-
Notifications
You must be signed in to change notification settings - Fork 2
/
parse.py
655 lines (588 loc) · 29.1 KB
/
parse.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
#!/usr/bin/python3
import argparse
import bisect
from collections import Counter, defaultdict
from configparser import ConfigParser
from dataclasses import dataclass
import itertools
import re
from typing import Any, NamedTuple
BITS_SUBGRP = 7
BITS_SUBTRIE = 9
GRP_START = 1 << (BITS_SUBTRIE - BITS_SUBGRP)
MAX_SUBGRPS = 1 << BITS_SUBGRP
MAX_SUBTRIES = (1 << BITS_SUBTRIE) - 1
MAX_GRPS = (1 << (16 - BITS_SUBGRP)) - GRP_START
print_decode_trie = False
class ParseException(Exception):
def __init__(self, message, str, col):
super().__init__(message)
self.str = str
self.col = col
class MaskedVal(NamedTuple):
val: int
mask: int
def __rshift__(self, other: int):
if type(other) != int:
return NotImplemented
return MaskedVal(self.val >> other, self.mask >> other)
def __and__(self, other):
if type(other) == int:
return MaskedVal(self.val & other, self.mask & other)
return NotImplemented
def bitmaskrange(fixed: MaskedVal, max: int):
idx = fixed.mask
while idx < max:
yield (idx & ~fixed.mask) | fixed.val
idx = (idx + 1) | fixed.mask
class DescEntry(NamedTuple):
idx: int
size: int
fixed: MaskedVal
name: str|None = None
flags: str = ""
Desc = list[DescEntry]
def parse_desc(desc: str) -> Desc:
res = []
idx = 0
remainder = desc
while remainder:
m = re.match(r'([01]+)|{(\d+):([a-zA-Z0-9|]+)(?:=([01x]+))?}', remainder)
if not m or (m[4] and len(m[4]) != int(m[2])):
raise ParseException(f"invalid syntax", desc, len(desc)-len(remainder))
bits = m[1] or m[4] or "x" * int(m[2])
mask = int(bits.translate(str.maketrans("01x", "110")), 2)
val = int(bits.translate(str.maketrans("01x", "010")), 2)
fixed = MaskedVal(val, mask)
name, _, flags = m[3].partition("|") if m[3] else (None, None, "")
res.append(DescEntry(32-idx-len(bits), len(bits), fixed, name, flags))
remainder = remainder[len(m[0]):]
idx += res[-1].size
if idx != 32:
raise ParseException(f"descriptor contains {idx} bits, not 32", desc, 0)
return res
RE_FMT = re.compile(r'([^{]+)' +
r'|{(\w+(?:<\d+(?::\d+)?>)?(?:,\w+(?:<\d+(?::\d+)?>)?)*)((?::\^?[^,=:}]+(?:,[^,=:}]+)*=[^:}]*|:![^,=:}]+(?:,[^,=:}]+)*)+)}')
RE_FMT_VAR = re.compile(r'^(\w+)(?:<(\d+)(?::(\d+))?>)?$')
class Fmt:
def __init__(self, fmt: str, desc: Desc):
self.desc = desc
self.comps = []
varmap = {e.name: i for i, e in enumerate(desc)}
def parse_var(var: str) -> tuple[int, int]: # shift, sz
var_comps = RE_FMT_VAR.match(var)
if var_comps is None:
raise Exception(f"couldn't match variable {var}")
de = desc[varmap[var_comps[1]]]
if var_comps[2]:
high = int(var_comps[2])
low = int(var_comps[3]) if var_comps[3] else high
return de.idx + low, high - low + 1
return de.idx, de.size
while fmt:
m = RE_FMT.match(fmt)
if not m:
raise ParseException(f"invalid format {fmt}", fmt, 0)
if m[1]:
self.comps.append(m[1])
else:
vars = [parse_var(var) for var in m[2].split(",")]
masks = [(1 << sz) - 1 << shift for shift, sz in vars]
assert sum(masks).bit_count() == sum(m.bit_count() for m in masks)
d = {}
for vals, _, text in (x.partition("=") for x in m[3][1:].split(":")):
negate, overwrite = vals[:1] == "!", vals[:1] == "^"
vals = vals[negate or overwrite:]
numvals: list[list[int]] = []
for val, (shift, sz) in zip(vals.split(","), vars):
if val == "*":
numvals.append(list(range(0, 1 << shift+sz, 1 << shift)))
else:
assert 0 <= int(val, 0) < (1 << sz)
numvals.append([int(val, 0) << shift])
assert len(numvals) == len(vars)
for k in itertools.product(*numvals):
if negate:
del d[sum(k)]
else:
assert (sum(k) in d) == overwrite
d[sum(k)] = text
self.comps.append((sum(masks), d))
fmt = fmt[len(m[0]):]
def eval(self, descvals: list[MaskedVal]) -> str|None:
# For given values, return the unique string the format evaluates to.
res = ""
for comp in self.comps:
if type(comp) == str:
res += comp
elif type(comp) == tuple:
# Performance optimization, both cases behave are identical.
if len(descvals) == 1 and descvals[0].mask & comp[0] == comp[0]:
results = [comp[1].get(descvals[0].val & comp[0])]
else:
results = list({comp[1].get(k) for v in descvals
for k in bitmaskrange(v & comp[0], comp[0] + 1)})
if len(results) != 1:
raise Exception(f"unable to specialize format {self}, candidates are {results}")
if results[0] is None:
return None
res += results[0]
return res
def expansions(self) -> dict[str, list[MaskedVal]]:
# Return a dict mapping expansions to all possible values for this.
specmsk = 0
for comp in self.comps:
if type(comp) == tuple:
specmsk |= comp[0]
baseval = sum(de.fixed.val << de.idx for de in self.desc)
basemsk = sum(de.fixed.mask << de.idx for de in self.desc) | specmsk
derivs: dict[str, list[MaskedVal]] = defaultdict(list)
for prod in bitmaskrange(MaskedVal(baseval, specmsk ^ 0xffffffff), 1 << 32):
if name := self.eval([mask := MaskedVal(prod, basemsk)]):
derivs[name].append(mask)
return dict(derivs)
class TrieDesc(NamedTuple):
fixed: MaskedVal
subgrp: int
name: str = ""
cond: str|None = None
class Trie(NamedTuple):
fmsk: int
fval: int
smsk_start: int
smsk_len: int
table: tuple[int, ...]
subtries: tuple[Any, ...] # Any is actually Trie
comment: str
@property
def table_size(self):
return len(self.table) + sum(st.table_size for st in self.subtries)
def codegen(self, fn_prefix: str, tables_as_switch: bool) -> str:
stnames = [f"{fn_prefix}_{i+1:x}" for i in range(len(self.subtries))]
res = "\n".join(st.codegen(n, tables_as_switch) for st, n in zip(self.subtries, stnames))
res += f"\n// {self.comment}"
res += f"\nstatic unsigned {fn_prefix}(uint32_t inst) {{\n"
if self.fmsk:
res += f" if ((inst&{self.fmsk:#x}) != {self.fval:#x}) return 0;\n"
if not self.subtries and len(set(self.table)) == 1:
res += f" return {hex(self.table[0])};\n"
res += "}\n"
return res
table_idx = f"(inst >> {self.smsk_start}) & {(1<<self.smsk_len)-1:#x}"
if tables_as_switch:
res += f" unsigned idx;\n"
res += f" switch ({table_idx}) {{\n"
for i, t in enumerate(self.table):
res += f" case {i:#x}: idx = {t:#x}; break;\n"
res += " }\n"
else:
res += " static const uint16_t table[] = {\n"
res += " " + ",".join(f"{t:#x}" for t in self.table) + " };\n"
res += f" unsigned idx = table[{table_idx}];\n"
if self.subtries:
res += " switch (idx) {\n"
for i in range(len(self.subtries)):
res += f" case {i+1:#x}: return {stnames[i]}(inst);"
res += f" // {self.subtries[i].comment}\n"
res += " }\n"
res += " return idx;\n"
res += "}\n"
return res
def make_table(descs: list[TrieDesc], for_keys, prev_msk=0):
cnts: list[list[int]] = [[0, 0] for _ in range(32)]
kmsks = 0xffffffff
for k in for_keys:
km = descs[k].fixed
kmsks &= km.mask
for i, l in enumerate(cnts):
if (km.mask&~prev_msk) & (1<<i):
l[km.val & (1<<i) != 0] += 1
# There is the case that bits end up being cleared in km.mask/kmsks without
# actually distinguishing different encodings (e.g., due to constraints).
# TODO: eliminate these.
fmsk = kmsks & sum(1<<i for i, cnt in enumerate(cnts) if 0 in cnt and sum(cnt) > 0)
fval = fmsk & sum((cnt[0] == 0) << i for i, cnt in enumerate(cnts))
debug_keys = ",".join(descs[k].name for k in for_keys) if prev_msk else "*"
if print_decode_trie:
indent = " " * bin(prev_msk).count("1") + ">"
print(indent, debug_keys, f"filter={fmsk:032b}:{fval:032b}", {i: cnt for i, cnt in enumerate(cnts) if sum(cnt)})
if all(0 in cnt for cnt in cnts):
# only one entry left
if len(for_keys) != 1:
raise Exception(f"redundant opcodes: {for_keys} ({debug_keys})")
return Trie(fmsk, fval, 0, 0, (descs[next(iter(for_keys))].subgrp,), (), debug_keys)
# Compute mask over first streak of differing bits
smsk_end = next(32-i for i, cnt in enumerate(cnts[::-1]) if 0 not in cnt)
smsk_len = next((i for i, cnt in enumerate(cnts[smsk_end-1::-1]) if 0 in cnt), smsk_end)
smsk_len = min(12, smsk_len)
smsk_start = smsk_end - smsk_len
smsk = ((1 << smsk_len) - 1) << smsk_start
if print_decode_trie:
print(indent, smsk_start, smsk_len, f"smsk={smsk:032b}")
table_keys: list[list[int]] = [[] for _ in range(1 << smsk_len)]
for k in for_keys:
for idx in bitmaskrange((descs[k].fixed & smsk) >> smsk_start, 1 << smsk_len):
table_keys[idx].append(k)
table = [0 for _ in range(1 << smsk_len)]
subtries = []
subtrie_map = {}
for i, keys in enumerate(table_keys):
if not keys:
continue
if len(keys) == 1 and descs[keys[0]].fixed.mask&~(prev_msk|fmsk|smsk) == 0:
table[i] = descs[keys[0]].subgrp
else:
keys_set = frozenset(keys)
assert len(keys_set) == len(keys)
if keys_set not in subtrie_map:
subtrie = make_table(descs, keys, prev_msk|fmsk|smsk)
if not subtrie.fmsk and len(set(subtrie.table)) == 1:
subtrie_map[keys_set] = subtrie.table[0]
else:
subtries.append(subtrie)
subtrie_map[keys_set] = len(subtries)
table[i] = subtrie_map[keys_set]
if len(subtries) > MAX_SUBTRIES:
print(len(subtries), ">", MAX_SUBTRIES, subtries)
raise Exception(f"too many subtries: {len(subtries)} > {MAX_SUBTRIES}")
return Trie(fmsk, fval, smsk_start, smsk_len, tuple(table), tuple(subtries), debug_keys)
# From Fadec
def superstring(strs):
# This faces the "shortest superstring" problem, which is NP-hard.
# Preprocessing: remove any strings which are already completely covered
realstrs = []
for s in sorted(strs, key=len, reverse=True):
for s2 in realstrs:
if s in s2:
break
else:
realstrs.append(s)
# Greedy heuristic generally yields acceptable results, though it depends on
# the order of the menmonics. More compact results are possible, but the
# expectable gains of an optimal result (probably with O(n!)) are small.
# First sort strings and later do a binary search for each possible prefix.
realstrs.sort()
merged = ""
while realstrs:
for i in range(min(16, len(merged)), 0, -1):
idx = bisect.bisect_left(realstrs, merged[-i:])
if idx < len(realstrs) and realstrs[idx][:i] == merged[-i:]:
merged += realstrs.pop(idx)[i:]
break
else:
merged += realstrs.pop()
return merged
def check_feature(features, featurestr):
return all(features[x] for x in featurestr.split("&&"))
class DecoderGenerator:
mnems: dict[str, int] = {}
grpnums: dict[str, int] = {}
trie_descs: list[TrieDesc] = []
decstr: dict[int, str] = {}
decoders: dict[int, str] = {}
docs: dict[str, dict[str, tuple[Desc, Fmt]]] = {}
@staticmethod
def generate_opdecoder(decodestr: str, desc: Desc) -> str:
res = "{"
ops = set(x[1:] for x in re.findall(r'@[a-zA-Z0-9_]+', decodestr))
for de in desc:
if de.name in ops:
ty = "int64_t" if de.flags == "s" else "uint64_t"
if de.fixed.mask == (1 << de.size) - 1:
val = f"{de.fixed.val}"
else:
val = f"inst>>{de.idx}&{(1<<de.size)-1:#x}"
if de.flags == "s":
val = f"sext({val}, {de.size})"
res += f"{ty} {de.name} = {val}; "
# Split decodestr, respecting parenthesized operands.
operands, parendepth = [], 0
for comp in decodestr.replace("@", "").strip().split():
if parendepth:
operands[-1] += " " + comp
else:
operands.append(comp)
parendepth += comp.count("(") - comp.count(")")
for i, op in enumerate(operands):
res += f"ddi->ops[{i}] = {op.rstrip(',')}; "
return res + "break;}"
def add_group(self, grp: str, grpdesc: Desc, props, features: dict[str, bool]):
if grp in self.grpnums:
raise Exception(f"redundant group {grp}")
grpnum = self.grpnums[grp] = len(self.grpnums) + GRP_START
if "decode" not in props:
print(f"group {grp} without decode")
return
decodefmt = Fmt(props["decode"], grpdesc)
featfmt = Fmt(props.get("feat", ""), grpdesc)
derivs = Fmt(props["mnem"], grpdesc).expansions()
assert len(derivs) <= MAX_SUBGRPS
# Most mnems share the same decode operands, map opdecoder to mnems
opdecoders = defaultdict(list)
for i, (name, masks) in enumerate(derivs.items()):
if name in self.mnems:
raise Exception(f"redundant mnem {name}")
mnem = self.mnems[name] = (grpnum << BITS_SUBGRP) + i
# Ensure that every mnemonic is tied to one feature (i.e., when
# adding a feature, an existing mnemonic won't get additional
# semantics). Mnemonic values are added nonetheless so that code
# can use the mnemonics even when they are disabled.
if not check_feature(features, featfmt.eval(masks)):
continue
for j, mask in enumerate(masks):
self.trie_descs.append(TrieDesc(mask, mnem, f"{grp}-{name}#{j}"))
decmnem, _, opstr = decodefmt.eval(masks).partition(" ")
self.decstr[mnem] = decmnem.replace("_", " ")
# TODO: maybe specialize decoder on fixed values?
opdecoders[self.generate_opdecoder(opstr, grpdesc)].append(name)
if len(opdecoders) == 1:
self.decoders[grp] = next(iter(opdecoders))
else:
switch = "switch (mnem) { default: __builtin_unreachable();\n"
for opdecoder, mnems in opdecoders.items():
switch += " " + " ".join(f"case DA64I_{mnem}:" for mnem in mnems)
switch += "\n " + opdecoder + "\n"
self.decoders[grp] = switch + "} break;"
def generate(self, tables_as_switch) -> tuple[str, dict[str, str]]:
trie = make_table(self.trie_descs, list(range(len(self.trie_descs))))
mnems_strs = [f"DA64I_{k}={v:#x},\n" for k, v in self.mnems.items()]
grpnum_strs = [f"DA64G_{k}={v:#x},\n" for k, v in self.grpnums.items()]
decodestr = superstring(self.decstr.values())
decstrtab = [f'[{m}] = {len(s) << 12 | decodestr.index(s):#x},\n'
for m, s in self.decstr.items()]
decoder = [f'case DA64G_{m}: {s}\n' for m, s in self.decoders.items()]
public = f"""
enum Da64InstKind {{
DA64I_UNKNOWN=0,
{"".join(mnems_strs)}
}};
enum Da64InstGroup {{
DA64G_UNKNOWN=0,
{"".join(grpnum_strs)}
}};
#define DA64_GROUP(mnem) ((mnem) >> {BITS_SUBGRP})
"""
return public, {
"DA64_CLASSIFIER": trie.codegen("da64_classify_impl", tables_as_switch),
"DA64_DECSTR": '"' + decodestr + '"',
"DA64_DECSTRTAB": "".join(decstrtab),
"DA64_DECODER": "".join(decoder),
}
@dataclass
class EncodeFunc:
paramtys: tuple[str, ...]
expansions: tuple[str, ...]
singleExpansion: bool = False
allowOverride: bool = False
cond: str|None = None
ENC_FUNCS = {
"const": EncodeFunc((), ("{val}",)),
"bool": EncodeFunc(("bool",), ("!!{0}",)),
"regzr": EncodeFunc((), ("31",)),
"reggp": EncodeFunc(("DA_GRegZR",), ("DA_REGVAL({0})",)),
"reggpsp": EncodeFunc(("DA_GRegSP",), ("DA_REGVAL({0})",)),
"reggpnozr": EncodeFunc(("DA_GReg",), ("DA_REGVAL({0})",)),
"reggpls64": EncodeFunc(("DA_GReg",), ("DA_REGVAL({0})",), cond="DA_REGVAL({0})<24&&!(DA_REGVAL({0})&1)"),
"regfp": EncodeFunc(("DA_VReg",), ("DA_REGVAL({0})",)),
"regfplim": EncodeFunc(("DA_VReg",), ("DA_REGVAL({0})<<16",), singleExpansion=True, cond="DA_REGVAL({0})<{max}"),
"prfop": EncodeFunc(("enum Da64PrfOp prfop",), ("{0}&31",)),
"cond": EncodeFunc(("Da64Cond",), ("{0}",)),
"invcond": EncodeFunc(("Da64Cond",), ("({0}^1)",)),
"imm": EncodeFunc(("unsigned",), ("{0}",), cond="{0}<{max}"),
"immadr": EncodeFunc(("uintptr_t codeaddr", "uintptr_t target"), ("({1}-{0})&3", "(({1}-{0})>>2)&0x7ffff"), cond="da_sext(({1}-{0}),21)==(intptr_t)({1}-{0})"),
"immadrp": EncodeFunc(
("uintptr_t codeaddr", "uintptr_t target"),
("((({1}&~0xfffull)-({0}&~0xfffull))>>12)&3", "((({1}&~0xfffull)-({0}&~0xfffull))>>14)&0x7ffff"),
cond="da_sext(({1}&~0xfffull)-({0}&~0xfffull),33)==(intptr_t)(({1}&~0xfffull)-({0}&~0xfffull))",
),
"immlsl": EncodeFunc(("unsigned lsl",), ("(-{0})&({max}-1)", "{max}-1-{0}"), cond="{0}<{max}"),
"immbfx": EncodeFunc(("unsigned lsb", "unsigned width"), ("{0}", "{0}+{1}-1"), cond="{0}<{max}&&{1}-1<{max}-{0}"),
"immbfi": EncodeFunc(("unsigned lsb", "unsigned width"), ("(-{0})&({max}-1)", "{1}-1"), cond="{0}<{max}&&{1}-1<{max}-{0}"),
# only base-imm instead of 2*base - imm, as relevant bit is already set.
"immshiftr": EncodeFunc(("unsigned imm",), ("({base}-{0})<<16",), singleExpansion=True, cond="{0}&&{0}<={base}"),
"immshiftl": EncodeFunc(("unsigned imm",), ("{0}<<16",), singleExpansion=True, cond="{0}&&{0}<={base}"),
"immldraut": EncodeFunc(("int64_t",), ("({0}>>12)&1", "({0}&0xff8)>>3"), cond="da_sext({0},13)=={0}&&({0}&7)==0"),
"immvidx": EncodeFunc(("unsigned",), ("{0}<<({size}+{shift})",), cond="{0}<(1u<<(4-{size}))", allowOverride=True),
"immrotmul": EncodeFunc(("unsigned",), ("{0}/90",), cond="{0}<360&&{0}%90==0"),
"immrotadd": EncodeFunc(("unsigned",), ("{0}/90-1",), cond="{0}==90||{0}==270"),
"velemidx": EncodeFunc(("unsigned elemidx",), ("({0}<<{size})>>3", "({0}<<{size})>>2", "({0}<<{size})>>1"), cond="{0}<(1u<<(4-{size}))", allowOverride=True),
"velemidxlim": EncodeFunc(("unsigned elemidx",), ("({0}<<{size})>>3", "({0}<<{size})>>2", "({0}<<{size})>>1"), cond="{0}<(1u<<(4-{size}))&&{0}<{lim}", allowOverride=True),
"velemidx0": EncodeFunc(("unsigned elemidx",), ("({0}<<{size})>>3", "({0}<<{size})>>2", "({0}<<{size})>>1", "(({0}<<{size})&1)<<3"), cond="{0}<(1u<<(4-{size}))", allowOverride=True),
"memsimdidx": EncodeFunc(("unsigned elemidx",), ("({0}<<{size})>>3", "({0}<<{size})>>2", "{0}<<{size}"), cond="{0}<(1u<<(4-{size}))", allowOverride=True),
"uimm": EncodeFunc(("uint64_t",), ("{0}>>{lsr}",), cond="({0}&(((1<<{bits})-1)<<{lsr}))=={0}"),
"simm": EncodeFunc(("int64_t",), ("{0}/(1<<{asr})",), cond="da_sext({0},{bits}+{asr})=={0}&&({0}&((1<<{asr})-1))==0"),
"tbz": EncodeFunc(("unsigned bit",), ("{0}>>5", "{0}&0x1f"), cond="{0}<64"),
"reladdr": EncodeFunc(("ptrdiff_t",), ("{0}",), cond="da_sext({0},{bits})=={0}"),
"fcvtfixscale": EncodeFunc(("unsigned fbits",), ("64-{0}",), cond="{0}<{max}"),
"immadd": EncodeFunc(("int{size}_t imm",), ("da_immadd({0})",), singleExpansion=True, cond="da_immadd({0}) != 0xffffffff"),
"immlogical": EncodeFunc(("uint{size}_t imm",), ("da_immlogical({0}, {size}>>6)",), singleExpansion=True, cond="da_immlogical({0}, {size}>>6) != 0xffffffff"),
"immfmov32": EncodeFunc(("float imm",), ("da_immfmov32({0})",), cond="da_immfmov32({0}) != 0xffffffff"),
"immfmov64": EncodeFunc(("double imm",), ("da_immfmov64({0})",), cond="da_immfmov64({0}) != 0xffffffff"),
"immsimd8movi": EncodeFunc(("uint64_t imm64",), ("da_immsimdmovi({0})",), singleExpansion=True, cond="da_immsimdmovi({0}) != 0xffffffff"),
"immsimd8lsl": EncodeFunc(("uint8_t imm", "unsigned lsl"), ("{0}>>5", "{0}&0x1f", "{1}>>2|1"), cond="{1}<(8*{maxshift}) && ({1}&7) == 0"),
"immsimd8fmov": EncodeFunc(("float imm",), ("da_immfmov32({0})>>5", "da_immfmov32({0})&0x1f"), cond="da_immfmov32({0}) != 0xffffffff"),
}
class EncoderGenerator:
decls: list[str] = []
defns: list[str] = []
encode_in_header = False
def __init__(self, encode_in_header=False):
self.encode_in_header = encode_in_header
def add_group(self, grp: str, grpdesc: Desc, props, features: dict[str, bool]):
featfmt = Fmt(props.get("feat", ""), grpdesc)
for key in props:
if key.startswith("encmnem"):
cat = key[7:]
mnemfmt = Fmt(props[f"encmnem{cat}"], grpdesc)
opsfmt = Fmt(props[f"encops{cat}"], grpdesc)
condfmt = Fmt(props.get(f"enccond{cat}", ""), grpdesc)
featfmt = Fmt(props.get(f"encfeat{cat}", props.get("feat", "")), grpdesc)
derivs = mnemfmt.expansions()
for name, masks in derivs.items():
assert len(masks) == 1, "encmnem with multiple descs per name"
ops = opsfmt.eval(masks)
cond = condfmt.eval(masks)
if not check_feature(features, featfmt.eval(masks)):
continue
self.add_mnem(grp, name, grpdesc, masks[0], ops, cond)
def add_mnem(self, grp: str, name: str, grpdesc: Desc, descmask: MaskedVal, ops: str, cond: str):
paramtys: list[str] = []
parnames: list[str] = []
conds: list[str] = []
assignments: dict[str, str] = {}
grandExpansions: list[str] = []
varaliases: dict[str, str] = {}
for varstr, _, funcspec in (comp.partition("=") for comp in ops.split()):
vars = varstr.split(",")
if funcspec[0] == "@":
aliases = funcspec[1:].split(",")
assert len(vars) == len(aliases), "var alias mismatch"
for var, alias in zip(vars, aliases):
varaliases[var] = alias
continue
funcname, _, funcops = funcspec.partition("/")
funcops = [op.partition("=") for op in funcops.split(",")]
funcops = {key: val for key, _, val in funcops if key}
func = ENC_FUNCS[funcname]
parnamebase = "".join(vars)
funcparnames = []
for ty, _, parname in (ty.partition(" ") for ty in func.paramtys):
funcparnames.append(parname or parnamebase)
paramtys.append(ty.format(**funcops))
parnames += funcparnames
if func.cond:
conds.append(func.cond.format(*funcparnames, **funcops))
if func.singleExpansion:
assert len(func.expansions) == 1
grandExpansions.append(func.expansions[0].format(*funcparnames, **funcops))
for var in vars:
assignments[var] = "0"
continue
assert len(vars) == len(func.expansions)
for var, expansion in zip(vars, func.expansions):
if var in assignments and not func.allowOverride:
raise Exception(f"duplicate assignment to {var}")
assignments[var] = expansion.format(*funcparnames, **funcops)
for alias, var in varaliases.items():
assignments[alias] = assignments[var]
if cond:
getval = lambda m: f"({assignments[m.group(1)]})"
conds.append(re.sub(r'@([a-zA-Z0-9_]+)', getval, cond))
# use XOR s.t. immadd can toggle between add and sub.
buildexprs = ["(" + "^".join([f"{descmask.val:#x}"] + grandExpansions) + ")"]
for de in grpdesc:
bitmask = (1 << de.size) - 1
if de.name in assignments:
buildexprs.append(f"((({assignments[de.name]})&{bitmask:#x})<<{de.idx})")
elif descmask.mask & (bitmask << de.idx) != (bitmask << de.idx):
print("Encode desc missing for", grp, name, de.name, assignments)
return
expr = "|".join(buildexprs)
if conds:
condexpr = "&&".join(f"({cond})" for cond in conds)
expr = f"!({condexpr}) ? 0 : {expr}"
substs = {"DA_GRegSP": "DA_GPSP", "DA_GRegZR": "DA_GPZR"}
if any(ty in substs for ty in paramtys):
implname = f"de64_impl_{name}"
macroparams = ",".join(parnames)
macroargs = ",".join(f"{substs.get(ty, '')}({parname})" for ty, parname in zip(paramtys, parnames))
self.decls.append(f"#define de64_{name}({macroparams}) {implname}({macroargs})")
else:
implname = f"de64_{name}"
paramstr = ", ".join(f"{ty} {parname}" for ty, parname in zip(paramtys, parnames))
if not paramstr:
paramstr = "void"
signature = f"uint32_t {implname}({paramstr})"
if self.encode_in_header:
self.decls.append(f"static inline {signature} {{ return {expr}; }}")
else:
self.decls.append(signature + ";")
self.defns.append(f"{signature} {{ return {expr}; }}")
def generate(self) -> tuple[str, dict[str, str]]:
return "\n".join(self.decls) + "\n", {
"DA64_ENCODER": "\n".join(self.defns),
}
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--features")
parser.add_argument("--coverage", action="store_true",
help="Generate decode tables as switch for measuring coverage")
parser.add_argument("--encode-in-header", action="store_true",
help="Move most encoding logic to header")
parser.add_argument("--feature-desc", help="Feature description file")
parser.add_argument("out_public", type=argparse.FileType("w"))
parser.add_argument("out_private", type=argparse.FileType("w"))
parser.add_argument("descfiles", nargs="+")
args = parser.parse_args()
features = {"": True}
if args.feature_desc:
featuredesc = ConfigParser(interpolation=None, delimiters=(":",), strict=True)
featuredesc.optionxform = str # make keys case-sensitive
with open(args.feature_desc, "r") as f:
featuredesc.read_file(f)
attrset = defaultdict(set)
for feature in featuredesc.sections():
features[feature] = False
attrset[feature.lower()].add(feature)
if "avail" in featuredesc[feature]:
attrset[featuredesc[feature]["avail"]].add(feature)
if "incomplete" not in featuredesc[feature]:
attrset["all"].add(feature)
for i in range(1, 10):
attrset[f"armv8.{i}a"] |= attrset[f"armv8.{i-1}a"]
for i in range(0, 5):
attrset[f"armv9.{i}a"] |= attrset[f"armv8.{i+5}a"]
for i in range(1, 5):
attrset[f"armv9.{i}a"] |= attrset[f"armv9.{i-1}a"]
attrset["really-all"] = set(features.keys())
for attr in (args.features or "all").strip().split(","):
if attrs := attrset.get(attr):
for attr in attrs:
features[attr] = True
else:
raise Exception(f"unknown feature {attr}")
config = ConfigParser(interpolation=None, delimiters=(":",), strict=True)
config.optionxform = str # make keys case-sensitive
for descfile in args.descfiles:
with open(descfile, "r") as f:
config.read_file(f)
decoder = DecoderGenerator()
encoder = EncoderGenerator(args.encode_in_header)
for grp in config.sections():
try:
props = config[grp]
grpdesc = parse_desc(props["desc"])
decoder.add_group(grp, grpdesc, props, features)
encoder.add_group(grp, grpdesc, props, features)
except ParseException as e:
print("error parsing", grp, e)
raise
public_features = "".join(f"#define DA64_HAVE_{f} {n:d}\n" for f, n in features.items())
public_decode, private_decode = decoder.generate(tables_as_switch=args.coverage)
public_encode, private_encode = encoder.generate()
private_dict = dict(private_decode, **private_encode)
private_str = "".join(f"#{'el' if i else ''}if defined({key})\n{val}\n"
for i, (key, val) in enumerate(private_dict.items()))
private_str += '#else\n#error "unknown table"\n#endif\n'
args.out_public.write(public_features + public_decode + public_encode)
args.out_private.write(private_str)