summaryrefslogtreecommitdiffstats
path: root/Tools/cases_generator/optimizer_generator.py
blob: f110a74e3ad3514ef66f399d05530672477b18bb (plain)
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
"""Generate the cases for the tier 2 optimizer.
Reads the instruction definitions from bytecodes.c and optimizer_bytecodes.c
Writes the cases to optimizer_cases.c.h, which is #included in Python/optimizer_analysis.c.
"""

import argparse
import os.path
import sys

from analyzer import (
    Analysis,
    Instruction,
    Uop,
    Part,
    analyze_files,
    Skip,
    StackItem,
    analysis_error,
)
from generators_common import (
    DEFAULT_INPUT,
    ROOT,
    write_header,
    emit_tokens,
    emit_to,
    replace_sync_sp,
)
from cwriter import CWriter
from typing import TextIO, Iterator
from lexer import Token
from stack import StackOffset, Stack, SizeMismatch, UNUSED

DEFAULT_OUTPUT = ROOT / "Python/optimizer_cases.c.h"
DEFAULT_ABSTRACT_INPUT = ROOT / "Python/optimizer_bytecodes.c"


def validate_uop(override: Uop, uop: Uop) -> None:
    # To do
    pass


def type_name(var: StackItem) -> str:
    if var.is_array():
        return f"_Py_UOpsSymType **"
    if var.type:
        return var.type
    return f"_Py_UOpsSymType *"


def declare_variables(uop: Uop, out: CWriter, skip_inputs: bool) -> None:
    variables = {"unused"}
    if not skip_inputs:
        for var in reversed(uop.stack.inputs):
            if var.name not in variables:
                variables.add(var.name)
                if var.condition:
                    out.emit(f"{type_name(var)}{var.name} = NULL;\n")
                else:
                    out.emit(f"{type_name(var)}{var.name};\n")
    for var in uop.stack.outputs:
        if var.peek:
            continue
        if var.name not in variables:
            variables.add(var.name)
            if var.condition:
                out.emit(f"{type_name(var)}{var.name} = NULL;\n")
            else:
                out.emit(f"{type_name(var)}{var.name};\n")


def decref_inputs(
    out: CWriter,
    tkn: Token,
    tkn_iter: Iterator[Token],
    uop: Uop,
    stack: Stack,
    inst: Instruction | None,
) -> None:
    next(tkn_iter)
    next(tkn_iter)
    next(tkn_iter)
    out.emit_at("", tkn)


def emit_default(out: CWriter, uop: Uop) -> None:
    for i, var in enumerate(uop.stack.outputs):
        if var.name != "unused" and not var.peek:
            if var.is_array():
                out.emit(f"for (int _i = {var.size}; --_i >= 0;) {{\n")
                out.emit(f"{var.name}[_i] = _Py_uop_sym_new_unknown(ctx);\n")
                out.emit(f"if ({var.name}[_i] == NULL) goto out_of_space;\n")
                out.emit("}\n")
            elif var.name == "null":
                out.emit(f"{var.name} = _Py_uop_sym_new_null(ctx);\n")
                out.emit(f"if ({var.name} == NULL) goto out_of_space;\n")
            else:
                out.emit(f"{var.name} = _Py_uop_sym_new_unknown(ctx);\n")
                out.emit(f"if ({var.name} == NULL) goto out_of_space;\n")


def write_uop(
    override: Uop | None,
    uop: Uop,
    out: CWriter,
    stack: Stack,
    debug: bool,
    skip_inputs: bool,
) -> None:
    try:
        prototype = override if override else uop
        is_override = override is not None
        out.start_line()
        for var in reversed(prototype.stack.inputs):
            res = stack.pop(var)
            if not skip_inputs:
                out.emit(res)
        if not prototype.properties.stores_sp:
            for i, var in enumerate(prototype.stack.outputs):
                res = stack.push(var)
                if not var.peek or is_override:
                    out.emit(res)
        if debug:
            args = []
            for var in prototype.stack.inputs:
                if not var.peek or is_override:
                    args.append(var.name)
            out.emit(f'DEBUG_PRINTF({", ".join(args)});\n')
        if override:
            for cache in uop.caches:
                if cache.name != "unused":
                    if cache.size == 4:
                        type = cast = "PyObject *"
                    else:
                        type = f"uint{cache.size*16}_t "
                        cast = f"uint{cache.size*16}_t"
                    out.emit(f"{type}{cache.name} = ({cast})this_instr->operand;\n")
        if override:
            replacement_funcs = {
                "DECREF_INPUTS": decref_inputs,
                "SYNC_SP": replace_sync_sp,
            }
            emit_tokens(out, override, stack, None, replacement_funcs)
        else:
            emit_default(out, uop)

        if prototype.properties.stores_sp:
            for i, var in enumerate(prototype.stack.outputs):
                if not var.peek or is_override:
                    out.emit(stack.push(var))
        out.start_line()
        stack.flush(out, cast_type="_Py_UOpsSymType *")
    except SizeMismatch as ex:
        raise analysis_error(ex.args[0], uop.body[0])


SKIPS = ("_EXTENDED_ARG",)


def generate_abstract_interpreter(
    filenames: list[str],
    abstract: Analysis,
    base: Analysis,
    outfile: TextIO,
    debug: bool,
) -> None:
    write_header(__file__, filenames, outfile)
    out = CWriter(outfile, 2, False)
    out.emit("\n")
    base_uop_names = set([uop.name for uop in base.uops.values()])
    for abstract_uop_name in abstract.uops:
        assert abstract_uop_name in base_uop_names,\
            f"All abstract uops should override base uops, but {abstract_uop_name} is not."

    for uop in base.uops.values():
        override: Uop | None = None
        if uop.name in abstract.uops:
            override = abstract.uops[uop.name]
            validate_uop(override, uop)
        if uop.properties.tier == 1:
            continue
        if uop.replicates:
            continue
        if uop.is_super():
            continue
        if not uop.is_viable():
            out.emit(f"/* {uop.name} is not a viable micro-op for tier 2 */\n\n")
            continue
        out.emit(f"case {uop.name}: {{\n")
        if override:
            declare_variables(override, out, skip_inputs=False)
        else:
            declare_variables(uop, out, skip_inputs=True)
        stack = Stack()
        write_uop(override, uop, out, stack, debug, skip_inputs=(override is None))
        out.start_line()
        out.emit("break;\n")
        out.emit("}")
        out.emit("\n\n")


def generate_tier2_abstract_from_files(
    filenames: list[str], outfilename: str, debug: bool=False
) -> None:
    assert len(filenames) == 2, "Need a base file and an abstract cases file."
    base = analyze_files([filenames[0]])
    abstract = analyze_files([filenames[1]])
    with open(outfilename, "w") as outfile:
        generate_abstract_interpreter(filenames, abstract, base, outfile, debug)


arg_parser = argparse.ArgumentParser(
    description="Generate the code for the tier 2 interpreter.",
    formatter_class=argparse.ArgumentDefaultsHelpFormatter,
)

arg_parser.add_argument(
    "-o", "--output", type=str, help="Generated code", default=DEFAULT_OUTPUT
)


arg_parser.add_argument("input", nargs=1, help="Abstract interpreter definition file")

arg_parser.add_argument(
    "base", nargs=argparse.REMAINDER, help="The base instruction definition file(s)"
)

arg_parser.add_argument("-d", "--debug", help="Insert debug calls", action="store_true")

if __name__ == "__main__":
    args = arg_parser.parse_args()
    if len(args.base) == 0:
        args.input.append(DEFAULT_INPUT)
        args.input.append(DEFAULT_ABSTRACT_INPUT)
    abstract = analyze_files(args.input)
    base = analyze_files(args.base)
    with open(args.output, "w") as outfile:
        generate_abstract_interpreter(args.input, abstract, base, outfile, args.debug)