Add more stats for freelist use and allocations. (GH-92211)

Author: markshannon

Diff for: Include/internal/pycore_code.h

@@ -292,7 +292,12 @@ typedef struct _call_stats {
 
 typedef struct _object_stats {
     uint64_t allocations;
+    uint64_t allocations512;
+    uint64_t allocations4k;
+    uint64_t allocations_big;
     uint64_t frees;
+    uint64_t to_freelist;
+    uint64_t from_freelist;
     uint64_t new_values;
     uint64_t dict_materialized_on_request;
     uint64_t dict_materialized_new_key;
@@ -313,6 +318,8 @@ extern PyStats _py_stats;
 #define OPCODE_EXE_INC(opname) _py_stats.opcode_stats[opname].execution_count++
 #define CALL_STAT_INC(name) _py_stats.call_stats.name++
 #define OBJECT_STAT_INC(name) _py_stats.object_stats.name++
+#define OBJECT_STAT_INC_COND(name, cond) \
+    do { if (cond) _py_stats.object_stats.name++; } while (0)
 
 extern void _Py_PrintSpecializationStats(int to_file);
 
@@ -325,6 +332,7 @@ PyAPI_FUNC(PyObject*) _Py_GetSpecializationStats(void);
 #define OPCODE_EXE_INC(opname) ((void)0)
 #define CALL_STAT_INC(name) ((void)0)
 #define OBJECT_STAT_INC(name) ((void)0)
+#define OBJECT_STAT_INC_COND(name, cond) ((void)0)
 #endif  // !Py_STATS
 
 // Cache values are only valid in memory, so use native endianness.

Diff for: Objects/dictobject.c

@@ -624,6 +624,7 @@ new_keys_object(uint8_t log2_size, bool unicode)
 #endif
     if (log2_size == PyDict_LOG_MINSIZE && unicode && state->keys_numfree > 0) {
         dk = state->keys_free_list[--state->keys_numfree];
+        OBJECT_STAT_INC(from_freelist);
     }
     else
 #endif
@@ -681,6 +682,7 @@ free_keys_object(PyDictKeysObject *keys)
             && state->keys_numfree < PyDict_MAXFREELIST
             && DK_IS_UNICODE(keys)) {
         state->keys_free_list[state->keys_numfree++] = keys;
+        OBJECT_STAT_INC(to_freelist);
         return;
     }
 #endif
@@ -726,6 +728,7 @@ new_dict(PyDictKeysObject *keys, PyDictValues *values, Py_ssize_t used, int free
         mp = state->free_list[--state->numfree];
         assert (mp != NULL);
         assert (Py_IS_TYPE(mp, &PyDict_Type));
+        OBJECT_STAT_INC(from_freelist);
         _Py_NewReference((PyObject *)mp);
     }
     else
@@ -1544,6 +1547,7 @@ dictresize(PyDictObject *mp, uint8_t log2_newsize, int unicode)
                     state->keys_numfree < PyDict_MAXFREELIST)
             {
                 state->keys_free_list[state->keys_numfree++] = oldkeys;
+                OBJECT_STAT_INC(to_freelist);
             }
             else
 #endif
@@ -2381,6 +2385,7 @@ dict_dealloc(PyDictObject *mp)
 #endif
     if (state->numfree < PyDict_MAXFREELIST && Py_IS_TYPE(mp, &PyDict_Type)) {
         state->free_list[state->numfree++] = mp;
+        OBJECT_STAT_INC(to_freelist);
     }
     else
 #endif

Diff for: Objects/floatobject.c

@@ -141,6 +141,7 @@ PyFloat_FromDouble(double fval)
 #endif
         state->free_list = (PyFloatObject *) Py_TYPE(op);
         state->numfree--;
+        OBJECT_STAT_INC(from_freelist);
     }
     else
 #endif
@@ -256,6 +257,7 @@ _PyFloat_ExactDealloc(PyObject *obj)
     state->numfree++;
     Py_SET_TYPE(op, (PyTypeObject *)state->free_list);
     state->free_list = op;
+    OBJECT_STAT_INC(to_freelist);
 #else
     PyObject_Free(op);
 #endif

Diff for: Objects/genobject.c

@@ -1942,6 +1942,7 @@ async_gen_wrapped_val_dealloc(_PyAsyncGenWrappedValue *o)
     if (state->value_numfree < _PyAsyncGen_MAXFREELIST) {
         assert(_PyAsyncGenWrappedValue_CheckExact(o));
         state->value_freelist[state->value_numfree++] = o;
+        OBJECT_STAT_INC(to_freelist);
     }
     else
 #endif
@@ -2018,6 +2019,7 @@ _PyAsyncGenValueWrapperNew(PyObject *val)
     if (state->value_numfree) {
         state->value_numfree--;
         o = state->value_freelist[state->value_numfree];
+        OBJECT_STAT_INC(from_freelist);
         assert(_PyAsyncGenWrappedValue_CheckExact(o));
         _Py_NewReference((PyObject*)o);
     }

Diff for: Objects/listobject.c

@@ -158,6 +158,7 @@ PyList_New(Py_ssize_t size)
     if (PyList_MAXFREELIST && state->numfree) {
         state->numfree--;
         op = state->free_list[state->numfree];
+        OBJECT_STAT_INC(from_freelist);
         _Py_NewReference((PyObject *)op);
     }
     else
@@ -353,6 +354,7 @@ list_dealloc(PyListObject *op)
 #endif
     if (state->numfree < PyList_MAXFREELIST && PyList_CheckExact(op)) {
         state->free_list[state->numfree++] = op;
+        OBJECT_STAT_INC(to_freelist);
     }
     else
 #endif

Diff for: Objects/obmalloc.c

@@ -616,6 +616,10 @@ PyMem_Malloc(size_t size)
     /* see PyMem_RawMalloc() */
     if (size > (size_t)PY_SSIZE_T_MAX)
         return NULL;
+    OBJECT_STAT_INC_COND(allocations512, size < 512);
+    OBJECT_STAT_INC_COND(allocations4k, size >= 512 && size < 4094);
+    OBJECT_STAT_INC_COND(allocations_big, size >= 4094);
+    OBJECT_STAT_INC(allocations);
     return _PyMem.malloc(_PyMem.ctx, size);
 }
 
@@ -625,6 +629,10 @@ PyMem_Calloc(size_t nelem, size_t elsize)
     /* see PyMem_RawMalloc() */
     if (elsize != 0 && nelem > (size_t)PY_SSIZE_T_MAX / elsize)
         return NULL;
+    OBJECT_STAT_INC_COND(allocations512, elsize < 512);
+    OBJECT_STAT_INC_COND(allocations4k, elsize >= 512 && elsize < 4094);
+    OBJECT_STAT_INC_COND(allocations_big, elsize >= 4094);
+    OBJECT_STAT_INC(allocations);
     return _PyMem.calloc(_PyMem.ctx, nelem, elsize);
 }
 
@@ -640,6 +648,7 @@ PyMem_Realloc(void *ptr, size_t new_size)
 void
 PyMem_Free(void *ptr)
 {
+    OBJECT_STAT_INC(frees);
     _PyMem.free(_PyMem.ctx, ptr);
 }
 
@@ -696,6 +705,9 @@ PyObject_Malloc(size_t size)
     /* see PyMem_RawMalloc() */
     if (size > (size_t)PY_SSIZE_T_MAX)
         return NULL;
+    OBJECT_STAT_INC_COND(allocations512, size < 512);
+    OBJECT_STAT_INC_COND(allocations4k, size >= 512 && size < 4094);
+    OBJECT_STAT_INC_COND(allocations_big, size >= 4094);
     OBJECT_STAT_INC(allocations);
     return _PyObject.malloc(_PyObject.ctx, size);
 }
@@ -706,6 +718,9 @@ PyObject_Calloc(size_t nelem, size_t elsize)
     /* see PyMem_RawMalloc() */
     if (elsize != 0 && nelem > (size_t)PY_SSIZE_T_MAX / elsize)
         return NULL;
+    OBJECT_STAT_INC_COND(allocations512, elsize < 512);
+    OBJECT_STAT_INC_COND(allocations4k, elsize >= 512 && elsize < 4094);
+    OBJECT_STAT_INC_COND(allocations_big, elsize >= 4094);
     OBJECT_STAT_INC(allocations);
     return _PyObject.calloc(_PyObject.ctx, nelem, elsize);
 }

Diff for: Objects/tupleobject.c

@@ -1195,6 +1195,7 @@ maybe_freelist_pop(Py_ssize_t size)
 #endif
             _Py_NewReference((PyObject *)op);
             /* END inlined _PyObject_InitVar() */
+            OBJECT_STAT_INC(from_freelist);
             return op;
         }
     }
@@ -1224,6 +1225,7 @@ maybe_freelist_push(PyTupleObject *op)
         op->ob_item[0] = (PyObject *) STATE.free_list[index];
         STATE.free_list[index] = op;
         STATE.numfree[index]++;
+        OBJECT_STAT_INC(to_freelist);
         return 1;
     }
 #endif

Diff for: Python/context.c

@@ -351,6 +351,7 @@ _context_alloc(void)
         state->numfree--;
         ctx = state->freelist;
         state->freelist = (PyContext *)ctx->ctx_weakreflist;
+        OBJECT_STAT_INC(from_freelist);
         ctx->ctx_weakreflist = NULL;
         _Py_NewReference((PyObject *)ctx);
     }
@@ -482,6 +483,7 @@ context_tp_dealloc(PyContext *self)
         state->numfree++;
         self->ctx_weakreflist = (PyObject *)state->freelist;
         state->freelist = self;
+        OBJECT_STAT_INC(to_freelist);
     }
     else
 #endif

Diff for: Python/specialize.c

@@ -183,7 +183,12 @@ print_call_stats(FILE *out, CallStats *stats)
 static void
 print_object_stats(FILE *out, ObjectStats *stats)
 {
+    fprintf(out, "Object allocations from freelist: %" PRIu64 "\n", stats->from_freelist);
+    fprintf(out, "Object frees to freelist: %" PRIu64 "\n", stats->to_freelist);
     fprintf(out, "Object allocations: %" PRIu64 "\n", stats->allocations);
+    fprintf(out, "Object allocations to 512 bytes: %" PRIu64 "\n", stats->allocations512);
+    fprintf(out, "Object allocations to 4 kbytes: %" PRIu64 "\n", stats->allocations4k);
+    fprintf(out, "Object allocations over 4 kbytes: %" PRIu64 "\n", stats->allocations_big);
     fprintf(out, "Object frees: %" PRIu64 "\n", stats->frees);
     fprintf(out, "Object new values: %" PRIu64 "\n", stats->new_values);
     fprintf(out, "Object materialize dict (on request): %" PRIu64 "\n", stats->dict_materialized_on_request);