diff options
| -rw-r--r-- | Include/internal/pycore_opcode_metadata.h | 12 | ||||
| -rw-r--r-- | Include/internal/pycore_uop_ids.h | 194 | ||||
| -rw-r--r-- | Include/internal/pycore_uop_metadata.h | 4 | ||||
| -rw-r--r-- | Lib/test/test_capi/test_opt.py | 59 | ||||
| -rw-r--r-- | Python/bytecodes.c | 21 | ||||
| -rw-r--r-- | Python/executor_cases.c.h | 6 | ||||
| -rw-r--r-- | Python/generated_cases.c.h | 98 | ||||
| -rw-r--r-- | Python/tier2_redundancy_eliminator_bytecodes.c | 8 | ||||
| -rw-r--r-- | Python/tier2_redundancy_eliminator_cases.c.h | 10 |
9 files changed, 249 insertions, 163 deletions
diff --git a/Include/internal/pycore_opcode_metadata.h b/Include/internal/pycore_opcode_metadata.h index f45e5f1..ab34366 100644 --- a/Include/internal/pycore_opcode_metadata.h +++ b/Include/internal/pycore_opcode_metadata.h @@ -999,9 +999,9 @@ const struct opcode_metadata _PyOpcode_opcode_metadata[268] = { [CHECK_EXC_MATCH] = { true, INSTR_FMT_IX, HAS_ERROR_FLAG | HAS_ESCAPES_FLAG }, [CLEANUP_THROW] = { true, INSTR_FMT_IX, HAS_ERROR_FLAG | HAS_ESCAPES_FLAG }, [COMPARE_OP] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG }, - [COMPARE_OP_FLOAT] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG }, - [COMPARE_OP_INT] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG }, - [COMPARE_OP_STR] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG }, + [COMPARE_OP_FLOAT] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_EXIT_FLAG | HAS_ESCAPES_FLAG }, + [COMPARE_OP_INT] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_EXIT_FLAG | HAS_ESCAPES_FLAG }, + [COMPARE_OP_STR] = { true, INSTR_FMT_IBC, HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_EXIT_FLAG | HAS_ESCAPES_FLAG }, [CONTAINS_OP] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG }, [CONVERT_VALUE] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_ERROR_FLAG }, [COPY] = { true, INSTR_FMT_IB, HAS_ARG_FLAG | HAS_PURE_FLAG }, @@ -1221,9 +1221,9 @@ _PyOpcode_macro_expansion[256] = { [CHECK_EG_MATCH] = { .nuops = 1, .uops = { { _CHECK_EG_MATCH, 0, 0 } } }, [CHECK_EXC_MATCH] = { .nuops = 1, .uops = { { _CHECK_EXC_MATCH, 0, 0 } } }, [COMPARE_OP] = { .nuops = 1, .uops = { { _COMPARE_OP, 0, 0 } } }, - [COMPARE_OP_FLOAT] = { .nuops = 1, .uops = { { _COMPARE_OP_FLOAT, 0, 0 } } }, - [COMPARE_OP_INT] = { .nuops = 1, .uops = { { _COMPARE_OP_INT, 0, 0 } } }, - [COMPARE_OP_STR] = { .nuops = 1, .uops = { { _COMPARE_OP_STR, 0, 0 } } }, + [COMPARE_OP_FLOAT] = { .nuops = 2, .uops = { { _GUARD_BOTH_FLOAT, 0, 0 }, { _COMPARE_OP_FLOAT, 0, 0 } } }, + [COMPARE_OP_INT] = { .nuops = 2, .uops = { { _GUARD_BOTH_INT, 0, 0 }, { _COMPARE_OP_INT, 0, 0 } } }, + [COMPARE_OP_STR] = { .nuops = 2, .uops = { { _GUARD_BOTH_UNICODE, 0, 0 }, { _COMPARE_OP_STR, 0, 0 } } }, [CONTAINS_OP] = { .nuops = 1, .uops = { { _CONTAINS_OP, 0, 0 } } }, [CONVERT_VALUE] = { .nuops = 1, .uops = { { _CONVERT_VALUE, 0, 0 } } }, [COPY] = { .nuops = 1, .uops = { { _COPY, 0, 0 } } }, diff --git a/Include/internal/pycore_uop_ids.h b/Include/internal/pycore_uop_ids.h index e098852..3c133d9 100644 --- a/Include/internal/pycore_uop_ids.h +++ b/Include/internal/pycore_uop_ids.h @@ -72,9 +72,9 @@ extern "C" { #define _CHECK_VALIDITY_AND_SET_IP 324 #define _COLD_EXIT 325 #define _COMPARE_OP 326 -#define _COMPARE_OP_FLOAT COMPARE_OP_FLOAT -#define _COMPARE_OP_INT COMPARE_OP_INT -#define _COMPARE_OP_STR COMPARE_OP_STR +#define _COMPARE_OP_FLOAT 327 +#define _COMPARE_OP_INT 328 +#define _COMPARE_OP_STR 329 #define _CONTAINS_OP CONTAINS_OP #define _CONVERT_VALUE CONVERT_VALUE #define _COPY COPY @@ -89,41 +89,41 @@ extern "C" { #define _DICT_UPDATE DICT_UPDATE #define _END_SEND END_SEND #define _EXIT_INIT_CHECK EXIT_INIT_CHECK -#define _FATAL_ERROR 327 +#define _FATAL_ERROR 330 #define _FORMAT_SIMPLE FORMAT_SIMPLE #define _FORMAT_WITH_SPEC FORMAT_WITH_SPEC -#define _FOR_ITER 328 +#define _FOR_ITER 331 #define _FOR_ITER_GEN FOR_ITER_GEN -#define _FOR_ITER_TIER_TWO 329 +#define _FOR_ITER_TIER_TWO 332 #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 330 -#define _GUARD_BOTH_INT 331 -#define _GUARD_BOTH_UNICODE 332 -#define _GUARD_BUILTINS_VERSION 333 -#define _GUARD_DORV_VALUES 334 -#define _GUARD_DORV_VALUES_INST_ATTR_FROM_DICT 335 -#define _GUARD_GLOBALS_VERSION 336 -#define _GUARD_IS_FALSE_POP 337 -#define _GUARD_IS_NONE_POP 338 -#define _GUARD_IS_NOT_NONE_POP 339 -#define _GUARD_IS_TRUE_POP 340 -#define _GUARD_KEYS_VERSION 341 -#define _GUARD_NOT_EXHAUSTED_LIST 342 -#define _GUARD_NOT_EXHAUSTED_RANGE 343 -#define _GUARD_NOT_EXHAUSTED_TUPLE 344 -#define _GUARD_TYPE_VERSION 345 -#define _INIT_CALL_BOUND_METHOD_EXACT_ARGS 346 -#define _INIT_CALL_PY_EXACT_ARGS 347 -#define _INIT_CALL_PY_EXACT_ARGS_0 348 -#define _INIT_CALL_PY_EXACT_ARGS_1 349 -#define _INIT_CALL_PY_EXACT_ARGS_2 350 -#define _INIT_CALL_PY_EXACT_ARGS_3 351 -#define _INIT_CALL_PY_EXACT_ARGS_4 352 +#define _GUARD_BOTH_FLOAT 333 +#define _GUARD_BOTH_INT 334 +#define _GUARD_BOTH_UNICODE 335 +#define _GUARD_BUILTINS_VERSION 336 +#define _GUARD_DORV_VALUES 337 +#define _GUARD_DORV_VALUES_INST_ATTR_FROM_DICT 338 +#define _GUARD_GLOBALS_VERSION 339 +#define _GUARD_IS_FALSE_POP 340 +#define _GUARD_IS_NONE_POP 341 +#define _GUARD_IS_NOT_NONE_POP 342 +#define _GUARD_IS_TRUE_POP 343 +#define _GUARD_KEYS_VERSION 344 +#define _GUARD_NOT_EXHAUSTED_LIST 345 +#define _GUARD_NOT_EXHAUSTED_RANGE 346 +#define _GUARD_NOT_EXHAUSTED_TUPLE 347 +#define _GUARD_TYPE_VERSION 348 +#define _INIT_CALL_BOUND_METHOD_EXACT_ARGS 349 +#define _INIT_CALL_PY_EXACT_ARGS 350 +#define _INIT_CALL_PY_EXACT_ARGS_0 351 +#define _INIT_CALL_PY_EXACT_ARGS_1 352 +#define _INIT_CALL_PY_EXACT_ARGS_2 353 +#define _INIT_CALL_PY_EXACT_ARGS_3 354 +#define _INIT_CALL_PY_EXACT_ARGS_4 355 #define _INSTRUMENTED_CALL INSTRUMENTED_CALL #define _INSTRUMENTED_CALL_FUNCTION_EX INSTRUMENTED_CALL_FUNCTION_EX #define _INSTRUMENTED_CALL_KW INSTRUMENTED_CALL_KW @@ -140,65 +140,65 @@ extern "C" { #define _INSTRUMENTED_RETURN_CONST INSTRUMENTED_RETURN_CONST #define _INSTRUMENTED_RETURN_VALUE INSTRUMENTED_RETURN_VALUE #define _INSTRUMENTED_YIELD_VALUE INSTRUMENTED_YIELD_VALUE -#define _INTERNAL_INCREMENT_OPT_COUNTER 353 -#define _IS_NONE 354 +#define _INTERNAL_INCREMENT_OPT_COUNTER 356 +#define _IS_NONE 357 #define _IS_OP IS_OP -#define _ITER_CHECK_LIST 355 -#define _ITER_CHECK_RANGE 356 -#define _ITER_CHECK_TUPLE 357 -#define _ITER_JUMP_LIST 358 -#define _ITER_JUMP_RANGE 359 -#define _ITER_JUMP_TUPLE 360 -#define _ITER_NEXT_LIST 361 -#define _ITER_NEXT_RANGE 362 -#define _ITER_NEXT_TUPLE 363 -#define _JUMP_TO_TOP 364 +#define _ITER_CHECK_LIST 358 +#define _ITER_CHECK_RANGE 359 +#define _ITER_CHECK_TUPLE 360 +#define _ITER_JUMP_LIST 361 +#define _ITER_JUMP_RANGE 362 +#define _ITER_JUMP_TUPLE 363 +#define _ITER_NEXT_LIST 364 +#define _ITER_NEXT_RANGE 365 +#define _ITER_NEXT_TUPLE 366 +#define _JUMP_TO_TOP 367 #define _LIST_APPEND LIST_APPEND #define _LIST_EXTEND LIST_EXTEND #define _LOAD_ASSERTION_ERROR LOAD_ASSERTION_ERROR -#define _LOAD_ATTR 365 -#define _LOAD_ATTR_CLASS 366 -#define _LOAD_ATTR_CLASS_0 367 -#define _LOAD_ATTR_CLASS_1 368 +#define _LOAD_ATTR 368 +#define _LOAD_ATTR_CLASS 369 +#define _LOAD_ATTR_CLASS_0 370 +#define _LOAD_ATTR_CLASS_1 371 #define _LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN LOAD_ATTR_GETATTRIBUTE_OVERRIDDEN -#define _LOAD_ATTR_INSTANCE_VALUE 369 -#define _LOAD_ATTR_INSTANCE_VALUE_0 370 -#define _LOAD_ATTR_INSTANCE_VALUE_1 371 -#define _LOAD_ATTR_METHOD_LAZY_DICT 372 -#define _LOAD_ATTR_METHOD_NO_DICT 373 -#define _LOAD_ATTR_METHOD_WITH_VALUES 374 -#define _LOAD_ATTR_MODULE 375 -#define _LOAD_ATTR_NONDESCRIPTOR_NO_DICT 376 -#define _LOAD_ATTR_NONDESCRIPTOR_WITH_VALUES 377 +#define _LOAD_ATTR_INSTANCE_VALUE 372 +#define _LOAD_ATTR_INSTANCE_VALUE_0 373 +#define _LOAD_ATTR_INSTANCE_VALUE_1 374 +#define _LOAD_ATTR_METHOD_LAZY_DICT 375 +#define _LOAD_ATTR_METHOD_NO_DICT 376 +#define _LOAD_ATTR_METHOD_WITH_VALUES 377 +#define _LOAD_ATTR_MODULE 378 +#define _LOAD_ATTR_NONDESCRIPTOR_NO_DICT 379 +#define _LOAD_ATTR_NONDESCRIPTOR_WITH_VALUES 380 #define _LOAD_ATTR_PROPERTY LOAD_ATTR_PROPERTY -#define _LOAD_ATTR_SLOT 378 -#define _LOAD_ATTR_SLOT_0 379 -#define _LOAD_ATTR_SLOT_1 380 -#define _LOAD_ATTR_WITH_HINT 381 +#define _LOAD_ATTR_SLOT 381 +#define _LOAD_ATTR_SLOT_0 382 +#define _LOAD_ATTR_SLOT_1 383 +#define _LOAD_ATTR_WITH_HINT 384 #define _LOAD_BUILD_CLASS LOAD_BUILD_CLASS #define _LOAD_CONST LOAD_CONST -#define _LOAD_CONST_INLINE 382 -#define _LOAD_CONST_INLINE_BORROW 383 -#define _LOAD_CONST_INLINE_BORROW_WITH_NULL 384 -#define _LOAD_CONST_INLINE_WITH_NULL 385 +#define _LOAD_CONST_INLINE 385 +#define _LOAD_CONST_INLINE_BORROW 386 +#define _LOAD_CONST_INLINE_BORROW_WITH_NULL 387 +#define _LOAD_CONST_INLINE_WITH_NULL 388 #define _LOAD_DEREF LOAD_DEREF -#define _LOAD_FAST 386 -#define _LOAD_FAST_0 387 -#define _LOAD_FAST_1 388 -#define _LOAD_FAST_2 389 -#define _LOAD_FAST_3 390 -#define _LOAD_FAST_4 391 -#define _LOAD_FAST_5 392 -#define _LOAD_FAST_6 393 -#define _LOAD_FAST_7 394 +#define _LOAD_FAST 389 +#define _LOAD_FAST_0 390 +#define _LOAD_FAST_1 391 +#define _LOAD_FAST_2 392 +#define _LOAD_FAST_3 393 +#define _LOAD_FAST_4 394 +#define _LOAD_FAST_5 395 +#define _LOAD_FAST_6 396 +#define _LOAD_FAST_7 397 #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 395 -#define _LOAD_GLOBAL_BUILTINS 396 -#define _LOAD_GLOBAL_MODULE 397 +#define _LOAD_GLOBAL 398 +#define _LOAD_GLOBAL_BUILTINS 399 +#define _LOAD_GLOBAL_MODULE 400 #define _LOAD_LOCALS LOAD_LOCALS #define _LOAD_NAME LOAD_NAME #define _LOAD_SUPER_ATTR_ATTR LOAD_SUPER_ATTR_ATTR @@ -212,46 +212,46 @@ extern "C" { #define _MATCH_SEQUENCE MATCH_SEQUENCE #define _NOP NOP #define _POP_EXCEPT POP_EXCEPT -#define _POP_FRAME 398 -#define _POP_JUMP_IF_FALSE 399 -#define _POP_JUMP_IF_TRUE 400 +#define _POP_FRAME 401 +#define _POP_JUMP_IF_FALSE 402 +#define _POP_JUMP_IF_TRUE 403 #define _POP_TOP POP_TOP #define _PUSH_EXC_INFO PUSH_EXC_INFO -#define _PUSH_FRAME 401 +#define _PUSH_FRAME 404 #define _PUSH_NULL PUSH_NULL #define _RESUME_CHECK RESUME_CHECK -#define _SAVE_RETURN_OFFSET 402 -#define _SEND 403 +#define _SAVE_RETURN_OFFSET 405 +#define _SEND 406 #define _SEND_GEN SEND_GEN #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 404 -#define _STORE_ATTR 405 -#define _STORE_ATTR_INSTANCE_VALUE 406 -#define _STORE_ATTR_SLOT 407 +#define _START_EXECUTOR 407 +#define _STORE_ATTR 408 +#define _STORE_ATTR_INSTANCE_VALUE 409 +#define _STORE_ATTR_SLOT 410 #define _STORE_ATTR_WITH_HINT STORE_ATTR_WITH_HINT #define _STORE_DEREF STORE_DEREF -#define _STORE_FAST 408 -#define _STORE_FAST_0 409 -#define _STORE_FAST_1 410 -#define _STORE_FAST_2 411 -#define _STORE_FAST_3 412 -#define _STORE_FAST_4 413 -#define _STORE_FAST_5 414 -#define _STORE_FAST_6 415 -#define _STORE_FAST_7 416 +#define _STORE_FAST 411 +#define _STORE_FAST_0 412 +#define _STORE_FAST_1 413 +#define _STORE_FAST_2 414 +#define _STORE_FAST_3 415 +#define _STORE_FAST_4 416 +#define _STORE_FAST_5 417 +#define _STORE_FAST_6 418 +#define _STORE_FAST_7 419 #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 STORE_SLICE -#define _STORE_SUBSCR 417 +#define _STORE_SUBSCR 420 #define _STORE_SUBSCR_DICT STORE_SUBSCR_DICT #define _STORE_SUBSCR_LIST_INT STORE_SUBSCR_LIST_INT #define _SWAP SWAP -#define _TO_BOOL 418 +#define _TO_BOOL 421 #define _TO_BOOL_ALWAYS_TRUE TO_BOOL_ALWAYS_TRUE #define _TO_BOOL_BOOL TO_BOOL_BOOL #define _TO_BOOL_INT TO_BOOL_INT @@ -262,12 +262,12 @@ extern "C" { #define _UNARY_NEGATIVE UNARY_NEGATIVE #define _UNARY_NOT UNARY_NOT #define _UNPACK_EX UNPACK_EX -#define _UNPACK_SEQUENCE 419 +#define _UNPACK_SEQUENCE 422 #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 MAX_UOP_ID 419 +#define MAX_UOP_ID 422 #ifdef __cplusplus } diff --git a/Include/internal/pycore_uop_metadata.h b/Include/internal/pycore_uop_metadata.h index c9def0e..35340fe 100644 --- a/Include/internal/pycore_uop_metadata.h +++ b/Include/internal/pycore_uop_metadata.h @@ -149,9 +149,9 @@ const uint16_t _PyUop_Flags[MAX_UOP_ID+1] = { [_STORE_ATTR_INSTANCE_VALUE] = HAS_ESCAPES_FLAG, [_STORE_ATTR_SLOT] = HAS_ESCAPES_FLAG, [_COMPARE_OP] = HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG, - [_COMPARE_OP_FLOAT] = HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG, + [_COMPARE_OP_FLOAT] = HAS_ARG_FLAG | HAS_ESCAPES_FLAG, [_COMPARE_OP_INT] = HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG, - [_COMPARE_OP_STR] = HAS_ARG_FLAG | HAS_DEOPT_FLAG | HAS_ESCAPES_FLAG, + [_COMPARE_OP_STR] = HAS_ARG_FLAG | HAS_ESCAPES_FLAG, [_IS_OP] = HAS_ARG_FLAG, [_CONTAINS_OP] = HAS_ARG_FLAG | HAS_ERROR_FLAG | HAS_ESCAPES_FLAG, [_CHECK_EG_MATCH] = HAS_ERROR_FLAG | HAS_ESCAPES_FLAG, diff --git a/Lib/test/test_capi/test_opt.py b/Lib/test/test_capi/test_opt.py index 9d19b6c..3ba38c7 100644 --- a/Lib/test/test_capi/test_opt.py +++ b/Lib/test/test_capi/test_opt.py @@ -574,7 +574,7 @@ class TestUopsOptimization(unittest.TestCase): def test_int_type_propagation(self): def testfunc(loops): num = 0 - while num < loops: + for i in range(loops): x = num + num a = x + 1 num += 1 @@ -593,7 +593,7 @@ class TestUopsOptimization(unittest.TestCase): return x + x def testfunc(loops): num = 0 - while num < loops: + for i in range(loops): x = num + num a = double(x) num += 1 @@ -617,7 +617,7 @@ class TestUopsOptimization(unittest.TestCase): return x + x def testfunc(loops): num = 0 - while num < loops: + for i in range(loops): a = double(num) x = a + a num += 1 @@ -821,6 +821,59 @@ class TestUopsOptimization(unittest.TestCase): # We'll also need to verify that propagation actually occurs. self.assertIn("_BINARY_OP_MULTIPLY_FLOAT", uops) + def test_compare_op_type_propagation_float(self): + def testfunc(n): + a = 1.0 + for _ in range(n): + x = a == a + x = a == a + x = a == a + x = a == a + return x + + res, ex = self._run_with_optimizer(testfunc, 32) + self.assertTrue(res) + self.assertIsNotNone(ex) + uops = {opname for opname, _, _ in ex} + guard_both_float_count = [opname for opname, _, _ in ex if opname == "_GUARD_BOTH_FLOAT"] + self.assertLessEqual(len(guard_both_float_count), 1) + self.assertIn("_COMPARE_OP_FLOAT", uops) + + def test_compare_op_type_propagation_int(self): + def testfunc(n): + a = 1 + for _ in range(n): + x = a == a + x = a == a + x = a == a + x = a == a + return x + + res, ex = self._run_with_optimizer(testfunc, 32) + self.assertTrue(res) + self.assertIsNotNone(ex) + uops = {opname for opname, _, _ in ex} + guard_both_float_count = [opname for opname, _, _ in ex if opname == "_GUARD_BOTH_INT"] + self.assertLessEqual(len(guard_both_float_count), 1) + self.assertIn("_COMPARE_OP_INT", uops) + + def test_compare_op_type_propagation_unicode(self): + def testfunc(n): + a = "" + for _ in range(n): + x = a == a + x = a == a + x = a == a + x = a == a + return x + + res, ex = self._run_with_optimizer(testfunc, 32) + self.assertTrue(res) + self.assertIsNotNone(ex) + uops = {opname for opname, _, _ in ex} + guard_both_float_count = [opname for opname, _, _ in ex if opname == "_GUARD_BOTH_UNICODE"] + self.assertLessEqual(len(guard_both_float_count), 1) + self.assertIn("_COMPARE_OP_STR", uops) if __name__ == "__main__": unittest.main() diff --git a/Python/bytecodes.c b/Python/bytecodes.c index 27c439b..10bb152 100644 --- a/Python/bytecodes.c +++ b/Python/bytecodes.c @@ -2200,9 +2200,16 @@ dummy_func( macro(COMPARE_OP) = _SPECIALIZE_COMPARE_OP + _COMPARE_OP; - inst(COMPARE_OP_FLOAT, (unused/1, left, right -- res)) { - DEOPT_IF(!PyFloat_CheckExact(left)); - DEOPT_IF(!PyFloat_CheckExact(right)); + macro(COMPARE_OP_FLOAT) = + _GUARD_BOTH_FLOAT + unused/1 + _COMPARE_OP_FLOAT; + + macro(COMPARE_OP_INT) = + _GUARD_BOTH_INT + unused/1 + _COMPARE_OP_INT; + + macro(COMPARE_OP_STR) = + _GUARD_BOTH_UNICODE + unused/1 + _COMPARE_OP_STR; + + op(_COMPARE_OP_FLOAT, (left, right -- res)) { STAT_INC(COMPARE_OP, hit); double dleft = PyFloat_AS_DOUBLE(left); double dright = PyFloat_AS_DOUBLE(right); @@ -2215,9 +2222,7 @@ dummy_func( } // Similar to COMPARE_OP_FLOAT - inst(COMPARE_OP_INT, (unused/1, left, right -- res)) { - DEOPT_IF(!PyLong_CheckExact(left)); - DEOPT_IF(!PyLong_CheckExact(right)); + op(_COMPARE_OP_INT, (left, right -- res)) { DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)left)); DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)right)); STAT_INC(COMPARE_OP, hit); @@ -2234,9 +2239,7 @@ dummy_func( } // Similar to COMPARE_OP_FLOAT, but for ==, != only - inst(COMPARE_OP_STR, (unused/1, left, right -- res)) { - DEOPT_IF(!PyUnicode_CheckExact(left)); - DEOPT_IF(!PyUnicode_CheckExact(right)); + op(_COMPARE_OP_STR, (left, right -- res)) { STAT_INC(COMPARE_OP, hit); int eq = _PyUnicode_Equal(left, right); assert((oparg >> 5) == Py_EQ || (oparg >> 5) == Py_NE); diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h index b46885e..445f98b 100644 --- a/Python/executor_cases.c.h +++ b/Python/executor_cases.c.h @@ -2100,8 +2100,6 @@ oparg = CURRENT_OPARG(); right = stack_pointer[-1]; left = stack_pointer[-2]; - if (!PyFloat_CheckExact(left)) goto deoptimize; - if (!PyFloat_CheckExact(right)) goto deoptimize; STAT_INC(COMPARE_OP, hit); double dleft = PyFloat_AS_DOUBLE(left); double dright = PyFloat_AS_DOUBLE(right); @@ -2123,8 +2121,6 @@ oparg = CURRENT_OPARG(); right = stack_pointer[-1]; left = stack_pointer[-2]; - if (!PyLong_CheckExact(left)) goto deoptimize; - if (!PyLong_CheckExact(right)) goto deoptimize; if (!_PyLong_IsCompact((PyLongObject *)left)) goto deoptimize; if (!_PyLong_IsCompact((PyLongObject *)right)) goto deoptimize; STAT_INC(COMPARE_OP, hit); @@ -2150,8 +2146,6 @@ oparg = CURRENT_OPARG(); right = stack_pointer[-1]; left = stack_pointer[-2]; - if (!PyUnicode_CheckExact(left)) goto deoptimize; - if (!PyUnicode_CheckExact(right)) goto deoptimize; STAT_INC(COMPARE_OP, hit); int eq = _PyUnicode_Equal(left, right); assert((oparg >> 5) == Py_EQ || (oparg >> 5) == Py_NE); diff --git a/Python/generated_cases.c.h b/Python/generated_cases.c.h index 324e53d..7899106 100644 --- a/Python/generated_cases.c.h +++ b/Python/generated_cases.c.h @@ -2026,20 +2026,26 @@ PyObject *right; PyObject *left; PyObject *res; - /* Skip 1 cache entry */ + // _GUARD_BOTH_FLOAT right = stack_pointer[-1]; left = stack_pointer[-2]; - DEOPT_IF(!PyFloat_CheckExact(left), COMPARE_OP); - DEOPT_IF(!PyFloat_CheckExact(right), COMPARE_OP); - STAT_INC(COMPARE_OP, hit); - double dleft = PyFloat_AS_DOUBLE(left); - double dright = PyFloat_AS_DOUBLE(right); - // 1 if NaN, 2 if <, 4 if >, 8 if ==; this matches low four bits of the oparg - int sign_ish = COMPARISON_BIT(dleft, dright); - _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc); - _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc); - res = (sign_ish & oparg) ? Py_True : Py_False; - // It's always a bool, so we don't care about oparg & 16. + { + DEOPT_IF(!PyFloat_CheckExact(left), COMPARE_OP); + DEOPT_IF(!PyFloat_CheckExact(right), COMPARE_OP); + } + /* Skip 1 cache entry */ + // _COMPARE_OP_FLOAT + { + STAT_INC(COMPARE_OP, hit); + double dleft = PyFloat_AS_DOUBLE(left); + double dright = PyFloat_AS_DOUBLE(right); + // 1 if NaN, 2 if <, 4 if >, 8 if ==; this matches low four bits of the oparg + int sign_ish = COMPARISON_BIT(dleft, dright); + _Py_DECREF_SPECIALIZED(left, _PyFloat_ExactDealloc); + _Py_DECREF_SPECIALIZED(right, _PyFloat_ExactDealloc); + res = (sign_ish & oparg) ? Py_True : Py_False; + // It's always a bool, so we don't care about oparg & 16. + } stack_pointer[-2] = res; stack_pointer += -1; DISPATCH(); @@ -2053,24 +2059,30 @@ PyObject *right; PyObject *left; PyObject *res; - /* Skip 1 cache entry */ + // _GUARD_BOTH_INT right = stack_pointer[-1]; left = stack_pointer[-2]; - DEOPT_IF(!PyLong_CheckExact(left), COMPARE_OP); - DEOPT_IF(!PyLong_CheckExact(right), COMPARE_OP); - DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)left), COMPARE_OP); - DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)right), COMPARE_OP); - STAT_INC(COMPARE_OP, hit); - assert(_PyLong_DigitCount((PyLongObject *)left) <= 1 && + { + DEOPT_IF(!PyLong_CheckExact(left), COMPARE_OP); + DEOPT_IF(!PyLong_CheckExact(right), COMPARE_OP); + } + /* Skip 1 cache entry */ + // _COMPARE_OP_INT + { + DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)left), COMPARE_OP); + DEOPT_IF(!_PyLong_IsCompact((PyLongObject *)right), COMPARE_OP); + STAT_INC(COMPARE_OP, hit); + assert(_PyLong_DigitCount((PyLongObject *)left) <= 1 && _PyLong_DigitCount((PyLongObject *)right) <= 1); - Py_ssize_t ileft = _PyLong_CompactValue((PyLongObject *)left); - Py_ssize_t iright = _PyLong_CompactValue((PyLongObject *)right); - // 2 if <, 4 if >, 8 if ==; this matches the low 4 bits of the oparg - int sign_ish = COMPARISON_BIT(ileft, iright); - _Py_DECREF_SPECIALIZED(left, (destructor)PyObject_Free); - _Py_DECREF_SPECIALIZED(right, (destructor)PyObject_Free); - res = (sign_ish & oparg) ? Py_True : Py_False; - // It's always a bool, so we don't care about oparg & 16. + Py_ssize_t ileft = _PyLong_CompactValue((PyLongObject *)left); + Py_ssize_t iright = _PyLong_CompactValue((PyLongObject *)right); + // 2 if <, 4 if >, 8 if ==; this matches the low 4 bits of the oparg + int sign_ish = COMPARISON_BIT(ileft, iright); + _Py_DECREF_SPECIALIZED(left, (destructor)PyObject_Free); + _Py_DECREF_SPECIALIZED(right, (destructor)PyObject_Free); + res = (sign_ish & oparg) ? Py_True : Py_False; + // It's always a bool, so we don't care about oparg & 16. + } stack_pointer[-2] = res; stack_pointer += -1; DISPATCH(); @@ -2084,21 +2096,27 @@ PyObject *right; PyObject *left; PyObject *res; - /* Skip 1 cache entry */ + // _GUARD_BOTH_UNICODE right = stack_pointer[-1]; left = stack_pointer[-2]; - DEOPT_IF(!PyUnicode_CheckExact(left), COMPARE_OP); - DEOPT_IF(!PyUnicode_CheckExact(right), COMPARE_OP); - STAT_INC(COMPARE_OP, hit); - int eq = _PyUnicode_Equal(left, right); - assert((oparg >> 5) == Py_EQ || (oparg >> 5) == Py_NE); - _Py_DECREF_SPECIALIZED(left, _PyUnicode_ExactDealloc); - _Py_DECREF_SPECIALIZED(right, _PyUnicode_ExactDealloc); - assert(eq == 0 || eq == 1); - assert((oparg & 0xf) == COMPARISON_NOT_EQUALS || (oparg & 0xf) == COMPARISON_EQUALS); - assert(COMPARISON_NOT_EQUALS + 1 == COMPARISON_EQUALS); - res = ((COMPARISON_NOT_EQUALS + eq) & oparg) ? Py_True : Py_False; - // It's always a bool, so we don't care about oparg & 16. + { + DEOPT_IF(!PyUnicode_CheckExact(left), COMPARE_OP); + DEOPT_IF(!PyUnicode_CheckExact(right), COMPARE_OP); + } + /* Skip 1 cache entry */ + // _COMPARE_OP_STR + { + STAT_INC(COMPARE_OP, hit); + int eq = _PyUnicode_Equal(left, right); + assert((oparg >> 5) == Py_EQ || (oparg >> 5) == Py_NE); + _Py_DECREF_SPECIALIZED(left, _PyUnicode_ExactDealloc); + _Py_DECREF_SPECIALIZED(right, _PyUnicode_ExactDealloc); + assert(eq == 0 || eq == 1); + assert((oparg & 0xf) == COMPARISON_NOT_EQUALS || (oparg & 0xf) == COMPARISON_EQUALS); + assert(COMPARISON_NOT_EQUALS + 1 == COMPARISON_EQUALS); + res = ((COMPARISON_NOT_EQUALS + eq) & oparg) ? Py_True : Py_False; + // It's always a bool, so we don't care about oparg & 16. + } stack_pointer[-2] = res; stack_pointer += -1; DISPATCH(); diff --git a/Python/tier2_redundancy_eliminator_bytecodes.c b/Python/tier2_redundancy_eliminator_bytecodes.c index 3f6e8ce..e9b556d 100644 --- a/Python/tier2_redundancy_eliminator_bytecodes.c +++ b/Python/tier2_redundancy_eliminator_bytecodes.c @@ -77,6 +77,14 @@ dummy_func(void) { sym_set_type(right, &PyFloat_Type); } + op(_GUARD_BOTH_UNICODE, (left, right -- left, right)) { + if (sym_matches_type(left, &PyUnicode_Type) && + sym_matches_type(right, &PyUnicode_Type)) { + REPLACE_OP(this_instr, _NOP, 0 ,0); + } + sym_set_type(left, &PyUnicode_Type); + sym_set_type(right, &PyUnicode_Type); + } op(_BINARY_OP_ADD_INT, (left, right -- res)) { if (is_const(left) && is_const(right)) { diff --git a/Python/tier2_redundancy_eliminator_cases.c.h b/Python/tier2_redundancy_eliminator_cases.c.h index 904700a..f41fe32 100644 --- a/Python/tier2_redundancy_eliminator_cases.c.h +++ b/Python/tier2_redundancy_eliminator_cases.c.h @@ -351,6 +351,16 @@ } case _GUARD_BOTH_UNICODE: { + _Py_UOpsSymType *right; + _Py_UOpsSymType *left; + right = stack_pointer[-1]; + left = stack_pointer[-2]; + if (sym_matches_type(left, &PyUnicode_Type) && + sym_matches_type(right, &PyUnicode_Type)) { + REPLACE_OP(this_instr, _NOP, 0 ,0); + } + sym_set_type(left, &PyUnicode_Type); + sym_set_type(right, &PyUnicode_Type); break; } |