-
-
Notifications
You must be signed in to change notification settings - Fork 29.2k
/
generate_cases.py
1723 lines (1506 loc) · 65.8 KB
/
generate_cases.py
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
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
"""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 dataclasses
import os
import posixpath
import re
import sys
import typing
import lexer as lx
import parser
from parser 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_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")
)
BEGIN_MARKER = "// BEGIN BYTECODES //"
END_MARKER = "// END BYTECODES //"
RE_PREDICTED = (
r"^\s*(?:GO_TO_INSTRUCTION\(|DEOPT_IF\(.*?,\s*)(\w+)\);\s*(?://.*)?$"
)
UNUSED = "unused"
BITS_PER_CODE_UNIT = 16
# 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,
}
RESERVED_WORDS = {
"co_consts" : "Use FRAME_CO_CONSTS.",
"co_names": "Use FRAME_CO_NAMES.",
}
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(
"-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,
)
def effect_size(effect: StackEffect) -> tuple[int, str]:
"""Return the 'size' impact of a stack effect.
Returns a tuple (numeric, symbolic) where:
- numeric is an int giving the statically analyzable size of the effect
- symbolic is a string representing a variable effect (e.g. 'oparg*2')
At most one of these will be non-zero / non-empty.
"""
if effect.size:
assert not effect.cond, "Array effects cannot have a condition"
return 0, effect.size
elif effect.cond:
if effect.cond in ("0", "1"):
return int(effect.cond), ""
return 0, f"{maybe_parenthesize(effect.cond)} ? 1 : 0"
else:
return 1, ""
def maybe_parenthesize(sym: str) -> str:
"""Add parentheses around a string if it contains an operator.
An exception is made for '*' which is common and harmless
in the context where the symbolic size is used.
"""
if re.match(r"^[\s\w*]+$", sym):
return sym
else:
return f"({sym})"
def list_effect_size(effects: list[StackEffect]) -> tuple[int, str]:
numeric = 0
symbolic: list[str] = []
for effect in effects:
diff, sym = effect_size(effect)
numeric += diff
if sym:
symbolic.append(maybe_parenthesize(sym))
return numeric, " + ".join(symbolic)
def string_effect_size(arg: tuple[int, str]) -> str:
numeric, symbolic = arg
if numeric and symbolic:
return f"{numeric} + {symbolic}"
elif symbolic:
return symbolic
else:
return str(numeric)
class Formatter:
"""Wraps an output stream with the ability to indent etc."""
stream: typing.TextIO
prefix: str
emit_line_directives: bool = False
lineno: int # Next line number, 1-based
filename: str # Slightly improved stream.filename
nominal_lineno: int
nominal_filename: str
def __init__(
self, stream: typing.TextIO, indent: int,
emit_line_directives: bool = False, comment: str = "//",
) -> None:
self.stream = stream
self.prefix = " " * indent
self.emit_line_directives = emit_line_directives
self.comment = comment
self.lineno = 1
self.filename = prettify_filename(self.stream.name)
self.nominal_lineno = 1
self.nominal_filename = self.filename
def write_raw(self, s: str) -> None:
self.stream.write(s)
newlines = s.count("\n")
self.lineno += newlines
self.nominal_lineno += newlines
def emit(self, arg: str) -> None:
if arg:
self.write_raw(f"{self.prefix}{arg}\n")
else:
self.write_raw("\n")
def set_lineno(self, lineno: int, filename: str) -> None:
if self.emit_line_directives:
if lineno != self.nominal_lineno or filename != self.nominal_filename:
self.emit(f'#line {lineno} "{filename}"')
self.nominal_lineno = lineno
self.nominal_filename = filename
def reset_lineno(self) -> None:
if self.lineno != self.nominal_lineno or self.filename != self.nominal_filename:
self.set_lineno(self.lineno + 1, self.filename)
@contextlib.contextmanager
def indent(self):
self.prefix += " "
yield
self.prefix = self.prefix[:-4]
@contextlib.contextmanager
def block(self, head: str, tail: str = ""):
if head:
self.emit(head + " {")
else:
self.emit("{")
with self.indent():
yield
self.emit("}" + tail)
def stack_adjust(
self,
input_effects: list[StackEffect],
output_effects: list[StackEffect],
):
shrink, isym = list_effect_size(input_effects)
grow, osym = list_effect_size(output_effects)
diff = grow - shrink
if isym and isym != osym:
self.emit(f"STACK_SHRINK({isym});")
if diff < 0:
self.emit(f"STACK_SHRINK({-diff});")
if diff > 0:
self.emit(f"STACK_GROW({diff});")
if osym and osym != isym:
self.emit(f"STACK_GROW({osym});")
def declare(self, dst: StackEffect, src: StackEffect | None):
if dst.name == UNUSED or dst.cond == "0":
return
typ = f"{dst.type}" if dst.type else "PyObject *"
if src:
cast = self.cast(dst, src)
init = f" = {cast}{src.name}"
elif dst.cond:
init = " = NULL"
else:
init = ""
sepa = "" if typ.endswith("*") else " "
self.emit(f"{typ}{sepa}{dst.name}{init};")
def assign(self, dst: StackEffect, src: StackEffect):
if src.name == UNUSED:
return
if src.size:
# Don't write sized arrays -- it's up to the user code.
return
cast = self.cast(dst, src)
if re.match(r"^REG\(oparg(\d+)\)$", dst.name):
self.emit(f"Py_XSETREF({dst.name}, {cast}{src.name});")
else:
stmt = f"{dst.name} = {cast}{src.name};"
if src.cond and src.cond != "1":
if src.cond == "0":
# It will not be executed
return
stmt = f"if ({src.cond}) {{ {stmt} }}"
self.emit(stmt)
def cast(self, dst: StackEffect, src: StackEffect) -> str:
return f"({dst.type or 'PyObject *'})" if src.type != dst.type else ""
@dataclasses.dataclass
class InstructionFlags:
"""Construct and manipulate instruction flags"""
HAS_ARG_FLAG: bool
HAS_CONST_FLAG: bool
HAS_NAME_FLAG: bool
HAS_JUMP_FLAG: bool
HAS_FREE_FLAG: bool
HAS_LOCAL_FLAG: bool
def __post_init__(self):
self.bitmask = {
name : (1 << i) for i, name in enumerate(self.names())
}
@staticmethod
def fromInstruction(instr: "AnyInstruction"):
has_free = (variable_used(instr, "PyCell_New") or
variable_used(instr, "PyCell_GET") or
variable_used(instr, "PyCell_SET"))
return InstructionFlags(
HAS_ARG_FLAG=variable_used(instr, "oparg"),
HAS_CONST_FLAG=variable_used(instr, "FRAME_CO_CONSTS"),
HAS_NAME_FLAG=variable_used(instr, "FRAME_CO_NAMES"),
HAS_JUMP_FLAG=variable_used(instr, "JUMPBY"),
HAS_FREE_FLAG=has_free,
HAS_LOCAL_FLAG=(variable_used(instr, "GETLOCAL") or
variable_used(instr, "SETLOCAL")) and
not has_free,
)
@staticmethod
def newEmpty():
return InstructionFlags(False, False, False, False, False, False)
def add(self, other: "InstructionFlags") -> None:
for name, value in dataclasses.asdict(other).items():
if value:
setattr(self, name, value)
def names(self, value=None):
if value is None:
return dataclasses.asdict(self).keys()
return [n for n, v in dataclasses.asdict(self).items() if v == value]
def bitmap(self) -> int:
flags = 0
for name in self.names():
if getattr(self, name):
flags |= self.bitmask[name]
return flags
@classmethod
def emit_macros(cls, out: Formatter):
flags = cls.newEmpty()
for name, value in flags.bitmask.items():
out.emit(f"#define {name} ({value})");
for name, value in flags.bitmask.items():
out.emit(
f"#define OPCODE_{name[:-len('_FLAG')]}(OP) "
f"(_PyOpcode_opcode_metadata[OP].flags & ({name}))")
@dataclasses.dataclass
class ActiveCacheEffect:
"""Wraps a CacheEffect that is actually used, in context."""
effect: parser.CacheEffect
offset: int
FORBIDDEN_NAMES_IN_UOPS = (
"resume_with_error",
"kwnames",
"next_instr",
"oparg1", # Proxy for super-instructions like LOAD_FAST_LOAD_FAST
"JUMPBY",
"DISPATCH",
"INSTRUMENTED_JUMP",
"throwflag",
"exception_unwind",
"import_from",
"import_name",
"_PyObject_CallNoArgs", # Proxy for BEFORE_WITH
)
# Interpreter tiers
TIER_ONE = 1 # Specializing adaptive interpreter (PEP 659)
TIER_TWO = 2 # Experimental tracing interpreter
Tiers: typing.TypeAlias = typing.Literal[1, 2]
@dataclasses.dataclass
class Instruction:
"""An instruction with additional data and code."""
# Parts of the underlying instruction definition
inst: parser.InstDef
kind: typing.Literal["inst", "op"]
name: str
block: parser.Block
block_text: list[str] # Block.text, less curlies, less PREDICT() calls
block_line: int # First line of block in original code
# Computed by constructor
always_exits: bool
cache_offset: int
cache_effects: list[parser.CacheEffect]
input_effects: list[StackEffect]
output_effects: list[StackEffect]
unmoved_names: frozenset[str]
instr_fmt: str
instr_flags: InstructionFlags
active_caches: list[ActiveCacheEffect]
# Set later
family: parser.Family | None = None
predicted: bool = False
def __init__(self, inst: parser.InstDef):
self.inst = inst
self.kind = inst.kind
self.name = inst.name
self.block = inst.block
self.block_text, self.check_eval_breaker, self.block_line = \
extract_block_text(self.block)
self.always_exits = always_exits(self.block_text)
self.cache_effects = [
effect for effect in inst.inputs if isinstance(effect, parser.CacheEffect)
]
self.cache_offset = sum(c.size for c in self.cache_effects)
self.input_effects = [
effect for effect in inst.inputs if isinstance(effect, StackEffect)
]
self.output_effects = inst.outputs # For consistency/completeness
unmoved_names: set[str] = set()
for ieffect, oeffect in zip(self.input_effects, self.output_effects):
if ieffect.name == oeffect.name:
unmoved_names.add(ieffect.name)
else:
break
self.unmoved_names = frozenset(unmoved_names)
self.instr_flags = InstructionFlags.fromInstruction(inst)
self.active_caches = []
offset = 0
for effect in self.cache_effects:
if effect.name != UNUSED:
self.active_caches.append(ActiveCacheEffect(effect, offset))
offset += effect.size
if self.instr_flags.HAS_ARG_FLAG:
fmt = "IB"
else:
fmt = "IX"
if offset:
fmt += "C" + "0"*(offset-1)
self.instr_fmt = fmt
def is_viable_uop(self) -> bool:
"""Whether this instruction is viable as a uop."""
dprint: typing.Callable[..., None] = lambda *args, **kwargs: None
# if self.name.startswith("CALL"):
# dprint = print
if self.name == "EXIT_TRACE":
return True # This has 'return frame' but it's okay
if self.always_exits:
dprint(f"Skipping {self.name} because it always exits")
return False
if len(self.active_caches) > 1:
# print(f"Skipping {self.name} because it has >1 cache entries")
return False
res = True
for forbidden in FORBIDDEN_NAMES_IN_UOPS:
# NOTE: To disallow unspecialized uops, use
# if variable_used(self.inst, forbidden):
if variable_used_unspecialized(self.inst, forbidden):
dprint(f"Skipping {self.name} because it uses {forbidden}")
res = False
return res
def write(self, out: Formatter, tier: Tiers = TIER_ONE) -> None:
"""Write one instruction, sans prologue and epilogue."""
# Write a static assertion that a family's cache size is correct
if family := self.family:
if self.name == family.name:
if cache_size := family.size:
out.emit(
f"static_assert({cache_size} == "
f'{self.cache_offset}, "incorrect cache size");'
)
# Write input stack effect variable declarations and initializations
ieffects = list(reversed(self.input_effects))
for i, ieffect in enumerate(ieffects):
isize = string_effect_size(
list_effect_size([ieff for ieff in ieffects[: i + 1]])
)
if ieffect.size:
src = StackEffect(f"(stack_pointer - {maybe_parenthesize(isize)})", "PyObject **")
elif ieffect.cond:
src = StackEffect(f"({ieffect.cond}) ? stack_pointer[-{maybe_parenthesize(isize)}] : NULL", "")
else:
src = StackEffect(f"stack_pointer[-{maybe_parenthesize(isize)}]", "")
out.declare(ieffect, src)
# Write output stack effect variable declarations
isize = string_effect_size(list_effect_size(self.input_effects))
input_names = {ieffect.name for ieffect in self.input_effects}
for i, oeffect in enumerate(self.output_effects):
if oeffect.name not in input_names:
if oeffect.size:
osize = string_effect_size(
list_effect_size([oeff for oeff in self.output_effects[:i]])
)
offset = "stack_pointer"
if isize != osize:
if isize != "0":
offset += f" - ({isize})"
if osize != "0":
offset += f" + {osize}"
src = StackEffect(offset, "PyObject **")
out.declare(oeffect, src)
else:
out.declare(oeffect, None)
# out.emit(f"next_instr += OPSIZE({self.inst.name}) - 1;")
self.write_body(out, 0, self.active_caches, tier=tier)
# Skip the rest if the block always exits
if self.always_exits:
return
# Write net stack growth/shrinkage
out.stack_adjust(
[ieff for ieff in self.input_effects],
[oeff for oeff in self.output_effects],
)
# Write output stack effect assignments
oeffects = list(reversed(self.output_effects))
for i, oeffect in enumerate(oeffects):
if oeffect.name in self.unmoved_names:
continue
osize = string_effect_size(
list_effect_size([oeff for oeff in oeffects[: i + 1]])
)
if oeffect.size:
dst = StackEffect(f"stack_pointer - {maybe_parenthesize(osize)}", "PyObject **")
else:
dst = StackEffect(f"stack_pointer[-{maybe_parenthesize(osize)}]", "")
out.assign(dst, oeffect)
# Write cache effect
if tier == TIER_ONE and self.cache_offset:
out.emit(f"next_instr += {self.cache_offset};")
def write_body(
self,
out: Formatter,
dedent: int,
active_caches: list[ActiveCacheEffect],
tier: Tiers = TIER_ONE,
) -> None:
"""Write the instruction body."""
# Write cache effect variable declarations and initializations
for active in active_caches:
ceffect = active.effect
bits = ceffect.size * BITS_PER_CODE_UNIT
if bits == 64:
# NOTE: We assume that 64-bit data in the cache
# is always an object pointer.
# If this becomes false, we need a way to specify
# syntactically what type the cache data is.
typ = "PyObject *"
func = "read_obj"
else:
typ = f"uint{bits}_t "
func = f"read_u{bits}"
if tier == TIER_ONE:
out.emit(
f"{typ}{ceffect.name} = {func}(&next_instr[{active.offset}].cache);"
)
else:
out.emit(f"{typ}{ceffect.name} = ({typ.strip()})operand;")
# Write the body, substituting a goto for ERROR_IF() and other stuff
assert dedent <= 0
extra = " " * -dedent
names_to_skip = self.unmoved_names | frozenset({UNUSED, "null"})
offset = 0
context = self.block.context
assert context is not None and context.owner is not None
filename = context.owner.filename
for line in self.block_text:
out.set_lineno(self.block_line + offset, filename)
offset += 1
if m := re.match(r"(\s*)ERROR_IF\((.+), (\w+)\);\s*(?://.*)?$", line):
space, cond, label = m.groups()
space = extra + space
# ERROR_IF() must pop the inputs from the stack.
# The code block is responsible for DECREF()ing them.
# NOTE: If the label doesn't exist, just add it to ceval.c.
# Don't pop common input/output effects at the bottom!
# These aren't DECREF'ed so they can stay.
ieffs = list(self.input_effects)
oeffs = list(self.output_effects)
while ieffs and oeffs and ieffs[0] == oeffs[0]:
ieffs.pop(0)
oeffs.pop(0)
ninputs, symbolic = list_effect_size(ieffs)
if ninputs:
label = f"pop_{ninputs}_{label}"
if symbolic:
out.write_raw(
f"{space}if ({cond}) {{ STACK_SHRINK({symbolic}); goto {label}; }}\n"
)
else:
out.write_raw(f"{space}if ({cond}) goto {label};\n")
elif m := re.match(r"(\s*)DECREF_INPUTS\(\);\s*(?://.*)?$", line):
out.reset_lineno()
space = extra + m.group(1)
for ieff in self.input_effects:
if ieff.name in names_to_skip:
continue
if ieff.size:
out.write_raw(
f"{space}for (int _i = {ieff.size}; --_i >= 0;) {{\n"
)
out.write_raw(f"{space} Py_DECREF({ieff.name}[_i]);\n")
out.write_raw(f"{space}}}\n")
else:
decref = "XDECREF" if ieff.cond else "DECREF"
out.write_raw(f"{space}Py_{decref}({ieff.name});\n")
else:
out.write_raw(extra + line)
out.reset_lineno()
InstructionOrCacheEffect = Instruction | parser.CacheEffect
StackEffectMapping = list[tuple[StackEffect, StackEffect]]
@dataclasses.dataclass
class Component:
instr: Instruction
input_mapping: StackEffectMapping
output_mapping: StackEffectMapping
active_caches: list[ActiveCacheEffect]
def write_body(self, out: Formatter) -> None:
with out.block(""):
input_names = {ieffect.name for _, ieffect in self.input_mapping}
for var, ieffect in self.input_mapping:
out.declare(ieffect, var)
for _, oeffect in self.output_mapping:
if oeffect.name not in input_names:
out.declare(oeffect, None)
self.instr.write_body(out, -4, self.active_caches)
for var, oeffect in self.output_mapping:
out.assign(var, oeffect)
MacroParts = list[Component | parser.CacheEffect]
@dataclasses.dataclass
class MacroInstruction:
"""A macro instruction."""
name: str
stack: list[StackEffect]
initial_sp: int
final_sp: int
instr_fmt: str
instr_flags: InstructionFlags
macro: parser.Macro
parts: MacroParts
cache_offset: int
predicted: bool = False
@dataclasses.dataclass
class PseudoInstruction:
"""A pseudo instruction."""
name: str
targets: list[Instruction]
instr_fmt: str
instr_flags: InstructionFlags
@dataclasses.dataclass
class OverriddenInstructionPlaceHolder:
name: str
AnyInstruction = Instruction | MacroInstruction | PseudoInstruction
INSTR_FMT_PREFIX = "INSTR_FMT_"
class Analyzer:
"""Parse input, analyze it, and write to output."""
input_filenames: list[str]
output_filename: str
metadata_filename: str
pymetadata_filename: str
executor_filename: str
errors: int = 0
emit_line_directives: bool = False
def __init__(
self,
input_filenames: list[str],
output_filename: str,
metadata_filename: str,
pymetadata_filename: str,
executor_filename: str,
):
"""Read the input file."""
self.input_filenames = input_filenames
self.output_filename = output_filename
self.metadata_filename = metadata_filename
self.pymetadata_filename = pymetadata_filename
self.executor_filename = executor_filename
def error(self, msg: str, node: parser.Node) -> None:
lineno = 0
filename = "<unknown file>"
if context := node.context:
filename = context.owner.filename
# Use line number of first non-comment in the node
for token in context.owner.tokens[context.begin : context.end]:
lineno = token.line
if token.kind != "COMMENT":
break
print(f"{filename}:{lineno}: {msg}", file=sys.stderr)
self.errors += 1
everything: list[
parser.InstDef | parser.Macro | parser.Pseudo | OverriddenInstructionPlaceHolder
]
instrs: dict[str, Instruction] # Includes ops
macros: dict[str, parser.Macro]
macro_instrs: dict[str, MacroInstruction]
families: dict[str, parser.Family]
pseudos: dict[str, parser.Pseudo]
pseudo_instrs: dict[str, PseudoInstruction]
def parse(self) -> None:
"""Parse the source text.
We only want the parser to see the stuff between the
begin and end markers.
"""
self.everything = []
self.instrs = {}
self.macros = {}
self.families = {}
self.pseudos = {}
instrs_idx: dict[str, int] = dict()
for filename in self.input_filenames:
self.parse_file(filename, instrs_idx)
files = " + ".join(self.input_filenames)
print(
f"Read {len(self.instrs)} instructions/ops, "
f"{len(self.macros)} macros, {len(self.pseudos)} pseudos, "
f"and {len(self.families)} families from {files}",
file=sys.stderr,
)
def parse_file(self, filename: str, instrs_idx: dict[str, int]) -> None:
with open(filename) as file:
src = file.read()
psr = parser.Parser(src, filename=prettify_filename(filename))
# Skip until begin marker
while tkn := psr.next(raw=True):
if tkn.text == BEGIN_MARKER:
break
else:
raise psr.make_syntax_error(
f"Couldn't find {BEGIN_MARKER!r} in {psr.filename}"
)
start = psr.getpos()
# Find end marker, then delete everything after it
while tkn := psr.next(raw=True):
if tkn.text == END_MARKER:
break
del psr.tokens[psr.getpos() - 1 :]
# Parse from start
psr.setpos(start)
thing: parser.InstDef | parser.Macro | parser.Pseudo | parser.Family | None
thing_first_token = psr.peek()
while thing := psr.definition():
if ws := [w for w in RESERVED_WORDS if variable_used(thing, w)]:
self.error(f"'{ws[0]}' is a reserved word. {RESERVED_WORDS[ws[0]]}", thing)
match thing:
case parser.InstDef(name=name):
if name in self.instrs:
if not thing.override:
raise psr.make_syntax_error(
f"Duplicate definition of '{name}' @ {thing.context} "
f"previous definition @ {self.instrs[name].inst.context}",
thing_first_token,
)
self.everything[instrs_idx[name]] = OverriddenInstructionPlaceHolder(name=name)
if name not in self.instrs and thing.override:
raise psr.make_syntax_error(
f"Definition of '{name}' @ {thing.context} is supposed to be "
"an override but no previous definition exists.",
thing_first_token,
)
self.instrs[name] = Instruction(thing)
instrs_idx[name] = len(self.everything)
self.everything.append(thing)
case parser.Macro(name):
self.macros[name] = thing
self.everything.append(thing)
case parser.Family(name):
self.families[name] = thing
case parser.Pseudo(name):
self.pseudos[name] = thing
self.everything.append(thing)
case _:
typing.assert_never(thing)
if not psr.eof():
raise psr.make_syntax_error(f"Extra stuff at the end of {filename}")
def analyze(self) -> None:
"""Analyze the inputs.
Raises SystemExit if there is an error.
"""
self.analyze_macros_and_pseudos()
self.find_predictions()
self.map_families()
self.check_families()
def find_predictions(self) -> None:
"""Find the instructions that need PREDICTED() labels."""
for instr in self.instrs.values():
targets: set[str] = set()
for line in instr.block_text:
if m := re.match(RE_PREDICTED, line):
targets.add(m.group(1))
for target in targets:
if target_instr := self.instrs.get(target):
target_instr.predicted = True
elif target_macro := self.macro_instrs.get(target):
target_macro.predicted = True
else:
self.error(
f"Unknown instruction {target!r} predicted in {instr.name!r}",
instr.inst, # TODO: Use better location
)
def map_families(self) -> None:
"""Link instruction names back to their family, if they have one."""
for family in self.families.values():
for member in [family.name] + family.members:
if member_instr := self.instrs.get(member):
if member_instr.family not in (family, None):
self.error(
f"Instruction {member} is a member of multiple families "
f"({member_instr.family.name}, {family.name}).",
family,
)
else:
member_instr.family = family
elif not self.macro_instrs.get(member):
self.error(
f"Unknown instruction {member!r} referenced in family {family.name!r}",
family,
)
def check_families(self) -> None:
"""Check each family:
- Must have at least 2 members (including head)
- Head and all members must be known instructions
- Head and all members must have the same cache, input and output effects
"""
for family in self.families.values():
if family.name not in self.macro_instrs and family.name not in self.instrs:
self.error(
f"Family {family.name!r} has unknown instruction {family.name!r}",
family,
)
members = [
member
for member in family.members
if member in self.instrs or member in self.macro_instrs
]
if members != family.members:
unknown = set(family.members) - set(members)
self.error(
f"Family {family.name!r} has unknown members: {unknown}", family
)
expected_effects = self.effect_counts(family.name)
for member in members:
member_effects = self.effect_counts(member)
if member_effects != expected_effects:
self.error(
f"Family {family.name!r} has inconsistent "
f"(cache, input, output) effects:\n"
f" {family.name} = {expected_effects}; "
f"{member} = {member_effects}",
family,
)
def effect_counts(self, name: str) -> tuple[int, int, int]:
if instr := self.instrs.get(name):
cache = instr.cache_offset
input = len(instr.input_effects)
output = len(instr.output_effects)
elif mac := self.macro_instrs.get(name):
cache = mac.cache_offset
input, output = 0, 0
for part in mac.parts:
if isinstance(part, Component):
# A component may pop what the previous component pushed,
# so we offset the input/output counts by that.
delta_i = len(part.instr.input_effects)
delta_o = len(part.instr.output_effects)
offset = min(delta_i, output)
input += delta_i - offset
output += delta_o - offset
else:
assert False, f"Unknown instruction {name!r}"
return cache, input, output
def analyze_macros_and_pseudos(self) -> None:
"""Analyze each macro and pseudo instruction."""
self.macro_instrs = {}
self.pseudo_instrs = {}
for name, macro in self.macros.items():
self.macro_instrs[name] = self.analyze_macro(macro)
for name, pseudo in self.pseudos.items():
self.pseudo_instrs[name] = self.analyze_pseudo(pseudo)
def analyze_macro(self, macro: parser.Macro) -> MacroInstruction:
components = self.check_macro_components(macro)
stack, initial_sp = self.stack_analysis(components)
sp = initial_sp
parts: MacroParts = []
flags = InstructionFlags.newEmpty()
offset = 0
for component in components:
match component:
case parser.CacheEffect() as ceffect:
parts.append(ceffect)
offset += ceffect.size
case Instruction() as instr:
part, sp, offset = self.analyze_instruction(instr, stack, sp, offset)
parts.append(part)
flags.add(instr.instr_flags)
case _:
typing.assert_never(component)
final_sp = sp
format = "IB"
if offset:
format += "C" + "0"*(offset-1)
return MacroInstruction(
macro.name, stack, initial_sp, final_sp, format, flags, macro, parts, offset
)
def analyze_pseudo(self, pseudo: parser.Pseudo) -> PseudoInstruction:
targets = [self.instrs[target] for target in pseudo.targets]
assert targets
# Make sure the targets have the same fmt
fmts = list(set([t.instr_fmt for t in targets]))
assert(len(fmts) == 1)
assert(len(list(set([t.instr_flags.bitmap() for t in targets]))) == 1)
return PseudoInstruction(pseudo.name, targets, fmts[0], targets[0].instr_flags)
def analyze_instruction(
self, instr: Instruction, stack: list[StackEffect], sp: int, offset: int
) -> tuple[Component, int, int]:
input_mapping: StackEffectMapping = []
for ieffect in reversed(instr.input_effects):
sp -= 1
input_mapping.append((stack[sp], ieffect))
output_mapping: StackEffectMapping = []
for oeffect in instr.output_effects:
output_mapping.append((stack[sp], oeffect))
sp += 1
active_effects: list[ActiveCacheEffect] = []
for ceffect in instr.cache_effects:
if ceffect.name != UNUSED:
active_effects.append(ActiveCacheEffect(ceffect, offset))
offset += ceffect.size
return Component(instr, input_mapping, output_mapping, active_effects), sp, offset
def check_macro_components(
self, macro: parser.Macro
) -> list[InstructionOrCacheEffect]:
components: list[InstructionOrCacheEffect] = []
for uop in macro.uops:
match uop:
case parser.OpName(name):
if name not in self.instrs:
self.error(f"Unknown instruction {name!r}", macro)
components.append(self.instrs[name])
case parser.CacheEffect():
components.append(uop)
case _:
typing.assert_never(uop)
return components
def stack_analysis(
self, components: typing.Iterable[InstructionOrCacheEffect]
) -> tuple[list[StackEffect], int]:
"""Analyze a macro.
Ignore cache effects.
Return the list of variables (as StackEffects) and the initial stack pointer.
"""
lowest = current = highest = 0
conditions: dict[int, str] = {} # Indexed by 'current'.
last_instr: Instruction | None = None
for thing in components:
if isinstance(thing, Instruction):
last_instr = thing