isa_parser.py revision 4273:449c6e09f6ca
1# Copyright (c) 2003-2005 The Regents of The University of Michigan 2# All rights reserved. 3# 4# Redistribution and use in source and binary forms, with or without 5# modification, are permitted provided that the following conditions are 6# met: redistributions of source code must retain the above copyright 7# notice, this list of conditions and the following disclaimer; 8# redistributions in binary form must reproduce the above copyright 9# notice, this list of conditions and the following disclaimer in the 10# documentation and/or other materials provided with the distribution; 11# neither the name of the copyright holders nor the names of its 12# contributors may be used to endorse or promote products derived from 13# this software without specific prior written permission. 14# 15# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 16# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 17# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 18# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 19# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 20# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 21# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 25# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26# 27# Authors: Steve Reinhardt 28# Korey Sewell 29 30import os 31import sys 32import re 33import string 34import traceback 35# get type names 36from types import * 37 38# Prepend the directory where the PLY lex & yacc modules are found 39# to the search path. Assumes we're compiling in a subdirectory 40# of 'build' in the current tree. 41sys.path[0:0] = [os.environ['M5_PLY']] 42 43import lex 44import yacc 45 46##################################################################### 47# 48# Lexer 49# 50# The PLY lexer module takes two things as input: 51# - A list of token names (the string list 'tokens') 52# - A regular expression describing a match for each token. The 53# regexp for token FOO can be provided in two ways: 54# - as a string variable named t_FOO 55# - as the doc string for a function named t_FOO. In this case, 56# the function is also executed, allowing an action to be 57# associated with each token match. 58# 59##################################################################### 60 61# Reserved words. These are listed separately as they are matched 62# using the same regexp as generic IDs, but distinguished in the 63# t_ID() function. The PLY documentation suggests this approach. 64reserved = ( 65 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT', 66 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS', 67 'OUTPUT', 'SIGNED', 'TEMPLATE' 68 ) 69 70# List of tokens. The lex module requires this. 71tokens = reserved + ( 72 # identifier 73 'ID', 74 75 # integer literal 76 'INTLIT', 77 78 # string literal 79 'STRLIT', 80 81 # code literal 82 'CODELIT', 83 84 # ( ) [ ] { } < > , ; : :: * 85 'LPAREN', 'RPAREN', 86 'LBRACKET', 'RBRACKET', 87 'LBRACE', 'RBRACE', 88 'LESS', 'GREATER', 'EQUALS', 89 'COMMA', 'SEMI', 'COLON', 'DBLCOLON', 90 'ASTERISK', 91 92 # C preprocessor directives 93 'CPPDIRECTIVE' 94 95# The following are matched but never returned. commented out to 96# suppress PLY warning 97 # newfile directive 98# 'NEWFILE', 99 100 # endfile directive 101# 'ENDFILE' 102) 103 104# Regular expressions for token matching 105t_LPAREN = r'\(' 106t_RPAREN = r'\)' 107t_LBRACKET = r'\[' 108t_RBRACKET = r'\]' 109t_LBRACE = r'\{' 110t_RBRACE = r'\}' 111t_LESS = r'\<' 112t_GREATER = r'\>' 113t_EQUALS = r'=' 114t_COMMA = r',' 115t_SEMI = r';' 116t_COLON = r':' 117t_DBLCOLON = r'::' 118t_ASTERISK = r'\*' 119 120# Identifiers and reserved words 121reserved_map = { } 122for r in reserved: 123 reserved_map[r.lower()] = r 124 125def t_ID(t): 126 r'[A-Za-z_]\w*' 127 t.type = reserved_map.get(t.value,'ID') 128 return t 129 130# Integer literal 131def t_INTLIT(t): 132 r'(0x[\da-fA-F]+)|\d+' 133 try: 134 t.value = int(t.value,0) 135 except ValueError: 136 error(t.lineno, 'Integer value "%s" too large' % t.value) 137 t.value = 0 138 return t 139 140# String literal. Note that these use only single quotes, and 141# can span multiple lines. 142def t_STRLIT(t): 143 r"(?m)'([^'])+'" 144 # strip off quotes 145 t.value = t.value[1:-1] 146 t.lineno += t.value.count('\n') 147 return t 148 149 150# "Code literal"... like a string literal, but delimiters are 151# '{{' and '}}' so they get formatted nicely under emacs c-mode 152def t_CODELIT(t): 153 r"(?m)\{\{([^\}]|}(?!\}))+\}\}" 154 # strip off {{ & }} 155 t.value = t.value[2:-2] 156 t.lineno += t.value.count('\n') 157 return t 158 159def t_CPPDIRECTIVE(t): 160 r'^\#[^\#].*\n' 161 t.lineno += t.value.count('\n') 162 return t 163 164def t_NEWFILE(t): 165 r'^\#\#newfile\s+"[\w/.-]*"' 166 fileNameStack.push((t.value[11:-1], t.lineno)) 167 t.lineno = 0 168 169def t_ENDFILE(t): 170 r'^\#\#endfile' 171 (old_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::ExtMachInst 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) 365 366# alternate form for structure member: 'def bitfield <ID> <ID>' 367def p_def_bitfield_2(t): 368 'def_bitfield : DEF nothing BITFIELD ID ID SEMI' 369 expr = 'machInst.%s' % t[5] 370 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 371 t[0] = GenCode(header_output = hash_define) 372 373def p_nothing(t): 374 'nothing : empty' 375 t[0] = '' 376 377def p_opt_signed_0(t): 378 'opt_signed : SIGNED' 379 t[0] = t[1] 380 381def p_opt_signed_1(t): 382 'opt_signed : empty' 383 t[0] = '' 384 385# Global map variable to hold templates 386templateMap = {} 387 388def p_def_template(t): 389 'def_template : DEF TEMPLATE ID CODELIT SEMI' 390 templateMap[t[3]] = Template(t[4]) 391 t[0] = GenCode() 392 393# An instruction format definition looks like 394# "def format <fmt>(<params>) {{...}};" 395def p_def_format(t): 396 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI' 397 (id, params, code) = (t[3], t[5], t[7]) 398 defFormat(id, params, code, t.lineno(1)) 399 t[0] = GenCode() 400 401# The formal parameter list for an instruction format is a possibly 402# empty list of comma-separated parameters. Positional (standard, 403# non-keyword) parameters must come first, followed by keyword 404# parameters, followed by a '*foo' parameter that gets excess 405# positional arguments (as in Python). Each of these three parameter 406# categories is optional. 407# 408# Note that we do not support the '**foo' parameter for collecting 409# otherwise undefined keyword args. Otherwise the parameter list is 410# (I believe) identical to what is supported in Python. 411# 412# The param list generates a tuple, where the first element is a list of 413# the positional params and the second element is a dict containing the 414# keyword params. 415def p_param_list_0(t): 416 'param_list : positional_param_list COMMA nonpositional_param_list' 417 t[0] = t[1] + t[3] 418 419def p_param_list_1(t): 420 '''param_list : positional_param_list 421 | nonpositional_param_list''' 422 t[0] = t[1] 423 424def p_positional_param_list_0(t): 425 'positional_param_list : empty' 426 t[0] = [] 427 428def p_positional_param_list_1(t): 429 'positional_param_list : ID' 430 t[0] = [t[1]] 431 432def p_positional_param_list_2(t): 433 'positional_param_list : positional_param_list COMMA ID' 434 t[0] = t[1] + [t[3]] 435 436def p_nonpositional_param_list_0(t): 437 'nonpositional_param_list : keyword_param_list COMMA excess_args_param' 438 t[0] = t[1] + t[3] 439 440def p_nonpositional_param_list_1(t): 441 '''nonpositional_param_list : keyword_param_list 442 | excess_args_param''' 443 t[0] = t[1] 444 445def p_keyword_param_list_0(t): 446 'keyword_param_list : keyword_param' 447 t[0] = [t[1]] 448 449def p_keyword_param_list_1(t): 450 'keyword_param_list : keyword_param_list COMMA keyword_param' 451 t[0] = t[1] + [t[3]] 452 453def p_keyword_param(t): 454 'keyword_param : ID EQUALS expr' 455 t[0] = t[1] + ' = ' + t[3].__repr__() 456 457def p_excess_args_param(t): 458 'excess_args_param : ASTERISK ID' 459 # Just concatenate them: '*ID'. Wrap in list to be consistent 460 # with positional_param_list and keyword_param_list. 461 t[0] = [t[1] + t[2]] 462 463# End of format definition-related rules. 464############## 465 466# 467# A decode block looks like: 468# decode <field1> [, <field2>]* [default <inst>] { ... } 469# 470def p_decode_block(t): 471 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE' 472 default_defaults = defaultStack.pop() 473 codeObj = t[5] 474 # use the "default defaults" only if there was no explicit 475 # default statement in decode_stmt_list 476 if not codeObj.has_decode_default: 477 codeObj += default_defaults 478 codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n') 479 t[0] = codeObj 480 481# The opt_default statement serves only to push the "default defaults" 482# onto defaultStack. This value will be used by nested decode blocks, 483# and used and popped off when the current decode_block is processed 484# (in p_decode_block() above). 485def p_opt_default_0(t): 486 'opt_default : empty' 487 # no default specified: reuse the one currently at the top of the stack 488 defaultStack.push(defaultStack.top()) 489 # no meaningful value returned 490 t[0] = None 491 492def p_opt_default_1(t): 493 'opt_default : DEFAULT inst' 494 # push the new default 495 codeObj = t[2] 496 codeObj.wrap_decode_block('\ndefault:\n', 'break;\n') 497 defaultStack.push(codeObj) 498 # no meaningful value returned 499 t[0] = None 500 501def p_decode_stmt_list_0(t): 502 'decode_stmt_list : decode_stmt' 503 t[0] = t[1] 504 505def p_decode_stmt_list_1(t): 506 'decode_stmt_list : decode_stmt decode_stmt_list' 507 if (t[1].has_decode_default and t[2].has_decode_default): 508 error(t.lineno(1), 'Two default cases in decode block') 509 t[0] = t[1] + t[2] 510 511# 512# Decode statement rules 513# 514# There are four types of statements allowed in a decode block: 515# 1. Format blocks 'format <foo> { ... }' 516# 2. Nested decode blocks 517# 3. Instruction definitions. 518# 4. C preprocessor directives. 519 520 521# Preprocessor directives found in a decode statement list are passed 522# through to the output, replicated to all of the output code 523# streams. This works well for ifdefs, so we can ifdef out both the 524# declarations and the decode cases generated by an instruction 525# definition. Handling them as part of the grammar makes it easy to 526# keep them in the right place with respect to the code generated by 527# the other statements. 528def p_decode_stmt_cpp(t): 529 'decode_stmt : CPPDIRECTIVE' 530 t[0] = GenCode(t[1], t[1], t[1], t[1]) 531 532# A format block 'format <foo> { ... }' sets the default instruction 533# format used to handle instruction definitions inside the block. 534# This format can be overridden by using an explicit format on the 535# instruction definition or with a nested format block. 536def p_decode_stmt_format(t): 537 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE' 538 # The format will be pushed on the stack when 'push_format_id' is 539 # processed (see below). Once the parser has recognized the full 540 # production (though the right brace), we're done with the format, 541 # so now we can pop it. 542 formatStack.pop() 543 t[0] = t[4] 544 545# This rule exists so we can set the current format (& push the stack) 546# when we recognize the format name part of the format block. 547def p_push_format_id(t): 548 'push_format_id : ID' 549 try: 550 formatStack.push(formatMap[t[1]]) 551 t[0] = ('', '// format %s' % t[1]) 552 except KeyError: 553 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 554 555# Nested decode block: if the value of the current field matches the 556# specified constant, do a nested decode on some other field. 557def p_decode_stmt_decode(t): 558 'decode_stmt : case_label COLON decode_block' 559 label = t[1] 560 codeObj = t[3] 561 # just wrap the decoding code from the block as a case in the 562 # outer switch statement. 563 codeObj.wrap_decode_block('\n%s:\n' % label) 564 codeObj.has_decode_default = (label == 'default') 565 t[0] = codeObj 566 567# Instruction definition (finally!). 568def p_decode_stmt_inst(t): 569 'decode_stmt : case_label COLON inst SEMI' 570 label = t[1] 571 codeObj = t[3] 572 codeObj.wrap_decode_block('\n%s:' % label, 'break;\n') 573 codeObj.has_decode_default = (label == 'default') 574 t[0] = codeObj 575 576# The case label is either a list of one or more constants or 'default' 577def p_case_label_0(t): 578 'case_label : intlit_list' 579 t[0] = ': '.join(map(lambda a: 'case %#x' % a, t[1])) 580 581def p_case_label_1(t): 582 'case_label : DEFAULT' 583 t[0] = 'default' 584 585# 586# The constant list for a decode case label must be non-empty, but may have 587# one or more comma-separated integer literals in it. 588# 589def p_intlit_list_0(t): 590 'intlit_list : INTLIT' 591 t[0] = [t[1]] 592 593def p_intlit_list_1(t): 594 'intlit_list : intlit_list COMMA INTLIT' 595 t[0] = t[1] 596 t[0].append(t[3]) 597 598# Define an instruction using the current instruction format (specified 599# by an enclosing format block). 600# "<mnemonic>(<args>)" 601def p_inst_0(t): 602 'inst : ID LPAREN arg_list RPAREN' 603 # Pass the ID and arg list to the current format class to deal with. 604 currentFormat = formatStack.top() 605 codeObj = currentFormat.defineInst(t[1], t[3], t.lineno(1)) 606 args = ','.join(map(str, t[3])) 607 args = re.sub('(?m)^', '//', args) 608 args = re.sub('^//', '', args) 609 comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args) 610 codeObj.prepend_all(comment) 611 t[0] = codeObj 612 613# Define an instruction using an explicitly specified format: 614# "<fmt>::<mnemonic>(<args>)" 615def p_inst_1(t): 616 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN' 617 try: 618 format = formatMap[t[1]] 619 except KeyError: 620 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 621 codeObj = format.defineInst(t[3], t[5], t.lineno(1)) 622 comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5]) 623 codeObj.prepend_all(comment) 624 t[0] = codeObj 625 626# The arg list generates a tuple, where the first element is a list of 627# the positional args and the second element is a dict containing the 628# keyword args. 629def p_arg_list_0(t): 630 'arg_list : positional_arg_list COMMA keyword_arg_list' 631 t[0] = ( t[1], t[3] ) 632 633def p_arg_list_1(t): 634 'arg_list : positional_arg_list' 635 t[0] = ( t[1], {} ) 636 637def p_arg_list_2(t): 638 'arg_list : keyword_arg_list' 639 t[0] = ( [], t[1] ) 640 641def p_positional_arg_list_0(t): 642 'positional_arg_list : empty' 643 t[0] = [] 644 645def p_positional_arg_list_1(t): 646 'positional_arg_list : expr' 647 t[0] = [t[1]] 648 649def p_positional_arg_list_2(t): 650 'positional_arg_list : positional_arg_list COMMA expr' 651 t[0] = t[1] + [t[3]] 652 653def p_keyword_arg_list_0(t): 654 'keyword_arg_list : keyword_arg' 655 t[0] = t[1] 656 657def p_keyword_arg_list_1(t): 658 'keyword_arg_list : keyword_arg_list COMMA keyword_arg' 659 t[0] = t[1] 660 t[0].update(t[3]) 661 662def p_keyword_arg(t): 663 'keyword_arg : ID EQUALS expr' 664 t[0] = { t[1] : t[3] } 665 666# 667# Basic expressions. These constitute the argument values of 668# "function calls" (i.e. instruction definitions in the decode block) 669# and default values for formal parameters of format functions. 670# 671# Right now, these are either strings, integers, or (recursively) 672# lists of exprs (using Python square-bracket list syntax). Note that 673# bare identifiers are trated as string constants here (since there 674# isn't really a variable namespace to refer to). 675# 676def p_expr_0(t): 677 '''expr : ID 678 | INTLIT 679 | STRLIT 680 | CODELIT''' 681 t[0] = t[1] 682 683def p_expr_1(t): 684 '''expr : LBRACKET list_expr RBRACKET''' 685 t[0] = t[2] 686 687def p_list_expr_0(t): 688 'list_expr : expr' 689 t[0] = [t[1]] 690 691def p_list_expr_1(t): 692 'list_expr : list_expr COMMA expr' 693 t[0] = t[1] + [t[3]] 694 695def p_list_expr_2(t): 696 'list_expr : empty' 697 t[0] = [] 698 699# 700# Empty production... use in other rules for readability. 701# 702def p_empty(t): 703 'empty :' 704 pass 705 706# Parse error handler. Note that the argument here is the offending 707# *token*, not a grammar symbol (hence the need to use t.value) 708def p_error(t): 709 if t: 710 error(t.lineno, "syntax error at '%s'" % t.value) 711 else: 712 error(0, "unknown syntax error", True) 713 714# END OF GRAMMAR RULES 715# 716# Now build the parser. 717yacc.yacc() 718 719 720##################################################################### 721# 722# Support Classes 723# 724##################################################################### 725 726# Expand template with CPU-specific references into a dictionary with 727# an entry for each CPU model name. The entry key is the model name 728# and the corresponding value is the template with the CPU-specific 729# refs substituted for that model. 730def expand_cpu_symbols_to_dict(template): 731 # Protect '%'s that don't go with CPU-specific terms 732 t = re.sub(r'%(?!\(CPU_)', '%%', template) 733 result = {} 734 for cpu in cpu_models: 735 result[cpu.name] = t % cpu.strings 736 return result 737 738# *If* the template has CPU-specific references, return a single 739# string containing a copy of the template for each CPU model with the 740# corresponding values substituted in. If the template has no 741# CPU-specific references, it is returned unmodified. 742def expand_cpu_symbols_to_string(template): 743 if template.find('%(CPU_') != -1: 744 return reduce(lambda x,y: x+y, 745 expand_cpu_symbols_to_dict(template).values()) 746 else: 747 return template 748 749# Protect CPU-specific references by doubling the corresponding '%'s 750# (in preparation for substituting a different set of references into 751# the template). 752def protect_cpu_symbols(template): 753 return re.sub(r'%(?=\(CPU_)', '%%', template) 754 755############### 756# GenCode class 757# 758# The GenCode class encapsulates generated code destined for various 759# output files. The header_output and decoder_output attributes are 760# strings containing code destined for decoder.hh and decoder.cc 761# respectively. The decode_block attribute contains code to be 762# incorporated in the decode function itself (that will also end up in 763# decoder.cc). The exec_output attribute is a dictionary with a key 764# for each CPU model name; the value associated with a particular key 765# is the string of code for that CPU model's exec.cc file. The 766# has_decode_default attribute is used in the decode block to allow 767# explicit default clauses to override default default clauses. 768 769class GenCode: 770 # Constructor. At this point we substitute out all CPU-specific 771 # symbols. For the exec output, these go into the per-model 772 # dictionary. For all other output types they get collapsed into 773 # a single string. 774 def __init__(self, 775 header_output = '', decoder_output = '', exec_output = '', 776 decode_block = '', has_decode_default = False): 777 self.header_output = expand_cpu_symbols_to_string(header_output) 778 self.decoder_output = expand_cpu_symbols_to_string(decoder_output) 779 if isinstance(exec_output, dict): 780 self.exec_output = exec_output 781 elif isinstance(exec_output, str): 782 # If the exec_output arg is a single string, we replicate 783 # it for each of the CPU models, substituting and 784 # %(CPU_foo)s params appropriately. 785 self.exec_output = expand_cpu_symbols_to_dict(exec_output) 786 self.decode_block = expand_cpu_symbols_to_string(decode_block) 787 self.has_decode_default = has_decode_default 788 789 # Override '+' operator: generate a new GenCode object that 790 # concatenates all the individual strings in the operands. 791 def __add__(self, other): 792 exec_output = {} 793 for cpu in cpu_models: 794 n = cpu.name 795 exec_output[n] = self.exec_output[n] + other.exec_output[n] 796 return GenCode(self.header_output + other.header_output, 797 self.decoder_output + other.decoder_output, 798 exec_output, 799 self.decode_block + other.decode_block, 800 self.has_decode_default or other.has_decode_default) 801 802 # Prepend a string (typically a comment) to all the strings. 803 def prepend_all(self, pre): 804 self.header_output = pre + self.header_output 805 self.decoder_output = pre + self.decoder_output 806 self.decode_block = pre + self.decode_block 807 for cpu in cpu_models: 808 self.exec_output[cpu.name] = pre + self.exec_output[cpu.name] 809 810 # Wrap the decode block in a pair of strings (e.g., 'case foo:' 811 # and 'break;'). Used to build the big nested switch statement. 812 def wrap_decode_block(self, pre, post = ''): 813 self.decode_block = pre + indent(self.decode_block) + post 814 815################ 816# Format object. 817# 818# A format object encapsulates an instruction format. It must provide 819# a defineInst() method that generates the code for an instruction 820# definition. 821 822exportContextSymbols = ('InstObjParams', 'makeList', 're', 'string') 823 824exportContext = {} 825 826def updateExportContext(): 827 exportContext.update(exportDict(*exportContextSymbols)) 828 exportContext.update(templateMap) 829 830def exportDict(*symNames): 831 return dict([(s, eval(s)) for s in symNames]) 832 833 834class Format: 835 def __init__(self, id, params, code): 836 # constructor: just save away arguments 837 self.id = id 838 self.params = params 839 label = 'def format ' + id 840 self.user_code = compile(fixPythonIndentation(code), label, 'exec') 841 param_list = string.join(params, ", ") 842 f = '''def defInst(_code, _context, %s): 843 my_locals = vars().copy() 844 exec _code in _context, my_locals 845 return my_locals\n''' % param_list 846 c = compile(f, label + ' wrapper', 'exec') 847 exec c 848 self.func = defInst 849 850 def defineInst(self, name, args, lineno): 851 context = {} 852 updateExportContext() 853 context.update(exportContext) 854 context.update({ 'name': name, 'Name': string.capitalize(name) }) 855 try: 856 vars = self.func(self.user_code, context, *args[0], **args[1]) 857 except Exception, exc: 858 error(lineno, 'error defining "%s": %s.' % (name, exc)) 859 for k in vars.keys(): 860 if k not in ('header_output', 'decoder_output', 861 'exec_output', 'decode_block'): 862 del vars[k] 863 return GenCode(**vars) 864 865# Special null format to catch an implicit-format instruction 866# definition outside of any format block. 867class NoFormat: 868 def __init__(self): 869 self.defaultInst = '' 870 871 def defineInst(self, name, args, lineno): 872 error(lineno, 873 'instruction definition "%s" with no active format!' % name) 874 875# This dictionary maps format name strings to Format objects. 876formatMap = {} 877 878# Define a new format 879def defFormat(id, params, code, lineno): 880 # make sure we haven't already defined this one 881 if formatMap.get(id, None) != None: 882 error(lineno, 'format %s redefined.' % id) 883 # create new object and store in global map 884 formatMap[id] = Format(id, params, code) 885 886 887############## 888# Stack: a simple stack object. Used for both formats (formatStack) 889# and default cases (defaultStack). Simply wraps a list to give more 890# stack-like syntax and enable initialization with an argument list 891# (as opposed to an argument that's a list). 892 893class Stack(list): 894 def __init__(self, *items): 895 list.__init__(self, items) 896 897 def push(self, item): 898 self.append(item); 899 900 def top(self): 901 return self[-1] 902 903# The global format stack. 904formatStack = Stack(NoFormat()) 905 906# The global default case stack. 907defaultStack = Stack( None ) 908 909# Global stack that tracks current file and line number. 910# Each element is a tuple (filename, lineno) that records the 911# *current* filename and the line number in the *previous* file where 912# it was included. 913fileNameStack = Stack() 914 915################### 916# Utility functions 917 918# 919# Indent every line in string 's' by two spaces 920# (except preprocessor directives). 921# Used to make nested code blocks look pretty. 922# 923def indent(s): 924 return re.sub(r'(?m)^(?!#)', ' ', s) 925 926# 927# Munge a somewhat arbitrarily formatted piece of Python code 928# (e.g. from a format 'let' block) into something whose indentation 929# will get by the Python parser. 930# 931# The two keys here are that Python will give a syntax error if 932# there's any whitespace at the beginning of the first line, and that 933# all lines at the same lexical nesting level must have identical 934# indentation. Unfortunately the way code literals work, an entire 935# let block tends to have some initial indentation. Rather than 936# trying to figure out what that is and strip it off, we prepend 'if 937# 1:' to make the let code the nested block inside the if (and have 938# the parser automatically deal with the indentation for us). 939# 940# We don't want to do this if (1) the code block is empty or (2) the 941# first line of the block doesn't have any whitespace at the front. 942 943def fixPythonIndentation(s): 944 # get rid of blank lines first 945 s = re.sub(r'(?m)^\s*\n', '', s); 946 if (s != '' and re.match(r'[ \t]', s[0])): 947 s = 'if 1:\n' + s 948 return s 949 950# Error handler. Just call exit. Output formatted to work under 951# Emacs compile-mode. Optional 'print_traceback' arg, if set to True, 952# prints a Python stack backtrace too (can be handy when trying to 953# debug the parser itself). 954def error(lineno, string, print_traceback = False): 955 spaces = "" 956 for (filename, line) in fileNameStack[0:-1]: 957 print spaces + "In file included from " + filename + ":" 958 spaces += " " 959 # Print a Python stack backtrace if requested. 960 if (print_traceback): 961 traceback.print_exc() 962 if lineno != 0: 963 line_str = "%d:" % lineno 964 else: 965 line_str = "" 966 sys.exit(spaces + "%s:%s %s" % (fileNameStack[-1][0], line_str, string)) 967 968 969##################################################################### 970# 971# Bitfield Operator Support 972# 973##################################################################### 974 975bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>') 976 977bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>') 978bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>') 979 980def substBitOps(code): 981 # first convert single-bit selectors to two-index form 982 # i.e., <n> --> <n:n> 983 code = bitOp1ArgRE.sub(r'<\1:\1>', code) 984 # simple case: selector applied to ID (name) 985 # i.e., foo<a:b> --> bits(foo, a, b) 986 code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code) 987 # if selector is applied to expression (ending in ')'), 988 # we need to search backward for matching '(' 989 match = bitOpExprRE.search(code) 990 while match: 991 exprEnd = match.start() 992 here = exprEnd - 1 993 nestLevel = 1 994 while nestLevel > 0: 995 if code[here] == '(': 996 nestLevel -= 1 997 elif code[here] == ')': 998 nestLevel += 1 999 here -= 1 1000 if here < 0: 1001 sys.exit("Didn't find '('!") 1002 exprStart = here+1 1003 newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1], 1004 match.group(1), match.group(2)) 1005 code = code[:exprStart] + newExpr + code[match.end():] 1006 match = bitOpExprRE.search(code) 1007 return code 1008 1009 1010#################### 1011# Template objects. 1012# 1013# Template objects are format strings that allow substitution from 1014# the attribute spaces of other objects (e.g. InstObjParams instances). 1015 1016labelRE = re.compile(r'[^%]%\(([^\)]+)\)[sd]') 1017 1018class Template: 1019 def __init__(self, t): 1020 self.template = t 1021 1022 def subst(self, d): 1023 myDict = None 1024 1025 # Protect non-Python-dict substitutions (e.g. if there's a printf 1026 # in the templated C++ code) 1027 template = protect_non_subst_percents(self.template) 1028 # CPU-model-specific substitutions are handled later (in GenCode). 1029 template = protect_cpu_symbols(template) 1030 1031 # Build a dict ('myDict') to use for the template substitution. 1032 # Start with the template namespace. Make a copy since we're 1033 # going to modify it. 1034 myDict = templateMap.copy() 1035 1036 if isinstance(d, InstObjParams): 1037 # If we're dealing with an InstObjParams object, we need 1038 # to be a little more sophisticated. The instruction-wide 1039 # parameters are already formed, but the parameters which 1040 # are only function wide still need to be generated. 1041 compositeCode = '' 1042 1043 myDict.update(d.__dict__) 1044 # The "operands" and "snippets" attributes of the InstObjParams 1045 # objects are for internal use and not substitution. 1046 del myDict['operands'] 1047 del myDict['snippets'] 1048 1049 snippetLabels = [l for l in labelRE.findall(template) 1050 if d.snippets.has_key(l)] 1051 1052 snippets = dict([(s, mungeSnippet(d.snippets[s])) 1053 for s in snippetLabels]) 1054 1055 myDict.update(snippets) 1056 1057 compositeCode = ' '.join(map(str, snippets.values())) 1058 1059 # Add in template itself in case it references any 1060 # operands explicitly (like Mem) 1061 compositeCode += ' ' + template 1062 1063 operands = SubOperandList(compositeCode, d.operands) 1064 1065 myDict['op_decl'] = operands.concatAttrStrings('op_decl') 1066 1067 is_src = lambda op: op.is_src 1068 is_dest = lambda op: op.is_dest 1069 1070 myDict['op_src_decl'] = \ 1071 operands.concatSomeAttrStrings(is_src, 'op_src_decl') 1072 myDict['op_dest_decl'] = \ 1073 operands.concatSomeAttrStrings(is_dest, 'op_dest_decl') 1074 1075 myDict['op_rd'] = operands.concatAttrStrings('op_rd') 1076 myDict['op_wb'] = operands.concatAttrStrings('op_wb') 1077 1078 if d.operands.memOperand: 1079 myDict['mem_acc_size'] = d.operands.memOperand.mem_acc_size 1080 myDict['mem_acc_type'] = d.operands.memOperand.mem_acc_type 1081 1082 elif isinstance(d, dict): 1083 # if the argument is a dictionary, we just use it. 1084 myDict.update(d) 1085 elif hasattr(d, '__dict__'): 1086 # if the argument is an object, we use its attribute map. 1087 myDict.update(d.__dict__) 1088 else: 1089 raise TypeError, "Template.subst() arg must be or have dictionary" 1090 return template % myDict 1091 1092 # Convert to string. This handles the case when a template with a 1093 # CPU-specific term gets interpolated into another template or into 1094 # an output block. 1095 def __str__(self): 1096 return expand_cpu_symbols_to_string(self.template) 1097 1098##################################################################### 1099# 1100# Code Parser 1101# 1102# The remaining code is the support for automatically extracting 1103# instruction characteristics from pseudocode. 1104# 1105##################################################################### 1106 1107# Force the argument to be a list. Useful for flags, where a caller 1108# can specify a singleton flag or a list of flags. Also usful for 1109# converting tuples to lists so they can be modified. 1110def makeList(arg): 1111 if isinstance(arg, list): 1112 return arg 1113 elif isinstance(arg, tuple): 1114 return list(arg) 1115 elif not arg: 1116 return [] 1117 else: 1118 return [ arg ] 1119 1120# Generate operandTypeMap from the user's 'def operand_types' 1121# statement. 1122def buildOperandTypeMap(userDict, lineno): 1123 global operandTypeMap 1124 operandTypeMap = {} 1125 for (ext, (desc, size)) in userDict.iteritems(): 1126 if desc == 'signed int': 1127 ctype = 'int%d_t' % size 1128 is_signed = 1 1129 elif desc == 'unsigned int': 1130 ctype = 'uint%d_t' % size 1131 is_signed = 0 1132 elif desc == 'float': 1133 is_signed = 1 # shouldn't really matter 1134 if size == 32: 1135 ctype = 'float' 1136 elif size == 64: 1137 ctype = 'double' 1138 elif desc == 'twin64 int': 1139 is_signed = 0 1140 ctype = 'Twin64_t' 1141 elif desc == 'twin32 int': 1142 is_signed = 0 1143 ctype = 'Twin32_t' 1144 if ctype == '': 1145 error(lineno, 'Unrecognized type description "%s" in userDict') 1146 operandTypeMap[ext] = (size, ctype, is_signed) 1147 1148# 1149# 1150# 1151# Base class for operand descriptors. An instance of this class (or 1152# actually a class derived from this one) represents a specific 1153# operand for a code block (e.g, "Rc.sq" as a dest). Intermediate 1154# derived classes encapsulates the traits of a particular operand type 1155# (e.g., "32-bit integer register"). 1156# 1157class Operand(object): 1158 def __init__(self, full_name, ext, is_src, is_dest): 1159 self.full_name = full_name 1160 self.ext = ext 1161 self.is_src = is_src 1162 self.is_dest = is_dest 1163 # The 'effective extension' (eff_ext) is either the actual 1164 # extension, if one was explicitly provided, or the default. 1165 if ext: 1166 self.eff_ext = ext 1167 else: 1168 self.eff_ext = self.dflt_ext 1169 1170 (self.size, self.ctype, self.is_signed) = operandTypeMap[self.eff_ext] 1171 1172 # note that mem_acc_size is undefined for non-mem operands... 1173 # template must be careful not to use it if it doesn't apply. 1174 if self.isMem(): 1175 self.mem_acc_size = self.makeAccSize() 1176 if self.ctype in ['Twin32_t', 'Twin64_t']: 1177 self.mem_acc_type = 'Twin' 1178 else: 1179 self.mem_acc_type = 'uint' 1180 1181 # Finalize additional fields (primarily code fields). This step 1182 # is done separately since some of these fields may depend on the 1183 # register index enumeration that hasn't been performed yet at the 1184 # time of __init__(). 1185 def finalize(self): 1186 self.flags = self.getFlags() 1187 self.constructor = self.makeConstructor() 1188 self.op_decl = self.makeDecl() 1189 1190 if self.is_src: 1191 self.op_rd = self.makeRead() 1192 self.op_src_decl = self.makeDecl() 1193 else: 1194 self.op_rd = '' 1195 self.op_src_decl = '' 1196 1197 if self.is_dest: 1198 self.op_wb = self.makeWrite() 1199 self.op_dest_decl = self.makeDecl() 1200 else: 1201 self.op_wb = '' 1202 self.op_dest_decl = '' 1203 1204 def isMem(self): 1205 return 0 1206 1207 def isReg(self): 1208 return 0 1209 1210 def isFloatReg(self): 1211 return 0 1212 1213 def isIntReg(self): 1214 return 0 1215 1216 def isControlReg(self): 1217 return 0 1218 1219 def getFlags(self): 1220 # note the empty slice '[:]' gives us a copy of self.flags[0] 1221 # instead of a reference to it 1222 my_flags = self.flags[0][:] 1223 if self.is_src: 1224 my_flags += self.flags[1] 1225 if self.is_dest: 1226 my_flags += self.flags[2] 1227 return my_flags 1228 1229 def makeDecl(self): 1230 # Note that initializations in the declarations are solely 1231 # to avoid 'uninitialized variable' errors from the compiler. 1232 return self.ctype + ' ' + self.base_name + ' = 0;\n'; 1233 1234class IntRegOperand(Operand): 1235 def isReg(self): 1236 return 1 1237 1238 def isIntReg(self): 1239 return 1 1240 1241 def makeConstructor(self): 1242 c = '' 1243 if self.is_src: 1244 c += '\n\t_srcRegIdx[%d] = %s;' % \ 1245 (self.src_reg_idx, self.reg_spec) 1246 if self.is_dest: 1247 c += '\n\t_destRegIdx[%d] = %s;' % \ 1248 (self.dest_reg_idx, self.reg_spec) 1249 return c 1250 1251 def makeRead(self): 1252 if (self.ctype == 'float' or self.ctype == 'double'): 1253 error(0, 'Attempt to read integer register as FP') 1254 if (self.size == self.dflt_size): 1255 return '%s = xc->readIntRegOperand(this, %d);\n' % \ 1256 (self.base_name, self.src_reg_idx) 1257 elif (self.size > self.dflt_size): 1258 int_reg_val = 'xc->readIntRegOperand(this, %d)' % \ 1259 (self.src_reg_idx) 1260 if (self.is_signed): 1261 int_reg_val = 'sext<%d>(%s)' % (self.dflt_size, int_reg_val) 1262 return '%s = %s;\n' % (self.base_name, int_reg_val) 1263 else: 1264 return '%s = bits(xc->readIntRegOperand(this, %d), %d, 0);\n' % \ 1265 (self.base_name, self.src_reg_idx, self.size-1) 1266 1267 def makeWrite(self): 1268 if (self.ctype == 'float' or self.ctype == 'double'): 1269 error(0, 'Attempt to write integer register as FP') 1270 if (self.size != self.dflt_size and self.is_signed): 1271 final_val = 'sext<%d>(%s)' % (self.size, self.base_name) 1272 else: 1273 final_val = self.base_name 1274 wb = ''' 1275 { 1276 %s final_val = %s; 1277 xc->setIntRegOperand(this, %d, final_val);\n 1278 if (traceData) { traceData->setData(final_val); } 1279 }''' % (self.dflt_ctype, final_val, self.dest_reg_idx) 1280 return wb 1281 1282class FloatRegOperand(Operand): 1283 def isReg(self): 1284 return 1 1285 1286 def isFloatReg(self): 1287 return 1 1288 1289 def makeConstructor(self): 1290 c = '' 1291 if self.is_src: 1292 c += '\n\t_srcRegIdx[%d] = %s + FP_Base_DepTag;' % \ 1293 (self.src_reg_idx, self.reg_spec) 1294 if self.is_dest: 1295 c += '\n\t_destRegIdx[%d] = %s + FP_Base_DepTag;' % \ 1296 (self.dest_reg_idx, self.reg_spec) 1297 return c 1298 1299 def makeRead(self): 1300 bit_select = 0 1301 width = 0; 1302 if (self.ctype == 'float'): 1303 func = 'readFloatRegOperand' 1304 width = 32; 1305 elif (self.ctype == 'double'): 1306 func = 'readFloatRegOperand' 1307 width = 64; 1308 else: 1309 func = 'readFloatRegOperandBits' 1310 if (self.ctype == 'uint32_t'): 1311 width = 32; 1312 elif (self.ctype == 'uint64_t'): 1313 width = 64; 1314 if (self.size != self.dflt_size): 1315 bit_select = 1 1316 if width: 1317 base = 'xc->%s(this, %d, %d)' % \ 1318 (func, self.src_reg_idx, width) 1319 else: 1320 base = 'xc->%s(this, %d)' % \ 1321 (func, self.src_reg_idx) 1322 if bit_select: 1323 return '%s = bits(%s, %d, 0);\n' % \ 1324 (self.base_name, base, self.size-1) 1325 else: 1326 return '%s = %s;\n' % (self.base_name, base) 1327 1328 def makeWrite(self): 1329 final_val = self.base_name 1330 final_ctype = self.ctype 1331 widthSpecifier = '' 1332 width = 0 1333 if (self.ctype == 'float'): 1334 width = 32 1335 func = 'setFloatRegOperand' 1336 elif (self.ctype == 'double'): 1337 width = 64 1338 func = 'setFloatRegOperand' 1339 elif (self.ctype == 'uint32_t'): 1340 func = 'setFloatRegOperandBits' 1341 width = 32 1342 elif (self.ctype == 'uint64_t'): 1343 func = 'setFloatRegOperandBits' 1344 width = 64 1345 else: 1346 func = 'setFloatRegOperandBits' 1347 final_ctype = 'uint%d_t' % self.dflt_size 1348 if (self.size != self.dflt_size and self.is_signed): 1349 final_val = 'sext<%d>(%s)' % (self.size, self.base_name) 1350 if width: 1351 widthSpecifier = ', %d' % width 1352 wb = ''' 1353 { 1354 %s final_val = %s; 1355 xc->%s(this, %d, final_val%s);\n 1356 if (traceData) { traceData->setData(final_val); } 1357 }''' % (final_ctype, final_val, func, self.dest_reg_idx, 1358 widthSpecifier) 1359 return wb 1360 1361class ControlRegOperand(Operand): 1362 def isReg(self): 1363 return 1 1364 1365 def isControlReg(self): 1366 return 1 1367 1368 def makeConstructor(self): 1369 c = '' 1370 if self.is_src: 1371 c += '\n\t_srcRegIdx[%d] = %s + Ctrl_Base_DepTag;' % \ 1372 (self.src_reg_idx, self.reg_spec) 1373 if self.is_dest: 1374 c += '\n\t_destRegIdx[%d] = %s + Ctrl_Base_DepTag;' % \ 1375 (self.dest_reg_idx, self.reg_spec) 1376 return c 1377 1378 def makeRead(self): 1379 bit_select = 0 1380 if (self.ctype == 'float' or self.ctype == 'double'): 1381 error(0, 'Attempt to read control register as FP') 1382 base = 'xc->readMiscRegOperand(this, %s)' % self.src_reg_idx 1383 if self.size == self.dflt_size: 1384 return '%s = %s;\n' % (self.base_name, base) 1385 else: 1386 return '%s = bits(%s, %d, 0);\n' % \ 1387 (self.base_name, base, self.size-1) 1388 1389 def makeWrite(self): 1390 if (self.ctype == 'float' or self.ctype == 'double'): 1391 error(0, 'Attempt to write control register as FP') 1392 wb = 'xc->setMiscRegOperand(this, %s, %s);\n' % \ 1393 (self.dest_reg_idx, self.base_name) 1394 wb += 'if (traceData) { traceData->setData(%s); }' % \ 1395 self.base_name 1396 return wb 1397 1398class MemOperand(Operand): 1399 def isMem(self): 1400 return 1 1401 1402 def makeConstructor(self): 1403 return '' 1404 1405 def makeDecl(self): 1406 # Note that initializations in the declarations are solely 1407 # to avoid 'uninitialized variable' errors from the compiler. 1408 # Declare memory data variable. 1409 if self.ctype in ['Twin32_t','Twin64_t']: 1410 return "%s %s; %s.a = 0; %s.b = 0;\n" % (self.ctype, self.base_name, 1411 self.base_name, self.base_name) 1412 c = '%s %s = 0;\n' % (self.ctype, self.base_name) 1413 return c 1414 1415 def makeRead(self): 1416 return '' 1417 1418 def makeWrite(self): 1419 return '' 1420 1421 # Return the memory access size *in bits*, suitable for 1422 # forming a type via "uint%d_t". Divide by 8 if you want bytes. 1423 def makeAccSize(self): 1424 return self.size 1425 1426 1427class NPCOperand(Operand): 1428 def makeConstructor(self): 1429 return '' 1430 1431 def makeRead(self): 1432 return '%s = xc->readNextPC();\n' % self.base_name 1433 1434 def makeWrite(self): 1435 return 'xc->setNextPC(%s);\n' % self.base_name 1436 1437class NNPCOperand(Operand): 1438 def makeConstructor(self): 1439 return '' 1440 1441 def makeRead(self): 1442 return '%s = xc->readNextNPC();\n' % self.base_name 1443 1444 def makeWrite(self): 1445 return 'xc->setNextNPC(%s);\n' % self.base_name 1446 1447def buildOperandNameMap(userDict, lineno): 1448 global operandNameMap 1449 operandNameMap = {} 1450 for (op_name, val) in userDict.iteritems(): 1451 (base_cls_name, dflt_ext, reg_spec, flags, sort_pri) = val 1452 (dflt_size, dflt_ctype, dflt_is_signed) = operandTypeMap[dflt_ext] 1453 # Canonical flag structure is a triple of lists, where each list 1454 # indicates the set of flags implied by this operand always, when 1455 # used as a source, and when used as a dest, respectively. 1456 # For simplicity this can be initialized using a variety of fairly 1457 # obvious shortcuts; we convert these to canonical form here. 1458 if not flags: 1459 # no flags specified (e.g., 'None') 1460 flags = ( [], [], [] ) 1461 elif isinstance(flags, str): 1462 # a single flag: assumed to be unconditional 1463 flags = ( [ flags ], [], [] ) 1464 elif isinstance(flags, list): 1465 # a list of flags: also assumed to be unconditional 1466 flags = ( flags, [], [] ) 1467 elif isinstance(flags, tuple): 1468 # it's a tuple: it should be a triple, 1469 # but each item could be a single string or a list 1470 (uncond_flags, src_flags, dest_flags) = flags 1471 flags = (makeList(uncond_flags), 1472 makeList(src_flags), makeList(dest_flags)) 1473 # Accumulate attributes of new operand class in tmp_dict 1474 tmp_dict = {} 1475 for attr in ('dflt_ext', 'reg_spec', 'flags', 'sort_pri', 1476 'dflt_size', 'dflt_ctype', 'dflt_is_signed'): 1477 tmp_dict[attr] = eval(attr) 1478 tmp_dict['base_name'] = op_name 1479 # New class name will be e.g. "IntReg_Ra" 1480 cls_name = base_cls_name + '_' + op_name 1481 # Evaluate string arg to get class object. Note that the 1482 # actual base class for "IntReg" is "IntRegOperand", i.e. we 1483 # have to append "Operand". 1484 try: 1485 base_cls = eval(base_cls_name + 'Operand') 1486 except NameError: 1487 error(lineno, 1488 'error: unknown operand base class "%s"' % base_cls_name) 1489 # The following statement creates a new class called 1490 # <cls_name> as a subclass of <base_cls> with the attributes 1491 # in tmp_dict, just as if we evaluated a class declaration. 1492 operandNameMap[op_name] = type(cls_name, (base_cls,), tmp_dict) 1493 1494 # Define operand variables. 1495 operands = userDict.keys() 1496 1497 operandsREString = (r''' 1498 (?<![\w\.]) # neg. lookbehind assertion: prevent partial matches 1499 ((%s)(?:\.(\w+))?) # match: operand with optional '.' then suffix 1500 (?![\w\.]) # neg. lookahead assertion: prevent partial matches 1501 ''' 1502 % string.join(operands, '|')) 1503 1504 global operandsRE 1505 operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE) 1506 1507 # Same as operandsREString, but extension is mandatory, and only two 1508 # groups are returned (base and ext, not full name as above). 1509 # Used for subtituting '_' for '.' to make C++ identifiers. 1510 operandsWithExtREString = (r'(?<![\w\.])(%s)\.(\w+)(?![\w\.])' 1511 % string.join(operands, '|')) 1512 1513 global operandsWithExtRE 1514 operandsWithExtRE = re.compile(operandsWithExtREString, re.MULTILINE) 1515 1516 1517class OperandList: 1518 1519 # Find all the operands in the given code block. Returns an operand 1520 # descriptor list (instance of class OperandList). 1521 def __init__(self, code): 1522 self.items = [] 1523 self.bases = {} 1524 # delete comments so we don't match on reg specifiers inside 1525 code = commentRE.sub('', code) 1526 # search for operands 1527 next_pos = 0 1528 while 1: 1529 match = operandsRE.search(code, next_pos) 1530 if not match: 1531 # no more matches: we're done 1532 break 1533 op = match.groups() 1534 # regexp groups are operand full name, base, and extension 1535 (op_full, op_base, op_ext) = op 1536 # if the token following the operand is an assignment, this is 1537 # a destination (LHS), else it's a source (RHS) 1538 is_dest = (assignRE.match(code, match.end()) != None) 1539 is_src = not is_dest 1540 # see if we've already seen this one 1541 op_desc = self.find_base(op_base) 1542 if op_desc: 1543 if op_desc.ext != op_ext: 1544 error(0, 'Inconsistent extensions for operand %s' % \ 1545 op_base) 1546 op_desc.is_src = op_desc.is_src or is_src 1547 op_desc.is_dest = op_desc.is_dest or is_dest 1548 else: 1549 # new operand: create new descriptor 1550 op_desc = operandNameMap[op_base](op_full, op_ext, 1551 is_src, is_dest) 1552 self.append(op_desc) 1553 # start next search after end of current match 1554 next_pos = match.end() 1555 self.sort() 1556 # enumerate source & dest register operands... used in building 1557 # constructor later 1558 self.numSrcRegs = 0 1559 self.numDestRegs = 0 1560 self.numFPDestRegs = 0 1561 self.numIntDestRegs = 0 1562 self.memOperand = None 1563 for op_desc in self.items: 1564 if op_desc.isReg(): 1565 if op_desc.is_src: 1566 op_desc.src_reg_idx = self.numSrcRegs 1567 self.numSrcRegs += 1 1568 if op_desc.is_dest: 1569 op_desc.dest_reg_idx = self.numDestRegs 1570 self.numDestRegs += 1 1571 if op_desc.isFloatReg(): 1572 self.numFPDestRegs += 1 1573 elif op_desc.isIntReg(): 1574 self.numIntDestRegs += 1 1575 elif op_desc.isMem(): 1576 if self.memOperand: 1577 error(0, "Code block has more than one memory operand.") 1578 self.memOperand = op_desc 1579 # now make a final pass to finalize op_desc fields that may depend 1580 # on the register enumeration 1581 for op_desc in self.items: 1582 op_desc.finalize() 1583 1584 def __len__(self): 1585 return len(self.items) 1586 1587 def __getitem__(self, index): 1588 return self.items[index] 1589 1590 def append(self, op_desc): 1591 self.items.append(op_desc) 1592 self.bases[op_desc.base_name] = op_desc 1593 1594 def find_base(self, base_name): 1595 # like self.bases[base_name], but returns None if not found 1596 # (rather than raising exception) 1597 return self.bases.get(base_name) 1598 1599 # internal helper function for concat[Some]Attr{Strings|Lists} 1600 def __internalConcatAttrs(self, attr_name, filter, result): 1601 for op_desc in self.items: 1602 if filter(op_desc): 1603 result += getattr(op_desc, attr_name) 1604 return result 1605 1606 # return a single string that is the concatenation of the (string) 1607 # values of the specified attribute for all operands 1608 def concatAttrStrings(self, attr_name): 1609 return self.__internalConcatAttrs(attr_name, lambda x: 1, '') 1610 1611 # like concatAttrStrings, but only include the values for the operands 1612 # for which the provided filter function returns true 1613 def concatSomeAttrStrings(self, filter, attr_name): 1614 return self.__internalConcatAttrs(attr_name, filter, '') 1615 1616 # return a single list that is the concatenation of the (list) 1617 # values of the specified attribute for all operands 1618 def concatAttrLists(self, attr_name): 1619 return self.__internalConcatAttrs(attr_name, lambda x: 1, []) 1620 1621 # like concatAttrLists, but only include the values for the operands 1622 # for which the provided filter function returns true 1623 def concatSomeAttrLists(self, filter, attr_name): 1624 return self.__internalConcatAttrs(attr_name, filter, []) 1625 1626 def sort(self): 1627 self.items.sort(lambda a, b: a.sort_pri - b.sort_pri) 1628 1629class SubOperandList(OperandList): 1630 1631 # Find all the operands in the given code block. Returns an operand 1632 # descriptor list (instance of class OperandList). 1633 def __init__(self, code, master_list): 1634 self.items = [] 1635 self.bases = {} 1636 # delete comments so we don't match on reg specifiers inside 1637 code = commentRE.sub('', code) 1638 # search for operands 1639 next_pos = 0 1640 while 1: 1641 match = operandsRE.search(code, next_pos) 1642 if not match: 1643 # no more matches: we're done 1644 break 1645 op = match.groups() 1646 # regexp groups are operand full name, base, and extension 1647 (op_full, op_base, op_ext) = op 1648 # find this op in the master list 1649 op_desc = master_list.find_base(op_base) 1650 if not op_desc: 1651 error(0, 'Found operand %s which is not in the master list!' \ 1652 ' This is an internal error' % \ 1653 op_base) 1654 else: 1655 # See if we've already found this operand 1656 op_desc = self.find_base(op_base) 1657 if not op_desc: 1658 # if not, add a reference to it to this sub list 1659 self.append(master_list.bases[op_base]) 1660 1661 # start next search after end of current match 1662 next_pos = match.end() 1663 self.sort() 1664 self.memOperand = None 1665 for op_desc in self.items: 1666 if op_desc.isMem(): 1667 if self.memOperand: 1668 error(0, "Code block has more than one memory operand.") 1669 self.memOperand = op_desc 1670 1671# Regular expression object to match C++ comments 1672# (used in findOperands()) 1673commentRE = re.compile(r'//.*\n') 1674 1675# Regular expression object to match assignment statements 1676# (used in findOperands()) 1677assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE) 1678 1679# Munge operand names in code string to make legal C++ variable names. 1680# This means getting rid of the type extension if any. 1681# (Will match base_name attribute of Operand object.) 1682def substMungedOpNames(code): 1683 return operandsWithExtRE.sub(r'\1', code) 1684 1685# Fix up code snippets for final substitution in templates. 1686def mungeSnippet(s): 1687 if isinstance(s, str): 1688 return substMungedOpNames(substBitOps(s)) 1689 else: 1690 return s 1691 1692def makeFlagConstructor(flag_list): 1693 if len(flag_list) == 0: 1694 return '' 1695 # filter out repeated flags 1696 flag_list.sort() 1697 i = 1 1698 while i < len(flag_list): 1699 if flag_list[i] == flag_list[i-1]: 1700 del flag_list[i] 1701 else: 1702 i += 1 1703 pre = '\n\tflags[' 1704 post = '] = true;' 1705 code = pre + string.join(flag_list, post + pre) + post 1706 return code 1707 1708# Assume all instruction flags are of the form 'IsFoo' 1709instFlagRE = re.compile(r'Is.*') 1710 1711# OpClass constants end in 'Op' except No_OpClass 1712opClassRE = re.compile(r'.*Op|No_OpClass') 1713 1714class InstObjParams: 1715 def __init__(self, mnem, class_name, base_class = '', 1716 snippets = {}, opt_args = []): 1717 self.mnemonic = mnem 1718 self.class_name = class_name 1719 self.base_class = base_class 1720 if not isinstance(snippets, dict): 1721 snippets = {'code' : snippets} 1722 compositeCode = ' '.join(map(str, snippets.values())) 1723 self.snippets = snippets 1724 1725 self.operands = OperandList(compositeCode) 1726 self.constructor = self.operands.concatAttrStrings('constructor') 1727 self.constructor += \ 1728 '\n\t_numSrcRegs = %d;' % self.operands.numSrcRegs 1729 self.constructor += \ 1730 '\n\t_numDestRegs = %d;' % self.operands.numDestRegs 1731 self.constructor += \ 1732 '\n\t_numFPDestRegs = %d;' % self.operands.numFPDestRegs 1733 self.constructor += \ 1734 '\n\t_numIntDestRegs = %d;' % self.operands.numIntDestRegs 1735 self.flags = self.operands.concatAttrLists('flags') 1736 1737 # Make a basic guess on the operand class (function unit type). 1738 # These are good enough for most cases, and can be overridden 1739 # later otherwise. 1740 if 'IsStore' in self.flags: 1741 self.op_class = 'MemWriteOp' 1742 elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags: 1743 self.op_class = 'MemReadOp' 1744 elif 'IsFloating' in self.flags: 1745 self.op_class = 'FloatAddOp' 1746 else: 1747 self.op_class = 'IntAluOp' 1748 1749 # Optional arguments are assumed to be either StaticInst flags 1750 # or an OpClass value. To avoid having to import a complete 1751 # list of these values to match against, we do it ad-hoc 1752 # with regexps. 1753 for oa in opt_args: 1754 if instFlagRE.match(oa): 1755 self.flags.append(oa) 1756 elif opClassRE.match(oa): 1757 self.op_class = oa 1758 else: 1759 error(0, 'InstObjParams: optional arg "%s" not recognized ' 1760 'as StaticInst::Flag or OpClass.' % oa) 1761 1762 # add flag initialization to contructor here to include 1763 # any flags added via opt_args 1764 self.constructor += makeFlagConstructor(self.flags) 1765 1766 # if 'IsFloating' is set, add call to the FP enable check 1767 # function (which should be provided by isa_desc via a declare) 1768 if 'IsFloating' in self.flags: 1769 self.fp_enable_check = 'fault = checkFpEnableFault(xc);' 1770 else: 1771 self.fp_enable_check = '' 1772 1773####################### 1774# 1775# Output file template 1776# 1777 1778file_template = ''' 1779/* 1780 * DO NOT EDIT THIS FILE!!! 1781 * 1782 * It was automatically generated from the ISA description in %(filename)s 1783 */ 1784 1785%(includes)s 1786 1787%(global_output)s 1788 1789namespace %(namespace)s { 1790 1791%(namespace_output)s 1792 1793} // namespace %(namespace)s 1794 1795%(decode_function)s 1796''' 1797 1798 1799# Update the output file only if the new contents are different from 1800# the current contents. Minimizes the files that need to be rebuilt 1801# after minor changes. 1802def update_if_needed(file, contents): 1803 update = False 1804 if os.access(file, os.R_OK): 1805 f = open(file, 'r') 1806 old_contents = f.read() 1807 f.close() 1808 if contents != old_contents: 1809 print 'Updating', file 1810 os.remove(file) # in case it's write-protected 1811 update = True 1812 else: 1813 print 'File', file, 'is unchanged' 1814 else: 1815 print 'Generating', file 1816 update = True 1817 if update: 1818 f = open(file, 'w') 1819 f.write(contents) 1820 f.close() 1821 1822# This regular expression matches '##include' directives 1823includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[\w/.-]*)".*$', 1824 re.MULTILINE) 1825 1826# Function to replace a matched '##include' directive with the 1827# contents of the specified file (with nested ##includes replaced 1828# recursively). 'matchobj' is an re match object (from a match of 1829# includeRE) and 'dirname' is the directory relative to which the file 1830# path should be resolved. 1831def replace_include(matchobj, dirname): 1832 fname = matchobj.group('filename') 1833 full_fname = os.path.normpath(os.path.join(dirname, fname)) 1834 contents = '##newfile "%s"\n%s\n##endfile\n' % \ 1835 (full_fname, read_and_flatten(full_fname)) 1836 return contents 1837 1838# Read a file and recursively flatten nested '##include' files. 1839def read_and_flatten(filename): 1840 current_dir = os.path.dirname(filename) 1841 try: 1842 contents = open(filename).read() 1843 except IOError: 1844 error(0, 'Error including file "%s"' % filename) 1845 fileNameStack.push((filename, 0)) 1846 # Find any includes and include them 1847 contents = includeRE.sub(lambda m: replace_include(m, current_dir), 1848 contents) 1849 fileNameStack.pop() 1850 return contents 1851 1852# 1853# Read in and parse the ISA description. 1854# 1855def parse_isa_desc(isa_desc_file, output_dir): 1856 # Read file and (recursively) all included files into a string. 1857 # PLY requires that the input be in a single string so we have to 1858 # do this up front. 1859 isa_desc = read_and_flatten(isa_desc_file) 1860 1861 # Initialize filename stack with outer file. 1862 fileNameStack.push((isa_desc_file, 0)) 1863 1864 # Parse it. 1865 (isa_name, namespace, global_code, namespace_code) = yacc.parse(isa_desc) 1866 1867 # grab the last three path components of isa_desc_file to put in 1868 # the output 1869 filename = '/'.join(isa_desc_file.split('/')[-3:]) 1870 1871 # generate decoder.hh 1872 includes = '#include "base/bitfield.hh" // for bitfield support' 1873 global_output = global_code.header_output 1874 namespace_output = namespace_code.header_output 1875 decode_function = '' 1876 update_if_needed(output_dir + '/decoder.hh', file_template % vars()) 1877 1878 # generate decoder.cc 1879 includes = '#include "decoder.hh"' 1880 global_output = global_code.decoder_output 1881 namespace_output = namespace_code.decoder_output 1882 # namespace_output += namespace_code.decode_block 1883 decode_function = namespace_code.decode_block 1884 update_if_needed(output_dir + '/decoder.cc', file_template % vars()) 1885 1886 # generate per-cpu exec files 1887 for cpu in cpu_models: 1888 includes = '#include "decoder.hh"\n' 1889 includes += cpu.includes 1890 global_output = global_code.exec_output[cpu.name] 1891 namespace_output = namespace_code.exec_output[cpu.name] 1892 decode_function = '' 1893 update_if_needed(output_dir + '/' + cpu.filename, 1894 file_template % vars()) 1895 1896# global list of CpuModel objects (see cpu_models.py) 1897cpu_models = [] 1898 1899# Called as script: get args from command line. 1900# Args are: <path to cpu_models.py> <isa desc file> <output dir> <cpu models> 1901if __name__ == '__main__': 1902 execfile(sys.argv[1]) # read in CpuModel definitions 1903 cpu_models = [CpuModel.dict[cpu] for cpu in sys.argv[4:]] 1904 parse_isa_desc(sys.argv[2], sys.argv[3]) 1905