Richard Henderson | 568ae7e | 2017-12-07 12:44:09 -0800 | [diff] [blame] | 1 | #!/usr/bin/env python |
| 2 | # Copyright (c) 2018 Linaro Limited |
| 3 | # |
| 4 | # This library is free software; you can redistribute it and/or |
| 5 | # modify it under the terms of the GNU Lesser General Public |
| 6 | # License as published by the Free Software Foundation; either |
| 7 | # version 2 of the License, or (at your option) any later version. |
| 8 | # |
| 9 | # This library is distributed in the hope that it will be useful, |
| 10 | # but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 12 | # Lesser General Public License for more details. |
| 13 | # |
| 14 | # You should have received a copy of the GNU Lesser General Public |
| 15 | # License along with this library; if not, see <http://www.gnu.org/licenses/>. |
| 16 | # |
| 17 | |
| 18 | # |
| 19 | # Generate a decoding tree from a specification file. |
| 20 | # |
| 21 | # The tree is built from instruction "patterns". A pattern may represent |
| 22 | # a single architectural instruction or a group of same, depending on what |
| 23 | # is convenient for further processing. |
| 24 | # |
| 25 | # Each pattern has "fixedbits" & "fixedmask", the combination of which |
| 26 | # describes the condition under which the pattern is matched: |
| 27 | # |
| 28 | # (insn & fixedmask) == fixedbits |
| 29 | # |
| 30 | # Each pattern may have "fields", which are extracted from the insn and |
| 31 | # passed along to the translator. Examples of such are registers, |
| 32 | # immediates, and sub-opcodes. |
| 33 | # |
| 34 | # In support of patterns, one may declare fields, argument sets, and |
| 35 | # formats, each of which may be re-used to simplify further definitions. |
| 36 | # |
| 37 | # *** Field syntax: |
| 38 | # |
| 39 | # field_def := '%' identifier ( unnamed_field )+ ( !function=identifier )? |
| 40 | # unnamed_field := number ':' ( 's' ) number |
| 41 | # |
| 42 | # For unnamed_field, the first number is the least-significant bit position of |
| 43 | # the field and the second number is the length of the field. If the 's' is |
| 44 | # present, the field is considered signed. If multiple unnamed_fields are |
| 45 | # present, they are concatenated. In this way one can define disjoint fields. |
| 46 | # |
| 47 | # If !function is specified, the concatenated result is passed through the |
| 48 | # named function, taking and returning an integral value. |
| 49 | # |
| 50 | # FIXME: the fields of the structure into which this result will be stored |
| 51 | # is restricted to "int". Which means that we cannot expand 64-bit items. |
| 52 | # |
| 53 | # Field examples: |
| 54 | # |
| 55 | # %disp 0:s16 -- sextract(i, 0, 16) |
| 56 | # %imm9 16:6 10:3 -- extract(i, 16, 6) << 3 | extract(i, 10, 3) |
| 57 | # %disp12 0:s1 1:1 2:10 -- sextract(i, 0, 1) << 11 |
| 58 | # | extract(i, 1, 1) << 10 |
| 59 | # | extract(i, 2, 10) |
| 60 | # %shimm8 5:s8 13:1 !function=expand_shimm8 |
| 61 | # -- expand_shimm8(sextract(i, 5, 8) << 1 |
| 62 | # | extract(i, 13, 1)) |
| 63 | # |
| 64 | # *** Argument set syntax: |
| 65 | # |
| 66 | # args_def := '&' identifier ( args_elt )+ |
| 67 | # args_elt := identifier |
| 68 | # |
| 69 | # Each args_elt defines an argument within the argument set. |
| 70 | # Each argument set will be rendered as a C structure "arg_$name" |
| 71 | # with each of the fields being one of the member arguments. |
| 72 | # |
| 73 | # Argument set examples: |
| 74 | # |
| 75 | # ®3 ra rb rc |
| 76 | # &loadstore reg base offset |
| 77 | # |
| 78 | # *** Format syntax: |
| 79 | # |
| 80 | # fmt_def := '@' identifier ( fmt_elt )+ |
| 81 | # fmt_elt := fixedbit_elt | field_elt | field_ref | args_ref |
| 82 | # fixedbit_elt := [01.-]+ |
| 83 | # field_elt := identifier ':' 's'? number |
| 84 | # field_ref := '%' identifier | identifier '=' '%' identifier |
| 85 | # args_ref := '&' identifier |
| 86 | # |
| 87 | # Defining a format is a handy way to avoid replicating groups of fields |
| 88 | # across many instruction patterns. |
| 89 | # |
| 90 | # A fixedbit_elt describes a contiguous sequence of bits that must |
| 91 | # be 1, 0, [.-] for don't care. The difference between '.' and '-' |
| 92 | # is that '.' means that the bit will be covered with a field or a |
| 93 | # final [01] from the pattern, and '-' means that the bit is really |
| 94 | # ignored by the cpu and will not be specified. |
| 95 | # |
| 96 | # A field_elt describes a simple field only given a width; the position of |
| 97 | # the field is implied by its position with respect to other fixedbit_elt |
| 98 | # and field_elt. |
| 99 | # |
| 100 | # If any fixedbit_elt or field_elt appear then all bits must be defined. |
| 101 | # Padding with a fixedbit_elt of all '.' is an easy way to accomplish that. |
| 102 | # |
| 103 | # A field_ref incorporates a field by reference. This is the only way to |
| 104 | # add a complex field to a format. A field may be renamed in the process |
| 105 | # via assignment to another identifier. This is intended to allow the |
| 106 | # same argument set be used with disjoint named fields. |
| 107 | # |
| 108 | # A single args_ref may specify an argument set to use for the format. |
| 109 | # The set of fields in the format must be a subset of the arguments in |
| 110 | # the argument set. If an argument set is not specified, one will be |
| 111 | # inferred from the set of fields. |
| 112 | # |
| 113 | # It is recommended, but not required, that all field_ref and args_ref |
| 114 | # appear at the end of the line, not interleaving with fixedbit_elf or |
| 115 | # field_elt. |
| 116 | # |
| 117 | # Format examples: |
| 118 | # |
| 119 | # @opr ...... ra:5 rb:5 ... 0 ....... rc:5 |
| 120 | # @opi ...... ra:5 lit:8 1 ....... rc:5 |
| 121 | # |
| 122 | # *** Pattern syntax: |
| 123 | # |
| 124 | # pat_def := identifier ( pat_elt )+ |
| 125 | # pat_elt := fixedbit_elt | field_elt | field_ref |
| 126 | # | args_ref | fmt_ref | const_elt |
| 127 | # fmt_ref := '@' identifier |
| 128 | # const_elt := identifier '=' number |
| 129 | # |
| 130 | # The fixedbit_elt and field_elt specifiers are unchanged from formats. |
| 131 | # A pattern that does not specify a named format will have one inferred |
| 132 | # from a referenced argument set (if present) and the set of fields. |
| 133 | # |
| 134 | # A const_elt allows a argument to be set to a constant value. This may |
| 135 | # come in handy when fields overlap between patterns and one has to |
| 136 | # include the values in the fixedbit_elt instead. |
| 137 | # |
| 138 | # The decoder will call a translator function for each pattern matched. |
| 139 | # |
| 140 | # Pattern examples: |
| 141 | # |
| 142 | # addl_r 010000 ..... ..... .... 0000000 ..... @opr |
| 143 | # addl_i 010000 ..... ..... .... 0000000 ..... @opi |
| 144 | # |
| 145 | # which will, in part, invoke |
| 146 | # |
| 147 | # trans_addl_r(ctx, &arg_opr, insn) |
| 148 | # and |
| 149 | # trans_addl_i(ctx, &arg_opi, insn) |
| 150 | # |
| 151 | |
Richard Henderson | 568ae7e | 2017-12-07 12:44:09 -0800 | [diff] [blame] | 152 | import os |
| 153 | import re |
| 154 | import sys |
| 155 | import getopt |
Richard Henderson | 568ae7e | 2017-12-07 12:44:09 -0800 | [diff] [blame] | 156 | |
| 157 | insnwidth = 32 |
| 158 | insnmask = 0xffffffff |
| 159 | fields = {} |
| 160 | arguments = {} |
| 161 | formats = {} |
| 162 | patterns = [] |
| 163 | |
| 164 | translate_prefix = 'trans' |
| 165 | translate_scope = 'static ' |
| 166 | input_file = '' |
| 167 | output_file = None |
| 168 | output_fd = None |
| 169 | insntype = 'uint32_t' |
| 170 | |
| 171 | re_ident = '[a-zA-Z][a-zA-Z0-9_]*' |
| 172 | |
| 173 | |
| 174 | def error(lineno, *args): |
| 175 | """Print an error message from file:line and args and exit.""" |
| 176 | global output_file |
| 177 | global output_fd |
| 178 | |
| 179 | if lineno: |
| 180 | r = '{0}:{1}: error:'.format(input_file, lineno) |
| 181 | elif input_file: |
| 182 | r = '{0}: error:'.format(input_file) |
| 183 | else: |
| 184 | r = 'error:' |
| 185 | for a in args: |
| 186 | r += ' ' + str(a) |
| 187 | r += '\n' |
| 188 | sys.stderr.write(r) |
| 189 | if output_file and output_fd: |
| 190 | output_fd.close() |
| 191 | os.remove(output_file) |
| 192 | exit(1) |
| 193 | |
| 194 | |
| 195 | def output(*args): |
| 196 | global output_fd |
| 197 | for a in args: |
| 198 | output_fd.write(a) |
| 199 | |
| 200 | |
| 201 | if sys.version_info >= (3, 0): |
| 202 | re_fullmatch = re.fullmatch |
| 203 | else: |
| 204 | def re_fullmatch(pat, str): |
| 205 | return re.match('^' + pat + '$', str) |
| 206 | |
| 207 | |
| 208 | def output_autogen(): |
| 209 | output('/* This file is autogenerated by scripts/decodetree.py. */\n\n') |
| 210 | |
| 211 | |
| 212 | def str_indent(c): |
| 213 | """Return a string with C spaces""" |
| 214 | return ' ' * c |
| 215 | |
| 216 | |
| 217 | def str_fields(fields): |
| 218 | """Return a string uniquely identifing FIELDS""" |
| 219 | r = '' |
| 220 | for n in sorted(fields.keys()): |
| 221 | r += '_' + n |
| 222 | return r[1:] |
| 223 | |
| 224 | |
| 225 | def str_match_bits(bits, mask): |
| 226 | """Return a string pretty-printing BITS/MASK""" |
| 227 | global insnwidth |
| 228 | |
| 229 | i = 1 << (insnwidth - 1) |
| 230 | space = 0x01010100 |
| 231 | r = '' |
| 232 | while i != 0: |
| 233 | if i & mask: |
| 234 | if i & bits: |
| 235 | r += '1' |
| 236 | else: |
| 237 | r += '0' |
| 238 | else: |
| 239 | r += '.' |
| 240 | if i & space: |
| 241 | r += ' ' |
| 242 | i >>= 1 |
| 243 | return r |
| 244 | |
| 245 | |
| 246 | def is_pow2(x): |
| 247 | """Return true iff X is equal to a power of 2.""" |
| 248 | return (x & (x - 1)) == 0 |
| 249 | |
| 250 | |
| 251 | def ctz(x): |
| 252 | """Return the number of times 2 factors into X.""" |
| 253 | r = 0 |
| 254 | while ((x >> r) & 1) == 0: |
| 255 | r += 1 |
| 256 | return r |
| 257 | |
| 258 | |
| 259 | def is_contiguous(bits): |
| 260 | shift = ctz(bits) |
| 261 | if is_pow2((bits >> shift) + 1): |
| 262 | return shift |
| 263 | else: |
| 264 | return -1 |
| 265 | |
| 266 | |
| 267 | def eq_fields_for_args(flds_a, flds_b): |
| 268 | if len(flds_a) != len(flds_b): |
| 269 | return False |
| 270 | for k, a in flds_a.items(): |
| 271 | if k not in flds_b: |
| 272 | return False |
| 273 | return True |
| 274 | |
| 275 | |
| 276 | def eq_fields_for_fmts(flds_a, flds_b): |
| 277 | if len(flds_a) != len(flds_b): |
| 278 | return False |
| 279 | for k, a in flds_a.items(): |
| 280 | if k not in flds_b: |
| 281 | return False |
| 282 | b = flds_b[k] |
| 283 | if a.__class__ != b.__class__ or a != b: |
| 284 | return False |
| 285 | return True |
| 286 | |
| 287 | |
| 288 | class Field: |
| 289 | """Class representing a simple instruction field""" |
| 290 | def __init__(self, sign, pos, len): |
| 291 | self.sign = sign |
| 292 | self.pos = pos |
| 293 | self.len = len |
| 294 | self.mask = ((1 << len) - 1) << pos |
| 295 | |
| 296 | def __str__(self): |
| 297 | if self.sign: |
| 298 | s = 's' |
| 299 | else: |
| 300 | s = '' |
| 301 | return str(pos) + ':' + s + str(len) |
| 302 | |
| 303 | def str_extract(self): |
| 304 | if self.sign: |
| 305 | extr = 'sextract32' |
| 306 | else: |
| 307 | extr = 'extract32' |
| 308 | return '{0}(insn, {1}, {2})'.format(extr, self.pos, self.len) |
| 309 | |
| 310 | def __eq__(self, other): |
| 311 | return self.sign == other.sign and self.sign == other.sign |
| 312 | |
| 313 | def __ne__(self, other): |
| 314 | return not self.__eq__(other) |
| 315 | # end Field |
| 316 | |
| 317 | |
| 318 | class MultiField: |
| 319 | """Class representing a compound instruction field""" |
| 320 | def __init__(self, subs, mask): |
| 321 | self.subs = subs |
| 322 | self.sign = subs[0].sign |
| 323 | self.mask = mask |
| 324 | |
| 325 | def __str__(self): |
| 326 | return str(self.subs) |
| 327 | |
| 328 | def str_extract(self): |
| 329 | ret = '0' |
| 330 | pos = 0 |
| 331 | for f in reversed(self.subs): |
| 332 | if pos == 0: |
| 333 | ret = f.str_extract() |
| 334 | else: |
| 335 | ret = 'deposit32({0}, {1}, {2}, {3})' \ |
| 336 | .format(ret, pos, 32 - pos, f.str_extract()) |
| 337 | pos += f.len |
| 338 | return ret |
| 339 | |
| 340 | def __ne__(self, other): |
| 341 | if len(self.subs) != len(other.subs): |
| 342 | return True |
| 343 | for a, b in zip(self.subs, other.subs): |
| 344 | if a.__class__ != b.__class__ or a != b: |
| 345 | return True |
| 346 | return False |
| 347 | |
| 348 | def __eq__(self, other): |
| 349 | return not self.__ne__(other) |
| 350 | # end MultiField |
| 351 | |
| 352 | |
| 353 | class ConstField: |
| 354 | """Class representing an argument field with constant value""" |
| 355 | def __init__(self, value): |
| 356 | self.value = value |
| 357 | self.mask = 0 |
| 358 | self.sign = value < 0 |
| 359 | |
| 360 | def __str__(self): |
| 361 | return str(self.value) |
| 362 | |
| 363 | def str_extract(self): |
| 364 | return str(self.value) |
| 365 | |
| 366 | def __cmp__(self, other): |
| 367 | return self.value - other.value |
| 368 | # end ConstField |
| 369 | |
| 370 | |
| 371 | class FunctionField: |
| 372 | """Class representing a field passed through an expander""" |
| 373 | def __init__(self, func, base): |
| 374 | self.mask = base.mask |
| 375 | self.sign = base.sign |
| 376 | self.base = base |
| 377 | self.func = func |
| 378 | |
| 379 | def __str__(self): |
| 380 | return self.func + '(' + str(self.base) + ')' |
| 381 | |
| 382 | def str_extract(self): |
| 383 | return self.func + '(' + self.base.str_extract() + ')' |
| 384 | |
| 385 | def __eq__(self, other): |
| 386 | return self.func == other.func and self.base == other.base |
| 387 | |
| 388 | def __ne__(self, other): |
| 389 | return not self.__eq__(other) |
| 390 | # end FunctionField |
| 391 | |
| 392 | |
| 393 | class Arguments: |
| 394 | """Class representing the extracted fields of a format""" |
| 395 | def __init__(self, nm, flds): |
| 396 | self.name = nm |
| 397 | self.fields = sorted(flds) |
| 398 | |
| 399 | def __str__(self): |
| 400 | return self.name + ' ' + str(self.fields) |
| 401 | |
| 402 | def struct_name(self): |
| 403 | return 'arg_' + self.name |
| 404 | |
| 405 | def output_def(self): |
| 406 | output('typedef struct {\n') |
| 407 | for n in self.fields: |
| 408 | output(' int ', n, ';\n') |
| 409 | output('} ', self.struct_name(), ';\n\n') |
| 410 | # end Arguments |
| 411 | |
| 412 | |
| 413 | class General: |
| 414 | """Common code between instruction formats and instruction patterns""" |
| 415 | def __init__(self, name, lineno, base, fixb, fixm, udfm, fldm, flds): |
| 416 | self.name = name |
| 417 | self.lineno = lineno |
| 418 | self.base = base |
| 419 | self.fixedbits = fixb |
| 420 | self.fixedmask = fixm |
| 421 | self.undefmask = udfm |
| 422 | self.fieldmask = fldm |
| 423 | self.fields = flds |
| 424 | |
| 425 | def __str__(self): |
| 426 | r = self.name |
| 427 | if self.base: |
| 428 | r = r + ' ' + self.base.name |
| 429 | else: |
| 430 | r = r + ' ' + str(self.fields) |
| 431 | r = r + ' ' + str_match_bits(self.fixedbits, self.fixedmask) |
| 432 | return r |
| 433 | |
| 434 | def str1(self, i): |
| 435 | return str_indent(i) + self.__str__() |
| 436 | # end General |
| 437 | |
| 438 | |
| 439 | class Format(General): |
| 440 | """Class representing an instruction format""" |
| 441 | |
| 442 | def extract_name(self): |
| 443 | return 'extract_' + self.name |
| 444 | |
| 445 | def output_extract(self): |
| 446 | output('static void ', self.extract_name(), '(', |
| 447 | self.base.struct_name(), ' *a, ', insntype, ' insn)\n{\n') |
| 448 | for n, f in self.fields.items(): |
| 449 | output(' a->', n, ' = ', f.str_extract(), ';\n') |
| 450 | output('}\n\n') |
| 451 | # end Format |
| 452 | |
| 453 | |
| 454 | class Pattern(General): |
| 455 | """Class representing an instruction pattern""" |
| 456 | |
| 457 | def output_decl(self): |
| 458 | global translate_scope |
| 459 | global translate_prefix |
| 460 | output('typedef ', self.base.base.struct_name(), |
| 461 | ' arg_', self.name, ';\n') |
Richard Henderson | 7680559 | 2018-03-02 10:45:35 +0000 | [diff] [blame] | 462 | output(translate_scope, 'bool ', translate_prefix, '_', self.name, |
Richard Henderson | 568ae7e | 2017-12-07 12:44:09 -0800 | [diff] [blame] | 463 | '(DisasContext *ctx, arg_', self.name, |
| 464 | ' *a, ', insntype, ' insn);\n') |
| 465 | |
| 466 | def output_code(self, i, extracted, outerbits, outermask): |
| 467 | global translate_prefix |
| 468 | ind = str_indent(i) |
| 469 | arg = self.base.base.name |
| 470 | output(ind, '/* line ', str(self.lineno), ' */\n') |
| 471 | if not extracted: |
| 472 | output(ind, self.base.extract_name(), '(&u.f_', arg, ', insn);\n') |
| 473 | for n, f in self.fields.items(): |
| 474 | output(ind, 'u.f_', arg, '.', n, ' = ', f.str_extract(), ';\n') |
Richard Henderson | 7680559 | 2018-03-02 10:45:35 +0000 | [diff] [blame] | 475 | output(ind, 'return ', translate_prefix, '_', self.name, |
Richard Henderson | 568ae7e | 2017-12-07 12:44:09 -0800 | [diff] [blame] | 476 | '(ctx, &u.f_', arg, ', insn);\n') |
Richard Henderson | 568ae7e | 2017-12-07 12:44:09 -0800 | [diff] [blame] | 477 | # end Pattern |
| 478 | |
| 479 | |
| 480 | def parse_field(lineno, name, toks): |
| 481 | """Parse one instruction field from TOKS at LINENO""" |
| 482 | global fields |
| 483 | global re_ident |
| 484 | global insnwidth |
| 485 | |
| 486 | # A "simple" field will have only one entry; |
| 487 | # a "multifield" will have several. |
| 488 | subs = [] |
| 489 | width = 0 |
| 490 | func = None |
| 491 | for t in toks: |
| 492 | if re_fullmatch('!function=' + re_ident, t): |
| 493 | if func: |
| 494 | error(lineno, 'duplicate function') |
| 495 | func = t.split('=') |
| 496 | func = func[1] |
| 497 | continue |
| 498 | |
| 499 | if re_fullmatch('[0-9]+:s[0-9]+', t): |
| 500 | # Signed field extract |
| 501 | subtoks = t.split(':s') |
| 502 | sign = True |
| 503 | elif re_fullmatch('[0-9]+:[0-9]+', t): |
| 504 | # Unsigned field extract |
| 505 | subtoks = t.split(':') |
| 506 | sign = False |
| 507 | else: |
| 508 | error(lineno, 'invalid field token "{0}"'.format(t)) |
| 509 | po = int(subtoks[0]) |
| 510 | le = int(subtoks[1]) |
| 511 | if po + le > insnwidth: |
| 512 | error(lineno, 'field {0} too large'.format(t)) |
| 513 | f = Field(sign, po, le) |
| 514 | subs.append(f) |
| 515 | width += le |
| 516 | |
| 517 | if width > insnwidth: |
| 518 | error(lineno, 'field too large') |
| 519 | if len(subs) == 1: |
| 520 | f = subs[0] |
| 521 | else: |
| 522 | mask = 0 |
| 523 | for s in subs: |
| 524 | if mask & s.mask: |
| 525 | error(lineno, 'field components overlap') |
| 526 | mask |= s.mask |
| 527 | f = MultiField(subs, mask) |
| 528 | if func: |
| 529 | f = FunctionField(func, f) |
| 530 | |
| 531 | if name in fields: |
| 532 | error(lineno, 'duplicate field', name) |
| 533 | fields[name] = f |
| 534 | # end parse_field |
| 535 | |
| 536 | |
| 537 | def parse_arguments(lineno, name, toks): |
| 538 | """Parse one argument set from TOKS at LINENO""" |
| 539 | global arguments |
| 540 | global re_ident |
| 541 | |
| 542 | flds = [] |
| 543 | for t in toks: |
| 544 | if not re_fullmatch(re_ident, t): |
| 545 | error(lineno, 'invalid argument set token "{0}"'.format(t)) |
| 546 | if t in flds: |
| 547 | error(lineno, 'duplicate argument "{0}"'.format(t)) |
| 548 | flds.append(t) |
| 549 | |
| 550 | if name in arguments: |
| 551 | error(lineno, 'duplicate argument set', name) |
| 552 | arguments[name] = Arguments(name, flds) |
| 553 | # end parse_arguments |
| 554 | |
| 555 | |
| 556 | def lookup_field(lineno, name): |
| 557 | global fields |
| 558 | if name in fields: |
| 559 | return fields[name] |
| 560 | error(lineno, 'undefined field', name) |
| 561 | |
| 562 | |
| 563 | def add_field(lineno, flds, new_name, f): |
| 564 | if new_name in flds: |
| 565 | error(lineno, 'duplicate field', new_name) |
| 566 | flds[new_name] = f |
| 567 | return flds |
| 568 | |
| 569 | |
| 570 | def add_field_byname(lineno, flds, new_name, old_name): |
| 571 | return add_field(lineno, flds, new_name, lookup_field(lineno, old_name)) |
| 572 | |
| 573 | |
| 574 | def infer_argument_set(flds): |
| 575 | global arguments |
| 576 | |
| 577 | for arg in arguments.values(): |
| 578 | if eq_fields_for_args(flds, arg.fields): |
| 579 | return arg |
| 580 | |
| 581 | name = str(len(arguments)) |
| 582 | arg = Arguments(name, flds.keys()) |
| 583 | arguments[name] = arg |
| 584 | return arg |
| 585 | |
| 586 | |
| 587 | def infer_format(arg, fieldmask, flds): |
| 588 | global arguments |
| 589 | global formats |
| 590 | |
| 591 | const_flds = {} |
| 592 | var_flds = {} |
| 593 | for n, c in flds.items(): |
| 594 | if c is ConstField: |
| 595 | const_flds[n] = c |
| 596 | else: |
| 597 | var_flds[n] = c |
| 598 | |
| 599 | # Look for an existing format with the same argument set and fields |
| 600 | for fmt in formats.values(): |
| 601 | if arg and fmt.base != arg: |
| 602 | continue |
| 603 | if fieldmask != fmt.fieldmask: |
| 604 | continue |
| 605 | if not eq_fields_for_fmts(flds, fmt.fields): |
| 606 | continue |
| 607 | return (fmt, const_flds) |
| 608 | |
| 609 | name = 'Fmt_' + str(len(formats)) |
| 610 | if not arg: |
| 611 | arg = infer_argument_set(flds) |
| 612 | |
| 613 | fmt = Format(name, 0, arg, 0, 0, 0, fieldmask, var_flds) |
| 614 | formats[name] = fmt |
| 615 | |
| 616 | return (fmt, const_flds) |
| 617 | # end infer_format |
| 618 | |
| 619 | |
| 620 | def parse_generic(lineno, is_format, name, toks): |
| 621 | """Parse one instruction format from TOKS at LINENO""" |
| 622 | global fields |
| 623 | global arguments |
| 624 | global formats |
| 625 | global patterns |
| 626 | global re_ident |
| 627 | global insnwidth |
| 628 | global insnmask |
| 629 | |
| 630 | fixedmask = 0 |
| 631 | fixedbits = 0 |
| 632 | undefmask = 0 |
| 633 | width = 0 |
| 634 | flds = {} |
| 635 | arg = None |
| 636 | fmt = None |
| 637 | for t in toks: |
| 638 | # '&Foo' gives a format an explcit argument set. |
| 639 | if t[0] == '&': |
| 640 | tt = t[1:] |
| 641 | if arg: |
| 642 | error(lineno, 'multiple argument sets') |
| 643 | if tt in arguments: |
| 644 | arg = arguments[tt] |
| 645 | else: |
| 646 | error(lineno, 'undefined argument set', t) |
| 647 | continue |
| 648 | |
| 649 | # '@Foo' gives a pattern an explicit format. |
| 650 | if t[0] == '@': |
| 651 | tt = t[1:] |
| 652 | if fmt: |
| 653 | error(lineno, 'multiple formats') |
| 654 | if tt in formats: |
| 655 | fmt = formats[tt] |
| 656 | else: |
| 657 | error(lineno, 'undefined format', t) |
| 658 | continue |
| 659 | |
| 660 | # '%Foo' imports a field. |
| 661 | if t[0] == '%': |
| 662 | tt = t[1:] |
| 663 | flds = add_field_byname(lineno, flds, tt, tt) |
| 664 | continue |
| 665 | |
| 666 | # 'Foo=%Bar' imports a field with a different name. |
| 667 | if re_fullmatch(re_ident + '=%' + re_ident, t): |
| 668 | (fname, iname) = t.split('=%') |
| 669 | flds = add_field_byname(lineno, flds, fname, iname) |
| 670 | continue |
| 671 | |
| 672 | # 'Foo=number' sets an argument field to a constant value |
| 673 | if re_fullmatch(re_ident + '=[0-9]+', t): |
| 674 | (fname, value) = t.split('=') |
| 675 | value = int(value) |
| 676 | flds = add_field(lineno, flds, fname, ConstField(value)) |
| 677 | continue |
| 678 | |
| 679 | # Pattern of 0s, 1s, dots and dashes indicate required zeros, |
| 680 | # required ones, or dont-cares. |
| 681 | if re_fullmatch('[01.-]+', t): |
| 682 | shift = len(t) |
| 683 | fms = t.replace('0', '1') |
| 684 | fms = fms.replace('.', '0') |
| 685 | fms = fms.replace('-', '0') |
| 686 | fbs = t.replace('.', '0') |
| 687 | fbs = fbs.replace('-', '0') |
| 688 | ubm = t.replace('1', '0') |
| 689 | ubm = ubm.replace('.', '0') |
| 690 | ubm = ubm.replace('-', '1') |
| 691 | fms = int(fms, 2) |
| 692 | fbs = int(fbs, 2) |
| 693 | ubm = int(ubm, 2) |
| 694 | fixedbits = (fixedbits << shift) | fbs |
| 695 | fixedmask = (fixedmask << shift) | fms |
| 696 | undefmask = (undefmask << shift) | ubm |
| 697 | # Otherwise, fieldname:fieldwidth |
| 698 | elif re_fullmatch(re_ident + ':s?[0-9]+', t): |
| 699 | (fname, flen) = t.split(':') |
| 700 | sign = False |
| 701 | if flen[0] == 's': |
| 702 | sign = True |
| 703 | flen = flen[1:] |
| 704 | shift = int(flen, 10) |
| 705 | f = Field(sign, insnwidth - width - shift, shift) |
| 706 | flds = add_field(lineno, flds, fname, f) |
| 707 | fixedbits <<= shift |
| 708 | fixedmask <<= shift |
| 709 | undefmask <<= shift |
| 710 | else: |
| 711 | error(lineno, 'invalid token "{0}"'.format(t)) |
| 712 | width += shift |
| 713 | |
| 714 | # We should have filled in all of the bits of the instruction. |
| 715 | if not (is_format and width == 0) and width != insnwidth: |
| 716 | error(lineno, 'definition has {0} bits'.format(width)) |
| 717 | |
| 718 | # Do not check for fields overlaping fields; one valid usage |
| 719 | # is to be able to duplicate fields via import. |
| 720 | fieldmask = 0 |
| 721 | for f in flds.values(): |
| 722 | fieldmask |= f.mask |
| 723 | |
| 724 | # Fix up what we've parsed to match either a format or a pattern. |
| 725 | if is_format: |
| 726 | # Formats cannot reference formats. |
| 727 | if fmt: |
| 728 | error(lineno, 'format referencing format') |
| 729 | # If an argument set is given, then there should be no fields |
| 730 | # without a place to store it. |
| 731 | if arg: |
| 732 | for f in flds.keys(): |
| 733 | if f not in arg.fields: |
| 734 | error(lineno, 'field {0} not in argument set {1}' |
| 735 | .format(f, arg.name)) |
| 736 | else: |
| 737 | arg = infer_argument_set(flds) |
| 738 | if name in formats: |
| 739 | error(lineno, 'duplicate format name', name) |
| 740 | fmt = Format(name, lineno, arg, fixedbits, fixedmask, |
| 741 | undefmask, fieldmask, flds) |
| 742 | formats[name] = fmt |
| 743 | else: |
| 744 | # Patterns can reference a format ... |
| 745 | if fmt: |
| 746 | # ... but not an argument simultaneously |
| 747 | if arg: |
| 748 | error(lineno, 'pattern specifies both format and argument set') |
| 749 | if fixedmask & fmt.fixedmask: |
| 750 | error(lineno, 'pattern fixed bits overlap format fixed bits') |
| 751 | fieldmask |= fmt.fieldmask |
| 752 | fixedbits |= fmt.fixedbits |
| 753 | fixedmask |= fmt.fixedmask |
| 754 | undefmask |= fmt.undefmask |
| 755 | else: |
| 756 | (fmt, flds) = infer_format(arg, fieldmask, flds) |
| 757 | arg = fmt.base |
| 758 | for f in flds.keys(): |
| 759 | if f not in arg.fields: |
| 760 | error(lineno, 'field {0} not in argument set {1}' |
| 761 | .format(f, arg.name)) |
| 762 | if f in fmt.fields.keys(): |
| 763 | error(lineno, 'field {0} set by format and pattern'.format(f)) |
| 764 | for f in arg.fields: |
| 765 | if f not in flds.keys() and f not in fmt.fields.keys(): |
| 766 | error(lineno, 'field {0} not initialized'.format(f)) |
| 767 | pat = Pattern(name, lineno, fmt, fixedbits, fixedmask, |
| 768 | undefmask, fieldmask, flds) |
| 769 | patterns.append(pat) |
| 770 | |
| 771 | # Validate the masks that we have assembled. |
| 772 | if fieldmask & fixedmask: |
| 773 | error(lineno, 'fieldmask overlaps fixedmask (0x{0:08x} & 0x{1:08x})' |
| 774 | .format(fieldmask, fixedmask)) |
| 775 | if fieldmask & undefmask: |
| 776 | error(lineno, 'fieldmask overlaps undefmask (0x{0:08x} & 0x{1:08x})' |
| 777 | .format(fieldmask, undefmask)) |
| 778 | if fixedmask & undefmask: |
| 779 | error(lineno, 'fixedmask overlaps undefmask (0x{0:08x} & 0x{1:08x})' |
| 780 | .format(fixedmask, undefmask)) |
| 781 | if not is_format: |
| 782 | allbits = fieldmask | fixedmask | undefmask |
| 783 | if allbits != insnmask: |
| 784 | error(lineno, 'bits left unspecified (0x{0:08x})' |
| 785 | .format(allbits ^ insnmask)) |
| 786 | # end parse_general |
| 787 | |
| 788 | |
| 789 | def parse_file(f): |
| 790 | """Parse all of the patterns within a file""" |
| 791 | |
| 792 | # Read all of the lines of the file. Concatenate lines |
| 793 | # ending in backslash; discard empty lines and comments. |
| 794 | toks = [] |
| 795 | lineno = 0 |
| 796 | for line in f: |
| 797 | lineno += 1 |
| 798 | |
| 799 | # Discard comments |
| 800 | end = line.find('#') |
| 801 | if end >= 0: |
| 802 | line = line[:end] |
| 803 | |
| 804 | t = line.split() |
| 805 | if len(toks) != 0: |
| 806 | # Next line after continuation |
| 807 | toks.extend(t) |
| 808 | elif len(t) == 0: |
| 809 | # Empty line |
| 810 | continue |
| 811 | else: |
| 812 | toks = t |
| 813 | |
| 814 | # Continuation? |
| 815 | if toks[-1] == '\\': |
| 816 | toks.pop() |
| 817 | continue |
| 818 | |
| 819 | if len(toks) < 2: |
| 820 | error(lineno, 'short line') |
| 821 | |
| 822 | name = toks[0] |
| 823 | del toks[0] |
| 824 | |
| 825 | # Determine the type of object needing to be parsed. |
| 826 | if name[0] == '%': |
| 827 | parse_field(lineno, name[1:], toks) |
| 828 | elif name[0] == '&': |
| 829 | parse_arguments(lineno, name[1:], toks) |
| 830 | elif name[0] == '@': |
| 831 | parse_generic(lineno, True, name[1:], toks) |
| 832 | else: |
| 833 | parse_generic(lineno, False, name, toks) |
| 834 | toks = [] |
| 835 | # end parse_file |
| 836 | |
| 837 | |
| 838 | class Tree: |
| 839 | """Class representing a node in a decode tree""" |
| 840 | |
| 841 | def __init__(self, fm, tm): |
| 842 | self.fixedmask = fm |
| 843 | self.thismask = tm |
| 844 | self.subs = [] |
| 845 | self.base = None |
| 846 | |
| 847 | def str1(self, i): |
| 848 | ind = str_indent(i) |
| 849 | r = '{0}{1:08x}'.format(ind, self.fixedmask) |
| 850 | if self.format: |
| 851 | r += ' ' + self.format.name |
| 852 | r += ' [\n' |
| 853 | for (b, s) in self.subs: |
| 854 | r += '{0} {1:08x}:\n'.format(ind, b) |
| 855 | r += s.str1(i + 4) + '\n' |
| 856 | r += ind + ']' |
| 857 | return r |
| 858 | |
| 859 | def __str__(self): |
| 860 | return self.str1(0) |
| 861 | |
| 862 | def output_code(self, i, extracted, outerbits, outermask): |
| 863 | ind = str_indent(i) |
| 864 | |
| 865 | # If we identified all nodes below have the same format, |
| 866 | # extract the fields now. |
| 867 | if not extracted and self.base: |
| 868 | output(ind, self.base.extract_name(), |
| 869 | '(&u.f_', self.base.base.name, ', insn);\n') |
| 870 | extracted = True |
| 871 | |
| 872 | # Attempt to aid the compiler in producing compact switch statements. |
| 873 | # If the bits in the mask are contiguous, extract them. |
| 874 | sh = is_contiguous(self.thismask) |
| 875 | if sh > 0: |
| 876 | # Propagate SH down into the local functions. |
| 877 | def str_switch(b, sh=sh): |
| 878 | return '(insn >> {0}) & 0x{1:x}'.format(sh, b >> sh) |
| 879 | |
| 880 | def str_case(b, sh=sh): |
| 881 | return '0x{0:x}'.format(b >> sh) |
| 882 | else: |
| 883 | def str_switch(b): |
| 884 | return 'insn & 0x{0:08x}'.format(b) |
| 885 | |
| 886 | def str_case(b): |
| 887 | return '0x{0:08x}'.format(b) |
| 888 | |
| 889 | output(ind, 'switch (', str_switch(self.thismask), ') {\n') |
| 890 | for b, s in sorted(self.subs): |
| 891 | assert (self.thismask & ~s.fixedmask) == 0 |
| 892 | innermask = outermask | self.thismask |
| 893 | innerbits = outerbits | b |
| 894 | output(ind, 'case ', str_case(b), ':\n') |
| 895 | output(ind, ' /* ', |
| 896 | str_match_bits(innerbits, innermask), ' */\n') |
| 897 | s.output_code(i + 4, extracted, innerbits, innermask) |
| 898 | output(ind, '}\n') |
| 899 | output(ind, 'return false;\n') |
| 900 | # end Tree |
| 901 | |
| 902 | |
| 903 | def build_tree(pats, outerbits, outermask): |
| 904 | # Find the intersection of all remaining fixedmask. |
| 905 | innermask = ~outermask |
| 906 | for i in pats: |
| 907 | innermask &= i.fixedmask |
| 908 | |
| 909 | if innermask == 0: |
| 910 | pnames = [] |
| 911 | for p in pats: |
| 912 | pnames.append(p.name + ':' + str(p.lineno)) |
| 913 | error(pats[0].lineno, 'overlapping patterns:', pnames) |
| 914 | |
| 915 | fullmask = outermask | innermask |
| 916 | |
| 917 | # Sort each element of pats into the bin selected by the mask. |
| 918 | bins = {} |
| 919 | for i in pats: |
| 920 | fb = i.fixedbits & innermask |
| 921 | if fb in bins: |
| 922 | bins[fb].append(i) |
| 923 | else: |
| 924 | bins[fb] = [i] |
| 925 | |
| 926 | # We must recurse if any bin has more than one element or if |
| 927 | # the single element in the bin has not been fully matched. |
| 928 | t = Tree(fullmask, innermask) |
| 929 | |
| 930 | for b, l in bins.items(): |
| 931 | s = l[0] |
| 932 | if len(l) > 1 or s.fixedmask & ~fullmask != 0: |
| 933 | s = build_tree(l, b | outerbits, fullmask) |
| 934 | t.subs.append((b, s)) |
| 935 | |
| 936 | return t |
| 937 | # end build_tree |
| 938 | |
| 939 | |
| 940 | def prop_format(tree): |
| 941 | """Propagate Format objects into the decode tree""" |
| 942 | |
| 943 | # Depth first search. |
| 944 | for (b, s) in tree.subs: |
| 945 | if isinstance(s, Tree): |
| 946 | prop_format(s) |
| 947 | |
| 948 | # If all entries in SUBS have the same format, then |
| 949 | # propagate that into the tree. |
| 950 | f = None |
| 951 | for (b, s) in tree.subs: |
| 952 | if f is None: |
| 953 | f = s.base |
| 954 | if f is None: |
| 955 | return |
| 956 | if f is not s.base: |
| 957 | return |
| 958 | tree.base = f |
| 959 | # end prop_format |
| 960 | |
| 961 | |
| 962 | def main(): |
| 963 | global arguments |
| 964 | global formats |
| 965 | global patterns |
| 966 | global translate_scope |
| 967 | global translate_prefix |
| 968 | global output_fd |
| 969 | global output_file |
| 970 | global input_file |
| 971 | global insnwidth |
| 972 | global insntype |
Bastian Koppelmann | 83d7c40 | 2018-03-19 12:58:46 +0100 | [diff] [blame] | 973 | global insnmask |
Richard Henderson | 568ae7e | 2017-12-07 12:44:09 -0800 | [diff] [blame] | 974 | |
| 975 | decode_function = 'decode' |
| 976 | decode_scope = 'static ' |
| 977 | |
| 978 | long_opts = ['decode=', 'translate=', 'output=', 'insnwidth='] |
| 979 | try: |
| 980 | (opts, args) = getopt.getopt(sys.argv[1:], 'o:w:', long_opts) |
| 981 | except getopt.GetoptError as err: |
| 982 | error(0, err) |
| 983 | for o, a in opts: |
| 984 | if o in ('-o', '--output'): |
| 985 | output_file = a |
| 986 | elif o == '--decode': |
| 987 | decode_function = a |
| 988 | decode_scope = '' |
| 989 | elif o == '--translate': |
| 990 | translate_prefix = a |
| 991 | translate_scope = '' |
| 992 | elif o in ('-w', '--insnwidth'): |
| 993 | insnwidth = int(a) |
| 994 | if insnwidth == 16: |
| 995 | insntype = 'uint16_t' |
| 996 | insnmask = 0xffff |
| 997 | elif insnwidth != 32: |
| 998 | error(0, 'cannot handle insns of width', insnwidth) |
| 999 | else: |
| 1000 | assert False, 'unhandled option' |
| 1001 | |
| 1002 | if len(args) < 1: |
| 1003 | error(0, 'missing input file') |
| 1004 | input_file = args[0] |
| 1005 | f = open(input_file, 'r') |
| 1006 | parse_file(f) |
| 1007 | f.close() |
| 1008 | |
| 1009 | t = build_tree(patterns, 0, 0) |
| 1010 | prop_format(t) |
| 1011 | |
| 1012 | if output_file: |
| 1013 | output_fd = open(output_file, 'w') |
| 1014 | else: |
| 1015 | output_fd = sys.stdout |
| 1016 | |
| 1017 | output_autogen() |
| 1018 | for n in sorted(arguments.keys()): |
| 1019 | f = arguments[n] |
| 1020 | f.output_def() |
| 1021 | |
| 1022 | # A single translate function can be invoked for different patterns. |
| 1023 | # Make sure that the argument sets are the same, and declare the |
| 1024 | # function only once. |
| 1025 | out_pats = {} |
| 1026 | for i in patterns: |
| 1027 | if i.name in out_pats: |
| 1028 | p = out_pats[i.name] |
| 1029 | if i.base.base != p.base.base: |
| 1030 | error(0, i.name, ' has conflicting argument sets') |
| 1031 | else: |
| 1032 | i.output_decl() |
| 1033 | out_pats[i.name] = i |
| 1034 | output('\n') |
| 1035 | |
| 1036 | for n in sorted(formats.keys()): |
| 1037 | f = formats[n] |
| 1038 | f.output_extract() |
| 1039 | |
| 1040 | output(decode_scope, 'bool ', decode_function, |
| 1041 | '(DisasContext *ctx, ', insntype, ' insn)\n{\n') |
| 1042 | |
| 1043 | i4 = str_indent(4) |
| 1044 | output(i4, 'union {\n') |
| 1045 | for n in sorted(arguments.keys()): |
| 1046 | f = arguments[n] |
| 1047 | output(i4, i4, f.struct_name(), ' f_', f.name, ';\n') |
| 1048 | output(i4, '} u;\n\n') |
| 1049 | |
| 1050 | t.output_code(4, False, 0, 0) |
| 1051 | |
| 1052 | output('}\n') |
| 1053 | |
| 1054 | if output_file: |
| 1055 | output_fd.close() |
| 1056 | # end main |
| 1057 | |
| 1058 | |
| 1059 | if __name__ == '__main__': |
| 1060 | main() |