Further optimize SliceIndex<str> impl for Range<usize>

We can shave a further 2 `icmp`s by inlining `is_char_boundary` and rearranging
the checks.

Author: Kmeakin

library/core/src/str/traits.rs

@@ -118,6 +118,29 @@ unsafe impl const SliceIndex<str> for ops::RangeFull {
     }
 }
 
+#[inline(always)]
+const fn check_range(slice: &str, range: crate::range::Range<usize>) -> bool {
+    let crate::range::Range { start, end } = range;
+    let bytes = slice.as_bytes();
+
+    if start > end || end > slice.len() {
+        return false;
+    }
+
+    if start == slice.len() {
+        // If `start == slice.len()`, then `end == slice.len()` must also be true.
+        return true;
+    }
+
+    // SAFETY:
+    // `start > end || end > slice.len()` is false, so `start <= end <= slice.len()` is true.
+    // `start == slice.len()` is false, so `start < slice.len()` is also true.
+    unsafe {
+        (start == 0 || bytes.as_ptr().add(start).read().is_utf8_char_boundary())
+            && (end == slice.len() || bytes.as_ptr().add(end).read().is_utf8_char_boundary())
+    }
+}
+
 /// Implements substring slicing with syntax `&self[begin .. end]` or `&mut
 /// self[begin .. end]`.
 ///
@@ -159,30 +182,11 @@ unsafe impl const SliceIndex<str> for ops::Range<usize> {
     type Output = str;
     #[inline]
     fn get(self, slice: &str) -> Option<&Self::Output> {
-        if self.start <= self.end
-            && slice.is_char_boundary(self.start)
-            && slice.is_char_boundary(self.end)
-        {
-            // SAFETY: just checked that `start` and `end` are on a char boundary,
-            // and we are passing in a safe reference, so the return value will also be one.
-            // We also checked char boundaries, so this is valid UTF-8.
-            Some(unsafe { &*self.get_unchecked(slice) })
-        } else {
-            None
-        }
+        range::Range::from(self).get(slice)
     }
     #[inline]
     fn get_mut(self, slice: &mut str) -> Option<&mut Self::Output> {
-        if self.start <= self.end
-            && slice.is_char_boundary(self.start)
-            && slice.is_char_boundary(self.end)
-        {
-            // SAFETY: just checked that `start` and `end` are on a char boundary.
-            // We know the pointer is unique because we got it from `slice`.
-            Some(unsafe { &mut *self.get_unchecked_mut(slice) })
-        } else {
-            None
-        }
+        range::Range::from(self).get_mut(slice)
     }
     #[inline]
     #[track_caller]
@@ -235,26 +239,11 @@ unsafe impl const SliceIndex<str> for ops::Range<usize> {
     }
     #[inline]
     fn index(self, slice: &str) -> &Self::Output {
-        let (start, end) = (self.start, self.end);
-        match self.get(slice) {
-            Some(s) => s,
-            None => super::slice_error_fail(slice, start, end),
-        }
+        range::Range::from(self).index(slice)
     }
     #[inline]
     fn index_mut(self, slice: &mut str) -> &mut Self::Output {
-        // is_char_boundary checks that the index is in [0, .len()]
-        // cannot reuse `get` as above, because of NLL trouble
-        if self.start <= self.end
-            && slice.is_char_boundary(self.start)
-            && slice.is_char_boundary(self.end)
-        {
-            // SAFETY: just checked that `start` and `end` are on a char boundary,
-            // and we are passing in a safe reference, so the return value will also be one.
-            unsafe { &mut *self.get_unchecked_mut(slice) }
-        } else {
-            super::slice_error_fail(slice, self.start, self.end)
-        }
+        range::Range::from(self).index_mut(slice)
     }
 }
 
@@ -264,10 +253,7 @@ unsafe impl const SliceIndex<str> for range::Range<usize> {
     type Output = str;
     #[inline]
     fn get(self, slice: &str) -> Option<&Self::Output> {
-        if self.start <= self.end
-            && slice.is_char_boundary(self.start)
-            && slice.is_char_boundary(self.end)
-        {
+        if check_range(slice, self) {
             // SAFETY: just checked that `start` and `end` are on a char boundary,
             // and we are passing in a safe reference, so the return value will also be one.
             // We also checked char boundaries, so this is valid UTF-8.
@@ -278,10 +264,7 @@ unsafe impl const SliceIndex<str> for range::Range<usize> {
     }
     #[inline]
     fn get_mut(self, slice: &mut str) -> Option<&mut Self::Output> {
-        if self.start <= self.end
-            && slice.is_char_boundary(self.start)
-            && slice.is_char_boundary(self.end)
-        {
+        if check_range(slice, self) {
             // SAFETY: just checked that `start` and `end` are on a char boundary.
             // We know the pointer is unique because we got it from `slice`.
             Some(unsafe { &mut *self.get_unchecked_mut(slice) })
@@ -350,10 +333,7 @@ unsafe impl const SliceIndex<str> for range::Range<usize> {
     fn index_mut(self, slice: &mut str) -> &mut Self::Output {
         // is_char_boundary checks that the index is in [0, .len()]
         // cannot reuse `get` as above, because of NLL trouble
-        if self.start <= self.end
-            && slice.is_char_boundary(self.start)
-            && slice.is_char_boundary(self.end)
-        {
+        if check_range(slice, self) {
             // SAFETY: just checked that `start` and `end` are on a char boundary,
             // and we are passing in a safe reference, so the return value will also be one.
             unsafe { &mut *self.get_unchecked_mut(slice) }

tests/codegen-llvm/str-range-indexing.rs

@@ -18,18 +18,19 @@ macro_rules! tests {
     };
 }
 
-// 9 comparisons required:
-// start <= end
-// && (start == 0 || (start >= len && start == len) || bytes[start] >= -0x40)
-// && (end   == 0 || (end   >= len && end   == len) || bytes[end]   >= -0x40)
+// 7 comparisons required:
+// start <= end && end <= len
+// && (start == len ||
+// (  (start == 0   || bytes[start] >= -0x40)
+// && (end   == len || bytes[end] >= -0x40)))
 
 // CHECK-LABEL: @get_range
-// CHECK-COUNT-9: %{{.+}} = icmp
+// CHECK-COUNT-7: %{{.+}} = icmp
 // CHECK-NOT: %{{.+}} = icmp
 // CHECK: ret
 
 // CHECK-LABEL: @index_range
-// CHECK-COUNT-9: %{{.+}} = icmp
+// CHECK-COUNT-7: %{{.+}} = icmp
 // CHECK-NOT: %{{.+}} = icmp
 // CHECK: ret
 tests!(Range<usize>, get_range, index_range);