isa_parser.py revision 10416
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 # Note that initializations in the declarations are solely 860 # to avoid 'uninitialized variable' errors from the compiler. 861 return '%s %s = 0;\n' % (ctype, self.base_name) 862 863 def isPCState(self): 864 return 1 865 866class OperandList(object): 867 '''Find all the operands in the given code block. Returns an operand 868 descriptor list (instance of class OperandList).''' 869 def __init__(self, parser, code): 870 self.items = [] 871 self.bases = {} 872 # delete strings and comments so we don't match on operands inside 873 for regEx in (stringRE, commentRE): 874 code = regEx.sub('', code) 875 # search for operands 876 next_pos = 0 877 while 1: 878 match = parser.operandsRE.search(code, next_pos) 879 if not match: 880 # no more matches: we're done 881 break 882 op = match.groups() 883 # regexp groups are operand full name, base, and extension 884 (op_full, op_base, op_ext) = op 885 # if the token following the operand is an assignment, this is 886 # a destination (LHS), else it's a source (RHS) 887 is_dest = (assignRE.match(code, match.end()) != None) 888 is_src = not is_dest 889 # see if we've already seen this one 890 op_desc = self.find_base(op_base) 891 if op_desc: 892 if op_desc.ext != op_ext: 893 error('Inconsistent extensions for operand %s' % \ 894 op_base) 895 op_desc.is_src = op_desc.is_src or is_src 896 op_desc.is_dest = op_desc.is_dest or is_dest 897 else: 898 # new operand: create new descriptor 899 op_desc = parser.operandNameMap[op_base](parser, 900 op_full, op_ext, is_src, is_dest) 901 self.append(op_desc) 902 # start next search after end of current match 903 next_pos = match.end() 904 self.sort() 905 # enumerate source & dest register operands... used in building 906 # constructor later 907 self.numSrcRegs = 0 908 self.numDestRegs = 0 909 self.numFPDestRegs = 0 910 self.numIntDestRegs = 0 911 self.numCCDestRegs = 0 912 self.numMiscDestRegs = 0 913 self.memOperand = None 914 915 # Flags to keep track if one or more operands are to be read/written 916 # conditionally. 917 self.predRead = False 918 self.predWrite = False 919 920 for op_desc in self.items: 921 if op_desc.isReg(): 922 if op_desc.is_src: 923 op_desc.src_reg_idx = self.numSrcRegs 924 self.numSrcRegs += 1 925 if op_desc.is_dest: 926 op_desc.dest_reg_idx = self.numDestRegs 927 self.numDestRegs += 1 928 if op_desc.isFloatReg(): 929 self.numFPDestRegs += 1 930 elif op_desc.isIntReg(): 931 self.numIntDestRegs += 1 932 elif op_desc.isCCReg(): 933 self.numCCDestRegs += 1 934 elif op_desc.isControlReg(): 935 self.numMiscDestRegs += 1 936 elif op_desc.isMem(): 937 if self.memOperand: 938 error("Code block has more than one memory operand.") 939 self.memOperand = op_desc 940 941 # Check if this operand has read/write predication. If true, then 942 # the microop will dynamically index source/dest registers. 943 self.predRead = self.predRead or op_desc.hasReadPred() 944 self.predWrite = self.predWrite or op_desc.hasWritePred() 945 946 if parser.maxInstSrcRegs < self.numSrcRegs: 947 parser.maxInstSrcRegs = self.numSrcRegs 948 if parser.maxInstDestRegs < self.numDestRegs: 949 parser.maxInstDestRegs = self.numDestRegs 950 if parser.maxMiscDestRegs < self.numMiscDestRegs: 951 parser.maxMiscDestRegs = self.numMiscDestRegs 952 953 # now make a final pass to finalize op_desc fields that may depend 954 # on the register enumeration 955 for op_desc in self.items: 956 op_desc.finalize(self.predRead, self.predWrite) 957 958 def __len__(self): 959 return len(self.items) 960 961 def __getitem__(self, index): 962 return self.items[index] 963 964 def append(self, op_desc): 965 self.items.append(op_desc) 966 self.bases[op_desc.base_name] = op_desc 967 968 def find_base(self, base_name): 969 # like self.bases[base_name], but returns None if not found 970 # (rather than raising exception) 971 return self.bases.get(base_name) 972 973 # internal helper function for concat[Some]Attr{Strings|Lists} 974 def __internalConcatAttrs(self, attr_name, filter, result): 975 for op_desc in self.items: 976 if filter(op_desc): 977 result += getattr(op_desc, attr_name) 978 return result 979 980 # return a single string that is the concatenation of the (string) 981 # values of the specified attribute for all operands 982 def concatAttrStrings(self, attr_name): 983 return self.__internalConcatAttrs(attr_name, lambda x: 1, '') 984 985 # like concatAttrStrings, but only include the values for the operands 986 # for which the provided filter function returns true 987 def concatSomeAttrStrings(self, filter, attr_name): 988 return self.__internalConcatAttrs(attr_name, filter, '') 989 990 # return a single list that is the concatenation of the (list) 991 # values of the specified attribute for all operands 992 def concatAttrLists(self, attr_name): 993 return self.__internalConcatAttrs(attr_name, lambda x: 1, []) 994 995 # like concatAttrLists, but only include the values for the operands 996 # for which the provided filter function returns true 997 def concatSomeAttrLists(self, filter, attr_name): 998 return self.__internalConcatAttrs(attr_name, filter, []) 999 1000 def sort(self): 1001 self.items.sort(lambda a, b: a.sort_pri - b.sort_pri) 1002 1003class SubOperandList(OperandList): 1004 '''Find all the operands in the given code block. Returns an operand 1005 descriptor list (instance of class OperandList).''' 1006 def __init__(self, parser, code, master_list): 1007 self.items = [] 1008 self.bases = {} 1009 # delete strings and comments so we don't match on operands inside 1010 for regEx in (stringRE, commentRE): 1011 code = regEx.sub('', code) 1012 # search for operands 1013 next_pos = 0 1014 while 1: 1015 match = parser.operandsRE.search(code, next_pos) 1016 if not match: 1017 # no more matches: we're done 1018 break 1019 op = match.groups() 1020 # regexp groups are operand full name, base, and extension 1021 (op_full, op_base, op_ext) = op 1022 # find this op in the master list 1023 op_desc = master_list.find_base(op_base) 1024 if not op_desc: 1025 error('Found operand %s which is not in the master list!' \ 1026 ' This is an internal error' % op_base) 1027 else: 1028 # See if we've already found this operand 1029 op_desc = self.find_base(op_base) 1030 if not op_desc: 1031 # if not, add a reference to it to this sub list 1032 self.append(master_list.bases[op_base]) 1033 1034 # start next search after end of current match 1035 next_pos = match.end() 1036 self.sort() 1037 self.memOperand = None 1038 # Whether the whole PC needs to be read so parts of it can be accessed 1039 self.readPC = False 1040 # Whether the whole PC needs to be written after parts of it were 1041 # changed 1042 self.setPC = False 1043 # Whether this instruction manipulates the whole PC or parts of it. 1044 # Mixing the two is a bad idea and flagged as an error. 1045 self.pcPart = None 1046 1047 # Flags to keep track if one or more operands are to be read/written 1048 # conditionally. 1049 self.predRead = False 1050 self.predWrite = False 1051 1052 for op_desc in self.items: 1053 if op_desc.isPCPart(): 1054 self.readPC = True 1055 if op_desc.is_dest: 1056 self.setPC = True 1057 1058 if op_desc.isPCState(): 1059 if self.pcPart is not None: 1060 if self.pcPart and not op_desc.isPCPart() or \ 1061 not self.pcPart and op_desc.isPCPart(): 1062 error("Mixed whole and partial PC state operands.") 1063 self.pcPart = op_desc.isPCPart() 1064 1065 if op_desc.isMem(): 1066 if self.memOperand: 1067 error("Code block has more than one memory operand.") 1068 self.memOperand = op_desc 1069 1070 # Check if this operand has read/write predication. If true, then 1071 # the microop will dynamically index source/dest registers. 1072 self.predRead = self.predRead or op_desc.hasReadPred() 1073 self.predWrite = self.predWrite or op_desc.hasWritePred() 1074 1075# Regular expression object to match C++ strings 1076stringRE = re.compile(r'"([^"\\]|\\.)*"') 1077 1078# Regular expression object to match C++ comments 1079# (used in findOperands()) 1080commentRE = re.compile(r'(^)?[^\S\n]*/(?:\*(.*?)\*/[^\S\n]*|/[^\n]*)($)?', 1081 re.DOTALL | re.MULTILINE) 1082 1083# Regular expression object to match assignment statements 1084# (used in findOperands()) 1085assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE) 1086 1087def makeFlagConstructor(flag_list): 1088 if len(flag_list) == 0: 1089 return '' 1090 # filter out repeated flags 1091 flag_list.sort() 1092 i = 1 1093 while i < len(flag_list): 1094 if flag_list[i] == flag_list[i-1]: 1095 del flag_list[i] 1096 else: 1097 i += 1 1098 pre = '\n\tflags[' 1099 post = '] = true;' 1100 code = pre + string.join(flag_list, post + pre) + post 1101 return code 1102 1103# Assume all instruction flags are of the form 'IsFoo' 1104instFlagRE = re.compile(r'Is.*') 1105 1106# OpClass constants end in 'Op' except No_OpClass 1107opClassRE = re.compile(r'.*Op|No_OpClass') 1108 1109class InstObjParams(object): 1110 def __init__(self, parser, mnem, class_name, base_class = '', 1111 snippets = {}, opt_args = []): 1112 self.mnemonic = mnem 1113 self.class_name = class_name 1114 self.base_class = base_class 1115 if not isinstance(snippets, dict): 1116 snippets = {'code' : snippets} 1117 compositeCode = ' '.join(map(str, snippets.values())) 1118 self.snippets = snippets 1119 1120 self.operands = OperandList(parser, compositeCode) 1121 1122 # The header of the constructor declares the variables to be used 1123 # in the body of the constructor. 1124 header = '' 1125 header += '\n\t_numSrcRegs = 0;' 1126 header += '\n\t_numDestRegs = 0;' 1127 header += '\n\t_numFPDestRegs = 0;' 1128 header += '\n\t_numIntDestRegs = 0;' 1129 header += '\n\t_numCCDestRegs = 0;' 1130 1131 self.constructor = header + \ 1132 self.operands.concatAttrStrings('constructor') 1133 1134 self.flags = self.operands.concatAttrLists('flags') 1135 1136 self.op_class = None 1137 1138 # Optional arguments are assumed to be either StaticInst flags 1139 # or an OpClass value. To avoid having to import a complete 1140 # list of these values to match against, we do it ad-hoc 1141 # with regexps. 1142 for oa in opt_args: 1143 if instFlagRE.match(oa): 1144 self.flags.append(oa) 1145 elif opClassRE.match(oa): 1146 self.op_class = oa 1147 else: 1148 error('InstObjParams: optional arg "%s" not recognized ' 1149 'as StaticInst::Flag or OpClass.' % oa) 1150 1151 # Make a basic guess on the operand class if not set. 1152 # These are good enough for most cases. 1153 if not self.op_class: 1154 if 'IsStore' in self.flags: 1155 self.op_class = 'MemWriteOp' 1156 elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags: 1157 self.op_class = 'MemReadOp' 1158 elif 'IsFloating' in self.flags: 1159 self.op_class = 'FloatAddOp' 1160 else: 1161 self.op_class = 'IntAluOp' 1162 1163 # add flag initialization to contructor here to include 1164 # any flags added via opt_args 1165 self.constructor += makeFlagConstructor(self.flags) 1166 1167 # if 'IsFloating' is set, add call to the FP enable check 1168 # function (which should be provided by isa_desc via a declare) 1169 if 'IsFloating' in self.flags: 1170 self.fp_enable_check = 'fault = checkFpEnableFault(xc);' 1171 else: 1172 self.fp_enable_check = '' 1173 1174############## 1175# Stack: a simple stack object. Used for both formats (formatStack) 1176# and default cases (defaultStack). Simply wraps a list to give more 1177# stack-like syntax and enable initialization with an argument list 1178# (as opposed to an argument that's a list). 1179 1180class Stack(list): 1181 def __init__(self, *items): 1182 list.__init__(self, items) 1183 1184 def push(self, item): 1185 self.append(item); 1186 1187 def top(self): 1188 return self[-1] 1189 1190####################### 1191# 1192# ISA Parser 1193# parses ISA DSL and emits C++ headers and source 1194# 1195 1196class ISAParser(Grammar): 1197 class CpuModel(object): 1198 def __init__(self, name, filename, includes, strings): 1199 self.name = name 1200 self.filename = filename 1201 self.includes = includes 1202 self.strings = strings 1203 1204 def __init__(self, output_dir): 1205 super(ISAParser, self).__init__() 1206 self.output_dir = output_dir 1207 1208 self.filename = None # for output file watermarking/scaremongering 1209 1210 self.cpuModels = [ 1211 ISAParser.CpuModel('ExecContext', 1212 'generic_cpu_exec.cc', 1213 '#include "cpu/exec_context.hh"', 1214 { "CPU_exec_context" : "ExecContext" }), 1215 ] 1216 1217 # variable to hold templates 1218 self.templateMap = {} 1219 1220 # This dictionary maps format name strings to Format objects. 1221 self.formatMap = {} 1222 1223 # Track open files and, if applicable, how many chunks it has been 1224 # split into so far. 1225 self.files = {} 1226 self.splits = {} 1227 1228 # isa_name / namespace identifier from namespace declaration. 1229 # before the namespace declaration, None. 1230 self.isa_name = None 1231 self.namespace = None 1232 1233 # The format stack. 1234 self.formatStack = Stack(NoFormat()) 1235 1236 # The default case stack. 1237 self.defaultStack = Stack(None) 1238 1239 # Stack that tracks current file and line number. Each 1240 # element is a tuple (filename, lineno) that records the 1241 # *current* filename and the line number in the *previous* 1242 # file where it was included. 1243 self.fileNameStack = Stack() 1244 1245 symbols = ('makeList', 're', 'string') 1246 self.exportContext = dict([(s, eval(s)) for s in symbols]) 1247 1248 self.maxInstSrcRegs = 0 1249 self.maxInstDestRegs = 0 1250 self.maxMiscDestRegs = 0 1251 1252 def __getitem__(self, i): # Allow object (self) to be 1253 return getattr(self, i) # passed to %-substitutions 1254 1255 # Change the file suffix of a base filename: 1256 # (e.g.) decoder.cc -> decoder-g.cc.inc for 'global' outputs 1257 def suffixize(self, s, sec): 1258 extn = re.compile('(\.[^\.]+)$') # isolate extension 1259 if self.namespace: 1260 return extn.sub(r'-ns\1.inc', s) # insert some text on either side 1261 else: 1262 return extn.sub(r'-g\1.inc', s) 1263 1264 # Get the file object for emitting code into the specified section 1265 # (header, decoder, exec, decode_block). 1266 def get_file(self, section): 1267 if section == 'decode_block': 1268 filename = 'decode-method.cc.inc' 1269 else: 1270 if section == 'header': 1271 file = 'decoder.hh' 1272 else: 1273 file = '%s.cc' % section 1274 filename = self.suffixize(file, section) 1275 try: 1276 return self.files[filename] 1277 except KeyError: pass 1278 1279 f = self.open(filename) 1280 self.files[filename] = f 1281 1282 # The splittable files are the ones with many independent 1283 # per-instruction functions - the decoder's instruction constructors 1284 # and the instruction execution (execute()) methods. These both have 1285 # the suffix -ns.cc.inc, meaning they are within the namespace part 1286 # of the ISA, contain object-emitting C++ source, and are included 1287 # into other top-level files. These are the files that need special 1288 # #define's to allow parts of them to be compiled separately. Rather 1289 # than splitting the emissions into separate files, the monolithic 1290 # output of the ISA parser is maintained, but the value (or lack 1291 # thereof) of the __SPLIT definition during C preprocessing will 1292 # select the different chunks. If no 'split' directives are used, 1293 # the cpp emissions have no effect. 1294 if re.search('-ns.cc.inc$', filename): 1295 print >>f, '#if !defined(__SPLIT) || (__SPLIT == 1)' 1296 self.splits[f] = 1 1297 # ensure requisite #include's 1298 elif filename in ['decoder-g.cc.inc', 'exec-g.cc.inc']: 1299 print >>f, '#include "decoder.hh"' 1300 elif filename == 'decoder-g.hh.inc': 1301 print >>f, '#include "base/bitfield.hh"' 1302 1303 return f 1304 1305 # Weave together the parts of the different output sections by 1306 # #include'ing them into some very short top-level .cc/.hh files. 1307 # These small files make it much clearer how this tool works, since 1308 # you directly see the chunks emitted as files that are #include'd. 1309 def write_top_level_files(self): 1310 dep = self.open('inc.d', bare=True) 1311 1312 # decoder header - everything depends on this 1313 file = 'decoder.hh' 1314 with self.open(file) as f: 1315 inc = [] 1316 1317 fn = 'decoder-g.hh.inc' 1318 assert(fn in self.files) 1319 f.write('#include "%s"\n' % fn) 1320 inc.append(fn) 1321 1322 fn = 'decoder-ns.hh.inc' 1323 assert(fn in self.files) 1324 f.write('namespace %s {\n#include "%s"\n}\n' 1325 % (self.namespace, fn)) 1326 inc.append(fn) 1327 1328 print >>dep, file+':', ' '.join(inc) 1329 1330 # decoder method - cannot be split 1331 file = 'decoder.cc' 1332 with self.open(file) as f: 1333 inc = [] 1334 1335 fn = 'decoder-g.cc.inc' 1336 assert(fn in self.files) 1337 f.write('#include "%s"\n' % fn) 1338 inc.append(fn) 1339 1340 fn = 'decode-method.cc.inc' 1341 # is guaranteed to have been written for parse to complete 1342 f.write('#include "%s"\n' % fn) 1343 inc.append(fn) 1344 1345 inc.append("decoder.hh") 1346 print >>dep, file+':', ' '.join(inc) 1347 1348 extn = re.compile('(\.[^\.]+)$') 1349 1350 # instruction constructors 1351 splits = self.splits[self.get_file('decoder')] 1352 file_ = 'inst-constrs.cc' 1353 for i in range(1, splits+1): 1354 if splits > 1: 1355 file = extn.sub(r'-%d\1' % i, file_) 1356 else: 1357 file = file_ 1358 with self.open(file) as f: 1359 inc = [] 1360 1361 fn = 'decoder-g.cc.inc' 1362 assert(fn in self.files) 1363 f.write('#include "%s"\n' % fn) 1364 inc.append(fn) 1365 1366 fn = 'decoder-ns.cc.inc' 1367 assert(fn in self.files) 1368 print >>f, 'namespace %s {' % self.namespace 1369 if splits > 1: 1370 print >>f, '#define __SPLIT %u' % i 1371 print >>f, '#include "%s"' % fn 1372 print >>f, '}' 1373 inc.append(fn) 1374 1375 inc.append("decoder.hh") 1376 print >>dep, file+':', ' '.join(inc) 1377 1378 # instruction execution per-CPU model 1379 splits = self.splits[self.get_file('exec')] 1380 for cpu in self.cpuModels: 1381 for i in range(1, splits+1): 1382 if splits > 1: 1383 file = extn.sub(r'_%d\1' % i, cpu.filename) 1384 else: 1385 file = cpu.filename 1386 with self.open(file) as f: 1387 inc = [] 1388 1389 fn = 'exec-g.cc.inc' 1390 assert(fn in self.files) 1391 f.write('#include "%s"\n' % fn) 1392 inc.append(fn) 1393 1394 f.write(cpu.includes+"\n") 1395 1396 fn = 'exec-ns.cc.inc' 1397 assert(fn in self.files) 1398 print >>f, 'namespace %s {' % self.namespace 1399 print >>f, '#define CPU_EXEC_CONTEXT %s' \ 1400 % cpu.strings['CPU_exec_context'] 1401 if splits > 1: 1402 print >>f, '#define __SPLIT %u' % i 1403 print >>f, '#include "%s"' % fn 1404 print >>f, '}' 1405 inc.append(fn) 1406 1407 inc.append("decoder.hh") 1408 print >>dep, file+':', ' '.join(inc) 1409 1410 # max_inst_regs.hh 1411 self.update('max_inst_regs.hh', 1412 '''namespace %(namespace)s { 1413 const int MaxInstSrcRegs = %(maxInstSrcRegs)d; 1414 const int MaxInstDestRegs = %(maxInstDestRegs)d; 1415 const int MaxMiscDestRegs = %(maxMiscDestRegs)d;\n}\n''' % self) 1416 print >>dep, 'max_inst_regs.hh:' 1417 1418 dep.close() 1419 1420 1421 scaremonger_template ='''// DO NOT EDIT 1422// This file was automatically generated from an ISA description: 1423// %(filename)s 1424 1425'''; 1426 1427 ##################################################################### 1428 # 1429 # Lexer 1430 # 1431 # The PLY lexer module takes two things as input: 1432 # - A list of token names (the string list 'tokens') 1433 # - A regular expression describing a match for each token. The 1434 # regexp for token FOO can be provided in two ways: 1435 # - as a string variable named t_FOO 1436 # - as the doc string for a function named t_FOO. In this case, 1437 # the function is also executed, allowing an action to be 1438 # associated with each token match. 1439 # 1440 ##################################################################### 1441 1442 # Reserved words. These are listed separately as they are matched 1443 # using the same regexp as generic IDs, but distinguished in the 1444 # t_ID() function. The PLY documentation suggests this approach. 1445 reserved = ( 1446 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT', 1447 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS', 1448 'OUTPUT', 'SIGNED', 'SPLIT', 'TEMPLATE' 1449 ) 1450 1451 # List of tokens. The lex module requires this. 1452 tokens = reserved + ( 1453 # identifier 1454 'ID', 1455 1456 # integer literal 1457 'INTLIT', 1458 1459 # string literal 1460 'STRLIT', 1461 1462 # code literal 1463 'CODELIT', 1464 1465 # ( ) [ ] { } < > , ; . : :: * 1466 'LPAREN', 'RPAREN', 1467 'LBRACKET', 'RBRACKET', 1468 'LBRACE', 'RBRACE', 1469 'LESS', 'GREATER', 'EQUALS', 1470 'COMMA', 'SEMI', 'DOT', 'COLON', 'DBLCOLON', 1471 'ASTERISK', 1472 1473 # C preprocessor directives 1474 'CPPDIRECTIVE' 1475 1476 # The following are matched but never returned. commented out to 1477 # suppress PLY warning 1478 # newfile directive 1479 # 'NEWFILE', 1480 1481 # endfile directive 1482 # 'ENDFILE' 1483 ) 1484 1485 # Regular expressions for token matching 1486 t_LPAREN = r'\(' 1487 t_RPAREN = r'\)' 1488 t_LBRACKET = r'\[' 1489 t_RBRACKET = r'\]' 1490 t_LBRACE = r'\{' 1491 t_RBRACE = r'\}' 1492 t_LESS = r'\<' 1493 t_GREATER = r'\>' 1494 t_EQUALS = r'=' 1495 t_COMMA = r',' 1496 t_SEMI = r';' 1497 t_DOT = r'\.' 1498 t_COLON = r':' 1499 t_DBLCOLON = r'::' 1500 t_ASTERISK = r'\*' 1501 1502 # Identifiers and reserved words 1503 reserved_map = { } 1504 for r in reserved: 1505 reserved_map[r.lower()] = r 1506 1507 def t_ID(self, t): 1508 r'[A-Za-z_]\w*' 1509 t.type = self.reserved_map.get(t.value, 'ID') 1510 return t 1511 1512 # Integer literal 1513 def t_INTLIT(self, t): 1514 r'-?(0x[\da-fA-F]+)|\d+' 1515 try: 1516 t.value = int(t.value,0) 1517 except ValueError: 1518 error(t, 'Integer value "%s" too large' % t.value) 1519 t.value = 0 1520 return t 1521 1522 # String literal. Note that these use only single quotes, and 1523 # can span multiple lines. 1524 def t_STRLIT(self, t): 1525 r"(?m)'([^'])+'" 1526 # strip off quotes 1527 t.value = t.value[1:-1] 1528 t.lexer.lineno += t.value.count('\n') 1529 return t 1530 1531 1532 # "Code literal"... like a string literal, but delimiters are 1533 # '{{' and '}}' so they get formatted nicely under emacs c-mode 1534 def t_CODELIT(self, t): 1535 r"(?m)\{\{([^\}]|}(?!\}))+\}\}" 1536 # strip off {{ & }} 1537 t.value = t.value[2:-2] 1538 t.lexer.lineno += t.value.count('\n') 1539 return t 1540 1541 def t_CPPDIRECTIVE(self, t): 1542 r'^\#[^\#].*\n' 1543 t.lexer.lineno += t.value.count('\n') 1544 return t 1545 1546 def t_NEWFILE(self, t): 1547 r'^\#\#newfile\s+"[^"]*"' 1548 self.fileNameStack.push((t.value[11:-1], t.lexer.lineno)) 1549 t.lexer.lineno = 0 1550 1551 def t_ENDFILE(self, t): 1552 r'^\#\#endfile' 1553 (old_filename, t.lexer.lineno) = self.fileNameStack.pop() 1554 1555 # 1556 # The functions t_NEWLINE, t_ignore, and t_error are 1557 # special for the lex module. 1558 # 1559 1560 # Newlines 1561 def t_NEWLINE(self, t): 1562 r'\n+' 1563 t.lexer.lineno += t.value.count('\n') 1564 1565 # Comments 1566 def t_comment(self, t): 1567 r'//.*' 1568 1569 # Completely ignored characters 1570 t_ignore = ' \t\x0c' 1571 1572 # Error handler 1573 def t_error(self, t): 1574 error(t, "illegal character '%s'" % t.value[0]) 1575 t.skip(1) 1576 1577 ##################################################################### 1578 # 1579 # Parser 1580 # 1581 # Every function whose name starts with 'p_' defines a grammar 1582 # rule. The rule is encoded in the function's doc string, while 1583 # the function body provides the action taken when the rule is 1584 # matched. The argument to each function is a list of the values 1585 # of the rule's symbols: t[0] for the LHS, and t[1..n] for the 1586 # symbols on the RHS. For tokens, the value is copied from the 1587 # t.value attribute provided by the lexer. For non-terminals, the 1588 # value is assigned by the producing rule; i.e., the job of the 1589 # grammar rule function is to set the value for the non-terminal 1590 # on the LHS (by assigning to t[0]). 1591 ##################################################################### 1592 1593 # The LHS of the first grammar rule is used as the start symbol 1594 # (in this case, 'specification'). Note that this rule enforces 1595 # that there will be exactly one namespace declaration, with 0 or 1596 # more global defs/decls before and after it. The defs & decls 1597 # before the namespace decl will be outside the namespace; those 1598 # after will be inside. The decoder function is always inside the 1599 # namespace. 1600 def p_specification(self, t): 1601 'specification : opt_defs_and_outputs top_level_decode_block' 1602 1603 for f in self.splits.iterkeys(): 1604 f.write('\n#endif\n') 1605 1606 for f in self.files.itervalues(): # close ALL the files; 1607 f.close() # not doing so can cause compilation to fail 1608 1609 self.write_top_level_files() 1610 1611 t[0] = True 1612 1613 # 'opt_defs_and_outputs' is a possibly empty sequence of def and/or 1614 # output statements. Its productions do the hard work of eventually 1615 # instantiating a GenCode, which are generally emitted (written to disk) 1616 # as soon as possible, except for the decode_block, which has to be 1617 # accumulated into one large function of nested switch/case blocks. 1618 def p_opt_defs_and_outputs_0(self, t): 1619 'opt_defs_and_outputs : empty' 1620 1621 def p_opt_defs_and_outputs_1(self, t): 1622 'opt_defs_and_outputs : defs_and_outputs' 1623 1624 def p_defs_and_outputs_0(self, t): 1625 'defs_and_outputs : def_or_output' 1626 1627 def p_defs_and_outputs_1(self, t): 1628 'defs_and_outputs : defs_and_outputs def_or_output' 1629 1630 # The list of possible definition/output statements. 1631 # They are all processed as they are seen. 1632 def p_def_or_output(self, t): 1633 '''def_or_output : name_decl 1634 | def_format 1635 | def_bitfield 1636 | def_bitfield_struct 1637 | def_template 1638 | def_operand_types 1639 | def_operands 1640 | output 1641 | global_let 1642 | split''' 1643 1644 # Utility function used by both invocations of splitting - explicit 1645 # 'split' keyword and split() function inside "let {{ }};" blocks. 1646 def split(self, sec, write=False): 1647 assert(sec != 'header' and "header cannot be split") 1648 1649 f = self.get_file(sec) 1650 self.splits[f] += 1 1651 s = '\n#endif\n#if __SPLIT == %u\n' % self.splits[f] 1652 if write: 1653 f.write(s) 1654 else: 1655 return s 1656 1657 # split output file to reduce compilation time 1658 def p_split(self, t): 1659 'split : SPLIT output_type SEMI' 1660 assert(self.isa_name and "'split' not allowed before namespace decl") 1661 1662 self.split(t[2], True) 1663 1664 def p_output_type(self, t): 1665 '''output_type : DECODER 1666 | HEADER 1667 | EXEC''' 1668 t[0] = t[1] 1669 1670 # ISA name declaration looks like "namespace <foo>;" 1671 def p_name_decl(self, t): 1672 'name_decl : NAMESPACE ID SEMI' 1673 assert(self.isa_name == None and "Only 1 namespace decl permitted") 1674 self.isa_name = t[2] 1675 self.namespace = t[2] + 'Inst' 1676 1677 # Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied 1678 # directly to the appropriate output section. 1679 1680 # Massage output block by substituting in template definitions and 1681 # bit operators. We handle '%'s embedded in the string that don't 1682 # indicate template substitutions (or CPU-specific symbols, which 1683 # get handled in GenCode) by doubling them first so that the 1684 # format operation will reduce them back to single '%'s. 1685 def process_output(self, s): 1686 s = self.protectNonSubstPercents(s) 1687 # protects cpu-specific symbols too 1688 s = self.protectCpuSymbols(s) 1689 return substBitOps(s % self.templateMap) 1690 1691 def p_output(self, t): 1692 'output : OUTPUT output_type CODELIT SEMI' 1693 kwargs = { t[2]+'_output' : self.process_output(t[3]) } 1694 GenCode(self, **kwargs).emit() 1695 1696 # global let blocks 'let {{...}}' (Python code blocks) are 1697 # executed directly when seen. Note that these execute in a 1698 # special variable context 'exportContext' to prevent the code 1699 # from polluting this script's namespace. 1700 def p_global_let(self, t): 1701 'global_let : LET CODELIT SEMI' 1702 def _split(sec): 1703 return self.split(sec) 1704 self.updateExportContext() 1705 self.exportContext["header_output"] = '' 1706 self.exportContext["decoder_output"] = '' 1707 self.exportContext["exec_output"] = '' 1708 self.exportContext["decode_block"] = '' 1709 self.exportContext["split"] = _split 1710 split_setup = ''' 1711def wrap(func): 1712 def split(sec): 1713 globals()[sec + '_output'] += func(sec) 1714 return split 1715split = wrap(split) 1716del wrap 1717''' 1718 # This tricky setup (immediately above) allows us to just write 1719 # (e.g.) "split('exec')" in the Python code and the split #ifdef's 1720 # will automatically be added to the exec_output variable. The inner 1721 # Python execution environment doesn't know about the split points, 1722 # so we carefully inject and wrap a closure that can retrieve the 1723 # next split's #define from the parser and add it to the current 1724 # emission-in-progress. 1725 try: 1726 exec split_setup+fixPythonIndentation(t[2]) in self.exportContext 1727 except Exception, exc: 1728 if debug: 1729 raise 1730 error(t, 'error: %s in global let block "%s".' % (exc, t[2])) 1731 GenCode(self, 1732 header_output=self.exportContext["header_output"], 1733 decoder_output=self.exportContext["decoder_output"], 1734 exec_output=self.exportContext["exec_output"], 1735 decode_block=self.exportContext["decode_block"]).emit() 1736 1737 # Define the mapping from operand type extensions to C++ types and 1738 # bit widths (stored in operandTypeMap). 1739 def p_def_operand_types(self, t): 1740 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI' 1741 try: 1742 self.operandTypeMap = eval('{' + t[3] + '}') 1743 except Exception, exc: 1744 if debug: 1745 raise 1746 error(t, 1747 'error: %s in def operand_types block "%s".' % (exc, t[3])) 1748 1749 # Define the mapping from operand names to operand classes and 1750 # other traits. Stored in operandNameMap. 1751 def p_def_operands(self, t): 1752 'def_operands : DEF OPERANDS CODELIT SEMI' 1753 if not hasattr(self, 'operandTypeMap'): 1754 error(t, 'error: operand types must be defined before operands') 1755 try: 1756 user_dict = eval('{' + t[3] + '}', self.exportContext) 1757 except Exception, exc: 1758 if debug: 1759 raise 1760 error(t, 'error: %s in def operands block "%s".' % (exc, t[3])) 1761 self.buildOperandNameMap(user_dict, t.lexer.lineno) 1762 1763 # A bitfield definition looks like: 1764 # 'def [signed] bitfield <ID> [<first>:<last>]' 1765 # This generates a preprocessor macro in the output file. 1766 def p_def_bitfield_0(self, t): 1767 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI' 1768 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8]) 1769 if (t[2] == 'signed'): 1770 expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr) 1771 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 1772 GenCode(self, header_output=hash_define).emit() 1773 1774 # alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]' 1775 def p_def_bitfield_1(self, t): 1776 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI' 1777 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6]) 1778 if (t[2] == 'signed'): 1779 expr = 'sext<%d>(%s)' % (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 structure member: 'def bitfield <ID> <ID>' 1784 def p_def_bitfield_struct(self, t): 1785 'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI' 1786 if (t[2] != ''): 1787 error(t, 'error: structure bitfields are always unsigned.') 1788 expr = 'machInst.%s' % t[5] 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 def p_id_with_dot_0(self, t): 1793 'id_with_dot : ID' 1794 t[0] = t[1] 1795 1796 def p_id_with_dot_1(self, t): 1797 'id_with_dot : ID DOT id_with_dot' 1798 t[0] = t[1] + t[2] + t[3] 1799 1800 def p_opt_signed_0(self, t): 1801 'opt_signed : SIGNED' 1802 t[0] = t[1] 1803 1804 def p_opt_signed_1(self, t): 1805 'opt_signed : empty' 1806 t[0] = '' 1807 1808 def p_def_template(self, t): 1809 'def_template : DEF TEMPLATE ID CODELIT SEMI' 1810 if t[3] in self.templateMap: 1811 print "warning: template %s already defined" % t[3] 1812 self.templateMap[t[3]] = Template(self, t[4]) 1813 1814 # An instruction format definition looks like 1815 # "def format <fmt>(<params>) {{...}};" 1816 def p_def_format(self, t): 1817 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI' 1818 (id, params, code) = (t[3], t[5], t[7]) 1819 self.defFormat(id, params, code, t.lexer.lineno) 1820 1821 # The formal parameter list for an instruction format is a 1822 # possibly empty list of comma-separated parameters. Positional 1823 # (standard, non-keyword) parameters must come first, followed by 1824 # keyword parameters, followed by a '*foo' parameter that gets 1825 # excess positional arguments (as in Python). Each of these three 1826 # parameter categories is optional. 1827 # 1828 # Note that we do not support the '**foo' parameter for collecting 1829 # otherwise undefined keyword args. Otherwise the parameter list 1830 # is (I believe) identical to what is supported in Python. 1831 # 1832 # The param list generates a tuple, where the first element is a 1833 # list of the positional params and the second element is a dict 1834 # containing the keyword params. 1835 def p_param_list_0(self, t): 1836 'param_list : positional_param_list COMMA nonpositional_param_list' 1837 t[0] = t[1] + t[3] 1838 1839 def p_param_list_1(self, t): 1840 '''param_list : positional_param_list 1841 | nonpositional_param_list''' 1842 t[0] = t[1] 1843 1844 def p_positional_param_list_0(self, t): 1845 'positional_param_list : empty' 1846 t[0] = [] 1847 1848 def p_positional_param_list_1(self, t): 1849 'positional_param_list : ID' 1850 t[0] = [t[1]] 1851 1852 def p_positional_param_list_2(self, t): 1853 'positional_param_list : positional_param_list COMMA ID' 1854 t[0] = t[1] + [t[3]] 1855 1856 def p_nonpositional_param_list_0(self, t): 1857 'nonpositional_param_list : keyword_param_list COMMA excess_args_param' 1858 t[0] = t[1] + t[3] 1859 1860 def p_nonpositional_param_list_1(self, t): 1861 '''nonpositional_param_list : keyword_param_list 1862 | excess_args_param''' 1863 t[0] = t[1] 1864 1865 def p_keyword_param_list_0(self, t): 1866 'keyword_param_list : keyword_param' 1867 t[0] = [t[1]] 1868 1869 def p_keyword_param_list_1(self, t): 1870 'keyword_param_list : keyword_param_list COMMA keyword_param' 1871 t[0] = t[1] + [t[3]] 1872 1873 def p_keyword_param(self, t): 1874 'keyword_param : ID EQUALS expr' 1875 t[0] = t[1] + ' = ' + t[3].__repr__() 1876 1877 def p_excess_args_param(self, t): 1878 'excess_args_param : ASTERISK ID' 1879 # Just concatenate them: '*ID'. Wrap in list to be consistent 1880 # with positional_param_list and keyword_param_list. 1881 t[0] = [t[1] + t[2]] 1882 1883 # End of format definition-related rules. 1884 ############## 1885 1886 # 1887 # A decode block looks like: 1888 # decode <field1> [, <field2>]* [default <inst>] { ... } 1889 # 1890 def p_top_level_decode_block(self, t): 1891 'top_level_decode_block : decode_block' 1892 codeObj = t[1] 1893 codeObj.wrap_decode_block(''' 1894StaticInstPtr 1895%(isa_name)s::Decoder::decodeInst(%(isa_name)s::ExtMachInst machInst) 1896{ 1897 using namespace %(namespace)s; 1898''' % self, '}') 1899 1900 codeObj.emit() 1901 1902 def p_decode_block(self, t): 1903 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE' 1904 default_defaults = self.defaultStack.pop() 1905 codeObj = t[5] 1906 # use the "default defaults" only if there was no explicit 1907 # default statement in decode_stmt_list 1908 if not codeObj.has_decode_default: 1909 codeObj += default_defaults 1910 codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n') 1911 t[0] = codeObj 1912 1913 # The opt_default statement serves only to push the "default 1914 # defaults" onto defaultStack. This value will be used by nested 1915 # decode blocks, and used and popped off when the current 1916 # decode_block is processed (in p_decode_block() above). 1917 def p_opt_default_0(self, t): 1918 'opt_default : empty' 1919 # no default specified: reuse the one currently at the top of 1920 # the stack 1921 self.defaultStack.push(self.defaultStack.top()) 1922 # no meaningful value returned 1923 t[0] = None 1924 1925 def p_opt_default_1(self, t): 1926 'opt_default : DEFAULT inst' 1927 # push the new default 1928 codeObj = t[2] 1929 codeObj.wrap_decode_block('\ndefault:\n', 'break;\n') 1930 self.defaultStack.push(codeObj) 1931 # no meaningful value returned 1932 t[0] = None 1933 1934 def p_decode_stmt_list_0(self, t): 1935 'decode_stmt_list : decode_stmt' 1936 t[0] = t[1] 1937 1938 def p_decode_stmt_list_1(self, t): 1939 'decode_stmt_list : decode_stmt decode_stmt_list' 1940 if (t[1].has_decode_default and t[2].has_decode_default): 1941 error(t, 'Two default cases in decode block') 1942 t[0] = t[1] + t[2] 1943 1944 # 1945 # Decode statement rules 1946 # 1947 # There are four types of statements allowed in a decode block: 1948 # 1. Format blocks 'format <foo> { ... }' 1949 # 2. Nested decode blocks 1950 # 3. Instruction definitions. 1951 # 4. C preprocessor directives. 1952 1953 1954 # Preprocessor directives found in a decode statement list are 1955 # passed through to the output, replicated to all of the output 1956 # code streams. This works well for ifdefs, so we can ifdef out 1957 # both the declarations and the decode cases generated by an 1958 # instruction definition. Handling them as part of the grammar 1959 # makes it easy to keep them in the right place with respect to 1960 # the code generated by the other statements. 1961 def p_decode_stmt_cpp(self, t): 1962 'decode_stmt : CPPDIRECTIVE' 1963 t[0] = GenCode(self, t[1], t[1], t[1], t[1]) 1964 1965 # A format block 'format <foo> { ... }' sets the default 1966 # instruction format used to handle instruction definitions inside 1967 # the block. This format can be overridden by using an explicit 1968 # format on the instruction definition or with a nested format 1969 # block. 1970 def p_decode_stmt_format(self, t): 1971 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE' 1972 # The format will be pushed on the stack when 'push_format_id' 1973 # is processed (see below). Once the parser has recognized 1974 # the full production (though the right brace), we're done 1975 # with the format, so now we can pop it. 1976 self.formatStack.pop() 1977 t[0] = t[4] 1978 1979 # This rule exists so we can set the current format (& push the 1980 # stack) when we recognize the format name part of the format 1981 # block. 1982 def p_push_format_id(self, t): 1983 'push_format_id : ID' 1984 try: 1985 self.formatStack.push(self.formatMap[t[1]]) 1986 t[0] = ('', '// format %s' % t[1]) 1987 except KeyError: 1988 error(t, 'instruction format "%s" not defined.' % t[1]) 1989 1990 # Nested decode block: if the value of the current field matches 1991 # the specified constant, do a nested decode on some other field. 1992 def p_decode_stmt_decode(self, t): 1993 'decode_stmt : case_label COLON decode_block' 1994 label = t[1] 1995 codeObj = t[3] 1996 # just wrap the decoding code from the block as a case in the 1997 # outer switch statement. 1998 codeObj.wrap_decode_block('\n%s:\n' % label) 1999 codeObj.has_decode_default = (label == 'default') 2000 t[0] = codeObj 2001 2002 # Instruction definition (finally!). 2003 def p_decode_stmt_inst(self, t): 2004 'decode_stmt : case_label COLON inst SEMI' 2005 label = t[1] 2006 codeObj = t[3] 2007 codeObj.wrap_decode_block('\n%s:' % label, 'break;\n') 2008 codeObj.has_decode_default = (label == 'default') 2009 t[0] = codeObj 2010 2011 # The case label is either a list of one or more constants or 2012 # 'default' 2013 def p_case_label_0(self, t): 2014 'case_label : intlit_list' 2015 def make_case(intlit): 2016 if intlit >= 2**32: 2017 return 'case ULL(%#x)' % intlit 2018 else: 2019 return 'case %#x' % intlit 2020 t[0] = ': '.join(map(make_case, t[1])) 2021 2022 def p_case_label_1(self, t): 2023 'case_label : DEFAULT' 2024 t[0] = 'default' 2025 2026 # 2027 # The constant list for a decode case label must be non-empty, but 2028 # may have one or more comma-separated integer literals in it. 2029 # 2030 def p_intlit_list_0(self, t): 2031 'intlit_list : INTLIT' 2032 t[0] = [t[1]] 2033 2034 def p_intlit_list_1(self, t): 2035 'intlit_list : intlit_list COMMA INTLIT' 2036 t[0] = t[1] 2037 t[0].append(t[3]) 2038 2039 # Define an instruction using the current instruction format 2040 # (specified by an enclosing format block). 2041 # "<mnemonic>(<args>)" 2042 def p_inst_0(self, t): 2043 'inst : ID LPAREN arg_list RPAREN' 2044 # Pass the ID and arg list to the current format class to deal with. 2045 currentFormat = self.formatStack.top() 2046 codeObj = currentFormat.defineInst(self, t[1], t[3], t.lexer.lineno) 2047 args = ','.join(map(str, t[3])) 2048 args = re.sub('(?m)^', '//', args) 2049 args = re.sub('^//', '', args) 2050 comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args) 2051 codeObj.prepend_all(comment) 2052 t[0] = codeObj 2053 2054 # Define an instruction using an explicitly specified format: 2055 # "<fmt>::<mnemonic>(<args>)" 2056 def p_inst_1(self, t): 2057 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN' 2058 try: 2059 format = self.formatMap[t[1]] 2060 except KeyError: 2061 error(t, 'instruction format "%s" not defined.' % t[1]) 2062 2063 codeObj = format.defineInst(self, t[3], t[5], t.lexer.lineno) 2064 comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5]) 2065 codeObj.prepend_all(comment) 2066 t[0] = codeObj 2067 2068 # The arg list generates a tuple, where the first element is a 2069 # list of the positional args and the second element is a dict 2070 # containing the keyword args. 2071 def p_arg_list_0(self, t): 2072 'arg_list : positional_arg_list COMMA keyword_arg_list' 2073 t[0] = ( t[1], t[3] ) 2074 2075 def p_arg_list_1(self, t): 2076 'arg_list : positional_arg_list' 2077 t[0] = ( t[1], {} ) 2078 2079 def p_arg_list_2(self, t): 2080 'arg_list : keyword_arg_list' 2081 t[0] = ( [], t[1] ) 2082 2083 def p_positional_arg_list_0(self, t): 2084 'positional_arg_list : empty' 2085 t[0] = [] 2086 2087 def p_positional_arg_list_1(self, t): 2088 'positional_arg_list : expr' 2089 t[0] = [t[1]] 2090 2091 def p_positional_arg_list_2(self, t): 2092 'positional_arg_list : positional_arg_list COMMA expr' 2093 t[0] = t[1] + [t[3]] 2094 2095 def p_keyword_arg_list_0(self, t): 2096 'keyword_arg_list : keyword_arg' 2097 t[0] = t[1] 2098 2099 def p_keyword_arg_list_1(self, t): 2100 'keyword_arg_list : keyword_arg_list COMMA keyword_arg' 2101 t[0] = t[1] 2102 t[0].update(t[3]) 2103 2104 def p_keyword_arg(self, t): 2105 'keyword_arg : ID EQUALS expr' 2106 t[0] = { t[1] : t[3] } 2107 2108 # 2109 # Basic expressions. These constitute the argument values of 2110 # "function calls" (i.e. instruction definitions in the decode 2111 # block) and default values for formal parameters of format 2112 # functions. 2113 # 2114 # Right now, these are either strings, integers, or (recursively) 2115 # lists of exprs (using Python square-bracket list syntax). Note 2116 # that bare identifiers are trated as string constants here (since 2117 # there isn't really a variable namespace to refer to). 2118 # 2119 def p_expr_0(self, t): 2120 '''expr : ID 2121 | INTLIT 2122 | STRLIT 2123 | CODELIT''' 2124 t[0] = t[1] 2125 2126 def p_expr_1(self, t): 2127 '''expr : LBRACKET list_expr RBRACKET''' 2128 t[0] = t[2] 2129 2130 def p_list_expr_0(self, t): 2131 'list_expr : expr' 2132 t[0] = [t[1]] 2133 2134 def p_list_expr_1(self, t): 2135 'list_expr : list_expr COMMA expr' 2136 t[0] = t[1] + [t[3]] 2137 2138 def p_list_expr_2(self, t): 2139 'list_expr : empty' 2140 t[0] = [] 2141 2142 # 2143 # Empty production... use in other rules for readability. 2144 # 2145 def p_empty(self, t): 2146 'empty :' 2147 pass 2148 2149 # Parse error handler. Note that the argument here is the 2150 # offending *token*, not a grammar symbol (hence the need to use 2151 # t.value) 2152 def p_error(self, t): 2153 if t: 2154 error(t, "syntax error at '%s'" % t.value) 2155 else: 2156 error("unknown syntax error") 2157 2158 # END OF GRAMMAR RULES 2159 2160 def updateExportContext(self): 2161 2162 # create a continuation that allows us to grab the current parser 2163 def wrapInstObjParams(*args): 2164 return InstObjParams(self, *args) 2165 self.exportContext['InstObjParams'] = wrapInstObjParams 2166 self.exportContext.update(self.templateMap) 2167 2168 def defFormat(self, id, params, code, lineno): 2169 '''Define a new format''' 2170 2171 # make sure we haven't already defined this one 2172 if id in self.formatMap: 2173 error(lineno, 'format %s redefined.' % id) 2174 2175 # create new object and store in global map 2176 self.formatMap[id] = Format(id, params, code) 2177 2178 def expandCpuSymbolsToDict(self, template): 2179 '''Expand template with CPU-specific references into a 2180 dictionary with an entry for each CPU model name. The entry 2181 key is the model name and the corresponding value is the 2182 template with the CPU-specific refs substituted for that 2183 model.''' 2184 2185 # Protect '%'s that don't go with CPU-specific terms 2186 t = re.sub(r'%(?!\(CPU_)', '%%', template) 2187 result = {} 2188 for cpu in self.cpuModels: 2189 result[cpu.name] = t % cpu.strings 2190 return result 2191 2192 def expandCpuSymbolsToString(self, template): 2193 '''*If* the template has CPU-specific references, return a 2194 single string containing a copy of the template for each CPU 2195 model with the corresponding values substituted in. If the 2196 template has no CPU-specific references, it is returned 2197 unmodified.''' 2198 2199 if template.find('%(CPU_') != -1: 2200 return reduce(lambda x,y: x+y, 2201 self.expandCpuSymbolsToDict(template).values()) 2202 else: 2203 return template 2204 2205 def protectCpuSymbols(self, template): 2206 '''Protect CPU-specific references by doubling the 2207 corresponding '%'s (in preparation for substituting a different 2208 set of references into the template).''' 2209 2210 return re.sub(r'%(?=\(CPU_)', '%%', template) 2211 2212 def protectNonSubstPercents(self, s): 2213 '''Protect any non-dict-substitution '%'s in a format string 2214 (i.e. those not followed by '(')''' 2215 2216 return re.sub(r'%(?!\()', '%%', s) 2217 2218 def buildOperandNameMap(self, user_dict, lineno): 2219 operand_name = {} 2220 for op_name, val in user_dict.iteritems(): 2221 2222 # Check if extra attributes have been specified. 2223 if len(val) > 9: 2224 error(lineno, 'error: too many attributes for operand "%s"' % 2225 base_cls_name) 2226 2227 # Pad val with None in case optional args are missing 2228 val += (None, None, None, None) 2229 base_cls_name, dflt_ext, reg_spec, flags, sort_pri, \ 2230 read_code, write_code, read_predicate, write_predicate = val[:9] 2231 2232 # Canonical flag structure is a triple of lists, where each list 2233 # indicates the set of flags implied by this operand always, when 2234 # used as a source, and when used as a dest, respectively. 2235 # For simplicity this can be initialized using a variety of fairly 2236 # obvious shortcuts; we convert these to canonical form here. 2237 if not flags: 2238 # no flags specified (e.g., 'None') 2239 flags = ( [], [], [] ) 2240 elif isinstance(flags, str): 2241 # a single flag: assumed to be unconditional 2242 flags = ( [ flags ], [], [] ) 2243 elif isinstance(flags, list): 2244 # a list of flags: also assumed to be unconditional 2245 flags = ( flags, [], [] ) 2246 elif isinstance(flags, tuple): 2247 # it's a tuple: it should be a triple, 2248 # but each item could be a single string or a list 2249 (uncond_flags, src_flags, dest_flags) = flags 2250 flags = (makeList(uncond_flags), 2251 makeList(src_flags), makeList(dest_flags)) 2252 2253 # Accumulate attributes of new operand class in tmp_dict 2254 tmp_dict = {} 2255 attrList = ['reg_spec', 'flags', 'sort_pri', 2256 'read_code', 'write_code', 2257 'read_predicate', 'write_predicate'] 2258 if dflt_ext: 2259 dflt_ctype = self.operandTypeMap[dflt_ext] 2260 attrList.extend(['dflt_ctype', 'dflt_ext']) 2261 for attr in attrList: 2262 tmp_dict[attr] = eval(attr) 2263 tmp_dict['base_name'] = op_name 2264 2265 # New class name will be e.g. "IntReg_Ra" 2266 cls_name = base_cls_name + '_' + op_name 2267 # Evaluate string arg to get class object. Note that the 2268 # actual base class for "IntReg" is "IntRegOperand", i.e. we 2269 # have to append "Operand". 2270 try: 2271 base_cls = eval(base_cls_name + 'Operand') 2272 except NameError: 2273 error(lineno, 2274 'error: unknown operand base class "%s"' % base_cls_name) 2275 # The following statement creates a new class called 2276 # <cls_name> as a subclass of <base_cls> with the attributes 2277 # in tmp_dict, just as if we evaluated a class declaration. 2278 operand_name[op_name] = type(cls_name, (base_cls,), tmp_dict) 2279 2280 self.operandNameMap = operand_name 2281 2282 # Define operand variables. 2283 operands = user_dict.keys() 2284 extensions = self.operandTypeMap.keys() 2285 2286 operandsREString = r''' 2287 (?<!\w) # neg. lookbehind assertion: prevent partial matches 2288 ((%s)(?:_(%s))?) # match: operand with optional '_' then suffix 2289 (?!\w) # neg. lookahead assertion: prevent partial matches 2290 ''' % (string.join(operands, '|'), string.join(extensions, '|')) 2291 2292 self.operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE) 2293 2294 # Same as operandsREString, but extension is mandatory, and only two 2295 # groups are returned (base and ext, not full name as above). 2296 # Used for subtituting '_' for '.' to make C++ identifiers. 2297 operandsWithExtREString = r'(?<!\w)(%s)_(%s)(?!\w)' \ 2298 % (string.join(operands, '|'), string.join(extensions, '|')) 2299 2300 self.operandsWithExtRE = \ 2301 re.compile(operandsWithExtREString, re.MULTILINE) 2302 2303 def substMungedOpNames(self, code): 2304 '''Munge operand names in code string to make legal C++ 2305 variable names. This means getting rid of the type extension 2306 if any. Will match base_name attribute of Operand object.)''' 2307 return self.operandsWithExtRE.sub(r'\1', code) 2308 2309 def mungeSnippet(self, s): 2310 '''Fix up code snippets for final substitution in templates.''' 2311 if isinstance(s, str): 2312 return self.substMungedOpNames(substBitOps(s)) 2313 else: 2314 return s 2315 2316 def open(self, name, bare=False): 2317 '''Open the output file for writing and include scary warning.''' 2318 filename = os.path.join(self.output_dir, name) 2319 f = open(filename, 'w') 2320 if f: 2321 if not bare: 2322 f.write(ISAParser.scaremonger_template % self) 2323 return f 2324 2325 def update(self, file, contents): 2326 '''Update the output file only. Scons should handle the case when 2327 the new contents are unchanged using its built-in hash feature.''' 2328 f = self.open(file) 2329 f.write(contents) 2330 f.close() 2331 2332 # This regular expression matches '##include' directives 2333 includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[^"]*)".*$', 2334 re.MULTILINE) 2335 2336 def replace_include(self, matchobj, dirname): 2337 """Function to replace a matched '##include' directive with the 2338 contents of the specified file (with nested ##includes 2339 replaced recursively). 'matchobj' is an re match object 2340 (from a match of includeRE) and 'dirname' is the directory 2341 relative to which the file path should be resolved.""" 2342 2343 fname = matchobj.group('filename') 2344 full_fname = os.path.normpath(os.path.join(dirname, fname)) 2345 contents = '##newfile "%s"\n%s\n##endfile\n' % \ 2346 (full_fname, self.read_and_flatten(full_fname)) 2347 return contents 2348 2349 def read_and_flatten(self, filename): 2350 """Read a file and recursively flatten nested '##include' files.""" 2351 2352 current_dir = os.path.dirname(filename) 2353 try: 2354 contents = open(filename).read() 2355 except IOError: 2356 error('Error including file "%s"' % filename) 2357 2358 self.fileNameStack.push((filename, 0)) 2359 2360 # Find any includes and include them 2361 def replace(matchobj): 2362 return self.replace_include(matchobj, current_dir) 2363 contents = self.includeRE.sub(replace, contents) 2364 2365 self.fileNameStack.pop() 2366 return contents 2367 2368 AlreadyGenerated = {} 2369 2370 def _parse_isa_desc(self, isa_desc_file): 2371 '''Read in and parse the ISA description.''' 2372 2373 # The build system can end up running the ISA parser twice: once to 2374 # finalize the build dependencies, and then to actually generate 2375 # the files it expects (in src/arch/$ARCH/generated). This code 2376 # doesn't do anything different either time, however; the SCons 2377 # invocations just expect different things. Since this code runs 2378 # within SCons, we can just remember that we've already run and 2379 # not perform a completely unnecessary run, since the ISA parser's 2380 # effect is idempotent. 2381 if isa_desc_file in ISAParser.AlreadyGenerated: 2382 return 2383 2384 # grab the last three path components of isa_desc_file 2385 self.filename = '/'.join(isa_desc_file.split('/')[-3:]) 2386 2387 # Read file and (recursively) all included files into a string. 2388 # PLY requires that the input be in a single string so we have to 2389 # do this up front. 2390 isa_desc = self.read_and_flatten(isa_desc_file) 2391 2392 # Initialize filename stack with outer file. 2393 self.fileNameStack.push((isa_desc_file, 0)) 2394 2395 # Parse. 2396 self.parse_string(isa_desc) 2397 2398 ISAParser.AlreadyGenerated[isa_desc_file] = None 2399 2400 def parse_isa_desc(self, *args, **kwargs): 2401 try: 2402 self._parse_isa_desc(*args, **kwargs) 2403 except ISAParserError, e: 2404 e.exit(self.fileNameStack) 2405 2406# Called as script: get args from command line. 2407# Args are: <isa desc file> <output dir> 2408if __name__ == '__main__': 2409 ISAParser(sys.argv[2]).parse_isa_desc(sys.argv[1]) 2410