Optimize Ord trait implementation for bool

Casting the booleans to `i8`s and converting their difference
into `Ordering` generates better assembly than casting them to
`u8`s and comparing them.

Author: Krishna Sai Veera Reddy

src/libcore/cmp.rs

@@ -1006,6 +1006,7 @@ pub fn max_by_key<T, F: FnMut(&T) -> K, K: Ord>(v1: T, v2: T, mut f: F) -> T {
 
 // Implementation of PartialEq, Eq, PartialOrd and Ord for primitive types
 mod impls {
+    use crate::hint::unreachable_unchecked;
     use crate::cmp::Ordering::{self, Less, Greater, Equal};
 
     macro_rules! partial_eq_impl {
@@ -1126,7 +1127,15 @@ mod impls {
     impl Ord for bool {
         #[inline]
         fn cmp(&self, other: &bool) -> Ordering {
-            (*self as u8).cmp(&(*other as u8))
+            // Casting to i8's and converting the difference to an Ordering generates
+            // more optimal assembly.
+            // See <https://github.com/rust-lang/rust/issues/66780> for more info.
+            match (*self as i8) - (*other as i8) {
+                -1 => Less,
+                0 => Equal,
+                1 => Greater,
+                _ => unsafe { unreachable_unchecked() },
+            }
         }
     }
 

src/libcore/tests/cmp.rs

@@ -9,6 +9,14 @@ fn test_int_totalord() {
     assert_eq!(12.cmp(&-5), Greater);
 }
 
+#[test]
+fn test_bool_totalord() {
+    assert_eq!(true.cmp(&false), Greater);
+    assert_eq!(false.cmp(&true), Less);
+    assert_eq!(true.cmp(&true), Equal);
+    assert_eq!(false.cmp(&false), Equal);
+}
+
 #[test]
 fn test_mut_int_totalord() {
     assert_eq!((&mut 5).cmp(&&mut 10), Less);

src/test/codegen/bool-cmp.rs

@@ -0,0 +1,17 @@
+// This is a test for optimal Ord trait implementation for bool.
+// See <https://github.com/rust-lang/rust/issues/66780> for more info.
+
+// compile-flags: -C opt-level=3
+
+#![crate_type = "lib"]
+
+use std::cmp::Ordering;
+
+// CHECK-LABEL: @cmp_bool
+#[no_mangle]
+pub fn cmp_bool(a: bool, b: bool) -> Ordering {
+// CHECK: zext i1
+// CHECK: zext i1
+// CHECK: sub nsw
+    a.cmp(&b)
+}