From 6293d00e7201f3f28b1f4512e57dc4f03855cabd Mon Sep 17 00:00:00 2001 From: Ken Jin Date: Sat, 9 Nov 2024 11:35:33 +0800 Subject: [PATCH] gh-120619: Strength reduce function guards, support 2-operand uop forms (GH-124846) Co-authored-by: Brandt Bucher --- Include/internal/pycore_optimizer.h | 3 +- Include/internal/pycore_uop_ids.h | 303 ++++++++++--------- Include/internal/pycore_uop_metadata.h | 4 + Lib/test/test_capi/test_opt.py | 24 ++ Python/bytecodes.c | 8 +- Python/ceval_macros.h | 3 +- Python/executor_cases.c.h | 108 ++++--- Python/generated_cases.c.h | 2 +- Python/optimizer.c | 22 +- Python/optimizer_analysis.c | 28 +- Python/optimizer_bytecodes.c | 23 +- Python/optimizer_cases.c.h | 69 +++-- Tools/cases_generator/analyzer.py | 2 +- Tools/cases_generator/optimizer_generator.py | 2 +- Tools/cases_generator/tier2_generator.py | 4 +- Tools/jit/_stencils.py | 24 +- Tools/jit/template.c | 20 +- 17 files changed, 379 insertions(+), 270 deletions(-) diff --git a/Include/internal/pycore_optimizer.h b/Include/internal/pycore_optimizer.h index f92c0a0cddf..6d70b42f708 100644 --- a/Include/internal/pycore_optimizer.h +++ b/Include/internal/pycore_optimizer.h @@ -58,7 +58,8 @@ typedef struct { uint16_t error_target; }; }; - uint64_t operand; // A cache entry + uint64_t operand0; // A cache entry + uint64_t operand1; } _PyUOpInstruction; typedef struct { diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h index 55416d2aae1..fab4ce6a25b 100644 --- a/Include/internal/pycore_uop_ids.h +++ b/Include/internal/pycore_uop_ids.h @@ -63,90 +63,91 @@ extern "C" { #define _CHECK_FUNCTION 333 #define _CHECK_FUNCTION_EXACT_ARGS 334 #define _CHECK_FUNCTION_VERSION 335 -#define _CHECK_FUNCTION_VERSION_KW 336 -#define _CHECK_IS_NOT_PY_CALLABLE 337 -#define _CHECK_IS_NOT_PY_CALLABLE_KW 338 -#define _CHECK_MANAGED_OBJECT_HAS_VALUES 339 -#define _CHECK_METHOD_VERSION 340 -#define _CHECK_METHOD_VERSION_KW 341 -#define _CHECK_PEP_523 342 -#define _CHECK_PERIODIC 343 -#define _CHECK_PERIODIC_IF_NOT_YIELD_FROM 344 -#define _CHECK_STACK_SPACE 345 -#define _CHECK_STACK_SPACE_OPERAND 346 -#define _CHECK_VALIDITY 347 -#define _CHECK_VALIDITY_AND_SET_IP 348 -#define _COMPARE_OP 349 -#define _COMPARE_OP_FLOAT 350 -#define _COMPARE_OP_INT 351 -#define _COMPARE_OP_STR 352 -#define _CONTAINS_OP 353 +#define _CHECK_FUNCTION_VERSION_INLINE 336 +#define _CHECK_FUNCTION_VERSION_KW 337 +#define _CHECK_IS_NOT_PY_CALLABLE 338 +#define _CHECK_IS_NOT_PY_CALLABLE_KW 339 +#define _CHECK_MANAGED_OBJECT_HAS_VALUES 340 +#define _CHECK_METHOD_VERSION 341 +#define _CHECK_METHOD_VERSION_KW 342 +#define _CHECK_PEP_523 343 +#define _CHECK_PERIODIC 344 +#define _CHECK_PERIODIC_IF_NOT_YIELD_FROM 345 +#define _CHECK_STACK_SPACE 346 +#define _CHECK_STACK_SPACE_OPERAND 347 +#define _CHECK_VALIDITY 348 +#define _CHECK_VALIDITY_AND_SET_IP 349 +#define _COMPARE_OP 350 +#define _COMPARE_OP_FLOAT 351 +#define _COMPARE_OP_INT 352 +#define _COMPARE_OP_STR 353 +#define _CONTAINS_OP 354 #define _CONTAINS_OP_DICT CONTAINS_OP_DICT #define _CONTAINS_OP_SET CONTAINS_OP_SET #define _CONVERT_VALUE CONVERT_VALUE #define _COPY COPY #define _COPY_FREE_VARS COPY_FREE_VARS -#define _CREATE_INIT_FRAME 354 +#define _CREATE_INIT_FRAME 355 #define _DELETE_ATTR DELETE_ATTR #define _DELETE_DEREF DELETE_DEREF #define _DELETE_FAST DELETE_FAST #define _DELETE_GLOBAL DELETE_GLOBAL #define _DELETE_NAME DELETE_NAME #define _DELETE_SUBSCR DELETE_SUBSCR -#define _DEOPT 355 +#define _DEOPT 356 #define _DICT_MERGE DICT_MERGE #define _DICT_UPDATE DICT_UPDATE -#define _DO_CALL 356 -#define _DO_CALL_FUNCTION_EX 357 -#define _DO_CALL_KW 358 -#define _DYNAMIC_EXIT 359 +#define _DO_CALL 357 +#define _DO_CALL_FUNCTION_EX 358 +#define _DO_CALL_KW 359 +#define _DYNAMIC_EXIT 360 #define _END_SEND END_SEND -#define _ERROR_POP_N 360 +#define _ERROR_POP_N 361 #define _EXIT_INIT_CHECK EXIT_INIT_CHECK -#define _EXPAND_METHOD 361 -#define _EXPAND_METHOD_KW 362 -#define _FATAL_ERROR 363 +#define _EXPAND_METHOD 362 +#define _EXPAND_METHOD_KW 363 +#define _FATAL_ERROR 364 #define _FORMAT_SIMPLE FORMAT_SIMPLE #define _FORMAT_WITH_SPEC FORMAT_WITH_SPEC -#define _FOR_ITER 364 -#define _FOR_ITER_GEN_FRAME 365 -#define _FOR_ITER_TIER_TWO 366 +#define _FOR_ITER 365 +#define _FOR_ITER_GEN_FRAME 366 +#define _FOR_ITER_TIER_TWO 367 #define _GET_AITER GET_AITER #define _GET_ANEXT GET_ANEXT #define _GET_AWAITABLE GET_AWAITABLE #define _GET_ITER GET_ITER #define _GET_LEN GET_LEN #define _GET_YIELD_FROM_ITER GET_YIELD_FROM_ITER -#define _GUARD_BOTH_FLOAT 367 -#define _GUARD_BOTH_INT 368 -#define _GUARD_BOTH_UNICODE 369 -#define _GUARD_BUILTINS_VERSION_PUSH_KEYS 370 -#define _GUARD_DORV_NO_DICT 371 -#define _GUARD_DORV_VALUES_INST_ATTR_FROM_DICT 372 -#define _GUARD_GLOBALS_VERSION 373 -#define _GUARD_GLOBALS_VERSION_PUSH_KEYS 374 -#define _GUARD_IS_FALSE_POP 375 -#define _GUARD_IS_NONE_POP 376 -#define _GUARD_IS_NOT_NONE_POP 377 -#define _GUARD_IS_TRUE_POP 378 -#define _GUARD_KEYS_VERSION 379 -#define _GUARD_NOS_FLOAT 380 -#define _GUARD_NOS_INT 381 -#define _GUARD_NOT_EXHAUSTED_LIST 382 -#define _GUARD_NOT_EXHAUSTED_RANGE 383 -#define _GUARD_NOT_EXHAUSTED_TUPLE 384 -#define _GUARD_TOS_FLOAT 385 -#define _GUARD_TOS_INT 386 -#define _GUARD_TYPE_VERSION 387 +#define _GUARD_BOTH_FLOAT 368 +#define _GUARD_BOTH_INT 369 +#define _GUARD_BOTH_UNICODE 370 +#define _GUARD_BUILTINS_VERSION_PUSH_KEYS 371 +#define _GUARD_DORV_NO_DICT 372 +#define _GUARD_DORV_VALUES_INST_ATTR_FROM_DICT 373 +#define _GUARD_GLOBALS_VERSION 374 +#define _GUARD_GLOBALS_VERSION_PUSH_KEYS 375 +#define _GUARD_IS_FALSE_POP 376 +#define _GUARD_IS_NONE_POP 377 +#define _GUARD_IS_NOT_NONE_POP 378 +#define _GUARD_IS_TRUE_POP 379 +#define _GUARD_KEYS_VERSION 380 +#define _GUARD_NOS_FLOAT 381 +#define _GUARD_NOS_INT 382 +#define _GUARD_NOT_EXHAUSTED_LIST 383 +#define _GUARD_NOT_EXHAUSTED_RANGE 384 +#define _GUARD_NOT_EXHAUSTED_TUPLE 385 +#define _GUARD_TOS_FLOAT 386 +#define _GUARD_TOS_INT 387 +#define _GUARD_TYPE_VERSION 388 #define _IMPORT_FROM IMPORT_FROM #define _IMPORT_NAME IMPORT_NAME -#define _INIT_CALL_BOUND_METHOD_EXACT_ARGS 388 -#define _INIT_CALL_PY_EXACT_ARGS 389 -#define _INIT_CALL_PY_EXACT_ARGS_0 390 -#define _INIT_CALL_PY_EXACT_ARGS_1 391 -#define _INIT_CALL_PY_EXACT_ARGS_2 392 -#define _INIT_CALL_PY_EXACT_ARGS_3 393 -#define _INIT_CALL_PY_EXACT_ARGS_4 394 +#define _INIT_CALL_BOUND_METHOD_EXACT_ARGS 389 +#define _INIT_CALL_PY_EXACT_ARGS 390 +#define _INIT_CALL_PY_EXACT_ARGS_0 391 +#define _INIT_CALL_PY_EXACT_ARGS_1 392 +#define _INIT_CALL_PY_EXACT_ARGS_2 393 +#define _INIT_CALL_PY_EXACT_ARGS_3 394 +#define _INIT_CALL_PY_EXACT_ARGS_4 395 #define _INSTRUMENTED_CALL_FUNCTION_EX INSTRUMENTED_CALL_FUNCTION_EX #define _INSTRUMENTED_CALL_KW INSTRUMENTED_CALL_KW #define _INSTRUMENTED_FOR_ITER INSTRUMENTED_FOR_ITER @@ -158,142 +159,142 @@ 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 _INTERNAL_INCREMENT_OPT_COUNTER 395 -#define _IS_NONE 396 +#define _INTERNAL_INCREMENT_OPT_COUNTER 396 +#define _IS_NONE 397 #define _IS_OP IS_OP -#define _ITER_CHECK_LIST 397 -#define _ITER_CHECK_RANGE 398 -#define _ITER_CHECK_TUPLE 399 -#define _ITER_JUMP_LIST 400 -#define _ITER_JUMP_RANGE 401 -#define _ITER_JUMP_TUPLE 402 -#define _ITER_NEXT_LIST 403 -#define _ITER_NEXT_RANGE 404 -#define _ITER_NEXT_TUPLE 405 -#define _JUMP_TO_TOP 406 +#define _ITER_CHECK_LIST 398 +#define _ITER_CHECK_RANGE 399 +#define _ITER_CHECK_TUPLE 400 +#define _ITER_JUMP_LIST 401 +#define _ITER_JUMP_RANGE 402 +#define _ITER_JUMP_TUPLE 403 +#define _ITER_NEXT_LIST 404 +#define _ITER_NEXT_RANGE 405 +#define _ITER_NEXT_TUPLE 406 +#define _JUMP_TO_TOP 407 #define _LIST_APPEND LIST_APPEND #define _LIST_EXTEND LIST_EXTEND -#define _LOAD_ATTR 407 -#define _LOAD_ATTR_CLASS 408 -#define _LOAD_ATTR_CLASS_0 409 -#define _LOAD_ATTR_CLASS_1 410 +#define _LOAD_ATTR 408 +#define _LOAD_ATTR_CLASS 409 +#define _LOAD_ATTR_CLASS_0 410 +#define _LOAD_ATTR_CLASS_1 411 #define _LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN -#define _LOAD_ATTR_INSTANCE_VALUE 411 -#define _LOAD_ATTR_INSTANCE_VALUE_0 412 -#define _LOAD_ATTR_INSTANCE_VALUE_1 413 -#define _LOAD_ATTR_METHOD_LAZY_DICT 414 -#define _LOAD_ATTR_METHOD_NO_DICT 415 -#define _LOAD_ATTR_METHOD_WITH_VALUES 416 -#define _LOAD_ATTR_MODULE 417 -#define _LOAD_ATTR_NONDESCRIPTOR_NO_DICT 418 -#define _LOAD_ATTR_NONDESCRIPTOR_WITH_VALUES 419 -#define _LOAD_ATTR_PROPERTY_FRAME 420 -#define _LOAD_ATTR_SLOT 421 -#define _LOAD_ATTR_SLOT_0 422 -#define _LOAD_ATTR_SLOT_1 423 -#define _LOAD_ATTR_WITH_HINT 424 +#define _LOAD_ATTR_INSTANCE_VALUE 412 +#define _LOAD_ATTR_INSTANCE_VALUE_0 413 +#define _LOAD_ATTR_INSTANCE_VALUE_1 414 +#define _LOAD_ATTR_METHOD_LAZY_DICT 415 +#define _LOAD_ATTR_METHOD_NO_DICT 416 +#define _LOAD_ATTR_METHOD_WITH_VALUES 417 +#define _LOAD_ATTR_MODULE 418 +#define _LOAD_ATTR_NONDESCRIPTOR_NO_DICT 419 +#define _LOAD_ATTR_NONDESCRIPTOR_WITH_VALUES 420 +#define _LOAD_ATTR_PROPERTY_FRAME 421 +#define _LOAD_ATTR_SLOT 422 +#define _LOAD_ATTR_SLOT_0 423 +#define _LOAD_ATTR_SLOT_1 424 +#define _LOAD_ATTR_WITH_HINT 425 #define _LOAD_BUILD_CLASS LOAD_BUILD_CLASS -#define _LOAD_BYTECODE 425 +#define _LOAD_BYTECODE 426 #define _LOAD_COMMON_CONSTANT LOAD_COMMON_CONSTANT #define _LOAD_CONST LOAD_CONST #define _LOAD_CONST_IMMORTAL LOAD_CONST_IMMORTAL -#define _LOAD_CONST_INLINE 426 -#define _LOAD_CONST_INLINE_BORROW 427 -#define _LOAD_CONST_INLINE_BORROW_WITH_NULL 428 -#define _LOAD_CONST_INLINE_WITH_NULL 429 +#define _LOAD_CONST_INLINE 427 +#define _LOAD_CONST_INLINE_BORROW 428 +#define _LOAD_CONST_INLINE_BORROW_WITH_NULL 429 +#define _LOAD_CONST_INLINE_WITH_NULL 430 #define _LOAD_DEREF LOAD_DEREF -#define _LOAD_FAST 430 -#define _LOAD_FAST_0 431 -#define _LOAD_FAST_1 432 -#define _LOAD_FAST_2 433 -#define _LOAD_FAST_3 434 -#define _LOAD_FAST_4 435 -#define _LOAD_FAST_5 436 -#define _LOAD_FAST_6 437 -#define _LOAD_FAST_7 438 +#define _LOAD_FAST 431 +#define _LOAD_FAST_0 432 +#define _LOAD_FAST_1 433 +#define _LOAD_FAST_2 434 +#define _LOAD_FAST_3 435 +#define _LOAD_FAST_4 436 +#define _LOAD_FAST_5 437 +#define _LOAD_FAST_6 438 +#define _LOAD_FAST_7 439 #define _LOAD_FAST_AND_CLEAR LOAD_FAST_AND_CLEAR #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 439 -#define _LOAD_GLOBAL_BUILTINS 440 -#define _LOAD_GLOBAL_BUILTINS_FROM_KEYS 441 -#define _LOAD_GLOBAL_MODULE 442 -#define _LOAD_GLOBAL_MODULE_FROM_KEYS 443 +#define _LOAD_GLOBAL 440 +#define _LOAD_GLOBAL_BUILTINS 441 +#define _LOAD_GLOBAL_BUILTINS_FROM_KEYS 442 +#define _LOAD_GLOBAL_MODULE 443 +#define _LOAD_GLOBAL_MODULE_FROM_KEYS 444 #define _LOAD_LOCALS LOAD_LOCALS #define _LOAD_NAME LOAD_NAME -#define _LOAD_SMALL_INT 444 -#define _LOAD_SMALL_INT_0 445 -#define _LOAD_SMALL_INT_1 446 -#define _LOAD_SMALL_INT_2 447 -#define _LOAD_SMALL_INT_3 448 +#define _LOAD_SMALL_INT 445 +#define _LOAD_SMALL_INT_0 446 +#define _LOAD_SMALL_INT_1 447 +#define _LOAD_SMALL_INT_2 448 +#define _LOAD_SMALL_INT_3 449 #define _LOAD_SPECIAL LOAD_SPECIAL #define _LOAD_SUPER_ATTR_ATTR LOAD_SUPER_ATTR_ATTR #define _LOAD_SUPER_ATTR_METHOD LOAD_SUPER_ATTR_METHOD -#define _MAKE_CALLARGS_A_TUPLE 449 +#define _MAKE_CALLARGS_A_TUPLE 450 #define _MAKE_CELL MAKE_CELL #define _MAKE_FUNCTION MAKE_FUNCTION -#define _MAKE_WARM 450 +#define _MAKE_WARM 451 #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 451 -#define _MAYBE_EXPAND_METHOD_KW 452 -#define _MONITOR_CALL 453 -#define _MONITOR_JUMP_BACKWARD 454 -#define _MONITOR_RESUME 455 +#define _MAYBE_EXPAND_METHOD 452 +#define _MAYBE_EXPAND_METHOD_KW 453 +#define _MONITOR_CALL 454 +#define _MONITOR_JUMP_BACKWARD 455 +#define _MONITOR_RESUME 456 #define _NOP NOP #define _POP_EXCEPT POP_EXCEPT -#define _POP_JUMP_IF_FALSE 456 -#define _POP_JUMP_IF_TRUE 457 +#define _POP_JUMP_IF_FALSE 457 +#define _POP_JUMP_IF_TRUE 458 #define _POP_TOP POP_TOP -#define _POP_TOP_LOAD_CONST_INLINE_BORROW 458 +#define _POP_TOP_LOAD_CONST_INLINE_BORROW 459 #define _PUSH_EXC_INFO PUSH_EXC_INFO -#define _PUSH_FRAME 459 +#define _PUSH_FRAME 460 #define _PUSH_NULL PUSH_NULL -#define _PY_FRAME_GENERAL 460 -#define _PY_FRAME_KW 461 -#define _QUICKEN_RESUME 462 -#define _REPLACE_WITH_TRUE 463 +#define _PY_FRAME_GENERAL 461 +#define _PY_FRAME_KW 462 +#define _QUICKEN_RESUME 463 +#define _REPLACE_WITH_TRUE 464 #define _RESUME_CHECK RESUME_CHECK #define _RETURN_GENERATOR RETURN_GENERATOR #define _RETURN_VALUE RETURN_VALUE -#define _SAVE_RETURN_OFFSET 464 -#define _SEND 465 -#define _SEND_GEN_FRAME 466 +#define _SAVE_RETURN_OFFSET 465 +#define _SEND 466 +#define _SEND_GEN_FRAME 467 #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 467 -#define _STORE_ATTR 468 -#define _STORE_ATTR_INSTANCE_VALUE 469 -#define _STORE_ATTR_SLOT 470 -#define _STORE_ATTR_WITH_HINT 471 +#define _START_EXECUTOR 468 +#define _STORE_ATTR 469 +#define _STORE_ATTR_INSTANCE_VALUE 470 +#define _STORE_ATTR_SLOT 471 +#define _STORE_ATTR_WITH_HINT 472 #define _STORE_DEREF STORE_DEREF -#define _STORE_FAST 472 -#define _STORE_FAST_0 473 -#define _STORE_FAST_1 474 -#define _STORE_FAST_2 475 -#define _STORE_FAST_3 476 -#define _STORE_FAST_4 477 -#define _STORE_FAST_5 478 -#define _STORE_FAST_6 479 -#define _STORE_FAST_7 480 +#define _STORE_FAST 473 +#define _STORE_FAST_0 474 +#define _STORE_FAST_1 475 +#define _STORE_FAST_2 476 +#define _STORE_FAST_3 477 +#define _STORE_FAST_4 478 +#define _STORE_FAST_5 479 +#define _STORE_FAST_6 480 +#define _STORE_FAST_7 481 #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 481 -#define _STORE_SUBSCR 482 +#define _STORE_SLICE 482 +#define _STORE_SUBSCR 483 #define _STORE_SUBSCR_DICT STORE_SUBSCR_DICT #define _STORE_SUBSCR_LIST_INT STORE_SUBSCR_LIST_INT #define _SWAP SWAP -#define _TIER2_RESUME_CHECK 483 -#define _TO_BOOL 484 +#define _TIER2_RESUME_CHECK 484 +#define _TO_BOOL 485 #define _TO_BOOL_BOOL TO_BOOL_BOOL #define _TO_BOOL_INT TO_BOOL_INT #define _TO_BOOL_LIST TO_BOOL_LIST @@ -303,13 +304,13 @@ extern "C" { #define _UNARY_NEGATIVE UNARY_NEGATIVE #define _UNARY_NOT UNARY_NOT #define _UNPACK_EX UNPACK_EX -#define _UNPACK_SEQUENCE 485 +#define _UNPACK_SEQUENCE 486 #define _UNPACK_SEQUENCE_LIST UNPACK_SEQUENCE_LIST #define _UNPACK_SEQUENCE_TUPLE UNPACK_SEQUENCE_TUPLE #define _UNPACK_SEQUENCE_TWO_TUPLE UNPACK_SEQUENCE_TWO_TUPLE #define _WITH_EXCEPT_START WITH_EXCEPT_START #define _YIELD_VALUE YIELD_VALUE -#define MAX_UOP_ID 485 +#define MAX_UOP_ID 486 #ifdef __cplusplus } diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h index 98a41d1f23f..1b2880cb6bb 100644 --- a/Include/internal/pycore_uop_metadata.h +++ b/Include/internal/pycore_uop_metadata.h @@ -213,6 +213,7 @@ const uint16_t _PyUop_Flags[MAX_UOP_ID+1] = { [_MAYBE_EXPAND_METHOD] = HAS_ARG_FLAG, [_PY_FRAME_GENERAL] = HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ERROR_NO_POP_FLAG | HAS_ESCAPES_FLAG, [_CHECK_FUNCTION_VERSION] = HAS_ARG_FLAG | HAS_EXIT_FLAG, + [_CHECK_FUNCTION_VERSION_INLINE] = HAS_EXIT_FLAG, [_CHECK_METHOD_VERSION] = HAS_ARG_FLAG | HAS_EXIT_FLAG, [_EXPAND_METHOD] = HAS_ARG_FLAG, [_CHECK_IS_NOT_PY_CALLABLE] = HAS_ARG_FLAG | HAS_EXIT_FLAG, @@ -353,6 +354,7 @@ const char *const _PyOpcode_uop_name[MAX_UOP_ID+1] = { [_CHECK_FUNCTION] = "_CHECK_FUNCTION", [_CHECK_FUNCTION_EXACT_ARGS] = "_CHECK_FUNCTION_EXACT_ARGS", [_CHECK_FUNCTION_VERSION] = "_CHECK_FUNCTION_VERSION", + [_CHECK_FUNCTION_VERSION_INLINE] = "_CHECK_FUNCTION_VERSION_INLINE", [_CHECK_FUNCTION_VERSION_KW] = "_CHECK_FUNCTION_VERSION_KW", [_CHECK_IS_NOT_PY_CALLABLE] = "_CHECK_IS_NOT_PY_CALLABLE", [_CHECK_IS_NOT_PY_CALLABLE_KW] = "_CHECK_IS_NOT_PY_CALLABLE_KW", @@ -965,6 +967,8 @@ int _PyUop_num_popped(int opcode, int oparg) return 2 + oparg; case _CHECK_FUNCTION_VERSION: return 0; + case _CHECK_FUNCTION_VERSION_INLINE: + return 0; case _CHECK_METHOD_VERSION: return 0; case _EXPAND_METHOD: diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py index 7b3d9e4fd11..9726353bcd6 100644 --- a/Lib/test/test_capi/test_opt.py +++ b/Lib/test/test_capi/test_opt.py @@ -1486,6 +1486,26 @@ class TestUopsOptimization(unittest.TestCase): fn(A()) + def test_func_guards_removed_or_reduced(self): + def testfunc(n): + for i in range(n): + # Only works on functions promoted to constants + global_identity(i) + + opt = _testinternalcapi.new_uop_optimizer() + with temporary_optimizer(opt): + testfunc(20) + + ex = get_first_executor(testfunc) + self.assertIsNotNone(ex) + uops = get_opnames(ex) + self.assertIn("_PUSH_FRAME", uops) + # Strength reduced version + self.assertIn("_CHECK_FUNCTION_VERSION_INLINE", uops) + self.assertNotIn("_CHECK_FUNCTION_VERSION", uops) + # Removed guard + self.assertNotIn("_CHECK_FUNCTION_EXACT_ARGS", uops) + def test_jit_error_pops(self): """ Tests that the correct number of pops are inserted into the @@ -1495,5 +1515,9 @@ class TestUopsOptimization(unittest.TestCase): with self.assertRaises(TypeError): {item for item in items} + +def global_identity(x): + return x + if __name__ == "__main__": unittest.main() diff --git a/Python/bytecodes.c b/Python/bytecodes.c index 7ae0f203696..04983fd861e 100644 --- a/Python/bytecodes.c +++ b/Python/bytecodes.c @@ -661,7 +661,7 @@ dummy_func( assert(next_instr->op.code == STORE_FAST); next_oparg = next_instr->op.arg; #else - next_oparg = CURRENT_OPERAND(); + next_oparg = CURRENT_OPERAND0(); #endif _PyStackRef *target_local = &GETLOCAL(next_oparg); DEOPT_IF(!PyStackRef_Is(*target_local, left)); @@ -3463,6 +3463,12 @@ dummy_func( EXIT_IF(func->func_version != func_version); } + tier2 op(_CHECK_FUNCTION_VERSION_INLINE, (func_version/2, callable_o/4 --)) { + assert(PyFunction_Check(callable_o)); + PyFunctionObject *func = (PyFunctionObject *)callable_o; + EXIT_IF(func->func_version != func_version); + } + macro(CALL_PY_GENERAL) = unused/1 + // Skip over the counter _CHECK_PEP_523 + diff --git a/Python/ceval_macros.h b/Python/ceval_macros.h index 5df55813a0d..603b71ea938 100644 --- a/Python/ceval_macros.h +++ b/Python/ceval_macros.h @@ -413,7 +413,8 @@ do { \ #define CURRENT_OPARG() (next_uop[-1].oparg) -#define CURRENT_OPERAND() (next_uop[-1].operand) +#define CURRENT_OPERAND0() (next_uop[-1].operand0) +#define CURRENT_OPERAND1() (next_uop[-1].operand1) #define JUMP_TO_JUMP_TARGET() goto jump_to_jump_target #define JUMP_TO_ERROR() goto jump_to_error_target diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h index 1d63402214d..494ace1bd85 100644 --- a/Python/executor_cases.c.h +++ b/Python/executor_cases.c.h @@ -831,7 +831,7 @@ assert(next_instr->op.code == STORE_FAST); next_oparg = next_instr->op.arg; #else - next_oparg = CURRENT_OPERAND(); + next_oparg = CURRENT_OPERAND0(); #endif _PyStackRef *target_local = &GETLOCAL(next_oparg); if (!PyStackRef_Is(*target_local, left)) { @@ -1864,7 +1864,7 @@ } case _GUARD_GLOBALS_VERSION: { - uint16_t version = (uint16_t)CURRENT_OPERAND(); + uint16_t version = (uint16_t)CURRENT_OPERAND0(); PyDictObject *dict = (PyDictObject *)GLOBALS(); if (!PyDict_CheckExact(dict)) { UOP_STAT_INC(uopcode, miss); @@ -1880,7 +1880,7 @@ case _GUARD_GLOBALS_VERSION_PUSH_KEYS: { PyDictKeysObject *globals_keys; - uint16_t version = (uint16_t)CURRENT_OPERAND(); + uint16_t version = (uint16_t)CURRENT_OPERAND0(); PyDictObject *dict = (PyDictObject *)GLOBALS(); if (!PyDict_CheckExact(dict)) { UOP_STAT_INC(uopcode, miss); @@ -1900,7 +1900,7 @@ case _GUARD_BUILTINS_VERSION_PUSH_KEYS: { PyDictKeysObject *builtins_keys; - uint16_t version = (uint16_t)CURRENT_OPERAND(); + uint16_t version = (uint16_t)CURRENT_OPERAND0(); PyDictObject *dict = (PyDictObject *)BUILTINS(); if (!PyDict_CheckExact(dict)) { UOP_STAT_INC(uopcode, miss); @@ -1924,7 +1924,7 @@ _PyStackRef null = PyStackRef_NULL; oparg = CURRENT_OPARG(); globals_keys = (PyDictKeysObject *)stack_pointer[-1].bits; - uint16_t index = (uint16_t)CURRENT_OPERAND(); + uint16_t index = (uint16_t)CURRENT_OPERAND0(); PyDictUnicodeEntry *entries = DK_UNICODE_ENTRIES(globals_keys); PyObject *res_o = entries[index].me_value; stack_pointer += -1; @@ -1950,7 +1950,7 @@ _PyStackRef null = PyStackRef_NULL; oparg = CURRENT_OPARG(); builtins_keys = (PyDictKeysObject *)stack_pointer[-1].bits; - uint16_t index = (uint16_t)CURRENT_OPERAND(); + uint16_t index = (uint16_t)CURRENT_OPERAND0(); PyDictUnicodeEntry *entries = DK_UNICODE_ENTRIES(builtins_keys); PyObject *res_o = entries[index].me_value; stack_pointer += -1; @@ -2523,7 +2523,7 @@ case _GUARD_TYPE_VERSION: { _PyStackRef owner; owner = stack_pointer[-1]; - uint32_t type_version = (uint32_t)CURRENT_OPERAND(); + uint32_t type_version = (uint32_t)CURRENT_OPERAND0(); PyTypeObject *tp = Py_TYPE(PyStackRef_AsPyObjectBorrow(owner)); assert(type_version != 0); if (tp->tp_version_tag != type_version) { @@ -2552,7 +2552,7 @@ _PyStackRef null = PyStackRef_NULL; (void)null; owner = stack_pointer[-1]; - uint16_t offset = (uint16_t)CURRENT_OPERAND(); + uint16_t offset = (uint16_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); PyObject **value_ptr = (PyObject**)(((char *)owner_o) + offset); PyObject *attr_o = *value_ptr; @@ -2575,7 +2575,7 @@ _PyStackRef null = PyStackRef_NULL; (void)null; owner = stack_pointer[-1]; - uint16_t offset = (uint16_t)CURRENT_OPERAND(); + uint16_t offset = (uint16_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); PyObject **value_ptr = (PyObject**)(((char *)owner_o) + offset); PyObject *attr_o = *value_ptr; @@ -2600,7 +2600,7 @@ case _CHECK_ATTR_MODULE: { _PyStackRef owner; owner = stack_pointer[-1]; - uint32_t dict_version = (uint32_t)CURRENT_OPERAND(); + uint32_t dict_version = (uint32_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); if (!PyModule_CheckExact(owner_o)) { UOP_STAT_INC(uopcode, miss); @@ -2621,7 +2621,7 @@ _PyStackRef null = PyStackRef_NULL; oparg = CURRENT_OPARG(); owner = stack_pointer[-1]; - uint16_t index = (uint16_t)CURRENT_OPERAND(); + uint16_t index = (uint16_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); PyDictObject *dict = (PyDictObject *)((PyModuleObject *)owner_o)->md_dict; assert(dict->ma_keys->dk_kind == DICT_KEYS_UNICODE); @@ -2664,7 +2664,7 @@ _PyStackRef null = PyStackRef_NULL; oparg = CURRENT_OPARG(); owner = stack_pointer[-1]; - uint16_t hint = (uint16_t)CURRENT_OPERAND(); + uint16_t hint = (uint16_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); PyObject *attr_o; PyDictObject *dict = _PyObject_GetManagedDict(owner_o); @@ -2705,7 +2705,7 @@ _PyStackRef null = PyStackRef_NULL; (void)null; owner = stack_pointer[-1]; - uint16_t index = (uint16_t)CURRENT_OPERAND(); + uint16_t index = (uint16_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); char *addr = (char *)owner_o + index; PyObject *attr_o = *(PyObject **)addr; @@ -2727,7 +2727,7 @@ _PyStackRef null = PyStackRef_NULL; (void)null; owner = stack_pointer[-1]; - uint16_t index = (uint16_t)CURRENT_OPERAND(); + uint16_t index = (uint16_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); char *addr = (char *)owner_o + index; PyObject *attr_o = *(PyObject **)addr; @@ -2751,7 +2751,7 @@ case _CHECK_ATTR_CLASS: { _PyStackRef owner; owner = stack_pointer[-1]; - uint32_t type_version = (uint32_t)CURRENT_OPERAND(); + uint32_t type_version = (uint32_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); if (!PyType_Check(owner_o)) { UOP_STAT_INC(uopcode, miss); @@ -2771,7 +2771,7 @@ _PyStackRef null = PyStackRef_NULL; (void)null; owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)CURRENT_OPERAND(); + PyObject *descr = (PyObject *)CURRENT_OPERAND0(); STAT_INC(LOAD_ATTR, hit); assert(descr != NULL); attr = PyStackRef_FromPyObjectNew(descr); @@ -2787,7 +2787,7 @@ _PyStackRef null = PyStackRef_NULL; (void)null; owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)CURRENT_OPERAND(); + PyObject *descr = (PyObject *)CURRENT_OPERAND0(); STAT_INC(LOAD_ATTR, hit); assert(descr != NULL); attr = PyStackRef_FromPyObjectNew(descr); @@ -2807,7 +2807,7 @@ _PyInterpreterFrame *new_frame; oparg = CURRENT_OPARG(); owner = stack_pointer[-1]; - PyObject *fget = (PyObject *)CURRENT_OPERAND(); + PyObject *fget = (PyObject *)CURRENT_OPERAND0(); assert((oparg & 1) == 0); assert(Py_IS_TYPE(fget, &PyFunction_Type)); PyFunctionObject *f = (PyFunctionObject *)fget; @@ -2859,7 +2859,7 @@ _PyStackRef value; owner = stack_pointer[-1]; value = stack_pointer[-2]; - uint16_t offset = (uint16_t)CURRENT_OPERAND(); + uint16_t offset = (uint16_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); STAT_INC(STORE_ATTR, hit); assert(_PyObject_GetManagedDict(owner_o) == NULL); @@ -2886,7 +2886,7 @@ oparg = CURRENT_OPARG(); owner = stack_pointer[-1]; value = stack_pointer[-2]; - uint16_t hint = (uint16_t)CURRENT_OPERAND(); + uint16_t hint = (uint16_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); assert(Py_TYPE(owner_o)->tp_flags & Py_TPFLAGS_MANAGED_DICT); PyDictObject *dict = _PyObject_GetManagedDict(owner_o); @@ -2937,7 +2937,7 @@ _PyStackRef value; owner = stack_pointer[-1]; value = stack_pointer[-2]; - uint16_t index = (uint16_t)CURRENT_OPERAND(); + uint16_t index = (uint16_t)CURRENT_OPERAND0(); PyObject *owner_o = PyStackRef_AsPyObjectBorrow(owner); char *addr = (char *)owner_o + index; STAT_INC(STORE_ATTR, hit); @@ -3780,7 +3780,7 @@ case _GUARD_KEYS_VERSION: { _PyStackRef owner; owner = stack_pointer[-1]; - uint32_t keys_version = (uint32_t)CURRENT_OPERAND(); + uint32_t keys_version = (uint32_t)CURRENT_OPERAND0(); PyTypeObject *owner_cls = Py_TYPE(PyStackRef_AsPyObjectBorrow(owner)); PyHeapTypeObject *owner_heap_type = (PyHeapTypeObject *)owner_cls; if (owner_heap_type->ht_cached_keys->dk_version != keys_version) { @@ -3796,7 +3796,7 @@ _PyStackRef self = PyStackRef_NULL; oparg = CURRENT_OPARG(); owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)CURRENT_OPERAND(); + PyObject *descr = (PyObject *)CURRENT_OPERAND0(); assert(oparg & 1); /* Cached method object */ STAT_INC(LOAD_ATTR, hit); @@ -3817,7 +3817,7 @@ _PyStackRef self = PyStackRef_NULL; oparg = CURRENT_OPARG(); owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)CURRENT_OPERAND(); + PyObject *descr = (PyObject *)CURRENT_OPERAND0(); assert(oparg & 1); assert(Py_TYPE(PyStackRef_AsPyObjectBorrow(owner))->tp_dictoffset == 0); STAT_INC(LOAD_ATTR, hit); @@ -3837,7 +3837,7 @@ _PyStackRef attr; oparg = CURRENT_OPARG(); owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)CURRENT_OPERAND(); + PyObject *descr = (PyObject *)CURRENT_OPERAND0(); assert((oparg & 1) == 0); STAT_INC(LOAD_ATTR, hit); assert(descr != NULL); @@ -3852,7 +3852,7 @@ _PyStackRef attr; oparg = CURRENT_OPARG(); owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)CURRENT_OPERAND(); + PyObject *descr = (PyObject *)CURRENT_OPERAND0(); assert((oparg & 1) == 0); assert(Py_TYPE(PyStackRef_AsPyObjectBorrow(owner))->tp_dictoffset == 0); STAT_INC(LOAD_ATTR, hit); @@ -3866,7 +3866,7 @@ case _CHECK_ATTR_METHOD_LAZY_DICT: { _PyStackRef owner; owner = stack_pointer[-1]; - uint16_t dictoffset = (uint16_t)CURRENT_OPERAND(); + uint16_t dictoffset = (uint16_t)CURRENT_OPERAND0(); char *ptr = ((char *)PyStackRef_AsPyObjectBorrow(owner)) + MANAGED_DICT_OFFSET + dictoffset; PyObject *dict = *(PyObject **)ptr; /* This object has a __dict__, just not yet created */ @@ -3883,7 +3883,7 @@ _PyStackRef self = PyStackRef_NULL; oparg = CURRENT_OPARG(); owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)CURRENT_OPERAND(); + PyObject *descr = (PyObject *)CURRENT_OPERAND0(); assert(oparg & 1); STAT_INC(LOAD_ATTR, hit); assert(descr != NULL); @@ -3967,7 +3967,7 @@ _PyStackRef *callable; oparg = CURRENT_OPARG(); callable = &stack_pointer[-2 - oparg]; - uint32_t func_version = (uint32_t)CURRENT_OPERAND(); + uint32_t func_version = (uint32_t)CURRENT_OPERAND0(); PyObject *callable_o = PyStackRef_AsPyObjectBorrow(callable[0]); if (!PyFunction_Check(callable_o)) { UOP_STAT_INC(uopcode, miss); @@ -3981,13 +3981,25 @@ break; } + case _CHECK_FUNCTION_VERSION_INLINE: { + uint32_t func_version = (uint32_t)CURRENT_OPERAND0(); + PyObject *callable_o = (PyObject *)CURRENT_OPERAND1(); + assert(PyFunction_Check(callable_o)); + PyFunctionObject *func = (PyFunctionObject *)callable_o; + if (func->func_version != func_version) { + UOP_STAT_INC(uopcode, miss); + JUMP_TO_JUMP_TARGET(); + } + break; + } + case _CHECK_METHOD_VERSION: { _PyStackRef *null; _PyStackRef *callable; oparg = CURRENT_OPARG(); null = &stack_pointer[-1 - oparg]; callable = &stack_pointer[-2 - oparg]; - uint32_t func_version = (uint32_t)CURRENT_OPERAND(); + uint32_t func_version = (uint32_t)CURRENT_OPERAND0(); PyObject *callable_o = PyStackRef_AsPyObjectBorrow(callable[0]); if (Py_TYPE(callable_o) != &PyMethod_Type) { UOP_STAT_INC(uopcode, miss); @@ -4443,7 +4455,7 @@ callable = &stack_pointer[-2 - oparg]; init = &stack_pointer[-2 - oparg]; self = &stack_pointer[-1 - oparg]; - uint32_t type_version = (uint32_t)CURRENT_OPERAND(); + uint32_t type_version = (uint32_t)CURRENT_OPERAND0(); PyObject *callable_o = PyStackRef_AsPyObjectBorrow(callable[0]); if (!PyStackRef_IsNull(null[0])) { UOP_STAT_INC(uopcode, miss); @@ -5201,7 +5213,7 @@ _PyStackRef *callable; oparg = CURRENT_OPARG(); callable = &stack_pointer[-3 - oparg]; - uint32_t func_version = (uint32_t)CURRENT_OPERAND(); + uint32_t func_version = (uint32_t)CURRENT_OPERAND0(); PyObject *callable_o = PyStackRef_AsPyObjectBorrow(callable[0]); if (!PyFunction_Check(callable_o)) { UOP_STAT_INC(uopcode, miss); @@ -5221,7 +5233,7 @@ oparg = CURRENT_OPARG(); null = &stack_pointer[-2 - oparg]; callable = &stack_pointer[-3 - oparg]; - uint32_t func_version = (uint32_t)CURRENT_OPERAND(); + uint32_t func_version = (uint32_t)CURRENT_OPERAND0(); PyObject *callable_o = PyStackRef_AsPyObjectBorrow(callable[0]); if (Py_TYPE(callable_o) != &PyMethod_Type) { UOP_STAT_INC(uopcode, miss); @@ -5659,13 +5671,13 @@ } case _SET_IP: { - PyObject *instr_ptr = (PyObject *)CURRENT_OPERAND(); + PyObject *instr_ptr = (PyObject *)CURRENT_OPERAND0(); frame->instr_ptr = (_Py_CODEUNIT *)instr_ptr; break; } case _CHECK_STACK_SPACE_OPERAND: { - uint32_t framesize = (uint32_t)CURRENT_OPERAND(); + uint32_t framesize = (uint32_t)CURRENT_OPERAND0(); assert(framesize <= INT_MAX); if (!_PyThreadState_HasStackSpace(tstate, framesize)) { UOP_STAT_INC(uopcode, miss); @@ -5690,7 +5702,7 @@ } case _EXIT_TRACE: { - PyObject *exit_p = (PyObject *)CURRENT_OPERAND(); + PyObject *exit_p = (PyObject *)CURRENT_OPERAND0(); _PyExitData *exit = (_PyExitData *)exit_p; PyCodeObject *code = _PyFrame_GetCode(frame); _PyFrame_SetStackPointer(frame, stack_pointer); @@ -5757,7 +5769,7 @@ case _LOAD_CONST_INLINE: { _PyStackRef value; - PyObject *ptr = (PyObject *)CURRENT_OPERAND(); + PyObject *ptr = (PyObject *)CURRENT_OPERAND0(); value = PyStackRef_FromPyObjectNew(ptr); stack_pointer[0] = value; stack_pointer += 1; @@ -5767,7 +5779,7 @@ case _LOAD_CONST_INLINE_BORROW: { _PyStackRef value; - PyObject *ptr = (PyObject *)CURRENT_OPERAND(); + PyObject *ptr = (PyObject *)CURRENT_OPERAND0(); value = PyStackRef_FromPyObjectImmortal(ptr); stack_pointer[0] = value; stack_pointer += 1; @@ -5779,7 +5791,7 @@ _PyStackRef pop; _PyStackRef value; pop = stack_pointer[-1]; - PyObject *ptr = (PyObject *)CURRENT_OPERAND(); + PyObject *ptr = (PyObject *)CURRENT_OPERAND0(); PyStackRef_CLOSE(pop); value = PyStackRef_FromPyObjectImmortal(ptr); stack_pointer[-1] = value; @@ -5789,7 +5801,7 @@ case _LOAD_CONST_INLINE_WITH_NULL: { _PyStackRef value; _PyStackRef null; - PyObject *ptr = (PyObject *)CURRENT_OPERAND(); + PyObject *ptr = (PyObject *)CURRENT_OPERAND0(); value = PyStackRef_FromPyObjectNew(ptr); null = PyStackRef_NULL; stack_pointer[0] = value; @@ -5802,7 +5814,7 @@ case _LOAD_CONST_INLINE_BORROW_WITH_NULL: { _PyStackRef value; _PyStackRef null; - PyObject *ptr = (PyObject *)CURRENT_OPERAND(); + PyObject *ptr = (PyObject *)CURRENT_OPERAND0(); value = PyStackRef_FromPyObjectImmortal(ptr); null = PyStackRef_NULL; stack_pointer[0] = value; @@ -5813,7 +5825,7 @@ } case _CHECK_FUNCTION: { - uint32_t func_version = (uint32_t)CURRENT_OPERAND(); + uint32_t func_version = (uint32_t)CURRENT_OPERAND0(); assert(PyStackRef_FunctionCheck(frame->f_funcobj)); PyFunctionObject *func = (PyFunctionObject *)PyStackRef_AsPyObjectBorrow(frame->f_funcobj); if (func->func_version != func_version) { @@ -5827,7 +5839,7 @@ _PyStackRef res; _PyStackRef null = PyStackRef_NULL; oparg = CURRENT_OPARG(); - uint16_t index = (uint16_t)CURRENT_OPERAND(); + uint16_t index = (uint16_t)CURRENT_OPERAND0(); PyDictObject *dict = (PyDictObject *)GLOBALS(); PyDictUnicodeEntry *entries = DK_UNICODE_ENTRIES(dict->ma_keys); PyObject *res_o = entries[index].me_value; @@ -5849,7 +5861,7 @@ _PyStackRef res; _PyStackRef null = PyStackRef_NULL; oparg = CURRENT_OPARG(); - uint16_t index = (uint16_t)CURRENT_OPERAND(); + uint16_t index = (uint16_t)CURRENT_OPERAND0(); PyDictObject *dict = (PyDictObject *)BUILTINS(); PyDictUnicodeEntry *entries = DK_UNICODE_ENTRIES(dict->ma_keys); PyObject *res_o = entries[index].me_value; @@ -5878,7 +5890,7 @@ } case _DYNAMIC_EXIT: { - PyObject *exit_p = (PyObject *)CURRENT_OPERAND(); + PyObject *exit_p = (PyObject *)CURRENT_OPERAND0(); tstate->previous_executor = (PyObject *)current_executor; _PyExitData *exit = (_PyExitData *)exit_p; _Py_CODEUNIT *target = frame->instr_ptr; @@ -5925,7 +5937,7 @@ } case _START_EXECUTOR: { - PyObject *executor = (PyObject *)CURRENT_OPERAND(); + PyObject *executor = (PyObject *)CURRENT_OPERAND0(); Py_DECREF(tstate->previous_executor); tstate->previous_executor = NULL; #ifndef _Py_JIT @@ -5951,7 +5963,7 @@ } case _CHECK_VALIDITY_AND_SET_IP: { - PyObject *instr_ptr = (PyObject *)CURRENT_OPERAND(); + PyObject *instr_ptr = (PyObject *)CURRENT_OPERAND0(); if (!current_executor->vm_data.valid) { UOP_STAT_INC(uopcode, miss); JUMP_TO_JUMP_TARGET(); @@ -5967,7 +5979,7 @@ case _ERROR_POP_N: { oparg = CURRENT_OPARG(); - uint32_t target = (uint32_t)CURRENT_OPERAND(); + uint32_t target = (uint32_t)CURRENT_OPERAND0(); stack_pointer += -oparg; assert(WITHIN_STACK_BOUNDS()); _PyFrame_SetStackPointer(frame, stack_pointer); diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h index 03b4d222492..77bf6ad3781 100644 --- a/Python/generated_cases.c.h +++ b/Python/generated_cases.c.h @@ -192,7 +192,7 @@ assert(next_instr->op.code == STORE_FAST); next_oparg = next_instr->op.arg; #else - next_oparg = CURRENT_OPERAND(); + next_oparg = CURRENT_OPERAND0(); #endif _PyStackRef *target_local = &GETLOCAL(next_oparg); DEOPT_IF(!PyStackRef_Is(*target_local, left), BINARY_OP); diff --git a/Python/optimizer.c b/Python/optimizer.c index b876b6c2bd7..bc2ecc098b0 100644 --- a/Python/optimizer.c +++ b/Python/optimizer.c @@ -288,13 +288,13 @@ _PyUOpPrint(const _PyUOpInstruction *uop) printf(" (%d, target=%d, operand=%#" PRIx64, uop->oparg, uop->target, - (uint64_t)uop->operand); + (uint64_t)uop->operand0); break; case UOP_FORMAT_JUMP: printf(" (%d, jump_target=%d, operand=%#" PRIx64, uop->oparg, uop->jump_target, - (uint64_t)uop->operand); + (uint64_t)uop->operand0); break; default: printf(" (%d, Unknown format)", uop->oparg); @@ -340,7 +340,7 @@ uop_item(_PyExecutorObject *self, Py_ssize_t index) Py_DECREF(oname); return NULL; } - PyObject *operand = PyLong_FromUnsignedLongLong(self->trace[index].operand); + PyObject *operand = PyLong_FromUnsignedLongLong(self->trace[index].operand0); if (operand == NULL) { Py_DECREF(target); Py_DECREF(oparg); @@ -463,7 +463,7 @@ add_to_trace( trace[trace_length].format = UOP_FORMAT_TARGET; trace[trace_length].target = target; trace[trace_length].oparg = oparg; - trace[trace_length].operand = operand; + trace[trace_length].operand0 = operand; return trace_length + 1; } @@ -970,7 +970,7 @@ static void make_exit(_PyUOpInstruction *inst, int opcode, int target) { inst->opcode = opcode; inst->oparg = 0; - inst->operand = 0; + inst->operand0 = 0; inst->format = UOP_FORMAT_TARGET; inst->target = target; } @@ -1033,7 +1033,7 @@ prepare_for_execution(_PyUOpInstruction *buffer, int length) current_error_target = target; make_exit(&buffer[next_spare], _ERROR_POP_N, 0); buffer[next_spare].oparg = popped; - buffer[next_spare].operand = target; + buffer[next_spare].operand0 = target; next_spare++; } buffer[i].error_target = current_error; @@ -1150,7 +1150,7 @@ make_executor_from_uops(_PyUOpInstruction *buffer, int length, const _PyBloomFil int next_exit = exit_count-1; _PyUOpInstruction *dest = (_PyUOpInstruction *)&executor->trace[length]; assert(buffer[0].opcode == _START_EXECUTOR); - buffer[0].operand = (uint64_t)executor; + buffer[0].operand0 = (uint64_t)executor; for (int i = length-1; i >= 0; i--) { int opcode = buffer[i].opcode; dest--; @@ -1159,13 +1159,13 @@ make_executor_from_uops(_PyUOpInstruction *buffer, int length, const _PyBloomFil if (opcode == _EXIT_TRACE) { _PyExitData *exit = &executor->exits[next_exit]; exit->target = buffer[i].target; - dest->operand = (uint64_t)exit; + dest->operand0 = (uint64_t)exit; next_exit--; } if (opcode == _DYNAMIC_EXIT) { _PyExitData *exit = &executor->exits[next_exit]; exit->target = 0; - dest->operand = (uint64_t)exit; + dest->operand0 = (uint64_t)exit; next_exit--; } } @@ -1312,7 +1312,7 @@ _PyOptimizer_NewUOpOptimizer(void) static void counter_dealloc(_PyExecutorObject *self) { /* The optimizer is the operand of the second uop. */ - PyObject *opt = (PyObject *)self->trace[1].operand; + PyObject *opt = (PyObject *)self->trace[1].operand0; Py_DECREF(opt); uop_dealloc(self); } @@ -1352,7 +1352,7 @@ counter_optimize( _Py_CODEUNIT *target = instr + 1 + _PyOpcode_Caches[JUMP_BACKWARD] - oparg; _PyUOpInstruction buffer[4] = { { .opcode = _START_EXECUTOR, .jump_target = 3, .format=UOP_FORMAT_JUMP }, - { .opcode = _LOAD_CONST_INLINE, .operand = (uintptr_t)self }, + { .opcode = _LOAD_CONST_INLINE, .operand0 = (uintptr_t)self }, { .opcode = _INTERNAL_INCREMENT_OPT_COUNTER }, { .opcode = _EXIT_TRACE, .target = (uint32_t)(target - _PyCode_CODE(code)), .format=UOP_FORMAT_TARGET } }; diff --git a/Python/optimizer_analysis.c b/Python/optimizer_analysis.c index 25166bc2dc5..a4a0472b64e 100644 --- a/Python/optimizer_analysis.c +++ b/Python/optimizer_analysis.c @@ -100,11 +100,11 @@ convert_global_to_const(_PyUOpInstruction *inst, PyObject *obj) PyDictObject *dict = (PyDictObject *)obj; assert(dict->ma_keys->dk_kind == DICT_KEYS_UNICODE); PyDictUnicodeEntry *entries = DK_UNICODE_ENTRIES(dict->ma_keys); - assert(inst->operand <= UINT16_MAX); - if ((int)inst->operand >= dict->ma_keys->dk_nentries) { + assert(inst->operand0 <= UINT16_MAX); + if ((int)inst->operand0 >= dict->ma_keys->dk_nentries) { return NULL; } - PyObject *res = entries[inst->operand].me_value; + PyObject *res = entries[inst->operand0].me_value; if (res == NULL) { return NULL; } @@ -114,7 +114,7 @@ convert_global_to_const(_PyUOpInstruction *inst, PyObject *obj) else { inst->opcode = (inst->oparg & 1) ? _LOAD_CONST_INLINE_WITH_NULL : _LOAD_CONST_INLINE; } - inst->operand = (uint64_t)res; + inst->operand0 = (uint64_t)res; return res; } @@ -125,7 +125,7 @@ incorrect_keys(_PyUOpInstruction *inst, PyObject *obj) return 1; } PyDictObject *dict = (PyDictObject *)obj; - if (dict->ma_keys->dk_version != inst->operand) { + if (dict->ma_keys->dk_version != inst->operand0) { return 1; } return 0; @@ -215,7 +215,7 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer, } else { buffer[pc].opcode = _CHECK_FUNCTION; - buffer[pc].operand = function_version; + buffer[pc].operand0 = function_version; function_checked |= 1; } // We're no longer pushing the builtins keys; rewrite the @@ -248,7 +248,7 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer, } else { buffer[pc].opcode = _CHECK_FUNCTION; - buffer[pc].operand = function_version; + buffer[pc].operand0 = function_version; function_checked |= 1; } if (opcode == _GUARD_GLOBALS_VERSION_PUSH_KEYS) { @@ -273,7 +273,7 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer, builtins_watched <<= 1; globals_watched <<= 1; function_checked <<= 1; - uint64_t operand = buffer[pc].operand; + uint64_t operand = buffer[pc].operand0; if (operand == 0 || (operand & 1)) { // It's either a code object or NULL, so bail return 1; @@ -301,7 +301,7 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer, builtins_watched >>= 1; globals_watched >>= 1; function_checked >>= 1; - uint64_t operand = buffer[pc].operand; + uint64_t operand = buffer[pc].operand0; if (operand == 0 || (operand & 1)) { // It's either a code object or NULL, so bail return 1; @@ -317,7 +317,7 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer, break; } case _CHECK_FUNCTION_EXACT_ARGS: - prechecked_function_version = (uint32_t)buffer[pc].operand; + prechecked_function_version = (uint32_t)buffer[pc].operand0; break; default: if (is_terminator(inst)) { @@ -343,7 +343,7 @@ remove_globals(_PyInterpreterFrame *frame, _PyUOpInstruction *buffer, #define REPLACE_OP(INST, OP, ARG, OPERAND) \ INST->opcode = OP; \ INST->oparg = ARG; \ - INST->operand = OPERAND; + INST->operand0 = OPERAND; /* Shortened forms for convenience, used in optimizer_bytecodes.c */ #define sym_is_not_null _Py_uop_sym_is_not_null @@ -409,7 +409,7 @@ get_code(_PyUOpInstruction *op) { assert(op->opcode == _PUSH_FRAME || op->opcode == _RETURN_VALUE || op->opcode == _RETURN_GENERATOR); PyCodeObject *co = NULL; - uint64_t operand = op->operand; + uint64_t operand = op->operand0; if (operand == 0) { return NULL; } @@ -429,7 +429,7 @@ static PyCodeObject * get_code_with_logging(_PyUOpInstruction *op) { PyCodeObject *co = NULL; - uint64_t push_operand = op->operand; + uint64_t push_operand = op->operand0; if (push_operand & 1) { co = (PyCodeObject *)(push_operand & ~1); DPRINTF(3, "code=%p ", co); @@ -534,7 +534,7 @@ optimize_uops( assert(max_space <= INT_MAX); assert(max_space <= INT32_MAX); first_valid_check_stack->opcode = _CHECK_STACK_SPACE_OPERAND; - first_valid_check_stack->operand = max_space; + first_valid_check_stack->operand0 = max_space; } return trace_len; diff --git a/Python/optimizer_bytecodes.c b/Python/optimizer_bytecodes.c index 71904c1bc73..42bdbd9ca8d 100644 --- a/Python/optimizer_bytecodes.c +++ b/Python/optimizer_bytecodes.c @@ -346,7 +346,7 @@ dummy_func(void) { res = sym_new_type(ctx, &PyUnicode_Type); } // _STORE_FAST: - GETLOCAL(this_instr->operand) = res; + GETLOCAL(this_instr->operand0) = res; } op(_BINARY_SUBSCR_INIT_CALL, (container, sub -- new_frame: _Py_UOpsAbstractFrame *)) { @@ -589,8 +589,27 @@ dummy_func(void) { self = sym_new_not_null(ctx); } - op(_CHECK_FUNCTION_EXACT_ARGS, (callable, self_or_null, unused[oparg] -- callable, self_or_null, unused[oparg])) { + op(_CHECK_FUNCTION_VERSION, (func_version/2, callable, self_or_null, unused[oparg] -- callable, self_or_null, unused[oparg])) { + (void)self_or_null; + if (sym_is_const(callable) && sym_matches_type(callable, &PyFunction_Type)) { + assert(PyFunction_Check(sym_get_const(callable))); + REPLACE_OP(this_instr, _CHECK_FUNCTION_VERSION_INLINE, 0, func_version); + this_instr->operand1 = (uintptr_t)sym_get_const(callable); + } sym_set_type(callable, &PyFunction_Type); + } + + op(_CHECK_FUNCTION_EXACT_ARGS, (callable, self_or_null, unused[oparg] -- callable, self_or_null, unused[oparg])) { + assert(sym_matches_type(callable, &PyFunction_Type)); + if (sym_is_const(callable)) { + if (sym_is_null(self_or_null) || sym_is_not_null(self_or_null)) { + PyFunctionObject *func = (PyFunctionObject *)sym_get_const(callable); + PyCodeObject *co = (PyCodeObject *)func->func_code; + if (co->co_argcount == oparg + !sym_is_null(self_or_null)) { + REPLACE_OP(this_instr, _NOP, 0 ,0); + } + } + } (void)self_or_null; } diff --git a/Python/optimizer_cases.c.h b/Python/optimizer_cases.c.h index 54821b23716..f77a5aa35bd 100644 --- a/Python/optimizer_cases.c.h +++ b/Python/optimizer_cases.c.h @@ -525,7 +525,7 @@ res = sym_new_type(ctx, &PyUnicode_Type); } // _STORE_FAST: - GETLOCAL(this_instr->operand) = res; + GETLOCAL(this_instr->operand0) = res; stack_pointer += -2; assert(WITHIN_STACK_BOUNDS()); break; @@ -888,7 +888,7 @@ case _GUARD_GLOBALS_VERSION_PUSH_KEYS: { _Py_UopsSymbol *globals_keys; - uint16_t version = (uint16_t)this_instr->operand; + uint16_t version = (uint16_t)this_instr->operand0; globals_keys = sym_new_unknown(ctx); (void)version; stack_pointer[0] = globals_keys; @@ -899,7 +899,7 @@ case _GUARD_BUILTINS_VERSION_PUSH_KEYS: { _Py_UopsSymbol *builtins_keys; - uint16_t version = (uint16_t)this_instr->operand; + uint16_t version = (uint16_t)this_instr->operand0; builtins_keys = sym_new_unknown(ctx); (void)version; stack_pointer[0] = builtins_keys; @@ -1090,7 +1090,7 @@ case _GUARD_TYPE_VERSION: { _Py_UopsSymbol *owner; owner = stack_pointer[-1]; - uint32_t type_version = (uint32_t)this_instr->operand; + uint32_t type_version = (uint32_t)this_instr->operand0; assert(type_version); if (sym_matches_type_version(owner, type_version)) { REPLACE_OP(this_instr, _NOP, 0, 0); @@ -1122,7 +1122,7 @@ _Py_UopsSymbol *attr; _Py_UopsSymbol *null = NULL; owner = stack_pointer[-1]; - uint16_t offset = (uint16_t)this_instr->operand; + uint16_t offset = (uint16_t)this_instr->operand0; attr = sym_new_not_null(ctx); null = sym_new_null(ctx); (void)offset; @@ -1137,7 +1137,7 @@ case _CHECK_ATTR_MODULE: { _Py_UopsSymbol *owner; owner = stack_pointer[-1]; - uint32_t dict_version = (uint32_t)this_instr->operand; + uint32_t dict_version = (uint32_t)this_instr->operand0; (void)dict_version; if (sym_is_const(owner)) { PyObject *cnst = sym_get_const(owner); @@ -1160,7 +1160,7 @@ _Py_UopsSymbol *attr; _Py_UopsSymbol *null = NULL; owner = stack_pointer[-1]; - uint16_t index = (uint16_t)this_instr->operand; + uint16_t index = (uint16_t)this_instr->operand0; (void)index; null = sym_new_null(ctx); attr = NULL; @@ -1202,7 +1202,7 @@ _Py_UopsSymbol *attr; _Py_UopsSymbol *null = NULL; owner = stack_pointer[-1]; - uint16_t hint = (uint16_t)this_instr->operand; + uint16_t hint = (uint16_t)this_instr->operand0; attr = sym_new_not_null(ctx); null = sym_new_null(ctx); (void)hint; @@ -1219,7 +1219,7 @@ _Py_UopsSymbol *attr; _Py_UopsSymbol *null = NULL; owner = stack_pointer[-1]; - uint16_t index = (uint16_t)this_instr->operand; + uint16_t index = (uint16_t)this_instr->operand0; attr = sym_new_not_null(ctx); null = sym_new_null(ctx); (void)index; @@ -1240,7 +1240,7 @@ _Py_UopsSymbol *attr; _Py_UopsSymbol *null = NULL; owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)this_instr->operand; + PyObject *descr = (PyObject *)this_instr->operand0; attr = sym_new_not_null(ctx); null = sym_new_null(ctx); (void)descr; @@ -1256,7 +1256,7 @@ _Py_UopsSymbol *owner; _Py_UOpsAbstractFrame *new_frame; owner = stack_pointer[-1]; - PyObject *fget = (PyObject *)this_instr->operand; + PyObject *fget = (PyObject *)this_instr->operand0; (void)fget; (void)owner; new_frame = NULL; @@ -1639,7 +1639,7 @@ _Py_UopsSymbol *attr; _Py_UopsSymbol *self = NULL; owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)this_instr->operand; + PyObject *descr = (PyObject *)this_instr->operand0; (void)descr; attr = sym_new_not_null(ctx); self = owner; @@ -1655,7 +1655,7 @@ _Py_UopsSymbol *attr; _Py_UopsSymbol *self = NULL; owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)this_instr->operand; + PyObject *descr = (PyObject *)this_instr->operand0; (void)descr; attr = sym_new_not_null(ctx); self = owner; @@ -1689,7 +1689,7 @@ _Py_UopsSymbol *attr; _Py_UopsSymbol *self = NULL; owner = stack_pointer[-1]; - PyObject *descr = (PyObject *)this_instr->operand; + PyObject *descr = (PyObject *)this_instr->operand0; (void)descr; attr = sym_new_not_null(ctx); self = owner; @@ -1749,6 +1749,22 @@ } case _CHECK_FUNCTION_VERSION: { + _Py_UopsSymbol *self_or_null; + _Py_UopsSymbol *callable; + self_or_null = stack_pointer[-1 - oparg]; + callable = stack_pointer[-2 - oparg]; + uint32_t func_version = (uint32_t)this_instr->operand0; + (void)self_or_null; + if (sym_is_const(callable) && sym_matches_type(callable, &PyFunction_Type)) { + assert(PyFunction_Check(sym_get_const(callable))); + REPLACE_OP(this_instr, _CHECK_FUNCTION_VERSION_INLINE, 0, func_version); + this_instr->operand1 = (uintptr_t)sym_get_const(callable); + } + sym_set_type(callable, &PyFunction_Type); + break; + } + + case _CHECK_FUNCTION_VERSION_INLINE: { break; } @@ -1816,7 +1832,16 @@ _Py_UopsSymbol *callable; self_or_null = stack_pointer[-1 - oparg]; callable = stack_pointer[-2 - oparg]; - sym_set_type(callable, &PyFunction_Type); + assert(sym_matches_type(callable, &PyFunction_Type)); + if (sym_is_const(callable)) { + if (sym_is_null(self_or_null) || sym_is_not_null(self_or_null)) { + PyFunctionObject *func = (PyFunctionObject *)sym_get_const(callable); + PyCodeObject *co = (PyCodeObject *)func->func_code; + if (co->co_argcount == oparg + !sym_is_null(self_or_null)) { + REPLACE_OP(this_instr, _NOP, 0 ,0); + } + } + } (void)self_or_null; break; } @@ -1939,7 +1964,7 @@ null = stack_pointer[-1 - oparg]; callable = stack_pointer[-2 - oparg]; args = &stack_pointer[-oparg]; - uint32_t type_version = (uint32_t)this_instr->operand; + uint32_t type_version = (uint32_t)this_instr->operand0; (void)type_version; (void)callable; (void)null; @@ -2399,7 +2424,7 @@ } case _CHECK_STACK_SPACE_OPERAND: { - uint32_t framesize = (uint32_t)this_instr->operand; + uint32_t framesize = (uint32_t)this_instr->operand0; (void)framesize; /* We should never see _CHECK_STACK_SPACE_OPERANDs. * They are only created at the end of this pass. */ @@ -2412,7 +2437,7 @@ } case _EXIT_TRACE: { - PyObject *exit_p = (PyObject *)this_instr->operand; + PyObject *exit_p = (PyObject *)this_instr->operand0; (void)exit_p; ctx->done = true; break; @@ -2424,7 +2449,7 @@ case _LOAD_CONST_INLINE: { _Py_UopsSymbol *value; - PyObject *ptr = (PyObject *)this_instr->operand; + PyObject *ptr = (PyObject *)this_instr->operand0; value = sym_new_const(ctx, ptr); stack_pointer[0] = value; stack_pointer += 1; @@ -2434,7 +2459,7 @@ case _LOAD_CONST_INLINE_BORROW: { _Py_UopsSymbol *value; - PyObject *ptr = (PyObject *)this_instr->operand; + PyObject *ptr = (PyObject *)this_instr->operand0; value = sym_new_const(ctx, ptr); stack_pointer[0] = value; stack_pointer += 1; @@ -2452,7 +2477,7 @@ case _LOAD_CONST_INLINE_WITH_NULL: { _Py_UopsSymbol *value; _Py_UopsSymbol *null; - PyObject *ptr = (PyObject *)this_instr->operand; + PyObject *ptr = (PyObject *)this_instr->operand0; value = sym_new_const(ctx, ptr); null = sym_new_null(ctx); stack_pointer[0] = value; @@ -2465,7 +2490,7 @@ case _LOAD_CONST_INLINE_BORROW_WITH_NULL: { _Py_UopsSymbol *value; _Py_UopsSymbol *null; - PyObject *ptr = (PyObject *)this_instr->operand; + PyObject *ptr = (PyObject *)this_instr->operand0; value = sym_new_const(ctx, ptr); null = sym_new_null(ctx); stack_pointer[0] = value; diff --git a/Tools/cases_generator/analyzer.py b/Tools/cases_generator/analyzer.py index a725ec10d4e..96b32445fb6 100644 --- a/Tools/cases_generator/analyzer.py +++ b/Tools/cases_generator/analyzer.py @@ -200,7 +200,7 @@ class Uop: return "has tier 1 control flow" if self.properties.needs_this: return "uses the 'this_instr' variable" - if len([c for c in self.caches if c.name != "unused"]) > 1: + if len([c for c in self.caches if c.name != "unused"]) > 2: return "has unused cache entries" if self.properties.error_with_pop and self.properties.error_without_pop: return "has both popping and not-popping errors" diff --git a/Tools/cases_generator/optimizer_generator.py b/Tools/cases_generator/optimizer_generator.py index 7a1dfe1b85b..d08b621aed5 100644 --- a/Tools/cases_generator/optimizer_generator.py +++ b/Tools/cases_generator/optimizer_generator.py @@ -143,7 +143,7 @@ def write_uop( else: type = f"uint{cache.size*16}_t " cast = f"uint{cache.size*16}_t" - out.emit(f"{type}{cache.name} = ({cast})this_instr->operand;\n") + out.emit(f"{type}{cache.name} = ({cast})this_instr->operand0;\n") if override: emitter = OptimizerEmitter(out) # No reference management of inputs needed. diff --git a/Tools/cases_generator/tier2_generator.py b/Tools/cases_generator/tier2_generator.py index ce761495cca..dd16a1a7eb2 100644 --- a/Tools/cases_generator/tier2_generator.py +++ b/Tools/cases_generator/tier2_generator.py @@ -181,14 +181,14 @@ def write_uop(uop: Uop, emitter: Emitter, stack: Stack) -> Stack: code_list, storage = Storage.for_uop(stack, uop) for code in code_list: emitter.emit(code) - for cache in uop.caches: + for idx, cache in enumerate(uop.caches): if cache.name != "unused": if cache.size == 4: type = cast = "PyObject *" else: type = f"uint{cache.size*16}_t " cast = f"uint{cache.size*16}_t" - emitter.emit(f"{type}{cache.name} = ({cast})CURRENT_OPERAND();\n") + emitter.emit(f"{type}{cache.name} = ({cast})CURRENT_OPERAND{idx}();\n") storage = emitter.emit_tokens(uop, storage, None) except StackError as ex: raise analysis_error(ex.args[0], uop.body[0]) from None diff --git a/Tools/jit/_stencils.py b/Tools/jit/_stencils.py index 2cd051b0a77..61be8fd3bbd 100644 --- a/Tools/jit/_stencils.py +++ b/Tools/jit/_stencils.py @@ -29,11 +29,16 @@ class HoleValue(enum.Enum): GOT = enum.auto() # The current uop's oparg (exposed as _JIT_OPARG): OPARG = enum.auto() - # The current uop's operand on 64-bit platforms (exposed as _JIT_OPERAND): - OPERAND = enum.auto() - # The current uop's operand on 32-bit platforms (exposed as _JIT_OPERAND_HI/LO): - OPERAND_HI = enum.auto() - OPERAND_LO = enum.auto() + # The current uop's operand0 on 64-bit platforms (exposed as _JIT_OPERAND0): + OPERAND0 = enum.auto() + # The current uop's operand0 on 32-bit platforms (exposed as _JIT_OPERAND0_HI/LO): + OPERAND0_HI = enum.auto() + OPERAND0_LO = enum.auto() + # The current uop's operand1 on 64-bit platforms (exposed as _JIT_OPERAND1): + OPERAND1 = enum.auto() + # The current uop's operand1 on 32-bit platforms (exposed as _JIT_OPERAND1_HI/LO): + OPERAND1_HI = enum.auto() + OPERAND1_LO = enum.auto() # The current uop's target (exposed as _JIT_TARGET): TARGET = enum.auto() # The base address of the machine code for the jump target (exposed as _JIT_JUMP_TARGET): @@ -99,9 +104,12 @@ _HOLE_EXPRS = { # These should all have been turned into DATA values by process_relocations: # HoleValue.GOT: "", HoleValue.OPARG: "instruction->oparg", - HoleValue.OPERAND: "instruction->operand", - HoleValue.OPERAND_HI: "(instruction->operand >> 32)", - HoleValue.OPERAND_LO: "(instruction->operand & UINT32_MAX)", + HoleValue.OPERAND0: "instruction->operand0", + HoleValue.OPERAND0_HI: "(instruction->operand0 >> 32)", + HoleValue.OPERAND0_LO: "(instruction->operand0 & UINT32_MAX)", + HoleValue.OPERAND1: "instruction->operand1", + HoleValue.OPERAND1_HI: "(instruction->operand1 >> 32)", + HoleValue.OPERAND1_LO: "(instruction->operand1 & UINT32_MAX)", HoleValue.TARGET: "instruction->target", HoleValue.JUMP_TARGET: "state->instruction_starts[instruction->jump_target]", HoleValue.ERROR_TARGET: "state->instruction_starts[instruction->error_target]", diff --git a/Tools/jit/template.c b/Tools/jit/template.c index 57c1006ab42..95c90bda70f 100644 --- a/Tools/jit/template.c +++ b/Tools/jit/template.c @@ -26,8 +26,11 @@ #undef CURRENT_OPARG #define CURRENT_OPARG() (_oparg) -#undef CURRENT_OPERAND -#define CURRENT_OPERAND() (_operand) +#undef CURRENT_OPERAND0 +#define CURRENT_OPERAND0() (_operand0) + +#undef CURRENT_OPERAND1 +#define CURRENT_OPERAND1() (_operand1) #undef DEOPT_IF #define DEOPT_IF(COND, INSTNAME) \ @@ -99,12 +102,17 @@ _JIT_ENTRY(_PyInterpreterFrame *frame, _PyStackRef *stack_pointer, PyThreadState // Other stuff we need handy: PATCH_VALUE(uint16_t, _oparg, _JIT_OPARG) #if SIZEOF_VOID_P == 8 - PATCH_VALUE(uint64_t, _operand, _JIT_OPERAND) + PATCH_VALUE(uint64_t, _operand0, _JIT_OPERAND0) + PATCH_VALUE(uint64_t, _operand1, _JIT_OPERAND1) #else assert(SIZEOF_VOID_P == 4); - PATCH_VALUE(uint32_t, _operand_hi, _JIT_OPERAND_HI) - PATCH_VALUE(uint32_t, _operand_lo, _JIT_OPERAND_LO) - uint64_t _operand = ((uint64_t)_operand_hi << 32) | _operand_lo; + PATCH_VALUE(uint32_t, _operand0_hi, _JIT_OPERAND0_HI) + PATCH_VALUE(uint32_t, _operand0_lo, _JIT_OPERAND0_LO) + uint64_t _operand0 = ((uint64_t)_operand0_hi << 32) | _operand0_lo; + + PATCH_VALUE(uint32_t, _operand1_hi, _JIT_OPERAND1_HI) + PATCH_VALUE(uint32_t, _operand1_lo, _JIT_OPERAND1_LO) + uint64_t _operand1 = ((uint64_t)_operand1_hi << 32) | _operand1_lo; #endif PATCH_VALUE(uint32_t, _target, _JIT_TARGET)