{{ message }}
forked from python/cpython
-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathceval_macros.h
More file actions
478 lines (407 loc) · 17 KB
/
Copy pathceval_macros.h
File metadata and controls
478 lines (407 loc) · 17 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
// Macros and other things needed by ceval.c, and bytecodes.c
/* Computed GOTOs, or
the-optimization-commonly-but-improperly-known-as-"threaded code"
using gcc's labels-as-values extension
(http://gcc.gnu.org/onlinedocs/gcc/Labels-as-Values.html).
The traditional bytecode evaluation loop uses a "switch" statement, which
decent compilers will optimize as a single indirect branch instruction
combined with a lookup table of jump addresses. However, since the
indirect jump instruction is shared by all opcodes, the CPU will have a
hard time making the right prediction for where to jump next (actually,
it will be always wrong except in the uncommon case of a sequence of
several identical opcodes).
"Threaded code" in contrast, uses an explicit jump table and an explicit
indirect jump instruction at the end of each opcode. Since the jump
instruction is at a different address for each opcode, the CPU will make a
separate prediction for each of these instructions, which is equivalent to
predicting the second opcode of each opcode pair. These predictions have
a much better chance to turn out valid, especially in small bytecode loops.
A mispredicted branch on a modern CPU flushes the whole pipeline and
can cost several CPU cycles (depending on the pipeline depth),
and potentially many more instructions (depending on the pipeline width).
A correctly predicted branch, however, is nearly free.
At the time of this writing, the "threaded code" version is up to 15-20%
faster than the normal "switch" version, depending on the compiler and the
CPU architecture.
NOTE: care must be taken that the compiler doesn't try to "optimize" the
indirect jumps by sharing them between all opcodes. Such optimizations
can be disabled on gcc by using the -fno-gcse flag (or possibly
-fno-crossjumping).
*/
/* Use macros rather than inline functions, to make it as clear as possible
* to the C compiler that the tracing check is a simple test then branch.
* We want to be sure that the compiler knows this before it generates
* the CFG.
*/
#ifdef WITH_DTRACE
#define OR_DTRACE_LINE | (PyDTrace_LINE_ENABLED() ? 255 : 0)
#else
#define OR_DTRACE_LINE
#endif
#ifdef HAVE_COMPUTED_GOTOS
#ifndef USE_COMPUTED_GOTOS
#define USE_COMPUTED_GOTOS 1
#endif
#else
#if defined(USE_COMPUTED_GOTOS) && USE_COMPUTED_GOTOS
#error "Computed gotos are not supported on this compiler."
#endif
#undef USE_COMPUTED_GOTOS
#define USE_COMPUTED_GOTOS 0
#endif
#ifdef Py_STATS
#define INSTRUCTION_STATS(op) \
do { \
OPCODE_EXE_INC(op); \
if (_Py_stats) _Py_stats->opcode_stats[lastopcode].pair_count[op]++; \
lastopcode = op; \
} while (0)
#else
#define INSTRUCTION_STATS(op) ((void)0)
#endif
#if USE_COMPUTED_GOTOS
# define TARGET(op) TARGET_##op:
# define DISPATCH_GOTO() goto *opcode_targets[opcode]
#else
# define TARGET(op) case op: TARGET_##op:
# define DISPATCH_GOTO() goto dispatch_opcode
#endif
/* PRE_DISPATCH_GOTO() does lltrace if enabled. Normally a no-op */
#ifdef LLTRACE
#define PRE_DISPATCH_GOTO() if (frame->lltrace >= 5) { \
lltrace_instruction(frame, stack_pointer, next_instr, opcode, oparg); }
#else
#define PRE_DISPATCH_GOTO() ((void)0)
#endif
#if LLTRACE
#define LLTRACE_RESUME_FRAME() \
do { \
int lltrace = maybe_lltrace_resume_frame(frame, GLOBALS()); \
frame->lltrace = lltrace; \
if (lltrace < 0) { \
goto exit_unwind; \
} \
} while (0)
#else
#define LLTRACE_RESUME_FRAME() ((void)0)
#endif
#ifdef Py_GIL_DISABLED
#define QSBR_QUIESCENT_STATE(tstate) _Py_qsbr_quiescent_state(((_PyThreadStateImpl *)tstate)->qsbr)
#else
#define QSBR_QUIESCENT_STATE(tstate)
#endif
/* Do interpreter dispatch accounting for tracing and instrumentation */
#define DISPATCH() \
{ \
assert(frame->stackpointer == NULL); \
NEXTOPARG(); \
PRE_DISPATCH_GOTO(); \
DISPATCH_GOTO(); \
}
#define DISPATCH_SAME_OPARG() \
{ \
opcode = next_instr->op.code; \
PRE_DISPATCH_GOTO(); \
DISPATCH_GOTO(); \
}
#define DISPATCH_INLINED(NEW_FRAME) \
do { \
assert(tstate->interp->eval_frame == NULL); \
_PyFrame_SetStackPointer(frame, stack_pointer); \
assert((NEW_FRAME)->previous == frame); \
frame = tstate->current_frame = (NEW_FRAME); \
CALL_STAT_INC(inlined_py_calls); \
goto start_frame; \
} while (0)
// Use this instead of 'goto error' so Tier 2 can go to a different label
#define GOTO_ERROR(LABEL) goto LABEL
/* Tuple access macros */
#ifndef Py_DEBUG
#define GETITEM(v, i) PyTuple_GET_ITEM((v), (i))
#else
static inline PyObject *
GETITEM(PyObject *v, Py_ssize_t i) {
assert(PyTuple_Check(v));
assert(i >= 0);
assert(i < PyTuple_GET_SIZE(v));
return PyTuple_GET_ITEM(v, i);
}
#endif
/* Code access macros */
/* The integer overflow is checked by an assertion below. */
#define INSTR_OFFSET() ((int)(next_instr - _PyFrame_GetBytecode(frame)))
#define NEXTOPARG() do { \
_Py_CODEUNIT word = {.cache = FT_ATOMIC_LOAD_UINT16_RELAXED(*(uint16_t*)next_instr)}; \
opcode = word.op.code; \
oparg = word.op.arg; \
} while (0)
/* JUMPBY makes the generator identify the instruction as a jump. SKIP_OVER is
* for advancing to the next instruction, taking into account cache entries
* and skipped instructions.
*/
#define JUMPBY(x) (next_instr += (x))
#define SKIP_OVER(x) (next_instr += (x))
/* OpCode prediction macros
Some opcodes tend to come in pairs thus making it possible to
predict the second code when the first is run. For example,
COMPARE_OP is often followed by POP_JUMP_IF_FALSE or POP_JUMP_IF_TRUE.
Verifying the prediction costs a single high-speed test of a register
variable against a constant. If the pairing was good, then the
processor's own internal branch predication has a high likelihood of
success, resulting in a nearly zero-overhead transition to the
next opcode. A successful prediction saves a trip through the eval-loop
including its unpredictable switch-case branch. Combined with the
processor's internal branch prediction, a successful PREDICT has the
effect of making the two opcodes run as if they were a single new opcode
with the bodies combined.
If collecting opcode statistics, your choices are to either keep the
predictions turned-on and interpret the results as if some opcodes
had been combined or turn-off predictions so that the opcode frequency
counter updates for both opcodes.
Opcode prediction is disabled with threaded code, since the latter allows
the CPU to record separate branch prediction information for each
opcode.
*/
#define PREDICT_ID(op) PRED_##op
#define PREDICTED(op) PREDICT_ID(op):
/* Stack manipulation macros */
/* The stack can grow at most MAXINT deep, as co_nlocals and
co_stacksize are ints. */
#define STACK_LEVEL() ((int)(stack_pointer - _PyFrame_Stackbase(frame)))
#define STACK_SIZE() (_PyFrame_GetCode(frame)->co_stacksize)
#define EMPTY() (STACK_LEVEL() == 0)
#define TOP() (stack_pointer[-1])
#define SECOND() (stack_pointer[-2])
#define THIRD() (stack_pointer[-3])
#define FOURTH() (stack_pointer[-4])
#define PEEK(n) (stack_pointer[-(n)])
#define POKE(n, v) (stack_pointer[-(n)] = (v))
#define SET_TOP(v) (stack_pointer[-1] = (v))
#define SET_SECOND(v) (stack_pointer[-2] = (v))
#define BASIC_STACKADJ(n) (stack_pointer += n)
#define BASIC_PUSH(v) (*stack_pointer++ = (v))
#define BASIC_POP() (*--stack_pointer)
#ifdef Py_DEBUG
#define PUSH(v) do { \
BASIC_PUSH(v); \
assert(STACK_LEVEL() <= STACK_SIZE()); \
} while (0)
#define POP() (assert(STACK_LEVEL() > 0), BASIC_POP())
#define STACK_GROW(n) do { \
assert(n >= 0); \
BASIC_STACKADJ(n); \
assert(STACK_LEVEL() <= STACK_SIZE()); \
} while (0)
#define STACK_SHRINK(n) do { \
assert(n >= 0); \
assert(STACK_LEVEL() >= n); \
BASIC_STACKADJ(-(n)); \
} while (0)
#else
#define PUSH(v) BASIC_PUSH(v)
#define POP() BASIC_POP()
#define STACK_GROW(n) BASIC_STACKADJ(n)
#define STACK_SHRINK(n) BASIC_STACKADJ(-(n))
#endif
#define WITHIN_STACK_BOUNDS() \
(frame->owner == FRAME_OWNED_BY_INTERPRETER || (STACK_LEVEL() >= 0 && STACK_LEVEL() <= STACK_SIZE()))
/* Data access macros */
#define FRAME_CO_CONSTS (_PyFrame_GetCode(frame)->co_consts)
#define FRAME_CO_NAMES (_PyFrame_GetCode(frame)->co_names)
/* Local variable macros */
#define LOCALS_ARRAY (frame->localsplus)
#define GETLOCAL(i) (frame->localsplus[i])
/* The SETLOCAL() macro must not DECREF the local variable in-place and
then store the new value; it must copy the old value to a temporary
value, then store the new value, and then DECREF the temporary value.
This is because it is possible that during the DECREF the frame is
accessed by other code (e.g. a __del__ method or gc.collect()) and the
variable would be pointing to already-freed memory. */
#define SETLOCAL(i, value) do { _PyStackRef tmp = GETLOCAL(i); \
GETLOCAL(i) = value; \
PyStackRef_XCLOSE(tmp); } while (0)
#define GO_TO_INSTRUCTION(op) goto PREDICT_ID(op)
#ifdef Py_STATS
#define UPDATE_MISS_STATS(INSTNAME) \
do { \
STAT_INC(opcode, miss); \
STAT_INC((INSTNAME), miss); \
/* The counter is always the first cache entry: */ \
if (ADAPTIVE_COUNTER_TRIGGERS(next_instr->cache)) { \
STAT_INC((INSTNAME), deopt); \
} \
} while (0)
#else
#define UPDATE_MISS_STATS(INSTNAME) ((void)0)
#endif
#define DEOPT_IF(COND, INSTNAME) \
if ((COND)) { \
/* This is only a single jump on release builds! */ \
UPDATE_MISS_STATS((INSTNAME)); \
assert(_PyOpcode_Deopt[opcode] == (INSTNAME)); \
GO_TO_INSTRUCTION(INSTNAME); \
}
// Try to lock an object in the free threading build, if it's not already
// locked. Use with a DEOPT_IF() to deopt if the object is already locked.
// These are no-ops in the default GIL build. The general pattern is:
//
// DEOPT_IF(!LOCK_OBJECT(op));
// if (/* condition fails */) {
// UNLOCK_OBJECT(op);
// DEOPT_IF(true);
// }
// ...
// UNLOCK_OBJECT(op);
//
// NOTE: The object must be unlocked on every exit code path and you should
// avoid any potentially escaping calls (like PyStackRef_CLOSE) while the
// object is locked.
#ifdef Py_GIL_DISABLED
# define LOCK_OBJECT(op) PyMutex_LockFast(&(_PyObject_CAST(op))->ob_mutex)
# define UNLOCK_OBJECT(op) PyMutex_Unlock(&(_PyObject_CAST(op))->ob_mutex)
#else
# define LOCK_OBJECT(op) (1)
# define UNLOCK_OBJECT(op) ((void)0)
#endif
#define GLOBALS() frame->f_globals
#define BUILTINS() frame->f_builtins
#define LOCALS() frame->f_locals
#define CONSTS() _PyFrame_GetCode(frame)->co_consts
#define NAMES() _PyFrame_GetCode(frame)->co_names
#define DTRACE_FUNCTION_ENTRY() \
if (PyDTrace_FUNCTION_ENTRY_ENABLED()) { \
dtrace_function_entry(frame); \
}
/* This takes a uint16_t instead of a _Py_BackoffCounter,
* because it is used directly on the cache entry in generated code,
* which is always an integral type. */
#define ADAPTIVE_COUNTER_TRIGGERS(COUNTER) \
backoff_counter_triggers(forge_backoff_counter((COUNTER)))
#define ADVANCE_ADAPTIVE_COUNTER(COUNTER) \
do { \
(COUNTER) = advance_backoff_counter((COUNTER)); \
} while (0);
#define PAUSE_ADAPTIVE_COUNTER(COUNTER) \
do { \
(COUNTER) = pause_backoff_counter((COUNTER)); \
} while (0);
#ifdef ENABLE_SPECIALIZATION_FT
/* Multiple threads may execute these concurrently if thread-local bytecode is
* disabled and they all execute the main copy of the bytecode. Specialization
* is disabled in that case so the value is unused, but the RMW cycle should be
* free of data races.
*/
#define RECORD_BRANCH_TAKEN(bitset, flag) \
FT_ATOMIC_STORE_UINT16_RELAXED( \
bitset, (FT_ATOMIC_LOAD_UINT16_RELAXED(bitset) << 1) | (flag))
#else
#define RECORD_BRANCH_TAKEN(bitset, flag)
#endif
#define UNBOUNDLOCAL_ERROR_MSG \
"cannot access local variable '%s' where it is not associated with a value"
#define UNBOUNDFREE_ERROR_MSG \
"cannot access free variable '%s' where it is not associated with a value" \
" in enclosing scope"
#define NAME_ERROR_MSG "name '%.200s' is not defined"
// If a trace function sets a new f_lineno and
// *then* raises, we use the destination when searching
// for an exception handler, displaying the traceback, and so on
#define INSTRUMENTED_JUMP(src, dest, event) \
do { \
if (tstate->tracing) {\
next_instr = dest; \
} else { \
_PyFrame_SetStackPointer(frame, stack_pointer); \
next_instr = _Py_call_instrumentation_jump(this_instr, tstate, event, frame, src, dest); \
stack_pointer = _PyFrame_GetStackPointer(frame); \
if (next_instr == NULL) { \
next_instr = (dest)+1; \
goto error; \
} \
} \
} while (0);
static inline int _Py_EnterRecursivePy(PyThreadState *tstate) {
return (tstate->py_recursion_remaining-- <= 0) &&
_Py_CheckRecursiveCallPy(tstate);
}
static inline void _Py_LeaveRecursiveCallPy(PyThreadState *tstate) {
tstate->py_recursion_remaining++;
}
/* Implementation of "macros" that modify the instruction pointer,
* stack pointer, or frame pointer.
* These need to treated differently by tier 1 and 2.
* The Tier 1 version is here; Tier 2 is inlined in ceval.c. */
#define LOAD_IP(OFFSET) do { \
next_instr = frame->instr_ptr + (OFFSET); \
} while (0)
/* There's no STORE_IP(), it's inlined by the code generator. */
#define LOAD_SP() \
stack_pointer = _PyFrame_GetStackPointer(frame)
#define SAVE_SP() \
_PyFrame_SetStackPointer(frame, stack_pointer)
/* Tier-switching macros. */
#ifdef _Py_JIT
#define GOTO_TIER_TWO(EXECUTOR) \
do { \
OPT_STAT_INC(traces_executed); \
jit_func jitted = (EXECUTOR)->jit_code; \
next_instr = jitted(frame, stack_pointer, tstate); \
Py_DECREF(tstate->previous_executor); \
tstate->previous_executor = NULL; \
frame = tstate->current_frame; \
if (next_instr == NULL) { \
goto resume_with_error; \
} \
stack_pointer = _PyFrame_GetStackPointer(frame); \
DISPATCH(); \
} while (0)
#else
#define GOTO_TIER_TWO(EXECUTOR) \
do { \
OPT_STAT_INC(traces_executed); \
next_uop = (EXECUTOR)->trace; \
assert(next_uop->opcode == _START_EXECUTOR); \
goto enter_tier_two; \
} while (0)
#endif
#define GOTO_TIER_ONE(TARGET) \
do { \
Py_DECREF(tstate->previous_executor); \
tstate->previous_executor = NULL; \
next_instr = target; \
DISPATCH(); \
} while (0)
#define CURRENT_OPARG() (next_uop[-1].oparg)
#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
#define GOTO_UNWIND() goto error_tier_two
#define EXIT_TO_TIER1() goto exit_to_tier1
#define EXIT_TO_TIER1_DYNAMIC() goto exit_to_tier1_dynamic;
/* Stackref macros */
/* How much scratch space to give stackref to PyObject* conversion. */
#define MAX_STACKREF_SCRATCH 10
#if defined(Py_GIL_DISABLED) || defined(Py_STACKREF_DEBUG)
#define STACKREFS_TO_PYOBJECTS(ARGS, ARG_COUNT, NAME) \
/* +1 because vectorcall might use -1 to write self */ \
PyObject *NAME##_temp[MAX_STACKREF_SCRATCH+1]; \
PyObject **NAME = _PyObjectArray_FromStackRefArray(ARGS, ARG_COUNT, NAME##_temp + 1);
#else
#define STACKREFS_TO_PYOBJECTS(ARGS, ARG_COUNT, NAME) \
PyObject **NAME = (PyObject **)ARGS; \
assert(NAME != NULL);
#endif
#if defined(Py_GIL_DISABLED) || defined(Py_STACKREF_DEBUG)
#define STACKREFS_TO_PYOBJECTS_CLEANUP(NAME) \
/* +1 because we +1 previously */ \
_PyObjectArray_Free(NAME - 1, NAME##_temp);
#else
#define STACKREFS_TO_PYOBJECTS_CLEANUP(NAME) \
(void)(NAME);
#endif
#if defined(Py_GIL_DISABLED) || defined(Py_STACKREF_DEBUG)
#define CONVERSION_FAILED(NAME) ((NAME) == NULL)
#else
#define CONVERSION_FAILED(NAME) (0)
#endif
You can’t perform that action at this time.