RangeMap: A int->int mapping with fast range operations

media/blink/interval_map.h

Index: media/blink/interval_map.h
diff --git a/media/blink/interval_map.h b/media/blink/interval_map.h
new file mode 100644
index 0000000000000000000000000000000000000000..556f9ce6b0c211d5b0c7d9c750bfeb26f49675ee
--- /dev/null
+++ b/media/blink/interval_map.h
@@ -0,0 +1,284 @@
+// Copyright 2015 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef MEDIA_BLINK_INTERVAL_MAP_H_
+#define MEDIA_BLINK_INTERVAL_MAP_H_
+
+#include <limits>
+#include <map>
+
+#include "base/logging.h"
+
+namespace media {
+
+// An IntervalMap<KeyType, ValueType> maps every value of KeyType to
+// a ValueType, and incrementing, decrementing and setting ranges of values
+// has been optimized. The default state is to map all values in
+// KeyType to ValueType(). (Which is usually zero.)
+//
+// Set/Increment operations should generally take
+// O(log(N)) + O(M) time where N is the number of ranges in the map and
+// M is the number of modified ranges.
+//
+// Internally, IntervalMap<> uses an std::map, where the beginning of each range
+// is stored along with the value for that range. Adjacent ranges which have the
+// same value are automatically merged. For instance, if you did:
+//
+// IntervalMap<int, int> tmp;
+// tmp.IncrementRange(2, 5, 2);
+// tmp.IncrementRange(4, 6, 1);
+//
+// Then:
+//  tmp[0] = 0
+//  tmp[1] = 0
+//  tmp[2] = 2
+//  tmp[3] = 2
+//  tmp[4] = 3
+//  tmp[5] = 1
+//  tmp[6] = 0
+//
+// If you iterate over tmp, you get the following ranges:
+//  -maxint .. 2 => 0
+//  2 .. 4 => 2
+//  4 .. 5 => 3
+//  5 .. 6 => 1
+//  6 .. maxint => 0
+//
+// Internally, this would be stored in a map as:
+//    -maxint:0, 2:2, 4:3, 5:1, 6:0
+//
+// TODO(hubbe): Consider consolidating with media::Ranges.
+
+// Simple range class.
+// Range ends are always non-inclusive.
+// Please note that end <= begin is a valid (but empty) range.
+template <typename T>
+struct Range {

[xhwang, 2015/11/12 18:24:58]

should this be renamed to Interval as well?

[hubbe, 2015/11/12 19:03:13]

Done.

+ public:
+  Range(const T& begin, const T& end) : begin(begin), end(end) {}
+
+  // May return empty ranges (begin >= end).
+  Range Intersect(const Range& other) const {
+    return Range(std::max(begin, other.begin), std::min(end, other.end));
+  }
+
+  bool Empty() const { return begin >= end; }
+
+  T begin;
+  T end;
+};
+
+template <typename KeyType,
+          typename ValueType,
+          class Compare = std::less<KeyType>,
+          class NumericLimits = std::numeric_limits<KeyType>>
+class IntervalMapConstIterator {

[xhwang, 2015/11/12 18:24:58]

Add a comment that this represents an interval of equal values...

[hubbe, 2015/11/12 19:03:13]

Done.

+ public:
+  typedef std::map<KeyType, ValueType, Compare> MapType;
+  IntervalMapConstIterator() {}
+  IntervalMapConstIterator(const MapType* map,
+                           typename MapType::const_iterator iter)
+      : map_(map), iter_(iter) {}
+
+  bool operator==(const IntervalMapConstIterator& other) const {
+    return iter_ == other.iter_;
+  }
+
+  bool operator!=(const IntervalMapConstIterator& other) const {
+    return iter_ != other.iter_;
+  }
+
+  // Returns the beginning of the current range.
+  KeyType range_begin() const {

[xhwang, 2015/11/12 18:24:58]

should we replace all "range" to "interval"?

[hubbe, 2015/11/12 19:03:13]

Done.

+    DCHECK(iter_ != map_->end());
+    return iter_->first;
+  }
+
+  // Returns the end of the current range, non-inclusive.
+  KeyType range_end() const {

[xhwang, 2015/11/12 18:24:58]

Wondering how range_end() is used? It seems the caller should be able to get
this by (iter++)->range_begin()?

[hubbe, 2015/11/12 19:03:12]

They could, except ++iter might point to end().
Requiring the user to replicate this code every time they want the end of the
interval seems annoying.

+    DCHECK(iter_ != map_->end());
+    typename MapType::const_iterator next = iter_;
+    ++next;
+    if (next == map_->end()) {
+      return NumericLimits::max();
+    } else {
+      return next->first;
+    }
+  }
+
+  // Returns the current range.
+  Range<KeyType> range() const {
+    return Range<KeyType>(range_begin(), range_end());
+  }
+
+  // Returns the value associated with the current range.
+  ValueType value() const {
+    DCHECK(iter_ != map_->end());
+    return iter_->second;
+  }
+
+  // Needed to make the following construct work:
+  // for (const auto& range_value_pair : interval_map)
+  // Note however that this will skip the "end" range, which
+  // is usually ok since it generally has the default value.
+  std::pair<Range<KeyType>, ValueType> operator*() const {
+    return std::make_pair(range(), value());
+  }
+
+  // Go to the next range.
+  // The beginning of the next range always matches the end of the current
+  // range. (But should always have a different value.)
+  // Not allowed if we're already at map_->end().
+  void operator++() {
+    DCHECK(iter_ != map_->end());
+    ++iter_;
+  }
+
+  // Go to the previous range.
+  // The end of the previous range always matches the beginning of the current
+  // range. (But should always have a different value.)
+  // Not allowed if we're already at map_->begin().
+  void operator--() {
+    DCHECK(iter_ != map_->begin());
+    --iter_;
+  }
+
+ private:
+  const MapType* map_;
+
+  // Pointer to the entry in the IntervalMap that specifies the
+  // beginning of the current range.
+  typename MapType::const_iterator iter_;
+};
+
+template <typename KeyType,
+          typename ValueType,
+          class Compare = std::less<KeyType>,
+          class NumericLimits = std::numeric_limits<KeyType>>
+class IntervalMap {
+ public:
+  typedef std::map<KeyType, ValueType, Compare> MapType;
+  typedef IntervalMapConstIterator<KeyType, ValueType, Compare, NumericLimits>
+      const_iterator;
+  IntervalMap() {
+    // Adding an explicit entry for the default range is not strictly needed,
+    // but simplifies the code a lot.
+    map_[NumericLimits::min()] = ValueType();
+  }
+
+  // Returns the value at a particular point.
+  // Defaults to ValueType().
+  ValueType operator[](const KeyType& k) const {
+    typename MapType::const_iterator i = map_.upper_bound(k);
+    if (i == map_.begin()) {
+      return 0;

[xhwang, 2015/11/12 18:24:58]

Not possible to happen? how about a DCHECK? We actually did this on l.228.

[hubbe, 2015/11/12 19:03:12]

Done.

+    } else {
+      --i;
+      return i->second;
+    }
+  }
+
+  // Increase [from..to) by |how_much|.
+  void IncrementRange(KeyType from, KeyType to, ValueType how_much) {
+    DCHECK_GT(to, from);
+    if (how_much == 0)
+      return;
+    typename MapType::iterator a = MakeEntry(from);
+    typename MapType::iterator b = MakeEntry(to);
+    for (typename MapType::iterator i = a; i != b; ++i) {
+      i->second += how_much;
+    }
+    RemoveDuplicates(a);
+    // b may be invalid
+    RemoveDuplicates(map_.lower_bound(to));
+  }
+
+  // Set [from..to) to |how_much|.
+  void SetRange(KeyType from, KeyType to, ValueType how_much) {
+    DCHECK_GT(to, from);
+    typename MapType::iterator a = MakeEntry(from);
+    typename MapType::iterator b = MakeEntry(to);
+    a->second = how_much;
+    while (true) {
+      typename MapType::iterator c = a;
+      ++c;
+      if (c == b) {
+        break;
+      } else {
+        map_.erase(c);
+      }
+    }
+    RemoveDuplicates(a);
+    // b may be invalid
+    RemoveDuplicates(map_.lower_bound(to));
+  }
+
+  // Returns an iterator to the first range.
+  // Note, there is always at least one range.
+  const_iterator begin() const { return const_iterator(&map(), map_.begin()); }
+
+  // Returns an end marker iterator.
+  const_iterator end() const { return const_iterator(&map(), map_.end()); }
+
+  // Returns an iterator to the range containing |k|.
+  // Always returns a valid iterator.
+  const_iterator find(KeyType k) const {
+    typename MapType::const_iterator first = map_.upper_bound(k);

[xhwang, 2015/11/12 18:24:58]

now we don't have "first" and "second" any more... maybe just iter to match the
ctor of IntervalMapConstIterator?

[hubbe, 2015/11/12 19:03:12]

Done.

+    DCHECK(first != map_.begin());
+    --first;
+    return const_iterator(&map(), first);
+  }
+
+  bool empty() const { return map().size() == 1; }
+
+ private:
+  const MapType& map() const { return map_; }
+
+  // Make an entry in map_ with the key |k| and return it's iterator.
+  // If such an entry already exists, just re-use it.
+  // If a new entry is created, it's value will be set to the same
+  // as the preceeding entry, or ValueType() if no preceeding entry exists.
+  // After calling this function, we'll need to call RemoveDuplicates()
+  // to clean up any duplicates that we made.
+  typename MapType::iterator MakeEntry(KeyType k) {
+    typename MapType::value_type tmp(k, 0);
+    std::pair<typename MapType::iterator, bool> insert_result;
+    insert_result = map_.insert(tmp);
+    if (insert_result.second) {
+      if (insert_result.first != map_.begin()) {
+        typename MapType::iterator i = insert_result.first;
+        --i;
+        insert_result.first->second = i->second;
+      }
+    }
+    return insert_result.first;
+  }
+
+  // Remove duplicates before and after |i|.
+  void RemoveDuplicates(typename MapType::iterator i) {
+    DCHECK(i != map_.begin());
+    if (i == map_.end())
+      return;
+
+    typename MapType::iterator first = i;
+    typename MapType::iterator second = i;
+    --first;
+    if (first->second == second->second) {
+      map_.erase(second);
+      second = first;
+    } else {
+      first = second;
+    }
+    ++second;
+    if (second != map_.end() && first->second == second->second) {
+      map_.erase(second);
+    }
+  }
+
+  MapType map_;
+};
+
+}  // namespace media
+
+#endif  // MEDIA_BLINK_INTERVAL_MAP_H_