isa_parser.py revision 13602:73512cfcca53
1# Copyright (c) 2014, 2016, 2019 ARM Limited 2# All rights reserved 3# 4# The license below extends only to copyright in the software and shall 5# not be construed as granting a license to any other intellectual 6# property including but not limited to intellectual property relating 7# to a hardware implementation of the functionality of the software 8# licensed hereunder. You may use the software subject to the license 9# terms below provided that you ensure that this notice is replicated 10# unmodified and in its entirety in all distributions of the software, 11# modified or unmodified, in source code or in binary form. 12# 13# Copyright (c) 2003-2005 The Regents of The University of Michigan 14# Copyright (c) 2013,2015 Advanced Micro Devices, Inc. 15# All rights reserved. 16# 17# Redistribution and use in source and binary forms, with or without 18# modification, are permitted provided that the following conditions are 19# met: redistributions of source code must retain the above copyright 20# notice, this list of conditions and the following disclaimer; 21# redistributions in binary form must reproduce the above copyright 22# notice, this list of conditions and the following disclaimer in the 23# documentation and/or other materials provided with the distribution; 24# neither the name of the copyright holders nor the names of its 25# contributors may be used to endorse or promote products derived from 26# this software without specific prior written permission. 27# 28# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 29# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 30# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 31# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 32# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 33# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 34# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 35# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 36# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 37# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 38# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 39# 40# Authors: Steve Reinhardt 41 42from __future__ import with_statement, print_function 43import os 44import sys 45import re 46import string 47import inspect, traceback 48# get type names 49from types import * 50 51from m5.util.grammar import Grammar 52 53debug=False 54 55################### 56# Utility functions 57 58# 59# Indent every line in string 's' by two spaces 60# (except preprocessor directives). 61# Used to make nested code blocks look pretty. 62# 63def indent(s): 64 return re.sub(r'(?m)^(?!#)', ' ', s) 65 66# 67# Munge a somewhat arbitrarily formatted piece of Python code 68# (e.g. from a format 'let' block) into something whose indentation 69# will get by the Python parser. 70# 71# The two keys here are that Python will give a syntax error if 72# there's any whitespace at the beginning of the first line, and that 73# all lines at the same lexical nesting level must have identical 74# indentation. Unfortunately the way code literals work, an entire 75# let block tends to have some initial indentation. Rather than 76# trying to figure out what that is and strip it off, we prepend 'if 77# 1:' to make the let code the nested block inside the if (and have 78# the parser automatically deal with the indentation for us). 79# 80# We don't want to do this if (1) the code block is empty or (2) the 81# first line of the block doesn't have any whitespace at the front. 82 83def fixPythonIndentation(s): 84 # get rid of blank lines first 85 s = re.sub(r'(?m)^\s*\n', '', s); 86 if (s != '' and re.match(r'[ \t]', s[0])): 87 s = 'if 1:\n' + s 88 return s 89 90class ISAParserError(Exception): 91 """Exception class for parser errors""" 92 def __init__(self, first, second=None): 93 if second is None: 94 self.lineno = 0 95 self.string = first 96 else: 97 self.lineno = first 98 self.string = second 99 100 def __str__(self): 101 return self.string 102 103def error(*args): 104 raise ISAParserError(*args) 105 106#################### 107# Template objects. 108# 109# Template objects are format strings that allow substitution from 110# the attribute spaces of other objects (e.g. InstObjParams instances). 111 112labelRE = re.compile(r'(?<!%)%\(([^\)]+)\)[sd]') 113 114class Template(object): 115 def __init__(self, parser, t): 116 self.parser = parser 117 self.template = t 118 119 def subst(self, d): 120 myDict = None 121 122 # Protect non-Python-dict substitutions (e.g. if there's a printf 123 # in the templated C++ code) 124 template = self.parser.protectNonSubstPercents(self.template) 125 126 # Build a dict ('myDict') to use for the template substitution. 127 # Start with the template namespace. Make a copy since we're 128 # going to modify it. 129 myDict = self.parser.templateMap.copy() 130 131 if isinstance(d, InstObjParams): 132 # If we're dealing with an InstObjParams object, we need 133 # to be a little more sophisticated. The instruction-wide 134 # parameters are already formed, but the parameters which 135 # are only function wide still need to be generated. 136 compositeCode = '' 137 138 myDict.update(d.__dict__) 139 # The "operands" and "snippets" attributes of the InstObjParams 140 # objects are for internal use and not substitution. 141 del myDict['operands'] 142 del myDict['snippets'] 143 144 snippetLabels = [l for l in labelRE.findall(template) 145 if d.snippets.has_key(l)] 146 147 snippets = dict([(s, self.parser.mungeSnippet(d.snippets[s])) 148 for s in snippetLabels]) 149 150 myDict.update(snippets) 151 152 compositeCode = ' '.join(map(str, snippets.values())) 153 154 # Add in template itself in case it references any 155 # operands explicitly (like Mem) 156 compositeCode += ' ' + template 157 158 operands = SubOperandList(self.parser, compositeCode, d.operands) 159 160 myDict['op_decl'] = operands.concatAttrStrings('op_decl') 161 if operands.readPC or operands.setPC: 162 myDict['op_decl'] += 'TheISA::PCState __parserAutoPCState;\n' 163 164 # In case there are predicated register reads and write, declare 165 # the variables for register indicies. It is being assumed that 166 # all the operands in the OperandList are also in the 167 # SubOperandList and in the same order. Otherwise, it is 168 # expected that predication would not be used for the operands. 169 if operands.predRead: 170 myDict['op_decl'] += 'uint8_t _sourceIndex = 0;\n' 171 if operands.predWrite: 172 myDict['op_decl'] += 'uint8_t M5_VAR_USED _destIndex = 0;\n' 173 174 is_src = lambda op: op.is_src 175 is_dest = lambda op: op.is_dest 176 177 myDict['op_src_decl'] = \ 178 operands.concatSomeAttrStrings(is_src, 'op_src_decl') 179 myDict['op_dest_decl'] = \ 180 operands.concatSomeAttrStrings(is_dest, 'op_dest_decl') 181 if operands.readPC: 182 myDict['op_src_decl'] += \ 183 'TheISA::PCState __parserAutoPCState;\n' 184 if operands.setPC: 185 myDict['op_dest_decl'] += \ 186 'TheISA::PCState __parserAutoPCState;\n' 187 188 myDict['op_rd'] = operands.concatAttrStrings('op_rd') 189 if operands.readPC: 190 myDict['op_rd'] = '__parserAutoPCState = xc->pcState();\n' + \ 191 myDict['op_rd'] 192 193 # Compose the op_wb string. If we're going to write back the 194 # PC state because we changed some of its elements, we'll need to 195 # do that as early as possible. That allows later uncoordinated 196 # modifications to the PC to layer appropriately. 197 reordered = list(operands.items) 198 reordered.reverse() 199 op_wb_str = '' 200 pcWbStr = 'xc->pcState(__parserAutoPCState);\n' 201 for op_desc in reordered: 202 if op_desc.isPCPart() and op_desc.is_dest: 203 op_wb_str = op_desc.op_wb + pcWbStr + op_wb_str 204 pcWbStr = '' 205 else: 206 op_wb_str = op_desc.op_wb + op_wb_str 207 myDict['op_wb'] = op_wb_str 208 209 elif isinstance(d, dict): 210 # if the argument is a dictionary, we just use it. 211 myDict.update(d) 212 elif hasattr(d, '__dict__'): 213 # if the argument is an object, we use its attribute map. 214 myDict.update(d.__dict__) 215 else: 216 raise TypeError, "Template.subst() arg must be or have dictionary" 217 return template % myDict 218 219 # Convert to string. 220 def __str__(self): 221 return self.template 222 223################ 224# Format object. 225# 226# A format object encapsulates an instruction format. It must provide 227# a defineInst() method that generates the code for an instruction 228# definition. 229 230class Format(object): 231 def __init__(self, id, params, code): 232 self.id = id 233 self.params = params 234 label = 'def format ' + id 235 self.user_code = compile(fixPythonIndentation(code), label, 'exec') 236 param_list = string.join(params, ", ") 237 f = '''def defInst(_code, _context, %s): 238 my_locals = vars().copy() 239 exec _code in _context, my_locals 240 return my_locals\n''' % param_list 241 c = compile(f, label + ' wrapper', 'exec') 242 exec c 243 self.func = defInst 244 245 def defineInst(self, parser, name, args, lineno): 246 parser.updateExportContext() 247 context = parser.exportContext.copy() 248 if len(name): 249 Name = name[0].upper() 250 if len(name) > 1: 251 Name += name[1:] 252 context.update({ 'name' : name, 'Name' : Name }) 253 try: 254 vars = self.func(self.user_code, context, *args[0], **args[1]) 255 except Exception, exc: 256 if debug: 257 raise 258 error(lineno, 'error defining "%s": %s.' % (name, exc)) 259 for k in vars.keys(): 260 if k not in ('header_output', 'decoder_output', 261 'exec_output', 'decode_block'): 262 del vars[k] 263 return GenCode(parser, **vars) 264 265# Special null format to catch an implicit-format instruction 266# definition outside of any format block. 267class NoFormat(object): 268 def __init__(self): 269 self.defaultInst = '' 270 271 def defineInst(self, parser, name, args, lineno): 272 error(lineno, 273 'instruction definition "%s" with no active format!' % name) 274 275############### 276# GenCode class 277# 278# The GenCode class encapsulates generated code destined for various 279# output files. The header_output and decoder_output attributes are 280# strings containing code destined for decoder.hh and decoder.cc 281# respectively. The decode_block attribute contains code to be 282# incorporated in the decode function itself (that will also end up in 283# decoder.cc). The exec_output attribute is the string of code for the 284# exec.cc file. The has_decode_default attribute is used in the decode block 285# to allow explicit default clauses to override default default clauses. 286 287class GenCode(object): 288 # Constructor. 289 def __init__(self, parser, 290 header_output = '', decoder_output = '', exec_output = '', 291 decode_block = '', has_decode_default = False): 292 self.parser = parser 293 self.header_output = header_output 294 self.decoder_output = decoder_output 295 self.exec_output = exec_output 296 self.decode_block = decode_block 297 self.has_decode_default = has_decode_default 298 299 # Write these code chunks out to the filesystem. They will be properly 300 # interwoven by the write_top_level_files(). 301 def emit(self): 302 if self.header_output: 303 self.parser.get_file('header').write(self.header_output) 304 if self.decoder_output: 305 self.parser.get_file('decoder').write(self.decoder_output) 306 if self.exec_output: 307 self.parser.get_file('exec').write(self.exec_output) 308 if self.decode_block: 309 self.parser.get_file('decode_block').write(self.decode_block) 310 311 # Override '+' operator: generate a new GenCode object that 312 # concatenates all the individual strings in the operands. 313 def __add__(self, other): 314 return GenCode(self.parser, 315 self.header_output + other.header_output, 316 self.decoder_output + other.decoder_output, 317 self.exec_output + other.exec_output, 318 self.decode_block + other.decode_block, 319 self.has_decode_default or other.has_decode_default) 320 321 # Prepend a string (typically a comment) to all the strings. 322 def prepend_all(self, pre): 323 self.header_output = pre + self.header_output 324 self.decoder_output = pre + self.decoder_output 325 self.decode_block = pre + self.decode_block 326 self.exec_output = pre + self.exec_output 327 328 # Wrap the decode block in a pair of strings (e.g., 'case foo:' 329 # and 'break;'). Used to build the big nested switch statement. 330 def wrap_decode_block(self, pre, post = ''): 331 self.decode_block = pre + indent(self.decode_block) + post 332 333##################################################################### 334# 335# Bitfield Operator Support 336# 337##################################################################### 338 339bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>') 340 341bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>') 342bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>') 343 344def substBitOps(code): 345 # first convert single-bit selectors to two-index form 346 # i.e., <n> --> <n:n> 347 code = bitOp1ArgRE.sub(r'<\1:\1>', code) 348 # simple case: selector applied to ID (name) 349 # i.e., foo<a:b> --> bits(foo, a, b) 350 code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code) 351 # if selector is applied to expression (ending in ')'), 352 # we need to search backward for matching '(' 353 match = bitOpExprRE.search(code) 354 while match: 355 exprEnd = match.start() 356 here = exprEnd - 1 357 nestLevel = 1 358 while nestLevel > 0: 359 if code[here] == '(': 360 nestLevel -= 1 361 elif code[here] == ')': 362 nestLevel += 1 363 here -= 1 364 if here < 0: 365 sys.exit("Didn't find '('!") 366 exprStart = here+1 367 newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1], 368 match.group(1), match.group(2)) 369 code = code[:exprStart] + newExpr + code[match.end():] 370 match = bitOpExprRE.search(code) 371 return code 372 373 374##################################################################### 375# 376# Code Parser 377# 378# The remaining code is the support for automatically extracting 379# instruction characteristics from pseudocode. 380# 381##################################################################### 382 383# Force the argument to be a list. Useful for flags, where a caller 384# can specify a singleton flag or a list of flags. Also usful for 385# converting tuples to lists so they can be modified. 386def makeList(arg): 387 if isinstance(arg, list): 388 return arg 389 elif isinstance(arg, tuple): 390 return list(arg) 391 elif not arg: 392 return [] 393 else: 394 return [ arg ] 395 396class Operand(object): 397 '''Base class for operand descriptors. An instance of this class 398 (or actually a class derived from this one) represents a specific 399 operand for a code block (e.g, "Rc.sq" as a dest). Intermediate 400 derived classes encapsulates the traits of a particular operand 401 type (e.g., "32-bit integer register").''' 402 403 def buildReadCode(self, func = None): 404 subst_dict = {"name": self.base_name, 405 "func": func, 406 "reg_idx": self.reg_spec, 407 "ctype": self.ctype} 408 if hasattr(self, 'src_reg_idx'): 409 subst_dict['op_idx'] = self.src_reg_idx 410 code = self.read_code % subst_dict 411 return '%s = %s;\n' % (self.base_name, code) 412 413 def buildWriteCode(self, func = None): 414 subst_dict = {"name": self.base_name, 415 "func": func, 416 "reg_idx": self.reg_spec, 417 "ctype": self.ctype, 418 "final_val": self.base_name} 419 if hasattr(self, 'dest_reg_idx'): 420 subst_dict['op_idx'] = self.dest_reg_idx 421 code = self.write_code % subst_dict 422 return ''' 423 { 424 %s final_val = %s; 425 %s; 426 if (traceData) { traceData->setData(final_val); } 427 }''' % (self.dflt_ctype, self.base_name, code) 428 429 def __init__(self, parser, full_name, ext, is_src, is_dest): 430 self.full_name = full_name 431 self.ext = ext 432 self.is_src = is_src 433 self.is_dest = is_dest 434 # The 'effective extension' (eff_ext) is either the actual 435 # extension, if one was explicitly provided, or the default. 436 if ext: 437 self.eff_ext = ext 438 elif hasattr(self, 'dflt_ext'): 439 self.eff_ext = self.dflt_ext 440 441 if hasattr(self, 'eff_ext'): 442 self.ctype = parser.operandTypeMap[self.eff_ext] 443 444 # Finalize additional fields (primarily code fields). This step 445 # is done separately since some of these fields may depend on the 446 # register index enumeration that hasn't been performed yet at the 447 # time of __init__(). The register index enumeration is affected 448 # by predicated register reads/writes. Hence, we forward the flags 449 # that indicate whether or not predication is in use. 450 def finalize(self, predRead, predWrite): 451 self.flags = self.getFlags() 452 self.constructor = self.makeConstructor(predRead, predWrite) 453 self.op_decl = self.makeDecl() 454 455 if self.is_src: 456 self.op_rd = self.makeRead(predRead) 457 self.op_src_decl = self.makeDecl() 458 else: 459 self.op_rd = '' 460 self.op_src_decl = '' 461 462 if self.is_dest: 463 self.op_wb = self.makeWrite(predWrite) 464 self.op_dest_decl = self.makeDecl() 465 else: 466 self.op_wb = '' 467 self.op_dest_decl = '' 468 469 def isMem(self): 470 return 0 471 472 def isReg(self): 473 return 0 474 475 def isFloatReg(self): 476 return 0 477 478 def isIntReg(self): 479 return 0 480 481 def isCCReg(self): 482 return 0 483 484 def isControlReg(self): 485 return 0 486 487 def isVecReg(self): 488 return 0 489 490 def isVecElem(self): 491 return 0 492 493 def isPCState(self): 494 return 0 495 496 def isPCPart(self): 497 return self.isPCState() and self.reg_spec 498 499 def hasReadPred(self): 500 return self.read_predicate != None 501 502 def hasWritePred(self): 503 return self.write_predicate != None 504 505 def getFlags(self): 506 # note the empty slice '[:]' gives us a copy of self.flags[0] 507 # instead of a reference to it 508 my_flags = self.flags[0][:] 509 if self.is_src: 510 my_flags += self.flags[1] 511 if self.is_dest: 512 my_flags += self.flags[2] 513 return my_flags 514 515 def makeDecl(self): 516 # Note that initializations in the declarations are solely 517 # to avoid 'uninitialized variable' errors from the compiler. 518 return self.ctype + ' ' + self.base_name + ' = 0;\n'; 519 520 521src_reg_constructor = '\n\t_srcRegIdx[_numSrcRegs++] = RegId(%s, %s);' 522dst_reg_constructor = '\n\t_destRegIdx[_numDestRegs++] = RegId(%s, %s);' 523 524 525class IntRegOperand(Operand): 526 reg_class = 'IntRegClass' 527 528 def isReg(self): 529 return 1 530 531 def isIntReg(self): 532 return 1 533 534 def makeConstructor(self, predRead, predWrite): 535 c_src = '' 536 c_dest = '' 537 538 if self.is_src: 539 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 540 if self.hasReadPred(): 541 c_src = '\n\tif (%s) {%s\n\t}' % \ 542 (self.read_predicate, c_src) 543 544 if self.is_dest: 545 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 546 c_dest += '\n\t_numIntDestRegs++;' 547 if self.hasWritePred(): 548 c_dest = '\n\tif (%s) {%s\n\t}' % \ 549 (self.write_predicate, c_dest) 550 551 return c_src + c_dest 552 553 def makeRead(self, predRead): 554 if (self.ctype == 'float' or self.ctype == 'double'): 555 error('Attempt to read integer register as FP') 556 if self.read_code != None: 557 return self.buildReadCode('readIntRegOperand') 558 559 int_reg_val = '' 560 if predRead: 561 int_reg_val = 'xc->readIntRegOperand(this, _sourceIndex++)' 562 if self.hasReadPred(): 563 int_reg_val = '(%s) ? %s : 0' % \ 564 (self.read_predicate, int_reg_val) 565 else: 566 int_reg_val = 'xc->readIntRegOperand(this, %d)' % self.src_reg_idx 567 568 return '%s = %s;\n' % (self.base_name, int_reg_val) 569 570 def makeWrite(self, predWrite): 571 if (self.ctype == 'float' or self.ctype == 'double'): 572 error('Attempt to write integer register as FP') 573 if self.write_code != None: 574 return self.buildWriteCode('setIntRegOperand') 575 576 if predWrite: 577 wp = 'true' 578 if self.hasWritePred(): 579 wp = self.write_predicate 580 581 wcond = 'if (%s)' % (wp) 582 windex = '_destIndex++' 583 else: 584 wcond = '' 585 windex = '%d' % self.dest_reg_idx 586 587 wb = ''' 588 %s 589 { 590 %s final_val = %s; 591 xc->setIntRegOperand(this, %s, final_val);\n 592 if (traceData) { traceData->setData(final_val); } 593 }''' % (wcond, self.ctype, self.base_name, windex) 594 595 return wb 596 597class FloatRegOperand(Operand): 598 reg_class = 'FloatRegClass' 599 600 def isReg(self): 601 return 1 602 603 def isFloatReg(self): 604 return 1 605 606 def makeConstructor(self, predRead, predWrite): 607 c_src = '' 608 c_dest = '' 609 610 if self.is_src: 611 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 612 613 if self.is_dest: 614 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 615 c_dest += '\n\t_numFPDestRegs++;' 616 617 return c_src + c_dest 618 619 def makeRead(self, predRead): 620 if self.read_code != None: 621 return self.buildReadCode('readFloatRegOperandBits') 622 623 if predRead: 624 rindex = '_sourceIndex++' 625 else: 626 rindex = '%d' % self.src_reg_idx 627 628 code = 'xc->readFloatRegOperandBits(this, %s)' % rindex 629 if self.ctype == 'float': 630 code = 'bitsToFloat32(%s)' % code 631 elif self.ctype == 'double': 632 code = 'bitsToFloat64(%s)' % code 633 return '%s = %s;\n' % (self.base_name, code) 634 635 def makeWrite(self, predWrite): 636 if self.write_code != None: 637 return self.buildWriteCode('setFloatRegOperandBits') 638 639 if predWrite: 640 wp = '_destIndex++' 641 else: 642 wp = '%d' % self.dest_reg_idx 643 644 val = 'final_val' 645 if self.ctype == 'float': 646 val = 'floatToBits32(%s)' % val 647 elif self.ctype == 'double': 648 val = 'floatToBits64(%s)' % val 649 650 wp = 'xc->setFloatRegOperandBits(this, %s, %s);' % (wp, val) 651 652 wb = ''' 653 { 654 %s final_val = %s; 655 %s\n 656 if (traceData) { traceData->setData(final_val); } 657 }''' % (self.ctype, self.base_name, wp) 658 return wb 659 660class VecRegOperand(Operand): 661 reg_class = 'VecRegClass' 662 663 def __init__(self, parser, full_name, ext, is_src, is_dest): 664 Operand.__init__(self, parser, full_name, ext, is_src, is_dest) 665 self.elemExt = None 666 self.parser = parser 667 668 def isReg(self): 669 return 1 670 671 def isVecReg(self): 672 return 1 673 674 def makeDeclElem(self, elem_op): 675 (elem_name, elem_ext) = elem_op 676 (elem_spec, dflt_elem_ext, zeroing) = self.elems[elem_name] 677 if elem_ext: 678 ext = elem_ext 679 else: 680 ext = dflt_elem_ext 681 ctype = self.parser.operandTypeMap[ext] 682 return '\n\t%s %s = 0;' % (ctype, elem_name) 683 684 def makeDecl(self): 685 if not self.is_dest and self.is_src: 686 c_decl = '\t/* Vars for %s*/' % (self.base_name) 687 if hasattr(self, 'active_elems'): 688 if self.active_elems: 689 for elem in self.active_elems: 690 c_decl += self.makeDeclElem(elem) 691 return c_decl + '\t/* End vars for %s */\n' % (self.base_name) 692 else: 693 return '' 694 695 def makeConstructor(self, predRead, predWrite): 696 c_src = '' 697 c_dest = '' 698 699 numAccessNeeded = 1 700 701 if self.is_src: 702 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 703 704 if self.is_dest: 705 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 706 c_dest += '\n\t_numVecDestRegs++;' 707 708 return c_src + c_dest 709 710 # Read destination register to write 711 def makeReadWElem(self, elem_op): 712 (elem_name, elem_ext) = elem_op 713 (elem_spec, dflt_elem_ext, zeroing) = self.elems[elem_name] 714 if elem_ext: 715 ext = elem_ext 716 else: 717 ext = dflt_elem_ext 718 ctype = self.parser.operandTypeMap[ext] 719 c_read = '\t\t%s& %s = %s[%s];\n' % \ 720 (ctype, elem_name, self.base_name, elem_spec) 721 return c_read 722 723 def makeReadW(self, predWrite): 724 func = 'getWritableVecRegOperand' 725 if self.read_code != None: 726 return self.buildReadCode(func) 727 728 if predWrite: 729 rindex = '_destIndex++' 730 else: 731 rindex = '%d' % self.dest_reg_idx 732 733 c_readw = '\t\t%s& tmp_d%s = xc->%s(this, %s);\n'\ 734 % ('TheISA::VecRegContainer', rindex, func, rindex) 735 if self.elemExt: 736 c_readw += '\t\tauto %s = tmp_d%s.as<%s>();\n' % (self.base_name, 737 rindex, self.parser.operandTypeMap[self.elemExt]) 738 if self.ext: 739 c_readw += '\t\tauto %s = tmp_d%s.as<%s>();\n' % (self.base_name, 740 rindex, self.parser.operandTypeMap[self.ext]) 741 if hasattr(self, 'active_elems'): 742 if self.active_elems: 743 for elem in self.active_elems: 744 c_readw += self.makeReadWElem(elem) 745 return c_readw 746 747 # Normal source operand read 748 def makeReadElem(self, elem_op, name): 749 (elem_name, elem_ext) = elem_op 750 (elem_spec, dflt_elem_ext, zeroing) = self.elems[elem_name] 751 752 if elem_ext: 753 ext = elem_ext 754 else: 755 ext = dflt_elem_ext 756 ctype = self.parser.operandTypeMap[ext] 757 c_read = '\t\t%s = %s[%s];\n' % \ 758 (elem_name, name, elem_spec) 759 return c_read 760 761 def makeRead(self, predRead): 762 func = 'readVecRegOperand' 763 if self.read_code != None: 764 return self.buildReadCode(func) 765 766 if predRead: 767 rindex = '_sourceIndex++' 768 else: 769 rindex = '%d' % self.src_reg_idx 770 771 name = self.base_name 772 if self.is_dest and self.is_src: 773 name += '_merger' 774 775 c_read = '\t\t%s& tmp_s%s = xc->%s(this, %s);\n' \ 776 % ('const TheISA::VecRegContainer', rindex, func, rindex) 777 # If the parser has detected that elements are being access, create 778 # the appropriate view 779 if self.elemExt: 780 c_read += '\t\tauto %s = tmp_s%s.as<%s>();\n' % \ 781 (name, rindex, self.parser.operandTypeMap[self.elemExt]) 782 if self.ext: 783 c_read += '\t\tauto %s = tmp_s%s.as<%s>();\n' % \ 784 (name, rindex, self.parser.operandTypeMap[self.ext]) 785 if hasattr(self, 'active_elems'): 786 if self.active_elems: 787 for elem in self.active_elems: 788 c_read += self.makeReadElem(elem, name) 789 return c_read 790 791 def makeWrite(self, predWrite): 792 func = 'setVecRegOperand' 793 if self.write_code != None: 794 return self.buildWriteCode(func) 795 796 wb = ''' 797 if (traceData) { 798 warn_once("Vectors not supported yet in tracedata"); 799 /*traceData->setData(final_val);*/ 800 } 801 ''' 802 return wb 803 804 def finalize(self, predRead, predWrite): 805 super(VecRegOperand, self).finalize(predRead, predWrite) 806 if self.is_dest: 807 self.op_rd = self.makeReadW(predWrite) + self.op_rd 808 809class VecElemOperand(Operand): 810 reg_class = 'VecElemClass' 811 812 def isReg(self): 813 return 1 814 815 def isVecElem(self): 816 return 1 817 818 def makeDecl(self): 819 if self.is_dest and not self.is_src: 820 return '\n\t%s %s;' % (self.ctype, self.base_name) 821 else: 822 return '' 823 824 def makeConstructor(self, predRead, predWrite): 825 c_src = '' 826 c_dest = '' 827 828 numAccessNeeded = 1 829 830 if self.is_src: 831 c_src = ('\n\t_srcRegIdx[_numSrcRegs++] = RegId(%s, %s, %s);' % 832 (self.reg_class, self.reg_spec, self.elem_spec)) 833 834 if self.is_dest: 835 c_dest = ('\n\t_destRegIdx[_numDestRegs++] = RegId(%s, %s, %s);' % 836 (self.reg_class, self.reg_spec, self.elem_spec)) 837 c_dest += '\n\t_numVecElemDestRegs++;' 838 return c_src + c_dest 839 840 def makeRead(self, predRead): 841 c_read = 'xc->readVecElemOperand(this, %d)' % self.src_reg_idx 842 843 if self.ctype == 'float': 844 c_read = 'bitsToFloat32(%s)' % c_read 845 elif self.ctype == 'double': 846 c_read = 'bitsToFloat64(%s)' % c_read 847 848 return '\n\t%s %s = %s;\n' % (self.ctype, self.base_name, c_read) 849 850 def makeWrite(self, predWrite): 851 if self.ctype == 'float': 852 c_write = 'floatToBits32(%s)' % self.base_name 853 elif self.ctype == 'double': 854 c_write = 'floatToBits64(%s)' % self.base_name 855 else: 856 c_write = self.base_name 857 858 c_write = ('\n\txc->setVecElemOperand(this, %d, %s);' % 859 (self.dest_reg_idx, c_write)) 860 861 return c_write 862 863class CCRegOperand(Operand): 864 reg_class = 'CCRegClass' 865 866 def isReg(self): 867 return 1 868 869 def isCCReg(self): 870 return 1 871 872 def makeConstructor(self, predRead, predWrite): 873 c_src = '' 874 c_dest = '' 875 876 if self.is_src: 877 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 878 if self.hasReadPred(): 879 c_src = '\n\tif (%s) {%s\n\t}' % \ 880 (self.read_predicate, c_src) 881 882 if self.is_dest: 883 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 884 c_dest += '\n\t_numCCDestRegs++;' 885 if self.hasWritePred(): 886 c_dest = '\n\tif (%s) {%s\n\t}' % \ 887 (self.write_predicate, c_dest) 888 889 return c_src + c_dest 890 891 def makeRead(self, predRead): 892 if (self.ctype == 'float' or self.ctype == 'double'): 893 error('Attempt to read condition-code register as FP') 894 if self.read_code != None: 895 return self.buildReadCode('readCCRegOperand') 896 897 int_reg_val = '' 898 if predRead: 899 int_reg_val = 'xc->readCCRegOperand(this, _sourceIndex++)' 900 if self.hasReadPred(): 901 int_reg_val = '(%s) ? %s : 0' % \ 902 (self.read_predicate, int_reg_val) 903 else: 904 int_reg_val = 'xc->readCCRegOperand(this, %d)' % self.src_reg_idx 905 906 return '%s = %s;\n' % (self.base_name, int_reg_val) 907 908 def makeWrite(self, predWrite): 909 if (self.ctype == 'float' or self.ctype == 'double'): 910 error('Attempt to write condition-code register as FP') 911 if self.write_code != None: 912 return self.buildWriteCode('setCCRegOperand') 913 914 if predWrite: 915 wp = 'true' 916 if self.hasWritePred(): 917 wp = self.write_predicate 918 919 wcond = 'if (%s)' % (wp) 920 windex = '_destIndex++' 921 else: 922 wcond = '' 923 windex = '%d' % self.dest_reg_idx 924 925 wb = ''' 926 %s 927 { 928 %s final_val = %s; 929 xc->setCCRegOperand(this, %s, final_val);\n 930 if (traceData) { traceData->setData(final_val); } 931 }''' % (wcond, self.ctype, self.base_name, windex) 932 933 return wb 934 935class ControlRegOperand(Operand): 936 reg_class = 'MiscRegClass' 937 938 def isReg(self): 939 return 1 940 941 def isControlReg(self): 942 return 1 943 944 def makeConstructor(self, predRead, predWrite): 945 c_src = '' 946 c_dest = '' 947 948 if self.is_src: 949 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 950 951 if self.is_dest: 952 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 953 954 return c_src + c_dest 955 956 def makeRead(self, predRead): 957 bit_select = 0 958 if (self.ctype == 'float' or self.ctype == 'double'): 959 error('Attempt to read control register as FP') 960 if self.read_code != None: 961 return self.buildReadCode('readMiscRegOperand') 962 963 if predRead: 964 rindex = '_sourceIndex++' 965 else: 966 rindex = '%d' % self.src_reg_idx 967 968 return '%s = xc->readMiscRegOperand(this, %s);\n' % \ 969 (self.base_name, rindex) 970 971 def makeWrite(self, predWrite): 972 if (self.ctype == 'float' or self.ctype == 'double'): 973 error('Attempt to write control register as FP') 974 if self.write_code != None: 975 return self.buildWriteCode('setMiscRegOperand') 976 977 if predWrite: 978 windex = '_destIndex++' 979 else: 980 windex = '%d' % self.dest_reg_idx 981 982 wb = 'xc->setMiscRegOperand(this, %s, %s);\n' % \ 983 (windex, self.base_name) 984 wb += 'if (traceData) { traceData->setData(%s); }' % \ 985 self.base_name 986 987 return wb 988 989class MemOperand(Operand): 990 def isMem(self): 991 return 1 992 993 def makeConstructor(self, predRead, predWrite): 994 return '' 995 996 def makeDecl(self): 997 # Declare memory data variable. 998 return '%s %s;\n' % (self.ctype, self.base_name) 999 1000 def makeRead(self, predRead): 1001 if self.read_code != None: 1002 return self.buildReadCode() 1003 return '' 1004 1005 def makeWrite(self, predWrite): 1006 if self.write_code != None: 1007 return self.buildWriteCode() 1008 return '' 1009 1010class PCStateOperand(Operand): 1011 def makeConstructor(self, predRead, predWrite): 1012 return '' 1013 1014 def makeRead(self, predRead): 1015 if self.reg_spec: 1016 # A component of the PC state. 1017 return '%s = __parserAutoPCState.%s();\n' % \ 1018 (self.base_name, self.reg_spec) 1019 else: 1020 # The whole PC state itself. 1021 return '%s = xc->pcState();\n' % self.base_name 1022 1023 def makeWrite(self, predWrite): 1024 if self.reg_spec: 1025 # A component of the PC state. 1026 return '__parserAutoPCState.%s(%s);\n' % \ 1027 (self.reg_spec, self.base_name) 1028 else: 1029 # The whole PC state itself. 1030 return 'xc->pcState(%s);\n' % self.base_name 1031 1032 def makeDecl(self): 1033 ctype = 'TheISA::PCState' 1034 if self.isPCPart(): 1035 ctype = self.ctype 1036 # Note that initializations in the declarations are solely 1037 # to avoid 'uninitialized variable' errors from the compiler. 1038 return '%s %s = 0;\n' % (ctype, self.base_name) 1039 1040 def isPCState(self): 1041 return 1 1042 1043class OperandList(object): 1044 '''Find all the operands in the given code block. Returns an operand 1045 descriptor list (instance of class OperandList).''' 1046 def __init__(self, parser, code): 1047 self.items = [] 1048 self.bases = {} 1049 # delete strings and comments so we don't match on operands inside 1050 for regEx in (stringRE, commentRE): 1051 code = regEx.sub('', code) 1052 # search for operands 1053 next_pos = 0 1054 while 1: 1055 match = parser.operandsRE.search(code, next_pos) 1056 if not match: 1057 # no more matches: we're done 1058 break 1059 op = match.groups() 1060 # regexp groups are operand full name, base, and extension 1061 (op_full, op_base, op_ext) = op 1062 # If is a elem operand, define or update the corresponding 1063 # vector operand 1064 isElem = False 1065 if op_base in parser.elemToVector: 1066 isElem = True 1067 elem_op = (op_base, op_ext) 1068 op_base = parser.elemToVector[op_base] 1069 op_ext = '' # use the default one 1070 # if the token following the operand is an assignment, this is 1071 # a destination (LHS), else it's a source (RHS) 1072 is_dest = (assignRE.match(code, match.end()) != None) 1073 is_src = not is_dest 1074 1075 # see if we've already seen this one 1076 op_desc = self.find_base(op_base) 1077 if op_desc: 1078 if op_ext and op_ext != '' and op_desc.ext != op_ext: 1079 error ('Inconsistent extensions for operand %s: %s - %s' \ 1080 % (op_base, op_desc.ext, op_ext)) 1081 op_desc.is_src = op_desc.is_src or is_src 1082 op_desc.is_dest = op_desc.is_dest or is_dest 1083 if isElem: 1084 (elem_base, elem_ext) = elem_op 1085 found = False 1086 for ae in op_desc.active_elems: 1087 (ae_base, ae_ext) = ae 1088 if ae_base == elem_base: 1089 if ae_ext != elem_ext: 1090 error('Inconsistent extensions for elem' 1091 ' operand %s' % elem_base) 1092 else: 1093 found = True 1094 if not found: 1095 op_desc.active_elems.append(elem_op) 1096 else: 1097 # new operand: create new descriptor 1098 op_desc = parser.operandNameMap[op_base](parser, 1099 op_full, op_ext, is_src, is_dest) 1100 # if operand is a vector elem, add the corresponding vector 1101 # operand if not already done 1102 if isElem: 1103 op_desc.elemExt = elem_op[1] 1104 op_desc.active_elems = [elem_op] 1105 self.append(op_desc) 1106 # start next search after end of current match 1107 next_pos = match.end() 1108 self.sort() 1109 # enumerate source & dest register operands... used in building 1110 # constructor later 1111 self.numSrcRegs = 0 1112 self.numDestRegs = 0 1113 self.numFPDestRegs = 0 1114 self.numIntDestRegs = 0 1115 self.numVecDestRegs = 0 1116 self.numCCDestRegs = 0 1117 self.numMiscDestRegs = 0 1118 self.memOperand = None 1119 1120 # Flags to keep track if one or more operands are to be read/written 1121 # conditionally. 1122 self.predRead = False 1123 self.predWrite = False 1124 1125 for op_desc in self.items: 1126 if op_desc.isReg(): 1127 if op_desc.is_src: 1128 op_desc.src_reg_idx = self.numSrcRegs 1129 self.numSrcRegs += 1 1130 if op_desc.is_dest: 1131 op_desc.dest_reg_idx = self.numDestRegs 1132 self.numDestRegs += 1 1133 if op_desc.isFloatReg(): 1134 self.numFPDestRegs += 1 1135 elif op_desc.isIntReg(): 1136 self.numIntDestRegs += 1 1137 elif op_desc.isVecReg(): 1138 self.numVecDestRegs += 1 1139 elif op_desc.isCCReg(): 1140 self.numCCDestRegs += 1 1141 elif op_desc.isControlReg(): 1142 self.numMiscDestRegs += 1 1143 elif op_desc.isMem(): 1144 if self.memOperand: 1145 error("Code block has more than one memory operand.") 1146 self.memOperand = op_desc 1147 1148 # Check if this operand has read/write predication. If true, then 1149 # the microop will dynamically index source/dest registers. 1150 self.predRead = self.predRead or op_desc.hasReadPred() 1151 self.predWrite = self.predWrite or op_desc.hasWritePred() 1152 1153 if parser.maxInstSrcRegs < self.numSrcRegs: 1154 parser.maxInstSrcRegs = self.numSrcRegs 1155 if parser.maxInstDestRegs < self.numDestRegs: 1156 parser.maxInstDestRegs = self.numDestRegs 1157 if parser.maxMiscDestRegs < self.numMiscDestRegs: 1158 parser.maxMiscDestRegs = self.numMiscDestRegs 1159 1160 # now make a final pass to finalize op_desc fields that may depend 1161 # on the register enumeration 1162 for op_desc in self.items: 1163 op_desc.finalize(self.predRead, self.predWrite) 1164 1165 def __len__(self): 1166 return len(self.items) 1167 1168 def __getitem__(self, index): 1169 return self.items[index] 1170 1171 def append(self, op_desc): 1172 self.items.append(op_desc) 1173 self.bases[op_desc.base_name] = op_desc 1174 1175 def find_base(self, base_name): 1176 # like self.bases[base_name], but returns None if not found 1177 # (rather than raising exception) 1178 return self.bases.get(base_name) 1179 1180 # internal helper function for concat[Some]Attr{Strings|Lists} 1181 def __internalConcatAttrs(self, attr_name, filter, result): 1182 for op_desc in self.items: 1183 if filter(op_desc): 1184 result += getattr(op_desc, attr_name) 1185 return result 1186 1187 # return a single string that is the concatenation of the (string) 1188 # values of the specified attribute for all operands 1189 def concatAttrStrings(self, attr_name): 1190 return self.__internalConcatAttrs(attr_name, lambda x: 1, '') 1191 1192 # like concatAttrStrings, but only include the values for the operands 1193 # for which the provided filter function returns true 1194 def concatSomeAttrStrings(self, filter, attr_name): 1195 return self.__internalConcatAttrs(attr_name, filter, '') 1196 1197 # return a single list that is the concatenation of the (list) 1198 # values of the specified attribute for all operands 1199 def concatAttrLists(self, attr_name): 1200 return self.__internalConcatAttrs(attr_name, lambda x: 1, []) 1201 1202 # like concatAttrLists, but only include the values for the operands 1203 # for which the provided filter function returns true 1204 def concatSomeAttrLists(self, filter, attr_name): 1205 return self.__internalConcatAttrs(attr_name, filter, []) 1206 1207 def sort(self): 1208 self.items.sort(lambda a, b: a.sort_pri - b.sort_pri) 1209 1210class SubOperandList(OperandList): 1211 '''Find all the operands in the given code block. Returns an operand 1212 descriptor list (instance of class OperandList).''' 1213 def __init__(self, parser, code, master_list): 1214 self.items = [] 1215 self.bases = {} 1216 # delete strings and comments so we don't match on operands inside 1217 for regEx in (stringRE, commentRE): 1218 code = regEx.sub('', code) 1219 # search for operands 1220 next_pos = 0 1221 while 1: 1222 match = parser.operandsRE.search(code, next_pos) 1223 if not match: 1224 # no more matches: we're done 1225 break 1226 op = match.groups() 1227 # regexp groups are operand full name, base, and extension 1228 (op_full, op_base, op_ext) = op 1229 # If is a elem operand, define or update the corresponding 1230 # vector operand 1231 if op_base in parser.elemToVector: 1232 elem_op = op_base 1233 op_base = parser.elemToVector[elem_op] 1234 # find this op in the master list 1235 op_desc = master_list.find_base(op_base) 1236 if not op_desc: 1237 error('Found operand %s which is not in the master list!' 1238 % op_base) 1239 else: 1240 # See if we've already found this operand 1241 op_desc = self.find_base(op_base) 1242 if not op_desc: 1243 # if not, add a reference to it to this sub list 1244 self.append(master_list.bases[op_base]) 1245 1246 # start next search after end of current match 1247 next_pos = match.end() 1248 self.sort() 1249 self.memOperand = None 1250 # Whether the whole PC needs to be read so parts of it can be accessed 1251 self.readPC = False 1252 # Whether the whole PC needs to be written after parts of it were 1253 # changed 1254 self.setPC = False 1255 # Whether this instruction manipulates the whole PC or parts of it. 1256 # Mixing the two is a bad idea and flagged as an error. 1257 self.pcPart = None 1258 1259 # Flags to keep track if one or more operands are to be read/written 1260 # conditionally. 1261 self.predRead = False 1262 self.predWrite = False 1263 1264 for op_desc in self.items: 1265 if op_desc.isPCPart(): 1266 self.readPC = True 1267 if op_desc.is_dest: 1268 self.setPC = True 1269 1270 if op_desc.isPCState(): 1271 if self.pcPart is not None: 1272 if self.pcPart and not op_desc.isPCPart() or \ 1273 not self.pcPart and op_desc.isPCPart(): 1274 error("Mixed whole and partial PC state operands.") 1275 self.pcPart = op_desc.isPCPart() 1276 1277 if op_desc.isMem(): 1278 if self.memOperand: 1279 error("Code block has more than one memory operand.") 1280 self.memOperand = op_desc 1281 1282 # Check if this operand has read/write predication. If true, then 1283 # the microop will dynamically index source/dest registers. 1284 self.predRead = self.predRead or op_desc.hasReadPred() 1285 self.predWrite = self.predWrite or op_desc.hasWritePred() 1286 1287# Regular expression object to match C++ strings 1288stringRE = re.compile(r'"([^"\\]|\\.)*"') 1289 1290# Regular expression object to match C++ comments 1291# (used in findOperands()) 1292commentRE = re.compile(r'(^)?[^\S\n]*/(?:\*(.*?)\*/[^\S\n]*|/[^\n]*)($)?', 1293 re.DOTALL | re.MULTILINE) 1294 1295# Regular expression object to match assignment statements (used in 1296# findOperands()). If the code immediately following the first 1297# appearance of the operand matches this regex, then the operand 1298# appears to be on the LHS of an assignment, and is thus a 1299# destination. basically we're looking for an '=' that's not '=='. 1300# The heinous tangle before that handles the case where the operand 1301# has an array subscript. 1302assignRE = re.compile(r'(\[[^\]]+\])?\s*=(?!=)', re.MULTILINE) 1303 1304def makeFlagConstructor(flag_list): 1305 if len(flag_list) == 0: 1306 return '' 1307 # filter out repeated flags 1308 flag_list.sort() 1309 i = 1 1310 while i < len(flag_list): 1311 if flag_list[i] == flag_list[i-1]: 1312 del flag_list[i] 1313 else: 1314 i += 1 1315 pre = '\n\tflags[' 1316 post = '] = true;' 1317 code = pre + string.join(flag_list, post + pre) + post 1318 return code 1319 1320# Assume all instruction flags are of the form 'IsFoo' 1321instFlagRE = re.compile(r'Is.*') 1322 1323# OpClass constants end in 'Op' except No_OpClass 1324opClassRE = re.compile(r'.*Op|No_OpClass') 1325 1326class InstObjParams(object): 1327 def __init__(self, parser, mnem, class_name, base_class = '', 1328 snippets = {}, opt_args = []): 1329 self.mnemonic = mnem 1330 self.class_name = class_name 1331 self.base_class = base_class 1332 if not isinstance(snippets, dict): 1333 snippets = {'code' : snippets} 1334 compositeCode = ' '.join(map(str, snippets.values())) 1335 self.snippets = snippets 1336 1337 self.operands = OperandList(parser, compositeCode) 1338 1339 # The header of the constructor declares the variables to be used 1340 # in the body of the constructor. 1341 header = '' 1342 header += '\n\t_numSrcRegs = 0;' 1343 header += '\n\t_numDestRegs = 0;' 1344 header += '\n\t_numFPDestRegs = 0;' 1345 header += '\n\t_numVecDestRegs = 0;' 1346 header += '\n\t_numVecElemDestRegs = 0;' 1347 header += '\n\t_numIntDestRegs = 0;' 1348 header += '\n\t_numCCDestRegs = 0;' 1349 1350 self.constructor = header + \ 1351 self.operands.concatAttrStrings('constructor') 1352 1353 self.flags = self.operands.concatAttrLists('flags') 1354 1355 self.op_class = None 1356 1357 # Optional arguments are assumed to be either StaticInst flags 1358 # or an OpClass value. To avoid having to import a complete 1359 # list of these values to match against, we do it ad-hoc 1360 # with regexps. 1361 for oa in opt_args: 1362 if instFlagRE.match(oa): 1363 self.flags.append(oa) 1364 elif opClassRE.match(oa): 1365 self.op_class = oa 1366 else: 1367 error('InstObjParams: optional arg "%s" not recognized ' 1368 'as StaticInst::Flag or OpClass.' % oa) 1369 1370 # Make a basic guess on the operand class if not set. 1371 # These are good enough for most cases. 1372 if not self.op_class: 1373 if 'IsStore' in self.flags: 1374 # The order matters here: 'IsFloating' and 'IsInteger' are 1375 # usually set in FP instructions because of the base 1376 # register 1377 if 'IsFloating' in self.flags: 1378 self.op_class = 'FloatMemWriteOp' 1379 else: 1380 self.op_class = 'MemWriteOp' 1381 elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags: 1382 # The order matters here: 'IsFloating' and 'IsInteger' are 1383 # usually set in FP instructions because of the base 1384 # register 1385 if 'IsFloating' in self.flags: 1386 self.op_class = 'FloatMemReadOp' 1387 else: 1388 self.op_class = 'MemReadOp' 1389 elif 'IsFloating' in self.flags: 1390 self.op_class = 'FloatAddOp' 1391 elif 'IsVector' in self.flags: 1392 self.op_class = 'SimdAddOp' 1393 else: 1394 self.op_class = 'IntAluOp' 1395 1396 # add flag initialization to contructor here to include 1397 # any flags added via opt_args 1398 self.constructor += makeFlagConstructor(self.flags) 1399 1400 # if 'IsFloating' is set, add call to the FP enable check 1401 # function (which should be provided by isa_desc via a declare) 1402 # if 'IsVector' is set, add call to the Vector enable check 1403 # function (which should be provided by isa_desc via a declare) 1404 if 'IsFloating' in self.flags: 1405 self.fp_enable_check = 'fault = checkFpEnableFault(xc);' 1406 elif 'IsVector' in self.flags: 1407 self.fp_enable_check = 'fault = checkVecEnableFault(xc);' 1408 else: 1409 self.fp_enable_check = '' 1410 1411############## 1412# Stack: a simple stack object. Used for both formats (formatStack) 1413# and default cases (defaultStack). Simply wraps a list to give more 1414# stack-like syntax and enable initialization with an argument list 1415# (as opposed to an argument that's a list). 1416 1417class Stack(list): 1418 def __init__(self, *items): 1419 list.__init__(self, items) 1420 1421 def push(self, item): 1422 self.append(item); 1423 1424 def top(self): 1425 return self[-1] 1426 1427# Format a file include stack backtrace as a string 1428def backtrace(filename_stack): 1429 fmt = "In file included from %s:" 1430 return "\n".join([fmt % f for f in filename_stack]) 1431 1432 1433####################### 1434# 1435# LineTracker: track filenames along with line numbers in PLY lineno fields 1436# PLY explicitly doesn't do anything with 'lineno' except propagate 1437# it. This class lets us tie filenames with the line numbers with a 1438# minimum of disruption to existing increment code. 1439# 1440 1441class LineTracker(object): 1442 def __init__(self, filename, lineno=1): 1443 self.filename = filename 1444 self.lineno = lineno 1445 1446 # Overload '+=' for increments. We need to create a new object on 1447 # each update else every token ends up referencing the same 1448 # constantly incrementing instance. 1449 def __iadd__(self, incr): 1450 return LineTracker(self.filename, self.lineno + incr) 1451 1452 def __str__(self): 1453 return "%s:%d" % (self.filename, self.lineno) 1454 1455 # In case there are places where someone really expects a number 1456 def __int__(self): 1457 return self.lineno 1458 1459 1460####################### 1461# 1462# ISA Parser 1463# parses ISA DSL and emits C++ headers and source 1464# 1465 1466class ISAParser(Grammar): 1467 def __init__(self, output_dir): 1468 super(ISAParser, self).__init__() 1469 self.output_dir = output_dir 1470 1471 self.filename = None # for output file watermarking/scaremongering 1472 1473 # variable to hold templates 1474 self.templateMap = {} 1475 1476 # This dictionary maps format name strings to Format objects. 1477 self.formatMap = {} 1478 1479 # Track open files and, if applicable, how many chunks it has been 1480 # split into so far. 1481 self.files = {} 1482 self.splits = {} 1483 1484 # isa_name / namespace identifier from namespace declaration. 1485 # before the namespace declaration, None. 1486 self.isa_name = None 1487 self.namespace = None 1488 1489 # The format stack. 1490 self.formatStack = Stack(NoFormat()) 1491 1492 # The default case stack. 1493 self.defaultStack = Stack(None) 1494 1495 # Stack that tracks current file and line number. Each 1496 # element is a tuple (filename, lineno) that records the 1497 # *current* filename and the line number in the *previous* 1498 # file where it was included. 1499 self.fileNameStack = Stack() 1500 1501 symbols = ('makeList', 're', 'string') 1502 self.exportContext = dict([(s, eval(s)) for s in symbols]) 1503 1504 self.maxInstSrcRegs = 0 1505 self.maxInstDestRegs = 0 1506 self.maxMiscDestRegs = 0 1507 1508 def __getitem__(self, i): # Allow object (self) to be 1509 return getattr(self, i) # passed to %-substitutions 1510 1511 # Change the file suffix of a base filename: 1512 # (e.g.) decoder.cc -> decoder-g.cc.inc for 'global' outputs 1513 def suffixize(self, s, sec): 1514 extn = re.compile('(\.[^\.]+)$') # isolate extension 1515 if self.namespace: 1516 return extn.sub(r'-ns\1.inc', s) # insert some text on either side 1517 else: 1518 return extn.sub(r'-g\1.inc', s) 1519 1520 # Get the file object for emitting code into the specified section 1521 # (header, decoder, exec, decode_block). 1522 def get_file(self, section): 1523 if section == 'decode_block': 1524 filename = 'decode-method.cc.inc' 1525 else: 1526 if section == 'header': 1527 file = 'decoder.hh' 1528 else: 1529 file = '%s.cc' % section 1530 filename = self.suffixize(file, section) 1531 try: 1532 return self.files[filename] 1533 except KeyError: pass 1534 1535 f = self.open(filename) 1536 self.files[filename] = f 1537 1538 # The splittable files are the ones with many independent 1539 # per-instruction functions - the decoder's instruction constructors 1540 # and the instruction execution (execute()) methods. These both have 1541 # the suffix -ns.cc.inc, meaning they are within the namespace part 1542 # of the ISA, contain object-emitting C++ source, and are included 1543 # into other top-level files. These are the files that need special 1544 # #define's to allow parts of them to be compiled separately. Rather 1545 # than splitting the emissions into separate files, the monolithic 1546 # output of the ISA parser is maintained, but the value (or lack 1547 # thereof) of the __SPLIT definition during C preprocessing will 1548 # select the different chunks. If no 'split' directives are used, 1549 # the cpp emissions have no effect. 1550 if re.search('-ns.cc.inc$', filename): 1551 print('#if !defined(__SPLIT) || (__SPLIT == 1)', file=f) 1552 self.splits[f] = 1 1553 # ensure requisite #include's 1554 elif filename == 'decoder-g.hh.inc': 1555 print('#include "base/bitfield.hh"', file=f) 1556 1557 return f 1558 1559 # Weave together the parts of the different output sections by 1560 # #include'ing them into some very short top-level .cc/.hh files. 1561 # These small files make it much clearer how this tool works, since 1562 # you directly see the chunks emitted as files that are #include'd. 1563 def write_top_level_files(self): 1564 # decoder header - everything depends on this 1565 file = 'decoder.hh' 1566 with self.open(file) as f: 1567 fn = 'decoder-g.hh.inc' 1568 assert(fn in self.files) 1569 f.write('#include "%s"\n' % fn) 1570 1571 fn = 'decoder-ns.hh.inc' 1572 assert(fn in self.files) 1573 f.write('namespace %s {\n#include "%s"\n}\n' 1574 % (self.namespace, fn)) 1575 1576 # decoder method - cannot be split 1577 file = 'decoder.cc' 1578 with self.open(file) as f: 1579 fn = 'base/compiler.hh' 1580 f.write('#include "%s"\n' % fn) 1581 1582 fn = 'decoder-g.cc.inc' 1583 assert(fn in self.files) 1584 f.write('#include "%s"\n' % fn) 1585 1586 fn = 'decoder.hh' 1587 f.write('#include "%s"\n' % fn) 1588 1589 fn = 'decode-method.cc.inc' 1590 # is guaranteed to have been written for parse to complete 1591 f.write('#include "%s"\n' % fn) 1592 1593 extn = re.compile('(\.[^\.]+)$') 1594 1595 # instruction constructors 1596 splits = self.splits[self.get_file('decoder')] 1597 file_ = 'inst-constrs.cc' 1598 for i in range(1, splits+1): 1599 if splits > 1: 1600 file = extn.sub(r'-%d\1' % i, file_) 1601 else: 1602 file = file_ 1603 with self.open(file) as f: 1604 fn = 'decoder-g.cc.inc' 1605 assert(fn in self.files) 1606 f.write('#include "%s"\n' % fn) 1607 1608 fn = 'decoder.hh' 1609 f.write('#include "%s"\n' % fn) 1610 1611 fn = 'decoder-ns.cc.inc' 1612 assert(fn in self.files) 1613 print('namespace %s {' % self.namespace, file=f) 1614 if splits > 1: 1615 print('#define __SPLIT %u' % i, file=f) 1616 print('#include "%s"' % fn, file=f) 1617 print('}', file=f) 1618 1619 # instruction execution 1620 splits = self.splits[self.get_file('exec')] 1621 for i in range(1, splits+1): 1622 file = 'generic_cpu_exec.cc' 1623 if splits > 1: 1624 file = extn.sub(r'_%d\1' % i, file) 1625 with self.open(file) as f: 1626 fn = 'exec-g.cc.inc' 1627 assert(fn in self.files) 1628 f.write('#include "%s"\n' % fn) 1629 f.write('#include "cpu/exec_context.hh"\n') 1630 f.write('#include "decoder.hh"\n') 1631 1632 fn = 'exec-ns.cc.inc' 1633 assert(fn in self.files) 1634 print('namespace %s {' % self.namespace, file=f) 1635 if splits > 1: 1636 print('#define __SPLIT %u' % i, file=f) 1637 print('#include "%s"' % fn, file=f) 1638 print('}', file=f) 1639 1640 # max_inst_regs.hh 1641 self.update('max_inst_regs.hh', 1642 '''namespace %(namespace)s { 1643 const int MaxInstSrcRegs = %(maxInstSrcRegs)d; 1644 const int MaxInstDestRegs = %(maxInstDestRegs)d; 1645 const int MaxMiscDestRegs = %(maxMiscDestRegs)d;\n}\n''' % self) 1646 1647 scaremonger_template ='''// DO NOT EDIT 1648// This file was automatically generated from an ISA description: 1649// %(filename)s 1650 1651'''; 1652 1653 ##################################################################### 1654 # 1655 # Lexer 1656 # 1657 # The PLY lexer module takes two things as input: 1658 # - A list of token names (the string list 'tokens') 1659 # - A regular expression describing a match for each token. The 1660 # regexp for token FOO can be provided in two ways: 1661 # - as a string variable named t_FOO 1662 # - as the doc string for a function named t_FOO. In this case, 1663 # the function is also executed, allowing an action to be 1664 # associated with each token match. 1665 # 1666 ##################################################################### 1667 1668 # Reserved words. These are listed separately as they are matched 1669 # using the same regexp as generic IDs, but distinguished in the 1670 # t_ID() function. The PLY documentation suggests this approach. 1671 reserved = ( 1672 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT', 1673 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS', 1674 'OUTPUT', 'SIGNED', 'SPLIT', 'TEMPLATE' 1675 ) 1676 1677 # List of tokens. The lex module requires this. 1678 tokens = reserved + ( 1679 # identifier 1680 'ID', 1681 1682 # integer literal 1683 'INTLIT', 1684 1685 # string literal 1686 'STRLIT', 1687 1688 # code literal 1689 'CODELIT', 1690 1691 # ( ) [ ] { } < > , ; . : :: * 1692 'LPAREN', 'RPAREN', 1693 'LBRACKET', 'RBRACKET', 1694 'LBRACE', 'RBRACE', 1695 'LESS', 'GREATER', 'EQUALS', 1696 'COMMA', 'SEMI', 'DOT', 'COLON', 'DBLCOLON', 1697 'ASTERISK', 1698 1699 # C preprocessor directives 1700 'CPPDIRECTIVE' 1701 1702 # The following are matched but never returned. commented out to 1703 # suppress PLY warning 1704 # newfile directive 1705 # 'NEWFILE', 1706 1707 # endfile directive 1708 # 'ENDFILE' 1709 ) 1710 1711 # Regular expressions for token matching 1712 t_LPAREN = r'\(' 1713 t_RPAREN = r'\)' 1714 t_LBRACKET = r'\[' 1715 t_RBRACKET = r'\]' 1716 t_LBRACE = r'\{' 1717 t_RBRACE = r'\}' 1718 t_LESS = r'\<' 1719 t_GREATER = r'\>' 1720 t_EQUALS = r'=' 1721 t_COMMA = r',' 1722 t_SEMI = r';' 1723 t_DOT = r'\.' 1724 t_COLON = r':' 1725 t_DBLCOLON = r'::' 1726 t_ASTERISK = r'\*' 1727 1728 # Identifiers and reserved words 1729 reserved_map = { } 1730 for r in reserved: 1731 reserved_map[r.lower()] = r 1732 1733 def t_ID(self, t): 1734 r'[A-Za-z_]\w*' 1735 t.type = self.reserved_map.get(t.value, 'ID') 1736 return t 1737 1738 # Integer literal 1739 def t_INTLIT(self, t): 1740 r'-?(0x[\da-fA-F]+)|\d+' 1741 try: 1742 t.value = int(t.value,0) 1743 except ValueError: 1744 error(t.lexer.lineno, 'Integer value "%s" too large' % t.value) 1745 t.value = 0 1746 return t 1747 1748 # String literal. Note that these use only single quotes, and 1749 # can span multiple lines. 1750 def t_STRLIT(self, t): 1751 r"(?m)'([^'])+'" 1752 # strip off quotes 1753 t.value = t.value[1:-1] 1754 t.lexer.lineno += t.value.count('\n') 1755 return t 1756 1757 1758 # "Code literal"... like a string literal, but delimiters are 1759 # '{{' and '}}' so they get formatted nicely under emacs c-mode 1760 def t_CODELIT(self, t): 1761 r"(?m)\{\{([^\}]|}(?!\}))+\}\}" 1762 # strip off {{ & }} 1763 t.value = t.value[2:-2] 1764 t.lexer.lineno += t.value.count('\n') 1765 return t 1766 1767 def t_CPPDIRECTIVE(self, t): 1768 r'^\#[^\#].*\n' 1769 t.lexer.lineno += t.value.count('\n') 1770 return t 1771 1772 def t_NEWFILE(self, t): 1773 r'^\#\#newfile\s+"[^"]*"\n' 1774 self.fileNameStack.push(t.lexer.lineno) 1775 t.lexer.lineno = LineTracker(t.value[11:-2]) 1776 1777 def t_ENDFILE(self, t): 1778 r'^\#\#endfile\n' 1779 t.lexer.lineno = self.fileNameStack.pop() 1780 1781 # 1782 # The functions t_NEWLINE, t_ignore, and t_error are 1783 # special for the lex module. 1784 # 1785 1786 # Newlines 1787 def t_NEWLINE(self, t): 1788 r'\n+' 1789 t.lexer.lineno += t.value.count('\n') 1790 1791 # Comments 1792 def t_comment(self, t): 1793 r'//.*' 1794 1795 # Completely ignored characters 1796 t_ignore = ' \t\x0c' 1797 1798 # Error handler 1799 def t_error(self, t): 1800 error(t.lexer.lineno, "illegal character '%s'" % t.value[0]) 1801 t.skip(1) 1802 1803 ##################################################################### 1804 # 1805 # Parser 1806 # 1807 # Every function whose name starts with 'p_' defines a grammar 1808 # rule. The rule is encoded in the function's doc string, while 1809 # the function body provides the action taken when the rule is 1810 # matched. The argument to each function is a list of the values 1811 # of the rule's symbols: t[0] for the LHS, and t[1..n] for the 1812 # symbols on the RHS. For tokens, the value is copied from the 1813 # t.value attribute provided by the lexer. For non-terminals, the 1814 # value is assigned by the producing rule; i.e., the job of the 1815 # grammar rule function is to set the value for the non-terminal 1816 # on the LHS (by assigning to t[0]). 1817 ##################################################################### 1818 1819 # The LHS of the first grammar rule is used as the start symbol 1820 # (in this case, 'specification'). Note that this rule enforces 1821 # that there will be exactly one namespace declaration, with 0 or 1822 # more global defs/decls before and after it. The defs & decls 1823 # before the namespace decl will be outside the namespace; those 1824 # after will be inside. The decoder function is always inside the 1825 # namespace. 1826 def p_specification(self, t): 1827 'specification : opt_defs_and_outputs top_level_decode_block' 1828 1829 for f in self.splits.iterkeys(): 1830 f.write('\n#endif\n') 1831 1832 for f in self.files.itervalues(): # close ALL the files; 1833 f.close() # not doing so can cause compilation to fail 1834 1835 self.write_top_level_files() 1836 1837 t[0] = True 1838 1839 # 'opt_defs_and_outputs' is a possibly empty sequence of def and/or 1840 # output statements. Its productions do the hard work of eventually 1841 # instantiating a GenCode, which are generally emitted (written to disk) 1842 # as soon as possible, except for the decode_block, which has to be 1843 # accumulated into one large function of nested switch/case blocks. 1844 def p_opt_defs_and_outputs_0(self, t): 1845 'opt_defs_and_outputs : empty' 1846 1847 def p_opt_defs_and_outputs_1(self, t): 1848 'opt_defs_and_outputs : defs_and_outputs' 1849 1850 def p_defs_and_outputs_0(self, t): 1851 'defs_and_outputs : def_or_output' 1852 1853 def p_defs_and_outputs_1(self, t): 1854 'defs_and_outputs : defs_and_outputs def_or_output' 1855 1856 # The list of possible definition/output statements. 1857 # They are all processed as they are seen. 1858 def p_def_or_output(self, t): 1859 '''def_or_output : name_decl 1860 | def_format 1861 | def_bitfield 1862 | def_bitfield_struct 1863 | def_template 1864 | def_operand_types 1865 | def_operands 1866 | output 1867 | global_let 1868 | split''' 1869 1870 # Utility function used by both invocations of splitting - explicit 1871 # 'split' keyword and split() function inside "let {{ }};" blocks. 1872 def split(self, sec, write=False): 1873 assert(sec != 'header' and "header cannot be split") 1874 1875 f = self.get_file(sec) 1876 self.splits[f] += 1 1877 s = '\n#endif\n#if __SPLIT == %u\n' % self.splits[f] 1878 if write: 1879 f.write(s) 1880 else: 1881 return s 1882 1883 # split output file to reduce compilation time 1884 def p_split(self, t): 1885 'split : SPLIT output_type SEMI' 1886 assert(self.isa_name and "'split' not allowed before namespace decl") 1887 1888 self.split(t[2], True) 1889 1890 def p_output_type(self, t): 1891 '''output_type : DECODER 1892 | HEADER 1893 | EXEC''' 1894 t[0] = t[1] 1895 1896 # ISA name declaration looks like "namespace <foo>;" 1897 def p_name_decl(self, t): 1898 'name_decl : NAMESPACE ID SEMI' 1899 assert(self.isa_name == None and "Only 1 namespace decl permitted") 1900 self.isa_name = t[2] 1901 self.namespace = t[2] + 'Inst' 1902 1903 # Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied 1904 # directly to the appropriate output section. 1905 1906 # Massage output block by substituting in template definitions and 1907 # bit operators. We handle '%'s embedded in the string that don't 1908 # indicate template substitutions by doubling them first so that the 1909 # format operation will reduce them back to single '%'s. 1910 def process_output(self, s): 1911 s = self.protectNonSubstPercents(s) 1912 return substBitOps(s % self.templateMap) 1913 1914 def p_output(self, t): 1915 'output : OUTPUT output_type CODELIT SEMI' 1916 kwargs = { t[2]+'_output' : self.process_output(t[3]) } 1917 GenCode(self, **kwargs).emit() 1918 1919 # global let blocks 'let {{...}}' (Python code blocks) are 1920 # executed directly when seen. Note that these execute in a 1921 # special variable context 'exportContext' to prevent the code 1922 # from polluting this script's namespace. 1923 def p_global_let(self, t): 1924 'global_let : LET CODELIT SEMI' 1925 def _split(sec): 1926 return self.split(sec) 1927 self.updateExportContext() 1928 self.exportContext["header_output"] = '' 1929 self.exportContext["decoder_output"] = '' 1930 self.exportContext["exec_output"] = '' 1931 self.exportContext["decode_block"] = '' 1932 self.exportContext["split"] = _split 1933 split_setup = ''' 1934def wrap(func): 1935 def split(sec): 1936 globals()[sec + '_output'] += func(sec) 1937 return split 1938split = wrap(split) 1939del wrap 1940''' 1941 # This tricky setup (immediately above) allows us to just write 1942 # (e.g.) "split('exec')" in the Python code and the split #ifdef's 1943 # will automatically be added to the exec_output variable. The inner 1944 # Python execution environment doesn't know about the split points, 1945 # so we carefully inject and wrap a closure that can retrieve the 1946 # next split's #define from the parser and add it to the current 1947 # emission-in-progress. 1948 try: 1949 exec split_setup+fixPythonIndentation(t[2]) in self.exportContext 1950 except Exception, exc: 1951 traceback.print_exc(file=sys.stdout) 1952 if debug: 1953 raise 1954 error(t.lineno(1), 'In global let block: %s' % exc) 1955 GenCode(self, 1956 header_output=self.exportContext["header_output"], 1957 decoder_output=self.exportContext["decoder_output"], 1958 exec_output=self.exportContext["exec_output"], 1959 decode_block=self.exportContext["decode_block"]).emit() 1960 1961 # Define the mapping from operand type extensions to C++ types and 1962 # bit widths (stored in operandTypeMap). 1963 def p_def_operand_types(self, t): 1964 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI' 1965 try: 1966 self.operandTypeMap = eval('{' + t[3] + '}') 1967 except Exception, exc: 1968 if debug: 1969 raise 1970 error(t.lineno(1), 1971 'In def operand_types: %s' % exc) 1972 1973 # Define the mapping from operand names to operand classes and 1974 # other traits. Stored in operandNameMap. 1975 def p_def_operands(self, t): 1976 'def_operands : DEF OPERANDS CODELIT SEMI' 1977 if not hasattr(self, 'operandTypeMap'): 1978 error(t.lineno(1), 1979 'error: operand types must be defined before operands') 1980 try: 1981 user_dict = eval('{' + t[3] + '}', self.exportContext) 1982 except Exception, exc: 1983 if debug: 1984 raise 1985 error(t.lineno(1), 'In def operands: %s' % exc) 1986 self.buildOperandNameMap(user_dict, t.lexer.lineno) 1987 1988 # A bitfield definition looks like: 1989 # 'def [signed] bitfield <ID> [<first>:<last>]' 1990 # This generates a preprocessor macro in the output file. 1991 def p_def_bitfield_0(self, t): 1992 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI' 1993 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8]) 1994 if (t[2] == 'signed'): 1995 expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr) 1996 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 1997 GenCode(self, header_output=hash_define).emit() 1998 1999 # alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]' 2000 def p_def_bitfield_1(self, t): 2001 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI' 2002 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6]) 2003 if (t[2] == 'signed'): 2004 expr = 'sext<%d>(%s)' % (1, expr) 2005 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 2006 GenCode(self, header_output=hash_define).emit() 2007 2008 # alternate form for structure member: 'def bitfield <ID> <ID>' 2009 def p_def_bitfield_struct(self, t): 2010 'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI' 2011 if (t[2] != ''): 2012 error(t.lineno(1), 2013 'error: structure bitfields are always unsigned.') 2014 expr = 'machInst.%s' % t[5] 2015 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 2016 GenCode(self, header_output=hash_define).emit() 2017 2018 def p_id_with_dot_0(self, t): 2019 'id_with_dot : ID' 2020 t[0] = t[1] 2021 2022 def p_id_with_dot_1(self, t): 2023 'id_with_dot : ID DOT id_with_dot' 2024 t[0] = t[1] + t[2] + t[3] 2025 2026 def p_opt_signed_0(self, t): 2027 'opt_signed : SIGNED' 2028 t[0] = t[1] 2029 2030 def p_opt_signed_1(self, t): 2031 'opt_signed : empty' 2032 t[0] = '' 2033 2034 def p_def_template(self, t): 2035 'def_template : DEF TEMPLATE ID CODELIT SEMI' 2036 if t[3] in self.templateMap: 2037 print("warning: template %s already defined" % t[3]) 2038 self.templateMap[t[3]] = Template(self, t[4]) 2039 2040 # An instruction format definition looks like 2041 # "def format <fmt>(<params>) {{...}};" 2042 def p_def_format(self, t): 2043 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI' 2044 (id, params, code) = (t[3], t[5], t[7]) 2045 self.defFormat(id, params, code, t.lexer.lineno) 2046 2047 # The formal parameter list for an instruction format is a 2048 # possibly empty list of comma-separated parameters. Positional 2049 # (standard, non-keyword) parameters must come first, followed by 2050 # keyword parameters, followed by a '*foo' parameter that gets 2051 # excess positional arguments (as in Python). Each of these three 2052 # parameter categories is optional. 2053 # 2054 # Note that we do not support the '**foo' parameter for collecting 2055 # otherwise undefined keyword args. Otherwise the parameter list 2056 # is (I believe) identical to what is supported in Python. 2057 # 2058 # The param list generates a tuple, where the first element is a 2059 # list of the positional params and the second element is a dict 2060 # containing the keyword params. 2061 def p_param_list_0(self, t): 2062 'param_list : positional_param_list COMMA nonpositional_param_list' 2063 t[0] = t[1] + t[3] 2064 2065 def p_param_list_1(self, t): 2066 '''param_list : positional_param_list 2067 | nonpositional_param_list''' 2068 t[0] = t[1] 2069 2070 def p_positional_param_list_0(self, t): 2071 'positional_param_list : empty' 2072 t[0] = [] 2073 2074 def p_positional_param_list_1(self, t): 2075 'positional_param_list : ID' 2076 t[0] = [t[1]] 2077 2078 def p_positional_param_list_2(self, t): 2079 'positional_param_list : positional_param_list COMMA ID' 2080 t[0] = t[1] + [t[3]] 2081 2082 def p_nonpositional_param_list_0(self, t): 2083 'nonpositional_param_list : keyword_param_list COMMA excess_args_param' 2084 t[0] = t[1] + t[3] 2085 2086 def p_nonpositional_param_list_1(self, t): 2087 '''nonpositional_param_list : keyword_param_list 2088 | excess_args_param''' 2089 t[0] = t[1] 2090 2091 def p_keyword_param_list_0(self, t): 2092 'keyword_param_list : keyword_param' 2093 t[0] = [t[1]] 2094 2095 def p_keyword_param_list_1(self, t): 2096 'keyword_param_list : keyword_param_list COMMA keyword_param' 2097 t[0] = t[1] + [t[3]] 2098 2099 def p_keyword_param(self, t): 2100 'keyword_param : ID EQUALS expr' 2101 t[0] = t[1] + ' = ' + t[3].__repr__() 2102 2103 def p_excess_args_param(self, t): 2104 'excess_args_param : ASTERISK ID' 2105 # Just concatenate them: '*ID'. Wrap in list to be consistent 2106 # with positional_param_list and keyword_param_list. 2107 t[0] = [t[1] + t[2]] 2108 2109 # End of format definition-related rules. 2110 ############## 2111 2112 # 2113 # A decode block looks like: 2114 # decode <field1> [, <field2>]* [default <inst>] { ... } 2115 # 2116 def p_top_level_decode_block(self, t): 2117 'top_level_decode_block : decode_block' 2118 codeObj = t[1] 2119 codeObj.wrap_decode_block(''' 2120StaticInstPtr 2121%(isa_name)s::Decoder::decodeInst(%(isa_name)s::ExtMachInst machInst) 2122{ 2123 using namespace %(namespace)s; 2124''' % self, '}') 2125 2126 codeObj.emit() 2127 2128 def p_decode_block(self, t): 2129 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE' 2130 default_defaults = self.defaultStack.pop() 2131 codeObj = t[5] 2132 # use the "default defaults" only if there was no explicit 2133 # default statement in decode_stmt_list 2134 if not codeObj.has_decode_default: 2135 codeObj += default_defaults 2136 codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n') 2137 t[0] = codeObj 2138 2139 # The opt_default statement serves only to push the "default 2140 # defaults" onto defaultStack. This value will be used by nested 2141 # decode blocks, and used and popped off when the current 2142 # decode_block is processed (in p_decode_block() above). 2143 def p_opt_default_0(self, t): 2144 'opt_default : empty' 2145 # no default specified: reuse the one currently at the top of 2146 # the stack 2147 self.defaultStack.push(self.defaultStack.top()) 2148 # no meaningful value returned 2149 t[0] = None 2150 2151 def p_opt_default_1(self, t): 2152 'opt_default : DEFAULT inst' 2153 # push the new default 2154 codeObj = t[2] 2155 codeObj.wrap_decode_block('\ndefault:\n', 'break;\n') 2156 self.defaultStack.push(codeObj) 2157 # no meaningful value returned 2158 t[0] = None 2159 2160 def p_decode_stmt_list_0(self, t): 2161 'decode_stmt_list : decode_stmt' 2162 t[0] = t[1] 2163 2164 def p_decode_stmt_list_1(self, t): 2165 'decode_stmt_list : decode_stmt decode_stmt_list' 2166 if (t[1].has_decode_default and t[2].has_decode_default): 2167 error(t.lineno(1), 'Two default cases in decode block') 2168 t[0] = t[1] + t[2] 2169 2170 # 2171 # Decode statement rules 2172 # 2173 # There are four types of statements allowed in a decode block: 2174 # 1. Format blocks 'format <foo> { ... }' 2175 # 2. Nested decode blocks 2176 # 3. Instruction definitions. 2177 # 4. C preprocessor directives. 2178 2179 2180 # Preprocessor directives found in a decode statement list are 2181 # passed through to the output, replicated to all of the output 2182 # code streams. This works well for ifdefs, so we can ifdef out 2183 # both the declarations and the decode cases generated by an 2184 # instruction definition. Handling them as part of the grammar 2185 # makes it easy to keep them in the right place with respect to 2186 # the code generated by the other statements. 2187 def p_decode_stmt_cpp(self, t): 2188 'decode_stmt : CPPDIRECTIVE' 2189 t[0] = GenCode(self, t[1], t[1], t[1], t[1]) 2190 2191 # A format block 'format <foo> { ... }' sets the default 2192 # instruction format used to handle instruction definitions inside 2193 # the block. This format can be overridden by using an explicit 2194 # format on the instruction definition or with a nested format 2195 # block. 2196 def p_decode_stmt_format(self, t): 2197 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE' 2198 # The format will be pushed on the stack when 'push_format_id' 2199 # is processed (see below). Once the parser has recognized 2200 # the full production (though the right brace), we're done 2201 # with the format, so now we can pop it. 2202 self.formatStack.pop() 2203 t[0] = t[4] 2204 2205 # This rule exists so we can set the current format (& push the 2206 # stack) when we recognize the format name part of the format 2207 # block. 2208 def p_push_format_id(self, t): 2209 'push_format_id : ID' 2210 try: 2211 self.formatStack.push(self.formatMap[t[1]]) 2212 t[0] = ('', '// format %s' % t[1]) 2213 except KeyError: 2214 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 2215 2216 # Nested decode block: if the value of the current field matches 2217 # the specified constant(s), do a nested decode on some other field. 2218 def p_decode_stmt_decode(self, t): 2219 'decode_stmt : case_list COLON decode_block' 2220 case_list = t[1] 2221 codeObj = t[3] 2222 # just wrap the decoding code from the block as a case in the 2223 # outer switch statement. 2224 codeObj.wrap_decode_block('\n%s\n' % ''.join(case_list), 2225 'M5_UNREACHABLE;\n') 2226 codeObj.has_decode_default = (case_list == ['default:']) 2227 t[0] = codeObj 2228 2229 # Instruction definition (finally!). 2230 def p_decode_stmt_inst(self, t): 2231 'decode_stmt : case_list COLON inst SEMI' 2232 case_list = t[1] 2233 codeObj = t[3] 2234 codeObj.wrap_decode_block('\n%s' % ''.join(case_list), 'break;\n') 2235 codeObj.has_decode_default = (case_list == ['default:']) 2236 t[0] = codeObj 2237 2238 # The constant list for a decode case label must be non-empty, and must 2239 # either be the keyword 'default', or made up of one or more 2240 # comma-separated integer literals or strings which evaluate to 2241 # constants when compiled as C++. 2242 def p_case_list_0(self, t): 2243 'case_list : DEFAULT' 2244 t[0] = ['default:'] 2245 2246 def prep_int_lit_case_label(self, lit): 2247 if lit >= 2**32: 2248 return 'case ULL(%#x): ' % lit 2249 else: 2250 return 'case %#x: ' % lit 2251 2252 def prep_str_lit_case_label(self, lit): 2253 return 'case %s: ' % lit 2254 2255 def p_case_list_1(self, t): 2256 'case_list : INTLIT' 2257 t[0] = [self.prep_int_lit_case_label(t[1])] 2258 2259 def p_case_list_2(self, t): 2260 'case_list : STRLIT' 2261 t[0] = [self.prep_str_lit_case_label(t[1])] 2262 2263 def p_case_list_3(self, t): 2264 'case_list : case_list COMMA INTLIT' 2265 t[0] = t[1] 2266 t[0].append(self.prep_int_lit_case_label(t[3])) 2267 2268 def p_case_list_4(self, t): 2269 'case_list : case_list COMMA STRLIT' 2270 t[0] = t[1] 2271 t[0].append(self.prep_str_lit_case_label(t[3])) 2272 2273 # Define an instruction using the current instruction format 2274 # (specified by an enclosing format block). 2275 # "<mnemonic>(<args>)" 2276 def p_inst_0(self, t): 2277 'inst : ID LPAREN arg_list RPAREN' 2278 # Pass the ID and arg list to the current format class to deal with. 2279 currentFormat = self.formatStack.top() 2280 codeObj = currentFormat.defineInst(self, t[1], t[3], t.lexer.lineno) 2281 args = ','.join(map(str, t[3])) 2282 args = re.sub('(?m)^', '//', args) 2283 args = re.sub('^//', '', args) 2284 comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args) 2285 codeObj.prepend_all(comment) 2286 t[0] = codeObj 2287 2288 # Define an instruction using an explicitly specified format: 2289 # "<fmt>::<mnemonic>(<args>)" 2290 def p_inst_1(self, t): 2291 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN' 2292 try: 2293 format = self.formatMap[t[1]] 2294 except KeyError: 2295 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 2296 2297 codeObj = format.defineInst(self, t[3], t[5], t.lexer.lineno) 2298 comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5]) 2299 codeObj.prepend_all(comment) 2300 t[0] = codeObj 2301 2302 # The arg list generates a tuple, where the first element is a 2303 # list of the positional args and the second element is a dict 2304 # containing the keyword args. 2305 def p_arg_list_0(self, t): 2306 'arg_list : positional_arg_list COMMA keyword_arg_list' 2307 t[0] = ( t[1], t[3] ) 2308 2309 def p_arg_list_1(self, t): 2310 'arg_list : positional_arg_list' 2311 t[0] = ( t[1], {} ) 2312 2313 def p_arg_list_2(self, t): 2314 'arg_list : keyword_arg_list' 2315 t[0] = ( [], t[1] ) 2316 2317 def p_positional_arg_list_0(self, t): 2318 'positional_arg_list : empty' 2319 t[0] = [] 2320 2321 def p_positional_arg_list_1(self, t): 2322 'positional_arg_list : expr' 2323 t[0] = [t[1]] 2324 2325 def p_positional_arg_list_2(self, t): 2326 'positional_arg_list : positional_arg_list COMMA expr' 2327 t[0] = t[1] + [t[3]] 2328 2329 def p_keyword_arg_list_0(self, t): 2330 'keyword_arg_list : keyword_arg' 2331 t[0] = t[1] 2332 2333 def p_keyword_arg_list_1(self, t): 2334 'keyword_arg_list : keyword_arg_list COMMA keyword_arg' 2335 t[0] = t[1] 2336 t[0].update(t[3]) 2337 2338 def p_keyword_arg(self, t): 2339 'keyword_arg : ID EQUALS expr' 2340 t[0] = { t[1] : t[3] } 2341 2342 # 2343 # Basic expressions. These constitute the argument values of 2344 # "function calls" (i.e. instruction definitions in the decode 2345 # block) and default values for formal parameters of format 2346 # functions. 2347 # 2348 # Right now, these are either strings, integers, or (recursively) 2349 # lists of exprs (using Python square-bracket list syntax). Note 2350 # that bare identifiers are trated as string constants here (since 2351 # there isn't really a variable namespace to refer to). 2352 # 2353 def p_expr_0(self, t): 2354 '''expr : ID 2355 | INTLIT 2356 | STRLIT 2357 | CODELIT''' 2358 t[0] = t[1] 2359 2360 def p_expr_1(self, t): 2361 '''expr : LBRACKET list_expr RBRACKET''' 2362 t[0] = t[2] 2363 2364 def p_list_expr_0(self, t): 2365 'list_expr : expr' 2366 t[0] = [t[1]] 2367 2368 def p_list_expr_1(self, t): 2369 'list_expr : list_expr COMMA expr' 2370 t[0] = t[1] + [t[3]] 2371 2372 def p_list_expr_2(self, t): 2373 'list_expr : empty' 2374 t[0] = [] 2375 2376 # 2377 # Empty production... use in other rules for readability. 2378 # 2379 def p_empty(self, t): 2380 'empty :' 2381 pass 2382 2383 # Parse error handler. Note that the argument here is the 2384 # offending *token*, not a grammar symbol (hence the need to use 2385 # t.value) 2386 def p_error(self, t): 2387 if t: 2388 error(t.lexer.lineno, "syntax error at '%s'" % t.value) 2389 else: 2390 error("unknown syntax error") 2391 2392 # END OF GRAMMAR RULES 2393 2394 def updateExportContext(self): 2395 2396 # create a continuation that allows us to grab the current parser 2397 def wrapInstObjParams(*args): 2398 return InstObjParams(self, *args) 2399 self.exportContext['InstObjParams'] = wrapInstObjParams 2400 self.exportContext.update(self.templateMap) 2401 2402 def defFormat(self, id, params, code, lineno): 2403 '''Define a new format''' 2404 2405 # make sure we haven't already defined this one 2406 if id in self.formatMap: 2407 error(lineno, 'format %s redefined.' % id) 2408 2409 # create new object and store in global map 2410 self.formatMap[id] = Format(id, params, code) 2411 2412 def protectNonSubstPercents(self, s): 2413 '''Protect any non-dict-substitution '%'s in a format string 2414 (i.e. those not followed by '(')''' 2415 2416 return re.sub(r'%(?!\()', '%%', s) 2417 2418 def buildOperandNameMap(self, user_dict, lineno): 2419 operand_name = {} 2420 for op_name, val in user_dict.iteritems(): 2421 2422 # Check if extra attributes have been specified. 2423 if len(val) > 9: 2424 error(lineno, 'error: too many attributes for operand "%s"' % 2425 base_cls_name) 2426 2427 # Pad val with None in case optional args are missing 2428 val += (None, None, None, None) 2429 base_cls_name, dflt_ext, reg_spec, flags, sort_pri, \ 2430 read_code, write_code, read_predicate, write_predicate = val[:9] 2431 2432 # Canonical flag structure is a triple of lists, where each list 2433 # indicates the set of flags implied by this operand always, when 2434 # used as a source, and when used as a dest, respectively. 2435 # For simplicity this can be initialized using a variety of fairly 2436 # obvious shortcuts; we convert these to canonical form here. 2437 if not flags: 2438 # no flags specified (e.g., 'None') 2439 flags = ( [], [], [] ) 2440 elif isinstance(flags, str): 2441 # a single flag: assumed to be unconditional 2442 flags = ( [ flags ], [], [] ) 2443 elif isinstance(flags, list): 2444 # a list of flags: also assumed to be unconditional 2445 flags = ( flags, [], [] ) 2446 elif isinstance(flags, tuple): 2447 # it's a tuple: it should be a triple, 2448 # but each item could be a single string or a list 2449 (uncond_flags, src_flags, dest_flags) = flags 2450 flags = (makeList(uncond_flags), 2451 makeList(src_flags), makeList(dest_flags)) 2452 2453 # Accumulate attributes of new operand class in tmp_dict 2454 tmp_dict = {} 2455 attrList = ['reg_spec', 'flags', 'sort_pri', 2456 'read_code', 'write_code', 2457 'read_predicate', 'write_predicate'] 2458 if dflt_ext: 2459 dflt_ctype = self.operandTypeMap[dflt_ext] 2460 attrList.extend(['dflt_ctype', 'dflt_ext']) 2461 # reg_spec is either just a string or a dictionary 2462 # (for elems of vector) 2463 if isinstance(reg_spec, tuple): 2464 (reg_spec, elem_spec) = reg_spec 2465 if isinstance(elem_spec, str): 2466 attrList.append('elem_spec') 2467 else: 2468 assert(isinstance(elem_spec, dict)) 2469 elems = elem_spec 2470 attrList.append('elems') 2471 for attr in attrList: 2472 tmp_dict[attr] = eval(attr) 2473 tmp_dict['base_name'] = op_name 2474 2475 # New class name will be e.g. "IntReg_Ra" 2476 cls_name = base_cls_name + '_' + op_name 2477 # Evaluate string arg to get class object. Note that the 2478 # actual base class for "IntReg" is "IntRegOperand", i.e. we 2479 # have to append "Operand". 2480 try: 2481 base_cls = eval(base_cls_name + 'Operand') 2482 except NameError: 2483 error(lineno, 2484 'error: unknown operand base class "%s"' % base_cls_name) 2485 # The following statement creates a new class called 2486 # <cls_name> as a subclass of <base_cls> with the attributes 2487 # in tmp_dict, just as if we evaluated a class declaration. 2488 operand_name[op_name] = type(cls_name, (base_cls,), tmp_dict) 2489 2490 self.operandNameMap = operand_name 2491 2492 # Define operand variables. 2493 operands = user_dict.keys() 2494 # Add the elems defined in the vector operands and 2495 # build a map elem -> vector (used in OperandList) 2496 elem_to_vec = {} 2497 for op in user_dict.keys(): 2498 if hasattr(self.operandNameMap[op], 'elems'): 2499 for elem in self.operandNameMap[op].elems.keys(): 2500 operands.append(elem) 2501 elem_to_vec[elem] = op 2502 self.elemToVector = elem_to_vec 2503 extensions = self.operandTypeMap.keys() 2504 2505 operandsREString = r''' 2506 (?<!\w) # neg. lookbehind assertion: prevent partial matches 2507 ((%s)(?:_(%s))?) # match: operand with optional '_' then suffix 2508 (?!\w) # neg. lookahead assertion: prevent partial matches 2509 ''' % (string.join(operands, '|'), string.join(extensions, '|')) 2510 2511 self.operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE) 2512 2513 # Same as operandsREString, but extension is mandatory, and only two 2514 # groups are returned (base and ext, not full name as above). 2515 # Used for subtituting '_' for '.' to make C++ identifiers. 2516 operandsWithExtREString = r'(?<!\w)(%s)_(%s)(?!\w)' \ 2517 % (string.join(operands, '|'), string.join(extensions, '|')) 2518 2519 self.operandsWithExtRE = \ 2520 re.compile(operandsWithExtREString, re.MULTILINE) 2521 2522 def substMungedOpNames(self, code): 2523 '''Munge operand names in code string to make legal C++ 2524 variable names. This means getting rid of the type extension 2525 if any. Will match base_name attribute of Operand object.)''' 2526 return self.operandsWithExtRE.sub(r'\1', code) 2527 2528 def mungeSnippet(self, s): 2529 '''Fix up code snippets for final substitution in templates.''' 2530 if isinstance(s, str): 2531 return self.substMungedOpNames(substBitOps(s)) 2532 else: 2533 return s 2534 2535 def open(self, name, bare=False): 2536 '''Open the output file for writing and include scary warning.''' 2537 filename = os.path.join(self.output_dir, name) 2538 f = open(filename, 'w') 2539 if f: 2540 if not bare: 2541 f.write(ISAParser.scaremonger_template % self) 2542 return f 2543 2544 def update(self, file, contents): 2545 '''Update the output file only. Scons should handle the case when 2546 the new contents are unchanged using its built-in hash feature.''' 2547 f = self.open(file) 2548 f.write(contents) 2549 f.close() 2550 2551 # This regular expression matches '##include' directives 2552 includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[^"]*)".*$', 2553 re.MULTILINE) 2554 2555 def replace_include(self, matchobj, dirname): 2556 """Function to replace a matched '##include' directive with the 2557 contents of the specified file (with nested ##includes 2558 replaced recursively). 'matchobj' is an re match object 2559 (from a match of includeRE) and 'dirname' is the directory 2560 relative to which the file path should be resolved.""" 2561 2562 fname = matchobj.group('filename') 2563 full_fname = os.path.normpath(os.path.join(dirname, fname)) 2564 contents = '##newfile "%s"\n%s\n##endfile\n' % \ 2565 (full_fname, self.read_and_flatten(full_fname)) 2566 return contents 2567 2568 def read_and_flatten(self, filename): 2569 """Read a file and recursively flatten nested '##include' files.""" 2570 2571 current_dir = os.path.dirname(filename) 2572 try: 2573 contents = open(filename).read() 2574 except IOError: 2575 error('Error including file "%s"' % filename) 2576 2577 self.fileNameStack.push(LineTracker(filename)) 2578 2579 # Find any includes and include them 2580 def replace(matchobj): 2581 return self.replace_include(matchobj, current_dir) 2582 contents = self.includeRE.sub(replace, contents) 2583 2584 self.fileNameStack.pop() 2585 return contents 2586 2587 AlreadyGenerated = {} 2588 2589 def _parse_isa_desc(self, isa_desc_file): 2590 '''Read in and parse the ISA description.''' 2591 2592 # The build system can end up running the ISA parser twice: once to 2593 # finalize the build dependencies, and then to actually generate 2594 # the files it expects (in src/arch/$ARCH/generated). This code 2595 # doesn't do anything different either time, however; the SCons 2596 # invocations just expect different things. Since this code runs 2597 # within SCons, we can just remember that we've already run and 2598 # not perform a completely unnecessary run, since the ISA parser's 2599 # effect is idempotent. 2600 if isa_desc_file in ISAParser.AlreadyGenerated: 2601 return 2602 2603 # grab the last three path components of isa_desc_file 2604 self.filename = '/'.join(isa_desc_file.split('/')[-3:]) 2605 2606 # Read file and (recursively) all included files into a string. 2607 # PLY requires that the input be in a single string so we have to 2608 # do this up front. 2609 isa_desc = self.read_and_flatten(isa_desc_file) 2610 2611 # Initialize lineno tracker 2612 self.lex.lineno = LineTracker(isa_desc_file) 2613 2614 # Parse. 2615 self.parse_string(isa_desc) 2616 2617 ISAParser.AlreadyGenerated[isa_desc_file] = None 2618 2619 def parse_isa_desc(self, *args, **kwargs): 2620 try: 2621 self._parse_isa_desc(*args, **kwargs) 2622 except ISAParserError, e: 2623 print(backtrace(self.fileNameStack)) 2624 print("At %s:" % e.lineno) 2625 print(e) 2626 sys.exit(1) 2627 2628# Called as script: get args from command line. 2629# Args are: <isa desc file> <output dir> 2630if __name__ == '__main__': 2631 ISAParser(sys.argv[2]).parse_isa_desc(sys.argv[1]) 2632