diff options
Diffstat (limited to 'vendor/memchr/src/arch/wasm32/simd128')
| -rw-r--r-- | vendor/memchr/src/arch/wasm32/simd128/memchr.rs | 1020 | ||||
| -rw-r--r-- | vendor/memchr/src/arch/wasm32/simd128/mod.rs | 6 | ||||
| -rw-r--r-- | vendor/memchr/src/arch/wasm32/simd128/packedpair.rs | 229 |
3 files changed, 0 insertions, 1255 deletions
diff --git a/vendor/memchr/src/arch/wasm32/simd128/memchr.rs b/vendor/memchr/src/arch/wasm32/simd128/memchr.rs deleted file mode 100644 index fa314c9..0000000 --- a/vendor/memchr/src/arch/wasm32/simd128/memchr.rs +++ /dev/null @@ -1,1020 +0,0 @@ -/*! -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 deleted file mode 100644 index b55d1f0..0000000 --- a/vendor/memchr/src/arch/wasm32/simd128/mod.rs +++ /dev/null @@ -1,6 +0,0 @@ -/*! -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 deleted file mode 100644 index b629377..0000000 --- a/vendor/memchr/src/arch/wasm32/simd128/packedpair.rs +++ /dev/null @@ -1,229 +0,0 @@ -/*! -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() - } -} |