Add alloc stats for more things

Add alloc stats to the base Box::operator new(), which is where non-builtin
types get allocated.

Also add stats for non-Python allocations (typically, conservative memory);
turns out that we are allocating a lot of conservative memory (both in terms
of allocations but also in terms of allocation size) for dicts.

microbenchmarks/dict_ubench.py

+def f():
+    for i in xrange(10000000):
+        {}
+f()

src/core/types.h

@@ -518,10 +518,12 @@ extern "C" PyObject* PystonType_GenericAlloc(BoxedClass* cls, Py_ssize_t nitems)
 
 #if STAT_ALLOCATIONS
 #define ALLOC_STATS(cls)                                                                                               \
-    std::string per_name_alloc_name = "alloc." + std::string(cls->tp_name);                                            \
-    std::string per_name_allocsize_name = "allocsize." + std::string(cls->tp_name);                                    \
-    Stats::log(Stats::getStatId(per_name_alloc_name));                                                                 \
-    Stats::log(Stats::getStatId(per_name_allocsize_name), size);
+    if (cls->tp_name) {                                                                                                \
+        std::string per_name_alloc_name = "alloc." + std::string(cls->tp_name);                                        \
+        std::string per_name_allocsize_name = "allocsize." + std::string(cls->tp_name);                                \
+        Stats::log(Stats::getStatId(per_name_alloc_name));                                                             \
+        Stats::log(Stats::getStatId(per_name_allocsize_name), size);                                                   \
+    }
 #define ALLOC_STATS_VAR(cls)                                                                                           \
     if (cls->tp_name) {                                                                                                \
         std::string per_name_alloc_name = "alloc." + std::string(cls->tp_name);                                        \
@@ -550,7 +552,6 @@ extern "C" PyObject* PystonType_GenericAlloc(BoxedClass* cls, Py_ssize_t nitems)
 // asserts in the 1-arg operator new function:
 #define DEFAULT_CLASS_SIMPLE(default_cls)                                                                              \
     void* operator new(size_t size, BoxedClass * cls) __attribute__((visibility("default"))) {                         \
-        ALLOC_STATS(cls);                                                                                              \
         return Box::operator new(size, cls);                                                                           \
     }                                                                                                                  \
     void* operator new(size_t size) __attribute__((visibility("default"))) {                                           \
@@ -601,7 +602,6 @@ extern "C" PyObject* PystonType_GenericAlloc(BoxedClass* cls, Py_ssize_t nitems)
     }                                                                                                                  \
                                                                                                                        \
     void* operator new(size_t size, BoxedClass * cls, size_t nitems) __attribute__((visibility("default"))) {          \
-        ALLOC_STATS_VAR(default_cls)                                                                                   \
         assert(cls->tp_itemsize == itemsize);                                                                          \
         return BoxVar::operator new(size, cls, nitems);                                                                \
     }                                                                                                                  \

src/gc/gc_alloc.h

@@ -28,7 +28,18 @@ namespace pyston {
 
 namespace gc {
 
-static StatCounter gc_alloc_bytes("zzz_gc_alloc_bytes");
+#if STAT_ALLOCATIONS
+static StatCounter gc_alloc_bytes("gc_alloc_bytes");
+static StatCounter gc_alloc_bytes_typed[] = {
+    StatCounter("gc_alloc_bytes_???"),          //
+    StatCounter("gc_alloc_bytes_python"),       //
+    StatCounter("gc_alloc_bytes_conservative"), //
+    StatCounter("gc_alloc_bytes_precise"),      //
+    StatCounter("gc_alloc_bytes_untracked"),    //
+    StatCounter("gc_alloc_bytes_hidden_class"), //
+};
+#endif
+
 extern "C" inline void* gc_alloc(size_t bytes, GCKind kind_id) {
     STAT_TIMER(t0, "us_timer_gc_alloc");
     size_t alloc_bytes = bytes + sizeof(GCAllocation);
@@ -94,6 +105,7 @@ extern "C" inline void* gc_alloc(size_t bytes, GCKind kind_id) {
 
 #if STAT_ALLOCATIONS
     gc_alloc_bytes.log(bytes);
+    gc_alloc_bytes_typed[(int)kind_id].log(bytes);
 #endif
 
     return r;

src/gc/heap.cpp

@@ -111,9 +111,12 @@ static unsigned bytesAllocatedSinceCollection;
 static __thread unsigned thread_bytesAllocatedSinceCollection;
 #define ALLOCBYTES_PER_COLLECTION 10000000
 
+static StatCounter gc_registered_bytes("gc_registered_bytes");
+
 void registerGCManagedBytes(size_t bytes) {
     thread_bytesAllocatedSinceCollection += bytes;
     if (unlikely(thread_bytesAllocatedSinceCollection > ALLOCBYTES_PER_COLLECTION / 4)) {
+        gc_registered_bytes.log(thread_bytesAllocatedSinceCollection);
         bytesAllocatedSinceCollection += thread_bytesAllocatedSinceCollection;
         thread_bytesAllocatedSinceCollection = 0;
 

src/runtime/types.cpp

@@ -180,6 +180,8 @@ extern "C" PyObject* _PyObject_New(PyTypeObject* tp) noexcept {
 
 // Analogue of PyType_GenericNew
 void* BoxVar::operator new(size_t size, BoxedClass* cls, size_t nitems) {
+    ALLOC_STATS_VAR(cls);
+
     assert(cls);
     ASSERT(cls->tp_basicsize >= size, "%s", cls->tp_name);
     assert(cls->tp_itemsize > 0);
@@ -191,6 +193,8 @@ void* BoxVar::operator new(size_t size, BoxedClass* cls, size_t nitems) {
 }
 
 void* Box::operator new(size_t size, BoxedClass* cls) {
+    ALLOC_STATS(cls);
+
     assert(cls);
     ASSERT(cls->tp_basicsize >= size, "%s", cls->tp_name);
     assert(cls->tp_itemsize == 0);