From 8b1ded5f7b8a86b40a4a0b494c0dce17dd837ec6 Mon Sep 17 00:00:00 2001 From: LloydZ <35182391+cocolato@users.noreply.github.com> Date: Thu, 18 Dec 2025 11:05:22 +0000 Subject: [PATCH] Eliminate redundant refcounting from _BINARY_OP_SUBSCR_LIST_INT --- Include/internal/pycore_opcode_metadata.h | 2 +- Include/internal/pycore_uop_ids.h | 2 +- Include/internal/pycore_uop_metadata.h | 6 ++--- Lib/test/test_capi/test_opt.py | 18 +++++++++++++++ Python/bytecodes.c | 8 ++++--- Python/executor_cases.c.h | 26 ++++++++------------- Python/generated_cases.c.h | 28 +++++++++++++++-------- Python/optimizer_bytecodes.c | 6 +++++ Python/optimizer_cases.c.h | 16 ++++++++++--- 9 files changed, 74 insertions(+), 38 deletions(-) diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h index a4f1f50e96d6d1..d36725804d270d 100644 --- a/Include/internal/pycore_opcode_metadata.h +++ b/Include/internal/pycore_opcode_metadata.h @@ -1338,7 +1338,7 @@ _PyOpcode_macro_expansion[256] = { [BINARY_OP_MULTIPLY_INT] = { .nuops = 5, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_INT, OPARG_SIMPLE, 0 }, { _BINARY_OP_MULTIPLY_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 }, { _POP_TOP_INT, OPARG_SIMPLE, 5 } } }, [BINARY_OP_SUBSCR_DICT] = { .nuops = 2, .uops = { { _GUARD_NOS_DICT, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_DICT, OPARG_SIMPLE, 5 } } }, [BINARY_OP_SUBSCR_GETITEM] = { .nuops = 4, .uops = { { _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 = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_LIST, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_LIST_INT, 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_STR_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_UNICODE, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_STR_INT, OPARG_SIMPLE, 5 } } }, [BINARY_OP_SUBSCR_TUPLE_INT] = { .nuops = 3, .uops = { { _GUARD_TOS_INT, OPARG_SIMPLE, 0 }, { _GUARD_NOS_TUPLE, OPARG_SIMPLE, 0 }, { _BINARY_OP_SUBSCR_TUPLE_INT, OPARG_SIMPLE, 5 } } }, diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h index 8678ccf5e1b823..1fe64cf3f8dfee 100644 --- a/Include/internal/pycore_uop_ids.h +++ b/Include/internal/pycore_uop_ids.h @@ -383,7 +383,7 @@ extern "C" { #define _BINARY_OP_SUBSCR_INIT_CALL_r11 576 #define _BINARY_OP_SUBSCR_INIT_CALL_r21 577 #define _BINARY_OP_SUBSCR_INIT_CALL_r31 578 -#define _BINARY_OP_SUBSCR_LIST_INT_r21 579 +#define _BINARY_OP_SUBSCR_LIST_INT_r23 579 #define _BINARY_OP_SUBSCR_LIST_SLICE_r21 580 #define _BINARY_OP_SUBSCR_STR_INT_r21 581 #define _BINARY_OP_SUBSCR_TUPLE_INT_r21 582 diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h index c6d3670b193123..30e58a3c7e709f 100644 --- a/Include/internal/pycore_uop_metadata.h +++ b/Include/internal/pycore_uop_metadata.h @@ -1108,7 +1108,7 @@ const _PyUopCachingInfo _PyUop_Caching[MAX_UOP_ID+1] = { .entries = { { -1, -1, -1 }, { -1, -1, -1 }, - { 1, 2, _BINARY_OP_SUBSCR_LIST_INT_r21 }, + { 3, 2, _BINARY_OP_SUBSCR_LIST_INT_r23 }, { -1, -1, -1 }, }, }, @@ -3423,7 +3423,7 @@ const uint16_t _PyUop_Uncached[MAX_UOP_REGS_ID+1] = { [_BINARY_OP_EXTEND_r21] = _BINARY_OP_EXTEND, [_BINARY_SLICE_r31] = _BINARY_SLICE, [_STORE_SLICE_r30] = _STORE_SLICE, - [_BINARY_OP_SUBSCR_LIST_INT_r21] = _BINARY_OP_SUBSCR_LIST_INT, + [_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_STR_INT_r21] = _BINARY_OP_SUBSCR_STR_INT, [_GUARD_NOS_TUPLE_r02] = _GUARD_NOS_TUPLE, @@ -3932,7 +3932,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_REGS_ID+1] = { [_BINARY_OP_SUBSCR_INIT_CALL_r21] = "_BINARY_OP_SUBSCR_INIT_CALL_r21", [_BINARY_OP_SUBSCR_INIT_CALL_r31] = "_BINARY_OP_SUBSCR_INIT_CALL_r31", [_BINARY_OP_SUBSCR_LIST_INT] = "_BINARY_OP_SUBSCR_LIST_INT", - [_BINARY_OP_SUBSCR_LIST_INT_r21] = "_BINARY_OP_SUBSCR_LIST_INT_r21", + [_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_STR_INT] = "_BINARY_OP_SUBSCR_STR_INT", diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py index bed8d1d52e93b2..116be23860501e 100644 --- a/Lib/test/test_capi/test_opt.py +++ b/Lib/test/test_capi/test_opt.py @@ -2972,6 +2972,24 @@ class Obj: for _ in range(TIER2_THRESHOLD+1): obj.attr = EvilAttr(obj.__dict__) + def test_binary_subscr_list_int(self): + def testfunc(n): + l = [1] + x = 0 + for _ in range(n): + y = l[0] + x += y + return x + + res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD) + self.assertEqual(res, TIER2_THRESHOLD) + self.assertIsNotNone(ex) + uops = get_opnames(ex) + + self.assertIn("_BINARY_OP_SUBSCR_LIST_INT", uops) + self.assertNotIn("_POP_TOP", uops) + self.assertNotIn("_POP_TOP_INT", uops) + self.assertIn("_POP_TOP_NOP", uops) def global_identity(x): return x diff --git a/Python/bytecodes.c b/Python/bytecodes.c index 25e730f7e6e8ec..df23d0ac364a01 100644 --- a/Python/bytecodes.c +++ b/Python/bytecodes.c @@ -891,9 +891,9 @@ dummy_func( macro(STORE_SLICE) = _SPECIALIZE_STORE_SLICE + _STORE_SLICE; macro(BINARY_OP_SUBSCR_LIST_INT) = - _GUARD_TOS_INT + _GUARD_NOS_LIST + unused/5 + _BINARY_OP_SUBSCR_LIST_INT; + _GUARD_TOS_INT + _GUARD_NOS_LIST + unused/5 + _BINARY_OP_SUBSCR_LIST_INT + _POP_TOP_INT + POP_TOP; - op(_BINARY_OP_SUBSCR_LIST_INT, (list_st, sub_st -- res)) { + op(_BINARY_OP_SUBSCR_LIST_INT, (list_st, sub_st -- res, ls, ss)) { PyObject *sub = PyStackRef_AsPyObjectBorrow(sub_st); PyObject *list = PyStackRef_AsPyObjectBorrow(list_st); @@ -916,7 +916,9 @@ dummy_func( res = PyStackRef_FromPyObjectNew(res_o); #endif STAT_INC(BINARY_OP, hit); - DECREF_INPUTS(); + ls = list_st; + ss = sub_st; + INPUTS_DEAD(); } macro(BINARY_OP_SUBSCR_LIST_SLICE) = diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h index 55c4495f2dc59d..7ea368314d3e88 100644 --- a/Python/executor_cases.c.h +++ b/Python/executor_cases.c.h @@ -4563,12 +4563,14 @@ break; } - case _BINARY_OP_SUBSCR_LIST_INT_r21: { + case _BINARY_OP_SUBSCR_LIST_INT_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; @@ -4619,23 +4621,13 @@ stack_pointer += 2; #endif STAT_INC(BINARY_OP, hit); - _PyFrame_SetStackPointer(frame, stack_pointer); - _PyStackRef tmp = list_st; - list_st = res; - stack_pointer[-2] = list_st; - PyStackRef_CLOSE(tmp); - tmp = sub_st; - sub_st = PyStackRef_NULL; - stack_pointer[-1] = sub_st; - PyStackRef_CLOSE(tmp); - stack_pointer = _PyFrame_GetStackPointer(frame); - stack_pointer += -1; - ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__); + 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); - stack_pointer += -1; + SET_CURRENT_CACHED_VALUES(3); + stack_pointer += -2; ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__); assert(WITHIN_STACK_BOUNDS_IGNORING_CACHE()); break; diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h index 3c18960a39d7d7..a3e1a2461aab75 100644 --- a/Python/generated_cases.c.h +++ b/Python/generated_cases.c.h @@ -750,6 +750,8 @@ _PyStackRef list_st; _PyStackRef sub_st; _PyStackRef res; + _PyStackRef ls; + _PyStackRef ss; // _GUARD_TOS_INT { value = stack_pointer[-1]; @@ -808,18 +810,24 @@ res = PyStackRef_FromPyObjectNew(res_o); #endif STAT_INC(BINARY_OP, hit); - _PyFrame_SetStackPointer(frame, stack_pointer); - _PyStackRef tmp = list_st; - list_st = res; - stack_pointer[-2] = list_st; - PyStackRef_CLOSE(tmp); - tmp = sub_st; - sub_st = PyStackRef_NULL; - stack_pointer[-1] = sub_st; - PyStackRef_CLOSE(tmp); - stack_pointer = _PyFrame_GetStackPointer(frame); + ls = list_st; + ss = sub_st; + } + // _POP_TOP_INT + { + value = ss; + assert(PyLong_CheckExact(PyStackRef_AsPyObjectBorrow(value))); + PyStackRef_CLOSE_SPECIALIZED(value, _PyLong_ExactDealloc); + } + // _POP_TOP + { + value = ls; + stack_pointer[-2] = res; stack_pointer += -1; ASSERT_WITHIN_STACK_BOUNDS(__FILE__, __LINE__); + _PyFrame_SetStackPointer(frame, stack_pointer); + PyStackRef_XCLOSE(value); + stack_pointer = _PyFrame_GetStackPointer(frame); } DISPATCH(); } diff --git a/Python/optimizer_bytecodes.c b/Python/optimizer_bytecodes.c index 31ec88e1c84912..2b56bfad840d5e 100644 --- a/Python/optimizer_bytecodes.c +++ b/Python/optimizer_bytecodes.c @@ -1409,6 +1409,12 @@ dummy_func(void) { } } + op(_BINARY_OP_SUBSCR_LIST_INT, (list_st, sub_st -- res, ls, ss)) { + res = sym_new_unknown(ctx); + ls = list_st; + ss = sub_st; + } + // END BYTECODES // diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h index 6fde87a17fbaa3..d4953c4f09a41a 100644 --- a/Python/optimizer_cases.c.h +++ b/Python/optimizer_cases.c.h @@ -753,11 +753,21 @@ } case _BINARY_OP_SUBSCR_LIST_INT: { + 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_unknown(ctx); + 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; }