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
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
//! This module defines various operations and types that are implemented in
//! one way for the serial compiler, and another way the parallel compiler.
//!
//! Operations
//! ----------
//! The parallel versions of operations use Rayon to execute code in parallel,
//! while the serial versions degenerate straightforwardly to serial execution.
//! The operations include `join`, `parallel`, `par_iter`, and `par_for_each`.
//!
//! Types
//! -----
//! The parallel versions of types provide various kinds of synchronization,
//! while the serial compiler versions do not.
//!
//! The following table shows how the types are implemented internally. Except
//! where noted otherwise, the type in column one is defined as a
//! newtype around the type from column two or three.
//!
//! | Type                    | Serial version      | Parallel version                |
//! | ----------------------- | ------------------- | ------------------------------- |
//! | `Lrc<T>`                | `rc::Rc<T>`         | `sync::Arc<T>`                  |
//! |` Weak<T>`               | `rc::Weak<T>`       | `sync::Weak<T>`                 |
//! | `LRef<'a, T>` [^2]      | `&'a mut T`         | `&'a T`                         |
//! |                         |                     |                                 |
//! | `AtomicBool`            | `Cell<bool>`        | `atomic::AtomicBool`            |
//! | `AtomicU32`             | `Cell<u32>`         | `atomic::AtomicU32`             |
//! | `AtomicU64`             | `Cell<u64>`         | `atomic::AtomicU64`             |
//! | `AtomicUsize`           | `Cell<usize>`       | `atomic::AtomicUsize`           |
//! |                         |                     |                                 |
//! | `Lock<T>`               | `RefCell<T>`        | `RefCell<T>` or                 |
//! |                         |                     | `parking_lot::Mutex<T>`         |
//! | `RwLock<T>`             | `RefCell<T>`        | `parking_lot::RwLock<T>`        |
//! | `MTLock<T>`        [^1] | `T`                 | `Lock<T>`                       |
//! | `MTLockRef<'a, T>` [^2] | `&'a mut MTLock<T>` | `&'a MTLock<T>`                 |
//! |                         |                     |                                 |
//! | `ParallelIterator`      | `Iterator`          | `rayon::iter::ParallelIterator` |
//!
//! [^1] `MTLock` is similar to `Lock`, but the serial version avoids the cost
//! of a `RefCell`. This is appropriate when interior mutability is not
//! required.
//!
//! [^2] `MTRef`, `MTLockRef` are type aliases.

pub use crate::marker::*;
use std::collections::HashMap;
use std::hash::{BuildHasher, Hash};

mod lock;
#[doc(no_inline)]
pub use lock::{Lock, LockGuard, Mode};

mod worker_local;
pub use worker_local::{Registry, WorkerLocal};

mod parallel;
#[cfg(parallel_compiler)]
pub use parallel::scope;
pub use parallel::{join, par_for_each_in, par_map, parallel_guard, try_par_for_each_in};

pub use vec::{AppendOnlyIndexVec, AppendOnlyVec};

mod vec;

mod freeze;
pub use freeze::{FreezeLock, FreezeReadGuard, FreezeWriteGuard};

mod mode {
    use std::sync::atomic::{AtomicU8, Ordering};

    const UNINITIALIZED: u8 = 0;
    const DYN_NOT_THREAD_SAFE: u8 = 1;
    const DYN_THREAD_SAFE: u8 = 2;

    static DYN_THREAD_SAFE_MODE: AtomicU8 = AtomicU8::new(UNINITIALIZED);

    // Whether thread safety is enabled (due to running under multiple threads).
    #[inline]
    pub fn is_dyn_thread_safe() -> bool {
        match DYN_THREAD_SAFE_MODE.load(Ordering::Relaxed) {
            DYN_NOT_THREAD_SAFE => false,
            DYN_THREAD_SAFE => true,
            _ => panic!("uninitialized dyn_thread_safe mode!"),
        }
    }

    // Whether thread safety might be enabled.
    #[inline]
    pub fn might_be_dyn_thread_safe() -> bool {
        DYN_THREAD_SAFE_MODE.load(Ordering::Relaxed) != DYN_NOT_THREAD_SAFE
    }

    // Only set by the `-Z threads` compile option
    pub fn set_dyn_thread_safe_mode(mode: bool) {
        let set: u8 = if mode { DYN_THREAD_SAFE } else { DYN_NOT_THREAD_SAFE };
        let previous = DYN_THREAD_SAFE_MODE.compare_exchange(
            UNINITIALIZED,
            set,
            Ordering::Relaxed,
            Ordering::Relaxed,
        );

        // Check that the mode was either uninitialized or was already set to the requested mode.
        assert!(previous.is_ok() || previous == Err(set));
    }
}

pub use mode::{is_dyn_thread_safe, set_dyn_thread_safe_mode};

cfg_match! {
    cfg(not(parallel_compiler)) => {
        use std::ops::Add;
        use std::cell::Cell;
        use std::sync::atomic::Ordering;

        pub unsafe auto trait Send {}
        pub unsafe auto trait Sync {}

        unsafe impl<T> Send for T {}
        unsafe impl<T> Sync for T {}

        /// This is a single threaded variant of `AtomicU64`, `AtomicUsize`, etc.
        /// It has explicit ordering arguments and is only intended for use with
        /// the native atomic types.
        /// You should use this type through the `AtomicU64`, `AtomicUsize`, etc, type aliases
        /// as it's not intended to be used separately.
        #[derive(Debug, Default)]
        pub struct Atomic<T: Copy>(Cell<T>);

        impl<T: Copy> Atomic<T> {
            #[inline]
            pub fn new(v: T) -> Self {
                Atomic(Cell::new(v))
            }

            #[inline]
            pub fn into_inner(self) -> T {
                self.0.into_inner()
            }

            #[inline]
            pub fn load(&self, _: Ordering) -> T {
                self.0.get()
            }

            #[inline]
            pub fn store(&self, val: T, _: Ordering) {
                self.0.set(val)
            }

            #[inline]
            pub fn swap(&self, val: T, _: Ordering) -> T {
                self.0.replace(val)
            }
        }

        impl Atomic<bool> {
            pub fn fetch_or(&self, val: bool, _: Ordering) -> bool {
                let old = self.0.get();
                self.0.set(val | old);
                old
            }
            pub fn fetch_and(&self, val: bool, _: Ordering) -> bool {
                let old = self.0.get();
                self.0.set(val & old);
                old
            }
        }

        impl<T: Copy + PartialEq> Atomic<T> {
            #[inline]
            pub fn compare_exchange(&self,
                                    current: T,
                                    new: T,
                                    _: Ordering,
                                    _: Ordering)
                                    -> Result<T, T> {
                let read = self.0.get();
                if read == current {
                    self.0.set(new);
                    Ok(read)
                } else {
                    Err(read)
                }
            }
        }

        impl<T: Add<Output=T> + Copy> Atomic<T> {
            #[inline]
            pub fn fetch_add(&self, val: T, _: Ordering) -> T {
                let old = self.0.get();
                self.0.set(old + val);
                old
            }
        }

        pub type AtomicUsize = Atomic<usize>;
        pub type AtomicBool = Atomic<bool>;
        pub type AtomicU32 = Atomic<u32>;
        pub type AtomicU64 = Atomic<u64>;

        pub use std::rc::Rc as Lrc;
        pub use std::rc::Weak as Weak;
        #[doc(no_inline)]
        pub use std::cell::Ref as ReadGuard;
        #[doc(no_inline)]
        pub use std::cell::Ref as MappedReadGuard;
        #[doc(no_inline)]
        pub use std::cell::RefMut as WriteGuard;
        #[doc(no_inline)]
        pub use std::cell::RefMut as MappedWriteGuard;
        #[doc(no_inline)]
        pub use std::cell::RefMut as MappedLockGuard;

        pub use std::cell::OnceCell as OnceLock;

        use std::cell::RefCell as InnerRwLock;

        pub type LRef<'a, T> = &'a mut T;

        #[derive(Debug, Default)]
        pub struct MTLock<T>(T);

        impl<T> MTLock<T> {
            #[inline(always)]
            pub fn new(inner: T) -> Self {
                MTLock(inner)
            }

            #[inline(always)]
            pub fn into_inner(self) -> T {
                self.0
            }

            #[inline(always)]
            pub fn get_mut(&mut self) -> &mut T {
                &mut self.0
            }

            #[inline(always)]
            pub fn lock(&self) -> &T {
                &self.0
            }

            #[inline(always)]
            pub fn lock_mut(&mut self) -> &mut T {
                &mut self.0
            }
        }

        // FIXME: Probably a bad idea (in the threaded case)
        impl<T: Clone> Clone for MTLock<T> {
            #[inline]
            fn clone(&self) -> Self {
                MTLock(self.0.clone())
            }
        }
    }
    _ => {
        pub use std::marker::Send as Send;
        pub use std::marker::Sync as Sync;

        pub use parking_lot::RwLockReadGuard as ReadGuard;
        pub use parking_lot::MappedRwLockReadGuard as MappedReadGuard;
        pub use parking_lot::RwLockWriteGuard as WriteGuard;
        pub use parking_lot::MappedRwLockWriteGuard as MappedWriteGuard;

        pub use parking_lot::MappedMutexGuard as MappedLockGuard;

        pub use std::sync::OnceLock;

        pub use std::sync::atomic::{AtomicBool, AtomicUsize, AtomicU32};

        // PowerPC and MIPS platforms with 32-bit pointers do not
        // have AtomicU64 type.
        #[cfg(not(any(target_arch = "powerpc", target_arch = "mips")))]
        pub use std::sync::atomic::AtomicU64;

        #[cfg(any(target_arch = "powerpc", target_arch = "mips"))]
        pub use portable_atomic::AtomicU64;

        pub use std::sync::Arc as Lrc;
        pub use std::sync::Weak as Weak;

        pub type LRef<'a, T> = &'a T;

        #[derive(Debug, Default)]
        pub struct MTLock<T>(Lock<T>);

        impl<T> MTLock<T> {
            #[inline(always)]
            pub fn new(inner: T) -> Self {
                MTLock(Lock::new(inner))
            }

            #[inline(always)]
            pub fn into_inner(self) -> T {
                self.0.into_inner()
            }

            #[inline(always)]
            pub fn get_mut(&mut self) -> &mut T {
                self.0.get_mut()
            }

            #[inline(always)]
            pub fn lock(&self) -> LockGuard<'_, T> {
                self.0.lock()
            }

            #[inline(always)]
            pub fn lock_mut(&self) -> LockGuard<'_, T> {
                self.lock()
            }
        }

        use parking_lot::RwLock as InnerRwLock;

        /// This makes locks panic if they are already held.
        /// It is only useful when you are running in a single thread
        const ERROR_CHECKING: bool = false;
    }
}

pub type MTLockRef<'a, T> = LRef<'a, MTLock<T>>;

#[derive(Default)]
#[cfg_attr(parallel_compiler, repr(align(64)))]
pub struct CacheAligned<T>(pub T);

pub trait HashMapExt<K, V> {
    /// Same as HashMap::insert, but it may panic if there's already an
    /// entry for `key` with a value not equal to `value`
    fn insert_same(&mut self, key: K, value: V);
}

impl<K: Eq + Hash, V: Eq, S: BuildHasher> HashMapExt<K, V> for HashMap<K, V, S> {
    fn insert_same(&mut self, key: K, value: V) {
        self.entry(key).and_modify(|old| assert!(*old == value)).or_insert(value);
    }
}

#[derive(Debug, Default)]
pub struct RwLock<T>(InnerRwLock<T>);

impl<T> RwLock<T> {
    #[inline(always)]
    pub fn new(inner: T) -> Self {
        RwLock(InnerRwLock::new(inner))
    }

    #[inline(always)]
    pub fn into_inner(self) -> T {
        self.0.into_inner()
    }

    #[inline(always)]
    pub fn get_mut(&mut self) -> &mut T {
        self.0.get_mut()
    }

    #[cfg(not(parallel_compiler))]
    #[inline(always)]
    #[track_caller]
    pub fn read(&self) -> ReadGuard<'_, T> {
        self.0.borrow()
    }

    #[cfg(parallel_compiler)]
    #[inline(always)]
    pub fn read(&self) -> ReadGuard<'_, T> {
        if ERROR_CHECKING {
            self.0.try_read().expect("lock was already held")
        } else {
            self.0.read()
        }
    }

    #[inline(always)]
    #[track_caller]
    pub fn with_read_lock<F: FnOnce(&T) -> R, R>(&self, f: F) -> R {
        f(&*self.read())
    }

    #[cfg(not(parallel_compiler))]
    #[inline(always)]
    pub fn try_write(&self) -> Result<WriteGuard<'_, T>, ()> {
        self.0.try_borrow_mut().map_err(|_| ())
    }

    #[cfg(parallel_compiler)]
    #[inline(always)]
    pub fn try_write(&self) -> Result<WriteGuard<'_, T>, ()> {
        self.0.try_write().ok_or(())
    }

    #[cfg(not(parallel_compiler))]
    #[inline(always)]
    #[track_caller]
    pub fn write(&self) -> WriteGuard<'_, T> {
        self.0.borrow_mut()
    }

    #[cfg(parallel_compiler)]
    #[inline(always)]
    pub fn write(&self) -> WriteGuard<'_, T> {
        if ERROR_CHECKING {
            self.0.try_write().expect("lock was already held")
        } else {
            self.0.write()
        }
    }

    #[inline(always)]
    #[track_caller]
    pub fn with_write_lock<F: FnOnce(&mut T) -> R, R>(&self, f: F) -> R {
        f(&mut *self.write())
    }

    #[inline(always)]
    #[track_caller]
    pub fn borrow(&self) -> ReadGuard<'_, T> {
        self.read()
    }

    #[inline(always)]
    #[track_caller]
    pub fn borrow_mut(&self) -> WriteGuard<'_, T> {
        self.write()
    }

    #[cfg(not(parallel_compiler))]
    #[inline(always)]
    pub fn leak(&self) -> &T {
        ReadGuard::leak(self.read())
    }

    #[cfg(parallel_compiler)]
    #[inline(always)]
    pub fn leak(&self) -> &T {
        let guard = self.read();
        let ret = unsafe { &*std::ptr::addr_of!(*guard) };
        std::mem::forget(guard);
        ret
    }
}

// FIXME: Probably a bad idea
impl<T: Clone> Clone for RwLock<T> {
    #[inline]
    fn clone(&self) -> Self {
        RwLock::new(self.borrow().clone())
    }
}