isa_parser.py revision 10326
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 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 """Error handler 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 if hasattr(first, 'lexer'): 98 first = first.lexer.lineno 99 self.lineno = first 100 self.string = second 101 102 def display(self, filename_stack, print_traceback=debug): 103 # Output formatted to work under Emacs compile-mode. Optional 104 # 'print_traceback' arg, if set to True, prints a Python stack 105 # backtrace too (can be handy when trying to debug the parser 106 # itself). 107 108 spaces = "" 109 for (filename, line) in filename_stack[:-1]: 110 print "%sIn file included from %s:" % (spaces, filename) 111 spaces += " " 112 113 # Print a Python stack backtrace if requested. 114 if print_traceback or not self.lineno: 115 traceback.print_exc() 116 117 line_str = "%s:" % (filename_stack[-1][0], ) 118 if self.lineno: 119 line_str += "%d:" % (self.lineno, ) 120 121 return "%s%s %s" % (spaces, line_str, self.string) 122 123 def exit(self, filename_stack, print_traceback=debug): 124 # Just call exit. 125 126 sys.exit(self.display(filename_stack, print_traceback)) 127 128def error(*args): 129 raise ISAParserError(*args) 130 131#################### 132# Template objects. 133# 134# Template objects are format strings that allow substitution from 135# the attribute spaces of other objects (e.g. InstObjParams instances). 136 137labelRE = re.compile(r'(?<!%)%\(([^\)]+)\)[sd]') 138 139class Template(object): 140 def __init__(self, parser, t): 141 self.parser = parser 142 self.template = t 143 144 def subst(self, d): 145 myDict = None 146 147 # Protect non-Python-dict substitutions (e.g. if there's a printf 148 # in the templated C++ code) 149 template = self.parser.protectNonSubstPercents(self.template) 150 # CPU-model-specific substitutions are handled later (in GenCode). 151 template = self.parser.protectCpuSymbols(template) 152 153 # Build a dict ('myDict') to use for the template substitution. 154 # Start with the template namespace. Make a copy since we're 155 # going to modify it. 156 myDict = self.parser.templateMap.copy() 157 158 if isinstance(d, InstObjParams): 159 # If we're dealing with an InstObjParams object, we need 160 # to be a little more sophisticated. The instruction-wide 161 # parameters are already formed, but the parameters which 162 # are only function wide still need to be generated. 163 compositeCode = '' 164 165 myDict.update(d.__dict__) 166 # The "operands" and "snippets" attributes of the InstObjParams 167 # objects are for internal use and not substitution. 168 del myDict['operands'] 169 del myDict['snippets'] 170 171 snippetLabels = [l for l in labelRE.findall(template) 172 if d.snippets.has_key(l)] 173 174 snippets = dict([(s, self.parser.mungeSnippet(d.snippets[s])) 175 for s in snippetLabels]) 176 177 myDict.update(snippets) 178 179 compositeCode = ' '.join(map(str, snippets.values())) 180 181 # Add in template itself in case it references any 182 # operands explicitly (like Mem) 183 compositeCode += ' ' + template 184 185 operands = SubOperandList(self.parser, compositeCode, d.operands) 186 187 myDict['op_decl'] = operands.concatAttrStrings('op_decl') 188 if operands.readPC or operands.setPC: 189 myDict['op_decl'] += 'TheISA::PCState __parserAutoPCState;\n' 190 191 # In case there are predicated register reads and write, declare 192 # the variables for register indicies. It is being assumed that 193 # all the operands in the OperandList are also in the 194 # SubOperandList and in the same order. Otherwise, it is 195 # expected that predication would not be used for the operands. 196 if operands.predRead: 197 myDict['op_decl'] += 'uint8_t _sourceIndex = 0;\n' 198 if operands.predWrite: 199 myDict['op_decl'] += 'uint8_t M5_VAR_USED _destIndex = 0;\n' 200 201 is_src = lambda op: op.is_src 202 is_dest = lambda op: op.is_dest 203 204 myDict['op_src_decl'] = \ 205 operands.concatSomeAttrStrings(is_src, 'op_src_decl') 206 myDict['op_dest_decl'] = \ 207 operands.concatSomeAttrStrings(is_dest, 'op_dest_decl') 208 if operands.readPC: 209 myDict['op_src_decl'] += \ 210 'TheISA::PCState __parserAutoPCState;\n' 211 if operands.setPC: 212 myDict['op_dest_decl'] += \ 213 'TheISA::PCState __parserAutoPCState;\n' 214 215 myDict['op_rd'] = operands.concatAttrStrings('op_rd') 216 if operands.readPC: 217 myDict['op_rd'] = '__parserAutoPCState = xc->pcState();\n' + \ 218 myDict['op_rd'] 219 220 # Compose the op_wb string. If we're going to write back the 221 # PC state because we changed some of its elements, we'll need to 222 # do that as early as possible. That allows later uncoordinated 223 # modifications to the PC to layer appropriately. 224 reordered = list(operands.items) 225 reordered.reverse() 226 op_wb_str = '' 227 pcWbStr = 'xc->pcState(__parserAutoPCState);\n' 228 for op_desc in reordered: 229 if op_desc.isPCPart() and op_desc.is_dest: 230 op_wb_str = op_desc.op_wb + pcWbStr + op_wb_str 231 pcWbStr = '' 232 else: 233 op_wb_str = op_desc.op_wb + op_wb_str 234 myDict['op_wb'] = op_wb_str 235 236 elif isinstance(d, dict): 237 # if the argument is a dictionary, we just use it. 238 myDict.update(d) 239 elif hasattr(d, '__dict__'): 240 # if the argument is an object, we use its attribute map. 241 myDict.update(d.__dict__) 242 else: 243 raise TypeError, "Template.subst() arg must be or have dictionary" 244 return template % myDict 245 246 # Convert to string. This handles the case when a template with a 247 # CPU-specific term gets interpolated into another template or into 248 # an output block. 249 def __str__(self): 250 return self.parser.expandCpuSymbolsToString(self.template) 251 252################ 253# Format object. 254# 255# A format object encapsulates an instruction format. It must provide 256# a defineInst() method that generates the code for an instruction 257# definition. 258 259class Format(object): 260 def __init__(self, id, params, code): 261 self.id = id 262 self.params = params 263 label = 'def format ' + id 264 self.user_code = compile(fixPythonIndentation(code), label, 'exec') 265 param_list = string.join(params, ", ") 266 f = '''def defInst(_code, _context, %s): 267 my_locals = vars().copy() 268 exec _code in _context, my_locals 269 return my_locals\n''' % param_list 270 c = compile(f, label + ' wrapper', 'exec') 271 exec c 272 self.func = defInst 273 274 def defineInst(self, parser, name, args, lineno): 275 parser.updateExportContext() 276 context = parser.exportContext.copy() 277 if len(name): 278 Name = name[0].upper() 279 if len(name) > 1: 280 Name += name[1:] 281 context.update({ 'name' : name, 'Name' : Name }) 282 try: 283 vars = self.func(self.user_code, context, *args[0], **args[1]) 284 except Exception, exc: 285 if debug: 286 raise 287 error(lineno, 'error defining "%s": %s.' % (name, exc)) 288 for k in vars.keys(): 289 if k not in ('header_output', 'decoder_output', 290 'exec_output', 'decode_block'): 291 del vars[k] 292 return GenCode(parser, **vars) 293 294# Special null format to catch an implicit-format instruction 295# definition outside of any format block. 296class NoFormat(object): 297 def __init__(self): 298 self.defaultInst = '' 299 300 def defineInst(self, parser, name, args, lineno): 301 error(lineno, 302 'instruction definition "%s" with no active format!' % name) 303 304############### 305# GenCode class 306# 307# The GenCode class encapsulates generated code destined for various 308# output files. The header_output and decoder_output attributes are 309# strings containing code destined for decoder.hh and decoder.cc 310# respectively. The decode_block attribute contains code to be 311# incorporated in the decode function itself (that will also end up in 312# decoder.cc). The exec_output attribute is a dictionary with a key 313# for each CPU model name; the value associated with a particular key 314# is the string of code for that CPU model's exec.cc file. The 315# has_decode_default attribute is used in the decode block to allow 316# explicit default clauses to override default default clauses. 317 318class GenCode(object): 319 # Constructor. At this point we substitute out all CPU-specific 320 # symbols. For the exec output, these go into the per-model 321 # dictionary. For all other output types they get collapsed into 322 # a single string. 323 def __init__(self, parser, 324 header_output = '', decoder_output = '', exec_output = '', 325 decode_block = '', has_decode_default = False): 326 self.parser = parser 327 self.header_output = parser.expandCpuSymbolsToString(header_output) 328 self.decoder_output = parser.expandCpuSymbolsToString(decoder_output) 329 self.exec_output = exec_output 330 self.decode_block = decode_block 331 self.has_decode_default = has_decode_default 332 333 # Write these code chunks out to the filesystem. They will be properly 334 # interwoven by the write_top_level_files(). 335 def emit(self): 336 if self.header_output: 337 self.parser.get_file('header').write(self.header_output) 338 if self.decoder_output: 339 self.parser.get_file('decoder').write(self.decoder_output) 340 if self.exec_output: 341 self.parser.get_file('exec').write(self.exec_output) 342 if self.decode_block: 343 self.parser.get_file('decode_block').write(self.decode_block) 344 345 # Override '+' operator: generate a new GenCode object that 346 # concatenates all the individual strings in the operands. 347 def __add__(self, other): 348 return GenCode(self.parser, 349 self.header_output + other.header_output, 350 self.decoder_output + other.decoder_output, 351 self.exec_output + other.exec_output, 352 self.decode_block + other.decode_block, 353 self.has_decode_default or other.has_decode_default) 354 355 # Prepend a string (typically a comment) to all the strings. 356 def prepend_all(self, pre): 357 self.header_output = pre + self.header_output 358 self.decoder_output = pre + self.decoder_output 359 self.decode_block = pre + self.decode_block 360 self.exec_output = pre + self.exec_output 361 362 # Wrap the decode block in a pair of strings (e.g., 'case foo:' 363 # and 'break;'). Used to build the big nested switch statement. 364 def wrap_decode_block(self, pre, post = ''): 365 self.decode_block = pre + indent(self.decode_block) + post 366 367##################################################################### 368# 369# Bitfield Operator Support 370# 371##################################################################### 372 373bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>') 374 375bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>') 376bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>') 377 378def substBitOps(code): 379 # first convert single-bit selectors to two-index form 380 # i.e., <n> --> <n:n> 381 code = bitOp1ArgRE.sub(r'<\1:\1>', code) 382 # simple case: selector applied to ID (name) 383 # i.e., foo<a:b> --> bits(foo, a, b) 384 code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code) 385 # if selector is applied to expression (ending in ')'), 386 # we need to search backward for matching '(' 387 match = bitOpExprRE.search(code) 388 while match: 389 exprEnd = match.start() 390 here = exprEnd - 1 391 nestLevel = 1 392 while nestLevel > 0: 393 if code[here] == '(': 394 nestLevel -= 1 395 elif code[here] == ')': 396 nestLevel += 1 397 here -= 1 398 if here < 0: 399 sys.exit("Didn't find '('!") 400 exprStart = here+1 401 newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1], 402 match.group(1), match.group(2)) 403 code = code[:exprStart] + newExpr + code[match.end():] 404 match = bitOpExprRE.search(code) 405 return code 406 407 408##################################################################### 409# 410# Code Parser 411# 412# The remaining code is the support for automatically extracting 413# instruction characteristics from pseudocode. 414# 415##################################################################### 416 417# Force the argument to be a list. Useful for flags, where a caller 418# can specify a singleton flag or a list of flags. Also usful for 419# converting tuples to lists so they can be modified. 420def makeList(arg): 421 if isinstance(arg, list): 422 return arg 423 elif isinstance(arg, tuple): 424 return list(arg) 425 elif not arg: 426 return [] 427 else: 428 return [ arg ] 429 430class Operand(object): 431 '''Base class for operand descriptors. An instance of this class 432 (or actually a class derived from this one) represents a specific 433 operand for a code block (e.g, "Rc.sq" as a dest). Intermediate 434 derived classes encapsulates the traits of a particular operand 435 type (e.g., "32-bit integer register").''' 436 437 def buildReadCode(self, func = None): 438 subst_dict = {"name": self.base_name, 439 "func": func, 440 "reg_idx": self.reg_spec, 441 "ctype": self.ctype} 442 if hasattr(self, 'src_reg_idx'): 443 subst_dict['op_idx'] = self.src_reg_idx 444 code = self.read_code % subst_dict 445 return '%s = %s;\n' % (self.base_name, code) 446 447 def buildWriteCode(self, func = None): 448 subst_dict = {"name": self.base_name, 449 "func": func, 450 "reg_idx": self.reg_spec, 451 "ctype": self.ctype, 452 "final_val": self.base_name} 453 if hasattr(self, 'dest_reg_idx'): 454 subst_dict['op_idx'] = self.dest_reg_idx 455 code = self.write_code % subst_dict 456 return ''' 457 { 458 %s final_val = %s; 459 %s; 460 if (traceData) { traceData->setData(final_val); } 461 }''' % (self.dflt_ctype, self.base_name, code) 462 463 def __init__(self, parser, full_name, ext, is_src, is_dest): 464 self.full_name = full_name 465 self.ext = ext 466 self.is_src = is_src 467 self.is_dest = is_dest 468 # The 'effective extension' (eff_ext) is either the actual 469 # extension, if one was explicitly provided, or the default. 470 if ext: 471 self.eff_ext = ext 472 elif hasattr(self, 'dflt_ext'): 473 self.eff_ext = self.dflt_ext 474 475 if hasattr(self, 'eff_ext'): 476 self.ctype = parser.operandTypeMap[self.eff_ext] 477 478 # Finalize additional fields (primarily code fields). This step 479 # is done separately since some of these fields may depend on the 480 # register index enumeration that hasn't been performed yet at the 481 # time of __init__(). The register index enumeration is affected 482 # by predicated register reads/writes. Hence, we forward the flags 483 # that indicate whether or not predication is in use. 484 def finalize(self, predRead, predWrite): 485 self.flags = self.getFlags() 486 self.constructor = self.makeConstructor(predRead, predWrite) 487 self.op_decl = self.makeDecl() 488 489 if self.is_src: 490 self.op_rd = self.makeRead(predRead) 491 self.op_src_decl = self.makeDecl() 492 else: 493 self.op_rd = '' 494 self.op_src_decl = '' 495 496 if self.is_dest: 497 self.op_wb = self.makeWrite(predWrite) 498 self.op_dest_decl = self.makeDecl() 499 else: 500 self.op_wb = '' 501 self.op_dest_decl = '' 502 503 def isMem(self): 504 return 0 505 506 def isReg(self): 507 return 0 508 509 def isFloatReg(self): 510 return 0 511 512 def isIntReg(self): 513 return 0 514 515 def isCCReg(self): 516 return 0 517 518 def isControlReg(self): 519 return 0 520 521 def isPCState(self): 522 return 0 523 524 def isPCPart(self): 525 return self.isPCState() and self.reg_spec 526 527 def hasReadPred(self): 528 return self.read_predicate != None 529 530 def hasWritePred(self): 531 return self.write_predicate != None 532 533 def getFlags(self): 534 # note the empty slice '[:]' gives us a copy of self.flags[0] 535 # instead of a reference to it 536 my_flags = self.flags[0][:] 537 if self.is_src: 538 my_flags += self.flags[1] 539 if self.is_dest: 540 my_flags += self.flags[2] 541 return my_flags 542 543 def makeDecl(self): 544 # Note that initializations in the declarations are solely 545 # to avoid 'uninitialized variable' errors from the compiler. 546 return self.ctype + ' ' + self.base_name + ' = 0;\n'; 547 548class IntRegOperand(Operand): 549 def isReg(self): 550 return 1 551 552 def isIntReg(self): 553 return 1 554 555 def makeConstructor(self, predRead, predWrite): 556 c_src = '' 557 c_dest = '' 558 559 if self.is_src: 560 c_src = '\n\t_srcRegIdx[_numSrcRegs++] = %s;' % (self.reg_spec) 561 if self.hasReadPred(): 562 c_src = '\n\tif (%s) {%s\n\t}' % \ 563 (self.read_predicate, c_src) 564 565 if self.is_dest: 566 c_dest = '\n\t_destRegIdx[_numDestRegs++] = %s;' % \ 567 (self.reg_spec) 568 c_dest += '\n\t_numIntDestRegs++;' 569 if self.hasWritePred(): 570 c_dest = '\n\tif (%s) {%s\n\t}' % \ 571 (self.write_predicate, c_dest) 572 573 return c_src + c_dest 574 575 def makeRead(self, predRead): 576 if (self.ctype == 'float' or self.ctype == 'double'): 577 error('Attempt to read integer register as FP') 578 if self.read_code != None: 579 return self.buildReadCode('readIntRegOperand') 580 581 int_reg_val = '' 582 if predRead: 583 int_reg_val = 'xc->readIntRegOperand(this, _sourceIndex++)' 584 if self.hasReadPred(): 585 int_reg_val = '(%s) ? %s : 0' % \ 586 (self.read_predicate, int_reg_val) 587 else: 588 int_reg_val = 'xc->readIntRegOperand(this, %d)' % self.src_reg_idx 589 590 return '%s = %s;\n' % (self.base_name, int_reg_val) 591 592 def makeWrite(self, predWrite): 593 if (self.ctype == 'float' or self.ctype == 'double'): 594 error('Attempt to write integer register as FP') 595 if self.write_code != None: 596 return self.buildWriteCode('setIntRegOperand') 597 598 if predWrite: 599 wp = 'true' 600 if self.hasWritePred(): 601 wp = self.write_predicate 602 603 wcond = 'if (%s)' % (wp) 604 windex = '_destIndex++' 605 else: 606 wcond = '' 607 windex = '%d' % self.dest_reg_idx 608 609 wb = ''' 610 %s 611 { 612 %s final_val = %s; 613 xc->setIntRegOperand(this, %s, final_val);\n 614 if (traceData) { traceData->setData(final_val); } 615 }''' % (wcond, self.ctype, self.base_name, windex) 616 617 return wb 618 619class FloatRegOperand(Operand): 620 def isReg(self): 621 return 1 622 623 def isFloatReg(self): 624 return 1 625 626 def makeConstructor(self, predRead, predWrite): 627 c_src = '' 628 c_dest = '' 629 630 if self.is_src: 631 c_src = '\n\t_srcRegIdx[_numSrcRegs++] = %s + FP_Reg_Base;' % \ 632 (self.reg_spec) 633 634 if self.is_dest: 635 c_dest = \ 636 '\n\t_destRegIdx[_numDestRegs++] = %s + FP_Reg_Base;' % \ 637 (self.reg_spec) 638 c_dest += '\n\t_numFPDestRegs++;' 639 640 return c_src + c_dest 641 642 def makeRead(self, predRead): 643 bit_select = 0 644 if (self.ctype == 'float' or self.ctype == 'double'): 645 func = 'readFloatRegOperand' 646 else: 647 func = 'readFloatRegOperandBits' 648 if self.read_code != None: 649 return self.buildReadCode(func) 650 651 if predRead: 652 rindex = '_sourceIndex++' 653 else: 654 rindex = '%d' % self.src_reg_idx 655 656 return '%s = xc->%s(this, %s);\n' % \ 657 (self.base_name, func, rindex) 658 659 def makeWrite(self, predWrite): 660 if (self.ctype == 'float' or self.ctype == 'double'): 661 func = 'setFloatRegOperand' 662 else: 663 func = 'setFloatRegOperandBits' 664 if self.write_code != None: 665 return self.buildWriteCode(func) 666 667 if predWrite: 668 wp = '_destIndex++' 669 else: 670 wp = '%d' % self.dest_reg_idx 671 wp = 'xc->%s(this, %s, final_val);' % (func, wp) 672 673 wb = ''' 674 { 675 %s final_val = %s; 676 %s\n 677 if (traceData) { traceData->setData(final_val); } 678 }''' % (self.ctype, self.base_name, wp) 679 return wb 680 681class CCRegOperand(Operand): 682 def isReg(self): 683 return 1 684 685 def isCCReg(self): 686 return 1 687 688 def makeConstructor(self, predRead, predWrite): 689 c_src = '' 690 c_dest = '' 691 692 if self.is_src: 693 c_src = '\n\t_srcRegIdx[_numSrcRegs++] = %s + CC_Reg_Base;' % \ 694 (self.reg_spec) 695 if self.hasReadPred(): 696 c_src = '\n\tif (%s) {%s\n\t}' % \ 697 (self.read_predicate, c_src) 698 699 if self.is_dest: 700 c_dest = \ 701 '\n\t_destRegIdx[_numDestRegs++] = %s + CC_Reg_Base;' % \ 702 (self.reg_spec) 703 c_dest += '\n\t_numCCDestRegs++;' 704 if self.hasWritePred(): 705 c_dest = '\n\tif (%s) {%s\n\t}' % \ 706 (self.write_predicate, c_dest) 707 708 return c_src + c_dest 709 710 def makeRead(self, predRead): 711 if (self.ctype == 'float' or self.ctype == 'double'): 712 error('Attempt to read condition-code register as FP') 713 if self.read_code != None: 714 return self.buildReadCode('readCCRegOperand') 715 716 int_reg_val = '' 717 if predRead: 718 int_reg_val = 'xc->readCCRegOperand(this, _sourceIndex++)' 719 if self.hasReadPred(): 720 int_reg_val = '(%s) ? %s : 0' % \ 721 (self.read_predicate, int_reg_val) 722 else: 723 int_reg_val = 'xc->readCCRegOperand(this, %d)' % self.src_reg_idx 724 725 return '%s = %s;\n' % (self.base_name, int_reg_val) 726 727 def makeWrite(self, predWrite): 728 if (self.ctype == 'float' or self.ctype == 'double'): 729 error('Attempt to write condition-code register as FP') 730 if self.write_code != None: 731 return self.buildWriteCode('setCCRegOperand') 732 733 if predWrite: 734 wp = 'true' 735 if self.hasWritePred(): 736 wp = self.write_predicate 737 738 wcond = 'if (%s)' % (wp) 739 windex = '_destIndex++' 740 else: 741 wcond = '' 742 windex = '%d' % self.dest_reg_idx 743 744 wb = ''' 745 %s 746 { 747 %s final_val = %s; 748 xc->setCCRegOperand(this, %s, final_val);\n 749 if (traceData) { traceData->setData(final_val); } 750 }''' % (wcond, self.ctype, self.base_name, windex) 751 752 return wb 753 754class ControlRegOperand(Operand): 755 def isReg(self): 756 return 1 757 758 def isControlReg(self): 759 return 1 760 761 def makeConstructor(self, predRead, predWrite): 762 c_src = '' 763 c_dest = '' 764 765 if self.is_src: 766 c_src = \ 767 '\n\t_srcRegIdx[_numSrcRegs++] = %s + Misc_Reg_Base;' % \ 768 (self.reg_spec) 769 770 if self.is_dest: 771 c_dest = \ 772 '\n\t_destRegIdx[_numDestRegs++] = %s + Misc_Reg_Base;' % \ 773 (self.reg_spec) 774 775 return c_src + c_dest 776 777 def makeRead(self, predRead): 778 bit_select = 0 779 if (self.ctype == 'float' or self.ctype == 'double'): 780 error('Attempt to read control register as FP') 781 if self.read_code != None: 782 return self.buildReadCode('readMiscRegOperand') 783 784 if predRead: 785 rindex = '_sourceIndex++' 786 else: 787 rindex = '%d' % self.src_reg_idx 788 789 return '%s = xc->readMiscRegOperand(this, %s);\n' % \ 790 (self.base_name, rindex) 791 792 def makeWrite(self, predWrite): 793 if (self.ctype == 'float' or self.ctype == 'double'): 794 error('Attempt to write control register as FP') 795 if self.write_code != None: 796 return self.buildWriteCode('setMiscRegOperand') 797 798 if predWrite: 799 windex = '_destIndex++' 800 else: 801 windex = '%d' % self.dest_reg_idx 802 803 wb = 'xc->setMiscRegOperand(this, %s, %s);\n' % \ 804 (windex, self.base_name) 805 wb += 'if (traceData) { traceData->setData(%s); }' % \ 806 self.base_name 807 808 return wb 809 810class MemOperand(Operand): 811 def isMem(self): 812 return 1 813 814 def makeConstructor(self, predRead, predWrite): 815 return '' 816 817 def makeDecl(self): 818 # Note that initializations in the declarations are solely 819 # to avoid 'uninitialized variable' errors from the compiler. 820 # Declare memory data variable. 821 return '%s %s = 0;\n' % (self.ctype, self.base_name) 822 823 def makeRead(self, predRead): 824 if self.read_code != None: 825 return self.buildReadCode() 826 return '' 827 828 def makeWrite(self, predWrite): 829 if self.write_code != None: 830 return self.buildWriteCode() 831 return '' 832 833class PCStateOperand(Operand): 834 def makeConstructor(self, predRead, predWrite): 835 return '' 836 837 def makeRead(self, predRead): 838 if self.reg_spec: 839 # A component of the PC state. 840 return '%s = __parserAutoPCState.%s();\n' % \ 841 (self.base_name, self.reg_spec) 842 else: 843 # The whole PC state itself. 844 return '%s = xc->pcState();\n' % self.base_name 845 846 def makeWrite(self, predWrite): 847 if self.reg_spec: 848 # A component of the PC state. 849 return '__parserAutoPCState.%s(%s);\n' % \ 850 (self.reg_spec, self.base_name) 851 else: 852 # The whole PC state itself. 853 return 'xc->pcState(%s);\n' % self.base_name 854 855 def makeDecl(self): 856 ctype = 'TheISA::PCState' 857 if self.isPCPart(): 858 ctype = self.ctype 859 return "%s %s;\n" % (ctype, self.base_name) 860 861 def isPCState(self): 862 return 1 863 864class OperandList(object): 865 '''Find all the operands in the given code block. Returns an operand 866 descriptor list (instance of class OperandList).''' 867 def __init__(self, parser, code): 868 self.items = [] 869 self.bases = {} 870 # delete strings and comments so we don't match on operands inside 871 for regEx in (stringRE, commentRE): 872 code = regEx.sub('', code) 873 # search for operands 874 next_pos = 0 875 while 1: 876 match = parser.operandsRE.search(code, next_pos) 877 if not match: 878 # no more matches: we're done 879 break 880 op = match.groups() 881 # regexp groups are operand full name, base, and extension 882 (op_full, op_base, op_ext) = op 883 # if the token following the operand is an assignment, this is 884 # a destination (LHS), else it's a source (RHS) 885 is_dest = (assignRE.match(code, match.end()) != None) 886 is_src = not is_dest 887 # see if we've already seen this one 888 op_desc = self.find_base(op_base) 889 if op_desc: 890 if op_desc.ext != op_ext: 891 error('Inconsistent extensions for operand %s' % \ 892 op_base) 893 op_desc.is_src = op_desc.is_src or is_src 894 op_desc.is_dest = op_desc.is_dest or is_dest 895 else: 896 # new operand: create new descriptor 897 op_desc = parser.operandNameMap[op_base](parser, 898 op_full, op_ext, is_src, is_dest) 899 self.append(op_desc) 900 # start next search after end of current match 901 next_pos = match.end() 902 self.sort() 903 # enumerate source & dest register operands... used in building 904 # constructor later 905 self.numSrcRegs = 0 906 self.numDestRegs = 0 907 self.numFPDestRegs = 0 908 self.numIntDestRegs = 0 909 self.numCCDestRegs = 0 910 self.numMiscDestRegs = 0 911 self.memOperand = None 912 913 # Flags to keep track if one or more operands are to be read/written 914 # conditionally. 915 self.predRead = False 916 self.predWrite = False 917 918 for op_desc in self.items: 919 if op_desc.isReg(): 920 if op_desc.is_src: 921 op_desc.src_reg_idx = self.numSrcRegs 922 self.numSrcRegs += 1 923 if op_desc.is_dest: 924 op_desc.dest_reg_idx = self.numDestRegs 925 self.numDestRegs += 1 926 if op_desc.isFloatReg(): 927 self.numFPDestRegs += 1 928 elif op_desc.isIntReg(): 929 self.numIntDestRegs += 1 930 elif op_desc.isCCReg(): 931 self.numCCDestRegs += 1 932 elif op_desc.isControlReg(): 933 self.numMiscDestRegs += 1 934 elif op_desc.isMem(): 935 if self.memOperand: 936 error("Code block has more than one memory operand.") 937 self.memOperand = op_desc 938 939 # Check if this operand has read/write predication. If true, then 940 # the microop will dynamically index source/dest registers. 941 self.predRead = self.predRead or op_desc.hasReadPred() 942 self.predWrite = self.predWrite or op_desc.hasWritePred() 943 944 if parser.maxInstSrcRegs < self.numSrcRegs: 945 parser.maxInstSrcRegs = self.numSrcRegs 946 if parser.maxInstDestRegs < self.numDestRegs: 947 parser.maxInstDestRegs = self.numDestRegs 948 if parser.maxMiscDestRegs < self.numMiscDestRegs: 949 parser.maxMiscDestRegs = self.numMiscDestRegs 950 951 # now make a final pass to finalize op_desc fields that may depend 952 # on the register enumeration 953 for op_desc in self.items: 954 op_desc.finalize(self.predRead, self.predWrite) 955 956 def __len__(self): 957 return len(self.items) 958 959 def __getitem__(self, index): 960 return self.items[index] 961 962 def append(self, op_desc): 963 self.items.append(op_desc) 964 self.bases[op_desc.base_name] = op_desc 965 966 def find_base(self, base_name): 967 # like self.bases[base_name], but returns None if not found 968 # (rather than raising exception) 969 return self.bases.get(base_name) 970 971 # internal helper function for concat[Some]Attr{Strings|Lists} 972 def __internalConcatAttrs(self, attr_name, filter, result): 973 for op_desc in self.items: 974 if filter(op_desc): 975 result += getattr(op_desc, attr_name) 976 return result 977 978 # return a single string that is the concatenation of the (string) 979 # values of the specified attribute for all operands 980 def concatAttrStrings(self, attr_name): 981 return self.__internalConcatAttrs(attr_name, lambda x: 1, '') 982 983 # like concatAttrStrings, but only include the values for the operands 984 # for which the provided filter function returns true 985 def concatSomeAttrStrings(self, filter, attr_name): 986 return self.__internalConcatAttrs(attr_name, filter, '') 987 988 # return a single list that is the concatenation of the (list) 989 # values of the specified attribute for all operands 990 def concatAttrLists(self, attr_name): 991 return self.__internalConcatAttrs(attr_name, lambda x: 1, []) 992 993 # like concatAttrLists, but only include the values for the operands 994 # for which the provided filter function returns true 995 def concatSomeAttrLists(self, filter, attr_name): 996 return self.__internalConcatAttrs(attr_name, filter, []) 997 998 def sort(self): 999 self.items.sort(lambda a, b: a.sort_pri - b.sort_pri) 1000 1001class SubOperandList(OperandList): 1002 '''Find all the operands in the given code block. Returns an operand 1003 descriptor list (instance of class OperandList).''' 1004 def __init__(self, parser, code, master_list): 1005 self.items = [] 1006 self.bases = {} 1007 # delete strings and comments so we don't match on operands inside 1008 for regEx in (stringRE, commentRE): 1009 code = regEx.sub('', code) 1010 # search for operands 1011 next_pos = 0 1012 while 1: 1013 match = parser.operandsRE.search(code, next_pos) 1014 if not match: 1015 # no more matches: we're done 1016 break 1017 op = match.groups() 1018 # regexp groups are operand full name, base, and extension 1019 (op_full, op_base, op_ext) = op 1020 # find this op in the master list 1021 op_desc = master_list.find_base(op_base) 1022 if not op_desc: 1023 error('Found operand %s which is not in the master list!' \ 1024 ' This is an internal error' % op_base) 1025 else: 1026 # See if we've already found this operand 1027 op_desc = self.find_base(op_base) 1028 if not op_desc: 1029 # if not, add a reference to it to this sub list 1030 self.append(master_list.bases[op_base]) 1031 1032 # start next search after end of current match 1033 next_pos = match.end() 1034 self.sort() 1035 self.memOperand = None 1036 # Whether the whole PC needs to be read so parts of it can be accessed 1037 self.readPC = False 1038 # Whether the whole PC needs to be written after parts of it were 1039 # changed 1040 self.setPC = False 1041 # Whether this instruction manipulates the whole PC or parts of it. 1042 # Mixing the two is a bad idea and flagged as an error. 1043 self.pcPart = None 1044 1045 # Flags to keep track if one or more operands are to be read/written 1046 # conditionally. 1047 self.predRead = False 1048 self.predWrite = False 1049 1050 for op_desc in self.items: 1051 if op_desc.isPCPart(): 1052 self.readPC = True 1053 if op_desc.is_dest: 1054 self.setPC = True 1055 1056 if op_desc.isPCState(): 1057 if self.pcPart is not None: 1058 if self.pcPart and not op_desc.isPCPart() or \ 1059 not self.pcPart and op_desc.isPCPart(): 1060 error("Mixed whole and partial PC state operands.") 1061 self.pcPart = op_desc.isPCPart() 1062 1063 if op_desc.isMem(): 1064 if self.memOperand: 1065 error("Code block has more than one memory operand.") 1066 self.memOperand = op_desc 1067 1068 # Check if this operand has read/write predication. If true, then 1069 # the microop will dynamically index source/dest registers. 1070 self.predRead = self.predRead or op_desc.hasReadPred() 1071 self.predWrite = self.predWrite or op_desc.hasWritePred() 1072 1073# Regular expression object to match C++ strings 1074stringRE = re.compile(r'"([^"\\]|\\.)*"') 1075 1076# Regular expression object to match C++ comments 1077# (used in findOperands()) 1078commentRE = re.compile(r'(^)?[^\S\n]*/(?:\*(.*?)\*/[^\S\n]*|/[^\n]*)($)?', 1079 re.DOTALL | re.MULTILINE) 1080 1081# Regular expression object to match assignment statements 1082# (used in findOperands()) 1083assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE) 1084 1085def makeFlagConstructor(flag_list): 1086 if len(flag_list) == 0: 1087 return '' 1088 # filter out repeated flags 1089 flag_list.sort() 1090 i = 1 1091 while i < len(flag_list): 1092 if flag_list[i] == flag_list[i-1]: 1093 del flag_list[i] 1094 else: 1095 i += 1 1096 pre = '\n\tflags[' 1097 post = '] = true;' 1098 code = pre + string.join(flag_list, post + pre) + post 1099 return code 1100 1101# Assume all instruction flags are of the form 'IsFoo' 1102instFlagRE = re.compile(r'Is.*') 1103 1104# OpClass constants end in 'Op' except No_OpClass 1105opClassRE = re.compile(r'.*Op|No_OpClass') 1106 1107class InstObjParams(object): 1108 def __init__(self, parser, mnem, class_name, base_class = '', 1109 snippets = {}, opt_args = []): 1110 self.mnemonic = mnem 1111 self.class_name = class_name 1112 self.base_class = base_class 1113 if not isinstance(snippets, dict): 1114 snippets = {'code' : snippets} 1115 compositeCode = ' '.join(map(str, snippets.values())) 1116 self.snippets = snippets 1117 1118 self.operands = OperandList(parser, compositeCode) 1119 1120 # The header of the constructor declares the variables to be used 1121 # in the body of the constructor. 1122 header = '' 1123 header += '\n\t_numSrcRegs = 0;' 1124 header += '\n\t_numDestRegs = 0;' 1125 header += '\n\t_numFPDestRegs = 0;' 1126 header += '\n\t_numIntDestRegs = 0;' 1127 header += '\n\t_numCCDestRegs = 0;' 1128 1129 self.constructor = header + \ 1130 self.operands.concatAttrStrings('constructor') 1131 1132 self.flags = self.operands.concatAttrLists('flags') 1133 1134 self.op_class = None 1135 1136 # Optional arguments are assumed to be either StaticInst flags 1137 # or an OpClass value. To avoid having to import a complete 1138 # list of these values to match against, we do it ad-hoc 1139 # with regexps. 1140 for oa in opt_args: 1141 if instFlagRE.match(oa): 1142 self.flags.append(oa) 1143 elif opClassRE.match(oa): 1144 self.op_class = oa 1145 else: 1146 error('InstObjParams: optional arg "%s" not recognized ' 1147 'as StaticInst::Flag or OpClass.' % oa) 1148 1149 # Make a basic guess on the operand class if not set. 1150 # These are good enough for most cases. 1151 if not self.op_class: 1152 if 'IsStore' in self.flags: 1153 self.op_class = 'MemWriteOp' 1154 elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags: 1155 self.op_class = 'MemReadOp' 1156 elif 'IsFloating' in self.flags: 1157 self.op_class = 'FloatAddOp' 1158 else: 1159 self.op_class = 'IntAluOp' 1160 1161 # add flag initialization to contructor here to include 1162 # any flags added via opt_args 1163 self.constructor += makeFlagConstructor(self.flags) 1164 1165 # if 'IsFloating' is set, add call to the FP enable check 1166 # function (which should be provided by isa_desc via a declare) 1167 if 'IsFloating' in self.flags: 1168 self.fp_enable_check = 'fault = checkFpEnableFault(xc);' 1169 else: 1170 self.fp_enable_check = '' 1171 1172############## 1173# Stack: a simple stack object. Used for both formats (formatStack) 1174# and default cases (defaultStack). Simply wraps a list to give more 1175# stack-like syntax and enable initialization with an argument list 1176# (as opposed to an argument that's a list). 1177 1178class Stack(list): 1179 def __init__(self, *items): 1180 list.__init__(self, items) 1181 1182 def push(self, item): 1183 self.append(item); 1184 1185 def top(self): 1186 return self[-1] 1187 1188####################### 1189# 1190# ISA Parser 1191# parses ISA DSL and emits C++ headers and source 1192# 1193 1194class ISAParser(Grammar): 1195 class CpuModel(object): 1196 def __init__(self, name, filename, includes, strings): 1197 self.name = name 1198 self.filename = filename 1199 self.includes = includes 1200 self.strings = strings 1201 1202 def __init__(self, output_dir): 1203 super(ISAParser, self).__init__() 1204 self.output_dir = output_dir 1205 1206 self.filename = None # for output file watermarking/scaremongering 1207 1208 self.cpuModels = [ 1209 ISAParser.CpuModel('ExecContext', 1210 'generic_cpu_exec.cc', 1211 '#include "cpu/exec_context.hh"', 1212 { "CPU_exec_context" : "ExecContext" }), 1213 ] 1214 1215 # variable to hold templates 1216 self.templateMap = {} 1217 1218 # This dictionary maps format name strings to Format objects. 1219 self.formatMap = {} 1220 1221 # Track open files and, if applicable, how many chunks it has been 1222 # split into so far. 1223 self.files = {} 1224 self.splits = {} 1225 1226 # isa_name / namespace identifier from namespace declaration. 1227 # before the namespace declaration, None. 1228 self.isa_name = None 1229 self.namespace = None 1230 1231 # The format stack. 1232 self.formatStack = Stack(NoFormat()) 1233 1234 # The default case stack. 1235 self.defaultStack = Stack(None) 1236 1237 # Stack that tracks current file and line number. Each 1238 # element is a tuple (filename, lineno) that records the 1239 # *current* filename and the line number in the *previous* 1240 # file where it was included. 1241 self.fileNameStack = Stack() 1242 1243 symbols = ('makeList', 're', 'string') 1244 self.exportContext = dict([(s, eval(s)) for s in symbols]) 1245 1246 self.maxInstSrcRegs = 0 1247 self.maxInstDestRegs = 0 1248 self.maxMiscDestRegs = 0 1249 1250 def __getitem__(self, i): # Allow object (self) to be 1251 return getattr(self, i) # passed to %-substitutions 1252 1253 # Change the file suffix of a base filename: 1254 # (e.g.) decoder.cc -> decoder-g.cc.inc for 'global' outputs 1255 def suffixize(self, s, sec): 1256 extn = re.compile('(\.[^\.]+)$') # isolate extension 1257 if self.namespace: 1258 return extn.sub(r'-ns\1.inc', s) # insert some text on either side 1259 else: 1260 return extn.sub(r'-g\1.inc', s) 1261 1262 # Get the file object for emitting code into the specified section 1263 # (header, decoder, exec, decode_block). 1264 def get_file(self, section): 1265 if section == 'decode_block': 1266 filename = 'decode-method.cc.inc' 1267 else: 1268 if section == 'header': 1269 file = 'decoder.hh' 1270 else: 1271 file = '%s.cc' % section 1272 filename = self.suffixize(file, section) 1273 try: 1274 return self.files[filename] 1275 except KeyError: pass 1276 1277 f = self.open(filename) 1278 self.files[filename] = f 1279 1280 # The splittable files are the ones with many independent 1281 # per-instruction functions - the decoder's instruction constructors 1282 # and the instruction execution (execute()) methods. These both have 1283 # the suffix -ns.cc.inc, meaning they are within the namespace part 1284 # of the ISA, contain object-emitting C++ source, and are included 1285 # into other top-level files. These are the files that need special 1286 # #define's to allow parts of them to be compiled separately. Rather 1287 # than splitting the emissions into separate files, the monolithic 1288 # output of the ISA parser is maintained, but the value (or lack 1289 # thereof) of the __SPLIT definition during C preprocessing will 1290 # select the different chunks. If no 'split' directives are used, 1291 # the cpp emissions have no effect. 1292 if re.search('-ns.cc.inc$', filename): 1293 print >>f, '#if !defined(__SPLIT) || (__SPLIT == 1)' 1294 self.splits[f] = 1 1295 # ensure requisite #include's 1296 elif filename in ['decoder-g.cc.inc', 'exec-g.cc.inc']: 1297 print >>f, '#include "decoder.hh"' 1298 elif filename == 'decoder-g.hh.inc': 1299 print >>f, '#include "base/bitfield.hh"' 1300 1301 return f 1302 1303 # Weave together the parts of the different output sections by 1304 # #include'ing them into some very short top-level .cc/.hh files. 1305 # These small files make it much clearer how this tool works, since 1306 # you directly see the chunks emitted as files that are #include'd. 1307 def write_top_level_files(self): 1308 dep = self.open('inc.d', bare=True) 1309 1310 # decoder header - everything depends on this 1311 file = 'decoder.hh' 1312 with self.open(file) as f: 1313 inc = [] 1314 1315 fn = 'decoder-g.hh.inc' 1316 assert(fn in self.files) 1317 f.write('#include "%s"\n' % fn) 1318 inc.append(fn) 1319 1320 fn = 'decoder-ns.hh.inc' 1321 assert(fn in self.files) 1322 f.write('namespace %s {\n#include "%s"\n}\n' 1323 % (self.namespace, fn)) 1324 inc.append(fn) 1325 1326 print >>dep, file+':', ' '.join(inc) 1327 1328 # decoder method - cannot be split 1329 file = 'decoder.cc' 1330 with self.open(file) as f: 1331 inc = [] 1332 1333 fn = 'decoder-g.cc.inc' 1334 assert(fn in self.files) 1335 f.write('#include "%s"\n' % fn) 1336 inc.append(fn) 1337 1338 fn = 'decode-method.cc.inc' 1339 # is guaranteed to have been written for parse to complete 1340 f.write('#include "%s"\n' % fn) 1341 inc.append(fn) 1342 1343 inc.append("decoder.hh") 1344 print >>dep, file+':', ' '.join(inc) 1345 1346 extn = re.compile('(\.[^\.]+)$') 1347 1348 # instruction constructors 1349 splits = self.splits[self.get_file('decoder')] 1350 file_ = 'inst-constrs.cc' 1351 for i in range(1, splits+1): 1352 if splits > 1: 1353 file = extn.sub(r'-%d\1' % i, file_) 1354 else: 1355 file = file_ 1356 with self.open(file) as f: 1357 inc = [] 1358 1359 fn = 'decoder-g.cc.inc' 1360 assert(fn in self.files) 1361 f.write('#include "%s"\n' % fn) 1362 inc.append(fn) 1363 1364 fn = 'decoder-ns.cc.inc' 1365 assert(fn in self.files) 1366 print >>f, 'namespace %s {' % self.namespace 1367 if splits > 1: 1368 print >>f, '#define __SPLIT %u' % i 1369 print >>f, '#include "%s"' % fn 1370 print >>f, '}' 1371 inc.append(fn) 1372 1373 inc.append("decoder.hh") 1374 print >>dep, file+':', ' '.join(inc) 1375 1376 # instruction execution per-CPU model 1377 splits = self.splits[self.get_file('exec')] 1378 for cpu in self.cpuModels: 1379 for i in range(1, splits+1): 1380 if splits > 1: 1381 file = extn.sub(r'_%d\1' % i, cpu.filename) 1382 else: 1383 file = cpu.filename 1384 with self.open(file) as f: 1385 inc = [] 1386 1387 fn = 'exec-g.cc.inc' 1388 assert(fn in self.files) 1389 f.write('#include "%s"\n' % fn) 1390 inc.append(fn) 1391 1392 f.write(cpu.includes+"\n") 1393 1394 fn = 'exec-ns.cc.inc' 1395 assert(fn in self.files) 1396 print >>f, 'namespace %s {' % self.namespace 1397 print >>f, '#define CPU_EXEC_CONTEXT %s' \ 1398 % cpu.strings['CPU_exec_context'] 1399 if splits > 1: 1400 print >>f, '#define __SPLIT %u' % i 1401 print >>f, '#include "%s"' % fn 1402 print >>f, '}' 1403 inc.append(fn) 1404 1405 inc.append("decoder.hh") 1406 print >>dep, file+':', ' '.join(inc) 1407 1408 # max_inst_regs.hh 1409 self.update('max_inst_regs.hh', 1410 '''namespace %(namespace)s { 1411 const int MaxInstSrcRegs = %(maxInstSrcRegs)d; 1412 const int MaxInstDestRegs = %(maxInstDestRegs)d; 1413 const int MaxMiscDestRegs = %(maxMiscDestRegs)d;\n}\n''' % self) 1414 print >>dep, 'max_inst_regs.hh:' 1415 1416 dep.close() 1417 1418 1419 scaremonger_template ='''// DO NOT EDIT 1420// This file was automatically generated from an ISA description: 1421// %(filename)s 1422 1423'''; 1424 1425 ##################################################################### 1426 # 1427 # Lexer 1428 # 1429 # The PLY lexer module takes two things as input: 1430 # - A list of token names (the string list 'tokens') 1431 # - A regular expression describing a match for each token. The 1432 # regexp for token FOO can be provided in two ways: 1433 # - as a string variable named t_FOO 1434 # - as the doc string for a function named t_FOO. In this case, 1435 # the function is also executed, allowing an action to be 1436 # associated with each token match. 1437 # 1438 ##################################################################### 1439 1440 # Reserved words. These are listed separately as they are matched 1441 # using the same regexp as generic IDs, but distinguished in the 1442 # t_ID() function. The PLY documentation suggests this approach. 1443 reserved = ( 1444 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT', 1445 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS', 1446 'OUTPUT', 'SIGNED', 'SPLIT', 'TEMPLATE' 1447 ) 1448 1449 # List of tokens. The lex module requires this. 1450 tokens = reserved + ( 1451 # identifier 1452 'ID', 1453 1454 # integer literal 1455 'INTLIT', 1456 1457 # string literal 1458 'STRLIT', 1459 1460 # code literal 1461 'CODELIT', 1462 1463 # ( ) [ ] { } < > , ; . : :: * 1464 'LPAREN', 'RPAREN', 1465 'LBRACKET', 'RBRACKET', 1466 'LBRACE', 'RBRACE', 1467 'LESS', 'GREATER', 'EQUALS', 1468 'COMMA', 'SEMI', 'DOT', 'COLON', 'DBLCOLON', 1469 'ASTERISK', 1470 1471 # C preprocessor directives 1472 'CPPDIRECTIVE' 1473 1474 # The following are matched but never returned. commented out to 1475 # suppress PLY warning 1476 # newfile directive 1477 # 'NEWFILE', 1478 1479 # endfile directive 1480 # 'ENDFILE' 1481 ) 1482 1483 # Regular expressions for token matching 1484 t_LPAREN = r'\(' 1485 t_RPAREN = r'\)' 1486 t_LBRACKET = r'\[' 1487 t_RBRACKET = r'\]' 1488 t_LBRACE = r'\{' 1489 t_RBRACE = r'\}' 1490 t_LESS = r'\<' 1491 t_GREATER = r'\>' 1492 t_EQUALS = r'=' 1493 t_COMMA = r',' 1494 t_SEMI = r';' 1495 t_DOT = r'\.' 1496 t_COLON = r':' 1497 t_DBLCOLON = r'::' 1498 t_ASTERISK = r'\*' 1499 1500 # Identifiers and reserved words 1501 reserved_map = { } 1502 for r in reserved: 1503 reserved_map[r.lower()] = r 1504 1505 def t_ID(self, t): 1506 r'[A-Za-z_]\w*' 1507 t.type = self.reserved_map.get(t.value, 'ID') 1508 return t 1509 1510 # Integer literal 1511 def t_INTLIT(self, t): 1512 r'-?(0x[\da-fA-F]+)|\d+' 1513 try: 1514 t.value = int(t.value,0) 1515 except ValueError: 1516 error(t, 'Integer value "%s" too large' % t.value) 1517 t.value = 0 1518 return t 1519 1520 # String literal. Note that these use only single quotes, and 1521 # can span multiple lines. 1522 def t_STRLIT(self, t): 1523 r"(?m)'([^'])+'" 1524 # strip off quotes 1525 t.value = t.value[1:-1] 1526 t.lexer.lineno += t.value.count('\n') 1527 return t 1528 1529 1530 # "Code literal"... like a string literal, but delimiters are 1531 # '{{' and '}}' so they get formatted nicely under emacs c-mode 1532 def t_CODELIT(self, t): 1533 r"(?m)\{\{([^\}]|}(?!\}))+\}\}" 1534 # strip off {{ & }} 1535 t.value = t.value[2:-2] 1536 t.lexer.lineno += t.value.count('\n') 1537 return t 1538 1539 def t_CPPDIRECTIVE(self, t): 1540 r'^\#[^\#].*\n' 1541 t.lexer.lineno += t.value.count('\n') 1542 return t 1543 1544 def t_NEWFILE(self, t): 1545 r'^\#\#newfile\s+"[^"]*"' 1546 self.fileNameStack.push((t.value[11:-1], t.lexer.lineno)) 1547 t.lexer.lineno = 0 1548 1549 def t_ENDFILE(self, t): 1550 r'^\#\#endfile' 1551 (old_filename, t.lexer.lineno) = self.fileNameStack.pop() 1552 1553 # 1554 # The functions t_NEWLINE, t_ignore, and t_error are 1555 # special for the lex module. 1556 # 1557 1558 # Newlines 1559 def t_NEWLINE(self, t): 1560 r'\n+' 1561 t.lexer.lineno += t.value.count('\n') 1562 1563 # Comments 1564 def t_comment(self, t): 1565 r'//.*' 1566 1567 # Completely ignored characters 1568 t_ignore = ' \t\x0c' 1569 1570 # Error handler 1571 def t_error(self, t): 1572 error(t, "illegal character '%s'" % t.value[0]) 1573 t.skip(1) 1574 1575 ##################################################################### 1576 # 1577 # Parser 1578 # 1579 # Every function whose name starts with 'p_' defines a grammar 1580 # rule. The rule is encoded in the function's doc string, while 1581 # the function body provides the action taken when the rule is 1582 # matched. The argument to each function is a list of the values 1583 # of the rule's symbols: t[0] for the LHS, and t[1..n] for the 1584 # symbols on the RHS. For tokens, the value is copied from the 1585 # t.value attribute provided by the lexer. For non-terminals, the 1586 # value is assigned by the producing rule; i.e., the job of the 1587 # grammar rule function is to set the value for the non-terminal 1588 # on the LHS (by assigning to t[0]). 1589 ##################################################################### 1590 1591 # The LHS of the first grammar rule is used as the start symbol 1592 # (in this case, 'specification'). Note that this rule enforces 1593 # that there will be exactly one namespace declaration, with 0 or 1594 # more global defs/decls before and after it. The defs & decls 1595 # before the namespace decl will be outside the namespace; those 1596 # after will be inside. The decoder function is always inside the 1597 # namespace. 1598 def p_specification(self, t): 1599 'specification : opt_defs_and_outputs top_level_decode_block' 1600 1601 for f in self.splits.iterkeys(): 1602 f.write('\n#endif\n') 1603 1604 for f in self.files.itervalues(): # close ALL the files; 1605 f.close() # not doing so can cause compilation to fail 1606 1607 self.write_top_level_files() 1608 1609 t[0] = True 1610 1611 # 'opt_defs_and_outputs' is a possibly empty sequence of def and/or 1612 # output statements. Its productions do the hard work of eventually 1613 # instantiating a GenCode, which are generally emitted (written to disk) 1614 # as soon as possible, except for the decode_block, which has to be 1615 # accumulated into one large function of nested switch/case blocks. 1616 def p_opt_defs_and_outputs_0(self, t): 1617 'opt_defs_and_outputs : empty' 1618 1619 def p_opt_defs_and_outputs_1(self, t): 1620 'opt_defs_and_outputs : defs_and_outputs' 1621 1622 def p_defs_and_outputs_0(self, t): 1623 'defs_and_outputs : def_or_output' 1624 1625 def p_defs_and_outputs_1(self, t): 1626 'defs_and_outputs : defs_and_outputs def_or_output' 1627 1628 # The list of possible definition/output statements. 1629 # They are all processed as they are seen. 1630 def p_def_or_output(self, t): 1631 '''def_or_output : name_decl 1632 | def_format 1633 | def_bitfield 1634 | def_bitfield_struct 1635 | def_template 1636 | def_operand_types 1637 | def_operands 1638 | output 1639 | global_let 1640 | split''' 1641 1642 # Utility function used by both invocations of splitting - explicit 1643 # 'split' keyword and split() function inside "let {{ }};" blocks. 1644 def split(self, sec, write=False): 1645 assert(sec != 'header' and "header cannot be split") 1646 1647 f = self.get_file(sec) 1648 self.splits[f] += 1 1649 s = '\n#endif\n#if __SPLIT == %u\n' % self.splits[f] 1650 if write: 1651 f.write(s) 1652 else: 1653 return s 1654 1655 # split output file to reduce compilation time 1656 def p_split(self, t): 1657 'split : SPLIT output_type SEMI' 1658 assert(self.isa_name and "'split' not allowed before namespace decl") 1659 1660 self.split(t[2], True) 1661 1662 def p_output_type(self, t): 1663 '''output_type : DECODER 1664 | HEADER 1665 | EXEC''' 1666 t[0] = t[1] 1667 1668 # ISA name declaration looks like "namespace <foo>;" 1669 def p_name_decl(self, t): 1670 'name_decl : NAMESPACE ID SEMI' 1671 assert(self.isa_name == None and "Only 1 namespace decl permitted") 1672 self.isa_name = t[2] 1673 self.namespace = t[2] + 'Inst' 1674 1675 # Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied 1676 # directly to the appropriate output section. 1677 1678 # Massage output block by substituting in template definitions and 1679 # bit operators. We handle '%'s embedded in the string that don't 1680 # indicate template substitutions (or CPU-specific symbols, which 1681 # get handled in GenCode) by doubling them first so that the 1682 # format operation will reduce them back to single '%'s. 1683 def process_output(self, s): 1684 s = self.protectNonSubstPercents(s) 1685 # protects cpu-specific symbols too 1686 s = self.protectCpuSymbols(s) 1687 return substBitOps(s % self.templateMap) 1688 1689 def p_output(self, t): 1690 'output : OUTPUT output_type CODELIT SEMI' 1691 kwargs = { t[2]+'_output' : self.process_output(t[3]) } 1692 GenCode(self, **kwargs).emit() 1693 1694 # global let blocks 'let {{...}}' (Python code blocks) are 1695 # executed directly when seen. Note that these execute in a 1696 # special variable context 'exportContext' to prevent the code 1697 # from polluting this script's namespace. 1698 def p_global_let(self, t): 1699 'global_let : LET CODELIT SEMI' 1700 def _split(sec): 1701 return self.split(sec) 1702 self.updateExportContext() 1703 self.exportContext["header_output"] = '' 1704 self.exportContext["decoder_output"] = '' 1705 self.exportContext["exec_output"] = '' 1706 self.exportContext["decode_block"] = '' 1707 self.exportContext["split"] = _split 1708 split_setup = ''' 1709def wrap(func): 1710 def split(sec): 1711 globals()[sec + '_output'] += func(sec) 1712 return split 1713split = wrap(split) 1714del wrap 1715''' 1716 # This tricky setup (immediately above) allows us to just write 1717 # (e.g.) "split('exec')" in the Python code and the split #ifdef's 1718 # will automatically be added to the exec_output variable. The inner 1719 # Python execution environment doesn't know about the split points, 1720 # so we carefully inject and wrap a closure that can retrieve the 1721 # next split's #define from the parser and add it to the current 1722 # emission-in-progress. 1723 try: 1724 exec split_setup+fixPythonIndentation(t[2]) in self.exportContext 1725 except Exception, exc: 1726 if debug: 1727 raise 1728 error(t, 'error: %s in global let block "%s".' % (exc, t[2])) 1729 GenCode(self, 1730 header_output=self.exportContext["header_output"], 1731 decoder_output=self.exportContext["decoder_output"], 1732 exec_output=self.exportContext["exec_output"], 1733 decode_block=self.exportContext["decode_block"]).emit() 1734 1735 # Define the mapping from operand type extensions to C++ types and 1736 # bit widths (stored in operandTypeMap). 1737 def p_def_operand_types(self, t): 1738 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI' 1739 try: 1740 self.operandTypeMap = eval('{' + t[3] + '}') 1741 except Exception, exc: 1742 if debug: 1743 raise 1744 error(t, 1745 'error: %s in def operand_types block "%s".' % (exc, t[3])) 1746 1747 # Define the mapping from operand names to operand classes and 1748 # other traits. Stored in operandNameMap. 1749 def p_def_operands(self, t): 1750 'def_operands : DEF OPERANDS CODELIT SEMI' 1751 if not hasattr(self, 'operandTypeMap'): 1752 error(t, 'error: operand types must be defined before operands') 1753 try: 1754 user_dict = eval('{' + t[3] + '}', self.exportContext) 1755 except Exception, exc: 1756 if debug: 1757 raise 1758 error(t, 'error: %s in def operands block "%s".' % (exc, t[3])) 1759 self.buildOperandNameMap(user_dict, t.lexer.lineno) 1760 1761 # A bitfield definition looks like: 1762 # 'def [signed] bitfield <ID> [<first>:<last>]' 1763 # This generates a preprocessor macro in the output file. 1764 def p_def_bitfield_0(self, t): 1765 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI' 1766 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8]) 1767 if (t[2] == 'signed'): 1768 expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr) 1769 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 1770 GenCode(self, header_output=hash_define).emit() 1771 1772 # alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]' 1773 def p_def_bitfield_1(self, t): 1774 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI' 1775 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6]) 1776 if (t[2] == 'signed'): 1777 expr = 'sext<%d>(%s)' % (1, expr) 1778 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 1779 GenCode(self, header_output=hash_define).emit() 1780 1781 # alternate form for structure member: 'def bitfield <ID> <ID>' 1782 def p_def_bitfield_struct(self, t): 1783 'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI' 1784 if (t[2] != ''): 1785 error(t, 'error: structure bitfields are always unsigned.') 1786 expr = 'machInst.%s' % t[5] 1787 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 1788 GenCode(self, header_output=hash_define).emit() 1789 1790 def p_id_with_dot_0(self, t): 1791 'id_with_dot : ID' 1792 t[0] = t[1] 1793 1794 def p_id_with_dot_1(self, t): 1795 'id_with_dot : ID DOT id_with_dot' 1796 t[0] = t[1] + t[2] + t[3] 1797 1798 def p_opt_signed_0(self, t): 1799 'opt_signed : SIGNED' 1800 t[0] = t[1] 1801 1802 def p_opt_signed_1(self, t): 1803 'opt_signed : empty' 1804 t[0] = '' 1805 1806 def p_def_template(self, t): 1807 'def_template : DEF TEMPLATE ID CODELIT SEMI' 1808 if t[3] in self.templateMap: 1809 print "warning: template %s already defined" % t[3] 1810 self.templateMap[t[3]] = Template(self, t[4]) 1811 1812 # An instruction format definition looks like 1813 # "def format <fmt>(<params>) {{...}};" 1814 def p_def_format(self, t): 1815 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI' 1816 (id, params, code) = (t[3], t[5], t[7]) 1817 self.defFormat(id, params, code, t.lexer.lineno) 1818 1819 # The formal parameter list for an instruction format is a 1820 # possibly empty list of comma-separated parameters. Positional 1821 # (standard, non-keyword) parameters must come first, followed by 1822 # keyword parameters, followed by a '*foo' parameter that gets 1823 # excess positional arguments (as in Python). Each of these three 1824 # parameter categories is optional. 1825 # 1826 # Note that we do not support the '**foo' parameter for collecting 1827 # otherwise undefined keyword args. Otherwise the parameter list 1828 # is (I believe) identical to what is supported in Python. 1829 # 1830 # The param list generates a tuple, where the first element is a 1831 # list of the positional params and the second element is a dict 1832 # containing the keyword params. 1833 def p_param_list_0(self, t): 1834 'param_list : positional_param_list COMMA nonpositional_param_list' 1835 t[0] = t[1] + t[3] 1836 1837 def p_param_list_1(self, t): 1838 '''param_list : positional_param_list 1839 | nonpositional_param_list''' 1840 t[0] = t[1] 1841 1842 def p_positional_param_list_0(self, t): 1843 'positional_param_list : empty' 1844 t[0] = [] 1845 1846 def p_positional_param_list_1(self, t): 1847 'positional_param_list : ID' 1848 t[0] = [t[1]] 1849 1850 def p_positional_param_list_2(self, t): 1851 'positional_param_list : positional_param_list COMMA ID' 1852 t[0] = t[1] + [t[3]] 1853 1854 def p_nonpositional_param_list_0(self, t): 1855 'nonpositional_param_list : keyword_param_list COMMA excess_args_param' 1856 t[0] = t[1] + t[3] 1857 1858 def p_nonpositional_param_list_1(self, t): 1859 '''nonpositional_param_list : keyword_param_list 1860 | excess_args_param''' 1861 t[0] = t[1] 1862 1863 def p_keyword_param_list_0(self, t): 1864 'keyword_param_list : keyword_param' 1865 t[0] = [t[1]] 1866 1867 def p_keyword_param_list_1(self, t): 1868 'keyword_param_list : keyword_param_list COMMA keyword_param' 1869 t[0] = t[1] + [t[3]] 1870 1871 def p_keyword_param(self, t): 1872 'keyword_param : ID EQUALS expr' 1873 t[0] = t[1] + ' = ' + t[3].__repr__() 1874 1875 def p_excess_args_param(self, t): 1876 'excess_args_param : ASTERISK ID' 1877 # Just concatenate them: '*ID'. Wrap in list to be consistent 1878 # with positional_param_list and keyword_param_list. 1879 t[0] = [t[1] + t[2]] 1880 1881 # End of format definition-related rules. 1882 ############## 1883 1884 # 1885 # A decode block looks like: 1886 # decode <field1> [, <field2>]* [default <inst>] { ... } 1887 # 1888 def p_top_level_decode_block(self, t): 1889 'top_level_decode_block : decode_block' 1890 codeObj = t[1] 1891 codeObj.wrap_decode_block(''' 1892StaticInstPtr 1893%(isa_name)s::Decoder::decodeInst(%(isa_name)s::ExtMachInst machInst) 1894{ 1895 using namespace %(namespace)s; 1896''' % self, '}') 1897 1898 codeObj.emit() 1899 1900 def p_decode_block(self, t): 1901 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE' 1902 default_defaults = self.defaultStack.pop() 1903 codeObj = t[5] 1904 # use the "default defaults" only if there was no explicit 1905 # default statement in decode_stmt_list 1906 if not codeObj.has_decode_default: 1907 codeObj += default_defaults 1908 codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n') 1909 t[0] = codeObj 1910 1911 # The opt_default statement serves only to push the "default 1912 # defaults" onto defaultStack. This value will be used by nested 1913 # decode blocks, and used and popped off when the current 1914 # decode_block is processed (in p_decode_block() above). 1915 def p_opt_default_0(self, t): 1916 'opt_default : empty' 1917 # no default specified: reuse the one currently at the top of 1918 # the stack 1919 self.defaultStack.push(self.defaultStack.top()) 1920 # no meaningful value returned 1921 t[0] = None 1922 1923 def p_opt_default_1(self, t): 1924 'opt_default : DEFAULT inst' 1925 # push the new default 1926 codeObj = t[2] 1927 codeObj.wrap_decode_block('\ndefault:\n', 'break;\n') 1928 self.defaultStack.push(codeObj) 1929 # no meaningful value returned 1930 t[0] = None 1931 1932 def p_decode_stmt_list_0(self, t): 1933 'decode_stmt_list : decode_stmt' 1934 t[0] = t[1] 1935 1936 def p_decode_stmt_list_1(self, t): 1937 'decode_stmt_list : decode_stmt decode_stmt_list' 1938 if (t[1].has_decode_default and t[2].has_decode_default): 1939 error(t, 'Two default cases in decode block') 1940 t[0] = t[1] + t[2] 1941 1942 # 1943 # Decode statement rules 1944 # 1945 # There are four types of statements allowed in a decode block: 1946 # 1. Format blocks 'format <foo> { ... }' 1947 # 2. Nested decode blocks 1948 # 3. Instruction definitions. 1949 # 4. C preprocessor directives. 1950 1951 1952 # Preprocessor directives found in a decode statement list are 1953 # passed through to the output, replicated to all of the output 1954 # code streams. This works well for ifdefs, so we can ifdef out 1955 # both the declarations and the decode cases generated by an 1956 # instruction definition. Handling them as part of the grammar 1957 # makes it easy to keep them in the right place with respect to 1958 # the code generated by the other statements. 1959 def p_decode_stmt_cpp(self, t): 1960 'decode_stmt : CPPDIRECTIVE' 1961 t[0] = GenCode(self, t[1], t[1], t[1], t[1]) 1962 1963 # A format block 'format <foo> { ... }' sets the default 1964 # instruction format used to handle instruction definitions inside 1965 # the block. This format can be overridden by using an explicit 1966 # format on the instruction definition or with a nested format 1967 # block. 1968 def p_decode_stmt_format(self, t): 1969 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE' 1970 # The format will be pushed on the stack when 'push_format_id' 1971 # is processed (see below). Once the parser has recognized 1972 # the full production (though the right brace), we're done 1973 # with the format, so now we can pop it. 1974 self.formatStack.pop() 1975 t[0] = t[4] 1976 1977 # This rule exists so we can set the current format (& push the 1978 # stack) when we recognize the format name part of the format 1979 # block. 1980 def p_push_format_id(self, t): 1981 'push_format_id : ID' 1982 try: 1983 self.formatStack.push(self.formatMap[t[1]]) 1984 t[0] = ('', '// format %s' % t[1]) 1985 except KeyError: 1986 error(t, 'instruction format "%s" not defined.' % t[1]) 1987 1988 # Nested decode block: if the value of the current field matches 1989 # the specified constant, do a nested decode on some other field. 1990 def p_decode_stmt_decode(self, t): 1991 'decode_stmt : case_label COLON decode_block' 1992 label = t[1] 1993 codeObj = t[3] 1994 # just wrap the decoding code from the block as a case in the 1995 # outer switch statement. 1996 codeObj.wrap_decode_block('\n%s:\n' % label) 1997 codeObj.has_decode_default = (label == 'default') 1998 t[0] = codeObj 1999 2000 # Instruction definition (finally!). 2001 def p_decode_stmt_inst(self, t): 2002 'decode_stmt : case_label COLON inst SEMI' 2003 label = t[1] 2004 codeObj = t[3] 2005 codeObj.wrap_decode_block('\n%s:' % label, 'break;\n') 2006 codeObj.has_decode_default = (label == 'default') 2007 t[0] = codeObj 2008 2009 # The case label is either a list of one or more constants or 2010 # 'default' 2011 def p_case_label_0(self, t): 2012 'case_label : intlit_list' 2013 def make_case(intlit): 2014 if intlit >= 2**32: 2015 return 'case ULL(%#x)' % intlit 2016 else: 2017 return 'case %#x' % intlit 2018 t[0] = ': '.join(map(make_case, t[1])) 2019 2020 def p_case_label_1(self, t): 2021 'case_label : DEFAULT' 2022 t[0] = 'default' 2023 2024 # 2025 # The constant list for a decode case label must be non-empty, but 2026 # may have one or more comma-separated integer literals in it. 2027 # 2028 def p_intlit_list_0(self, t): 2029 'intlit_list : INTLIT' 2030 t[0] = [t[1]] 2031 2032 def p_intlit_list_1(self, t): 2033 'intlit_list : intlit_list COMMA INTLIT' 2034 t[0] = t[1] 2035 t[0].append(t[3]) 2036 2037 # Define an instruction using the current instruction format 2038 # (specified by an enclosing format block). 2039 # "<mnemonic>(<args>)" 2040 def p_inst_0(self, t): 2041 'inst : ID LPAREN arg_list RPAREN' 2042 # Pass the ID and arg list to the current format class to deal with. 2043 currentFormat = self.formatStack.top() 2044 codeObj = currentFormat.defineInst(self, t[1], t[3], t.lexer.lineno) 2045 args = ','.join(map(str, t[3])) 2046 args = re.sub('(?m)^', '//', args) 2047 args = re.sub('^//', '', args) 2048 comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args) 2049 codeObj.prepend_all(comment) 2050 t[0] = codeObj 2051 2052 # Define an instruction using an explicitly specified format: 2053 # "<fmt>::<mnemonic>(<args>)" 2054 def p_inst_1(self, t): 2055 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN' 2056 try: 2057 format = self.formatMap[t[1]] 2058 except KeyError: 2059 error(t, 'instruction format "%s" not defined.' % t[1]) 2060 2061 codeObj = format.defineInst(self, t[3], t[5], t.lexer.lineno) 2062 comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5]) 2063 codeObj.prepend_all(comment) 2064 t[0] = codeObj 2065 2066 # The arg list generates a tuple, where the first element is a 2067 # list of the positional args and the second element is a dict 2068 # containing the keyword args. 2069 def p_arg_list_0(self, t): 2070 'arg_list : positional_arg_list COMMA keyword_arg_list' 2071 t[0] = ( t[1], t[3] ) 2072 2073 def p_arg_list_1(self, t): 2074 'arg_list : positional_arg_list' 2075 t[0] = ( t[1], {} ) 2076 2077 def p_arg_list_2(self, t): 2078 'arg_list : keyword_arg_list' 2079 t[0] = ( [], t[1] ) 2080 2081 def p_positional_arg_list_0(self, t): 2082 'positional_arg_list : empty' 2083 t[0] = [] 2084 2085 def p_positional_arg_list_1(self, t): 2086 'positional_arg_list : expr' 2087 t[0] = [t[1]] 2088 2089 def p_positional_arg_list_2(self, t): 2090 'positional_arg_list : positional_arg_list COMMA expr' 2091 t[0] = t[1] + [t[3]] 2092 2093 def p_keyword_arg_list_0(self, t): 2094 'keyword_arg_list : keyword_arg' 2095 t[0] = t[1] 2096 2097 def p_keyword_arg_list_1(self, t): 2098 'keyword_arg_list : keyword_arg_list COMMA keyword_arg' 2099 t[0] = t[1] 2100 t[0].update(t[3]) 2101 2102 def p_keyword_arg(self, t): 2103 'keyword_arg : ID EQUALS expr' 2104 t[0] = { t[1] : t[3] } 2105 2106 # 2107 # Basic expressions. These constitute the argument values of 2108 # "function calls" (i.e. instruction definitions in the decode 2109 # block) and default values for formal parameters of format 2110 # functions. 2111 # 2112 # Right now, these are either strings, integers, or (recursively) 2113 # lists of exprs (using Python square-bracket list syntax). Note 2114 # that bare identifiers are trated as string constants here (since 2115 # there isn't really a variable namespace to refer to). 2116 # 2117 def p_expr_0(self, t): 2118 '''expr : ID 2119 | INTLIT 2120 | STRLIT 2121 | CODELIT''' 2122 t[0] = t[1] 2123 2124 def p_expr_1(self, t): 2125 '''expr : LBRACKET list_expr RBRACKET''' 2126 t[0] = t[2] 2127 2128 def p_list_expr_0(self, t): 2129 'list_expr : expr' 2130 t[0] = [t[1]] 2131 2132 def p_list_expr_1(self, t): 2133 'list_expr : list_expr COMMA expr' 2134 t[0] = t[1] + [t[3]] 2135 2136 def p_list_expr_2(self, t): 2137 'list_expr : empty' 2138 t[0] = [] 2139 2140 # 2141 # Empty production... use in other rules for readability. 2142 # 2143 def p_empty(self, t): 2144 'empty :' 2145 pass 2146 2147 # Parse error handler. Note that the argument here is the 2148 # offending *token*, not a grammar symbol (hence the need to use 2149 # t.value) 2150 def p_error(self, t): 2151 if t: 2152 error(t, "syntax error at '%s'" % t.value) 2153 else: 2154 error("unknown syntax error") 2155 2156 # END OF GRAMMAR RULES 2157 2158 def updateExportContext(self): 2159 2160 # create a continuation that allows us to grab the current parser 2161 def wrapInstObjParams(*args): 2162 return InstObjParams(self, *args) 2163 self.exportContext['InstObjParams'] = wrapInstObjParams 2164 self.exportContext.update(self.templateMap) 2165 2166 def defFormat(self, id, params, code, lineno): 2167 '''Define a new format''' 2168 2169 # make sure we haven't already defined this one 2170 if id in self.formatMap: 2171 error(lineno, 'format %s redefined.' % id) 2172 2173 # create new object and store in global map 2174 self.formatMap[id] = Format(id, params, code) 2175 2176 def expandCpuSymbolsToDict(self, template): 2177 '''Expand template with CPU-specific references into a 2178 dictionary with an entry for each CPU model name. The entry 2179 key is the model name and the corresponding value is the 2180 template with the CPU-specific refs substituted for that 2181 model.''' 2182 2183 # Protect '%'s that don't go with CPU-specific terms 2184 t = re.sub(r'%(?!\(CPU_)', '%%', template) 2185 result = {} 2186 for cpu in self.cpuModels: 2187 result[cpu.name] = t % cpu.strings 2188 return result 2189 2190 def expandCpuSymbolsToString(self, template): 2191 '''*If* the template has CPU-specific references, return a 2192 single string containing a copy of the template for each CPU 2193 model with the corresponding values substituted in. If the 2194 template has no CPU-specific references, it is returned 2195 unmodified.''' 2196 2197 if template.find('%(CPU_') != -1: 2198 return reduce(lambda x,y: x+y, 2199 self.expandCpuSymbolsToDict(template).values()) 2200 else: 2201 return template 2202 2203 def protectCpuSymbols(self, template): 2204 '''Protect CPU-specific references by doubling the 2205 corresponding '%'s (in preparation for substituting a different 2206 set of references into the template).''' 2207 2208 return re.sub(r'%(?=\(CPU_)', '%%', template) 2209 2210 def protectNonSubstPercents(self, s): 2211 '''Protect any non-dict-substitution '%'s in a format string 2212 (i.e. those not followed by '(')''' 2213 2214 return re.sub(r'%(?!\()', '%%', s) 2215 2216 def buildOperandNameMap(self, user_dict, lineno): 2217 operand_name = {} 2218 for op_name, val in user_dict.iteritems(): 2219 2220 # Check if extra attributes have been specified. 2221 if len(val) > 9: 2222 error(lineno, 'error: too many attributes for operand "%s"' % 2223 base_cls_name) 2224 2225 # Pad val with None in case optional args are missing 2226 val += (None, None, None, None) 2227 base_cls_name, dflt_ext, reg_spec, flags, sort_pri, \ 2228 read_code, write_code, read_predicate, write_predicate = val[:9] 2229 2230 # Canonical flag structure is a triple of lists, where each list 2231 # indicates the set of flags implied by this operand always, when 2232 # used as a source, and when used as a dest, respectively. 2233 # For simplicity this can be initialized using a variety of fairly 2234 # obvious shortcuts; we convert these to canonical form here. 2235 if not flags: 2236 # no flags specified (e.g., 'None') 2237 flags = ( [], [], [] ) 2238 elif isinstance(flags, str): 2239 # a single flag: assumed to be unconditional 2240 flags = ( [ flags ], [], [] ) 2241 elif isinstance(flags, list): 2242 # a list of flags: also assumed to be unconditional 2243 flags = ( flags, [], [] ) 2244 elif isinstance(flags, tuple): 2245 # it's a tuple: it should be a triple, 2246 # but each item could be a single string or a list 2247 (uncond_flags, src_flags, dest_flags) = flags 2248 flags = (makeList(uncond_flags), 2249 makeList(src_flags), makeList(dest_flags)) 2250 2251 # Accumulate attributes of new operand class in tmp_dict 2252 tmp_dict = {} 2253 attrList = ['reg_spec', 'flags', 'sort_pri', 2254 'read_code', 'write_code', 2255 'read_predicate', 'write_predicate'] 2256 if dflt_ext: 2257 dflt_ctype = self.operandTypeMap[dflt_ext] 2258 attrList.extend(['dflt_ctype', 'dflt_ext']) 2259 for attr in attrList: 2260 tmp_dict[attr] = eval(attr) 2261 tmp_dict['base_name'] = op_name 2262 2263 # New class name will be e.g. "IntReg_Ra" 2264 cls_name = base_cls_name + '_' + op_name 2265 # Evaluate string arg to get class object. Note that the 2266 # actual base class for "IntReg" is "IntRegOperand", i.e. we 2267 # have to append "Operand". 2268 try: 2269 base_cls = eval(base_cls_name + 'Operand') 2270 except NameError: 2271 error(lineno, 2272 'error: unknown operand base class "%s"' % base_cls_name) 2273 # The following statement creates a new class called 2274 # <cls_name> as a subclass of <base_cls> with the attributes 2275 # in tmp_dict, just as if we evaluated a class declaration. 2276 operand_name[op_name] = type(cls_name, (base_cls,), tmp_dict) 2277 2278 self.operandNameMap = operand_name 2279 2280 # Define operand variables. 2281 operands = user_dict.keys() 2282 extensions = self.operandTypeMap.keys() 2283 2284 operandsREString = r''' 2285 (?<!\w) # neg. lookbehind assertion: prevent partial matches 2286 ((%s)(?:_(%s))?) # match: operand with optional '_' then suffix 2287 (?!\w) # neg. lookahead assertion: prevent partial matches 2288 ''' % (string.join(operands, '|'), string.join(extensions, '|')) 2289 2290 self.operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE) 2291 2292 # Same as operandsREString, but extension is mandatory, and only two 2293 # groups are returned (base and ext, not full name as above). 2294 # Used for subtituting '_' for '.' to make C++ identifiers. 2295 operandsWithExtREString = r'(?<!\w)(%s)_(%s)(?!\w)' \ 2296 % (string.join(operands, '|'), string.join(extensions, '|')) 2297 2298 self.operandsWithExtRE = \ 2299 re.compile(operandsWithExtREString, re.MULTILINE) 2300 2301 def substMungedOpNames(self, code): 2302 '''Munge operand names in code string to make legal C++ 2303 variable names. This means getting rid of the type extension 2304 if any. Will match base_name attribute of Operand object.)''' 2305 return self.operandsWithExtRE.sub(r'\1', code) 2306 2307 def mungeSnippet(self, s): 2308 '''Fix up code snippets for final substitution in templates.''' 2309 if isinstance(s, str): 2310 return self.substMungedOpNames(substBitOps(s)) 2311 else: 2312 return s 2313 2314 def open(self, name, bare=False): 2315 '''Open the output file for writing and include scary warning.''' 2316 filename = os.path.join(self.output_dir, name) 2317 f = open(filename, 'w') 2318 if f: 2319 if not bare: 2320 f.write(ISAParser.scaremonger_template % self) 2321 return f 2322 2323 def update(self, file, contents): 2324 '''Update the output file only. Scons should handle the case when 2325 the new contents are unchanged using its built-in hash feature.''' 2326 f = self.open(file) 2327 f.write(contents) 2328 f.close() 2329 2330 # This regular expression matches '##include' directives 2331 includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[^"]*)".*$', 2332 re.MULTILINE) 2333 2334 def replace_include(self, matchobj, dirname): 2335 """Function to replace a matched '##include' directive with the 2336 contents of the specified file (with nested ##includes 2337 replaced recursively). 'matchobj' is an re match object 2338 (from a match of includeRE) and 'dirname' is the directory 2339 relative to which the file path should be resolved.""" 2340 2341 fname = matchobj.group('filename') 2342 full_fname = os.path.normpath(os.path.join(dirname, fname)) 2343 contents = '##newfile "%s"\n%s\n##endfile\n' % \ 2344 (full_fname, self.read_and_flatten(full_fname)) 2345 return contents 2346 2347 def read_and_flatten(self, filename): 2348 """Read a file and recursively flatten nested '##include' files.""" 2349 2350 current_dir = os.path.dirname(filename) 2351 try: 2352 contents = open(filename).read() 2353 except IOError: 2354 error('Error including file "%s"' % filename) 2355 2356 self.fileNameStack.push((filename, 0)) 2357 2358 # Find any includes and include them 2359 def replace(matchobj): 2360 return self.replace_include(matchobj, current_dir) 2361 contents = self.includeRE.sub(replace, contents) 2362 2363 self.fileNameStack.pop() 2364 return contents 2365 2366 AlreadyGenerated = {} 2367 2368 def _parse_isa_desc(self, isa_desc_file): 2369 '''Read in and parse the ISA description.''' 2370 2371 # The build system can end up running the ISA parser twice: once to 2372 # finalize the build dependencies, and then to actually generate 2373 # the files it expects (in src/arch/$ARCH/generated). This code 2374 # doesn't do anything different either time, however; the SCons 2375 # invocations just expect different things. Since this code runs 2376 # within SCons, we can just remember that we've already run and 2377 # not perform a completely unnecessary run, since the ISA parser's 2378 # effect is idempotent. 2379 if isa_desc_file in ISAParser.AlreadyGenerated: 2380 return 2381 2382 # grab the last three path components of isa_desc_file 2383 self.filename = '/'.join(isa_desc_file.split('/')[-3:]) 2384 2385 # Read file and (recursively) all included files into a string. 2386 # PLY requires that the input be in a single string so we have to 2387 # do this up front. 2388 isa_desc = self.read_and_flatten(isa_desc_file) 2389 2390 # Initialize filename stack with outer file. 2391 self.fileNameStack.push((isa_desc_file, 0)) 2392 2393 # Parse. 2394 self.parse_string(isa_desc) 2395 2396 ISAParser.AlreadyGenerated[isa_desc_file] = None 2397 2398 def parse_isa_desc(self, *args, **kwargs): 2399 try: 2400 self._parse_isa_desc(*args, **kwargs) 2401 except ISAParserError, e: 2402 e.exit(self.fileNameStack) 2403 2404# Called as script: get args from command line. 2405# Args are: <isa desc file> <output dir> 2406if __name__ == '__main__': 2407 ISAParser(sys.argv[2]).parse_isa_desc(sys.argv[1]) 2408