Auto merge of rust-lang#136771 - scottmcm:poke-slice-iter-next, r=joboet

Simplify `slice::Iter::next` enough that it inlines

Inspired by this zulip conversation: <https://rust-lang.zulipchat.com/#narrow/channel/189540-t-compiler.2Fwg-mir-opt/topic/Feedback.20on.20a.20MIR.20optimization.20idea/near/498579990>

~~Draft for now because it needs rust-lang#136735 to get the codegen tests to pass.~~

Author: bors

core/src/slice/iter/macros.rs

@@ -154,16 +154,39 @@ macro_rules! iterator {
 
             #[inline]
             fn next(&mut self) -> Option<$elem> {
-                // could be implemented with slices, but this avoids bounds checks
+                // intentionally not using the helpers because this is
+                // one of the most mono'd things in the library.
 
-                // SAFETY: The call to `next_unchecked` is
-                // safe since we check if the iterator is empty first.
+                let ptr = self.ptr;
+                let end_or_len = self.end_or_len;
+                // SAFETY: See inner comments. (For some reason having multiple
+                // block breaks inlining this -- if you can fix that please do!)
                 unsafe {
-                    if is_empty!(self) {
-                        None
+                    if T::IS_ZST {
+                        let len = end_or_len.addr();
+                        if len == 0 {
+                            return None;
+                        }
+                        // SAFETY: just checked that it's not zero, so subtracting one
+                        // cannot wrap.  (Ideally this would be `checked_sub`, which
+                        // does the same thing internally, but as of 2025-02 that
+                        // doesn't optimize quite as small in MIR.)
+                        self.end_or_len = without_provenance_mut(len.unchecked_sub(1));
                     } else {
-                        Some(self.next_unchecked())
+                        // SAFETY: by type invariant, the `end_or_len` field is always
+                        // non-null for a non-ZST pointee.  (This transmute ensures we
+                        // get `!nonnull` metadata on the load of the field.)
+                        if ptr == crate::intrinsics::transmute::<$ptr, NonNull<T>>(end_or_len) {
+                            return None;
+                        }
+                        // SAFETY: since it's not empty, per the check above, moving
+                        // forward one keeps us inside the slice, and this is valid.
+                        self.ptr = ptr.add(1);
                     }
+                    // SAFETY: Now that we know it wasn't empty and we've moved past
+                    // the first one (to avoid giving a duplicate `&mut` next time),
+                    // we can give out a reference to it.
+                    Some({ptr}.$into_ref())
                 }
             }