| 1 |
from dataclasses import dataclass |
| 2 |
import itertools |
| 3 |
import lexer |
| 4 |
import parser |
| 5 |
import re |
| 6 |
from typing import Optional, Callable |
| 7 |
|
| 8 |
from parser import Stmt, SimpleStmt, BlockStmt, IfStmt, WhileStmt |
| 9 |
|
| 10 |
@dataclass |
| 11 |
class EscapingCall: |
| 12 |
stmt: SimpleStmt |
| 13 |
call: lexer.Token |
| 14 |
kills: lexer.Token | None |
| 15 |
|
| 16 |
@dataclass |
| 17 |
class Properties: |
| 18 |
escaping_calls: dict[SimpleStmt, EscapingCall] |
| 19 |
escapes: bool |
| 20 |
error_with_pop: bool |
| 21 |
error_without_pop: bool |
| 22 |
deopts: bool |
| 23 |
oparg: bool |
| 24 |
jumps: bool |
| 25 |
eval_breaker: bool |
| 26 |
needs_this: bool |
| 27 |
always_exits: bool |
| 28 |
stores_sp: bool |
| 29 |
uses_co_consts: bool |
| 30 |
uses_co_names: bool |
| 31 |
uses_locals: bool |
| 32 |
has_free: bool |
| 33 |
side_exit: bool |
| 34 |
pure: bool |
| 35 |
uses_opcode: bool |
| 36 |
tier: int | None = None |
| 37 |
const_oparg: int = -1 |
| 38 |
needs_prev: bool = False |
| 39 |
no_save_ip: bool = False |
| 40 |
|
| 41 |
def dump(self, indent: str) -> None: |
| 42 |
simple_properties = self.__dict__.copy() |
| 43 |
del simple_properties["escaping_calls"] |
| 44 |
text = "escaping_calls:\n" |
| 45 |
for tkns in self.escaping_calls.values(): |
| 46 |
text += f"{indent} {tkns}\n" |
| 47 |
text += ", ".join([f"{key}: {value}" for (key, value) in simple_properties.items()]) |
| 48 |
print(indent, text, sep="") |
| 49 |
|
| 50 |
@staticmethod |
| 51 |
def from_list(properties: list["Properties"]) -> "Properties": |
| 52 |
escaping_calls: dict[SimpleStmt, EscapingCall] = {} |
| 53 |
for p in properties: |
| 54 |
escaping_calls.update(p.escaping_calls) |
| 55 |
return Properties( |
| 56 |
escaping_calls=escaping_calls, |
| 57 |
escapes = any(p.escapes for p in properties), |
| 58 |
error_with_pop=any(p.error_with_pop for p in properties), |
| 59 |
error_without_pop=any(p.error_without_pop for p in properties), |
| 60 |
deopts=any(p.deopts for p in properties), |
| 61 |
oparg=any(p.oparg for p in properties), |
| 62 |
jumps=any(p.jumps for p in properties), |
| 63 |
eval_breaker=any(p.eval_breaker for p in properties), |
| 64 |
needs_this=any(p.needs_this for p in properties), |
| 65 |
always_exits=any(p.always_exits for p in properties), |
| 66 |
stores_sp=any(p.stores_sp for p in properties), |
| 67 |
uses_co_consts=any(p.uses_co_consts for p in properties), |
| 68 |
uses_co_names=any(p.uses_co_names for p in properties), |
| 69 |
uses_locals=any(p.uses_locals for p in properties), |
| 70 |
uses_opcode=any(p.uses_opcode for p in properties), |
| 71 |
has_free=any(p.has_free for p in properties), |
| 72 |
side_exit=any(p.side_exit for p in properties), |
| 73 |
pure=all(p.pure for p in properties), |
| 74 |
needs_prev=any(p.needs_prev for p in properties), |
| 75 |
no_save_ip=all(p.no_save_ip for p in properties), |
| 76 |
) |
| 77 |
|
| 78 |
@property |
| 79 |
def infallible(self) -> bool: |
| 80 |
return not self.error_with_pop and not self.error_without_pop |
| 81 |
|
| 82 |
SKIP_PROPERTIES = Properties( |
| 83 |
escaping_calls={}, |
| 84 |
escapes=False, |
| 85 |
error_with_pop=False, |
| 86 |
error_without_pop=False, |
| 87 |
deopts=False, |
| 88 |
oparg=False, |
| 89 |
jumps=False, |
| 90 |
eval_breaker=False, |
| 91 |
needs_this=False, |
| 92 |
always_exits=False, |
| 93 |
stores_sp=False, |
| 94 |
uses_co_consts=False, |
| 95 |
uses_co_names=False, |
| 96 |
uses_locals=False, |
| 97 |
uses_opcode=False, |
| 98 |
has_free=False, |
| 99 |
side_exit=False, |
| 100 |
pure=True, |
| 101 |
no_save_ip=False, |
| 102 |
) |
| 103 |
|
| 104 |
|
| 105 |
@dataclass |
| 106 |
class Skip: |
| 107 |
"Unused cache entry" |
| 108 |
size: int |
| 109 |
|
| 110 |
@property |
| 111 |
def name(self) -> str: |
| 112 |
return f"unused/{self.size}" |
| 113 |
|
| 114 |
@property |
| 115 |
def properties(self) -> Properties: |
| 116 |
return SKIP_PROPERTIES |
| 117 |
|
| 118 |
|
| 119 |
class Flush: |
| 120 |
@property |
| 121 |
def properties(self) -> Properties: |
| 122 |
return SKIP_PROPERTIES |
| 123 |
|
| 124 |
@property |
| 125 |
def name(self) -> str: |
| 126 |
return "flush" |
| 127 |
|
| 128 |
@property |
| 129 |
def size(self) -> int: |
| 130 |
return 0 |
| 131 |
|
| 132 |
|
| 133 |
|
| 134 |
|
| 135 |
@dataclass |
| 136 |
class StackItem: |
| 137 |
name: str |
| 138 |
type: str | None |
| 139 |
size: str |
| 140 |
peek: bool = False |
| 141 |
used: bool = False |
| 142 |
|
| 143 |
def __str__(self) -> str: |
| 144 |
size = f"[{self.size}]" if self.size else "" |
| 145 |
type = "" if self.type is None else f"{self.type} " |
| 146 |
return f"{type}{self.name}{size} {self.peek}" |
| 147 |
|
| 148 |
def is_array(self) -> bool: |
| 149 |
return self.size != "" |
| 150 |
|
| 151 |
def get_size(self) -> str: |
| 152 |
return self.size if self.size else "1" |
| 153 |
|
| 154 |
|
| 155 |
@dataclass |
| 156 |
class StackEffect: |
| 157 |
inputs: list[StackItem] |
| 158 |
outputs: list[StackItem] |
| 159 |
|
| 160 |
def __str__(self) -> str: |
| 161 |
return f"({', '.join([str(i) for i in self.inputs])} -- {', '.join([str(i) for i in self.outputs])})" |
| 162 |
|
| 163 |
|
| 164 |
@dataclass |
| 165 |
class CacheEntry: |
| 166 |
name: str |
| 167 |
size: int |
| 168 |
|
| 169 |
def __str__(self) -> str: |
| 170 |
return f"{self.name}/{self.size}" |
| 171 |
|
| 172 |
|
| 173 |
@dataclass |
| 174 |
class Uop: |
| 175 |
name: str |
| 176 |
context: parser.Context | None |
| 177 |
annotations: list[str] |
| 178 |
stack: StackEffect |
| 179 |
caches: list[CacheEntry] |
| 180 |
local_stores: list[lexer.Token] |
| 181 |
body: BlockStmt |
| 182 |
properties: Properties |
| 183 |
_size: int = -1 |
| 184 |
implicitly_created: bool = False |
| 185 |
replicated = 0 |
| 186 |
replicates: "Uop | None" = None |
| 187 |
# Size of the instruction(s), only set for uops containing the INSTRUCTION_SIZE macro |
| 188 |
instruction_size: int | None = None |
| 189 |
|
| 190 |
def dump(self, indent: str) -> None: |
| 191 |
print( |
| 192 |
indent, self.name, ", ".join(self.annotations) if self.annotations else "" |
| 193 |
) |
| 194 |
print(indent, self.stack, ", ".join([str(c) for c in self.caches])) |
| 195 |
self.properties.dump(" " + indent) |
| 196 |
|
| 197 |
@property |
| 198 |
def size(self) -> int: |
| 199 |
if self._size < 0: |
| 200 |
self._size = sum(c.size for c in self.caches) |
| 201 |
return self._size |
| 202 |
|
| 203 |
def why_not_viable(self) -> str | None: |
| 204 |
if self.name == "_SAVE_RETURN_OFFSET": |
| 205 |
return None # Adjusts next_instr, but only in tier 1 code |
| 206 |
if "INSTRUMENTED" in self.name: |
| 207 |
return "is instrumented" |
| 208 |
if "replaced" in self.annotations: |
| 209 |
return "is replaced" |
| 210 |
if self.name in ("INTERPRETER_EXIT", "JUMP_BACKWARD"): |
| 211 |
return "has tier 1 control flow" |
| 212 |
if self.properties.needs_this: |
| 213 |
return "uses the 'this_instr' variable" |
| 214 |
if len([c for c in self.caches if c.name != "unused"]) > 2: |
| 215 |
return "has too many cache entries" |
| 216 |
if self.properties.error_with_pop and self.properties.error_without_pop: |
| 217 |
return "has both popping and not-popping errors" |
| 218 |
return None |
| 219 |
|
| 220 |
def is_viable(self) -> bool: |
| 221 |
return self.why_not_viable() is None |
| 222 |
|
| 223 |
def is_super(self) -> bool: |
| 224 |
for tkn in self.body.tokens(): |
| 225 |
if tkn.kind == "IDENTIFIER" and tkn.text == "oparg1": |
| 226 |
return True |
| 227 |
return False |
| 228 |
|
| 229 |
|
| 230 |
class Label: |
| 231 |
|
| 232 |
def __init__(self, name: str, spilled: bool, body: BlockStmt, properties: Properties): |
| 233 |
self.name = name |
| 234 |
self.spilled = spilled |
| 235 |
self.body = body |
| 236 |
self.properties = properties |
| 237 |
|
| 238 |
size:int = 0 |
| 239 |
local_stores: list[lexer.Token] = [] |
| 240 |
instruction_size = None |
| 241 |
|
| 242 |
def __str__(self) -> str: |
| 243 |
return f"label({self.name})" |
| 244 |
|
| 245 |
|
| 246 |
Part = Uop | Skip | Flush |
| 247 |
CodeSection = Uop | Label |
| 248 |
|
| 249 |
|
| 250 |
@dataclass |
| 251 |
class Instruction: |
| 252 |
where: lexer.Token |
| 253 |
name: str |
| 254 |
parts: list[Part] |
| 255 |
_properties: Properties | None |
| 256 |
is_target: bool = False |
| 257 |
family: Optional["Family"] = None |
| 258 |
opcode: int = -1 |
| 259 |
|
| 260 |
@property |
| 261 |
def properties(self) -> Properties: |
| 262 |
if self._properties is None: |
| 263 |
self._properties = self._compute_properties() |
| 264 |
return self._properties |
| 265 |
|
| 266 |
def _compute_properties(self) -> Properties: |
| 267 |
return Properties.from_list([part.properties for part in self.parts]) |
| 268 |
|
| 269 |
def dump(self, indent: str) -> None: |
| 270 |
print(indent, self.name, "=", ", ".join([part.name for part in self.parts])) |
| 271 |
self.properties.dump(" " + indent) |
| 272 |
|
| 273 |
@property |
| 274 |
def size(self) -> int: |
| 275 |
return 1 + sum(part.size for part in self.parts) |
| 276 |
|
| 277 |
def is_super(self) -> bool: |
| 278 |
if len(self.parts) != 1: |
| 279 |
return False |
| 280 |
uop = self.parts[0] |
| 281 |
if isinstance(uop, Uop): |
| 282 |
return uop.is_super() |
| 283 |
else: |
| 284 |
return False |
| 285 |
|
| 286 |
|
| 287 |
@dataclass |
| 288 |
class PseudoInstruction: |
| 289 |
name: str |
| 290 |
stack: StackEffect |
| 291 |
targets: list[Instruction] |
| 292 |
as_sequence: bool |
| 293 |
flags: list[str] |
| 294 |
opcode: int = -1 |
| 295 |
|
| 296 |
def dump(self, indent: str) -> None: |
| 297 |
print(indent, self.name, "->", " or ".join([t.name for t in self.targets])) |
| 298 |
|
| 299 |
@property |
| 300 |
def properties(self) -> Properties: |
| 301 |
return Properties.from_list([i.properties for i in self.targets]) |
| 302 |
|
| 303 |
|
| 304 |
@dataclass |
| 305 |
class Family: |
| 306 |
name: str |
| 307 |
size: str |
| 308 |
members: list[Instruction] |
| 309 |
|
| 310 |
def dump(self, indent: str) -> None: |
| 311 |
print(indent, self.name, "= ", ", ".join([m.name for m in self.members])) |
| 312 |
|
| 313 |
|
| 314 |
@dataclass |
| 315 |
class Analysis: |
| 316 |
instructions: dict[str, Instruction] |
| 317 |
uops: dict[str, Uop] |
| 318 |
families: dict[str, Family] |
| 319 |
pseudos: dict[str, PseudoInstruction] |
| 320 |
labels: dict[str, Label] |
| 321 |
opmap: dict[str, int] |
| 322 |
have_arg: int |
| 323 |
min_instrumented: int |
| 324 |
|
| 325 |
|
| 326 |
def analysis_error(message: str, tkn: lexer.Token) -> SyntaxError: |
| 327 |
# To do -- support file and line output |
| 328 |
# Construct a SyntaxError instance from message and token |
| 329 |
return lexer.make_syntax_error(message, tkn.filename, tkn.line, tkn.column, "") |
| 330 |
|
| 331 |
|
| 332 |
def override_error( |
| 333 |
name: str, |
| 334 |
context: parser.Context | None, |
| 335 |
prev_context: parser.Context | None, |
| 336 |
token: lexer.Token, |
| 337 |
) -> SyntaxError: |
| 338 |
return analysis_error( |
| 339 |
f"Duplicate definition of '{name}' @ {context} " |
| 340 |
f"previous definition @ {prev_context}", |
| 341 |
token, |
| 342 |
) |
| 343 |
|
| 344 |
|
| 345 |
def convert_stack_item( |
| 346 |
item: parser.StackEffect, replace_op_arg_1: str | None |
| 347 |
) -> StackItem: |
| 348 |
return StackItem(item.name, item.type, item.size) |
| 349 |
|
| 350 |
def check_unused(stack: list[StackItem], input_names: dict[str, lexer.Token]) -> None: |
| 351 |
"Unused items cannot be on the stack above used, non-peek items" |
| 352 |
seen_unused = False |
| 353 |
for item in reversed(stack): |
| 354 |
if item.name == "unused": |
| 355 |
seen_unused = True |
| 356 |
elif item.peek: |
| 357 |
break |
| 358 |
elif seen_unused: |
| 359 |
raise analysis_error(f"Cannot have used input '{item.name}' below an unused value on the stack", input_names[item.name]) |
| 360 |
|
| 361 |
|
| 362 |
def analyze_stack( |
| 363 |
op: parser.InstDef | parser.Pseudo, replace_op_arg_1: str | None = None |
| 364 |
) -> StackEffect: |
| 365 |
inputs: list[StackItem] = [ |
| 366 |
convert_stack_item(i, replace_op_arg_1) |
| 367 |
for i in op.inputs |
| 368 |
if isinstance(i, parser.StackEffect) |
| 369 |
] |
| 370 |
outputs: list[StackItem] = [ |
| 371 |
convert_stack_item(i, replace_op_arg_1) for i in op.outputs |
| 372 |
] |
| 373 |
# Mark variables with matching names at the base of the stack as "peek" |
| 374 |
modified = False |
| 375 |
input_names: dict[str, lexer.Token] = { i.name : i.first_token for i in op.inputs if i.name != "unused" } |
| 376 |
for input, output in itertools.zip_longest(inputs, outputs): |
| 377 |
if output is None: |
| 378 |
pass |
| 379 |
elif input is None: |
| 380 |
if output.name in input_names: |
| 381 |
raise analysis_error( |
| 382 |
f"Reuse of variable '{output.name}' at different stack location", |
| 383 |
input_names[output.name]) |
| 384 |
elif input.name == output.name: |
| 385 |
if not modified: |
| 386 |
input.peek = output.peek = True |
| 387 |
else: |
| 388 |
modified = True |
| 389 |
if output.name in input_names: |
| 390 |
raise analysis_error( |
| 391 |
f"Reuse of variable '{output.name}' at different stack location", |
| 392 |
input_names[output.name]) |
| 393 |
if isinstance(op, parser.InstDef): |
| 394 |
output_names = [out.name for out in outputs] |
| 395 |
for input in inputs: |
| 396 |
if ( |
| 397 |
variable_used(op, input.name) |
| 398 |
or variable_used(op, "DECREF_INPUTS") |
| 399 |
or (not input.peek and input.name in output_names) |
| 400 |
): |
| 401 |
input.used = True |
| 402 |
for output in outputs: |
| 403 |
if variable_used(op, output.name): |
| 404 |
output.used = True |
| 405 |
check_unused(inputs, input_names) |
| 406 |
return StackEffect(inputs, outputs) |
| 407 |
|
| 408 |
|
| 409 |
def analyze_caches(inputs: list[parser.InputEffect]) -> list[CacheEntry]: |
| 410 |
caches: list[parser.CacheEffect] = [ |
| 411 |
i for i in inputs if isinstance(i, parser.CacheEffect) |
| 412 |
] |
| 413 |
if caches: |
| 414 |
# Middle entries are allowed to be unused. Check first and last caches. |
| 415 |
for index in (0, -1): |
| 416 |
cache = caches[index] |
| 417 |
if cache.name == "unused": |
| 418 |
position = "First" if index == 0 else "Last" |
| 419 |
msg = f"{position} cache entry in op is unused. Move to enclosing macro." |
| 420 |
raise analysis_error(msg, cache.tokens[0]) |
| 421 |
return [CacheEntry(i.name, int(i.size)) for i in caches] |
| 422 |
|
| 423 |
|
| 424 |
def find_variable_stores(node: parser.InstDef) -> list[lexer.Token]: |
| 425 |
res: list[lexer.Token] = [] |
| 426 |
outnames = { out.name for out in node.outputs } |
| 427 |
innames = { out.name for out in node.inputs } |
| 428 |
|
| 429 |
def find_stores_in_tokens(tokens: list[lexer.Token], callback: Callable[[lexer.Token], None]) -> None: |
| 430 |
while tokens and tokens[0].kind == "COMMENT": |
| 431 |
tokens = tokens[1:] |
| 432 |
if len(tokens) < 4: |
| 433 |
return |
| 434 |
if tokens[1].kind == "EQUALS": |
| 435 |
if tokens[0].kind == "IDENTIFIER": |
| 436 |
name = tokens[0].text |
| 437 |
if name in outnames or name in innames: |
| 438 |
callback(tokens[0]) |
| 439 |
#Passing the address of a local is also a definition |
| 440 |
for idx, tkn in enumerate(tokens): |
| 441 |
if tkn.kind == "AND": |
| 442 |
name_tkn = tokens[idx+1] |
| 443 |
if name_tkn.text in outnames: |
| 444 |
callback(name_tkn) |
| 445 |
|
| 446 |
def visit(stmt: Stmt) -> None: |
| 447 |
if isinstance(stmt, IfStmt): |
| 448 |
def error(tkn: lexer.Token) -> None: |
| 449 |
raise analysis_error("Cannot define variable in 'if' condition", tkn) |
| 450 |
find_stores_in_tokens(stmt.condition, error) |
| 451 |
elif isinstance(stmt, SimpleStmt): |
| 452 |
find_stores_in_tokens(stmt.contents, res.append) |
| 453 |
|
| 454 |
node.block.accept(visit) |
| 455 |
return res |
| 456 |
|
| 457 |
|
| 458 |
#def analyze_deferred_refs(node: parser.InstDef) -> dict[lexer.Token, str | None]: |
| 459 |
#"""Look for PyStackRef_FromPyObjectNew() calls""" |
| 460 |
|
| 461 |
#def in_frame_push(idx: int) -> bool: |
| 462 |
#for tkn in reversed(node.block.tokens[: idx - 1]): |
| 463 |
#if tkn.kind in {"SEMI", "LBRACE", "RBRACE"}: |
| 464 |
#return False |
| 465 |
#if tkn.kind == "IDENTIFIER" and tkn.text == "_PyFrame_PushUnchecked": |
| 466 |
#return True |
| 467 |
#return False |
| 468 |
|
| 469 |
#refs: dict[lexer.Token, str | None] = {} |
| 470 |
#for idx, tkn in enumerate(node.block.tokens): |
| 471 |
#if tkn.kind != "IDENTIFIER" or tkn.text != "PyStackRef_FromPyObjectNew": |
| 472 |
#continue |
| 473 |
|
| 474 |
#if idx == 0 or node.block.tokens[idx - 1].kind != "EQUALS": |
| 475 |
#if in_frame_push(idx): |
| 476 |
## PyStackRef_FromPyObjectNew() is called in _PyFrame_PushUnchecked() |
| 477 |
#refs[tkn] = None |
| 478 |
#continue |
| 479 |
#raise analysis_error("Expected '=' before PyStackRef_FromPyObjectNew", tkn) |
| 480 |
|
| 481 |
#lhs = find_assignment_target(node, idx - 1) |
| 482 |
#if len(lhs) == 0: |
| 483 |
#raise analysis_error( |
| 484 |
#"PyStackRef_FromPyObjectNew() must be assigned to an output", tkn |
| 485 |
#) |
| 486 |
|
| 487 |
#if lhs[0].kind == "TIMES" or any( |
| 488 |
#t.kind == "ARROW" or t.kind == "LBRACKET" for t in lhs[1:] |
| 489 |
#): |
| 490 |
## Don't handle: *ptr = ..., ptr->field = ..., or ptr[field] = ... |
| 491 |
## Assume that they are visible to the GC. |
| 492 |
#refs[tkn] = None |
| 493 |
#continue |
| 494 |
|
| 495 |
#if len(lhs) != 1 or lhs[0].kind != "IDENTIFIER": |
| 496 |
#raise analysis_error( |
| 497 |
#"PyStackRef_FromPyObjectNew() must be assigned to an output", tkn |
| 498 |
#) |
| 499 |
|
| 500 |
#name = lhs[0].text |
| 501 |
#match = ( |
| 502 |
#any(var.name == name for var in node.inputs) |
| 503 |
#or any(var.name == name for var in node.outputs) |
| 504 |
#) |
| 505 |
#if not match: |
| 506 |
#raise analysis_error( |
| 507 |
#f"PyStackRef_FromPyObjectNew() must be assigned to an input or output, not '{name}'", |
| 508 |
#tkn, |
| 509 |
#) |
| 510 |
|
| 511 |
#refs[tkn] = name |
| 512 |
|
| 513 |
#return refs |
| 514 |
|
| 515 |
|
| 516 |
def variable_used(node: parser.CodeDef, name: str) -> bool: |
| 517 |
"""Determine whether a variable with a given name is used in a node.""" |
| 518 |
return any( |
| 519 |
token.kind == "IDENTIFIER" and token.text == name for token in node.block.tokens() |
| 520 |
) |
| 521 |
|
| 522 |
|
| 523 |
def oparg_used(node: parser.CodeDef) -> bool: |
| 524 |
"""Determine whether `oparg` is used in a node.""" |
| 525 |
return any( |
| 526 |
token.kind == "IDENTIFIER" and token.text == "oparg" for token in node.tokens |
| 527 |
) |
| 528 |
|
| 529 |
|
| 530 |
def tier_variable(node: parser.CodeDef) -> int | None: |
| 531 |
"""Determine whether a tier variable is used in a node.""" |
| 532 |
if isinstance(node, parser.LabelDef): |
| 533 |
return None |
| 534 |
for token in node.tokens: |
| 535 |
if token.kind == "ANNOTATION": |
| 536 |
if token.text == "specializing": |
| 537 |
return 1 |
| 538 |
if re.fullmatch(r"tier\d", token.text): |
| 539 |
return int(token.text[-1]) |
| 540 |
return None |
| 541 |
|
| 542 |
|
| 543 |
def has_error_with_pop(op: parser.CodeDef) -> bool: |
| 544 |
return ( |
| 545 |
variable_used(op, "ERROR_IF") |
| 546 |
or variable_used(op, "exception_unwind") |
| 547 |
) |
| 548 |
|
| 549 |
|
| 550 |
def has_error_without_pop(op: parser.CodeDef) -> bool: |
| 551 |
return ( |
| 552 |
variable_used(op, "ERROR_NO_POP") |
| 553 |
or variable_used(op, "exception_unwind") |
| 554 |
) |
| 555 |
|
| 556 |
|
| 557 |
NON_ESCAPING_FUNCTIONS = ( |
| 558 |
"PyCFunction_GET_FLAGS", |
| 559 |
"PyCFunction_GET_FUNCTION", |
| 560 |
"PyCFunction_GET_SELF", |
| 561 |
"PyCell_GetRef", |
| 562 |
"PyCell_New", |
| 563 |
"PyCell_SwapTakeRef", |
| 564 |
"PyExceptionInstance_Class", |
| 565 |
"PyException_GetCause", |
| 566 |
"PyException_GetContext", |
| 567 |
"PyException_GetTraceback", |
| 568 |
"PyFloat_AS_DOUBLE", |
| 569 |
"PyFloat_FromDouble", |
| 570 |
"PyFunction_GET_CODE", |
| 571 |
"PyFunction_GET_GLOBALS", |
| 572 |
"PyList_GET_ITEM", |
| 573 |
"PyList_GET_SIZE", |
| 574 |
"PyList_SET_ITEM", |
| 575 |
"PyLong_AsLong", |
| 576 |
"PyLong_FromLong", |
| 577 |
"PyLong_FromSsize_t", |
| 578 |
"PySlice_New", |
| 579 |
"PyStackRef_AsPyObjectBorrow", |
| 580 |
"PyStackRef_AsPyObjectNew", |
| 581 |
"PyStackRef_FromPyObjectNewMortal", |
| 582 |
"PyStackRef_AsPyObjectSteal", |
| 583 |
"PyStackRef_Borrow", |
| 584 |
"PyStackRef_CLEAR", |
| 585 |
"PyStackRef_CLOSE_SPECIALIZED", |
| 586 |
"PyStackRef_DUP", |
| 587 |
"PyStackRef_False", |
| 588 |
"PyStackRef_FromPyObjectImmortal", |
| 589 |
"PyStackRef_FromPyObjectNew", |
| 590 |
"PyStackRef_FromPyObjectSteal", |
| 591 |
"PyStackRef_IsExactly", |
| 592 |
"PyStackRef_FromPyObjectStealMortal", |
| 593 |
"PyStackRef_IsNone", |
| 594 |
"PyStackRef_Is", |
| 595 |
"PyStackRef_IsHeapSafe", |
| 596 |
"PyStackRef_IsTrue", |
| 597 |
"PyStackRef_IsFalse", |
| 598 |
"PyStackRef_IsNull", |
| 599 |
"PyStackRef_MakeHeapSafe", |
| 600 |
"PyStackRef_None", |
| 601 |
"PyStackRef_TYPE", |
| 602 |
"PyStackRef_True", |
| 603 |
"PyTuple_GET_ITEM", |
| 604 |
"PyTuple_GET_SIZE", |
| 605 |
"PyType_HasFeature", |
| 606 |
"PyUnicode_Concat", |
| 607 |
"PyUnicode_GET_LENGTH", |
| 608 |
"PyUnicode_READ_CHAR", |
| 609 |
"Py_ARRAY_LENGTH", |
| 610 |
"Py_FatalError", |
| 611 |
"Py_INCREF", |
| 612 |
"Py_IS_TYPE", |
| 613 |
"Py_NewRef", |
| 614 |
"Py_REFCNT", |
| 615 |
"Py_SIZE", |
| 616 |
"Py_TYPE", |
| 617 |
"Py_UNREACHABLE", |
| 618 |
"Py_Unicode_GET_LENGTH", |
| 619 |
"_PyCode_CODE", |
| 620 |
"_PyDictValues_AddToInsertionOrder", |
| 621 |
"_PyErr_Occurred", |
| 622 |
"_PyFloat_FromDouble_ConsumeInputs", |
| 623 |
"_PyFrame_GetBytecode", |
| 624 |
"_PyFrame_GetCode", |
| 625 |
"_PyFrame_IsIncomplete", |
| 626 |
"_PyFrame_PushUnchecked", |
| 627 |
"_PyFrame_SetStackPointer", |
| 628 |
"_PyFrame_StackPush", |
| 629 |
"_PyFunction_SetVersion", |
| 630 |
"_PyGen_GetGeneratorFromFrame", |
| 631 |
"_PyInterpreterState_GET", |
| 632 |
"_PyList_AppendTakeRef", |
| 633 |
"_PyList_ITEMS", |
| 634 |
"_PyLong_CompactValue", |
| 635 |
"_PyLong_DigitCount", |
| 636 |
"_PyLong_IsCompact", |
| 637 |
"_PyLong_IsNegative", |
| 638 |
"_PyLong_IsNonNegativeCompact", |
| 639 |
"_PyLong_IsZero", |
| 640 |
"_PyManagedDictPointer_IsValues", |
| 641 |
"_PyObject_GC_IS_SHARED", |
| 642 |
"_PyObject_GC_IS_TRACKED", |
| 643 |
"_PyObject_GC_MAY_BE_TRACKED", |
| 644 |
"_PyObject_GC_TRACK", |
| 645 |
"_PyObject_GetManagedDict", |
| 646 |
"_PyObject_InlineValues", |
| 647 |
"_PyObject_IsUniquelyReferenced", |
| 648 |
"_PyObject_ManagedDictPointer", |
| 649 |
"_PyThreadState_HasStackSpace", |
| 650 |
"_PyTuple_FromStackRefStealOnSuccess", |
| 651 |
"_PyTuple_ITEMS", |
| 652 |
"_PyType_HasFeature", |
| 653 |
"_PyType_NewManagedObject", |
| 654 |
"_PyUnicode_Equal", |
| 655 |
"_PyUnicode_JoinArray", |
| 656 |
"_Py_CHECK_EMSCRIPTEN_SIGNALS_PERIODICALLY", |
| 657 |
"_Py_DECREF_NO_DEALLOC", |
| 658 |
"_Py_ID", |
| 659 |
"_Py_IsImmortal", |
| 660 |
"_Py_IsOwnedByCurrentThread", |
| 661 |
"_Py_LeaveRecursiveCallPy", |
| 662 |
"_Py_LeaveRecursiveCallTstate", |
| 663 |
"_Py_NewRef", |
| 664 |
"_Py_SINGLETON", |
| 665 |
"_Py_STR", |
| 666 |
"_Py_TryIncrefCompare", |
| 667 |
"_Py_TryIncrefCompareStackRef", |
| 668 |
"_Py_atomic_compare_exchange_uint8", |
| 669 |
"_Py_atomic_load_ptr_acquire", |
| 670 |
"_Py_atomic_load_uintptr_relaxed", |
| 671 |
"_Py_set_eval_breaker_bit", |
| 672 |
"advance_backoff_counter", |
| 673 |
"assert", |
| 674 |
"backoff_counter_triggers", |
| 675 |
"initial_temperature_backoff_counter", |
| 676 |
"JUMP_TO_LABEL", |
| 677 |
"restart_backoff_counter", |
| 678 |
"_Py_ReachedRecursionLimit", |
| 679 |
"PyStackRef_IsTaggedInt", |
| 680 |
"PyStackRef_TagInt", |
| 681 |
"PyStackRef_UntagInt", |
| 682 |
) |
| 683 |
|
| 684 |
def check_escaping_calls(instr: parser.CodeDef, escapes: dict[SimpleStmt, EscapingCall]) -> None: |
| 685 |
error: lexer.Token | None = None |
| 686 |
calls = {e.call for e in escapes.values()} |
| 687 |
|
| 688 |
def visit(stmt: Stmt) -> None: |
| 689 |
nonlocal error |
| 690 |
if isinstance(stmt, IfStmt) or isinstance(stmt, WhileStmt): |
| 691 |
for tkn in stmt.condition: |
| 692 |
if tkn in calls: |
| 693 |
error = tkn |
| 694 |
elif isinstance(stmt, SimpleStmt): |
| 695 |
in_if = 0 |
| 696 |
tkn_iter = iter(stmt.contents) |
| 697 |
for tkn in tkn_iter: |
| 698 |
if tkn.kind == "IDENTIFIER" and tkn.text in ("DEOPT_IF", "ERROR_IF", "EXIT_IF"): |
| 699 |
in_if = 1 |
| 700 |
next(tkn_iter) |
| 701 |
elif tkn.kind == "LPAREN": |
| 702 |
if in_if: |
| 703 |
in_if += 1 |
| 704 |
elif tkn.kind == "RPAREN": |
| 705 |
if in_if: |
| 706 |
in_if -= 1 |
| 707 |
elif tkn in calls and in_if: |
| 708 |
error = tkn |
| 709 |
|
| 710 |
|
| 711 |
instr.block.accept(visit) |
| 712 |
if error is not None: |
| 713 |
raise analysis_error(f"Escaping call '{error.text} in condition", error) |
| 714 |
|
| 715 |
def find_escaping_api_calls(instr: parser.CodeDef) -> dict[SimpleStmt, EscapingCall]: |
| 716 |
result: dict[SimpleStmt, EscapingCall] = {} |
| 717 |
|
| 718 |
def visit(stmt: Stmt) -> None: |
| 719 |
if not isinstance(stmt, SimpleStmt): |
| 720 |
return |
| 721 |
tokens = stmt.contents |
| 722 |
for idx, tkn in enumerate(tokens): |
| 723 |
try: |
| 724 |
next_tkn = tokens[idx+1] |
| 725 |
except IndexError: |
| 726 |
break |
| 727 |
if next_tkn.kind != lexer.LPAREN: |
| 728 |
continue |
| 729 |
if tkn.kind == lexer.IDENTIFIER: |
| 730 |
if tkn.text.upper() == tkn.text: |
| 731 |
# simple macro |
| 732 |
continue |
| 733 |
#if not tkn.text.startswith(("Py", "_Py", "monitor")): |
| 734 |
# continue |
| 735 |
if tkn.text.startswith(("sym_", "optimize_")): |
| 736 |
# Optimize functions |
| 737 |
continue |
| 738 |
if tkn.text.endswith("Check"): |
| 739 |
continue |
| 740 |
if tkn.text.startswith("Py_Is"): |
| 741 |
continue |
| 742 |
if tkn.text.endswith("CheckExact"): |
| 743 |
continue |
| 744 |
if tkn.text in NON_ESCAPING_FUNCTIONS: |
| 745 |
continue |
| 746 |
elif tkn.kind == "RPAREN": |
| 747 |
prev = tokens[idx-1] |
| 748 |
if prev.text.endswith("_t") or prev.text == "*" or prev.text == "int": |
| 749 |
#cast |
| 750 |
continue |
| 751 |
elif tkn.kind != "RBRACKET": |
| 752 |
continue |
| 753 |
if tkn.text in ("PyStackRef_CLOSE", "PyStackRef_XCLOSE"): |
| 754 |
if len(tokens) <= idx+2: |
| 755 |
raise analysis_error("Unexpected end of file", next_tkn) |
| 756 |
kills = tokens[idx+2] |
| 757 |
if kills.kind != "IDENTIFIER": |
| 758 |
raise analysis_error(f"Expected identifier, got '{kills.text}'", kills) |
| 759 |
else: |
| 760 |
kills = None |
| 761 |
result[stmt] = EscapingCall(stmt, tkn, kills) |
| 762 |
|
| 763 |
instr.block.accept(visit) |
| 764 |
check_escaping_calls(instr, result) |
| 765 |
return result |
| 766 |
|
| 767 |
|
| 768 |
EXITS = { |
| 769 |
"DISPATCH", |
| 770 |
"Py_UNREACHABLE", |
| 771 |
"DISPATCH_INLINED", |
| 772 |
"DISPATCH_GOTO", |
| 773 |
} |
| 774 |
|
| 775 |
|
| 776 |
def always_exits(op: parser.CodeDef) -> bool: |
| 777 |
depth = 0 |
| 778 |
tkn_iter = iter(op.tokens) |
| 779 |
for tkn in tkn_iter: |
| 780 |
if tkn.kind == "LBRACE": |
| 781 |
depth += 1 |
| 782 |
elif tkn.kind == "RBRACE": |
| 783 |
depth -= 1 |
| 784 |
elif depth > 1: |
| 785 |
continue |
| 786 |
elif tkn.kind == "GOTO" or tkn.kind == "RETURN": |
| 787 |
return True |
| 788 |
elif tkn.kind == "KEYWORD": |
| 789 |
if tkn.text in EXITS: |
| 790 |
return True |
| 791 |
elif tkn.kind == "IDENTIFIER": |
| 792 |
if tkn.text in EXITS: |
| 793 |
return True |
| 794 |
if tkn.text == "DEOPT_IF" or tkn.text == "ERROR_IF": |
| 795 |
next(tkn_iter) # '(' |
| 796 |
t = next(tkn_iter) |
| 797 |
if t.text in ("true", "1"): |
| 798 |
return True |
| 799 |
return False |
| 800 |
|
| 801 |
|
| 802 |
def stack_effect_only_peeks(instr: parser.InstDef) -> bool: |
| 803 |
stack_inputs = [s for s in instr.inputs if not isinstance(s, parser.CacheEffect)] |
| 804 |
if len(stack_inputs) != len(instr.outputs): |
| 805 |
return False |
| 806 |
if len(stack_inputs) == 0: |
| 807 |
return False |
| 808 |
return all( |
| 809 |
(s.name == other.name and s.type == other.type and s.size == other.size) |
| 810 |
for s, other in zip(stack_inputs, instr.outputs) |
| 811 |
) |
| 812 |
|
| 813 |
|
| 814 |
def compute_properties(op: parser.CodeDef) -> Properties: |
| 815 |
escaping_calls = find_escaping_api_calls(op) |
| 816 |
has_free = ( |
| 817 |
variable_used(op, "PyCell_New") |
| 818 |
or variable_used(op, "PyCell_GetRef") |
| 819 |
or variable_used(op, "PyCell_SetTakeRef") |
| 820 |
or variable_used(op, "PyCell_SwapTakeRef") |
| 821 |
) |
| 822 |
deopts_if = variable_used(op, "DEOPT_IF") |
| 823 |
exits_if = variable_used(op, "EXIT_IF") |
| 824 |
if deopts_if and exits_if: |
| 825 |
tkn = op.tokens[0] |
| 826 |
raise lexer.make_syntax_error( |
| 827 |
"Op cannot contain both EXIT_IF and DEOPT_IF", |
| 828 |
tkn.filename, |
| 829 |
tkn.line, |
| 830 |
tkn.column, |
| 831 |
op.name, |
| 832 |
) |
| 833 |
error_with_pop = has_error_with_pop(op) |
| 834 |
error_without_pop = has_error_without_pop(op) |
| 835 |
escapes = bool(escaping_calls) |
| 836 |
pure = False if isinstance(op, parser.LabelDef) else "pure" in op.annotations |
| 837 |
no_save_ip = False if isinstance(op, parser.LabelDef) else "no_save_ip" in op.annotations |
| 838 |
return Properties( |
| 839 |
escaping_calls=escaping_calls, |
| 840 |
escapes=escapes, |
| 841 |
error_with_pop=error_with_pop, |
| 842 |
error_without_pop=error_without_pop, |
| 843 |
deopts=deopts_if, |
| 844 |
side_exit=exits_if, |
| 845 |
oparg=oparg_used(op), |
| 846 |
jumps=variable_used(op, "JUMPBY"), |
| 847 |
eval_breaker="CHECK_PERIODIC" in op.name, |
| 848 |
needs_this=variable_used(op, "this_instr"), |
| 849 |
always_exits=always_exits(op), |
| 850 |
stores_sp=variable_used(op, "SYNC_SP"), |
| 851 |
uses_co_consts=variable_used(op, "FRAME_CO_CONSTS"), |
| 852 |
uses_co_names=variable_used(op, "FRAME_CO_NAMES"), |
| 853 |
uses_locals=variable_used(op, "GETLOCAL") and not has_free, |
| 854 |
uses_opcode=variable_used(op, "opcode"), |
| 855 |
has_free=has_free, |
| 856 |
pure=pure, |
| 857 |
no_save_ip=no_save_ip, |
| 858 |
tier=tier_variable(op), |
| 859 |
needs_prev=variable_used(op, "prev_instr"), |
| 860 |
) |
| 861 |
|
| 862 |
|
| 863 |
def make_uop( |
| 864 |
name: str, |
| 865 |
op: parser.InstDef, |
| 866 |
inputs: list[parser.InputEffect], |
| 867 |
uops: dict[str, Uop], |
| 868 |
) -> Uop: |
| 869 |
result = Uop( |
| 870 |
name=name, |
| 871 |
context=op.context, |
| 872 |
annotations=op.annotations, |
| 873 |
stack=analyze_stack(op), |
| 874 |
caches=analyze_caches(inputs), |
| 875 |
local_stores=find_variable_stores(op), |
| 876 |
body=op.block, |
| 877 |
properties=compute_properties(op), |
| 878 |
) |
| 879 |
for anno in op.annotations: |
| 880 |
if anno.startswith("replicate"): |
| 881 |
result.replicated = int(anno[10:-1]) |
| 882 |
break |
| 883 |
else: |
| 884 |
return result |
| 885 |
for oparg in range(result.replicated): |
| 886 |
name_x = name + "_" + str(oparg) |
| 887 |
properties = compute_properties(op) |
| 888 |
properties.oparg = False |
| 889 |
properties.const_oparg = oparg |
| 890 |
rep = Uop( |
| 891 |
name=name_x, |
| 892 |
context=op.context, |
| 893 |
annotations=op.annotations, |
| 894 |
stack=analyze_stack(op), |
| 895 |
caches=analyze_caches(inputs), |
| 896 |
local_stores=find_variable_stores(op), |
| 897 |
body=op.block, |
| 898 |
properties=properties, |
| 899 |
) |
| 900 |
rep.replicates = result |
| 901 |
uops[name_x] = rep |
| 902 |
|
| 903 |
return result |
| 904 |
|
| 905 |
|
| 906 |
def add_op(op: parser.InstDef, uops: dict[str, Uop]) -> None: |
| 907 |
assert op.kind == "op" |
| 908 |
if op.name in uops: |
| 909 |
if "override" not in op.annotations: |
| 910 |
raise override_error( |
| 911 |
op.name, op.context, uops[op.name].context, op.tokens[0] |
| 912 |
) |
| 913 |
uops[op.name] = make_uop(op.name, op, op.inputs, uops) |
| 914 |
|
| 915 |
|
| 916 |
def add_instruction( |
| 917 |
where: lexer.Token, |
| 918 |
name: str, |
| 919 |
parts: list[Part], |
| 920 |
instructions: dict[str, Instruction], |
| 921 |
) -> None: |
| 922 |
instructions[name] = Instruction(where, name, parts, None) |
| 923 |
|
| 924 |
|
| 925 |
def desugar_inst( |
| 926 |
inst: parser.InstDef, instructions: dict[str, Instruction], uops: dict[str, Uop] |
| 927 |
) -> None: |
| 928 |
assert inst.kind == "inst" |
| 929 |
name = inst.name |
| 930 |
op_inputs: list[parser.InputEffect] = [] |
| 931 |
parts: list[Part] = [] |
| 932 |
uop_index = -1 |
| 933 |
# Move unused cache entries to the Instruction, removing them from the Uop. |
| 934 |
for input in inst.inputs: |
| 935 |
if isinstance(input, parser.CacheEffect) and input.name == "unused": |
| 936 |
parts.append(Skip(input.size)) |
| 937 |
else: |
| 938 |
op_inputs.append(input) |
| 939 |
if uop_index < 0: |
| 940 |
uop_index = len(parts) |
| 941 |
# Place holder for the uop. |
| 942 |
parts.append(Skip(0)) |
| 943 |
uop = make_uop("_" + inst.name, inst, op_inputs, uops) |
| 944 |
uop.implicitly_created = True |
| 945 |
uops[inst.name] = uop |
| 946 |
if uop_index < 0: |
| 947 |
parts.append(uop) |
| 948 |
else: |
| 949 |
parts[uop_index] = uop |
| 950 |
add_instruction(inst.first_token, name, parts, instructions) |
| 951 |
|
| 952 |
|
| 953 |
def add_macro( |
| 954 |
macro: parser.Macro, instructions: dict[str, Instruction], uops: dict[str, Uop] |
| 955 |
) -> None: |
| 956 |
parts: list[Part] = [] |
| 957 |
for part in macro.uops: |
| 958 |
match part: |
| 959 |
case parser.OpName(): |
| 960 |
if part.name == "flush": |
| 961 |
parts.append(Flush()) |
| 962 |
else: |
| 963 |
if part.name not in uops: |
| 964 |
raise analysis_error( |
| 965 |
f"No Uop named {part.name}", macro.tokens[0] |
| 966 |
) |
| 967 |
parts.append(uops[part.name]) |
| 968 |
case parser.CacheEffect(): |
| 969 |
parts.append(Skip(part.size)) |
| 970 |
case _: |
| 971 |
assert False |
| 972 |
assert parts |
| 973 |
add_instruction(macro.first_token, macro.name, parts, instructions) |
| 974 |
|
| 975 |
|
| 976 |
def add_family( |
| 977 |
pfamily: parser.Family, |
| 978 |
instructions: dict[str, Instruction], |
| 979 |
families: dict[str, Family], |
| 980 |
) -> None: |
| 981 |
family = Family( |
| 982 |
pfamily.name, |
| 983 |
pfamily.size, |
| 984 |
[instructions[member_name] for member_name in pfamily.members], |
| 985 |
) |
| 986 |
for member in family.members: |
| 987 |
member.family = family |
| 988 |
# The head of the family is an implicit jump target for DEOPTs |
| 989 |
instructions[family.name].is_target = True |
| 990 |
families[family.name] = family |
| 991 |
|
| 992 |
|
| 993 |
def add_pseudo( |
| 994 |
pseudo: parser.Pseudo, |
| 995 |
instructions: dict[str, Instruction], |
| 996 |
pseudos: dict[str, PseudoInstruction], |
| 997 |
) -> None: |
| 998 |
pseudos[pseudo.name] = PseudoInstruction( |
| 999 |
pseudo.name, |
| 1000 |
analyze_stack(pseudo), |
| 1001 |
[instructions[target] for target in pseudo.targets], |
| 1002 |
pseudo.as_sequence, |
| 1003 |
pseudo.flags, |
| 1004 |
) |
| 1005 |
|
| 1006 |
|
| 1007 |
def add_label( |
| 1008 |
label: parser.LabelDef, |
| 1009 |
labels: dict[str, Label], |
| 1010 |
) -> None: |
| 1011 |
properties = compute_properties(label) |
| 1012 |
labels[label.name] = Label(label.name, label.spilled, label.block, properties) |
| 1013 |
|
| 1014 |
|
| 1015 |
def assign_opcodes( |
| 1016 |
instructions: dict[str, Instruction], |
| 1017 |
families: dict[str, Family], |
| 1018 |
pseudos: dict[str, PseudoInstruction], |
| 1019 |
) -> tuple[dict[str, int], int, int]: |
| 1020 |
"""Assigns opcodes, then returns the opmap, |
| 1021 |
have_arg and min_instrumented values""" |
| 1022 |
instmap: dict[str, int] = {} |
| 1023 |
|
| 1024 |
# 0 is reserved for cache entries. This helps debugging. |
| 1025 |
instmap["CACHE"] = 0 |
| 1026 |
|
| 1027 |
# 17 is reserved as it is the initial value for the specializing counter. |
| 1028 |
# This helps catch cases where we attempt to execute a cache. |
| 1029 |
instmap["RESERVED"] = 17 |
| 1030 |
|
| 1031 |
# 128 is RESUME - it is hard coded as such in Tools/build/deepfreeze.py |
| 1032 |
instmap["RESUME"] = 128 |
| 1033 |
|
| 1034 |
# This is an historical oddity. |
| 1035 |
instmap["BINARY_OP_INPLACE_ADD_UNICODE"] = 3 |
| 1036 |
|
| 1037 |
instmap["INSTRUMENTED_LINE"] = 254 |
| 1038 |
instmap["ENTER_EXECUTOR"] = 255 |
| 1039 |
|
| 1040 |
instrumented = [name for name in instructions if name.startswith("INSTRUMENTED")] |
| 1041 |
|
| 1042 |
specialized: set[str] = set() |
| 1043 |
no_arg: list[str] = [] |
| 1044 |
has_arg: list[str] = [] |
| 1045 |
|
| 1046 |
for family in families.values(): |
| 1047 |
specialized.update(inst.name for inst in family.members) |
| 1048 |
|
| 1049 |
for inst in instructions.values(): |
| 1050 |
name = inst.name |
| 1051 |
if name in specialized: |
| 1052 |
continue |
| 1053 |
if name in instrumented: |
| 1054 |
continue |
| 1055 |
if inst.properties.oparg: |
| 1056 |
has_arg.append(name) |
| 1057 |
else: |
| 1058 |
no_arg.append(name) |
| 1059 |
|
| 1060 |
# Specialized ops appear in their own section |
| 1061 |
# Instrumented opcodes are at the end of the valid range |
| 1062 |
min_internal = instmap["RESUME"] + 1 |
| 1063 |
min_instrumented = 254 - (len(instrumented) - 1) |
| 1064 |
assert min_internal + len(specialized) < min_instrumented |
| 1065 |
|
| 1066 |
next_opcode = 1 |
| 1067 |
|
| 1068 |
def add_instruction(name: str) -> None: |
| 1069 |
nonlocal next_opcode |
| 1070 |
if name in instmap: |
| 1071 |
return # Pre-defined name |
| 1072 |
while next_opcode in instmap.values(): |
| 1073 |
next_opcode += 1 |
| 1074 |
instmap[name] = next_opcode |
| 1075 |
next_opcode += 1 |
| 1076 |
|
| 1077 |
for name in sorted(no_arg): |
| 1078 |
add_instruction(name) |
| 1079 |
for name in sorted(has_arg): |
| 1080 |
add_instruction(name) |
| 1081 |
# For compatibility |
| 1082 |
next_opcode = min_internal |
| 1083 |
for name in sorted(specialized): |
| 1084 |
add_instruction(name) |
| 1085 |
next_opcode = min_instrumented |
| 1086 |
for name in instrumented: |
| 1087 |
add_instruction(name) |
| 1088 |
|
| 1089 |
for name in instructions: |
| 1090 |
instructions[name].opcode = instmap[name] |
| 1091 |
|
| 1092 |
for op, name in enumerate(sorted(pseudos), 256): |
| 1093 |
instmap[name] = op |
| 1094 |
pseudos[name].opcode = op |
| 1095 |
|
| 1096 |
return instmap, len(no_arg), min_instrumented |
| 1097 |
|
| 1098 |
|
| 1099 |
def get_instruction_size_for_uop(instructions: dict[str, Instruction], uop: Uop) -> int | None: |
| 1100 |
"""Return the size of the instruction that contains the given uop or |
| 1101 |
`None` if the uop does not contains the `INSTRUCTION_SIZE` macro. |
| 1102 |
|
| 1103 |
If there is more than one instruction that contains the uop, |
| 1104 |
ensure that they all have the same size. |
| 1105 |
""" |
| 1106 |
for tkn in uop.body.tokens(): |
| 1107 |
if tkn.text == "INSTRUCTION_SIZE": |
| 1108 |
break |
| 1109 |
else: |
| 1110 |
return None |
| 1111 |
|
| 1112 |
size = None |
| 1113 |
for inst in instructions.values(): |
| 1114 |
if uop in inst.parts: |
| 1115 |
if size is None: |
| 1116 |
size = inst.size |
| 1117 |
if size != inst.size: |
| 1118 |
raise analysis_error( |
| 1119 |
"All instructions containing a uop with the `INSTRUCTION_SIZE` macro " |
| 1120 |
f"must have the same size: {size} != {inst.size}", |
| 1121 |
tkn |
| 1122 |
) |
| 1123 |
if size is None: |
| 1124 |
raise analysis_error(f"No instruction containing the uop '{uop.name}' was found", tkn) |
| 1125 |
return size |
| 1126 |
|
| 1127 |
|
| 1128 |
def analyze_forest(forest: list[parser.AstNode]) -> Analysis: |
| 1129 |
instructions: dict[str, Instruction] = {} |
| 1130 |
uops: dict[str, Uop] = {} |
| 1131 |
families: dict[str, Family] = {} |
| 1132 |
pseudos: dict[str, PseudoInstruction] = {} |
| 1133 |
labels: dict[str, Label] = {} |
| 1134 |
for node in forest: |
| 1135 |
match node: |
| 1136 |
case parser.InstDef(name): |
| 1137 |
if node.kind == "inst": |
| 1138 |
desugar_inst(node, instructions, uops) |
| 1139 |
else: |
| 1140 |
assert node.kind == "op" |
| 1141 |
add_op(node, uops) |
| 1142 |
case parser.Macro(): |
| 1143 |
pass |
| 1144 |
case parser.Family(): |
| 1145 |
pass |
| 1146 |
case parser.Pseudo(): |
| 1147 |
pass |
| 1148 |
case parser.LabelDef(): |
| 1149 |
pass |
| 1150 |
case _: |
| 1151 |
assert False |
| 1152 |
for node in forest: |
| 1153 |
if isinstance(node, parser.Macro): |
| 1154 |
add_macro(node, instructions, uops) |
| 1155 |
for node in forest: |
| 1156 |
match node: |
| 1157 |
case parser.Family(): |
| 1158 |
add_family(node, instructions, families) |
| 1159 |
case parser.Pseudo(): |
| 1160 |
add_pseudo(node, instructions, pseudos) |
| 1161 |
case parser.LabelDef(): |
| 1162 |
add_label(node, labels) |
| 1163 |
case _: |
| 1164 |
pass |
| 1165 |
for uop in uops.values(): |
| 1166 |
uop.instruction_size = get_instruction_size_for_uop(instructions, uop) |
| 1167 |
# Special case BINARY_OP_INPLACE_ADD_UNICODE |
| 1168 |
# BINARY_OP_INPLACE_ADD_UNICODE is not a normal family member, |
| 1169 |
# as it is the wrong size, but we need it to maintain an |
| 1170 |
# historical optimization. |
| 1171 |
if "BINARY_OP_INPLACE_ADD_UNICODE" in instructions: |
| 1172 |
inst = instructions["BINARY_OP_INPLACE_ADD_UNICODE"] |
| 1173 |
inst.family = families["BINARY_OP"] |
| 1174 |
families["BINARY_OP"].members.append(inst) |
| 1175 |
opmap, first_arg, min_instrumented = assign_opcodes(instructions, families, pseudos) |
| 1176 |
return Analysis( |
| 1177 |
instructions, uops, families, pseudos, labels, opmap, first_arg, min_instrumented |
| 1178 |
) |
| 1179 |
|
| 1180 |
|
| 1181 |
def analyze_files(filenames: list[str]) -> Analysis: |
| 1182 |
return analyze_forest(parser.parse_files(filenames)) |
| 1183 |
|
| 1184 |
|
| 1185 |
def dump_analysis(analysis: Analysis) -> None: |
| 1186 |
print("Uops:") |
| 1187 |
for u in analysis.uops.values(): |
| 1188 |
u.dump(" ") |
| 1189 |
print("Instructions:") |
| 1190 |
for i in analysis.instructions.values(): |
| 1191 |
i.dump(" ") |
| 1192 |
print("Families:") |
| 1193 |
for f in analysis.families.values(): |
| 1194 |
f.dump(" ") |
| 1195 |
print("Pseudos:") |
| 1196 |
for p in analysis.pseudos.values(): |
| 1197 |
p.dump(" ") |
| 1198 |
|
| 1199 |
|
| 1200 |
if __name__ == "__main__": |
| 1201 |
import sys |
| 1202 |
|
| 1203 |
if len(sys.argv) < 2: |
| 1204 |
print("No input") |
| 1205 |
else: |
| 1206 |
filenames = sys.argv[1:] |
| 1207 |
dump_analysis(analyze_files(filenames)) |