Object stats

src/snmalloc/backend_helpers/statsrange.h

@@ -17,8 +17,7 @@ namespace snmalloc
     {
       using ContainsParent<ParentRange>::parent;
 
-      static inline std::atomic<size_t> current_usage{};
-      static inline std::atomic<size_t> peak_usage{};
+      static inline Stat usage{};
 
     public:
       static constexpr bool Aligned = ParentRange::Aligned;
@@ -29,36 +28,28 @@ namespace snmalloc
 
       constexpr Type() = default;
 
-      CapPtr<void, ChunkBounds> alloc_range(size_t size)
+      capptr::Arena<void> alloc_range(size_t size)
       {
-        auto result = parent.alloc_range(size);
-        if (result != nullptr)
-        {
-          auto prev = current_usage.fetch_add(size);
-          auto curr = peak_usage.load();
-          while (curr < prev + size)
-          {
-            if (peak_usage.compare_exchange_weak(curr, prev + size))
-              break;
-          }
-        }
-        return result;
+        auto r = parent.alloc_range(size);
+        if (r != nullptr)
+          usage += size;
+        return r;
       }
 
-      void dealloc_range(CapPtr<void, ChunkBounds> base, size_t size)
+      void dealloc_range(capptr::Arena<void> base, size_t size)
       {
-        current_usage -= size;
+        usage -= size;
         parent.dealloc_range(base, size);
       }
 
       size_t get_current_usage()
       {
-        return current_usage.load();
+        return usage.get_curr();
       }
 
       size_t get_peak_usage()
       {
-        return peak_usage.load();
+        return usage.get_peak();
       }
     };
   };

src/snmalloc/ds_core/ds_core.h

@@ -15,3 +15,4 @@
 #include "ptrwrap.h"
 #include "redblacktree.h"
 #include "seqset.h"
+#include "stats.h"

src/snmalloc/ds_core/stats.h

@@ -0,0 +1,68 @@
+#include "defines.h"
+
+#include <atomic>
+#include <cstddef>
+
+namespace snmalloc
+{
+  /**
+   * Very basic statistic that tracks current and peak values.
+   */
+  class Stat
+  {
+  private:
+    std::atomic<size_t> curr{0};
+    std::atomic<size_t> peak{0};
+
+  public:
+    void increase(size_t amount)
+    {
+      size_t c = (curr += amount);
+      size_t p = peak.load(std::memory_order_relaxed);
+      while (c > p)
+      {
+        if (peak.compare_exchange_strong(p, c))
+          break;
+      }
+    }
+
+    void decrease(size_t amount)
+    {
+      size_t prev = curr.fetch_sub(amount);
+      // TODO Fix this to be true.
+      //      SNMALLOC_ASSERT_MSG(prev >= amount, "prev = {}, amount = {}",
+      //      prev, amount);
+      UNUSED(prev);
+    }
+
+    size_t get_curr()
+    {
+      return curr.load(std::memory_order_relaxed);
+    }
+
+    size_t get_peak()
+    {
+      return peak.load(std::memory_order_relaxed);
+    }
+
+    void operator+=(size_t amount)
+    {
+      increase(amount);
+    }
+
+    void operator-=(size_t amount)
+    {
+      decrease(amount);
+    }
+
+    void operator++()
+    {
+      increase(1);
+    }
+
+    void operator--()
+    {
+      decrease(1);
+    }
+  };
+} // namespace snmalloc

src/snmalloc/mem/allocstats.h

@@ -0,0 +1,65 @@
+#include "../ds_core/ds_core.h"
+#include "sizeclasstable.h"
+
+#include <array>
+
+namespace snmalloc
+{
+  class MonotoneStat
+  {
+    size_t value{0};
+
+  public:
+    void operator++(int)
+    {
+      value++;
+    }
+
+    void operator+=(const MonotoneStat& other)
+    {
+      value += other.value;
+    }
+
+    size_t operator*()
+    {
+      return value;
+    }
+  };
+
+  struct AllocStat
+  {
+    MonotoneStat objects_allocated{};
+    MonotoneStat objects_deallocated{};
+    MonotoneStat slabs_allocated{};
+    MonotoneStat slabs_deallocated{};
+  };
+
+  class AllocStats
+  {
+    std::array<AllocStat, SIZECLASS_REP_SIZE> sizeclass{};
+
+  public:
+    AllocStat& operator[](sizeclass_t index)
+    {
+      auto i = index.raw();
+      return sizeclass[i];
+    }
+
+    AllocStat& operator[](smallsizeclass_t index)
+    {
+      return sizeclass[sizeclass_t::from_small_class(index).raw()];
+    }
+
+    void operator+=(const AllocStats& other)
+    {
+      for (size_t i = 0; i < SIZECLASS_REP_SIZE; i++)
+      {
+        sizeclass[i].objects_allocated += other.sizeclass[i].objects_allocated;
+        sizeclass[i].objects_deallocated +=
+          other.sizeclass[i].objects_deallocated;
+        sizeclass[i].slabs_allocated += other.sizeclass[i].slabs_allocated;
+        sizeclass[i].slabs_deallocated += other.sizeclass[i].slabs_deallocated;
+      }
+    }
+  };
+} // namespace snmalloc

src/snmalloc/mem/corealloc.h

@@ -111,6 +111,11 @@ namespace snmalloc
      */
     Ticker<typename Config::Pal> ticker;
 
+    /**
+     * Tracks this allocators memory usage
+     */
+    AllocStats stats;
+
     /**
      * The message queue needs to be accessible from other threads
      *
@@ -364,6 +369,8 @@ namespace snmalloc
         // don't touch the cache lines at this point in snmalloc_check_client.
         auto start = clear_slab(meta, sizeclass);
 
+        stats[sizeclass].slabs_deallocated++;
+
         Config::Backend::dealloc_chunk(
           get_backend_local_state(),
           *meta,
@@ -400,6 +407,8 @@ namespace snmalloc
         // Remove from set of fully used slabs.
         meta->node.remove();
 
+        stats[entry.get_sizeclass()].slabs_deallocated++;
+
         Config::Backend::dealloc_chunk(
           get_backend_local_state(), *meta, p, size);
 
@@ -478,14 +487,18 @@ namespace snmalloc
                            SNMALLOC_FAST_PATH_LAMBDA {
                              return capptr_domesticate<Config>(local_state, p);
                            };
-      auto cb = [this,
-                 &need_post](freelist::HeadPtr msg) SNMALLOC_FAST_PATH_LAMBDA {
+
+      size_t received_bytes = 0;
+
+      auto cb = [this, &need_post, &received_bytes](
+                  freelist::HeadPtr msg) SNMALLOC_FAST_PATH_LAMBDA {
 #ifdef SNMALLOC_TRACING
         message<1024>("Handling remote");
 #endif
 
         auto& entry =
           Config::Backend::template get_metaentry(snmalloc::address_cast(msg));
+        received_bytes += sizeclass_full_to_size(entry.get_sizeclass());
 
         handle_dealloc_remote(entry, msg.as_void(), need_post);
 
@@ -514,6 +527,9 @@ namespace snmalloc
         post();
       }
 
+      // Push size to global statistics
+      RemoteDeallocCache::remote_inflight -= received_bytes;
+
       return action(args...);
     }
 
@@ -542,10 +558,7 @@ namespace snmalloc
       }
       else
       {
-        if (
-          !need_post &&
-          !attached_cache->remote_dealloc_cache.reserve_space(entry))
-          need_post = true;
+        need_post = attached_cache->remote_dealloc_cache.reserve_space(entry);
         attached_cache->remote_dealloc_cache
           .template dealloc<sizeof(CoreAllocator)>(
             entry.get_remote()->trunc_id(), p.as_void());
@@ -668,13 +681,14 @@ namespace snmalloc
       // pointers
       auto& entry =
         Config::Backend::template get_metaentry(snmalloc::address_cast(p));
-      if (SNMALLOC_LIKELY(dealloc_local_object_fast(entry, p, entropy)))
+      if (SNMALLOC_LIKELY(dealloc_local_object_fast<false>(entry, p, entropy)))
         return;
 
       dealloc_local_object_slow(p, entry);
     }
 
-    SNMALLOC_FAST_PATH static bool dealloc_local_object_fast(
+    template<bool Statistics = true>
+    SNMALLOC_FAST_PATH bool dealloc_local_object_fast(
       const PagemapEntry& entry,
       CapPtr<void, capptr::bounds::Alloc> p,
       LocalEntropy& entropy)
@@ -695,6 +709,10 @@ namespace snmalloc
       // Update the head and the next pointer in the free list.
       meta->free_queue.add(cp, key, entropy);
 
+      if constexpr (Statistics)
+      {
+        stats[entry.get_sizeclass()].objects_deallocated++;
+      }
       return SNMALLOC_LIKELY(!meta->return_object());
     }
 
@@ -741,6 +759,7 @@ namespace snmalloc
         }
 
         auto r = finish_alloc<zero_mem, Config>(p, sizeclass);
+        stats[sizeclass].objects_allocated++;
         return ticker.check_tick(r);
       }
       return small_alloc_slow<zero_mem>(sizeclass, fast_free_list);
@@ -813,6 +832,9 @@ namespace snmalloc
       }
 
       auto r = finish_alloc<zero_mem, Config>(p, sizeclass);
+
+      stats[sizeclass].objects_allocated++;
+      stats[sizeclass].slabs_allocated++;
       return ticker.check_tick(r);
     }
 
@@ -834,17 +856,19 @@ namespace snmalloc
       {
         auto p_wild = message_queue().destroy();
         auto p_tame = domesticate(p_wild);
-
+        size_t received_bytes = 0;
         while (p_tame != nullptr)
         {
           bool need_post = true; // Always going to post, so ignore.
           auto n_tame =
             p_tame->atomic_read_next(RemoteAllocator::key_global, domesticate);
           const PagemapEntry& entry =
             Config::Backend::get_metaentry(snmalloc::address_cast(p_tame));
+          received_bytes += sizeclass_full_to_size(entry.get_sizeclass());
           handle_dealloc_remote(entry, p_tame.as_void(), need_post);
           p_tame = n_tame;
         }
+        RemoteDeallocCache::remote_inflight -= received_bytes;
       }
       else
       {
@@ -986,6 +1010,11 @@ namespace snmalloc
 
       return debug_is_empty_impl(result);
     }
+
+    const AllocStats& get_stats()
+    {
+      return stats;
+    }
   };
 
   /**