8000 bpo-46841: Inline cache for `BINARY_SUBSCR`. by markshannon · Pull Request #31618 · python/cpython · GitHub
[go: up one dir, main page]
Skip to content

bpo-46841: Inline cache for BINARY_SUBSCR. #31618

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 11 commits into from
Mar 1, 2022
Merged

bpo-46841: Inline cache for BINARY_SUBSCR. #31618

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
465f0c0
Use inline cache for BINARY_SUBSCR. Work in progress.
markshannon Feb 28, 2022
8381b7f
Merge branch 'main' into inline-cache-binary-subscr
markshannon Feb 28, 2022
a9f7043
Finish inlining cache for .
markshannon Feb 28, 2022
a330574
Tidy up
markshannon Feb 28, 2022
e4b0ba0
Address review comments.
markshannon Mar 1, 2022
05787b3
Remove per-code-object-cache and add small per-heap-type cache.
markshannon Mar 1, 2022
956ed1e
Remove unused parameter.
markshannon Mar 1, 2022
0e282f3
Remove unused code.
markshannon Mar 1, 2022
d0c9077
Merge branch 'main' into inline-cache-binary-subscr
markshannon Mar 1, 2022
f667ddb
Clarify comments.
markshannon Mar 1, 2022
2271337
Fix typo.
markshannon Mar 1, 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
Prev Previous commit
Next Next commit
Merge branch 'main' into inline-cache-binary-subscr
  • Loading branch information
@markshannon
markshannon committed Feb 28, 2022
commit 8381b7ff668d1550f387c9238fc5df023ff36d88
11 changes: 0 additions & 11 deletions Include/internal/pycore_code.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -98,8 +98,6 @@ typedef struct {
_Py_CODEUNIT func_version;
} _PyBinarySubscrCache;

#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 @@ -127,15 +125,6 @@ _GetSpecializedCacheEntry(const _Py_CODEUNIT *first_instr, Py_ssize_t n)
return &last_cache_plus_one[-1-n].entry;
}

/* Returns a borrowed reference */
static inline PyObject*
_PyQuickenedGetObject(const _Py_CODEUNIT *first_instr, uint16_t index)
{
SpecializedCacheOrInstruction *last_cache_plus_one = (SpecializedCacheOrInstruction *)first_instr;
assert(&last_cache_plus_one->code[0] == first_instr);
return last_cache_plus_one[-1-index].entry.obj.obj;
}

/* Following two functions form a pair.
*
* oparg_from_offset_and_index() is used to compute the oparg
Expand Down
7 changes: 1 addition & 6 deletions Include/opcode.h
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear 10000 on GitHub.

4 changes: 4 additions & 0 deletions Objects/codeobject.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -348,6 +348,7 @@ init_code(PyCodeObject *co, struct _PyCodeConstructor *con)
/* not set */
co->co_weakreflist = NULL;
co->co_extra = NULL;
co->_co_obj_cache = NULL;

co->co_warmup = QUICKENING_INITIAL_WARMUP_VALUE;
co->co_quickened = NULL;
Expand Down Expand Up @@ -1369,6 +1370,9 @@ code_dealloc(PyCodeObject *co)
PyMem_Free(co->co_quickened);
_Py_QuickenedCount--;
}
if (co->_co_obj_cache) {
PyMem_Free(co->_co_obj_cache);
}
PyObject_Free(co);
}

Expand Down
2 changes: 1 addition & 1 deletion Python/ceval.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2197,7 +2197,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
PyObject *sub = TOP();
PyObject *container = SECOND();
_PyBinarySubscrCache *cache = (_PyBinarySubscrCache *)next_instr;
PyObject *cached = _PyQuickenedGetObject(first_instr, cache->object);
PyObject *cached = frame->f_code->_co_obj_cache[cache->object];
assert(PyFunction_Check(cached));
PyFunctionObject *getitem = (PyFunctionObject *)cached;
uint32_t type_version = read32(&cache->type_version);
Expand Down
44 changes: 28 additions & 16 deletions Python/specialize.c
View file
Open in desktop 10000
Original file line number Diff line number Diff line change
Expand Up @@ -69,7 +69,7 @@ static uint8_t cache_requirements[256] = {

/* The number of object cache entries required for a "family" of instructions. */
static const uint8_t object_cache_requirements[256] = {
[BINARY_SUBSCR] = 5,
[BINARY_SUBSCR] = 1,
};

Py_ssize_t _Py_QuickenedCount = 0;
Expand Down Expand Up @@ -290,14 +290,6 @@ _Py_PrintSpecializationStats(int to_file)
#define SPECIALIZATION_FAIL(opcode, kind) ((void)0)
#endif

static void
_PyQuickenedSetObject(const _Py_CODEUNIT *first_instr, uint16_t index, PyObject *obj)
{
SpecializedCacheOrInstruction *last_cache_plus_one = (SpecializedCacheOrInstruction *)first_instr;
assert(&last_cache_plus_one->code[0] == first_instr);
last_cache_plus_one[-1-index].entry.obj.obj = obj;
}

static SpecializedCacheOrInstruction *
allocate(int cache_count, int instruction_count)
{
Expand Down Expand Up @@ -364,17 +356,25 @@ entries_needed(const _Py_CODEUNIT *code, int len)
int previous_opcode = -1;
for (int i = 0; i < len; i++) {
uint8_t opcode = _Py_OPCODE(code[i]);
if (object_cache_requirements[opcode]) {
cache_offset += object_cache_requirements[opcode];
}
else if (previous_opcode != EXTENDED_ARG) {
if (previous_opcode != EXTENDED_ARG) {
oparg_from_instruction_and_update_offset(i, opcode, 0, &cache_offset);
}
previous_opcode = opcode;
}
return cache_offset + 1; // One extra for the count entry
}

static int
object_slots_needed(const _Py_CODEUNIT *code, int len)
{
int count = 0;
for (int i = 0; i < len; i++) {
uint8_t opcode = _Py_OPCODE(code[i]);
count += object_cache_requirements[opcode];
}
return count;
}

static inline _Py_CODEUNIT *
first_instruction(SpecializedCacheOrInstruction *quickened)
{
Expand All @@ -392,6 +392,7 @@ optimize(SpecializedCacheOrInstruction *quickened, int len)
{
_Py_CODEUNIT *instructions = first_instruction(quickened);
int cache_offset = 0;
int object_offset = 0;
int previous_opcode = -1;
int previous_oparg = 0;
for(int i = 0; i < len; i++) {
Expand All @@ -403,9 +404,11 @@ optimize(SpecializedCacheOrInstruction *quickened, int len)
instructions[i] = _Py_MAKECODEUNIT(adaptive_opcode, oparg);
if (object_cache_requirements[opcode]) {
assert(_PyOpcode_InlineCacheEntries[opcode] >= 2);
instructions[i+2] = cache_offset;
cache_offset += object_cache_requirements[opcode];
instructions[i+2] = object_offset;
object_offset += object_cache_requirements[opcode];
}
previous_opcode = -1;
i += _PyOpcode_InlineCacheEntries[opcode];
}
else if (previous_opcode != EXTENDED_ARG) {
int new_oparg = oparg_from_instruction_and_update_offset(
Expand Down Expand Up @@ -484,9 +487,16 @@ _Py_Quicken(PyCodeObject *code) {
code->co_warmup = QUICKENING_WARMUP_COLDEST;
return 0;
}
int obj_count = object_slots_needed(code->co_firstinstr, instr_count);
code->_co_obj_cache = PyMem_Malloc(obj_count*sizeof(PyObject *));
code->_co_obj_cache_len = obj_count;
if (code->_co_obj_cache == NULL) {
return -1;
}
int entry_count = entries_needed(code->co_firstinstr, instr_count);
SpecializedCacheOrInstruction *quickened = allocate(entry_count, instr_count);
if (quickened == NULL) {
PyMem_Free(code->_co_obj_cache);
return -1;
}
_Py_CODEUNIT *new_instructions = first_instruction(quickened);
Expand Down Expand Up @@ -1406,7 +1416,9 @@ _Py_Specialize_BinarySubscr(
goto fail;
}
cache->func_version = version;
_PyQuickenedSetObject(code->co_firstinstr, cache->object, descriptor);
assert(code->_co_obj_cache != NULL);
assert(cache->object >= 0 && cache->object < code->_co_obj_cache_len);
code->_co_obj_cache[cache->object] = descriptor;
*instr = _Py_MAKECODEUNIT(BINARY_SUBSCR_GETITEM, _Py_OPARG(*instr));
goto success;
}
Expand Down
You are viewing a condensed version of this merge commit. You can view the full changes here.
0