2022-11-17 21:06:07 -04:00
|
|
|
"""Generate the main interpreter switch.
|
|
|
|
Reads the instruction definitions from bytecodes.c.
|
|
|
|
Writes the cases to generated_cases.c.h, which is #included in ceval.c.
|
|
|
|
"""
|
2022-11-03 01:31:26 -03:00
|
|
|
|
|
|
|
import argparse
|
2023-08-14 15:36:29 -03:00
|
|
|
import contextlib
|
2023-08-16 19:25:18 -03:00
|
|
|
import itertools
|
2022-11-04 21:40:43 -03:00
|
|
|
import os
|
2023-02-20 10:56:48 -04:00
|
|
|
import posixpath
|
2022-11-03 01:31:26 -03:00
|
|
|
import sys
|
2023-08-21 14:15:52 -03:00
|
|
|
import textwrap
|
2022-11-17 21:06:07 -04:00
|
|
|
import typing
|
2023-08-18 10:42:45 -03:00
|
|
|
from collections.abc import Iterator
|
2022-11-03 01:31:26 -03:00
|
|
|
|
2023-08-04 13:35:56 -03:00
|
|
|
import stacking # Early import to avoid circular import
|
2023-08-20 20:40:41 -03:00
|
|
|
from _typing_backports import assert_never
|
2023-07-24 13:38:23 -03:00
|
|
|
from analysis import Analyzer
|
2023-08-04 13:35:56 -03:00
|
|
|
from formatting import Formatter, list_effect_size
|
2023-07-24 13:38:23 -03:00
|
|
|
from flags import InstructionFlags, variable_used
|
|
|
|
from instructions import (
|
|
|
|
AnyInstruction,
|
2023-08-15 15:04:17 -03:00
|
|
|
AbstractInstruction,
|
2023-07-24 13:38:23 -03:00
|
|
|
Component,
|
|
|
|
Instruction,
|
|
|
|
MacroInstruction,
|
|
|
|
MacroParts,
|
|
|
|
PseudoInstruction,
|
|
|
|
OverriddenInstructionPlaceHolder,
|
2023-08-16 20:26:43 -03:00
|
|
|
TIER_ONE,
|
2023-07-24 13:38:23 -03:00
|
|
|
TIER_TWO,
|
|
|
|
)
|
|
|
|
import parsing
|
|
|
|
from parsing import StackEffect
|
2022-11-03 01:31:26 -03:00
|
|
|
|
2023-07-20 20:08:52 -03:00
|
|
|
|
2023-01-17 19:59:19 -04:00
|
|
|
HERE = os.path.dirname(__file__)
|
|
|
|
ROOT = os.path.join(HERE, "../..")
|
2023-02-20 10:56:48 -04:00
|
|
|
THIS = os.path.relpath(__file__, ROOT).replace(os.path.sep, posixpath.sep)
|
2023-01-17 19:59:19 -04:00
|
|
|
|
|
|
|
DEFAULT_INPUT = os.path.relpath(os.path.join(ROOT, "Python/bytecodes.c"))
|
|
|
|
DEFAULT_OUTPUT = os.path.relpath(os.path.join(ROOT, "Python/generated_cases.c.h"))
|
2023-08-16 19:25:18 -03:00
|
|
|
DEFAULT_OPCODE_IDS_H_OUTPUT = os.path.relpath(
|
|
|
|
os.path.join(ROOT, "Include/opcode_ids.h")
|
|
|
|
)
|
|
|
|
DEFAULT_OPCODE_TARGETS_H_OUTPUT = os.path.relpath(
|
|
|
|
os.path.join(ROOT, "Python/opcode_targets.h")
|
|
|
|
)
|
2023-01-05 17:01:07 -04:00
|
|
|
DEFAULT_METADATA_OUTPUT = os.path.relpath(
|
2023-07-12 07:30:25 -03:00
|
|
|
os.path.join(ROOT, "Include/internal/pycore_opcode_metadata.h")
|
2023-01-05 17:01:07 -04:00
|
|
|
)
|
2023-06-19 19:47:04 -03:00
|
|
|
DEFAULT_PYMETADATA_OUTPUT = os.path.relpath(
|
|
|
|
os.path.join(ROOT, "Lib/_opcode_metadata.py")
|
|
|
|
)
|
2023-06-26 23:02:57 -03:00
|
|
|
DEFAULT_EXECUTOR_OUTPUT = os.path.relpath(
|
|
|
|
os.path.join(ROOT, "Python/executor_cases.c.h")
|
|
|
|
)
|
2023-08-15 15:04:17 -03:00
|
|
|
DEFAULT_ABSTRACT_INTERPRETER_OUTPUT = os.path.relpath(
|
|
|
|
os.path.join(ROOT, "Python/abstract_interp_cases.c.h")
|
|
|
|
)
|
2022-11-09 14:50:09 -04:00
|
|
|
|
2023-07-07 14:42:10 -03:00
|
|
|
# Constants used instead of size for macro expansions.
|
|
|
|
# Note: 1, 2, 4 must match actual cache entry sizes.
|
|
|
|
OPARG_SIZES = {
|
|
|
|
"OPARG_FULL": 0,
|
|
|
|
"OPARG_CACHE_1": 1,
|
|
|
|
"OPARG_CACHE_2": 2,
|
|
|
|
"OPARG_CACHE_4": 4,
|
|
|
|
"OPARG_TOP": 5,
|
|
|
|
"OPARG_BOTTOM": 6,
|
2023-09-11 19:39:19 -03:00
|
|
|
"OPARG_SET_IP": 7,
|
2023-07-07 14:42:10 -03:00
|
|
|
}
|
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
INSTR_FMT_PREFIX = "INSTR_FMT_"
|
2023-06-13 17:42:03 -03:00
|
|
|
|
2023-08-15 15:04:17 -03:00
|
|
|
# TODO: generate all these after updating the DSL
|
|
|
|
SPECIALLY_HANDLED_ABSTRACT_INSTR = {
|
|
|
|
"LOAD_FAST",
|
|
|
|
"LOAD_FAST_CHECK",
|
|
|
|
"LOAD_FAST_AND_CLEAR",
|
|
|
|
"LOAD_CONST",
|
|
|
|
"STORE_FAST",
|
|
|
|
"STORE_FAST_MAYBE_NULL",
|
|
|
|
"COPY",
|
|
|
|
# Arithmetic
|
|
|
|
"_BINARY_OP_MULTIPLY_INT",
|
|
|
|
"_BINARY_OP_ADD_INT",
|
|
|
|
"_BINARY_OP_SUBTRACT_INT",
|
|
|
|
}
|
|
|
|
|
2022-12-02 23:57:30 -04:00
|
|
|
arg_parser = argparse.ArgumentParser(
|
|
|
|
description="Generate the code for the interpreter switch.",
|
|
|
|
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
|
|
|
|
)
|
|
|
|
arg_parser.add_argument(
|
|
|
|
"-o", "--output", type=str, help="Generated code", default=DEFAULT_OUTPUT
|
|
|
|
)
|
2023-08-16 19:25:18 -03:00
|
|
|
arg_parser.add_argument(
|
|
|
|
"-n",
|
|
|
|
"--opcode_ids_h",
|
|
|
|
type=str,
|
|
|
|
help="Header file with opcode number definitions",
|
|
|
|
default=DEFAULT_OPCODE_IDS_H_OUTPUT,
|
|
|
|
)
|
|
|
|
arg_parser.add_argument(
|
|
|
|
"-t",
|
|
|
|
"--opcode_targets_h",
|
|
|
|
type=str,
|
|
|
|
help="File with opcode targets for computed gotos",
|
|
|
|
default=DEFAULT_OPCODE_TARGETS_H_OUTPUT,
|
|
|
|
)
|
2023-01-05 17:01:07 -04:00
|
|
|
arg_parser.add_argument(
|
2023-07-24 13:38:23 -03:00
|
|
|
"-m",
|
|
|
|
"--metadata",
|
|
|
|
type=str,
|
|
|
|
help="Generated C metadata",
|
|
|
|
default=DEFAULT_METADATA_OUTPUT,
|
2023-06-19 19:47:04 -03:00
|
|
|
)
|
|
|
|
arg_parser.add_argument(
|
2023-07-24 13:38:23 -03:00
|
|
|
"-p",
|
|
|
|
"--pymetadata",
|
|
|
|
type=str,
|
|
|
|
help="Generated Python metadata",
|
|
|
|
default=DEFAULT_PYMETADATA_OUTPUT,
|
2023-01-05 17:01:07 -04:00
|
|
|
)
|
2023-03-15 12:37:36 -03:00
|
|
|
arg_parser.add_argument(
|
|
|
|
"-l", "--emit-line-directives", help="Emit #line directives", action="store_true"
|
|
|
|
)
|
2023-03-04 00:59:21 -04:00
|
|
|
arg_parser.add_argument(
|
|
|
|
"input", nargs=argparse.REMAINDER, help="Instruction definition file(s)"
|
|
|
|
)
|
2023-06-26 23:02:57 -03:00
|
|
|
arg_parser.add_argument(
|
|
|
|
"-e",
|
|
|
|
"--executor-cases",
|
|
|
|
type=str,
|
|
|
|
help="Write executor cases to this file",
|
|
|
|
default=DEFAULT_EXECUTOR_OUTPUT,
|
|
|
|
)
|
2023-08-15 15:04:17 -03:00
|
|
|
arg_parser.add_argument(
|
|
|
|
"-a",
|
|
|
|
"--abstract-interpreter-cases",
|
|
|
|
type=str,
|
|
|
|
help="Write abstract interpreter cases to this file",
|
|
|
|
default=DEFAULT_ABSTRACT_INTERPRETER_OUTPUT,
|
|
|
|
)
|
2022-11-03 01:31:26 -03:00
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
class Generator(Analyzer):
|
2023-01-25 16:41:03 -04:00
|
|
|
def get_stack_effect_info(
|
2023-07-24 13:38:23 -03:00
|
|
|
self, thing: parsing.InstDef | parsing.Macro | parsing.Pseudo
|
2023-08-20 20:40:41 -03:00
|
|
|
) -> tuple[AnyInstruction | None, str, str]:
|
2023-01-29 21:28:39 -04:00
|
|
|
def effect_str(effects: list[StackEffect]) -> str:
|
|
|
|
n_effect, sym_effect = list_effect_size(effects)
|
2023-01-25 16:41:03 -04:00
|
|
|
if sym_effect:
|
|
|
|
return f"{sym_effect} + {n_effect}" if n_effect else sym_effect
|
|
|
|
return str(n_effect)
|
|
|
|
|
2023-01-29 21:28:39 -04:00
|
|
|
instr: AnyInstruction | None
|
2023-08-25 14:08:29 -03:00
|
|
|
popped: str | None = None
|
|
|
|
pushed: str | None = None
|
2023-01-25 16:41:03 -04:00
|
|
|
match thing:
|
2023-07-24 13:38:23 -03:00
|
|
|
case parsing.InstDef():
|
2023-09-15 12:39:05 -03:00
|
|
|
instr = self.instrs[thing.name]
|
|
|
|
popped = effect_str(instr.input_effects)
|
|
|
|
pushed = effect_str(instr.output_effects)
|
2023-07-24 13:38:23 -03:00
|
|
|
case parsing.Macro():
|
2023-01-25 16:41:03 -04:00
|
|
|
instr = self.macro_instrs[thing.name]
|
2023-08-04 13:35:56 -03:00
|
|
|
popped, pushed = stacking.get_stack_effect_info_for_macro(instr)
|
2023-07-24 13:38:23 -03:00
|
|
|
case parsing.Pseudo():
|
2023-06-14 14:06:50 -03:00
|
|
|
instr = self.pseudo_instrs[thing.name]
|
2023-09-20 13:58:23 -03:00
|
|
|
# Calculate stack effect, and check that it's the same
|
2023-06-11 18:31:59 -03:00
|
|
|
# for all targets.
|
2023-08-25 14:08:29 -03:00
|
|
|
for target in self.pseudos[thing.name].targets:
|
2023-06-11 18:31:59 -03:00
|
|
|
target_instr = self.instrs.get(target)
|
|
|
|
# Currently target is always an instr. This could change
|
|
|
|
# in the future, e.g., if we have a pseudo targetting a
|
|
|
|
# macro instruction.
|
|
|
|
assert target_instr
|
|
|
|
target_popped = effect_str(target_instr.input_effects)
|
|
|
|
target_pushed = effect_str(target_instr.output_effects)
|
2023-08-25 14:08:29 -03:00
|
|
|
if popped is None:
|
2023-06-11 18:31:59 -03:00
|
|
|
popped, pushed = target_popped, target_pushed
|
|
|
|
else:
|
|
|
|
assert popped == target_popped
|
|
|
|
assert pushed == target_pushed
|
2023-01-25 16:41:03 -04:00
|
|
|
case _:
|
2023-08-20 20:40:41 -03:00
|
|
|
assert_never(thing)
|
2023-08-25 14:08:29 -03:00
|
|
|
assert popped is not None and pushed is not None
|
2023-01-25 16:41:03 -04:00
|
|
|
return instr, popped, pushed
|
|
|
|
|
2023-08-14 15:36:29 -03:00
|
|
|
@contextlib.contextmanager
|
2023-08-18 10:42:45 -03:00
|
|
|
def metadata_item(self, signature: str, open: str, close: str) -> Iterator[None]:
|
2023-08-14 15:36:29 -03:00
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit(f"extern {signature};")
|
|
|
|
self.out.emit("#ifdef NEED_OPCODE_METADATA")
|
|
|
|
with self.out.block(f"{signature} {open}", close):
|
|
|
|
yield
|
|
|
|
self.out.emit("#endif // NEED_OPCODE_METADATA")
|
|
|
|
|
2023-01-25 16:41:03 -04:00
|
|
|
def write_stack_effect_functions(self) -> None:
|
2023-01-29 21:28:39 -04:00
|
|
|
popped_data: list[tuple[AnyInstruction, str]] = []
|
|
|
|
pushed_data: list[tuple[AnyInstruction, str]] = []
|
2023-01-25 16:41:03 -04:00
|
|
|
for thing in self.everything:
|
2023-03-04 00:59:21 -04:00
|
|
|
if isinstance(thing, OverriddenInstructionPlaceHolder):
|
|
|
|
continue
|
2023-09-15 12:39:05 -03:00
|
|
|
if isinstance(thing, parsing.Macro) and thing.name in self.instrs:
|
|
|
|
continue
|
2023-01-25 16:41:03 -04:00
|
|
|
instr, popped, pushed = self.get_stack_effect_info(thing)
|
2023-01-26 13:15:05 -04:00
|
|
|
if instr is not None:
|
2023-01-29 21:28:39 -04:00
|
|
|
popped_data.append((instr, popped))
|
|
|
|
pushed_data.append((instr, pushed))
|
2023-01-25 16:41:03 -04:00
|
|
|
|
2023-01-30 21:27:51 -04:00
|
|
|
def write_function(
|
|
|
|
direction: str, data: list[tuple[AnyInstruction, str]]
|
|
|
|
) -> None:
|
2023-08-14 15:36:29 -03:00
|
|
|
with self.metadata_item(
|
2023-08-18 10:42:45 -03:00
|
|
|
f"int _PyOpcode_num_{direction}(int opcode, int oparg, bool jump)",
|
|
|
|
"",
|
|
|
|
"",
|
2023-08-14 15:36:29 -03:00
|
|
|
):
|
|
|
|
with self.out.block("switch(opcode)"):
|
|
|
|
for instr, effect in data:
|
|
|
|
self.out.emit(f"case {instr.name}:")
|
|
|
|
self.out.emit(f" return {effect};")
|
|
|
|
self.out.emit("default:")
|
|
|
|
self.out.emit(" return -1;")
|
2023-01-25 16:41:03 -04:00
|
|
|
|
2023-01-29 21:28:39 -04:00
|
|
|
write_function("popped", popped_data)
|
|
|
|
write_function("pushed", pushed_data)
|
2023-02-08 20:23:19 -04:00
|
|
|
self.out.emit("")
|
2023-01-25 16:41:03 -04:00
|
|
|
|
2023-03-04 00:59:21 -04:00
|
|
|
def from_source_files(self) -> str:
|
2023-07-20 20:08:52 -03:00
|
|
|
filenames = []
|
|
|
|
for filename in self.input_filenames:
|
|
|
|
try:
|
|
|
|
filename = os.path.relpath(filename, ROOT)
|
|
|
|
except ValueError:
|
2023-07-24 13:38:23 -03:00
|
|
|
# May happen on Windows if root and temp on different volumes
|
2023-07-20 20:08:52 -03:00
|
|
|
pass
|
2023-08-22 16:49:39 -03:00
|
|
|
filenames.append(filename.replace(os.path.sep, posixpath.sep))
|
2023-07-20 20:08:52 -03:00
|
|
|
paths = f"\n{self.out.comment} ".join(filenames)
|
2023-06-19 19:47:04 -03:00
|
|
|
return f"{self.out.comment} from:\n{self.out.comment} {paths}\n"
|
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
def write_provenance_header(self) -> None:
|
2023-06-19 19:47:04 -03:00
|
|
|
self.out.write_raw(f"{self.out.comment} This file is generated by {THIS}\n")
|
|
|
|
self.out.write_raw(self.from_source_files())
|
|
|
|
self.out.write_raw(f"{self.out.comment} Do not edit!\n")
|
2023-03-04 00:59:21 -04:00
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
def assign_opcode_ids(self) -> None:
|
2023-08-16 19:25:18 -03:00
|
|
|
"""Assign IDs to opcodes"""
|
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
ops: list[tuple[bool, str]] = [] # (has_arg, name) for each opcode
|
2023-08-16 19:25:18 -03:00
|
|
|
instrumented_ops: list[str] = []
|
|
|
|
|
2023-09-15 12:39:05 -03:00
|
|
|
specialized_ops: set[str] = set()
|
2023-09-12 07:36:17 -03:00
|
|
|
for name, family in self.families.items():
|
|
|
|
specialized_ops.update(family.members)
|
|
|
|
|
2023-09-15 12:39:05 -03:00
|
|
|
for instr in self.macro_instrs.values():
|
2023-08-16 19:25:18 -03:00
|
|
|
name = instr.name
|
2023-09-12 07:36:17 -03:00
|
|
|
if name in specialized_ops:
|
|
|
|
continue
|
2023-08-18 10:42:45 -03:00
|
|
|
if name.startswith("INSTRUMENTED_"):
|
2023-08-16 19:25:18 -03:00
|
|
|
instrumented_ops.append(name)
|
|
|
|
else:
|
|
|
|
ops.append((instr.instr_flags.HAS_ARG_FLAG, name))
|
|
|
|
|
|
|
|
# Special case: this instruction is implemented in ceval.c
|
|
|
|
# rather than bytecodes.c, so we need to add it explicitly
|
|
|
|
# here (at least until we add something to bytecodes.c to
|
|
|
|
# declare external instructions).
|
2023-08-18 10:42:45 -03:00
|
|
|
instrumented_ops.append("INSTRUMENTED_LINE")
|
2023-08-16 19:25:18 -03:00
|
|
|
|
|
|
|
# assert lists are unique
|
|
|
|
assert len(set(ops)) == len(ops)
|
|
|
|
assert len(set(instrumented_ops)) == len(instrumented_ops)
|
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
opname: list[str | None] = [None] * 512
|
|
|
|
opmap: dict[str, int] = {}
|
|
|
|
markers: dict[str, int] = {}
|
2023-08-16 19:25:18 -03:00
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
def map_op(op: int, name: str) -> None:
|
2023-08-16 19:25:18 -03:00
|
|
|
assert op < len(opname)
|
2023-09-12 07:36:17 -03:00
|
|
|
assert opname[op] is None, (op, name)
|
2023-08-16 19:25:18 -03:00
|
|
|
assert name not in opmap
|
|
|
|
opname[op] = name
|
|
|
|
opmap[name] = op
|
|
|
|
|
|
|
|
# 0 is reserved for cache entries. This helps debugging.
|
2023-08-18 10:42:45 -03:00
|
|
|
map_op(0, "CACHE")
|
2023-08-16 19:25:18 -03:00
|
|
|
|
|
|
|
# 17 is reserved as it is the initial value for the specializing counter.
|
|
|
|
# This helps catch cases where we attempt to execute a cache.
|
2023-08-18 10:42:45 -03:00
|
|
|
map_op(17, "RESERVED")
|
2023-08-16 19:25:18 -03:00
|
|
|
|
2023-09-12 07:36:17 -03:00
|
|
|
# 149 is RESUME - it is hard coded as such in Tools/build/deepfreeze.py
|
|
|
|
map_op(149, "RESUME")
|
2023-08-16 19:25:18 -03:00
|
|
|
|
2023-09-12 07:36:17 -03:00
|
|
|
# Specialized ops appear in their own section
|
|
|
|
# Instrumented opcodes are at the end of the valid range
|
|
|
|
min_internal = 150
|
|
|
|
min_instrumented = 254 - (len(instrumented_ops) - 1)
|
|
|
|
assert min_internal + len(specialized_ops) < min_instrumented
|
2023-08-16 19:25:18 -03:00
|
|
|
|
2023-09-12 07:36:17 -03:00
|
|
|
next_opcode = 1
|
2023-08-16 19:25:18 -03:00
|
|
|
for has_arg, name in sorted(ops):
|
|
|
|
if name in opmap:
|
|
|
|
continue # an anchored name, like CACHE
|
|
|
|
map_op(next_opcode, name)
|
2023-08-18 10:42:45 -03:00
|
|
|
if has_arg and "HAVE_ARGUMENT" not in markers:
|
|
|
|
markers["HAVE_ARGUMENT"] = next_opcode
|
2023-08-16 19:25:18 -03:00
|
|
|
|
2023-09-12 07:36:17 -03:00
|
|
|
while opname[next_opcode] is not None:
|
|
|
|
next_opcode += 1
|
|
|
|
|
2023-09-15 12:39:05 -03:00
|
|
|
assert next_opcode < min_internal, next_opcode
|
2023-09-12 07:36:17 -03:00
|
|
|
|
|
|
|
for i, op in enumerate(sorted(specialized_ops)):
|
|
|
|
map_op(min_internal + i, op)
|
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
markers["MIN_INSTRUMENTED_OPCODE"] = min_instrumented
|
2023-08-16 19:25:18 -03:00
|
|
|
for i, op in enumerate(instrumented_ops):
|
|
|
|
map_op(min_instrumented + i, op)
|
|
|
|
|
|
|
|
# Pseudo opcodes are after the valid range
|
|
|
|
for i, op in enumerate(sorted(self.pseudos)):
|
|
|
|
map_op(256 + i, op)
|
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
assert 255 not in opmap.values() # 255 is reserved
|
2023-08-16 19:25:18 -03:00
|
|
|
self.opmap = opmap
|
|
|
|
self.markers = markers
|
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
def write_opcode_ids(
|
|
|
|
self, opcode_ids_h_filename: str, opcode_targets_filename: str
|
|
|
|
) -> None:
|
2023-08-16 19:25:18 -03:00
|
|
|
"""Write header file that defined the opcode IDs"""
|
|
|
|
|
|
|
|
with open(opcode_ids_h_filename, "w") as f:
|
|
|
|
# Create formatter
|
|
|
|
self.out = Formatter(f, 0)
|
|
|
|
|
|
|
|
self.write_provenance_header()
|
|
|
|
|
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit("#ifndef Py_OPCODE_IDS_H")
|
|
|
|
self.out.emit("#define Py_OPCODE_IDS_H")
|
|
|
|
self.out.emit("#ifdef __cplusplus")
|
2023-08-18 10:42:45 -03:00
|
|
|
self.out.emit('extern "C" {')
|
2023-08-16 19:25:18 -03:00
|
|
|
self.out.emit("#endif")
|
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit("/* Instruction opcodes for compiled code */")
|
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
def define(name: str, opcode: int) -> None:
|
2023-08-16 19:25:18 -03:00
|
|
|
self.out.emit(f"#define {name:<38} {opcode:>3}")
|
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
all_pairs: list[tuple[int, int, str]] = []
|
2023-08-16 19:25:18 -03:00
|
|
|
# the second item in the tuple sorts the markers before the ops
|
|
|
|
all_pairs.extend((i, 1, name) for (name, i) in self.markers.items())
|
|
|
|
all_pairs.extend((i, 2, name) for (name, i) in self.opmap.items())
|
|
|
|
for i, _, name in sorted(all_pairs):
|
|
|
|
assert name is not None
|
|
|
|
define(name, i)
|
|
|
|
|
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit("#ifdef __cplusplus")
|
|
|
|
self.out.emit("}")
|
|
|
|
self.out.emit("#endif")
|
|
|
|
self.out.emit("#endif /* !Py_OPCODE_IDS_H */")
|
|
|
|
|
|
|
|
with open(opcode_targets_filename, "w") as f:
|
|
|
|
# Create formatter
|
|
|
|
self.out = Formatter(f, 0)
|
|
|
|
|
|
|
|
with self.out.block("static void *opcode_targets[256] =", ";"):
|
|
|
|
targets = ["_unknown_opcode"] * 256
|
|
|
|
for name, op in self.opmap.items():
|
|
|
|
if op < 256:
|
|
|
|
targets[op] = f"TARGET_{name}"
|
|
|
|
f.write(",\n".join([f" &&{s}" for s in targets]))
|
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
def write_metadata(self, metadata_filename: str, pymetadata_filename: str) -> None:
|
2023-01-05 17:01:07 -04:00
|
|
|
"""Write instruction metadata to output file."""
|
2023-01-09 19:53:01 -04:00
|
|
|
|
|
|
|
# Compute the set of all instruction formats.
|
|
|
|
all_formats: set[str] = set()
|
|
|
|
for thing in self.everything:
|
2023-08-25 14:08:29 -03:00
|
|
|
format: str | None = None
|
2023-01-09 19:53:01 -04:00
|
|
|
match thing:
|
2023-03-04 00:59:21 -04:00
|
|
|
case OverriddenInstructionPlaceHolder():
|
|
|
|
continue
|
2023-07-24 13:38:23 -03:00
|
|
|
case parsing.InstDef():
|
2023-01-09 19:53:01 -04:00
|
|
|
format = self.instrs[thing.name].instr_fmt
|
2023-07-24 13:38:23 -03:00
|
|
|
case parsing.Macro():
|
2023-01-09 19:53:01 -04:00
|
|
|
format = self.macro_instrs[thing.name].instr_fmt
|
2023-07-24 13:38:23 -03:00
|
|
|
case parsing.Pseudo():
|
2023-08-25 14:08:29 -03:00
|
|
|
for target in self.pseudos[thing.name].targets:
|
2023-06-11 18:31:59 -03:00
|
|
|
target_instr = self.instrs.get(target)
|
|
|
|
assert target_instr
|
2023-08-25 14:08:29 -03:00
|
|
|
if format is None:
|
2023-06-11 18:31:59 -03:00
|
|
|
format = target_instr.instr_fmt
|
|
|
|
else:
|
|
|
|
assert format == target_instr.instr_fmt
|
2023-01-09 19:53:01 -04:00
|
|
|
case _:
|
2023-08-20 20:40:41 -03:00
|
|
|
assert_never(thing)
|
2023-08-25 14:08:29 -03:00
|
|
|
assert format is not None
|
2023-01-09 19:53:01 -04:00
|
|
|
all_formats.add(format)
|
2023-08-04 13:35:56 -03:00
|
|
|
|
|
|
|
# Turn it into a sorted list of enum values.
|
2023-01-09 19:53:01 -04:00
|
|
|
format_enums = [INSTR_FMT_PREFIX + format for format in sorted(all_formats)]
|
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
with open(metadata_filename, "w") as f:
|
2023-03-15 12:37:36 -03:00
|
|
|
# Create formatter
|
|
|
|
self.out = Formatter(f, 0)
|
|
|
|
|
2023-06-19 19:47:04 -03:00
|
|
|
self.write_provenance_header()
|
2023-01-05 17:01:07 -04:00
|
|
|
|
2023-09-15 12:39:05 -03:00
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit("#ifndef Py_BUILD_CORE")
|
|
|
|
self.out.emit('# error "this header requires Py_BUILD_CORE define"')
|
|
|
|
self.out.emit("#endif")
|
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit("#include <stdbool.h> // bool")
|
2023-07-14 14:41:52 -03:00
|
|
|
|
2023-06-11 18:31:59 -03:00
|
|
|
self.write_pseudo_instrs()
|
2023-01-05 17:01:07 -04:00
|
|
|
|
2023-06-28 15:28:07 -03:00
|
|
|
self.out.emit("")
|
|
|
|
self.write_uop_items(lambda name, counter: f"#define {name} {counter}")
|
2023-06-26 23:02:57 -03:00
|
|
|
|
2023-01-25 16:41:03 -04:00
|
|
|
self.write_stack_effect_functions()
|
|
|
|
|
2023-08-04 13:35:56 -03:00
|
|
|
# Write the enum definition for instruction formats.
|
|
|
|
with self.out.block("enum InstructionFormat", ";"):
|
|
|
|
for enum in format_enums:
|
|
|
|
self.out.emit(enum + ",")
|
2023-06-21 20:14:43 -03:00
|
|
|
|
2023-07-14 14:41:52 -03:00
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit(
|
|
|
|
"#define IS_VALID_OPCODE(OP) \\\n"
|
|
|
|
" (((OP) >= 0) && ((OP) < OPCODE_METADATA_SIZE) && \\\n"
|
2023-07-24 13:38:23 -03:00
|
|
|
" (_PyOpcode_opcode_metadata[(OP)].valid_entry))"
|
|
|
|
)
|
2023-07-14 14:41:52 -03:00
|
|
|
|
|
|
|
self.out.emit("")
|
2023-06-21 20:14:43 -03:00
|
|
|
InstructionFlags.emit_macros(self.out)
|
2023-06-13 17:42:03 -03:00
|
|
|
|
2023-07-14 14:41:52 -03:00
|
|
|
self.out.emit("")
|
2023-06-26 23:02:57 -03:00
|
|
|
with self.out.block("struct opcode_metadata", ";"):
|
2023-01-05 17:01:07 -04:00
|
|
|
self.out.emit("bool valid_entry;")
|
2023-01-09 19:53:01 -04:00
|
|
|
self.out.emit("enum InstructionFormat instr_format;")
|
2023-06-13 17:42:03 -03:00
|
|
|
self.out.emit("int flags;")
|
2023-06-26 23:02:57 -03:00
|
|
|
self.out.emit("")
|
|
|
|
|
|
|
|
with self.out.block("struct opcode_macro_expansion", ";"):
|
|
|
|
self.out.emit("int nuops;")
|
2023-07-24 13:38:23 -03:00
|
|
|
self.out.emit(
|
2023-08-24 21:36:00 -03:00
|
|
|
"struct { int16_t uop; int8_t size; int8_t offset; } uops[12];"
|
2023-07-24 13:38:23 -03:00
|
|
|
)
|
2023-06-26 23:02:57 -03:00
|
|
|
self.out.emit("")
|
|
|
|
|
2023-07-07 14:42:10 -03:00
|
|
|
for key, value in OPARG_SIZES.items():
|
|
|
|
self.out.emit(f"#define {key} {value}")
|
2023-02-08 20:23:19 -04:00
|
|
|
self.out.emit("")
|
2023-07-07 14:42:10 -03:00
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
self.out.emit(
|
|
|
|
"#define OPCODE_METADATA_FMT(OP) "
|
|
|
|
"(_PyOpcode_opcode_metadata[(OP)].instr_format)"
|
|
|
|
)
|
2023-06-11 18:31:59 -03:00
|
|
|
self.out.emit("#define SAME_OPCODE_METADATA(OP1, OP2) \\")
|
2023-07-24 13:38:23 -03:00
|
|
|
self.out.emit(
|
|
|
|
" (OPCODE_METADATA_FMT(OP1) == OPCODE_METADATA_FMT(OP2))"
|
|
|
|
)
|
2023-06-11 18:31:59 -03:00
|
|
|
self.out.emit("")
|
2023-02-08 20:23:19 -04:00
|
|
|
|
|
|
|
# Write metadata array declaration
|
2023-07-14 14:41:52 -03:00
|
|
|
self.out.emit("#define OPCODE_METADATA_SIZE 512")
|
|
|
|
self.out.emit("#define OPCODE_UOP_NAME_SIZE 512")
|
|
|
|
self.out.emit("#define OPCODE_MACRO_EXPANSION_SIZE 256")
|
2023-06-26 23:02:57 -03:00
|
|
|
|
2023-08-14 15:36:29 -03:00
|
|
|
with self.metadata_item(
|
2023-07-24 13:38:23 -03:00
|
|
|
"const struct opcode_metadata "
|
2023-08-14 15:36:29 -03:00
|
|
|
"_PyOpcode_opcode_metadata[OPCODE_METADATA_SIZE]",
|
|
|
|
"=",
|
2023-08-18 10:42:45 -03:00
|
|
|
";",
|
2023-08-14 15:36:29 -03:00
|
|
|
):
|
|
|
|
# Write metadata for each instruction
|
|
|
|
for thing in self.everything:
|
|
|
|
match thing:
|
|
|
|
case OverriddenInstructionPlaceHolder():
|
|
|
|
continue
|
|
|
|
case parsing.InstDef():
|
2023-08-29 13:51:51 -03:00
|
|
|
self.write_metadata_for_inst(self.instrs[thing.name])
|
2023-08-14 15:36:29 -03:00
|
|
|
case parsing.Macro():
|
2023-09-15 12:39:05 -03:00
|
|
|
if thing.name not in self.instrs:
|
|
|
|
self.write_metadata_for_macro(
|
|
|
|
self.macro_instrs[thing.name]
|
|
|
|
)
|
2023-08-14 15:36:29 -03:00
|
|
|
case parsing.Pseudo():
|
2023-08-18 10:42:45 -03:00
|
|
|
self.write_metadata_for_pseudo(
|
|
|
|
self.pseudo_instrs[thing.name]
|
|
|
|
)
|
2023-08-14 15:36:29 -03:00
|
|
|
case _:
|
2023-08-20 20:40:41 -03:00
|
|
|
assert_never(thing)
|
2023-01-05 17:01:07 -04:00
|
|
|
|
2023-08-14 15:36:29 -03:00
|
|
|
with self.metadata_item(
|
2023-07-14 14:41:52 -03:00
|
|
|
"const struct opcode_macro_expansion "
|
2023-08-14 15:36:29 -03:00
|
|
|
"_PyOpcode_macro_expansion[OPCODE_MACRO_EXPANSION_SIZE]",
|
|
|
|
"=",
|
2023-08-18 10:42:45 -03:00
|
|
|
";",
|
2023-06-26 23:02:57 -03:00
|
|
|
):
|
|
|
|
# Write macro expansion for each non-pseudo instruction
|
2023-09-15 12:39:05 -03:00
|
|
|
for mac in self.macro_instrs.values():
|
|
|
|
if is_super_instruction(mac):
|
|
|
|
# Special-case the heck out of super-instructions
|
|
|
|
self.write_super_expansions(mac.name)
|
|
|
|
else:
|
|
|
|
self.write_macro_expansions(
|
|
|
|
mac.name, mac.parts, mac.cache_offset
|
|
|
|
)
|
2023-06-26 23:02:57 -03:00
|
|
|
|
2023-08-14 15:36:29 -03:00
|
|
|
with self.metadata_item(
|
|
|
|
"const char * const _PyOpcode_uop_name[OPCODE_UOP_NAME_SIZE]", "=", ";"
|
2023-07-24 13:38:23 -03:00
|
|
|
):
|
|
|
|
self.write_uop_items(lambda name, counter: f'[{name}] = "{name}",')
|
2023-06-28 15:28:07 -03:00
|
|
|
|
2023-08-16 19:25:18 -03:00
|
|
|
with self.metadata_item(
|
2023-08-18 10:42:45 -03:00
|
|
|
f"const char *const _PyOpcode_OpName[{1 + max(self.opmap.values())}]",
|
|
|
|
"=",
|
|
|
|
";",
|
2023-08-16 19:25:18 -03:00
|
|
|
):
|
|
|
|
for name in self.opmap:
|
|
|
|
self.out.emit(f'[{name}] = "{name}",')
|
|
|
|
|
2023-08-23 14:39:00 -03:00
|
|
|
with self.metadata_item(
|
|
|
|
f"const uint8_t _PyOpcode_Caches[256]",
|
|
|
|
"=",
|
|
|
|
";",
|
|
|
|
):
|
2023-09-11 15:20:24 -03:00
|
|
|
family_member_names: set[str] = set()
|
|
|
|
for family in self.families.values():
|
|
|
|
family_member_names.update(family.members)
|
2023-09-15 12:39:05 -03:00
|
|
|
for mac in self.macro_instrs.values():
|
2023-09-11 15:20:24 -03:00
|
|
|
if (
|
2023-09-15 12:39:05 -03:00
|
|
|
mac.cache_offset > 0
|
|
|
|
and mac.name not in family_member_names
|
|
|
|
and not mac.name.startswith("INSTRUMENTED_")
|
2023-09-11 15:20:24 -03:00
|
|
|
):
|
|
|
|
self.out.emit(f"[{mac.name}] = {mac.cache_offset},")
|
2023-08-23 14:39:00 -03:00
|
|
|
# Irregular case:
|
2023-09-15 12:39:05 -03:00
|
|
|
self.out.emit("[JUMP_BACKWARD] = 1,")
|
2023-08-23 14:39:00 -03:00
|
|
|
|
2023-08-16 19:25:18 -03:00
|
|
|
deoptcodes = {}
|
|
|
|
for name, op in self.opmap.items():
|
|
|
|
if op < 256:
|
|
|
|
deoptcodes[name] = name
|
|
|
|
for name, family in self.families.items():
|
|
|
|
for m in family.members:
|
|
|
|
deoptcodes[m] = name
|
|
|
|
# special case:
|
2023-08-18 10:42:45 -03:00
|
|
|
deoptcodes["BINARY_OP_INPLACE_ADD_UNICODE"] = "BINARY_OP"
|
2023-08-16 19:25:18 -03:00
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
with self.metadata_item(f"const uint8_t _PyOpcode_Deopt[256]", "=", ";"):
|
2023-08-16 19:25:18 -03:00
|
|
|
for opt, deopt in sorted(deoptcodes.items()):
|
|
|
|
self.out.emit(f"[{opt}] = {deopt},")
|
|
|
|
|
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit("#define EXTRA_CASES \\")
|
|
|
|
valid_opcodes = set(self.opmap.values())
|
|
|
|
with self.out.indent():
|
|
|
|
for op in range(256):
|
|
|
|
if op not in valid_opcodes:
|
|
|
|
self.out.emit(f"case {op}: \\")
|
|
|
|
self.out.emit(" ;\n")
|
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
with open(pymetadata_filename, "w") as f:
|
2023-06-19 19:47:04 -03:00
|
|
|
# Create formatter
|
2023-07-24 13:38:23 -03:00
|
|
|
self.out = Formatter(f, 0, comment="#")
|
2023-06-19 19:47:04 -03:00
|
|
|
|
|
|
|
self.write_provenance_header()
|
|
|
|
|
2023-08-16 19:25:18 -03:00
|
|
|
# emit specializations
|
|
|
|
specialized_ops = set()
|
|
|
|
|
2023-06-19 19:47:04 -03:00
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit("_specializations = {")
|
|
|
|
for name, family in self.families.items():
|
|
|
|
with self.out.indent():
|
2023-07-24 13:38:23 -03:00
|
|
|
self.out.emit(f'"{family.name}": [')
|
2023-06-19 19:47:04 -03:00
|
|
|
with self.out.indent():
|
2023-07-16 12:16:34 -03:00
|
|
|
for m in family.members:
|
2023-07-24 13:38:23 -03:00
|
|
|
self.out.emit(f'"{m}",')
|
2023-08-16 19:25:18 -03:00
|
|
|
specialized_ops.update(family.members)
|
2023-06-19 19:47:04 -03:00
|
|
|
self.out.emit(f"],")
|
|
|
|
self.out.emit("}")
|
|
|
|
|
|
|
|
# Handle special case
|
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit("# An irregular case:")
|
|
|
|
self.out.emit(
|
2023-07-24 13:38:23 -03:00
|
|
|
'_specializations["BINARY_OP"].append('
|
|
|
|
'"BINARY_OP_INPLACE_ADD_UNICODE")'
|
|
|
|
)
|
2023-08-16 19:25:18 -03:00
|
|
|
specialized_ops.add("BINARY_OP_INPLACE_ADD_UNICODE")
|
2023-06-19 19:47:04 -03:00
|
|
|
|
2023-08-16 19:25:18 -03:00
|
|
|
ops = sorted((id, name) for (name, id) in self.opmap.items())
|
|
|
|
# emit specialized opmap
|
2023-06-19 19:47:04 -03:00
|
|
|
self.out.emit("")
|
2023-08-16 19:25:18 -03:00
|
|
|
with self.out.block("_specialized_opmap ="):
|
|
|
|
for op, name in ops:
|
|
|
|
if name in specialized_ops:
|
|
|
|
self.out.emit(f"'{name}': {op},")
|
|
|
|
|
|
|
|
# emit opmap
|
|
|
|
self.out.emit("")
|
|
|
|
with self.out.block("opmap ="):
|
|
|
|
for op, name in ops:
|
|
|
|
if name not in specialized_ops:
|
|
|
|
self.out.emit(f"'{name}': {op},")
|
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
for name in ["MIN_INSTRUMENTED_OPCODE", "HAVE_ARGUMENT"]:
|
2023-08-16 19:25:18 -03:00
|
|
|
self.out.emit(f"{name} = {self.markers[name]}")
|
|
|
|
|
2023-06-11 18:31:59 -03:00
|
|
|
def write_pseudo_instrs(self) -> None:
|
|
|
|
"""Write the IS_PSEUDO_INSTR macro"""
|
2023-07-01 07:28:07 -03:00
|
|
|
self.out.emit("\n\n#define IS_PSEUDO_INSTR(OP) ( \\")
|
2023-06-11 18:31:59 -03:00
|
|
|
for op in self.pseudos:
|
|
|
|
self.out.emit(f" ((OP) == {op}) || \\")
|
2023-07-01 07:28:07 -03:00
|
|
|
self.out.emit(f" 0)")
|
2023-06-11 18:31:59 -03:00
|
|
|
|
2023-06-28 15:28:07 -03:00
|
|
|
def write_uop_items(self, make_text: typing.Callable[[str, int], str]) -> None:
|
2023-06-26 23:02:57 -03:00
|
|
|
"""Write '#define XXX NNN' for each uop"""
|
2023-06-28 15:28:07 -03:00
|
|
|
counter = 300 # TODO: Avoid collision with pseudo instructions
|
2023-07-13 16:14:51 -03:00
|
|
|
seen = set()
|
2023-07-10 20:04:26 -03:00
|
|
|
|
2023-06-26 23:02:57 -03:00
|
|
|
def add(name: str) -> None:
|
2023-07-13 16:14:51 -03:00
|
|
|
if name in seen:
|
|
|
|
return
|
2023-06-26 23:02:57 -03:00
|
|
|
nonlocal counter
|
2023-06-28 15:28:07 -03:00
|
|
|
self.out.emit(make_text(name, counter))
|
2023-06-26 23:02:57 -03:00
|
|
|
counter += 1
|
2023-07-13 16:14:51 -03:00
|
|
|
seen.add(name)
|
2023-07-10 20:04:26 -03:00
|
|
|
|
2023-07-13 16:14:51 -03:00
|
|
|
# These two are first by convention
|
2023-09-11 19:39:19 -03:00
|
|
|
add("_EXIT_TRACE")
|
|
|
|
add("_SET_IP")
|
2023-07-10 20:04:26 -03:00
|
|
|
|
2023-06-26 23:02:57 -03:00
|
|
|
for instr in self.instrs.values():
|
2023-09-15 12:39:05 -03:00
|
|
|
# Skip ops that are also macros -- those are desugared inst()s
|
|
|
|
if instr.name not in self.macros:
|
2023-06-26 23:02:57 -03:00
|
|
|
add(instr.name)
|
|
|
|
|
2023-08-16 20:26:43 -03:00
|
|
|
def write_macro_expansions(
|
|
|
|
self, name: str, parts: MacroParts, cache_offset: int
|
|
|
|
) -> None:
|
2023-06-28 15:28:07 -03:00
|
|
|
"""Write the macro expansions for a macro-instruction."""
|
|
|
|
# TODO: Refactor to share code with write_cody(), is_viaible_uop(), etc.
|
|
|
|
offset = 0 # Cache effect offset
|
|
|
|
expansions: list[tuple[str, int, int]] = [] # [(name, size, offset), ...]
|
2023-06-29 17:02:25 -03:00
|
|
|
for part in parts:
|
2023-06-28 15:28:07 -03:00
|
|
|
if isinstance(part, Component):
|
|
|
|
# All component instructions must be viable uops
|
|
|
|
if not part.instr.is_viable_uop():
|
2023-08-04 13:35:56 -03:00
|
|
|
# This note just reminds us about macros that cannot
|
|
|
|
# be expanded to Tier 2 uops. It is not an error.
|
|
|
|
# It is sometimes emitted for macros that have a
|
|
|
|
# manual translation in translate_bytecode_to_trace()
|
|
|
|
# in Python/optimizer.c.
|
2023-09-15 12:39:05 -03:00
|
|
|
if len(parts) > 1 or part.instr.name != name:
|
|
|
|
self.note(
|
|
|
|
f"Part {part.instr.name} of {name} is not a viable uop",
|
|
|
|
part.instr.inst,
|
|
|
|
)
|
2023-06-28 15:28:07 -03:00
|
|
|
return
|
2023-07-17 16:12:33 -03:00
|
|
|
if not part.active_caches:
|
2023-09-11 19:39:19 -03:00
|
|
|
if part.instr.name == "_SET_IP":
|
|
|
|
size, offset = OPARG_SIZES["OPARG_SET_IP"], cache_offset
|
2023-08-16 20:26:43 -03:00
|
|
|
else:
|
|
|
|
size, offset = OPARG_SIZES["OPARG_FULL"], 0
|
2023-06-28 15:28:07 -03:00
|
|
|
else:
|
|
|
|
# If this assert triggers, is_viable_uops() lied
|
2023-06-29 17:02:25 -03:00
|
|
|
assert len(part.active_caches) == 1, (name, part.instr.name)
|
2023-06-28 15:28:07 -03:00
|
|
|
cache = part.active_caches[0]
|
|
|
|
size, offset = cache.effect.size, cache.offset
|
|
|
|
expansions.append((part.instr.name, size, offset))
|
2023-06-29 17:02:25 -03:00
|
|
|
assert len(expansions) > 0, f"Macro {name} has empty expansion?!"
|
2023-07-07 14:42:10 -03:00
|
|
|
self.write_expansions(name, expansions)
|
|
|
|
|
|
|
|
def write_super_expansions(self, name: str) -> None:
|
|
|
|
"""Write special macro expansions for super-instructions.
|
|
|
|
|
|
|
|
If you get an assertion failure here, you probably have accidentally
|
|
|
|
violated one of the assumptions here.
|
|
|
|
|
|
|
|
- A super-instruction's name is of the form FIRST_SECOND where
|
|
|
|
FIRST and SECOND are regular instructions whose name has the
|
|
|
|
form FOO_BAR. Thus, there must be exactly 3 underscores.
|
|
|
|
Example: LOAD_CONST_STORE_FAST.
|
|
|
|
|
|
|
|
- A super-instruction's body uses `oparg1 and `oparg2`, and no
|
|
|
|
other instruction's body uses those variable names.
|
|
|
|
|
|
|
|
- A super-instruction has no active (used) cache entries.
|
|
|
|
|
|
|
|
In the expansion, the first instruction's operand is all but the
|
|
|
|
bottom 4 bits of the super-instruction's oparg, and the second
|
|
|
|
instruction's operand is the bottom 4 bits. We use the special
|
|
|
|
size codes OPARG_TOP and OPARG_BOTTOM for these.
|
|
|
|
"""
|
|
|
|
pieces = name.split("_")
|
|
|
|
assert len(pieces) == 4, f"{name} doesn't look like a super-instr"
|
|
|
|
name1 = "_".join(pieces[:2])
|
|
|
|
name2 = "_".join(pieces[2:])
|
|
|
|
assert name1 in self.instrs, f"{name1} doesn't match any instr"
|
|
|
|
assert name2 in self.instrs, f"{name2} doesn't match any instr"
|
|
|
|
instr1 = self.instrs[name1]
|
|
|
|
instr2 = self.instrs[name2]
|
|
|
|
assert not instr1.active_caches, f"{name1} has active caches"
|
|
|
|
assert not instr2.active_caches, f"{name2} has active caches"
|
2023-08-04 13:35:56 -03:00
|
|
|
expansions: list[tuple[str, int, int]] = [
|
2023-07-07 14:42:10 -03:00
|
|
|
(name1, OPARG_SIZES["OPARG_TOP"], 0),
|
|
|
|
(name2, OPARG_SIZES["OPARG_BOTTOM"], 0),
|
|
|
|
]
|
|
|
|
self.write_expansions(name, expansions)
|
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
def write_expansions(
|
|
|
|
self, name: str, expansions: list[tuple[str, int, int]]
|
|
|
|
) -> None:
|
|
|
|
pieces = [
|
|
|
|
f"{{ {name}, {size}, {offset} }}" for name, size, offset in expansions
|
|
|
|
]
|
2023-06-28 15:28:07 -03:00
|
|
|
self.out.emit(
|
2023-06-29 17:02:25 -03:00
|
|
|
f"[{name}] = "
|
2023-07-07 14:42:10 -03:00
|
|
|
f"{{ .nuops = {len(pieces)}, .uops = {{ {', '.join(pieces)} }} }},"
|
2023-06-28 15:28:07 -03:00
|
|
|
)
|
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
def emit_metadata_entry(self, name: str, fmt: str, flags: InstructionFlags) -> None:
|
2023-06-21 20:14:43 -03:00
|
|
|
flag_names = flags.names(value=True)
|
|
|
|
if not flag_names:
|
|
|
|
flag_names.append("0")
|
2023-01-05 17:01:07 -04:00
|
|
|
self.out.emit(
|
2023-08-14 15:36:29 -03:00
|
|
|
f"[{name}] = {{ true, {INSTR_FMT_PREFIX}{fmt},"
|
2023-06-21 20:14:43 -03:00
|
|
|
f" {' | '.join(flag_names)} }},"
|
2023-01-05 17:01:07 -04:00
|
|
|
)
|
|
|
|
|
2023-06-13 17:42:03 -03:00
|
|
|
def write_metadata_for_inst(self, instr: Instruction) -> None:
|
|
|
|
"""Write metadata for a single instruction."""
|
2023-06-21 20:14:43 -03:00
|
|
|
self.emit_metadata_entry(instr.name, instr.instr_fmt, instr.instr_flags)
|
2023-06-13 17:42:03 -03:00
|
|
|
|
2023-01-05 17:01:07 -04:00
|
|
|
def write_metadata_for_macro(self, mac: MacroInstruction) -> None:
|
|
|
|
"""Write metadata for a macro-instruction."""
|
2023-06-21 20:14:43 -03:00
|
|
|
self.emit_metadata_entry(mac.name, mac.instr_fmt, mac.instr_flags)
|
2023-01-05 17:01:07 -04:00
|
|
|
|
2023-06-11 18:31:59 -03:00
|
|
|
def write_metadata_for_pseudo(self, ps: PseudoInstruction) -> None:
|
|
|
|
"""Write metadata for a macro-instruction."""
|
2023-06-21 20:14:43 -03:00
|
|
|
self.emit_metadata_entry(ps.name, ps.instr_fmt, ps.instr_flags)
|
2023-06-11 18:31:59 -03:00
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
def write_instructions(
|
|
|
|
self, output_filename: str, emit_line_directives: bool
|
|
|
|
) -> None:
|
2022-11-17 21:06:07 -04:00
|
|
|
"""Write instructions to output file."""
|
2023-07-24 13:38:23 -03:00
|
|
|
with open(output_filename, "w") as f:
|
2023-03-15 12:37:36 -03:00
|
|
|
# Create formatter
|
2023-07-24 13:38:23 -03:00
|
|
|
self.out = Formatter(f, 8, emit_line_directives)
|
2022-11-17 21:06:07 -04:00
|
|
|
|
2023-06-26 23:02:57 -03:00
|
|
|
self.write_provenance_header()
|
2022-12-02 23:57:30 -04:00
|
|
|
|
2022-12-08 19:54:07 -04:00
|
|
|
# Write and count instructions of all kinds
|
2022-12-02 23:57:30 -04:00
|
|
|
n_macros = 0
|
2022-12-08 19:54:07 -04:00
|
|
|
for thing in self.everything:
|
|
|
|
match thing:
|
2023-03-04 00:59:21 -04:00
|
|
|
case OverriddenInstructionPlaceHolder():
|
|
|
|
self.write_overridden_instr_place_holder(thing)
|
2023-07-24 13:38:23 -03:00
|
|
|
case parsing.InstDef():
|
2023-09-15 12:39:05 -03:00
|
|
|
pass
|
2023-07-24 13:38:23 -03:00
|
|
|
case parsing.Macro():
|
2022-12-08 19:54:07 -04:00
|
|
|
n_macros += 1
|
2023-08-04 13:35:56 -03:00
|
|
|
mac = self.macro_instrs[thing.name]
|
2023-08-16 20:26:43 -03:00
|
|
|
stacking.write_macro_instr(
|
|
|
|
mac, self.out, self.families.get(mac.name)
|
|
|
|
)
|
2023-07-24 13:38:23 -03:00
|
|
|
case parsing.Pseudo():
|
2023-08-04 13:35:56 -03:00
|
|
|
pass
|
2022-12-08 19:54:07 -04:00
|
|
|
case _:
|
2023-08-20 20:40:41 -03:00
|
|
|
assert_never(thing)
|
2022-11-22 20:04:57 -04:00
|
|
|
|
2022-12-02 23:57:30 -04:00
|
|
|
print(
|
2023-09-15 12:39:05 -03:00
|
|
|
f"Wrote {n_macros} cases to {output_filename}",
|
2022-12-02 23:57:30 -04:00
|
|
|
file=sys.stderr,
|
|
|
|
)
|
2022-11-22 20:04:57 -04:00
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
def write_executor_instructions(
|
|
|
|
self, executor_filename: str, emit_line_directives: bool
|
|
|
|
) -> None:
|
2023-06-26 23:02:57 -03:00
|
|
|
"""Generate cases for the Tier 2 interpreter."""
|
2023-08-04 13:35:56 -03:00
|
|
|
n_uops = 0
|
2023-07-24 13:38:23 -03:00
|
|
|
with open(executor_filename, "w") as f:
|
|
|
|
self.out = Formatter(f, 8, emit_line_directives)
|
2023-06-26 23:02:57 -03:00
|
|
|
self.write_provenance_header()
|
2023-09-15 12:39:05 -03:00
|
|
|
for instr in self.instrs.values():
|
|
|
|
if instr.is_viable_uop():
|
|
|
|
n_uops += 1
|
|
|
|
self.out.emit("")
|
|
|
|
with self.out.block(f"case {instr.name}:"):
|
|
|
|
stacking.write_single_instr(instr, self.out, tier=TIER_TWO)
|
|
|
|
if instr.check_eval_breaker:
|
|
|
|
self.out.emit("CHECK_EVAL_BREAKER();")
|
|
|
|
self.out.emit("break;")
|
2023-06-26 23:02:57 -03:00
|
|
|
print(
|
2023-09-15 12:39:05 -03:00
|
|
|
f"Wrote {n_uops} cases to {executor_filename}",
|
2023-06-26 23:02:57 -03:00
|
|
|
file=sys.stderr,
|
|
|
|
)
|
|
|
|
|
2023-08-15 15:04:17 -03:00
|
|
|
def write_abstract_interpreter_instructions(
|
|
|
|
self, abstract_interpreter_filename: str, emit_line_directives: bool
|
|
|
|
) -> None:
|
|
|
|
"""Generate cases for the Tier 2 abstract interpreter/analzyer."""
|
|
|
|
with open(abstract_interpreter_filename, "w") as f:
|
|
|
|
self.out = Formatter(f, 8, emit_line_directives)
|
|
|
|
self.write_provenance_header()
|
2023-09-15 12:39:05 -03:00
|
|
|
for instr in self.instrs.values():
|
|
|
|
instr = AbstractInstruction(instr.inst)
|
|
|
|
if (
|
|
|
|
instr.is_viable_uop()
|
|
|
|
and instr.name not in SPECIALLY_HANDLED_ABSTRACT_INSTR
|
|
|
|
):
|
|
|
|
self.out.emit("")
|
|
|
|
with self.out.block(f"case {instr.name}:"):
|
|
|
|
instr.write(self.out, tier=TIER_TWO)
|
|
|
|
self.out.emit("break;")
|
2023-08-15 15:04:17 -03:00
|
|
|
print(
|
|
|
|
f"Wrote some stuff to {abstract_interpreter_filename}",
|
|
|
|
file=sys.stderr,
|
|
|
|
)
|
|
|
|
|
2023-07-24 13:38:23 -03:00
|
|
|
def write_overridden_instr_place_holder(
|
|
|
|
self, place_holder: OverriddenInstructionPlaceHolder
|
|
|
|
) -> None:
|
2023-03-04 00:59:21 -04:00
|
|
|
self.out.emit("")
|
|
|
|
self.out.emit(
|
2023-07-24 13:38:23 -03:00
|
|
|
f"{self.out.comment} TARGET({place_holder.name}) overridden by later definition"
|
|
|
|
)
|
2023-03-04 00:59:21 -04:00
|
|
|
|
2023-09-15 12:39:05 -03:00
|
|
|
|
|
|
|
def is_super_instruction(mac: MacroInstruction) -> bool:
|
|
|
|
if (
|
|
|
|
len(mac.parts) == 1
|
|
|
|
and isinstance(mac.parts[0], Component)
|
|
|
|
and variable_used(mac.parts[0].instr.inst, "oparg1")
|
|
|
|
):
|
|
|
|
assert variable_used(mac.parts[0].instr.inst, "oparg2")
|
|
|
|
return True
|
|
|
|
else:
|
|
|
|
return False
|
2022-12-08 19:54:07 -04:00
|
|
|
|
2022-11-03 01:31:26 -03:00
|
|
|
|
2023-08-18 10:42:45 -03:00
|
|
|
def main() -> None:
|
2022-11-17 21:06:07 -04:00
|
|
|
"""Parse command line, parse input, analyze, write output."""
|
|
|
|
args = arg_parser.parse_args() # Prints message and sys.exit(2) on error
|
2023-03-04 00:59:21 -04:00
|
|
|
if len(args.input) == 0:
|
|
|
|
args.input.append(DEFAULT_INPUT)
|
2023-06-19 19:47:04 -03:00
|
|
|
|
|
|
|
# Raises OSError if input unreadable
|
2023-07-24 13:38:23 -03:00
|
|
|
a = Generator(args.input)
|
2023-06-19 19:47:04 -03:00
|
|
|
|
2022-11-17 21:06:07 -04:00
|
|
|
a.parse() # Raises SyntaxError on failure
|
2022-12-02 23:57:30 -04:00
|
|
|
a.analyze() # Prints messages and sets a.errors on failure
|
2022-11-17 21:06:07 -04:00
|
|
|
if a.errors:
|
|
|
|
sys.exit(f"Found {a.errors} errors")
|
2023-07-24 13:38:23 -03:00
|
|
|
|
|
|
|
# These raise OSError if output can't be written
|
|
|
|
a.write_instructions(args.output, args.emit_line_directives)
|
2023-08-16 19:25:18 -03:00
|
|
|
|
|
|
|
a.assign_opcode_ids()
|
|
|
|
a.write_opcode_ids(args.opcode_ids_h, args.opcode_targets_h)
|
2023-07-24 13:38:23 -03:00
|
|
|
a.write_metadata(args.metadata, args.pymetadata)
|
|
|
|
a.write_executor_instructions(args.executor_cases, args.emit_line_directives)
|
2023-08-18 10:42:45 -03:00
|
|
|
a.write_abstract_interpreter_instructions(
|
|
|
|
args.abstract_interpreter_cases, args.emit_line_directives
|
|
|
|
)
|
2022-11-03 01:31:26 -03:00
|
|
|
|
|
|
|
|
|
|
|
if __name__ == "__main__":
|
|
|
|
main()
|