Auto merge of #132282 - Noratrieb:it-is-the-end-of-serial, r=cjgillot

Delete the `cfg(not(parallel))` serial compiler

Since it's inception a long time ago, the parallel compiler and its cfgs have been a maintenance burden. This was a necessary evil the allow iteration while not degrading performance because of synchronization overhead.

But this time is over. Thanks to the amazing work by the parallel working group (and the dyn sync crimes), the parallel compiler has now been fast enough to be shipped by default in nightly for quite a while now.
Stable and beta have still been on the serial compiler, because they can't use `-Zthreads` anyways.
But this is quite suboptimal:
- the maintenance burden still sucks
- we're not testing the serial compiler in nightly

Because of these reasons, it's time to end it. The serial compiler has served us well in the years since it was split from the parallel one, but it's over now.

Let the knight slay one head of the two-headed dragon!

#113349

Note that the default is still 1 thread, as more than 1 thread is still fairly broken.

cc `@onur-ozkan` to see if i did the bootstrap field removal correctly, `@SparrowLii` on the sync parts

compiler/rustc/Cargo.toml

@@ -31,5 +31,4 @@ jemalloc = ['dep:jemalloc-sys']
 llvm = ['rustc_driver_impl/llvm']
 max_level_info = ['rustc_driver_impl/max_level_info']
 rustc_randomized_layouts = ['rustc_driver_impl/rustc_randomized_layouts']
-rustc_use_parallel_compiler = ['rustc_driver_impl/rustc_use_parallel_compiler']
 # tidy-alphabetical-end

compiler/rustc_ast/src/tokenstream.rs

@@ -38,7 +38,6 @@ pub enum TokenTree {
 }
 
 // Ensure all fields of `TokenTree` are `DynSend` and `DynSync`.
-#[cfg(parallel_compiler)]
 fn _dummy()
 where
     Token: sync::DynSend + sync::DynSync,

compiler/rustc_baked_icu_data/Cargo.toml

@@ -8,11 +8,6 @@ edition = "2021"
 icu_list = "1.2"
 icu_locid = "1.2"
 icu_locid_transform = "1.3.2"
-icu_provider = "1.2"
+icu_provider = { version = "1.2", features = ["sync"] }
 zerovec = "0.10.0"
 # tidy-alphabetical-end
-
-[features]
-# tidy-alphabetical-start
-rustc_use_parallel_compiler = ['icu_provider/sync']
-# tidy-alphabetical-end

compiler/rustc_data_structures/Cargo.toml

@@ -10,11 +10,11 @@ bitflags = "2.4.1"
 either = "1.0"
 elsa = "=1.7.1"
 ena = "0.14.3"
-indexmap = { version = "2.4.0" }
+indexmap = { version = "2.4.0", features = ["rustc-rayon"] }
 jobserver_crate = { version = "0.1.28", package = "jobserver" }
 measureme = "11"
 rustc-hash = "2.0.0"
-rustc-rayon = { version = "0.5.0", optional = true }
+rustc-rayon = "0.5.0"
 rustc-stable-hash = { version = "0.1.0", features = ["nightly"] }
 rustc_arena = { path = "../rustc_arena" }
 rustc_graphviz = { path = "../rustc_graphviz" }
@@ -53,8 +53,3 @@ memmap2 = "0.2.1"
 
 [target.'cfg(not(target_has_atomic = "64"))'.dependencies]
 portable-atomic = "1.5.1"
-
-[features]
-# tidy-alphabetical-start
-rustc_use_parallel_compiler = ["indexmap/rustc-rayon", "dep:rustc-rayon"]
-# tidy-alphabetical-end

compiler/rustc_data_structures/src/lib.rs

@@ -10,7 +10,6 @@
 #![allow(internal_features)]
 #![allow(rustc::default_hash_types)]
 #![allow(rustc::potential_query_instability)]
-#![cfg_attr(not(parallel_compiler), feature(cell_leak))]
 #![deny(unsafe_op_in_unsafe_fn)]
 #![doc(html_root_url = "https://doc.rust-lang.org/nightly/nightly-rustc/")]
 #![doc(rust_logo)]

compiler/rustc_data_structures/src/marker.rs

@@ -1,25 +1,12 @@
-cfg_match! {
+#[rustc_on_unimplemented(message = "`{Self}` doesn't implement `DynSend`. \
-    cfg(not(parallel_compiler)) => {
+            Add it to `rustc_data_structures::marker` or use `IntoDynSyncSend` if it's already `Send`")]
-        pub auto trait DynSend {}
-        pub auto trait DynSync {}
-
-        impl<T> DynSend for T {}
-        impl<T> DynSync for T {}
-    }
-    _ => {
-        #[rustc_on_unimplemented(
-            message = "`{Self}` doesn't implement `DynSend`. \
-            Add it to `rustc_data_structures::marker` or use `IntoDynSyncSend` if it's already `Send`"
-        )]
 // This is an auto trait for types which can be sent across threads if `sync::is_dyn_thread_safe()`
 // is true. These types can be wrapped in a `FromDyn` to get a `Send` type. Wrapping a
 // `Send` type in `IntoDynSyncSend` will create a `DynSend` type.
 pub unsafe auto trait DynSend {}
 
-        #[rustc_on_unimplemented(
+#[rustc_on_unimplemented(message = "`{Self}` doesn't implement `DynSync`. \
-            message = "`{Self}` doesn't implement `DynSync`. \
+            Add it to `rustc_data_structures::marker` or use `IntoDynSyncSend` if it's already `Sync`")]
-            Add it to `rustc_data_structures::marker` or use `IntoDynSyncSend` if it's already `Sync`"
-        )]
 // This is an auto trait for types which can be shared across threads if `sync::is_dyn_thread_safe()`
 // is true. These types can be wrapped in a `FromDyn` to get a `Sync` type. Wrapping a
 // `Sync` type in `IntoDynSyncSend` will create a `DynSync` type.
@@ -62,14 +49,8 @@ cfg_match! {
 
 // These structures are already `Send`.
 already_send!(
-            [std::backtrace::Backtrace]
+    [std::backtrace::Backtrace][std::io::Stdout][std::io::Stderr][std::io::Error][std::fs::File]
-            [std::io::Stdout]
+        [rustc_arena::DroplessArena][crate::memmap::Mmap][crate::profiling::SelfProfiler]
-            [std::io::Stderr]
-            [std::io::Error]
-            [std::fs::File]
-            [rustc_arena::DroplessArena]
-            [crate::memmap::Mmap]
-            [crate::profiling::SelfProfiler]
         [crate::owned_slice::OwnedSlice]
 );
 
@@ -134,29 +115,18 @@ cfg_match! {
 
 // These structures are already `Sync`.
 already_sync!(
-            [std::sync::atomic::AtomicBool]
+    [std::sync::atomic::AtomicBool][std::sync::atomic::AtomicUsize][std::sync::atomic::AtomicU8]
-            [std::sync::atomic::AtomicUsize]
+        [std::sync::atomic::AtomicU32][std::backtrace::Backtrace][std::io::Error][std::fs::File]
-            [std::sync::atomic::AtomicU8]
+        [jobserver_crate::Client][crate::memmap::Mmap][crate::profiling::SelfProfiler]
-            [std::sync::atomic::AtomicU32]
-            [std::backtrace::Backtrace]
-            [std::io::Error]
-            [std::fs::File]
-            [jobserver_crate::Client]
-            [crate::memmap::Mmap]
-            [crate::profiling::SelfProfiler]
         [crate::owned_slice::OwnedSlice]
 );
 
 // Use portable AtomicU64 for targets without native 64-bit atomics
 #[cfg(target_has_atomic = "64")]
-        already_sync!(
+already_sync!([std::sync::atomic::AtomicU64]);
-            [std::sync::atomic::AtomicU64]
-        );
 
 #[cfg(not(target_has_atomic = "64"))]
-        already_sync!(
+already_sync!([portable_atomic::AtomicU64]);
-            [portable_atomic::AtomicU64]
-        );
 
 macro_rules! impl_dyn_sync {
     ($($($attr: meta)* [$ty: ty where $($generics2: tt)*])*) => {
@@ -186,8 +156,6 @@ cfg_match! {
     [smallvec::SmallVec<A> where A: smallvec::Array + DynSync]
     [thin_vec::ThinVec<T> where T: DynSync]
 );
-    }
-}
 
 pub fn assert_dyn_sync<T: ?Sized + DynSync>() {}
 pub fn assert_dyn_send<T: ?Sized + DynSend>() {}
@@ -203,7 +171,6 @@ impl<T> FromDyn<T> {
         // Check that `sync::is_dyn_thread_safe()` is true on creation so we can
         // implement `Send` and `Sync` for this structure when `T`
         // implements `DynSend` and `DynSync` respectively.
-        #[cfg(parallel_compiler)]
         assert!(crate::sync::is_dyn_thread_safe());
         FromDyn(val)
     }
@@ -215,11 +182,9 @@ impl<T> FromDyn<T> {
 }
 
 // `FromDyn` is `Send` if `T` is `DynSend`, since it ensures that sync::is_dyn_thread_safe() is true.
-#[cfg(parallel_compiler)]
 unsafe impl<T: DynSend> Send for FromDyn<T> {}
 
 // `FromDyn` is `Sync` if `T` is `DynSync`, since it ensures that sync::is_dyn_thread_safe() is true.
-#[cfg(parallel_compiler)]
 unsafe impl<T: DynSync> Sync for FromDyn<T> {}
 
 impl<T> std::ops::Deref for FromDyn<T> {
@@ -237,9 +202,7 @@ impl<T> std::ops::Deref for FromDyn<T> {
 #[derive(Copy, Clone)]
 pub struct IntoDynSyncSend<T: ?Sized>(pub T);
 
-#[cfg(parallel_compiler)]
 unsafe impl<T: ?Sized + Send> DynSend for IntoDynSyncSend<T> {}
-#[cfg(parallel_compiler)]
 unsafe impl<T: ?Sized + Sync> DynSync for IntoDynSyncSend<T> {}
 
 impl<T> std::ops::Deref for IntoDynSyncSend<T> {

compiler/rustc_data_structures/src/owned_slice.rs

@@ -139,11 +139,9 @@ impl Borrow<[u8]> for OwnedSlice {
 }
 
 // Safety: `OwnedSlice` is conceptually `(&'self.1 [u8], Arc<dyn Send + Sync>)`, which is `Send`
-#[cfg(parallel_compiler)]
 unsafe impl sync::Send for OwnedSlice {}
 
 // Safety: `OwnedSlice` is conceptually `(&'self.1 [u8], Arc<dyn Send + Sync>)`, which is `Sync`
-#[cfg(parallel_compiler)]
 unsafe impl sync::Sync for OwnedSlice {}
 
 #[cfg(test)]

compiler/rustc_data_structures/src/sharded.rs

@@ -3,27 +3,22 @@ use std::collections::hash_map::RawEntryMut;
 use std::hash::{Hash, Hasher};
 use std::{iter, mem};
 
-#[cfg(parallel_compiler)]
 use either::Either;
 
 use crate::fx::{FxHashMap, FxHasher};
-#[cfg(parallel_compiler)]
+use crate::sync::{CacheAligned, Lock, LockGuard, Mode, is_dyn_thread_safe};
-use crate::sync::{CacheAligned, is_dyn_thread_safe};
-use crate::sync::{Lock, LockGuard, Mode};
 
 // 32 shards is sufficient to reduce contention on an 8-core Ryzen 7 1700,
 // but this should be tested on higher core count CPUs. How the `Sharded` type gets used
 // may also affect the ideal number of shards.
 const SHARD_BITS: usize = 5;
 
-#[cfg(parallel_compiler)]
 const SHARDS: usize = 1 << SHARD_BITS;
 
 /// An array of cache-line aligned inner locked structures with convenience methods.
 /// A single field is used when the compiler uses only one thread.
 pub enum Sharded<T> {
     Single(Lock<T>),
-    #[cfg(parallel_compiler)]
     Shards(Box<[CacheAligned<Lock<T>>; SHARDS]>),
 }
 
@@ -37,7 +32,6 @@ impl<T: Default> Default for Sharded<T> {
 impl<T> Sharded<T> {
     #[inline]
     pub fn new(mut value: impl FnMut() -> T) -> Self {
-        #[cfg(parallel_compiler)]
         if is_dyn_thread_safe() {
             return Sharded::Shards(Box::new(
                 [(); SHARDS].map(|()| CacheAligned(Lock::new(value()))),
@@ -52,7 +46,6 @@ impl<T> Sharded<T> {
     pub fn get_shard_by_value<K: Hash + ?Sized>(&self, _val: &K) -> &Lock<T> {
         match self {
             Self::Single(single) => single,
-            #[cfg(parallel_compiler)]
             Self::Shards(..) => self.get_shard_by_hash(make_hash(_val)),
         }
     }
@@ -66,7 +59,6 @@ impl<T> Sharded<T> {
     pub fn get_shard_by_index(&self, _i: usize) -> &Lock<T> {
         match self {
             Self::Single(single) => single,
-            #[cfg(parallel_compiler)]
             Self::Shards(shards) => {
                 // SAFETY: The index gets ANDed with the shard mask, ensuring it is always inbounds.
                 unsafe { &shards.get_unchecked(_i & (SHARDS - 1)).0 }
@@ -87,7 +79,6 @@ impl<T> Sharded<T> {
                 // `might_be_dyn_thread_safe` was also false.
                 unsafe { single.lock_assume(Mode::NoSync) }
             }
-            #[cfg(parallel_compiler)]
             Self::Shards(..) => self.lock_shard_by_hash(make_hash(_val)),
         }
     }
@@ -110,7 +101,6 @@ impl<T> Sharded<T> {
                 // `might_be_dyn_thread_safe` was also false.
                 unsafe { single.lock_assume(Mode::NoSync) }
             }
-            #[cfg(parallel_compiler)]
             Self::Shards(shards) => {
                 // Synchronization is enabled so use the `lock_assume_sync` method optimized
                 // for that case.
@@ -127,11 +117,7 @@ impl<T> Sharded<T> {
     #[inline]
     pub fn lock_shards(&self) -> impl Iterator<Item = LockGuard<'_, T>> {
         match self {
-            #[cfg(not(parallel_compiler))]
-            Self::Single(single) => iter::once(single.lock()),
-            #[cfg(parallel_compiler)]
             Self::Single(single) => Either::Left(iter::once(single.lock())),
-            #[cfg(parallel_compiler)]
             Self::Shards(shards) => Either::Right(shards.iter().map(|shard| shard.0.lock())),
         }
     }
@@ -139,11 +125,7 @@ impl<T> Sharded<T> {
     #[inline]
     pub fn try_lock_shards(&self) -> impl Iterator<Item = Option<LockGuard<'_, T>>> {
         match self {
-            #[cfg(not(parallel_compiler))]
-            Self::Single(single) => iter::once(single.try_lock()),
-            #[cfg(parallel_compiler)]
             Self::Single(single) => Either::Left(iter::once(single.try_lock())),
-            #[cfg(parallel_compiler)]
             Self::Shards(shards) => Either::Right(shards.iter().map(|shard| shard.0.try_lock())),
         }
     }
@@ -151,7 +133,6 @@ impl<T> Sharded<T> {
 
 #[inline]
 pub fn shards() -> usize {
-    #[cfg(parallel_compiler)]
     if is_dyn_thread_safe() {
         return SHARDS;
     }

compiler/rustc_data_structures/src/sync.rs

@@ -54,9 +54,7 @@ mod worker_local;
 pub use worker_local::{Registry, WorkerLocal};
 
 mod parallel;
-#[cfg(parallel_compiler)]
+pub use parallel::{join, par_for_each_in, par_map, parallel_guard, scope, try_par_for_each_in};
-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;
@@ -104,182 +102,24 @@ mod mode {
     }
 }
 
-pub use mode::{is_dyn_thread_safe, set_dyn_thread_safe_mode};
+// FIXME(parallel_compiler): Get rid of these aliases across the compiler.
-
-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};
 
+pub use std::marker::{Send, Sync};
 // Use portable AtomicU64 for targets without native 64-bit atomics
 #[cfg(target_has_atomic = "64")]
 pub use std::sync::atomic::AtomicU64;
+pub use std::sync::atomic::{AtomicBool, AtomicU32, AtomicUsize};
+pub use std::sync::{Arc as Lrc, OnceLock, Weak};
 
+pub use mode::{is_dyn_thread_safe, set_dyn_thread_safe_mode};
+pub use parking_lot::{
+    MappedMutexGuard as MappedLockGuard, MappedRwLockReadGuard as MappedReadGuard,
+    MappedRwLockWriteGuard as MappedWriteGuard, RwLockReadGuard as ReadGuard,
+    RwLockWriteGuard as WriteGuard,
+};
 #[cfg(not(target_has_atomic = "64"))]
 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)]
@@ -317,13 +157,11 @@ cfg_match! {
 /// 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)))]
+#[repr(align(64))]
 pub struct CacheAligned<T>(pub T);
 
 pub trait HashMapExt<K, V> {
@@ -357,14 +195,6 @@ impl<T> RwLock<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 {
@@ -380,26 +210,11 @@ impl<T> RwLock<T> {
         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 {
@@ -427,13 +242,6 @@ impl<T> RwLock<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();

compiler/rustc_data_structures/src/sync/freeze.rs

@@ -5,9 +5,7 @@ use std::ops::{Deref, DerefMut};
 use std::ptr::NonNull;
 use std::sync::atomic::Ordering;
 
-use crate::sync::{AtomicBool, ReadGuard, RwLock, WriteGuard};
+use crate::sync::{AtomicBool, DynSend, DynSync, ReadGuard, RwLock, WriteGuard};
-#[cfg(parallel_compiler)]
-use crate::sync::{DynSend, DynSync};
 
 /// A type which allows mutation using a lock until
 /// the value is frozen and can be accessed lock-free.
@@ -22,7 +20,6 @@ pub struct FreezeLock<T> {
     lock: RwLock<()>,
 }
 
-#[cfg(parallel_compiler)]
 unsafe impl<T: DynSync + DynSend> DynSync for FreezeLock<T> {}
 
 impl<T> FreezeLock<T> {

compiler/rustc_data_structures/src/sync/lock.rs

@@ -1,24 +1,16 @@
 //! This module implements a lock which only uses synchronization if `might_be_dyn_thread_safe` is true.
 //! It implements `DynSend` and `DynSync` instead of the typical `Send` and `Sync` traits.
-//!
-//! When `cfg(parallel_compiler)` is not set, the lock is instead a wrapper around `RefCell`.
 
 #![allow(dead_code)]
 
 use std::fmt;
 
-#[cfg(parallel_compiler)]
-pub use maybe_sync::*;
-#[cfg(not(parallel_compiler))]
-pub use no_sync::*;
-
 #[derive(Clone, Copy, PartialEq)]
 pub enum Mode {
     NoSync,
     Sync,
 }
 
-mod maybe_sync {
 use std::cell::{Cell, UnsafeCell};
 use std::intrinsics::unlikely;
 use std::marker::PhantomData;
@@ -28,10 +20,7 @@ mod maybe_sync {
 use parking_lot::RawMutex;
 use parking_lot::lock_api::RawMutex as _;
 
-    use super::Mode;
+use crate::sync::{DynSend, DynSync, mode};
-    use crate::sync::mode;
-    #[cfg(parallel_compiler)]
-    use crate::sync::{DynSend, DynSync};
 
 /// A guard holding mutable access to a `Lock` which is in a locked state.
 #[must_use = "if unused the Lock will immediately unlock"]
@@ -180,56 +169,8 @@ mod maybe_sync {
     }
 }
 
-    #[cfg(parallel_compiler)]
 unsafe impl<T: DynSend> DynSend for Lock<T> {}
-    #[cfg(parallel_compiler)]
 unsafe impl<T: DynSend> DynSync for Lock<T> {}
-}
-
-mod no_sync {
-    use std::cell::RefCell;
-    #[doc(no_inline)]
-    pub use std::cell::RefMut as LockGuard;
-
-    use super::Mode;
-
-    pub struct Lock<T>(RefCell<T>);
-
-    impl<T> Lock<T> {
-        #[inline(always)]
-        pub fn new(inner: T) -> Self {
-            Lock(RefCell::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 try_lock(&self) -> Option<LockGuard<'_, T>> {
-            self.0.try_borrow_mut().ok()
-        }
-
-        #[inline(always)]
-        #[track_caller]
-        // This is unsafe to match the API for the `parallel_compiler` case.
-        pub unsafe fn lock_assume(&self, _mode: Mode) -> LockGuard<'_, T> {
-            self.0.borrow_mut()
-        }
-
-        #[inline(always)]
-        #[track_caller]
-        pub fn lock(&self) -> LockGuard<'_, T> {
-            self.0.borrow_mut()
-        }
-    }
-}
 
 impl<T> Lock<T> {
     #[inline(always)]

compiler/rustc_data_structures/src/sync/parallel.rs

@@ -6,14 +6,11 @@
 use std::any::Any;
 use std::panic::{AssertUnwindSafe, catch_unwind, resume_unwind};
 
-#[cfg(not(parallel_compiler))]
-pub use disabled::*;
-#[cfg(parallel_compiler)]
-pub use enabled::*;
 use parking_lot::Mutex;
+use rayon::iter::{FromParallelIterator, IntoParallelIterator, ParallelIterator};
 
 use crate::FatalErrorMarker;
-use crate::sync::IntoDynSyncSend;
+use crate::sync::{DynSend, DynSync, FromDyn, IntoDynSyncSend, mode};
 
 /// A guard used to hold panics that occur during a parallel section to later by unwound.
 /// This is used for the parallel compiler to prevent fatal errors from non-deterministically
@@ -49,20 +46,7 @@ pub fn parallel_guard<R>(f: impl FnOnce(&ParallelGuard) -> R) -> R {
     ret
 }
 
-mod disabled {
+pub fn serial_join<A, B, RA, RB>(oper_a: A, oper_b: B) -> (RA, RB)
-    use crate::sync::parallel_guard;
-
-    #[macro_export]
-    #[cfg(not(parallel_compiler))]
-    macro_rules! parallel {
-        ($($blocks:block),*) => {{
-            $crate::sync::parallel_guard(|guard| {
-                $(guard.run(|| $blocks);)*
-            });
-        }}
-    }
-
-    pub fn join<A, B, RA, RB>(oper_a: A, oper_b: B) -> (RA, RB)
 where
     A: FnOnce() -> RA,
     B: FnOnce() -> RB,
@@ -75,35 +59,6 @@ mod disabled {
     (a.unwrap(), b.unwrap())
 }
 
-    pub fn par_for_each_in<T: IntoIterator>(t: T, mut for_each: impl FnMut(T::Item)) {
-        parallel_guard(|guard| {
-            t.into_iter().for_each(|i| {
-                guard.run(|| for_each(i));
-            });
-        })
-    }
-
-    pub fn try_par_for_each_in<T: IntoIterator, E>(
-        t: T,
-        mut for_each: impl FnMut(T::Item) -> Result<(), E>,
-    ) -> Result<(), E> {
-        parallel_guard(|guard| {
-            t.into_iter().filter_map(|i| guard.run(|| for_each(i))).fold(Ok(()), Result::and)
-        })
-    }
-
-    pub fn par_map<T: IntoIterator, R, C: FromIterator<R>>(
-        t: T,
-        mut map: impl FnMut(<<T as IntoIterator>::IntoIter as Iterator>::Item) -> R,
-    ) -> C {
-        parallel_guard(|guard| t.into_iter().filter_map(|i| guard.run(|| map(i))).collect())
-    }
-}
-
-#[cfg(parallel_compiler)]
-mod enabled {
-    use crate::sync::{DynSend, DynSync, FromDyn, mode, parallel_guard};
-
 /// Runs a list of blocks in parallel. The first block is executed immediately on
 /// the current thread. Use that for the longest running block.
 #[macro_export]
@@ -166,12 +121,10 @@ mod enabled {
         });
         (a.unwrap().into_inner(), b.unwrap().into_inner())
     } else {
-            super::disabled::join(oper_a, oper_b)
+        serial_join(oper_a, oper_b)
     }
 }
 
-    use rayon::iter::{FromParallelIterator, IntoParallelIterator, ParallelIterator};
-
 pub fn par_for_each_in<I, T: IntoIterator<Item = I> + IntoParallelIterator<Item = I>>(
     t: T,
     for_each: impl Fn(I) + DynSync + DynSend,
@@ -227,4 +180,3 @@ mod enabled {
         }
     })
 }
-}

compiler/rustc_data_structures/src/sync/vec.rs

@@ -4,40 +4,23 @@ use rustc_index::Idx;
 
 #[derive(Default)]
 pub struct AppendOnlyIndexVec<I: Idx, T: Copy> {
-    #[cfg(not(parallel_compiler))]
-    vec: elsa::vec::FrozenVec<T>,
-    #[cfg(parallel_compiler)]
     vec: elsa::sync::LockFreeFrozenVec<T>,
     _marker: PhantomData<fn(&I)>,
 }
 
 impl<I: Idx, T: Copy> AppendOnlyIndexVec<I, T> {
     pub fn new() -> Self {
-        Self {
+        Self { vec: elsa::sync::LockFreeFrozenVec::new(), _marker: PhantomData }
-            #[cfg(not(parallel_compiler))]
-            vec: elsa::vec::FrozenVec::new(),
-            #[cfg(parallel_compiler)]
-            vec: elsa::sync::LockFreeFrozenVec::new(),
-            _marker: PhantomData,
-        }
     }
 
     pub fn push(&self, val: T) -> I {
-        #[cfg(not(parallel_compiler))]
-        let i = self.vec.len();
-        #[cfg(not(parallel_compiler))]
-        self.vec.push(val);
-        #[cfg(parallel_compiler)]
         let i = self.vec.push(val);
         I::new(i)
     }
 
     pub fn get(&self, i: I) -> Option<T> {
         let i = i.index();
-        #[cfg(not(parallel_compiler))]
+        self.vec.get(i)
-        return self.vec.get_copy(i);
-        #[cfg(parallel_compiler)]
-        return self.vec.get(i);
     }
 }
 

compiler/rustc_data_structures/src/sync/worker_local.rs

@@ -5,8 +5,9 @@ use std::ptr;
 use std::sync::Arc;
 
 use parking_lot::Mutex;
-#[cfg(parallel_compiler)]
+
-use {crate::outline, crate::sync::CacheAligned};
+use crate::outline;
+use crate::sync::CacheAligned;
 
 /// A pointer to the `RegistryData` which uniquely identifies a registry.
 /// This identifier can be reused if the registry gets freed.
@@ -21,7 +22,6 @@ impl RegistryId {
     ///
     /// Note that there's a race possible where the identifier in `THREAD_DATA` could be reused
     /// so this can succeed from a different registry.
-    #[cfg(parallel_compiler)]
     fn verify(self) -> usize {
         let (id, index) = THREAD_DATA.with(|data| (data.registry_id.get(), data.index.get()));
 
@@ -102,11 +102,7 @@ impl Registry {
 /// worker local value through the `Deref` impl on the registry associated with the thread it was
 /// created on. It will panic otherwise.
 pub struct WorkerLocal<T> {
-    #[cfg(not(parallel_compiler))]
-    local: T,
-    #[cfg(parallel_compiler)]
     locals: Box<[CacheAligned<T>]>,
-    #[cfg(parallel_compiler)]
     registry: Registry,
 }
 
@@ -114,7 +110,6 @@ pub struct WorkerLocal<T> {
 // or it will panic for threads without an associated local. So there isn't a need for `T` to do
 // it's own synchronization. The `verify` method on `RegistryId` has an issue where the id
 // can be reused, but `WorkerLocal` has a reference to `Registry` which will prevent any reuse.
-#[cfg(parallel_compiler)]
 unsafe impl<T: Send> Sync for WorkerLocal<T> {}
 
 impl<T> WorkerLocal<T> {
@@ -122,47 +117,24 @@ impl<T> WorkerLocal<T> {
     /// value this worker local should take for each thread in the registry.
     #[inline]
     pub fn new<F: FnMut(usize) -> T>(mut initial: F) -> WorkerLocal<T> {
-        #[cfg(parallel_compiler)]
-        {
         let registry = Registry::current();
         WorkerLocal {
-                locals: (0..registry.0.thread_limit.get())
+            locals: (0..registry.0.thread_limit.get()).map(|i| CacheAligned(initial(i))).collect(),
-                    .map(|i| CacheAligned(initial(i)))
-                    .collect(),
             registry,
         }
     }
-        #[cfg(not(parallel_compiler))]
-        {
-            WorkerLocal { local: initial(0) }
-        }
-    }
 
     /// Returns the worker-local values for each thread
     #[inline]
     pub fn into_inner(self) -> impl Iterator<Item = T> {
-        #[cfg(parallel_compiler)]
-        {
         self.locals.into_vec().into_iter().map(|local| local.0)
     }
-        #[cfg(not(parallel_compiler))]
-        {
-            std::iter::once(self.local)
-        }
-    }
 }
 
 impl<T> Deref for WorkerLocal<T> {
     type Target = T;
 
     #[inline(always)]
-    #[cfg(not(parallel_compiler))]
-    fn deref(&self) -> &T {
-        &self.local
-    }
-
-    #[inline(always)]
-    #[cfg(parallel_compiler)]
     fn deref(&self) -> &T {
         // This is safe because `verify` will only return values less than
         // `self.registry.thread_limit` which is the size of the `self.locals` array.

compiler/rustc_driver_impl/Cargo.toml

@@ -77,9 +77,4 @@ rustc_randomized_layouts = [
     'rustc_index/rustc_randomized_layouts',
     'rustc_middle/rustc_randomized_layouts'
 ]
-rustc_use_parallel_compiler = [
-    'rustc_data_structures/rustc_use_parallel_compiler',
-    'rustc_interface/rustc_use_parallel_compiler',
-    'rustc_middle/rustc_use_parallel_compiler'
-]
 # tidy-alphabetical-end

compiler/rustc_error_messages/Cargo.toml

@@ -19,8 +19,3 @@ rustc_span = { path = "../rustc_span" }
 tracing = "0.1"
 unic-langid = { version = "0.9.0", features = ["macros"] }
 # tidy-alphabetical-end
-
-[features]
-# tidy-alphabetical-start
-rustc_use_parallel_compiler = ['rustc_baked_icu_data/rustc_use_parallel_compiler']
-# tidy-alphabetical-end

compiler/rustc_error_messages/src/lib.rs

@@ -8,12 +8,9 @@
 // tidy-alphabetical-end
 
 use std::borrow::Cow;
-#[cfg(not(parallel_compiler))]
-use std::cell::LazyCell as Lazy;
 use std::error::Error;
 use std::path::{Path, PathBuf};
-#[cfg(parallel_compiler)]
+use std::sync::LazyLock;
-use std::sync::LazyLock as Lazy;
 use std::{fmt, fs, io};
 
 use fluent_bundle::FluentResource;
@@ -21,9 +18,6 @@ pub use fluent_bundle::types::FluentType;
 pub use fluent_bundle::{self, FluentArgs, FluentError, FluentValue};
 use fluent_syntax::parser::ParserError;
 use icu_provider_adapters::fallback::{LocaleFallbackProvider, LocaleFallbacker};
-#[cfg(not(parallel_compiler))]
-use intl_memoizer::IntlLangMemoizer;
-#[cfg(parallel_compiler)]
 use intl_memoizer::concurrent::IntlLangMemoizer;
 use rustc_data_structures::sync::{IntoDynSyncSend, Lrc};
 use rustc_macros::{Decodable, Encodable};
@@ -34,12 +28,6 @@ pub use unic_langid::{LanguageIdentifier, langid};
 pub type FluentBundle =
     IntoDynSyncSend<fluent_bundle::bundle::FluentBundle<FluentResource, IntlLangMemoizer>>;
 
-#[cfg(not(parallel_compiler))]
-fn new_bundle(locales: Vec<LanguageIdentifier>) -> FluentBundle {
-    IntoDynSyncSend(fluent_bundle::bundle::FluentBundle::new(locales))
-}
-
-#[cfg(parallel_compiler)]
 fn new_bundle(locales: Vec<LanguageIdentifier>) -> FluentBundle {
     IntoDynSyncSend(fluent_bundle::bundle::FluentBundle::new_concurrent(locales))
 }
@@ -217,7 +205,7 @@ fn register_functions(bundle: &mut FluentBundle) {
 
 /// Type alias for the result of `fallback_fluent_bundle` - a reference-counted pointer to a lazily
 /// evaluated fluent bundle.
-pub type LazyFallbackBundle = Lrc<Lazy<FluentBundle, impl FnOnce() -> FluentBundle>>;
+pub type LazyFallbackBundle = Lrc<LazyLock<FluentBundle, impl FnOnce() -> FluentBundle>>;
 
 /// Return the default `FluentBundle` with standard "en-US" diagnostic messages.
 #[instrument(level = "trace", skip(resources))]
@@ -225,7 +213,7 @@ pub fn fallback_fluent_bundle(
     resources: Vec<&'static str>,
     with_directionality_markers: bool,
 ) -> LazyFallbackBundle {
-    Lrc::new(Lazy::new(move || {
+    Lrc::new(LazyLock::new(move || {
         let mut fallback_bundle = new_bundle(vec![langid!("en-US")]);
 
         register_functions(&mut fallback_bundle);
@@ -548,15 +536,6 @@ pub fn fluent_value_from_str_list_sep_by_and(l: Vec<Cow<'_, str>>) -> FluentValu
             Cow::Owned(result)
         }
 
-        #[cfg(not(parallel_compiler))]
-        fn as_string_threadsafe(
-            &self,
-            _intls: &intl_memoizer::concurrent::IntlLangMemoizer,
-        ) -> Cow<'static, str> {
-            unreachable!("`as_string_threadsafe` is not used in non-parallel rustc")
-        }
-
-        #[cfg(parallel_compiler)]
         fn as_string_threadsafe(
             &self,
             intls: &intl_memoizer::concurrent::IntlLangMemoizer,

compiler/rustc_errors/Cargo.toml

@@ -36,8 +36,3 @@ features = [
     "Win32_Security",
     "Win32_System_Threading",
 ]
-
-[features]
-# tidy-alphabetical-start
-rustc_use_parallel_compiler = ['rustc_error_messages/rustc_use_parallel_compiler']
-# tidy-alphabetical-end

compiler/rustc_errors/src/tests.rs

@@ -26,16 +26,11 @@ fn make_dummy(ftl: &'static str) -> Dummy {
 
     let langid_en = langid!("en-US");
 
-    #[cfg(parallel_compiler)]
     let mut bundle: FluentBundle =
         IntoDynSyncSend(crate::fluent_bundle::bundle::FluentBundle::new_concurrent(vec![
             langid_en,
         ]));
 
-    #[cfg(not(parallel_compiler))]
-    let mut bundle: FluentBundle =
-        IntoDynSyncSend(crate::fluent_bundle::bundle::FluentBundle::new(vec![langid_en]));
-
     bundle.add_resource(resource).expect("Failed to add FTL resources to the bundle.");
 
     Dummy { bundle }

compiler/rustc_interface/Cargo.toml

@@ -5,8 +5,8 @@ edition = "2021"
 
 [dependencies]
 # tidy-alphabetical-start
-rustc-rayon = { version = "0.5.0", optional = true }
+rustc-rayon = { version = "0.5.0" }
-rustc-rayon-core = { version = "0.5.0", optional = true }
+rustc-rayon-core = { version = "0.5.0" }
 rustc_ast = { path = "../rustc_ast" }
 rustc_ast_lowering = { path = "../rustc_ast_lowering" }
 rustc_ast_passes = { path = "../rustc_ast_passes" }
@@ -54,10 +54,4 @@ tracing = "0.1"
 [features]
 # tidy-alphabetical-start
 llvm = ['dep:rustc_codegen_llvm']
-rustc_use_parallel_compiler = [
-    'dep:rustc-rayon',
-    'dep:rustc-rayon-core',
-    'rustc_query_impl/rustc_use_parallel_compiler',
-    'rustc_errors/rustc_use_parallel_compiler'
-]
 # tidy-alphabetical-end

compiler/rustc_interface/src/util.rs

@@ -6,7 +6,6 @@ use std::{env, iter, thread};
 
 use rustc_ast as ast;
 use rustc_codegen_ssa::traits::CodegenBackend;
-#[cfg(parallel_compiler)]
 use rustc_data_structures::sync;
 use rustc_metadata::{DylibError, load_symbol_from_dylib};
 use rustc_middle::ty::CurrentGcx;
@@ -117,19 +116,6 @@ fn run_in_thread_with_globals<F: FnOnce(CurrentGcx) -> R + Send, R: Send>(
     })
 }
 
-#[cfg(not(parallel_compiler))]
-pub(crate) fn run_in_thread_pool_with_globals<F: FnOnce(CurrentGcx) -> R + Send, R: Send>(
-    thread_builder_diag: &EarlyDiagCtxt,
-    edition: Edition,
-    _threads: usize,
-    sm_inputs: SourceMapInputs,
-    f: F,
-) -> R {
-    let thread_stack_size = init_stack_size(thread_builder_diag);
-    run_in_thread_with_globals(thread_stack_size, edition, sm_inputs, f)
-}
-
-#[cfg(parallel_compiler)]
 pub(crate) fn run_in_thread_pool_with_globals<F: FnOnce(CurrentGcx) -> R + Send, R: Send>(
     thread_builder_diag: &EarlyDiagCtxt,
     edition: Edition,

compiler/rustc_middle/Cargo.toml

@@ -11,7 +11,7 @@ either = "1.5.0"
 field-offset = "0.3.5"
 gsgdt = "0.1.2"
 polonius-engine = "0.13.0"
-rustc-rayon-core = { version = "0.5.0", optional = true }
+rustc-rayon-core = { version = "0.5.0" }
 rustc_abi = { path = "../rustc_abi" }
 rustc_apfloat = "0.2.0"
 rustc_arena = { path = "../rustc_arena" }
@@ -43,5 +43,4 @@ tracing = "0.1"
 [features]
 # tidy-alphabetical-start
 rustc_randomized_layouts = []
-rustc_use_parallel_compiler = ["dep:rustc-rayon-core"]
 # tidy-alphabetical-end

compiler/rustc_middle/src/ty/context.rs

@@ -22,9 +22,9 @@ use rustc_data_structures::profiling::SelfProfilerRef;
 use rustc_data_structures::sharded::{IntoPointer, ShardedHashMap};
 use rustc_data_structures::stable_hasher::{HashStable, StableHasher};
 use rustc_data_structures::steal::Steal;
-use rustc_data_structures::sync::{self, FreezeReadGuard, Lock, Lrc, RwLock, WorkerLocal};
+use rustc_data_structures::sync::{
-#[cfg(parallel_compiler)]
+    self, DynSend, DynSync, FreezeReadGuard, Lock, Lrc, RwLock, WorkerLocal,
-use rustc_data_structures::sync::{DynSend, DynSync};
+};
 use rustc_data_structures::unord::UnordSet;
 use rustc_errors::{
     Applicability, Diag, DiagCtxtHandle, ErrorGuaranteed, LintDiagnostic, MultiSpan,
@@ -1259,9 +1259,7 @@ pub struct TyCtxt<'tcx> {
 }
 
 // Explicitly implement `DynSync` and `DynSend` for `TyCtxt` to short circuit trait resolution.
-#[cfg(parallel_compiler)]
 unsafe impl DynSend for TyCtxt<'_> {}
-#[cfg(parallel_compiler)]
 unsafe impl DynSync for TyCtxt<'_> {}
 fn _assert_tcx_fields() {
     sync::assert_dyn_sync::<&'_ GlobalCtxt<'_>>();
@@ -1383,9 +1381,7 @@ pub struct CurrentGcx {
     value: Lrc<RwLock<Option<*const ()>>>,
 }
 
-#[cfg(parallel_compiler)]
 unsafe impl DynSend for CurrentGcx {}
-#[cfg(parallel_compiler)]
 unsafe impl DynSync for CurrentGcx {}
 
 impl CurrentGcx {

compiler/rustc_middle/src/ty/context/tls.rs

@@ -1,5 +1,3 @@
-#[cfg(not(parallel_compiler))]
-use std::cell::Cell;
 use std::{mem, ptr};
 
 use rustc_data_structures::sync::{self, Lock};
@@ -50,16 +48,8 @@ impl<'a, 'tcx> ImplicitCtxt<'a, 'tcx> {
 }
 
 // Import the thread-local variable from Rayon, which is preserved for Rayon jobs.
-#[cfg(parallel_compiler)]
 use rayon_core::tlv::TLV;
 
-// Otherwise define our own
-#[cfg(not(parallel_compiler))]
-thread_local! {
-    /// A thread local variable that stores a pointer to the current `ImplicitCtxt`.
-    static TLV: Cell<*const ()> = const { Cell::new(ptr::null()) };
-}
-
 #[inline]
 fn erase(context: &ImplicitCtxt<'_, '_>) -> *const () {
     context as *const _ as *const ()

compiler/rustc_middle/src/ty/generic_args.rs

@@ -143,12 +143,10 @@ impl<'tcx> rustc_type_ir::inherent::IntoKind for GenericArg<'tcx> {
     }
 }
 
-#[cfg(parallel_compiler)]
 unsafe impl<'tcx> rustc_data_structures::sync::DynSend for GenericArg<'tcx> where
     &'tcx (Ty<'tcx>, ty::Region<'tcx>, ty::Const<'tcx>): rustc_data_structures::sync::DynSend
 {
 }
-#[cfg(parallel_compiler)]
 unsafe impl<'tcx> rustc_data_structures::sync::DynSync for GenericArg<'tcx> where
     &'tcx (Ty<'tcx>, ty::Region<'tcx>, ty::Const<'tcx>): rustc_data_structures::sync::DynSync
 {

compiler/rustc_middle/src/ty/list.rs

@@ -5,7 +5,6 @@ use std::ops::Deref;
 use std::{fmt, iter, mem, ptr, slice};
 
 use rustc_data_structures::aligned::{Aligned, align_of};
-#[cfg(parallel_compiler)]
 use rustc_data_structures::sync::DynSync;
 use rustc_serialize::{Encodable, Encoder};
 
@@ -259,7 +258,6 @@ impl<'a, H, T: Copy> IntoIterator for &'a RawList<H, T> {
 unsafe impl<H: Sync, T: Sync> Sync for RawList<H, T> {}
 
 // We need this since `List` uses extern type `OpaqueListContents`.
-#[cfg(parallel_compiler)]
 unsafe impl<H: DynSync, T: DynSync> DynSync for RawList<H, T> {}
 
 // Safety:

compiler/rustc_middle/src/ty/mod.rs

@@ -487,12 +487,10 @@ impl<'tcx> rustc_type_ir::inherent::IntoKind for Term<'tcx> {
     }
 }
 
-#[cfg(parallel_compiler)]
 unsafe impl<'tcx> rustc_data_structures::sync::DynSend for Term<'tcx> where
     &'tcx (Ty<'tcx>, Const<'tcx>): rustc_data_structures::sync::DynSend
 {
 }
-#[cfg(parallel_compiler)]
 unsafe impl<'tcx> rustc_data_structures::sync::DynSync for Term<'tcx> where
     &'tcx (Ty<'tcx>, Const<'tcx>): rustc_data_structures::sync::DynSync
 {

compiler/rustc_query_impl/Cargo.toml

@@ -19,8 +19,3 @@ rustc_span = { path = "../rustc_span" }
 thin-vec = "0.2.12"
 tracing = "0.1"
 # tidy-alphabetical-end
-
-[features]
-# tidy-alphabetical-start
-rustc_use_parallel_compiler = ["rustc_query_system/rustc_use_parallel_compiler"]
-# tidy-alphabetical-end

compiler/rustc_query_system/Cargo.toml

@@ -6,7 +6,7 @@ edition = "2021"
 [dependencies]
 # tidy-alphabetical-start
 parking_lot = "0.12"
-rustc-rayon-core = { version = "0.5.0", optional = true }
+rustc-rayon-core = { version = "0.5.0" }
 rustc_ast = { path = "../rustc_ast" }
 rustc_data_structures = { path = "../rustc_data_structures" }
 rustc_errors = { path = "../rustc_errors" }
@@ -23,8 +23,3 @@ smallvec = { version = "1.8.1", features = ["union", "may_dangle"] }
 thin-vec = "0.2.12"
 tracing = "0.1"
 # tidy-alphabetical-end
-
-[features]
-# tidy-alphabetical-start
-rustc_use_parallel_compiler = ["dep:rustc-rayon-core"]
-# tidy-alphabetical-end

compiler/rustc_query_system/src/dep_graph/graph.rs

@@ -837,12 +837,6 @@ impl<D: Deps> DepGraphData<D> {
     ) -> Option<DepNodeIndex> {
         let frame = MarkFrame { index: prev_dep_node_index, parent: frame };
 
-        #[cfg(not(parallel_compiler))]
-        {
-            debug_assert!(!self.dep_node_exists(dep_node));
-            debug_assert!(self.colors.get(prev_dep_node_index).is_none());
-        }
-
         // We never try to mark eval_always nodes as green
         debug_assert!(!qcx.dep_context().is_eval_always(dep_node.kind));
 
@@ -871,13 +865,6 @@ impl<D: Deps> DepGraphData<D> {
         // Maybe store a list on disk and encode this fact in the DepNodeState
         let side_effects = qcx.load_side_effects(prev_dep_node_index);
 
-        #[cfg(not(parallel_compiler))]
-        debug_assert!(
-            self.colors.get(prev_dep_node_index).is_none(),
-            "DepGraph::try_mark_previous_green() - Duplicate DepNodeColor \
-                      insertion for {dep_node:?}"
-        );
-
         if side_effects.maybe_any() {
             qcx.dep_context().dep_graph().with_query_deserialization(|| {
                 self.emit_side_effects(qcx, dep_node_index, side_effects)

compiler/rustc_query_system/src/query/job.rs

@@ -1,21 +1,16 @@
 use std::hash::Hash;
 use std::io::Write;
+use std::iter;
 use std::num::NonZero;
+use std::sync::Arc;
 
-use rustc_data_structures::fx::FxHashMap;
+use parking_lot::{Condvar, Mutex};
+use rustc_data_structures::fx::{FxHashMap, FxHashSet};
+use rustc_data_structures::jobserver;
 use rustc_errors::{Diag, DiagCtxtHandle};
 use rustc_hir::def::DefKind;
 use rustc_session::Session;
-use rustc_span::Span;
+use rustc_span::{DUMMY_SP, Span};
-#[cfg(parallel_compiler)]
-use {
-    parking_lot::{Condvar, Mutex},
-    rustc_data_structures::fx::FxHashSet,
-    rustc_data_structures::jobserver,
-    rustc_span::DUMMY_SP,
-    std::iter,
-    std::sync::Arc,
-};
 
 use crate::dep_graph::DepContext;
 use crate::error::CycleStack;
@@ -41,17 +36,14 @@ impl QueryJobId {
         map.get(&self).unwrap().query.clone()
     }
 
-    #[cfg(parallel_compiler)]
     fn span(self, map: &QueryMap) -> Span {
         map.get(&self).unwrap().job.span
     }
 
-    #[cfg(parallel_compiler)]
     fn parent(self, map: &QueryMap) -> Option<QueryJobId> {
         map.get(&self).unwrap().job.parent
     }
 
-    #[cfg(parallel_compiler)]
     fn latch(self, map: &QueryMap) -> Option<&QueryLatch> {
         map.get(&self).unwrap().job.latch.as_ref()
     }
@@ -75,7 +67,6 @@ pub struct QueryJob {
     pub parent: Option<QueryJobId>,
 
     /// The latch that is used to wait on this job.
-    #[cfg(parallel_compiler)]
     latch: Option<QueryLatch>,
 }
 
@@ -83,16 +74,9 @@ impl QueryJob {
     /// Creates a new query job.
     #[inline]
     pub fn new(id: QueryJobId, span: Span, parent: Option<QueryJobId>) -> Self {
-        QueryJob {
+        QueryJob { id, span, parent, latch: None }
-            id,
-            span,
-            parent,
-            #[cfg(parallel_compiler)]
-            latch: None,
-        }
     }
 
-    #[cfg(parallel_compiler)]
     pub(super) fn latch(&mut self) -> QueryLatch {
         if self.latch.is_none() {
             self.latch = Some(QueryLatch::new());
@@ -106,14 +90,11 @@ impl QueryJob {
     /// as there are no concurrent jobs which could be waiting on us
     #[inline]
     pub fn signal_complete(self) {
-        #[cfg(parallel_compiler)]
-        {
         if let Some(latch) = self.latch {
             latch.set();
         }
     }
 }
-}
 
 impl QueryJobId {
     pub(super) fn find_cycle_in_stack(
@@ -176,7 +157,6 @@ impl QueryJobId {
     }
 }
 
-#[cfg(parallel_compiler)]
 #[derive(Debug)]
 struct QueryWaiter {
     query: Option<QueryJobId>,
@@ -185,7 +165,6 @@ struct QueryWaiter {
     cycle: Mutex<Option<CycleError>>,
 }
 
-#[cfg(parallel_compiler)]
 impl QueryWaiter {
     fn notify(&self, registry: &rayon_core::Registry) {
         rayon_core::mark_unblocked(registry);
@@ -193,20 +172,17 @@ impl QueryWaiter {
     }
 }
 
-#[cfg(parallel_compiler)]
 #[derive(Debug)]
 struct QueryLatchInfo {
     complete: bool,
     waiters: Vec<Arc<QueryWaiter>>,
 }
 
-#[cfg(parallel_compiler)]
 #[derive(Clone, Debug)]
 pub(super) struct QueryLatch {
     info: Arc<Mutex<QueryLatchInfo>>,
 }
 
-#[cfg(parallel_compiler)]
 impl QueryLatch {
     fn new() -> Self {
         QueryLatch {
@@ -273,7 +249,6 @@ impl QueryLatch {
 }
 
 /// A resumable waiter of a query. The usize is the index into waiters in the query's latch
-#[cfg(parallel_compiler)]
 type Waiter = (QueryJobId, usize);
 
 /// Visits all the non-resumable and resumable waiters of a query.
@@ -285,7 +260,6 @@ type Waiter = (QueryJobId, usize);
 /// For visits of resumable waiters it returns Some(Some(Waiter)) which has the
 /// required information to resume the waiter.
 /// If all `visit` calls returns None, this function also returns None.
-#[cfg(parallel_compiler)]
 fn visit_waiters<F>(query_map: &QueryMap, query: QueryJobId, mut visit: F) -> Option<Option<Waiter>>
 where
     F: FnMut(Span, QueryJobId) -> Option<Option<Waiter>>,
@@ -316,7 +290,6 @@ where
 /// `span` is the reason for the `query` to execute. This is initially DUMMY_SP.
 /// If a cycle is detected, this initial value is replaced with the span causing
 /// the cycle.
-#[cfg(parallel_compiler)]
 fn cycle_check(
     query_map: &QueryMap,
     query: QueryJobId,
@@ -357,7 +330,6 @@ fn cycle_check(
 /// Finds out if there's a path to the compiler root (aka. code which isn't in a query)
 /// from `query` without going through any of the queries in `visited`.
 /// This is achieved with a depth first search.
-#[cfg(parallel_compiler)]
 fn connected_to_root(
     query_map: &QueryMap,
     query: QueryJobId,
@@ -380,7 +352,6 @@ fn connected_to_root(
 }
 
 // Deterministically pick an query from a list
-#[cfg(parallel_compiler)]
 fn pick_query<'a, T, F>(query_map: &QueryMap, queries: &'a [T], f: F) -> &'a T
 where
     F: Fn(&T) -> (Span, QueryJobId),
@@ -406,7 +377,6 @@ where
 /// the function return true.
 /// If a cycle was not found, the starting query is removed from `jobs` and
 /// the function returns false.
-#[cfg(parallel_compiler)]
 fn remove_cycle(
     query_map: &QueryMap,
     jobs: &mut Vec<QueryJobId>,
@@ -511,7 +481,6 @@ fn remove_cycle(
 /// uses a query latch and then resuming that waiter.
 /// There may be multiple cycles involved in a deadlock, so this searches
 /// all active queries for cycles before finally resuming all the waiters at once.
-#[cfg(parallel_compiler)]
 pub fn break_query_cycles(query_map: QueryMap, registry: &rayon_core::Registry) {
     let mut wakelist = Vec::new();
     let mut jobs: Vec<QueryJobId> = query_map.keys().cloned().collect();

compiler/rustc_query_system/src/query/mod.rs

@@ -2,10 +2,9 @@ mod plumbing;
 pub use self::plumbing::*;
 
 mod job;
-#[cfg(parallel_compiler)]
-pub use self::job::break_query_cycles;
 pub use self::job::{
-    QueryInfo, QueryJob, QueryJobId, QueryJobInfo, QueryMap, print_query_stack, report_cycle,
+    QueryInfo, QueryJob, QueryJobId, QueryJobInfo, QueryMap, break_query_cycles, print_query_stack,
+    report_cycle,
 };
 
 mod caches;
@@ -38,7 +37,6 @@ pub struct QueryStackFrame {
     pub dep_kind: DepKind,
     /// This hash is used to deterministically pick
     /// a query to remove cycles in the parallel compiler.
-    #[cfg(parallel_compiler)]
     hash: Hash64,
 }
 
@@ -51,18 +49,9 @@ impl QueryStackFrame {
         def_kind: Option<DefKind>,
         dep_kind: DepKind,
         ty_def_id: Option<DefId>,
-        _hash: impl FnOnce() -> Hash64,
+        hash: impl FnOnce() -> Hash64,
     ) -> Self {
-        Self {
+        Self { description, span, def_id, def_kind, ty_def_id, dep_kind, hash: hash() }
-            description,
-            span,
-            def_id,
-            def_kind,
-            ty_def_id,
-            dep_kind,
-            #[cfg(parallel_compiler)]
-            hash: _hash(),
-        }
     }
 
     // FIXME(eddyb) Get more valid `Span`s on queries.

compiler/rustc_query_system/src/query/plumbing.rs

@@ -13,7 +13,6 @@ use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::sharded::Sharded;
 use rustc_data_structures::stack::ensure_sufficient_stack;
 use rustc_data_structures::sync::Lock;
-#[cfg(parallel_compiler)]
 use rustc_data_structures::{outline, sync};
 use rustc_errors::{Diag, FatalError, StashKey};
 use rustc_span::{DUMMY_SP, Span};
@@ -25,9 +24,7 @@ use crate::HandleCycleError;
 use crate::dep_graph::{DepContext, DepGraphData, DepNode, DepNodeIndex, DepNodeParams};
 use crate::ich::StableHashingContext;
 use crate::query::caches::QueryCache;
-#[cfg(parallel_compiler)]
+use crate::query::job::{QueryInfo, QueryJob, QueryJobId, QueryJobInfo, QueryLatch, report_cycle};
-use crate::query::job::QueryLatch;
-use crate::query::job::{QueryInfo, QueryJob, QueryJobId, QueryJobInfo, report_cycle};
 use crate::query::{
     QueryContext, QueryMap, QuerySideEffects, QueryStackFrame, SerializedDepNodeIndex,
 };
@@ -263,7 +260,6 @@ where
 }
 
 #[inline(always)]
-#[cfg(parallel_compiler)]
 fn wait_for_query<Q, Qcx>(
     query: Q,
     qcx: Qcx,
@@ -334,7 +330,7 @@ where
     // re-executing the query since `try_start` only checks that the query is not currently
     // executing, but another thread may have already completed the query and stores it result
     // in the query cache.
-    if cfg!(parallel_compiler) && qcx.dep_context().sess().threads() > 1 {
+    if qcx.dep_context().sess().threads() > 1 {
         if let Some((value, index)) = query.query_cache(qcx).lookup(&key) {
             qcx.dep_context().profiler().query_cache_hit(index.into());
             return (value, Some(index));
@@ -359,7 +355,6 @@ where
         Entry::Occupied(mut entry) => {
             match entry.get_mut() {
                 QueryResult::Started(job) => {
-                    #[cfg(parallel_compiler)]
                     if sync::is_dyn_thread_safe() {
                         // Get the latch out
                         let latch = job.latch();

compiler/rustc_span/src/hygiene.rs

@@ -1300,7 +1300,6 @@ pub fn register_expn_id(
     let expn_id = ExpnId { krate, local_id };
     HygieneData::with(|hygiene_data| {
         let _old_data = hygiene_data.foreign_expn_data.insert(expn_id, data);
-        debug_assert!(_old_data.is_none() || cfg!(parallel_compiler));
         let _old_hash = hygiene_data.foreign_expn_hashes.insert(expn_id, hash);
         debug_assert!(_old_hash.is_none() || _old_hash == Some(hash));
         let _old_id = hygiene_data.expn_hash_to_expn_id.insert(hash, expn_id);
@@ -1423,18 +1422,7 @@ pub fn decode_syntax_context<D: Decoder, F: FnOnce(&mut D, u32) -> SyntaxContext
             ctxt_data = old.clone();
         }
 
-        let dummy = std::mem::replace(
+        hygiene_data.syntax_context_data[ctxt.as_u32() as usize] = ctxt_data;
-            &mut hygiene_data.syntax_context_data[ctxt.as_u32() as usize],
-            ctxt_data,
-        );
-        if cfg!(not(parallel_compiler)) {
-            // Make sure nothing weird happened while `decode_data` was running.
-            // We used `kw::Empty` for the dummy value and we expect nothing to be
-            // modifying the dummy entry.
-            // This does not hold for the parallel compiler as another thread may
-            // have inserted the fully decoded data.
-            assert_eq!(dummy.dollar_crate_name, kw::Empty);
-        }
     });
 
     // Mark the context as completed

compiler/rustc_span/src/lib.rs

@@ -521,14 +521,6 @@ impl SpanData {
     }
 }
 
-// The interner is pointed to by a thread local value which is only set on the main thread
-// with parallelization is disabled. So we don't allow `Span` to transfer between threads
-// to avoid panics and other errors, even though it would be memory safe to do so.
-#[cfg(not(parallel_compiler))]
-impl !Send for Span {}
-#[cfg(not(parallel_compiler))]
-impl !Sync for Span {}
-
 impl PartialOrd for Span {
     fn partial_cmp(&self, rhs: &Self) -> Option<Ordering> {
         PartialOrd::partial_cmp(&self.data(), &rhs.data())

config.example.toml

@@ -594,8 +594,7 @@
 
 # Build a multi-threaded rustc. This allows users to use parallel rustc
 # via the unstable option `-Z threads=n`.
-# Since stable/beta channels only allow using stable features,
+# This option is deprecated and always true.
-# `parallel-compiler = false` should be set for these channels.
 #parallel-compiler = true
 
 # The default linker that will be hard-coded into the generated

src/bootstrap/src/core/builder/cargo.rs

@@ -1198,15 +1198,6 @@ impl Builder<'_> {
             rustflags.arg("-Zinline-mir-preserve-debug");
         }
 
-        if self.config.rustc_parallel
-            && matches!(mode, Mode::ToolRustc | Mode::Rustc | Mode::Codegen)
-        {
-            // keep in sync with `bootstrap/lib.rs:Build::rustc_features`
-            // `cfg` option for rustc, `features` option for cargo, for conditional compilation
-            rustflags.arg("--cfg=parallel_compiler");
-            rustdocflags.arg("--cfg=parallel_compiler");
-        }
-
         Cargo {
             command: cargo,
             compiler,

src/bootstrap/src/core/config/config.rs

@@ -278,7 +278,6 @@ pub struct Config {
     pub rust_strip: bool,
     pub rust_frame_pointers: bool,
     pub rust_stack_protector: Option<String>,
-    pub rustc_parallel: bool,
     pub rustc_default_linker: Option<String>,
     pub rust_optimize_tests: bool,
     pub rust_dist_src: bool,
@@ -1224,7 +1223,6 @@ impl Config {
             bindir: "bin".into(),
             dist_include_mingw_linker: true,
             dist_compression_profile: "fast".into(),
-            rustc_parallel: true,
 
             stdout_is_tty: std::io::stdout().is_terminal(),
             stderr_is_tty: std::io::stderr().is_terminal(),
@@ -1773,8 +1771,14 @@ impl Config {
 
             config.rust_randomize_layout = randomize_layout.unwrap_or_default();
             config.llvm_tools_enabled = llvm_tools.unwrap_or(true);
-            config.rustc_parallel =
+
-                parallel_compiler.unwrap_or(config.channel == "dev" || config.channel == "nightly");
+            // FIXME: Remove this option at the end of 2024.
+            if parallel_compiler.is_some() {
+                println!(
+                    "WARNING: The `rust.parallel-compiler` option is deprecated and does nothing. The parallel compiler (with one thread) is now the default"
+                );
+            }
+
             config.llvm_enzyme =
                 llvm_enzyme.unwrap_or(config.channel == "dev" || config.channel == "nightly");
             config.rustc_default_linker = default_linker;

src/bootstrap/src/lib.rs

@@ -80,11 +80,8 @@ const EXTRA_CHECK_CFGS: &[(Option<Mode>, &str, Option<&[&'static str]>)] = &[
     (Some(Mode::Rustc), "llvm_enzyme", None),
     (Some(Mode::Codegen), "llvm_enzyme", None),
     (Some(Mode::ToolRustc), "llvm_enzyme", None),
-    (Some(Mode::Rustc), "parallel_compiler", None),
-    (Some(Mode::ToolRustc), "parallel_compiler", None),
     (Some(Mode::ToolRustc), "rust_analyzer", None),
     (Some(Mode::ToolStd), "rust_analyzer", None),
-    (Some(Mode::Codegen), "parallel_compiler", None),
     // Any library specific cfgs like `target_os`, `target_arch` should be put in
     // priority the `[lints.rust.unexpected_cfgs.check-cfg]` table
     // in the appropriate `library/{std,alloc,core}/Cargo.toml`
@@ -695,9 +692,6 @@ impl Build {
             features.push("llvm");
         }
         // keep in sync with `bootstrap/compile.rs:rustc_cargo_env`
-        if self.config.rustc_parallel && check("rustc_use_parallel_compiler") {
-            features.push("rustc_use_parallel_compiler");
-        }
         if self.config.rust_randomize_layout {
             features.push("rustc_randomized_layouts");
         }

src/bootstrap/src/utils/change_tracker.rs

@@ -290,6 +290,11 @@ pub const CONFIG_CHANGE_HISTORY: &[ChangeInfo] = &[
         severity: ChangeSeverity::Info,
         summary: "New option `llvm.offload` to control whether the llvm offload runtime for GPU support is built. Implicitly enables the openmp runtime as dependency.",
     },
+    ChangeInfo {
+        change_id: 132282,
+        severity: ChangeSeverity::Warning,
+        summary: "Deprecated `rust.parallel_compiler` as the compiler now always defaults to being parallel (with 1 thread)",
+    },
     ChangeInfo {
         change_id: 132494,
         severity: ChangeSeverity::Info,

src/ci/run.sh

@@ -120,9 +120,6 @@ if [ "$DEPLOY$DEPLOY_ALT" = "1" ]; then
   if [ "$NO_LLVM_ASSERTIONS" = "1" ]; then
     RUST_CONFIGURE_ARGS="$RUST_CONFIGURE_ARGS --disable-llvm-assertions"
   elif [ "$DEPLOY_ALT" != "" ]; then
-    if [ "$ALT_PARALLEL_COMPILER" = "" ]; then
-      RUST_CONFIGURE_ARGS="$RUST_CONFIGURE_ARGS --set rust.parallel-compiler=false"
-    fi
     RUST_CONFIGURE_ARGS="$RUST_CONFIGURE_ARGS --enable-llvm-assertions"
     RUST_CONFIGURE_ARGS="$RUST_CONFIGURE_ARGS --set rust.verify-llvm-ir"
   fi
@@ -263,7 +260,7 @@ fi
 
 if [ "$RUN_CHECK_WITH_PARALLEL_QUERIES" != "" ]; then
   rm -f config.toml
-  $SRC/configure --set change-id=99999999 --set rust.parallel-compiler
+  $SRC/configure --set change-id=99999999
 
   # Save the build metrics before we wipe the directory
   if [ "$HAS_METRICS" = 1 ]; then

src/librustdoc/lib.rs

@@ -206,7 +206,7 @@ fn init_logging(early_dcx: &EarlyDiagCtxt) {
         .with_verbose_exit(true)
         .with_verbose_entry(true)
         .with_indent_amount(2);
-    #[cfg(all(parallel_compiler, debug_assertions))]
+    #[cfg(debug_assertions)]
     let layer = layer.with_thread_ids(true).with_thread_names(true);
 
     use tracing_subscriber::layer::SubscriberExt;