x86_64: Remove needless test in CAS

Before:

```
xchg    r10, rbx
lock            cmpxchg16b      xmmword ptr [rdi]
sete    r8b
mov     rbx, r10

xor     ecx, ecx
test    r8b, r8b
sete    cl
```

After:

```
xchg    r10, rbx
lock            cmpxchg16b      xmmword ptr [rdi]
setne   r8b
mov     rbx, r10

xor     r8b, 1
movzx   ecx, r8b
```

Note: There is xor in "After" because it is an assembly generated for a
function that returns Result<u128, u128>
(discriminant of Result: 0==Ok,1==Err, bool: 0==false,1=true).

Author: taiki-e

src/imp/atomic128/x86_64.rs

@@ -10,7 +10,7 @@
 // - atomic-maybe-uninit https://github.com/taiki-e/atomic-maybe-uninit
 //
 // Generated asm:
-// - x86_64 (+cmpxchg16b) https://godbolt.org/z/YnYE9qT6b
+// - x86_64 (+cmpxchg16b) https://godbolt.org/z/r5x9M8PdK
 
 include!("macros.rs");
 
@@ -124,6 +124,7 @@ unsafe fn cmpxchg16b(dst: *mut u128, old: u128, new: u128) -> (u128, bool) {
         cmpxchg16b!("edi");
         #[cfg(target_pointer_width = "64")]
         cmpxchg16b!("rdi");
+        crate::utils::assert_unchecked(r == 0 || r == 1); // needed to remove extra test
         (U128 { pair: Pair { lo: prev_lo, hi: prev_hi } }.whole, r != 0)
     }
 }

src/imp/x86.rs

@@ -148,6 +148,7 @@ macro_rules! atomic_bit_opts {
                         // Do not use `preserves_flags` because BTS modifies the CF flag.
                         options(nostack),
                     );
+                    crate::utils::assert_unchecked(r == 0 || r == 1); // may help remove extra test
                     r != 0
                 }
             }
@@ -172,6 +173,7 @@ macro_rules! atomic_bit_opts {
                         // Do not use `preserves_flags` because BTR modifies the CF flag.
                         options(nostack),
                     );
+                    crate::utils::assert_unchecked(r == 0 || r == 1); // may help remove extra test
                     r != 0
                 }
             }
@@ -196,6 +198,7 @@ macro_rules! atomic_bit_opts {
                         // Do not use `preserves_flags` because BTC modifies the CF flag.
                         options(nostack),
                     );
+                    crate::utils::assert_unchecked(r == 0 || r == 1); // may help remove extra test
                     r != 0
                 }
             }

src/utils.rs

@@ -246,6 +246,22 @@ macro_rules! items {
     };
 }
 
+#[allow(dead_code)]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+// Stable version of https://doc.rust-lang.org/nightly/std/hint/fn.assert_unchecked.html.
+#[inline(always)]
+#[cfg_attr(all(debug_assertions, not(portable_atomic_no_track_caller)), track_caller)]
+pub(crate) unsafe fn assert_unchecked(cond: bool) {
+    if !cond {
+        if cfg!(debug_assertions) {
+            unreachable!()
+        } else {
+            // SAFETY: the caller promised `cond` is true.
+            unsafe { core::hint::unreachable_unchecked() }
+        }
+    }
+}
+
 // https://github.com/rust-lang/rust/blob/1.70.0/library/core/src/sync/atomic.rs#L3155
 #[inline]
 #[cfg_attr(all(debug_assertions, not(portable_atomic_no_track_caller)), track_caller)]