1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311
//! # Clone only when it's necessary //! //! This library provides an efficient way to clone values in a rayon thread pool, but usually //! just once per thread. It cuts down on computation time for potentially expensive cloning //! operations. //! //! Additional clones can rarely occur when rayon schedules execution of another instance of the //! same job, recursively. But in the end, there should not be more than 2N clones, for N threads. //! //! # Examples //! //! ``` //! use rayon_tlsctx::ThreadLocalCtx; //! use rayon::iter::*; //! //! const NUM_COPIES: usize = 16; //! //! let mut buf: Vec<u16> = (0..!0).collect(); //! //! // Create a thread local context with value 0. //! let ctx = ThreadLocalCtx::new(|| { //! // Simulate expensive operation. //! // Since we are building unlocked context, //! // the sleeps will occur concurrently. //! std::thread::sleep_ms(200); //! 0 //! }); //! //! let pool = rayon::ThreadPoolBuilder::new().num_threads(64).build().unwrap(); //! //! // Run inside a custom thread pool. //! pool.install(|| { //! // Sum the buffer `NUM_COPIES` times and accumulate the results //! // into the threaded pool of counts. Note that the counts may be //! // Unevenly distributed. //! (0..NUM_COPIES) //! .into_par_iter() //! .flat_map(|_| buf.par_iter()) //! .for_each(|i| { //! let mut cnt = unsafe { ctx.get() }; //! *cnt += *i as usize; //! }); //! }); //! //! //! let buf_sum = buf.into_iter().fold(0, |acc, i| acc + i as usize); //! //! // What matters is that the final sum matches the expected value. //! assert_eq!(ctx.into_iter().sum::<usize>(), buf_sum * NUM_COPIES); //! ``` use std::cell::Cell; use std::ops::{Deref, DerefMut}; use std::sync::Mutex; /// A thread local storage container for Rayon jobs /// /// This context can be used to efficiently clone `inner`, only when it's necessary. pub struct ThreadLocalCtx<T, F> { inner: F, init_mutex: Mutex<Vec<ThreadLocalNode<T>>>, cloned: Cell<*mut ThreadLocalNode<T>>, } unsafe impl<T, F: Send + Sync> Sync for ThreadLocalCtx<T, F> {} unsafe impl<T, F: Send + Sync> Send for ThreadLocalCtx<T, F> {} impl<T, F: Fn() -> T> ThreadLocalCtx<T, F> { /// Create a new `TlsCtx` /// /// # Examples /// /// Creating a thread-local byte buffer: /// ``` /// use rayon_tlsctx::ThreadLocalCtx; /// let ctx = ThreadLocalCtx::new(Vec::<u8>::new); /// ``` pub fn new(inner: F) -> Self { Self { inner, //: Mutex::new(inner), init_mutex: Mutex::new(vec![]), cloned: Cell::new(std::ptr::null_mut()), } } /// Create a new `TlsCtx`. /// /// This context utilises a lock for cloning values, making it usable for non-sync types. /// /// Cloning an initialised buffer for each thread: /// ``` /// use rayon_tlsctx::ThreadLocalCtx; /// use rayon::iter::*; /// use std::cell::Cell; /// # use rand::prelude::*; /// # let mut rng = rand::thread_rng(); /// /// let mut buf: Vec<u16> = (0..!0).collect(); /// buf.shuffle(&mut rng); /// /// let buf = (buf, Cell::new(0)); /// /// // Must use new_locked, because cloning a cell across threads is not allowed /// let ctx = ThreadLocalCtx::new_locked(move || buf.clone()); /// /// (0..16).into_par_iter().for_each(|_| unsafe { ctx.get(); }); /// ``` pub fn new_locked(inner: F) -> ThreadLocalCtx<T, impl Fn() -> T> { let locked_inner = Mutex::new(inner); let inner = move || (locked_inner.lock().unwrap())(); ThreadLocalCtx { inner, init_mutex: Mutex::new(vec![]), cloned: Cell::new(std::ptr::null_mut()), } } /// Get a thread local context reference with dynamically checked borrow rules /// /// # Remarks /// /// It is advised to manually drop the borrowed context if it is to be reborrowed within inner scope. /// This is because chances of heap allocations significantly increase. /// /// # Safety /// /// Only one thread pool should use this scope (throughout the duration of its lifetime). The thread /// pool size can not grow midway through. /// /// # Examples /// /// ``` /// use rayon_tlsctx::ThreadLocalCtx; /// use rayon::iter::*; /// /// const NUM_COPIES: usize = 16; /// /// let mut buf: Vec<u16> = (0..!0).collect(); /// /// // Create a thread local context with value 0. /// let ctx = ThreadLocalCtx::new(|| 0); /// /// // Sum the buffer `NUM_COPIES` times and accumulate the results /// // into the threaded pool of counts. Note that the counts may be /// // Unevenly distributed. /// (0..NUM_COPIES) /// .into_par_iter() /// .flat_map(|_| buf.par_iter()) /// .for_each(|i| { /// let mut cnt = unsafe { ctx.get() }; /// *cnt += *i as usize; /// }); /// /// let buf_sum = buf.into_iter().fold(0, |acc, i| acc + i as usize); /// /// // What matters is that the final sum matches the expected value. /// assert_eq!(ctx.into_iter().sum::<usize>(), buf_sum * NUM_COPIES); /// ``` pub unsafe fn get(&self) -> ThreadLocalMut<T> { if self.cloned.get().is_null() { let mut data = self.init_mutex.lock().unwrap(); if self.cloned.get().is_null() { *data = (0..=rayon::current_num_threads()) .map(|_| Default::default()) .collect(); self.cloned.set(data.as_mut_ptr()); } } let tid = rayon::current_thread_index().map(|i| i + 1).unwrap_or(0); let freenode = (*self.cloned.get().add(tid)).get_free_node(); match freenode { ThreadLocalNode { value: _, borrowed: true, next_box: _, } => panic!("Already borrowed the value on thread {}!", tid), ThreadLocalNode { value: Some(val), borrowed, next_box: _, } => { *borrowed = true; ThreadLocalMut { val, parent: borrowed, } } ThreadLocalNode { value, borrowed, next_box: _, } => { *borrowed = true; let cloned = (self.inner)(); *value = Some(cloned); ThreadLocalMut { val: value.as_mut().unwrap(), parent: borrowed, } } } } } /// Thread local node, can have children if there is some work stealing happening. struct ThreadLocalNode<T> { value: Option<T>, borrowed: bool, next_box: Option<Box<ThreadLocalNode<T>>>, } impl<T> Default for ThreadLocalNode<T> { fn default() -> Self { Self { value: None, borrowed: false, next_box: None, } } } impl<T> ThreadLocalNode<T> { fn get_free_node(&mut self) -> &mut Self { if !self.borrowed { self } else { match &mut self.next_box { Some(next) => next.get_free_node(), x => { *x = Some(Default::default()); x.as_mut().unwrap() } } } } } /// Final iterator for ThreadLocalCtx. pub struct ThreadNodeIterator<T> { nodes: std::vec::IntoIter<ThreadLocalNode<T>>, cur: Option<ThreadLocalNode<T>>, } impl<T> Iterator for ThreadNodeIterator<T> { type Item = T; fn next(&mut self) -> Option<T> { if self.cur.is_none() { self.cur = self.nodes.next(); } if let Some(cur) = self.cur.take() { let next = cur.next_box; self.cur = next.map(|i| *i); if let Some(v) = cur.value { Some(v) } else { self.next() } } else { None } } } /// Consume the context and retrieve all created items. impl<T, F> IntoIterator for ThreadLocalCtx<T, F> { type Item = T; type IntoIter = ThreadNodeIterator<T>; fn into_iter(self) -> Self::IntoIter { ThreadNodeIterator { nodes: self.init_mutex.into_inner().unwrap().into_iter(), cur: None, } } } /// Borrowed thread local variable. /// /// This structure tracks borrow rules at runtime, it may be necessary to manually /// drop the object, if multiple rayon loops are involved. pub struct ThreadLocalMut<'a, T> { val: &'a mut T, parent: &'a mut bool, } impl<'a, T> Deref for ThreadLocalMut<'a, T> { type Target = T; fn deref(&self) -> &T { self.val } } impl<'a, T> DerefMut for ThreadLocalMut<'a, T> { fn deref_mut(&mut self) -> &mut T { self.val } } impl<'a, T> Drop for ThreadLocalMut<'a, T> { fn drop(&mut self) { *self.parent = false; } }