Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Move stats for the method cache into the Py_STAT machinery #100255

Merged
merged 4 commits into from
Dec 15, 2022
Merged

Move stats for the method cache into the Py_STAT machinery #100255

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
56734b3
Add stats for the global type cache.
markshannon Dec 14, 2022
0ceb725
Print out stats
markshannon Dec 14, 2022
67fc05f
Add type cache stats for dunders and collisions.
markshannon Dec 14, 2022
6d6aa01
Rename 'type cache' to 'method cache' in stats output.
markshannon Dec 14, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 0 additions & 6 deletions Include/internal/pycore_typeobject.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,15 +36,9 @@ struct type_cache_entry {
};

#define MCACHE_SIZE_EXP 12
#define MCACHE_STATS 0

struct type_cache {
struct type_cache_entry hashtable[1 << MCACHE_SIZE_EXP];
#if MCACHE_STATS
size_t hits;
size_t misses;
size_t collisions;
#endif
};

/* For now we hard-code this to a value for which we are confident
Expand Down
7 changes: 7 additions & 0 deletions Include/pystats.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,8 +65,15 @@ typedef struct _object_stats {
uint64_t dict_materialized_new_key;
uint64_t dict_materialized_too_big;
uint64_t dict_materialized_str_subclass;
uint64_t type_cache_hits;
uint64_t type_cache_misses;
uint64_t type_cache_dunder_hits;
uint64_t type_cache_dunder_misses;
uint64_t type_cache_collisions;
} ObjectStats;

#

typedef struct _stats {
OpcodeStats opcode_stats[256];
CallStats call_stats;
Expand Down
64 changes: 23 additions & 41 deletions Objects/typeobject.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -327,18 +327,6 @@ static unsigned int
_PyType_ClearCache(PyInterpreterState *interp)
{
struct type_cache *cache = &interp->types.type_cache;
#if MCACHE_STATS
size_t total = cache->hits + cache->collisions + cache->misses;
fprintf(stderr, "-- Method cache hits = %zd (%d%%)\n",
cache->hits, (int) (100.0 * cache->hits / total));
fprintf(stderr, "-- Method cache true misses = %zd (%d%%)\n",
cache->misses, (int) (100.0 * cache->misses / total));
fprintf(stderr, "-- Method cache collisions = %zd (%d%%)\n",
cache->collisions, (int) (100.0 * cache->collisions / total));
fprintf(stderr, "-- Method cache size = %zd KiB\n",
sizeof(cache->hashtable) / 1024);
#endif

// Set to None, rather than NULL, so _PyType_Lookup() can
// use Py_SETREF() rather than using slower Py_XSETREF().
type_cache_clear(cache, Py_None);
Expand Down Expand Up @@ -4147,6 +4135,24 @@ find_name_in_mro(PyTypeObject *type, PyObject *name, int *error)
return res;
}

/* Check if the "readied" PyUnicode name
is a double-underscore special name. */
static int
is_dunder_name(PyObject *name)
{
Py_ssize_t length = PyUnicode_GET_LENGTH(name);
int kind = PyUnicode_KIND(name);
/* Special names contain at least "__x__" and are always ASCII. */
if (length > 4 && kind == PyUnicode_1BYTE_KIND) {
const Py_UCS1 *characters = PyUnicode_1BYTE_DATA(name);
return (
((characters[length-2] == '_') && (characters[length-1] == '_')) &&
((characters[0] == '_') && (characters[1] == '_'))
);
}
return 0;
}

/* Internal API to look for a name through the MRO.
This returns a borrowed reference, and doesn't set an exception! */
PyObject *
Expand All @@ -4160,12 +4166,13 @@ _PyType_Lookup(PyTypeObject *type, PyObject *name)
struct type_cache_entry *entry = &cache->hashtable[h];
if (entry->version == type->tp_version_tag &&
entry->name == name) {
#if MCACHE_STATS
cache->hits++;
#endif
assert(_PyType_HasFeature(type, Py_TPFLAGS_VALID_VERSION_TAG));
OBJECT_STAT_INC_COND(type_cache_hits, !is_dunder_name(name));
OBJECT_STAT_INC_COND(type_cache_dunder_hits, is_dunder_name(name));
return entry->value;
}
OBJECT_STAT_INC_COND(type_cache_misses, !is_dunder_name(name));
OBJECT_STAT_INC_COND(type_cache_dunder_misses, is_dunder_name(name));
Comment on lines +4174 to +4175
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I assume there's no user of these stats who will care that the definition of a "miss" now includes collisions whereas before it excluded collisions? Obviously the same information is derivable either way.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The old stats were only available if MCACHE_STATS was defined, which could only be done by editing the source or build. There is no configure option. So, I don't see how there can be a user.


/* We may end up clearing live exceptions below, so make sure it's ours. */
assert(!PyErr_Occurred());
Expand Down Expand Up @@ -4193,14 +4200,7 @@ _PyType_Lookup(PyTypeObject *type, PyObject *name)
entry->version = type->tp_version_tag;
entry->value = res; /* borrowed */
assert(_PyASCIIObject_CAST(name)->hash != -1);
#if MCACHE_STATS
if (entry->name != Py_None && entry->name != name) {
cache->collisions++;
}
else {
cache->misses++;
}
#endif
OBJECT_STAT_INC_COND(type_cache_collisions, entry->name != Py_None && entry->name != name);
assert(_PyType_HasFeature(type, Py_TPFLAGS_VALID_VERSION_TAG));
Py_SETREF(entry->name, Py_NewRef(name));
}
Expand All @@ -4217,24 +4217,6 @@ _PyType_LookupId(PyTypeObject *type, _Py_Identifier *name)
return _PyType_Lookup(type, oname);
}

/* Check if the "readied" PyUnicode name
is a double-underscore special name. */
static int
is_dunder_name(PyObject *name)
{
Py_ssize_t length = PyUnicode_GET_LENGTH(name);
int kind = PyUnicode_KIND(name);
/* Special names contain at least "__x__" and are always ASCII. */
if (length > 4 && kind == PyUnicode_1BYTE_KIND) {
const Py_UCS1 *characters = PyUnicode_1BYTE_DATA(name);
return (
((characters[length-2] == '_') && (characters[length-1] == '_')) &&
((characters[0] == '_') && (characters[1] == '_'))
);
}
return 0;
}

/* This is similar to PyObject_GenericGetAttr(),
but uses _PyType_Lookup() instead of just looking in type->tp_dict. */
static PyObject *
Expand Down
5 changes: 5 additions & 0 deletions Python/specialize.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -187,6 +187,11 @@ print_object_stats(FILE *out, ObjectStats *stats)
fprintf(out, "Object materialize dict (new key): %" PRIu64 "\n", stats->dict_materialized_new_key);
fprintf(out, "Object materialize dict (too big): %" PRIu64 "\n", stats->dict_materialized_too_big);
fprintf(out, "Object materialize dict (str subclass): %" PRIu64 "\n", stats->dict_materialized_str_subclass);
fprintf(out, "Object method cache hits: %" PRIu64 "\n", stats->type_cache_hits);
fprintf(out, "Object method cache misses: %" PRIu64 "\n", stats->type_cache_misses);
fprintf(out, "Object method cache collisions: %" PRIu64 "\n", stats->type_cache_collisions);
fprintf(out, "Object method cache dunder hits: %" PRIu64 "\n", stats->type_cache_dunder_hits);
fprintf(out, "Object method cache dunder misses: %" PRIu64 "\n", stats->type_cache_dunder_misses);
}

static void
Expand Down