Simplify internals of {Rc,Arc}::default

This commit simplifies the internal implementation of `Default` for
these two pointer types to have the same performance characteristics as
before (a side effect of changes in 131460) while avoid use of internal
private APIs of Rc/Arc. To preserve the same codegen as before some
non-generic functions needed to be tagged as `#[inline]` as well, but
otherwise the same IR is produced before/after this change.

The motivation of this commit is I was studying up on the state of
initialization of `Arc` and `Rc` and figured it'd be nicer to reduce the
use of internal APIs and instead use public stable APIs where possible,
even in the implementation itself.

library/alloc/src/rc.rs

@@ -289,6 +289,7 @@ struct RcInner<T: ?Sized> {
 }
 
 /// Calculate layout for `RcInner<T>` using the inner value's layout
+#[inline]
 fn rc_inner_layout_for_value_layout(layout: Layout) -> Layout {
     // Calculate layout using the given value layout.
     // Previously, layout was calculated on the expression
@@ -2518,15 +2519,25 @@ impl<T: Default> Default for Rc<T> {
     /// ```
     #[inline]
     fn default() -> Self {
+        // First create an uninitialized allocation before creating an instance
+        // of `T`. This avoids having `T` on the stack and avoids the need to
+        // codegen a call to the destructor for `T` leading to generally better
+        // codegen. See #131460 for some more details.
+        let mut rc = Rc::new_uninit();
+
+        // SAFETY: this is a freshly allocated `Rc` so it's guaranteed there are
+        // no other strong or weak pointers other than `rc` itself.
         unsafe {
-            Self::from_inner(
-                Box::leak(Box::write(
-                    Box::new_uninit(),
-                    RcInner { strong: Cell::new(1), weak: Cell::new(1), value: T::default() },
-                ))
-                .into(),
-            )
+            let raw = Rc::get_mut_unchecked(&mut rc);
+
+            // Note that `ptr::write` here is used specifically instead of
+            // `MaybeUninit::write` to avoid creating an extra stack copy of `T`
+            // in debug mode. See #136043 for more context.
+            ptr::write(raw.as_mut_ptr(), T::default());
         }
+
+        // SAFETY: this allocation was just initialized above.
+        unsafe { rc.assume_init() }
     }
 }
 

library/alloc/src/sync.rs

@@ -392,6 +392,7 @@ struct ArcInner<T: ?Sized> {
 }
 
 /// Calculate layout for `ArcInner<T>` using the inner value's layout
+#[inline]
 fn arcinner_layout_for_value_layout(layout: Layout) -> Layout {
     // Calculate layout using the given value layout.
     // Previously, layout was calculated on the expression
@@ -3724,19 +3725,25 @@ impl<T: Default> Default for Arc<T> {
     /// assert_eq!(*x, 0);
     /// ```
     fn default() -> Arc<T> {
+        // First create an uninitialized allocation before creating an instance
+        // of `T`. This avoids having `T` on the stack and avoids the need to
+        // codegen a call to the destructor for `T` leading to generally better
+        // codegen. See #131460 for some more details.
+        let mut arc = Arc::new_uninit();
+
+        // SAFETY: this is a freshly allocated `Arc` so it's guaranteed there
+        // are no other strong or weak pointers other than `arc` itself.
         unsafe {
-            Self::from_inner(
-                Box::leak(Box::write(
-                    Box::new_uninit(),
-                    ArcInner {
-                        strong: atomic::AtomicUsize::new(1),
-                        weak: atomic::AtomicUsize::new(1),
-                        data: T::default(),
-                    },
-                ))
-                .into(),
-            )
+            let raw = Arc::get_mut_unchecked(&mut arc);
+
+            // Note that `ptr::write` here is used specifically instead of
+            // `MaybeUninit::write` to avoid creating an extra stack copy of `T`
+            // in debug mode. See #136043 for more context.
+            ptr::write(raw.as_mut_ptr(), T::default());
         }
+
+        // SAFETY: this allocation was just initialized above.
+        unsafe { arc.assume_init() }
     }
 }
 

tests/codegen-llvm/issues/issue-111603.rs

@@ -10,9 +10,9 @@
 pub fn new_from_array(x: u64) -> Arc<[u64]> {
     // Ensure that we only generate one alloca for the array.
 
-    // CHECK: alloca
+    // CHECK: %[[A:.+]] = alloca
     // CHECK-SAME: [8000 x i8]
-    // CHECK-NOT: alloca
+    // CHECK-NOT: %[[B:.+]] = alloca
     let array = [x; 1000];
     Arc::new(array)
 }
@@ -20,8 +20,9 @@ pub fn new_from_array(x: u64) -> Arc<[u64]> {
 // CHECK-LABEL: @new_uninit
 #[no_mangle]
 pub fn new_uninit(x: u64) -> Arc<[u64; 1000]> {
-    // CHECK: call alloc::sync::arcinner_layout_for_value_layout
-    // CHECK-NOT: call alloc::sync::arcinner_layout_for_value_layout
+    // CHECK: %[[A:.+]] = alloca
+    // CHECK-SAME: [8000 x i8]
+    // CHECK-NOT: %[[B:.+]] = alloca
     let mut arc = Arc::new_uninit();
     unsafe { Arc::get_mut_unchecked(&mut arc) }.write([x; 1000]);
     unsafe { arc.assume_init() }
@@ -30,8 +31,7 @@ pub fn new_from_array(x: u64) -> Arc<[u64]> {
 // CHECK-LABEL: @new_uninit_slice
 #[no_mangle]
 pub fn new_uninit_slice(x: u64) -> Arc<[u64]> {
-    // CHECK: call alloc::sync::arcinner_layout_for_value_layout
-    // CHECK-NOT: call alloc::sync::arcinner_layout_for_value_layout
+    // CHECK-NOT: %[[B:.+]] = alloca
     let mut arc = Arc::new_uninit_slice(1000);
     for elem in unsafe { Arc::get_mut_unchecked(&mut arc) } {
         elem.write(x);