From e9e2bb9acd29f7d1426baf1c055d60d739340a66 Mon Sep 17 00:00:00 2001 From: Ken Jin <28750310+Fidget-Spinner@users.noreply.github.com> Date: Mon, 27 Oct 2025 22:27:46 +0000 Subject: [PATCH] Sink LOAD_IP into guards --- Include/internal/pycore_uop_ids.h | 318 ++++++++++++----------- Include/internal/pycore_uop_metadata.h | 14 +- Python/bytecodes.c | 16 +- Python/executor_cases.c.h | 34 ++- Python/optimizer.c | 21 +- Python/optimizer_cases.c.h | 10 +- Tools/cases_generator/tier2_generator.py | 14 - 7 files changed, 243 insertions(+), 184 deletions(-) diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h index 9f4ae71a313a..067da4209f01 100644 --- a/Include/internal/pycore_uop_ids.h +++ b/Include/internal/pycore_uop_ids.h @@ -137,47 +137,49 @@ extern "C" { #define _GUARD_DORV_NO_DICT 392 #define _GUARD_DORV_VALUES_INST_ATTR_FROM_DICT 393 #define _GUARD_GLOBALS_VERSION 394 -#define _GUARD_IP 395 -#define _GUARD_IS_FALSE_POP 396 -#define _GUARD_IS_NONE_POP 397 -#define _GUARD_IS_NOT_NONE_POP 398 -#define _GUARD_IS_TRUE_POP 399 -#define _GUARD_KEYS_VERSION 400 -#define _GUARD_NOS_DICT 401 -#define _GUARD_NOS_FLOAT 402 -#define _GUARD_NOS_INT 403 -#define _GUARD_NOS_LIST 404 -#define _GUARD_NOS_NOT_NULL 405 -#define _GUARD_NOS_NULL 406 -#define _GUARD_NOS_OVERFLOWED 407 -#define _GUARD_NOS_TUPLE 408 -#define _GUARD_NOS_UNICODE 409 -#define _GUARD_NOT_EXHAUSTED_LIST 410 -#define _GUARD_NOT_EXHAUSTED_RANGE 411 -#define _GUARD_NOT_EXHAUSTED_TUPLE 412 -#define _GUARD_THIRD_NULL 413 -#define _GUARD_TOS_ANY_SET 414 -#define _GUARD_TOS_DICT 415 -#define _GUARD_TOS_FLOAT 416 -#define _GUARD_TOS_INT 417 -#define _GUARD_TOS_LIST 418 -#define _GUARD_TOS_OVERFLOWED 419 -#define _GUARD_TOS_SLICE 420 -#define _GUARD_TOS_TUPLE 421 -#define _GUARD_TOS_UNICODE 422 -#define _GUARD_TYPE_VERSION 423 -#define _GUARD_TYPE_VERSION_AND_LOCK 424 -#define _HANDLE_PENDING_AND_DEOPT 425 +#define _GUARD_IP_PUSH_FRAME 395 +#define _GUARD_IP_RETURN_VALUE 396 +#define _GUARD_IP_YIELD_VALUE 397 +#define _GUARD_IS_FALSE_POP 398 +#define _GUARD_IS_NONE_POP 399 +#define _GUARD_IS_NOT_NONE_POP 400 +#define _GUARD_IS_TRUE_POP 401 +#define _GUARD_KEYS_VERSION 402 +#define _GUARD_NOS_DICT 403 +#define _GUARD_NOS_FLOAT 404 +#define _GUARD_NOS_INT 405 +#define _GUARD_NOS_LIST 406 +#define _GUARD_NOS_NOT_NULL 407 +#define _GUARD_NOS_NULL 408 +#define _GUARD_NOS_OVERFLOWED 409 +#define _GUARD_NOS_TUPLE 410 +#define _GUARD_NOS_UNICODE 411 +#define _GUARD_NOT_EXHAUSTED_LIST 412 +#define _GUARD_NOT_EXHAUSTED_RANGE 413 +#define _GUARD_NOT_EXHAUSTED_TUPLE 414 +#define _GUARD_THIRD_NULL 415 +#define _GUARD_TOS_ANY_SET 416 +#define _GUARD_TOS_DICT 417 +#define _GUARD_TOS_FLOAT 418 +#define _GUARD_TOS_INT 419 +#define _GUARD_TOS_LIST 420 +#define _GUARD_TOS_OVERFLOWED 421 +#define _GUARD_TOS_SLICE 422 +#define _GUARD_TOS_TUPLE 423 +#define _GUARD_TOS_UNICODE 424 +#define _GUARD_TYPE_VERSION 425 +#define _GUARD_TYPE_VERSION_AND_LOCK 426 +#define _HANDLE_PENDING_AND_DEOPT 427 #define _IMPORT_FROM IMPORT_FROM #define _IMPORT_NAME IMPORT_NAME -#define _INIT_CALL_BOUND_METHOD_EXACT_ARGS 426 -#define _INIT_CALL_PY_EXACT_ARGS 427 -#define _INIT_CALL_PY_EXACT_ARGS_0 428 -#define _INIT_CALL_PY_EXACT_ARGS_1 429 -#define _INIT_CALL_PY_EXACT_ARGS_2 430 -#define _INIT_CALL_PY_EXACT_ARGS_3 431 -#define _INIT_CALL_PY_EXACT_ARGS_4 432 -#define _INSERT_NULL 433 +#define _INIT_CALL_BOUND_METHOD_EXACT_ARGS 428 +#define _INIT_CALL_PY_EXACT_ARGS 429 +#define _INIT_CALL_PY_EXACT_ARGS_0 430 +#define _INIT_CALL_PY_EXACT_ARGS_1 431 +#define _INIT_CALL_PY_EXACT_ARGS_2 432 +#define _INIT_CALL_PY_EXACT_ARGS_3 433 +#define _INIT_CALL_PY_EXACT_ARGS_4 434 +#define _INSERT_NULL 435 #define _INSTRUMENTED_FOR_ITER INSTRUMENTED_FOR_ITER #define _INSTRUMENTED_INSTRUCTION INSTRUMENTED_INSTRUCTION #define _INSTRUMENTED_JUMP_FORWARD INSTRUMENTED_JUMP_FORWARD @@ -187,178 +189,178 @@ extern "C" { #define _INSTRUMENTED_POP_JUMP_IF_NONE INSTRUMENTED_POP_JUMP_IF_NONE #define _INSTRUMENTED_POP_JUMP_IF_NOT_NONE INSTRUMENTED_POP_JUMP_IF_NOT_NONE #define _INSTRUMENTED_POP_JUMP_IF_TRUE INSTRUMENTED_POP_JUMP_IF_TRUE -#define _IS_NONE 434 +#define _IS_NONE 436 #define _IS_OP IS_OP -#define _ITER_CHECK_LIST 435 -#define _ITER_CHECK_RANGE 436 -#define _ITER_CHECK_TUPLE 437 -#define _ITER_JUMP_LIST 438 -#define _ITER_JUMP_RANGE 439 -#define _ITER_JUMP_TUPLE 440 -#define _ITER_NEXT_LIST 441 -#define _ITER_NEXT_LIST_TIER_TWO 442 -#define _ITER_NEXT_RANGE 443 -#define _ITER_NEXT_TUPLE 444 +#define _ITER_CHECK_LIST 437 +#define _ITER_CHECK_RANGE 438 +#define _ITER_CHECK_TUPLE 439 +#define _ITER_JUMP_LIST 440 +#define _ITER_JUMP_RANGE 441 +#define _ITER_JUMP_TUPLE 442 +#define _ITER_NEXT_LIST 443 +#define _ITER_NEXT_LIST_TIER_TWO 444 +#define _ITER_NEXT_RANGE 445 +#define _ITER_NEXT_TUPLE 446 #define _JUMP_BACKWARD_NO_INTERRUPT JUMP_BACKWARD_NO_INTERRUPT -#define _JUMP_TO_TOP 445 +#define _JUMP_TO_TOP 447 #define _LIST_APPEND LIST_APPEND #define _LIST_EXTEND LIST_EXTEND -#define _LOAD_ATTR 446 -#define _LOAD_ATTR_CLASS 447 +#define _LOAD_ATTR 448 +#define _LOAD_ATTR_CLASS 449 #define _LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN -#define _LOAD_ATTR_INSTANCE_VALUE 448 -#define _LOAD_ATTR_METHOD_LAZY_DICT 449 -#define _LOAD_ATTR_METHOD_NO_DICT 450 -#define _LOAD_ATTR_METHOD_WITH_VALUES 451 -#define _LOAD_ATTR_MODULE 452 -#define _LOAD_ATTR_NONDESCRIPTOR_NO_DICT 453 -#define _LOAD_ATTR_NONDESCRIPTOR_WITH_VALUES 454 -#define _LOAD_ATTR_PROPERTY_FRAME 455 -#define _LOAD_ATTR_SLOT 456 -#define _LOAD_ATTR_WITH_HINT 457 +#define _LOAD_ATTR_INSTANCE_VALUE 450 +#define _LOAD_ATTR_METHOD_LAZY_DICT 451 +#define _LOAD_ATTR_METHOD_NO_DICT 452 +#define _LOAD_ATTR_METHOD_WITH_VALUES 453 +#define _LOAD_ATTR_MODULE 454 +#define _LOAD_ATTR_NONDESCRIPTOR_NO_DICT 455 +#define _LOAD_ATTR_NONDESCRIPTOR_WITH_VALUES 456 +#define _LOAD_ATTR_PROPERTY_FRAME 457 +#define _LOAD_ATTR_SLOT 458 +#define _LOAD_ATTR_WITH_HINT 459 #define _LOAD_BUILD_CLASS LOAD_BUILD_CLASS -#define _LOAD_BYTECODE 458 +#define _LOAD_BYTECODE 460 #define _LOAD_COMMON_CONSTANT LOAD_COMMON_CONSTANT #define _LOAD_CONST LOAD_CONST -#define _LOAD_CONST_INLINE 459 -#define _LOAD_CONST_INLINE_BORROW 460 -#define _LOAD_CONST_UNDER_INLINE 461 -#define _LOAD_CONST_UNDER_INLINE_BORROW 462 +#define _LOAD_CONST_INLINE 461 +#define _LOAD_CONST_INLINE_BORROW 462 +#define _LOAD_CONST_UNDER_INLINE 463 +#define _LOAD_CONST_UNDER_INLINE_BORROW 464 #define _LOAD_DEREF LOAD_DEREF -#define _LOAD_FAST 463 -#define _LOAD_FAST_0 464 -#define _LOAD_FAST_1 465 -#define _LOAD_FAST_2 466 -#define _LOAD_FAST_3 467 -#define _LOAD_FAST_4 468 -#define _LOAD_FAST_5 469 -#define _LOAD_FAST_6 470 -#define _LOAD_FAST_7 471 +#define _LOAD_FAST 465 +#define _LOAD_FAST_0 466 +#define _LOAD_FAST_1 467 +#define _LOAD_FAST_2 468 +#define _LOAD_FAST_3 469 +#define _LOAD_FAST_4 470 +#define _LOAD_FAST_5 471 +#define _LOAD_FAST_6 472 +#define _LOAD_FAST_7 473 #define _LOAD_FAST_AND_CLEAR LOAD_FAST_AND_CLEAR -#define _LOAD_FAST_BORROW 472 -#define _LOAD_FAST_BORROW_0 473 -#define _LOAD_FAST_BORROW_1 474 -#define _LOAD_FAST_BORROW_2 475 -#define _LOAD_FAST_BORROW_3 476 -#define _LOAD_FAST_BORROW_4 477 -#define _LOAD_FAST_BORROW_5 478 -#define _LOAD_FAST_BORROW_6 479 -#define _LOAD_FAST_BORROW_7 480 +#define _LOAD_FAST_BORROW 474 +#define _LOAD_FAST_BORROW_0 475 +#define _LOAD_FAST_BORROW_1 476 +#define _LOAD_FAST_BORROW_2 477 +#define _LOAD_FAST_BORROW_3 478 +#define _LOAD_FAST_BORROW_4 479 +#define _LOAD_FAST_BORROW_5 480 +#define _LOAD_FAST_BORROW_6 481 +#define _LOAD_FAST_BORROW_7 482 #define _LOAD_FAST_BORROW_LOAD_FAST_BORROW LOAD_FAST_BORROW_LOAD_FAST_BORROW #define _LOAD_FAST_CHECK LOAD_FAST_CHECK #define _LOAD_FAST_LOAD_FAST LOAD_FAST_LOAD_FAST #define _LOAD_FROM_DICT_OR_DEREF LOAD_FROM_DICT_OR_DEREF #define _LOAD_FROM_DICT_OR_GLOBALS LOAD_FROM_DICT_OR_GLOBALS -#define _LOAD_GLOBAL 481 -#define _LOAD_GLOBAL_BUILTINS 482 -#define _LOAD_GLOBAL_MODULE 483 +#define _LOAD_GLOBAL 483 +#define _LOAD_GLOBAL_BUILTINS 484 +#define _LOAD_GLOBAL_MODULE 485 #define _LOAD_LOCALS LOAD_LOCALS #define _LOAD_NAME LOAD_NAME -#define _LOAD_SMALL_INT 484 -#define _LOAD_SMALL_INT_0 485 -#define _LOAD_SMALL_INT_1 486 -#define _LOAD_SMALL_INT_2 487 -#define _LOAD_SMALL_INT_3 488 -#define _LOAD_SPECIAL 489 +#define _LOAD_SMALL_INT 486 +#define _LOAD_SMALL_INT_0 487 +#define _LOAD_SMALL_INT_1 488 +#define _LOAD_SMALL_INT_2 489 +#define _LOAD_SMALL_INT_3 490 +#define _LOAD_SPECIAL 491 #define _LOAD_SUPER_ATTR_ATTR LOAD_SUPER_ATTR_ATTR #define _LOAD_SUPER_ATTR_METHOD LOAD_SUPER_ATTR_METHOD -#define _MAKE_CALLARGS_A_TUPLE 490 +#define _MAKE_CALLARGS_A_TUPLE 492 #define _MAKE_CELL MAKE_CELL #define _MAKE_FUNCTION MAKE_FUNCTION -#define _MAKE_WARM 491 +#define _MAKE_WARM 493 #define _MAP_ADD MAP_ADD #define _MATCH_CLASS MATCH_CLASS #define _MATCH_KEYS MATCH_KEYS #define _MATCH_MAPPING MATCH_MAPPING #define _MATCH_SEQUENCE MATCH_SEQUENCE -#define _MAYBE_EXPAND_METHOD 492 -#define _MAYBE_EXPAND_METHOD_KW 493 -#define _MONITOR_CALL 494 -#define _MONITOR_CALL_KW 495 -#define _MONITOR_JUMP_BACKWARD 496 -#define _MONITOR_RESUME 497 +#define _MAYBE_EXPAND_METHOD 494 +#define _MAYBE_EXPAND_METHOD_KW 495 +#define _MONITOR_CALL 496 +#define _MONITOR_CALL_KW 497 +#define _MONITOR_JUMP_BACKWARD 498 +#define _MONITOR_RESUME 499 #define _NOP NOP -#define _POP_CALL 498 -#define _POP_CALL_LOAD_CONST_INLINE_BORROW 499 -#define _POP_CALL_ONE 500 -#define _POP_CALL_ONE_LOAD_CONST_INLINE_BORROW 501 -#define _POP_CALL_TWO 502 -#define _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW 503 +#define _POP_CALL 500 +#define _POP_CALL_LOAD_CONST_INLINE_BORROW 501 +#define _POP_CALL_ONE 502 +#define _POP_CALL_ONE_LOAD_CONST_INLINE_BORROW 503 +#define _POP_CALL_TWO 504 +#define _POP_CALL_TWO_LOAD_CONST_INLINE_BORROW 505 #define _POP_EXCEPT POP_EXCEPT #define _POP_ITER POP_ITER -#define _POP_JUMP_IF_FALSE 504 -#define _POP_JUMP_IF_TRUE 505 +#define _POP_JUMP_IF_FALSE 506 +#define _POP_JUMP_IF_TRUE 507 #define _POP_TOP POP_TOP -#define _POP_TOP_FLOAT 506 -#define _POP_TOP_INT 507 -#define _POP_TOP_LOAD_CONST_INLINE 508 -#define _POP_TOP_LOAD_CONST_INLINE_BORROW 509 -#define _POP_TOP_NOP 510 -#define _POP_TOP_UNICODE 511 -#define _POP_TWO 512 -#define _POP_TWO_LOAD_CONST_INLINE_BORROW 513 +#define _POP_TOP_FLOAT 508 +#define _POP_TOP_INT 509 +#define _POP_TOP_LOAD_CONST_INLINE 510 +#define _POP_TOP_LOAD_CONST_INLINE_BORROW 511 +#define _POP_TOP_NOP 512 +#define _POP_TOP_UNICODE 513 +#define _POP_TWO 514 +#define _POP_TWO_LOAD_CONST_INLINE_BORROW 515 #define _PUSH_EXC_INFO PUSH_EXC_INFO -#define _PUSH_FRAME 514 +#define _PUSH_FRAME 516 #define _PUSH_NULL PUSH_NULL -#define _PUSH_NULL_CONDITIONAL 515 -#define _PY_FRAME_GENERAL 516 -#define _PY_FRAME_KW 517 -#define _QUICKEN_RESUME 518 -#define _REPLACE_WITH_TRUE 519 +#define _PUSH_NULL_CONDITIONAL 517 +#define _PY_FRAME_GENERAL 518 +#define _PY_FRAME_KW 519 +#define _QUICKEN_RESUME 520 +#define _REPLACE_WITH_TRUE 521 #define _RESUME_CHECK RESUME_CHECK #define _RETURN_GENERATOR RETURN_GENERATOR #define _RETURN_VALUE RETURN_VALUE -#define _SAVE_RETURN_OFFSET 520 -#define _SEND 521 -#define _SEND_GEN_FRAME 522 +#define _SAVE_RETURN_OFFSET 522 +#define _SEND 523 +#define _SEND_GEN_FRAME 524 #define _SETUP_ANNOTATIONS SETUP_ANNOTATIONS #define _SET_ADD SET_ADD #define _SET_FUNCTION_ATTRIBUTE SET_FUNCTION_ATTRIBUTE #define _SET_UPDATE SET_UPDATE -#define _START_EXECUTOR 523 -#define _STORE_ATTR 524 -#define _STORE_ATTR_INSTANCE_VALUE 525 -#define _STORE_ATTR_SLOT 526 -#define _STORE_ATTR_WITH_HINT 527 +#define _START_EXECUTOR 525 +#define _STORE_ATTR 526 +#define _STORE_ATTR_INSTANCE_VALUE 527 +#define _STORE_ATTR_SLOT 528 +#define _STORE_ATTR_WITH_HINT 529 #define _STORE_DEREF STORE_DEREF -#define _STORE_FAST 528 -#define _STORE_FAST_0 529 -#define _STORE_FAST_1 530 -#define _STORE_FAST_2 531 -#define _STORE_FAST_3 532 -#define _STORE_FAST_4 533 -#define _STORE_FAST_5 534 -#define _STORE_FAST_6 535 -#define _STORE_FAST_7 536 +#define _STORE_FAST 530 +#define _STORE_FAST_0 531 +#define _STORE_FAST_1 532 +#define _STORE_FAST_2 533 +#define _STORE_FAST_3 534 +#define _STORE_FAST_4 535 +#define _STORE_FAST_5 536 +#define _STORE_FAST_6 537 +#define _STORE_FAST_7 538 #define _STORE_FAST_LOAD_FAST STORE_FAST_LOAD_FAST #define _STORE_FAST_STORE_FAST STORE_FAST_STORE_FAST #define _STORE_GLOBAL STORE_GLOBAL #define _STORE_NAME STORE_NAME -#define _STORE_SLICE 537 -#define _STORE_SUBSCR 538 -#define _STORE_SUBSCR_DICT 539 -#define _STORE_SUBSCR_LIST_INT 540 -#define _SWAP 541 -#define _SWAP_2 542 -#define _SWAP_3 543 -#define _TIER2_RESUME_CHECK 544 -#define _TO_BOOL 545 +#define _STORE_SLICE 539 +#define _STORE_SUBSCR 540 +#define _STORE_SUBSCR_DICT 541 +#define _STORE_SUBSCR_LIST_INT 542 +#define _SWAP 543 +#define _SWAP_2 544 +#define _SWAP_3 545 +#define _TIER2_RESUME_CHECK 546 +#define _TO_BOOL 547 #define _TO_BOOL_BOOL TO_BOOL_BOOL #define _TO_BOOL_INT TO_BOOL_INT -#define _TO_BOOL_LIST 546 +#define _TO_BOOL_LIST 548 #define _TO_BOOL_NONE TO_BOOL_NONE -#define _TO_BOOL_STR 547 +#define _TO_BOOL_STR 549 #define _UNARY_INVERT UNARY_INVERT #define _UNARY_NEGATIVE UNARY_NEGATIVE #define _UNARY_NOT UNARY_NOT #define _UNPACK_EX UNPACK_EX -#define _UNPACK_SEQUENCE 548 -#define _UNPACK_SEQUENCE_LIST 549 -#define _UNPACK_SEQUENCE_TUPLE 550 -#define _UNPACK_SEQUENCE_TWO_TUPLE 551 +#define _UNPACK_SEQUENCE 550 +#define _UNPACK_SEQUENCE_LIST 551 +#define _UNPACK_SEQUENCE_TUPLE 552 +#define _UNPACK_SEQUENCE_TWO_TUPLE 553 #define _WITH_EXCEPT_START WITH_EXCEPT_START #define _YIELD_VALUE YIELD_VALUE -#define MAX_UOP_ID 551 +#define MAX_UOP_ID 553 #ifdef __cplusplus } diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h index 82a0fa926525..67c5827ffd0c 100644 --- a/Include/internal/pycore_uop_metadata.h +++ b/Include/internal/pycore_uop_metadata.h @@ -337,7 +337,9 @@ const uint16_t _PyUop_Flags[MAX_UOP_ID+1] = { [_ERROR_POP_N] = HAS_ARG_FLAG, [_TIER2_RESUME_CHECK] = HAS_PERIODIC_FLAG, [_COLD_EXIT] = 0, - [_GUARD_IP] = HAS_EXIT_FLAG, + [_GUARD_IP_PUSH_FRAME] = HAS_EXIT_FLAG, + [_GUARD_IP_YIELD_VALUE] = HAS_EXIT_FLAG, + [_GUARD_IP_RETURN_VALUE] = HAS_EXIT_FLAG, [_DYNAMIC_EXIT] = HAS_ESCAPES_FLAG, }; @@ -474,7 +476,9 @@ const char *const _PyOpcode_uop_name[MAX_UOP_ID+1] = { [_GUARD_DORV_NO_DICT] = "_GUARD_DORV_NO_DICT", [_GUARD_DORV_VALUES_INST_ATTR_FROM_DICT] = "_GUARD_DORV_VALUES_INST_ATTR_FROM_DICT", [_GUARD_GLOBALS_VERSION] = "_GUARD_GLOBALS_VERSION", - [_GUARD_IP] = "_GUARD_IP", + [_GUARD_IP_PUSH_FRAME] = "_GUARD_IP_PUSH_FRAME", + [_GUARD_IP_RETURN_VALUE] = "_GUARD_IP_RETURN_VALUE", + [_GUARD_IP_YIELD_VALUE] = "_GUARD_IP_YIELD_VALUE", [_GUARD_IS_FALSE_POP] = "_GUARD_IS_FALSE_POP", [_GUARD_IS_NONE_POP] = "_GUARD_IS_NONE_POP", [_GUARD_IS_NOT_NONE_POP] = "_GUARD_IS_NOT_NONE_POP", @@ -1309,7 +1313,11 @@ int _PyUop_num_popped(int opcode, int oparg) return 0; case _COLD_EXIT: return 0; - case _GUARD_IP: + case _GUARD_IP_PUSH_FRAME: + return 0; + case _GUARD_IP_YIELD_VALUE: + return 0; + case _GUARD_IP_RETURN_VALUE: return 0; case _DYNAMIC_EXIT: return 0; diff --git a/Python/bytecodes.c b/Python/bytecodes.c index 971660ace56a..b3977e5613b4 100644 --- a/Python/bytecodes.c +++ b/Python/bytecodes.c @@ -5443,10 +5443,24 @@ dummy_func( TIER2_TO_TIER2(exit->executor); } - tier2 op(_GUARD_IP, (ip/4 --)) { + tier2 op(_GUARD_IP_PUSH_FRAME, (ip/4 --)) { EXIT_IF(frame->instr_ptr != (_Py_CODEUNIT *)ip); } + tier2 op(_GUARD_IP_YIELD_VALUE, (ip/4 --)) { + if (frame->instr_ptr + 1 + INLINE_CACHE_ENTRIES_SEND != (_Py_CODEUNIT *)ip) { + frame->instr_ptr += 1 + INLINE_CACHE_ENTRIES_SEND; + EXIT_IF(true); + } + } + + tier2 op(_GUARD_IP_RETURN_VALUE, (ip/4 --)) { + if (frame->instr_ptr + frame->return_offset != (_Py_CODEUNIT *)ip) { + frame->instr_ptr += frame->return_offset; + EXIT_IF(true); + } + } + // Note: this is different than _COLD_EXIT/_EXIT_TRACE, as it may lead to multiple executors // from a single exit! tier2 op(_DYNAMIC_EXIT, (exit_p/4 --)) { diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h index 5273670008c0..7f915e31b121 100644 --- a/Python/executor_cases.c.h +++ b/Python/executor_cases.c.h @@ -1930,7 +1930,7 @@ frame = tstate->current_frame = dying->previous; _PyEval_FrameClearAndPop(tstate, dying); stack_pointer = _PyFrame_GetStackPointer(frame); - frame->instr_ptr += (frame->return_offset); + LOAD_IP(frame->return_offset); res = temp; LLTRACE_RESUME_FRAME(); stack_pointer[0] = res; @@ -2096,7 +2096,7 @@ _PyOpcode_Deopt[frame->instr_ptr->op.code] == ENTER_EXECUTOR); #endif stack_pointer = _PyFrame_GetStackPointer(frame); - frame->instr_ptr += (1 + INLINE_CACHE_ENTRIES_SEND); + LOAD_IP(1 + INLINE_CACHE_ENTRIES_SEND); value = PyStackRef_MakeHeapSafe(temp); LLTRACE_RESUME_FRAME(); stack_pointer[0] = value; @@ -5347,7 +5347,7 @@ frame = tstate->current_frame = temp; tstate->py_recursion_remaining--; LOAD_SP(); - frame->instr_ptr += (0); + LOAD_IP(0); LLTRACE_RESUME_FRAME(); break; } @@ -6752,7 +6752,7 @@ _PyInterpreterFrame *prev = frame->previous; _PyThreadState_PopFrame(tstate, frame); frame = tstate->current_frame = prev; - frame->instr_ptr += (frame->return_offset); + LOAD_IP(frame->return_offset); stack_pointer = _PyFrame_GetStackPointer(frame); res = PyStackRef_FromPyObjectStealMortal((PyObject *)gen); LLTRACE_RESUME_FRAME(); @@ -7493,7 +7493,7 @@ break; } - case _GUARD_IP: { + case _GUARD_IP_PUSH_FRAME: { PyObject *ip = (PyObject *)CURRENT_OPERAND0(); if (frame->instr_ptr != (_Py_CODEUNIT *)ip) { UOP_STAT_INC(uopcode, miss); @@ -7502,6 +7502,30 @@ break; } + case _GUARD_IP_YIELD_VALUE: { + PyObject *ip = (PyObject *)CURRENT_OPERAND0(); + if (frame->instr_ptr + 1 + INLINE_CACHE_ENTRIES_SEND != (_Py_CODEUNIT *)ip) { + frame->instr_ptr += 1 + INLINE_CACHE_ENTRIES_SEND; + if (true) { + UOP_STAT_INC(uopcode, miss); + JUMP_TO_JUMP_TARGET(); + } + } + break; + } + + case _GUARD_IP_RETURN_VALUE: { + PyObject *ip = (PyObject *)CURRENT_OPERAND0(); + if (frame->instr_ptr + frame->return_offset != (_Py_CODEUNIT *)ip) { + frame->instr_ptr += frame->return_offset; + if (true) { + UOP_STAT_INC(uopcode, miss); + JUMP_TO_JUMP_TARGET(); + } + } + break; + } + case _DYNAMIC_EXIT: { PyObject *exit_p = (PyObject *)CURRENT_OPERAND0(); _Py_CODEUNIT *target = frame->instr_ptr; diff --git a/Python/optimizer.c b/Python/optimizer.c index 12bb803d18d8..e6955edbc194 100644 --- a/Python/optimizer.c +++ b/Python/optimizer.c @@ -901,7 +901,21 @@ _PyJit_translate_single_bytecode_to_trace( } // End switch (opcode) if (needs_guard_ip) { - ADD_TO_TRACE(_GUARD_IP, 0, (uintptr_t)next_instr, 0); + switch (trace[trace_length-1].opcode) { + case _PUSH_FRAME: + ADD_TO_TRACE(_GUARD_IP_PUSH_FRAME, 0, (uintptr_t)next_instr, 0); + break; + case _RETURN_GENERATOR: + case _RETURN_VALUE: + ADD_TO_TRACE(_GUARD_IP_RETURN_VALUE, 0, (uintptr_t)next_instr, 0); + break; + case _YIELD_VALUE: + ADD_TO_TRACE(_GUARD_IP_YIELD_VALUE, 0, (uintptr_t)next_instr, 0); + break; + default: + DPRINTF(1, "Unknown uop needing guard ip %s\n", _PyOpcode_uop_name[trace[trace_length-1].opcode]); + Py_UNREACHABLE(); + } } // Loop back to the start int is_first_instr = tstate->interp->jit_state.close_loop_instr == next_instr || tstate->interp->jit_state.insert_exec_instr == next_instr; @@ -1074,7 +1088,10 @@ prepare_for_execution(_PyUOpInstruction *buffer, int length) if (opcode == _FOR_ITER_TIER_TWO) { exit_op = _DYNAMIC_EXIT; } - else if (opcode == _GUARD_IP) { + else if ( + opcode == _GUARD_IP_PUSH_FRAME || + opcode == _GUARD_IP_RETURN_VALUE || + opcode == _GUARD_IP_YIELD_VALUE) { exit_op = _DYNAMIC_EXIT; unique_target = true; } diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h index 5f7d600b3cd2..9fc77dd4b6ed 100644 --- a/Python/optimizer_cases.c.h +++ b/Python/optimizer_cases.c.h @@ -3426,7 +3426,15 @@ break; } - case _GUARD_IP: { + case _GUARD_IP_PUSH_FRAME: { + break; + } + + case _GUARD_IP_YIELD_VALUE: { + break; + } + + case _GUARD_IP_RETURN_VALUE: { break; } diff --git a/Tools/cases_generator/tier2_generator.py b/Tools/cases_generator/tier2_generator.py index e95a070b163f..4bf04154df16 100644 --- a/Tools/cases_generator/tier2_generator.py +++ b/Tools/cases_generator/tier2_generator.py @@ -65,7 +65,6 @@ class Tier2Emitter(Emitter): self._replacers["oparg"] = self.oparg self._replacers["JUMPBY"] = self.jumpby self._replacers["DISPATCH"] = self.dispatch - self._replacers["LOAD_IP"] = self.load_ip def goto_error(self, offset: int, storage: Storage) -> str: # To do: Add jump targets for popping values. @@ -170,19 +169,6 @@ class Tier2Emitter(Emitter): next(tkn_iter) return False - def load_ip( - self, - tkn: Token, - tkn_iter: TokenIterator, - uop: CodeSection, - storage: Storage, - inst: Instruction | None, - ) -> bool: - self.out.start_line() - self.emit("frame->instr_ptr += ") - emit_to(self.out, tkn_iter, "SEMI") - self.emit(";\n") - return True def write_uop(uop: Uop, emitter: Emitter, stack: Stack) -> Stack: locals: dict[str, Local] = {} -- 2.47.3