Skip to content

bpo-46841: Move the cache for LOAD_GLOBAL inline. #31575

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

Merged
merged 14 commits into from
Feb 28, 2022
Merged

bpo-46841: Move the cache for LOAD_GLOBAL inline. #31575

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
a209634
Experiment: Make 'inline' cache for LOAD_GLOBAL actually inline.
markshannon Feb 18, 2022
8ceabb5
Emit cache entry space in assembler.
markshannon Feb 23, 2022
c8e38a3
Merge main and fix duplication of _PyOpcode_InlineCacheEntries table.
markshannon Feb 25, 2022
98f73a1
Use native endianness for 32 bit cache entries.
markshannon Feb 25, 2022
87b63f9
Remove unused code.
markshannon Feb 25, 2022
cca4d6b
Add news
markshannon Feb 25, 2022
26d0d70
Fix up test_dis
markshannon Feb 25, 2022
c49f156
Restore specialization of LOAD_GLOBAL.
markshannon Feb 25, 2022
f65ae9f
Turn off test_dis test output generation.
markshannon Feb 25, 2022
054fcf6
Merge branch 'main' into inline-cache-load-globals
markshannon Feb 25, 2022
c6c0527
Clean up messy merge.
markshannon Feb 28, 2022
0a853af
Address review comments.
markshannon Feb 28, 2022
92a4b04
Bump version number.
markshannon Feb 28, 2022
138b2d6
Merge branch 'main' into inline-cache-load-globals
markshannon Feb 28, 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
57 changes: 48 additions & 9 deletions Include/internal/pycore_code.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,11 +26,6 @@ typedef struct {
uint32_t dk_version;
} _PyAttrCache;

typedef struct {
uint32_t module_keys_version;
uint32_t builtin_keys_version;
} _PyLoadGlobalCache;

typedef struct {
/* Borrowed ref in LOAD_METHOD */
PyObject *obj;
Expand All @@ -57,23 +52,35 @@ typedef union {
_PyEntryZero zero;
_PyAdaptiveEntry adaptive;
_PyAttrCache attr;
_PyLoadGlobalCache load_global;
_PyObjectCache obj;
_PyCallCache call;
} SpecializedCacheEntry;

#define INSTRUCTIONS_PER_ENTRY (sizeof(SpecializedCacheEntry)/sizeof(_Py_CODEUNIT))

/* Inline caches */

#define CACHE_ENTRIES(cache) (sizeof(cache)/sizeof(_Py_CODEUNIT))

typedef struct {
_Py_CODEUNIT counter;
_Py_CODEUNIT index;
_Py_CODEUNIT module_keys_version;
_Py_CODEUNIT _m1;
_Py_CODEUNIT builtin_keys_version;
} _PyLoadGlobalCache;

#define INLINE_CACHE_ENTRIES_LOAD_GLOBAL CACHE_ENTRIES(_PyLoadGlobalCache)

typedef struct {
_Py_CODEUNIT counter;
} _PyBinaryOpCache;

#define INLINE_CACHE_ENTRIES_BINARY_OP CACHE_ENTRIES(_PyBinaryOpCache)
typedef struct {
_Py_CODEUNIT counter;
} _PyUnpackSequenceCache;

#define INLINE_CACHE_ENTRIES_BINARY_OP \
(sizeof(_PyBinaryOpCache) / sizeof(_Py_CODEUNIT))

#define INLINE_CACHE_ENTRIES_UNPACK_SEQUENCE \
(sizeof(_PyUnpackSequenceCache) / sizeof(_Py_CODEUNIT))
Expand Down Expand Up @@ -307,7 +314,7 @@ cache_backoff(_PyAdaptiveEntry *entry) {

extern int _Py_Specialize_LoadAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
extern int _Py_Specialize_StoreAttr(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
extern int _Py_Specialize_LoadGlobal(PyObject *globals, PyObject *builtins, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
extern int _Py_Specialize_LoadGlobal(PyObject *globals, PyObject *builtins, _Py_CODEUNIT *instr, PyObject *name);
extern int _Py_Specialize_LoadMethod(PyObject *owner, _Py_CODEUNIT *instr, PyObject *name, SpecializedCacheEntry *cache);
extern int _Py_Specialize_BinarySubscr(PyObject *sub, PyObject *container, _Py_CODEUNIT *instr, SpecializedCacheEntry *cache);
extern int _Py_Specialize_StoreSubscr(PyObject *container, PyObject *sub, _Py_CODEUNIT *instr);
Expand Down Expand Up @@ -388,6 +395,38 @@ extern PyObject* _Py_GetSpecializationStats(void);
#define OBJECT_STAT_INC(name) ((void)0)
#endif

// Cache values are only valid in memory, so use native endianness.
#ifdef WORDS_BIGENDIAN

static inline void
write32(uint16_t *p, uint32_t val)
{
p[0] = val >> 16;
p[1] = (uint16_t)val;
}

static inline uint32_t
read32(uint16_t *p)
{
return (p[0] << 16) | p[1];
}

#else

static inline void
write32(uint16_t *p, uint32_t val)
{
p[0] = (uint16_t)val;
p[1] = val >> 16;
}

static inline uint32_t
read32(uint16_t *p)
{
return p[0] | (p[1] << 16);
}

#endif

#ifdef __cplusplus
}
Expand Down
20 changes: 12 additions & 8 deletions Include/opcode.h
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

4 changes: 2 additions & 2 deletions Lib/importlib/_bootstrap_external.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -387,8 +387,8 @@ def _write_atomic(path, data, mode=0o666):
# Python 3.11a5 3478 (New CALL opcodes)
# Python 3.11a5 3479 (Add PUSH_NULL opcode)
# Python 3.11a5 3480 (New CALL opcodes, second iteration)
# Python 3.11a5 3481 (Use inline CACHE instructions)
# Python 3.11a5 3482 (Use inline caching for UNPACK_SEQUENCE)
# Python 3.11a5 3481 (Use inline cache for BINARY_OP)
# Python 3.11a5 3482 (Use inline caching for UNPACK_SEQUENCE and LOAD_GLOBAL)
Comment on lines -390 to +391
Copy link
Member

@brandtbucher brandtbucher Feb 28, 2022 •
edited
Loading

Choose a reason for hiding this comment

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

Hm, looks like the magic number wasn't actually changed... bad merge?

Copy link
Member Author

Choose a reason for hiding this comment

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

I figured we didn't need more than one version number per day.

Copy link
Member Author

Choose a reason for hiding this comment

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

#31618 has another version number bump.

Copy link
Member

Choose a reason for hiding this comment

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

Ah, I see what you did now. Thanks for clarifying.


# Python 3.12 will start with magic number 3500

Expand Down
3 changes: 2 additions & 1 deletion Lib/opcode.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -135,7 +135,7 @@ def jabs_op(name, op, entries=0):
jabs_op('JUMP_ABSOLUTE', 113) # ""
jabs_op('POP_JUMP_IF_FALSE', 114) # ""
jabs_op('POP_JUMP_IF_TRUE', 115) # ""
name_op('LOAD_GLOBAL', 116) # Index in name list
name_op('LOAD_GLOBAL', 116, 5) # Index in name list
def_op('IS_OP', 117)
def_op('CONTAINS_OP', 118)
def_op('RERAISE', 119)
Expand Down Expand Up @@ -198,6 +198,7 @@ def jabs_op(name, op, entries=0):
def_op('KW_NAMES', 172)
hasconst.append(172)


del def_op, name_op, jrel_op, jabs_op

_nb_ops = [
Expand Down
Loading