gh-134584: Optimize _BINARY_OP_SUBSCR_LIST_SLICE (GH-144659)

This commit is contained in:
Sacul
2026-02-10 22:33:32 +08:00
committed by GitHub
parent 40a82abe93
commit 6c8ca1c378
11 changed files with 108 additions and 60 deletions
+2 -2
View File
@@ -1103,7 +1103,7 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[267] = {
[BINARY_OP_SUBSCR_DICT] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_SUBSCR_GETITEM] = { true, INSTR_FMT_IXC0000, HAS_DEOPT_FLAG | HAS_SYNC_SP_FLAG | HAS_NEEDS_GUARD_IP_FLAG | HAS_RECORDS_VALUE_FLAG },
[BINARY_OP_SUBSCR_LIST_INT] = { true, INSTR_FMT_IXC0000, HAS_DEOPT_FLAG | HAS_EXIT_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_SUBSCR_LIST_SLICE] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_SUBSCR_LIST_SLICE] = { true, INSTR_FMT_IXC0000, HAS_EXIT_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_SUBSCR_STR_INT] = { true, INSTR_FMT_IXC0000, HAS_DEOPT_FLAG | HAS_EXIT_FLAG },
[BINARY_OP_SUBSCR_TUPLE_INT] = { true, INSTR_FMT_IXC0000, HAS_DEOPT_FLAG | HAS_EXIT_FLAG | HAS_ESCAPES_FLAG },
[BINARY_OP_SUBSCR_USTR_INT] = { true, INSTR_FMT_IXC0000, HAS_DEOPT_FLAG | HAS_EXIT_FLAG },
@@ -1356,7 +1356,7 @@ _PyOpcode_macro_expansion[256] = {
[BINARY_OP_SUBSCR_DICT] = { .nuops = 4, .uops = { { _GUARD_NOS_DICT, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_DICT, OPARG_SIMPLE, 5 }, { _POP_TOP, OPARG_SIMPLE, 5 }, { _POP_TOP, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_GETITEM] = { .nuops = 5, .uops = { { _RECORD_NOS, OPARG_SIMPLE, 0 }, { _CHECK_PEP_523, OPARG_SIMPLE, 5 }, { _BINARY_OP_SUBSCR_CHECK_FUNC, OPARG_SIMPLE, 5 }, { _BINARY_OP_SUBSCR_INIT_CALL, OPARG_SIMPLE, 5 }, { _PUSH_FRAME, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_LIST_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_LIST_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_LIST_SLICE] = { .nuops = 3, .uops = { { _GUARD_TOS_SLICE, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_LIST_SLICE, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_LIST_SLICE] = { .nuops = 5, .uops = { { _GUARD_TOS_SLICE, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_LIST_SLICE, OPARG_SIMPLE, 5 }, { _POP_TOP, OPARG_SIMPLE, 5 }, { _POP_TOP, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_STR_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_COMPACT_ASCII, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_STR_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_UNICODE, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_TUPLE_INT] = { .nuops = 6, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_TUPLE, OPARG_SIMPLE, 0 }, { _GUARD_BINARY_OP_SUBSCR_TUPLE_INT_BOUNDS, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_TUPLE_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP, OPARG_SIMPLE, 5 } } },
[BINARY_OP_SUBSCR_USTR_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_UNICODE, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_USTR_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_UNICODE, OPARG_SIMPLE, 5 } } },
+1 -1
View File
@@ -414,7 +414,7 @@ extern "C" {
#define _BINARY_OP_SUBSCR_INIT_CALL_r21 611
#define _BINARY_OP_SUBSCR_INIT_CALL_r31 612
#define _BINARY_OP_SUBSCR_LIST_INT_r23 613
#define _BINARY_OP_SUBSCR_LIST_SLICE_r21 614
#define _BINARY_OP_SUBSCR_LIST_SLICE_r23 614
#define _BINARY_OP_SUBSCR_STR_INT_r23 615
#define _BINARY_OP_SUBSCR_TUPLE_INT_r03 616
#define _BINARY_OP_SUBSCR_TUPLE_INT_r13 617
+4 -4
View File
@@ -119,7 +119,7 @@ const uint32_t _PyUop_Flags[MAX_UOP_ID+1] = {
[_BINARY_SLICE] = HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
[_STORE_SLICE] = HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
[_BINARY_OP_SUBSCR_LIST_INT] = HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG,
[_BINARY_OP_SUBSCR_LIST_SLICE] = HAS_ERROR_FLAG | HAS_ESCAPES_FLAG,
[_BINARY_OP_SUBSCR_LIST_SLICE] = HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG,
[_BINARY_OP_SUBSCR_STR_INT] = HAS_DEOPT_FLAG,
[_BINARY_OP_SUBSCR_USTR_INT] = HAS_DEOPT_FLAG,
[_GUARD_NOS_TUPLE] = HAS_EXIT_FLAG,
@@ -1159,7 +1159,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = {
.entries = {
{ -1, -1, -1 },
{ -1, -1, -1 },
{ 1, 2, _BINARY_OP_SUBSCR_LIST_SLICE_r21 },
{ 3, 2, _BINARY_OP_SUBSCR_LIST_SLICE_r23 },
{ -1, -1, -1 },
},
},
@@ -3702,7 +3702,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = {
[_BINARY_SLICE_r31] = _BINARY_SLICE,
[_STORE_SLICE_r30] = _STORE_SLICE,
[_BINARY_OP_SUBSCR_LIST_INT_r23] = _BINARY_OP_SUBSCR_LIST_INT,
[_BINARY_OP_SUBSCR_LIST_SLICE_r21] = _BINARY_OP_SUBSCR_LIST_SLICE,
[_BINARY_OP_SUBSCR_LIST_SLICE_r23] = _BINARY_OP_SUBSCR_LIST_SLICE,
[_BINARY_OP_SUBSCR_STR_INT_r23] = _BINARY_OP_SUBSCR_STR_INT,
[_BINARY_OP_SUBSCR_USTR_INT_r23] = _BINARY_OP_SUBSCR_USTR_INT,
[_GUARD_NOS_TUPLE_r02] = _GUARD_NOS_TUPLE,
@@ -4297,7 +4297,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = {
[_BINARY_OP_SUBSCR_LIST_INT] = "_BINARY_OP_SUBSCR_LIST_INT",
[_BINARY_OP_SUBSCR_LIST_INT_r23] = "_BINARY_OP_SUBSCR_LIST_INT_r23",
[_BINARY_OP_SUBSCR_LIST_SLICE] = "_BINARY_OP_SUBSCR_LIST_SLICE",
[_BINARY_OP_SUBSCR_LIST_SLICE_r21] = "_BINARY_OP_SUBSCR_LIST_SLICE_r21",
[_BINARY_OP_SUBSCR_LIST_SLICE_r23] = "_BINARY_OP_SUBSCR_LIST_SLICE_r23",
[_BINARY_OP_SUBSCR_STR_INT] = "_BINARY_OP_SUBSCR_STR_INT",
[_BINARY_OP_SUBSCR_STR_INT_r23] = "_BINARY_OP_SUBSCR_STR_INT_r23",
[_BINARY_OP_SUBSCR_TUPLE_INT] = "_BINARY_OP_SUBSCR_TUPLE_INT",
+17
View File
@@ -3660,6 +3660,23 @@ class TestUopsOptimization(unittest.TestCase):
self.assertLessEqual(count_ops(ex, "_POP_TOP_INT"), 1)
self.assertIn("_POP_TOP_NOP", uops)
def test_binary_subscr_list_slice(self):
def testfunc(n):
x = 0
for _ in range(n):
l = [1, 2, 3]
x += l[0:1][0]
return x
res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
self.assertEqual(res, TIER2_THRESHOLD)
uops = get_opnames(ex)
self.assertIn("_BINARY_OP_SUBSCR_LIST_SLICE", uops)
self.assertNotIn("_GUARD_TOS_LIST", uops)
self.assertEqual(count_ops(ex, "_POP_TOP"), 3)
self.assertEqual(count_ops(ex, "_POP_TOP_NOP"), 4)
def test_is_op(self):
def test_is_false(n):
a = object()
@@ -0,0 +1 @@
Optimize and eliminate ref-counting in ``_BINARY_OP_SUBSCR_LIST_SLICE``
+23 -15
View File
@@ -892,6 +892,9 @@
_PyStackRef list_st;
_PyStackRef sub_st;
_PyStackRef res;
_PyStackRef ls;
_PyStackRef ss;
_PyStackRef value;
// _GUARD_TOS_SLICE
{
tos = stack_pointer[-1];
@@ -925,26 +928,31 @@
PyObject *res_o = _PyList_SliceSubscript(list, sub);
stack_pointer = _PyFrame_GetStackPointer(frame);
STAT_INC(BINARY_OP, hit);
_PyFrame_SetStackPointer(frame, stack_pointer);
_PyStackRef tmp = sub_st;
sub_st = PyStackRef_NULL;
stack_pointer[-1] = sub_st;
PyStackRef_CLOSE(tmp);
tmp = list_st;
list_st = PyStackRef_NULL;
stack_pointer[-2] = list_st;
PyStackRef_CLOSE(tmp);
stack_pointer = _PyFrame_GetStackPointer(frame);
stack_pointer += -2;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
if (res_o == NULL) {
JUMP_TO_LABEL(error);
}
res = PyStackRef_FromPyObjectSteal(res_o);
ls = list_st;
ss = sub_st;
}
// _POP_TOP
{
value = ss;
stack_pointer[-2] = res;
stack_pointer[-1] = ls;
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_XCLOSE(value);
stack_pointer = _PyFrame_GetStackPointer(frame);
}
// _POP_TOP
{
value = ls;
stack_pointer += -1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_XCLOSE(value);
stack_pointer = _PyFrame_GetStackPointer(frame);
}
stack_pointer[0] = res;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
DISPATCH();
}
+8 -4
View File
@@ -938,9 +938,9 @@ dummy_func(
}
macro(BINARY_OP_SUBSCR_LIST_SLICE) =
_GUARD_TOS_SLICE + _GUARD_NOS_LIST + unused/5 + _BINARY_OP_SUBSCR_LIST_SLICE;
_GUARD_TOS_SLICE + _GUARD_NOS_LIST + unused/5 + _BINARY_OP_SUBSCR_LIST_SLICE + POP_TOP + POP_TOP;
op(_BINARY_OP_SUBSCR_LIST_SLICE, (list_st, sub_st -- res)) {
op(_BINARY_OP_SUBSCR_LIST_SLICE, (list_st, sub_st -- res, ls, ss)) {
PyObject *sub = PyStackRef_AsPyObjectBorrow(sub_st);
PyObject *list = PyStackRef_AsPyObjectBorrow(list_st);
@@ -949,9 +949,13 @@ dummy_func(
PyObject *res_o = _PyList_SliceSubscript(list, sub);
STAT_INC(BINARY_OP, hit);
DECREF_INPUTS();
ERROR_IF(res_o == NULL);
if (res_o == NULL) {
ERROR_NO_POP();
}
res = PyStackRef_FromPyObjectSteal(res_o);
ls = list_st;
ss = sub_st;
INPUTS_DEAD();
}
macro(BINARY_OP_SUBSCR_STR_INT) =
+10 -16
View File
@@ -5368,12 +5368,14 @@
break;
}
case _BINARY_OP_SUBSCR_LIST_SLICE_r21: {
case _BINARY_OP_SUBSCR_LIST_SLICE_r23: {
CHECK_CURRENT_CACHED_VALUES(2);
assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());
_PyStackRef sub_st;
_PyStackRef list_st;
_PyStackRef res;
_PyStackRef ls;
_PyStackRef ss;
_PyStackRef _stack_item_0 = _tos_cache0;
_PyStackRef _stack_item_1 = _tos_cache1;
sub_st = _stack_item_1;
@@ -5390,27 +5392,19 @@
PyObject *res_o = _PyList_SliceSubscript(list, sub);
stack_pointer = _PyFrame_GetStackPointer(frame);
STAT_INC(BINARY_OP, hit);
_PyFrame_SetStackPointer(frame, stack_pointer);
_PyStackRef tmp = sub_st;
sub_st = PyStackRef_NULL;
stack_pointer[-1] = sub_st;
PyStackRef_CLOSE(tmp);
tmp = list_st;
list_st = PyStackRef_NULL;
stack_pointer[-2] = list_st;
PyStackRef_CLOSE(tmp);
stack_pointer = _PyFrame_GetStackPointer(frame);
stack_pointer += -2;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
if (res_o == NULL) {
SET_CURRENT_CACHED_VALUES(0);
JUMP_TO_ERROR();
}
res = PyStackRef_FromPyObjectSteal(res_o);
ls = list_st;
ss = sub_st;
_tos_cache2 = ss;
_tos_cache1 = ls;
_tos_cache0 = res;
_tos_cache1 = PyStackRef_ZERO_BITS;
_tos_cache2 = PyStackRef_ZERO_BITS;
SET_CURRENT_CACHED_VALUES(1);
SET_CURRENT_CACHED_VALUES(3);
stack_pointer += -2;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE());
break;
}
+23 -15
View File
@@ -892,6 +892,9 @@
_PyStackRef list_st;
_PyStackRef sub_st;
_PyStackRef res;
_PyStackRef ls;
_PyStackRef ss;
_PyStackRef value;
// _GUARD_TOS_SLICE
{
tos = stack_pointer[-1];
@@ -925,26 +928,31 @@
PyObject *res_o = _PyList_SliceSubscript(list, sub);
stack_pointer = _PyFrame_GetStackPointer(frame);
STAT_INC(BINARY_OP, hit);
_PyFrame_SetStackPointer(frame, stack_pointer);
_PyStackRef tmp = sub_st;
sub_st = PyStackRef_NULL;
stack_pointer[-1] = sub_st;
PyStackRef_CLOSE(tmp);
tmp = list_st;
list_st = PyStackRef_NULL;
stack_pointer[-2] = list_st;
PyStackRef_CLOSE(tmp);
stack_pointer = _PyFrame_GetStackPointer(frame);
stack_pointer += -2;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
if (res_o == NULL) {
JUMP_TO_LABEL(error);
}
res = PyStackRef_FromPyObjectSteal(res_o);
ls = list_st;
ss = sub_st;
}
// _POP_TOP
{
value = ss;
stack_pointer[-2] = res;
stack_pointer[-1] = ls;
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_XCLOSE(value);
stack_pointer = _PyFrame_GetStackPointer(frame);
}
// _POP_TOP
{
value = ls;
stack_pointer += -1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
_PyFrame_SetStackPointer(frame, stack_pointer);
PyStackRef_XCLOSE(value);
stack_pointer = _PyFrame_GetStackPointer(frame);
}
stack_pointer[0] = res;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
DISPATCH();
}
+6
View File
@@ -422,6 +422,12 @@ dummy_func(void) {
ss = sub_st;
}
op(_BINARY_OP_SUBSCR_LIST_SLICE, (list_st, sub_st -- res, ls, ss)) {
res = sym_new_type(ctx, &PyList_Type);
ls = list_st;
ss = sub_st;
}
op(_TO_BOOL, (value -- res)) {
int already_bool = optimize_to_bool(this_instr, ctx, value, &res, false);
if (!already_bool) {
+13 -3
View File
@@ -957,11 +957,21 @@
}
case _BINARY_OP_SUBSCR_LIST_SLICE: {
JitOptRef sub_st;
JitOptRef list_st;
JitOptRef res;
res = sym_new_not_null(ctx);
CHECK_STACK_BOUNDS(-1);
JitOptRef ls;
JitOptRef ss;
sub_st = stack_pointer[-1];
list_st = stack_pointer[-2];
res = sym_new_type(ctx, &PyList_Type);
ls = list_st;
ss = sub_st;
CHECK_STACK_BOUNDS(1);
stack_pointer[-2] = res;
stack_pointer += -1;
stack_pointer[-1] = ls;
stack_pointer[0] = ss;
stack_pointer += 1;
ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__);
break;
}