Re: [PATCH][Hashtable 6/6] PR 68303 small size optimization

["François Dumont" <frs.dumont@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 5735 bytes --]

On 17/07/20 2:58 pm, Jonathan Wakely wrote:
> On 17/11/19 22:31 +0100, François Dumont wrote:
>> This is an implementation of PR 68303.
>>
>> I try to use this idea as much as possible to avoid computation of 
>> hash codes.
>>
>> Note that tests are not showing any gain. I guess hash computation 
>> must be quite bad to get a benefit from it. So I am only activating 
>> it when hash code is not cached and/or when computation is not fast.
>
> If the tests don't show any benefit, why bother making the change?

I eventually managed to demonstrate this optimization through a 
performance test case.

>
> Does it help the example in the PR?

No, the code attached to the PR just show what the user has done to put 
in place this optim on his side.

What I needed was a slow hash code computation compared to the equal 
operation. I realized that I had to use longer string to achieve this.

Moreover making this optim dependant on _Hashtable_traits::__hash_cached 
was just wrong as we cannot use the cached hash code here as the input 
is a key instance, not a node.

I introduce _Hashtable_hash_traits<_Hash> to offer a single 
customization point as this optim depends highly on the difference 
between a hash code computation and a comparison. Maybe I should put it 
at std namespace scope to ease partial specialization ?

Performance test results before the patch:

unordered_small_size.cc      std::unordered_set<int>: 1st insert      
40r   32u    8s 264000112mem    0pf
unordered_small_size.cc      std::unordered_set<int>: find/erase      
22r   22u    0s -191999808mem    0pf
unordered_small_size.cc      std::unordered_set<int>: 2nd insert      
36r   36u    0s 191999776mem    0pf
unordered_small_size.cc      std::unordered_set<int>: erase key       
25r   25u    0s -191999808mem    0pf
unordered_small_size.cc      std::unordered_set<string>: 1st insert    
  404r  244u  156s -1989936256mem    0pf
unordered_small_size.cc      std::unordered_set<string>: find/erase    
  315r  315u    0s 2061942272mem    0pf
unordered_small_size.cc      std::unordered_set<string>: 2nd insert    
  233r  233u    0s -2061942208mem    0pf
unordered_small_size.cc      std::unordered_set<string>: erase key     
  299r  298u    0s 2061942208mem    0pf

after the patch:

unordered_small_size.cc      std::unordered_set<int>: 1st insert      
41r   33u    7s 264000112mem    0pf
unordered_small_size.cc      std::unordered_set<int>: find/erase      
24r   25u    1s -191999808mem    0pf
unordered_small_size.cc      std::unordered_set<int>: 2nd insert      
34r   34u    0s 191999776mem    0pf
unordered_small_size.cc      std::unordered_set<int>: erase key       
25r   25u    0s -191999808mem    0pf
unordered_small_size.cc      std::unordered_set<string>: 1st insert    
  399r  232u  165s -1989936256mem    0pf
unordered_small_size.cc      std::unordered_set<string>: find/erase    
  196r  197u    0s 2061942272mem    0pf
unordered_small_size.cc      std::unordered_set<string>: 2nd insert    
  221r  222u    0s -2061942208mem    0pf
unordered_small_size.cc      std::unordered_set<string>: erase key     
  178r  178u    0s 2061942208mem    0pf

     libstdc++: Optimize operations on small size hashtable [PR 68303]

     When hasher is identified as slow and the number of elements is 
limited in the
     container use a brute-force loop on those elements to look for a 
given key using
     the key_equal functor. For the moment the default threshold below 
which the
     container is considered as small is 20.

     libstdc++-v3/ChangeLog:

             PR libstdc++/68303
             * include/bits/hashtable_policy.h
             (_Hashtable_hash_traits<_Hash>): New.
             (_Hash_code_base<>::_M_hash_code(const 
_Hash_node_value<>&)): New.
             (_Hashtable_base<>::_M_key_equals): New.
             (_Hashtable_base<>::_M_equals): Use latter.
             (_Hashtable_base<>::_M_key_equals_tr): New.
             (_Hashtable_base<>::_M_equals_tr): Use latter.
             * include/bits/hashtable.h
             (_Hashtable<>::__small_size_threshold()): New, use 
_Hashtable_hash_traits.
             (_Hashtable<>::find): Loop through elements to look for key 
if size is lower
             than __small_size_threshold().
             (_Hashtable<>::_M_emplace(true_type, _Args&&...)): Likewise.
             (_Hashtable<>::_M_insert_unique(_Kt&&, _Args&&, const 
_NodeGenerator&)): Likewise.
(_Hashtable<>::_M_compute_hash_code(const_iterator, const key_type&)): New.
             (_Hashtable<>::_M_emplace(const_iterator, false_type, 
_Args&&...)): Use latter.
             (_Hashtable<>::_M_find_before_node(const key_type&)): New.
             (_Hashtable<>::_M_erase(true_type, const key_type&)): Use 
latter.
             (_Hashtable<>::_M_erase(false_type, const key_type&)): 
Likewise.
             * src/c++11/hashtable_c++0x.cc: Include 
<bits/functional_hash.h>.
             * testsuite/util/testsuite_performane.h
             (report_performance): Use 9 width to display memory.
             * 
testsuite/performance/23_containers/insert_erase/unordered_small_size.cc:
             New performance test case.

Tested under Linux x86_64.

Ok to commit ?

François


[-- Attachment #2: pr68303.patch --]
[-- Type: text/x-patch, Size: 20515 bytes --]

diff --git a/libstdc++-v3/include/bits/hashtable.h b/libstdc++-v3/include/bits/hashtable.h
index 92516b81ae5..fedb5a7ec3d 100644
--- a/libstdc++-v3/include/bits/hashtable.h
+++ b/libstdc++-v3/include/bits/hashtable.h
@@ -426,6 +426,13 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
       _M_uses_single_bucket() const
       { return _M_uses_single_bucket(_M_buckets); }
 
+      static constexpr size_t
+      __small_size_threshold()
+      {
+	return
+	  __detail::_Hashtable_hash_traits<_Hash>::__small_size_threshold();
+      }
+
       __hashtable_alloc&
       _M_base_alloc() { return *this; }
 
@@ -795,6 +802,9 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
       _M_bucket_index(__hash_code __c) const
       { return __hash_code_base::_M_bucket_index(__c, _M_bucket_count); }
 
+      __node_base_ptr
+      _M_find_before_node(const key_type&);
+
       // Find and insert helper functions and types
       // Find the node before the one matching the criteria.
       __node_base_ptr
@@ -838,6 +848,9 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
       __node_base_ptr
       _M_get_previous_node(size_type __bkt, __node_ptr __n);
 
+      pair<const_iterator, __hash_code>
+      _M_compute_hash_code(const_iterator __hint, const key_type& __k) const;
+
       // Insert node __n with hash code __code, in bucket __bkt if no
       // rehash (assumes no element with same key already present).
       // Takes ownership of __n if insertion succeeds, throws otherwise.
@@ -1124,7 +1137,6 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
       void _M_rehash(size_type __bkt_count, const __rehash_state& __state);
     };
 
-
   // Definitions of class template _Hashtable's out-of-line member functions.
   template<typename _Key, typename _Value, typename _Alloc,
 	   typename _ExtractKey, typename _Equal,
@@ -1617,6 +1629,14 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
     find(const key_type& __k)
     -> iterator
     {
+      if (size() <= __small_size_threshold())
+	{
+	  for (auto __it = begin(); __it != end(); ++__it)
+	    if (this->_M_key_equals(__k, *__it._M_cur))
+	      return __it;
+	  return end();
+	}
+
       __hash_code __code = this->_M_hash_code(__k);
       std::size_t __bkt = _M_bucket_index(__code);
       return iterator(_M_find_node(__bkt, __k, __code));
@@ -1632,6 +1652,14 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
     find(const key_type& __k) const
     -> const_iterator
     {
+      if (size() <= __small_size_threshold())
+	{
+	  for (auto __it = begin(); __it != end(); ++__it)
+	    if (this->_M_key_equals(__k, *__it._M_cur))
+	      return __it;
+	  return end();
+	}
+
       __hash_code __code = this->_M_hash_code(__k);
       std::size_t __bkt = _M_bucket_index(__code);
       return const_iterator(_M_find_node(__bkt, __k, __code));
@@ -1844,6 +1872,35 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
       }
 #endif
 
+  // Find the node before the one whose key compares equal to k.
+  // Return nullptr if no node is found.
+  template<typename _Key, typename _Value, typename _Alloc,
+	   typename _ExtractKey, typename _Equal,
+	   typename _Hash, typename _RangeHash, typename _Unused,
+	   typename _RehashPolicy, typename _Traits>
+    auto
+    _Hashtable<_Key, _Value, _Alloc, _ExtractKey, _Equal,
+	       _Hash, _RangeHash, _Unused, _RehashPolicy, _Traits>::
+    _M_find_before_node(const key_type& __k)
+    -> __node_base_ptr
+    {
+      __node_base_ptr __prev_p = &_M_before_begin;
+      if (!__prev_p->_M_nxt)
+	return nullptr;
+
+      for (__node_ptr __p = static_cast<__node_ptr>(__prev_p->_M_nxt);
+	   __p != nullptr;
+	   __p = __p->_M_next())
+	{
+	  if (this->_M_key_equals(__k, *__p))
+	    return __prev_p;
+
+	  __prev_p = __p;
+	}
+
+      return nullptr;
+    }
+
   // Find the node before the one whose key compares equal to k in the bucket
   // bkt. Return nullptr if no node is found.
   template<typename _Key, typename _Value, typename _Alloc,
@@ -1992,11 +2049,20 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
 	// First build the node to get access to the hash code
 	_Scoped_node __node { this, std::forward<_Args>(__args)...  };
 	const key_type& __k = _ExtractKey{}(__node._M_node->_M_v());
+	if (size() <= __small_size_threshold())
+	  {
+	    for (auto __it = begin(); __it != end(); ++__it)
+	      if (this->_M_key_equals(__k, *__it._M_cur))
+		// There is already an equivalent node, no insertion
+		return { __it, false };
+	  }
+
 	__hash_code __code = this->_M_hash_code(__k);
 	size_type __bkt = _M_bucket_index(__code);
+	if (size() > __small_size_threshold())
 	  if (__node_ptr __p = _M_find_node(__bkt, __k, __code))
 	    // There is already an equivalent node, no insertion
-	  return std::make_pair(iterator(__p), false);
+	    return { iterator(__p), false };
 
 	// Insert the node
 	auto __pos = _M_insert_unique_node(__bkt, __code, __node._M_node);
@@ -2020,13 +2086,41 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
 	_Scoped_node __node { this, std::forward<_Args>(__args)...  };
 	const key_type& __k = _ExtractKey{}(__node._M_node->_M_v());
 
-	__hash_code __code = this->_M_hash_code(__k);
+	auto __res = this->_M_compute_hash_code(__hint, __k);
 	auto __pos
-	  = _M_insert_multi_node(__hint._M_cur, __code, __node._M_node);
+	  = _M_insert_multi_node(__res.first._M_cur, __res.second,
+				 __node._M_node);
 	__node._M_node = nullptr;
 	return __pos;
       }
 
+  template<typename _Key, typename _Value, typename _Alloc,
+	   typename _ExtractKey, typename _Equal,
+	   typename _Hash, typename _RangeHash, typename _Unused,
+	   typename _RehashPolicy, typename _Traits>
+    auto
+    _Hashtable<_Key, _Value, _Alloc, _ExtractKey, _Equal,
+	       _Hash, _RangeHash, _Unused, _RehashPolicy, _Traits>::
+    _M_compute_hash_code(const_iterator __hint, const key_type& __k) const
+    -> pair<const_iterator, __hash_code>
+    {
+      if (size() <= __small_size_threshold())
+	{
+	  if (__hint != cend())
+	    {
+	      for (auto __it = __hint; __it != cend(); ++__it)
+		if (this->_M_key_equals(__k, *__it._M_cur))
+		  return { __it, this->_M_hash_code(*__it._M_cur) };
+	    }
+
+	  for (auto __it = cbegin(); __it != __hint; ++__it)
+	    if (this->_M_key_equals(__k, *__it._M_cur))
+	      return { __it, this->_M_hash_code(*__it._M_cur) };
+	}
+
+      return { __hint, this->_M_hash_code(__k) };
+    }
+
   template<typename _Key, typename _Value, typename _Alloc,
 	   typename _ExtractKey, typename _Equal,
 	   typename _Hash, typename _RangeHash, typename _Unused,
@@ -2125,9 +2219,15 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
 		       const _NodeGenerator& __node_gen)
       -> pair<iterator, bool>
       {
+	if (size() <= __small_size_threshold())
+	  for (auto __it = begin(); __it != end(); ++__it)
+	    if (this->_M_key_equals_tr(__k, *__it._M_cur))
+	      return { __it, false };
+
 	__hash_code __code = this->_M_hash_code_tr(__k);
 	size_type __bkt = _M_bucket_index(__code);
 
+	if (size() > __small_size_threshold())
 	  if (__node_ptr __node = _M_find_node_tr(__bkt, __k, __code))
 	    return { iterator(__node), false };
 
@@ -2161,11 +2261,12 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
 	_Scoped_node __node{ __node_gen(std::forward<_Arg>(__v)), this };
 
 	// Second compute the hash code so that we don't rehash if it throws.
-	__hash_code __code
-	  = this->_M_hash_code(_ExtractKey{}(__node._M_node->_M_v()));
+	auto __res = this->_M_compute_hash_code(
+	  __hint, _ExtractKey{}(__node._M_node->_M_v()));
 
 	auto __pos
-	  = _M_insert_multi_node(__hint._M_cur, __code, __node._M_node);
+	  = _M_insert_multi_node(__res.first._M_cur, __res.second,
+				 __node._M_node);
 	__node._M_node = nullptr;
 	return __pos;
       }
@@ -2227,17 +2328,34 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
 	       _Hash, _RangeHash, _Unused, _RehashPolicy, _Traits>::
     _M_erase(true_type /* __uks */, const key_type& __k)
     -> size_type
+    {
+      __node_base_ptr __prev_n;
+      __node_ptr __n;
+      std::size_t __bkt;
+      if (size() <= __small_size_threshold())
+	{
+	  __prev_n = _M_find_before_node(__k);
+	  if (!__prev_n)
+	    return 0;
+
+	  // We found a matching node, erase it.
+	  __n = static_cast<__node_ptr>(__prev_n->_M_nxt);
+	  __bkt = _M_bucket_index(*__n);
+	}
+      else
 	{
 	  __hash_code __code = this->_M_hash_code(__k);
-      std::size_t __bkt = _M_bucket_index(__code);
+	  __bkt = _M_bucket_index(__code);
 
 	  // Look for the node before the first matching node.
-      __node_base_ptr __prev_n = _M_find_before_node(__bkt, __k, __code);
+	  __prev_n = _M_find_before_node(__bkt, __k, __code);
 	  if (!__prev_n)
 	    return 0;
 
 	  // We found a matching node, erase it.
-      __node_ptr __n = static_cast<__node_ptr>(__prev_n->_M_nxt);
+	  __n = static_cast<__node_ptr>(__prev_n->_M_nxt);
+	}
+
       _M_erase(__bkt, __prev_n, __n);
       return 1;
     }
@@ -2251,22 +2369,39 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
 	       _Hash, _RangeHash, _Unused, _RehashPolicy, _Traits>::
     _M_erase(false_type /* __uks */, const key_type& __k)
     -> size_type
+    {
+      std::size_t __bkt;
+      __node_base_ptr __prev_n;
+      __node_ptr __n;
+      if (size() <= __small_size_threshold())
+	{
+	  __prev_n = _M_find_before_node(__k);
+	  if (!__prev_n)
+	    return 0;
+
+	  // We found a matching node, erase it.
+	  __n = static_cast<__node_ptr>(__prev_n->_M_nxt);
+	  __bkt = _M_bucket_index(*__n);
+	}
+      else
 	{
 	  __hash_code __code = this->_M_hash_code(__k);
-      std::size_t __bkt = _M_bucket_index(__code);
+	  __bkt = _M_bucket_index(__code);
 
 	  // Look for the node before the first matching node.
-      __node_base_ptr __prev_n = _M_find_before_node(__bkt, __k, __code);
+	  __prev_n = _M_find_before_node(__bkt, __k, __code);
 	  if (!__prev_n)
 	    return 0;
 
+	  __n = static_cast<__node_ptr>(__prev_n->_M_nxt);
+	}
+
       // _GLIBCXX_RESOLVE_LIB_DEFECTS
       // 526. Is it undefined if a function in the standard changes
       // in parameters?
       // We use one loop to find all matching nodes and another to deallocate
       // them so that the key stays valid during the first loop. It might be
       // invalidated indirectly when destroying nodes.
-      __node_ptr __n = static_cast<__node_ptr>(__prev_n->_M_nxt);
       __node_ptr __n_last = __n->_M_next();
       while (__n_last && this->_M_node_equals(*__n, *__n_last))
 	__n_last = __n_last->_M_next();
diff --git a/libstdc++-v3/include/bits/hashtable_policy.h b/libstdc++-v3/include/bits/hashtable_policy.h
index 2130c958262..93d1962c74f 100644
--- a/libstdc++-v3/include/bits/hashtable_policy.h
+++ b/libstdc++-v3/include/bits/hashtable_policy.h
@@ -59,19 +59,19 @@ namespace __detail
 
   // Helper function: return distance(first, last) for forward
   // iterators, or 0/1 for input iterators.
-  template<class _Iterator>
+  template<typename _Iterator>
     inline typename std::iterator_traits<_Iterator>::difference_type
     __distance_fw(_Iterator __first, _Iterator __last,
 		  std::input_iterator_tag)
     { return __first != __last ? 1 : 0; }
 
-  template<class _Iterator>
+  template<typename _Iterator>
     inline typename std::iterator_traits<_Iterator>::difference_type
     __distance_fw(_Iterator __first, _Iterator __last,
 		  std::forward_iterator_tag)
     { return std::distance(__first, __last); }
 
-  template<class _Iterator>
+  template<typename _Iterator>
     inline typename std::iterator_traits<_Iterator>::difference_type
     __distance_fw(_Iterator __first, _Iterator __last)
     { return __distance_fw(__first, __last,
@@ -242,6 +242,20 @@ namespace __detail
       using __unique_keys = __bool_constant<_Unique_keys>;
     };
 
+  /**
+   *  struct _Hashtable_hash_traits
+   *
+   *  Important traits for hash tables depending on associated hasher.
+   *
+   */
+  template<typename _Hash>
+    struct _Hashtable_hash_traits
+    {
+	static constexpr std::size_t
+	__small_size_threshold()
+	{ return std::__is_fast_hash<_Hash>::value ? 0 : 20; }
+    };
+
   /**
    *  struct _Hash_node_base
    *
@@ -1121,10 +1135,12 @@ namespace __detail
 			_Hash, _RangeHash, _Unused, _RehashPolicy, _Traits,
 			true_type /* Has load factor */>
     {
+    private:
       using __hashtable = _Hashtable<_Key, _Value, _Alloc, _ExtractKey,
 				     _Equal, _Hash, _RangeHash, _Unused,
 				     _RehashPolicy, _Traits>;
 
+    public:
       float
       max_load_factor() const noexcept
       {
@@ -1266,6 +1282,14 @@ namespace __detail
 	  return _M_hash()(__k);
 	}
 
+      __hash_code
+      _M_hash_code(const _Hash_node_value<_Value, false>& __n) const
+      { return _M_hash_code(_ExtractKey{}(__n._M_v())); }
+
+      __hash_code
+      _M_hash_code(const _Hash_node_value<_Value, true>& __n) const
+      { return __n._M_hash_code; }
+
       std::size_t
       _M_bucket_index(__hash_code __c, std::size_t __bkt_count) const
       { return _RangeHash{}(__c, __bkt_count); }
@@ -1276,17 +1300,14 @@ namespace __detail
 	noexcept( noexcept(declval<const _Hash&>()(declval<const _Key&>()))
 		  && noexcept(declval<const _RangeHash&>()((__hash_code)0,
 							   (std::size_t)0)) )
-      {
-	return _RangeHash{}(_M_hash_code(_ExtractKey{}(__n._M_v())),
-			    __bkt_count);
-      }
+      { return _M_bucket_index(_M_hash_code(__n), __bkt_count); }
 
       std::size_t
       _M_bucket_index(const _Hash_node_value<_Value, true>& __n,
 		      std::size_t __bkt_count) const
 	noexcept( noexcept(declval<const _RangeHash&>()((__hash_code)0,
 							(std::size_t)0)) )
-      { return _RangeHash{}(__n._M_hash_code, __bkt_count); }
+      { return _M_bucket_index(__n._M_hash_code, __bkt_count); }
 
       void
       _M_store_code(_Hash_node_code_cache<false>&, __hash_code) const
@@ -1646,18 +1667,20 @@ namespace __detail
       { }
 
       bool
-      _M_equals(const _Key& __k, __hash_code __c,
-		const _Hash_node_value<_Value, __hash_cached::value>& __n) const
+      _M_key_equals(const _Key& __k,
+		    const _Hash_node_value<_Value,
+					   __hash_cached::value>& __n) const
 	{
-	static_assert(__is_invocable<const _Equal&, const _Key&, const _Key&>{},
+	  static_assert(
+	    __is_invocable<const _Equal&, const _Key&, const _Key&>{},
 	    "key equality predicate must be invocable with two arguments of "
 	    "key type");
-	return _S_equals(__c, __n) && _M_eq()(__k, _ExtractKey{}(__n._M_v()));
+	  return _M_eq()(__k, _ExtractKey{}(__n._M_v()));
 	}
 
       template<typename _Kt>
 	bool
-	_M_equals_tr(const _Kt& __k, __hash_code __c,
+	_M_key_equals_tr(const _Kt& __k,
 			 const _Hash_node_value<_Value,
 					     __hash_cached::value>& __n) const
 	{
@@ -1665,16 +1688,28 @@ namespace __detail
 	    __is_invocable<const _Equal&, const _Kt&, const _Key&>{},
 	    "key equality predicate must be invocable with two arguments of "
 	    "key type");
-	  return _S_equals(__c, __n) && _M_eq()(__k, _ExtractKey{}(__n._M_v()));
+	  return _M_eq()(__k, _ExtractKey{}(__n._M_v()));
 	}
 
+      bool
+      _M_equals(const _Key& __k, __hash_code __c,
+		const _Hash_node_value<_Value, __hash_cached::value>& __n) const
+      { return _S_equals(__c, __n) && _M_key_equals(__k, __n); }
+
+      template<typename _Kt>
+	bool
+	_M_equals_tr(const _Kt& __k, __hash_code __c,
+		     const _Hash_node_value<_Value,
+					    __hash_cached::value>& __n) const
+	{ return _S_equals(__c, __n) && _M_key_equals_tr(__k, __n); }
+
       bool
       _M_node_equals(
 	const _Hash_node_value<_Value, __hash_cached::value>& __lhn,
 	const _Hash_node_value<_Value, __hash_cached::value>& __rhn) const
       {
 	return _S_node_equals(__lhn, __rhn)
-	  && _M_eq()(_ExtractKey{}(__lhn._M_v()), _ExtractKey{}(__rhn._M_v()));
+	  && _M_key_equals(_ExtractKey{}(__lhn._M_v()), __rhn);
       }
 
       void
diff --git a/libstdc++-v3/src/c++11/hashtable_c++0x.cc b/libstdc++-v3/src/c++11/hashtable_c++0x.cc
index b9dec61b2ea..116192dfcf1 100644
--- a/libstdc++-v3/src/c++11/hashtable_c++0x.cc
+++ b/libstdc++-v3/src/c++11/hashtable_c++0x.cc
@@ -30,6 +30,7 @@
 #include <tuple>
 #include <ext/aligned_buffer.h>
 #include <ext/alloc_traits.h>
+#include <bits/functional_hash.h>
 #include <bits/hashtable_policy.h>
 
 namespace std _GLIBCXX_VISIBILITY(default)
diff --git a/libstdc++-v3/testsuite/performance/23_containers/insert_erase/unordered_small_size.cc b/libstdc++-v3/testsuite/performance/23_containers/insert_erase/unordered_small_size.cc
new file mode 100644
index 00000000000..ae63c15b5da
--- /dev/null
+++ b/libstdc++-v3/testsuite/performance/23_containers/insert_erase/unordered_small_size.cc
@@ -0,0 +1,125 @@
+// { dg-do run { target c++11 } }
+
+// Copyright (C) 2021 Free Software Foundation, Inc.
+//
+// This file is part of the GNU ISO C++ Library.  This library is free
+// software; you can redistribute it and/or modify it under the
+// terms of the GNU General Public License as published by the
+// Free Software Foundation; either version 3, or (at your option)
+// any later version.
+
+// This library is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License for more details.
+
+// You should have received a copy of the GNU General Public License along
+// with this library; see the file COPYING3.  If not see
+// <http://www.gnu.org/licenses/>.
+
+#include <string>
+#include <sstream>
+#include <vector>
+#include <unordered_set>
+#include <testsuite_performance.h>
+
+namespace
+{
+  const int nb_elements = 20;
+  const int nb_insts = 150000;
+
+  template<typename _ElemType>
+    void bench(const char* desc, const std::vector<_ElemType>& elems)
+    {
+      using namespace __gnu_test;
+
+      time_counter time;
+      resource_counter resource;
+
+      std::vector<std::unordered_set<_ElemType>> insts(nb_insts);
+      for (int j = 0; j != nb_insts; ++j)
+	insts.emplace_back();
+
+      start_counters(time, resource);
+
+      for (auto& us : insts)
+	for (int i = 0; i != nb_elements; ++i)
+	  us.insert(elems[i]);
+
+      stop_counters(time, resource);
+
+      std::ostringstream ostr;
+      ostr << desc << " 1st insert";
+      report_performance(__FILE__, ostr.str().c_str(), time, resource);
+
+      start_counters(time, resource);
+
+      for (auto& us : insts)
+	for (int i = nb_elements - 1; i >= 0; --i)
+	  {
+	    auto it = us.find(elems[i]);
+	    if (it != us.end())
+	      us.erase(it);
+	  }
+
+      stop_counters(time, resource);
+
+      ostr.str("");
+      ostr << desc << " find/erase";
+      report_performance(__FILE__, ostr.str().c_str(), time, resource);
+
+      start_counters(time, resource);
+
+      for (auto& us : insts)
+	{
+	  us.insert(elems[0]);
+	  for (int i = nb_elements - 1; i >= 0; --i)
+	    us.insert(elems[i]);
+	}
+
+      stop_counters(time, resource);
+      ostr.str("");
+      ostr << desc << " 2nd insert";
+      report_performance(__FILE__, ostr.str().c_str(), time, resource);
+
+      start_counters(time, resource);
+
+      for (auto& us : insts)
+	for (int j = nb_elements - 1; j >= 0; --j)
+	  us.erase(elems[j]);
+
+      stop_counters(time, resource);
+      ostr.str("");
+      ostr << desc << " erase key ";
+      report_performance(__FILE__, ostr.str().c_str(), time, resource);
+    }
+}
+
+int main()
+{
+  {
+    std::vector<int> elems;
+    elems.reserve(nb_elements);
+    for (int i = 0; i != nb_elements; ++i)
+      elems.push_back(i);
+
+    bench("std::unordered_set<int>:    ", elems);
+  }
+
+  {
+    std::vector<std::string> elems;
+    {
+      elems.reserve(nb_elements);
+      for (int i = 0; i != nb_elements; ++i)
+	{
+	  std::ostringstream ostr;
+	  ostr << "string #" << i << ' ' << std::string(1000, 'a' + i);
+	  elems.push_back(ostr.str());
+	}
+    }
+
+    bench("std::unordered_set<string>: ", elems);
+  }
+
+  return 0;
+}
diff --git a/libstdc++-v3/testsuite/util/testsuite_performance.h b/libstdc++-v3/testsuite/util/testsuite_performance.h
index cba3a0d4b17..4ca15ab0e71 100644
--- a/libstdc++-v3/testsuite/util/testsuite_performance.h
+++ b/libstdc++-v3/testsuite/util/testsuite_performance.h
@@ -239,7 +239,7 @@ namespace __gnu_test
     out << std::setw(4) << t.real_time() << "r" << space;
     out << std::setw(4) << t.user_time() << "u" << space;
     out << std::setw(4) << t.system_time() << "s" << space;
-    out << std::setw(8) << r.allocated_memory() << "mem" << space;
+    out << std::setw(9) << r.allocated_memory() << "mem" << space;
     out << std::setw(4) << r.hard_page_fault() << "pf" << space;
 
     out << std::endl;