isa_parser.py revision 11163
1# Copyright (c) 2014 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 isPCState(self): 497 return 0 498 499 def isPCPart(self): 500 return self.isPCState() and self.reg_spec 501 502 def hasReadPred(self): 503 return self.read_predicate != None 504 505 def hasWritePred(self): 506 return self.write_predicate != None 507 508 def getFlags(self): 509 # note the empty slice '[:]' gives us a copy of self.flags[0] 510 # instead of a reference to it 511 my_flags = self.flags[0][:] 512 if self.is_src: 513 my_flags += self.flags[1] 514 if self.is_dest: 515 my_flags += self.flags[2] 516 return my_flags 517 518 def makeDecl(self): 519 # Note that initializations in the declarations are solely 520 # to avoid 'uninitialized variable' errors from the compiler. 521 return self.ctype + ' ' + self.base_name + ' = 0;\n'; 522 523class IntRegOperand(Operand): 524 def isReg(self): 525 return 1 526 527 def isIntReg(self): 528 return 1 529 530 def makeConstructor(self, predRead, predWrite): 531 c_src = '' 532 c_dest = '' 533 534 if self.is_src: 535 c_src = '\n\t_srcRegIdx[_numSrcRegs++] = %s;' % (self.reg_spec) 536 if self.hasReadPred(): 537 c_src = '\n\tif (%s) {%s\n\t}' % \ 538 (self.read_predicate, c_src) 539 540 if self.is_dest: 541 c_dest = '\n\t_destRegIdx[_numDestRegs++] = %s;' % \ 542 (self.reg_spec) 543 c_dest += '\n\t_numIntDestRegs++;' 544 if self.hasWritePred(): 545 c_dest = '\n\tif (%s) {%s\n\t}' % \ 546 (self.write_predicate, c_dest) 547 548 return c_src + c_dest 549 550 def makeRead(self, predRead): 551 if (self.ctype == 'float' or self.ctype == 'double'): 552 error('Attempt to read integer register as FP') 553 if self.read_code != None: 554 return self.buildReadCode('readIntRegOperand') 555 556 int_reg_val = '' 557 if predRead: 558 int_reg_val = 'xc->readIntRegOperand(this, _sourceIndex++)' 559 if self.hasReadPred(): 560 int_reg_val = '(%s) ? %s : 0' % \ 561 (self.read_predicate, int_reg_val) 562 else: 563 int_reg_val = 'xc->readIntRegOperand(this, %d)' % self.src_reg_idx 564 565 return '%s = %s;\n' % (self.base_name, int_reg_val) 566 567 def makeWrite(self, predWrite): 568 if (self.ctype == 'float' or self.ctype == 'double'): 569 error('Attempt to write integer register as FP') 570 if self.write_code != None: 571 return self.buildWriteCode('setIntRegOperand') 572 573 if predWrite: 574 wp = 'true' 575 if self.hasWritePred(): 576 wp = self.write_predicate 577 578 wcond = 'if (%s)' % (wp) 579 windex = '_destIndex++' 580 else: 581 wcond = '' 582 windex = '%d' % self.dest_reg_idx 583 584 wb = ''' 585 %s 586 { 587 %s final_val = %s; 588 xc->setIntRegOperand(this, %s, final_val);\n 589 if (traceData) { traceData->setData(final_val); } 590 }''' % (wcond, self.ctype, self.base_name, windex) 591 592 return wb 593 594class FloatRegOperand(Operand): 595 def isReg(self): 596 return 1 597 598 def isFloatReg(self): 599 return 1 600 601 def makeConstructor(self, predRead, predWrite): 602 c_src = '' 603 c_dest = '' 604 605 if self.is_src: 606 c_src = '\n\t_srcRegIdx[_numSrcRegs++] = %s + FP_Reg_Base;' % \ 607 (self.reg_spec) 608 609 if self.is_dest: 610 c_dest = \ 611 '\n\t_destRegIdx[_numDestRegs++] = %s + FP_Reg_Base;' % \ 612 (self.reg_spec) 613 c_dest += '\n\t_numFPDestRegs++;' 614 615 return c_src + c_dest 616 617 def makeRead(self, predRead): 618 bit_select = 0 619 if (self.ctype == 'float' or self.ctype == 'double'): 620 func = 'readFloatRegOperand' 621 else: 622 func = 'readFloatRegOperandBits' 623 if self.read_code != None: 624 return self.buildReadCode(func) 625 626 if predRead: 627 rindex = '_sourceIndex++' 628 else: 629 rindex = '%d' % self.src_reg_idx 630 631 return '%s = xc->%s(this, %s);\n' % \ 632 (self.base_name, func, rindex) 633 634 def makeWrite(self, predWrite): 635 if (self.ctype == 'float' or self.ctype == 'double'): 636 func = 'setFloatRegOperand' 637 else: 638 func = 'setFloatRegOperandBits' 639 if self.write_code != None: 640 return self.buildWriteCode(func) 641 642 if predWrite: 643 wp = '_destIndex++' 644 else: 645 wp = '%d' % self.dest_reg_idx 646 wp = 'xc->%s(this, %s, final_val);' % (func, wp) 647 648 wb = ''' 649 { 650 %s final_val = %s; 651 %s\n 652 if (traceData) { traceData->setData(final_val); } 653 }''' % (self.ctype, self.base_name, wp) 654 return wb 655 656class CCRegOperand(Operand): 657 def isReg(self): 658 return 1 659 660 def isCCReg(self): 661 return 1 662 663 def makeConstructor(self, predRead, predWrite): 664 c_src = '' 665 c_dest = '' 666 667 if self.is_src: 668 c_src = '\n\t_srcRegIdx[_numSrcRegs++] = %s + CC_Reg_Base;' % \ 669 (self.reg_spec) 670 if self.hasReadPred(): 671 c_src = '\n\tif (%s) {%s\n\t}' % \ 672 (self.read_predicate, c_src) 673 674 if self.is_dest: 675 c_dest = \ 676 '\n\t_destRegIdx[_numDestRegs++] = %s + CC_Reg_Base;' % \ 677 (self.reg_spec) 678 c_dest += '\n\t_numCCDestRegs++;' 679 if self.hasWritePred(): 680 c_dest = '\n\tif (%s) {%s\n\t}' % \ 681 (self.write_predicate, c_dest) 682 683 return c_src + c_dest 684 685 def makeRead(self, predRead): 686 if (self.ctype == 'float' or self.ctype == 'double'): 687 error('Attempt to read condition-code register as FP') 688 if self.read_code != None: 689 return self.buildReadCode('readCCRegOperand') 690 691 int_reg_val = '' 692 if predRead: 693 int_reg_val = 'xc->readCCRegOperand(this, _sourceIndex++)' 694 if self.hasReadPred(): 695 int_reg_val = '(%s) ? %s : 0' % \ 696 (self.read_predicate, int_reg_val) 697 else: 698 int_reg_val = 'xc->readCCRegOperand(this, %d)' % self.src_reg_idx 699 700 return '%s = %s;\n' % (self.base_name, int_reg_val) 701 702 def makeWrite(self, predWrite): 703 if (self.ctype == 'float' or self.ctype == 'double'): 704 error('Attempt to write condition-code register as FP') 705 if self.write_code != None: 706 return self.buildWriteCode('setCCRegOperand') 707 708 if predWrite: 709 wp = 'true' 710 if self.hasWritePred(): 711 wp = self.write_predicate 712 713 wcond = 'if (%s)' % (wp) 714 windex = '_destIndex++' 715 else: 716 wcond = '' 717 windex = '%d' % self.dest_reg_idx 718 719 wb = ''' 720 %s 721 { 722 %s final_val = %s; 723 xc->setCCRegOperand(this, %s, final_val);\n 724 if (traceData) { traceData->setData(final_val); } 725 }''' % (wcond, self.ctype, self.base_name, windex) 726 727 return wb 728 729class ControlRegOperand(Operand): 730 def isReg(self): 731 return 1 732 733 def isControlReg(self): 734 return 1 735 736 def makeConstructor(self, predRead, predWrite): 737 c_src = '' 738 c_dest = '' 739 740 if self.is_src: 741 c_src = \ 742 '\n\t_srcRegIdx[_numSrcRegs++] = %s + Misc_Reg_Base;' % \ 743 (self.reg_spec) 744 745 if self.is_dest: 746 c_dest = \ 747 '\n\t_destRegIdx[_numDestRegs++] = %s + Misc_Reg_Base;' % \ 748 (self.reg_spec) 749 750 return c_src + c_dest 751 752 def makeRead(self, predRead): 753 bit_select = 0 754 if (self.ctype == 'float' or self.ctype == 'double'): 755 error('Attempt to read control register as FP') 756 if self.read_code != None: 757 return self.buildReadCode('readMiscRegOperand') 758 759 if predRead: 760 rindex = '_sourceIndex++' 761 else: 762 rindex = '%d' % self.src_reg_idx 763 764 return '%s = xc->readMiscRegOperand(this, %s);\n' % \ 765 (self.base_name, rindex) 766 767 def makeWrite(self, predWrite): 768 if (self.ctype == 'float' or self.ctype == 'double'): 769 error('Attempt to write control register as FP') 770 if self.write_code != None: 771 return self.buildWriteCode('setMiscRegOperand') 772 773 if predWrite: 774 windex = '_destIndex++' 775 else: 776 windex = '%d' % self.dest_reg_idx 777 778 wb = 'xc->setMiscRegOperand(this, %s, %s);\n' % \ 779 (windex, self.base_name) 780 wb += 'if (traceData) { traceData->setData(%s); }' % \ 781 self.base_name 782 783 return wb 784 785class MemOperand(Operand): 786 def isMem(self): 787 return 1 788 789 def makeConstructor(self, predRead, predWrite): 790 return '' 791 792 def makeDecl(self): 793 # Note that initializations in the declarations are solely 794 # to avoid 'uninitialized variable' errors from the compiler. 795 # Declare memory data variable. 796 return '%s %s = 0;\n' % (self.ctype, self.base_name) 797 798 def makeRead(self, predRead): 799 if self.read_code != None: 800 return self.buildReadCode() 801 return '' 802 803 def makeWrite(self, predWrite): 804 if self.write_code != None: 805 return self.buildWriteCode() 806 return '' 807 808class PCStateOperand(Operand): 809 def makeConstructor(self, predRead, predWrite): 810 return '' 811 812 def makeRead(self, predRead): 813 if self.reg_spec: 814 # A component of the PC state. 815 return '%s = __parserAutoPCState.%s();\n' % \ 816 (self.base_name, self.reg_spec) 817 else: 818 # The whole PC state itself. 819 return '%s = xc->pcState();\n' % self.base_name 820 821 def makeWrite(self, predWrite): 822 if self.reg_spec: 823 # A component of the PC state. 824 return '__parserAutoPCState.%s(%s);\n' % \ 825 (self.reg_spec, self.base_name) 826 else: 827 # The whole PC state itself. 828 return 'xc->pcState(%s);\n' % self.base_name 829 830 def makeDecl(self): 831 ctype = 'TheISA::PCState' 832 if self.isPCPart(): 833 ctype = self.ctype 834 # Note that initializations in the declarations are solely 835 # to avoid 'uninitialized variable' errors from the compiler. 836 return '%s %s = 0;\n' % (ctype, self.base_name) 837 838 def isPCState(self): 839 return 1 840 841class OperandList(object): 842 '''Find all the operands in the given code block. Returns an operand 843 descriptor list (instance of class OperandList).''' 844 def __init__(self, parser, code): 845 self.items = [] 846 self.bases = {} 847 # delete strings and comments so we don't match on operands inside 848 for regEx in (stringRE, commentRE): 849 code = regEx.sub('', code) 850 # search for operands 851 next_pos = 0 852 while 1: 853 match = parser.operandsRE.search(code, next_pos) 854 if not match: 855 # no more matches: we're done 856 break 857 op = match.groups() 858 # regexp groups are operand full name, base, and extension 859 (op_full, op_base, op_ext) = op 860 # if the token following the operand is an assignment, this is 861 # a destination (LHS), else it's a source (RHS) 862 is_dest = (assignRE.match(code, match.end()) != None) 863 is_src = not is_dest 864 # see if we've already seen this one 865 op_desc = self.find_base(op_base) 866 if op_desc: 867 if op_desc.ext != op_ext: 868 error('Inconsistent extensions for operand %s' % \ 869 op_base) 870 op_desc.is_src = op_desc.is_src or is_src 871 op_desc.is_dest = op_desc.is_dest or is_dest 872 else: 873 # new operand: create new descriptor 874 op_desc = parser.operandNameMap[op_base](parser, 875 op_full, op_ext, is_src, is_dest) 876 self.append(op_desc) 877 # start next search after end of current match 878 next_pos = match.end() 879 self.sort() 880 # enumerate source & dest register operands... used in building 881 # constructor later 882 self.numSrcRegs = 0 883 self.numDestRegs = 0 884 self.numFPDestRegs = 0 885 self.numIntDestRegs = 0 886 self.numCCDestRegs = 0 887 self.numMiscDestRegs = 0 888 self.memOperand = None 889 890 # Flags to keep track if one or more operands are to be read/written 891 # conditionally. 892 self.predRead = False 893 self.predWrite = False 894 895 for op_desc in self.items: 896 if op_desc.isReg(): 897 if op_desc.is_src: 898 op_desc.src_reg_idx = self.numSrcRegs 899 self.numSrcRegs += 1 900 if op_desc.is_dest: 901 op_desc.dest_reg_idx = self.numDestRegs 902 self.numDestRegs += 1 903 if op_desc.isFloatReg(): 904 self.numFPDestRegs += 1 905 elif op_desc.isIntReg(): 906 self.numIntDestRegs += 1 907 elif op_desc.isCCReg(): 908 self.numCCDestRegs += 1 909 elif op_desc.isControlReg(): 910 self.numMiscDestRegs += 1 911 elif op_desc.isMem(): 912 if self.memOperand: 913 error("Code block has more than one memory operand.") 914 self.memOperand = op_desc 915 916 # Check if this operand has read/write predication. If true, then 917 # the microop will dynamically index source/dest registers. 918 self.predRead = self.predRead or op_desc.hasReadPred() 919 self.predWrite = self.predWrite or op_desc.hasWritePred() 920 921 if parser.maxInstSrcRegs < self.numSrcRegs: 922 parser.maxInstSrcRegs = self.numSrcRegs 923 if parser.maxInstDestRegs < self.numDestRegs: 924 parser.maxInstDestRegs = self.numDestRegs 925 if parser.maxMiscDestRegs < self.numMiscDestRegs: 926 parser.maxMiscDestRegs = self.numMiscDestRegs 927 928 # now make a final pass to finalize op_desc fields that may depend 929 # on the register enumeration 930 for op_desc in self.items: 931 op_desc.finalize(self.predRead, self.predWrite) 932 933 def __len__(self): 934 return len(self.items) 935 936 def __getitem__(self, index): 937 return self.items[index] 938 939 def append(self, op_desc): 940 self.items.append(op_desc) 941 self.bases[op_desc.base_name] = op_desc 942 943 def find_base(self, base_name): 944 # like self.bases[base_name], but returns None if not found 945 # (rather than raising exception) 946 return self.bases.get(base_name) 947 948 # internal helper function for concat[Some]Attr{Strings|Lists} 949 def __internalConcatAttrs(self, attr_name, filter, result): 950 for op_desc in self.items: 951 if filter(op_desc): 952 result += getattr(op_desc, attr_name) 953 return result 954 955 # return a single string that is the concatenation of the (string) 956 # values of the specified attribute for all operands 957 def concatAttrStrings(self, attr_name): 958 return self.__internalConcatAttrs(attr_name, lambda x: 1, '') 959 960 # like concatAttrStrings, but only include the values for the operands 961 # for which the provided filter function returns true 962 def concatSomeAttrStrings(self, filter, attr_name): 963 return self.__internalConcatAttrs(attr_name, filter, '') 964 965 # return a single list that is the concatenation of the (list) 966 # values of the specified attribute for all operands 967 def concatAttrLists(self, attr_name): 968 return self.__internalConcatAttrs(attr_name, lambda x: 1, []) 969 970 # like concatAttrLists, but only include the values for the operands 971 # for which the provided filter function returns true 972 def concatSomeAttrLists(self, filter, attr_name): 973 return self.__internalConcatAttrs(attr_name, filter, []) 974 975 def sort(self): 976 self.items.sort(lambda a, b: a.sort_pri - b.sort_pri) 977 978class SubOperandList(OperandList): 979 '''Find all the operands in the given code block. Returns an operand 980 descriptor list (instance of class OperandList).''' 981 def __init__(self, parser, code, master_list): 982 self.items = [] 983 self.bases = {} 984 # delete strings and comments so we don't match on operands inside 985 for regEx in (stringRE, commentRE): 986 code = regEx.sub('', code) 987 # search for operands 988 next_pos = 0 989 while 1: 990 match = parser.operandsRE.search(code, next_pos) 991 if not match: 992 # no more matches: we're done 993 break 994 op = match.groups() 995 # regexp groups are operand full name, base, and extension 996 (op_full, op_base, op_ext) = op 997 # find this op in the master list 998 op_desc = master_list.find_base(op_base) 999 if not op_desc: 1000 error('Found operand %s which is not in the master list!' 1001 % op_base) 1002 else: 1003 # See if we've already found this operand 1004 op_desc = self.find_base(op_base) 1005 if not op_desc: 1006 # if not, add a reference to it to this sub list 1007 self.append(master_list.bases[op_base]) 1008 1009 # start next search after end of current match 1010 next_pos = match.end() 1011 self.sort() 1012 self.memOperand = None 1013 # Whether the whole PC needs to be read so parts of it can be accessed 1014 self.readPC = False 1015 # Whether the whole PC needs to be written after parts of it were 1016 # changed 1017 self.setPC = False 1018 # Whether this instruction manipulates the whole PC or parts of it. 1019 # Mixing the two is a bad idea and flagged as an error. 1020 self.pcPart = None 1021 1022 # Flags to keep track if one or more operands are to be read/written 1023 # conditionally. 1024 self.predRead = False 1025 self.predWrite = False 1026 1027 for op_desc in self.items: 1028 if op_desc.isPCPart(): 1029 self.readPC = True 1030 if op_desc.is_dest: 1031 self.setPC = True 1032 1033 if op_desc.isPCState(): 1034 if self.pcPart is not None: 1035 if self.pcPart and not op_desc.isPCPart() or \ 1036 not self.pcPart and op_desc.isPCPart(): 1037 error("Mixed whole and partial PC state operands.") 1038 self.pcPart = op_desc.isPCPart() 1039 1040 if op_desc.isMem(): 1041 if self.memOperand: 1042 error("Code block has more than one memory operand.") 1043 self.memOperand = op_desc 1044 1045 # Check if this operand has read/write predication. If true, then 1046 # the microop will dynamically index source/dest registers. 1047 self.predRead = self.predRead or op_desc.hasReadPred() 1048 self.predWrite = self.predWrite or op_desc.hasWritePred() 1049 1050# Regular expression object to match C++ strings 1051stringRE = re.compile(r'"([^"\\]|\\.)*"') 1052 1053# Regular expression object to match C++ comments 1054# (used in findOperands()) 1055commentRE = re.compile(r'(^)?[^\S\n]*/(?:\*(.*?)\*/[^\S\n]*|/[^\n]*)($)?', 1056 re.DOTALL | re.MULTILINE) 1057 1058# Regular expression object to match assignment statements 1059# (used in findOperands()) 1060assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE) 1061 1062def makeFlagConstructor(flag_list): 1063 if len(flag_list) == 0: 1064 return '' 1065 # filter out repeated flags 1066 flag_list.sort() 1067 i = 1 1068 while i < len(flag_list): 1069 if flag_list[i] == flag_list[i-1]: 1070 del flag_list[i] 1071 else: 1072 i += 1 1073 pre = '\n\tflags[' 1074 post = '] = true;' 1075 code = pre + string.join(flag_list, post + pre) + post 1076 return code 1077 1078# Assume all instruction flags are of the form 'IsFoo' 1079instFlagRE = re.compile(r'Is.*') 1080 1081# OpClass constants end in 'Op' except No_OpClass 1082opClassRE = re.compile(r'.*Op|No_OpClass') 1083 1084class InstObjParams(object): 1085 def __init__(self, parser, mnem, class_name, base_class = '', 1086 snippets = {}, opt_args = []): 1087 self.mnemonic = mnem 1088 self.class_name = class_name 1089 self.base_class = base_class 1090 if not isinstance(snippets, dict): 1091 snippets = {'code' : snippets} 1092 compositeCode = ' '.join(map(str, snippets.values())) 1093 self.snippets = snippets 1094 1095 self.operands = OperandList(parser, compositeCode) 1096 1097 # The header of the constructor declares the variables to be used 1098 # in the body of the constructor. 1099 header = '' 1100 header += '\n\t_numSrcRegs = 0;' 1101 header += '\n\t_numDestRegs = 0;' 1102 header += '\n\t_numFPDestRegs = 0;' 1103 header += '\n\t_numIntDestRegs = 0;' 1104 header += '\n\t_numCCDestRegs = 0;' 1105 1106 self.constructor = header + \ 1107 self.operands.concatAttrStrings('constructor') 1108 1109 self.flags = self.operands.concatAttrLists('flags') 1110 1111 self.op_class = None 1112 1113 # Optional arguments are assumed to be either StaticInst flags 1114 # or an OpClass value. To avoid having to import a complete 1115 # list of these values to match against, we do it ad-hoc 1116 # with regexps. 1117 for oa in opt_args: 1118 if instFlagRE.match(oa): 1119 self.flags.append(oa) 1120 elif opClassRE.match(oa): 1121 self.op_class = oa 1122 else: 1123 error('InstObjParams: optional arg "%s" not recognized ' 1124 'as StaticInst::Flag or OpClass.' % oa) 1125 1126 # Make a basic guess on the operand class if not set. 1127 # These are good enough for most cases. 1128 if not self.op_class: 1129 if 'IsStore' in self.flags: 1130 self.op_class = 'MemWriteOp' 1131 elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags: 1132 self.op_class = 'MemReadOp' 1133 elif 'IsFloating' in self.flags: 1134 self.op_class = 'FloatAddOp' 1135 else: 1136 self.op_class = 'IntAluOp' 1137 1138 # add flag initialization to contructor here to include 1139 # any flags added via opt_args 1140 self.constructor += makeFlagConstructor(self.flags) 1141 1142 # if 'IsFloating' is set, add call to the FP enable check 1143 # function (which should be provided by isa_desc via a declare) 1144 if 'IsFloating' in self.flags: 1145 self.fp_enable_check = 'fault = checkFpEnableFault(xc);' 1146 else: 1147 self.fp_enable_check = '' 1148 1149############## 1150# Stack: a simple stack object. Used for both formats (formatStack) 1151# and default cases (defaultStack). Simply wraps a list to give more 1152# stack-like syntax and enable initialization with an argument list 1153# (as opposed to an argument that's a list). 1154 1155class Stack(list): 1156 def __init__(self, *items): 1157 list.__init__(self, items) 1158 1159 def push(self, item): 1160 self.append(item); 1161 1162 def top(self): 1163 return self[-1] 1164 1165# Format a file include stack backtrace as a string 1166def backtrace(filename_stack): 1167 fmt = "In file included from %s:" 1168 return "\n".join([fmt % f for f in filename_stack]) 1169 1170 1171####################### 1172# 1173# LineTracker: track filenames along with line numbers in PLY lineno fields 1174# PLY explicitly doesn't do anything with 'lineno' except propagate 1175# it. This class lets us tie filenames with the line numbers with a 1176# minimum of disruption to existing increment code. 1177# 1178 1179class LineTracker(object): 1180 def __init__(self, filename, lineno=1): 1181 self.filename = filename 1182 self.lineno = lineno 1183 1184 # Overload '+=' for increments. We need to create a new object on 1185 # each update else every token ends up referencing the same 1186 # constantly incrementing instance. 1187 def __iadd__(self, incr): 1188 return LineTracker(self.filename, self.lineno + incr) 1189 1190 def __str__(self): 1191 return "%s:%d" % (self.filename, self.lineno) 1192 1193 # In case there are places where someone really expects a number 1194 def __int__(self): 1195 return self.lineno 1196 1197 1198####################### 1199# 1200# ISA Parser 1201# parses ISA DSL and emits C++ headers and source 1202# 1203 1204class ISAParser(Grammar): 1205 class CpuModel(object): 1206 def __init__(self, name, filename, includes, strings): 1207 self.name = name 1208 self.filename = filename 1209 self.includes = includes 1210 self.strings = strings 1211 1212 def __init__(self, output_dir): 1213 super(ISAParser, self).__init__() 1214 self.output_dir = output_dir 1215 1216 self.filename = None # for output file watermarking/scaremongering 1217 1218 self.cpuModels = [ 1219 ISAParser.CpuModel('ExecContext', 1220 'generic_cpu_exec.cc', 1221 '#include "cpu/exec_context.hh"', 1222 { "CPU_exec_context" : "ExecContext" }), 1223 ] 1224 1225 # variable to hold templates 1226 self.templateMap = {} 1227 1228 # This dictionary maps format name strings to Format objects. 1229 self.formatMap = {} 1230 1231 # Track open files and, if applicable, how many chunks it has been 1232 # split into so far. 1233 self.files = {} 1234 self.splits = {} 1235 1236 # isa_name / namespace identifier from namespace declaration. 1237 # before the namespace declaration, None. 1238 self.isa_name = None 1239 self.namespace = None 1240 1241 # The format stack. 1242 self.formatStack = Stack(NoFormat()) 1243 1244 # The default case stack. 1245 self.defaultStack = Stack(None) 1246 1247 # Stack that tracks current file and line number. Each 1248 # element is a tuple (filename, lineno) that records the 1249 # *current* filename and the line number in the *previous* 1250 # file where it was included. 1251 self.fileNameStack = Stack() 1252 1253 symbols = ('makeList', 're', 'string') 1254 self.exportContext = dict([(s, eval(s)) for s in symbols]) 1255 1256 self.maxInstSrcRegs = 0 1257 self.maxInstDestRegs = 0 1258 self.maxMiscDestRegs = 0 1259 1260 def __getitem__(self, i): # Allow object (self) to be 1261 return getattr(self, i) # passed to %-substitutions 1262 1263 # Change the file suffix of a base filename: 1264 # (e.g.) decoder.cc -> decoder-g.cc.inc for 'global' outputs 1265 def suffixize(self, s, sec): 1266 extn = re.compile('(\.[^\.]+)$') # isolate extension 1267 if self.namespace: 1268 return extn.sub(r'-ns\1.inc', s) # insert some text on either side 1269 else: 1270 return extn.sub(r'-g\1.inc', s) 1271 1272 # Get the file object for emitting code into the specified section 1273 # (header, decoder, exec, decode_block). 1274 def get_file(self, section): 1275 if section == 'decode_block': 1276 filename = 'decode-method.cc.inc' 1277 else: 1278 if section == 'header': 1279 file = 'decoder.hh' 1280 else: 1281 file = '%s.cc' % section 1282 filename = self.suffixize(file, section) 1283 try: 1284 return self.files[filename] 1285 except KeyError: pass 1286 1287 f = self.open(filename) 1288 self.files[filename] = f 1289 1290 # The splittable files are the ones with many independent 1291 # per-instruction functions - the decoder's instruction constructors 1292 # and the instruction execution (execute()) methods. These both have 1293 # the suffix -ns.cc.inc, meaning they are within the namespace part 1294 # of the ISA, contain object-emitting C++ source, and are included 1295 # into other top-level files. These are the files that need special 1296 # #define's to allow parts of them to be compiled separately. Rather 1297 # than splitting the emissions into separate files, the monolithic 1298 # output of the ISA parser is maintained, but the value (or lack 1299 # thereof) of the __SPLIT definition during C preprocessing will 1300 # select the different chunks. If no 'split' directives are used, 1301 # the cpp emissions have no effect. 1302 if re.search('-ns.cc.inc$', filename): 1303 print >>f, '#if !defined(__SPLIT) || (__SPLIT == 1)' 1304 self.splits[f] = 1 1305 # ensure requisite #include's 1306 elif filename in ['decoder-g.cc.inc', 'exec-g.cc.inc']: 1307 print >>f, '#include "decoder.hh"' 1308 elif filename == 'decoder-g.hh.inc': 1309 print >>f, '#include "base/bitfield.hh"' 1310 1311 return f 1312 1313 # Weave together the parts of the different output sections by 1314 # #include'ing them into some very short top-level .cc/.hh files. 1315 # These small files make it much clearer how this tool works, since 1316 # you directly see the chunks emitted as files that are #include'd. 1317 def write_top_level_files(self): 1318 dep = self.open('inc.d', bare=True) 1319 1320 # decoder header - everything depends on this 1321 file = 'decoder.hh' 1322 with self.open(file) as f: 1323 inc = [] 1324 1325 fn = 'decoder-g.hh.inc' 1326 assert(fn in self.files) 1327 f.write('#include "%s"\n' % fn) 1328 inc.append(fn) 1329 1330 fn = 'decoder-ns.hh.inc' 1331 assert(fn in self.files) 1332 f.write('namespace %s {\n#include "%s"\n}\n' 1333 % (self.namespace, fn)) 1334 inc.append(fn) 1335 1336 print >>dep, file+':', ' '.join(inc) 1337 1338 # decoder method - cannot be split 1339 file = 'decoder.cc' 1340 with self.open(file) as f: 1341 inc = [] 1342 1343 fn = 'decoder-g.cc.inc' 1344 assert(fn in self.files) 1345 f.write('#include "%s"\n' % fn) 1346 inc.append(fn) 1347 1348 fn = 'decode-method.cc.inc' 1349 # is guaranteed to have been written for parse to complete 1350 f.write('#include "%s"\n' % fn) 1351 inc.append(fn) 1352 1353 inc.append("decoder.hh") 1354 print >>dep, file+':', ' '.join(inc) 1355 1356 extn = re.compile('(\.[^\.]+)$') 1357 1358 # instruction constructors 1359 splits = self.splits[self.get_file('decoder')] 1360 file_ = 'inst-constrs.cc' 1361 for i in range(1, splits+1): 1362 if splits > 1: 1363 file = extn.sub(r'-%d\1' % i, file_) 1364 else: 1365 file = file_ 1366 with self.open(file) as f: 1367 inc = [] 1368 1369 fn = 'decoder-g.cc.inc' 1370 assert(fn in self.files) 1371 f.write('#include "%s"\n' % fn) 1372 inc.append(fn) 1373 1374 fn = 'decoder-ns.cc.inc' 1375 assert(fn in self.files) 1376 print >>f, 'namespace %s {' % self.namespace 1377 if splits > 1: 1378 print >>f, '#define __SPLIT %u' % i 1379 print >>f, '#include "%s"' % fn 1380 print >>f, '}' 1381 inc.append(fn) 1382 1383 inc.append("decoder.hh") 1384 print >>dep, file+':', ' '.join(inc) 1385 1386 # instruction execution per-CPU model 1387 splits = self.splits[self.get_file('exec')] 1388 for cpu in self.cpuModels: 1389 for i in range(1, splits+1): 1390 if splits > 1: 1391 file = extn.sub(r'_%d\1' % i, cpu.filename) 1392 else: 1393 file = cpu.filename 1394 with self.open(file) as f: 1395 inc = [] 1396 1397 fn = 'exec-g.cc.inc' 1398 assert(fn in self.files) 1399 f.write('#include "%s"\n' % fn) 1400 inc.append(fn) 1401 1402 f.write(cpu.includes+"\n") 1403 1404 fn = 'exec-ns.cc.inc' 1405 assert(fn in self.files) 1406 print >>f, 'namespace %s {' % self.namespace 1407 print >>f, '#define CPU_EXEC_CONTEXT %s' \ 1408 % cpu.strings['CPU_exec_context'] 1409 if splits > 1: 1410 print >>f, '#define __SPLIT %u' % i 1411 print >>f, '#include "%s"' % fn 1412 print >>f, '}' 1413 inc.append(fn) 1414 1415 inc.append("decoder.hh") 1416 print >>dep, file+':', ' '.join(inc) 1417 1418 # max_inst_regs.hh 1419 self.update('max_inst_regs.hh', 1420 '''namespace %(namespace)s { 1421 const int MaxInstSrcRegs = %(maxInstSrcRegs)d; 1422 const int MaxInstDestRegs = %(maxInstDestRegs)d; 1423 const int MaxMiscDestRegs = %(maxMiscDestRegs)d;\n}\n''' % self) 1424 print >>dep, 'max_inst_regs.hh:' 1425 1426 dep.close() 1427 1428 1429 scaremonger_template ='''// DO NOT EDIT 1430// This file was automatically generated from an ISA description: 1431// %(filename)s 1432 1433'''; 1434 1435 ##################################################################### 1436 # 1437 # Lexer 1438 # 1439 # The PLY lexer module takes two things as input: 1440 # - A list of token names (the string list 'tokens') 1441 # - A regular expression describing a match for each token. The 1442 # regexp for token FOO can be provided in two ways: 1443 # - as a string variable named t_FOO 1444 # - as the doc string for a function named t_FOO. In this case, 1445 # the function is also executed, allowing an action to be 1446 # associated with each token match. 1447 # 1448 ##################################################################### 1449 1450 # Reserved words. These are listed separately as they are matched 1451 # using the same regexp as generic IDs, but distinguished in the 1452 # t_ID() function. The PLY documentation suggests this approach. 1453 reserved = ( 1454 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT', 1455 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS', 1456 'OUTPUT', 'SIGNED', 'SPLIT', 'TEMPLATE' 1457 ) 1458 1459 # List of tokens. The lex module requires this. 1460 tokens = reserved + ( 1461 # identifier 1462 'ID', 1463 1464 # integer literal 1465 'INTLIT', 1466 1467 # string literal 1468 'STRLIT', 1469 1470 # code literal 1471 'CODELIT', 1472 1473 # ( ) [ ] { } < > , ; . : :: * 1474 'LPAREN', 'RPAREN', 1475 'LBRACKET', 'RBRACKET', 1476 'LBRACE', 'RBRACE', 1477 'LESS', 'GREATER', 'EQUALS', 1478 'COMMA', 'SEMI', 'DOT', 'COLON', 'DBLCOLON', 1479 'ASTERISK', 1480 1481 # C preprocessor directives 1482 'CPPDIRECTIVE' 1483 1484 # The following are matched but never returned. commented out to 1485 # suppress PLY warning 1486 # newfile directive 1487 # 'NEWFILE', 1488 1489 # endfile directive 1490 # 'ENDFILE' 1491 ) 1492 1493 # Regular expressions for token matching 1494 t_LPAREN = r'\(' 1495 t_RPAREN = r'\)' 1496 t_LBRACKET = r'\[' 1497 t_RBRACKET = r'\]' 1498 t_LBRACE = r'\{' 1499 t_RBRACE = r'\}' 1500 t_LESS = r'\<' 1501 t_GREATER = r'\>' 1502 t_EQUALS = r'=' 1503 t_COMMA = r',' 1504 t_SEMI = r';' 1505 t_DOT = r'\.' 1506 t_COLON = r':' 1507 t_DBLCOLON = r'::' 1508 t_ASTERISK = r'\*' 1509 1510 # Identifiers and reserved words 1511 reserved_map = { } 1512 for r in reserved: 1513 reserved_map[r.lower()] = r 1514 1515 def t_ID(self, t): 1516 r'[A-Za-z_]\w*' 1517 t.type = self.reserved_map.get(t.value, 'ID') 1518 return t 1519 1520 # Integer literal 1521 def t_INTLIT(self, t): 1522 r'-?(0x[\da-fA-F]+)|\d+' 1523 try: 1524 t.value = int(t.value,0) 1525 except ValueError: 1526 error(t.lexer.lineno, 'Integer value "%s" too large' % t.value) 1527 t.value = 0 1528 return t 1529 1530 # String literal. Note that these use only single quotes, and 1531 # can span multiple lines. 1532 def t_STRLIT(self, t): 1533 r"(?m)'([^'])+'" 1534 # strip off quotes 1535 t.value = t.value[1:-1] 1536 t.lexer.lineno += t.value.count('\n') 1537 return t 1538 1539 1540 # "Code literal"... like a string literal, but delimiters are 1541 # '{{' and '}}' so they get formatted nicely under emacs c-mode 1542 def t_CODELIT(self, t): 1543 r"(?m)\{\{([^\}]|}(?!\}))+\}\}" 1544 # strip off {{ & }} 1545 t.value = t.value[2:-2] 1546 t.lexer.lineno += t.value.count('\n') 1547 return t 1548 1549 def t_CPPDIRECTIVE(self, t): 1550 r'^\#[^\#].*\n' 1551 t.lexer.lineno += t.value.count('\n') 1552 return t 1553 1554 def t_NEWFILE(self, t): 1555 r'^\#\#newfile\s+"[^"]*"\n' 1556 self.fileNameStack.push(t.lexer.lineno) 1557 t.lexer.lineno = LineTracker(t.value[11:-2]) 1558 1559 def t_ENDFILE(self, t): 1560 r'^\#\#endfile\n' 1561 t.lexer.lineno = self.fileNameStack.pop() 1562 1563 # 1564 # The functions t_NEWLINE, t_ignore, and t_error are 1565 # special for the lex module. 1566 # 1567 1568 # Newlines 1569 def t_NEWLINE(self, t): 1570 r'\n+' 1571 t.lexer.lineno += t.value.count('\n') 1572 1573 # Comments 1574 def t_comment(self, t): 1575 r'//.*' 1576 1577 # Completely ignored characters 1578 t_ignore = ' \t\x0c' 1579 1580 # Error handler 1581 def t_error(self, t): 1582 error(t.lexer.lineno, "illegal character '%s'" % t.value[0]) 1583 t.skip(1) 1584 1585 ##################################################################### 1586 # 1587 # Parser 1588 # 1589 # Every function whose name starts with 'p_' defines a grammar 1590 # rule. The rule is encoded in the function's doc string, while 1591 # the function body provides the action taken when the rule is 1592 # matched. The argument to each function is a list of the values 1593 # of the rule's symbols: t[0] for the LHS, and t[1..n] for the 1594 # symbols on the RHS. For tokens, the value is copied from the 1595 # t.value attribute provided by the lexer. For non-terminals, the 1596 # value is assigned by the producing rule; i.e., the job of the 1597 # grammar rule function is to set the value for the non-terminal 1598 # on the LHS (by assigning to t[0]). 1599 ##################################################################### 1600 1601 # The LHS of the first grammar rule is used as the start symbol 1602 # (in this case, 'specification'). Note that this rule enforces 1603 # that there will be exactly one namespace declaration, with 0 or 1604 # more global defs/decls before and after it. The defs & decls 1605 # before the namespace decl will be outside the namespace; those 1606 # after will be inside. The decoder function is always inside the 1607 # namespace. 1608 def p_specification(self, t): 1609 'specification : opt_defs_and_outputs top_level_decode_block' 1610 1611 for f in self.splits.iterkeys(): 1612 f.write('\n#endif\n') 1613 1614 for f in self.files.itervalues(): # close ALL the files; 1615 f.close() # not doing so can cause compilation to fail 1616 1617 self.write_top_level_files() 1618 1619 t[0] = True 1620 1621 # 'opt_defs_and_outputs' is a possibly empty sequence of def and/or 1622 # output statements. Its productions do the hard work of eventually 1623 # instantiating a GenCode, which are generally emitted (written to disk) 1624 # as soon as possible, except for the decode_block, which has to be 1625 # accumulated into one large function of nested switch/case blocks. 1626 def p_opt_defs_and_outputs_0(self, t): 1627 'opt_defs_and_outputs : empty' 1628 1629 def p_opt_defs_and_outputs_1(self, t): 1630 'opt_defs_and_outputs : defs_and_outputs' 1631 1632 def p_defs_and_outputs_0(self, t): 1633 'defs_and_outputs : def_or_output' 1634 1635 def p_defs_and_outputs_1(self, t): 1636 'defs_and_outputs : defs_and_outputs def_or_output' 1637 1638 # The list of possible definition/output statements. 1639 # They are all processed as they are seen. 1640 def p_def_or_output(self, t): 1641 '''def_or_output : name_decl 1642 | def_format 1643 | def_bitfield 1644 | def_bitfield_struct 1645 | def_template 1646 | def_operand_types 1647 | def_operands 1648 | output 1649 | global_let 1650 | split''' 1651 1652 # Utility function used by both invocations of splitting - explicit 1653 # 'split' keyword and split() function inside "let {{ }};" blocks. 1654 def split(self, sec, write=False): 1655 assert(sec != 'header' and "header cannot be split") 1656 1657 f = self.get_file(sec) 1658 self.splits[f] += 1 1659 s = '\n#endif\n#if __SPLIT == %u\n' % self.splits[f] 1660 if write: 1661 f.write(s) 1662 else: 1663 return s 1664 1665 # split output file to reduce compilation time 1666 def p_split(self, t): 1667 'split : SPLIT output_type SEMI' 1668 assert(self.isa_name and "'split' not allowed before namespace decl") 1669 1670 self.split(t[2], True) 1671 1672 def p_output_type(self, t): 1673 '''output_type : DECODER 1674 | HEADER 1675 | EXEC''' 1676 t[0] = t[1] 1677 1678 # ISA name declaration looks like "namespace <foo>;" 1679 def p_name_decl(self, t): 1680 'name_decl : NAMESPACE ID SEMI' 1681 assert(self.isa_name == None and "Only 1 namespace decl permitted") 1682 self.isa_name = t[2] 1683 self.namespace = t[2] + 'Inst' 1684 1685 # Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied 1686 # directly to the appropriate output section. 1687 1688 # Massage output block by substituting in template definitions and 1689 # bit operators. We handle '%'s embedded in the string that don't 1690 # indicate template substitutions (or CPU-specific symbols, which 1691 # get handled in GenCode) by doubling them first so that the 1692 # format operation will reduce them back to single '%'s. 1693 def process_output(self, s): 1694 s = self.protectNonSubstPercents(s) 1695 # protects cpu-specific symbols too 1696 s = self.protectCpuSymbols(s) 1697 return substBitOps(s % self.templateMap) 1698 1699 def p_output(self, t): 1700 'output : OUTPUT output_type CODELIT SEMI' 1701 kwargs = { t[2]+'_output' : self.process_output(t[3]) } 1702 GenCode(self, **kwargs).emit() 1703 1704 # global let blocks 'let {{...}}' (Python code blocks) are 1705 # executed directly when seen. Note that these execute in a 1706 # special variable context 'exportContext' to prevent the code 1707 # from polluting this script's namespace. 1708 def p_global_let(self, t): 1709 'global_let : LET CODELIT SEMI' 1710 def _split(sec): 1711 return self.split(sec) 1712 self.updateExportContext() 1713 self.exportContext["header_output"] = '' 1714 self.exportContext["decoder_output"] = '' 1715 self.exportContext["exec_output"] = '' 1716 self.exportContext["decode_block"] = '' 1717 self.exportContext["split"] = _split 1718 split_setup = ''' 1719def wrap(func): 1720 def split(sec): 1721 globals()[sec + '_output'] += func(sec) 1722 return split 1723split = wrap(split) 1724del wrap 1725''' 1726 # This tricky setup (immediately above) allows us to just write 1727 # (e.g.) "split('exec')" in the Python code and the split #ifdef's 1728 # will automatically be added to the exec_output variable. The inner 1729 # Python execution environment doesn't know about the split points, 1730 # so we carefully inject and wrap a closure that can retrieve the 1731 # next split's #define from the parser and add it to the current 1732 # emission-in-progress. 1733 try: 1734 exec split_setup+fixPythonIndentation(t[2]) in self.exportContext 1735 except Exception, exc: 1736 if debug: 1737 raise 1738 error(t.lineno(1), 'In global let block: %s' % exc) 1739 GenCode(self, 1740 header_output=self.exportContext["header_output"], 1741 decoder_output=self.exportContext["decoder_output"], 1742 exec_output=self.exportContext["exec_output"], 1743 decode_block=self.exportContext["decode_block"]).emit() 1744 1745 # Define the mapping from operand type extensions to C++ types and 1746 # bit widths (stored in operandTypeMap). 1747 def p_def_operand_types(self, t): 1748 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI' 1749 try: 1750 self.operandTypeMap = eval('{' + t[3] + '}') 1751 except Exception, exc: 1752 if debug: 1753 raise 1754 error(t.lineno(1), 1755 'In def operand_types: %s' % exc) 1756 1757 # Define the mapping from operand names to operand classes and 1758 # other traits. Stored in operandNameMap. 1759 def p_def_operands(self, t): 1760 'def_operands : DEF OPERANDS CODELIT SEMI' 1761 if not hasattr(self, 'operandTypeMap'): 1762 error(t.lineno(1), 1763 'error: operand types must be defined before operands') 1764 try: 1765 user_dict = eval('{' + t[3] + '}', self.exportContext) 1766 except Exception, exc: 1767 if debug: 1768 raise 1769 error(t.lineno(1), 'In def operands: %s' % exc) 1770 self.buildOperandNameMap(user_dict, t.lexer.lineno) 1771 1772 # A bitfield definition looks like: 1773 # 'def [signed] bitfield <ID> [<first>:<last>]' 1774 # This generates a preprocessor macro in the output file. 1775 def p_def_bitfield_0(self, t): 1776 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI' 1777 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8]) 1778 if (t[2] == 'signed'): 1779 expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr) 1780 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 1781 GenCode(self, header_output=hash_define).emit() 1782 1783 # alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]' 1784 def p_def_bitfield_1(self, t): 1785 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI' 1786 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6]) 1787 if (t[2] == 'signed'): 1788 expr = 'sext<%d>(%s)' % (1, expr) 1789 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 1790 GenCode(self, header_output=hash_define).emit() 1791 1792 # alternate form for structure member: 'def bitfield <ID> <ID>' 1793 def p_def_bitfield_struct(self, t): 1794 'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI' 1795 if (t[2] != ''): 1796 error(t.lineno(1), 1797 'error: structure bitfields are always unsigned.') 1798 expr = 'machInst.%s' % t[5] 1799 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 1800 GenCode(self, header_output=hash_define).emit() 1801 1802 def p_id_with_dot_0(self, t): 1803 'id_with_dot : ID' 1804 t[0] = t[1] 1805 1806 def p_id_with_dot_1(self, t): 1807 'id_with_dot : ID DOT id_with_dot' 1808 t[0] = t[1] + t[2] + t[3] 1809 1810 def p_opt_signed_0(self, t): 1811 'opt_signed : SIGNED' 1812 t[0] = t[1] 1813 1814 def p_opt_signed_1(self, t): 1815 'opt_signed : empty' 1816 t[0] = '' 1817 1818 def p_def_template(self, t): 1819 'def_template : DEF TEMPLATE ID CODELIT SEMI' 1820 if t[3] in self.templateMap: 1821 print "warning: template %s already defined" % t[3] 1822 self.templateMap[t[3]] = Template(self, t[4]) 1823 1824 # An instruction format definition looks like 1825 # "def format <fmt>(<params>) {{...}};" 1826 def p_def_format(self, t): 1827 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI' 1828 (id, params, code) = (t[3], t[5], t[7]) 1829 self.defFormat(id, params, code, t.lexer.lineno) 1830 1831 # The formal parameter list for an instruction format is a 1832 # possibly empty list of comma-separated parameters. Positional 1833 # (standard, non-keyword) parameters must come first, followed by 1834 # keyword parameters, followed by a '*foo' parameter that gets 1835 # excess positional arguments (as in Python). Each of these three 1836 # parameter categories is optional. 1837 # 1838 # Note that we do not support the '**foo' parameter for collecting 1839 # otherwise undefined keyword args. Otherwise the parameter list 1840 # is (I believe) identical to what is supported in Python. 1841 # 1842 # The param list generates a tuple, where the first element is a 1843 # list of the positional params and the second element is a dict 1844 # containing the keyword params. 1845 def p_param_list_0(self, t): 1846 'param_list : positional_param_list COMMA nonpositional_param_list' 1847 t[0] = t[1] + t[3] 1848 1849 def p_param_list_1(self, t): 1850 '''param_list : positional_param_list 1851 | nonpositional_param_list''' 1852 t[0] = t[1] 1853 1854 def p_positional_param_list_0(self, t): 1855 'positional_param_list : empty' 1856 t[0] = [] 1857 1858 def p_positional_param_list_1(self, t): 1859 'positional_param_list : ID' 1860 t[0] = [t[1]] 1861 1862 def p_positional_param_list_2(self, t): 1863 'positional_param_list : positional_param_list COMMA ID' 1864 t[0] = t[1] + [t[3]] 1865 1866 def p_nonpositional_param_list_0(self, t): 1867 'nonpositional_param_list : keyword_param_list COMMA excess_args_param' 1868 t[0] = t[1] + t[3] 1869 1870 def p_nonpositional_param_list_1(self, t): 1871 '''nonpositional_param_list : keyword_param_list 1872 | excess_args_param''' 1873 t[0] = t[1] 1874 1875 def p_keyword_param_list_0(self, t): 1876 'keyword_param_list : keyword_param' 1877 t[0] = [t[1]] 1878 1879 def p_keyword_param_list_1(self, t): 1880 'keyword_param_list : keyword_param_list COMMA keyword_param' 1881 t[0] = t[1] + [t[3]] 1882 1883 def p_keyword_param(self, t): 1884 'keyword_param : ID EQUALS expr' 1885 t[0] = t[1] + ' = ' + t[3].__repr__() 1886 1887 def p_excess_args_param(self, t): 1888 'excess_args_param : ASTERISK ID' 1889 # Just concatenate them: '*ID'. Wrap in list to be consistent 1890 # with positional_param_list and keyword_param_list. 1891 t[0] = [t[1] + t[2]] 1892 1893 # End of format definition-related rules. 1894 ############## 1895 1896 # 1897 # A decode block looks like: 1898 # decode <field1> [, <field2>]* [default <inst>] { ... } 1899 # 1900 def p_top_level_decode_block(self, t): 1901 'top_level_decode_block : decode_block' 1902 codeObj = t[1] 1903 codeObj.wrap_decode_block(''' 1904StaticInstPtr 1905%(isa_name)s::Decoder::decodeInst(%(isa_name)s::ExtMachInst machInst) 1906{ 1907 using namespace %(namespace)s; 1908''' % self, '}') 1909 1910 codeObj.emit() 1911 1912 def p_decode_block(self, t): 1913 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE' 1914 default_defaults = self.defaultStack.pop() 1915 codeObj = t[5] 1916 # use the "default defaults" only if there was no explicit 1917 # default statement in decode_stmt_list 1918 if not codeObj.has_decode_default: 1919 codeObj += default_defaults 1920 codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n') 1921 t[0] = codeObj 1922 1923 # The opt_default statement serves only to push the "default 1924 # defaults" onto defaultStack. This value will be used by nested 1925 # decode blocks, and used and popped off when the current 1926 # decode_block is processed (in p_decode_block() above). 1927 def p_opt_default_0(self, t): 1928 'opt_default : empty' 1929 # no default specified: reuse the one currently at the top of 1930 # the stack 1931 self.defaultStack.push(self.defaultStack.top()) 1932 # no meaningful value returned 1933 t[0] = None 1934 1935 def p_opt_default_1(self, t): 1936 'opt_default : DEFAULT inst' 1937 # push the new default 1938 codeObj = t[2] 1939 codeObj.wrap_decode_block('\ndefault:\n', 'break;\n') 1940 self.defaultStack.push(codeObj) 1941 # no meaningful value returned 1942 t[0] = None 1943 1944 def p_decode_stmt_list_0(self, t): 1945 'decode_stmt_list : decode_stmt' 1946 t[0] = t[1] 1947 1948 def p_decode_stmt_list_1(self, t): 1949 'decode_stmt_list : decode_stmt decode_stmt_list' 1950 if (t[1].has_decode_default and t[2].has_decode_default): 1951 error(t.lineno(1), 'Two default cases in decode block') 1952 t[0] = t[1] + t[2] 1953 1954 # 1955 # Decode statement rules 1956 # 1957 # There are four types of statements allowed in a decode block: 1958 # 1. Format blocks 'format <foo> { ... }' 1959 # 2. Nested decode blocks 1960 # 3. Instruction definitions. 1961 # 4. C preprocessor directives. 1962 1963 1964 # Preprocessor directives found in a decode statement list are 1965 # passed through to the output, replicated to all of the output 1966 # code streams. This works well for ifdefs, so we can ifdef out 1967 # both the declarations and the decode cases generated by an 1968 # instruction definition. Handling them as part of the grammar 1969 # makes it easy to keep them in the right place with respect to 1970 # the code generated by the other statements. 1971 def p_decode_stmt_cpp(self, t): 1972 'decode_stmt : CPPDIRECTIVE' 1973 t[0] = GenCode(self, t[1], t[1], t[1], t[1]) 1974 1975 # A format block 'format <foo> { ... }' sets the default 1976 # instruction format used to handle instruction definitions inside 1977 # the block. This format can be overridden by using an explicit 1978 # format on the instruction definition or with a nested format 1979 # block. 1980 def p_decode_stmt_format(self, t): 1981 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE' 1982 # The format will be pushed on the stack when 'push_format_id' 1983 # is processed (see below). Once the parser has recognized 1984 # the full production (though the right brace), we're done 1985 # with the format, so now we can pop it. 1986 self.formatStack.pop() 1987 t[0] = t[4] 1988 1989 # This rule exists so we can set the current format (& push the 1990 # stack) when we recognize the format name part of the format 1991 # block. 1992 def p_push_format_id(self, t): 1993 'push_format_id : ID' 1994 try: 1995 self.formatStack.push(self.formatMap[t[1]]) 1996 t[0] = ('', '// format %s' % t[1]) 1997 except KeyError: 1998 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 1999 2000 # Nested decode block: if the value of the current field matches 2001 # the specified constant(s), do a nested decode on some other field. 2002 def p_decode_stmt_decode(self, t): 2003 'decode_stmt : case_list COLON decode_block' 2004 case_list = t[1] 2005 codeObj = t[3] 2006 # just wrap the decoding code from the block as a case in the 2007 # outer switch statement. 2008 codeObj.wrap_decode_block('\n%s\n' % ''.join(case_list)) 2009 codeObj.has_decode_default = (case_list == ['default:']) 2010 t[0] = codeObj 2011 2012 # Instruction definition (finally!). 2013 def p_decode_stmt_inst(self, t): 2014 'decode_stmt : case_list COLON inst SEMI' 2015 case_list = t[1] 2016 codeObj = t[3] 2017 codeObj.wrap_decode_block('\n%s' % ''.join(case_list), 'break;\n') 2018 codeObj.has_decode_default = (case_list == ['default:']) 2019 t[0] = codeObj 2020 2021 # The constant list for a decode case label must be non-empty, and must 2022 # either be the keyword 'default', or made up of one or more 2023 # comma-separated integer literals or strings which evaluate to 2024 # constants when compiled as C++. 2025 def p_case_list_0(self, t): 2026 'case_list : DEFAULT' 2027 t[0] = ['default:'] 2028 2029 def prep_int_lit_case_label(self, lit): 2030 if lit >= 2**32: 2031 return 'case ULL(%#x): ' % lit 2032 else: 2033 return 'case %#x: ' % lit 2034 2035 def prep_str_lit_case_label(self, lit): 2036 return 'case %s: ' % lit 2037 2038 def p_case_list_1(self, t): 2039 'case_list : INTLIT' 2040 t[0] = [self.prep_int_lit_case_label(t[1])] 2041 2042 def p_case_list_2(self, t): 2043 'case_list : STRLIT' 2044 t[0] = [self.prep_str_lit_case_label(t[1])] 2045 2046 def p_case_list_3(self, t): 2047 'case_list : case_list COMMA INTLIT' 2048 t[0] = t[1] 2049 t[0].append(self.prep_int_lit_case_label(t[3])) 2050 2051 def p_case_list_4(self, t): 2052 'case_list : case_list COMMA STRLIT' 2053 t[0] = t[1] 2054 t[0].append(self.prep_str_lit_case_label(t[3])) 2055 2056 # Define an instruction using the current instruction format 2057 # (specified by an enclosing format block). 2058 # "<mnemonic>(<args>)" 2059 def p_inst_0(self, t): 2060 'inst : ID LPAREN arg_list RPAREN' 2061 # Pass the ID and arg list to the current format class to deal with. 2062 currentFormat = self.formatStack.top() 2063 codeObj = currentFormat.defineInst(self, t[1], t[3], t.lexer.lineno) 2064 args = ','.join(map(str, t[3])) 2065 args = re.sub('(?m)^', '//', args) 2066 args = re.sub('^//', '', args) 2067 comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args) 2068 codeObj.prepend_all(comment) 2069 t[0] = codeObj 2070 2071 # Define an instruction using an explicitly specified format: 2072 # "<fmt>::<mnemonic>(<args>)" 2073 def p_inst_1(self, t): 2074 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN' 2075 try: 2076 format = self.formatMap[t[1]] 2077 except KeyError: 2078 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 2079 2080 codeObj = format.defineInst(self, t[3], t[5], t.lexer.lineno) 2081 comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5]) 2082 codeObj.prepend_all(comment) 2083 t[0] = codeObj 2084 2085 # The arg list generates a tuple, where the first element is a 2086 # list of the positional args and the second element is a dict 2087 # containing the keyword args. 2088 def p_arg_list_0(self, t): 2089 'arg_list : positional_arg_list COMMA keyword_arg_list' 2090 t[0] = ( t[1], t[3] ) 2091 2092 def p_arg_list_1(self, t): 2093 'arg_list : positional_arg_list' 2094 t[0] = ( t[1], {} ) 2095 2096 def p_arg_list_2(self, t): 2097 'arg_list : keyword_arg_list' 2098 t[0] = ( [], t[1] ) 2099 2100 def p_positional_arg_list_0(self, t): 2101 'positional_arg_list : empty' 2102 t[0] = [] 2103 2104 def p_positional_arg_list_1(self, t): 2105 'positional_arg_list : expr' 2106 t[0] = [t[1]] 2107 2108 def p_positional_arg_list_2(self, t): 2109 'positional_arg_list : positional_arg_list COMMA expr' 2110 t[0] = t[1] + [t[3]] 2111 2112 def p_keyword_arg_list_0(self, t): 2113 'keyword_arg_list : keyword_arg' 2114 t[0] = t[1] 2115 2116 def p_keyword_arg_list_1(self, t): 2117 'keyword_arg_list : keyword_arg_list COMMA keyword_arg' 2118 t[0] = t[1] 2119 t[0].update(t[3]) 2120 2121 def p_keyword_arg(self, t): 2122 'keyword_arg : ID EQUALS expr' 2123 t[0] = { t[1] : t[3] } 2124 2125 # 2126 # Basic expressions. These constitute the argument values of 2127 # "function calls" (i.e. instruction definitions in the decode 2128 # block) and default values for formal parameters of format 2129 # functions. 2130 # 2131 # Right now, these are either strings, integers, or (recursively) 2132 # lists of exprs (using Python square-bracket list syntax). Note 2133 # that bare identifiers are trated as string constants here (since 2134 # there isn't really a variable namespace to refer to). 2135 # 2136 def p_expr_0(self, t): 2137 '''expr : ID 2138 | INTLIT 2139 | STRLIT 2140 | CODELIT''' 2141 t[0] = t[1] 2142 2143 def p_expr_1(self, t): 2144 '''expr : LBRACKET list_expr RBRACKET''' 2145 t[0] = t[2] 2146 2147 def p_list_expr_0(self, t): 2148 'list_expr : expr' 2149 t[0] = [t[1]] 2150 2151 def p_list_expr_1(self, t): 2152 'list_expr : list_expr COMMA expr' 2153 t[0] = t[1] + [t[3]] 2154 2155 def p_list_expr_2(self, t): 2156 'list_expr : empty' 2157 t[0] = [] 2158 2159 # 2160 # Empty production... use in other rules for readability. 2161 # 2162 def p_empty(self, t): 2163 'empty :' 2164 pass 2165 2166 # Parse error handler. Note that the argument here is the 2167 # offending *token*, not a grammar symbol (hence the need to use 2168 # t.value) 2169 def p_error(self, t): 2170 if t: 2171 error(t.lexer.lineno, "syntax error at '%s'" % t.value) 2172 else: 2173 error("unknown syntax error") 2174 2175 # END OF GRAMMAR RULES 2176 2177 def updateExportContext(self): 2178 2179 # create a continuation that allows us to grab the current parser 2180 def wrapInstObjParams(*args): 2181 return InstObjParams(self, *args) 2182 self.exportContext['InstObjParams'] = wrapInstObjParams 2183 self.exportContext.update(self.templateMap) 2184 2185 def defFormat(self, id, params, code, lineno): 2186 '''Define a new format''' 2187 2188 # make sure we haven't already defined this one 2189 if id in self.formatMap: 2190 error(lineno, 'format %s redefined.' % id) 2191 2192 # create new object and store in global map 2193 self.formatMap[id] = Format(id, params, code) 2194 2195 def expandCpuSymbolsToDict(self, template): 2196 '''Expand template with CPU-specific references into a 2197 dictionary with an entry for each CPU model name. The entry 2198 key is the model name and the corresponding value is the 2199 template with the CPU-specific refs substituted for that 2200 model.''' 2201 2202 # Protect '%'s that don't go with CPU-specific terms 2203 t = re.sub(r'%(?!\(CPU_)', '%%', template) 2204 result = {} 2205 for cpu in self.cpuModels: 2206 result[cpu.name] = t % cpu.strings 2207 return result 2208 2209 def expandCpuSymbolsToString(self, template): 2210 '''*If* the template has CPU-specific references, return a 2211 single string containing a copy of the template for each CPU 2212 model with the corresponding values substituted in. If the 2213 template has no CPU-specific references, it is returned 2214 unmodified.''' 2215 2216 if template.find('%(CPU_') != -1: 2217 return reduce(lambda x,y: x+y, 2218 self.expandCpuSymbolsToDict(template).values()) 2219 else: 2220 return template 2221 2222 def protectCpuSymbols(self, template): 2223 '''Protect CPU-specific references by doubling the 2224 corresponding '%'s (in preparation for substituting a different 2225 set of references into the template).''' 2226 2227 return re.sub(r'%(?=\(CPU_)', '%%', template) 2228 2229 def protectNonSubstPercents(self, s): 2230 '''Protect any non-dict-substitution '%'s in a format string 2231 (i.e. those not followed by '(')''' 2232 2233 return re.sub(r'%(?!\()', '%%', s) 2234 2235 def buildOperandNameMap(self, user_dict, lineno): 2236 operand_name = {} 2237 for op_name, val in user_dict.iteritems(): 2238 2239 # Check if extra attributes have been specified. 2240 if len(val) > 9: 2241 error(lineno, 'error: too many attributes for operand "%s"' % 2242 base_cls_name) 2243 2244 # Pad val with None in case optional args are missing 2245 val += (None, None, None, None) 2246 base_cls_name, dflt_ext, reg_spec, flags, sort_pri, \ 2247 read_code, write_code, read_predicate, write_predicate = val[:9] 2248 2249 # Canonical flag structure is a triple of lists, where each list 2250 # indicates the set of flags implied by this operand always, when 2251 # used as a source, and when used as a dest, respectively. 2252 # For simplicity this can be initialized using a variety of fairly 2253 # obvious shortcuts; we convert these to canonical form here. 2254 if not flags: 2255 # no flags specified (e.g., 'None') 2256 flags = ( [], [], [] ) 2257 elif isinstance(flags, str): 2258 # a single flag: assumed to be unconditional 2259 flags = ( [ flags ], [], [] ) 2260 elif isinstance(flags, list): 2261 # a list of flags: also assumed to be unconditional 2262 flags = ( flags, [], [] ) 2263 elif isinstance(flags, tuple): 2264 # it's a tuple: it should be a triple, 2265 # but each item could be a single string or a list 2266 (uncond_flags, src_flags, dest_flags) = flags 2267 flags = (makeList(uncond_flags), 2268 makeList(src_flags), makeList(dest_flags)) 2269 2270 # Accumulate attributes of new operand class in tmp_dict 2271 tmp_dict = {} 2272 attrList = ['reg_spec', 'flags', 'sort_pri', 2273 'read_code', 'write_code', 2274 'read_predicate', 'write_predicate'] 2275 if dflt_ext: 2276 dflt_ctype = self.operandTypeMap[dflt_ext] 2277 attrList.extend(['dflt_ctype', 'dflt_ext']) 2278 for attr in attrList: 2279 tmp_dict[attr] = eval(attr) 2280 tmp_dict['base_name'] = op_name 2281 2282 # New class name will be e.g. "IntReg_Ra" 2283 cls_name = base_cls_name + '_' + op_name 2284 # Evaluate string arg to get class object. Note that the 2285 # actual base class for "IntReg" is "IntRegOperand", i.e. we 2286 # have to append "Operand". 2287 try: 2288 base_cls = eval(base_cls_name + 'Operand') 2289 except NameError: 2290 error(lineno, 2291 'error: unknown operand base class "%s"' % base_cls_name) 2292 # The following statement creates a new class called 2293 # <cls_name> as a subclass of <base_cls> with the attributes 2294 # in tmp_dict, just as if we evaluated a class declaration. 2295 operand_name[op_name] = type(cls_name, (base_cls,), tmp_dict) 2296 2297 self.operandNameMap = operand_name 2298 2299 # Define operand variables. 2300 operands = user_dict.keys() 2301 extensions = self.operandTypeMap.keys() 2302 2303 operandsREString = r''' 2304 (?<!\w) # neg. lookbehind assertion: prevent partial matches 2305 ((%s)(?:_(%s))?) # match: operand with optional '_' then suffix 2306 (?!\w) # neg. lookahead assertion: prevent partial matches 2307 ''' % (string.join(operands, '|'), string.join(extensions, '|')) 2308 2309 self.operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE) 2310 2311 # Same as operandsREString, but extension is mandatory, and only two 2312 # groups are returned (base and ext, not full name as above). 2313 # Used for subtituting '_' for '.' to make C++ identifiers. 2314 operandsWithExtREString = r'(?<!\w)(%s)_(%s)(?!\w)' \ 2315 % (string.join(operands, '|'), string.join(extensions, '|')) 2316 2317 self.operandsWithExtRE = \ 2318 re.compile(operandsWithExtREString, re.MULTILINE) 2319 2320 def substMungedOpNames(self, code): 2321 '''Munge operand names in code string to make legal C++ 2322 variable names. This means getting rid of the type extension 2323 if any. Will match base_name attribute of Operand object.)''' 2324 return self.operandsWithExtRE.sub(r'\1', code) 2325 2326 def mungeSnippet(self, s): 2327 '''Fix up code snippets for final substitution in templates.''' 2328 if isinstance(s, str): 2329 return self.substMungedOpNames(substBitOps(s)) 2330 else: 2331 return s 2332 2333 def open(self, name, bare=False): 2334 '''Open the output file for writing and include scary warning.''' 2335 filename = os.path.join(self.output_dir, name) 2336 f = open(filename, 'w') 2337 if f: 2338 if not bare: 2339 f.write(ISAParser.scaremonger_template % self) 2340 return f 2341 2342 def update(self, file, contents): 2343 '''Update the output file only. Scons should handle the case when 2344 the new contents are unchanged using its built-in hash feature.''' 2345 f = self.open(file) 2346 f.write(contents) 2347 f.close() 2348 2349 # This regular expression matches '##include' directives 2350 includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[^"]*)".*$', 2351 re.MULTILINE) 2352 2353 def replace_include(self, matchobj, dirname): 2354 """Function to replace a matched '##include' directive with the 2355 contents of the specified file (with nested ##includes 2356 replaced recursively). 'matchobj' is an re match object 2357 (from a match of includeRE) and 'dirname' is the directory 2358 relative to which the file path should be resolved.""" 2359 2360 fname = matchobj.group('filename') 2361 full_fname = os.path.normpath(os.path.join(dirname, fname)) 2362 contents = '##newfile "%s"\n%s\n##endfile\n' % \ 2363 (full_fname, self.read_and_flatten(full_fname)) 2364 return contents 2365 2366 def read_and_flatten(self, filename): 2367 """Read a file and recursively flatten nested '##include' files.""" 2368 2369 current_dir = os.path.dirname(filename) 2370 try: 2371 contents = open(filename).read() 2372 except IOError: 2373 error('Error including file "%s"' % filename) 2374 2375 self.fileNameStack.push(LineTracker(filename)) 2376 2377 # Find any includes and include them 2378 def replace(matchobj): 2379 return self.replace_include(matchobj, current_dir) 2380 contents = self.includeRE.sub(replace, contents) 2381 2382 self.fileNameStack.pop() 2383 return contents 2384 2385 AlreadyGenerated = {} 2386 2387 def _parse_isa_desc(self, isa_desc_file): 2388 '''Read in and parse the ISA description.''' 2389 2390 # The build system can end up running the ISA parser twice: once to 2391 # finalize the build dependencies, and then to actually generate 2392 # the files it expects (in src/arch/$ARCH/generated). This code 2393 # doesn't do anything different either time, however; the SCons 2394 # invocations just expect different things. Since this code runs 2395 # within SCons, we can just remember that we've already run and 2396 # not perform a completely unnecessary run, since the ISA parser's 2397 # effect is idempotent. 2398 if isa_desc_file in ISAParser.AlreadyGenerated: 2399 return 2400 2401 # grab the last three path components of isa_desc_file 2402 self.filename = '/'.join(isa_desc_file.split('/')[-3:]) 2403 2404 # Read file and (recursively) all included files into a string. 2405 # PLY requires that the input be in a single string so we have to 2406 # do this up front. 2407 isa_desc = self.read_and_flatten(isa_desc_file) 2408 2409 # Initialize lineno tracker 2410 self.lex.lineno = LineTracker(isa_desc_file) 2411 2412 # Parse. 2413 self.parse_string(isa_desc) 2414 2415 ISAParser.AlreadyGenerated[isa_desc_file] = None 2416 2417 def parse_isa_desc(self, *args, **kwargs): 2418 try: 2419 self._parse_isa_desc(*args, **kwargs) 2420 except ISAParserError, e: 2421 print backtrace(self.fileNameStack) 2422 print "At %s:" % e.lineno 2423 print e 2424 sys.exit(1) 2425 2426# Called as script: get args from command line. 2427# Args are: <isa desc file> <output dir> 2428if __name__ == '__main__': 2429 ISAParser(sys.argv[2]).parse_isa_desc(sys.argv[1]) 2430