isa_parser.py revision 2155
1#! /usr/bin/env python 2 3# Copyright (c) 2003-2005 The Regents of The University of Michigan 4# All rights reserved. 5# 6# Redistribution and use in source and binary forms, with or without 7# modification, are permitted provided that the following conditions are 8# met: redistributions of source code must retain the above copyright 9# notice, this list of conditions and the following disclaimer; 10# redistributions in binary form must reproduce the above copyright 11# notice, this list of conditions and the following disclaimer in the 12# documentation and/or other materials provided with the distribution; 13# neither the name of the copyright holders nor the names of its 14# contributors may be used to endorse or promote products derived from 15# this software without specific prior written permission. 16# 17# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 18# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 19# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 20# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 21# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 22# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 23# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 27# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 29import os 30import sys 31import re 32import string 33import traceback 34# get type names 35from types import * 36 37# Prepend the directory where the PLY lex & yacc modules are found 38# to the search path. Assumes we're compiling in a subdirectory 39# of 'build' in the current tree. 40sys.path[0:0] = [os.environ['M5_EXT'] + '/ply'] 41 42import lex 43import yacc 44 45##################################################################### 46# 47# Lexer 48# 49# The PLY lexer module takes two things as input: 50# - A list of token names (the string list 'tokens') 51# - A regular expression describing a match for each token. The 52# regexp for token FOO can be provided in two ways: 53# - as a string variable named t_FOO 54# - as the doc string for a function named t_FOO. In this case, 55# the function is also executed, allowing an action to be 56# associated with each token match. 57# 58##################################################################### 59 60# Reserved words. These are listed separately as they are matched 61# using the same regexp as generic IDs, but distinguished in the 62# t_ID() function. The PLY documentation suggests this approach. 63reserved = ( 64 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT', 65 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS', 66 'OUTPUT', 'SIGNED', 'TEMPLATE' 67 ) 68 69# List of tokens. The lex module requires this. 70tokens = reserved + ( 71 # identifier 72 'ID', 73 74 # integer literal 75 'INTLIT', 76 77 # string literal 78 'STRLIT', 79 80 # code literal 81 'CODELIT', 82 83 # ( ) [ ] { } < > , ; : :: * 84 'LPAREN', 'RPAREN', 85 'LBRACKET', 'RBRACKET', 86 'LBRACE', 'RBRACE', 87 'LESS', 'GREATER', 'EQUALS', 88 'COMMA', 'SEMI', 'COLON', 'DBLCOLON', 89 'ASTERISK', 90 91 # C preprocessor directives 92 'CPPDIRECTIVE' 93 94# The following are matched but never returned. commented out to 95# suppress PLY warning 96 # newfile directive 97# 'NEWFILE', 98 99 # endfile directive 100# 'ENDFILE' 101) 102 103# Regular expressions for token matching 104t_LPAREN = r'\(' 105t_RPAREN = r'\)' 106t_LBRACKET = r'\[' 107t_RBRACKET = r'\]' 108t_LBRACE = r'\{' 109t_RBRACE = r'\}' 110t_LESS = r'\<' 111t_GREATER = r'\>' 112t_EQUALS = r'=' 113t_COMMA = r',' 114t_SEMI = r';' 115t_COLON = r':' 116t_DBLCOLON = r'::' 117t_ASTERISK = r'\*' 118 119# Identifiers and reserved words 120reserved_map = { } 121for r in reserved: 122 reserved_map[r.lower()] = r 123 124def t_ID(t): 125 r'[A-Za-z_]\w*' 126 t.type = reserved_map.get(t.value,'ID') 127 return t 128 129# Integer literal 130def t_INTLIT(t): 131 r'(0x[\da-fA-F]+)|\d+' 132 try: 133 t.value = int(t.value,0) 134 except ValueError: 135 error(t.lineno, 'Integer value "%s" too large' % t.value) 136 t.value = 0 137 return t 138 139# String literal. Note that these use only single quotes, and 140# can span multiple lines. 141def t_STRLIT(t): 142 r"(?m)'([^'])+'" 143 # strip off quotes 144 t.value = t.value[1:-1] 145 t.lineno += t.value.count('\n') 146 return t 147 148 149# "Code literal"... like a string literal, but delimiters are 150# '{{' and '}}' so they get formatted nicely under emacs c-mode 151def t_CODELIT(t): 152 r"(?m)\{\{([^\}]|}(?!\}))+\}\}" 153 # strip off {{ & }} 154 t.value = t.value[2:-2] 155 t.lineno += t.value.count('\n') 156 return t 157 158def t_CPPDIRECTIVE(t): 159 r'^\#[^\#].*\n' 160 t.lineno += t.value.count('\n') 161 return t 162 163def t_NEWFILE(t): 164 r'^\#\#newfile\s+"[\w/.-]*"' 165 global fileNameStack 166 fileNameStack.append((t.value[11:-1], t.lineno)) 167 t.lineno = 0 168 169def t_ENDFILE(t): 170 r'^\#\#endfile' 171 (filename, t.lineno) = fileNameStack.pop() 172 173# 174# The functions t_NEWLINE, t_ignore, and t_error are 175# special for the lex module. 176# 177 178# Newlines 179def t_NEWLINE(t): 180 r'\n+' 181 t.lineno += t.value.count('\n') 182 183# Comments 184def t_comment(t): 185 r'//.*' 186 187# Completely ignored characters 188t_ignore = ' \t\x0c' 189 190# Error handler 191def t_error(t): 192 error(t.lineno, "illegal character '%s'" % t.value[0]) 193 t.skip(1) 194 195# Build the lexer 196lex.lex() 197 198##################################################################### 199# 200# Parser 201# 202# Every function whose name starts with 'p_' defines a grammar rule. 203# The rule is encoded in the function's doc string, while the 204# function body provides the action taken when the rule is matched. 205# The argument to each function is a list of the values of the 206# rule's symbols: t[0] for the LHS, and t[1..n] for the symbols 207# on the RHS. For tokens, the value is copied from the t.value 208# attribute provided by the lexer. For non-terminals, the value 209# is assigned by the producing rule; i.e., the job of the grammar 210# rule function is to set the value for the non-terminal on the LHS 211# (by assigning to t[0]). 212##################################################################### 213 214# The LHS of the first grammar rule is used as the start symbol 215# (in this case, 'specification'). Note that this rule enforces 216# that there will be exactly one namespace declaration, with 0 or more 217# global defs/decls before and after it. The defs & decls before 218# the namespace decl will be outside the namespace; those after 219# will be inside. The decoder function is always inside the namespace. 220def p_specification(t): 221 'specification : opt_defs_and_outputs name_decl opt_defs_and_outputs decode_block' 222 global_code = t[1] 223 isa_name = t[2] 224 namespace = isa_name + "Inst" 225 # wrap the decode block as a function definition 226 t[4].wrap_decode_block(''' 227StaticInstPtr 228%(isa_name)s::decodeInst(%(isa_name)s::MachInst machInst) 229{ 230 using namespace %(namespace)s; 231''' % vars(), '}') 232 # both the latter output blocks and the decode block are in the namespace 233 namespace_code = t[3] + t[4] 234 # pass it all back to the caller of yacc.parse() 235 t[0] = (isa_name, namespace, global_code, namespace_code) 236 237# ISA name declaration looks like "namespace <foo>;" 238def p_name_decl(t): 239 'name_decl : NAMESPACE ID SEMI' 240 t[0] = t[2] 241 242# 'opt_defs_and_outputs' is a possibly empty sequence of 243# def and/or output statements. 244def p_opt_defs_and_outputs_0(t): 245 'opt_defs_and_outputs : empty' 246 t[0] = GenCode() 247 248def p_opt_defs_and_outputs_1(t): 249 'opt_defs_and_outputs : defs_and_outputs' 250 t[0] = t[1] 251 252def p_defs_and_outputs_0(t): 253 'defs_and_outputs : def_or_output' 254 t[0] = t[1] 255 256def p_defs_and_outputs_1(t): 257 'defs_and_outputs : defs_and_outputs def_or_output' 258 t[0] = t[1] + t[2] 259 260# The list of possible definition/output statements. 261def p_def_or_output(t): 262 '''def_or_output : def_format 263 | def_bitfield 264 | def_template 265 | def_operand_types 266 | def_operands 267 | output_header 268 | output_decoder 269 | output_exec 270 | global_let''' 271 t[0] = t[1] 272 273# Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied 274# directly to the appropriate output section. 275 276 277# Protect any non-dict-substitution '%'s in a format string 278# (i.e. those not followed by '(') 279def protect_non_subst_percents(s): 280 return re.sub(r'%(?!\()', '%%', s) 281 282# Massage output block by substituting in template definitions and bit 283# operators. We handle '%'s embedded in the string that don't 284# indicate template substitutions (or CPU-specific symbols, which get 285# handled in GenCode) by doubling them first so that the format 286# operation will reduce them back to single '%'s. 287def process_output(s): 288 s = protect_non_subst_percents(s) 289 # protects cpu-specific symbols too 290 s = protect_cpu_symbols(s) 291 return substBitOps(s % templateMap) 292 293def p_output_header(t): 294 'output_header : OUTPUT HEADER CODELIT SEMI' 295 t[0] = GenCode(header_output = process_output(t[3])) 296 297def p_output_decoder(t): 298 'output_decoder : OUTPUT DECODER CODELIT SEMI' 299 t[0] = GenCode(decoder_output = process_output(t[3])) 300 301def p_output_exec(t): 302 'output_exec : OUTPUT EXEC CODELIT SEMI' 303 t[0] = GenCode(exec_output = process_output(t[3])) 304 305# global let blocks 'let {{...}}' (Python code blocks) are executed 306# directly when seen. Note that these execute in a special variable 307# context 'exportContext' to prevent the code from polluting this 308# script's namespace. 309def p_global_let(t): 310 'global_let : LET CODELIT SEMI' 311 updateExportContext() 312 try: 313 exec fixPythonIndentation(t[2]) in exportContext 314 except Exception, exc: 315 error(t.lineno(1), 316 'error: %s in global let block "%s".' % (exc, t[2])) 317 t[0] = GenCode() # contributes nothing to the output C++ file 318 319# Define the mapping from operand type extensions to C++ types and bit 320# widths (stored in operandTypeMap). 321def p_def_operand_types(t): 322 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI' 323 try: 324 userDict = eval('{' + t[3] + '}') 325 except Exception, exc: 326 error(t.lineno(1), 327 'error: %s in def operand_types block "%s".' % (exc, t[3])) 328 buildOperandTypeMap(userDict, t.lineno(1)) 329 t[0] = GenCode() # contributes nothing to the output C++ file 330 331# Define the mapping from operand names to operand classes and other 332# traits. Stored in operandNameMap. 333def p_def_operands(t): 334 'def_operands : DEF OPERANDS CODELIT SEMI' 335 if not globals().has_key('operandTypeMap'): 336 error(t.lineno(1), 337 'error: operand types must be defined before operands') 338 try: 339 userDict = eval('{' + t[3] + '}') 340 except Exception, exc: 341 error(t.lineno(1), 342 'error: %s in def operands block "%s".' % (exc, t[3])) 343 buildOperandNameMap(userDict, t.lineno(1)) 344 t[0] = GenCode() # contributes nothing to the output C++ file 345 346# A bitfield definition looks like: 347# 'def [signed] bitfield <ID> [<first>:<last>]' 348# This generates a preprocessor macro in the output file. 349def p_def_bitfield_0(t): 350 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI' 351 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8]) 352 if (t[2] == 'signed'): 353 expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr) 354 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 355 t[0] = GenCode(header_output = hash_define) 356 357# alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]' 358def p_def_bitfield_1(t): 359 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI' 360 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6]) 361 if (t[2] == 'signed'): 362 expr = 'sext<%d>(%s)' % (1, expr) 363 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 364 t[0] = GenCode(header_output = hash_define) 365 366def p_opt_signed_0(t): 367 'opt_signed : SIGNED' 368 t[0] = t[1] 369 370def p_opt_signed_1(t): 371 'opt_signed : empty' 372 t[0] = '' 373 374# Global map variable to hold templates 375templateMap = {} 376 377def p_def_template(t): 378 'def_template : DEF TEMPLATE ID CODELIT SEMI' 379 templateMap[t[3]] = Template(t[4]) 380 t[0] = GenCode() 381 382# An instruction format definition looks like 383# "def format <fmt>(<params>) {{...}};" 384def p_def_format(t): 385 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI' 386 (id, params, code) = (t[3], t[5], t[7]) 387 defFormat(id, params, code, t.lineno(1)) 388 t[0] = GenCode() 389 390# The formal parameter list for an instruction format is a possibly 391# empty list of comma-separated parameters. Positional (standard, 392# non-keyword) parameters must come first, followed by keyword 393# parameters, followed by a '*foo' parameter that gets excess 394# positional arguments (as in Python). Each of these three parameter 395# categories is optional. 396# 397# Note that we do not support the '**foo' parameter for collecting 398# otherwise undefined keyword args. Otherwise the parameter list is 399# (I believe) identical to what is supported in Python. 400# 401# The param list generates a tuple, where the first element is a list of 402# the positional params and the second element is a dict containing the 403# keyword params. 404def p_param_list_0(t): 405 'param_list : positional_param_list COMMA nonpositional_param_list' 406 t[0] = t[1] + t[3] 407 408def p_param_list_1(t): 409 '''param_list : positional_param_list 410 | nonpositional_param_list''' 411 t[0] = t[1] 412 413def p_positional_param_list_0(t): 414 'positional_param_list : empty' 415 t[0] = [] 416 417def p_positional_param_list_1(t): 418 'positional_param_list : ID' 419 t[0] = [t[1]] 420 421def p_positional_param_list_2(t): 422 'positional_param_list : positional_param_list COMMA ID' 423 t[0] = t[1] + [t[3]] 424 425def p_nonpositional_param_list_0(t): 426 'nonpositional_param_list : keyword_param_list COMMA excess_args_param' 427 t[0] = t[1] + t[3] 428 429def p_nonpositional_param_list_1(t): 430 '''nonpositional_param_list : keyword_param_list 431 | excess_args_param''' 432 t[0] = t[1] 433 434def p_keyword_param_list_0(t): 435 'keyword_param_list : keyword_param' 436 t[0] = [t[1]] 437 438def p_keyword_param_list_1(t): 439 'keyword_param_list : keyword_param_list COMMA keyword_param' 440 t[0] = t[1] + [t[3]] 441 442def p_keyword_param(t): 443 'keyword_param : ID EQUALS expr' 444 t[0] = t[1] + ' = ' + t[3].__repr__() 445 446def p_excess_args_param(t): 447 'excess_args_param : ASTERISK ID' 448 # Just concatenate them: '*ID'. Wrap in list to be consistent 449 # with positional_param_list and keyword_param_list. 450 t[0] = [t[1] + t[2]] 451 452# End of format definition-related rules. 453############## 454 455# 456# A decode block looks like: 457# decode <field1> [, <field2>]* [default <inst>] { ... } 458# 459def p_decode_block(t): 460 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE' 461 default_defaults = defaultStack.pop() 462 codeObj = t[5] 463 # use the "default defaults" only if there was no explicit 464 # default statement in decode_stmt_list 465 if not codeObj.has_decode_default: 466 codeObj += default_defaults 467 codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n') 468 t[0] = codeObj 469 470# The opt_default statement serves only to push the "default defaults" 471# onto defaultStack. This value will be used by nested decode blocks, 472# and used and popped off when the current decode_block is processed 473# (in p_decode_block() above). 474def p_opt_default_0(t): 475 'opt_default : empty' 476 # no default specified: reuse the one currently at the top of the stack 477 defaultStack.push(defaultStack.top()) 478 # no meaningful value returned 479 t[0] = None 480 481def p_opt_default_1(t): 482 'opt_default : DEFAULT inst' 483 # push the new default 484 codeObj = t[2] 485 codeObj.wrap_decode_block('\ndefault:\n', 'break;\n') 486 defaultStack.push(codeObj) 487 # no meaningful value returned 488 t[0] = None 489 490def p_decode_stmt_list_0(t): 491 'decode_stmt_list : decode_stmt' 492 t[0] = t[1] 493 494def p_decode_stmt_list_1(t): 495 'decode_stmt_list : decode_stmt decode_stmt_list' 496 if (t[1].has_decode_default and t[2].has_decode_default): 497 error(t.lineno(1), 'Two default cases in decode block') 498 t[0] = t[1] + t[2] 499 500# 501# Decode statement rules 502# 503# There are four types of statements allowed in a decode block: 504# 1. Format blocks 'format <foo> { ... }' 505# 2. Nested decode blocks 506# 3. Instruction definitions. 507# 4. C preprocessor directives. 508 509 510# Preprocessor directives found in a decode statement list are passed 511# through to the output, replicated to all of the output code 512# streams. This works well for ifdefs, so we can ifdef out both the 513# declarations and the decode cases generated by an instruction 514# definition. Handling them as part of the grammar makes it easy to 515# keep them in the right place with respect to the code generated by 516# the other statements. 517def p_decode_stmt_cpp(t): 518 'decode_stmt : CPPDIRECTIVE' 519 t[0] = GenCode(t[1], t[1], t[1], t[1]) 520 521# A format block 'format <foo> { ... }' sets the default instruction 522# format used to handle instruction definitions inside the block. 523# This format can be overridden by using an explicit format on the 524# instruction definition or with a nested format block. 525def p_decode_stmt_format(t): 526 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE' 527 # The format will be pushed on the stack when 'push_format_id' is 528 # processed (see below). Once the parser has recognized the full 529 # production (though the right brace), we're done with the format, 530 # so now we can pop it. 531 formatStack.pop() 532 t[0] = t[4] 533 534# This rule exists so we can set the current format (& push the stack) 535# when we recognize the format name part of the format block. 536def p_push_format_id(t): 537 'push_format_id : ID' 538 try: 539 formatStack.push(formatMap[t[1]]) 540 t[0] = ('', '// format %s' % t[1]) 541 except KeyError: 542 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 543 544# Nested decode block: if the value of the current field matches the 545# specified constant, do a nested decode on some other field. 546def p_decode_stmt_decode(t): 547 'decode_stmt : case_label COLON decode_block' 548 label = t[1] 549 codeObj = t[3] 550 # just wrap the decoding code from the block as a case in the 551 # outer switch statement. 552 codeObj.wrap_decode_block('\n%s:\n' % label) 553 codeObj.has_decode_default = (label == 'default') 554 t[0] = codeObj 555 556# Instruction definition (finally!). 557def p_decode_stmt_inst(t): 558 'decode_stmt : case_label COLON inst SEMI' 559 label = t[1] 560 codeObj = t[3] 561 codeObj.wrap_decode_block('\n%s:' % label, 'break;\n') 562 codeObj.has_decode_default = (label == 'default') 563 t[0] = codeObj 564 565# The case label is either a list of one or more constants or 'default' 566def p_case_label_0(t): 567 'case_label : intlit_list' 568 t[0] = ': '.join(map(lambda a: 'case %#x' % a, t[1])) 569 570def p_case_label_1(t): 571 'case_label : DEFAULT' 572 t[0] = 'default' 573 574# 575# The constant list for a decode case label must be non-empty, but may have 576# one or more comma-separated integer literals in it. 577# 578def p_intlit_list_0(t): 579 'intlit_list : INTLIT' 580 t[0] = [t[1]] 581 582def p_intlit_list_1(t): 583 'intlit_list : intlit_list COMMA INTLIT' 584 t[0] = t[1] 585 t[0].append(t[3]) 586 587# Define an instruction using the current instruction format (specified 588# by an enclosing format block). 589# "<mnemonic>(<args>)" 590def p_inst_0(t): 591 'inst : ID LPAREN arg_list RPAREN' 592 # Pass the ID and arg list to the current format class to deal with. 593 currentFormat = formatStack.top() 594 codeObj = currentFormat.defineInst(t[1], t[3], t.lineno(1)) 595 args = ','.join(map(str, t[3])) 596 args = re.sub('(?m)^', '//', args) 597 args = re.sub('^//', '', args) 598 comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args) 599 codeObj.prepend_all(comment) 600 t[0] = codeObj 601 602# Define an instruction using an explicitly specified format: 603# "<fmt>::<mnemonic>(<args>)" 604def p_inst_1(t): 605 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN' 606 try: 607 format = formatMap[t[1]] 608 except KeyError: 609 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 610 codeObj = format.defineInst(t[3], t[5], t.lineno(1)) 611 comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5]) 612 codeObj.prepend_all(comment) 613 t[0] = codeObj 614 615# The arg list generates a tuple, where the first element is a list of 616# the positional args and the second element is a dict containing the 617# keyword args. 618def p_arg_list_0(t): 619 'arg_list : positional_arg_list COMMA keyword_arg_list' 620 t[0] = ( t[1], t[3] ) 621 622def p_arg_list_1(t): 623 'arg_list : positional_arg_list' 624 t[0] = ( t[1], {} ) 625 626def p_arg_list_2(t): 627 'arg_list : keyword_arg_list' 628 t[0] = ( [], t[1] ) 629 630def p_positional_arg_list_0(t): 631 'positional_arg_list : empty' 632 t[0] = [] 633 634def p_positional_arg_list_1(t): 635 'positional_arg_list : expr' 636 t[0] = [t[1]] 637 638def p_positional_arg_list_2(t): 639 'positional_arg_list : positional_arg_list COMMA expr' 640 t[0] = t[1] + [t[3]] 641 642def p_keyword_arg_list_0(t): 643 'keyword_arg_list : keyword_arg' 644 t[0] = t[1] 645 646def p_keyword_arg_list_1(t): 647 'keyword_arg_list : keyword_arg_list COMMA keyword_arg' 648 t[0] = t[1] 649 t[0].update(t[3]) 650 651def p_keyword_arg(t): 652 'keyword_arg : ID EQUALS expr' 653 t[0] = { t[1] : t[3] } 654 655# 656# Basic expressions. These constitute the argument values of 657# "function calls" (i.e. instruction definitions in the decode block) 658# and default values for formal parameters of format functions. 659# 660# Right now, these are either strings, integers, or (recursively) 661# lists of exprs (using Python square-bracket list syntax). Note that 662# bare identifiers are trated as string constants here (since there 663# isn't really a variable namespace to refer to). 664# 665def p_expr_0(t): 666 '''expr : ID 667 | INTLIT 668 | STRLIT 669 | CODELIT''' 670 t[0] = t[1] 671 672def p_expr_1(t): 673 '''expr : LBRACKET list_expr RBRACKET''' 674 t[0] = t[2] 675 676def p_list_expr_0(t): 677 'list_expr : expr' 678 t[0] = [t[1]] 679 680def p_list_expr_1(t): 681 'list_expr : list_expr COMMA expr' 682 t[0] = t[1] + [t[3]] 683 684def p_list_expr_2(t): 685 'list_expr : empty' 686 t[0] = [] 687 688# 689# Empty production... use in other rules for readability. 690# 691def p_empty(t): 692 'empty :' 693 pass 694 695# Parse error handler. Note that the argument here is the offending 696# *token*, not a grammar symbol (hence the need to use t.value) 697def p_error(t): 698 if t: 699 error(t.lineno, "syntax error at '%s'" % t.value) 700 else: 701 error_bt(0, "unknown syntax error") 702 703# END OF GRAMMAR RULES 704# 705# Now build the parser. 706yacc.yacc() 707 708 709##################################################################### 710# 711# Support Classes 712# 713##################################################################### 714 715# Expand template with CPU-specific references into a dictionary with 716# an entry for each CPU model name. The entry key is the model name 717# and the corresponding value is the template with the CPU-specific 718# refs substituted for that model. 719def expand_cpu_symbols_to_dict(template): 720 # Protect '%'s that don't go with CPU-specific terms 721 t = re.sub(r'%(?!\(CPU_)', '%%', template) 722 result = {} 723 for cpu in cpu_models: 724 result[cpu.name] = t % cpu.strings 725 return result 726 727# *If* the template has CPU-specific references, return a single 728# string containing a copy of the template for each CPU model with the 729# corresponding values substituted in. If the template has no 730# CPU-specific references, it is returned unmodified. 731def expand_cpu_symbols_to_string(template): 732 if template.find('%(CPU_') != -1: 733 return reduce(lambda x,y: x+y, 734 expand_cpu_symbols_to_dict(template).values()) 735 else: 736 return template 737 738# Protect CPU-specific references by doubling the corresponding '%'s 739# (in preparation for substituting a different set of references into 740# the template). 741def protect_cpu_symbols(template): 742 return re.sub(r'%(?=\(CPU_)', '%%', template) 743 744############### 745# GenCode class 746# 747# The GenCode class encapsulates generated code destined for various 748# output files. The header_output and decoder_output attributes are 749# strings containing code destined for decoder.hh and decoder.cc 750# respectively. The decode_block attribute contains code to be 751# incorporated in the decode function itself (that will also end up in 752# decoder.cc). The exec_output attribute is a dictionary with a key 753# for each CPU model name; the value associated with a particular key 754# is the string of code for that CPU model's exec.cc file. The 755# has_decode_default attribute is used in the decode block to allow 756# explicit default clauses to override default default clauses. 757 758class GenCode: 759 # Constructor. At this point we substitute out all CPU-specific 760 # symbols. For the exec output, these go into the per-model 761 # dictionary. For all other output types they get collapsed into 762 # a single string. 763 def __init__(self, 764 header_output = '', decoder_output = '', exec_output = '', 765 decode_block = '', has_decode_default = False): 766 self.header_output = expand_cpu_symbols_to_string(header_output) 767 self.decoder_output = expand_cpu_symbols_to_string(decoder_output) 768 if isinstance(exec_output, dict): 769 self.exec_output = exec_output 770 elif isinstance(exec_output, str): 771 # If the exec_output arg is a single string, we replicate 772 # it for each of the CPU models, substituting and 773 # %(CPU_foo)s params appropriately. 774 self.exec_output = expand_cpu_symbols_to_dict(exec_output) 775 self.decode_block = expand_cpu_symbols_to_string(decode_block) 776 self.has_decode_default = has_decode_default 777 778 # Override '+' operator: generate a new GenCode object that 779 # concatenates all the individual strings in the operands. 780 def __add__(self, other): 781 exec_output = {} 782 for cpu in cpu_models: 783 n = cpu.name 784 exec_output[n] = self.exec_output[n] + other.exec_output[n] 785 return GenCode(self.header_output + other.header_output, 786 self.decoder_output + other.decoder_output, 787 exec_output, 788 self.decode_block + other.decode_block, 789 self.has_decode_default or other.has_decode_default) 790 791 # Prepend a string (typically a comment) to all the strings. 792 def prepend_all(self, pre): 793 self.header_output = pre + self.header_output 794 self.decoder_output = pre + self.decoder_output 795 self.decode_block = pre + self.decode_block 796 for cpu in cpu_models: 797 self.exec_output[cpu.name] = pre + self.exec_output[cpu.name] 798 799 # Wrap the decode block in a pair of strings (e.g., 'case foo:' 800 # and 'break;'). Used to build the big nested switch statement. 801 def wrap_decode_block(self, pre, post = ''): 802 self.decode_block = pre + indent(self.decode_block) + post 803 804################ 805# Format object. 806# 807# A format object encapsulates an instruction format. It must provide 808# a defineInst() method that generates the code for an instruction 809# definition. 810 811exportContextSymbols = ('InstObjParams', 'CodeBlock', 812 'makeList', 're', 'string') 813 814exportContext = {} 815 816def updateExportContext(): 817 exportContext.update(exportDict(*exportContextSymbols)) 818 exportContext.update(templateMap) 819 820def exportDict(*symNames): 821 return dict([(s, eval(s)) for s in symNames]) 822 823 824class Format: 825 def __init__(self, id, params, code): 826 # constructor: just save away arguments 827 self.id = id 828 self.params = params 829 label = 'def format ' + id 830 self.user_code = compile(fixPythonIndentation(code), label, 'exec') 831 param_list = string.join(params, ", ") 832 f = '''def defInst(_code, _context, %s): 833 my_locals = vars().copy() 834 exec _code in _context, my_locals 835 return my_locals\n''' % param_list 836 c = compile(f, label + ' wrapper', 'exec') 837 exec c 838 self.func = defInst 839 840 def defineInst(self, name, args, lineno): 841 context = {} 842 updateExportContext() 843 context.update(exportContext) 844 context.update({ 'name': name, 'Name': string.capitalize(name) }) 845 try: 846 vars = self.func(self.user_code, context, *args[0], **args[1]) 847 except Exception, exc: 848 error(lineno, 'error defining "%s": %s.' % (name, exc)) 849 for k in vars.keys(): 850 if k not in ('header_output', 'decoder_output', 851 'exec_output', 'decode_block'): 852 del vars[k] 853 return GenCode(**vars) 854 855# Special null format to catch an implicit-format instruction 856# definition outside of any format block. 857class NoFormat: 858 def __init__(self): 859 self.defaultInst = '' 860 861 def defineInst(self, name, args, lineno): 862 error(lineno, 863 'instruction definition "%s" with no active format!' % name) 864 865# This dictionary maps format name strings to Format objects. 866formatMap = {} 867 868# Define a new format 869def defFormat(id, params, code, lineno): 870 # make sure we haven't already defined this one 871 if formatMap.get(id, None) != None: 872 error(lineno, 'format %s redefined.' % id) 873 # create new object and store in global map 874 formatMap[id] = Format(id, params, code) 875 876 877############## 878# Stack: a simple stack object. Used for both formats (formatStack) 879# and default cases (defaultStack). Simply wraps a list to give more 880# stack-like syntax and enable initialization with an argument list 881# (as opposed to an argument that's a list). 882 883class Stack(list): 884 def __init__(self, *items): 885 list.__init__(self, items) 886 887 def push(self, item): 888 self.append(item); 889 890 def top(self): 891 return self[-1] 892 893# The global format stack. 894formatStack = Stack(NoFormat()) 895 896# The global default case stack. 897defaultStack = Stack( None ) 898 899################### 900# Utility functions 901 902# 903# Indent every line in string 's' by two spaces 904# (except preprocessor directives). 905# Used to make nested code blocks look pretty. 906# 907def indent(s): 908 return re.sub(r'(?m)^(?!#)', ' ', s) 909 910# 911# Munge a somewhat arbitrarily formatted piece of Python code 912# (e.g. from a format 'let' block) into something whose indentation 913# will get by the Python parser. 914# 915# The two keys here are that Python will give a syntax error if 916# there's any whitespace at the beginning of the first line, and that 917# all lines at the same lexical nesting level must have identical 918# indentation. Unfortunately the way code literals work, an entire 919# let block tends to have some initial indentation. Rather than 920# trying to figure out what that is and strip it off, we prepend 'if 921# 1:' to make the let code the nested block inside the if (and have 922# the parser automatically deal with the indentation for us). 923# 924# We don't want to do this if (1) the code block is empty or (2) the 925# first line of the block doesn't have any whitespace at the front. 926 927def fixPythonIndentation(s): 928 # get rid of blank lines first 929 s = re.sub(r'(?m)^\s*\n', '', s); 930 if (s != '' and re.match(r'[ \t]', s[0])): 931 s = 'if 1:\n' + s 932 return s 933 934# Error handler. Just call exit. Output formatted to work under 935# Emacs compile-mode. This function should be called when errors due 936# to user input are detected (as opposed to parser bugs). 937def error(lineno, string): 938 spaces = "" 939 for (filename, line) in fileNameStack[0:-1]: 940 print spaces + "In file included from " + filename 941 spaces += " " 942 # Uncomment the following line to get a Python stack backtrace for 943 # these errors too. Can be handy when trying to debug the parser. 944 # traceback.print_exc() 945 sys.exit(spaces + "%s:%d: %s" % (fileNameStack[-1][0], lineno, string)) 946 947# Like error(), but include a Python stack backtrace (for processing 948# Python exceptions). This function should be called for errors that 949# appear to be bugs in the parser itself. 950def error_bt(lineno, string): 951 traceback.print_exc() 952 print >> sys.stderr, "%s:%d: %s" % (input_filename, lineno, string) 953 sys.exit(1) 954 955 956##################################################################### 957# 958# Bitfield Operator Support 959# 960##################################################################### 961 962bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>') 963 964bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>') 965bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>') 966 967def substBitOps(code): 968 # first convert single-bit selectors to two-index form 969 # i.e., <n> --> <n:n> 970 code = bitOp1ArgRE.sub(r'<\1:\1>', code) 971 # simple case: selector applied to ID (name) 972 # i.e., foo<a:b> --> bits(foo, a, b) 973 code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code) 974 # if selector is applied to expression (ending in ')'), 975 # we need to search backward for matching '(' 976 match = bitOpExprRE.search(code) 977 while match: 978 exprEnd = match.start() 979 here = exprEnd - 1 980 nestLevel = 1 981 while nestLevel > 0: 982 if code[here] == '(': 983 nestLevel -= 1 984 elif code[here] == ')': 985 nestLevel += 1 986 here -= 1 987 if here < 0: 988 sys.exit("Didn't find '('!") 989 exprStart = here+1 990 newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1], 991 match.group(1), match.group(2)) 992 code = code[:exprStart] + newExpr + code[match.end():] 993 match = bitOpExprRE.search(code) 994 return code 995 996 997#################### 998# Template objects. 999# 1000# Template objects are format strings that allow substitution from 1001# the attribute spaces of other objects (e.g. InstObjParams instances). 1002 1003class Template: 1004 def __init__(self, t): 1005 self.template = t 1006 1007 def subst(self, d): 1008 # Start with the template namespace. Make a copy since we're 1009 # going to modify it. 1010 myDict = templateMap.copy() 1011 # if the argument is a dictionary, we just use it. 1012 if isinstance(d, dict): 1013 myDict.update(d) 1014 # if the argument is an object, we use its attribute map. 1015 elif hasattr(d, '__dict__'): 1016 myDict.update(d.__dict__) 1017 else: 1018 raise TypeError, "Template.subst() arg must be or have dictionary" 1019 # Protect non-Python-dict substitutions (e.g. if there's a printf 1020 # in the templated C++ code) 1021 template = protect_non_subst_percents(self.template) 1022 # CPU-model-specific substitutions are handled later (in GenCode). 1023 template = protect_cpu_symbols(template) 1024 return template % myDict 1025 1026 # Convert to string. This handles the case when a template with a 1027 # CPU-specific term gets interpolated into another template or into 1028 # an output block. 1029 def __str__(self): 1030 return expand_cpu_symbols_to_string(self.template) 1031 1032##################################################################### 1033# 1034# Code Parser 1035# 1036# The remaining code is the support for automatically extracting 1037# instruction characteristics from pseudocode. 1038# 1039##################################################################### 1040 1041# Force the argument to be a list. Useful for flags, where a caller 1042# can specify a singleton flag or a list of flags. Also usful for 1043# converting tuples to lists so they can be modified. 1044def makeList(arg): 1045 if isinstance(arg, list): 1046 return arg 1047 elif isinstance(arg, tuple): 1048 return list(arg) 1049 elif not arg: 1050 return [] 1051 else: 1052 return [ arg ] 1053 1054# Generate operandTypeMap from the user's 'def operand_types' 1055# statement. 1056def buildOperandTypeMap(userDict, lineno): 1057 global operandTypeMap 1058 operandTypeMap = {} 1059 for (ext, (desc, size)) in userDict.iteritems(): 1060 if desc == 'signed int': 1061 ctype = 'int%d_t' % size 1062 is_signed = 1 1063 elif desc == 'unsigned int': 1064 ctype = 'uint%d_t' % size 1065 is_signed = 0 1066 elif desc == 'float': 1067 is_signed = 1 # shouldn't really matter 1068 if size == 32: 1069 ctype = 'float' 1070 elif size == 64: 1071 ctype = 'double' 1072 if ctype == '': 1073 error(0, 'Unrecognized type description "%s" in userDict') 1074 operandTypeMap[ext] = (size, ctype, is_signed) 1075 1076# 1077# 1078# 1079# Base class for operand descriptors. An instance of this class (or 1080# actually a class derived from this one) represents a specific 1081# operand for a code block (e.g, "Rc.sq" as a dest). Intermediate 1082# derived classes encapsulates the traits of a particular operand type 1083# (e.g., "32-bit integer register"). 1084# 1085class Operand(object): 1086 def __init__(self, full_name, ext, is_src, is_dest): 1087 self.full_name = full_name 1088 self.ext = ext 1089 self.is_src = is_src 1090 self.is_dest = is_dest 1091 # The 'effective extension' (eff_ext) is either the actual 1092 # extension, if one was explicitly provided, or the default. 1093 if ext: 1094 self.eff_ext = ext 1095 else: 1096 self.eff_ext = self.dflt_ext 1097 1098 (self.size, self.ctype, self.is_signed) = operandTypeMap[self.eff_ext] 1099 1100 # note that mem_acc_size is undefined for non-mem operands... 1101 # template must be careful not to use it if it doesn't apply. 1102 if self.isMem(): 1103 self.mem_acc_size = self.makeAccSize() 1104 self.mem_acc_type = self.ctype 1105 1106 # Finalize additional fields (primarily code fields). This step 1107 # is done separately since some of these fields may depend on the 1108 # register index enumeration that hasn't been performed yet at the 1109 # time of __init__(). 1110 def finalize(self): 1111 self.flags = self.getFlags() 1112 self.constructor = self.makeConstructor() 1113 self.op_decl = self.makeDecl() 1114 1115 if self.is_src: 1116 self.op_rd = self.makeRead() 1117 self.op_src_decl = self.makeDecl() 1118 else: 1119 self.op_rd = '' 1120 self.op_src_decl = '' 1121 1122 if self.is_dest: 1123 self.op_wb = self.makeWrite() 1124 self.op_dest_decl = self.makeDecl() 1125 else: 1126 self.op_wb = '' 1127 self.op_dest_decl = '' 1128 1129 def isMem(self): 1130 return 0 1131 1132 def isReg(self): 1133 return 0 1134 1135 def isFloatReg(self): 1136 return 0 1137 1138 def isIntReg(self): 1139 return 0 1140 1141 def isControlReg(self): 1142 return 0 1143 1144 def getFlags(self): 1145 # note the empty slice '[:]' gives us a copy of self.flags[0] 1146 # instead of a reference to it 1147 my_flags = self.flags[0][:] 1148 if self.is_src: 1149 my_flags += self.flags[1] 1150 if self.is_dest: 1151 my_flags += self.flags[2] 1152 return my_flags 1153 1154 def makeDecl(self): 1155 # Note that initializations in the declarations are solely 1156 # to avoid 'uninitialized variable' errors from the compiler. 1157 return self.ctype + ' ' + self.base_name + ' = 0;\n'; 1158 1159class IntRegOperand(Operand): 1160 def isReg(self): 1161 return 1 1162 1163 def isIntReg(self): 1164 return 1 1165 1166 def makeConstructor(self): 1167 c = '' 1168 if self.is_src: 1169 c += '\n\t_srcRegIdx[%d] = %s;' % \ 1170 (self.src_reg_idx, self.reg_spec) 1171 if self.is_dest: 1172 c += '\n\t_destRegIdx[%d] = %s;' % \ 1173 (self.dest_reg_idx, self.reg_spec) 1174 return c 1175 1176 def makeRead(self): 1177 if (self.ctype == 'float' or self.ctype == 'double'): 1178 error(0, 'Attempt to read integer register as FP') 1179 if (self.size == self.dflt_size): 1180 return '%s = xc->readIntReg(this, %d);\n' % \ 1181 (self.base_name, self.src_reg_idx) 1182 else: 1183 return '%s = bits(xc->readIntReg(this, %d), %d, 0);\n' % \ 1184 (self.base_name, self.src_reg_idx, self.size-1) 1185 1186 def makeWrite(self): 1187 if (self.ctype == 'float' or self.ctype == 'double'): 1188 error(0, 'Attempt to write integer register as FP') 1189 if (self.size != self.dflt_size and self.is_signed): 1190 final_val = 'sext<%d>(%s)' % (self.size, self.base_name) 1191 else: 1192 final_val = self.base_name 1193 wb = ''' 1194 { 1195 %s final_val = %s; 1196 xc->setIntReg(this, %d, final_val);\n 1197 if (traceData) { traceData->setData(final_val); } 1198 }''' % (self.dflt_ctype, final_val, self.dest_reg_idx) 1199 return wb 1200 1201class FloatRegOperand(Operand): 1202 def isReg(self): 1203 return 1 1204 1205 def isFloatReg(self): 1206 return 1 1207 1208 def makeConstructor(self): 1209 c = '' 1210 if self.is_src: 1211 c += '\n\t_srcRegIdx[%d] = %s + FP_Base_DepTag;' % \ 1212 (self.src_reg_idx, self.reg_spec) 1213 if self.is_dest: 1214 c += '\n\t_destRegIdx[%d] = %s + FP_Base_DepTag;' % \ 1215 (self.dest_reg_idx, self.reg_spec) 1216 return c 1217 1218 def makeRead(self): 1219 bit_select = 0 1220 if (self.ctype == 'float'): 1221 func = 'readFloatRegSingle' 1222 elif (self.ctype == 'double'): 1223 func = 'readFloatRegDouble' 1224 else: 1225 func = 'readFloatRegInt' 1226 if (self.size != self.dflt_size): 1227 bit_select = 1 1228 base = 'xc->%s(this, %d)' % \ 1229 (func, self.src_reg_idx) 1230 if bit_select: 1231 return '%s = bits(%s, %d, 0);\n' % \ 1232 (self.base_name, base, self.size-1) 1233 else: 1234 return '%s = %s;\n' % (self.base_name, base) 1235 1236 def makeWrite(self): 1237 final_val = self.base_name 1238 final_ctype = self.ctype 1239 if (self.ctype == 'float'): 1240 func = 'setFloatRegSingle' 1241 elif (self.ctype == 'double'): 1242 func = 'setFloatRegDouble' 1243 else: 1244 func = 'setFloatRegInt' 1245 final_ctype = 'uint%d_t' % self.dflt_size 1246 if (self.size != self.dflt_size and self.is_signed): 1247 final_val = 'sext<%d>(%s)' % (self.size, self.base_name) 1248 wb = ''' 1249 { 1250 %s final_val = %s; 1251 xc->%s(this, %d, final_val);\n 1252 if (traceData) { traceData->setData(final_val); } 1253 }''' % (final_ctype, final_val, func, self.dest_reg_idx) 1254 return wb 1255 1256class ControlRegOperand(Operand): 1257 def isReg(self): 1258 return 1 1259 1260 def isControlReg(self): 1261 return 1 1262 1263 def makeConstructor(self): 1264 c = '' 1265 if self.is_src: 1266 c += '\n\t_srcRegIdx[%d] = %s_DepTag;' % \ 1267 (self.src_reg_idx, self.reg_spec) 1268 if self.is_dest: 1269 c += '\n\t_destRegIdx[%d] = %s_DepTag;' % \ 1270 (self.dest_reg_idx, self.reg_spec) 1271 return c 1272 1273 def makeRead(self): 1274 bit_select = 0 1275 if (self.ctype == 'float' or self.ctype == 'double'): 1276 error(0, 'Attempt to read control register as FP') 1277 base = 'xc->read%s()' % self.reg_spec 1278 if self.size == self.dflt_size: 1279 return '%s = %s;\n' % (self.base_name, base) 1280 else: 1281 return '%s = bits(%s, %d, 0);\n' % \ 1282 (self.base_name, base, self.size-1) 1283 1284 def makeWrite(self): 1285 if (self.ctype == 'float' or self.ctype == 'double'): 1286 error(0, 'Attempt to write control register as FP') 1287 wb = 'xc->set%s(%s);\n' % (self.reg_spec, self.base_name) 1288 wb += 'if (traceData) { traceData->setData(%s); }' % \ 1289 self.base_name 1290 return wb 1291 1292class MemOperand(Operand): 1293 def isMem(self): 1294 return 1 1295 1296 def makeConstructor(self): 1297 return '' 1298 1299 def makeDecl(self): 1300 # Note that initializations in the declarations are solely 1301 # to avoid 'uninitialized variable' errors from the compiler. 1302 # Declare memory data variable. 1303 c = '%s %s = 0;\n' % (self.ctype, self.base_name) 1304 return c 1305 1306 def makeRead(self): 1307 return '' 1308 1309 def makeWrite(self): 1310 return '' 1311 1312 # Return the memory access size *in bits*, suitable for 1313 # forming a type via "uint%d_t". Divide by 8 if you want bytes. 1314 def makeAccSize(self): 1315 return self.size 1316 1317 1318class NPCOperand(Operand): 1319 def makeConstructor(self): 1320 return '' 1321 1322 def makeRead(self): 1323 return '%s = xc->readPC() + 4;\n' % self.base_name 1324 1325 def makeWrite(self): 1326 return 'xc->setNextPC(%s);\n' % self.base_name 1327 1328class NNPCOperand(Operand): 1329 def makeConstructor(self): 1330 return '' 1331 1332 def makeRead(self): 1333 return '%s = xc->readPC() + 8;\n' % self.base_name 1334 1335 def makeWrite(self): 1336 return 'xc->setNextNPC(%s);\n' % self.base_name 1337 1338def buildOperandNameMap(userDict, lineno): 1339 global operandNameMap 1340 operandNameMap = {} 1341 for (op_name, val) in userDict.iteritems(): 1342 (base_cls_name, dflt_ext, reg_spec, flags, sort_pri) = val 1343 (dflt_size, dflt_ctype, dflt_is_signed) = operandTypeMap[dflt_ext] 1344 # Canonical flag structure is a triple of lists, where each list 1345 # indicates the set of flags implied by this operand always, when 1346 # used as a source, and when used as a dest, respectively. 1347 # For simplicity this can be initialized using a variety of fairly 1348 # obvious shortcuts; we convert these to canonical form here. 1349 if not flags: 1350 # no flags specified (e.g., 'None') 1351 flags = ( [], [], [] ) 1352 elif isinstance(flags, str): 1353 # a single flag: assumed to be unconditional 1354 flags = ( [ flags ], [], [] ) 1355 elif isinstance(flags, list): 1356 # a list of flags: also assumed to be unconditional 1357 flags = ( flags, [], [] ) 1358 elif isinstance(flags, tuple): 1359 # it's a tuple: it should be a triple, 1360 # but each item could be a single string or a list 1361 (uncond_flags, src_flags, dest_flags) = flags 1362 flags = (makeList(uncond_flags), 1363 makeList(src_flags), makeList(dest_flags)) 1364 # Accumulate attributes of new operand class in tmp_dict 1365 tmp_dict = {} 1366 for attr in ('dflt_ext', 'reg_spec', 'flags', 'sort_pri', 1367 'dflt_size', 'dflt_ctype', 'dflt_is_signed'): 1368 tmp_dict[attr] = eval(attr) 1369 tmp_dict['base_name'] = op_name 1370 # New class name will be e.g. "IntReg_Ra" 1371 cls_name = base_cls_name + '_' + op_name 1372 # Evaluate string arg to get class object. Note that the 1373 # actual base class for "IntReg" is "IntRegOperand", i.e. we 1374 # have to append "Operand". 1375 try: 1376 base_cls = eval(base_cls_name + 'Operand') 1377 except NameError: 1378 error(lineno, 1379 'error: unknown operand base class "%s"' % base_cls_name) 1380 # The following statement creates a new class called 1381 # <cls_name> as a subclass of <base_cls> with the attributes 1382 # in tmp_dict, just as if we evaluated a class declaration. 1383 operandNameMap[op_name] = type(cls_name, (base_cls,), tmp_dict) 1384 1385 # Define operand variables. 1386 operands = userDict.keys() 1387 1388 operandsREString = (r''' 1389 (?<![\w\.]) # neg. lookbehind assertion: prevent partial matches 1390 ((%s)(?:\.(\w+))?) # match: operand with optional '.' then suffix 1391 (?![\w\.]) # neg. lookahead assertion: prevent partial matches 1392 ''' 1393 % string.join(operands, '|')) 1394 1395 global operandsRE 1396 operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE) 1397 1398 # Same as operandsREString, but extension is mandatory, and only two 1399 # groups are returned (base and ext, not full name as above). 1400 # Used for subtituting '_' for '.' to make C++ identifiers. 1401 operandsWithExtREString = (r'(?<![\w\.])(%s)\.(\w+)(?![\w\.])' 1402 % string.join(operands, '|')) 1403 1404 global operandsWithExtRE 1405 operandsWithExtRE = re.compile(operandsWithExtREString, re.MULTILINE) 1406 1407 1408class OperandList: 1409 1410 # Find all the operands in the given code block. Returns an operand 1411 # descriptor list (instance of class OperandList). 1412 def __init__(self, code): 1413 self.items = [] 1414 self.bases = {} 1415 # delete comments so we don't match on reg specifiers inside 1416 code = commentRE.sub('', code) 1417 # search for operands 1418 next_pos = 0 1419 while 1: 1420 match = operandsRE.search(code, next_pos) 1421 if not match: 1422 # no more matches: we're done 1423 break 1424 op = match.groups() 1425 # regexp groups are operand full name, base, and extension 1426 (op_full, op_base, op_ext) = op 1427 # if the token following the operand is an assignment, this is 1428 # a destination (LHS), else it's a source (RHS) 1429 is_dest = (assignRE.match(code, match.end()) != None) 1430 is_src = not is_dest 1431 # see if we've already seen this one 1432 op_desc = self.find_base(op_base) 1433 if op_desc: 1434 if op_desc.ext != op_ext: 1435 error(0, 'Inconsistent extensions for operand %s' % \ 1436 op_base) 1437 op_desc.is_src = op_desc.is_src or is_src 1438 op_desc.is_dest = op_desc.is_dest or is_dest 1439 else: 1440 # new operand: create new descriptor 1441 op_desc = operandNameMap[op_base](op_full, op_ext, 1442 is_src, is_dest) 1443 self.append(op_desc) 1444 # start next search after end of current match 1445 next_pos = match.end() 1446 self.sort() 1447 # enumerate source & dest register operands... used in building 1448 # constructor later 1449 self.numSrcRegs = 0 1450 self.numDestRegs = 0 1451 self.numFPDestRegs = 0 1452 self.numIntDestRegs = 0 1453 self.memOperand = None 1454 for op_desc in self.items: 1455 if op_desc.isReg(): 1456 if op_desc.is_src: 1457 op_desc.src_reg_idx = self.numSrcRegs 1458 self.numSrcRegs += 1 1459 if op_desc.is_dest: 1460 op_desc.dest_reg_idx = self.numDestRegs 1461 self.numDestRegs += 1 1462 if op_desc.isFloatReg(): 1463 self.numFPDestRegs += 1 1464 elif op_desc.isIntReg(): 1465 self.numIntDestRegs += 1 1466 elif op_desc.isMem(): 1467 if self.memOperand: 1468 error(0, "Code block has more than one memory operand.") 1469 self.memOperand = op_desc 1470 # now make a final pass to finalize op_desc fields that may depend 1471 # on the register enumeration 1472 for op_desc in self.items: 1473 op_desc.finalize() 1474 1475 def __len__(self): 1476 return len(self.items) 1477 1478 def __getitem__(self, index): 1479 return self.items[index] 1480 1481 def append(self, op_desc): 1482 self.items.append(op_desc) 1483 self.bases[op_desc.base_name] = op_desc 1484 1485 def find_base(self, base_name): 1486 # like self.bases[base_name], but returns None if not found 1487 # (rather than raising exception) 1488 return self.bases.get(base_name) 1489 1490 # internal helper function for concat[Some]Attr{Strings|Lists} 1491 def __internalConcatAttrs(self, attr_name, filter, result): 1492 for op_desc in self.items: 1493 if filter(op_desc): 1494 result += getattr(op_desc, attr_name) 1495 return result 1496 1497 # return a single string that is the concatenation of the (string) 1498 # values of the specified attribute for all operands 1499 def concatAttrStrings(self, attr_name): 1500 return self.__internalConcatAttrs(attr_name, lambda x: 1, '') 1501 1502 # like concatAttrStrings, but only include the values for the operands 1503 # for which the provided filter function returns true 1504 def concatSomeAttrStrings(self, filter, attr_name): 1505 return self.__internalConcatAttrs(attr_name, filter, '') 1506 1507 # return a single list that is the concatenation of the (list) 1508 # values of the specified attribute for all operands 1509 def concatAttrLists(self, attr_name): 1510 return self.__internalConcatAttrs(attr_name, lambda x: 1, []) 1511 1512 # like concatAttrLists, but only include the values for the operands 1513 # for which the provided filter function returns true 1514 def concatSomeAttrLists(self, filter, attr_name): 1515 return self.__internalConcatAttrs(attr_name, filter, []) 1516 1517 def sort(self): 1518 self.items.sort(lambda a, b: a.sort_pri - b.sort_pri) 1519 1520# Regular expression object to match C++ comments 1521# (used in findOperands()) 1522commentRE = re.compile(r'//.*\n') 1523 1524# Regular expression object to match assignment statements 1525# (used in findOperands()) 1526assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE) 1527 1528# Munge operand names in code string to make legal C++ variable names. 1529# This means getting rid of the type extension if any. 1530# (Will match base_name attribute of Operand object.) 1531def substMungedOpNames(code): 1532 return operandsWithExtRE.sub(r'\1', code) 1533 1534def joinLists(t): 1535 return map(string.join, t) 1536 1537def makeFlagConstructor(flag_list): 1538 if len(flag_list) == 0: 1539 return '' 1540 # filter out repeated flags 1541 flag_list.sort() 1542 i = 1 1543 while i < len(flag_list): 1544 if flag_list[i] == flag_list[i-1]: 1545 del flag_list[i] 1546 else: 1547 i += 1 1548 pre = '\n\tflags[' 1549 post = '] = true;' 1550 code = pre + string.join(flag_list, post + pre) + post 1551 return code 1552 1553class CodeBlock: 1554 def __init__(self, code): 1555 self.orig_code = code 1556 self.operands = OperandList(code) 1557 self.code = substMungedOpNames(substBitOps(code)) 1558 self.constructor = self.operands.concatAttrStrings('constructor') 1559 self.constructor += \ 1560 '\n\t_numSrcRegs = %d;' % self.operands.numSrcRegs 1561 self.constructor += \ 1562 '\n\t_numDestRegs = %d;' % self.operands.numDestRegs 1563 self.constructor += \ 1564 '\n\t_numFPDestRegs = %d;' % self.operands.numFPDestRegs 1565 self.constructor += \ 1566 '\n\t_numIntDestRegs = %d;' % self.operands.numIntDestRegs 1567 1568 self.op_decl = self.operands.concatAttrStrings('op_decl') 1569 1570 is_src = lambda op: op.is_src 1571 is_dest = lambda op: op.is_dest 1572 1573 self.op_src_decl = \ 1574 self.operands.concatSomeAttrStrings(is_src, 'op_src_decl') 1575 self.op_dest_decl = \ 1576 self.operands.concatSomeAttrStrings(is_dest, 'op_dest_decl') 1577 1578 self.op_rd = self.operands.concatAttrStrings('op_rd') 1579 self.op_wb = self.operands.concatAttrStrings('op_wb') 1580 1581 self.flags = self.operands.concatAttrLists('flags') 1582 1583 if self.operands.memOperand: 1584 self.mem_acc_size = self.operands.memOperand.mem_acc_size 1585 self.mem_acc_type = self.operands.memOperand.mem_acc_type 1586 1587 # Make a basic guess on the operand class (function unit type). 1588 # These are good enough for most cases, and will be overridden 1589 # later otherwise. 1590 if 'IsStore' in self.flags: 1591 self.op_class = 'MemWriteOp' 1592 elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags: 1593 self.op_class = 'MemReadOp' 1594 elif 'IsFloating' in self.flags: 1595 self.op_class = 'FloatAddOp' 1596 else: 1597 self.op_class = 'IntAluOp' 1598 1599# Assume all instruction flags are of the form 'IsFoo' 1600instFlagRE = re.compile(r'Is.*') 1601 1602# OpClass constants end in 'Op' except No_OpClass 1603opClassRE = re.compile(r'.*Op|No_OpClass') 1604 1605class InstObjParams: 1606 def __init__(self, mnem, class_name, base_class = '', 1607 code_block = None, opt_args = []): 1608 self.mnemonic = mnem 1609 self.class_name = class_name 1610 self.base_class = base_class 1611 if code_block: 1612 for code_attr in code_block.__dict__.keys(): 1613 setattr(self, code_attr, getattr(code_block, code_attr)) 1614 else: 1615 self.constructor = '' 1616 self.flags = [] 1617 # Optional arguments are assumed to be either StaticInst flags 1618 # or an OpClass value. To avoid having to import a complete 1619 # list of these values to match against, we do it ad-hoc 1620 # with regexps. 1621 for oa in opt_args: 1622 if instFlagRE.match(oa): 1623 self.flags.append(oa) 1624 elif opClassRE.match(oa): 1625 self.op_class = oa 1626 else: 1627 error(0, 'InstObjParams: optional arg "%s" not recognized ' 1628 'as StaticInst::Flag or OpClass.' % oa) 1629 1630 # add flag initialization to contructor here to include 1631 # any flags added via opt_args 1632 self.constructor += makeFlagConstructor(self.flags) 1633 1634 # if 'IsFloating' is set, add call to the FP enable check 1635 # function (which should be provided by isa_desc via a declare) 1636 if 'IsFloating' in self.flags: 1637 self.fp_enable_check = 'fault = checkFpEnableFault(xc);' 1638 else: 1639 self.fp_enable_check = '' 1640 1641####################### 1642# 1643# Output file template 1644# 1645 1646file_template = ''' 1647/* 1648 * DO NOT EDIT THIS FILE!!! 1649 * 1650 * It was automatically generated from the ISA description in %(filename)s 1651 */ 1652 1653%(includes)s 1654 1655%(global_output)s 1656 1657namespace %(namespace)s { 1658 1659%(namespace_output)s 1660 1661} // namespace %(namespace)s 1662 1663%(decode_function)s 1664''' 1665 1666 1667# Update the output file only if the new contents are different from 1668# the current contents. Minimizes the files that need to be rebuilt 1669# after minor changes. 1670def update_if_needed(file, contents): 1671 update = False 1672 if os.access(file, os.R_OK): 1673 f = open(file, 'r') 1674 old_contents = f.read() 1675 f.close() 1676 if contents != old_contents: 1677 print 'Updating', file 1678 os.remove(file) # in case it's write-protected 1679 update = True 1680 else: 1681 print 'File', file, 'is unchanged' 1682 else: 1683 print 'Generating', file 1684 update = True 1685 if update: 1686 f = open(file, 'w') 1687 f.write(contents) 1688 f.close() 1689 1690# This regular expression matches include directives 1691includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[\w/.-]*)".*$', 1692 re.MULTILINE) 1693 1694def preprocess_isa_desc(isa_desc): 1695 # Find any includes and include them 1696 pos = 0 1697 while 1: 1698 m = includeRE.search(isa_desc, pos) 1699 if not m: 1700 break 1701 filename = m.group('filename') 1702 print 'Including file "%s"' % filename 1703 try: 1704 isa_desc = isa_desc[:m.start()] + \ 1705 '##newfile "' + filename + '"\n' + \ 1706 open(filename).read() + \ 1707 '##endfile\n' + \ 1708 isa_desc[m.end():] 1709 except IOError: 1710 error(0, 'Error including file "%s"' % (filename)) 1711 pos = m.start() 1712 return isa_desc 1713 1714# 1715# Read in and parse the ISA description. 1716# 1717def parse_isa_desc(isa_desc_file, output_dir): 1718 # set a global var for the input filename... used in error messages 1719 global input_filename 1720 input_filename = isa_desc_file 1721 global fileNameStack 1722 fileNameStack = [(input_filename, 1)] 1723 1724 # Suck the ISA description file in. 1725 input = open(isa_desc_file) 1726 isa_desc = input.read() 1727 input.close() 1728 1729 # Perform Preprocessing 1730 isa_desc = preprocess_isa_desc(isa_desc) 1731 1732 # Parse it. 1733 (isa_name, namespace, global_code, namespace_code) = yacc.parse(isa_desc) 1734 1735 # grab the last three path components of isa_desc_file to put in 1736 # the output 1737 filename = '/'.join(isa_desc_file.split('/')[-3:]) 1738 1739 # generate decoder.hh 1740 includes = '#include "base/bitfield.hh" // for bitfield support' 1741 global_output = global_code.header_output 1742 namespace_output = namespace_code.header_output 1743 decode_function = '' 1744 update_if_needed(output_dir + '/decoder.hh', file_template % vars()) 1745 1746 # generate decoder.cc 1747 includes = '#include "decoder.hh"' 1748 global_output = global_code.decoder_output 1749 namespace_output = namespace_code.decoder_output 1750 # namespace_output += namespace_code.decode_block 1751 decode_function = namespace_code.decode_block 1752 update_if_needed(output_dir + '/decoder.cc', file_template % vars()) 1753 1754 # generate per-cpu exec files 1755 for cpu in cpu_models: 1756 includes = '#include "decoder.hh"\n' 1757 includes += cpu.includes 1758 global_output = global_code.exec_output[cpu.name] 1759 namespace_output = namespace_code.exec_output[cpu.name] 1760 decode_function = '' 1761 update_if_needed(output_dir + '/' + cpu.filename, 1762 file_template % vars()) 1763 1764# global list of CpuModel objects (see cpu_models.py) 1765cpu_models = [] 1766 1767# Called as script: get args from command line. 1768# Args are: <path to cpu_models.py> <isa desc file> <output dir> <cpu models> 1769if __name__ == '__main__': 1770 execfile(sys.argv[1]) # read in CpuModel definitions 1771 cpu_models = [CpuModel.dict[cpu] for cpu in sys.argv[4:]] 1772 parse_isa_desc(sys.argv[2], sys.argv[3]) 1773