[libcxx-commits] [libcxx] [libc++] Speed up set_intersection() by fast-forwarding over ranges of non-matching elements with one-sided binary search. (PR #75230)

[Louis Dionne via libcxx-commits]

================
@@ -733,8 +733,24 @@ struct common_input_iterator {
 // * `equals_count`, which records the total number of calls to an op== or op!=. If compared
 //   against a sentinel object, that sentinel object must call the `record_equality_comparison`
 //   function so that the comparison is counted correctly.
-template <class It>
+template <class It,
+          class StrideCountType        = std::iter_difference_t<It>,
+          class StrideDisplacementType = std::iter_difference_t<It>>
 class stride_counting_iterator {
----------------
ldionne wrote:

I would prefer instead changing `stride_counting_iterator` in the following way:

```c++
struct IteratorOperations {
  std::size_t increments;
  std::size_t decrements;
  std::size_t comparisons;
  // etc...
};

template <...>
struct operation_counting_iterator {
  explicit operation_counting_iterator(Iter underlying, IteratorOperations* counter /* may be nullptr */);
};

// then use it as
IteratorOperations ops;
operation_counting_iterator it(x.begin(), &ops);
```

Since `stride_counting_iterator` doesn't have that much use, I think it shouldn't be too hard to change.

Note that if you want, this could be done in a separate patch prior to this patch. Depending on how large this ends up being.

https://github.com/llvm/llvm-project/pull/75230