs new file mode 100644 index 0000000..7b32912 --- /dev/null +++ b/vendor/memchr/src/arch/aarch64/mod.rs @@ -0,0 +1,7 @@ +/*! +Vector algorithms for the `aarch64` target. +*/ + +pub mod neon; + +pub(crate) mod memchr; diff --git a/vendor/memchr/src/arch/aarch64/neon/memchr.rs b/vendor/memchr/src/arch/aarch64/neon/memchr.rs new file mode 100644 index 0000000..5fcc762 --- /dev/null +++ b/vendor/memchr/src/arch/aarch64/neon/memchr.rs @@ -0,0 +1,1031 @@ +/*! +This module defines 128-bit vector implementations of `memchr` and friends. + +The main types in this module are [`One`], [`Two`] and [`Three`]. They are for +searching for one, two or three distinct bytes, respectively, in a haystack. +Each type also has corresponding double ended iterators. These searchers are +typically much faster than scalar routines accomplishing the same task. + +The `One` searcher also provides a [`One::count`] routine for efficiently +counting the number of times a single byte occurs in a haystack. This is +useful, for example, for counting the number of lines in a haystack. This +routine exists because it is usually faster, especially with a high match +count, then using [`One::find`] repeatedly. ([`OneIter`] specializes its +`Iterator::count` implementation to use this routine.) + +Only one, two and three bytes are supported because three bytes is about +the point where one sees diminishing returns. Beyond this point and it's +probably (but not necessarily) better to just use a simple `[bool; 256]` array +or similar. However, it depends mightily on the specific work-load and the +expected match frequency. +*/ + +use core::arch::aarch64::uint8x16_t; + +use crate::{arch::generic::memchr as generic, ext::Pointer, vector::Vector}; + +/// Finds all occurrences of a single byte in a haystack. +#[derive(Clone, Copy, Debug)] +pub struct One(generic::One<uint8x16_t>); + +impl One { + /// Create a new searcher that finds occurrences of the needle byte given. + /// + /// This particular searcher is specialized to use neon vector instructions + /// that typically make it quite fast. + /// + /// If neon is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn new(needle: u8) -> Option<One> { + if One::is_available() { + // SAFETY: we check that neon is available above. + unsafe { Some(One::new_unchecked(needle)) } + } else { + None + } + } + + /// Create a new finder specific to neon vectors and routines without + /// checking that neon is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute `neon` instructions + /// in the current environment. + /// + /// Note that it is a common misconception that if one compiles for an + /// `x86_64` target, then they therefore automatically have access to neon + /// instructions. While this is almost always the case, it isn't true in + /// 100% of cases. + #[target_feature(enable = "neon")] + #[inline] + pub unsafe fn new_unchecked(needle: u8) -> One { + One(generic::One::new(needle)) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`One::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `One::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "neon")] + { + true + } + #[cfg(not(target_feature = "neon"))] + { + false + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `rfind_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Counts all occurrences of this byte in the given haystack. + #[inline] + pub fn count(&self, haystack: &[u8]) -> usize { + // SAFETY: All of our pointers are derived directly from a borrowed + // slice, which is guaranteed to be valid. + unsafe { + let start = haystack.as_ptr(); + let end = start.add(haystack.len()); + self.count_raw(start, end) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < uint8x16_t::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::fwd_byte_by_byte(start, end, |b| { + b == self.0.needle1() + }); + } + // SAFETY: Building a `One` means it's safe to call 'neon' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.find_raw_impl(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < uint8x16_t::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::rev_byte_by_byte(start, end, |b| { + b == self.0.needle1() + }); + } + // SAFETY: Building a `One` means it's safe to call 'neon' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.rfind_raw_impl(start, end) + } + + /// Like `count`, but accepts and returns raw pointers. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn count_raw(&self, start: *const u8, end: *const u8) -> usize { + if start >= end { + return 0; + } + if end.distance(start) < uint8x16_t::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::count_byte_by_byte(start, end, |b| { + b == self.0.needle1() + }); + } + // SAFETY: Building a `One` means it's safe to call 'neon' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.count_raw_impl(start, end) + } + + /// Execute a search using neon vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::find_raw`], except the distance between `start` and + /// `end` must be at least the size of a neon vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `neon` routines.) + #[target_feature(enable = "neon")] + #[inline] + unsafe fn find_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.find_raw(start, end) + } + + /// Execute a search using neon vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of a neon vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `neon` routines.) + #[target_feature(enable = "neon")] + #[inline] + unsafe fn rfind_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.rfind_raw(start, end) + } + + /// Execute a count using neon vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::count_raw`], except the distance between `start` and + /// `end` must be at least the size of a neon vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `neon` routines.) + #[target_feature(enable = "neon")] + #[inline] + unsafe fn count_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> usize { + self.0.count_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle byte in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> OneIter<'a, 'h> { + OneIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of a single byte in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`One::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`One`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct OneIter<'a, 'h> { + searcher: &'a One, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for OneIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn count(self) -> usize { + self.it.count(|s, e| { + // SAFETY: We rely on our generic iterator to return valid start + // and end pointers. + unsafe { self.searcher.count_raw(s, e) } + }) + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for OneIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for OneIter<'a, 'h> {} + +/// Finds all occurrences of two bytes in a haystack. +/// +/// That is, this reports matches of one of two possible bytes. For example, +/// searching for `a` or `b` in `afoobar` would report matches at offsets `0`, +/// `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Two(generic::Two<uint8x16_t>); + +impl Two { + /// Create a new searcher that finds occurrences of the needle bytes given. + /// + /// This particular searcher is specialized to use neon vector instructions + /// that typically make it quite fast. + /// + /// If neon is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn new(needle1: u8, needle2: u8) -> Option<Two> { + if Two::is_available() { + // SAFETY: we check that neon is available above. + unsafe { Some(Two::new_unchecked(needle1, needle2)) } + } else { + None + } + } + + /// Create a new finder specific to neon vectors and routines without + /// checking that neon is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute `neon` instructions + /// in the current environment. + /// + /// Note that it is a common misconception that if one compiles for an + /// `x86_64` target, then they therefore automatically have access to neon + /// instructions. While this is almost always the case, it isn't true in + /// 100% of cases. + #[target_feature(enable = "neon")] + #[inline] + pub unsafe fn new_unchecked(needle1: u8, needle2: u8) -> Two { + Two(generic::Two::new(needle1, needle2)) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Two::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `Two::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "neon")] + { + true + } + #[cfg(not(target_feature = "neon"))] + { + false + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `rfind_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < uint8x16_t::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::fwd_byte_by_byte(start, end, |b| { + b == self.0.needle1() || b == self.0.needle2() + }); + } + // SAFETY: Building a `Two` means it's safe to call 'neon' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.find_raw_impl(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < uint8x16_t::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::rev_byte_by_byte(start, end, |b| { + b == self.0.needle1() || b == self.0.needle2() + }); + } + // SAFETY: Building a `Two` means it's safe to call 'neon' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.rfind_raw_impl(start, end) + } + + /// Execute a search using neon vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::find_raw`], except the distance between `start` and + /// `end` must be at least the size of a neon vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `neon` routines.) + #[target_feature(enable = "neon")] + #[inline] + unsafe fn find_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.find_raw(start, end) + } + + /// Execute a search using neon vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of a neon vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `neon` routines.) + #[target_feature(enable = "neon")] + #[inline] + unsafe fn rfind_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.rfind_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle bytes in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> TwoIter<'a, 'h> { + TwoIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of two possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Two::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Two`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct TwoIter<'a, 'h> { + searcher: &'a Two, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for TwoIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for TwoIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for TwoIter<'a, 'h> {} + +/// Finds all occurrences of three bytes in a haystack. +/// +/// That is, this reports matches of one of three possible bytes. For example, +/// searching for `a`, `b` or `o` in `afoobar` would report matches at offsets +/// `0`, `2`, `3`, `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Three(generic::Three<uint8x16_t>); + +impl Three { + /// Create a new searcher that finds occurrences of the needle bytes given. + /// + /// This particular searcher is specialized to use neon vector instructions + /// that typically make it quite fast. + /// + /// If neon is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn new(needle1: u8, needle2: u8, needle3: u8) -> Option<Three> { + if Three::is_available() { + // SAFETY: we check that neon is available above. + unsafe { Some(Three::new_unchecked(needle1, needle2, needle3)) } + } else { + None + } + } + + /// Create a new finder specific to neon vectors and routines without + /// checking that neon is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute `neon` instructions + /// in the current environment. + /// + /// Note that it is a common misconception that if one compiles for an + /// `x86_64` target, then they therefore automatically have access to neon + /// instructions. While this is almost always the case, it isn't true in + /// 100% of cases. + #[target_feature(enable = "neon")] + #[inline] + pub unsafe fn new_unchecked( + needle1: u8, + needle2: u8, + needle3: u8, + ) -> Three { + Three(generic::Three::new(needle1, needle2, needle3)) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Three::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `Three::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "neon")] + { + true + } + #[cfg(not(target_feature = "neon"))] + { + false + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `rfind_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < uint8x16_t::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::fwd_byte_by_byte(start, end, |b| { + b == self.0.needle1() + || b == self.0.needle2() + || b == self.0.needle3() + }); + } + // SAFETY: Building a `Three` means it's safe to call 'neon' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.find_raw_impl(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < uint8x16_t::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::rev_byte_by_byte(start, end, |b| { + b == self.0.needle1() + || b == self.0.needle2() + || b == self.0.needle3() + }); + } + // SAFETY: Building a `Three` means it's safe to call 'neon' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.rfind_raw_impl(start, end) + } + + /// Execute a search using neon vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::find_raw`], except the distance between `start` and + /// `end` must be at least the size of a neon vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `neon` routines.) + #[target_feature(enable = "neon")] + #[inline] + unsafe fn find_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.find_raw(start, end) + } + + /// Execute a search using neon vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of a neon vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `neon` routines.) + #[target_feature(enable = "neon")] + #[inline] + unsafe fn rfind_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.rfind_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle byte in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> ThreeIter<'a, 'h> { + ThreeIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of three possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Three::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Three`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct ThreeIter<'a, 'h> { + searcher: &'a Three, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for ThreeIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for ThreeIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for ThreeIter<'a, 'h> {} + +#[cfg(test)] +mod tests { + use super::*; + + define_memchr_quickcheck!(super); + + #[test] + fn forward_one() { + crate::tests::memchr::Runner::new(1).forward_iter( + |haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_one() { + crate::tests::memchr::Runner::new(1).reverse_iter( + |haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn count_one() { + crate::tests::memchr::Runner::new(1).count_iter(|haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).count()) + }) + } + + #[test] + fn forward_two() { + crate::tests::memchr::Runner::new(2).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2)?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_two() { + crate::tests::memchr::Runner::new(2).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2)?.iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn forward_three() { + crate::tests::memchr::Runner::new(3).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3)?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_three() { + crate::tests::memchr::Runner::new(3).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3)?.iter(haystack).rev().collect()) + }, + ) + } +} diff --git a/vendor/memchr/src/arch/aarch64/neon/mod.rs b/vendor/memchr/src/arch/aarch64/neon/mod.rs new file mode 100644 index 0000000..ccf9cf8 --- /dev/null +++ b/vendor/memchr/src/arch/aarch64/neon/mod.rs @@ -0,0 +1,6 @@ +/*! +Algorithms for the `aarch64` target using 128-bit vectors via NEON. +*/ + +pub mod memchr; +pub mod packedpair; diff --git a/vendor/memchr/src/arch/aarch64/neon/packedpair.rs b/vendor/memchr/src/arch/aarch64/neon/packedpair.rs new file mode 100644 index 0000000..6884882 --- /dev/null +++ b/vendor/memchr/src/arch/aarch64/neon/packedpair.rs @@ -0,0 +1,236 @@ +/*! +A 128-bit vector implementation of the "packed pair" SIMD algorithm. + +The "packed pair" algorithm is based on the [generic SIMD] algorithm. The main +difference is that it (by default) uses a background distribution of byte +frequencies to heuristically select the pair of bytes to search for. + +[generic SIMD]: http://0x80.pl/articles/simd-strfind.html#first-and-last +*/ + +use core::arch::aarch64::uint8x16_t; + +use crate::arch::{all::packedpair::Pair, generic::packedpair}; + +/// A "packed pair" finder that uses 128-bit vector operations. +/// +/// This finder picks two bytes that it believes have high predictive power +/// for indicating an overall match of a needle. Depending on whether +/// `Finder::find` or `Finder::find_prefilter` is used, it reports offsets +/// where the needle matches or could match. In the prefilter case, candidates +/// are reported whenever the [`Pair`] of bytes given matches. +#[derive(Clone, Copy, Debug)] +pub struct Finder(packedpair::Finder<uint8x16_t>); + +/// A "packed pair" finder that uses 128-bit vector operations. +/// +/// This finder picks two bytes that it believes have high predictive power +/// for indicating an overall match of a needle. Depending on whether +/// `Finder::find` or `Finder::find_prefilter` is used, it reports offsets +/// where the needle matches or could match. In the prefilter case, candidates +/// are reported whenever the [`Pair`] of bytes given matches. +impl Finder { + /// Create a new pair searcher. The searcher returned can either report + /// exact matches of `needle` or act as a prefilter and report candidate + /// positions of `needle`. + /// + /// If neon is unavailable in the current environment or if a [`Pair`] + /// could not be constructed from the needle given, then `None` is + /// returned. + #[inline] + pub fn new(needle: &[u8]) -> Option<Finder> { + Finder::with_pair(needle, Pair::new(needle)?) + } + + /// Create a new "packed pair" finder using the pair of bytes given. + /// + /// This constructor permits callers to control precisely which pair of + /// bytes is used as a predicate. + /// + /// If neon is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn with_pair(needle: &[u8], pair: Pair) -> Option<Finder> { + if Finder::is_available() { + // SAFETY: we check that sse2 is available above. We are also + // guaranteed to have needle.len() > 1 because we have a valid + // Pair. + unsafe { Some(Finder::with_pair_impl(needle, pair)) } + } else { + None + } + } + + /// Create a new `Finder` specific to neon vectors and routines. + /// + /// # Safety + /// + /// Same as the safety for `packedpair::Finder::new`, and callers must also + /// ensure that neon is available. + #[target_feature(enable = "neon")] + #[inline] + unsafe fn with_pair_impl(needle: &[u8], pair: Pair) -> Finder { + let finder = packedpair::Finder::<uint8x16_t>::new(needle, pair); + Finder(finder) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Finder::with_pair`] will + /// return a `Some` value. Similarly, when it is false, it is guaranteed + /// that `Finder::with_pair` will return a `None` value. Notice that this + /// does not guarantee that [`Finder::new`] will return a `Finder`. Namely, + /// even when `Finder::is_available` is true, it is not guaranteed that a + /// valid [`Pair`] can be found from the needle given. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "neon")] + { + true + } + #[cfg(not(target_feature = "neon"))] + { + false + } + } + + /// Execute a search using neon vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + #[inline] + pub fn find(&self, haystack: &[u8], needle: &[u8]) -> Option<usize> { + // SAFETY: Building a `Finder` means it's safe to call 'neon' routines. + unsafe { self.find_impl(haystack, needle) } + } + + /// Execute a search using neon vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + #[inline] + pub fn find_prefilter(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: Building a `Finder` means it's safe to call 'neon' routines. + unsafe { self.find_prefilter_impl(haystack) } + } + + /// Execute a search using neon vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Finder`, which can only be constructed + /// when it is safe to call `neon` routines.) + #[target_feature(enable = "neon")] + #[inline] + unsafe fn find_impl( + &self, + haystack: &[u8], + needle: &[u8], + ) -> Option<usize> { + self.0.find(haystack, needle) + } + + /// Execute a prefilter search using neon vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Finder`, which can only be constructed + /// when it is safe to call `neon` routines.) + #[target_feature(enable = "neon")] + #[inline] + unsafe fn find_prefilter_impl(&self, haystack: &[u8]) -> Option<usize> { + self.0.find_prefilter(haystack) + } + + /// Returns the pair of offsets (into the needle) used to check as a + /// predicate before confirming whether a needle exists at a particular + /// position. + #[inline] + pub fn pair(&self) -> &Pair { + self.0.pair() + } + + /// Returns the minimum haystack length that this `Finder` can search. + /// + /// Using a haystack with length smaller than this in a search will result + /// in a panic. The reason for this restriction is that this finder is + /// meant to be a low-level component that is part of a larger substring + /// strategy. In that sense, it avoids trying to handle all cases and + /// instead only handles the cases that it can handle very well. + #[inline] + pub fn min_haystack_len(&self) -> usize { + self.0.min_haystack_len() + } +} + +#[cfg(test)] +mod tests { + use super::*; + + fn find(haystack: &[u8], needle: &[u8]) -> Option<Option<usize>> { + let f = Finder::new(needle)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find(haystack, needle)) + } + + define_substring_forward_quickcheck!(find); + + #[test] + fn forward_substring() { + crate::tests::substring::Runner::new().fwd(find).run() + } + + #[test] + fn forward_packedpair() { + fn find( + haystack: &[u8], + needle: &[u8], + index1: u8, + index2: u8, + ) -> Option<Option<usize>> { + let pair = Pair::with_indices(needle, index1, index2)?; + let f = Finder::with_pair(needle, pair)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find(haystack, needle)) + } + crate::tests::packedpair::Runner::new().fwd(find).run() + } + + #[test] + fn forward_packedpair_prefilter() { + fn find( + haystack: &[u8], + needle: &[u8], + index1: u8, + index2: u8, + ) -> Option<Option<usize>> { + let pair = Pair::with_indices(needle, index1, index2)?; + let f = Finder::with_pair(needle, pair)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find_prefilter(haystack)) + } + crate::tests::packedpair::Runner::new().fwd(find).run() + } +} diff --git a/vendor/memchr/src/arch/all/memchr.rs b/vendor/memchr/src/arch/all/memchr.rs new file mode 100644 index 0000000..435b1be --- /dev/null +++ b/vendor/memchr/src/arch/all/memchr.rs @@ -0,0 +1,996 @@ +/*! +Provides architecture independent implementations of `memchr` and friends. + +The main types in this module are [`One`], [`Two`] and [`Three`]. They are for +searching for one, two or three distinct bytes, respectively, in a haystack. +Each type also has corresponding double ended iterators. These searchers +are typically slower than hand-coded vector routines accomplishing the same +task, but are also typically faster than naive scalar code. These routines +effectively work by treating a `usize` as a vector of 8-bit lanes, and thus +achieves some level of data parallelism even without explicit vector support. + +The `One` searcher also provides a [`One::count`] routine for efficiently +counting the number of times a single byte occurs in a haystack. This is +useful, for example, for counting the number of lines in a haystack. This +routine exists because it is usually faster, especially with a high match +count, then using [`One::find`] repeatedly. ([`OneIter`] specializes its +`Iterator::count` implementation to use this routine.) + +Only one, two and three bytes are supported because three bytes is about +the point where one sees diminishing returns. Beyond this point and it's +probably (but not necessarily) better to just use a simple `[bool; 256]` array +or similar. However, it depends mightily on the specific work-load and the +expected match frequency. +*/ + +use crate::{arch::generic::memchr as generic, ext::Pointer}; + +/// The number of bytes in a single `usize` value. +const USIZE_BYTES: usize = (usize::BITS / 8) as usize; +/// The bits that must be zero for a `*const usize` to be properly aligned. +const USIZE_ALIGN: usize = USIZE_BYTES - 1; + +/// Finds all occurrences of a single byte in a haystack. +#[derive(Clone, Copy, Debug)] +pub struct One { + s1: u8, + v1: usize, +} + +impl One { + /// The number of bytes we examine per each iteration of our search loop. + const LOOP_BYTES: usize = 2 * USIZE_BYTES; + + /// Create a new searcher that finds occurrences of the byte given. + #[inline] + pub fn new(needle: u8) -> One { + One { s1: needle, v1: splat(needle) } + } + + /// A test-only routine so that we can bundle a bunch of quickcheck + /// properties into a single macro. Basically, this provides a constructor + /// that makes it identical to most other memchr implementations, which + /// have fallible constructors. + #[cfg(test)] + pub(crate) fn try_new(needle: u8) -> Option<One> { + Some(One::new(needle)) + } + + /// Return the first occurrence of the needle in the given haystack. If no + /// such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value for a non-empty haystack is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of the needle in the given haystack. If no + /// such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value for a non-empty haystack is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Counts all occurrences of this byte in the given haystack. + #[inline] + pub fn count(&self, haystack: &[u8]) -> usize { + // SAFETY: All of our pointers are derived directly from a borrowed + // slice, which is guaranteed to be valid. + unsafe { + let start = haystack.as_ptr(); + let end = start.add(haystack.len()); + self.count_raw(start, end) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let confirm = |b| self.confirm(b); + let len = end.distance(start); + if len < USIZE_BYTES { + return generic::fwd_byte_by_byte(start, end, confirm); + } + + // The start of the search may not be aligned to `*const usize`, + // so we do an unaligned load here. + let chunk = start.cast::<usize>().read_unaligned(); + if self.has_needle(chunk) { + return generic::fwd_byte_by_byte(start, end, confirm); + } + + // And now we start our search at a guaranteed aligned position. + // The first iteration of the loop below will overlap with the the + // unaligned chunk above in cases where the search starts at an + // unaligned offset, but that's okay as we're only here if that + // above didn't find a match. + let mut cur = + start.add(USIZE_BYTES - (start.as_usize() & USIZE_ALIGN)); + debug_assert!(cur > start); + if len <= One::LOOP_BYTES { + return generic::fwd_byte_by_byte(cur, end, confirm); + } + debug_assert!(end.sub(One::LOOP_BYTES) >= start); + while cur <= end.sub(One::LOOP_BYTES) { + debug_assert_eq!(0, cur.as_usize() % USIZE_BYTES); + + let a = cur.cast::<usize>().read(); + let b = cur.add(USIZE_BYTES).cast::<usize>().read(); + if self.has_needle(a) || self.has_needle(b) { + break; + } + cur = cur.add(One::LOOP_BYTES); + } + generic::fwd_byte_by_byte(cur, end, confirm) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let confirm = |b| self.confirm(b); + let len = end.distance(start); + if len < USIZE_BYTES { + return generic::rev_byte_by_byte(start, end, confirm); + } + + let chunk = end.sub(USIZE_BYTES).cast::<usize>().read_unaligned(); + if self.has_needle(chunk) { + return generic::rev_byte_by_byte(start, end, confirm); + } + + let mut cur = end.sub(end.as_usize() & USIZE_ALIGN); + debug_assert!(start <= cur && cur <= end); + if len <= One::LOOP_BYTES { + return generic::rev_byte_by_byte(start, cur, confirm); + } + while cur >= start.add(One::LOOP_BYTES) { + debug_assert_eq!(0, cur.as_usize() % USIZE_BYTES); + + let a = cur.sub(2 * USIZE_BYTES).cast::<usize>().read(); + let b = cur.sub(1 * USIZE_BYTES).cast::<usize>().read(); + if self.has_needle(a) || self.has_needle(b) { + break; + } + cur = cur.sub(One::LOOP_BYTES); + } + generic::rev_byte_by_byte(start, cur, confirm) + } + + /// Counts all occurrences of this byte in the given haystack represented + /// by raw pointers. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `0` will always be returned. + #[inline] + pub unsafe fn count_raw(&self, start: *const u8, end: *const u8) -> usize { + if start >= end { + return 0; + } + // Sadly I couldn't get the SWAR approach to work here, so we just do + // one byte at a time for now. PRs to improve this are welcome. + let mut ptr = start; + let mut count = 0; + while ptr < end { + count += (ptr.read() == self.s1) as usize; + ptr = ptr.offset(1); + } + count + } + + /// Returns an iterator over all occurrences of the needle byte in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> OneIter<'a, 'h> { + OneIter { searcher: self, it: generic::Iter::new(haystack) } + } + + #[inline(always)] + fn has_needle(&self, chunk: usize) -> bool { + has_zero_byte(self.v1 ^ chunk) + } + + #[inline(always)] + fn confirm(&self, haystack_byte: u8) -> bool { + self.s1 == haystack_byte + } +} + +/// An iterator over all occurrences of a single byte in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`One::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`One`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct OneIter<'a, 'h> { + /// The underlying memchr searcher. + searcher: &'a One, + /// Generic iterator implementation. + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for OneIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn count(self) -> usize { + self.it.count(|s, e| { + // SAFETY: We rely on our generic iterator to return valid start + // and end pointers. + unsafe { self.searcher.count_raw(s, e) } + }) + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for OneIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +/// Finds all occurrences of two bytes in a haystack. +/// +/// That is, this reports matches of one of two possible bytes. For example, +/// searching for `a` or `b` in `afoobar` would report matches at offsets `0`, +/// `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Two { + s1: u8, + s2: u8, + v1: usize, + v2: usize, +} + +impl Two { + /// Create a new searcher that finds occurrences of the two needle bytes + /// given. + #[inline] + pub fn new(needle1: u8, needle2: u8) -> Two { + Two { + s1: needle1, + s2: needle2, + v1: splat(needle1), + v2: splat(needle2), + } + } + + /// A test-only routine so that we can bundle a bunch of quickcheck + /// properties into a single macro. Basically, this provides a constructor + /// that makes it identical to most other memchr implementations, which + /// have fallible constructors. + #[cfg(test)] + pub(crate) fn try_new(needle1: u8, needle2: u8) -> Option<Two> { + Some(Two::new(needle1, needle2)) + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value for a non-empty haystack is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value for a non-empty haystack is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let confirm = |b| self.confirm(b); + let len = end.distance(start); + if len < USIZE_BYTES { + return generic::fwd_byte_by_byte(start, end, confirm); + } + + // The start of the search may not be aligned to `*const usize`, + // so we do an unaligned load here. + let chunk = start.cast::<usize>().read_unaligned(); + if self.has_needle(chunk) { + return generic::fwd_byte_by_byte(start, end, confirm); + } + + // And now we start our search at a guaranteed aligned position. + // The first iteration of the loop below will overlap with the the + // unaligned chunk above in cases where the search starts at an + // unaligned offset, but that's okay as we're only here if that + // above didn't find a match. + let mut cur = + start.add(USIZE_BYTES - (start.as_usize() & USIZE_ALIGN)); + debug_assert!(cur > start); + debug_assert!(end.sub(USIZE_BYTES) >= start); + while cur <= end.sub(USIZE_BYTES) { + debug_assert_eq!(0, cur.as_usize() % USIZE_BYTES); + + let chunk = cur.cast::<usize>().read(); + if self.has_needle(chunk) { + break; + } + cur = cur.add(USIZE_BYTES); + } + generic::fwd_byte_by_byte(cur, end, confirm) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let confirm = |b| self.confirm(b); + let len = end.distance(start); + if len < USIZE_BYTES { + return generic::rev_byte_by_byte(start, end, confirm); + } + + let chunk = end.sub(USIZE_BYTES).cast::<usize>().read_unaligned(); + if self.has_needle(chunk) { + return generic::rev_byte_by_byte(start, end, confirm); + } + + let mut cur = end.sub(end.as_usize() & USIZE_ALIGN); + debug_assert!(start <= cur && cur <= end); + while cur >= start.add(USIZE_BYTES) { + debug_assert_eq!(0, cur.as_usize() % USIZE_BYTES); + + let chunk = cur.sub(USIZE_BYTES).cast::<usize>().read(); + if self.has_needle(chunk) { + break; + } + cur = cur.sub(USIZE_BYTES); + } + generic::rev_byte_by_byte(start, cur, confirm) + } + + /// Returns an iterator over all occurrences of one of the needle bytes in + /// the given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> TwoIter<'a, 'h> { + TwoIter { searcher: self, it: generic::Iter::new(haystack) } + } + + #[inline(always)] + fn has_needle(&self, chunk: usize) -> bool { + has_zero_byte(self.v1 ^ chunk) || has_zero_byte(self.v2 ^ chunk) + } + + #[inline(always)] + fn confirm(&self, haystack_byte: u8) -> bool { + self.s1 == haystack_byte || self.s2 == haystack_byte + } +} + +/// An iterator over all occurrences of two possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Two::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Two`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct TwoIter<'a, 'h> { + /// The underlying memchr searcher. + searcher: &'a Two, + /// Generic iterator implementation. + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for TwoIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for TwoIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +/// Finds all occurrences of three bytes in a haystack. +/// +/// That is, this reports matches of one of three possible bytes. For example, +/// searching for `a`, `b` or `o` in `afoobar` would report matches at offsets +/// `0`, `2`, `3`, `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Three { + s1: u8, + s2: u8, + s3: u8, + v1: usize, + v2: usize, + v3: usize, +} + +impl Three { + /// Create a new searcher that finds occurrences of the three needle bytes + /// given. + #[inline] + pub fn new(needle1: u8, needle2: u8, needle3: u8) -> Three { + Three { + s1: needle1, + s2: needle2, + s3: needle3, + v1: splat(needle1), + v2: splat(needle2), + v3: splat(needle3), + } + } + + /// A test-only routine so that we can bundle a bunch of quickcheck + /// properties into a single macro. Basically, this provides a constructor + /// that makes it identical to most other memchr implementations, which + /// have fallible constructors. + #[cfg(test)] + pub(crate) fn try_new( + needle1: u8, + needle2: u8, + needle3: u8, + ) -> Option<Three> { + Some(Three::new(needle1, needle2, needle3)) + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value for a non-empty haystack is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value for a non-empty haystack is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let confirm = |b| self.confirm(b); + let len = end.distance(start); + if len < USIZE_BYTES { + return generic::fwd_byte_by_byte(start, end, confirm); + } + + // The start of the search may not be aligned to `*const usize`, + // so we do an unaligned load here. + let chunk = start.cast::<usize>().read_unaligned(); + if self.has_needle(chunk) { + return generic::fwd_byte_by_byte(start, end, confirm); + } + + // And now we start our search at a guaranteed aligned position. + // The first iteration of the loop below will overlap with the the + // unaligned chunk above in cases where the search starts at an + // unaligned offset, but that's okay as we're only here if that + // above didn't find a match. + let mut cur = + start.add(USIZE_BYTES - (start.as_usize() & USIZE_ALIGN)); + debug_assert!(cur > start); + debug_assert!(end.sub(USIZE_BYTES) >= start); + while cur <= end.sub(USIZE_BYTES) { + debug_assert_eq!(0, cur.as_usize() % USIZE_BYTES); + + let chunk = cur.cast::<usize>().read(); + if self.has_needle(chunk) { + break; + } + cur = cur.add(USIZE_BYTES); + } + generic::fwd_byte_by_byte(cur, end, confirm) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let confirm = |b| self.confirm(b); + let len = end.distance(start); + if len < USIZE_BYTES { + return generic::rev_byte_by_byte(start, end, confirm); + } + + let chunk = end.sub(USIZE_BYTES).cast::<usize>().read_unaligned(); + if self.has_needle(chunk) { + return generic::rev_byte_by_byte(start, end, confirm); + } + + let mut cur = end.sub(end.as_usize() & USIZE_ALIGN); + debug_assert!(start <= cur && cur <= end); + while cur >= start.add(USIZE_BYTES) { + debug_assert_eq!(0, cur.as_usize() % USIZE_BYTES); + + let chunk = cur.sub(USIZE_BYTES).cast::<usize>().read(); + if self.has_needle(chunk) { + break; + } + cur = cur.sub(USIZE_BYTES); + } + generic::rev_byte_by_byte(start, cur, confirm) + } + + /// Returns an iterator over all occurrences of one of the needle bytes in + /// the given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> ThreeIter<'a, 'h> { + ThreeIter { searcher: self, it: generic::Iter::new(haystack) } + } + + #[inline(always)] + fn has_needle(&self, chunk: usize) -> bool { + has_zero_byte(self.v1 ^ chunk) + || has_zero_byte(self.v2 ^ chunk) + || has_zero_byte(self.v3 ^ chunk) + } + + #[inline(always)] + fn confirm(&self, haystack_byte: u8) -> bool { + self.s1 == haystack_byte + || self.s2 == haystack_byte + || self.s3 == haystack_byte + } +} + +/// An iterator over all occurrences of three possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Three::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Three`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct ThreeIter<'a, 'h> { + /// The underlying memchr searcher. + searcher: &'a Three, + /// Generic iterator implementation. + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for ThreeIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for ThreeIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +/// Return `true` if `x` contains any zero byte. +/// +/// That is, this routine treats `x` as a register of 8-bit lanes and returns +/// true when any of those lanes is `0`. +/// +/// From "Matters Computational" by J. Arndt. +#[inline(always)] +fn has_zero_byte(x: usize) -> bool { + // "The idea is to subtract one from each of the bytes and then look for + // bytes where the borrow propagated all the way to the most significant + // bit." + const LO: usize = splat(0x01); + const HI: usize = splat(0x80); + + (x.wrapping_sub(LO) & !x & HI) != 0 +} + +/// Repeat the given byte into a word size number. That is, every 8 bits +/// is equivalent to the given byte. For example, if `b` is `\x4E` or +/// `01001110` in binary, then the returned value on a 32-bit system would be: +/// `01001110_01001110_01001110_01001110`. +#[inline(always)] +const fn splat(b: u8) -> usize { + // TODO: use `usize::from` once it can be used in const context. + (b as usize) * (usize::MAX / 255) +} + +#[cfg(test)] +mod tests { + use super::*; + + define_memchr_quickcheck!(super, try_new); + + #[test] + fn forward_one() { + crate::tests::memchr::Runner::new(1).forward_iter( + |haystack, needles| { + Some(One::new(needles[0]).iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_one() { + crate::tests::memchr::Runner::new(1).reverse_iter( + |haystack, needles| { + Some(One::new(needles[0]).iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn count_one() { + crate::tests::memchr::Runner::new(1).count_iter(|haystack, needles| { + Some(One::new(needles[0]).iter(haystack).count()) + }) + } + + #[test] + fn forward_two() { + crate::tests::memchr::Runner::new(2).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2).iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_two() { + crate::tests::memchr::Runner::new(2).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2).iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn forward_three() { + crate::tests::memchr::Runner::new(3).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3).iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_three() { + crate::tests::memchr::Runner::new(3).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3).iter(haystack).rev().collect()) + }, + ) + } + + // This was found by quickcheck in the course of refactoring this crate + // after memchr 2.5.0. + #[test] + fn regression_double_ended_iterator() { + let finder = One::new(b'a'); + let haystack = "a"; + let mut it = finder.iter(haystack.as_bytes()); + assert_eq!(Some(0), it.next()); + assert_eq!(None, it.next_back()); + } + + // This regression test was caught by ripgrep's test suite on i686 when + // upgrading to memchr 2.6. Namely, something about the \x0B bytes here + // screws with the SWAR counting approach I was using. This regression test + // prompted me to remove the SWAR counting approach and just replace it + // with a byte-at-a-time loop. + #[test] + fn regression_count_new_lines() { + let haystack = "01234567\x0b\n\x0b\n\x0b\n\x0b\nx"; + let count = One::new(b'\n').count(haystack.as_bytes()); + assert_eq!(4, count); + } +} diff --git a/vendor/memchr/src/arch/all/mod.rs b/vendor/memchr/src/arch/all/mod.rs new file mode 100644 index 0000000..559cb75 --- /dev/null +++ b/vendor/memchr/src/arch/all/mod.rs @@ -0,0 +1,234 @@ +/*! +Contains architecture independent routines. + +These routines are often used as a "fallback" implementation when the more +specialized architecture dependent routines are unavailable. +*/ + +pub mod memchr; +pub mod packedpair; +pub mod rabinkarp; +#[cfg(feature = "alloc")] +pub mod shiftor; +pub mod twoway; + +/// Returns true if and only if `needle` is a prefix of `haystack`. +/// +/// This uses a latency optimized variant of `memcmp` internally which *might* +/// make this faster for very short strings. +/// +/// # Inlining +/// +/// This routine is marked `inline(always)`. If you want to call this function +/// in a way that is not always inlined, you'll need to wrap a call to it in +/// another function that is marked as `inline(never)` or just `inline`. +#[inline(always)] +pub fn is_prefix(haystack: &[u8], needle: &[u8]) -> bool { + needle.len() <= haystack.len() + && is_equal(&haystack[..needle.len()], needle) +} + +/// Returns true if and only if `needle` is a suffix of `haystack`. +/// +/// This uses a latency optimized variant of `memcmp` internally which *might* +/// make this faster for very short strings. +/// +/// # Inlining +/// +/// This routine is marked `inline(always)`. If you want to call this function +/// in a way that is not always inlined, you'll need to wrap a call to it in +/// another function that is marked as `inline(never)` or just `inline`. +#[inline(always)] +pub fn is_suffix(haystack: &[u8], needle: &[u8]) -> bool { + needle.len() <= haystack.len() + && is_equal(&haystack[haystack.len() - needle.len()..], needle) +} + +/// Compare corresponding bytes in `x` and `y` for equality. +/// +/// That is, this returns true if and only if `x.len() == y.len()` and +/// `x[i] == y[i]` for all `0 <= i < x.len()`. +/// +/// # Inlining +/// +/// This routine is marked `inline(always)`. If you want to call this function +/// in a way that is not always inlined, you'll need to wrap a call to it in +/// another function that is marked as `inline(never)` or just `inline`. +/// +/// # Motivation +/// +/// Why not use slice equality instead? Well, slice equality usually results in +/// a call out to the current platform's `libc` which might not be inlineable +/// or have other overhead. This routine isn't guaranteed to be a win, but it +/// might be in some cases. +#[inline(always)] +pub fn is_equal(x: &[u8], y: &[u8]) -> bool { + if x.len() != y.len() { + return false; + } + // SAFETY: Our pointers are derived directly from borrowed slices which + // uphold all of our safety guarantees except for length. We account for + // length with the check above. + unsafe { is_equal_raw(x.as_ptr(), y.as_ptr(), x.len()) } +} + +/// Compare `n` bytes at the given pointers for equality. +/// +/// This returns true if and only if `*x.add(i) == *y.add(i)` for all +/// `0 <= i < n`. +/// +/// # Inlining +/// +/// This routine is marked `inline(always)`. If you want to call this function +/// in a way that is not always inlined, you'll need to wrap a call to it in +/// another function that is marked as `inline(never)` or just `inline`. +/// +/// # Motivation +/// +/// Why not use slice equality instead? Well, slice equality usually results in +/// a call out to the current platform's `libc` which might not be inlineable +/// or have other overhead. This routine isn't guaranteed to be a win, but it +/// might be in some cases. +/// +/// # Safety +/// +/// * Both `x` and `y` must be valid for reads of up to `n` bytes. +/// * Both `x` and `y` must point to an initialized value. +/// * Both `x` and `y` must each point to an allocated object and +/// must either be in bounds or at most one byte past the end of the +/// allocated object. `x` and `y` do not need to point to the same allocated +/// object, but they may. +/// * Both `x` and `y` must be _derived from_ a pointer to their respective +/// allocated objects. +/// * The distance between `x` and `x+n` must not overflow `isize`. Similarly +/// for `y` and `y+n`. +/// * The distance being in bounds must not rely on "wrapping around" the +/// address space. +#[inline(always)] +pub unsafe fn is_equal_raw( + mut x: *const u8, + mut y: *const u8, + mut n: usize, +) -> bool { + // When we have 4 or more bytes to compare, then proceed in chunks of 4 at + // a time using unaligned loads. + // + // Also, why do 4 byte loads instead of, say, 8 byte loads? The reason is + // that this particular version of memcmp is likely to be called with tiny + // needles. That means that if we do 8 byte loads, then a higher proportion + // of memcmp calls will use the slower variant above. With that said, this + // is a hypothesis and is only loosely supported by benchmarks. There's + // likely some improvement that could be made here. The main thing here + // though is to optimize for latency, not throughput. + + // SAFETY: The caller is responsible for ensuring the pointers we get are + // valid and readable for at least `n` bytes. We also do unaligned loads, + // so there's no need to ensure we're aligned. (This is justified by this + // routine being specifically for short strings.) + while n >= 4 { + let vx = x.cast::<u32>().read_unaligned(); + let vy = y.cast::<u32>().read_unaligned(); + if vx != vy { + return false; + } + x = x.add(4); + y = y.add(4); + n -= 4; + } + // If we don't have enough bytes to do 4-byte at a time loads, then + // do partial loads. Note that I used to have a byte-at-a-time + // loop here and that turned out to be quite a bit slower for the + // memmem/pathological/defeat-simple-vector-alphabet benchmark. + if n >= 2 { + let vx = x.cast::<u16>().read_unaligned(); + let vy = y.cast::<u16>().read_unaligned(); + if vx != vy { + return false; + } + x = x.add(2); + y = y.add(2); + n -= 2; + } + if n > 0 { + if x.read() != y.read() { + return false; + } + } + true +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn equals_different_lengths() { + assert!(!is_equal(b"", b"a")); + assert!(!is_equal(b"a", b"")); + assert!(!is_equal(b"ab", b"a")); + assert!(!is_equal(b"a", b"ab")); + } + + #[test] + fn equals_mismatch() { + let one_mismatch = [ + (&b"a"[..], &b"x"[..]), + (&b"ab"[..], &b"ax"[..]), + (&b"abc"[..], &b"abx"[..]), + (&b"abcd"[..], &b"abcx"[..]), + (&b"abcde"[..], &b"abcdx"[..]), + (&b"abcdef"[..], &b"abcdex"[..]), + (&b"abcdefg"[..], &b"abcdefx"[..]), + (&b"abcdefgh"[..], &b"abcdefgx"[..]), + (&b"abcdefghi"[..], &b"abcdefghx"[..]), + (&b"abcdefghij"[..], &b"abcdefghix"[..]), + (&b"abcdefghijk"[..], &b"abcdefghijx"[..]), + (&b"abcdefghijkl"[..], &b"abcdefghijkx"[..]), + (&b"abcdefghijklm"[..], &b"abcdefghijklx"[..]), + (&b"abcdefghijklmn"[..], &b"abcdefghijklmx"[..]), + ]; + for (x, y) in one_mismatch { + assert_eq!(x.len(), y.len(), "lengths should match"); + assert!(!is_equal(x, y)); + assert!(!is_equal(y, x)); + } + } + + #[test] + fn equals_yes() { + assert!(is_equal(b"", b"")); + assert!(is_equal(b"a", b"a")); + assert!(is_equal(b"ab", b"ab")); + assert!(is_equal(b"abc", b"abc")); + assert!(is_equal(b"abcd", b"abcd")); + assert!(is_equal(b"abcde", b"abcde")); + assert!(is_equal(b"abcdef", b"abcdef")); + assert!(is_equal(b"abcdefg", b"abcdefg")); + assert!(is_equal(b"abcdefgh", b"abcdefgh")); + assert!(is_equal(b"abcdefghi", b"abcdefghi")); + } + + #[test] + fn prefix() { + assert!(is_prefix(b"", b"")); + assert!(is_prefix(b"a", b"")); + assert!(is_prefix(b"ab", b"")); + assert!(is_prefix(b"foo", b"foo")); + assert!(is_prefix(b"foobar", b"foo")); + + assert!(!is_prefix(b"foo", b"fob")); + assert!(!is_prefix(b"foobar", b"fob")); + } + + #[test] + fn suffix() { + assert!(is_suffix(b"", b"")); + assert!(is_suffix(b"a", b"")); + assert!(is_suffix(b"ab", b"")); + assert!(is_suffix(b"foo", b"foo")); + assert!(is_suffix(b"foobar", b"bar")); + + assert!(!is_suffix(b"foo", b"goo")); + assert!(!is_suffix(b"foobar", b"gar")); + } +} diff --git a/vendor/memchr/src/arch/all/packedpair/default_rank.rs b/vendor/memchr/src/arch/all/packedpair/default_rank.rs new file mode 100644 index 0000000..6aa3895 --- /dev/null +++ b/vendor/memchr/src/arch/all/packedpair/default_rank.rs @@ -0,0 +1,258 @@ +pub(crate) const RANK: [u8; 256] = [ + 55, // '\x00' + 52, // '\x01' + 51, // '\x02' + 50, // '\x03' + 49, // '\x04' + 48, // '\x05' + 47, // '\x06' + 46, // '\x07' + 45, // '\x08' + 103, // '\t' + 242, // '\n' + 66, // '\x0b' + 67, // '\x0c' + 229, // '\r' + 44, // '\x0e' + 43, // '\x0f' + 42, // '\x10' + 41, // '\x11' + 40, // '\x12' + 39, // '\x13' + 38, // '\x14' + 37, // '\x15' + 36, // '\x16' + 35, // '\x17' + 34, // '\x18' + 33, // '\x19' + 56, // '\x1a' + 32, // '\x1b' + 31, // '\x1c' + 30, // '\x1d' + 29, // '\x1e' + 28, // '\x1f' + 255, // ' ' + 148, // '!' + 164, // '"' + 149, // '#' + 136, // '$' + 160, // '%' + 155, // '&' + 173, // "'" + 221, // '(' + 222, // ')' + 134, // '*' + 122, // '+' + 232, // ',' + 202, // '-' + 215, // '.' + 224, // '/' + 208, // '0' + 220, // '1' + 204, // '2' + 187, // '3' + 183, // '4' + 179, // '5' + 177, // '6' + 168, // '7' + 178, // '8' + 200, // '9' + 226, // ':' + 195, // ';' + 154, // '<' + 184, // '=' + 174, // '>' + 126, // '?' + 120, // '@' + 191, // 'A' + 157, // 'B' + 194, // 'C' + 170, // 'D' + 189, // 'E' + 162, // 'F' + 161, // 'G' + 150, // 'H' + 193, // 'I' + 142, // 'J' + 137, // 'K' + 171, // 'L' + 176, // 'M' + 185, // 'N' + 167, // 'O' + 186, // 'P' + 112, // 'Q' + 175, // 'R' + 192, // 'S' + 188, // 'T' + 156, // 'U' + 140, // 'V' + 143, // 'W' + 123, // 'X' + 133, // 'Y' + 128, // 'Z' + 147, // '[' + 138, // '\\' + 146, // ']' + 114, // '^' + 223, // '_' + 151, // '`' + 249, // 'a' + 216, // 'b' + 238, // 'c' + 236, // 'd' + 253, // 'e' + 227, // 'f' + 218, // 'g' + 230, // 'h' + 247, // 'i' + 135, // 'j' + 180, // 'k' + 241, // 'l' + 233, // 'm' + 246, // 'n' + 244, // 'o' + 231, // 'p' + 139, // 'q' + 245, // 'r' + 243, // 's' + 251, // 't' + 235, // 'u' + 201, // 'v' + 196, // 'w' + 240, // 'x' + 214, // 'y' + 152, // 'z' + 182, // '{' + 205, // '|' + 181, // '}' + 127, // '~' + 27, // '\x7f' + 212, // '\x80' + 211, // '\x81' + 210, // '\x82' + 213, // '\x83' + 228, // '\x84' + 197, // '\x85' + 169, // '\x86' + 159, // '\x87' + 131, // '\x88' + 172, // '\x89' + 105, // '\x8a' + 80, // '\x8b' + 98, // '\x8c' + 96, // '\x8d' + 97, // '\x8e' + 81, // '\x8f' + 207, // '\x90' + 145, // '\x91' + 116, // '\x92' + 115, // '\x93' + 144, // '\x94' + 130, // '\x95' + 153, // '\x96' + 121, // '\x97' + 107, // '\x98' + 132, // '\x99' + 109, // '\x9a' + 110, // '\x9b' + 124, // '\x9c' + 111, // '\x9d' + 82, // '\x9e' + 108, // '\x9f' + 118, // '\xa0' + 141, // '¡' + 113, // '¢' + 129, // '£' + 119, // '¤' + 125, // '¥' + 165, // '¦' + 117, // '§' + 92, // '¨' + 106, // '©' + 83, // 'ª' + 72, // '«' + 99, // '¬' + 93, // '\xad' + 65, // '®' + 79, // '¯' + 166, // '°' + 237, // '±' + 163, // '²' + 199, // '³' + 190, // '´' + 225, // 'µ' + 209, // '¶' + 203, // '·' + 198, // '¸' + 217, // '¹' + 219, // 'º' + 206, // '»' + 234, // '¼' + 248, // '½' + 158, // '¾' + 239, // '¿' + 255, // 'À' + 255, // 'Á' + 255, // 'Â' + 255, // 'Ã' + 255, // 'Ä' + 255, // 'Å' + 255, // 'Æ' + 255, // 'Ç' + 255, // 'È' + 255, // 'É' + 255, // 'Ê' + 255, // 'Ë' + 255, // 'Ì' + 255, // 'Í' + 255, // 'Î' + 255, // 'Ï' + 255, // 'Ð' + 255, // 'Ñ' + 255, // 'Ò' + 255, // 'Ó' + 255, // 'Ô' + 255, // 'Õ' + 255, // 'Ö' + 255, // '×' + 255, // 'Ø' + 255, // 'Ù' + 255, // 'Ú' + 255, // 'Û' + 255, // 'Ü' + 255, // 'Ý' + 255, // 'Þ' + 255, // 'ß' + 255, // 'à' + 255, // 'á' + 255, // 'â' + 255, // 'ã' + 255, // 'ä' + 255, // 'å' + 255, // 'æ' + 255, // 'ç' + 255, // 'è' + 255, // 'é' + 255, // 'ê' + 255, // 'ë' + 255, // 'ì' + 255, // 'í' + 255, // 'î' + 255, // 'ï' + 255, // 'ð' + 255, // 'ñ' + 255, // 'ò' + 255, // 'ó' + 255, // 'ô' + 255, // 'õ' + 255, // 'ö' + 255, // '÷' + 255, // 'ø' + 255, // 'ù' + 255, // 'ú' + 255, // 'û' + 255, // 'ü' + 255, // 'ý' + 255, // 'þ' + 255, // 'ÿ' +]; diff --git a/vendor/memchr/src/arch/all/packedpair/mod.rs b/vendor/memchr/src/arch/all/packedpair/mod.rs new file mode 100644 index 0000000..148a985 --- /dev/null +++ b/vendor/memchr/src/arch/all/packedpair/mod.rs @@ -0,0 +1,359 @@ +/*! +Provides an architecture independent implementation of the "packed pair" +algorithm. + +The "packed pair" algorithm is based on the [generic SIMD] algorithm. The main +difference is that it (by default) uses a background distribution of byte +frequencies to heuristically select the pair of bytes to search for. Note that +this module provides an architecture independent version that doesn't do as +good of a job keeping the search for candidates inside a SIMD hot path. It +however can be good enough in many circumstances. + +[generic SIMD]: http://0x80.pl/articles/simd-strfind.html#first-and-last +*/ + +use crate::memchr; + +mod default_rank; + +/// An architecture independent "packed pair" finder. +/// +/// This finder picks two bytes that it believes have high predictive power for +/// indicating an overall match of a needle. At search time, it reports offsets +/// where the needle could match based on whether the pair of bytes it chose +/// match. +/// +/// This is architecture independent because it utilizes `memchr` to find the +/// occurrence of one of the bytes in the pair, and then checks whether the +/// second byte matches. If it does, in the case of [`Finder::find_prefilter`], +/// the location at which the needle could match is returned. +/// +/// It is generally preferred to use architecture specific routines for a +/// "packed pair" prefilter, but this can be a useful fallback when the +/// architecture independent routines are unavailable. +#[derive(Clone, Copy, Debug)] +pub struct Finder { + pair: Pair, + byte1: u8, + byte2: u8, +} + +impl Finder { + /// Create a new prefilter that reports possible locations where the given + /// needle matches. + #[inline] + pub fn new(needle: &[u8]) -> Option<Finder> { + Finder::with_pair(needle, Pair::new(needle)?) + } + + /// Create a new prefilter using the pair given. + /// + /// If the prefilter could not be constructed, then `None` is returned. + /// + /// This constructor permits callers to control precisely which pair of + /// bytes is used as a predicate. + #[inline] + pub fn with_pair(needle: &[u8], pair: Pair) -> Option<Finder> { + let byte1 = needle[usize::from(pair.index1())]; + let byte2 = needle[usize::from(pair.index2())]; + // Currently this can never fail so we could just return a Finder, + // but it's conceivable this could change. + Some(Finder { pair, byte1, byte2 }) + } + + /// Run this finder on the given haystack as a prefilter. + /// + /// If a candidate match is found, then an offset where the needle *could* + /// begin in the haystack is returned. + #[inline] + pub fn find_prefilter(&self, haystack: &[u8]) -> Option<usize> { + let mut i = 0; + let index1 = usize::from(self.pair.index1()); + let index2 = usize::from(self.pair.index2()); + loop { + // Use a fast vectorized implementation to skip to the next + // occurrence of the rarest byte (heuristically chosen) in the + // needle. + i += memchr(self.byte1, &haystack[i..])?; + let found = i; + i += 1; + + // If we can't align our first byte match with the haystack, then a + // match is impossible. + let aligned1 = match found.checked_sub(index1) { + None => continue, + Some(aligned1) => aligned1, + }; + + // Now align the second byte match with the haystack. A mismatch + // means that a match is impossible. + let aligned2 = match aligned1.checked_add(index2) { + None => continue, + Some(aligned_index2) => aligned_index2, + }; + if haystack.get(aligned2).map_or(true, |&b| b != self.byte2) { + continue; + } + + // We've done what we can. There might be a match here. + return Some(aligned1); + } + } + + /// Returns the pair of offsets (into the needle) used to check as a + /// predicate before confirming whether a needle exists at a particular + /// position. + #[inline] + pub fn pair(&self) -> &Pair { + &self.pair + } +} + +/// A pair of byte offsets into a needle to use as a predicate. +/// +/// This pair is used as a predicate to quickly filter out positions in a +/// haystack in which a needle cannot match. In some cases, this pair can even +/// be used in vector algorithms such that the vector algorithm only switches +/// over to scalar code once this pair has been found. +/// +/// A pair of offsets can be used in both substring search implementations and +/// in prefilters. The former will report matches of a needle in a haystack +/// where as the latter will only report possible matches of a needle. +/// +/// The offsets are limited each to a maximum of 255 to keep memory usage low. +/// Moreover, it's rarely advantageous to create a predicate using offsets +/// greater than 255 anyway. +/// +/// The only guarantee enforced on the pair of offsets is that they are not +/// equivalent. It is not necessarily the case that `index1 < index2` for +/// example. By convention, `index1` corresponds to the byte in the needle +/// that is believed to be most the predictive. Note also that because of the +/// requirement that the indices be both valid for the needle used to build +/// the pair and not equal, it follows that a pair can only be constructed for +/// needles with length at least 2. +#[derive(Clone, Copy, Debug)] +pub struct Pair { + index1: u8, + index2: u8, +} + +impl Pair { + /// Create a new pair of offsets from the given needle. + /// + /// If a pair could not be created (for example, if the needle is too + /// short), then `None` is returned. + /// + /// This chooses the pair in the needle that is believed to be as + /// predictive of an overall match of the needle as possible. + #[inline] + pub fn new(needle: &[u8]) -> Option<Pair> { + Pair::with_ranker(needle, DefaultFrequencyRank) + } + + /// Create a new pair of offsets from the given needle and ranker. + /// + /// This permits the caller to choose a background frequency distribution + /// with which bytes are selected. The idea is to select a pair of bytes + /// that is believed to strongly predict a match in the haystack. This + /// usually means selecting bytes that occur rarely in a haystack. + /// + /// If a pair could not be created (for example, if the needle is too + /// short), then `None` is returned. + #[inline] + pub fn with_ranker<R: HeuristicFrequencyRank>( + needle: &[u8], + ranker: R, + ) -> Option<Pair> { + if needle.len() <= 1 { + return None; + } + // Find the rarest two bytes. We make them distinct indices by + // construction. (The actual byte value may be the same in degenerate + // cases, but that's OK.) + let (mut rare1, mut index1) = (needle[0], 0); + let (mut rare2, mut index2) = (needle[1], 1); + if ranker.rank(rare2) < ranker.rank(rare1) { + core::mem::swap(&mut rare1, &mut rare2); + core::mem::swap(&mut index1, &mut index2); + } + let max = usize::from(core::u8::MAX); + for (i, &b) in needle.iter().enumerate().take(max).skip(2) { + if ranker.rank(b) < ranker.rank(rare1) { + rare2 = rare1; + index2 = index1; + rare1 = b; + index1 = u8::try_from(i).unwrap(); + } else if b != rare1 && ranker.rank(b) < ranker.rank(rare2) { + rare2 = b; + index2 = u8::try_from(i).unwrap(); + } + } + // While not strictly required for how a Pair is normally used, we + // really don't want these to be equivalent. If they were, it would + // reduce the effectiveness of candidate searching using these rare + // bytes by increasing the rate of false positives. + assert_ne!(index1, index2); + Some(Pair { index1, index2 }) + } + + /// Create a new pair using the offsets given for the needle given. + /// + /// This bypasses any sort of heuristic process for choosing the offsets + /// and permits the caller to choose the offsets themselves. + /// + /// Indices are limited to valid `u8` values so that a `Pair` uses less + /// memory. It is not possible to create a `Pair` with offsets bigger than + /// `u8::MAX`. It's likely that such a thing is not needed, but if it is, + /// it's suggested to build your own bespoke algorithm because you're + /// likely working on a very niche case. (File an issue if this suggestion + /// does not make sense to you.) + /// + /// If a pair could not be created (for example, if the needle is too + /// short), then `None` is returned. + #[inline] + pub fn with_indices( + needle: &[u8], + index1: u8, + index2: u8, + ) -> Option<Pair> { + // While not strictly required for how a Pair is normally used, we + // really don't want these to be equivalent. If they were, it would + // reduce the effectiveness of candidate searching using these rare + // bytes by increasing the rate of false positives. + if index1 == index2 { + return None; + } + // Similarly, invalid indices means the Pair is invalid too. + if usize::from(index1) >= needle.len() { + return None; + } + if usize::from(index2) >= needle.len() { + return None; + } + Some(Pair { index1, index2 }) + } + + /// Returns the first offset of the pair. + #[inline] + pub fn index1(&self) -> u8 { + self.index1 + } + + /// Returns the second offset of the pair. + #[inline] + pub fn index2(&self) -> u8 { + self.index2 + } +} + +/// This trait allows the user to customize the heuristic used to determine the +/// relative frequency of a given byte in the dataset being searched. +/// +/// The use of this trait can have a dramatic impact on performance depending +/// on the type of data being searched. The details of why are explained in the +/// docs of [`crate::memmem::Prefilter`]. To summarize, the core algorithm uses +/// a prefilter to quickly identify candidate matches that are later verified +/// more slowly. This prefilter is implemented in terms of trying to find +/// `rare` bytes at specific offsets that will occur less frequently in the +/// dataset. While the concept of a `rare` byte is similar for most datasets, +/// there are some specific datasets (like binary executables) that have +/// dramatically different byte distributions. For these datasets customizing +/// the byte frequency heuristic can have a massive impact on performance, and +/// might even need to be done at runtime. +/// +/// The default implementation of `HeuristicFrequencyRank` reads from the +/// static frequency table defined in `src/memmem/byte_frequencies.rs`. This +/// is optimal for most inputs, so if you are unsure of the impact of using a +/// custom `HeuristicFrequencyRank` you should probably just use the default. +/// +/// # Example +/// +/// ``` +/// use memchr::{ +/// arch::all::packedpair::HeuristicFrequencyRank, +/// memmem::FinderBuilder, +/// }; +/// +/// /// A byte-frequency table that is good for scanning binary executables. +/// struct Binary; +/// +/// impl HeuristicFrequencyRank for Binary { +/// fn rank(&self, byte: u8) -> u8 { +/// const TABLE: [u8; 256] = [ +/// 255, 128, 61, 43, 50, 41, 27, 28, 57, 15, 21, 13, 24, 17, 17, +/// 89, 58, 16, 11, 7, 14, 23, 7, 6, 24, 9, 6, 5, 9, 4, 7, 16, +/// 68, 11, 9, 6, 88, 7, 4, 4, 23, 9, 4, 8, 8, 5, 10, 4, 30, 11, +/// 9, 24, 11, 5, 5, 5, 19, 11, 6, 17, 9, 9, 6, 8, +/// 48, 58, 11, 14, 53, 40, 9, 9, 254, 35, 3, 6, 52, 23, 6, 6, 27, +/// 4, 7, 11, 14, 13, 10, 11, 11, 5, 2, 10, 16, 12, 6, 19, +/// 19, 20, 5, 14, 16, 31, 19, 7, 14, 20, 4, 4, 19, 8, 18, 20, 24, +/// 1, 25, 19, 58, 29, 10, 5, 15, 20, 2, 2, 9, 4, 3, 5, +/// 51, 11, 4, 53, 23, 39, 6, 4, 13, 81, 4, 186, 5, 67, 3, 2, 15, +/// 0, 0, 1, 3, 2, 0, 0, 5, 0, 0, 0, 2, 0, 0, 0, +/// 12, 2, 1, 1, 3, 1, 1, 1, 6, 1, 2, 1, 3, 1, 1, 2, 9, 1, 1, 0, +/// 2, 2, 4, 4, 11, 6, 7, 3, 6, 9, 4, 5, +/// 46, 18, 8, 18, 17, 3, 8, 20, 16, 10, 3, 7, 175, 4, 6, 7, 13, +/// 3, 7, 3, 3, 1, 3, 3, 10, 3, 1, 5, 2, 0, 1, 2, +/// 16, 3, 5, 1, 6, 1, 1, 2, 58, 20, 3, 14, 12, 2, 1, 3, 16, 3, 5, +/// 8, 3, 1, 8, 6, 17, 6, 5, 3, 8, 6, 13, 175, +/// ]; +/// TABLE[byte as usize] +/// } +/// } +/// // Create a new finder with the custom heuristic. +/// let finder = FinderBuilder::new() +/// .build_forward_with_ranker(Binary, b"\x00\x00\xdd\xdd"); +/// // Find needle with custom heuristic. +/// assert!(finder.find(b"\x00\x00\x00\xdd\xdd").is_some()); +/// ``` +pub trait HeuristicFrequencyRank { + /// Return the heuristic frequency rank of the given byte. A lower rank + /// means the byte is believed to occur less frequently in the haystack. + /// + /// Some uses of this heuristic may treat arbitrary absolute rank values as + /// significant. For example, an implementation detail in this crate may + /// determine that heuristic prefilters are inappropriate if every byte in + /// the needle has a "high" rank. + fn rank(&self, byte: u8) -> u8; +} + +/// The default byte frequency heuristic that is good for most haystacks. +pub(crate) struct DefaultFrequencyRank; + +impl HeuristicFrequencyRank for DefaultFrequencyRank { + fn rank(&self, byte: u8) -> u8 { + self::default_rank::RANK[usize::from(byte)] + } +} + +/// This permits passing any implementation of `HeuristicFrequencyRank` as a +/// borrowed version of itself. +impl<'a, R> HeuristicFrequencyRank for &'a R +where + R: HeuristicFrequencyRank, +{ + fn rank(&self, byte: u8) -> u8 { + (**self).rank(byte) + } +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn forward_packedpair() { + fn find( + haystack: &[u8], + needle: &[u8], + _index1: u8, + _index2: u8, + ) -> Option<Option<usize>> { + // We ignore the index positions requested since it winds up making + // this test too slow overall. + let f = Finder::new(needle)?; + Some(f.find_prefilter(haystack)) + } + crate::tests::packedpair::Runner::new().fwd(find).run() + } +} diff --git a/vendor/memchr/src/arch/all/rabinkarp.rs b/vendor/memchr/src/arch/all/rabinkarp.rs new file mode 100644 index 0000000..e0bafba --- /dev/null +++ b/vendor/memchr/src/arch/all/rabinkarp.rs @@ -0,0 +1,390 @@ +/*! +An implementation of the [Rabin-Karp substring search algorithm][rabinkarp]. + +Rabin-Karp works by creating a hash of the needle provided and then computing +a rolling hash for each needle sized window in the haystack. When the rolling +hash matches the hash of the needle, a byte-wise comparison is done to check +if a match exists. The worst case time complexity of Rabin-Karp is `O(m * +n)` where `m ~ len(needle)` and `n ~ len(haystack)`. Its worst case space +complexity is constant. + +The main utility of Rabin-Karp is that the searcher can be constructed very +quickly with very little memory. This makes it especially useful when searching +for small needles in small haystacks, as it might finish its search before a +beefier algorithm (like Two-Way) even starts. + +[rabinkarp]: https://en.wikipedia.org/wiki/Rabin%E2%80%93Karp_algorithm +*/ + +/* +(This was the comment I wrote for this module originally when it was not +exposed. The comment still looks useful, but it's a bit in the weeds, so it's +not public itself.) + +This module implements the classical Rabin-Karp substring search algorithm, +with no extra frills. While its use would seem to break our time complexity +guarantee of O(m+n) (RK's time complexity is O(mn)), we are careful to only +ever use RK on a constant subset of haystacks. The main point here is that +RK has good latency properties for small needles/haystacks. It's very quick +to compute a needle hash and zip through the haystack when compared to +initializing Two-Way, for example. And this is especially useful for cases +where the haystack is just too short for vector instructions to do much good. + +The hashing function used here is the same one recommended by ESMAJ. + +Another choice instead of Rabin-Karp would be Shift-Or. But its latency +isn't quite as good since its preprocessing time is a bit more expensive +(both in practice and in theory). However, perhaps Shift-Or has a place +somewhere else for short patterns. I think the main problem is that it +requires space proportional to the alphabet and the needle. If we, for +example, supported needles up to length 16, then the total table size would be +len(alphabet)*size_of::<u16>()==512 bytes. Which isn't exactly small, and it's +probably bad to put that on the stack. So ideally, we'd throw it on the heap, +but we'd really like to write as much code without using alloc/std as possible. +But maybe it's worth the special casing. It's a TODO to benchmark. + +Wikipedia has a decent explanation, if a bit heavy on the theory: +https://en.wikipedia.org/wiki/Rabin%E2%80%93Karp_algorithm + +But ESMAJ provides something a bit more concrete: +http://www-igm.univ-mlv.fr/~lecroq/string/node5.html + +Finally, aho-corasick uses Rabin-Karp for multiple pattern match in some cases: +https://github.com/BurntSushi/aho-corasick/blob/3852632f10587db0ff72ef29e88d58bf305a0946/src/packed/rabinkarp.rs +*/ + +use crate::ext::Pointer; + +/// A forward substring searcher using the Rabin-Karp algorithm. +/// +/// Note that, as a lower level API, a `Finder` does not have access to the +/// needle it was constructed with. For this reason, executing a search +/// with a `Finder` requires passing both the needle and the haystack, +/// where the needle is exactly equivalent to the one given to the `Finder` +/// at construction time. This design was chosen so that callers can have +/// more precise control over where and how many times a needle is stored. +/// For example, in cases where Rabin-Karp is just one of several possible +/// substring search algorithms. +#[derive(Clone, Debug)] +pub struct Finder { + /// The actual hash. + hash: Hash, + /// The factor needed to multiply a byte by in order to subtract it from + /// the hash. It is defined to be 2^(n-1) (using wrapping exponentiation), + /// where n is the length of the needle. This is how we "remove" a byte + /// from the hash once the hash window rolls past it. + hash_2pow: u32, +} + +impl Finder { + /// Create a new Rabin-Karp forward searcher for the given `needle`. + /// + /// The needle may be empty. The empty needle matches at every byte offset. + /// + /// Note that callers must pass the same needle to all search calls using + /// this `Finder`. + #[inline] + pub fn new(needle: &[u8]) -> Finder { + let mut s = Finder { hash: Hash::new(), hash_2pow: 1 }; + let first_byte = match needle.get(0) { + None => return s, + Some(&first_byte) => first_byte, + }; + s.hash.add(first_byte); + for b in needle.iter().copied().skip(1) { + s.hash.add(b); + s.hash_2pow = s.hash_2pow.wrapping_shl(1); + } + s + } + + /// Return the first occurrence of the `needle` in the `haystack` + /// given. If no such occurrence exists, then `None` is returned. + /// + /// The `needle` provided must match the needle given to this finder at + /// construction time. + /// + /// The maximum value this can return is `haystack.len()`, which can only + /// occur when the needle and haystack both have length zero. Otherwise, + /// for non-empty haystacks, the maximum value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8], needle: &[u8]) -> Option<usize> { + unsafe { + let hstart = haystack.as_ptr(); + let hend = hstart.add(haystack.len()); + let nstart = needle.as_ptr(); + let nend = nstart.add(needle.len()); + let found = self.find_raw(hstart, hend, nstart, nend)?; + Some(found.distance(hstart)) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `<= end`. The pointer returned is only ever equivalent + /// to `end` when both the needle and haystack are empty. (That is, the + /// empty string matches the empty string.) + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// Note that `start` and `end` below refer to both pairs of pointers given + /// to this routine. That is, the conditions apply to both `hstart`/`hend` + /// and `nstart`/`nend`. + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// * It must be the case that `start <= end`. + #[inline] + pub unsafe fn find_raw( + &self, + hstart: *const u8, + hend: *const u8, + nstart: *const u8, + nend: *const u8, + ) -> Option<*const u8> { + let hlen = hend.distance(hstart); + let nlen = nend.distance(nstart); + if nlen > hlen { + return None; + } + let mut cur = hstart; + let end = hend.sub(nlen); + let mut hash = Hash::forward(cur, cur.add(nlen)); + loop { + if self.hash == hash && is_equal_raw(cur, nstart, nlen) { + return Some(cur); + } + if cur >= end { + return None; + } + hash.roll(self, cur.read(), cur.add(nlen).read()); + cur = cur.add(1); + } + } +} + +/// A reverse substring searcher using the Rabin-Karp algorithm. +#[derive(Clone, Debug)] +pub struct FinderRev(Finder); + +impl FinderRev { + /// Create a new Rabin-Karp reverse searcher for the given `needle`. + #[inline] + pub fn new(needle: &[u8]) -> FinderRev { + let mut s = FinderRev(Finder { hash: Hash::new(), hash_2pow: 1 }); + let last_byte = match needle.last() { + None => return s, + Some(&last_byte) => last_byte, + }; + s.0.hash.add(last_byte); + for b in needle.iter().rev().copied().skip(1) { + s.0.hash.add(b); + s.0.hash_2pow = s.0.hash_2pow.wrapping_shl(1); + } + s + } + + /// Return the last occurrence of the `needle` in the `haystack` + /// given. If no such occurrence exists, then `None` is returned. + /// + /// The `needle` provided must match the needle given to this finder at + /// construction time. + /// + /// The maximum value this can return is `haystack.len()`, which can only + /// occur when the needle and haystack both have length zero. Otherwise, + /// for non-empty haystacks, the maximum value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8], needle: &[u8]) -> Option<usize> { + unsafe { + let hstart = haystack.as_ptr(); + let hend = hstart.add(haystack.len()); + let nstart = needle.as_ptr(); + let nend = nstart.add(needle.len()); + let found = self.rfind_raw(hstart, hend, nstart, nend)?; + Some(found.distance(hstart)) + } + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `<= end`. The pointer returned is only ever equivalent + /// to `end` when both the needle and haystack are empty. (That is, the + /// empty string matches the empty string.) + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// Note that `start` and `end` below refer to both pairs of pointers given + /// to this routine. That is, the conditions apply to both `hstart`/`hend` + /// and `nstart`/`nend`. + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// * It must be the case that `start <= end`. + #[inline] + pub unsafe fn rfind_raw( + &self, + hstart: *const u8, + hend: *const u8, + nstart: *const u8, + nend: *const u8, + ) -> Option<*const u8> { + let hlen = hend.distance(hstart); + let nlen = nend.distance(nstart); + if nlen > hlen { + return None; + } + let mut cur = hend.sub(nlen); + let start = hstart; + let mut hash = Hash::reverse(cur, cur.add(nlen)); + loop { + if self.0.hash == hash && is_equal_raw(cur, nstart, nlen) { + return Some(cur); + } + if cur <= start { + return None; + } + cur = cur.sub(1); + hash.roll(&self.0, cur.add(nlen).read(), cur.read()); + } + } +} + +/// Whether RK is believed to be very fast for the given needle/haystack. +#[inline] +pub(crate) fn is_fast(haystack: &[u8], _needle: &[u8]) -> bool { + haystack.len() < 16 +} + +/// A Rabin-Karp hash. This might represent the hash of a needle, or the hash +/// of a rolling window in the haystack. +#[derive(Clone, Copy, Debug, Default, Eq, PartialEq)] +struct Hash(u32); + +impl Hash { + /// Create a new hash that represents the empty string. + #[inline(always)] + fn new() -> Hash { + Hash(0) + } + + /// Create a new hash from the bytes given for use in forward searches. + /// + /// # Safety + /// + /// The given pointers must be valid to read from within their range. + #[inline(always)] + unsafe fn forward(mut start: *const u8, end: *const u8) -> Hash { + let mut hash = Hash::new(); + while start < end { + hash.add(start.read()); + start = start.add(1); + } + hash + } + + /// Create a new hash from the bytes given for use in reverse searches. + /// + /// # Safety + /// + /// The given pointers must be valid to read from within their range. + #[inline(always)] + unsafe fn reverse(start: *const u8, mut end: *const u8) -> Hash { + let mut hash = Hash::new(); + while start < end { + end = end.sub(1); + hash.add(end.read()); + } + hash + } + + /// Add 'new' and remove 'old' from this hash. The given needle hash should + /// correspond to the hash computed for the needle being searched for. + /// + /// This is meant to be used when the rolling window of the haystack is + /// advanced. + #[inline(always)] + fn roll(&mut self, finder: &Finder, old: u8, new: u8) { + self.del(finder, old); + self.add(new); + } + + /// Add a byte to this hash. + #[inline(always)] + fn add(&mut self, byte: u8) { + self.0 = self.0.wrapping_shl(1).wrapping_add(u32::from(byte)); + } + + /// Remove a byte from this hash. The given needle hash should correspond + /// to the hash computed for the needle being searched for. + #[inline(always)] + fn del(&mut self, finder: &Finder, byte: u8) { + let factor = finder.hash_2pow; + self.0 = self.0.wrapping_sub(u32::from(byte).wrapping_mul(factor)); + } +} + +/// Returns true when `x[i] == y[i]` for all `0 <= i < n`. +/// +/// We forcefully don't inline this to hint at the compiler that it is unlikely +/// to be called. This causes the inner rabinkarp loop above to be a bit +/// tighter and leads to some performance improvement. See the +/// memmem/krate/prebuilt/sliceslice-words/words benchmark. +/// +/// # Safety +/// +/// Same as `crate::arch::all::is_equal_raw`. +#[cold] +#[inline(never)] +unsafe fn is_equal_raw(x: *const u8, y: *const u8, n: usize) -> bool { + crate::arch::all::is_equal_raw(x, y, n) +} + +#[cfg(test)] +mod tests { + use super::*; + + define_substring_forward_quickcheck!(|h, n| Some( + Finder::new(n).find(h, n) + )); + define_substring_reverse_quickcheck!(|h, n| Some( + FinderRev::new(n).rfind(h, n) + )); + + #[test] + fn forward() { + crate::tests::substring::Runner::new() + .fwd(|h, n| Some(Finder::new(n).find(h, n))) + .run(); + } + + #[test] + fn reverse() { + crate::tests::substring::Runner::new() + .rev(|h, n| Some(FinderRev::new(n).rfind(h, n))) + .run(); + } +} diff --git a/vendor/memchr/src/arch/all/shiftor.rs b/vendor/memchr/src/arch/all/shiftor.rs new file mode 100644 index 0000000..b690564 --- /dev/null +++ b/vendor/memchr/src/arch/all/shiftor.rs @@ -0,0 +1,89 @@ +/*! +An implementation of the [Shift-Or substring search algorithm][shiftor]. + +[shiftor]: https://en.wikipedia.org/wiki/Bitap_algorithm +*/ + +use alloc::boxed::Box; + +/// The type of our mask. +/// +/// While we don't expose anyway to configure this in the public API, if one +/// really needs less memory usage or support for longer needles, then it is +/// suggested to copy the code from this module and modify it to fit your +/// needs. The code below is written to be correct regardless of whether Mask +/// is a u8, u16, u32, u64 or u128. +type Mask = u16; + +/// A forward substring searcher using the Shift-Or algorithm. +#[derive(Debug)] +pub struct Finder { + masks: Box<[Mask; 256]>, + needle_len: usize, +} + +impl Finder { + const MAX_NEEDLE_LEN: usize = (Mask::BITS - 1) as usize; + + /// Create a new Shift-Or forward searcher for the given `needle`. + /// + /// The needle may be empty. The empty needle matches at every byte offset. + #[inline] + pub fn new(needle: &[u8]) -> Option<Finder> { + let needle_len = needle.len(); + if needle_len > Finder::MAX_NEEDLE_LEN { + // A match is found when bit 7 is set in 'result' in the search + // routine below. So our needle can't be bigger than 7. We could + // permit bigger needles by using u16, u32 or u64 for our mask + // entries. But this is all we need for this example. + return None; + } + let mut searcher = Finder { masks: Box::from([!0; 256]), needle_len }; + for (i, &byte) in needle.iter().enumerate() { + searcher.masks[usize::from(byte)] &= !(1 << i); + } + Some(searcher) + } + + /// Return the first occurrence of the needle given to `Finder::new` in + /// the `haystack` given. If no such occurrence exists, then `None` is + /// returned. + /// + /// Unlike most other substring search implementations in this crate, this + /// finder does not require passing the needle at search time. A match can + /// be determined without the needle at all since the required information + /// is already encoded into this finder at construction time. + /// + /// The maximum value this can return is `haystack.len()`, which can only + /// occur when the needle and haystack both have length zero. Otherwise, + /// for non-empty haystacks, the maximum value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + if self.needle_len == 0 { + return Some(0); + } + let mut result = !1; + for (i, &byte) in haystack.iter().enumerate() { + result |= self.masks[usize::from(byte)]; + result <<= 1; + if result & (1 << self.needle_len) == 0 { + return Some(i + 1 - self.needle_len); + } + } + None + } +} + +#[cfg(test)] +mod tests { + use super::*; + + define_substring_forward_quickcheck!(|h, n| Some(Finder::new(n)?.find(h))); + + #[test] + fn forward() { + crate::tests::substring::Runner::new() + .fwd(|h, n| Some(Finder::new(n)?.find(h))) + .run(); + } +} diff --git a/vendor/memchr/src/arch/all/twoway.rs b/vendor/memchr/src/arch/all/twoway.rs new file mode 100644 index 0000000..0df3b4a --- /dev/null +++ b/vendor/memchr/src/arch/all/twoway.rs @@ -0,0 +1,877 @@ +/*! +An implementation of the [Two-Way substring search algorithm][two-way]. + +[`Finder`] can be built for forward searches, while [`FinderRev`] can be built +for reverse searches. + +Two-Way makes for a nice general purpose substring search algorithm because of +its time and space complexity properties. It also performs well in practice. +Namely, with `m = len(needle)` and `n = len(haystack)`, Two-Way takes `O(m)` +time to create a finder, `O(1)` space and `O(n)` search time. In other words, +the preprocessing step is quick, doesn't require any heap memory and the worst +case search time is guaranteed to be linear in the haystack regardless of the +size of the needle. + +While vector algorithms will usually beat Two-Way handedly, vector algorithms +also usually have pathological or edge cases that are better handled by Two-Way. +Moreover, not all targets support vector algorithms or implementations for them +simply may not exist yet. + +Two-Way can be found in the `memmem` implementations in at least [GNU libc] and +[musl]. + +[two-way]: https://en.wikipedia.org/wiki/Two-way_string-matching_algorithm +[GNU libc]: https://www.gnu.org/software/libc/ +[musl]: https://www.musl-libc.org/ +*/ + +use core::cmp; + +use crate::{ + arch::all::{is_prefix, is_suffix}, + memmem::Pre, +}; + +/// A forward substring searcher that uses the Two-Way algorithm. +#[derive(Clone, Copy, Debug)] +pub struct Finder(TwoWay); + +/// A reverse substring searcher that uses the Two-Way algorithm. +#[derive(Clone, Copy, Debug)] +pub struct FinderRev(TwoWay); + +/// An implementation of the TwoWay substring search algorithm. +/// +/// This searcher supports forward and reverse search, although not +/// simultaneously. It runs in `O(n + m)` time and `O(1)` space, where +/// `n ~ len(needle)` and `m ~ len(haystack)`. +/// +/// The implementation here roughly matches that which was developed by +/// Crochemore and Perrin in their 1991 paper "Two-way string-matching." The +/// changes in this implementation are 1) the use of zero-based indices, 2) a +/// heuristic skip table based on the last byte (borrowed from Rust's standard +/// library) and 3) the addition of heuristics for a fast skip loop. For (3), +/// callers can pass any kind of prefilter they want, but usually it's one +/// based on a heuristic that uses an approximate background frequency of bytes +/// to choose rare bytes to quickly look for candidate match positions. Note +/// though that currently, this prefilter functionality is not exposed directly +/// in the public API. (File an issue if you want it and provide a use case +/// please.) +/// +/// The heuristic for fast skipping is automatically shut off if it's +/// detected to be ineffective at search time. Generally, this only occurs in +/// pathological cases. But this is generally necessary in order to preserve +/// a `O(n + m)` time bound. +/// +/// The code below is fairly complex and not obviously correct at all. It's +/// likely necessary to read the Two-Way paper cited above in order to fully +/// grok this code. The essence of it is: +/// +/// 1. Do something to detect a "critical" position in the needle. +/// 2. For the current position in the haystack, look if `needle[critical..]` +/// matches at that position. +/// 3. If so, look if `needle[..critical]` matches. +/// 4. If a mismatch occurs, shift the search by some amount based on the +/// critical position and a pre-computed shift. +/// +/// This type is wrapped in the forward and reverse finders that expose +/// consistent forward or reverse APIs. +#[derive(Clone, Copy, Debug)] +struct TwoWay { + /// A small bitset used as a quick prefilter (in addition to any prefilter + /// given by the caller). Namely, a bit `i` is set if and only if `b%64==i` + /// for any `b == needle[i]`. + /// + /// When used as a prefilter, if the last byte at the current candidate + /// position is NOT in this set, then we can skip that entire candidate + /// position (the length of the needle). This is essentially the shift + /// trick found in Boyer-Moore, but only applied to bytes that don't appear + /// in the needle. + /// + /// N.B. This trick was inspired by something similar in std's + /// implementation of Two-Way. + byteset: ApproximateByteSet, + /// A critical position in needle. Specifically, this position corresponds + /// to beginning of either the minimal or maximal suffix in needle. (N.B. + /// See SuffixType below for why "minimal" isn't quite the correct word + /// here.) + /// + /// This is the position at which every search begins. Namely, search + /// starts by scanning text to the right of this position, and only if + /// there's a match does the text to the left of this position get scanned. + critical_pos: usize, + /// The amount we shift by in the Two-Way search algorithm. This + /// corresponds to the "small period" and "large period" cases. + shift: Shift, +} + +impl Finder { + /// Create a searcher that finds occurrences of the given `needle`. + /// + /// An empty `needle` results in a match at every position in a haystack, + /// including at `haystack.len()`. + #[inline] + pub fn new(needle: &[u8]) -> Finder { + let byteset = ApproximateByteSet::new(needle); + let min_suffix = Suffix::forward(needle, SuffixKind::Minimal); + let max_suffix = Suffix::forward(needle, SuffixKind::Maximal); + let (period_lower_bound, critical_pos) = + if min_suffix.pos > max_suffix.pos { + (min_suffix.period, min_suffix.pos) + } else { + (max_suffix.period, max_suffix.pos) + }; + let shift = Shift::forward(needle, period_lower_bound, critical_pos); + Finder(TwoWay { byteset, critical_pos, shift }) + } + + /// Returns the first occurrence of `needle` in the given `haystack`, or + /// `None` if no such occurrence could be found. + /// + /// The `needle` given must be the same as the `needle` provided to + /// [`Finder::new`]. + /// + /// An empty `needle` results in a match at every position in a haystack, + /// including at `haystack.len()`. + #[inline] + pub fn find(&self, haystack: &[u8], needle: &[u8]) -> Option<usize> { + self.find_with_prefilter(None, haystack, needle) + } + + /// This is like [`Finder::find`], but it accepts a prefilter for + /// accelerating searches. + /// + /// Currently this is not exposed in the public API because, at the time + /// of writing, I didn't want to spend time thinking about how to expose + /// the prefilter infrastructure (if at all). If you have a compelling use + /// case for exposing this routine, please create an issue. Do *not* open + /// a PR that just exposes `Pre` and friends. Exporting this routine will + /// require API design. + #[inline(always)] + pub(crate) fn find_with_prefilter( + &self, + pre: Option<Pre<'_>>, + haystack: &[u8], + needle: &[u8], + ) -> Option<usize> { + match self.0.shift { + Shift::Small { period } => { + self.find_small_imp(pre, haystack, needle, period) + } + Shift::Large { shift } => { + self.find_large_imp(pre, haystack, needle, shift) + } + } + } + + // Each of the two search implementations below can be accelerated by a + // prefilter, but it is not always enabled. To avoid its overhead when + // its disabled, we explicitly inline each search implementation based on + // whether a prefilter will be used or not. The decision on which to use + // is made in the parent meta searcher. + + #[inline(always)] + fn find_small_imp( + &self, + mut pre: Option<Pre<'_>>, + haystack: &[u8], + needle: &[u8], + period: usize, + ) -> Option<usize> { + let mut pos = 0; + let mut shift = 0; + let last_byte_pos = match needle.len().checked_sub(1) { + None => return Some(pos), + Some(last_byte) => last_byte, + }; + while pos + needle.len() <= haystack.len() { + let mut i = cmp::max(self.0.critical_pos, shift); + if let Some(pre) = pre.as_mut() { + if pre.is_effective() { + pos += pre.find(&haystack[pos..])?; + shift = 0; + i = self.0.critical_pos; + if pos + needle.len() > haystack.len() { + return None; + } + } + } + if !self.0.byteset.contains(haystack[pos + last_byte_pos]) { + pos += needle.len(); + shift = 0; + continue; + } + while i < needle.len() && needle[i] == haystack[pos + i] { + i += 1; + } + if i < needle.len() { + pos += i - self.0.critical_pos + 1; + shift = 0; + } else { + let mut j = self.0.critical_pos; + while j > shift && needle[j] == haystack[pos + j] { + j -= 1; + } + if j <= shift && needle[shift] == haystack[pos + shift] { + return Some(pos); + } + pos += period; + shift = needle.len() - period; + } + } + None + } + + #[inline(always)] + fn find_large_imp( + &self, + mut pre: Option<Pre<'_>>, + haystack: &[u8], + needle: &[u8], + shift: usize, + ) -> Option<usize> { + let mut pos = 0; + let last_byte_pos = match needle.len().checked_sub(1) { + None => return Some(pos), + Some(last_byte) => last_byte, + }; + 'outer: while pos + needle.len() <= haystack.len() { + if let Some(pre) = pre.as_mut() { + if pre.is_effective() { + pos += pre.find(&haystack[pos..])?; + if pos + needle.len() > haystack.len() { + return None; + } + } + } + + if !self.0.byteset.contains(haystack[pos + last_byte_pos]) { + pos += needle.len(); + continue; + } + let mut i = self.0.critical_pos; + while i < needle.len() && needle[i] == haystack[pos + i] { + i += 1; + } + if i < needle.len() { + pos += i - self.0.critical_pos + 1; + } else { + for j in (0..self.0.critical_pos).rev() { + if needle[j] != haystack[pos + j] { + pos += shift; + continue 'outer; + } + } + return Some(pos); + } + } + None + } +} + +impl FinderRev { + /// Create a searcher that finds occurrences of the given `needle`. + /// + /// An empty `needle` results in a match at every position in a haystack, + /// including at `haystack.len()`. + #[inline] + pub fn new(needle: &[u8]) -> FinderRev { + let byteset = ApproximateByteSet::new(needle); + let min_suffix = Suffix::reverse(needle, SuffixKind::Minimal); + let max_suffix = Suffix::reverse(needle, SuffixKind::Maximal); + let (period_lower_bound, critical_pos) = + if min_suffix.pos < max_suffix.pos { + (min_suffix.period, min_suffix.pos) + } else { + (max_suffix.period, max_suffix.pos) + }; + let shift = Shift::reverse(needle, period_lower_bound, critical_pos); + FinderRev(TwoWay { byteset, critical_pos, shift }) + } + + /// Returns the last occurrence of `needle` in the given `haystack`, or + /// `None` if no such occurrence could be found. + /// + /// The `needle` given must be the same as the `needle` provided to + /// [`FinderRev::new`]. + /// + /// An empty `needle` results in a match at every position in a haystack, + /// including at `haystack.len()`. + #[inline] + pub fn rfind(&self, haystack: &[u8], needle: &[u8]) -> Option<usize> { + // For the reverse case, we don't use a prefilter. It's plausible that + // perhaps we should, but it's a lot of additional code to do it, and + // it's not clear that it's actually worth it. If you have a really + // compelling use case for this, please file an issue. + match self.0.shift { + Shift::Small { period } => { + self.rfind_small_imp(haystack, needle, period) + } + Shift::Large { shift } => { + self.rfind_large_imp(haystack, needle, shift) + } + } + } + + #[inline(always)] + fn rfind_small_imp( + &self, + haystack: &[u8], + needle: &[u8], + period: usize, + ) -> Option<usize> { + let nlen = needle.len(); + let mut pos = haystack.len(); + let mut shift = nlen; + let first_byte = match needle.get(0) { + None => return Some(pos), + Some(&first_byte) => first_byte, + }; + while pos >= nlen { + if !self.0.byteset.contains(haystack[pos - nlen]) { + pos -= nlen; + shift = nlen; + continue; + } + let mut i = cmp::min(self.0.critical_pos, shift); + while i > 0 && needle[i - 1] == haystack[pos - nlen + i - 1] { + i -= 1; + } + if i > 0 || first_byte != haystack[pos - nlen] { + pos -= self.0.critical_pos - i + 1; + shift = nlen; + } else { + let mut j = self.0.critical_pos; + while j < shift && needle[j] == haystack[pos - nlen + j] { + j += 1; + } + if j >= shift { + return Some(pos - nlen); + } + pos -= period; + shift = period; + } + } + None + } + + #[inline(always)] + fn rfind_large_imp( + &self, + haystack: &[u8], + needle: &[u8], + shift: usize, + ) -> Option<usize> { + let nlen = needle.len(); + let mut pos = haystack.len(); + let first_byte = match needle.get(0) { + None => return Some(pos), + Some(&first_byte) => first_byte, + }; + while pos >= nlen { + if !self.0.byteset.contains(haystack[pos - nlen]) { + pos -= nlen; + continue; + } + let mut i = self.0.critical_pos; + while i > 0 && needle[i - 1] == haystack[pos - nlen + i - 1] { + i -= 1; + } + if i > 0 || first_byte != haystack[pos - nlen] { + pos -= self.0.critical_pos - i + 1; + } else { + let mut j = self.0.critical_pos; + while j < nlen && needle[j] == haystack[pos - nlen + j] { + j += 1; + } + if j == nlen { + return Some(pos - nlen); + } + pos -= shift; + } + } + None + } +} + +/// A representation of the amount we're allowed to shift by during Two-Way +/// search. +/// +/// When computing a critical factorization of the needle, we find the position +/// of the critical factorization by finding the needle's maximal (or minimal) +/// suffix, along with the period of that suffix. It turns out that the period +/// of that suffix is a lower bound on the period of the needle itself. +/// +/// This lower bound is equivalent to the actual period of the needle in +/// some cases. To describe that case, we denote the needle as `x` where +/// `x = uv` and `v` is the lexicographic maximal suffix of `v`. The lower +/// bound given here is always the period of `v`, which is `<= period(x)`. The +/// case where `period(v) == period(x)` occurs when `len(u) < (len(x) / 2)` and +/// where `u` is a suffix of `v[0..period(v)]`. +/// +/// This case is important because the search algorithm for when the +/// periods are equivalent is slightly different than the search algorithm +/// for when the periods are not equivalent. In particular, when they aren't +/// equivalent, we know that the period of the needle is no less than half its +/// length. In this case, we shift by an amount less than or equal to the +/// period of the needle (determined by the maximum length of the components +/// of the critical factorization of `x`, i.e., `max(len(u), len(v))`).. +/// +/// The above two cases are represented by the variants below. Each entails +/// a different instantiation of the Two-Way search algorithm. +/// +/// N.B. If we could find a way to compute the exact period in all cases, +/// then we could collapse this case analysis and simplify the algorithm. The +/// Two-Way paper suggests this is possible, but more reading is required to +/// grok why the authors didn't pursue that path. +#[derive(Clone, Copy, Debug)] +enum Shift { + Small { period: usize }, + Large { shift: usize }, +} + +impl Shift { + /// Compute the shift for a given needle in the forward direction. + /// + /// This requires a lower bound on the period and a critical position. + /// These can be computed by extracting both the minimal and maximal + /// lexicographic suffixes, and choosing the right-most starting position. + /// The lower bound on the period is then the period of the chosen suffix. + fn forward( + needle: &[u8], + period_lower_bound: usize, + critical_pos: usize, + ) -> Shift { + let large = cmp::max(critical_pos, needle.len() - critical_pos); + if critical_pos * 2 >= needle.len() { + return Shift::Large { shift: large }; + } + + let (u, v) = needle.split_at(critical_pos); + if !is_suffix(&v[..period_lower_bound], u) { + return Shift::Large { shift: large }; + } + Shift::Small { period: period_lower_bound } + } + + /// Compute the shift for a given needle in the reverse direction. + /// + /// This requires a lower bound on the period and a critical position. + /// These can be computed by extracting both the minimal and maximal + /// lexicographic suffixes, and choosing the left-most starting position. + /// The lower bound on the period is then the period of the chosen suffix. + fn reverse( + needle: &[u8], + period_lower_bound: usize, + critical_pos: usize, + ) -> Shift { + let large = cmp::max(critical_pos, needle.len() - critical_pos); + if (needle.len() - critical_pos) * 2 >= needle.len() { + return Shift::Large { shift: large }; + } + + let (v, u) = needle.split_at(critical_pos); + if !is_prefix(&v[v.len() - period_lower_bound..], u) { + return Shift::Large { shift: large }; + } + Shift::Small { period: period_lower_bound } + } +} + +/// A suffix extracted from a needle along with its period. +#[derive(Debug)] +struct Suffix { + /// The starting position of this suffix. + /// + /// If this is a forward suffix, then `&bytes[pos..]` can be used. If this + /// is a reverse suffix, then `&bytes[..pos]` can be used. That is, for + /// forward suffixes, this is an inclusive starting position, where as for + /// reverse suffixes, this is an exclusive ending position. + pos: usize, + /// The period of this suffix. + /// + /// Note that this is NOT necessarily the period of the string from which + /// this suffix comes from. (It is always less than or equal to the period + /// of the original string.) + period: usize, +} + +impl Suffix { + fn forward(needle: &[u8], kind: SuffixKind) -> Suffix { + // suffix represents our maximal (or minimal) suffix, along with + // its period. + let mut suffix = Suffix { pos: 0, period: 1 }; + // The start of a suffix in `needle` that we are considering as a + // more maximal (or minimal) suffix than what's in `suffix`. + let mut candidate_start = 1; + // The current offset of our suffixes that we're comparing. + // + // When the characters at this offset are the same, then we mush on + // to the next position since no decision is possible. When the + // candidate's character is greater (or lesser) than the corresponding + // character than our current maximal (or minimal) suffix, then the + // current suffix is changed over to the candidate and we restart our + // search. Otherwise, the candidate suffix is no good and we restart + // our search on the next candidate. + // + // The three cases above correspond to the three cases in the loop + // below. + let mut offset = 0; + + while candidate_start + offset < needle.len() { + let current = needle[suffix.pos + offset]; + let candidate = needle[candidate_start + offset]; + match kind.cmp(current, candidate) { + SuffixOrdering::Accept => { + suffix = Suffix { pos: candidate_start, period: 1 }; + candidate_start += 1; + offset = 0; + } + SuffixOrdering::Skip => { + candidate_start += offset + 1; + offset = 0; + suffix.period = candidate_start - suffix.pos; + } + SuffixOrdering::Push => { + if offset + 1 == suffix.period { + candidate_start += suffix.period; + offset = 0; + } else { + offset += 1; + } + } + } + } + suffix + } + + fn reverse(needle: &[u8], kind: SuffixKind) -> Suffix { + // See the comments in `forward` for how this works. + let mut suffix = Suffix { pos: needle.len(), period: 1 }; + if needle.len() == 1 { + return suffix; + } + let mut candidate_start = match needle.len().checked_sub(1) { + None => return suffix, + Some(candidate_start) => candidate_start, + }; + let mut offset = 0; + + while offset < candidate_start { + let current = needle[suffix.pos - offset - 1]; + let candidate = needle[candidate_start - offset - 1]; + match kind.cmp(current, candidate) { + SuffixOrdering::Accept => { + suffix = Suffix { pos: candidate_start, period: 1 }; + candidate_start -= 1; + offset = 0; + } + SuffixOrdering::Skip => { + candidate_start -= offset + 1; + offset = 0; + suffix.period = suffix.pos - candidate_start; + } + SuffixOrdering::Push => { + if offset + 1 == suffix.period { + candidate_start -= suffix.period; + offset = 0; + } else { + offset += 1; + } + } + } + } + suffix + } +} + +/// The kind of suffix to extract. +#[derive(Clone, Copy, Debug)] +enum SuffixKind { + /// Extract the smallest lexicographic suffix from a string. + /// + /// Technically, this doesn't actually pick the smallest lexicographic + /// suffix. e.g., Given the choice between `a` and `aa`, this will choose + /// the latter over the former, even though `a < aa`. The reasoning for + /// this isn't clear from the paper, but it still smells like a minimal + /// suffix. + Minimal, + /// Extract the largest lexicographic suffix from a string. + /// + /// Unlike `Minimal`, this really does pick the maximum suffix. e.g., Given + /// the choice between `z` and `zz`, this will choose the latter over the + /// former. + Maximal, +} + +/// The result of comparing corresponding bytes between two suffixes. +#[derive(Clone, Copy, Debug)] +enum SuffixOrdering { + /// This occurs when the given candidate byte indicates that the candidate + /// suffix is better than the current maximal (or minimal) suffix. That is, + /// the current candidate suffix should supplant the current maximal (or + /// minimal) suffix. + Accept, + /// This occurs when the given candidate byte excludes the candidate suffix + /// from being better than the current maximal (or minimal) suffix. That + /// is, the current candidate suffix should be dropped and the next one + /// should be considered. + Skip, + /// This occurs when no decision to accept or skip the candidate suffix + /// can be made, e.g., when corresponding bytes are equivalent. In this + /// case, the next corresponding bytes should be compared. + Push, +} + +impl SuffixKind { + /// Returns true if and only if the given candidate byte indicates that + /// it should replace the current suffix as the maximal (or minimal) + /// suffix. + fn cmp(self, current: u8, candidate: u8) -> SuffixOrdering { + use self::SuffixOrdering::*; + + match self { + SuffixKind::Minimal if candidate < current => Accept, + SuffixKind::Minimal if candidate > current => Skip, + SuffixKind::Minimal => Push, + SuffixKind::Maximal if candidate > current => Accept, + SuffixKind::Maximal if candidate < current => Skip, + SuffixKind::Maximal => Push, + } + } +} + +/// A bitset used to track whether a particular byte exists in a needle or not. +/// +/// Namely, bit 'i' is set if and only if byte%64==i for any byte in the +/// needle. If a particular byte in the haystack is NOT in this set, then one +/// can conclude that it is also not in the needle, and thus, one can advance +/// in the haystack by needle.len() bytes. +#[derive(Clone, Copy, Debug)] +struct ApproximateByteSet(u64); + +impl ApproximateByteSet { + /// Create a new set from the given needle. + fn new(needle: &[u8]) -> ApproximateByteSet { + let mut bits = 0; + for &b in needle { + bits |= 1 << (b % 64); + } + ApproximateByteSet(bits) + } + + /// Return true if and only if the given byte might be in this set. This + /// may return a false positive, but will never return a false negative. + #[inline(always)] + fn contains(&self, byte: u8) -> bool { + self.0 & (1 << (byte % 64)) != 0 + } +} + +#[cfg(test)] +mod tests { + use alloc::vec::Vec; + + use super::*; + + /// Convenience wrapper for computing the suffix as a byte string. + fn get_suffix_forward(needle: &[u8], kind: SuffixKind) -> (&[u8], usize) { + let s = Suffix::forward(needle, kind); + (&needle[s.pos..], s.period) + } + + /// Convenience wrapper for computing the reverse suffix as a byte string. + fn get_suffix_reverse(needle: &[u8], kind: SuffixKind) -> (&[u8], usize) { + let s = Suffix::reverse(needle, kind); + (&needle[..s.pos], s.period) + } + + /// Return all of the non-empty suffixes in the given byte string. + fn suffixes(bytes: &[u8]) -> Vec<&[u8]> { + (0..bytes.len()).map(|i| &bytes[i..]).collect() + } + + /// Return the lexicographically maximal suffix of the given byte string. + fn naive_maximal_suffix_forward(needle: &[u8]) -> &[u8] { + let mut sufs = suffixes(needle); + sufs.sort(); + sufs.pop().unwrap() + } + + /// Return the lexicographically maximal suffix of the reverse of the given + /// byte string. + fn naive_maximal_suffix_reverse(needle: &[u8]) -> Vec<u8> { + let mut reversed = needle.to_vec(); + reversed.reverse(); + let mut got = naive_maximal_suffix_forward(&reversed).to_vec(); + got.reverse(); + got + } + + define_substring_forward_quickcheck!(|h, n| Some( + Finder::new(n).find(h, n) + )); + define_substring_reverse_quickcheck!(|h, n| Some( + FinderRev::new(n).rfind(h, n) + )); + + #[test] + fn forward() { + crate::tests::substring::Runner::new() + .fwd(|h, n| Some(Finder::new(n).find(h, n))) + .run(); + } + + #[test] + fn reverse() { + crate::tests::substring::Runner::new() + .rev(|h, n| Some(FinderRev::new(n).rfind(h, n))) + .run(); + } + + #[test] + fn suffix_forward() { + macro_rules! assert_suffix_min { + ($given:expr, $expected:expr, $period:expr) => { + let (got_suffix, got_period) = + get_suffix_forward($given.as_bytes(), SuffixKind::Minimal); + let got_suffix = core::str::from_utf8(got_suffix).unwrap(); + assert_eq!(($expected, $period), (got_suffix, got_period)); + }; + } + + macro_rules! assert_suffix_max { + ($given:expr, $expected:expr, $period:expr) => { + let (got_suffix, got_period) = + get_suffix_forward($given.as_bytes(), SuffixKind::Maximal); + let got_suffix = core::str::from_utf8(got_suffix).unwrap(); + assert_eq!(($expected, $period), (got_suffix, got_period)); + }; + } + + assert_suffix_min!("a", "a", 1); + assert_suffix_max!("a", "a", 1); + + assert_suffix_min!("ab", "ab", 2); + assert_suffix_max!("ab", "b", 1); + + assert_suffix_min!("ba", "a", 1); + assert_suffix_max!("ba", "ba", 2); + + assert_suffix_min!("abc", "abc", 3); + assert_suffix_max!("abc", "c", 1); + + assert_suffix_min!("acb", "acb", 3); + assert_suffix_max!("acb", "cb", 2); + + assert_suffix_min!("cba", "a", 1); + assert_suffix_max!("cba", "cba", 3); + + assert_suffix_min!("abcabc", "abcabc", 3); + assert_suffix_max!("abcabc", "cabc", 3); + + assert_suffix_min!("abcabcabc", "abcabcabc", 3); + assert_suffix_max!("abcabcabc", "cabcabc", 3); + + assert_suffix_min!("abczz", "abczz", 5); + assert_suffix_max!("abczz", "zz", 1); + + assert_suffix_min!("zzabc", "abc", 3); + assert_suffix_max!("zzabc", "zzabc", 5); + + assert_suffix_min!("aaa", "aaa", 1); + assert_suffix_max!("aaa", "aaa", 1); + + assert_suffix_min!("foobar", "ar", 2); + assert_suffix_max!("foobar", "r", 1); + } + + #[test] + fn suffix_reverse() { + macro_rules! assert_suffix_min { + ($given:expr, $expected:expr, $period:expr) => { + let (got_suffix, got_period) = + get_suffix_reverse($given.as_bytes(), SuffixKind::Minimal); + let got_suffix = core::str::from_utf8(got_suffix).unwrap(); + assert_eq!(($expected, $period), (got_suffix, got_period)); + }; + } + + macro_rules! assert_suffix_max { + ($given:expr, $expected:expr, $period:expr) => { + let (got_suffix, got_period) = + get_suffix_reverse($given.as_bytes(), SuffixKind::Maximal); + let got_suffix = core::str::from_utf8(got_suffix).unwrap(); + assert_eq!(($expected, $period), (got_suffix, got_period)); + }; + } + + assert_suffix_min!("a", "a", 1); + assert_suffix_max!("a", "a", 1); + + assert_suffix_min!("ab", "a", 1); + assert_suffix_max!("ab", "ab", 2); + + assert_suffix_min!("ba", "ba", 2); + assert_suffix_max!("ba", "b", 1); + + assert_suffix_min!("abc", "a", 1); + assert_suffix_max!("abc", "abc", 3); + + assert_suffix_min!("acb", "a", 1); + assert_suffix_max!("acb", "ac", 2); + + assert_suffix_min!("cba", "cba", 3); + assert_suffix_max!("cba", "c", 1); + + assert_suffix_min!("abcabc", "abca", 3); + assert_suffix_max!("abcabc", "abcabc", 3); + + assert_suffix_min!("abcabcabc", "abcabca", 3); + assert_suffix_max!("abcabcabc", "abcabcabc", 3); + + assert_suffix_min!("abczz", "a", 1); + assert_suffix_max!("abczz", "abczz", 5); + + assert_suffix_min!("zzabc", "zza", 3); + assert_suffix_max!("zzabc", "zz", 1); + + assert_suffix_min!("aaa", "aaa", 1); + assert_suffix_max!("aaa", "aaa", 1); + } + + #[cfg(not(miri))] + quickcheck::quickcheck! { + fn qc_suffix_forward_maximal(bytes: Vec<u8>) -> bool { + if bytes.is_empty() { + return true; + } + + let (got, _) = get_suffix_forward(&bytes, SuffixKind::Maximal); + let expected = naive_maximal_suffix_forward(&bytes); + got == expected + } + + fn qc_suffix_reverse_maximal(bytes: Vec<u8>) -> bool { + if bytes.is_empty() { + return true; + } + + let (got, _) = get_suffix_reverse(&bytes, SuffixKind::Maximal); + let expected = naive_maximal_suffix_reverse(&bytes); + expected == got + } + } + + // This is a regression test caught by quickcheck that exercised a bug in + // the reverse small period handling. The bug was that we were using 'if j + // == shift' to determine if a match occurred, but the correct guard is 'if + // j >= shift', which matches the corresponding guard in the forward impl. + #[test] + fn regression_rev_small_period() { + let rfind = |h, n| FinderRev::new(n).rfind(h, n); + let haystack = "ababaz"; + let needle = "abab"; + assert_eq!(Some(0), rfind(haystack.as_bytes(), needle.as_bytes())); + } +} diff --git a/vendor/memchr/src/arch/generic/memchr.rs b/vendor/memchr/src/arch/generic/memchr.rs new file mode 100644 index 0000000..580b3cc --- /dev/null +++ b/vendor/memchr/src/arch/generic/memchr.rs @@ -0,0 +1,1214 @@ +/*! +Generic crate-internal routines for the `memchr` family of functions. +*/ + +// What follows is a vector algorithm generic over the specific vector +// type to detect the position of one, two or three needles in a haystack. +// From what I know, this is a "classic" algorithm, although I don't +// believe it has been published in any peer reviewed journal. I believe +// it can be found in places like glibc and Go's standard library. It +// appears to be well known and is elaborated on in more detail here: +// https://gms.tf/stdfind-and-memchr-optimizations.html +// +// While the routine below is fairly long and perhaps intimidating, the basic +// idea is actually very simple and can be expressed straight-forwardly in +// pseudo code. The psuedo code below is written for 128 bit vectors, but the +// actual code below works for anything that implements the Vector trait. +// +// needle = (n1 << 15) | (n1 << 14) | ... | (n1 << 1) | n1 +// // Note: shift amount is in bytes +// +// while i <= haystack.len() - 16: +// // A 16 byte vector. Each byte in chunk corresponds to a byte in +// // the haystack. +// chunk = haystack[i:i+16] +// // Compare bytes in needle with bytes in chunk. The result is a 16 +// // byte chunk where each byte is 0xFF if the corresponding bytes +// // in needle and chunk were equal, or 0x00 otherwise. +// eqs = cmpeq(needle, chunk) +// // Return a 32 bit integer where the most significant 16 bits +// // are always 0 and the lower 16 bits correspond to whether the +// // most significant bit in the correspond byte in `eqs` is set. +// // In other words, `mask as u16` has bit i set if and only if +// // needle[i] == chunk[i]. +// mask = movemask(eqs) +// +// // Mask is 0 if there is no match, and non-zero otherwise. +// if mask != 0: +// // trailing_zeros tells us the position of the least significant +// // bit that is set. +// return i + trailing_zeros(mask) +// +// // haystack length may not be a multiple of 16, so search the rest. +// while i < haystack.len(): +// if haystack[i] == n1: +// return i +// +// // No match found. +// return NULL +// +// In fact, we could loosely translate the above code to Rust line-for-line +// and it would be a pretty fast algorithm. But, we pull out all the stops +// to go as fast as possible: +// +// 1. We use aligned loads. That is, we do some finagling to make sure our +// primary loop not only proceeds in increments of 16 bytes, but that +// the address of haystack's pointer that we dereference is aligned to +// 16 bytes. 16 is a magic number here because it is the size of SSE2 +// 128-bit vector. (For the AVX2 algorithm, 32 is the magic number.) +// Therefore, to get aligned loads, our pointer's address must be evenly +// divisible by 16. +// 2. Our primary loop proceeds 64 bytes at a time instead of 16. It's +// kind of like loop unrolling, but we combine the equality comparisons +// using a vector OR such that we only need to extract a single mask to +// determine whether a match exists or not. If so, then we do some +// book-keeping to determine the precise location but otherwise mush on. +// 3. We use our "chunk" comparison routine in as many places as possible, +// even if it means using unaligned loads. In particular, if haystack +// starts with an unaligned address, then we do an unaligned load to +// search the first 16 bytes. We then start our primary loop at the +// smallest subsequent aligned address, which will actually overlap with +// previously searched bytes. But we're OK with that. We do a similar +// dance at the end of our primary loop. Finally, to avoid a +// byte-at-a-time loop at the end, we do a final 16 byte unaligned load +// that may overlap with a previous load. This is OK because it converts +// a loop into a small number of very fast vector instructions. The overlap +// is OK because we know the place where the overlap occurs does not +// contain a match. +// +// And that's pretty all there is to it. Note that since the below is +// generic and since it's meant to be inlined into routines with a +// `#[target_feature(enable = "...")]` annotation, we must mark all routines as +// both unsafe and `#[inline(always)]`. +// +// The fact that the code below is generic does somewhat inhibit us. For +// example, I've noticed that introducing an unlineable `#[cold]` function to +// handle the match case in the loop generates tighter assembly, but there is +// no way to do this in the generic code below because the generic code doesn't +// know what `target_feature` annotation to apply to the unlineable function. +// We could make such functions part of the `Vector` trait, but we instead live +// with the slightly sub-optimal codegen for now since it doesn't seem to have +// a noticeable perf difference. + +use crate::{ + ext::Pointer, + vector::{MoveMask, Vector}, +}; + +/// Finds all occurrences of a single byte in a haystack. +#[derive(Clone, Copy, Debug)] +pub(crate) struct One<V> { + s1: u8, + v1: V, +} + +impl<V: Vector> One<V> { + /// The number of bytes we examine per each iteration of our search loop. + const LOOP_SIZE: usize = 4 * V::BYTES; + + /// Create a new searcher that finds occurrences of the byte given. + #[inline(always)] + pub(crate) unsafe fn new(needle: u8) -> One<V> { + One { s1: needle, v1: V::splat(needle) } + } + + /// Returns the needle given to `One::new`. + #[inline(always)] + pub(crate) fn needle1(&self) -> u8 { + self.s1 + } + + /// Return a pointer to the first occurrence of the needle in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// # Safety + /// + /// * It must be the case that `start < end` and that the distance between + /// them is at least equal to `V::BYTES`. That is, it must always be valid + /// to do at least an unaligned load of `V` at `start`. + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + #[inline(always)] + pub(crate) unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + // If we want to support vectors bigger than 256 bits, we probably + // need to move up to using a u64 for the masks used below. Currently + // they are 32 bits, which means we're SOL for vectors that need masks + // bigger than 32 bits. Overall unclear until there's a use case. + debug_assert!(V::BYTES <= 32, "vector cannot be bigger than 32 bytes"); + + let topos = V::Mask::first_offset; + let len = end.distance(start); + debug_assert!( + len >= V::BYTES, + "haystack has length {}, but must be at least {}", + len, + V::BYTES + ); + + // Search a possibly unaligned chunk at `start`. This covers any part + // of the haystack prior to where aligned loads can start. + if let Some(cur) = self.search_chunk(start, topos) { + return Some(cur); + } + // Set `cur` to the first V-aligned pointer greater than `start`. + let mut cur = start.add(V::BYTES - (start.as_usize() & V::ALIGN)); + debug_assert!(cur > start && end.sub(V::BYTES) >= start); + if len >= Self::LOOP_SIZE { + while cur <= end.sub(Self::LOOP_SIZE) { + debug_assert_eq!(0, cur.as_usize() % V::BYTES); + + let a = V::load_aligned(cur); + let b = V::load_aligned(cur.add(1 * V::BYTES)); + let c = V::load_aligned(cur.add(2 * V::BYTES)); + let d = V::load_aligned(cur.add(3 * V::BYTES)); + let eqa = self.v1.cmpeq(a); + let eqb = self.v1.cmpeq(b); + let eqc = self.v1.cmpeq(c); + let eqd = self.v1.cmpeq(d); + let or1 = eqa.or(eqb); + let or2 = eqc.or(eqd); + let or3 = or1.or(or2); + if or3.movemask_will_have_non_zero() { + let mask = eqa.movemask(); + if mask.has_non_zero() { + return Some(cur.add(topos(mask))); + } + + let mask = eqb.movemask(); + if mask.has_non_zero() { + return Some(cur.add(1 * V::BYTES).add(topos(mask))); + } + + let mask = eqc.movemask(); + if mask.has_non_zero() { + return Some(cur.add(2 * V::BYTES).add(topos(mask))); + } + + let mask = eqd.movemask(); + debug_assert!(mask.has_non_zero()); + return Some(cur.add(3 * V::BYTES).add(topos(mask))); + } + cur = cur.add(Self::LOOP_SIZE); + } + } + // Handle any leftovers after the aligned loop above. We use unaligned + // loads here, but I believe we are guaranteed that they are aligned + // since `cur` is aligned. + while cur <= end.sub(V::BYTES) { + debug_assert!(end.distance(cur) >= V::BYTES); + if let Some(cur) = self.search_chunk(cur, topos) { + return Some(cur); + } + cur = cur.add(V::BYTES); + } + // Finally handle any remaining bytes less than the size of V. In this + // case, our pointer may indeed be unaligned and the load may overlap + // with the previous one. But that's okay since we know the previous + // load didn't lead to a match (otherwise we wouldn't be here). + if cur < end { + debug_assert!(end.distance(cur) < V::BYTES); + cur = cur.sub(V::BYTES - end.distance(cur)); + debug_assert_eq!(end.distance(cur), V::BYTES); + return self.search_chunk(cur, topos); + } + None + } + + /// Return a pointer to the last occurrence of the needle in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// # Safety + /// + /// * It must be the case that `start < end` and that the distance between + /// them is at least equal to `V::BYTES`. That is, it must always be valid + /// to do at least an unaligned load of `V` at `start`. + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + #[inline(always)] + pub(crate) unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + // If we want to support vectors bigger than 256 bits, we probably + // need to move up to using a u64 for the masks used below. Currently + // they are 32 bits, which means we're SOL for vectors that need masks + // bigger than 32 bits. Overall unclear until there's a use case. + debug_assert!(V::BYTES <= 32, "vector cannot be bigger than 32 bytes"); + + let topos = V::Mask::last_offset; + let len = end.distance(start); + debug_assert!( + len >= V::BYTES, + "haystack has length {}, but must be at least {}", + len, + V::BYTES + ); + + if let Some(cur) = self.search_chunk(end.sub(V::BYTES), topos) { + return Some(cur); + } + let mut cur = end.sub(end.as_usize() & V::ALIGN); + debug_assert!(start <= cur && cur <= end); + if len >= Self::LOOP_SIZE { + while cur >= start.add(Self::LOOP_SIZE) { + debug_assert_eq!(0, cur.as_usize() % V::BYTES); + + cur = cur.sub(Self::LOOP_SIZE); + let a = V::load_aligned(cur); + let b = V::load_aligned(cur.add(1 * V::BYTES)); + let c = V::load_aligned(cur.add(2 * V::BYTES)); + let d = V::load_aligned(cur.add(3 * V::BYTES)); + let eqa = self.v1.cmpeq(a); + let eqb = self.v1.cmpeq(b); + let eqc = self.v1.cmpeq(c); + let eqd = self.v1.cmpeq(d); + let or1 = eqa.or(eqb); + let or2 = eqc.or(eqd); + let or3 = or1.or(or2); + if or3.movemask_will_have_non_zero() { + let mask = eqd.movemask(); + if mask.has_non_zero() { + return Some(cur.add(3 * V::BYTES).add(topos(mask))); + } + + let mask = eqc.movemask(); + if mask.has_non_zero() { + return Some(cur.add(2 * V::BYTES).add(topos(mask))); + } + + let mask = eqb.movemask(); + if mask.has_non_zero() { + return Some(cur.add(1 * V::BYTES).add(topos(mask))); + } + + let mask = eqa.movemask(); + debug_assert!(mask.has_non_zero()); + return Some(cur.add(topos(mask))); + } + } + } + while cur >= start.add(V::BYTES) { + debug_assert!(cur.distance(start) >= V::BYTES); + cur = cur.sub(V::BYTES); + if let Some(cur) = self.search_chunk(cur, topos) { + return Some(cur); + } + } + if cur > start { + debug_assert!(cur.distance(start) < V::BYTES); + return self.search_chunk(start, topos); + } + None + } + + /// Return a count of all matching bytes in the given haystack. + /// + /// # Safety + /// + /// * It must be the case that `start < end` and that the distance between + /// them is at least equal to `V::BYTES`. That is, it must always be valid + /// to do at least an unaligned load of `V` at `start`. + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + #[inline(always)] + pub(crate) unsafe fn count_raw( + &self, + start: *const u8, + end: *const u8, + ) -> usize { + debug_assert!(V::BYTES <= 32, "vector cannot be bigger than 32 bytes"); + + let confirm = |b| b == self.needle1(); + let len = end.distance(start); + debug_assert!( + len >= V::BYTES, + "haystack has length {}, but must be at least {}", + len, + V::BYTES + ); + + // Set `cur` to the first V-aligned pointer greater than `start`. + let mut cur = start.add(V::BYTES - (start.as_usize() & V::ALIGN)); + // Count any matching bytes before we start our aligned loop. + let mut count = count_byte_by_byte(start, cur, confirm); + debug_assert!(cur > start && end.sub(V::BYTES) >= start); + if len >= Self::LOOP_SIZE { + while cur <= end.sub(Self::LOOP_SIZE) { + debug_assert_eq!(0, cur.as_usize() % V::BYTES); + + let a = V::load_aligned(cur); + let b = V::load_aligned(cur.add(1 * V::BYTES)); + let c = V::load_aligned(cur.add(2 * V::BYTES)); + let d = V::load_aligned(cur.add(3 * V::BYTES)); + let eqa = self.v1.cmpeq(a); + let eqb = self.v1.cmpeq(b); + let eqc = self.v1.cmpeq(c); + let eqd = self.v1.cmpeq(d); + count += eqa.movemask().count_ones(); + count += eqb.movemask().count_ones(); + count += eqc.movemask().count_ones(); + count += eqd.movemask().count_ones(); + cur = cur.add(Self::LOOP_SIZE); + } + } + // Handle any leftovers after the aligned loop above. We use unaligned + // loads here, but I believe we are guaranteed that they are aligned + // since `cur` is aligned. + while cur <= end.sub(V::BYTES) { + debug_assert!(end.distance(cur) >= V::BYTES); + let chunk = V::load_unaligned(cur); + count += self.v1.cmpeq(chunk).movemask().count_ones(); + cur = cur.add(V::BYTES); + } + // And finally count any leftovers that weren't caught above. + count += count_byte_by_byte(cur, end, confirm); + count + } + + /// Search `V::BYTES` starting at `cur` via an unaligned load. + /// + /// `mask_to_offset` should be a function that converts a `movemask` to + /// an offset such that `cur.add(offset)` corresponds to a pointer to the + /// match location if one is found. Generally it is expected to use either + /// `mask_to_first_offset` or `mask_to_last_offset`, depending on whether + /// one is implementing a forward or reverse search, respectively. + /// + /// # Safety + /// + /// `cur` must be a valid pointer and it must be valid to do an unaligned + /// load of size `V::BYTES` at `cur`. + #[inline(always)] + unsafe fn search_chunk( + &self, + cur: *const u8, + mask_to_offset: impl Fn(V::Mask) -> usize, + ) -> Option<*const u8> { + let chunk = V::load_unaligned(cur); + let mask = self.v1.cmpeq(chunk).movemask(); + if mask.has_non_zero() { + Some(cur.add(mask_to_offset(mask))) + } else { + None + } + } +} + +/// Finds all occurrences of two bytes in a haystack. +/// +/// That is, this reports matches of one of two possible bytes. For example, +/// searching for `a` or `b` in `afoobar` would report matches at offsets `0`, +/// `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub(crate) struct Two<V> { + s1: u8, + s2: u8, + v1: V, + v2: V, +} + +impl<V: Vector> Two<V> { + /// The number of bytes we examine per each iteration of our search loop. + const LOOP_SIZE: usize = 2 * V::BYTES; + + /// Create a new searcher that finds occurrences of the byte given. + #[inline(always)] + pub(crate) unsafe fn new(needle1: u8, needle2: u8) -> Two<V> { + Two { + s1: needle1, + s2: needle2, + v1: V::splat(needle1), + v2: V::splat(needle2), + } + } + + /// Returns the first needle given to `Two::new`. + #[inline(always)] + pub(crate) fn needle1(&self) -> u8 { + self.s1 + } + + /// Returns the second needle given to `Two::new`. + #[inline(always)] + pub(crate) fn needle2(&self) -> u8 { + self.s2 + } + + /// Return a pointer to the first occurrence of one of the needles in the + /// given haystack. If no such occurrence exists, then `None` is returned. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// # Safety + /// + /// * It must be the case that `start < end` and that the distance between + /// them is at least equal to `V::BYTES`. That is, it must always be valid + /// to do at least an unaligned load of `V` at `start`. + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + #[inline(always)] + pub(crate) unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + // If we want to support vectors bigger than 256 bits, we probably + // need to move up to using a u64 for the masks used below. Currently + // they are 32 bits, which means we're SOL for vectors that need masks + // bigger than 32 bits. Overall unclear until there's a use case. + debug_assert!(V::BYTES <= 32, "vector cannot be bigger than 32 bytes"); + + let topos = V::Mask::first_offset; + let len = end.distance(start); + debug_assert!( + len >= V::BYTES, + "haystack has length {}, but must be at least {}", + len, + V::BYTES + ); + + // Search a possibly unaligned chunk at `start`. This covers any part + // of the haystack prior to where aligned loads can start. + if let Some(cur) = self.search_chunk(start, topos) { + return Some(cur); + } + // Set `cur` to the first V-aligned pointer greater than `start`. + let mut cur = start.add(V::BYTES - (start.as_usize() & V::ALIGN)); + debug_assert!(cur > start && end.sub(V::BYTES) >= start); + if len >= Self::LOOP_SIZE { + while cur <= end.sub(Self::LOOP_SIZE) { + debug_assert_eq!(0, cur.as_usize() % V::BYTES); + + let a = V::load_aligned(cur); + let b = V::load_aligned(cur.add(V::BYTES)); + let eqa1 = self.v1.cmpeq(a); + let eqb1 = self.v1.cmpeq(b); + let eqa2 = self.v2.cmpeq(a); + let eqb2 = self.v2.cmpeq(b); + let or1 = eqa1.or(eqb1); + let or2 = eqa2.or(eqb2); + let or3 = or1.or(or2); + if or3.movemask_will_have_non_zero() { + let mask = eqa1.movemask().or(eqa2.movemask()); + if mask.has_non_zero() { + return Some(cur.add(topos(mask))); + } + + let mask = eqb1.movemask().or(eqb2.movemask()); + debug_assert!(mask.has_non_zero()); + return Some(cur.add(V::BYTES).add(topos(mask))); + } + cur = cur.add(Self::LOOP_SIZE); + } + } + // Handle any leftovers after the aligned loop above. We use unaligned + // loads here, but I believe we are guaranteed that they are aligned + // since `cur` is aligned. + while cur <= end.sub(V::BYTES) { + debug_assert!(end.distance(cur) >= V::BYTES); + if let Some(cur) = self.search_chunk(cur, topos) { + return Some(cur); + } + cur = cur.add(V::BYTES); + } + // Finally handle any remaining bytes less than the size of V. In this + // case, our pointer may indeed be unaligned and the load may overlap + // with the previous one. But that's okay since we know the previous + // load didn't lead to a match (otherwise we wouldn't be here). + if cur < end { + debug_assert!(end.distance(cur) < V::BYTES); + cur = cur.sub(V::BYTES - end.distance(cur)); + debug_assert_eq!(end.distance(cur), V::BYTES); + return self.search_chunk(cur, topos); + } + None + } + + /// Return a pointer to the last occurrence of the needle in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// # Safety + /// + /// * It must be the case that `start < end` and that the distance between + /// them is at least equal to `V::BYTES`. That is, it must always be valid + /// to do at least an unaligned load of `V` at `start`. + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + #[inline(always)] + pub(crate) unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + // If we want to support vectors bigger than 256 bits, we probably + // need to move up to using a u64 for the masks used below. Currently + // they are 32 bits, which means we're SOL for vectors that need masks + // bigger than 32 bits. Overall unclear until there's a use case. + debug_assert!(V::BYTES <= 32, "vector cannot be bigger than 32 bytes"); + + let topos = V::Mask::last_offset; + let len = end.distance(start); + debug_assert!( + len >= V::BYTES, + "haystack has length {}, but must be at least {}", + len, + V::BYTES + ); + + if let Some(cur) = self.search_chunk(end.sub(V::BYTES), topos) { + return Some(cur); + } + let mut cur = end.sub(end.as_usize() & V::ALIGN); + debug_assert!(start <= cur && cur <= end); + if len >= Self::LOOP_SIZE { + while cur >= start.add(Self::LOOP_SIZE) { + debug_assert_eq!(0, cur.as_usize() % V::BYTES); + + cur = cur.sub(Self::LOOP_SIZE); + let a = V::load_aligned(cur); + let b = V::load_aligned(cur.add(V::BYTES)); + let eqa1 = self.v1.cmpeq(a); + let eqb1 = self.v1.cmpeq(b); + let eqa2 = self.v2.cmpeq(a); + let eqb2 = self.v2.cmpeq(b); + let or1 = eqa1.or(eqb1); + let or2 = eqa2.or(eqb2); + let or3 = or1.or(or2); + if or3.movemask_will_have_non_zero() { + let mask = eqb1.movemask().or(eqb2.movemask()); + if mask.has_non_zero() { + return Some(cur.add(V::BYTES).add(topos(mask))); + } + + let mask = eqa1.movemask().or(eqa2.movemask()); + debug_assert!(mask.has_non_zero()); + return Some(cur.add(topos(mask))); + } + } + } + while cur >= start.add(V::BYTES) { + debug_assert!(cur.distance(start) >= V::BYTES); + cur = cur.sub(V::BYTES); + if let Some(cur) = self.search_chunk(cur, topos) { + return Some(cur); + } + } + if cur > start { + debug_assert!(cur.distance(start) < V::BYTES); + return self.search_chunk(start, topos); + } + None + } + + /// Search `V::BYTES` starting at `cur` via an unaligned load. + /// + /// `mask_to_offset` should be a function that converts a `movemask` to + /// an offset such that `cur.add(offset)` corresponds to a pointer to the + /// match location if one is found. Generally it is expected to use either + /// `mask_to_first_offset` or `mask_to_last_offset`, depending on whether + /// one is implementing a forward or reverse search, respectively. + /// + /// # Safety + /// + /// `cur` must be a valid pointer and it must be valid to do an unaligned + /// load of size `V::BYTES` at `cur`. + #[inline(always)] + unsafe fn search_chunk( + &self, + cur: *const u8, + mask_to_offset: impl Fn(V::Mask) -> usize, + ) -> Option<*const u8> { + let chunk = V::load_unaligned(cur); + let eq1 = self.v1.cmpeq(chunk); + let eq2 = self.v2.cmpeq(chunk); + let mask = eq1.or(eq2).movemask(); + if mask.has_non_zero() { + let mask1 = eq1.movemask(); + let mask2 = eq2.movemask(); + Some(cur.add(mask_to_offset(mask1.or(mask2)))) + } else { + None + } + } +} + +/// Finds all occurrences of two bytes in a haystack. +/// +/// That is, this reports matches of one of two possible bytes. For example, +/// searching for `a` or `b` in `afoobar` would report matches at offsets `0`, +/// `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub(crate) struct Three<V> { + s1: u8, + s2: u8, + s3: u8, + v1: V, + v2: V, + v3: V, +} + +impl<V: Vector> Three<V> { + /// The number of bytes we examine per each iteration of our search loop. + const LOOP_SIZE: usize = 2 * V::BYTES; + + /// Create a new searcher that finds occurrences of the byte given. + #[inline(always)] + pub(crate) unsafe fn new( + needle1: u8, + needle2: u8, + needle3: u8, + ) -> Three<V> { + Three { + s1: needle1, + s2: needle2, + s3: needle3, + v1: V::splat(needle1), + v2: V::splat(needle2), + v3: V::splat(needle3), + } + } + + /// Returns the first needle given to `Three::new`. + #[inline(always)] + pub(crate) fn needle1(&self) -> u8 { + self.s1 + } + + /// Returns the second needle given to `Three::new`. + #[inline(always)] + pub(crate) fn needle2(&self) -> u8 { + self.s2 + } + + /// Returns the third needle given to `Three::new`. + #[inline(always)] + pub(crate) fn needle3(&self) -> u8 { + self.s3 + } + + /// Return a pointer to the first occurrence of one of the needles in the + /// given haystack. If no such occurrence exists, then `None` is returned. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// # Safety + /// + /// * It must be the case that `start < end` and that the distance between + /// them is at least equal to `V::BYTES`. That is, it must always be valid + /// to do at least an unaligned load of `V` at `start`. + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + #[inline(always)] + pub(crate) unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + // If we want to support vectors bigger than 256 bits, we probably + // need to move up to using a u64 for the masks used below. Currently + // they are 32 bits, which means we're SOL for vectors that need masks + // bigger than 32 bits. Overall unclear until there's a use case. + debug_assert!(V::BYTES <= 32, "vector cannot be bigger than 32 bytes"); + + let topos = V::Mask::first_offset; + let len = end.distance(start); + debug_assert!( + len >= V::BYTES, + "haystack has length {}, but must be at least {}", + len, + V::BYTES + ); + + // Search a possibly unaligned chunk at `start`. This covers any part + // of the haystack prior to where aligned loads can start. + if let Some(cur) = self.search_chunk(start, topos) { + return Some(cur); + } + // Set `cur` to the first V-aligned pointer greater than `start`. + let mut cur = start.add(V::BYTES - (start.as_usize() & V::ALIGN)); + debug_assert!(cur > start && end.sub(V::BYTES) >= start); + if len >= Self::LOOP_SIZE { + while cur <= end.sub(Self::LOOP_SIZE) { + debug_assert_eq!(0, cur.as_usize() % V::BYTES); + + let a = V::load_aligned(cur); + let b = V::load_aligned(cur.add(V::BYTES)); + let eqa1 = self.v1.cmpeq(a); + let eqb1 = self.v1.cmpeq(b); + let eqa2 = self.v2.cmpeq(a); + let eqb2 = self.v2.cmpeq(b); + let eqa3 = self.v3.cmpeq(a); + let eqb3 = self.v3.cmpeq(b); + let or1 = eqa1.or(eqb1); + let or2 = eqa2.or(eqb2); + let or3 = eqa3.or(eqb3); + let or4 = or1.or(or2); + let or5 = or3.or(or4); + if or5.movemask_will_have_non_zero() { + let mask = eqa1 + .movemask() + .or(eqa2.movemask()) + .or(eqa3.movemask()); + if mask.has_non_zero() { + return Some(cur.add(topos(mask))); + } + + let mask = eqb1 + .movemask() + .or(eqb2.movemask()) + .or(eqb3.movemask()); + debug_assert!(mask.has_non_zero()); + return Some(cur.add(V::BYTES).add(topos(mask))); + } + cur = cur.add(Self::LOOP_SIZE); + } + } + // Handle any leftovers after the aligned loop above. We use unaligned + // loads here, but I believe we are guaranteed that they are aligned + // since `cur` is aligned. + while cur <= end.sub(V::BYTES) { + debug_assert!(end.distance(cur) >= V::BYTES); + if let Some(cur) = self.search_chunk(cur, topos) { + return Some(cur); + } + cur = cur.add(V::BYTES); + } + // Finally handle any remaining bytes less than the size of V. In this + // case, our pointer may indeed be unaligned and the load may overlap + // with the previous one. But that's okay since we know the previous + // load didn't lead to a match (otherwise we wouldn't be here). + if cur < end { + debug_assert!(end.distance(cur) < V::BYTES); + cur = cur.sub(V::BYTES - end.distance(cur)); + debug_assert_eq!(end.distance(cur), V::BYTES); + return self.search_chunk(cur, topos); + } + None + } + + /// Return a pointer to the last occurrence of the needle in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// # Safety + /// + /// * It must be the case that `start < end` and that the distance between + /// them is at least equal to `V::BYTES`. That is, it must always be valid + /// to do at least an unaligned load of `V` at `start`. + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + #[inline(always)] + pub(crate) unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + // If we want to support vectors bigger than 256 bits, we probably + // need to move up to using a u64 for the masks used below. Currently + // they are 32 bits, which means we're SOL for vectors that need masks + // bigger than 32 bits. Overall unclear until there's a use case. + debug_assert!(V::BYTES <= 32, "vector cannot be bigger than 32 bytes"); + + let topos = V::Mask::last_offset; + let len = end.distance(start); + debug_assert!( + len >= V::BYTES, + "haystack has length {}, but must be at least {}", + len, + V::BYTES + ); + + if let Some(cur) = self.search_chunk(end.sub(V::BYTES), topos) { + return Some(cur); + } + let mut cur = end.sub(end.as_usize() & V::ALIGN); + debug_assert!(start <= cur && cur <= end); + if len >= Self::LOOP_SIZE { + while cur >= start.add(Self::LOOP_SIZE) { + debug_assert_eq!(0, cur.as_usize() % V::BYTES); + + cur = cur.sub(Self::LOOP_SIZE); + let a = V::load_aligned(cur); + let b = V::load_aligned(cur.add(V::BYTES)); + let eqa1 = self.v1.cmpeq(a); + let eqb1 = self.v1.cmpeq(b); + let eqa2 = self.v2.cmpeq(a); + let eqb2 = self.v2.cmpeq(b); + let eqa3 = self.v3.cmpeq(a); + let eqb3 = self.v3.cmpeq(b); + let or1 = eqa1.or(eqb1); + let or2 = eqa2.or(eqb2); + let or3 = eqa3.or(eqb3); + let or4 = or1.or(or2); + let or5 = or3.or(or4); + if or5.movemask_will_have_non_zero() { + let mask = eqb1 + .movemask() + .or(eqb2.movemask()) + .or(eqb3.movemask()); + if mask.has_non_zero() { + return Some(cur.add(V::BYTES).add(topos(mask))); + } + + let mask = eqa1 + .movemask() + .or(eqa2.movemask()) + .or(eqa3.movemask()); + debug_assert!(mask.has_non_zero()); + return Some(cur.add(topos(mask))); + } + } + } + while cur >= start.add(V::BYTES) { + debug_assert!(cur.distance(start) >= V::BYTES); + cur = cur.sub(V::BYTES); + if let Some(cur) = self.search_chunk(cur, topos) { + return Some(cur); + } + } + if cur > start { + debug_assert!(cur.distance(start) < V::BYTES); + return self.search_chunk(start, topos); + } + None + } + + /// Search `V::BYTES` starting at `cur` via an unaligned load. + /// + /// `mask_to_offset` should be a function that converts a `movemask` to + /// an offset such that `cur.add(offset)` corresponds to a pointer to the + /// match location if one is found. Generally it is expected to use either + /// `mask_to_first_offset` or `mask_to_last_offset`, depending on whether + /// one is implementing a forward or reverse search, respectively. + /// + /// # Safety + /// + /// `cur` must be a valid pointer and it must be valid to do an unaligned + /// load of size `V::BYTES` at `cur`. + #[inline(always)] + unsafe fn search_chunk( + &self, + cur: *const u8, + mask_to_offset: impl Fn(V::Mask) -> usize, + ) -> Option<*const u8> { + let chunk = V::load_unaligned(cur); + let eq1 = self.v1.cmpeq(chunk); + let eq2 = self.v2.cmpeq(chunk); + let eq3 = self.v3.cmpeq(chunk); + let mask = eq1.or(eq2).or(eq3).movemask(); + if mask.has_non_zero() { + let mask1 = eq1.movemask(); + let mask2 = eq2.movemask(); + let mask3 = eq3.movemask(); + Some(cur.add(mask_to_offset(mask1.or(mask2).or(mask3)))) + } else { + None + } + } +} + +/// An iterator over all occurrences of a set of bytes in a haystack. +/// +/// This iterator implements the routines necessary to provide a +/// `DoubleEndedIterator` impl, which means it can also be used to find +/// occurrences in reverse order. +/// +/// The lifetime parameters are as follows: +/// +/// * `'h` refers to the lifetime of the haystack being searched. +/// +/// This type is intended to be used to implement all iterators for the +/// `memchr` family of functions. It handles a tiny bit of marginally tricky +/// raw pointer math, but otherwise expects the caller to provide `find_raw` +/// and `rfind_raw` routines for each call of `next` and `next_back`, +/// respectively. +#[derive(Clone, Debug)] +pub(crate) struct Iter<'h> { + /// The original starting point into the haystack. We use this to convert + /// pointers to offsets. + original_start: *const u8, + /// The current starting point into the haystack. That is, where the next + /// search will begin. + start: *const u8, + /// The current ending point into the haystack. That is, where the next + /// reverse search will begin. + end: *const u8, + /// A marker for tracking the lifetime of the start/cur_start/cur_end + /// pointers above, which all point into the haystack. + haystack: core::marker::PhantomData<&'h [u8]>, +} + +// SAFETY: Iter contains no shared references to anything that performs any +// interior mutations. Also, the lifetime guarantees that Iter will not outlive +// the haystack. +unsafe impl<'h> Send for Iter<'h> {} + +// SAFETY: Iter perform no interior mutations, therefore no explicit +// synchronization is necessary. Also, the lifetime guarantees that Iter will +// not outlive the haystack. +unsafe impl<'h> Sync for Iter<'h> {} + +impl<'h> Iter<'h> { + /// Create a new generic memchr iterator. + #[inline(always)] + pub(crate) fn new(haystack: &'h [u8]) -> Iter<'h> { + Iter { + original_start: haystack.as_ptr(), + start: haystack.as_ptr(), + end: haystack.as_ptr().wrapping_add(haystack.len()), + haystack: core::marker::PhantomData, + } + } + + /// Returns the next occurrence in the forward direction. + /// + /// # Safety + /// + /// Callers must ensure that if a pointer is returned from the closure + /// provided, then it must be greater than or equal to the start pointer + /// and less than the end pointer. + #[inline(always)] + pub(crate) unsafe fn next( + &mut self, + mut find_raw: impl FnMut(*const u8, *const u8) -> Option<*const u8>, + ) -> Option<usize> { + // SAFETY: Pointers are derived directly from the same &[u8] haystack. + // We only ever modify start/end corresponding to a matching offset + // found between start and end. Thus all changes to start/end maintain + // our safety requirements. + // + // The only other assumption we rely on is that the pointer returned + // by `find_raw` satisfies `self.start <= found < self.end`, and that + // safety contract is forwarded to the caller. + let found = find_raw(self.start, self.end)?; + let result = found.distance(self.original_start); + self.start = found.add(1); + Some(result) + } + + /// Returns the number of remaining elements in this iterator. + #[inline(always)] + pub(crate) fn count( + self, + mut count_raw: impl FnMut(*const u8, *const u8) -> usize, + ) -> usize { + // SAFETY: Pointers are derived directly from the same &[u8] haystack. + // We only ever modify start/end corresponding to a matching offset + // found between start and end. Thus all changes to start/end maintain + // our safety requirements. + count_raw(self.start, self.end) + } + + /// Returns the next occurrence in reverse. + /// + /// # Safety + /// + /// Callers must ensure that if a pointer is returned from the closure + /// provided, then it must be greater than or equal to the start pointer + /// and less than the end pointer. + #[inline(always)] + pub(crate) unsafe fn next_back( + &mut self, + mut rfind_raw: impl FnMut(*const u8, *const u8) -> Option<*const u8>, + ) -> Option<usize> { + // SAFETY: Pointers are derived directly from the same &[u8] haystack. + // We only ever modify start/end corresponding to a matching offset + // found between start and end. Thus all changes to start/end maintain + // our safety requirements. + // + // The only other assumption we rely on is that the pointer returned + // by `rfind_raw` satisfies `self.start <= found < self.end`, and that + // safety contract is forwarded to the caller. + let found = rfind_raw(self.start, self.end)?; + let result = found.distance(self.original_start); + self.end = found; + Some(result) + } + + /// Provides an implementation of `Iterator::size_hint`. + #[inline(always)] + pub(crate) fn size_hint(&self) -> (usize, Option<usize>) { + (0, Some(self.end.as_usize().saturating_sub(self.start.as_usize()))) + } +} + +/// Search a slice using a function that operates on raw pointers. +/// +/// Given a function to search a contiguous sequence of memory for the location +/// of a non-empty set of bytes, this will execute that search on a slice of +/// bytes. The pointer returned by the given function will be converted to an +/// offset relative to the starting point of the given slice. That is, if a +/// match is found, the offset returned by this routine is guaranteed to be a +/// valid index into `haystack`. +/// +/// Callers may use this for a forward or reverse search. +/// +/// # Safety +/// +/// Callers must ensure that if a pointer is returned by `find_raw`, then the +/// pointer must be greater than or equal to the starting pointer and less than +/// the end pointer. +#[inline(always)] +pub(crate) unsafe fn search_slice_with_raw( + haystack: &[u8], + mut find_raw: impl FnMut(*const u8, *const u8) -> Option<*const u8>, +) -> Option<usize> { + // SAFETY: We rely on `find_raw` to return a correct and valid pointer, but + // otherwise, `start` and `end` are valid due to the guarantees provided by + // a &[u8]. + let start = haystack.as_ptr(); + let end = start.add(haystack.len()); + let found = find_raw(start, end)?; + Some(found.distance(start)) +} + +/// Performs a forward byte-at-a-time loop until either `ptr >= end_ptr` or +/// until `confirm(*ptr)` returns `true`. If the former occurs, then `None` is +/// returned. If the latter occurs, then the pointer at which `confirm` returns +/// `true` is returned. +/// +/// # Safety +/// +/// Callers must provide valid pointers and they must satisfy `start_ptr <= +/// ptr` and `ptr <= end_ptr`. +#[inline(always)] +pub(crate) unsafe fn fwd_byte_by_byte<F: Fn(u8) -> bool>( + start: *const u8, + end: *const u8, + confirm: F, +) -> Option<*const u8> { + debug_assert!(start <= end); + let mut ptr = start; + while ptr < end { + if confirm(*ptr) { + return Some(ptr); + } + ptr = ptr.offset(1); + } + None +} + +/// Performs a reverse byte-at-a-time loop until either `ptr < start_ptr` or +/// until `confirm(*ptr)` returns `true`. If the former occurs, then `None` is +/// returned. If the latter occurs, then the pointer at which `confirm` returns +/// `true` is returned. +/// +/// # Safety +/// +/// Callers must provide valid pointers and they must satisfy `start_ptr <= +/// ptr` and `ptr <= end_ptr`. +#[inline(always)] +pub(crate) unsafe fn rev_byte_by_byte<F: Fn(u8) -> bool>( + start: *const u8, + end: *const u8, + confirm: F, +) -> Option<*const u8> { + debug_assert!(start <= end); + + let mut ptr = end; + while ptr > start { + ptr = ptr.offset(-1); + if confirm(*ptr) { + return Some(ptr); + } + } + None +} + +/// Performs a forward byte-at-a-time loop until `ptr >= end_ptr` and returns +/// the number of times `confirm(*ptr)` returns `true`. +/// +/// # Safety +/// +/// Callers must provide valid pointers and they must satisfy `start_ptr <= +/// ptr` and `ptr <= end_ptr`. +#[inline(always)] +pub(crate) unsafe fn count_byte_by_byte<F: Fn(u8) -> bool>( + start: *const u8, + end: *const u8, + confirm: F, +) -> usize { + debug_assert!(start <= end); + let mut ptr = start; + let mut count = 0; + while ptr < end { + if confirm(*ptr) { + count += 1; + } + ptr = ptr.offset(1); + } + count +} diff --git a/vendor/memchr/src/arch/generic/mod.rs b/vendor/memchr/src/arch/generic/mod.rs new file mode 100644 index 0000000..63ee3f0 --- /dev/null +++ b/vendor/memchr/src/arch/generic/mod.rs @@ -0,0 +1,14 @@ +/*! +This module defines "generic" routines that can be specialized to specific +architectures. + +We don't expose this module primarily because it would require exposing all +of the internal infrastructure required to write these generic routines. +That infrastructure should be treated as an implementation detail so that +it is allowed to evolve. Instead, what we expose are architecture specific +instantiations of these generic implementations. The generic code just lets us +write the code once (usually). +*/ + +pub(crate) mod memchr; +pub(crate) mod packedpair; diff --git a/vendor/memchr/src/arch/generic/packedpair.rs b/vendor/memchr/src/arch/generic/packedpair.rs new file mode 100644 index 0000000..8d97cf2 --- /dev/null +++ b/vendor/memchr/src/arch/generic/packedpair.rs @@ -0,0 +1,317 @@ +/*! +Generic crate-internal routines for the "packed pair" SIMD algorithm. + +The "packed pair" algorithm is based on the [generic SIMD] algorithm. The main +difference is that it (by default) uses a background distribution of byte +frequencies to heuristically select the pair of bytes to search for. + +[generic SIMD]: http://0x80.pl/articles/simd-strfind.html#first-and-last +*/ + +use crate::{ + arch::all::{is_equal_raw, packedpair::Pair}, + ext::Pointer, + vector::{MoveMask, Vector}, +}; + +/// A generic architecture dependent "packed pair" finder. +/// +/// This finder picks two bytes that it believes have high predictive power +/// for indicating an overall match of a needle. Depending on whether +/// `Finder::find` or `Finder::find_prefilter` is used, it reports offsets +/// where the needle matches or could match. In the prefilter case, candidates +/// are reported whenever the [`Pair`] of bytes given matches. +/// +/// This is architecture dependent because it uses specific vector operations +/// to look for occurrences of the pair of bytes. +/// +/// This type is not meant to be exported and is instead meant to be used as +/// the implementation for architecture specific facades. Why? Because it's a +/// bit of a quirky API that requires `inline(always)` annotations. And pretty +/// much everything has safety obligations due (at least) to the caller needing +/// to inline calls into routines marked with +/// `#[target_feature(enable = "...")]`. +#[derive(Clone, Copy, Debug)] +pub(crate) struct Finder<V> { + pair: Pair, + v1: V, + v2: V, + min_haystack_len: usize, +} + +impl<V: Vector> Finder<V> { + /// Create a new pair searcher. The searcher returned can either report + /// exact matches of `needle` or act as a prefilter and report candidate + /// positions of `needle`. + /// + /// # Safety + /// + /// Callers must ensure that whatever vector type this routine is called + /// with is supported by the current environment. + /// + /// Callers must also ensure that `needle.len() >= 2`. + #[inline(always)] + pub(crate) unsafe fn new(needle: &[u8], pair: Pair) -> Finder<V> { + let max_index = pair.index1().max(pair.index2()); + let min_haystack_len = + core::cmp::max(needle.len(), usize::from(max_index) + V::BYTES); + let v1 = V::splat(needle[usize::from(pair.index1())]); + let v2 = V::splat(needle[usize::from(pair.index2())]); + Finder { pair, v1, v2, min_haystack_len } + } + + /// Searches the given haystack for the given needle. The needle given + /// should be the same as the needle that this finder was initialized + /// with. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// Since this is meant to be used with vector functions, callers need to + /// specialize this inside of a function with a `target_feature` attribute. + /// Therefore, callers must ensure that whatever target feature is being + /// used supports the vector functions that this function is specialized + /// for. (For the specific vector functions used, see the Vector trait + /// implementations.) + #[inline(always)] + pub(crate) unsafe fn find( + &self, + haystack: &[u8], + needle: &[u8], + ) -> Option<usize> { + assert!( + haystack.len() >= self.min_haystack_len, + "haystack too small, should be at least {} but got {}", + self.min_haystack_len, + haystack.len(), + ); + + let all = V::Mask::all_zeros_except_least_significant(0); + let start = haystack.as_ptr(); + let end = start.add(haystack.len()); + let max = end.sub(self.min_haystack_len); + let mut cur = start; + + // N.B. I did experiment with unrolling the loop to deal with size(V) + // bytes at a time and 2*size(V) bytes at a time. The double unroll + // was marginally faster while the quadruple unroll was unambiguously + // slower. In the end, I decided the complexity from unrolling wasn't + // worth it. I used the memmem/krate/prebuilt/huge-en/ benchmarks to + // compare. + while cur <= max { + if let Some(chunki) = self.find_in_chunk(needle, cur, end, all) { + return Some(matched(start, cur, chunki)); + } + cur = cur.add(V::BYTES); + } + if cur < end { + let remaining = end.distance(cur); + debug_assert!( + remaining < self.min_haystack_len, + "remaining bytes should be smaller than the minimum haystack \ + length of {}, but there are {} bytes remaining", + self.min_haystack_len, + remaining, + ); + if remaining < needle.len() { + return None; + } + debug_assert!( + max < cur, + "after main loop, cur should have exceeded max", + ); + let overlap = cur.distance(max); + debug_assert!( + overlap > 0, + "overlap ({}) must always be non-zero", + overlap, + ); + debug_assert!( + overlap < V::BYTES, + "overlap ({}) cannot possibly be >= than a vector ({})", + overlap, + V::BYTES, + ); + // The mask has all of its bits set except for the first N least + // significant bits, where N=overlap. This way, any matches that + // occur in find_in_chunk within the overlap are automatically + // ignored. + let mask = V::Mask::all_zeros_except_least_significant(overlap); + cur = max; + let m = self.find_in_chunk(needle, cur, end, mask); + if let Some(chunki) = m { + return Some(matched(start, cur, chunki)); + } + } + None + } + + /// Searches the given haystack for offsets that represent candidate + /// matches of the `needle` given to this finder's constructor. The offsets + /// returned, if they are a match, correspond to the starting offset of + /// `needle` in the given `haystack`. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// Since this is meant to be used with vector functions, callers need to + /// specialize this inside of a function with a `target_feature` attribute. + /// Therefore, callers must ensure that whatever target feature is being + /// used supports the vector functions that this function is specialized + /// for. (For the specific vector functions used, see the Vector trait + /// implementations.) + #[inline(always)] + pub(crate) unsafe fn find_prefilter( + &self, + haystack: &[u8], + ) -> Option<usize> { + assert!( + haystack.len() >= self.min_haystack_len, + "haystack too small, should be at least {} but got {}", + self.min_haystack_len, + haystack.len(), + ); + + let start = haystack.as_ptr(); + let end = start.add(haystack.len()); + let max = end.sub(self.min_haystack_len); + let mut cur = start; + + // N.B. I did experiment with unrolling the loop to deal with size(V) + // bytes at a time and 2*size(V) bytes at a time. The double unroll + // was marginally faster while the quadruple unroll was unambiguously + // slower. In the end, I decided the complexity from unrolling wasn't + // worth it. I used the memmem/krate/prebuilt/huge-en/ benchmarks to + // compare. + while cur <= max { + if let Some(chunki) = self.find_prefilter_in_chunk(cur) { + return Some(matched(start, cur, chunki)); + } + cur = cur.add(V::BYTES); + } + if cur < end { + // This routine immediately quits if a candidate match is found. + // That means that if we're here, no candidate matches have been + // found at or before 'ptr'. Thus, we don't need to mask anything + // out even though we might technically search part of the haystack + // that we've already searched (because we know it can't match). + cur = max; + if let Some(chunki) = self.find_prefilter_in_chunk(cur) { + return Some(matched(start, cur, chunki)); + } + } + None + } + + /// Search for an occurrence of our byte pair from the needle in the chunk + /// pointed to by cur, with the end of the haystack pointed to by end. + /// When an occurrence is found, memcmp is run to check if a match occurs + /// at the corresponding position. + /// + /// `mask` should have bits set corresponding the positions in the chunk + /// in which matches are considered. This is only used for the last vector + /// load where the beginning of the vector might have overlapped with the + /// last load in the main loop. The mask lets us avoid visiting positions + /// that have already been discarded as matches. + /// + /// # Safety + /// + /// It must be safe to do an unaligned read of size(V) bytes starting at + /// both (cur + self.index1) and (cur + self.index2). It must also be safe + /// to do unaligned loads on cur up to (end - needle.len()). + #[inline(always)] + unsafe fn find_in_chunk( + &self, + needle: &[u8], + cur: *const u8, + end: *const u8, + mask: V::Mask, + ) -> Option<usize> { + let index1 = usize::from(self.pair.index1()); + let index2 = usize::from(self.pair.index2()); + let chunk1 = V::load_unaligned(cur.add(index1)); + let chunk2 = V::load_unaligned(cur.add(index2)); + let eq1 = chunk1.cmpeq(self.v1); + let eq2 = chunk2.cmpeq(self.v2); + + let mut offsets = eq1.and(eq2).movemask().and(mask); + while offsets.has_non_zero() { + let offset = offsets.first_offset(); + let cur = cur.add(offset); + if end.sub(needle.len()) < cur { + return None; + } + if is_equal_raw(needle.as_ptr(), cur, needle.len()) { + return Some(offset); + } + offsets = offsets.clear_least_significant_bit(); + } + None + } + + /// Search for an occurrence of our byte pair from the needle in the chunk + /// pointed to by cur, with the end of the haystack pointed to by end. + /// When an occurrence is found, memcmp is run to check if a match occurs + /// at the corresponding position. + /// + /// # Safety + /// + /// It must be safe to do an unaligned read of size(V) bytes starting at + /// both (cur + self.index1) and (cur + self.index2). It must also be safe + /// to do unaligned reads on cur up to (end - needle.len()). + #[inline(always)] + unsafe fn find_prefilter_in_chunk(&self, cur: *const u8) -> Option<usize> { + let index1 = usize::from(self.pair.index1()); + let index2 = usize::from(self.pair.index2()); + let chunk1 = V::load_unaligned(cur.add(index1)); + let chunk2 = V::load_unaligned(cur.add(index2)); + let eq1 = chunk1.cmpeq(self.v1); + let eq2 = chunk2.cmpeq(self.v2); + + let offsets = eq1.and(eq2).movemask(); + if !offsets.has_non_zero() { + return None; + } + Some(offsets.first_offset()) + } + + /// Returns the pair of offsets (into the needle) used to check as a + /// predicate before confirming whether a needle exists at a particular + /// position. + #[inline] + pub(crate) fn pair(&self) -> &Pair { + &self.pair + } + + /// Returns the minimum haystack length that this `Finder` can search. + /// + /// Providing a haystack to this `Finder` shorter than this length is + /// guaranteed to result in a panic. + #[inline(always)] + pub(crate) fn min_haystack_len(&self) -> usize { + self.min_haystack_len + } +} + +/// Accepts a chunk-relative offset and returns a haystack relative offset. +/// +/// This used to be marked `#[cold]` and `#[inline(never)]`, but I couldn't +/// observe a consistent measureable difference between that and just inlining +/// it. So we go with inlining it. +/// +/// # Safety +/// +/// Same at `ptr::offset_from` in addition to `cur >= start`. +#[inline(always)] +unsafe fn matched(start: *const u8, cur: *const u8, chunki: usize) -> usize { + cur.distance(start) + chunki +} + +// If you're looking for tests, those are run for each instantiation of the +// above code. So for example, see arch::x86_64::sse2::packedpair. diff --git a/vendor/memchr/src/arch/mod.rs b/vendor/memchr/src/arch/mod.rs new file mode 100644 index 0000000..2f63a1a --- /dev/null +++ b/vendor/memchr/src/arch/mod.rs @@ -0,0 +1,16 @@ +/*! +A module with low-level architecture dependent routines. + +These routines are useful as primitives for tasks not covered by the higher +level crate API. +*/ + +pub mod all; +pub(crate) mod generic; + +#[cfg(target_arch = "aarch64")] +pub mod aarch64; +#[cfg(target_arch = "wasm32")] +pub mod wasm32; +#[cfg(target_arch = "x86_64")] +pub mod x86_64; diff --git a/vendor/memchr/src/arch/wasm32/memchr.rs b/vendor/memchr/src/arch/wasm32/memchr.rs new file mode 100644 index 0000000..b0bbd1c --- /dev/null +++ b/vendor/memchr/src/arch/wasm32/memchr.rs @@ -0,0 +1,137 @@ +/*! +Wrapper routines for `memchr` and friends. + +These routines choose the best implementation at compile time. (This is +different from `x86_64` because it is expected that `simd128` is almost always +available for `wasm32` targets.) +*/ + +macro_rules! defraw { + ($ty:ident, $find:ident, $start:ident, $end:ident, $($needles:ident),+) => {{ + #[cfg(target_feature = "simd128")] + { + use crate::arch::wasm32::simd128::memchr::$ty; + + debug!("chose simd128 for {}", stringify!($ty)); + debug_assert!($ty::is_available()); + // SAFETY: We know that wasm memchr is always available whenever + // code is compiled for `wasm32` with the `simd128` target feature + // enabled. + $ty::new_unchecked($($needles),+).$find($start, $end) + } + #[cfg(not(target_feature = "simd128"))] + { + use crate::arch::all::memchr::$ty; + + debug!( + "no simd128 feature available, using fallback for {}", + stringify!($ty), + ); + $ty::new($($needles),+).$find($start, $end) + } + }} +} + +/// memchr, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::find_raw`. +#[inline(always)] +pub(crate) unsafe fn memchr_raw( + n1: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(One, find_raw, start, end, n1) +} + +/// memrchr, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::rfind_raw`. +#[inline(always)] +pub(crate) unsafe fn memrchr_raw( + n1: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(One, rfind_raw, start, end, n1) +} + +/// memchr2, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Two::find_raw`. +#[inline(always)] +pub(crate) unsafe fn memchr2_raw( + n1: u8, + n2: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(Two, find_raw, start, end, n1, n2) +} + +/// memrchr2, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Two::rfind_raw`. +#[inline(always)] +pub(crate) unsafe fn memrchr2_raw( + n1: u8, + n2: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(Two, rfind_raw, start, end, n1, n2) +} + +/// memchr3, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Three::find_raw`. +#[inline(always)] +pub(crate) unsafe fn memchr3_raw( + n1: u8, + n2: u8, + n3: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(Three, find_raw, start, end, n1, n2, n3) +} + +/// memrchr3, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Three::rfind_raw`. +#[inline(always)] +pub(crate) unsafe fn memrchr3_raw( + n1: u8, + n2: u8, + n3: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + defraw!(Three, rfind_raw, start, end, n1, n2, n3) +} + +/// Count all matching bytes, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::count_raw`. +#[inline(always)] +pub(crate) unsafe fn count_raw( + n1: u8, + start: *const u8, + end: *const u8, +) -> usize { + defraw!(One, count_raw, start, end, n1) +} diff --git a/vendor/memchr/src/arch/wasm32/mod.rs b/vendor/memchr/src/arch/wasm32/mod.rs new file mode 100644 index 0000000..209f876 --- /dev/null +++ b/vendor/memchr/src/arch/wasm32/mod.rs @@ -0,0 +1,7 @@ +/*! +Vector algorithms for the `wasm32` target. +*/ + +pub mod simd128; + +pub(crate) mod memchr; diff --git a/vendor/memchr/src/arch/wasm32/simd128/memchr.rs b/vendor/memchr/src/arch/wasm32/simd128/memchr.rs new file mode 100644 index 0000000..fa314c9 --- /dev/null +++ b/vendor/memchr/src/arch/wasm32/simd128/memchr.rs @@ -0,0 +1,1020 @@ +/*! +This module defines 128-bit vector implementations of `memchr` and friends. + +The main types in this module are [`One`], [`Two`] and [`Three`]. They are for +searching for one, two or three distinct bytes, respectively, in a haystack. +Each type also has corresponding double ended iterators. These searchers are +typically much faster than scalar routines accomplishing the same task. + +The `One` searcher also provides a [`One::count`] routine for efficiently +counting the number of times a single byte occurs in a haystack. This is +useful, for example, for counting the number of lines in a haystack. This +routine exists because it is usually faster, especially with a high match +count, then using [`One::find`] repeatedly. ([`OneIter`] specializes its +`Iterator::count` implementation to use this routine.) + +Only one, two and three bytes are supported because three bytes is about +the point where one sees diminishing returns. Beyond this point and it's +probably (but not necessarily) better to just use a simple `[bool; 256]` array +or similar. However, it depends mightily on the specific work-load and the +expected match frequency. +*/ + +use core::arch::wasm32::v128; + +use crate::{arch::generic::memchr as generic, ext::Pointer, vector::Vector}; + +/// Finds all occurrences of a single byte in a haystack. +#[derive(Clone, Copy, Debug)] +pub struct One(generic::One<v128>); + +impl One { + /// Create a new searcher that finds occurrences of the needle byte given. + /// + /// This particular searcher is specialized to use simd128 vector + /// instructions that typically make it quite fast. + /// + /// If simd128 is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn new(needle: u8) -> Option<One> { + if One::is_available() { + // SAFETY: we check that simd128 is available above. + unsafe { Some(One::new_unchecked(needle)) } + } else { + None + } + } + + /// Create a new finder specific to simd128 vectors and routines without + /// checking that simd128 is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute `simd128` + /// instructions in the current environment. + #[target_feature(enable = "simd128")] + #[inline] + pub unsafe fn new_unchecked(needle: u8) -> One { + One(generic::One::new(needle)) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`One::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `One::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "simd128")] + { + true + } + #[cfg(not(target_feature = "simd128"))] + { + false + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `rfind_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Counts all occurrences of this byte in the given haystack. + #[inline] + pub fn count(&self, haystack: &[u8]) -> usize { + // SAFETY: All of our pointers are derived directly from a borrowed + // slice, which is guaranteed to be valid. + unsafe { + let start = haystack.as_ptr(); + let end = start.add(haystack.len()); + self.count_raw(start, end) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < v128::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::fwd_byte_by_byte(start, end, |b| { + b == self.0.needle1() + }); + } + // SAFETY: Building a `One` means it's safe to call 'simd128' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.find_raw_impl(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < v128::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::rev_byte_by_byte(start, end, |b| { + b == self.0.needle1() + }); + } + // SAFETY: Building a `One` means it's safe to call 'simd128' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.rfind_raw_impl(start, end) + } + + /// Counts all occurrences of this byte in the given haystack represented + /// by raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn count_raw(&self, start: *const u8, end: *const u8) -> usize { + if start >= end { + return 0; + } + if end.distance(start) < v128::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::count_byte_by_byte(start, end, |b| { + b == self.0.needle1() + }); + } + // SAFETY: Building a `One` means it's safe to call 'simd128' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.count_raw_impl(start, end) + } + + /// Execute a search using simd128 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::find_raw`], except the distance between `start` and + /// `end` must be at least the size of a simd128 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `simd128` routines.) + #[target_feature(enable = "simd128")] + #[inline] + unsafe fn find_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.find_raw(start, end) + } + + /// Execute a search using simd128 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of a simd128 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `simd128` routines.) + #[target_feature(enable = "simd128")] + #[inline] + unsafe fn rfind_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.rfind_raw(start, end) + } + + /// Execute a count using simd128 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::count_raw`], except the distance between `start` and + /// `end` must be at least the size of a simd128 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `simd128` routines.) + #[target_feature(enable = "simd128")] + #[inline] + unsafe fn count_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> usize { + self.0.count_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle byte in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> OneIter<'a, 'h> { + OneIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of a single byte in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`One::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`One`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct OneIter<'a, 'h> { + searcher: &'a One, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for OneIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn count(self) -> usize { + self.it.count(|s, e| { + // SAFETY: We rely on our generic iterator to return valid start + // and end pointers. + unsafe { self.searcher.count_raw(s, e) } + }) + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for OneIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for OneIter<'a, 'h> {} + +/// Finds all occurrences of two bytes in a haystack. +/// +/// That is, this reports matches of one of two possible bytes. For example, +/// searching for `a` or `b` in `afoobar` would report matches at offsets `0`, +/// `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Two(generic::Two<v128>); + +impl Two { + /// Create a new searcher that finds occurrences of the needle bytes given. + /// + /// This particular searcher is specialized to use simd128 vector + /// instructions that typically make it quite fast. + /// + /// If simd128 is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn new(needle1: u8, needle2: u8) -> Option<Two> { + if Two::is_available() { + // SAFETY: we check that simd128 is available above. + unsafe { Some(Two::new_unchecked(needle1, needle2)) } + } else { + None + } + } + + /// Create a new finder specific to simd128 vectors and routines without + /// checking that simd128 is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute `simd128` + /// instructions in the current environment. + #[target_feature(enable = "simd128")] + #[inline] + pub unsafe fn new_unchecked(needle1: u8, needle2: u8) -> Two { + Two(generic::Two::new(needle1, needle2)) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Two::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `Two::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "simd128")] + { + true + } + #[cfg(not(target_feature = "simd128"))] + { + false + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `rfind_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < v128::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::fwd_byte_by_byte(start, end, |b| { + b == self.0.needle1() || b == self.0.needle2() + }); + } + // SAFETY: Building a `Two` means it's safe to call 'simd128' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.find_raw_impl(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < v128::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::rev_byte_by_byte(start, end, |b| { + b == self.0.needle1() || b == self.0.needle2() + }); + } + // SAFETY: Building a `Two` means it's safe to call 'simd128' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.rfind_raw_impl(start, end) + } + + /// Execute a search using simd128 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::find_raw`], except the distance between `start` and + /// `end` must be at least the size of a simd128 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `simd128` routines.) + #[target_feature(enable = "simd128")] + #[inline] + unsafe fn find_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.find_raw(start, end) + } + + /// Execute a search using simd128 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of a simd128 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `simd128` routines.) + #[target_feature(enable = "simd128")] + #[inline] + unsafe fn rfind_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.rfind_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle bytes in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> TwoIter<'a, 'h> { + TwoIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of two possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Two::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Two`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct TwoIter<'a, 'h> { + searcher: &'a Two, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for TwoIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for TwoIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for TwoIter<'a, 'h> {} + +/// Finds all occurrences of three bytes in a haystack. +/// +/// That is, this reports matches of one of three possible bytes. For example, +/// searching for `a`, `b` or `o` in `afoobar` would report matches at offsets +/// `0`, `2`, `3`, `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Three(generic::Three<v128>); + +impl Three { + /// Create a new searcher that finds occurrences of the needle bytes given. + /// + /// This particular searcher is specialized to use simd128 vector + /// instructions that typically make it quite fast. + /// + /// If simd128 is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn new(needle1: u8, needle2: u8, needle3: u8) -> Option<Three> { + if Three::is_available() { + // SAFETY: we check that simd128 is available above. + unsafe { Some(Three::new_unchecked(needle1, needle2, needle3)) } + } else { + None + } + } + + /// Create a new finder specific to simd128 vectors and routines without + /// checking that simd128 is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute `simd128` + /// instructions in the current environment. + #[target_feature(enable = "simd128")] + #[inline] + pub unsafe fn new_unchecked( + needle1: u8, + needle2: u8, + needle3: u8, + ) -> Three { + Three(generic::Three::new(needle1, needle2, needle3)) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Three::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `Three::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "simd128")] + { + true + } + #[cfg(not(target_feature = "simd128"))] + { + false + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `rfind_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < v128::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::fwd_byte_by_byte(start, end, |b| { + b == self.0.needle1() + || b == self.0.needle2() + || b == self.0.needle3() + }); + } + // SAFETY: Building a `Three` means it's safe to call 'simd128' + // routines. Also, we've checked that our haystack is big enough to run + // on the vector routine. Pointer validity is caller's responsibility. + self.find_raw_impl(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < v128::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::rev_byte_by_byte(start, end, |b| { + b == self.0.needle1() + || b == self.0.needle2() + || b == self.0.needle3() + }); + } + // SAFETY: Building a `Three` means it's safe to call 'simd128' + // routines. Also, we've checked that our haystack is big enough to run + // on the vector routine. Pointer validity is caller's responsibility. + self.rfind_raw_impl(start, end) + } + + /// Execute a search using simd128 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::find_raw`], except the distance between `start` and + /// `end` must be at least the size of a simd128 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `simd128` routines.) + #[target_feature(enable = "simd128")] + #[inline] + unsafe fn find_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.find_raw(start, end) + } + + /// Execute a search using simd128 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of a simd128 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `simd128` routines.) + #[target_feature(enable = "simd128")] + #[inline] + unsafe fn rfind_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.rfind_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle byte in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> ThreeIter<'a, 'h> { + ThreeIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of three possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Three::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Three`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct ThreeIter<'a, 'h> { + searcher: &'a Three, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for ThreeIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for ThreeIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for ThreeIter<'a, 'h> {} + +#[cfg(test)] +mod tests { + use super::*; + + define_memchr_quickcheck!(super); + + #[test] + fn forward_one() { + crate::tests::memchr::Runner::new(1).forward_iter( + |haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_one() { + crate::tests::memchr::Runner::new(1).reverse_iter( + |haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn count_one() { + crate::tests::memchr::Runner::new(1).count_iter(|haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).count()) + }) + } + + #[test] + fn forward_two() { + crate::tests::memchr::Runner::new(2).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2)?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_two() { + crate::tests::memchr::Runner::new(2).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2)?.iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn forward_three() { + crate::tests::memchr::Runner::new(3).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3)?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_three() { + crate::tests::memchr::Runner::new(3).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3)?.iter(haystack).rev().collect()) + }, + ) + } +} diff --git a/vendor/memchr/src/arch/wasm32/simd128/mod.rs b/vendor/memchr/src/arch/wasm32/simd128/mod.rs new file mode 100644 index 0000000..b55d1f0 --- /dev/null +++ b/vendor/memchr/src/arch/wasm32/simd128/mod.rs @@ -0,0 +1,6 @@ +/*! +Algorithms for the `wasm32` target using 128-bit vectors via simd128. +*/ + +pub mod memchr; +pub mod packedpair; diff --git a/vendor/memchr/src/arch/wasm32/simd128/packedpair.rs b/vendor/memchr/src/arch/wasm32/simd128/packedpair.rs new file mode 100644 index 0000000..b629377 --- /dev/null +++ b/vendor/memchr/src/arch/wasm32/simd128/packedpair.rs @@ -0,0 +1,229 @@ +/*! +A 128-bit vector implementation of the "packed pair" SIMD algorithm. + +The "packed pair" algorithm is based on the [generic SIMD] algorithm. The main +difference is that it (by default) uses a background distribution of byte +frequencies to heuristically select the pair of bytes to search for. + +[generic SIMD]: http://0x80.pl/articles/simd-strfind.html#first-and-last +*/ + +use core::arch::wasm32::v128; + +use crate::arch::{all::packedpair::Pair, generic::packedpair}; + +/// A "packed pair" finder that uses 128-bit vector operations. +/// +/// This finder picks two bytes that it believes have high predictive power +/// for indicating an overall match of a needle. Depending on whether +/// `Finder::find` or `Finder::find_prefilter` is used, it reports offsets +/// where the needle matches or could match. In the prefilter case, candidates +/// are reported whenever the [`Pair`] of bytes given matches. +#[derive(Clone, Copy, Debug)] +pub struct Finder(packedpair::Finder<v128>); + +impl Finder { + /// Create a new pair searcher. The searcher returned can either report + /// exact matches of `needle` or act as a prefilter and report candidate + /// positions of `needle`. + /// + /// If simd128 is unavailable in the current environment or if a [`Pair`] + /// could not be constructed from the needle given, then `None` is + /// returned. + #[inline] + pub fn new(needle: &[u8]) -> Option<Finder> { + Finder::with_pair(needle, Pair::new(needle)?) + } + + /// Create a new "packed pair" finder using the pair of bytes given. + /// + /// This constructor permits callers to control precisely which pair of + /// bytes is used as a predicate. + /// + /// If simd128 is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn with_pair(needle: &[u8], pair: Pair) -> Option<Finder> { + if Finder::is_available() { + // SAFETY: we check that simd128 is available above. We are also + // guaranteed to have needle.len() > 1 because we have a valid + // Pair. + unsafe { Some(Finder::with_pair_impl(needle, pair)) } + } else { + None + } + } + + /// Create a new `Finder` specific to simd128 vectors and routines. + /// + /// # Safety + /// + /// Same as the safety for `packedpair::Finder::new`, and callers must also + /// ensure that simd128 is available. + #[target_feature(enable = "simd128")] + #[inline] + unsafe fn with_pair_impl(needle: &[u8], pair: Pair) -> Finder { + let finder = packedpair::Finder::<v128>::new(needle, pair); + Finder(finder) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Finder::with_pair`] will + /// return a `Some` value. Similarly, when it is false, it is guaranteed + /// that `Finder::with_pair` will return a `None` value. Notice that this + /// does not guarantee that [`Finder::new`] will return a `Finder`. Namely, + /// even when `Finder::is_available` is true, it is not guaranteed that a + /// valid [`Pair`] can be found from the needle given. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "simd128")] + { + true + } + #[cfg(not(target_feature = "simd128"))] + { + false + } + } + + /// Execute a search using wasm32 v128 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + #[inline] + pub fn find(&self, haystack: &[u8], needle: &[u8]) -> Option<usize> { + self.find_impl(haystack, needle) + } + + /// Execute a search using wasm32 v128 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + #[inline] + pub fn find_prefilter(&self, haystack: &[u8]) -> Option<usize> { + self.find_prefilter_impl(haystack) + } + + /// Execute a search using wasm32 v128 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Finder`, which can only be constructed + /// when it is safe to call `simd128` routines.) + #[target_feature(enable = "simd128")] + #[inline] + fn find_impl(&self, haystack: &[u8], needle: &[u8]) -> Option<usize> { + // SAFETY: The target feature safety obligation is automatically + // fulfilled by virtue of being a method on `Finder`, which can only be + // constructed when it is safe to call `simd128` routines. + unsafe { self.0.find(haystack, needle) } + } + + /// Execute a prefilter search using wasm32 v128 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Finder`, which can only be constructed + /// when it is safe to call `simd128` routines.) + #[target_feature(enable = "simd128")] + #[inline] + fn find_prefilter_impl(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: The target feature safety obligation is automatically + // fulfilled by virtue of being a method on `Finder`, which can only be + // constructed when it is safe to call `simd128` routines. + unsafe { self.0.find_prefilter(haystack) } + } + + /// Returns the pair of offsets (into the needle) used to check as a + /// predicate before confirming whether a needle exists at a particular + /// position. + #[inline] + pub fn pair(&self) -> &Pair { + self.0.pair() + } + + /// Returns the minimum haystack length that this `Finder` can search. + /// + /// Using a haystack with length smaller than this in a search will result + /// in a panic. The reason for this restriction is that this finder is + /// meant to be a low-level component that is part of a larger substring + /// strategy. In that sense, it avoids trying to handle all cases and + /// instead only handles the cases that it can handle very well. + #[inline] + pub fn min_haystack_len(&self) -> usize { + self.0.min_haystack_len() + } +} + +#[cfg(test)] +mod tests { + use super::*; + + fn find(haystack: &[u8], needle: &[u8]) -> Option<Option<usize>> { + let f = Finder::new(needle)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find(haystack, needle)) + } + + define_substring_forward_quickcheck!(find); + + #[test] + fn forward_substring() { + crate::tests::substring::Runner::new().fwd(find).run() + } + + #[test] + fn forward_packedpair() { + fn find( + haystack: &[u8], + needle: &[u8], + index1: u8, + index2: u8, + ) -> Option<Option<usize>> { + let pair = Pair::with_indices(needle, index1, index2)?; + let f = Finder::with_pair(needle, pair)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find(haystack, needle)) + } + crate::tests::packedpair::Runner::new().fwd(find).run() + } + + #[test] + fn forward_packedpair_prefilter() { + fn find( + haystack: &[u8], + needle: &[u8], + index1: u8, + index2: u8, + ) -> Option<Option<usize>> { + let pair = Pair::with_indices(needle, index1, index2)?; + let f = Finder::with_pair(needle, pair)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find_prefilter(haystack)) + } + crate::tests::packedpair::Runner::new().fwd(find).run() + } +} diff --git a/vendor/memchr/src/arch/x86_64/avx2/memchr.rs b/vendor/memchr/src/arch/x86_64/avx2/memchr.rs new file mode 100644 index 0000000..59f8c7f --- /dev/null +++ b/vendor/memchr/src/arch/x86_64/avx2/memchr.rs @@ -0,0 +1,1352 @@ +/*! +This module defines 256-bit vector implementations of `memchr` and friends. + +The main types in this module are [`One`], [`Two`] and [`Three`]. They are for +searching for one, two or three distinct bytes, respectively, in a haystack. +Each type also has corresponding double ended iterators. These searchers are +typically much faster than scalar routines accomplishing the same task. + +The `One` searcher also provides a [`One::count`] routine for efficiently +counting the number of times a single byte occurs in a haystack. This is +useful, for example, for counting the number of lines in a haystack. This +routine exists because it is usually faster, especially with a high match +count, then using [`One::find`] repeatedly. ([`OneIter`] specializes its +`Iterator::count` implementation to use this routine.) + +Only one, two and three bytes are supported because three bytes is about +the point where one sees diminishing returns. Beyond this point and it's +probably (but not necessarily) better to just use a simple `[bool; 256]` array +or similar. However, it depends mightily on the specific work-load and the +expected match frequency. +*/ + +use core::arch::x86_64::{__m128i, __m256i}; + +use crate::{arch::generic::memchr as generic, ext::Pointer, vector::Vector}; + +/// Finds all occurrences of a single byte in a haystack. +#[derive(Clone, Copy, Debug)] +pub struct One { + /// Used for haystacks less than 32 bytes. + sse2: generic::One<__m128i>, + /// Used for haystacks bigger than 32 bytes. + avx2: generic::One<__m256i>, +} + +impl One { + /// Create a new searcher that finds occurrences of the needle byte given. + /// + /// This particular searcher is specialized to use AVX2 vector instructions + /// that typically make it quite fast. (SSE2 is used for haystacks that + /// are too short to accommodate an AVX2 vector.) + /// + /// If either SSE2 or AVX2 is unavailable in the current environment, then + /// `None` is returned. + #[inline] + pub fn new(needle: u8) -> Option<One> { + if One::is_available() { + // SAFETY: we check that sse2 and avx2 are available above. + unsafe { Some(One::new_unchecked(needle)) } + } else { + None + } + } + + /// Create a new finder specific to AVX2 vectors and routines without + /// checking that either SSE2 or AVX2 is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute both `sse2` and + /// `avx2` instructions in the current environment. + /// + /// Note that it is a common misconception that if one compiles for an + /// `x86_64` target, then they therefore automatically have access to SSE2 + /// instructions. While this is almost always the case, it isn't true in + /// 100% of cases. + #[target_feature(enable = "sse2", enable = "avx2")] + #[inline] + pub unsafe fn new_unchecked(needle: u8) -> One { + One { + sse2: generic::One::new(needle), + avx2: generic::One::new(needle), + } + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`One::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `One::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(not(target_feature = "sse2"))] + { + false + } + #[cfg(target_feature = "sse2")] + { + #[cfg(target_feature = "avx2")] + { + true + } + #[cfg(not(target_feature = "avx2"))] + { + #[cfg(feature = "std")] + { + std::is_x86_feature_detected!("avx2") + } + #[cfg(not(feature = "std"))] + { + false + } + } + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Counts all occurrences of this byte in the given haystack. + #[inline] + pub fn count(&self, haystack: &[u8]) -> usize { + // SAFETY: All of our pointers are derived directly from a borrowed + // slice, which is guaranteed to be valid. + unsafe { + let start = haystack.as_ptr(); + let end = start.add(haystack.len()); + self.count_raw(start, end) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let len = end.distance(start); + if len < __m256i::BYTES { + return if len < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end + // pointers. + generic::fwd_byte_by_byte(start, end, |b| { + b == self.sse2.needle1() + }) + } else { + // SAFETY: We require the caller to pass valid start/end + // pointers. + self.find_raw_sse2(start, end) + }; + } + // SAFETY: Building a `One` means it's safe to call both 'sse2' and + // 'avx2' routines. Also, we've checked that our haystack is big + // enough to run on the vector routine. Pointer validity is caller's + // responsibility. + // + // Note that we could call `self.avx2.find_raw` directly here. But that + // means we'd have to annotate this routine with `target_feature`. + // Which is fine, because this routine is `unsafe` anyway and the + // `target_feature` obligation is met by virtue of building a `One`. + // The real problem is that a routine with a `target_feature` + // annotation generally can't be inlined into caller code unless + // the caller code has the same target feature annotations. Namely, + // the common case (at time of writing) is for calling code to not + // have the `avx2` target feature enabled *at compile time*. Without + // `target_feature` on this routine, it can be inlined which will + // handle some of the short-haystack cases above without touching the + // architecture specific code. + self.find_raw_avx2(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let len = end.distance(start); + if len < __m256i::BYTES { + return if len < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end + // pointers. + generic::rev_byte_by_byte(start, end, |b| { + b == self.sse2.needle1() + }) + } else { + // SAFETY: We require the caller to pass valid start/end + // pointers. + self.rfind_raw_sse2(start, end) + }; + } + // SAFETY: Building a `One` means it's safe to call both 'sse2' and + // 'avx2' routines. Also, we've checked that our haystack is big + // enough to run on the vector routine. Pointer validity is caller's + // responsibility. + // + // See note in forward routine above for why we don't just call + // `self.avx2.rfind_raw` directly here. + self.rfind_raw_avx2(start, end) + } + + /// Counts all occurrences of this byte in the given haystack represented + /// by raw pointers. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `0` will always be returned. + #[inline] + pub unsafe fn count_raw(&self, start: *const u8, end: *const u8) -> usize { + if start >= end { + return 0; + } + let len = end.distance(start); + if len < __m256i::BYTES { + return if len < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end + // pointers. + generic::count_byte_by_byte(start, end, |b| { + b == self.sse2.needle1() + }) + } else { + // SAFETY: We require the caller to pass valid start/end + // pointers. + self.count_raw_sse2(start, end) + }; + } + // SAFETY: Building a `One` means it's safe to call both 'sse2' and + // 'avx2' routines. Also, we've checked that our haystack is big + // enough to run on the vector routine. Pointer validity is caller's + // responsibility. + self.count_raw_avx2(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::find_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn find_raw_sse2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.sse2.find_raw(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn rfind_raw_sse2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.sse2.rfind_raw(start, end) + } + + /// Execute a count using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::count_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn count_raw_sse2( + &self, + start: *const u8, + end: *const u8, + ) -> usize { + self.sse2.count_raw(start, end) + } + + /// Execute a search using AVX2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::find_raw`], except the distance between `start` and + /// `end` must be at least the size of an AVX2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "avx2")] + #[inline] + unsafe fn find_raw_avx2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.avx2.find_raw(start, end) + } + + /// Execute a search using AVX2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of an AVX2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "avx2")] + #[inline] + unsafe fn rfind_raw_avx2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.avx2.rfind_raw(start, end) + } + + /// Execute a count using AVX2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::count_raw`], except the distance between `start` and + /// `end` must be at least the size of an AVX2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "avx2")] + #[inline] + unsafe fn count_raw_avx2( + &self, + start: *const u8, + end: *const u8, + ) -> usize { + self.avx2.count_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle byte in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> OneIter<'a, 'h> { + OneIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of a single byte in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`One::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`One`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct OneIter<'a, 'h> { + searcher: &'a One, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for OneIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn count(self) -> usize { + self.it.count(|s, e| { + // SAFETY: We rely on our generic iterator to return valid start + // and end pointers. + unsafe { self.searcher.count_raw(s, e) } + }) + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for OneIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for OneIter<'a, 'h> {} + +/// Finds all occurrences of two bytes in a haystack. +/// +/// That is, this reports matches of one of two possible bytes. For example, +/// searching for `a` or `b` in `afoobar` would report matches at offsets `0`, +/// `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Two { + /// Used for haystacks less than 32 bytes. + sse2: generic::Two<__m128i>, + /// Used for haystacks bigger than 32 bytes. + avx2: generic::Two<__m256i>, +} + +impl Two { + /// Create a new searcher that finds occurrences of the needle bytes given. + /// + /// This particular searcher is specialized to use AVX2 vector instructions + /// that typically make it quite fast. (SSE2 is used for haystacks that + /// are too short to accommodate an AVX2 vector.) + /// + /// If either SSE2 or AVX2 is unavailable in the current environment, then + /// `None` is returned. + #[inline] + pub fn new(needle1: u8, needle2: u8) -> Option<Two> { + if Two::is_available() { + // SAFETY: we check that sse2 and avx2 are available above. + unsafe { Some(Two::new_unchecked(needle1, needle2)) } + } else { + None + } + } + + /// Create a new finder specific to AVX2 vectors and routines without + /// checking that either SSE2 or AVX2 is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute both `sse2` and + /// `avx2` instructions in the current environment. + /// + /// Note that it is a common misconception that if one compiles for an + /// `x86_64` target, then they therefore automatically have access to SSE2 + /// instructions. While this is almost always the case, it isn't true in + /// 100% of cases. + #[target_feature(enable = "sse2", enable = "avx2")] + #[inline] + pub unsafe fn new_unchecked(needle1: u8, needle2: u8) -> Two { + Two { + sse2: generic::Two::new(needle1, needle2), + avx2: generic::Two::new(needle1, needle2), + } + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Two::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `Two::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(not(target_feature = "sse2"))] + { + false + } + #[cfg(target_feature = "sse2")] + { + #[cfg(target_feature = "avx2")] + { + true + } + #[cfg(not(target_feature = "avx2"))] + { + #[cfg(feature = "std")] + { + std::is_x86_feature_detected!("avx2") + } + #[cfg(not(feature = "std"))] + { + false + } + } + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let len = end.distance(start); + if len < __m256i::BYTES { + return if len < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end + // pointers. + generic::fwd_byte_by_byte(start, end, |b| { + b == self.sse2.needle1() || b == self.sse2.needle2() + }) + } else { + // SAFETY: We require the caller to pass valid start/end + // pointers. + self.find_raw_sse2(start, end) + }; + } + // SAFETY: Building a `Two` means it's safe to call both 'sse2' and + // 'avx2' routines. Also, we've checked that our haystack is big + // enough to run on the vector routine. Pointer validity is caller's + // responsibility. + // + // Note that we could call `self.avx2.find_raw` directly here. But that + // means we'd have to annotate this routine with `target_feature`. + // Which is fine, because this routine is `unsafe` anyway and the + // `target_feature` obligation is met by virtue of building a `Two`. + // The real problem is that a routine with a `target_feature` + // annotation generally can't be inlined into caller code unless + // the caller code has the same target feature annotations. Namely, + // the common case (at time of writing) is for calling code to not + // have the `avx2` target feature enabled *at compile time*. Without + // `target_feature` on this routine, it can be inlined which will + // handle some of the short-haystack cases above without touching the + // architecture specific code. + self.find_raw_avx2(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let len = end.distance(start); + if len < __m256i::BYTES { + return if len < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end + // pointers. + generic::rev_byte_by_byte(start, end, |b| { + b == self.sse2.needle1() || b == self.sse2.needle2() + }) + } else { + // SAFETY: We require the caller to pass valid start/end + // pointers. + self.rfind_raw_sse2(start, end) + }; + } + // SAFETY: Building a `Two` means it's safe to call both 'sse2' and + // 'avx2' routines. Also, we've checked that our haystack is big + // enough to run on the vector routine. Pointer validity is caller's + // responsibility. + // + // See note in forward routine above for why we don't just call + // `self.avx2.rfind_raw` directly here. + self.rfind_raw_avx2(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::find_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn find_raw_sse2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.sse2.find_raw(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn rfind_raw_sse2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.sse2.rfind_raw(start, end) + } + + /// Execute a search using AVX2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::find_raw`], except the distance between `start` and + /// `end` must be at least the size of an AVX2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "avx2")] + #[inline] + unsafe fn find_raw_avx2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.avx2.find_raw(start, end) + } + + /// Execute a search using AVX2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of an AVX2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "avx2")] + #[inline] + unsafe fn rfind_raw_avx2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.avx2.rfind_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle bytes in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> TwoIter<'a, 'h> { + TwoIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of two possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Two::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Two`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct TwoIter<'a, 'h> { + searcher: &'a Two, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for TwoIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for TwoIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for TwoIter<'a, 'h> {} + +/// Finds all occurrences of three bytes in a haystack. +/// +/// That is, this reports matches of one of three possible bytes. For example, +/// searching for `a`, `b` or `o` in `afoobar` would report matches at offsets +/// `0`, `2`, `3`, `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Three { + /// Used for haystacks less than 32 bytes. + sse2: generic::Three<__m128i>, + /// Used for haystacks bigger than 32 bytes. + avx2: generic::Three<__m256i>, +} + +impl Three { + /// Create a new searcher that finds occurrences of the needle bytes given. + /// + /// This particular searcher is specialized to use AVX2 vector instructions + /// that typically make it quite fast. (SSE2 is used for haystacks that + /// are too short to accommodate an AVX2 vector.) + /// + /// If either SSE2 or AVX2 is unavailable in the current environment, then + /// `None` is returned. + #[inline] + pub fn new(needle1: u8, needle2: u8, needle3: u8) -> Option<Three> { + if Three::is_available() { + // SAFETY: we check that sse2 and avx2 are available above. + unsafe { Some(Three::new_unchecked(needle1, needle2, needle3)) } + } else { + None + } + } + + /// Create a new finder specific to AVX2 vectors and routines without + /// checking that either SSE2 or AVX2 is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute both `sse2` and + /// `avx2` instructions in the current environment. + /// + /// Note that it is a common misconception that if one compiles for an + /// `x86_64` target, then they therefore automatically have access to SSE2 + /// instructions. While this is almost always the case, it isn't true in + /// 100% of cases. + #[target_feature(enable = "sse2", enable = "avx2")] + #[inline] + pub unsafe fn new_unchecked( + needle1: u8, + needle2: u8, + needle3: u8, + ) -> Three { + Three { + sse2: generic::Three::new(needle1, needle2, needle3), + avx2: generic::Three::new(needle1, needle2, needle3), + } + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Three::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `Three::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(not(target_feature = "sse2"))] + { + false + } + #[cfg(target_feature = "sse2")] + { + #[cfg(target_feature = "avx2")] + { + true + } + #[cfg(not(target_feature = "avx2"))] + { + #[cfg(feature = "std")] + { + std::is_x86_feature_detected!("avx2") + } + #[cfg(not(feature = "std"))] + { + false + } + } + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let len = end.distance(start); + if len < __m256i::BYTES { + return if len < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end + // pointers. + generic::fwd_byte_by_byte(start, end, |b| { + b == self.sse2.needle1() + || b == self.sse2.needle2() + || b == self.sse2.needle3() + }) + } else { + // SAFETY: We require the caller to pass valid start/end + // pointers. + self.find_raw_sse2(start, end) + }; + } + // SAFETY: Building a `Three` means it's safe to call both 'sse2' and + // 'avx2' routines. Also, we've checked that our haystack is big + // enough to run on the vector routine. Pointer validity is caller's + // responsibility. + // + // Note that we could call `self.avx2.find_raw` directly here. But that + // means we'd have to annotate this routine with `target_feature`. + // Which is fine, because this routine is `unsafe` anyway and the + // `target_feature` obligation is met by virtue of building a `Three`. + // The real problem is that a routine with a `target_feature` + // annotation generally can't be inlined into caller code unless + // the caller code has the same target feature annotations. Namely, + // the common case (at time of writing) is for calling code to not + // have the `avx2` target feature enabled *at compile time*. Without + // `target_feature` on this routine, it can be inlined which will + // handle some of the short-haystack cases above without touching the + // architecture specific code. + self.find_raw_avx2(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + let len = end.distance(start); + if len < __m256i::BYTES { + return if len < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end + // pointers. + generic::rev_byte_by_byte(start, end, |b| { + b == self.sse2.needle1() + || b == self.sse2.needle2() + || b == self.sse2.needle3() + }) + } else { + // SAFETY: We require the caller to pass valid start/end + // pointers. + self.rfind_raw_sse2(start, end) + }; + } + // SAFETY: Building a `Three` means it's safe to call both 'sse2' and + // 'avx2' routines. Also, we've checked that our haystack is big + // enough to run on the vector routine. Pointer validity is caller's + // responsibility. + // + // See note in forward routine above for why we don't just call + // `self.avx2.rfind_raw` directly here. + self.rfind_raw_avx2(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::find_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn find_raw_sse2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.sse2.find_raw(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn rfind_raw_sse2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.sse2.rfind_raw(start, end) + } + + /// Execute a search using AVX2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::find_raw`], except the distance between `start` and + /// `end` must be at least the size of an AVX2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "avx2")] + #[inline] + unsafe fn find_raw_avx2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.avx2.find_raw(start, end) + } + + /// Execute a search using AVX2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of an AVX2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `sse2`/`avx2` routines.) + #[target_feature(enable = "avx2")] + #[inline] + unsafe fn rfind_raw_avx2( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.avx2.rfind_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle bytes in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> ThreeIter<'a, 'h> { + ThreeIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of three possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Three::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Three`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct ThreeIter<'a, 'h> { + searcher: &'a Three, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for ThreeIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for ThreeIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for ThreeIter<'a, 'h> {} + +#[cfg(test)] +mod tests { + use super::*; + + define_memchr_quickcheck!(super); + + #[test] + fn forward_one() { + crate::tests::memchr::Runner::new(1).forward_iter( + |haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_one() { + crate::tests::memchr::Runner::new(1).reverse_iter( + |haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn count_one() { + crate::tests::memchr::Runner::new(1).count_iter(|haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).count()) + }) + } + + #[test] + fn forward_two() { + crate::tests::memchr::Runner::new(2).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2)?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_two() { + crate::tests::memchr::Runner::new(2).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2)?.iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn forward_three() { + crate::tests::memchr::Runner::new(3).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3)?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_three() { + crate::tests::memchr::Runner::new(3).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3)?.iter(haystack).rev().collect()) + }, + ) + } +} diff --git a/vendor/memchr/src/arch/x86_64/avx2/mod.rs b/vendor/memchr/src/arch/x86_64/avx2/mod.rs new file mode 100644 index 0000000..ee4097d --- /dev/null +++ b/vendor/memchr/src/arch/x86_64/avx2/mod.rs @@ -0,0 +1,6 @@ +/*! +Algorithms for the `x86_64` target using 256-bit vectors via AVX2. +*/ + +pub mod memchr; +pub mod packedpair; diff --git a/vendor/memchr/src/arch/x86_64/avx2/packedpair.rs b/vendor/memchr/src/arch/x86_64/avx2/packedpair.rs new file mode 100644 index 0000000..efae7b6 --- /dev/null +++ b/vendor/memchr/src/arch/x86_64/avx2/packedpair.rs @@ -0,0 +1,272 @@ +/*! +A 256-bit vector implementation of the "packed pair" SIMD algorithm. + +The "packed pair" algorithm is based on the [generic SIMD] algorithm. The main +difference is that it (by default) uses a background distribution of byte +frequencies to heuristically select the pair of bytes to search for. + +[generic SIMD]: http://0x80.pl/articles/simd-strfind.html#first-and-last +*/ + +use core::arch::x86_64::{__m128i, __m256i}; + +use crate::arch::{all::packedpair::Pair, generic::packedpair}; + +/// A "packed pair" finder that uses 256-bit vector operations. +/// +/// This finder picks two bytes that it believes have high predictive power +/// for indicating an overall match of a needle. Depending on whether +/// `Finder::find` or `Finder::find_prefilter` is used, it reports offsets +/// where the needle matches or could match. In the prefilter case, candidates +/// are reported whenever the [`Pair`] of bytes given matches. +#[derive(Clone, Copy, Debug)] +pub struct Finder { + sse2: packedpair::Finder<__m128i>, + avx2: packedpair::Finder<__m256i>, +} + +impl Finder { + /// Create a new pair searcher. The searcher returned can either report + /// exact matches of `needle` or act as a prefilter and report candidate + /// positions of `needle`. + /// + /// If AVX2 is unavailable in the current environment or if a [`Pair`] + /// could not be constructed from the needle given, then `None` is + /// returned. + #[inline] + pub fn new(needle: &[u8]) -> Option<Finder> { + Finder::with_pair(needle, Pair::new(needle)?) + } + + /// Create a new "packed pair" finder using the pair of bytes given. + /// + /// This constructor permits callers to control precisely which pair of + /// bytes is used as a predicate. + /// + /// If AVX2 is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn with_pair(needle: &[u8], pair: Pair) -> Option<Finder> { + if Finder::is_available() { + // SAFETY: we check that sse2/avx2 is available above. We are also + // guaranteed to have needle.len() > 1 because we have a valid + // Pair. + unsafe { Some(Finder::with_pair_impl(needle, pair)) } + } else { + None + } + } + + /// Create a new `Finder` specific to SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as the safety for `packedpair::Finder::new`, and callers must also + /// ensure that both SSE2 and AVX2 are available. + #[target_feature(enable = "sse2", enable = "avx2")] + #[inline] + unsafe fn with_pair_impl(needle: &[u8], pair: Pair) -> Finder { + let sse2 = packedpair::Finder::<__m128i>::new(needle, pair); + let avx2 = packedpair::Finder::<__m256i>::new(needle, pair); + Finder { sse2, avx2 } + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Finder::with_pair`] will + /// return a `Some` value. Similarly, when it is false, it is guaranteed + /// that `Finder::with_pair` will return a `None` value. Notice that this + /// does not guarantee that [`Finder::new`] will return a `Finder`. Namely, + /// even when `Finder::is_available` is true, it is not guaranteed that a + /// valid [`Pair`] can be found from the needle given. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(not(target_feature = "sse2"))] + { + false + } + #[cfg(target_feature = "sse2")] + { + #[cfg(target_feature = "avx2")] + { + true + } + #[cfg(not(target_feature = "avx2"))] + { + #[cfg(feature = "std")] + { + std::is_x86_feature_detected!("avx2") + } + #[cfg(not(feature = "std"))] + { + false + } + } + } + } + + /// Execute a search using AVX2 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + #[inline] + pub fn find(&self, haystack: &[u8], needle: &[u8]) -> Option<usize> { + // SAFETY: Building a `Finder` means it's safe to call 'sse2' routines. + unsafe { self.find_impl(haystack, needle) } + } + + /// Run this finder on the given haystack as a prefilter. + /// + /// If a candidate match is found, then an offset where the needle *could* + /// begin in the haystack is returned. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + #[inline] + pub fn find_prefilter(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: Building a `Finder` means it's safe to call 'sse2' routines. + unsafe { self.find_prefilter_impl(haystack) } + } + + /// Execute a search using AVX2 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Finder`, which can only be constructed + /// when it is safe to call `sse2` and `avx2` routines.) + #[target_feature(enable = "sse2", enable = "avx2")] + #[inline] + unsafe fn find_impl( + &self, + haystack: &[u8], + needle: &[u8], + ) -> Option<usize> { + if haystack.len() < self.avx2.min_haystack_len() { + self.sse2.find(haystack, needle) + } else { + self.avx2.find(haystack, needle) + } + } + + /// Execute a prefilter search using AVX2 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Finder`, which can only be constructed + /// when it is safe to call `sse2` and `avx2` routines.) + #[target_feature(enable = "sse2", enable = "avx2")] + #[inline] + unsafe fn find_prefilter_impl(&self, haystack: &[u8]) -> Option<usize> { + if haystack.len() < self.avx2.min_haystack_len() { + self.sse2.find_prefilter(haystack) + } else { + self.avx2.find_prefilter(haystack) + } + } + + /// Returns the pair of offsets (into the needle) used to check as a + /// predicate before confirming whether a needle exists at a particular + /// position. + #[inline] + pub fn pair(&self) -> &Pair { + self.avx2.pair() + } + + /// Returns the minimum haystack length that this `Finder` can search. + /// + /// Using a haystack with length smaller than this in a search will result + /// in a panic. The reason for this restriction is that this finder is + /// meant to be a low-level component that is part of a larger substring + /// strategy. In that sense, it avoids trying to handle all cases and + /// instead only handles the cases that it can handle very well. + #[inline] + pub fn min_haystack_len(&self) -> usize { + // The caller doesn't need to care about AVX2's min_haystack_len + // since this implementation will automatically switch to the SSE2 + // implementation if the haystack is too short for AVX2. Therefore, the + // caller only needs to care about SSE2's min_haystack_len. + // + // This does assume that SSE2's min_haystack_len is less than or + // equal to AVX2's min_haystack_len. In practice, this is true and + // there is no way it could be false based on how this Finder is + // implemented. Namely, both SSE2 and AVX2 use the same `Pair`. If + // they used different pairs, then it's possible (although perhaps + // pathological) for SSE2's min_haystack_len to be bigger than AVX2's. + self.sse2.min_haystack_len() + } +} + +#[cfg(test)] +mod tests { + use super::*; + + fn find(haystack: &[u8], needle: &[u8]) -> Option<Option<usize>> { + let f = Finder::new(needle)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find(haystack, needle)) + } + + define_substring_forward_quickcheck!(find); + + #[test] + fn forward_substring() { + crate::tests::substring::Runner::new().fwd(find).run() + } + + #[test] + fn forward_packedpair() { + fn find( + haystack: &[u8], + needle: &[u8], + index1: u8, + index2: u8, + ) -> Option<Option<usize>> { + let pair = Pair::with_indices(needle, index1, index2)?; + let f = Finder::with_pair(needle, pair)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find(haystack, needle)) + } + crate::tests::packedpair::Runner::new().fwd(find).run() + } + + #[test] + fn forward_packedpair_prefilter() { + fn find( + haystack: &[u8], + needle: &[u8], + index1: u8, + index2: u8, + ) -> Option<Option<usize>> { + if !cfg!(target_feature = "sse2") { + return None; + } + let pair = Pair::with_indices(needle, index1, index2)?; + let f = Finder::with_pair(needle, pair)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find_prefilter(haystack)) + } + crate::tests::packedpair::Runner::new().fwd(find).run() + } +} diff --git a/vendor/memchr/src/arch/x86_64/memchr.rs b/vendor/memchr/src/arch/x86_64/memchr.rs new file mode 100644 index 0000000..fcb1399 --- /dev/null +++ b/vendor/memchr/src/arch/x86_64/memchr.rs @@ -0,0 +1,335 @@ +/*! +Wrapper routines for `memchr` and friends. + +These routines efficiently dispatch to the best implementation based on what +the CPU supports. +*/ + +/// Provides a way to run a memchr-like function while amortizing the cost of +/// runtime CPU feature detection. +/// +/// This works by loading a function pointer from an atomic global. Initially, +/// this global is set to a function that does CPU feature detection. For +/// example, if AVX2 is enabled, then the AVX2 implementation is used. +/// Otherwise, at least on x86_64, the SSE2 implementation is used. (And +/// in some niche cases, if SSE2 isn't available, then the architecture +/// independent fallback implementation is used.) +/// +/// After the first call to this function, the atomic global is replaced with +/// the specific AVX2, SSE2 or fallback routine chosen. Subsequent calls then +/// will directly call the chosen routine instead of needing to go through the +/// CPU feature detection branching again. +/// +/// This particular macro is specifically written to provide the implementation +/// of functions with the following signature: +/// +/// ```ignore +/// fn memchr(needle1: u8, start: *const u8, end: *const u8) -> Option<usize>; +/// ``` +/// +/// Where you can also have `memchr2` and `memchr3`, but with `needle2` and +/// `needle3`, respectively. The `start` and `end` parameters correspond to the +/// start and end of the haystack, respectively. +/// +/// We use raw pointers here instead of the more obvious `haystack: &[u8]` so +/// that the function is compatible with our lower level iterator logic that +/// operates on raw pointers. We use this macro to implement "raw" memchr +/// routines with the signature above, and then define memchr routines using +/// regular slices on top of them. +/// +/// Note that we use `#[cfg(target_feature = "sse2")]` below even though +/// it shouldn't be strictly necessary because without it, it seems to +/// cause the compiler to blow up. I guess it can't handle a function +/// pointer being created with a sse target feature? Dunno. See the +/// `build-for-x86-64-but-non-sse-target` CI job if you want to experiment with +/// this. +/// +/// # Safety +/// +/// Primarily callers must that `$fnty` is a correct function pointer type and +/// not something else. +/// +/// Callers must also ensure that `$memchrty::$memchrfind` corresponds to a +/// routine that returns a valid function pointer when a match is found. That +/// is, a pointer that is `>= start` and `< end`. +/// +/// Callers must also ensure that the `$hay_start` and `$hay_end` identifiers +/// correspond to valid pointers. +macro_rules! unsafe_ifunc { + ( + $memchrty:ident, + $memchrfind:ident, + $fnty:ty, + $retty:ty, + $hay_start:ident, + $hay_end:ident, + $($needle:ident),+ + ) => {{ + #![allow(unused_unsafe)] + + use core::sync::atomic::{AtomicPtr, Ordering}; + + type Fn = *mut (); + type RealFn = $fnty; + static FN: AtomicPtr<()> = AtomicPtr::new(detect as Fn); + + #[cfg(target_feature = "sse2")] + #[target_feature(enable = "sse2", enable = "avx2")] + unsafe fn find_avx2( + $($needle: u8),+, + $hay_start: *const u8, + $hay_end: *const u8, + ) -> $retty { + use crate::arch::x86_64::avx2::memchr::$memchrty; + $memchrty::new_unchecked($($needle),+) + .$memchrfind($hay_start, $hay_end) + } + + #[cfg(target_feature = "sse2")] + #[target_feature(enable = "sse2")] + unsafe fn find_sse2( + $($needle: u8),+, + $hay_start: *const u8, + $hay_end: *const u8, + ) -> $retty { + use crate::arch::x86_64::sse2::memchr::$memchrty; + $memchrty::new_unchecked($($needle),+) + .$memchrfind($hay_start, $hay_end) + } + + unsafe fn find_fallback( + $($needle: u8),+, + $hay_start: *const u8, + $hay_end: *const u8, + ) -> $retty { + use crate::arch::all::memchr::$memchrty; + $memchrty::new($($needle),+).$memchrfind($hay_start, $hay_end) + } + + unsafe fn detect( + $($needle: u8),+, + $hay_start: *const u8, + $hay_end: *const u8, + ) -> $retty { + let fun = { + #[cfg(not(target_feature = "sse2"))] + { + debug!( + "no sse2 feature available, using fallback for {}", + stringify!($memchrty), + ); + find_fallback as RealFn + } + #[cfg(target_feature = "sse2")] + { + use crate::arch::x86_64::{sse2, avx2}; + if avx2::memchr::$memchrty::is_available() { + debug!("chose AVX2 for {}", stringify!($memchrty)); + find_avx2 as RealFn + } else if sse2::memchr::$memchrty::is_available() { + debug!("chose SSE2 for {}", stringify!($memchrty)); + find_sse2 as RealFn + } else { + debug!("chose fallback for {}", stringify!($memchrty)); + find_fallback as RealFn + } + } + }; + FN.store(fun as Fn, Ordering::Relaxed); + // SAFETY: The only thing we need to uphold here is the + // `#[target_feature]` requirements. Since we check is_available + // above before using the corresponding implementation, we are + // guaranteed to only call code that is supported on the current + // CPU. + fun($($needle),+, $hay_start, $hay_end) + } + + // SAFETY: By virtue of the caller contract, RealFn is a function + // pointer, which is always safe to transmute with a *mut (). Also, + // since we use $memchrty::is_available, it is guaranteed to be safe + // to call $memchrty::$memchrfind. + unsafe { + let fun = FN.load(Ordering::Relaxed); + core::mem::transmute::<Fn, RealFn>(fun)( + $($needle),+, + $hay_start, + $hay_end, + ) + } + }}; +} + +// The routines below dispatch to AVX2, SSE2 or a fallback routine based on +// what's available in the current environment. The secret sauce here is that +// we only check for which one to use approximately once, and then "cache" that +// choice into a global function pointer. Subsequent invocations then just call +// the appropriate function directly. + +/// memchr, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::find_raw`. +#[inline(always)] +pub(crate) fn memchr_raw( + n1: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + // SAFETY: We provide a valid function pointer type. + unsafe_ifunc!( + One, + find_raw, + unsafe fn(u8, *const u8, *const u8) -> Option<*const u8>, + Option<*const u8>, + start, + end, + n1 + ) +} + +/// memrchr, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::rfind_raw`. +#[inline(always)] +pub(crate) fn memrchr_raw( + n1: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + // SAFETY: We provide a valid function pointer type. + unsafe_ifunc!( + One, + rfind_raw, + unsafe fn(u8, *const u8, *const u8) -> Option<*const u8>, + Option<*const u8>, + start, + end, + n1 + ) +} + +/// memchr2, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Two::find_raw`. +#[inline(always)] +pub(crate) fn memchr2_raw( + n1: u8, + n2: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + // SAFETY: We provide a valid function pointer type. + unsafe_ifunc!( + Two, + find_raw, + unsafe fn(u8, u8, *const u8, *const u8) -> Option<*const u8>, + Option<*const u8>, + start, + end, + n1, + n2 + ) +} + +/// memrchr2, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Two::rfind_raw`. +#[inline(always)] +pub(crate) fn memrchr2_raw( + n1: u8, + n2: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + // SAFETY: We provide a valid function pointer type. + unsafe_ifunc!( + Two, + rfind_raw, + unsafe fn(u8, u8, *const u8, *const u8) -> Option<*const u8>, + Option<*const u8>, + start, + end, + n1, + n2 + ) +} + +/// memchr3, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Three::find_raw`. +#[inline(always)] +pub(crate) fn memchr3_raw( + n1: u8, + n2: u8, + n3: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + // SAFETY: We provide a valid function pointer type. + unsafe_ifunc!( + Three, + find_raw, + unsafe fn(u8, u8, u8, *const u8, *const u8) -> Option<*const u8>, + Option<*const u8>, + start, + end, + n1, + n2, + n3 + ) +} + +/// memrchr3, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Three::rfind_raw`. +#[inline(always)] +pub(crate) fn memrchr3_raw( + n1: u8, + n2: u8, + n3: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + // SAFETY: We provide a valid function pointer type. + unsafe_ifunc!( + Three, + rfind_raw, + unsafe fn(u8, u8, u8, *const u8, *const u8) -> Option<*const u8>, + Option<*const u8>, + start, + end, + n1, + n2, + n3 + ) +} + +/// Count all matching bytes, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::count_raw`. +#[inline(always)] +pub(crate) fn count_raw(n1: u8, start: *const u8, end: *const u8) -> usize { + // SAFETY: We provide a valid function pointer type. + unsafe_ifunc!( + One, + count_raw, + unsafe fn(u8, *const u8, *const u8) -> usize, + usize, + start, + end, + n1 + ) +} diff --git a/vendor/memchr/src/arch/x86_64/mod.rs b/vendor/memchr/src/arch/x86_64/mod.rs new file mode 100644 index 0000000..5dad721 --- /dev/null +++ b/vendor/memchr/src/arch/x86_64/mod.rs @@ -0,0 +1,8 @@ +/*! +Vector algorithms for the `x86_64` target. +*/ + +pub mod avx2; +pub mod sse2; + +pub(crate) mod memchr; diff --git a/vendor/memchr/src/arch/x86_64/sse2/memchr.rs b/vendor/memchr/src/arch/x86_64/sse2/memchr.rs new file mode 100644 index 0000000..c6f75df --- /dev/null +++ b/vendor/memchr/src/arch/x86_64/sse2/memchr.rs @@ -0,0 +1,1077 @@ +/*! +This module defines 128-bit vector implementations of `memchr` and friends. + +The main types in this module are [`One`], [`Two`] and [`Three`]. They are for +searching for one, two or three distinct bytes, respectively, in a haystack. +Each type also has corresponding double ended iterators. These searchers are +typically much faster than scalar routines accomplishing the same task. + +The `One` searcher also provides a [`One::count`] routine for efficiently +counting the number of times a single byte occurs in a haystack. This is +useful, for example, for counting the number of lines in a haystack. This +routine exists because it is usually faster, especially with a high match +count, then using [`One::find`] repeatedly. ([`OneIter`] specializes its +`Iterator::count` implementation to use this routine.) + +Only one, two and three bytes are supported because three bytes is about +the point where one sees diminishing returns. Beyond this point and it's +probably (but not necessarily) better to just use a simple `[bool; 256]` array +or similar. However, it depends mightily on the specific work-load and the +expected match frequency. +*/ + +use core::arch::x86_64::__m128i; + +use crate::{arch::generic::memchr as generic, ext::Pointer, vector::Vector}; + +/// Finds all occurrences of a single byte in a haystack. +#[derive(Clone, Copy, Debug)] +pub struct One(generic::One<__m128i>); + +impl One { + /// Create a new searcher that finds occurrences of the needle byte given. + /// + /// This particular searcher is specialized to use SSE2 vector instructions + /// that typically make it quite fast. + /// + /// If SSE2 is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn new(needle: u8) -> Option<One> { + if One::is_available() { + // SAFETY: we check that sse2 is available above. + unsafe { Some(One::new_unchecked(needle)) } + } else { + None + } + } + + /// Create a new finder specific to SSE2 vectors and routines without + /// checking that SSE2 is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute `sse2` instructions + /// in the current environment. + /// + /// Note that it is a common misconception that if one compiles for an + /// `x86_64` target, then they therefore automatically have access to SSE2 + /// instructions. While this is almost always the case, it isn't true in + /// 100% of cases. + #[target_feature(enable = "sse2")] + #[inline] + pub unsafe fn new_unchecked(needle: u8) -> One { + One(generic::One::new(needle)) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`One::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `One::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "sse2")] + { + true + } + #[cfg(not(target_feature = "sse2"))] + { + false + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `rfind_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Counts all occurrences of this byte in the given haystack. + #[inline] + pub fn count(&self, haystack: &[u8]) -> usize { + // SAFETY: All of our pointers are derived directly from a borrowed + // slice, which is guaranteed to be valid. + unsafe { + let start = haystack.as_ptr(); + let end = start.add(haystack.len()); + self.count_raw(start, end) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::fwd_byte_by_byte(start, end, |b| { + b == self.0.needle1() + }); + } + // SAFETY: Building a `One` means it's safe to call 'sse2' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + // + // Note that we could call `self.0.find_raw` directly here. But that + // means we'd have to annotate this routine with `target_feature`. + // Which is fine, because this routine is `unsafe` anyway and the + // `target_feature` obligation is met by virtue of building a `One`. + // The real problem is that a routine with a `target_feature` + // annotation generally can't be inlined into caller code unless the + // caller code has the same target feature annotations. Which is maybe + // okay for SSE2, but we do the same thing for AVX2 where caller code + // probably usually doesn't have AVX2 enabled. That means that this + // routine can be inlined which will handle some of the short-haystack + // cases above without touching the architecture specific code. + self.find_raw_impl(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::rev_byte_by_byte(start, end, |b| { + b == self.0.needle1() + }); + } + // SAFETY: Building a `One` means it's safe to call 'sse2' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + // + // See note in forward routine above for why we don't just call + // `self.0.rfind_raw` directly here. + self.rfind_raw_impl(start, end) + } + + /// Counts all occurrences of this byte in the given haystack represented + /// by raw pointers. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `0` will always be returned. + #[inline] + pub unsafe fn count_raw(&self, start: *const u8, end: *const u8) -> usize { + if start >= end { + return 0; + } + if end.distance(start) < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::count_byte_by_byte(start, end, |b| { + b == self.0.needle1() + }); + } + // SAFETY: Building a `One` means it's safe to call 'sse2' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + self.count_raw_impl(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::find_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `sse2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn find_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.find_raw(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `sse2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn rfind_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.rfind_raw(start, end) + } + + /// Execute a count using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`One::count_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `One`, which can only be constructed + /// when it is safe to call `sse2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn count_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> usize { + self.0.count_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle byte in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> OneIter<'a, 'h> { + OneIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of a single byte in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`One::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`One`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct OneIter<'a, 'h> { + searcher: &'a One, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for OneIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn count(self) -> usize { + self.it.count(|s, e| { + // SAFETY: We rely on our generic iterator to return valid start + // and end pointers. + unsafe { self.searcher.count_raw(s, e) } + }) + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for OneIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for OneIter<'a, 'h> {} + +/// Finds all occurrences of two bytes in a haystack. +/// +/// That is, this reports matches of one of two possible bytes. For example, +/// searching for `a` or `b` in `afoobar` would report matches at offsets `0`, +/// `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Two(generic::Two<__m128i>); + +impl Two { + /// Create a new searcher that finds occurrences of the needle bytes given. + /// + /// This particular searcher is specialized to use SSE2 vector instructions + /// that typically make it quite fast. + /// + /// If SSE2 is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn new(needle1: u8, needle2: u8) -> Option<Two> { + if Two::is_available() { + // SAFETY: we check that sse2 is available above. + unsafe { Some(Two::new_unchecked(needle1, needle2)) } + } else { + None + } + } + + /// Create a new finder specific to SSE2 vectors and routines without + /// checking that SSE2 is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute `sse2` instructions + /// in the current environment. + /// + /// Note that it is a common misconception that if one compiles for an + /// `x86_64` target, then they therefore automatically have access to SSE2 + /// instructions. While this is almost always the case, it isn't true in + /// 100% of cases. + #[target_feature(enable = "sse2")] + #[inline] + pub unsafe fn new_unchecked(needle1: u8, needle2: u8) -> Two { + Two(generic::Two::new(needle1, needle2)) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Two::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `Two::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "sse2")] + { + true + } + #[cfg(not(target_feature = "sse2"))] + { + false + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `rfind_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::fwd_byte_by_byte(start, end, |b| { + b == self.0.needle1() || b == self.0.needle2() + }); + } + // SAFETY: Building a `Two` means it's safe to call 'sse2' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + // + // Note that we could call `self.0.find_raw` directly here. But that + // means we'd have to annotate this routine with `target_feature`. + // Which is fine, because this routine is `unsafe` anyway and the + // `target_feature` obligation is met by virtue of building a `Two`. + // The real problem is that a routine with a `target_feature` + // annotation generally can't be inlined into caller code unless the + // caller code has the same target feature annotations. Which is maybe + // okay for SSE2, but we do the same thing for AVX2 where caller code + // probably usually doesn't have AVX2 enabled. That means that this + // routine can be inlined which will handle some of the short-haystack + // cases above without touching the architecture specific code. + self.find_raw_impl(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::rev_byte_by_byte(start, end, |b| { + b == self.0.needle1() || b == self.0.needle2() + }); + } + // SAFETY: Building a `Two` means it's safe to call 'sse2' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + // + // See note in forward routine above for why we don't just call + // `self.0.rfind_raw` directly here. + self.rfind_raw_impl(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::find_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `sse2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn find_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.find_raw(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Two::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Two`, which can only be constructed + /// when it is safe to call `sse2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn rfind_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.rfind_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle bytes in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> TwoIter<'a, 'h> { + TwoIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of two possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Two::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Two`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct TwoIter<'a, 'h> { + searcher: &'a Two, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for TwoIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for TwoIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for TwoIter<'a, 'h> {} + +/// Finds all occurrences of three bytes in a haystack. +/// +/// That is, this reports matches of one of three possible bytes. For example, +/// searching for `a`, `b` or `o` in `afoobar` would report matches at offsets +/// `0`, `2`, `3`, `4` and `5`. +#[derive(Clone, Copy, Debug)] +pub struct Three(generic::Three<__m128i>); + +impl Three { + /// Create a new searcher that finds occurrences of the needle bytes given. + /// + /// This particular searcher is specialized to use SSE2 vector instructions + /// that typically make it quite fast. + /// + /// If SSE2 is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn new(needle1: u8, needle2: u8, needle3: u8) -> Option<Three> { + if Three::is_available() { + // SAFETY: we check that sse2 is available above. + unsafe { Some(Three::new_unchecked(needle1, needle2, needle3)) } + } else { + None + } + } + + /// Create a new finder specific to SSE2 vectors and routines without + /// checking that SSE2 is available. + /// + /// # Safety + /// + /// Callers must guarantee that it is safe to execute `sse2` instructions + /// in the current environment. + /// + /// Note that it is a common misconception that if one compiles for an + /// `x86_64` target, then they therefore automatically have access to SSE2 + /// instructions. While this is almost always the case, it isn't true in + /// 100% of cases. + #[target_feature(enable = "sse2")] + #[inline] + pub unsafe fn new_unchecked( + needle1: u8, + needle2: u8, + needle3: u8, + ) -> Three { + Three(generic::Three::new(needle1, needle2, needle3)) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Three::new`] will return + /// a `Some` value. Similarly, when it is false, it is guaranteed that + /// `Three::new` will return a `None` value. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(target_feature = "sse2")] + { + true + } + #[cfg(not(target_feature = "sse2"))] + { + false + } + } + + /// Return the first occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `find_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.find_raw(s, e) + }) + } + } + + /// Return the last occurrence of one of the needle bytes in the given + /// haystack. If no such occurrence exists, then `None` is returned. + /// + /// The occurrence is reported as an offset into `haystack`. Its maximum + /// value is `haystack.len() - 1`. + #[inline] + pub fn rfind(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: `rfind_raw` guarantees that if a pointer is returned, it + // falls within the bounds of the start and end pointers. + unsafe { + generic::search_slice_with_raw(haystack, |s, e| { + self.rfind_raw(s, e) + }) + } + } + + /// Like `find`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn find_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::fwd_byte_by_byte(start, end, |b| { + b == self.0.needle1() + || b == self.0.needle2() + || b == self.0.needle3() + }); + } + // SAFETY: Building a `Three` means it's safe to call 'sse2' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + // + // Note that we could call `self.0.find_raw` directly here. But that + // means we'd have to annotate this routine with `target_feature`. + // Which is fine, because this routine is `unsafe` anyway and the + // `target_feature` obligation is met by virtue of building a `Three`. + // The real problem is that a routine with a `target_feature` + // annotation generally can't be inlined into caller code unless the + // caller code has the same target feature annotations. Which is maybe + // okay for SSE2, but we do the same thing for AVX2 where caller code + // probably usually doesn't have AVX2 enabled. That means that this + // routine can be inlined which will handle some of the short-haystack + // cases above without touching the architecture specific code. + self.find_raw_impl(start, end) + } + + /// Like `rfind`, but accepts and returns raw pointers. + /// + /// When a match is found, the pointer returned is guaranteed to be + /// `>= start` and `< end`. + /// + /// This routine is useful if you're already using raw pointers and would + /// like to avoid converting back to a slice before executing a search. + /// + /// # Safety + /// + /// * Both `start` and `end` must be valid for reads. + /// * Both `start` and `end` must point to an initialized value. + /// * Both `start` and `end` must point to the same allocated object and + /// must either be in bounds or at most one byte past the end of the + /// allocated object. + /// * Both `start` and `end` must be _derived from_ a pointer to the same + /// object. + /// * The distance between `start` and `end` must not overflow `isize`. + /// * The distance being in bounds must not rely on "wrapping around" the + /// address space. + /// + /// Note that callers may pass a pair of pointers such that `start >= end`. + /// In that case, `None` will always be returned. + #[inline] + pub unsafe fn rfind_raw( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + if start >= end { + return None; + } + if end.distance(start) < __m128i::BYTES { + // SAFETY: We require the caller to pass valid start/end pointers. + return generic::rev_byte_by_byte(start, end, |b| { + b == self.0.needle1() + || b == self.0.needle2() + || b == self.0.needle3() + }); + } + // SAFETY: Building a `Three` means it's safe to call 'sse2' routines. + // Also, we've checked that our haystack is big enough to run on the + // vector routine. Pointer validity is caller's responsibility. + // + // See note in forward routine above for why we don't just call + // `self.0.rfind_raw` directly here. + self.rfind_raw_impl(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::find_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `sse2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn find_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.find_raw(start, end) + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as [`Three::rfind_raw`], except the distance between `start` and + /// `end` must be at least the size of an SSE2 vector (in bytes). + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Three`, which can only be constructed + /// when it is safe to call `sse2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn rfind_raw_impl( + &self, + start: *const u8, + end: *const u8, + ) -> Option<*const u8> { + self.0.rfind_raw(start, end) + } + + /// Returns an iterator over all occurrences of the needle byte in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn iter<'a, 'h>(&'a self, haystack: &'h [u8]) -> ThreeIter<'a, 'h> { + ThreeIter { searcher: self, it: generic::Iter::new(haystack) } + } +} + +/// An iterator over all occurrences of three possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`Three::iter`] method. +/// +/// The lifetime parameters are as follows: +/// +/// * `'a` refers to the lifetime of the underlying [`Three`] searcher. +/// * `'h` refers to the lifetime of the haystack being searched. +#[derive(Clone, Debug)] +pub struct ThreeIter<'a, 'h> { + searcher: &'a Three, + it: generic::Iter<'h>, +} + +impl<'a, 'h> Iterator for ThreeIter<'a, 'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'find_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next(|s, e| self.searcher.find_raw(s, e)) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'a, 'h> DoubleEndedIterator for ThreeIter<'a, 'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: We rely on the generic iterator to provide valid start + // and end pointers, but we guarantee that any pointer returned by + // 'rfind_raw' falls within the bounds of the start and end pointer. + unsafe { self.it.next_back(|s, e| self.searcher.rfind_raw(s, e)) } + } +} + +impl<'a, 'h> core::iter::FusedIterator for ThreeIter<'a, 'h> {} + +#[cfg(test)] +mod tests { + use super::*; + + define_memchr_quickcheck!(super); + + #[test] + fn forward_one() { + crate::tests::memchr::Runner::new(1).forward_iter( + |haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_one() { + crate::tests::memchr::Runner::new(1).reverse_iter( + |haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn count_one() { + crate::tests::memchr::Runner::new(1).count_iter(|haystack, needles| { + Some(One::new(needles[0])?.iter(haystack).count()) + }) + } + + #[test] + fn forward_two() { + crate::tests::memchr::Runner::new(2).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2)?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_two() { + crate::tests::memchr::Runner::new(2).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(Two::new(n1, n2)?.iter(haystack).rev().collect()) + }, + ) + } + + #[test] + fn forward_three() { + crate::tests::memchr::Runner::new(3).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3)?.iter(haystack).collect()) + }, + ) + } + + #[test] + fn reverse_three() { + crate::tests::memchr::Runner::new(3).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(Three::new(n1, n2, n3)?.iter(haystack).rev().collect()) + }, + ) + } +} diff --git a/vendor/memchr/src/arch/x86_64/sse2/mod.rs b/vendor/memchr/src/arch/x86_64/sse2/mod.rs new file mode 100644 index 0000000..bcb8307 --- /dev/null +++ b/vendor/memchr/src/arch/x86_64/sse2/mod.rs @@ -0,0 +1,6 @@ +/*! +Algorithms for the `x86_64` target using 128-bit vectors via SSE2. +*/ + +pub mod memchr; +pub mod packedpair; diff --git a/vendor/memchr/src/arch/x86_64/sse2/packedpair.rs b/vendor/memchr/src/arch/x86_64/sse2/packedpair.rs new file mode 100644 index 0000000..c8b5b99 --- /dev/null +++ b/vendor/memchr/src/arch/x86_64/sse2/packedpair.rs @@ -0,0 +1,232 @@ +/*! +A 128-bit vector implementation of the "packed pair" SIMD algorithm. + +The "packed pair" algorithm is based on the [generic SIMD] algorithm. The main +difference is that it (by default) uses a background distribution of byte +frequencies to heuristically select the pair of bytes to search for. + +[generic SIMD]: http://0x80.pl/articles/simd-strfind.html#first-and-last +*/ + +use core::arch::x86_64::__m128i; + +use crate::arch::{all::packedpair::Pair, generic::packedpair}; + +/// A "packed pair" finder that uses 128-bit vector operations. +/// +/// This finder picks two bytes that it believes have high predictive power +/// for indicating an overall match of a needle. Depending on whether +/// `Finder::find` or `Finder::find_prefilter` is used, it reports offsets +/// where the needle matches or could match. In the prefilter case, candidates +/// are reported whenever the [`Pair`] of bytes given matches. +#[derive(Clone, Copy, Debug)] +pub struct Finder(packedpair::Finder<__m128i>); + +impl Finder { + /// Create a new pair searcher. The searcher returned can either report + /// exact matches of `needle` or act as a prefilter and report candidate + /// positions of `needle`. + /// + /// If SSE2 is unavailable in the current environment or if a [`Pair`] + /// could not be constructed from the needle given, then `None` is + /// returned. + #[inline] + pub fn new(needle: &[u8]) -> Option<Finder> { + Finder::with_pair(needle, Pair::new(needle)?) + } + + /// Create a new "packed pair" finder using the pair of bytes given. + /// + /// This constructor permits callers to control precisely which pair of + /// bytes is used as a predicate. + /// + /// If SSE2 is unavailable in the current environment, then `None` is + /// returned. + #[inline] + pub fn with_pair(needle: &[u8], pair: Pair) -> Option<Finder> { + if Finder::is_available() { + // SAFETY: we check that sse2 is available above. We are also + // guaranteed to have needle.len() > 1 because we have a valid + // Pair. + unsafe { Some(Finder::with_pair_impl(needle, pair)) } + } else { + None + } + } + + /// Create a new `Finder` specific to SSE2 vectors and routines. + /// + /// # Safety + /// + /// Same as the safety for `packedpair::Finder::new`, and callers must also + /// ensure that SSE2 is available. + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn with_pair_impl(needle: &[u8], pair: Pair) -> Finder { + let finder = packedpair::Finder::<__m128i>::new(needle, pair); + Finder(finder) + } + + /// Returns true when this implementation is available in the current + /// environment. + /// + /// When this is true, it is guaranteed that [`Finder::with_pair`] will + /// return a `Some` value. Similarly, when it is false, it is guaranteed + /// that `Finder::with_pair` will return a `None` value. Notice that this + /// does not guarantee that [`Finder::new`] will return a `Finder`. Namely, + /// even when `Finder::is_available` is true, it is not guaranteed that a + /// valid [`Pair`] can be found from the needle given. + /// + /// Note also that for the lifetime of a single program, if this returns + /// true then it will always return true. + #[inline] + pub fn is_available() -> bool { + #[cfg(not(target_feature = "sse2"))] + { + false + } + #[cfg(target_feature = "sse2")] + { + true + } + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + #[inline] + pub fn find(&self, haystack: &[u8], needle: &[u8]) -> Option<usize> { + // SAFETY: Building a `Finder` means it's safe to call 'sse2' routines. + unsafe { self.find_impl(haystack, needle) } + } + + /// Run this finder on the given haystack as a prefilter. + /// + /// If a candidate match is found, then an offset where the needle *could* + /// begin in the haystack is returned. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + #[inline] + pub fn find_prefilter(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: Building a `Finder` means it's safe to call 'sse2' routines. + unsafe { self.find_prefilter_impl(haystack) } + } + + /// Execute a search using SSE2 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Finder`, which can only be constructed + /// when it is safe to call `sse2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn find_impl( + &self, + haystack: &[u8], + needle: &[u8], + ) -> Option<usize> { + self.0.find(haystack, needle) + } + + /// Execute a prefilter search using SSE2 vectors and routines. + /// + /// # Panics + /// + /// When `haystack.len()` is less than [`Finder::min_haystack_len`]. + /// + /// # Safety + /// + /// (The target feature safety obligation is automatically fulfilled by + /// virtue of being a method on `Finder`, which can only be constructed + /// when it is safe to call `sse2` routines.) + #[target_feature(enable = "sse2")] + #[inline] + unsafe fn find_prefilter_impl(&self, haystack: &[u8]) -> Option<usize> { + self.0.find_prefilter(haystack) + } + + /// Returns the pair of offsets (into the needle) used to check as a + /// predicate before confirming whether a needle exists at a particular + /// position. + #[inline] + pub fn pair(&self) -> &Pair { + self.0.pair() + } + + /// Returns the minimum haystack length that this `Finder` can search. + /// + /// Using a haystack with length smaller than this in a search will result + /// in a panic. The reason for this restriction is that this finder is + /// meant to be a low-level component that is part of a larger substring + /// strategy. In that sense, it avoids trying to handle all cases and + /// instead only handles the cases that it can handle very well. + #[inline] + pub fn min_haystack_len(&self) -> usize { + self.0.min_haystack_len() + } +} + +#[cfg(test)] +mod tests { + use super::*; + + fn find(haystack: &[u8], needle: &[u8]) -> Option<Option<usize>> { + let f = Finder::new(needle)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find(haystack, needle)) + } + + define_substring_forward_quickcheck!(find); + + #[test] + fn forward_substring() { + crate::tests::substring::Runner::new().fwd(find).run() + } + + #[test] + fn forward_packedpair() { + fn find( + haystack: &[u8], + needle: &[u8], + index1: u8, + index2: u8, + ) -> Option<Option<usize>> { + let pair = Pair::with_indices(needle, index1, index2)?; + let f = Finder::with_pair(needle, pair)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find(haystack, needle)) + } + crate::tests::packedpair::Runner::new().fwd(find).run() + } + + #[test] + fn forward_packedpair_prefilter() { + fn find( + haystack: &[u8], + needle: &[u8], + index1: u8, + index2: u8, + ) -> Option<Option<usize>> { + let pair = Pair::with_indices(needle, index1, index2)?; + let f = Finder::with_pair(needle, pair)?; + if haystack.len() < f.min_haystack_len() { + return None; + } + Some(f.find_prefilter(haystack)) + } + crate::tests::packedpair::Runner::new().fwd(find).run() + } +} diff --git a/vendor/memchr/src/cow.rs b/vendor/memchr/src/cow.rs new file mode 100644 index 0000000..f291645 --- /dev/null +++ b/vendor/memchr/src/cow.rs @@ -0,0 +1,107 @@ +use core::ops; + +/// A specialized copy-on-write byte string. +/// +/// The purpose of this type is to permit usage of a "borrowed or owned +/// byte string" in a way that keeps std/no-std compatibility. That is, in +/// no-std/alloc mode, this type devolves into a simple &[u8] with no owned +/// variant available. We can't just use a plain Cow because Cow is not in +/// core. +#[derive(Clone, Debug)] +pub struct CowBytes<'a>(Imp<'a>); + +// N.B. We don't use alloc::borrow::Cow here since we can get away with a +// Box<[u8]> for our use case, which is 1/3 smaller than the Vec<u8> that +// a Cow<[u8]> would use. +#[cfg(feature = "alloc")] +#[derive(Clone, Debug)] +enum Imp<'a> { + Borrowed(&'a [u8]), + Owned(alloc::boxed::Box<[u8]>), +} + +#[cfg(not(feature = "alloc"))] +#[derive(Clone, Debug)] +struct Imp<'a>(&'a [u8]); + +impl<'a> ops::Deref for CowBytes<'a> { + type Target = [u8]; + + #[inline(always)] + fn deref(&self) -> &[u8] { + self.as_slice() + } +} + +impl<'a> CowBytes<'a> { + /// Create a new borrowed CowBytes. + #[inline(always)] + pub(crate) fn new<B: ?Sized + AsRef<[u8]>>(bytes: &'a B) -> CowBytes<'a> { + CowBytes(Imp::new(bytes.as_ref())) + } + + /// Create a new owned CowBytes. + #[cfg(feature = "alloc")] + #[inline(always)] + fn new_owned(bytes: alloc::boxed::Box<[u8]>) -> CowBytes<'static> { + CowBytes(Imp::Owned(bytes)) + } + + /// Return a borrowed byte string, regardless of whether this is an owned + /// or borrowed byte string internally. + #[inline(always)] + pub(crate) fn as_slice(&self) -> &[u8] { + self.0.as_slice() + } + + /// Return an owned version of this copy-on-write byte string. + /// + /// If this is already an owned byte string internally, then this is a + /// no-op. Otherwise, the internal byte string is copied. + #[cfg(feature = "alloc")] + #[inline(always)] + pub(crate) fn into_owned(self) -> CowBytes<'static> { + match self.0 { + Imp::Borrowed(b) => { + CowBytes::new_owned(alloc::boxed::Box::from(b)) + } + Imp::Owned(b) => CowBytes::new_owned(b), + } + } +} + +impl<'a> Imp<'a> { + #[inline(always)] + pub fn new(bytes: &'a [u8]) -> Imp<'a> { + #[cfg(feature = "alloc")] + { + Imp::Borrowed(bytes) + } + #[cfg(not(feature = "alloc"))] + { + Imp(bytes) + } + } + + #[cfg(feature = "alloc")] + #[inline(always)] + pub fn as_slice(&self) -> &[u8] { + #[cfg(feature = "alloc")] + { + match self { + Imp::Owned(ref x) => x, + Imp::Borrowed(x) => x, + } + } + #[cfg(not(feature = "alloc"))] + { + self.0 + } + } + + #[cfg(not(feature = "alloc"))] + #[inline(always)] + pub fn as_slice(&self) -> &[u8] { + self.0 + } +} diff --git a/vendor/memchr/src/ext.rs b/vendor/memchr/src/ext.rs new file mode 100644 index 0000000..1bb21dd --- /dev/null +++ b/vendor/memchr/src/ext.rs @@ -0,0 +1,52 @@ +/// A trait for adding some helper routines to pointers. +pub(crate) trait Pointer { + /// Returns the distance, in units of `T`, between `self` and `origin`. + /// + /// # Safety + /// + /// Same as `ptr::offset_from` in addition to `self >= origin`. + unsafe fn distance(self, origin: Self) -> usize; + + /// Casts this pointer to `usize`. + /// + /// Callers should not convert the `usize` back to a pointer if at all + /// possible. (And if you believe it's necessary, open an issue to discuss + /// why. Otherwise, it has the potential to violate pointer provenance.) + /// The purpose of this function is just to be able to do arithmetic, i.e., + /// computing offsets or alignments. + fn as_usize(self) -> usize; +} + +impl<T> Pointer for *const T { + unsafe fn distance(self, origin: *const T) -> usize { + // TODO: Replace with `ptr::sub_ptr` once stabilized. + usize::try_from(self.offset_from(origin)).unwrap_unchecked() + } + + fn as_usize(self) -> usize { + self as usize + } +} + +impl<T> Pointer for *mut T { + unsafe fn distance(self, origin: *mut T) -> usize { + (self as *const T).distance(origin as *const T) + } + + fn as_usize(self) -> usize { + (self as *const T).as_usize() + } +} + +/// A trait for adding some helper routines to raw bytes. +pub(crate) trait Byte { + /// Converts this byte to a `char` if it's ASCII. Otherwise panics. + fn to_char(self) -> char; +} + +impl Byte for u8 { + fn to_char(self) -> char { + assert!(self.is_ascii()); + char::from(self) + } +} diff --git a/vendor/memchr/src/lib.rs b/vendor/memchr/src/lib.rs new file mode 100644 index 0000000..de366fb --- /dev/null +++ b/vendor/memchr/src/lib.rs @@ -0,0 +1,221 @@ +/*! +This library provides heavily optimized routines for string search primitives. + +# Overview + +This section gives a brief high level overview of what this crate offers. + +* The top-level module provides routines for searching for 1, 2 or 3 bytes + in the forward or reverse direction. When searching for more than one byte, + positions are considered a match if the byte at that position matches any + of the bytes. +* The [`memmem`] sub-module provides forward and reverse substring search + routines. + +In all such cases, routines operate on `&[u8]` without regard to encoding. This +is exactly what you want when searching either UTF-8 or arbitrary bytes. + +# Example: using `memchr` + +This example shows how to use `memchr` to find the first occurrence of `z` in +a haystack: + +``` +use memchr::memchr; + +let haystack = b"foo bar baz quuz"; +assert_eq!(Some(10), memchr(b'z', haystack)); +``` + +# Example: matching one of three possible bytes + +This examples shows how to use `memrchr3` to find occurrences of `a`, `b` or +`c`, starting at the end of the haystack. + +``` +use memchr::memchr3_iter; + +let haystack = b"xyzaxyzbxyzc"; + +let mut it = memchr3_iter(b'a', b'b', b'c', haystack).rev(); +assert_eq!(Some(11), it.next()); +assert_eq!(Some(7), it.next()); +assert_eq!(Some(3), it.next()); +assert_eq!(None, it.next()); +``` + +# Example: iterating over substring matches + +This example shows how to use the [`memmem`] sub-module to find occurrences of +a substring in a haystack. + +``` +use memchr::memmem; + +let haystack = b"foo bar foo baz foo"; + +let mut it = memmem::find_iter(haystack, "foo"); +assert_eq!(Some(0), it.next()); +assert_eq!(Some(8), it.next()); +assert_eq!(Some(16), it.next()); +assert_eq!(None, it.next()); +``` + +# Example: repeating a search for the same needle + +It may be possible for the overhead of constructing a substring searcher to be +measurable in some workloads. In cases where the same needle is used to search +many haystacks, it is possible to do construction once and thus to avoid it for +subsequent searches. This can be done with a [`memmem::Finder`]: + +``` +use memchr::memmem; + +let finder = memmem::Finder::new("foo"); + +assert_eq!(Some(4), finder.find(b"baz foo quux")); +assert_eq!(None, finder.find(b"quux baz bar")); +``` + +# Why use this crate? + +At first glance, the APIs provided by this crate might seem weird. Why provide +a dedicated routine like `memchr` for something that could be implemented +clearly and trivially in one line: + +``` +fn memchr(needle: u8, haystack: &[u8]) -> Option<usize> { + haystack.iter().position(|&b| b == needle) +} +``` + +Or similarly, why does this crate provide substring search routines when Rust's +core library already provides them? + +``` +fn search(haystack: &str, needle: &str) -> Option<usize> { + haystack.find(needle) +} +``` + +The primary reason for both of them to exist is performance. When it comes to +performance, at a high level at least, there are two primary ways to look at +it: + +* **Throughput**: For this, think about it as, "given some very large haystack + and a byte that never occurs in that haystack, how long does it take to + search through it and determine that it, in fact, does not occur?" +* **Latency**: For this, think about it as, "given a tiny haystack---just a + few bytes---how long does it take to determine if a byte is in it?" + +The `memchr` routine in this crate has _slightly_ worse latency than the +solution presented above, however, its throughput can easily be over an +order of magnitude faster. This is a good general purpose trade off to make. +You rarely lose, but often gain big. + +**NOTE:** The name `memchr` comes from the corresponding routine in `libc`. A +key advantage of using this library is that its performance is not tied to its +quality of implementation in the `libc` you happen to be using, which can vary +greatly from platform to platform. + +But what about substring search? This one is a bit more complicated. The +primary reason for its existence is still indeed performance, but it's also +useful because Rust's core library doesn't actually expose any substring +search routine on arbitrary bytes. The only substring search routine that +exists works exclusively on valid UTF-8. + +So if you have valid UTF-8, is there a reason to use this over the standard +library substring search routine? Yes. This routine is faster on almost every +metric, including latency. The natural question then, is why isn't this +implementation in the standard library, even if only for searching on UTF-8? +The reason is that the implementation details for using SIMD in the standard +library haven't quite been worked out yet. + +**NOTE:** Currently, only `x86_64`, `wasm32` and `aarch64` targets have vector +accelerated implementations of `memchr` (and friends) and `memmem`. + +# Crate features + +* **std** - When enabled (the default), this will permit features specific to +the standard library. Currently, the only thing used from the standard library +is runtime SIMD CPU feature detection. This means that this feature must be +enabled to get AVX2 accelerated routines on `x86_64` targets without enabling +the `avx2` feature at compile time, for example. When `std` is not enabled, +this crate will still attempt to use SSE2 accelerated routines on `x86_64`. It +will also use AVX2 accelerated routines when the `avx2` feature is enabled at +compile time. In general, enable this feature if you can. +* **alloc** - When enabled (the default), APIs in this crate requiring some +kind of allocation will become available. For example, the +[`memmem::Finder::into_owned`](crate::memmem::Finder::into_owned) API and the +[`arch::all::shiftor`](crate::arch::all::shiftor) substring search +implementation. Otherwise, this crate is designed from the ground up to be +usable in core-only contexts, so the `alloc` feature doesn't add much +currently. Notably, disabling `std` but enabling `alloc` will **not** result +in the use of AVX2 on `x86_64` targets unless the `avx2` feature is enabled +at compile time. (With `std` enabled, AVX2 can be used even without the `avx2` +feature enabled at compile time by way of runtime CPU feature detection.) +* **logging** - When enabled (disabled by default), the `log` crate is used +to emit log messages about what kinds of `memchr` and `memmem` algorithms +are used. Namely, both `memchr` and `memmem` have a number of different +implementation choices depending on the target and CPU, and the log messages +can help show what specific implementations are being used. Generally, this is +useful for debugging performance issues. +* **libc** - **DEPRECATED**. Previously, this enabled the use of the target's +`memchr` function from whatever `libc` was linked into the program. This +feature is now a no-op because this crate's implementation of `memchr` should +now be sufficiently fast on a number of platforms that `libc` should no longer +be needed. (This feature is somewhat of a holdover from this crate's origins. +Originally, this crate was literally just a safe wrapper function around the +`memchr` function from `libc`.) +*/ + +#![deny(missing_docs)] +#![no_std] +// It's just not worth trying to squash all dead code warnings. Pretty +// unfortunate IMO. Not really sure how to fix this other than to either +// live with it or sprinkle a whole mess of `cfg` annotations everywhere. +#![cfg_attr( + not(any( + all(target_arch = "x86_64", target_feature = "sse2"), + target_arch = "wasm32", + target_arch = "aarch64", + )), + allow(dead_code) +)] +// Same deal for miri. +#![cfg_attr(miri, allow(dead_code, unused_macros))] + +// Supporting 8-bit (or others) would be fine. If you need it, please submit a +// bug report at https://github.com/BurntSushi/memchr +#[cfg(not(any( + target_pointer_width = "16", + target_pointer_width = "32", + target_pointer_width = "64" +)))] +compile_error!("memchr currently not supported on non-{16,32,64}"); + +#[cfg(any(test, feature = "std"))] +extern crate std; + +#[cfg(any(test, feature = "alloc"))] +extern crate alloc; + +pub use crate::memchr::{ + memchr, memchr2, memchr2_iter, memchr3, memchr3_iter, memchr_iter, + memrchr, memrchr2, memrchr2_iter, memrchr3, memrchr3_iter, memrchr_iter, + Memchr, Memchr2, Memchr3, +}; + +#[macro_use] +mod macros; + +#[cfg(test)] +#[macro_use] +mod tests; + +pub mod arch; +mod cow; +mod ext; +mod memchr; +pub mod memmem; +mod vector; diff --git a/vendor/memchr/src/macros.rs b/vendor/memchr/src/macros.rs new file mode 100644 index 0000000..31b4ca3 --- /dev/null +++ b/vendor/memchr/src/macros.rs @@ -0,0 +1,20 @@ +// Some feature combinations result in some of these macros never being used. +// Which is fine. Just squash the warnings. +#![allow(unused_macros)] + +macro_rules! log { + ($($tt:tt)*) => { + #[cfg(feature = "logging")] + { + $($tt)* + } + } +} + +macro_rules! debug { + ($($tt:tt)*) => { log!(log::debug!($($tt)*)) } +} + +macro_rules! trace { + ($($tt:tt)*) => { log!(log::trace!($($tt)*)) } +} diff --git a/vendor/memchr/src/memchr.rs b/vendor/memchr/src/memchr.rs new file mode 100644 index 0000000..68adb9a --- /dev/null +++ b/vendor/memchr/src/memchr.rs @@ -0,0 +1,903 @@ +use core::iter::Rev; + +use crate::arch::generic::memchr as generic; + +/// Search for the first occurrence of a byte in a slice. +/// +/// This returns the index corresponding to the first occurrence of `needle` in +/// `haystack`, or `None` if one is not found. If an index is returned, it is +/// guaranteed to be less than `haystack.len()`. +/// +/// While this is semantically the same as something like +/// `haystack.iter().position(|&b| b == needle)`, this routine will attempt to +/// use highly optimized vector operations that can be an order of magnitude +/// faster (or more). +/// +/// # Example +/// +/// This shows how to find the first position of a byte in a byte string. +/// +/// ``` +/// use memchr::memchr; +/// +/// let haystack = b"the quick brown fox"; +/// assert_eq!(memchr(b'k', haystack), Some(8)); +/// ``` +#[inline] +pub fn memchr(needle: u8, haystack: &[u8]) -> Option<usize> { + // SAFETY: memchr_raw, when a match is found, always returns a valid + // pointer between start and end. + unsafe { + generic::search_slice_with_raw(haystack, |start, end| { + memchr_raw(needle, start, end) + }) + } +} + +/// Search for the last occurrence of a byte in a slice. +/// +/// This returns the index corresponding to the last occurrence of `needle` in +/// `haystack`, or `None` if one is not found. If an index is returned, it is +/// guaranteed to be less than `haystack.len()`. +/// +/// While this is semantically the same as something like +/// `haystack.iter().rposition(|&b| b == needle)`, this routine will attempt to +/// use highly optimized vector operations that can be an order of magnitude +/// faster (or more). +/// +/// # Example +/// +/// This shows how to find the last position of a byte in a byte string. +/// +/// ``` +/// use memchr::memrchr; +/// +/// let haystack = b"the quick brown fox"; +/// assert_eq!(memrchr(b'o', haystack), Some(17)); +/// ``` +#[inline] +pub fn memrchr(needle: u8, haystack: &[u8]) -> Option<usize> { + // SAFETY: memrchr_raw, when a match is found, always returns a valid + // pointer between start and end. + unsafe { + generic::search_slice_with_raw(haystack, |start, end| { + memrchr_raw(needle, start, end) + }) + } +} + +/// Search for the first occurrence of two possible bytes in a haystack. +/// +/// This returns the index corresponding to the first occurrence of one of the +/// needle bytes in `haystack`, or `None` if one is not found. If an index is +/// returned, it is guaranteed to be less than `haystack.len()`. +/// +/// While this is semantically the same as something like +/// `haystack.iter().position(|&b| b == needle1 || b == needle2)`, this routine +/// will attempt to use highly optimized vector operations that can be an order +/// of magnitude faster (or more). +/// +/// # Example +/// +/// This shows how to find the first position of one of two possible bytes in a +/// haystack. +/// +/// ``` +/// use memchr::memchr2; +/// +/// let haystack = b"the quick brown fox"; +/// assert_eq!(memchr2(b'k', b'q', haystack), Some(4)); +/// ``` +#[inline] +pub fn memchr2(needle1: u8, needle2: u8, haystack: &[u8]) -> Option<usize> { + // SAFETY: memchr2_raw, when a match is found, always returns a valid + // pointer between start and end. + unsafe { + generic::search_slice_with_raw(haystack, |start, end| { + memchr2_raw(needle1, needle2, start, end) + }) + } +} + +/// Search for the last occurrence of two possible bytes in a haystack. +/// +/// This returns the index corresponding to the last occurrence of one of the +/// needle bytes in `haystack`, or `None` if one is not found. If an index is +/// returned, it is guaranteed to be less than `haystack.len()`. +/// +/// While this is semantically the same as something like +/// `haystack.iter().rposition(|&b| b == needle1 || b == needle2)`, this +/// routine will attempt to use highly optimized vector operations that can be +/// an order of magnitude faster (or more). +/// +/// # Example +/// +/// This shows how to find the last position of one of two possible bytes in a +/// haystack. +/// +/// ``` +/// use memchr::memrchr2; +/// +/// let haystack = b"the quick brown fox"; +/// assert_eq!(memrchr2(b'k', b'o', haystack), Some(17)); +/// ``` +#[inline] +pub fn memrchr2(needle1: u8, needle2: u8, haystack: &[u8]) -> Option<usize> { + // SAFETY: memrchr2_raw, when a match is found, always returns a valid + // pointer between start and end. + unsafe { + generic::search_slice_with_raw(haystack, |start, end| { + memrchr2_raw(needle1, needle2, start, end) + }) + } +} + +/// Search for the first occurrence of three possible bytes in a haystack. +/// +/// This returns the index corresponding to the first occurrence of one of the +/// needle bytes in `haystack`, or `None` if one is not found. If an index is +/// returned, it is guaranteed to be less than `haystack.len()`. +/// +/// While this is semantically the same as something like +/// `haystack.iter().position(|&b| b == needle1 || b == needle2 || b == needle3)`, +/// this routine will attempt to use highly optimized vector operations that +/// can be an order of magnitude faster (or more). +/// +/// # Example +/// +/// This shows how to find the first position of one of three possible bytes in +/// a haystack. +/// +/// ``` +/// use memchr::memchr3; +/// +/// let haystack = b"the quick brown fox"; +/// assert_eq!(memchr3(b'k', b'q', b'u', haystack), Some(4)); +/// ``` +#[inline] +pub fn memchr3( + needle1: u8, + needle2: u8, + needle3: u8, + haystack: &[u8], +) -> Option<usize> { + // SAFETY: memchr3_raw, when a match is found, always returns a valid + // pointer between start and end. + unsafe { + generic::search_slice_with_raw(haystack, |start, end| { + memchr3_raw(needle1, needle2, needle3, start, end) + }) + } +} + +/// Search for the last occurrence of three possible bytes in a haystack. +/// +/// This returns the index corresponding to the last occurrence of one of the +/// needle bytes in `haystack`, or `None` if one is not found. If an index is +/// returned, it is guaranteed to be less than `haystack.len()`. +/// +/// While this is semantically the same as something like +/// `haystack.iter().rposition(|&b| b == needle1 || b == needle2 || b == needle3)`, +/// this routine will attempt to use highly optimized vector operations that +/// can be an order of magnitude faster (or more). +/// +/// # Example +/// +/// This shows how to find the last position of one of three possible bytes in +/// a haystack. +/// +/// ``` +/// use memchr::memrchr3; +/// +/// let haystack = b"the quick brown fox"; +/// assert_eq!(memrchr3(b'k', b'o', b'n', haystack), Some(17)); +/// ``` +#[inline] +pub fn memrchr3( + needle1: u8, + needle2: u8, + needle3: u8, + haystack: &[u8], +) -> Option<usize> { + // SAFETY: memrchr3_raw, when a match is found, always returns a valid + // pointer between start and end. + unsafe { + generic::search_slice_with_raw(haystack, |start, end| { + memrchr3_raw(needle1, needle2, needle3, start, end) + }) + } +} + +/// Returns an iterator over all occurrences of the needle in a haystack. +/// +/// The iterator returned implements `DoubleEndedIterator`. This means it +/// can also be used to find occurrences in reverse order. +#[inline] +pub fn memchr_iter<'h>(needle: u8, haystack: &'h [u8]) -> Memchr<'h> { + Memchr::new(needle, haystack) +} + +/// Returns an iterator over all occurrences of the needle in a haystack, in +/// reverse. +#[inline] +pub fn memrchr_iter(needle: u8, haystack: &[u8]) -> Rev<Memchr<'_>> { + Memchr::new(needle, haystack).rev() +} + +/// Returns an iterator over all occurrences of the needles in a haystack. +/// +/// The iterator returned implements `DoubleEndedIterator`. This means it +/// can also be used to find occurrences in reverse order. +#[inline] +pub fn memchr2_iter<'h>( + needle1: u8, + needle2: u8, + haystack: &'h [u8], +) -> Memchr2<'h> { + Memchr2::new(needle1, needle2, haystack) +} + +/// Returns an iterator over all occurrences of the needles in a haystack, in +/// reverse. +#[inline] +pub fn memrchr2_iter( + needle1: u8, + needle2: u8, + haystack: &[u8], +) -> Rev<Memchr2<'_>> { + Memchr2::new(needle1, needle2, haystack).rev() +} + +/// Returns an iterator over all occurrences of the needles in a haystack. +/// +/// The iterator returned implements `DoubleEndedIterator`. This means it +/// can also be used to find occurrences in reverse order. +#[inline] +pub fn memchr3_iter<'h>( + needle1: u8, + needle2: u8, + needle3: u8, + haystack: &'h [u8], +) -> Memchr3<'h> { + Memchr3::new(needle1, needle2, needle3, haystack) +} + +/// Returns an iterator over all occurrences of the needles in a haystack, in +/// reverse. +#[inline] +pub fn memrchr3_iter( + needle1: u8, + needle2: u8, + needle3: u8, + haystack: &[u8], +) -> Rev<Memchr3<'_>> { + Memchr3::new(needle1, needle2, needle3, haystack).rev() +} + +/// An iterator over all occurrences of a single byte in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`memchr_iter`] or `[memrchr_iter`] +/// functions. It can also be created with the [`Memchr::new`] method. +/// +/// The lifetime parameter `'h` refers to the lifetime of the haystack being +/// searched. +#[derive(Clone, Debug)] +pub struct Memchr<'h> { + needle1: u8, + it: crate::arch::generic::memchr::Iter<'h>, +} + +impl<'h> Memchr<'h> { + /// Returns an iterator over all occurrences of the needle byte in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn new(needle1: u8, haystack: &'h [u8]) -> Memchr<'h> { + Memchr { + needle1, + it: crate::arch::generic::memchr::Iter::new(haystack), + } + } +} + +impl<'h> Iterator for Memchr<'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: All of our implementations of memchr ensure that any + // pointers returns will fall within the start and end bounds, and this + // upholds the safety contract of `self.it.next`. + unsafe { + // NOTE: I attempted to define an enum of previously created + // searchers and then switch on those here instead of just + // calling `memchr_raw` (or `One::new(..).find_raw(..)`). But + // that turned out to have a fair bit of extra overhead when + // searching very small haystacks. + self.it.next(|s, e| memchr_raw(self.needle1, s, e)) + } + } + + #[inline] + fn count(self) -> usize { + self.it.count(|s, e| { + // SAFETY: We rely on our generic iterator to return valid start + // and end pointers. + unsafe { count_raw(self.needle1, s, e) } + }) + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'h> DoubleEndedIterator for Memchr<'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: All of our implementations of memchr ensure that any + // pointers returns will fall within the start and end bounds, and this + // upholds the safety contract of `self.it.next_back`. + unsafe { self.it.next_back(|s, e| memrchr_raw(self.needle1, s, e)) } + } +} + +impl<'h> core::iter::FusedIterator for Memchr<'h> {} + +/// An iterator over all occurrences of two possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`memchr2_iter`] or `[memrchr2_iter`] +/// functions. It can also be created with the [`Memchr2::new`] method. +/// +/// The lifetime parameter `'h` refers to the lifetime of the haystack being +/// searched. +#[derive(Clone, Debug)] +pub struct Memchr2<'h> { + needle1: u8, + needle2: u8, + it: crate::arch::generic::memchr::Iter<'h>, +} + +impl<'h> Memchr2<'h> { + /// Returns an iterator over all occurrences of the needle bytes in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn new(needle1: u8, needle2: u8, haystack: &'h [u8]) -> Memchr2<'h> { + Memchr2 { + needle1, + needle2, + it: crate::arch::generic::memchr::Iter::new(haystack), + } + } +} + +impl<'h> Iterator for Memchr2<'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: All of our implementations of memchr ensure that any + // pointers returns will fall within the start and end bounds, and this + // upholds the safety contract of `self.it.next`. + unsafe { + self.it.next(|s, e| memchr2_raw(self.needle1, self.needle2, s, e)) + } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'h> DoubleEndedIterator for Memchr2<'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: All of our implementations of memchr ensure that any + // pointers returns will fall within the start and end bounds, and this + // upholds the safety contract of `self.it.next_back`. + unsafe { + self.it.next_back(|s, e| { + memrchr2_raw(self.needle1, self.needle2, s, e) + }) + } + } +} + +impl<'h> core::iter::FusedIterator for Memchr2<'h> {} + +/// An iterator over all occurrences of three possible bytes in a haystack. +/// +/// This iterator implements `DoubleEndedIterator`, which means it can also be +/// used to find occurrences in reverse order. +/// +/// This iterator is created by the [`memchr2_iter`] or `[memrchr2_iter`] +/// functions. It can also be created with the [`Memchr3::new`] method. +/// +/// The lifetime parameter `'h` refers to the lifetime of the haystack being +/// searched. +#[derive(Clone, Debug)] +pub struct Memchr3<'h> { + needle1: u8, + needle2: u8, + needle3: u8, + it: crate::arch::generic::memchr::Iter<'h>, +} + +impl<'h> Memchr3<'h> { + /// Returns an iterator over all occurrences of the needle bytes in the + /// given haystack. + /// + /// The iterator returned implements `DoubleEndedIterator`. This means it + /// can also be used to find occurrences in reverse order. + #[inline] + pub fn new( + needle1: u8, + needle2: u8, + needle3: u8, + haystack: &'h [u8], + ) -> Memchr3<'h> { + Memchr3 { + needle1, + needle2, + needle3, + it: crate::arch::generic::memchr::Iter::new(haystack), + } + } +} + +impl<'h> Iterator for Memchr3<'h> { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + // SAFETY: All of our implementations of memchr ensure that any + // pointers returns will fall within the start and end bounds, and this + // upholds the safety contract of `self.it.next`. + unsafe { + self.it.next(|s, e| { + memchr3_raw(self.needle1, self.needle2, self.needle3, s, e) + }) + } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.it.size_hint() + } +} + +impl<'h> DoubleEndedIterator for Memchr3<'h> { + #[inline] + fn next_back(&mut self) -> Option<usize> { + // SAFETY: All of our implementations of memchr ensure that any + // pointers returns will fall within the start and end bounds, and this + // upholds the safety contract of `self.it.next_back`. + unsafe { + self.it.next_back(|s, e| { + memrchr3_raw(self.needle1, self.needle2, self.needle3, s, e) + }) + } + } +} + +impl<'h> core::iter::FusedIterator for Memchr3<'h> {} + +/// memchr, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::find_raw`. +#[inline] +unsafe fn memchr_raw( + needle: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + #[cfg(target_arch = "x86_64")] + { + // x86_64 does CPU feature detection at runtime in order to use AVX2 + // instructions even when the `avx2` feature isn't enabled at compile + // time. This function also handles using a fallback if neither AVX2 + // nor SSE2 (unusual) are available. + crate::arch::x86_64::memchr::memchr_raw(needle, start, end) + } + #[cfg(target_arch = "wasm32")] + { + crate::arch::wasm32::memchr::memchr_raw(needle, start, end) + } + #[cfg(target_arch = "aarch64")] + { + crate::arch::aarch64::memchr::memchr_raw(needle, start, end) + } + #[cfg(not(any( + target_arch = "x86_64", + target_arch = "wasm32", + target_arch = "aarch64" + )))] + { + crate::arch::all::memchr::One::new(needle).find_raw(start, end) + } +} + +/// memrchr, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::rfind_raw`. +#[inline] +unsafe fn memrchr_raw( + needle: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + #[cfg(target_arch = "x86_64")] + { + crate::arch::x86_64::memchr::memrchr_raw(needle, start, end) + } + #[cfg(target_arch = "wasm32")] + { + crate::arch::wasm32::memchr::memrchr_raw(needle, start, end) + } + #[cfg(target_arch = "aarch64")] + { + crate::arch::aarch64::memchr::memrchr_raw(needle, start, end) + } + #[cfg(not(any( + target_arch = "x86_64", + target_arch = "wasm32", + target_arch = "aarch64" + )))] + { + crate::arch::all::memchr::One::new(needle).rfind_raw(start, end) + } +} + +/// memchr2, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Two::find_raw`. +#[inline] +unsafe fn memchr2_raw( + needle1: u8, + needle2: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + #[cfg(target_arch = "x86_64")] + { + crate::arch::x86_64::memchr::memchr2_raw(needle1, needle2, start, end) + } + #[cfg(target_arch = "wasm32")] + { + crate::arch::wasm32::memchr::memchr2_raw(needle1, needle2, start, end) + } + #[cfg(target_arch = "aarch64")] + { + crate::arch::aarch64::memchr::memchr2_raw(needle1, needle2, start, end) + } + #[cfg(not(any( + target_arch = "x86_64", + target_arch = "wasm32", + target_arch = "aarch64" + )))] + { + crate::arch::all::memchr::Two::new(needle1, needle2) + .find_raw(start, end) + } +} + +/// memrchr2, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Two::rfind_raw`. +#[inline] +unsafe fn memrchr2_raw( + needle1: u8, + needle2: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + #[cfg(target_arch = "x86_64")] + { + crate::arch::x86_64::memchr::memrchr2_raw(needle1, needle2, start, end) + } + #[cfg(target_arch = "wasm32")] + { + crate::arch::wasm32::memchr::memrchr2_raw(needle1, needle2, start, end) + } + #[cfg(target_arch = "aarch64")] + { + crate::arch::aarch64::memchr::memrchr2_raw( + needle1, needle2, start, end, + ) + } + #[cfg(not(any( + target_arch = "x86_64", + target_arch = "wasm32", + target_arch = "aarch64" + )))] + { + crate::arch::all::memchr::Two::new(needle1, needle2) + .rfind_raw(start, end) + } +} + +/// memchr3, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Three::find_raw`. +#[inline] +unsafe fn memchr3_raw( + needle1: u8, + needle2: u8, + needle3: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + #[cfg(target_arch = "x86_64")] + { + crate::arch::x86_64::memchr::memchr3_raw( + needle1, needle2, needle3, start, end, + ) + } + #[cfg(target_arch = "wasm32")] + { + crate::arch::wasm32::memchr::memchr3_raw( + needle1, needle2, needle3, start, end, + ) + } + #[cfg(target_arch = "aarch64")] + { + crate::arch::aarch64::memchr::memchr3_raw( + needle1, needle2, needle3, start, end, + ) + } + #[cfg(not(any( + target_arch = "x86_64", + target_arch = "wasm32", + target_arch = "aarch64" + )))] + { + crate::arch::all::memchr::Three::new(needle1, needle2, needle3) + .find_raw(start, end) + } +} + +/// memrchr3, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `Three::rfind_raw`. +#[inline] +unsafe fn memrchr3_raw( + needle1: u8, + needle2: u8, + needle3: u8, + start: *const u8, + end: *const u8, +) -> Option<*const u8> { + #[cfg(target_arch = "x86_64")] + { + crate::arch::x86_64::memchr::memrchr3_raw( + needle1, needle2, needle3, start, end, + ) + } + #[cfg(target_arch = "wasm32")] + { + crate::arch::wasm32::memchr::memrchr3_raw( + needle1, needle2, needle3, start, end, + ) + } + #[cfg(target_arch = "aarch64")] + { + crate::arch::aarch64::memchr::memrchr3_raw( + needle1, needle2, needle3, start, end, + ) + } + #[cfg(not(any( + target_arch = "x86_64", + target_arch = "wasm32", + target_arch = "aarch64" + )))] + { + crate::arch::all::memchr::Three::new(needle1, needle2, needle3) + .rfind_raw(start, end) + } +} + +/// Count all matching bytes, but using raw pointers to represent the haystack. +/// +/// # Safety +/// +/// Pointers must be valid. See `One::count_raw`. +#[inline] +unsafe fn count_raw(needle: u8, start: *const u8, end: *const u8) -> usize { + #[cfg(target_arch = "x86_64")] + { + crate::arch::x86_64::memchr::count_raw(needle, start, end) + } + #[cfg(target_arch = "wasm32")] + { + crate::arch::wasm32::memchr::count_raw(needle, start, end) + } + #[cfg(target_arch = "aarch64")] + { + crate::arch::aarch64::memchr::count_raw(needle, start, end) + } + #[cfg(not(any( + target_arch = "x86_64", + target_arch = "wasm32", + target_arch = "aarch64" + )))] + { + crate::arch::all::memchr::One::new(needle).count_raw(start, end) + } +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn forward1_iter() { + crate::tests::memchr::Runner::new(1).forward_iter( + |haystack, needles| { + Some(memchr_iter(needles[0], haystack).collect()) + }, + ) + } + + #[test] + fn forward1_oneshot() { + crate::tests::memchr::Runner::new(1).forward_oneshot( + |haystack, needles| Some(memchr(needles[0], haystack)), + ) + } + + #[test] + fn reverse1_iter() { + crate::tests::memchr::Runner::new(1).reverse_iter( + |haystack, needles| { + Some(memrchr_iter(needles[0], haystack).collect()) + }, + ) + } + + #[test] + fn reverse1_oneshot() { + crate::tests::memchr::Runner::new(1).reverse_oneshot( + |haystack, needles| Some(memrchr(needles[0], haystack)), + ) + } + + #[test] + fn count1_iter() { + crate::tests::memchr::Runner::new(1).count_iter(|haystack, needles| { + Some(memchr_iter(needles[0], haystack).count()) + }) + } + + #[test] + fn forward2_iter() { + crate::tests::memchr::Runner::new(2).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(memchr2_iter(n1, n2, haystack).collect()) + }, + ) + } + + #[test] + fn forward2_oneshot() { + crate::tests::memchr::Runner::new(2).forward_oneshot( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(memchr2(n1, n2, haystack)) + }, + ) + } + + #[test] + fn reverse2_iter() { + crate::tests::memchr::Runner::new(2).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(memrchr2_iter(n1, n2, haystack).collect()) + }, + ) + } + + #[test] + fn reverse2_oneshot() { + crate::tests::memchr::Runner::new(2).reverse_oneshot( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + Some(memrchr2(n1, n2, haystack)) + }, + ) + } + + #[test] + fn forward3_iter() { + crate::tests::memchr::Runner::new(3).forward_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(memchr3_iter(n1, n2, n3, haystack).collect()) + }, + ) + } + + #[test] + fn forward3_oneshot() { + crate::tests::memchr::Runner::new(3).forward_oneshot( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(memchr3(n1, n2, n3, haystack)) + }, + ) + } + + #[test] + fn reverse3_iter() { + crate::tests::memchr::Runner::new(3).reverse_iter( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(memrchr3_iter(n1, n2, n3, haystack).collect()) + }, + ) + } + + #[test] + fn reverse3_oneshot() { + crate::tests::memchr::Runner::new(3).reverse_oneshot( + |haystack, needles| { + let n1 = needles.get(0).copied()?; + let n2 = needles.get(1).copied()?; + let n3 = needles.get(2).copied()?; + Some(memrchr3(n1, n2, n3, haystack)) + }, + ) + } + + // Prior to memchr 2.6, the memchr iterators both implemented Send and + // Sync. But in memchr 2.6, the iterator changed to use raw pointers + // internally and I didn't add explicit Send/Sync impls. This ended up + // regressing the API. This test ensures we don't do that again. + // + // See: https://github.com/BurntSushi/memchr/issues/133 + #[test] + fn sync_regression() { + use core::panic::{RefUnwindSafe, UnwindSafe}; + + fn assert_send_sync<T: Send + Sync + UnwindSafe + RefUnwindSafe>() {} + assert_send_sync::<Memchr>(); + assert_send_sync::<Memchr2>(); + assert_send_sync::<Memchr3>() + } +} diff --git a/vendor/memchr/src/memmem/mod.rs b/vendor/memchr/src/memmem/mod.rs new file mode 100644 index 0000000..4f04943 --- /dev/null +++ b/vendor/memchr/src/memmem/mod.rs @@ -0,0 +1,737 @@ +/*! +This module provides forward and reverse substring search routines. + +Unlike the standard library's substring search routines, these work on +arbitrary bytes. For all non-empty needles, these routines will report exactly +the same values as the corresponding routines in the standard library. For +the empty needle, the standard library reports matches only at valid UTF-8 +boundaries, where as these routines will report matches at every position. + +Other than being able to work on arbitrary bytes, the primary reason to prefer +these routines over the standard library routines is that these will generally +be faster. In some cases, significantly so. + +# Example: iterating over substring matches + +This example shows how to use [`find_iter`] to find occurrences of a substring +in a haystack. + +``` +use memchr::memmem; + +let haystack = b"foo bar foo baz foo"; + +let mut it = memmem::find_iter(haystack, "foo"); +assert_eq!(Some(0), it.next()); +assert_eq!(Some(8), it.next()); +assert_eq!(Some(16), it.next()); +assert_eq!(None, it.next()); +``` + +# Example: iterating over substring matches in reverse + +This example shows how to use [`rfind_iter`] to find occurrences of a substring +in a haystack starting from the end of the haystack. + +**NOTE:** This module does not implement double ended iterators, so reverse +searches aren't done by calling `rev` on a forward iterator. + +``` +use memchr::memmem; + +let haystack = b"foo bar foo baz foo"; + +let mut it = memmem::rfind_iter(haystack, "foo"); +assert_eq!(Some(16), it.next()); +assert_eq!(Some(8), it.next()); +assert_eq!(Some(0), it.next()); +assert_eq!(None, it.next()); +``` + +# Example: repeating a search for the same needle + +It may be possible for the overhead of constructing a substring searcher to be +measurable in some workloads. In cases where the same needle is used to search +many haystacks, it is possible to do construction once and thus to avoid it for +subsequent searches. This can be done with a [`Finder`] (or a [`FinderRev`] for +reverse searches). + +``` +use memchr::memmem; + +let finder = memmem::Finder::new("foo"); + +assert_eq!(Some(4), finder.find(b"baz foo quux")); +assert_eq!(None, finder.find(b"quux baz bar")); +``` +*/ + +pub use crate::memmem::searcher::PrefilterConfig as Prefilter; + +// This is exported here for use in the crate::arch::all::twoway +// implementation. This is essentially an abstraction breaker. Namely, the +// public API of twoway doesn't support providing a prefilter, but its crate +// internal API does. The main reason for this is that I didn't want to do the +// API design required to support it without a concrete use case. +pub(crate) use crate::memmem::searcher::Pre; + +use crate::{ + arch::all::{ + packedpair::{DefaultFrequencyRank, HeuristicFrequencyRank}, + rabinkarp, + }, + cow::CowBytes, + memmem::searcher::{PrefilterState, Searcher, SearcherRev}, +}; + +mod searcher; + +/// Returns an iterator over all non-overlapping occurrences of a substring in +/// a haystack. +/// +/// # Complexity +/// +/// This routine is guaranteed to have worst case linear time complexity +/// with respect to both the needle and the haystack. That is, this runs +/// in `O(needle.len() + haystack.len())` time. +/// +/// This routine is also guaranteed to have worst case constant space +/// complexity. +/// +/// # Examples +/// +/// Basic usage: +/// +/// ``` +/// use memchr::memmem; +/// +/// let haystack = b"foo bar foo baz foo"; +/// let mut it = memmem::find_iter(haystack, b"foo"); +/// assert_eq!(Some(0), it.next()); +/// assert_eq!(Some(8), it.next()); +/// assert_eq!(Some(16), it.next()); +/// assert_eq!(None, it.next()); +/// ``` +#[inline] +pub fn find_iter<'h, 'n, N: 'n + ?Sized + AsRef<[u8]>>( + haystack: &'h [u8], + needle: &'n N, +) -> FindIter<'h, 'n> { + FindIter::new(haystack, Finder::new(needle)) +} + +/// Returns a reverse iterator over all non-overlapping occurrences of a +/// substring in a haystack. +/// +/// # Complexity +/// +/// This routine is guaranteed to have worst case linear time complexity +/// with respect to both the needle and the haystack. That is, this runs +/// in `O(needle.len() + haystack.len())` time. +/// +/// This routine is also guaranteed to have worst case constant space +/// complexity. +/// +/// # Examples +/// +/// Basic usage: +/// +/// ``` +/// use memchr::memmem; +/// +/// let haystack = b"foo bar foo baz foo"; +/// let mut it = memmem::rfind_iter(haystack, b"foo"); +/// assert_eq!(Some(16), it.next()); +/// assert_eq!(Some(8), it.next()); +/// assert_eq!(Some(0), it.next()); +/// assert_eq!(None, it.next()); +/// ``` +#[inline] +pub fn rfind_iter<'h, 'n, N: 'n + ?Sized + AsRef<[u8]>>( + haystack: &'h [u8], + needle: &'n N, +) -> FindRevIter<'h, 'n> { + FindRevIter::new(haystack, FinderRev::new(needle)) +} + +/// Returns the index of the first occurrence of the given needle. +/// +/// Note that if you're are searching for the same needle in many different +/// small haystacks, it may be faster to initialize a [`Finder`] once, +/// and reuse it for each search. +/// +/// # Complexity +/// +/// This routine is guaranteed to have worst case linear time complexity +/// with respect to both the needle and the haystack. That is, this runs +/// in `O(needle.len() + haystack.len())` time. +/// +/// This routine is also guaranteed to have worst case constant space +/// complexity. +/// +/// # Examples +/// +/// Basic usage: +/// +/// ``` +/// use memchr::memmem; +/// +/// let haystack = b"foo bar baz"; +/// assert_eq!(Some(0), memmem::find(haystack, b"foo")); +/// assert_eq!(Some(4), memmem::find(haystack, b"bar")); +/// assert_eq!(None, memmem::find(haystack, b"quux")); +/// ``` +#[inline] +pub fn find(haystack: &[u8], needle: &[u8]) -> Option<usize> { + if haystack.len() < 64 { + rabinkarp::Finder::new(needle).find(haystack, needle) + } else { + Finder::new(needle).find(haystack) + } +} + +/// Returns the index of the last occurrence of the given needle. +/// +/// Note that if you're are searching for the same needle in many different +/// small haystacks, it may be faster to initialize a [`FinderRev`] once, +/// and reuse it for each search. +/// +/// # Complexity +/// +/// This routine is guaranteed to have worst case linear time complexity +/// with respect to both the needle and the haystack. That is, this runs +/// in `O(needle.len() + haystack.len())` time. +/// +/// This routine is also guaranteed to have worst case constant space +/// complexity. +/// +/// # Examples +/// +/// Basic usage: +/// +/// ``` +/// use memchr::memmem; +/// +/// let haystack = b"foo bar baz"; +/// assert_eq!(Some(0), memmem::rfind(haystack, b"foo")); +/// assert_eq!(Some(4), memmem::rfind(haystack, b"bar")); +/// assert_eq!(Some(8), memmem::rfind(haystack, b"ba")); +/// assert_eq!(None, memmem::rfind(haystack, b"quux")); +/// ``` +#[inline] +pub fn rfind(haystack: &[u8], needle: &[u8]) -> Option<usize> { + if haystack.len() < 64 { + rabinkarp::FinderRev::new(needle).rfind(haystack, needle) + } else { + FinderRev::new(needle).rfind(haystack) + } +} + +/// An iterator over non-overlapping substring matches. +/// +/// Matches are reported by the byte offset at which they begin. +/// +/// `'h` is the lifetime of the haystack while `'n` is the lifetime of the +/// needle. +#[derive(Debug, Clone)] +pub struct FindIter<'h, 'n> { + haystack: &'h [u8], + prestate: PrefilterState, + finder: Finder<'n>, + pos: usize, +} + +impl<'h, 'n> FindIter<'h, 'n> { + #[inline(always)] + pub(crate) fn new( + haystack: &'h [u8], + finder: Finder<'n>, + ) -> FindIter<'h, 'n> { + let prestate = PrefilterState::new(); + FindIter { haystack, prestate, finder, pos: 0 } + } + + /// Convert this iterator into its owned variant, such that it no longer + /// borrows the finder and needle. + /// + /// If this is already an owned iterator, then this is a no-op. Otherwise, + /// this copies the needle. + /// + /// This is only available when the `alloc` feature is enabled. + #[cfg(feature = "alloc")] + #[inline] + pub fn into_owned(self) -> FindIter<'h, 'static> { + FindIter { + haystack: self.haystack, + prestate: self.prestate, + finder: self.finder.into_owned(), + pos: self.pos, + } + } +} + +impl<'h, 'n> Iterator for FindIter<'h, 'n> { + type Item = usize; + + fn next(&mut self) -> Option<usize> { + let needle = self.finder.needle(); + let haystack = self.haystack.get(self.pos..)?; + let idx = + self.finder.searcher.find(&mut self.prestate, haystack, needle)?; + + let pos = self.pos + idx; + self.pos = pos + needle.len().max(1); + + Some(pos) + } + + fn size_hint(&self) -> (usize, Option<usize>) { + // The largest possible number of non-overlapping matches is the + // quotient of the haystack and the needle (or the length of the + // haystack, if the needle is empty) + match self.haystack.len().checked_sub(self.pos) { + None => (0, Some(0)), + Some(haystack_len) => match self.finder.needle().len() { + // Empty needles always succeed and match at every point + // (including the very end) + 0 => ( + haystack_len.saturating_add(1), + haystack_len.checked_add(1), + ), + needle_len => (0, Some(haystack_len / needle_len)), + }, + } + } +} + +/// An iterator over non-overlapping substring matches in reverse. +/// +/// Matches are reported by the byte offset at which they begin. +/// +/// `'h` is the lifetime of the haystack while `'n` is the lifetime of the +/// needle. +#[derive(Clone, Debug)] +pub struct FindRevIter<'h, 'n> { + haystack: &'h [u8], + finder: FinderRev<'n>, + /// When searching with an empty needle, this gets set to `None` after + /// we've yielded the last element at `0`. + pos: Option<usize>, +} + +impl<'h, 'n> FindRevIter<'h, 'n> { + #[inline(always)] + pub(crate) fn new( + haystack: &'h [u8], + finder: FinderRev<'n>, + ) -> FindRevIter<'h, 'n> { + let pos = Some(haystack.len()); + FindRevIter { haystack, finder, pos } + } + + /// Convert this iterator into its owned variant, such that it no longer + /// borrows the finder and needle. + /// + /// If this is already an owned iterator, then this is a no-op. Otherwise, + /// this copies the needle. + /// + /// This is only available when the `std` feature is enabled. + #[cfg(feature = "alloc")] + #[inline] + pub fn into_owned(self) -> FindRevIter<'h, 'static> { + FindRevIter { + haystack: self.haystack, + finder: self.finder.into_owned(), + pos: self.pos, + } + } +} + +impl<'h, 'n> Iterator for FindRevIter<'h, 'n> { + type Item = usize; + + fn next(&mut self) -> Option<usize> { + let pos = match self.pos { + None => return None, + Some(pos) => pos, + }; + let result = self.finder.rfind(&self.haystack[..pos]); + match result { + None => None, + Some(i) => { + if pos == i { + self.pos = pos.checked_sub(1); + } else { + self.pos = Some(i); + } + Some(i) + } + } + } +} + +/// A single substring searcher fixed to a particular needle. +/// +/// The purpose of this type is to permit callers to construct a substring +/// searcher that can be used to search haystacks without the overhead of +/// constructing the searcher in the first place. This is a somewhat niche +/// concern when it's necessary to re-use the same needle to search multiple +/// different haystacks with as little overhead as possible. In general, using +/// [`find`] is good enough, but `Finder` is useful when you can meaningfully +/// observe searcher construction time in a profile. +/// +/// When the `std` feature is enabled, then this type has an `into_owned` +/// version which permits building a `Finder` that is not connected to +/// the lifetime of its needle. +#[derive(Clone, Debug)] +pub struct Finder<'n> { + needle: CowBytes<'n>, + searcher: Searcher, +} + +impl<'n> Finder<'n> { + /// Create a new finder for the given needle. + #[inline] + pub fn new<B: ?Sized + AsRef<[u8]>>(needle: &'n B) -> Finder<'n> { + FinderBuilder::new().build_forward(needle) + } + + /// Returns the index of the first occurrence of this needle in the given + /// haystack. + /// + /// # Complexity + /// + /// This routine is guaranteed to have worst case linear time complexity + /// with respect to both the needle and the haystack. That is, this runs + /// in `O(needle.len() + haystack.len())` time. + /// + /// This routine is also guaranteed to have worst case constant space + /// complexity. + /// + /// # Examples + /// + /// Basic usage: + /// + /// ``` + /// use memchr::memmem::Finder; + /// + /// let haystack = b"foo bar baz"; + /// assert_eq!(Some(0), Finder::new("foo").find(haystack)); + /// assert_eq!(Some(4), Finder::new("bar").find(haystack)); + /// assert_eq!(None, Finder::new("quux").find(haystack)); + /// ``` + #[inline] + pub fn find(&self, haystack: &[u8]) -> Option<usize> { + let mut prestate = PrefilterState::new(); + let needle = self.needle.as_slice(); + self.searcher.find(&mut prestate, haystack, needle) + } + + /// Returns an iterator over all occurrences of a substring in a haystack. + /// + /// # Complexity + /// + /// This routine is guaranteed to have worst case linear time complexity + /// with respect to both the needle and the haystack. That is, this runs + /// in `O(needle.len() + haystack.len())` time. + /// + /// This routine is also guaranteed to have worst case constant space + /// complexity. + /// + /// # Examples + /// + /// Basic usage: + /// + /// ``` + /// use memchr::memmem::Finder; + /// + /// let haystack = b"foo bar foo baz foo"; + /// let finder = Finder::new(b"foo"); + /// let mut it = finder.find_iter(haystack); + /// assert_eq!(Some(0), it.next()); + /// assert_eq!(Some(8), it.next()); + /// assert_eq!(Some(16), it.next()); + /// assert_eq!(None, it.next()); + /// ``` + #[inline] + pub fn find_iter<'a, 'h>( + &'a self, + haystack: &'h [u8], + ) -> FindIter<'h, 'a> { + FindIter::new(haystack, self.as_ref()) + } + + /// Convert this finder into its owned variant, such that it no longer + /// borrows the needle. + /// + /// If this is already an owned finder, then this is a no-op. Otherwise, + /// this copies the needle. + /// + /// This is only available when the `alloc` feature is enabled. + #[cfg(feature = "alloc")] + #[inline] + pub fn into_owned(self) -> Finder<'static> { + Finder { + needle: self.needle.into_owned(), + searcher: self.searcher.clone(), + } + } + + /// Convert this finder into its borrowed variant. + /// + /// This is primarily useful if your finder is owned and you'd like to + /// store its borrowed variant in some intermediate data structure. + /// + /// Note that the lifetime parameter of the returned finder is tied to the + /// lifetime of `self`, and may be shorter than the `'n` lifetime of the + /// needle itself. Namely, a finder's needle can be either borrowed or + /// owned, so the lifetime of the needle returned must necessarily be the + /// shorter of the two. + #[inline] + pub fn as_ref(&self) -> Finder<'_> { + Finder { + needle: CowBytes::new(self.needle()), + searcher: self.searcher.clone(), + } + } + + /// Returns the needle that this finder searches for. + /// + /// Note that the lifetime of the needle returned is tied to the lifetime + /// of the finder, and may be shorter than the `'n` lifetime. Namely, a + /// finder's needle can be either borrowed or owned, so the lifetime of the + /// needle returned must necessarily be the shorter of the two. + #[inline] + pub fn needle(&self) -> &[u8] { + self.needle.as_slice() + } +} + +/// A single substring reverse searcher fixed to a particular needle. +/// +/// The purpose of this type is to permit callers to construct a substring +/// searcher that can be used to search haystacks without the overhead of +/// constructing the searcher in the first place. This is a somewhat niche +/// concern when it's necessary to re-use the same needle to search multiple +/// different haystacks with as little overhead as possible. In general, +/// using [`rfind`] is good enough, but `FinderRev` is useful when you can +/// meaningfully observe searcher construction time in a profile. +/// +/// When the `std` feature is enabled, then this type has an `into_owned` +/// version which permits building a `FinderRev` that is not connected to +/// the lifetime of its needle. +#[derive(Clone, Debug)] +pub struct FinderRev<'n> { + needle: CowBytes<'n>, + searcher: SearcherRev, +} + +impl<'n> FinderRev<'n> { + /// Create a new reverse finder for the given needle. + #[inline] + pub fn new<B: ?Sized + AsRef<[u8]>>(needle: &'n B) -> FinderRev<'n> { + FinderBuilder::new().build_reverse(needle) + } + + /// Returns the index of the last occurrence of this needle in the given + /// haystack. + /// + /// The haystack may be any type that can be cheaply converted into a + /// `&[u8]`. This includes, but is not limited to, `&str` and `&[u8]`. + /// + /// # Complexity + /// + /// This routine is guaranteed to have worst case linear time complexity + /// with respect to both the needle and the haystack. That is, this runs + /// in `O(needle.len() + haystack.len())` time. + /// + /// This routine is also guaranteed to have worst case constant space + /// complexity. + /// + /// # Examples + /// + /// Basic usage: + /// + /// ``` + /// use memchr::memmem::FinderRev; + /// + /// let haystack = b"foo bar baz"; + /// assert_eq!(Some(0), FinderRev::new("foo").rfind(haystack)); + /// assert_eq!(Some(4), FinderRev::new("bar").rfind(haystack)); + /// assert_eq!(None, FinderRev::new("quux").rfind(haystack)); + /// ``` + pub fn rfind<B: AsRef<[u8]>>(&self, haystack: B) -> Option<usize> { + self.searcher.rfind(haystack.as_ref(), self.needle.as_slice()) + } + + /// Returns a reverse iterator over all occurrences of a substring in a + /// haystack. + /// + /// # Complexity + /// + /// This routine is guaranteed to have worst case linear time complexity + /// with respect to both the needle and the haystack. That is, this runs + /// in `O(needle.len() + haystack.len())` time. + /// + /// This routine is also guaranteed to have worst case constant space + /// complexity. + /// + /// # Examples + /// + /// Basic usage: + /// + /// ``` + /// use memchr::memmem::FinderRev; + /// + /// let haystack = b"foo bar foo baz foo"; + /// let finder = FinderRev::new(b"foo"); + /// let mut it = finder.rfind_iter(haystack); + /// assert_eq!(Some(16), it.next()); + /// assert_eq!(Some(8), it.next()); + /// assert_eq!(Some(0), it.next()); + /// assert_eq!(None, it.next()); + /// ``` + #[inline] + pub fn rfind_iter<'a, 'h>( + &'a self, + haystack: &'h [u8], + ) -> FindRevIter<'h, 'a> { + FindRevIter::new(haystack, self.as_ref()) + } + + /// Convert this finder into its owned variant, such that it no longer + /// borrows the needle. + /// + /// If this is already an owned finder, then this is a no-op. Otherwise, + /// this copies the needle. + /// + /// This is only available when the `std` feature is enabled. + #[cfg(feature = "alloc")] + #[inline] + pub fn into_owned(self) -> FinderRev<'static> { + FinderRev { + needle: self.needle.into_owned(), + searcher: self.searcher.clone(), + } + } + + /// Convert this finder into its borrowed variant. + /// + /// This is primarily useful if your finder is owned and you'd like to + /// store its borrowed variant in some intermediate data structure. + /// + /// Note that the lifetime parameter of the returned finder is tied to the + /// lifetime of `self`, and may be shorter than the `'n` lifetime of the + /// needle itself. Namely, a finder's needle can be either borrowed or + /// owned, so the lifetime of the needle returned must necessarily be the + /// shorter of the two. + #[inline] + pub fn as_ref(&self) -> FinderRev<'_> { + FinderRev { + needle: CowBytes::new(self.needle()), + searcher: self.searcher.clone(), + } + } + + /// Returns the needle that this finder searches for. + /// + /// Note that the lifetime of the needle returned is tied to the lifetime + /// of the finder, and may be shorter than the `'n` lifetime. Namely, a + /// finder's needle can be either borrowed or owned, so the lifetime of the + /// needle returned must necessarily be the shorter of the two. + #[inline] + pub fn needle(&self) -> &[u8] { + self.needle.as_slice() + } +} + +/// A builder for constructing non-default forward or reverse memmem finders. +/// +/// A builder is primarily useful for configuring a substring searcher. +/// Currently, the only configuration exposed is the ability to disable +/// heuristic prefilters used to speed up certain searches. +#[derive(Clone, Debug, Default)] +pub struct FinderBuilder { + prefilter: Prefilter, +} + +impl FinderBuilder { + /// Create a new finder builder with default settings. + pub fn new() -> FinderBuilder { + FinderBuilder::default() + } + + /// Build a forward finder using the given needle from the current + /// settings. + pub fn build_forward<'n, B: ?Sized + AsRef<[u8]>>( + &self, + needle: &'n B, + ) -> Finder<'n> { + self.build_forward_with_ranker(DefaultFrequencyRank, needle) + } + + /// Build a forward finder using the given needle and a custom heuristic for + /// determining the frequency of a given byte in the dataset. + /// See [`HeuristicFrequencyRank`] for more details. + pub fn build_forward_with_ranker< + 'n, + R: HeuristicFrequencyRank, + B: ?Sized + AsRef<[u8]>, + >( + &self, + ranker: R, + needle: &'n B, + ) -> Finder<'n> { + let needle = needle.as_ref(); + Finder { + needle: CowBytes::new(needle), + searcher: Searcher::new(self.prefilter, ranker, needle), + } + } + + /// Build a reverse finder using the given needle from the current + /// settings. + pub fn build_reverse<'n, B: ?Sized + AsRef<[u8]>>( + &self, + needle: &'n B, + ) -> FinderRev<'n> { + let needle = needle.as_ref(); + FinderRev { + needle: CowBytes::new(needle), + searcher: SearcherRev::new(needle), + } + } + + /// Configure the prefilter setting for the finder. + /// + /// See the documentation for [`Prefilter`] for more discussion on why + /// you might want to configure this. + pub fn prefilter(&mut self, prefilter: Prefilter) -> &mut FinderBuilder { + self.prefilter = prefilter; + self + } +} + +#[cfg(test)] +mod tests { + use super::*; + + define_substring_forward_quickcheck!(|h, n| Some(Finder::new(n).find(h))); + define_substring_reverse_quickcheck!(|h, n| Some( + FinderRev::new(n).rfind(h) + )); + + #[test] + fn forward() { + crate::tests::substring::Runner::new() + .fwd(|h, n| Some(Finder::new(n).find(h))) + .run(); + } + + #[test] + fn reverse() { + crate::tests::substring::Runner::new() + .rev(|h, n| Some(FinderRev::new(n).rfind(h))) + .run(); + } +} diff --git a/vendor/memchr/src/memmem/searcher.rs b/vendor/memchr/src/memmem/searcher.rs new file mode 100644 index 0000000..98b9bd6 --- /dev/null +++ b/vendor/memchr/src/memmem/searcher.rs @@ -0,0 +1,1030 @@ +use crate::arch::all::{ + packedpair::{HeuristicFrequencyRank, Pair}, + rabinkarp, twoway, +}; + +#[cfg(target_arch = "aarch64")] +use crate::arch::aarch64::neon::packedpair as neon; +#[cfg(target_arch = "wasm32")] +use crate::arch::wasm32::simd128::packedpair as simd128; +#[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] +use crate::arch::x86_64::{ + avx2::packedpair as avx2, sse2::packedpair as sse2, +}; + +/// A "meta" substring searcher. +/// +/// To a first approximation, this chooses what it believes to be the "best" +/// substring search implemnetation based on the needle at construction time. +/// Then, every call to `find` will execute that particular implementation. To +/// a second approximation, multiple substring search algorithms may be used, +/// depending on the haystack. For example, for supremely short haystacks, +/// Rabin-Karp is typically used. +/// +/// See the documentation on `Prefilter` for an explanation of the dispatching +/// mechanism. The quick summary is that an enum has too much overhead and +/// we can't use dynamic dispatch via traits because we need to work in a +/// core-only environment. (Dynamic dispatch works in core-only, but you +/// need `&dyn Trait` and we really need a `Box<dyn Trait>` here. The latter +/// requires `alloc`.) So instead, we use a union and an appropriately paired +/// free function to read from the correct field on the union and execute the +/// chosen substring search implementation. +#[derive(Clone)] +pub(crate) struct Searcher { + call: SearcherKindFn, + kind: SearcherKind, + rabinkarp: rabinkarp::Finder, +} + +impl Searcher { + /// Creates a new "meta" substring searcher that attempts to choose the + /// best algorithm based on the needle, heuristics and what the current + /// target supports. + #[inline] + pub(crate) fn new<R: HeuristicFrequencyRank>( + prefilter: PrefilterConfig, + ranker: R, + needle: &[u8], + ) -> Searcher { + let rabinkarp = rabinkarp::Finder::new(needle); + if needle.len() <= 1 { + return if needle.is_empty() { + trace!("building empty substring searcher"); + Searcher { + call: searcher_kind_empty, + kind: SearcherKind { empty: () }, + rabinkarp, + } + } else { + trace!("building one-byte substring searcher"); + debug_assert_eq!(1, needle.len()); + Searcher { + call: searcher_kind_one_byte, + kind: SearcherKind { one_byte: needle[0] }, + rabinkarp, + } + }; + } + let pair = match Pair::with_ranker(needle, &ranker) { + Some(pair) => pair, + None => return Searcher::twoway(needle, rabinkarp, None), + }; + debug_assert_ne!( + pair.index1(), + pair.index2(), + "pair offsets should not be equivalent" + ); + #[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] + { + if let Some(pp) = avx2::Finder::with_pair(needle, pair) { + if do_packed_search(needle) { + trace!("building x86_64 AVX2 substring searcher"); + let kind = SearcherKind { avx2: pp }; + Searcher { call: searcher_kind_avx2, kind, rabinkarp } + } else if prefilter.is_none() { + Searcher::twoway(needle, rabinkarp, None) + } else { + let prestrat = Prefilter::avx2(pp, needle); + Searcher::twoway(needle, rabinkarp, Some(prestrat)) + } + } else if let Some(pp) = sse2::Finder::with_pair(needle, pair) { + if do_packed_search(needle) { + trace!("building x86_64 SSE2 substring searcher"); + let kind = SearcherKind { sse2: pp }; + Searcher { call: searcher_kind_sse2, kind, rabinkarp } + } else if prefilter.is_none() { + Searcher::twoway(needle, rabinkarp, None) + } else { + let prestrat = Prefilter::sse2(pp, needle); + Searcher::twoway(needle, rabinkarp, Some(prestrat)) + } + } else if prefilter.is_none() { + Searcher::twoway(needle, rabinkarp, None) + } else { + // We're pretty unlikely to get to this point, but it is + // possible to be running on x86_64 without SSE2. Namely, it's + // really up to the OS whether it wants to support vector + // registers or not. + let prestrat = Prefilter::fallback(ranker, pair, needle); + Searcher::twoway(needle, rabinkarp, prestrat) + } + } + #[cfg(target_arch = "wasm32")] + { + if let Some(pp) = simd128::Finder::with_pair(needle, pair) { + if do_packed_search(needle) { + trace!("building wasm32 simd128 substring searcher"); + let kind = SearcherKind { simd128: pp }; + Searcher { call: searcher_kind_simd128, kind, rabinkarp } + } else if prefilter.is_none() { + Searcher::twoway(needle, rabinkarp, None) + } else { + let prestrat = Prefilter::simd128(pp, needle); + Searcher::twoway(needle, rabinkarp, Some(prestrat)) + } + } else if prefilter.is_none() { + Searcher::twoway(needle, rabinkarp, None) + } else { + let prestrat = Prefilter::fallback(ranker, pair, needle); + Searcher::twoway(needle, rabinkarp, prestrat) + } + } + #[cfg(target_arch = "aarch64")] + { + if let Some(pp) = neon::Finder::with_pair(needle, pair) { + if do_packed_search(needle) { + trace!("building aarch64 neon substring searcher"); + let kind = SearcherKind { neon: pp }; + Searcher { call: searcher_kind_neon, kind, rabinkarp } + } else if prefilter.is_none() { + Searcher::twoway(needle, rabinkarp, None) + } else { + let prestrat = Prefilter::neon(pp, needle); + Searcher::twoway(needle, rabinkarp, Some(prestrat)) + } + } else if prefilter.is_none() { + Searcher::twoway(needle, rabinkarp, None) + } else { + let prestrat = Prefilter::fallback(ranker, pair, needle); + Searcher::twoway(needle, rabinkarp, prestrat) + } + } + #[cfg(not(any( + all(target_arch = "x86_64", target_feature = "sse2"), + target_arch = "wasm32", + target_arch = "aarch64" + )))] + { + if prefilter.is_none() { + Searcher::twoway(needle, rabinkarp, None) + } else { + let prestrat = Prefilter::fallback(ranker, pair, needle); + Searcher::twoway(needle, rabinkarp, prestrat) + } + } + } + + /// Creates a new searcher that always uses the Two-Way algorithm. This is + /// typically used when vector algorithms are unavailable or inappropriate. + /// (For example, when the needle is "too long.") + /// + /// If a prefilter is given, then the searcher returned will be accelerated + /// by the prefilter. + #[inline] + fn twoway( + needle: &[u8], + rabinkarp: rabinkarp::Finder, + prestrat: Option<Prefilter>, + ) -> Searcher { + let finder = twoway::Finder::new(needle); + match prestrat { + None => { + trace!("building scalar two-way substring searcher"); + let kind = SearcherKind { two_way: finder }; + Searcher { call: searcher_kind_two_way, kind, rabinkarp } + } + Some(prestrat) => { + trace!( + "building scalar two-way \ + substring searcher with a prefilter" + ); + let two_way_with_prefilter = + TwoWayWithPrefilter { finder, prestrat }; + let kind = SearcherKind { two_way_with_prefilter }; + Searcher { + call: searcher_kind_two_way_with_prefilter, + kind, + rabinkarp, + } + } + } + } + + /// Searches the given haystack for the given needle. The needle given + /// should be the same as the needle that this finder was initialized + /// with. + /// + /// Inlining this can lead to big wins for latency, and #[inline] doesn't + /// seem to be enough in some cases. + #[inline(always)] + pub(crate) fn find( + &self, + prestate: &mut PrefilterState, + haystack: &[u8], + needle: &[u8], + ) -> Option<usize> { + if haystack.len() < needle.len() { + None + } else { + // SAFETY: By construction, we've ensured that the function + // in `self.call` is properly paired with the union used in + // `self.kind`. + unsafe { (self.call)(self, prestate, haystack, needle) } + } + } +} + +impl core::fmt::Debug for Searcher { + fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result { + f.debug_struct("Searcher") + .field("call", &"<searcher function>") + .field("kind", &"<searcher kind union>") + .field("rabinkarp", &self.rabinkarp) + .finish() + } +} + +/// A union indicating one of several possible substring search implementations +/// that are in active use. +/// +/// This union should only be read by one of the functions prefixed with +/// `searcher_kind_`. Namely, the correct function is meant to be paired with +/// the union by the caller, such that the function always reads from the +/// designated union field. +#[derive(Clone, Copy)] +union SearcherKind { + empty: (), + one_byte: u8, + two_way: twoway::Finder, + two_way_with_prefilter: TwoWayWithPrefilter, + #[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] + sse2: crate::arch::x86_64::sse2::packedpair::Finder, + #[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] + avx2: crate::arch::x86_64::avx2::packedpair::Finder, + #[cfg(target_arch = "wasm32")] + simd128: crate::arch::wasm32::simd128::packedpair::Finder, + #[cfg(target_arch = "aarch64")] + neon: crate::arch::aarch64::neon::packedpair::Finder, +} + +/// A two-way substring searcher with a prefilter. +#[derive(Copy, Clone, Debug)] +struct TwoWayWithPrefilter { + finder: twoway::Finder, + prestrat: Prefilter, +} + +/// The type of a substring search function. +/// +/// # Safety +/// +/// When using a function of this type, callers must ensure that the correct +/// function is paired with the value populated in `SearcherKind` union. +type SearcherKindFn = unsafe fn( + searcher: &Searcher, + prestate: &mut PrefilterState, + haystack: &[u8], + needle: &[u8], +) -> Option<usize>; + +/// Reads from the `empty` field of `SearcherKind` to handle the case of +/// searching for the empty needle. Works on all platforms. +/// +/// # Safety +/// +/// Callers must ensure that the `searcher.kind.empty` union field is set. +unsafe fn searcher_kind_empty( + _searcher: &Searcher, + _prestate: &mut PrefilterState, + _haystack: &[u8], + _needle: &[u8], +) -> Option<usize> { + Some(0) +} + +/// Reads from the `one_byte` field of `SearcherKind` to handle the case of +/// searching for a single byte needle. Works on all platforms. +/// +/// # Safety +/// +/// Callers must ensure that the `searcher.kind.one_byte` union field is set. +unsafe fn searcher_kind_one_byte( + searcher: &Searcher, + _prestate: &mut PrefilterState, + haystack: &[u8], + _needle: &[u8], +) -> Option<usize> { + let needle = searcher.kind.one_byte; + crate::memchr(needle, haystack) +} + +/// Reads from the `two_way` field of `SearcherKind` to handle the case of +/// searching for an arbitrary needle without prefilter acceleration. Works on +/// all platforms. +/// +/// # Safety +/// +/// Callers must ensure that the `searcher.kind.two_way` union field is set. +unsafe fn searcher_kind_two_way( + searcher: &Searcher, + _prestate: &mut PrefilterState, + haystack: &[u8], + needle: &[u8], +) -> Option<usize> { + if rabinkarp::is_fast(haystack, needle) { + searcher.rabinkarp.find(haystack, needle) + } else { + searcher.kind.two_way.find(haystack, needle) + } +} + +/// Reads from the `two_way_with_prefilter` field of `SearcherKind` to handle +/// the case of searching for an arbitrary needle with prefilter acceleration. +/// Works on all platforms. +/// +/// # Safety +/// +/// Callers must ensure that the `searcher.kind.two_way_with_prefilter` union +/// field is set. +unsafe fn searcher_kind_two_way_with_prefilter( + searcher: &Searcher, + prestate: &mut PrefilterState, + haystack: &[u8], + needle: &[u8], +) -> Option<usize> { + if rabinkarp::is_fast(haystack, needle) { + searcher.rabinkarp.find(haystack, needle) + } else { + let TwoWayWithPrefilter { ref finder, ref prestrat } = + searcher.kind.two_way_with_prefilter; + let pre = Pre { prestate, prestrat }; + finder.find_with_prefilter(Some(pre), haystack, needle) + } +} + +/// Reads from the `sse2` field of `SearcherKind` to execute the x86_64 SSE2 +/// vectorized substring search implementation. +/// +/// # Safety +/// +/// Callers must ensure that the `searcher.kind.sse2` union field is set. +#[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] +unsafe fn searcher_kind_sse2( + searcher: &Searcher, + _prestate: &mut PrefilterState, + haystack: &[u8], + needle: &[u8], +) -> Option<usize> { + let finder = &searcher.kind.sse2; + if haystack.len() < finder.min_haystack_len() { + searcher.rabinkarp.find(haystack, needle) + } else { + finder.find(haystack, needle) + } +} + +/// Reads from the `avx2` field of `SearcherKind` to execute the x86_64 AVX2 +/// vectorized substring search implementation. +/// +/// # Safety +/// +/// Callers must ensure that the `searcher.kind.avx2` union field is set. +#[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] +unsafe fn searcher_kind_avx2( + searcher: &Searcher, + _prestate: &mut PrefilterState, + haystack: &[u8], + needle: &[u8], +) -> Option<usize> { + let finder = &searcher.kind.avx2; + if haystack.len() < finder.min_haystack_len() { + searcher.rabinkarp.find(haystack, needle) + } else { + finder.find(haystack, needle) + } +} + +/// Reads from the `simd128` field of `SearcherKind` to execute the wasm32 +/// simd128 vectorized substring search implementation. +/// +/// # Safety +/// +/// Callers must ensure that the `searcher.kind.simd128` union field is set. +#[cfg(target_arch = "wasm32")] +unsafe fn searcher_kind_simd128( + searcher: &Searcher, + _prestate: &mut PrefilterState, + haystack: &[u8], + needle: &[u8], +) -> Option<usize> { + let finder = &searcher.kind.simd128; + if haystack.len() < finder.min_haystack_len() { + searcher.rabinkarp.find(haystack, needle) + } else { + finder.find(haystack, needle) + } +} + +/// Reads from the `neon` field of `SearcherKind` to execute the aarch64 neon +/// vectorized substring search implementation. +/// +/// # Safety +/// +/// Callers must ensure that the `searcher.kind.neon` union field is set. +#[cfg(target_arch = "aarch64")] +unsafe fn searcher_kind_neon( + searcher: &Searcher, + _prestate: &mut PrefilterState, + haystack: &[u8], + needle: &[u8], +) -> Option<usize> { + let finder = &searcher.kind.neon; + if haystack.len() < finder.min_haystack_len() { + searcher.rabinkarp.find(haystack, needle) + } else { + finder.find(haystack, needle) + } +} + +/// A reverse substring searcher. +#[derive(Clone, Debug)] +pub(crate) struct SearcherRev { + kind: SearcherRevKind, + rabinkarp: rabinkarp::FinderRev, +} + +/// The kind of the reverse searcher. +/// +/// For the reverse case, we don't do any SIMD acceleration or prefilters. +/// There is no specific technical reason why we don't, but rather don't do it +/// because it's not clear it's worth the extra code to do so. If you have a +/// use case for it, please file an issue. +/// +/// We also don't do the union trick as we do with the forward case and +/// prefilters. Basically for the same reason we don't have prefilters or +/// vector algorithms for reverse searching: it's not clear it's worth doing. +/// Please file an issue if you have a compelling use case for fast reverse +/// substring search. +#[derive(Clone, Debug)] +enum SearcherRevKind { + Empty, + OneByte { needle: u8 }, + TwoWay { finder: twoway::FinderRev }, +} + +impl SearcherRev { + /// Creates a new searcher for finding occurrences of the given needle in + /// reverse. That is, it reports the last (instead of the first) occurrence + /// of a needle in a haystack. + #[inline] + pub(crate) fn new(needle: &[u8]) -> SearcherRev { + let kind = if needle.len() <= 1 { + if needle.is_empty() { + trace!("building empty reverse substring searcher"); + SearcherRevKind::Empty + } else { + trace!("building one-byte reverse substring searcher"); + debug_assert_eq!(1, needle.len()); + SearcherRevKind::OneByte { needle: needle[0] } + } + } else { + trace!("building scalar two-way reverse substring searcher"); + let finder = twoway::FinderRev::new(needle); + SearcherRevKind::TwoWay { finder } + }; + let rabinkarp = rabinkarp::FinderRev::new(needle); + SearcherRev { kind, rabinkarp } + } + + /// Searches the given haystack for the last occurrence of the given + /// needle. The needle given should be the same as the needle that this + /// finder was initialized with. + #[inline] + pub(crate) fn rfind( + &self, + haystack: &[u8], + needle: &[u8], + ) -> Option<usize> { + if haystack.len() < needle.len() { + return None; + } + match self.kind { + SearcherRevKind::Empty => Some(haystack.len()), + SearcherRevKind::OneByte { needle } => { + crate::memrchr(needle, haystack) + } + SearcherRevKind::TwoWay { ref finder } => { + if rabinkarp::is_fast(haystack, needle) { + self.rabinkarp.rfind(haystack, needle) + } else { + finder.rfind(haystack, needle) + } + } + } + } +} + +/// Prefilter controls whether heuristics are used to accelerate searching. +/// +/// A prefilter refers to the idea of detecting candidate matches very quickly, +/// and then confirming whether those candidates are full matches. This +/// idea can be quite effective since it's often the case that looking for +/// candidates can be a lot faster than running a complete substring search +/// over the entire input. Namely, looking for candidates can be done with +/// extremely fast vectorized code. +/// +/// The downside of a prefilter is that it assumes false positives (which are +/// candidates generated by a prefilter that aren't matches) are somewhat rare +/// relative to the frequency of full matches. That is, if a lot of false +/// positives are generated, then it's possible for search time to be worse +/// than if the prefilter wasn't enabled in the first place. +/// +/// Another downside of a prefilter is that it can result in highly variable +/// performance, where some cases are extraordinarily fast and others aren't. +/// Typically, variable performance isn't a problem, but it may be for your use +/// case. +/// +/// The use of prefilters in this implementation does use a heuristic to detect +/// when a prefilter might not be carrying its weight, and will dynamically +/// disable its use. Nevertheless, this configuration option gives callers +/// the ability to disable prefilters if you have knowledge that they won't be +/// useful. +#[derive(Clone, Copy, Debug)] +#[non_exhaustive] +pub enum PrefilterConfig { + /// Never used a prefilter in substring search. + None, + /// Automatically detect whether a heuristic prefilter should be used. If + /// it is used, then heuristics will be used to dynamically disable the + /// prefilter if it is believed to not be carrying its weight. + Auto, +} + +impl Default for PrefilterConfig { + fn default() -> PrefilterConfig { + PrefilterConfig::Auto + } +} + +impl PrefilterConfig { + /// Returns true when this prefilter is set to the `None` variant. + fn is_none(&self) -> bool { + matches!(*self, PrefilterConfig::None) + } +} + +/// The implementation of a prefilter. +/// +/// This type encapsulates dispatch to one of several possible choices for a +/// prefilter. Generally speaking, all prefilters have the same approximate +/// algorithm: they choose a couple of bytes from the needle that are believed +/// to be rare, use a fast vector algorithm to look for those bytes and return +/// positions as candidates for some substring search algorithm (currently only +/// Two-Way) to confirm as a match or not. +/// +/// The differences between the algorithms are actually at the vector +/// implementation level. Namely, we need different routines based on both +/// which target architecture we're on and what CPU features are supported. +/// +/// The straight-forwardly obvious approach here is to use an enum, and make +/// `Prefilter::find` do case analysis to determine which algorithm was +/// selected and invoke it. However, I've observed that this leads to poor +/// codegen in some cases, especially in latency sensitive benchmarks. That is, +/// this approach comes with overhead that I wasn't able to eliminate. +/// +/// The second obvious approach is to use dynamic dispatch with traits. Doing +/// that in this context where `Prefilter` owns the selection generally +/// requires heap allocation, and this code is designed to run in core-only +/// environments. +/// +/// So we settle on using a union (that's `PrefilterKind`) and a function +/// pointer (that's `PrefilterKindFn`). We select the right function pointer +/// based on which field in the union we set, and that function in turn +/// knows which field of the union to access. The downside of this approach +/// is that it forces us to think about safety, but the upside is that +/// there are some nice latency improvements to benchmarks. (Especially the +/// `memmem/sliceslice/short` benchmark.) +/// +/// In cases where we've selected a vector algorithm and the haystack given +/// is too short, we fallback to the scalar version of `memchr` on the +/// `rarest_byte`. (The scalar version of `memchr` is still better than a naive +/// byte-at-a-time loop because it will read in `usize`-sized chunks at a +/// time.) +#[derive(Clone, Copy)] +struct Prefilter { + call: PrefilterKindFn, + kind: PrefilterKind, + rarest_byte: u8, + rarest_offset: u8, +} + +impl Prefilter { + /// Return a "fallback" prefilter, but only if it is believed to be + /// effective. + #[inline] + fn fallback<R: HeuristicFrequencyRank>( + ranker: R, + pair: Pair, + needle: &[u8], + ) -> Option<Prefilter> { + /// The maximum frequency rank permitted for the fallback prefilter. + /// If the rarest byte in the needle has a frequency rank above this + /// value, then no prefilter is used if the fallback prefilter would + /// otherwise be selected. + const MAX_FALLBACK_RANK: u8 = 250; + + trace!("building fallback prefilter"); + let rarest_offset = pair.index1(); + let rarest_byte = needle[usize::from(rarest_offset)]; + let rarest_rank = ranker.rank(rarest_byte); + if rarest_rank > MAX_FALLBACK_RANK { + None + } else { + let finder = crate::arch::all::packedpair::Finder::with_pair( + needle, + pair.clone(), + )?; + let call = prefilter_kind_fallback; + let kind = PrefilterKind { fallback: finder }; + Some(Prefilter { call, kind, rarest_byte, rarest_offset }) + } + } + + /// Return a prefilter using a x86_64 SSE2 vector algorithm. + #[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] + #[inline] + fn sse2(finder: sse2::Finder, needle: &[u8]) -> Prefilter { + trace!("building x86_64 SSE2 prefilter"); + let rarest_offset = finder.pair().index1(); + let rarest_byte = needle[usize::from(rarest_offset)]; + Prefilter { + call: prefilter_kind_sse2, + kind: PrefilterKind { sse2: finder }, + rarest_byte, + rarest_offset, + } + } + + /// Return a prefilter using a x86_64 AVX2 vector algorithm. + #[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] + #[inline] + fn avx2(finder: avx2::Finder, needle: &[u8]) -> Prefilter { + trace!("building x86_64 AVX2 prefilter"); + let rarest_offset = finder.pair().index1(); + let rarest_byte = needle[usize::from(rarest_offset)]; + Prefilter { + call: prefilter_kind_avx2, + kind: PrefilterKind { avx2: finder }, + rarest_byte, + rarest_offset, + } + } + + /// Return a prefilter using a wasm32 simd128 vector algorithm. + #[cfg(target_arch = "wasm32")] + #[inline] + fn simd128(finder: simd128::Finder, needle: &[u8]) -> Prefilter { + trace!("building wasm32 simd128 prefilter"); + let rarest_offset = finder.pair().index1(); + let rarest_byte = needle[usize::from(rarest_offset)]; + Prefilter { + call: prefilter_kind_simd128, + kind: PrefilterKind { simd128: finder }, + rarest_byte, + rarest_offset, + } + } + + /// Return a prefilter using a aarch64 neon vector algorithm. + #[cfg(target_arch = "aarch64")] + #[inline] + fn neon(finder: neon::Finder, needle: &[u8]) -> Prefilter { + trace!("building aarch64 neon prefilter"); + let rarest_offset = finder.pair().index1(); + let rarest_byte = needle[usize::from(rarest_offset)]; + Prefilter { + call: prefilter_kind_neon, + kind: PrefilterKind { neon: finder }, + rarest_byte, + rarest_offset, + } + } + + /// Return a *candidate* position for a match. + /// + /// When this returns an offset, it implies that a match could begin at + /// that offset, but it may not. That is, it is possible for a false + /// positive to be returned. + /// + /// When `None` is returned, then it is guaranteed that there are no + /// matches for the needle in the given haystack. That is, it is impossible + /// for a false negative to be returned. + /// + /// The purpose of this routine is to look for candidate matching positions + /// as quickly as possible before running a (likely) slower confirmation + /// step. + #[inline] + fn find(&self, haystack: &[u8]) -> Option<usize> { + // SAFETY: By construction, we've ensured that the function in + // `self.call` is properly paired with the union used in `self.kind`. + unsafe { (self.call)(self, haystack) } + } + + /// A "simple" prefilter that just looks for the occurrence of the rarest + /// byte from the needle. This is generally only used for very small + /// haystacks. + #[inline] + fn find_simple(&self, haystack: &[u8]) -> Option<usize> { + // We don't use crate::memchr here because the haystack should be small + // enough that memchr won't be able to use vector routines anyway. So + // we just skip straight to the fallback implementation which is likely + // faster. (A byte-at-a-time loop is only used when the haystack is + // smaller than `size_of::<usize>()`.) + crate::arch::all::memchr::One::new(self.rarest_byte) + .find(haystack) + .map(|i| i.saturating_sub(usize::from(self.rarest_offset))) + } +} + +impl core::fmt::Debug for Prefilter { + fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result { + f.debug_struct("Prefilter") + .field("call", &"<prefilter function>") + .field("kind", &"<prefilter kind union>") + .field("rarest_byte", &self.rarest_byte) + .field("rarest_offset", &self.rarest_offset) + .finish() + } +} + +/// A union indicating one of several possible prefilters that are in active +/// use. +/// +/// This union should only be read by one of the functions prefixed with +/// `prefilter_kind_`. Namely, the correct function is meant to be paired with +/// the union by the caller, such that the function always reads from the +/// designated union field. +#[derive(Clone, Copy)] +union PrefilterKind { + fallback: crate::arch::all::packedpair::Finder, + #[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] + sse2: crate::arch::x86_64::sse2::packedpair::Finder, + #[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] + avx2: crate::arch::x86_64::avx2::packedpair::Finder, + #[cfg(target_arch = "wasm32")] + simd128: crate::arch::wasm32::simd128::packedpair::Finder, + #[cfg(target_arch = "aarch64")] + neon: crate::arch::aarch64::neon::packedpair::Finder, +} + +/// The type of a prefilter function. +/// +/// # Safety +/// +/// When using a function of this type, callers must ensure that the correct +/// function is paired with the value populated in `PrefilterKind` union. +type PrefilterKindFn = + unsafe fn(strat: &Prefilter, haystack: &[u8]) -> Option<usize>; + +/// Reads from the `fallback` field of `PrefilterKind` to execute the fallback +/// prefilter. Works on all platforms. +/// +/// # Safety +/// +/// Callers must ensure that the `strat.kind.fallback` union field is set. +unsafe fn prefilter_kind_fallback( + strat: &Prefilter, + haystack: &[u8], +) -> Option<usize> { + strat.kind.fallback.find_prefilter(haystack) +} + +/// Reads from the `sse2` field of `PrefilterKind` to execute the x86_64 SSE2 +/// prefilter. +/// +/// # Safety +/// +/// Callers must ensure that the `strat.kind.sse2` union field is set. +#[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] +unsafe fn prefilter_kind_sse2( + strat: &Prefilter, + haystack: &[u8], +) -> Option<usize> { + let finder = &strat.kind.sse2; + if haystack.len() < finder.min_haystack_len() { + strat.find_simple(haystack) + } else { + finder.find_prefilter(haystack) + } +} + +/// Reads from the `avx2` field of `PrefilterKind` to execute the x86_64 AVX2 +/// prefilter. +/// +/// # Safety +/// +/// Callers must ensure that the `strat.kind.avx2` union field is set. +#[cfg(all(target_arch = "x86_64", target_feature = "sse2"))] +unsafe fn prefilter_kind_avx2( + strat: &Prefilter, + haystack: &[u8], +) -> Option<usize> { + let finder = &strat.kind.avx2; + if haystack.len() < finder.min_haystack_len() { + strat.find_simple(haystack) + } else { + finder.find_prefilter(haystack) + } +} + +/// Reads from the `simd128` field of `PrefilterKind` to execute the wasm32 +/// simd128 prefilter. +/// +/// # Safety +/// +/// Callers must ensure that the `strat.kind.simd128` union field is set. +#[cfg(target_arch = "wasm32")] +unsafe fn prefilter_kind_simd128( + strat: &Prefilter, + haystack: &[u8], +) -> Option<usize> { + let finder = &strat.kind.simd128; + if haystack.len() < finder.min_haystack_len() { + strat.find_simple(haystack) + } else { + finder.find_prefilter(haystack) + } +} + +/// Reads from the `neon` field of `PrefilterKind` to execute the aarch64 neon +/// prefilter. +/// +/// # Safety +/// +/// Callers must ensure that the `strat.kind.neon` union field is set. +#[cfg(target_arch = "aarch64")] +unsafe fn prefilter_kind_neon( + strat: &Prefilter, + haystack: &[u8], +) -> Option<usize> { + let finder = &strat.kind.neon; + if haystack.len() < finder.min_haystack_len() { + strat.find_simple(haystack) + } else { + finder.find_prefilter(haystack) + } +} + +/// PrefilterState tracks state associated with the effectiveness of a +/// prefilter. It is used to track how many bytes, on average, are skipped by +/// the prefilter. If this average dips below a certain threshold over time, +/// then the state renders the prefilter inert and stops using it. +/// +/// A prefilter state should be created for each search. (Where creating an +/// iterator is treated as a single search.) A prefilter state should only be +/// created from a `Freqy`. e.g., An inert `Freqy` will produce an inert +/// `PrefilterState`. +#[derive(Clone, Copy, Debug)] +pub(crate) struct PrefilterState { + /// The number of skips that has been executed. This is always 1 greater + /// than the actual number of skips. The special sentinel value of 0 + /// indicates that the prefilter is inert. This is useful to avoid + /// additional checks to determine whether the prefilter is still + /// "effective." Once a prefilter becomes inert, it should no longer be + /// used (according to our heuristics). + skips: u32, + /// The total number of bytes that have been skipped. + skipped: u32, +} + +impl PrefilterState { + /// The minimum number of skip attempts to try before considering whether + /// a prefilter is effective or not. + const MIN_SKIPS: u32 = 50; + + /// The minimum amount of bytes that skipping must average. + /// + /// This value was chosen based on varying it and checking + /// the microbenchmarks. In particular, this can impact the + /// pathological/repeated-{huge,small} benchmarks quite a bit if it's set + /// too low. + const MIN_SKIP_BYTES: u32 = 8; + + /// Create a fresh prefilter state. + #[inline] + pub(crate) fn new() -> PrefilterState { + PrefilterState { skips: 1, skipped: 0 } + } + + /// Update this state with the number of bytes skipped on the last + /// invocation of the prefilter. + #[inline] + fn update(&mut self, skipped: usize) { + self.skips = self.skips.saturating_add(1); + // We need to do this dance since it's technically possible for + // `skipped` to overflow a `u32`. (And we use a `u32` to reduce the + // size of a prefilter state.) + self.skipped = match u32::try_from(skipped) { + Err(_) => core::u32::MAX, + Ok(skipped) => self.skipped.saturating_add(skipped), + }; + } + + /// Return true if and only if this state indicates that a prefilter is + /// still effective. + #[inline] + fn is_effective(&mut self) -> bool { + if self.is_inert() { + return false; + } + if self.skips() < PrefilterState::MIN_SKIPS { + return true; + } + if self.skipped >= PrefilterState::MIN_SKIP_BYTES * self.skips() { + return true; + } + + // We're inert. + self.skips = 0; + false + } + + /// Returns true if the prefilter this state represents should no longer + /// be used. + #[inline] + fn is_inert(&self) -> bool { + self.skips == 0 + } + + /// Returns the total number of times the prefilter has been used. + #[inline] + fn skips(&self) -> u32 { + // Remember, `0` is a sentinel value indicating inertness, so we + // always need to subtract `1` to get our actual number of skips. + self.skips.saturating_sub(1) + } +} + +/// A combination of prefilter effectiveness state and the prefilter itself. +#[derive(Debug)] +pub(crate) struct Pre<'a> { + /// State that tracks the effectiveness of a prefilter. + prestate: &'a mut PrefilterState, + /// The actual prefilter. + prestrat: &'a Prefilter, +} + +impl<'a> Pre<'a> { + /// Call this prefilter on the given haystack with the given needle. + #[inline] + pub(crate) fn find(&mut self, haystack: &[u8]) -> Option<usize> { + let result = self.prestrat.find(haystack); + self.prestate.update(result.unwrap_or(haystack.len())); + result + } + + /// Return true if and only if this prefilter should be used. + #[inline] + pub(crate) fn is_effective(&mut self) -> bool { + self.prestate.is_effective() + } +} + +/// Returns true if the needle has the right characteristics for a vector +/// algorithm to handle the entirety of substring search. +/// +/// Vector algorithms can be used for prefilters for other substring search +/// algorithms (like Two-Way), but they can also be used for substring search +/// on their own. When used for substring search, vector algorithms will +/// quickly identify candidate match positions (just like in the prefilter +/// case), but instead of returning the candidate position they will try to +/// confirm the match themselves. Confirmation happens via `memcmp`. This +/// works well for short needles, but can break down when many false candidate +/// positions are generated for large needles. Thus, we only permit vector +/// algorithms to own substring search when the needle is of a certain length. +#[inline] +fn do_packed_search(needle: &[u8]) -> bool { + /// The minimum length of a needle required for this algorithm. The minimum + /// is 2 since a length of 1 should just use memchr and a length of 0 isn't + /// a case handled by this searcher. + const MIN_LEN: usize = 2; + + /// The maximum length of a needle required for this algorithm. + /// + /// In reality, there is no hard max here. The code below can handle any + /// length needle. (Perhaps that suggests there are missing optimizations.) + /// Instead, this is a heuristic and a bound guaranteeing our linear time + /// complexity. + /// + /// It is a heuristic because when a candidate match is found, memcmp is + /// run. For very large needles with lots of false positives, memcmp can + /// make the code run quite slow. + /// + /// It is a bound because the worst case behavior with memcmp is + /// multiplicative in the size of the needle and haystack, and we want + /// to keep that additive. This bound ensures we still meet that bound + /// theoretically, since it's just a constant. We aren't acting in bad + /// faith here, memcmp on tiny needles is so fast that even in pathological + /// cases (see pathological vector benchmarks), this is still just as fast + /// or faster in practice. + /// + /// This specific number was chosen by tweaking a bit and running + /// benchmarks. The rare-medium-needle, for example, gets about 5% faster + /// by using this algorithm instead of a prefilter-accelerated Two-Way. + /// There's also a theoretical desire to keep this number reasonably + /// low, to mitigate the impact of pathological cases. I did try 64, and + /// some benchmarks got a little better, and others (particularly the + /// pathological ones), got a lot worse. So... 32 it is? + const MAX_LEN: usize = 32; + MIN_LEN <= needle.len() && needle.len() <= MAX_LEN +} diff --git a/vendor/memchr/src/tests/memchr/mod.rs b/vendor/memchr/src/tests/memchr/mod.rs new file mode 100644 index 0000000..0564ad4 --- /dev/null +++ b/vendor/memchr/src/tests/memchr/mod.rs @@ -0,0 +1,307 @@ +use alloc::{ + string::{String, ToString}, + vec, + vec::Vec, +}; + +use crate::ext::Byte; + +pub(crate) mod naive; +#[macro_use] +pub(crate) mod prop; + +const SEEDS: &'static [Seed] = &[ + Seed { haystack: "a", needles: &[b'a'], positions: &[0] }, + Seed { haystack: "aa", needles: &[b'a'], positions: &[0, 1] }, + Seed { haystack: "aaa", needles: &[b'a'], positions: &[0, 1, 2] }, + Seed { haystack: "", needles: &[b'a'], positions: &[] }, + Seed { haystack: "z", needles: &[b'a'], positions: &[] }, + Seed { haystack: "zz", needles: &[b'a'], positions: &[] }, + Seed { haystack: "zza", needles: &[b'a'], positions: &[2] }, + Seed { haystack: "zaza", needles: &[b'a'], positions: &[1, 3] }, + Seed { haystack: "zzza", needles: &[b'a'], positions: &[3] }, + Seed { haystack: "\x00a", needles: &[b'a'], positions: &[1] }, + Seed { haystack: "\x00", needles: &[b'\x00'], positions: &[0] }, + Seed { haystack: "\x00\x00", needles: &[b'\x00'], positions: &[0, 1] }, + Seed { haystack: "\x00a\x00", needles: &[b'\x00'], positions: &[0, 2] }, + Seed { haystack: "zzzzzzzzzzzzzzzza", needles: &[b'a'], positions: &[16] }, + Seed { + haystack: "zzzzzzzzzzzzzzzzzzzzzzzzzzzzzzzza", + needles: &[b'a'], + positions: &[32], + }, + // two needles (applied to memchr2 + memchr3) + Seed { haystack: "az", needles: &[b'a', b'z'], positions: &[0, 1] }, + Seed { haystack: "az", needles: &[b'a', b'z'], positions: &[0, 1] }, + Seed { haystack: "az", needles: &[b'x', b'y'], positions: &[] }, + Seed { haystack: "az", needles: &[b'a', b'y'], positions: &[0] }, + Seed { haystack: "az", needles: &[b'x', b'z'], positions: &[1] }, + Seed { haystack: "yyyyaz", needles: &[b'a', b'z'], positions: &[4, 5] }, + Seed { haystack: "yyyyaz", needles: &[b'z', b'a'], positions: &[4, 5] }, + // three needles (applied to memchr3) + Seed { + haystack: "xyz", + needles: &[b'x', b'y', b'z'], + positions: &[0, 1, 2], + }, + Seed { + haystack: "zxy", + needles: &[b'x', b'y', b'z'], + positions: &[0, 1, 2], + }, + Seed { haystack: "zxy", needles: &[b'x', b'a', b'z'], positions: &[0, 1] }, + Seed { haystack: "zxy", needles: &[b't', b'a', b'z'], positions: &[0] }, + Seed { haystack: "yxz", needles: &[b't', b'a', b'z'], positions: &[2] }, +]; + +/// Runs a host of substring search tests. +/// +/// This has support for "partial" substring search implementations only work +/// for a subset of needles/haystacks. For example, the "packed pair" substring +/// search implementation only works for haystacks of some minimum length based +/// of the pair of bytes selected and the size of the vector used. +pub(crate) struct Runner { + needle_len: usize, +} + +impl Runner { + /// Create a new test runner for forward and reverse byte search + /// implementations. + /// + /// The `needle_len` given must be at most `3` and at least `1`. It + /// corresponds to the number of needle bytes to search for. + pub(crate) fn new(needle_len: usize) -> Runner { + assert!(needle_len >= 1, "needle_len must be at least 1"); + assert!(needle_len <= 3, "needle_len must be at most 3"); + Runner { needle_len } + } + + /// Run all tests. This panics on the first failure. + /// + /// If the implementation being tested returns `None` for a particular + /// haystack/needle combination, then that test is skipped. + pub(crate) fn forward_iter<F>(self, mut test: F) + where + F: FnMut(&[u8], &[u8]) -> Option<Vec<usize>> + 'static, + { + for seed in SEEDS.iter() { + if seed.needles.len() > self.needle_len { + continue; + } + for t in seed.generate() { + let results = match test(t.haystack.as_bytes(), &t.needles) { + None => continue, + Some(results) => results, + }; + assert_eq!( + t.expected, + results, + "needles: {:?}, haystack: {:?}", + t.needles + .iter() + .map(|&b| b.to_char()) + .collect::<Vec<char>>(), + t.haystack, + ); + } + } + } + + /// Run all tests in the reverse direction. This panics on the first + /// failure. + /// + /// If the implementation being tested returns `None` for a particular + /// haystack/needle combination, then that test is skipped. + pub(crate) fn reverse_iter<F>(self, mut test: F) + where + F: FnMut(&[u8], &[u8]) -> Option<Vec<usize>> + 'static, + { + for seed in SEEDS.iter() { + if seed.needles.len() > self.needle_len { + continue; + } + for t in seed.generate() { + let mut results = match test(t.haystack.as_bytes(), &t.needles) + { + None => continue, + Some(results) => results, + }; + results.reverse(); + assert_eq!( + t.expected, + results, + "needles: {:?}, haystack: {:?}", + t.needles + .iter() + .map(|&b| b.to_char()) + .collect::<Vec<char>>(), + t.haystack, + ); + } + } + } + + /// Run all tests as counting tests. This panics on the first failure. + /// + /// That is, this only checks that the number of matches is correct and + /// not whether the offsets of each match are. + pub(crate) fn count_iter<F>(self, mut test: F) + where + F: FnMut(&[u8], &[u8]) -> Option<usize> + 'static, + { + for seed in SEEDS.iter() { + if seed.needles.len() > self.needle_len { + continue; + } + for t in seed.generate() { + let got = match test(t.haystack.as_bytes(), &t.needles) { + None => continue, + Some(got) => got, + }; + assert_eq!( + t.expected.len(), + got, + "needles: {:?}, haystack: {:?}", + t.needles + .iter() + .map(|&b| b.to_char()) + .collect::<Vec<char>>(), + t.haystack, + ); + } + } + } + + /// Like `Runner::forward`, but for a function that returns only the next + /// match and not all matches. + /// + /// If the function returns `None`, then it is skipped. + pub(crate) fn forward_oneshot<F>(self, mut test: F) + where + F: FnMut(&[u8], &[u8]) -> Option<Option<usize>> + 'static, + { + self.forward_iter(move |haystack, needles| { + let mut start = 0; + let mut results = vec![]; + while let Some(i) = test(&haystack[start..], needles)? { + results.push(start + i); + start += i + 1; + } + Some(results) + }) + } + + /// Like `Runner::reverse`, but for a function that returns only the last + /// match and not all matches. + /// + /// If the function returns `None`, then it is skipped. + pub(crate) fn reverse_oneshot<F>(self, mut test: F) + where + F: FnMut(&[u8], &[u8]) -> Option<Option<usize>> + 'static, + { + self.reverse_iter(move |haystack, needles| { + let mut end = haystack.len(); + let mut results = vec![]; + while let Some(i) = test(&haystack[..end], needles)? { + results.push(i); + end = i; + } + Some(results) + }) + } +} + +/// A single test for memr?chr{,2,3}. +#[derive(Clone, Debug)] +struct Test { + /// The string to search in. + haystack: String, + /// The needles to look for. + needles: Vec<u8>, + /// The offsets that are expected to be found for all needles in the + /// forward direction. + expected: Vec<usize>, +} + +impl Test { + fn new(seed: &Seed) -> Test { + Test { + haystack: seed.haystack.to_string(), + needles: seed.needles.to_vec(), + expected: seed.positions.to_vec(), + } + } +} + +/// Data that can be expanded into many memchr tests by padding out the corpus. +#[derive(Clone, Debug)] +struct Seed { + /// The thing to search. We use `&str` instead of `&[u8]` because they + /// are nicer to write in tests, and we don't miss much since memchr + /// doesn't care about UTF-8. + /// + /// Corpora cannot contain either '%' or '#'. We use these bytes when + /// expanding test cases into many test cases, and we assume they are not + /// used. If they are used, `memchr_tests` will panic. + haystack: &'static str, + /// The needles to search for. This is intended to be an alternation of + /// needles. The number of needles may cause this test to be skipped for + /// some memchr variants. For example, a test with 2 needles cannot be used + /// to test `memchr`, but can be used to test `memchr2` and `memchr3`. + /// However, a test with only 1 needle can be used to test all of `memchr`, + /// `memchr2` and `memchr3`. We achieve this by filling in the needles with + /// bytes that we never used in the corpus (such as '#'). + needles: &'static [u8], + /// The positions expected to match for all of the needles. + positions: &'static [usize], +} + +impl Seed { + /// Controls how much we expand the haystack on either side for each test. + /// We lower this on Miri because otherwise running the tests would take + /// forever. + const EXPAND_LEN: usize = { + #[cfg(not(miri))] + { + 515 + } + #[cfg(miri)] + { + 6 + } + }; + + /// Expand this test into many variations of the same test. + /// + /// In particular, this will generate more tests with larger corpus sizes. + /// The expected positions are updated to maintain the integrity of the + /// test. + /// + /// This is important in testing a memchr implementation, because there are + /// often different cases depending on the length of the corpus. + /// + /// Note that we extend the corpus by adding `%` bytes, which we + /// don't otherwise use as a needle. + fn generate(&self) -> impl Iterator<Item = Test> { + let mut more = vec![]; + + // Add bytes to the start of the corpus. + for i in 0..Seed::EXPAND_LEN { + let mut t = Test::new(self); + let mut new: String = core::iter::repeat('%').take(i).collect(); + new.push_str(&t.haystack); + t.haystack = new; + t.expected = t.expected.into_iter().map(|p| p + i).collect(); + more.push(t); + } + // Add bytes to the end of the corpus. + for i in 1..Seed::EXPAND_LEN { + let mut t = Test::new(self); + let padding: String = core::iter::repeat('%').take(i).collect(); + t.haystack.push_str(&padding); + more.push(t); + } + + more.into_iter() + } +} diff --git a/vendor/memchr/src/tests/memchr/naive.rs b/vendor/memchr/src/tests/memchr/naive.rs new file mode 100644 index 0000000..6ebcdae --- /dev/null +++ b/vendor/memchr/src/tests/memchr/naive.rs @@ -0,0 +1,33 @@ +pub(crate) fn memchr(n1: u8, haystack: &[u8]) -> Option<usize> { + haystack.iter().position(|&b| b == n1) +} + +pub(crate) fn memchr2(n1: u8, n2: u8, haystack: &[u8]) -> Option<usize> { + haystack.iter().position(|&b| b == n1 || b == n2) +} + +pub(crate) fn memchr3( + n1: u8, + n2: u8, + n3: u8, + haystack: &[u8], +) -> Option<usize> { + haystack.iter().position(|&b| b == n1 || b == n2 || b == n3) +} + +pub(crate) fn memrchr(n1: u8, haystack: &[u8]) -> Option<usize> { + haystack.iter().rposition(|&b| b == n1) +} + +pub(crate) fn memrchr2(n1: u8, n2: u8, haystack: &[u8]) -> Option<usize> { + haystack.iter().rposition(|&b| b == n1 || b == n2) +} + +pub(crate) fn memrchr3( + n1: u8, + n2: u8, + n3: u8, + haystack: &[u8], +) -> Option<usize> { + haystack.iter().rposition(|&b| b == n1 || b == n2 || b == n3) +} diff --git a/vendor/memchr/src/tests/memchr/prop.rs b/vendor/memchr/src/tests/memchr/prop.rs new file mode 100644 index 0000000..b988260 --- /dev/null +++ b/vendor/memchr/src/tests/memchr/prop.rs @@ -0,0 +1,321 @@ +#[cfg(miri)] +#[macro_export] +macro_rules! define_memchr_quickcheck { + ($($tt:tt)*) => {}; +} + +#[cfg(not(miri))] +#[macro_export] +macro_rules! define_memchr_quickcheck { + ($mod:ident) => { + define_memchr_quickcheck!($mod, new); + }; + ($mod:ident, $cons:ident) => { + use alloc::vec::Vec; + + use quickcheck::TestResult; + + use crate::tests::memchr::{ + naive, + prop::{double_ended_take, naive1_iter, naive2_iter, naive3_iter}, + }; + + quickcheck::quickcheck! { + fn qc_memchr_matches_naive(n1: u8, corpus: Vec<u8>) -> TestResult { + let expected = naive::memchr(n1, &corpus); + let got = match $mod::One::$cons(n1) { + None => return TestResult::discard(), + Some(f) => f.find(&corpus), + }; + TestResult::from_bool(expected == got) + } + + fn qc_memrchr_matches_naive(n1: u8, corpus: Vec<u8>) -> TestResult { + let expected = naive::memrchr(n1, &corpus); + let got = match $mod::One::$cons(n1) { + None => return TestResult::discard(), + Some(f) => f.rfind(&corpus), + }; + TestResult::from_bool(expected == got) + } + + fn qc_memchr2_matches_naive(n1: u8, n2: u8, corpus: Vec<u8>) -> TestResult { + let expected = naive::memchr2(n1, n2, &corpus); + let got = match $mod::Two::$cons(n1, n2) { + None => return TestResult::discard(), + Some(f) => f.find(&corpus), + }; + TestResult::from_bool(expected == got) + } + + fn qc_memrchr2_matches_naive(n1: u8, n2: u8, corpus: Vec<u8>) -> TestResult { + let expected = naive::memrchr2(n1, n2, &corpus); + let got = match $mod::Two::$cons(n1, n2) { + None => return TestResult::discard(), + Some(f) => f.rfind(&corpus), + }; + TestResult::from_bool(expected == got) + } + + fn qc_memchr3_matches_naive( + n1: u8, n2: u8, n3: u8, + corpus: Vec<u8> + ) -> TestResult { + let expected = naive::memchr3(n1, n2, n3, &corpus); + let got = match $mod::Three::$cons(n1, n2, n3) { + None => return TestResult::discard(), + Some(f) => f.find(&corpus), + }; + TestResult::from_bool(expected == got) + } + + fn qc_memrchr3_matches_naive( + n1: u8, n2: u8, n3: u8, + corpus: Vec<u8> + ) -> TestResult { + let expected = naive::memrchr3(n1, n2, n3, &corpus); + let got = match $mod::Three::$cons(n1, n2, n3) { + None => return TestResult::discard(), + Some(f) => f.rfind(&corpus), + }; + TestResult::from_bool(expected == got) + } + + fn qc_memchr_double_ended_iter( + needle: u8, data: Vec<u8>, take_side: Vec<bool> + ) -> TestResult { + // make nonempty + let mut take_side = take_side; + if take_side.is_empty() { take_side.push(true) }; + + let finder = match $mod::One::$cons(needle) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let iter = finder.iter(&data); + let got = double_ended_take( + iter, + take_side.iter().cycle().cloned(), + ); + let expected = naive1_iter(needle, &data); + + TestResult::from_bool(got.iter().cloned().eq(expected)) + } + + fn qc_memchr2_double_ended_iter( + needle1: u8, needle2: u8, data: Vec<u8>, take_side: Vec<bool> + ) -> TestResult { + // make nonempty + let mut take_side = take_side; + if take_side.is_empty() { take_side.push(true) }; + + let finder = match $mod::Two::$cons(needle1, needle2) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let iter = finder.iter(&data); + let got = double_ended_take( + iter, + take_side.iter().cycle().cloned(), + ); + let expected = naive2_iter(needle1, needle2, &data); + + TestResult::from_bool(got.iter().cloned().eq(expected)) + } + + fn qc_memchr3_double_ended_iter( + needle1: u8, needle2: u8, needle3: u8, + data: Vec<u8>, take_side: Vec<bool> + ) -> TestResult { + // make nonempty + let mut take_side = take_side; + if take_side.is_empty() { take_side.push(true) }; + + let finder = match $mod::Three::$cons(needle1, needle2, needle3) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let iter = finder.iter(&data); + let got = double_ended_take( + iter, + take_side.iter().cycle().cloned(), + ); + let expected = naive3_iter(needle1, needle2, needle3, &data); + + TestResult::from_bool(got.iter().cloned().eq(expected)) + } + + fn qc_memchr1_iter(data: Vec<u8>) -> TestResult { + let needle = 0; + let finder = match $mod::One::$cons(needle) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let got = finder.iter(&data); + let expected = naive1_iter(needle, &data); + TestResult::from_bool(got.eq(expected)) + } + + fn qc_memchr1_rev_iter(data: Vec<u8>) -> TestResult { + let needle = 0; + + let finder = match $mod::One::$cons(needle) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let got = finder.iter(&data).rev(); + let expected = naive1_iter(needle, &data).rev(); + TestResult::from_bool(got.eq(expected)) + } + + fn qc_memchr2_iter(data: Vec<u8>) -> TestResult { + let needle1 = 0; + let needle2 = 1; + + let finder = match $mod::Two::$cons(needle1, needle2) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let got = finder.iter(&data); + let expected = naive2_iter(needle1, needle2, &data); + TestResult::from_bool(got.eq(expected)) + } + + fn qc_memchr2_rev_iter(data: Vec<u8>) -> TestResult { + let needle1 = 0; + let needle2 = 1; + + let finder = match $mod::Two::$cons(needle1, needle2) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let got = finder.iter(&data).rev(); + let expected = naive2_iter(needle1, needle2, &data).rev(); + TestResult::from_bool(got.eq(expected)) + } + + fn qc_memchr3_iter(data: Vec<u8>) -> TestResult { + let needle1 = 0; + let needle2 = 1; + let needle3 = 2; + + let finder = match $mod::Three::$cons(needle1, needle2, needle3) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let got = finder.iter(&data); + let expected = naive3_iter(needle1, needle2, needle3, &data); + TestResult::from_bool(got.eq(expected)) + } + + fn qc_memchr3_rev_iter(data: Vec<u8>) -> TestResult { + let needle1 = 0; + let needle2 = 1; + let needle3 = 2; + + let finder = match $mod::Three::$cons(needle1, needle2, needle3) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let got = finder.iter(&data).rev(); + let expected = naive3_iter(needle1, needle2, needle3, &data).rev(); + TestResult::from_bool(got.eq(expected)) + } + + fn qc_memchr1_iter_size_hint(data: Vec<u8>) -> TestResult { + // test that the size hint is within reasonable bounds + let needle = 0; + let finder = match $mod::One::$cons(needle) { + None => return TestResult::discard(), + Some(finder) => finder, + }; + let mut iter = finder.iter(&data); + let mut real_count = data + .iter() + .filter(|&&elt| elt == needle) + .count(); + + while let Some(index) = iter.next() { + real_count -= 1; + let (lower, upper) = iter.size_hint(); + assert!(lower <= real_count); + assert!(upper.unwrap() >= real_count); + assert!(upper.unwrap() <= data.len() - index); + } + TestResult::passed() + } + } + }; +} + +// take items from a DEI, taking front for each true and back for each false. +// Return a vector with the concatenation of the fronts and the reverse of the +// backs. +#[cfg(not(miri))] +pub(crate) fn double_ended_take<I, J>( + mut iter: I, + take_side: J, +) -> alloc::vec::Vec<I::Item> +where + I: DoubleEndedIterator, + J: Iterator<Item = bool>, +{ + let mut found_front = alloc::vec![]; + let mut found_back = alloc::vec![]; + + for take_front in take_side { + if take_front { + if let Some(pos) = iter.next() { + found_front.push(pos); + } else { + break; + } + } else { + if let Some(pos) = iter.next_back() { + found_back.push(pos); + } else { + break; + } + }; + } + + let mut all_found = found_front; + all_found.extend(found_back.into_iter().rev()); + all_found +} + +// return an iterator of the 0-based indices of haystack that match the needle +#[cfg(not(miri))] +pub(crate) fn naive1_iter<'a>( + n1: u8, + haystack: &'a [u8], +) -> impl DoubleEndedIterator<Item = usize> + 'a { + haystack.iter().enumerate().filter(move |&(_, &b)| b == n1).map(|t| t.0) +} + +#[cfg(not(miri))] +pub(crate) fn naive2_iter<'a>( + n1: u8, + n2: u8, + haystack: &'a [u8], +) -> impl DoubleEndedIterator<Item = usize> + 'a { + haystack + .iter() + .enumerate() + .filter(move |&(_, &b)| b == n1 || b == n2) + .map(|t| t.0) +} + +#[cfg(not(miri))] +pub(crate) fn naive3_iter<'a>( + n1: u8, + n2: u8, + n3: u8, + haystack: &'a [u8], +) -> impl DoubleEndedIterator<Item = usize> + 'a { + haystack + .iter() + .enumerate() + .filter(move |&(_, &b)| b == n1 || b == n2 || b == n3) + .map(|t| t.0) +} diff --git a/vendor/memchr/src/tests/mod.rs b/vendor/memchr/src/tests/mod.rs new file mode 100644 index 0000000..259b678 --- /dev/null +++ b/vendor/memchr/src/tests/mod.rs @@ -0,0 +1,15 @@ +#[macro_use] +pub(crate) mod memchr; +pub(crate) mod packedpair; +#[macro_use] +pub(crate) mod substring; + +// For debugging, particularly in CI, print out the byte order of the current +// target. +#[test] +fn byte_order() { + #[cfg(target_endian = "little")] + std::eprintln!("LITTLE ENDIAN"); + #[cfg(target_endian = "big")] + std::eprintln!("BIG ENDIAN"); +} diff --git a/vendor/memchr/src/tests/packedpair.rs b/vendor/memchr/src/tests/packedpair.rs new file mode 100644 index 0000000..204635b --- /dev/null +++ b/vendor/memchr/src/tests/packedpair.rs @@ -0,0 +1,216 @@ +use alloc::{boxed::Box, vec, vec::Vec}; + +/// A set of "packed pair" test seeds. Each seed serves as the base for the +/// generation of many other tests. In essence, the seed captures the pair of +/// bytes we used for a predicate and first byte among our needle. The tests +/// generated from each seed essentially vary the length of the needle and +/// haystack, while using the rare/first byte configuration from the seed. +/// +/// The purpose of this is to test many different needle/haystack lengths. +/// In particular, some of the vector optimizations might only have bugs +/// in haystacks of a certain size. +const SEEDS: &[Seed] = &[ + // Why not use different 'first' bytes? It seemed like a good idea to be + // able to configure it, but when I wrote the test generator below, it + // didn't seem necessary to use for reasons that I forget. + Seed { first: b'x', index1: b'y', index2: b'z' }, + Seed { first: b'x', index1: b'x', index2: b'z' }, + Seed { first: b'x', index1: b'y', index2: b'x' }, + Seed { first: b'x', index1: b'x', index2: b'x' }, + Seed { first: b'x', index1: b'y', index2: b'y' }, +]; + +/// Runs a host of "packed pair" search tests. +/// +/// These tests specifically look for the occurrence of a possible substring +/// match based on a pair of bytes matching at the right offsets. +pub(crate) struct Runner { + fwd: Option< + Box< + dyn FnMut(&[u8], &[u8], u8, u8) -> Option<Option<usize>> + 'static, + >, + >, +} + +impl Runner { + /// Create a new test runner for "packed pair" substring search. + pub(crate) fn new() -> Runner { + Runner { fwd: None } + } + + /// Run all tests. This panics on the first failure. + /// + /// If the implementation being tested returns `None` for a particular + /// haystack/needle combination, then that test is skipped. + /// + /// This runs tests on both the forward and reverse implementations given. + /// If either (or both) are missing, then tests for that implementation are + /// skipped. + pub(crate) fn run(self) { + if let Some(mut fwd) = self.fwd { + for seed in SEEDS.iter() { + for t in seed.generate() { + match fwd(&t.haystack, &t.needle, t.index1, t.index2) { + None => continue, + Some(result) => { + assert_eq!( + t.fwd, result, + "FORWARD, needle: {:?}, haystack: {:?}, \ + index1: {:?}, index2: {:?}", + t.needle, t.haystack, t.index1, t.index2, + ) + } + } + } + } + } + } + + /// Set the implementation for forward "packed pair" substring search. + /// + /// If the closure returns `None`, then it is assumed that the given + /// test cannot be applied to the particular implementation and it is + /// skipped. For example, if a particular implementation only supports + /// needles or haystacks for some minimum length. + /// + /// If this is not set, then forward "packed pair" search is not tested. + pub(crate) fn fwd( + mut self, + search: impl FnMut(&[u8], &[u8], u8, u8) -> Option<Option<usize>> + 'static, + ) -> Runner { + self.fwd = Some(Box::new(search)); + self + } +} + +/// A test that represents the input and expected output to a "packed pair" +/// search function. The test should be able to run with any "packed pair" +/// implementation and get the expected output. +struct Test { + haystack: Vec<u8>, + needle: Vec<u8>, + index1: u8, + index2: u8, + fwd: Option<usize>, +} + +impl Test { + /// Create a new "packed pair" test from a seed and some given offsets to + /// the pair of bytes to use as a predicate in the seed's needle. + /// + /// If a valid test could not be constructed, then None is returned. + /// (Currently, we take the approach of massaging tests to be valid + /// instead of rejecting them outright.) + fn new( + seed: Seed, + index1: usize, + index2: usize, + haystack_len: usize, + needle_len: usize, + fwd: Option<usize>, + ) -> Option<Test> { + let mut index1: u8 = index1.try_into().unwrap(); + let mut index2: u8 = index2.try_into().unwrap(); + // The '#' byte is never used in a haystack (unless we're expecting + // a match), while the '@' byte is never used in a needle. + let mut haystack = vec![b'@'; haystack_len]; + let mut needle = vec![b'#'; needle_len]; + needle[0] = seed.first; + needle[index1 as usize] = seed.index1; + needle[index2 as usize] = seed.index2; + // If we're expecting a match, then make sure the needle occurs + // in the haystack at the expected position. + if let Some(i) = fwd { + haystack[i..i + needle.len()].copy_from_slice(&needle); + } + // If the operations above lead to rare offsets pointing to the + // non-first occurrence of a byte, then adjust it. This might lead + // to redundant tests, but it's simpler than trying to change the + // generation process I think. + if let Some(i) = crate::memchr(seed.index1, &needle) { + index1 = u8::try_from(i).unwrap(); + } + if let Some(i) = crate::memchr(seed.index2, &needle) { + index2 = u8::try_from(i).unwrap(); + } + Some(Test { haystack, needle, index1, index2, fwd }) + } +} + +/// Data that describes a single prefilter test seed. +#[derive(Clone, Copy)] +struct Seed { + first: u8, + index1: u8, + index2: u8, +} + +impl Seed { + const NEEDLE_LENGTH_LIMIT: usize = { + #[cfg(not(miri))] + { + 33 + } + #[cfg(miri)] + { + 5 + } + }; + + const HAYSTACK_LENGTH_LIMIT: usize = { + #[cfg(not(miri))] + { + 65 + } + #[cfg(miri)] + { + 8 + } + }; + + /// Generate a series of prefilter tests from this seed. + fn generate(self) -> impl Iterator<Item = Test> { + let len_start = 2; + // The iterator below generates *a lot* of tests. The number of + // tests was chosen somewhat empirically to be "bearable" when + // running the test suite. + // + // We use an iterator here because the collective haystacks of all + // these test cases add up to enough memory to OOM a conservative + // sandbox or a small laptop. + (len_start..=Seed::NEEDLE_LENGTH_LIMIT).flat_map(move |needle_len| { + let index_start = len_start - 1; + (index_start..needle_len).flat_map(move |index1| { + (index1..needle_len).flat_map(move |index2| { + (needle_len..=Seed::HAYSTACK_LENGTH_LIMIT).flat_map( + move |haystack_len| { + Test::new( + self, + index1, + index2, + haystack_len, + needle_len, + None, + ) + .into_iter() + .chain( + (0..=(haystack_len - needle_len)).flat_map( + move |output| { + Test::new( + self, + index1, + index2, + haystack_len, + needle_len, + Some(output), + ) + }, + ), + ) + }, + ) + }) + }) + }) + } +} diff --git a/vendor/memchr/src/tests/substring/mod.rs b/vendor/memchr/src/tests/substring/mod.rs new file mode 100644 index 0000000..dd10cbd --- /dev/null +++ b/vendor/memchr/src/tests/substring/mod.rs @@ -0,0 +1,232 @@ +/*! +This module defines tests and test helpers for substring implementations. +*/ + +use alloc::{ + boxed::Box, + format, + string::{String, ToString}, +}; + +pub(crate) mod naive; +#[macro_use] +pub(crate) mod prop; + +const SEEDS: &'static [Seed] = &[ + Seed::new("", "", Some(0), Some(0)), + Seed::new("", "a", Some(0), Some(1)), + Seed::new("", "ab", Some(0), Some(2)), + Seed::new("", "abc", Some(0), Some(3)), + Seed::new("a", "", None, None), + Seed::new("a", "a", Some(0), Some(0)), + Seed::new("a", "aa", Some(0), Some(1)), + Seed::new("a", "ba", Some(1), Some(1)), + Seed::new("a", "bba", Some(2), Some(2)), + Seed::new("a", "bbba", Some(3), Some(3)), + Seed::new("a", "bbbab", Some(3), Some(3)), + Seed::new("a", "bbbabb", Some(3), Some(3)), + Seed::new("a", "bbbabbb", Some(3), Some(3)), + Seed::new("a", "bbbbbb", None, None), + Seed::new("ab", "", None, None), + Seed::new("ab", "a", None, None), + Seed::new("ab", "b", None, None), + Seed::new("ab", "ab", Some(0), Some(0)), + Seed::new("ab", "aab", Some(1), Some(1)), + Seed::new("ab", "aaab", Some(2), Some(2)), + Seed::new("ab", "abaab", Some(0), Some(3)), + Seed::new("ab", "baaab", Some(3), Some(3)), + Seed::new("ab", "acb", None, None), + Seed::new("ab", "abba", Some(0), Some(0)), + Seed::new("abc", "ab", None, None), + Seed::new("abc", "abc", Some(0), Some(0)), + Seed::new("abc", "abcz", Some(0), Some(0)), + Seed::new("abc", "abczz", Some(0), Some(0)), + Seed::new("abc", "zabc", Some(1), Some(1)), + Seed::new("abc", "zzabc", Some(2), Some(2)), + Seed::new("abc", "azbc", None, None), + Seed::new("abc", "abzc", None, None), + Seed::new("abczdef", "abczdefzzzzzzzzzzzzzzzzzzzz", Some(0), Some(0)), + Seed::new("abczdef", "zzzzzzzzzzzzzzzzzzzzabczdef", Some(20), Some(20)), + Seed::new( + "xyz", + "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaxyz", + Some(32), + Some(32), + ), + Seed::new("\u{0}\u{15}", "\u{0}\u{15}\u{15}\u{0}", Some(0), Some(0)), + Seed::new("\u{0}\u{1e}", "\u{1e}\u{0}", None, None), +]; + +/// Runs a host of substring search tests. +/// +/// This has support for "partial" substring search implementations only work +/// for a subset of needles/haystacks. For example, the "packed pair" substring +/// search implementation only works for haystacks of some minimum length based +/// of the pair of bytes selected and the size of the vector used. +pub(crate) struct Runner { + fwd: Option< + Box<dyn FnMut(&[u8], &[u8]) -> Option<Option<usize>> + 'static>, + >, + rev: Option< + Box<dyn FnMut(&[u8], &[u8]) -> Option<Option<usize>> + 'static>, + >, +} + +impl Runner { + /// Create a new test runner for forward and reverse substring search + /// implementations. + pub(crate) fn new() -> Runner { + Runner { fwd: None, rev: None } + } + + /// Run all tests. This panics on the first failure. + /// + /// If the implementation being tested returns `None` for a particular + /// haystack/needle combination, then that test is skipped. + /// + /// This runs tests on both the forward and reverse implementations given. + /// If either (or both) are missing, then tests for that implementation are + /// skipped. + pub(crate) fn run(self) { + if let Some(mut fwd) = self.fwd { + for seed in SEEDS.iter() { + for t in seed.generate() { + match fwd(t.haystack.as_bytes(), t.needle.as_bytes()) { + None => continue, + Some(result) => { + assert_eq!( + t.fwd, result, + "FORWARD, needle: {:?}, haystack: {:?}", + t.needle, t.haystack, + ); + } + } + } + } + } + if let Some(mut rev) = self.rev { + for seed in SEEDS.iter() { + for t in seed.generate() { + match rev(t.haystack.as_bytes(), t.needle.as_bytes()) { + None => continue, + Some(result) => { + assert_eq!( + t.rev, result, + "REVERSE, needle: {:?}, haystack: {:?}", + t.needle, t.haystack, + ); + } + } + } + } + } + } + + /// Set the implementation for forward substring search. + /// + /// If the closure returns `None`, then it is assumed that the given + /// test cannot be applied to the particular implementation and it is + /// skipped. For example, if a particular implementation only supports + /// needles or haystacks for some minimum length. + /// + /// If this is not set, then forward substring search is not tested. + pub(crate) fn fwd( + mut self, + search: impl FnMut(&[u8], &[u8]) -> Option<Option<usize>> + 'static, + ) -> Runner { + self.fwd = Some(Box::new(search)); + self + } + + /// Set the implementation for reverse substring search. + /// + /// If the closure returns `None`, then it is assumed that the given + /// test cannot be applied to the particular implementation and it is + /// skipped. For example, if a particular implementation only supports + /// needles or haystacks for some minimum length. + /// + /// If this is not set, then reverse substring search is not tested. + pub(crate) fn rev( + mut self, + search: impl FnMut(&[u8], &[u8]) -> Option<Option<usize>> + 'static, + ) -> Runner { + self.rev = Some(Box::new(search)); + self + } +} + +/// A single substring test for forward and reverse searches. +#[derive(Clone, Debug)] +struct Test { + needle: String, + haystack: String, + fwd: Option<usize>, + rev: Option<usize>, +} + +/// A single substring test for forward and reverse searches. +/// +/// Each seed is valid on its own, but it also serves as a starting point +/// to generate more tests. Namely, we pad out the haystacks with other +/// characters so that we get more complete coverage. This is especially useful +/// for testing vector algorithms that tend to have weird special cases for +/// alignment and loop unrolling. +/// +/// Padding works by assuming certain characters never otherwise appear in a +/// needle or a haystack. Neither should contain a `#` character. +#[derive(Clone, Copy, Debug)] +struct Seed { + needle: &'static str, + haystack: &'static str, + fwd: Option<usize>, + rev: Option<usize>, +} + +impl Seed { + const MAX_PAD: usize = 34; + + const fn new( + needle: &'static str, + haystack: &'static str, + fwd: Option<usize>, + rev: Option<usize>, + ) -> Seed { + Seed { needle, haystack, fwd, rev } + } + + fn generate(self) -> impl Iterator<Item = Test> { + assert!(!self.needle.contains('#'), "needle must not contain '#'"); + assert!(!self.haystack.contains('#'), "haystack must not contain '#'"); + (0..=Seed::MAX_PAD) + // Generate tests for padding at the beginning of haystack. + .map(move |pad| { + let needle = self.needle.to_string(); + let prefix = "#".repeat(pad); + let haystack = format!("{}{}", prefix, self.haystack); + let fwd = if needle.is_empty() { + Some(0) + } else { + self.fwd.map(|i| pad + i) + }; + let rev = if needle.is_empty() { + Some(haystack.len()) + } else { + self.rev.map(|i| pad + i) + }; + Test { needle, haystack, fwd, rev } + }) + // Generate tests for padding at the end of haystack. + .chain((1..=Seed::MAX_PAD).map(move |pad| { + let needle = self.needle.to_string(); + let suffix = "#".repeat(pad); + let haystack = format!("{}{}", self.haystack, suffix); + let fwd = if needle.is_empty() { Some(0) } else { self.fwd }; + let rev = if needle.is_empty() { + Some(haystack.len()) + } else { + self.rev + }; + Test { needle, haystack, fwd, rev } + })) + } +} diff --git a/vendor/memchr/src/tests/substring/naive.rs b/vendor/memchr/src/tests/substring/naive.rs new file mode 100644 index 0000000..1bc6009 --- /dev/null +++ b/vendor/memchr/src/tests/substring/naive.rs @@ -0,0 +1,45 @@ +/*! +This module defines "naive" implementations of substring search. + +These are sometimes useful to compare with "real" substring implementations. +The idea is that they are so simple that they are unlikely to be incorrect. +*/ + +/// Naively search forwards for the given needle in the given haystack. +pub(crate) fn find(haystack: &[u8], needle: &[u8]) -> Option<usize> { + let end = haystack.len().checked_sub(needle.len()).map_or(0, |i| i + 1); + for i in 0..end { + if needle == &haystack[i..i + needle.len()] { + return Some(i); + } + } + None +} + +/// Naively search in reverse for the given needle in the given haystack. +pub(crate) fn rfind(haystack: &[u8], needle: &[u8]) -> Option<usize> { + let end = haystack.len().checked_sub(needle.len()).map_or(0, |i| i + 1); + for i in (0..end).rev() { + if needle == &haystack[i..i + needle.len()] { + return Some(i); + } + } + None +} + +#[cfg(test)] +mod tests { + use crate::tests::substring; + + use super::*; + + #[test] + fn forward() { + substring::Runner::new().fwd(|h, n| Some(find(h, n))).run() + } + + #[test] + fn reverse() { + substring::Runner::new().rev(|h, n| Some(rfind(h, n))).run() + } +} diff --git a/vendor/memchr/src/tests/substring/prop.rs b/vendor/memchr/src/tests/substring/prop.rs new file mode 100644 index 0000000..a8352ec --- /dev/null +++ b/vendor/memchr/src/tests/substring/prop.rs @@ -0,0 +1,126 @@ +/*! +This module defines a few quickcheck properties for substring search. + +It also provides a forward and reverse macro for conveniently defining +quickcheck tests that run these properties over any substring search +implementation. +*/ + +use crate::tests::substring::naive; + +/// $fwd is a `impl FnMut(haystack, needle) -> Option<Option<usize>>`. When the +/// routine returns `None`, then it's skipped, which is useful for substring +/// implementations that don't work for all inputs. +#[macro_export] +macro_rules! define_substring_forward_quickcheck { + ($fwd:expr) => { + #[cfg(not(miri))] + quickcheck::quickcheck! { + fn qc_fwd_prefix_is_substring(bs: alloc::vec::Vec<u8>) -> bool { + crate::tests::substring::prop::prefix_is_substring(&bs, $fwd) + } + + fn qc_fwd_suffix_is_substring(bs: alloc::vec::Vec<u8>) -> bool { + crate::tests::substring::prop::suffix_is_substring(&bs, $fwd) + } + + fn qc_fwd_matches_naive( + haystack: alloc::vec::Vec<u8>, + needle: alloc::vec::Vec<u8> + ) -> bool { + crate::tests::substring::prop::same_as_naive( + false, + &haystack, + &needle, + $fwd, + ) + } + } + }; +} + +/// $rev is a `impl FnMut(haystack, needle) -> Option<Option<usize>>`. When the +/// routine returns `None`, then it's skipped, which is useful for substring +/// implementations that don't work for all inputs. +#[macro_export] +macro_rules! define_substring_reverse_quickcheck { + ($rev:expr) => { + #[cfg(not(miri))] + quickcheck::quickcheck! { + fn qc_rev_prefix_is_substring(bs: alloc::vec::Vec<u8>) -> bool { + crate::tests::substring::prop::prefix_is_substring(&bs, $rev) + } + + fn qc_rev_suffix_is_substring(bs: alloc::vec::Vec<u8>) -> bool { + crate::tests::substring::prop::suffix_is_substring(&bs, $rev) + } + + fn qc_rev_matches_naive( + haystack: alloc::vec::Vec<u8>, + needle: alloc::vec::Vec<u8> + ) -> bool { + crate::tests::substring::prop::same_as_naive( + true, + &haystack, + &needle, + $rev, + ) + } + } + }; +} + +/// Check that every prefix of the given byte string is a substring. +pub(crate) fn prefix_is_substring( + bs: &[u8], + mut search: impl FnMut(&[u8], &[u8]) -> Option<Option<usize>>, +) -> bool { + for i in 0..bs.len().saturating_sub(1) { + let prefix = &bs[..i]; + let result = match search(bs, prefix) { + None => continue, + Some(result) => result, + }; + if !result.is_some() { + return false; + } + } + true +} + +/// Check that every suffix of the given byte string is a substring. +pub(crate) fn suffix_is_substring( + bs: &[u8], + mut search: impl FnMut(&[u8], &[u8]) -> Option<Option<usize>>, +) -> bool { + for i in 0..bs.len().saturating_sub(1) { + let suffix = &bs[i..]; + let result = match search(bs, suffix) { + None => continue, + Some(result) => result, + }; + if !result.is_some() { + return false; + } + } + true +} + +/// Check that naive substring search matches the result of the given search +/// algorithm. +pub(crate) fn same_as_naive( + reverse: bool, + haystack: &[u8], + needle: &[u8], + mut search: impl FnMut(&[u8], &[u8]) -> Option<Option<usize>>, +) -> bool { + let result = match search(haystack, needle) { + None => return true, + Some(result) => result, + }; + if reverse { + result == naive::rfind(haystack, needle) + } else { + result == naive::find(haystack, needle) + } +} diff --git a/vendor/memchr/src/tests/x86_64-soft_float.json b/vendor/memchr/src/tests/x86_64-soft_float.json new file mode 100644 index 0000000..b77649e --- /dev/null +++ b/vendor/memchr/src/tests/x86_64-soft_float.json @@ -0,0 +1,15 @@ +{ + "llvm-target": "x86_64-unknown-none", + "target-endian": "little", + "target-pointer-width": "64", + "target-c-int-width": "32", + "os": "none", + "arch": "x86_64", + "data-layout": "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128", + "linker-flavor": "ld.lld", + "linker": "rust-lld", + "features": "-mmx,-sse,-sse2,-sse3,-ssse3,-sse4.1,-sse4.2,-3dnow,-3dnowa,-avx,-avx2,+soft-float", + "executables": true, + "disable-redzone": true, + "panic-strategy": "abort" +} diff --git a/vendor/memchr/src/vector.rs b/vendor/memchr/src/vector.rs new file mode 100644 index 0000000..f360176 --- /dev/null +++ b/vendor/memchr/src/vector.rs @@ -0,0 +1,515 @@ +/// A trait for describing vector operations used by vectorized searchers. +/// +/// The trait is highly constrained to low level vector operations needed. +/// In general, it was invented mostly to be generic over x86's __m128i and +/// __m256i types. At time of writing, it also supports wasm and aarch64 +/// 128-bit vector types as well. +/// +/// # Safety +/// +/// All methods are not safe since they are intended to be implemented using +/// vendor intrinsics, which are also not safe. Callers must ensure that the +/// appropriate target features are enabled in the calling function, and that +/// the current CPU supports them. All implementations should avoid marking the +/// routines with #[target_feature] and instead mark them as #[inline(always)] +/// to ensure they get appropriately inlined. (inline(always) cannot be used +/// with target_feature.) +pub(crate) trait Vector: Copy + core::fmt::Debug { + /// The number of bits in the vector. + const BITS: usize; + /// The number of bytes in the vector. That is, this is the size of the + /// vector in memory. + const BYTES: usize; + /// The bits that must be zero in order for a `*const u8` pointer to be + /// correctly aligned to read vector values. + const ALIGN: usize; + + /// The type of the value returned by `Vector::movemask`. + /// + /// This supports abstracting over the specific representation used in + /// order to accommodate different representations in different ISAs. + type Mask: MoveMask; + + /// Create a vector with 8-bit lanes with the given byte repeated into each + /// lane. + unsafe fn splat(byte: u8) -> Self; + + /// Read a vector-size number of bytes from the given pointer. The pointer + /// must be aligned to the size of the vector. + /// + /// # Safety + /// + /// Callers must guarantee that at least `BYTES` bytes are readable from + /// `data` and that `data` is aligned to a `BYTES` boundary. + unsafe fn load_aligned(data: *const u8) -> Self; + + /// Read a vector-size number of bytes from the given pointer. The pointer + /// does not need to be aligned. + /// + /// # Safety + /// + /// Callers must guarantee that at least `BYTES` bytes are readable from + /// `data`. + unsafe fn load_unaligned(data: *const u8) -> Self; + + /// _mm_movemask_epi8 or _mm256_movemask_epi8 + unsafe fn movemask(self) -> Self::Mask; + /// _mm_cmpeq_epi8 or _mm256_cmpeq_epi8 + unsafe fn cmpeq(self, vector2: Self) -> Self; + /// _mm_and_si128 or _mm256_and_si256 + unsafe fn and(self, vector2: Self) -> Self; + /// _mm_or or _mm256_or_si256 + unsafe fn or(self, vector2: Self) -> Self; + /// Returns true if and only if `Self::movemask` would return a mask that + /// contains at least one non-zero bit. + unsafe fn movemask_will_have_non_zero(self) -> bool { + self.movemask().has_non_zero() + } +} + +/// A trait that abstracts over a vector-to-scalar operation called +/// "move mask." +/// +/// On x86-64, this is `_mm_movemask_epi8` for SSE2 and `_mm256_movemask_epi8` +/// for AVX2. It takes a vector of `u8` lanes and returns a scalar where the +/// `i`th bit is set if and only if the most significant bit in the `i`th lane +/// of the vector is set. The simd128 ISA for wasm32 also supports this +/// exact same operation natively. +/// +/// ... But aarch64 doesn't. So we have to fake it with more instructions and +/// a slightly different representation. We could do extra work to unify the +/// representations, but then would require additional costs in the hot path +/// for `memchr` and `packedpair`. So instead, we abstraction over the specific +/// representation with this trait an ddefine the operations we actually need. +pub(crate) trait MoveMask: Copy + core::fmt::Debug { + /// Return a mask that is all zeros except for the least significant `n` + /// lanes in a corresponding vector. + fn all_zeros_except_least_significant(n: usize) -> Self; + + /// Returns true if and only if this mask has a a non-zero bit anywhere. + fn has_non_zero(self) -> bool; + + /// Returns the number of bits set to 1 in this mask. + fn count_ones(self) -> usize; + + /// Does a bitwise `and` operation between `self` and `other`. + fn and(self, other: Self) -> Self; + + /// Does a bitwise `or` operation between `self` and `other`. + fn or(self, other: Self) -> Self; + + /// Returns a mask that is equivalent to `self` but with the least + /// significant 1-bit set to 0. + fn clear_least_significant_bit(self) -> Self; + + /// Returns the offset of the first non-zero lane this mask represents. + fn first_offset(self) -> usize; + + /// Returns the offset of the last non-zero lane this mask represents. + fn last_offset(self) -> usize; +} + +/// This is a "sensible" movemask implementation where each bit represents +/// whether the most significant bit is set in each corresponding lane of a +/// vector. This is used on x86-64 and wasm, but such a mask is more expensive +/// to get on aarch64 so we use something a little different. +/// +/// We call this "sensible" because this is what we get using native sse/avx +/// movemask instructions. But neon has no such native equivalent. +#[derive(Clone, Copy, Debug)] +pub(crate) struct SensibleMoveMask(u32); + +impl SensibleMoveMask { + /// Get the mask in a form suitable for computing offsets. + /// + /// Basically, this normalizes to little endian. On big endian, this swaps + /// the bytes. + #[inline(always)] + fn get_for_offset(self) -> u32 { + #[cfg(target_endian = "big")] + { + self.0.swap_bytes() + } + #[cfg(target_endian = "little")] + { + self.0 + } + } +} + +impl MoveMask for SensibleMoveMask { + #[inline(always)] + fn all_zeros_except_least_significant(n: usize) -> SensibleMoveMask { + debug_assert!(n < 32); + SensibleMoveMask(!((1 << n) - 1)) + } + + #[inline(always)] + fn has_non_zero(self) -> bool { + self.0 != 0 + } + + #[inline(always)] + fn count_ones(self) -> usize { + self.0.count_ones() as usize + } + + #[inline(always)] + fn and(self, other: SensibleMoveMask) -> SensibleMoveMask { + SensibleMoveMask(self.0 & other.0) + } + + #[inline(always)] + fn or(self, other: SensibleMoveMask) -> SensibleMoveMask { + SensibleMoveMask(self.0 | other.0) + } + + #[inline(always)] + fn clear_least_significant_bit(self) -> SensibleMoveMask { + SensibleMoveMask(self.0 & (self.0 - 1)) + } + + #[inline(always)] + fn first_offset(self) -> usize { + // We are dealing with little endian here (and if we aren't, we swap + // the bytes so we are in practice), where the most significant byte + // is at a higher address. That means the least significant bit that + // is set corresponds to the position of our first matching byte. + // That position corresponds to the number of zeros after the least + // significant bit. + self.get_for_offset().trailing_zeros() as usize + } + + #[inline(always)] + fn last_offset(self) -> usize { + // We are dealing with little endian here (and if we aren't, we swap + // the bytes so we are in practice), where the most significant byte is + // at a higher address. That means the most significant bit that is set + // corresponds to the position of our last matching byte. The position + // from the end of the mask is therefore the number of leading zeros + // in a 32 bit integer, and the position from the start of the mask is + // therefore 32 - (leading zeros) - 1. + 32 - self.get_for_offset().leading_zeros() as usize - 1 + } +} + +#[cfg(target_arch = "x86_64")] +mod x86sse2 { + use core::arch::x86_64::*; + + use super::{SensibleMoveMask, Vector}; + + impl Vector for __m128i { + const BITS: usize = 128; + const BYTES: usize = 16; + const ALIGN: usize = Self::BYTES - 1; + + type Mask = SensibleMoveMask; + + #[inline(always)] + unsafe fn splat(byte: u8) -> __m128i { + _mm_set1_epi8(byte as i8) + } + + #[inline(always)] + unsafe fn load_aligned(data: *const u8) -> __m128i { + _mm_load_si128(data as *const __m128i) + } + + #[inline(always)] + unsafe fn load_unaligned(data: *const u8) -> __m128i { + _mm_loadu_si128(data as *const __m128i) + } + + #[inline(always)] + unsafe fn movemask(self) -> SensibleMoveMask { + SensibleMoveMask(_mm_movemask_epi8(self) as u32) + } + + #[inline(always)] + unsafe fn cmpeq(self, vector2: Self) -> __m128i { + _mm_cmpeq_epi8(self, vector2) + } + + #[inline(always)] + unsafe fn and(self, vector2: Self) -> __m128i { + _mm_and_si128(self, vector2) + } + + #[inline(always)] + unsafe fn or(self, vector2: Self) -> __m128i { + _mm_or_si128(self, vector2) + } + } +} + +#[cfg(target_arch = "x86_64")] +mod x86avx2 { + use core::arch::x86_64::*; + + use super::{SensibleMoveMask, Vector}; + + impl Vector for __m256i { + const BITS: usize = 256; + const BYTES: usize = 32; + const ALIGN: usize = Self::BYTES - 1; + + type Mask = SensibleMoveMask; + + #[inline(always)] + unsafe fn splat(byte: u8) -> __m256i { + _mm256_set1_epi8(byte as i8) + } + + #[inline(always)] + unsafe fn load_aligned(data: *const u8) -> __m256i { + _mm256_load_si256(data as *const __m256i) + } + + #[inline(always)] + unsafe fn load_unaligned(data: *const u8) -> __m256i { + _mm256_loadu_si256(data as *const __m256i) + } + + #[inline(always)] + unsafe fn movemask(self) -> SensibleMoveMask { + SensibleMoveMask(_mm256_movemask_epi8(self) as u32) + } + + #[inline(always)] + unsafe fn cmpeq(self, vector2: Self) -> __m256i { + _mm256_cmpeq_epi8(self, vector2) + } + + #[inline(always)] + unsafe fn and(self, vector2: Self) -> __m256i { + _mm256_and_si256(self, vector2) + } + + #[inline(always)] + unsafe fn or(self, vector2: Self) -> __m256i { + _mm256_or_si256(self, vector2) + } + } +} + +#[cfg(target_arch = "aarch64")] +mod aarch64neon { + use core::arch::aarch64::*; + + use super::{MoveMask, Vector}; + + impl Vector for uint8x16_t { + const BITS: usize = 128; + const BYTES: usize = 16; + const ALIGN: usize = Self::BYTES - 1; + + type Mask = NeonMoveMask; + + #[inline(always)] + unsafe fn splat(byte: u8) -> uint8x16_t { + vdupq_n_u8(byte) + } + + #[inline(always)] + unsafe fn load_aligned(data: *const u8) -> uint8x16_t { + // I've tried `data.cast::<uint8x16_t>().read()` instead, but + // couldn't observe any benchmark differences. + Self::load_unaligned(data) + } + + #[inline(always)] + unsafe fn load_unaligned(data: *const u8) -> uint8x16_t { + vld1q_u8(data) + } + + #[inline(always)] + unsafe fn movemask(self) -> NeonMoveMask { + let asu16s = vreinterpretq_u16_u8(self); + let mask = vshrn_n_u16(asu16s, 4); + let asu64 = vreinterpret_u64_u8(mask); + let scalar64 = vget_lane_u64(asu64, 0); + NeonMoveMask(scalar64 & 0x8888888888888888) + } + + #[inline(always)] + unsafe fn cmpeq(self, vector2: Self) -> uint8x16_t { + vceqq_u8(self, vector2) + } + + #[inline(always)] + unsafe fn and(self, vector2: Self) -> uint8x16_t { + vandq_u8(self, vector2) + } + + #[inline(always)] + unsafe fn or(self, vector2: Self) -> uint8x16_t { + vorrq_u8(self, vector2) + } + + /// This is the only interesting implementation of this routine. + /// Basically, instead of doing the "shift right narrow" dance, we use + /// adajacent folding max to determine whether there are any non-zero + /// bytes in our mask. If there are, *then* we'll do the "shift right + /// narrow" dance. In benchmarks, this does lead to slightly better + /// throughput, but the win doesn't appear huge. + #[inline(always)] + unsafe fn movemask_will_have_non_zero(self) -> bool { + let low = vreinterpretq_u64_u8(vpmaxq_u8(self, self)); + vgetq_lane_u64(low, 0) != 0 + } + } + + /// Neon doesn't have a `movemask` that works like the one in x86-64, so we + /// wind up using a different method[1]. The different method also produces + /// a mask, but 4 bits are set in the neon case instead of a single bit set + /// in the x86-64 case. We do an extra step to zero out 3 of the 4 bits, + /// but we still wind up with at least 3 zeroes between each set bit. This + /// generally means that we need to do some division by 4 before extracting + /// offsets. + /// + /// In fact, the existence of this type is the entire reason that we have + /// the `MoveMask` trait in the first place. This basically lets us keep + /// the different representations of masks without being forced to unify + /// them into a single representation, which could result in extra and + /// unnecessary work. + /// + /// [1]: https://community.arm.com/arm-community-blogs/b/infrastructure-solutions-blog/posts/porting-x86-vector-bitmask-optimizations-to-arm-neon + #[derive(Clone, Copy, Debug)] + pub(crate) struct NeonMoveMask(u64); + + impl NeonMoveMask { + /// Get the mask in a form suitable for computing offsets. + /// + /// Basically, this normalizes to little endian. On big endian, this + /// swaps the bytes. + #[inline(always)] + fn get_for_offset(self) -> u64 { + #[cfg(target_endian = "big")] + { + self.0.swap_bytes() + } + #[cfg(target_endian = "little")] + { + self.0 + } + } + } + + impl MoveMask for NeonMoveMask { + #[inline(always)] + fn all_zeros_except_least_significant(n: usize) -> NeonMoveMask { + debug_assert!(n < 16); + NeonMoveMask(!(((1 << n) << 2) - 1)) + } + + #[inline(always)] + fn has_non_zero(self) -> bool { + self.0 != 0 + } + + #[inline(always)] + fn count_ones(self) -> usize { + self.0.count_ones() as usize + } + + #[inline(always)] + fn and(self, other: NeonMoveMask) -> NeonMoveMask { + NeonMoveMask(self.0 & other.0) + } + + #[inline(always)] + fn or(self, other: NeonMoveMask) -> NeonMoveMask { + NeonMoveMask(self.0 | other.0) + } + + #[inline(always)] + fn clear_least_significant_bit(self) -> NeonMoveMask { + NeonMoveMask(self.0 & (self.0 - 1)) + } + + #[inline(always)] + fn first_offset(self) -> usize { + // We are dealing with little endian here (and if we aren't, + // we swap the bytes so we are in practice), where the most + // significant byte is at a higher address. That means the least + // significant bit that is set corresponds to the position of our + // first matching byte. That position corresponds to the number of + // zeros after the least significant bit. + // + // Note that unlike `SensibleMoveMask`, this mask has its bits + // spread out over 64 bits instead of 16 bits (for a 128 bit + // vector). Namely, where as x86-64 will turn + // + // 0x00 0xFF 0x00 0x00 0xFF + // + // into 10010, our neon approach will turn it into + // + // 10000000000010000000 + // + // And this happens because neon doesn't have a native `movemask` + // instruction, so we kind of fake it[1]. Thus, we divide the + // number of trailing zeros by 4 to get the "real" offset. + // + // [1]: https://community.arm.com/arm-community-blogs/b/infrastructure-solutions-blog/posts/porting-x86-vector-bitmask-optimizations-to-arm-neon + (self.get_for_offset().trailing_zeros() >> 2) as usize + } + + #[inline(always)] + fn last_offset(self) -> usize { + // See comment in `first_offset` above. This is basically the same, + // but coming from the other direction. + 16 - (self.get_for_offset().leading_zeros() >> 2) as usize - 1 + } + } +} + +#[cfg(target_arch = "wasm32")] +mod wasm_simd128 { + use core::arch::wasm32::*; + + use super::{SensibleMoveMask, Vector}; + + impl Vector for v128 { + const BITS: usize = 128; + const BYTES: usize = 16; + const ALIGN: usize = Self::BYTES - 1; + + type Mask = SensibleMoveMask; + + #[inline(always)] + unsafe fn splat(byte: u8) -> v128 { + u8x16_splat(byte) + } + + #[inline(always)] + unsafe fn load_aligned(data: *const u8) -> v128 { + *data.cast() + } + + #[inline(always)] + unsafe fn load_unaligned(data: *const u8) -> v128 { + v128_load(data.cast()) + } + + #[inline(always)] + unsafe fn movemask(self) -> SensibleMoveMask { + SensibleMoveMask(u8x16_bitmask(self).into()) + } + + #[inline(always)] + unsafe fn cmpeq(self, vector2: Self) -> v128 { + u8x16_eq(self, vector2) + } + + #[inline(always)] + unsafe fn and(self, vector2: Self) -> v128 { + v128_and(self, vector2) + } + + #[inline(always)] + unsafe fn or(self, vector2: Self) -> v128 { + v128_or(self, vector2) + } + } +} |