isa_parser.py revision 12204:2ba80db6075c
1# Copyright (c) 2014, 2016 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 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 # CPU-model-specific substitutions are handled later (in GenCode). 126 template = self.parser.protectCpuSymbols(template) 127 128 # Build a dict ('myDict') to use for the template substitution. 129 # Start with the template namespace. Make a copy since we're 130 # going to modify it. 131 myDict = self.parser.templateMap.copy() 132 133 if isinstance(d, InstObjParams): 134 # If we're dealing with an InstObjParams object, we need 135 # to be a little more sophisticated. The instruction-wide 136 # parameters are already formed, but the parameters which 137 # are only function wide still need to be generated. 138 compositeCode = '' 139 140 myDict.update(d.__dict__) 141 # The "operands" and "snippets" attributes of the InstObjParams 142 # objects are for internal use and not substitution. 143 del myDict['operands'] 144 del myDict['snippets'] 145 146 snippetLabels = [l for l in labelRE.findall(template) 147 if d.snippets.has_key(l)] 148 149 snippets = dict([(s, self.parser.mungeSnippet(d.snippets[s])) 150 for s in snippetLabels]) 151 152 myDict.update(snippets) 153 154 compositeCode = ' '.join(map(str, snippets.values())) 155 156 # Add in template itself in case it references any 157 # operands explicitly (like Mem) 158 compositeCode += ' ' + template 159 160 operands = SubOperandList(self.parser, compositeCode, d.operands) 161 162 myDict['op_decl'] = operands.concatAttrStrings('op_decl') 163 if operands.readPC or operands.setPC: 164 myDict['op_decl'] += 'TheISA::PCState __parserAutoPCState;\n' 165 166 # In case there are predicated register reads and write, declare 167 # the variables for register indicies. It is being assumed that 168 # all the operands in the OperandList are also in the 169 # SubOperandList and in the same order. Otherwise, it is 170 # expected that predication would not be used for the operands. 171 if operands.predRead: 172 myDict['op_decl'] += 'uint8_t _sourceIndex = 0;\n' 173 if operands.predWrite: 174 myDict['op_decl'] += 'uint8_t M5_VAR_USED _destIndex = 0;\n' 175 176 is_src = lambda op: op.is_src 177 is_dest = lambda op: op.is_dest 178 179 myDict['op_src_decl'] = \ 180 operands.concatSomeAttrStrings(is_src, 'op_src_decl') 181 myDict['op_dest_decl'] = \ 182 operands.concatSomeAttrStrings(is_dest, 'op_dest_decl') 183 if operands.readPC: 184 myDict['op_src_decl'] += \ 185 'TheISA::PCState __parserAutoPCState;\n' 186 if operands.setPC: 187 myDict['op_dest_decl'] += \ 188 'TheISA::PCState __parserAutoPCState;\n' 189 190 myDict['op_rd'] = operands.concatAttrStrings('op_rd') 191 if operands.readPC: 192 myDict['op_rd'] = '__parserAutoPCState = xc->pcState();\n' + \ 193 myDict['op_rd'] 194 195 # Compose the op_wb string. If we're going to write back the 196 # PC state because we changed some of its elements, we'll need to 197 # do that as early as possible. That allows later uncoordinated 198 # modifications to the PC to layer appropriately. 199 reordered = list(operands.items) 200 reordered.reverse() 201 op_wb_str = '' 202 pcWbStr = 'xc->pcState(__parserAutoPCState);\n' 203 for op_desc in reordered: 204 if op_desc.isPCPart() and op_desc.is_dest: 205 op_wb_str = op_desc.op_wb + pcWbStr + op_wb_str 206 pcWbStr = '' 207 else: 208 op_wb_str = op_desc.op_wb + op_wb_str 209 myDict['op_wb'] = op_wb_str 210 211 elif isinstance(d, dict): 212 # if the argument is a dictionary, we just use it. 213 myDict.update(d) 214 elif hasattr(d, '__dict__'): 215 # if the argument is an object, we use its attribute map. 216 myDict.update(d.__dict__) 217 else: 218 raise TypeError, "Template.subst() arg must be or have dictionary" 219 return template % myDict 220 221 # Convert to string. This handles the case when a template with a 222 # CPU-specific term gets interpolated into another template or into 223 # an output block. 224 def __str__(self): 225 return self.parser.expandCpuSymbolsToString(self.template) 226 227################ 228# Format object. 229# 230# A format object encapsulates an instruction format. It must provide 231# a defineInst() method that generates the code for an instruction 232# definition. 233 234class Format(object): 235 def __init__(self, id, params, code): 236 self.id = id 237 self.params = params 238 label = 'def format ' + id 239 self.user_code = compile(fixPythonIndentation(code), label, 'exec') 240 param_list = string.join(params, ", ") 241 f = '''def defInst(_code, _context, %s): 242 my_locals = vars().copy() 243 exec _code in _context, my_locals 244 return my_locals\n''' % param_list 245 c = compile(f, label + ' wrapper', 'exec') 246 exec c 247 self.func = defInst 248 249 def defineInst(self, parser, name, args, lineno): 250 parser.updateExportContext() 251 context = parser.exportContext.copy() 252 if len(name): 253 Name = name[0].upper() 254 if len(name) > 1: 255 Name += name[1:] 256 context.update({ 'name' : name, 'Name' : Name }) 257 try: 258 vars = self.func(self.user_code, context, *args[0], **args[1]) 259 except Exception, exc: 260 if debug: 261 raise 262 error(lineno, 'error defining "%s": %s.' % (name, exc)) 263 for k in vars.keys(): 264 if k not in ('header_output', 'decoder_output', 265 'exec_output', 'decode_block'): 266 del vars[k] 267 return GenCode(parser, **vars) 268 269# Special null format to catch an implicit-format instruction 270# definition outside of any format block. 271class NoFormat(object): 272 def __init__(self): 273 self.defaultInst = '' 274 275 def defineInst(self, parser, name, args, lineno): 276 error(lineno, 277 'instruction definition "%s" with no active format!' % name) 278 279############### 280# GenCode class 281# 282# The GenCode class encapsulates generated code destined for various 283# output files. The header_output and decoder_output attributes are 284# strings containing code destined for decoder.hh and decoder.cc 285# respectively. The decode_block attribute contains code to be 286# incorporated in the decode function itself (that will also end up in 287# decoder.cc). The exec_output attribute is a dictionary with a key 288# for each CPU model name; the value associated with a particular key 289# is the string of code for that CPU model's exec.cc file. The 290# has_decode_default attribute is used in the decode block to allow 291# explicit default clauses to override default default clauses. 292 293class GenCode(object): 294 # Constructor. At this point we substitute out all CPU-specific 295 # symbols. For the exec output, these go into the per-model 296 # dictionary. For all other output types they get collapsed into 297 # a single string. 298 def __init__(self, parser, 299 header_output = '', decoder_output = '', exec_output = '', 300 decode_block = '', has_decode_default = False): 301 self.parser = parser 302 self.header_output = parser.expandCpuSymbolsToString(header_output) 303 self.decoder_output = parser.expandCpuSymbolsToString(decoder_output) 304 self.exec_output = exec_output 305 self.decode_block = decode_block 306 self.has_decode_default = has_decode_default 307 308 # Write these code chunks out to the filesystem. They will be properly 309 # interwoven by the write_top_level_files(). 310 def emit(self): 311 if self.header_output: 312 self.parser.get_file('header').write(self.header_output) 313 if self.decoder_output: 314 self.parser.get_file('decoder').write(self.decoder_output) 315 if self.exec_output: 316 self.parser.get_file('exec').write(self.exec_output) 317 if self.decode_block: 318 self.parser.get_file('decode_block').write(self.decode_block) 319 320 # Override '+' operator: generate a new GenCode object that 321 # concatenates all the individual strings in the operands. 322 def __add__(self, other): 323 return GenCode(self.parser, 324 self.header_output + other.header_output, 325 self.decoder_output + other.decoder_output, 326 self.exec_output + other.exec_output, 327 self.decode_block + other.decode_block, 328 self.has_decode_default or other.has_decode_default) 329 330 # Prepend a string (typically a comment) to all the strings. 331 def prepend_all(self, pre): 332 self.header_output = pre + self.header_output 333 self.decoder_output = pre + self.decoder_output 334 self.decode_block = pre + self.decode_block 335 self.exec_output = pre + self.exec_output 336 337 # Wrap the decode block in a pair of strings (e.g., 'case foo:' 338 # and 'break;'). Used to build the big nested switch statement. 339 def wrap_decode_block(self, pre, post = ''): 340 self.decode_block = pre + indent(self.decode_block) + post 341 342##################################################################### 343# 344# Bitfield Operator Support 345# 346##################################################################### 347 348bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>') 349 350bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>') 351bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>') 352 353def substBitOps(code): 354 # first convert single-bit selectors to two-index form 355 # i.e., <n> --> <n:n> 356 code = bitOp1ArgRE.sub(r'<\1:\1>', code) 357 # simple case: selector applied to ID (name) 358 # i.e., foo<a:b> --> bits(foo, a, b) 359 code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code) 360 # if selector is applied to expression (ending in ')'), 361 # we need to search backward for matching '(' 362 match = bitOpExprRE.search(code) 363 while match: 364 exprEnd = match.start() 365 here = exprEnd - 1 366 nestLevel = 1 367 while nestLevel > 0: 368 if code[here] == '(': 369 nestLevel -= 1 370 elif code[here] == ')': 371 nestLevel += 1 372 here -= 1 373 if here < 0: 374 sys.exit("Didn't find '('!") 375 exprStart = here+1 376 newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1], 377 match.group(1), match.group(2)) 378 code = code[:exprStart] + newExpr + code[match.end():] 379 match = bitOpExprRE.search(code) 380 return code 381 382 383##################################################################### 384# 385# Code Parser 386# 387# The remaining code is the support for automatically extracting 388# instruction characteristics from pseudocode. 389# 390##################################################################### 391 392# Force the argument to be a list. Useful for flags, where a caller 393# can specify a singleton flag or a list of flags. Also usful for 394# converting tuples to lists so they can be modified. 395def makeList(arg): 396 if isinstance(arg, list): 397 return arg 398 elif isinstance(arg, tuple): 399 return list(arg) 400 elif not arg: 401 return [] 402 else: 403 return [ arg ] 404 405class Operand(object): 406 '''Base class for operand descriptors. An instance of this class 407 (or actually a class derived from this one) represents a specific 408 operand for a code block (e.g, "Rc.sq" as a dest). Intermediate 409 derived classes encapsulates the traits of a particular operand 410 type (e.g., "32-bit integer register").''' 411 412 def buildReadCode(self, func = None): 413 subst_dict = {"name": self.base_name, 414 "func": func, 415 "reg_idx": self.reg_spec, 416 "ctype": self.ctype} 417 if hasattr(self, 'src_reg_idx'): 418 subst_dict['op_idx'] = self.src_reg_idx 419 code = self.read_code % subst_dict 420 return '%s = %s;\n' % (self.base_name, code) 421 422 def buildWriteCode(self, func = None): 423 subst_dict = {"name": self.base_name, 424 "func": func, 425 "reg_idx": self.reg_spec, 426 "ctype": self.ctype, 427 "final_val": self.base_name} 428 if hasattr(self, 'dest_reg_idx'): 429 subst_dict['op_idx'] = self.dest_reg_idx 430 code = self.write_code % subst_dict 431 return ''' 432 { 433 %s final_val = %s; 434 %s; 435 if (traceData) { traceData->setData(final_val); } 436 }''' % (self.dflt_ctype, self.base_name, code) 437 438 def __init__(self, parser, full_name, ext, is_src, is_dest): 439 self.full_name = full_name 440 self.ext = ext 441 self.is_src = is_src 442 self.is_dest = is_dest 443 # The 'effective extension' (eff_ext) is either the actual 444 # extension, if one was explicitly provided, or the default. 445 if ext: 446 self.eff_ext = ext 447 elif hasattr(self, 'dflt_ext'): 448 self.eff_ext = self.dflt_ext 449 450 if hasattr(self, 'eff_ext'): 451 self.ctype = parser.operandTypeMap[self.eff_ext] 452 453 # Finalize additional fields (primarily code fields). This step 454 # is done separately since some of these fields may depend on the 455 # register index enumeration that hasn't been performed yet at the 456 # time of __init__(). The register index enumeration is affected 457 # by predicated register reads/writes. Hence, we forward the flags 458 # that indicate whether or not predication is in use. 459 def finalize(self, predRead, predWrite): 460 self.flags = self.getFlags() 461 self.constructor = self.makeConstructor(predRead, predWrite) 462 self.op_decl = self.makeDecl() 463 464 if self.is_src: 465 self.op_rd = self.makeRead(predRead) 466 self.op_src_decl = self.makeDecl() 467 else: 468 self.op_rd = '' 469 self.op_src_decl = '' 470 471 if self.is_dest: 472 self.op_wb = self.makeWrite(predWrite) 473 self.op_dest_decl = self.makeDecl() 474 else: 475 self.op_wb = '' 476 self.op_dest_decl = '' 477 478 def isMem(self): 479 return 0 480 481 def isReg(self): 482 return 0 483 484 def isFloatReg(self): 485 return 0 486 487 def isIntReg(self): 488 return 0 489 490 def isCCReg(self): 491 return 0 492 493 def isControlReg(self): 494 return 0 495 496 def isVecReg(self): 497 return 0 498 499 def isVecElem(self): 500 return 0 501 502 def isPCState(self): 503 return 0 504 505 def isPCPart(self): 506 return self.isPCState() and self.reg_spec 507 508 def hasReadPred(self): 509 return self.read_predicate != None 510 511 def hasWritePred(self): 512 return self.write_predicate != None 513 514 def getFlags(self): 515 # note the empty slice '[:]' gives us a copy of self.flags[0] 516 # instead of a reference to it 517 my_flags = self.flags[0][:] 518 if self.is_src: 519 my_flags += self.flags[1] 520 if self.is_dest: 521 my_flags += self.flags[2] 522 return my_flags 523 524 def makeDecl(self): 525 # Note that initializations in the declarations are solely 526 # to avoid 'uninitialized variable' errors from the compiler. 527 return self.ctype + ' ' + self.base_name + ' = 0;\n'; 528 529 530src_reg_constructor = '\n\t_srcRegIdx[_numSrcRegs++] = RegId(%s, %s);' 531dst_reg_constructor = '\n\t_destRegIdx[_numDestRegs++] = RegId(%s, %s);' 532 533 534class IntRegOperand(Operand): 535 reg_class = 'IntRegClass' 536 537 def isReg(self): 538 return 1 539 540 def isIntReg(self): 541 return 1 542 543 def makeConstructor(self, predRead, predWrite): 544 c_src = '' 545 c_dest = '' 546 547 if self.is_src: 548 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 549 if self.hasReadPred(): 550 c_src = '\n\tif (%s) {%s\n\t}' % \ 551 (self.read_predicate, c_src) 552 553 if self.is_dest: 554 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 555 c_dest += '\n\t_numIntDestRegs++;' 556 if self.hasWritePred(): 557 c_dest = '\n\tif (%s) {%s\n\t}' % \ 558 (self.write_predicate, c_dest) 559 560 return c_src + c_dest 561 562 def makeRead(self, predRead): 563 if (self.ctype == 'float' or self.ctype == 'double'): 564 error('Attempt to read integer register as FP') 565 if self.read_code != None: 566 return self.buildReadCode('readIntRegOperand') 567 568 int_reg_val = '' 569 if predRead: 570 int_reg_val = 'xc->readIntRegOperand(this, _sourceIndex++)' 571 if self.hasReadPred(): 572 int_reg_val = '(%s) ? %s : 0' % \ 573 (self.read_predicate, int_reg_val) 574 else: 575 int_reg_val = 'xc->readIntRegOperand(this, %d)' % self.src_reg_idx 576 577 return '%s = %s;\n' % (self.base_name, int_reg_val) 578 579 def makeWrite(self, predWrite): 580 if (self.ctype == 'float' or self.ctype == 'double'): 581 error('Attempt to write integer register as FP') 582 if self.write_code != None: 583 return self.buildWriteCode('setIntRegOperand') 584 585 if predWrite: 586 wp = 'true' 587 if self.hasWritePred(): 588 wp = self.write_predicate 589 590 wcond = 'if (%s)' % (wp) 591 windex = '_destIndex++' 592 else: 593 wcond = '' 594 windex = '%d' % self.dest_reg_idx 595 596 wb = ''' 597 %s 598 { 599 %s final_val = %s; 600 xc->setIntRegOperand(this, %s, final_val);\n 601 if (traceData) { traceData->setData(final_val); } 602 }''' % (wcond, self.ctype, self.base_name, windex) 603 604 return wb 605 606class FloatRegOperand(Operand): 607 reg_class = 'FloatRegClass' 608 609 def isReg(self): 610 return 1 611 612 def isFloatReg(self): 613 return 1 614 615 def makeConstructor(self, predRead, predWrite): 616 c_src = '' 617 c_dest = '' 618 619 if self.is_src: 620 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 621 622 if self.is_dest: 623 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 624 c_dest += '\n\t_numFPDestRegs++;' 625 626 return c_src + c_dest 627 628 def makeRead(self, predRead): 629 bit_select = 0 630 if (self.ctype == 'float' or self.ctype == 'double'): 631 func = 'readFloatRegOperand' 632 else: 633 func = 'readFloatRegOperandBits' 634 if self.read_code != None: 635 return self.buildReadCode(func) 636 637 if predRead: 638 rindex = '_sourceIndex++' 639 else: 640 rindex = '%d' % self.src_reg_idx 641 642 return '%s = xc->%s(this, %s);\n' % \ 643 (self.base_name, func, rindex) 644 645 def makeWrite(self, predWrite): 646 if (self.ctype == 'float' or self.ctype == 'double'): 647 func = 'setFloatRegOperand' 648 else: 649 func = 'setFloatRegOperandBits' 650 if self.write_code != None: 651 return self.buildWriteCode(func) 652 653 if predWrite: 654 wp = '_destIndex++' 655 else: 656 wp = '%d' % self.dest_reg_idx 657 wp = 'xc->%s(this, %s, final_val);' % (func, wp) 658 659 wb = ''' 660 { 661 %s final_val = %s; 662 %s\n 663 if (traceData) { traceData->setData(final_val); } 664 }''' % (self.ctype, self.base_name, wp) 665 return wb 666 667class VecRegOperand(Operand): 668 reg_class = 'VecRegClass' 669 670 def __init__(self, parser, full_name, ext, is_src, is_dest): 671 Operand.__init__(self, parser, full_name, ext, is_src, is_dest) 672 self.elemExt = None 673 self.parser = parser 674 675 def isReg(self): 676 return 1 677 678 def isVecReg(self): 679 return 1 680 681 def makeDeclElem(self, elem_op): 682 (elem_name, elem_ext) = elem_op 683 (elem_spec, dflt_elem_ext, zeroing) = self.elems[elem_name] 684 if elem_ext: 685 ext = elem_ext 686 else: 687 ext = dflt_elem_ext 688 ctype = self.parser.operandTypeMap[ext] 689 return '\n\t%s %s = 0;' % (ctype, elem_name) 690 691 def makeDecl(self): 692 if not self.is_dest and self.is_src: 693 c_decl = '\t/* Vars for %s*/' % (self.base_name) 694 if hasattr(self, 'active_elems'): 695 if self.active_elems: 696 for elem in self.active_elems: 697 c_decl += self.makeDeclElem(elem) 698 return c_decl + '\t/* End vars for %s */\n' % (self.base_name) 699 else: 700 return '' 701 702 def makeConstructor(self, predRead, predWrite): 703 c_src = '' 704 c_dest = '' 705 706 numAccessNeeded = 1 707 708 if self.is_src: 709 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 710 711 if self.is_dest: 712 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 713 c_dest += '\n\t_numVecDestRegs++;' 714 715 return c_src + c_dest 716 717 # Read destination register to write 718 def makeReadWElem(self, elem_op): 719 (elem_name, elem_ext) = elem_op 720 (elem_spec, dflt_elem_ext, zeroing) = self.elems[elem_name] 721 if elem_ext: 722 ext = elem_ext 723 else: 724 ext = dflt_elem_ext 725 ctype = self.parser.operandTypeMap[ext] 726 c_read = '\t\t%s& %s = %s[%s];\n' % \ 727 (ctype, elem_name, self.base_name, elem_spec) 728 return c_read 729 730 def makeReadW(self, predWrite): 731 func = 'getWritableVecRegOperand' 732 if self.read_code != None: 733 return self.buildReadCode(func) 734 735 if predWrite: 736 rindex = '_destIndex++' 737 else: 738 rindex = '%d' % self.dest_reg_idx 739 740 c_readw = '\t\t%s& tmp_d%s = xc->%s(this, %s);\n'\ 741 % ('TheISA::VecRegContainer', rindex, func, rindex) 742 if self.elemExt: 743 c_readw += '\t\tauto %s = tmp_d%s.as<%s>();\n' % (self.base_name, 744 rindex, self.parser.operandTypeMap[self.elemExt]) 745 if self.ext: 746 c_readw += '\t\tauto %s = tmp_d%s.as<%s>();\n' % (self.base_name, 747 rindex, self.parser.operandTypeMap[self.ext]) 748 if hasattr(self, 'active_elems'): 749 if self.active_elems: 750 for elem in self.active_elems: 751 c_readw += self.makeReadWElem(elem) 752 return c_readw 753 754 # Normal source operand read 755 def makeReadElem(self, elem_op, name): 756 (elem_name, elem_ext) = elem_op 757 (elem_spec, dflt_elem_ext, zeroing) = self.elems[elem_name] 758 759 if elem_ext: 760 ext = elem_ext 761 else: 762 ext = dflt_elem_ext 763 ctype = self.parser.operandTypeMap[ext] 764 c_read = '\t\t%s = %s[%s];\n' % \ 765 (elem_name, name, elem_spec) 766 return c_read 767 768 def makeRead(self, predRead): 769 func = 'readVecRegOperand' 770 if self.read_code != None: 771 return self.buildReadCode(func) 772 773 if predRead: 774 rindex = '_sourceIndex++' 775 else: 776 rindex = '%d' % self.src_reg_idx 777 778 name = self.base_name 779 if self.is_dest and self.is_src: 780 name += '_merger' 781 782 c_read = '\t\t%s& tmp_s%s = xc->%s(this, %s);\n' \ 783 % ('const TheISA::VecRegContainer', rindex, func, rindex) 784 # If the parser has detected that elements are being access, create 785 # the appropriate view 786 if self.elemExt: 787 c_read += '\t\tauto %s = tmp_s%s.as<%s>();\n' % \ 788 (name, rindex, self.parser.operandTypeMap[self.elemExt]) 789 if self.ext: 790 c_read += '\t\tauto %s = tmp_s%s.as<%s>();\n' % \ 791 (name, rindex, self.parser.operandTypeMap[self.ext]) 792 if hasattr(self, 'active_elems'): 793 if self.active_elems: 794 for elem in self.active_elems: 795 c_read += self.makeReadElem(elem, name) 796 return c_read 797 798 def makeWrite(self, predWrite): 799 func = 'setVecRegOperand' 800 if self.write_code != None: 801 return self.buildWriteCode(func) 802 803 wb = ''' 804 if (traceData) { 805 warn_once("Vectors not supported yet in tracedata"); 806 /*traceData->setData(final_val);*/ 807 } 808 ''' 809 return wb 810 811 def finalize(self, predRead, predWrite): 812 super(VecRegOperand, self).finalize(predRead, predWrite) 813 if self.is_dest: 814 self.op_rd = self.makeReadW(predWrite) + self.op_rd 815 816class VecElemOperand(Operand): 817 reg_class = 'VectorElemClass' 818 819 def isReg(self): 820 return 1 821 822 def isVecElem(self): 823 return 1 824 825 def makeDecl(self): 826 if self.is_dest and not self.is_src: 827 return '\n\t%s %s;' % (self.ctype, self.base_name) 828 else: 829 return '' 830 831 def makeConstructor(self, predRead, predWrite): 832 c_src = '' 833 c_dest = '' 834 835 numAccessNeeded = 1 836 regId = 'RegId(%s, %s * numVecElemPerVecReg + elemIdx, %s)' % \ 837 (self.reg_class, self.reg_spec) 838 839 if self.is_src: 840 c_src = ('\n\t_srcRegIdx[_numSrcRegs++] = RegId(%s, %s, %s);' % 841 (self.reg_class, self.reg_spec, self.elem_spec)) 842 843 if self.is_dest: 844 c_dest = ('\n\t_destRegIdx[_numDestRegs++] = RegId(%s, %s, %s);' % 845 (self.reg_class, self.reg_spec, self.elem_spec)) 846 c_dest += '\n\t_numVecElemDestRegs++;' 847 return c_src + c_dest 848 849 def makeRead(self, predRead): 850 c_read = ('\n/* Elem is kept inside the operand description */' + 851 '\n\tVecElem %s = xc->readVecElemOperand(this, %d);' % 852 (self.base_name, self.src_reg_idx)) 853 return c_read 854 855 def makeWrite(self, predWrite): 856 c_write = ('\n/* Elem is kept inside the operand description */' + 857 '\n\txc->setVecElemOperand(this, %d, %s);' % 858 (self.dest_reg_idx, self.base_name)) 859 return c_write 860 861class CCRegOperand(Operand): 862 reg_class = 'CCRegClass' 863 864 def isReg(self): 865 return 1 866 867 def isCCReg(self): 868 return 1 869 870 def makeConstructor(self, predRead, predWrite): 871 c_src = '' 872 c_dest = '' 873 874 if self.is_src: 875 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 876 if self.hasReadPred(): 877 c_src = '\n\tif (%s) {%s\n\t}' % \ 878 (self.read_predicate, c_src) 879 880 if self.is_dest: 881 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 882 c_dest += '\n\t_numCCDestRegs++;' 883 if self.hasWritePred(): 884 c_dest = '\n\tif (%s) {%s\n\t}' % \ 885 (self.write_predicate, c_dest) 886 887 return c_src + c_dest 888 889 def makeRead(self, predRead): 890 if (self.ctype == 'float' or self.ctype == 'double'): 891 error('Attempt to read condition-code register as FP') 892 if self.read_code != None: 893 return self.buildReadCode('readCCRegOperand') 894 895 int_reg_val = '' 896 if predRead: 897 int_reg_val = 'xc->readCCRegOperand(this, _sourceIndex++)' 898 if self.hasReadPred(): 899 int_reg_val = '(%s) ? %s : 0' % \ 900 (self.read_predicate, int_reg_val) 901 else: 902 int_reg_val = 'xc->readCCRegOperand(this, %d)' % self.src_reg_idx 903 904 return '%s = %s;\n' % (self.base_name, int_reg_val) 905 906 def makeWrite(self, predWrite): 907 if (self.ctype == 'float' or self.ctype == 'double'): 908 error('Attempt to write condition-code register as FP') 909 if self.write_code != None: 910 return self.buildWriteCode('setCCRegOperand') 911 912 if predWrite: 913 wp = 'true' 914 if self.hasWritePred(): 915 wp = self.write_predicate 916 917 wcond = 'if (%s)' % (wp) 918 windex = '_destIndex++' 919 else: 920 wcond = '' 921 windex = '%d' % self.dest_reg_idx 922 923 wb = ''' 924 %s 925 { 926 %s final_val = %s; 927 xc->setCCRegOperand(this, %s, final_val);\n 928 if (traceData) { traceData->setData(final_val); } 929 }''' % (wcond, self.ctype, self.base_name, windex) 930 931 return wb 932 933class ControlRegOperand(Operand): 934 reg_class = 'MiscRegClass' 935 936 def isReg(self): 937 return 1 938 939 def isControlReg(self): 940 return 1 941 942 def makeConstructor(self, predRead, predWrite): 943 c_src = '' 944 c_dest = '' 945 946 if self.is_src: 947 c_src = src_reg_constructor % (self.reg_class, self.reg_spec) 948 949 if self.is_dest: 950 c_dest = dst_reg_constructor % (self.reg_class, self.reg_spec) 951 952 return c_src + c_dest 953 954 def makeRead(self, predRead): 955 bit_select = 0 956 if (self.ctype == 'float' or self.ctype == 'double'): 957 error('Attempt to read control register as FP') 958 if self.read_code != None: 959 return self.buildReadCode('readMiscRegOperand') 960 961 if predRead: 962 rindex = '_sourceIndex++' 963 else: 964 rindex = '%d' % self.src_reg_idx 965 966 return '%s = xc->readMiscRegOperand(this, %s);\n' % \ 967 (self.base_name, rindex) 968 969 def makeWrite(self, predWrite): 970 if (self.ctype == 'float' or self.ctype == 'double'): 971 error('Attempt to write control register as FP') 972 if self.write_code != None: 973 return self.buildWriteCode('setMiscRegOperand') 974 975 if predWrite: 976 windex = '_destIndex++' 977 else: 978 windex = '%d' % self.dest_reg_idx 979 980 wb = 'xc->setMiscRegOperand(this, %s, %s);\n' % \ 981 (windex, self.base_name) 982 wb += 'if (traceData) { traceData->setData(%s); }' % \ 983 self.base_name 984 985 return wb 986 987class MemOperand(Operand): 988 def isMem(self): 989 return 1 990 991 def makeConstructor(self, predRead, predWrite): 992 return '' 993 994 def makeDecl(self): 995 # Declare memory data variable. 996 return '%s %s;\n' % (self.ctype, self.base_name) 997 998 def makeRead(self, predRead): 999 if self.read_code != None: 1000 return self.buildReadCode() 1001 return '' 1002 1003 def makeWrite(self, predWrite): 1004 if self.write_code != None: 1005 return self.buildWriteCode() 1006 return '' 1007 1008class PCStateOperand(Operand): 1009 def makeConstructor(self, predRead, predWrite): 1010 return '' 1011 1012 def makeRead(self, predRead): 1013 if self.reg_spec: 1014 # A component of the PC state. 1015 return '%s = __parserAutoPCState.%s();\n' % \ 1016 (self.base_name, self.reg_spec) 1017 else: 1018 # The whole PC state itself. 1019 return '%s = xc->pcState();\n' % self.base_name 1020 1021 def makeWrite(self, predWrite): 1022 if self.reg_spec: 1023 # A component of the PC state. 1024 return '__parserAutoPCState.%s(%s);\n' % \ 1025 (self.reg_spec, self.base_name) 1026 else: 1027 # The whole PC state itself. 1028 return 'xc->pcState(%s);\n' % self.base_name 1029 1030 def makeDecl(self): 1031 ctype = 'TheISA::PCState' 1032 if self.isPCPart(): 1033 ctype = self.ctype 1034 # Note that initializations in the declarations are solely 1035 # to avoid 'uninitialized variable' errors from the compiler. 1036 return '%s %s = 0;\n' % (ctype, self.base_name) 1037 1038 def isPCState(self): 1039 return 1 1040 1041class OperandList(object): 1042 '''Find all the operands in the given code block. Returns an operand 1043 descriptor list (instance of class OperandList).''' 1044 def __init__(self, parser, code): 1045 self.items = [] 1046 self.bases = {} 1047 # delete strings and comments so we don't match on operands inside 1048 for regEx in (stringRE, commentRE): 1049 code = regEx.sub('', code) 1050 # search for operands 1051 next_pos = 0 1052 while 1: 1053 match = parser.operandsRE.search(code, next_pos) 1054 if not match: 1055 # no more matches: we're done 1056 break 1057 op = match.groups() 1058 # regexp groups are operand full name, base, and extension 1059 (op_full, op_base, op_ext) = op 1060 # If is a elem operand, define or update the corresponding 1061 # vector operand 1062 isElem = False 1063 if op_base in parser.elemToVector: 1064 isElem = True 1065 elem_op = (op_base, op_ext) 1066 op_base = parser.elemToVector[op_base] 1067 op_ext = '' # use the default one 1068 # if the token following the operand is an assignment, this is 1069 # a destination (LHS), else it's a source (RHS) 1070 is_dest = (assignRE.match(code, match.end()) != None) 1071 is_src = not is_dest 1072 1073 # see if we've already seen this one 1074 op_desc = self.find_base(op_base) 1075 if op_desc: 1076 if op_ext and op_ext != '' and op_desc.ext != op_ext: 1077 error ('Inconsistent extensions for operand %s: %s - %s' \ 1078 % (op_base, op_desc.ext, op_ext)) 1079 op_desc.is_src = op_desc.is_src or is_src 1080 op_desc.is_dest = op_desc.is_dest or is_dest 1081 if isElem: 1082 (elem_base, elem_ext) = elem_op 1083 found = False 1084 for ae in op_desc.active_elems: 1085 (ae_base, ae_ext) = ae 1086 if ae_base == elem_base: 1087 if ae_ext != elem_ext: 1088 error('Inconsistent extensions for elem' 1089 ' operand %s' % elem_base) 1090 else: 1091 found = True 1092 if not found: 1093 op_desc.active_elems.append(elem_op) 1094 else: 1095 # new operand: create new descriptor 1096 op_desc = parser.operandNameMap[op_base](parser, 1097 op_full, op_ext, is_src, is_dest) 1098 # if operand is a vector elem, add the corresponding vector 1099 # operand if not already done 1100 if isElem: 1101 op_desc.elemExt = elem_op[1] 1102 op_desc.active_elems = [elem_op] 1103 self.append(op_desc) 1104 # start next search after end of current match 1105 next_pos = match.end() 1106 self.sort() 1107 # enumerate source & dest register operands... used in building 1108 # constructor later 1109 self.numSrcRegs = 0 1110 self.numDestRegs = 0 1111 self.numFPDestRegs = 0 1112 self.numIntDestRegs = 0 1113 self.numVecDestRegs = 0 1114 self.numCCDestRegs = 0 1115 self.numMiscDestRegs = 0 1116 self.memOperand = None 1117 1118 # Flags to keep track if one or more operands are to be read/written 1119 # conditionally. 1120 self.predRead = False 1121 self.predWrite = False 1122 1123 for op_desc in self.items: 1124 if op_desc.isReg(): 1125 if op_desc.is_src: 1126 op_desc.src_reg_idx = self.numSrcRegs 1127 self.numSrcRegs += 1 1128 if op_desc.is_dest: 1129 op_desc.dest_reg_idx = self.numDestRegs 1130 self.numDestRegs += 1 1131 if op_desc.isFloatReg(): 1132 self.numFPDestRegs += 1 1133 elif op_desc.isIntReg(): 1134 self.numIntDestRegs += 1 1135 elif op_desc.isVecReg(): 1136 self.numVecDestRegs += 1 1137 elif op_desc.isCCReg(): 1138 self.numCCDestRegs += 1 1139 elif op_desc.isControlReg(): 1140 self.numMiscDestRegs += 1 1141 elif op_desc.isMem(): 1142 if self.memOperand: 1143 error("Code block has more than one memory operand.") 1144 self.memOperand = op_desc 1145 1146 # Check if this operand has read/write predication. If true, then 1147 # the microop will dynamically index source/dest registers. 1148 self.predRead = self.predRead or op_desc.hasReadPred() 1149 self.predWrite = self.predWrite or op_desc.hasWritePred() 1150 1151 if parser.maxInstSrcRegs < self.numSrcRegs: 1152 parser.maxInstSrcRegs = self.numSrcRegs 1153 if parser.maxInstDestRegs < self.numDestRegs: 1154 parser.maxInstDestRegs = self.numDestRegs 1155 if parser.maxMiscDestRegs < self.numMiscDestRegs: 1156 parser.maxMiscDestRegs = self.numMiscDestRegs 1157 1158 # now make a final pass to finalize op_desc fields that may depend 1159 # on the register enumeration 1160 for op_desc in self.items: 1161 op_desc.finalize(self.predRead, self.predWrite) 1162 1163 def __len__(self): 1164 return len(self.items) 1165 1166 def __getitem__(self, index): 1167 return self.items[index] 1168 1169 def append(self, op_desc): 1170 self.items.append(op_desc) 1171 self.bases[op_desc.base_name] = op_desc 1172 1173 def find_base(self, base_name): 1174 # like self.bases[base_name], but returns None if not found 1175 # (rather than raising exception) 1176 return self.bases.get(base_name) 1177 1178 # internal helper function for concat[Some]Attr{Strings|Lists} 1179 def __internalConcatAttrs(self, attr_name, filter, result): 1180 for op_desc in self.items: 1181 if filter(op_desc): 1182 result += getattr(op_desc, attr_name) 1183 return result 1184 1185 # return a single string that is the concatenation of the (string) 1186 # values of the specified attribute for all operands 1187 def concatAttrStrings(self, attr_name): 1188 return self.__internalConcatAttrs(attr_name, lambda x: 1, '') 1189 1190 # like concatAttrStrings, but only include the values for the operands 1191 # for which the provided filter function returns true 1192 def concatSomeAttrStrings(self, filter, attr_name): 1193 return self.__internalConcatAttrs(attr_name, filter, '') 1194 1195 # return a single list that is the concatenation of the (list) 1196 # values of the specified attribute for all operands 1197 def concatAttrLists(self, attr_name): 1198 return self.__internalConcatAttrs(attr_name, lambda x: 1, []) 1199 1200 # like concatAttrLists, but only include the values for the operands 1201 # for which the provided filter function returns true 1202 def concatSomeAttrLists(self, filter, attr_name): 1203 return self.__internalConcatAttrs(attr_name, filter, []) 1204 1205 def sort(self): 1206 self.items.sort(lambda a, b: a.sort_pri - b.sort_pri) 1207 1208class SubOperandList(OperandList): 1209 '''Find all the operands in the given code block. Returns an operand 1210 descriptor list (instance of class OperandList).''' 1211 def __init__(self, parser, code, master_list): 1212 self.items = [] 1213 self.bases = {} 1214 # delete strings and comments so we don't match on operands inside 1215 for regEx in (stringRE, commentRE): 1216 code = regEx.sub('', code) 1217 # search for operands 1218 next_pos = 0 1219 while 1: 1220 match = parser.operandsRE.search(code, next_pos) 1221 if not match: 1222 # no more matches: we're done 1223 break 1224 op = match.groups() 1225 # regexp groups are operand full name, base, and extension 1226 (op_full, op_base, op_ext) = op 1227 # If is a elem operand, define or update the corresponding 1228 # vector operand 1229 if op_base in parser.elemToVector: 1230 elem_op = op_base 1231 op_base = parser.elemToVector[elem_op] 1232 # find this op in the master list 1233 op_desc = master_list.find_base(op_base) 1234 if not op_desc: 1235 error('Found operand %s which is not in the master list!' 1236 % op_base) 1237 else: 1238 # See if we've already found this operand 1239 op_desc = self.find_base(op_base) 1240 if not op_desc: 1241 # if not, add a reference to it to this sub list 1242 self.append(master_list.bases[op_base]) 1243 1244 # start next search after end of current match 1245 next_pos = match.end() 1246 self.sort() 1247 self.memOperand = None 1248 # Whether the whole PC needs to be read so parts of it can be accessed 1249 self.readPC = False 1250 # Whether the whole PC needs to be written after parts of it were 1251 # changed 1252 self.setPC = False 1253 # Whether this instruction manipulates the whole PC or parts of it. 1254 # Mixing the two is a bad idea and flagged as an error. 1255 self.pcPart = None 1256 1257 # Flags to keep track if one or more operands are to be read/written 1258 # conditionally. 1259 self.predRead = False 1260 self.predWrite = False 1261 1262 for op_desc in self.items: 1263 if op_desc.isPCPart(): 1264 self.readPC = True 1265 if op_desc.is_dest: 1266 self.setPC = True 1267 1268 if op_desc.isPCState(): 1269 if self.pcPart is not None: 1270 if self.pcPart and not op_desc.isPCPart() or \ 1271 not self.pcPart and op_desc.isPCPart(): 1272 error("Mixed whole and partial PC state operands.") 1273 self.pcPart = op_desc.isPCPart() 1274 1275 if op_desc.isMem(): 1276 if self.memOperand: 1277 error("Code block has more than one memory operand.") 1278 self.memOperand = op_desc 1279 1280 # Check if this operand has read/write predication. If true, then 1281 # the microop will dynamically index source/dest registers. 1282 self.predRead = self.predRead or op_desc.hasReadPred() 1283 self.predWrite = self.predWrite or op_desc.hasWritePred() 1284 1285# Regular expression object to match C++ strings 1286stringRE = re.compile(r'"([^"\\]|\\.)*"') 1287 1288# Regular expression object to match C++ comments 1289# (used in findOperands()) 1290commentRE = re.compile(r'(^)?[^\S\n]*/(?:\*(.*?)\*/[^\S\n]*|/[^\n]*)($)?', 1291 re.DOTALL | re.MULTILINE) 1292 1293# Regular expression object to match assignment statements (used in 1294# findOperands()). If the code immediately following the first 1295# appearance of the operand matches this regex, then the operand 1296# appears to be on the LHS of an assignment, and is thus a 1297# destination. basically we're looking for an '=' that's not '=='. 1298# The heinous tangle before that handles the case where the operand 1299# has an array subscript. 1300assignRE = re.compile(r'(\[[^\]]+\])?\s*=(?!=)', re.MULTILINE) 1301 1302def makeFlagConstructor(flag_list): 1303 if len(flag_list) == 0: 1304 return '' 1305 # filter out repeated flags 1306 flag_list.sort() 1307 i = 1 1308 while i < len(flag_list): 1309 if flag_list[i] == flag_list[i-1]: 1310 del flag_list[i] 1311 else: 1312 i += 1 1313 pre = '\n\tflags[' 1314 post = '] = true;' 1315 code = pre + string.join(flag_list, post + pre) + post 1316 return code 1317 1318# Assume all instruction flags are of the form 'IsFoo' 1319instFlagRE = re.compile(r'Is.*') 1320 1321# OpClass constants end in 'Op' except No_OpClass 1322opClassRE = re.compile(r'.*Op|No_OpClass') 1323 1324class InstObjParams(object): 1325 def __init__(self, parser, mnem, class_name, base_class = '', 1326 snippets = {}, opt_args = []): 1327 self.mnemonic = mnem 1328 self.class_name = class_name 1329 self.base_class = base_class 1330 if not isinstance(snippets, dict): 1331 snippets = {'code' : snippets} 1332 compositeCode = ' '.join(map(str, snippets.values())) 1333 self.snippets = snippets 1334 1335 self.operands = OperandList(parser, compositeCode) 1336 1337 # The header of the constructor declares the variables to be used 1338 # in the body of the constructor. 1339 header = '' 1340 header += '\n\t_numSrcRegs = 0;' 1341 header += '\n\t_numDestRegs = 0;' 1342 header += '\n\t_numFPDestRegs = 0;' 1343 header += '\n\t_numVecDestRegs = 0;' 1344 header += '\n\t_numVecElemDestRegs = 0;' 1345 header += '\n\t_numIntDestRegs = 0;' 1346 header += '\n\t_numCCDestRegs = 0;' 1347 1348 self.constructor = header + \ 1349 self.operands.concatAttrStrings('constructor') 1350 1351 self.flags = self.operands.concatAttrLists('flags') 1352 1353 self.op_class = None 1354 1355 # Optional arguments are assumed to be either StaticInst flags 1356 # or an OpClass value. To avoid having to import a complete 1357 # list of these values to match against, we do it ad-hoc 1358 # with regexps. 1359 for oa in opt_args: 1360 if instFlagRE.match(oa): 1361 self.flags.append(oa) 1362 elif opClassRE.match(oa): 1363 self.op_class = oa 1364 else: 1365 error('InstObjParams: optional arg "%s" not recognized ' 1366 'as StaticInst::Flag or OpClass.' % oa) 1367 1368 # Make a basic guess on the operand class if not set. 1369 # These are good enough for most cases. 1370 if not self.op_class: 1371 if 'IsStore' in self.flags: 1372 # The order matters here: 'IsFloating' and 'IsInteger' are 1373 # usually set in FP instructions because of the base 1374 # register 1375 if 'IsFloating' in self.flags: 1376 self.op_class = 'FloatMemWriteOp' 1377 else: 1378 self.op_class = 'MemWriteOp' 1379 elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags: 1380 # The order matters here: 'IsFloating' and 'IsInteger' are 1381 # usually set in FP instructions because of the base 1382 # register 1383 if 'IsFloating' in self.flags: 1384 self.op_class = 'FloatMemReadOp' 1385 else: 1386 self.op_class = 'MemReadOp' 1387 elif 'IsFloating' in self.flags: 1388 self.op_class = 'FloatAddOp' 1389 elif 'IsVector' in self.flags: 1390 self.op_class = 'SimdAddOp' 1391 else: 1392 self.op_class = 'IntAluOp' 1393 1394 # add flag initialization to contructor here to include 1395 # any flags added via opt_args 1396 self.constructor += makeFlagConstructor(self.flags) 1397 1398 # if 'IsFloating' is set, add call to the FP enable check 1399 # function (which should be provided by isa_desc via a declare) 1400 # if 'IsVector' is set, add call to the Vector enable check 1401 # function (which should be provided by isa_desc via a declare) 1402 if 'IsFloating' in self.flags: 1403 self.fp_enable_check = 'fault = checkFpEnableFault(xc);' 1404 elif 'IsVector' in self.flags: 1405 self.fp_enable_check = 'fault = checkVecEnableFault(xc);' 1406 else: 1407 self.fp_enable_check = '' 1408 1409############## 1410# Stack: a simple stack object. Used for both formats (formatStack) 1411# and default cases (defaultStack). Simply wraps a list to give more 1412# stack-like syntax and enable initialization with an argument list 1413# (as opposed to an argument that's a list). 1414 1415class Stack(list): 1416 def __init__(self, *items): 1417 list.__init__(self, items) 1418 1419 def push(self, item): 1420 self.append(item); 1421 1422 def top(self): 1423 return self[-1] 1424 1425# Format a file include stack backtrace as a string 1426def backtrace(filename_stack): 1427 fmt = "In file included from %s:" 1428 return "\n".join([fmt % f for f in filename_stack]) 1429 1430 1431####################### 1432# 1433# LineTracker: track filenames along with line numbers in PLY lineno fields 1434# PLY explicitly doesn't do anything with 'lineno' except propagate 1435# it. This class lets us tie filenames with the line numbers with a 1436# minimum of disruption to existing increment code. 1437# 1438 1439class LineTracker(object): 1440 def __init__(self, filename, lineno=1): 1441 self.filename = filename 1442 self.lineno = lineno 1443 1444 # Overload '+=' for increments. We need to create a new object on 1445 # each update else every token ends up referencing the same 1446 # constantly incrementing instance. 1447 def __iadd__(self, incr): 1448 return LineTracker(self.filename, self.lineno + incr) 1449 1450 def __str__(self): 1451 return "%s:%d" % (self.filename, self.lineno) 1452 1453 # In case there are places where someone really expects a number 1454 def __int__(self): 1455 return self.lineno 1456 1457 1458####################### 1459# 1460# ISA Parser 1461# parses ISA DSL and emits C++ headers and source 1462# 1463 1464class ISAParser(Grammar): 1465 class CpuModel(object): 1466 def __init__(self, name, filename, includes, strings): 1467 self.name = name 1468 self.filename = filename 1469 self.includes = includes 1470 self.strings = strings 1471 1472 def __init__(self, output_dir): 1473 super(ISAParser, self).__init__() 1474 self.output_dir = output_dir 1475 1476 self.filename = None # for output file watermarking/scaremongering 1477 1478 self.cpuModels = [ 1479 ISAParser.CpuModel('ExecContext', 1480 'generic_cpu_exec.cc', 1481 '#include "cpu/exec_context.hh"', 1482 { "CPU_exec_context" : "ExecContext" }), 1483 ] 1484 1485 # variable to hold templates 1486 self.templateMap = {} 1487 1488 # This dictionary maps format name strings to Format objects. 1489 self.formatMap = {} 1490 1491 # Track open files and, if applicable, how many chunks it has been 1492 # split into so far. 1493 self.files = {} 1494 self.splits = {} 1495 1496 # isa_name / namespace identifier from namespace declaration. 1497 # before the namespace declaration, None. 1498 self.isa_name = None 1499 self.namespace = None 1500 1501 # The format stack. 1502 self.formatStack = Stack(NoFormat()) 1503 1504 # The default case stack. 1505 self.defaultStack = Stack(None) 1506 1507 # Stack that tracks current file and line number. Each 1508 # element is a tuple (filename, lineno) that records the 1509 # *current* filename and the line number in the *previous* 1510 # file where it was included. 1511 self.fileNameStack = Stack() 1512 1513 symbols = ('makeList', 're', 'string') 1514 self.exportContext = dict([(s, eval(s)) for s in symbols]) 1515 1516 self.maxInstSrcRegs = 0 1517 self.maxInstDestRegs = 0 1518 self.maxMiscDestRegs = 0 1519 1520 def __getitem__(self, i): # Allow object (self) to be 1521 return getattr(self, i) # passed to %-substitutions 1522 1523 # Change the file suffix of a base filename: 1524 # (e.g.) decoder.cc -> decoder-g.cc.inc for 'global' outputs 1525 def suffixize(self, s, sec): 1526 extn = re.compile('(\.[^\.]+)$') # isolate extension 1527 if self.namespace: 1528 return extn.sub(r'-ns\1.inc', s) # insert some text on either side 1529 else: 1530 return extn.sub(r'-g\1.inc', s) 1531 1532 # Get the file object for emitting code into the specified section 1533 # (header, decoder, exec, decode_block). 1534 def get_file(self, section): 1535 if section == 'decode_block': 1536 filename = 'decode-method.cc.inc' 1537 else: 1538 if section == 'header': 1539 file = 'decoder.hh' 1540 else: 1541 file = '%s.cc' % section 1542 filename = self.suffixize(file, section) 1543 try: 1544 return self.files[filename] 1545 except KeyError: pass 1546 1547 f = self.open(filename) 1548 self.files[filename] = f 1549 1550 # The splittable files are the ones with many independent 1551 # per-instruction functions - the decoder's instruction constructors 1552 # and the instruction execution (execute()) methods. These both have 1553 # the suffix -ns.cc.inc, meaning they are within the namespace part 1554 # of the ISA, contain object-emitting C++ source, and are included 1555 # into other top-level files. These are the files that need special 1556 # #define's to allow parts of them to be compiled separately. Rather 1557 # than splitting the emissions into separate files, the monolithic 1558 # output of the ISA parser is maintained, but the value (or lack 1559 # thereof) of the __SPLIT definition during C preprocessing will 1560 # select the different chunks. If no 'split' directives are used, 1561 # the cpp emissions have no effect. 1562 if re.search('-ns.cc.inc$', filename): 1563 print >>f, '#if !defined(__SPLIT) || (__SPLIT == 1)' 1564 self.splits[f] = 1 1565 # ensure requisite #include's 1566 elif filename == 'decoder-g.hh.inc': 1567 print >>f, '#include "base/bitfield.hh"' 1568 1569 return f 1570 1571 # Weave together the parts of the different output sections by 1572 # #include'ing them into some very short top-level .cc/.hh files. 1573 # These small files make it much clearer how this tool works, since 1574 # you directly see the chunks emitted as files that are #include'd. 1575 def write_top_level_files(self): 1576 dep = self.open('inc.d', bare=True) 1577 1578 # decoder header - everything depends on this 1579 file = 'decoder.hh' 1580 with self.open(file) as f: 1581 inc = [] 1582 1583 fn = 'decoder-g.hh.inc' 1584 assert(fn in self.files) 1585 f.write('#include "%s"\n' % fn) 1586 inc.append(fn) 1587 1588 fn = 'decoder-ns.hh.inc' 1589 assert(fn in self.files) 1590 f.write('namespace %s {\n#include "%s"\n}\n' 1591 % (self.namespace, fn)) 1592 inc.append(fn) 1593 1594 print >>dep, file+':', ' '.join(inc) 1595 1596 # decoder method - cannot be split 1597 file = 'decoder.cc' 1598 with self.open(file) as f: 1599 inc = [] 1600 1601 fn = 'decoder-g.cc.inc' 1602 assert(fn in self.files) 1603 f.write('#include "%s"\n' % fn) 1604 inc.append(fn) 1605 1606 fn = 'decoder.hh' 1607 f.write('#include "%s"\n' % fn) 1608 inc.append(fn) 1609 1610 fn = 'decode-method.cc.inc' 1611 # is guaranteed to have been written for parse to complete 1612 f.write('#include "%s"\n' % fn) 1613 inc.append(fn) 1614 1615 print >>dep, file+':', ' '.join(inc) 1616 1617 extn = re.compile('(\.[^\.]+)$') 1618 1619 # instruction constructors 1620 splits = self.splits[self.get_file('decoder')] 1621 file_ = 'inst-constrs.cc' 1622 for i in range(1, splits+1): 1623 if splits > 1: 1624 file = extn.sub(r'-%d\1' % i, file_) 1625 else: 1626 file = file_ 1627 with self.open(file) as f: 1628 inc = [] 1629 1630 fn = 'decoder-g.cc.inc' 1631 assert(fn in self.files) 1632 f.write('#include "%s"\n' % fn) 1633 inc.append(fn) 1634 1635 fn = 'decoder.hh' 1636 f.write('#include "%s"\n' % fn) 1637 inc.append(fn) 1638 1639 fn = 'decoder-ns.cc.inc' 1640 assert(fn in self.files) 1641 print >>f, 'namespace %s {' % self.namespace 1642 if splits > 1: 1643 print >>f, '#define __SPLIT %u' % i 1644 print >>f, '#include "%s"' % fn 1645 print >>f, '}' 1646 inc.append(fn) 1647 1648 print >>dep, file+':', ' '.join(inc) 1649 1650 # instruction execution per-CPU model 1651 splits = self.splits[self.get_file('exec')] 1652 for cpu in self.cpuModels: 1653 for i in range(1, splits+1): 1654 if splits > 1: 1655 file = extn.sub(r'_%d\1' % i, cpu.filename) 1656 else: 1657 file = cpu.filename 1658 with self.open(file) as f: 1659 inc = [] 1660 1661 fn = 'exec-g.cc.inc' 1662 assert(fn in self.files) 1663 f.write('#include "%s"\n' % fn) 1664 inc.append(fn) 1665 1666 f.write(cpu.includes+"\n") 1667 1668 fn = 'decoder.hh' 1669 f.write('#include "%s"\n' % fn) 1670 inc.append(fn) 1671 1672 fn = 'exec-ns.cc.inc' 1673 assert(fn in self.files) 1674 print >>f, 'namespace %s {' % self.namespace 1675 print >>f, '#define CPU_EXEC_CONTEXT %s' \ 1676 % cpu.strings['CPU_exec_context'] 1677 if splits > 1: 1678 print >>f, '#define __SPLIT %u' % i 1679 print >>f, '#include "%s"' % fn 1680 print >>f, '}' 1681 inc.append(fn) 1682 1683 inc.append("decoder.hh") 1684 print >>dep, file+':', ' '.join(inc) 1685 1686 # max_inst_regs.hh 1687 self.update('max_inst_regs.hh', 1688 '''namespace %(namespace)s { 1689 const int MaxInstSrcRegs = %(maxInstSrcRegs)d; 1690 const int MaxInstDestRegs = %(maxInstDestRegs)d; 1691 const int MaxMiscDestRegs = %(maxMiscDestRegs)d;\n}\n''' % self) 1692 print >>dep, 'max_inst_regs.hh:' 1693 1694 dep.close() 1695 1696 1697 scaremonger_template ='''// DO NOT EDIT 1698// This file was automatically generated from an ISA description: 1699// %(filename)s 1700 1701'''; 1702 1703 ##################################################################### 1704 # 1705 # Lexer 1706 # 1707 # The PLY lexer module takes two things as input: 1708 # - A list of token names (the string list 'tokens') 1709 # - A regular expression describing a match for each token. The 1710 # regexp for token FOO can be provided in two ways: 1711 # - as a string variable named t_FOO 1712 # - as the doc string for a function named t_FOO. In this case, 1713 # the function is also executed, allowing an action to be 1714 # associated with each token match. 1715 # 1716 ##################################################################### 1717 1718 # Reserved words. These are listed separately as they are matched 1719 # using the same regexp as generic IDs, but distinguished in the 1720 # t_ID() function. The PLY documentation suggests this approach. 1721 reserved = ( 1722 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT', 1723 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS', 1724 'OUTPUT', 'SIGNED', 'SPLIT', 'TEMPLATE' 1725 ) 1726 1727 # List of tokens. The lex module requires this. 1728 tokens = reserved + ( 1729 # identifier 1730 'ID', 1731 1732 # integer literal 1733 'INTLIT', 1734 1735 # string literal 1736 'STRLIT', 1737 1738 # code literal 1739 'CODELIT', 1740 1741 # ( ) [ ] { } < > , ; . : :: * 1742 'LPAREN', 'RPAREN', 1743 'LBRACKET', 'RBRACKET', 1744 'LBRACE', 'RBRACE', 1745 'LESS', 'GREATER', 'EQUALS', 1746 'COMMA', 'SEMI', 'DOT', 'COLON', 'DBLCOLON', 1747 'ASTERISK', 1748 1749 # C preprocessor directives 1750 'CPPDIRECTIVE' 1751 1752 # The following are matched but never returned. commented out to 1753 # suppress PLY warning 1754 # newfile directive 1755 # 'NEWFILE', 1756 1757 # endfile directive 1758 # 'ENDFILE' 1759 ) 1760 1761 # Regular expressions for token matching 1762 t_LPAREN = r'\(' 1763 t_RPAREN = r'\)' 1764 t_LBRACKET = r'\[' 1765 t_RBRACKET = r'\]' 1766 t_LBRACE = r'\{' 1767 t_RBRACE = r'\}' 1768 t_LESS = r'\<' 1769 t_GREATER = r'\>' 1770 t_EQUALS = r'=' 1771 t_COMMA = r',' 1772 t_SEMI = r';' 1773 t_DOT = r'\.' 1774 t_COLON = r':' 1775 t_DBLCOLON = r'::' 1776 t_ASTERISK = r'\*' 1777 1778 # Identifiers and reserved words 1779 reserved_map = { } 1780 for r in reserved: 1781 reserved_map[r.lower()] = r 1782 1783 def t_ID(self, t): 1784 r'[A-Za-z_]\w*' 1785 t.type = self.reserved_map.get(t.value, 'ID') 1786 return t 1787 1788 # Integer literal 1789 def t_INTLIT(self, t): 1790 r'-?(0x[\da-fA-F]+)|\d+' 1791 try: 1792 t.value = int(t.value,0) 1793 except ValueError: 1794 error(t.lexer.lineno, 'Integer value "%s" too large' % t.value) 1795 t.value = 0 1796 return t 1797 1798 # String literal. Note that these use only single quotes, and 1799 # can span multiple lines. 1800 def t_STRLIT(self, t): 1801 r"(?m)'([^'])+'" 1802 # strip off quotes 1803 t.value = t.value[1:-1] 1804 t.lexer.lineno += t.value.count('\n') 1805 return t 1806 1807 1808 # "Code literal"... like a string literal, but delimiters are 1809 # '{{' and '}}' so they get formatted nicely under emacs c-mode 1810 def t_CODELIT(self, t): 1811 r"(?m)\{\{([^\}]|}(?!\}))+\}\}" 1812 # strip off {{ & }} 1813 t.value = t.value[2:-2] 1814 t.lexer.lineno += t.value.count('\n') 1815 return t 1816 1817 def t_CPPDIRECTIVE(self, t): 1818 r'^\#[^\#].*\n' 1819 t.lexer.lineno += t.value.count('\n') 1820 return t 1821 1822 def t_NEWFILE(self, t): 1823 r'^\#\#newfile\s+"[^"]*"\n' 1824 self.fileNameStack.push(t.lexer.lineno) 1825 t.lexer.lineno = LineTracker(t.value[11:-2]) 1826 1827 def t_ENDFILE(self, t): 1828 r'^\#\#endfile\n' 1829 t.lexer.lineno = self.fileNameStack.pop() 1830 1831 # 1832 # The functions t_NEWLINE, t_ignore, and t_error are 1833 # special for the lex module. 1834 # 1835 1836 # Newlines 1837 def t_NEWLINE(self, t): 1838 r'\n+' 1839 t.lexer.lineno += t.value.count('\n') 1840 1841 # Comments 1842 def t_comment(self, t): 1843 r'//.*' 1844 1845 # Completely ignored characters 1846 t_ignore = ' \t\x0c' 1847 1848 # Error handler 1849 def t_error(self, t): 1850 error(t.lexer.lineno, "illegal character '%s'" % t.value[0]) 1851 t.skip(1) 1852 1853 ##################################################################### 1854 # 1855 # Parser 1856 # 1857 # Every function whose name starts with 'p_' defines a grammar 1858 # rule. The rule is encoded in the function's doc string, while 1859 # the function body provides the action taken when the rule is 1860 # matched. The argument to each function is a list of the values 1861 # of the rule's symbols: t[0] for the LHS, and t[1..n] for the 1862 # symbols on the RHS. For tokens, the value is copied from the 1863 # t.value attribute provided by the lexer. For non-terminals, the 1864 # value is assigned by the producing rule; i.e., the job of the 1865 # grammar rule function is to set the value for the non-terminal 1866 # on the LHS (by assigning to t[0]). 1867 ##################################################################### 1868 1869 # The LHS of the first grammar rule is used as the start symbol 1870 # (in this case, 'specification'). Note that this rule enforces 1871 # that there will be exactly one namespace declaration, with 0 or 1872 # more global defs/decls before and after it. The defs & decls 1873 # before the namespace decl will be outside the namespace; those 1874 # after will be inside. The decoder function is always inside the 1875 # namespace. 1876 def p_specification(self, t): 1877 'specification : opt_defs_and_outputs top_level_decode_block' 1878 1879 for f in self.splits.iterkeys(): 1880 f.write('\n#endif\n') 1881 1882 for f in self.files.itervalues(): # close ALL the files; 1883 f.close() # not doing so can cause compilation to fail 1884 1885 self.write_top_level_files() 1886 1887 t[0] = True 1888 1889 # 'opt_defs_and_outputs' is a possibly empty sequence of def and/or 1890 # output statements. Its productions do the hard work of eventually 1891 # instantiating a GenCode, which are generally emitted (written to disk) 1892 # as soon as possible, except for the decode_block, which has to be 1893 # accumulated into one large function of nested switch/case blocks. 1894 def p_opt_defs_and_outputs_0(self, t): 1895 'opt_defs_and_outputs : empty' 1896 1897 def p_opt_defs_and_outputs_1(self, t): 1898 'opt_defs_and_outputs : defs_and_outputs' 1899 1900 def p_defs_and_outputs_0(self, t): 1901 'defs_and_outputs : def_or_output' 1902 1903 def p_defs_and_outputs_1(self, t): 1904 'defs_and_outputs : defs_and_outputs def_or_output' 1905 1906 # The list of possible definition/output statements. 1907 # They are all processed as they are seen. 1908 def p_def_or_output(self, t): 1909 '''def_or_output : name_decl 1910 | def_format 1911 | def_bitfield 1912 | def_bitfield_struct 1913 | def_template 1914 | def_operand_types 1915 | def_operands 1916 | output 1917 | global_let 1918 | split''' 1919 1920 # Utility function used by both invocations of splitting - explicit 1921 # 'split' keyword and split() function inside "let {{ }};" blocks. 1922 def split(self, sec, write=False): 1923 assert(sec != 'header' and "header cannot be split") 1924 1925 f = self.get_file(sec) 1926 self.splits[f] += 1 1927 s = '\n#endif\n#if __SPLIT == %u\n' % self.splits[f] 1928 if write: 1929 f.write(s) 1930 else: 1931 return s 1932 1933 # split output file to reduce compilation time 1934 def p_split(self, t): 1935 'split : SPLIT output_type SEMI' 1936 assert(self.isa_name and "'split' not allowed before namespace decl") 1937 1938 self.split(t[2], True) 1939 1940 def p_output_type(self, t): 1941 '''output_type : DECODER 1942 | HEADER 1943 | EXEC''' 1944 t[0] = t[1] 1945 1946 # ISA name declaration looks like "namespace <foo>;" 1947 def p_name_decl(self, t): 1948 'name_decl : NAMESPACE ID SEMI' 1949 assert(self.isa_name == None and "Only 1 namespace decl permitted") 1950 self.isa_name = t[2] 1951 self.namespace = t[2] + 'Inst' 1952 1953 # Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied 1954 # directly to the appropriate output section. 1955 1956 # Massage output block by substituting in template definitions and 1957 # bit operators. We handle '%'s embedded in the string that don't 1958 # indicate template substitutions (or CPU-specific symbols, which 1959 # get handled in GenCode) by doubling them first so that the 1960 # format operation will reduce them back to single '%'s. 1961 def process_output(self, s): 1962 s = self.protectNonSubstPercents(s) 1963 # protects cpu-specific symbols too 1964 s = self.protectCpuSymbols(s) 1965 return substBitOps(s % self.templateMap) 1966 1967 def p_output(self, t): 1968 'output : OUTPUT output_type CODELIT SEMI' 1969 kwargs = { t[2]+'_output' : self.process_output(t[3]) } 1970 GenCode(self, **kwargs).emit() 1971 1972 # global let blocks 'let {{...}}' (Python code blocks) are 1973 # executed directly when seen. Note that these execute in a 1974 # special variable context 'exportContext' to prevent the code 1975 # from polluting this script's namespace. 1976 def p_global_let(self, t): 1977 'global_let : LET CODELIT SEMI' 1978 def _split(sec): 1979 return self.split(sec) 1980 self.updateExportContext() 1981 self.exportContext["header_output"] = '' 1982 self.exportContext["decoder_output"] = '' 1983 self.exportContext["exec_output"] = '' 1984 self.exportContext["decode_block"] = '' 1985 self.exportContext["split"] = _split 1986 split_setup = ''' 1987def wrap(func): 1988 def split(sec): 1989 globals()[sec + '_output'] += func(sec) 1990 return split 1991split = wrap(split) 1992del wrap 1993''' 1994 # This tricky setup (immediately above) allows us to just write 1995 # (e.g.) "split('exec')" in the Python code and the split #ifdef's 1996 # will automatically be added to the exec_output variable. The inner 1997 # Python execution environment doesn't know about the split points, 1998 # so we carefully inject and wrap a closure that can retrieve the 1999 # next split's #define from the parser and add it to the current 2000 # emission-in-progress. 2001 try: 2002 exec split_setup+fixPythonIndentation(t[2]) in self.exportContext 2003 except Exception, exc: 2004 if debug: 2005 raise 2006 error(t.lineno(1), 'In global let block: %s' % exc) 2007 GenCode(self, 2008 header_output=self.exportContext["header_output"], 2009 decoder_output=self.exportContext["decoder_output"], 2010 exec_output=self.exportContext["exec_output"], 2011 decode_block=self.exportContext["decode_block"]).emit() 2012 2013 # Define the mapping from operand type extensions to C++ types and 2014 # bit widths (stored in operandTypeMap). 2015 def p_def_operand_types(self, t): 2016 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI' 2017 try: 2018 self.operandTypeMap = eval('{' + t[3] + '}') 2019 except Exception, exc: 2020 if debug: 2021 raise 2022 error(t.lineno(1), 2023 'In def operand_types: %s' % exc) 2024 2025 # Define the mapping from operand names to operand classes and 2026 # other traits. Stored in operandNameMap. 2027 def p_def_operands(self, t): 2028 'def_operands : DEF OPERANDS CODELIT SEMI' 2029 if not hasattr(self, 'operandTypeMap'): 2030 error(t.lineno(1), 2031 'error: operand types must be defined before operands') 2032 try: 2033 user_dict = eval('{' + t[3] + '}', self.exportContext) 2034 except Exception, exc: 2035 if debug: 2036 raise 2037 error(t.lineno(1), 'In def operands: %s' % exc) 2038 self.buildOperandNameMap(user_dict, t.lexer.lineno) 2039 2040 # A bitfield definition looks like: 2041 # 'def [signed] bitfield <ID> [<first>:<last>]' 2042 # This generates a preprocessor macro in the output file. 2043 def p_def_bitfield_0(self, t): 2044 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI' 2045 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8]) 2046 if (t[2] == 'signed'): 2047 expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr) 2048 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 2049 GenCode(self, header_output=hash_define).emit() 2050 2051 # alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]' 2052 def p_def_bitfield_1(self, t): 2053 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI' 2054 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6]) 2055 if (t[2] == 'signed'): 2056 expr = 'sext<%d>(%s)' % (1, expr) 2057 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 2058 GenCode(self, header_output=hash_define).emit() 2059 2060 # alternate form for structure member: 'def bitfield <ID> <ID>' 2061 def p_def_bitfield_struct(self, t): 2062 'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI' 2063 if (t[2] != ''): 2064 error(t.lineno(1), 2065 'error: structure bitfields are always unsigned.') 2066 expr = 'machInst.%s' % t[5] 2067 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 2068 GenCode(self, header_output=hash_define).emit() 2069 2070 def p_id_with_dot_0(self, t): 2071 'id_with_dot : ID' 2072 t[0] = t[1] 2073 2074 def p_id_with_dot_1(self, t): 2075 'id_with_dot : ID DOT id_with_dot' 2076 t[0] = t[1] + t[2] + t[3] 2077 2078 def p_opt_signed_0(self, t): 2079 'opt_signed : SIGNED' 2080 t[0] = t[1] 2081 2082 def p_opt_signed_1(self, t): 2083 'opt_signed : empty' 2084 t[0] = '' 2085 2086 def p_def_template(self, t): 2087 'def_template : DEF TEMPLATE ID CODELIT SEMI' 2088 if t[3] in self.templateMap: 2089 print "warning: template %s already defined" % t[3] 2090 self.templateMap[t[3]] = Template(self, t[4]) 2091 2092 # An instruction format definition looks like 2093 # "def format <fmt>(<params>) {{...}};" 2094 def p_def_format(self, t): 2095 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI' 2096 (id, params, code) = (t[3], t[5], t[7]) 2097 self.defFormat(id, params, code, t.lexer.lineno) 2098 2099 # The formal parameter list for an instruction format is a 2100 # possibly empty list of comma-separated parameters. Positional 2101 # (standard, non-keyword) parameters must come first, followed by 2102 # keyword parameters, followed by a '*foo' parameter that gets 2103 # excess positional arguments (as in Python). Each of these three 2104 # parameter categories is optional. 2105 # 2106 # Note that we do not support the '**foo' parameter for collecting 2107 # otherwise undefined keyword args. Otherwise the parameter list 2108 # is (I believe) identical to what is supported in Python. 2109 # 2110 # The param list generates a tuple, where the first element is a 2111 # list of the positional params and the second element is a dict 2112 # containing the keyword params. 2113 def p_param_list_0(self, t): 2114 'param_list : positional_param_list COMMA nonpositional_param_list' 2115 t[0] = t[1] + t[3] 2116 2117 def p_param_list_1(self, t): 2118 '''param_list : positional_param_list 2119 | nonpositional_param_list''' 2120 t[0] = t[1] 2121 2122 def p_positional_param_list_0(self, t): 2123 'positional_param_list : empty' 2124 t[0] = [] 2125 2126 def p_positional_param_list_1(self, t): 2127 'positional_param_list : ID' 2128 t[0] = [t[1]] 2129 2130 def p_positional_param_list_2(self, t): 2131 'positional_param_list : positional_param_list COMMA ID' 2132 t[0] = t[1] + [t[3]] 2133 2134 def p_nonpositional_param_list_0(self, t): 2135 'nonpositional_param_list : keyword_param_list COMMA excess_args_param' 2136 t[0] = t[1] + t[3] 2137 2138 def p_nonpositional_param_list_1(self, t): 2139 '''nonpositional_param_list : keyword_param_list 2140 | excess_args_param''' 2141 t[0] = t[1] 2142 2143 def p_keyword_param_list_0(self, t): 2144 'keyword_param_list : keyword_param' 2145 t[0] = [t[1]] 2146 2147 def p_keyword_param_list_1(self, t): 2148 'keyword_param_list : keyword_param_list COMMA keyword_param' 2149 t[0] = t[1] + [t[3]] 2150 2151 def p_keyword_param(self, t): 2152 'keyword_param : ID EQUALS expr' 2153 t[0] = t[1] + ' = ' + t[3].__repr__() 2154 2155 def p_excess_args_param(self, t): 2156 'excess_args_param : ASTERISK ID' 2157 # Just concatenate them: '*ID'. Wrap in list to be consistent 2158 # with positional_param_list and keyword_param_list. 2159 t[0] = [t[1] + t[2]] 2160 2161 # End of format definition-related rules. 2162 ############## 2163 2164 # 2165 # A decode block looks like: 2166 # decode <field1> [, <field2>]* [default <inst>] { ... } 2167 # 2168 def p_top_level_decode_block(self, t): 2169 'top_level_decode_block : decode_block' 2170 codeObj = t[1] 2171 codeObj.wrap_decode_block(''' 2172StaticInstPtr 2173%(isa_name)s::Decoder::decodeInst(%(isa_name)s::ExtMachInst machInst) 2174{ 2175 using namespace %(namespace)s; 2176''' % self, '}') 2177 2178 codeObj.emit() 2179 2180 def p_decode_block(self, t): 2181 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE' 2182 default_defaults = self.defaultStack.pop() 2183 codeObj = t[5] 2184 # use the "default defaults" only if there was no explicit 2185 # default statement in decode_stmt_list 2186 if not codeObj.has_decode_default: 2187 codeObj += default_defaults 2188 codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n') 2189 t[0] = codeObj 2190 2191 # The opt_default statement serves only to push the "default 2192 # defaults" onto defaultStack. This value will be used by nested 2193 # decode blocks, and used and popped off when the current 2194 # decode_block is processed (in p_decode_block() above). 2195 def p_opt_default_0(self, t): 2196 'opt_default : empty' 2197 # no default specified: reuse the one currently at the top of 2198 # the stack 2199 self.defaultStack.push(self.defaultStack.top()) 2200 # no meaningful value returned 2201 t[0] = None 2202 2203 def p_opt_default_1(self, t): 2204 'opt_default : DEFAULT inst' 2205 # push the new default 2206 codeObj = t[2] 2207 codeObj.wrap_decode_block('\ndefault:\n', 'break;\n') 2208 self.defaultStack.push(codeObj) 2209 # no meaningful value returned 2210 t[0] = None 2211 2212 def p_decode_stmt_list_0(self, t): 2213 'decode_stmt_list : decode_stmt' 2214 t[0] = t[1] 2215 2216 def p_decode_stmt_list_1(self, t): 2217 'decode_stmt_list : decode_stmt decode_stmt_list' 2218 if (t[1].has_decode_default and t[2].has_decode_default): 2219 error(t.lineno(1), 'Two default cases in decode block') 2220 t[0] = t[1] + t[2] 2221 2222 # 2223 # Decode statement rules 2224 # 2225 # There are four types of statements allowed in a decode block: 2226 # 1. Format blocks 'format <foo> { ... }' 2227 # 2. Nested decode blocks 2228 # 3. Instruction definitions. 2229 # 4. C preprocessor directives. 2230 2231 2232 # Preprocessor directives found in a decode statement list are 2233 # passed through to the output, replicated to all of the output 2234 # code streams. This works well for ifdefs, so we can ifdef out 2235 # both the declarations and the decode cases generated by an 2236 # instruction definition. Handling them as part of the grammar 2237 # makes it easy to keep them in the right place with respect to 2238 # the code generated by the other statements. 2239 def p_decode_stmt_cpp(self, t): 2240 'decode_stmt : CPPDIRECTIVE' 2241 t[0] = GenCode(self, t[1], t[1], t[1], t[1]) 2242 2243 # A format block 'format <foo> { ... }' sets the default 2244 # instruction format used to handle instruction definitions inside 2245 # the block. This format can be overridden by using an explicit 2246 # format on the instruction definition or with a nested format 2247 # block. 2248 def p_decode_stmt_format(self, t): 2249 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE' 2250 # The format will be pushed on the stack when 'push_format_id' 2251 # is processed (see below). Once the parser has recognized 2252 # the full production (though the right brace), we're done 2253 # with the format, so now we can pop it. 2254 self.formatStack.pop() 2255 t[0] = t[4] 2256 2257 # This rule exists so we can set the current format (& push the 2258 # stack) when we recognize the format name part of the format 2259 # block. 2260 def p_push_format_id(self, t): 2261 'push_format_id : ID' 2262 try: 2263 self.formatStack.push(self.formatMap[t[1]]) 2264 t[0] = ('', '// format %s' % t[1]) 2265 except KeyError: 2266 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 2267 2268 # Nested decode block: if the value of the current field matches 2269 # the specified constant(s), do a nested decode on some other field. 2270 def p_decode_stmt_decode(self, t): 2271 'decode_stmt : case_list COLON decode_block' 2272 case_list = t[1] 2273 codeObj = t[3] 2274 # just wrap the decoding code from the block as a case in the 2275 # outer switch statement. 2276 codeObj.wrap_decode_block('\n%s\n' % ''.join(case_list)) 2277 codeObj.has_decode_default = (case_list == ['default:']) 2278 t[0] = codeObj 2279 2280 # Instruction definition (finally!). 2281 def p_decode_stmt_inst(self, t): 2282 'decode_stmt : case_list COLON inst SEMI' 2283 case_list = t[1] 2284 codeObj = t[3] 2285 codeObj.wrap_decode_block('\n%s' % ''.join(case_list), 'break;\n') 2286 codeObj.has_decode_default = (case_list == ['default:']) 2287 t[0] = codeObj 2288 2289 # The constant list for a decode case label must be non-empty, and must 2290 # either be the keyword 'default', or made up of one or more 2291 # comma-separated integer literals or strings which evaluate to 2292 # constants when compiled as C++. 2293 def p_case_list_0(self, t): 2294 'case_list : DEFAULT' 2295 t[0] = ['default:'] 2296 2297 def prep_int_lit_case_label(self, lit): 2298 if lit >= 2**32: 2299 return 'case ULL(%#x): ' % lit 2300 else: 2301 return 'case %#x: ' % lit 2302 2303 def prep_str_lit_case_label(self, lit): 2304 return 'case %s: ' % lit 2305 2306 def p_case_list_1(self, t): 2307 'case_list : INTLIT' 2308 t[0] = [self.prep_int_lit_case_label(t[1])] 2309 2310 def p_case_list_2(self, t): 2311 'case_list : STRLIT' 2312 t[0] = [self.prep_str_lit_case_label(t[1])] 2313 2314 def p_case_list_3(self, t): 2315 'case_list : case_list COMMA INTLIT' 2316 t[0] = t[1] 2317 t[0].append(self.prep_int_lit_case_label(t[3])) 2318 2319 def p_case_list_4(self, t): 2320 'case_list : case_list COMMA STRLIT' 2321 t[0] = t[1] 2322 t[0].append(self.prep_str_lit_case_label(t[3])) 2323 2324 # Define an instruction using the current instruction format 2325 # (specified by an enclosing format block). 2326 # "<mnemonic>(<args>)" 2327 def p_inst_0(self, t): 2328 'inst : ID LPAREN arg_list RPAREN' 2329 # Pass the ID and arg list to the current format class to deal with. 2330 currentFormat = self.formatStack.top() 2331 codeObj = currentFormat.defineInst(self, t[1], t[3], t.lexer.lineno) 2332 args = ','.join(map(str, t[3])) 2333 args = re.sub('(?m)^', '//', args) 2334 args = re.sub('^//', '', args) 2335 comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args) 2336 codeObj.prepend_all(comment) 2337 t[0] = codeObj 2338 2339 # Define an instruction using an explicitly specified format: 2340 # "<fmt>::<mnemonic>(<args>)" 2341 def p_inst_1(self, t): 2342 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN' 2343 try: 2344 format = self.formatMap[t[1]] 2345 except KeyError: 2346 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 2347 2348 codeObj = format.defineInst(self, t[3], t[5], t.lexer.lineno) 2349 comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5]) 2350 codeObj.prepend_all(comment) 2351 t[0] = codeObj 2352 2353 # The arg list generates a tuple, where the first element is a 2354 # list of the positional args and the second element is a dict 2355 # containing the keyword args. 2356 def p_arg_list_0(self, t): 2357 'arg_list : positional_arg_list COMMA keyword_arg_list' 2358 t[0] = ( t[1], t[3] ) 2359 2360 def p_arg_list_1(self, t): 2361 'arg_list : positional_arg_list' 2362 t[0] = ( t[1], {} ) 2363 2364 def p_arg_list_2(self, t): 2365 'arg_list : keyword_arg_list' 2366 t[0] = ( [], t[1] ) 2367 2368 def p_positional_arg_list_0(self, t): 2369 'positional_arg_list : empty' 2370 t[0] = [] 2371 2372 def p_positional_arg_list_1(self, t): 2373 'positional_arg_list : expr' 2374 t[0] = [t[1]] 2375 2376 def p_positional_arg_list_2(self, t): 2377 'positional_arg_list : positional_arg_list COMMA expr' 2378 t[0] = t[1] + [t[3]] 2379 2380 def p_keyword_arg_list_0(self, t): 2381 'keyword_arg_list : keyword_arg' 2382 t[0] = t[1] 2383 2384 def p_keyword_arg_list_1(self, t): 2385 'keyword_arg_list : keyword_arg_list COMMA keyword_arg' 2386 t[0] = t[1] 2387 t[0].update(t[3]) 2388 2389 def p_keyword_arg(self, t): 2390 'keyword_arg : ID EQUALS expr' 2391 t[0] = { t[1] : t[3] } 2392 2393 # 2394 # Basic expressions. These constitute the argument values of 2395 # "function calls" (i.e. instruction definitions in the decode 2396 # block) and default values for formal parameters of format 2397 # functions. 2398 # 2399 # Right now, these are either strings, integers, or (recursively) 2400 # lists of exprs (using Python square-bracket list syntax). Note 2401 # that bare identifiers are trated as string constants here (since 2402 # there isn't really a variable namespace to refer to). 2403 # 2404 def p_expr_0(self, t): 2405 '''expr : ID 2406 | INTLIT 2407 | STRLIT 2408 | CODELIT''' 2409 t[0] = t[1] 2410 2411 def p_expr_1(self, t): 2412 '''expr : LBRACKET list_expr RBRACKET''' 2413 t[0] = t[2] 2414 2415 def p_list_expr_0(self, t): 2416 'list_expr : expr' 2417 t[0] = [t[1]] 2418 2419 def p_list_expr_1(self, t): 2420 'list_expr : list_expr COMMA expr' 2421 t[0] = t[1] + [t[3]] 2422 2423 def p_list_expr_2(self, t): 2424 'list_expr : empty' 2425 t[0] = [] 2426 2427 # 2428 # Empty production... use in other rules for readability. 2429 # 2430 def p_empty(self, t): 2431 'empty :' 2432 pass 2433 2434 # Parse error handler. Note that the argument here is the 2435 # offending *token*, not a grammar symbol (hence the need to use 2436 # t.value) 2437 def p_error(self, t): 2438 if t: 2439 error(t.lexer.lineno, "syntax error at '%s'" % t.value) 2440 else: 2441 error("unknown syntax error") 2442 2443 # END OF GRAMMAR RULES 2444 2445 def updateExportContext(self): 2446 2447 # create a continuation that allows us to grab the current parser 2448 def wrapInstObjParams(*args): 2449 return InstObjParams(self, *args) 2450 self.exportContext['InstObjParams'] = wrapInstObjParams 2451 self.exportContext.update(self.templateMap) 2452 2453 def defFormat(self, id, params, code, lineno): 2454 '''Define a new format''' 2455 2456 # make sure we haven't already defined this one 2457 if id in self.formatMap: 2458 error(lineno, 'format %s redefined.' % id) 2459 2460 # create new object and store in global map 2461 self.formatMap[id] = Format(id, params, code) 2462 2463 def expandCpuSymbolsToDict(self, template): 2464 '''Expand template with CPU-specific references into a 2465 dictionary with an entry for each CPU model name. The entry 2466 key is the model name and the corresponding value is the 2467 template with the CPU-specific refs substituted for that 2468 model.''' 2469 2470 # Protect '%'s that don't go with CPU-specific terms 2471 t = re.sub(r'%(?!\(CPU_)', '%%', template) 2472 result = {} 2473 for cpu in self.cpuModels: 2474 result[cpu.name] = t % cpu.strings 2475 return result 2476 2477 def expandCpuSymbolsToString(self, template): 2478 '''*If* the template has CPU-specific references, return a 2479 single string containing a copy of the template for each CPU 2480 model with the corresponding values substituted in. If the 2481 template has no CPU-specific references, it is returned 2482 unmodified.''' 2483 2484 if template.find('%(CPU_') != -1: 2485 return reduce(lambda x,y: x+y, 2486 self.expandCpuSymbolsToDict(template).values()) 2487 else: 2488 return template 2489 2490 def protectCpuSymbols(self, template): 2491 '''Protect CPU-specific references by doubling the 2492 corresponding '%'s (in preparation for substituting a different 2493 set of references into the template).''' 2494 2495 return re.sub(r'%(?=\(CPU_)', '%%', template) 2496 2497 def protectNonSubstPercents(self, s): 2498 '''Protect any non-dict-substitution '%'s in a format string 2499 (i.e. those not followed by '(')''' 2500 2501 return re.sub(r'%(?!\()', '%%', s) 2502 2503 def buildOperandNameMap(self, user_dict, lineno): 2504 operand_name = {} 2505 for op_name, val in user_dict.iteritems(): 2506 2507 # Check if extra attributes have been specified. 2508 if len(val) > 9: 2509 error(lineno, 'error: too many attributes for operand "%s"' % 2510 base_cls_name) 2511 2512 # Pad val with None in case optional args are missing 2513 val += (None, None, None, None) 2514 base_cls_name, dflt_ext, reg_spec, flags, sort_pri, \ 2515 read_code, write_code, read_predicate, write_predicate = val[:9] 2516 2517 # Canonical flag structure is a triple of lists, where each list 2518 # indicates the set of flags implied by this operand always, when 2519 # used as a source, and when used as a dest, respectively. 2520 # For simplicity this can be initialized using a variety of fairly 2521 # obvious shortcuts; we convert these to canonical form here. 2522 if not flags: 2523 # no flags specified (e.g., 'None') 2524 flags = ( [], [], [] ) 2525 elif isinstance(flags, str): 2526 # a single flag: assumed to be unconditional 2527 flags = ( [ flags ], [], [] ) 2528 elif isinstance(flags, list): 2529 # a list of flags: also assumed to be unconditional 2530 flags = ( flags, [], [] ) 2531 elif isinstance(flags, tuple): 2532 # it's a tuple: it should be a triple, 2533 # but each item could be a single string or a list 2534 (uncond_flags, src_flags, dest_flags) = flags 2535 flags = (makeList(uncond_flags), 2536 makeList(src_flags), makeList(dest_flags)) 2537 2538 # Accumulate attributes of new operand class in tmp_dict 2539 tmp_dict = {} 2540 attrList = ['reg_spec', 'flags', 'sort_pri', 2541 'read_code', 'write_code', 2542 'read_predicate', 'write_predicate'] 2543 if dflt_ext: 2544 dflt_ctype = self.operandTypeMap[dflt_ext] 2545 attrList.extend(['dflt_ctype', 'dflt_ext']) 2546 # reg_spec is either just a string or a dictionary 2547 # (for elems of vector) 2548 if isinstance(reg_spec, tuple): 2549 (reg_spec, elem_spec) = reg_spec 2550 if isinstance(elem_spec, str): 2551 attrList.append('elem_spec') 2552 else: 2553 assert(isinstance(elem_spec, dict)) 2554 elems = elem_spec 2555 attrList.append('elems') 2556 for attr in attrList: 2557 tmp_dict[attr] = eval(attr) 2558 tmp_dict['base_name'] = op_name 2559 2560 # New class name will be e.g. "IntReg_Ra" 2561 cls_name = base_cls_name + '_' + op_name 2562 # Evaluate string arg to get class object. Note that the 2563 # actual base class for "IntReg" is "IntRegOperand", i.e. we 2564 # have to append "Operand". 2565 try: 2566 base_cls = eval(base_cls_name + 'Operand') 2567 except NameError: 2568 error(lineno, 2569 'error: unknown operand base class "%s"' % base_cls_name) 2570 # The following statement creates a new class called 2571 # <cls_name> as a subclass of <base_cls> with the attributes 2572 # in tmp_dict, just as if we evaluated a class declaration. 2573 operand_name[op_name] = type(cls_name, (base_cls,), tmp_dict) 2574 2575 self.operandNameMap = operand_name 2576 2577 # Define operand variables. 2578 operands = user_dict.keys() 2579 # Add the elems defined in the vector operands and 2580 # build a map elem -> vector (used in OperandList) 2581 elem_to_vec = {} 2582 for op in user_dict.keys(): 2583 if hasattr(self.operandNameMap[op], 'elems'): 2584 for elem in self.operandNameMap[op].elems.keys(): 2585 operands.append(elem) 2586 elem_to_vec[elem] = op 2587 self.elemToVector = elem_to_vec 2588 extensions = self.operandTypeMap.keys() 2589 2590 operandsREString = r''' 2591 (?<!\w) # neg. lookbehind assertion: prevent partial matches 2592 ((%s)(?:_(%s))?) # match: operand with optional '_' then suffix 2593 (?!\w) # neg. lookahead assertion: prevent partial matches 2594 ''' % (string.join(operands, '|'), string.join(extensions, '|')) 2595 2596 self.operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE) 2597 2598 # Same as operandsREString, but extension is mandatory, and only two 2599 # groups are returned (base and ext, not full name as above). 2600 # Used for subtituting '_' for '.' to make C++ identifiers. 2601 operandsWithExtREString = r'(?<!\w)(%s)_(%s)(?!\w)' \ 2602 % (string.join(operands, '|'), string.join(extensions, '|')) 2603 2604 self.operandsWithExtRE = \ 2605 re.compile(operandsWithExtREString, re.MULTILINE) 2606 2607 def substMungedOpNames(self, code): 2608 '''Munge operand names in code string to make legal C++ 2609 variable names. This means getting rid of the type extension 2610 if any. Will match base_name attribute of Operand object.)''' 2611 return self.operandsWithExtRE.sub(r'\1', code) 2612 2613 def mungeSnippet(self, s): 2614 '''Fix up code snippets for final substitution in templates.''' 2615 if isinstance(s, str): 2616 return self.substMungedOpNames(substBitOps(s)) 2617 else: 2618 return s 2619 2620 def open(self, name, bare=False): 2621 '''Open the output file for writing and include scary warning.''' 2622 filename = os.path.join(self.output_dir, name) 2623 f = open(filename, 'w') 2624 if f: 2625 if not bare: 2626 f.write(ISAParser.scaremonger_template % self) 2627 return f 2628 2629 def update(self, file, contents): 2630 '''Update the output file only. Scons should handle the case when 2631 the new contents are unchanged using its built-in hash feature.''' 2632 f = self.open(file) 2633 f.write(contents) 2634 f.close() 2635 2636 # This regular expression matches '##include' directives 2637 includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[^"]*)".*$', 2638 re.MULTILINE) 2639 2640 def replace_include(self, matchobj, dirname): 2641 """Function to replace a matched '##include' directive with the 2642 contents of the specified file (with nested ##includes 2643 replaced recursively). 'matchobj' is an re match object 2644 (from a match of includeRE) and 'dirname' is the directory 2645 relative to which the file path should be resolved.""" 2646 2647 fname = matchobj.group('filename') 2648 full_fname = os.path.normpath(os.path.join(dirname, fname)) 2649 contents = '##newfile "%s"\n%s\n##endfile\n' % \ 2650 (full_fname, self.read_and_flatten(full_fname)) 2651 return contents 2652 2653 def read_and_flatten(self, filename): 2654 """Read a file and recursively flatten nested '##include' files.""" 2655 2656 current_dir = os.path.dirname(filename) 2657 try: 2658 contents = open(filename).read() 2659 except IOError: 2660 error('Error including file "%s"' % filename) 2661 2662 self.fileNameStack.push(LineTracker(filename)) 2663 2664 # Find any includes and include them 2665 def replace(matchobj): 2666 return self.replace_include(matchobj, current_dir) 2667 contents = self.includeRE.sub(replace, contents) 2668 2669 self.fileNameStack.pop() 2670 return contents 2671 2672 AlreadyGenerated = {} 2673 2674 def _parse_isa_desc(self, isa_desc_file): 2675 '''Read in and parse the ISA description.''' 2676 2677 # The build system can end up running the ISA parser twice: once to 2678 # finalize the build dependencies, and then to actually generate 2679 # the files it expects (in src/arch/$ARCH/generated). This code 2680 # doesn't do anything different either time, however; the SCons 2681 # invocations just expect different things. Since this code runs 2682 # within SCons, we can just remember that we've already run and 2683 # not perform a completely unnecessary run, since the ISA parser's 2684 # effect is idempotent. 2685 if isa_desc_file in ISAParser.AlreadyGenerated: 2686 return 2687 2688 # grab the last three path components of isa_desc_file 2689 self.filename = '/'.join(isa_desc_file.split('/')[-3:]) 2690 2691 # Read file and (recursively) all included files into a string. 2692 # PLY requires that the input be in a single string so we have to 2693 # do this up front. 2694 isa_desc = self.read_and_flatten(isa_desc_file) 2695 2696 # Initialize lineno tracker 2697 self.lex.lineno = LineTracker(isa_desc_file) 2698 2699 # Parse. 2700 self.parse_string(isa_desc) 2701 2702 ISAParser.AlreadyGenerated[isa_desc_file] = None 2703 2704 def parse_isa_desc(self, *args, **kwargs): 2705 try: 2706 self._parse_isa_desc(*args, **kwargs) 2707 except ISAParserError, e: 2708 print backtrace(self.fileNameStack) 2709 print "At %s:" % e.lineno 2710 print e 2711 sys.exit(1) 2712 2713# Called as script: get args from command line. 2714# Args are: <isa desc file> <output dir> 2715if __name__ == '__main__': 2716 ISAParser(sys.argv[2]).parse_isa_desc(sys.argv[1]) 2717