Added simple stats tracking memory usage (#241)

These statistics can be maintained with effectively zero cost to
realistic applications.  They do not track the precise amount of
memory used, but are an over-approximation.

Author: mjp41

src/mem/largealloc.h

@@ -70,7 +70,17 @@ namespace snmalloc
      */
     AddressSpaceManager<PAL> address_space = {};
 
+    /**
+     * High-water mark of used memory.
+     */
+    std::atomic<size_t> peak_memory_used_bytes{0};
+
   public:
+    /**
+     * Memory current available in large_stacks
+     */
+    std::atomic<size_t> available_large_chunks_in_bytes{0};
+
     /**
      * Stack of large allocations that have been returned for reuse.
      */
@@ -184,20 +194,34 @@ namespace snmalloc
     {
       // Cache line align
       size_t size = bits::align_up(sizeof(T), 64);
-      size = bits::max(size, alignment);
-      void* p = address_space.template reserve<true>(bits::next_pow2(size));
+      size = bits::next_pow2(bits::max(size, alignment));
+      void* p = address_space.template reserve<true>(size);
       if (p == nullptr)
         return nullptr;
+
+      peak_memory_used_bytes += size;
+
       return new (p) T(std::forward<Args...>(args)...);
     }
 
     template<bool committed>
     void* reserve(size_t large_class) noexcept
     {
       size_t size = bits::one_at_bit(SUPERSLAB_BITS) << large_class;
-
+      peak_memory_used_bytes += size;
       return address_space.template reserve<committed>(size);
     }
+
+    /**
+     * Returns a pair of current memory usage and peak memory usage.
+     * Both statistics are very coarse-grained.
+     */
+    std::pair<size_t, size_t> memory_usage()
+    {
+      size_t avail = available_large_chunks_in_bytes;
+      size_t peak = peak_memory_used_bytes;
+      return {peak - avail, peak};
+    }
   };
 
   using Stats = AllocStats<NUM_SIZECLASSES, NUM_LARGE_CLASSES>;
@@ -239,6 +263,7 @@ namespace snmalloc
       else
       {
         stats.superslab_pop();
+        memory_provider.available_large_chunks_in_bytes -= rsize;
 
         // Cross-reference alloc.h's large_dealloc decommitment condition.
         bool decommitted =
@@ -284,18 +309,19 @@ namespace snmalloc
           "without low memory notifications");
       }
 
+      size_t rsize = bits::one_at_bit(SUPERSLAB_BITS) << large_class;
+
       // Cross-reference largealloc's alloc() decommitted condition.
       if (
         (decommit_strategy != DecommitNone) &&
         (large_class != 0 || decommit_strategy == DecommitSuper))
       {
-        size_t rsize = bits::one_at_bit(SUPERSLAB_BITS) << large_class;
-
         memory_provider.notify_not_using(
           pointer_offset(p, OS_PAGE_SIZE), rsize - OS_PAGE_SIZE);
       }
 
       stats.superslab_push();
+      memory_provider.available_large_chunks_in_bytes += rsize;
       memory_provider.large_stack[large_class].push(static_cast<Largeslab*>(p));
     }
   };

src/override/malloc-extensions.cc

@@ -0,0 +1,12 @@
+#include "malloc-extensions.h"
+
+#include "../snmalloc.h"
+
+using namespace snmalloc;
+
+void get_malloc_info_v1(malloc_info_v1* stats)
+{
+  auto next_memory_usage = default_memory_provider().memory_usage();
+  stats->current_memory_usage = next_memory_usage.first;
+  stats->peak_memory_usage = next_memory_usage.second;
+}

src/override/malloc-extensions.h

@@ -0,0 +1,34 @@
+/**
+ * Malloc extensions
+ *
+ * This file contains additional non-standard API surface for snmalloc.
+ * The API is subject to changes, but will be clearly noted in release
+ * notes.
+ */
+
+/**
+ * Structure for returning memory used by snmalloc.
+ *
+ * The statistics are very coarse grained as they only track
+ * usage at the superslab/chunk level. Meta-data and object
+ * data is not tracked independantly.
+ */
+struct malloc_info_v1
+{
+  /**
+   * Current memory usage of the allocator. Extremely coarse
+   * grained for efficient calculation.
+   */
+  size_t current_memory_usage;
+
+  /**
+   * High-water mark of current_memory_usage.
+   */
+  size_t peak_memory_usage;
+};
+
+/**
+ * Populates a malloc_info_v1 structure for the latest values
+ * from snmalloc.
+ */
+void get_malloc_info_v1(malloc_info_v1* stats);

src/test/func/memory_usage/memory_usage.cc

@@ -0,0 +1,103 @@
+/**
+ * Memory usage test
+ * Query memory usage repeatedly
+ */
+
+#include <iostream>
+#include <test/setup.h>
+#include <vector>
+
+#define SNMALLOC_NAME_MANGLE(a) our_##a
+#include "../../../override/malloc-extensions.cc"
+#include "../../../override/malloc.cc"
+
+using namespace snmalloc;
+
+bool print_memory_usage()
+{
+  static malloc_info_v1 last_memory_usage;
+  malloc_info_v1 next_memory_usage;
+
+  get_malloc_info_v1(&next_memory_usage);
+
+  if (
+    (next_memory_usage.current_memory_usage !=
+     last_memory_usage.current_memory_usage) ||
+    (next_memory_usage.peak_memory_usage !=
+     last_memory_usage.peak_memory_usage))
+  {
+    std::cout << "Memory Usages Changed to ("
+              << next_memory_usage.current_memory_usage << ", "
+              << next_memory_usage.peak_memory_usage << ")" << std::endl;
+    last_memory_usage = next_memory_usage;
+    return true;
+  }
+  return false;
+}
+
+std::vector<void*> allocs;
+
+/**
+ * Add allocs until the statistics have changed n times.
+ */
+void add_n_allocs(size_t n)
+{
+  while (true)
+  {
+    allocs.push_back(our_malloc(1024));
+    if (print_memory_usage())
+    {
+      n--;
+      if (n == 0)
+        break;
+    }
+  }
+}
+
+/**
+ * Remove allocs until the statistics have changed n times.
+ */
+void remove_n_allocs(size_t n)
+{
+  while (true)
+  {
+    our_free(allocs.back());
+    allocs.pop_back();
+    if (print_memory_usage())
+    {
+      n--;
+      if (n == 0)
+        break;
+    }
+  }
+}
+
+int main(int argc, char** argv)
+{
+  UNUSED(argc);
+  UNUSED(argv);
+
+  setup();
+
+  add_n_allocs(5);
+  std::cout << "Init complete!" << std::endl;
+
+  for (int i = 0; i < 10; i++)
+  {
+    remove_n_allocs(1);
+    std::cout << "Phase " << i << " remove complete!" << std::endl;
+    add_n_allocs(2);
+    std::cout << "Phase " << i << " add complete!" << std::endl;
+  }
+
+  for (int i = 0; i < 10; i++)
+  {
+    remove_n_allocs(2);
+    std::cout << "Phase " << i << " remove complete!" << std::endl;
+    add_n_allocs(1);
+    std::cout << "Phase " << i << " add complete!" << std::endl;
+  }
+
+  remove_n_allocs(3);
+  std::cout << "Teardown complete!" << std::endl;
+}