1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
|
"""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.
"""
import argparse
import contextlib
import itertools
import os
import posixpath
import sys
import textwrap
import typing
from collections.abc import Iterator
import stacking # Early import to avoid circular import
from _typing_backports import assert_never
from analysis import Analyzer
from formatting import Formatter, list_effect_size
from flags import InstructionFlags, variable_used
from instructions import (
AnyInstruction,
AbstractInstruction,
Component,
Instruction,
MacroInstruction,
MacroParts,
PseudoInstruction,
OverriddenInstructionPlaceHolder,
TIER_ONE,
TIER_TWO,
)
import parsing
from parsing import StackEffect
HERE = os.path.dirname(__file__)
ROOT = os.path.join(HERE, "../..")
THIS = os.path.relpath(__file__, ROOT).replace(os.path.sep, posixpath.sep)
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"))
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")
)
DEFAULT_METADATA_OUTPUT = os.path.relpath(
os.path.join(ROOT, "Include/internal/pycore_opcode_metadata.h")
)
DEFAULT_PYMETADATA_OUTPUT = os.path.relpath(
os.path.join(ROOT, "Lib/_opcode_metadata.py")
)
DEFAULT_EXECUTOR_OUTPUT = os.path.relpath(
os.path.join(ROOT, "Python/executor_cases.c.h")
)
DEFAULT_ABSTRACT_INTERPRETER_OUTPUT = os.path.relpath(
os.path.join(ROOT, "Python/abstract_interp_cases.c.h")
)
# 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,
"OPARG_SAVE_IP": 7,
}
INSTR_FMT_PREFIX = "INSTR_FMT_"
# 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",
}
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
)
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,
)
arg_parser.add_argument(
"-m",
"--metadata",
type=str,
help="Generated C metadata",
default=DEFAULT_METADATA_OUTPUT,
)
arg_parser.add_argument(
"-p",
"--pymetadata",
type=str,
help="Generated Python metadata",
default=DEFAULT_PYMETADATA_OUTPUT,
)
arg_parser.add_argument(
"-l", "--emit-line-directives", help="Emit #line directives", action="store_true"
)
arg_parser.add_argument(
"input", nargs=argparse.REMAINDER, help="Instruction definition file(s)"
)
arg_parser.add_argument(
"-e",
"--executor-cases",
type=str,
help="Write executor cases to this file",
default=DEFAULT_EXECUTOR_OUTPUT,
)
arg_parser.add_argument(
"-a",
"--abstract-interpreter-cases",
type=str,
help="Write abstract interpreter cases to this file",
default=DEFAULT_ABSTRACT_INTERPRETER_OUTPUT,
)
class Generator(Analyzer):
def get_stack_effect_info(
self, thing: parsing.InstDef | parsing.Macro | parsing.Pseudo
) -> tuple[AnyInstruction | None, str, str]:
def effect_str(effects: list[StackEffect]) -> str:
n_effect, sym_effect = list_effect_size(effects)
if sym_effect:
return f"{sym_effect} + {n_effect}" if n_effect else sym_effect
return str(n_effect)
instr: AnyInstruction | None
popped: str | None = None
pushed: str | None = None
match thing:
case parsing.InstDef():
if thing.kind != "op" or self.instrs[thing.name].is_viable_uop():
instr = self.instrs[thing.name]
popped = effect_str(instr.input_effects)
pushed = effect_str(instr.output_effects)
else:
instr = None
popped = ""
pushed = ""
case parsing.Macro():
instr = self.macro_instrs[thing.name]
popped, pushed = stacking.get_stack_effect_info_for_macro(instr)
case parsing.Pseudo():
instr = self.pseudo_instrs[thing.name]
# Calculate stack effect, and check that it's the the same
# for all targets.
for target in self.pseudos[thing.name].targets:
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)
if popped is None:
popped, pushed = target_popped, target_pushed
else:
assert popped == target_popped
assert pushed == target_pushed
case _:
assert_never(thing)
assert popped is not None and pushed is not None
return instr, popped, pushed
@contextlib.contextmanager
def metadata_item(self, signature: str, open: str, close: str) -> Iterator[None]:
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")
def write_stack_effect_functions(self) -> None:
popped_data: list[tuple[AnyInstruction, str]] = []
pushed_data: list[tuple[AnyInstruction, str]] = []
for thing in self.everything:
if isinstance(thing, OverriddenInstructionPlaceHolder):
continue
instr, popped, pushed = self.get_stack_effect_info(thing)
if instr is not None:
popped_data.append((instr, popped))
pushed_data.append((instr, pushed))
def write_function(
direction: str, data: list[tuple[AnyInstruction, str]]
) -> None:
with self.metadata_item(
f"int _PyOpcode_num_{direction}(int opcode, int oparg, bool jump)",
"",
"",
):
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;")
write_function("popped", popped_data)
write_function("pushed", pushed_data)
self.out.emit("")
def from_source_files(self) -> str:
filenames = []
for filename in self.input_filenames:
try:
filename = os.path.relpath(filename, ROOT)
except ValueError:
# May happen on Windows if root and temp on different volumes
pass
filenames.append(filename.replace(os.path.sep, posixpath.sep))
paths = f"\n{self.out.comment} ".join(filenames)
return f"{self.out.comment} from:\n{self.out.comment} {paths}\n"
def write_provenance_header(self) -> None:
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")
def assign_opcode_ids(self) -> None:
"""Assign IDs to opcodes"""
ops: list[tuple[bool, str]] = [] # (has_arg, name) for each opcode
instrumented_ops: list[str] = []
for instr in itertools.chain(
[instr for instr in self.instrs.values() if instr.kind != "op"],
self.macro_instrs.values(),
):
assert isinstance(instr, (Instruction, MacroInstruction, PseudoInstruction))
name = instr.name
if name.startswith("INSTRUMENTED_"):
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).
instrumented_ops.append("INSTRUMENTED_LINE")
# assert lists are unique
assert len(set(ops)) == len(ops)
assert len(set(instrumented_ops)) == len(instrumented_ops)
opname: list[str | None] = [None] * 512
opmap: dict[str, int] = {}
markers: dict[str, int] = {}
def map_op(op: int, name: str) -> None:
assert op < len(opname)
assert opname[op] is None
assert name not in opmap
opname[op] = name
opmap[name] = op
# 0 is reserved for cache entries. This helps debugging.
map_op(0, "CACHE")
# 17 is reserved as it is the initial value for the specializing counter.
# This helps catch cases where we attempt to execute a cache.
map_op(17, "RESERVED")
# 166 is RESUME - it is hard coded as such in Tools/build/deepfreeze.py
map_op(166, "RESUME")
next_opcode = 1
for has_arg, name in sorted(ops):
if name in opmap:
continue # an anchored name, like CACHE
while opname[next_opcode] is not None:
next_opcode += 1
assert next_opcode < 255
map_op(next_opcode, name)
if has_arg and "HAVE_ARGUMENT" not in markers:
markers["HAVE_ARGUMENT"] = next_opcode
# Instrumented opcodes are at the end of the valid range
min_instrumented = 254 - (len(instrumented_ops) - 1)
assert next_opcode <= min_instrumented
markers["MIN_INSTRUMENTED_OPCODE"] = min_instrumented
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)
assert 255 not in opmap.values() # 255 is reserved
self.opmap = opmap
self.markers = markers
def write_opcode_ids(
self, opcode_ids_h_filename: str, opcode_targets_filename: str
) -> None:
"""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")
self.out.emit('extern "C" {')
self.out.emit("#endif")
self.out.emit("")
self.out.emit("/* Instruction opcodes for compiled code */")
def define(name: str, opcode: int) -> None:
self.out.emit(f"#define {name:<38} {opcode:>3}")
all_pairs: list[tuple[int, int, str]] = []
# 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]))
def write_metadata(self, metadata_filename: str, pymetadata_filename: str) -> None:
"""Write instruction metadata to output file."""
# Compute the set of all instruction formats.
all_formats: set[str] = set()
for thing in self.everything:
format: str | None = None
match thing:
case OverriddenInstructionPlaceHolder():
continue
case parsing.InstDef():
format = self.instrs[thing.name].instr_fmt
case parsing.Macro():
format = self.macro_instrs[thing.name].instr_fmt
case parsing.Pseudo():
for target in self.pseudos[thing.name].targets:
target_instr = self.instrs.get(target)
assert target_instr
if format is None:
format = target_instr.instr_fmt
else:
assert format == target_instr.instr_fmt
case _:
assert_never(thing)
assert format is not None
all_formats.add(format)
# Turn it into a sorted list of enum values.
format_enums = [INSTR_FMT_PREFIX + format for format in sorted(all_formats)]
with open(metadata_filename, "w") as f:
# Create formatter
self.out = Formatter(f, 0)
self.write_provenance_header()
self.out.emit("\n" + textwrap.dedent("""
#ifndef Py_BUILD_CORE
# error "this header requires Py_BUILD_CORE define"
#endif
""").strip())
self.out.emit("\n#include <stdbool.h> // bool")
self.write_pseudo_instrs()
self.out.emit("")
self.write_uop_items(lambda name, counter: f"#define {name} {counter}")
self.write_stack_effect_functions()
# Write the enum definition for instruction formats.
with self.out.block("enum InstructionFormat", ";"):
for enum in format_enums:
self.out.emit(enum + ",")
self.out.emit("")
self.out.emit(
"#define IS_VALID_OPCODE(OP) \\\n"
" (((OP) >= 0) && ((OP) < OPCODE_METADATA_SIZE) && \\\n"
" (_PyOpcode_opcode_metadata[(OP)].valid_entry))"
)
self.out.emit("")
InstructionFlags.emit_macros(self.out)
self.out.emit("")
with self.out.block("struct opcode_metadata", ";"):
self.out.emit("bool valid_entry;")
self.out.emit("enum InstructionFormat instr_format;")
self.out.emit("int flags;")
self.out.emit("")
with self.out.block("struct opcode_macro_expansion", ";"):
self.out.emit("int nuops;")
self.out.emit(
"struct { int16_t uop; int8_t size; int8_t offset; } uops[12];"
)
self.out.emit("")
for key, value in OPARG_SIZES.items():
self.out.emit(f"#define {key} {value}")
self.out.emit("")
self.out.emit(
"#define OPCODE_METADATA_FMT(OP) "
"(_PyOpcode_opcode_metadata[(OP)].instr_format)"
)
self.out.emit("#define SAME_OPCODE_METADATA(OP1, OP2) \\")
self.out.emit(
" (OPCODE_METADATA_FMT(OP1) == OPCODE_METADATA_FMT(OP2))"
)
self.out.emit("")
# Write metadata array declaration
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")
with self.metadata_item(
"const struct opcode_metadata "
"_PyOpcode_opcode_metadata[OPCODE_METADATA_SIZE]",
"=",
";",
):
# Write metadata for each instruction
for thing in self.everything:
match thing:
case OverriddenInstructionPlaceHolder():
continue
case parsing.InstDef():
self.write_metadata_for_inst(self.instrs[thing.name])
case parsing.Macro():
self.write_metadata_for_macro(self.macro_instrs[thing.name])
case parsing.Pseudo():
self.write_metadata_for_pseudo(
self.pseudo_instrs[thing.name]
)
case _:
assert_never(thing)
with self.metadata_item(
"const struct opcode_macro_expansion "
"_PyOpcode_macro_expansion[OPCODE_MACRO_EXPANSION_SIZE]",
"=",
";",
):
# Write macro expansion for each non-pseudo instruction
for thing in self.everything:
match thing:
case OverriddenInstructionPlaceHolder():
pass
case parsing.InstDef(name=name):
instr = self.instrs[name]
# Since an 'op' is not a bytecode, it has no expansion; but 'inst' is
if instr.kind == "inst" and instr.is_viable_uop():
# Construct a dummy Component -- input/output mappings are not used
part = Component(instr, instr.active_caches)
self.write_macro_expansions(
instr.name, [part], instr.cache_offset
)
elif instr.kind == "inst" and variable_used(
instr.inst, "oparg1"
):
assert variable_used(
instr.inst, "oparg2"
), "Half super-instr?"
self.write_super_expansions(instr.name)
case parsing.Macro():
mac = self.macro_instrs[thing.name]
self.write_macro_expansions(
mac.name, mac.parts, mac.cache_offset
)
case parsing.Pseudo():
pass
case _:
assert_never(thing)
with self.metadata_item(
"const char * const _PyOpcode_uop_name[OPCODE_UOP_NAME_SIZE]", "=", ";"
):
self.write_uop_items(lambda name, counter: f'[{name}] = "{name}",')
with self.metadata_item(
f"const char *const _PyOpcode_OpName[{1 + max(self.opmap.values())}]",
"=",
";",
):
for name in self.opmap:
self.out.emit(f'[{name}] = "{name}",')
with self.metadata_item(
f"const uint8_t _PyOpcode_Caches[256]",
"=",
";",
):
family_member_names: set[str] = set()
for family in self.families.values():
family_member_names.update(family.members)
for instr in self.instrs.values():
if (
instr.name not in family_member_names
and instr.cache_offset > 0
and instr.kind == "inst"
and not instr.name.startswith("INSTRUMENTED_")
):
self.out.emit(f"[{instr.name}] = {instr.cache_offset},")
for mac in self.macro_instrs.values():
if mac.name not in family_member_names and mac.cache_offset > 0:
self.out.emit(f"[{mac.name}] = {mac.cache_offset},")
# Irregular case:
self.out.emit('[JUMP_BACKWARD] = 1,')
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:
deoptcodes["BINARY_OP_INPLACE_ADD_UNICODE"] = "BINARY_OP"
with self.metadata_item(f"const uint8_t _PyOpcode_Deopt[256]", "=", ";"):
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")
with open(pymetadata_filename, "w") as f:
# Create formatter
self.out = Formatter(f, 0, comment="#")
self.write_provenance_header()
# emit specializations
specialized_ops = set()
self.out.emit("")
self.out.emit("_specializations = {")
for name, family in self.families.items():
with self.out.indent():
self.out.emit(f'"{family.name}": [')
with self.out.indent():
for m in family.members:
self.out.emit(f'"{m}",')
specialized_ops.update(family.members)
self.out.emit(f"],")
self.out.emit("}")
# Handle special case
self.out.emit("")
self.out.emit("# An irregular case:")
self.out.emit(
'_specializations["BINARY_OP"].append('
'"BINARY_OP_INPLACE_ADD_UNICODE")'
)
specialized_ops.add("BINARY_OP_INPLACE_ADD_UNICODE")
ops = sorted((id, name) for (name, id) in self.opmap.items())
# emit specialized opmap
self.out.emit("")
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},")
for name in ["MIN_INSTRUMENTED_OPCODE", "HAVE_ARGUMENT"]:
self.out.emit(f"{name} = {self.markers[name]}")
def write_pseudo_instrs(self) -> None:
"""Write the IS_PSEUDO_INSTR macro"""
self.out.emit("\n\n#define IS_PSEUDO_INSTR(OP) ( \\")
for op in self.pseudos:
self.out.emit(f" ((OP) == {op}) || \\")
self.out.emit(f" 0)")
def write_uop_items(self, make_text: typing.Callable[[str, int], str]) -> None:
"""Write '#define XXX NNN' for each uop"""
counter = 300 # TODO: Avoid collision with pseudo instructions
seen = set()
def add(name: str) -> None:
if name in seen:
return
nonlocal counter
self.out.emit(make_text(name, counter))
counter += 1
seen.add(name)
# These two are first by convention
add("EXIT_TRACE")
add("SAVE_IP")
for instr in self.instrs.values():
if instr.kind == "op":
add(instr.name)
def write_macro_expansions(
self, name: str, parts: MacroParts, cache_offset: int
) -> None:
"""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), ...]
for part in parts:
if isinstance(part, Component):
# All component instructions must be viable uops
if not part.instr.is_viable_uop():
# 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.
self.note(
f"Part {part.instr.name} of {name} is not a viable uop",
part.instr.inst,
)
return
if not part.active_caches:
if part.instr.name == "SAVE_IP":
size, offset = OPARG_SIZES["OPARG_SAVE_IP"], cache_offset
else:
size, offset = OPARG_SIZES["OPARG_FULL"], 0
else:
# If this assert triggers, is_viable_uops() lied
assert len(part.active_caches) == 1, (name, part.instr.name)
cache = part.active_caches[0]
size, offset = cache.effect.size, cache.offset
expansions.append((part.instr.name, size, offset))
assert len(expansions) > 0, f"Macro {name} has empty expansion?!"
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"
expansions: list[tuple[str, int, int]] = [
(name1, OPARG_SIZES["OPARG_TOP"], 0),
(name2, OPARG_SIZES["OPARG_BOTTOM"], 0),
]
self.write_expansions(name, expansions)
def write_expansions(
self, name: str, expansions: list[tuple[str, int, int]]
) -> None:
pieces = [
f"{{ {name}, {size}, {offset} }}" for name, size, offset in expansions
]
self.out.emit(
f"[{name}] = "
f"{{ .nuops = {len(pieces)}, .uops = {{ {', '.join(pieces)} }} }},"
)
def emit_metadata_entry(self, name: str, fmt: str, flags: InstructionFlags) -> None:
flag_names = flags.names(value=True)
if not flag_names:
flag_names.append("0")
self.out.emit(
f"[{name}] = {{ true, {INSTR_FMT_PREFIX}{fmt},"
f" {' | '.join(flag_names)} }},"
)
def write_metadata_for_inst(self, instr: Instruction) -> None:
"""Write metadata for a single instruction."""
self.emit_metadata_entry(instr.name, instr.instr_fmt, instr.instr_flags)
def write_metadata_for_macro(self, mac: MacroInstruction) -> None:
"""Write metadata for a macro-instruction."""
self.emit_metadata_entry(mac.name, mac.instr_fmt, mac.instr_flags)
def write_metadata_for_pseudo(self, ps: PseudoInstruction) -> None:
"""Write metadata for a macro-instruction."""
self.emit_metadata_entry(ps.name, ps.instr_fmt, ps.instr_flags)
def write_instructions(
self, output_filename: str, emit_line_directives: bool
) -> None:
"""Write instructions to output file."""
with open(output_filename, "w") as f:
# Create formatter
self.out = Formatter(f, 8, emit_line_directives)
self.write_provenance_header()
# Write and count instructions of all kinds
n_instrs = 0
n_macros = 0
for thing in self.everything:
match thing:
case OverriddenInstructionPlaceHolder():
self.write_overridden_instr_place_holder(thing)
case parsing.InstDef():
if thing.kind != "op":
n_instrs += 1
self.write_instr(self.instrs[thing.name])
case parsing.Macro():
n_macros += 1
mac = self.macro_instrs[thing.name]
stacking.write_macro_instr(
mac, self.out, self.families.get(mac.name)
)
# self.write_macro(self.macro_instrs[thing.name])
case parsing.Pseudo():
pass
case _:
assert_never(thing)
print(
f"Wrote {n_instrs} instructions and {n_macros} macros "
f"to {output_filename}",
file=sys.stderr,
)
def write_executor_instructions(
self, executor_filename: str, emit_line_directives: bool
) -> None:
"""Generate cases for the Tier 2 interpreter."""
n_instrs = 0
n_uops = 0
with open(executor_filename, "w") as f:
self.out = Formatter(f, 8, emit_line_directives)
self.write_provenance_header()
for thing in self.everything:
match thing:
case OverriddenInstructionPlaceHolder():
# TODO: Is this helpful?
self.write_overridden_instr_place_holder(thing)
case parsing.InstDef():
instr = self.instrs[thing.name]
if instr.is_viable_uop():
if instr.kind == "op":
n_uops += 1
else:
n_instrs += 1
self.out.emit("")
with self.out.block(f"case {thing.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;")
# elif instr.kind != "op":
# print(f"NOTE: {thing.name} is not a viable uop")
case parsing.Macro():
pass
case parsing.Pseudo():
pass
case _:
assert_never(thing)
print(
f"Wrote {n_instrs} instructions and {n_uops} ops to {executor_filename}",
file=sys.stderr,
)
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()
for thing in self.everything:
match thing:
case OverriddenInstructionPlaceHolder():
pass
case parsing.InstDef():
instr = AbstractInstruction(self.instrs[thing.name].inst)
if (
instr.is_viable_uop()
and instr.name not in SPECIALLY_HANDLED_ABSTRACT_INSTR
):
self.out.emit("")
with self.out.block(f"case {thing.name}:"):
instr.write(self.out, tier=TIER_TWO)
self.out.emit("break;")
case parsing.Macro():
pass
case parsing.Pseudo():
pass
case _:
assert_never(thing)
print(
f"Wrote some stuff to {abstract_interpreter_filename}",
file=sys.stderr,
)
def write_overridden_instr_place_holder(
self, place_holder: OverriddenInstructionPlaceHolder
) -> None:
self.out.emit("")
self.out.emit(
f"{self.out.comment} TARGET({place_holder.name}) overridden by later definition"
)
def write_instr(self, instr: Instruction) -> None:
name = instr.name
self.out.emit("")
if instr.inst.override:
self.out.emit("{self.out.comment} Override")
with self.out.block(f"TARGET({name})"):
if instr.predicted:
self.out.emit(f"PREDICTED({name});")
self.out.static_assert_family_size(
instr.name, instr.family, instr.cache_offset
)
stacking.write_single_instr(instr, self.out, tier=TIER_ONE)
if not instr.always_exits:
if instr.cache_offset:
self.out.emit(f"next_instr += {instr.cache_offset};")
if instr.check_eval_breaker:
self.out.emit("CHECK_EVAL_BREAKER();")
self.out.emit(f"DISPATCH();")
def main() -> None:
"""Parse command line, parse input, analyze, write output."""
args = arg_parser.parse_args() # Prints message and sys.exit(2) on error
if len(args.input) == 0:
args.input.append(DEFAULT_INPUT)
# Raises OSError if input unreadable
a = Generator(args.input)
a.parse() # Raises SyntaxError on failure
a.analyze() # Prints messages and sets a.errors on failure
if a.errors:
sys.exit(f"Found {a.errors} errors")
# These raise OSError if output can't be written
a.write_instructions(args.output, args.emit_line_directives)
a.assign_opcode_ids()
a.write_opcode_ids(args.opcode_ids_h, args.opcode_targets_h)
a.write_metadata(args.metadata, args.pymetadata)
a.write_executor_instructions(args.executor_cases, args.emit_line_directives)
a.write_abstract_interpreter_instructions(
args.abstract_interpreter_cases, args.emit_line_directives
)
if __name__ == "__main__":
main()
|