isa_parser.py revision 4466:f5aac470c214
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', 'DOT', '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_DOT = r'\.' 117t_COLON = r':' 118t_DBLCOLON = r'::' 119t_ASTERISK = r'\*' 120 121# Identifiers and reserved words 122reserved_map = { } 123for r in reserved: 124 reserved_map[r.lower()] = r 125 126def t_ID(t): 127 r'[A-Za-z_]\w*' 128 t.type = reserved_map.get(t.value,'ID') 129 return t 130 131# Integer literal 132def t_INTLIT(t): 133 r'(0x[\da-fA-F]+)|\d+' 134 try: 135 t.value = int(t.value,0) 136 except ValueError: 137 error(t.lineno, 'Integer value "%s" too large' % t.value) 138 t.value = 0 139 return t 140 141# String literal. Note that these use only single quotes, and 142# can span multiple lines. 143def t_STRLIT(t): 144 r"(?m)'([^'])+'" 145 # strip off quotes 146 t.value = t.value[1:-1] 147 t.lineno += t.value.count('\n') 148 return t 149 150 151# "Code literal"... like a string literal, but delimiters are 152# '{{' and '}}' so they get formatted nicely under emacs c-mode 153def t_CODELIT(t): 154 r"(?m)\{\{([^\}]|}(?!\}))+\}\}" 155 # strip off {{ & }} 156 t.value = t.value[2:-2] 157 t.lineno += t.value.count('\n') 158 return t 159 160def t_CPPDIRECTIVE(t): 161 r'^\#[^\#].*\n' 162 t.lineno += t.value.count('\n') 163 return t 164 165def t_NEWFILE(t): 166 r'^\#\#newfile\s+"[\w/.-]*"' 167 fileNameStack.push((t.value[11:-1], t.lineno)) 168 t.lineno = 0 169 170def t_ENDFILE(t): 171 r'^\#\#endfile' 172 (old_filename, t.lineno) = fileNameStack.pop() 173 174# 175# The functions t_NEWLINE, t_ignore, and t_error are 176# special for the lex module. 177# 178 179# Newlines 180def t_NEWLINE(t): 181 r'\n+' 182 t.lineno += t.value.count('\n') 183 184# Comments 185def t_comment(t): 186 r'//.*' 187 188# Completely ignored characters 189t_ignore = ' \t\x0c' 190 191# Error handler 192def t_error(t): 193 error(t.lineno, "illegal character '%s'" % t.value[0]) 194 t.skip(1) 195 196# Build the lexer 197lexer = lex.lex() 198 199##################################################################### 200# 201# Parser 202# 203# Every function whose name starts with 'p_' defines a grammar rule. 204# The rule is encoded in the function's doc string, while the 205# function body provides the action taken when the rule is matched. 206# The argument to each function is a list of the values of the 207# rule's symbols: t[0] for the LHS, and t[1..n] for the symbols 208# on the RHS. For tokens, the value is copied from the t.value 209# attribute provided by the lexer. For non-terminals, the value 210# is assigned by the producing rule; i.e., the job of the grammar 211# rule function is to set the value for the non-terminal on the LHS 212# (by assigning to t[0]). 213##################################################################### 214 215# The LHS of the first grammar rule is used as the start symbol 216# (in this case, 'specification'). Note that this rule enforces 217# that there will be exactly one namespace declaration, with 0 or more 218# global defs/decls before and after it. The defs & decls before 219# the namespace decl will be outside the namespace; those after 220# will be inside. The decoder function is always inside the namespace. 221def p_specification(t): 222 'specification : opt_defs_and_outputs name_decl opt_defs_and_outputs decode_block' 223 global_code = t[1] 224 isa_name = t[2] 225 namespace = isa_name + "Inst" 226 # wrap the decode block as a function definition 227 t[4].wrap_decode_block(''' 228StaticInstPtr 229%(isa_name)s::decodeInst(%(isa_name)s::ExtMachInst machInst) 230{ 231 using namespace %(namespace)s; 232''' % vars(), '}') 233 # both the latter output blocks and the decode block are in the namespace 234 namespace_code = t[3] + t[4] 235 # pass it all back to the caller of yacc.parse() 236 t[0] = (isa_name, namespace, global_code, namespace_code) 237 238# ISA name declaration looks like "namespace <foo>;" 239def p_name_decl(t): 240 'name_decl : NAMESPACE ID SEMI' 241 t[0] = t[2] 242 243# 'opt_defs_and_outputs' is a possibly empty sequence of 244# def and/or output statements. 245def p_opt_defs_and_outputs_0(t): 246 'opt_defs_and_outputs : empty' 247 t[0] = GenCode() 248 249def p_opt_defs_and_outputs_1(t): 250 'opt_defs_and_outputs : defs_and_outputs' 251 t[0] = t[1] 252 253def p_defs_and_outputs_0(t): 254 'defs_and_outputs : def_or_output' 255 t[0] = t[1] 256 257def p_defs_and_outputs_1(t): 258 'defs_and_outputs : defs_and_outputs def_or_output' 259 t[0] = t[1] + t[2] 260 261# The list of possible definition/output statements. 262def p_def_or_output(t): 263 '''def_or_output : def_format 264 | def_bitfield 265 | def_bitfield_struct 266 | def_template 267 | def_operand_types 268 | def_operands 269 | output_header 270 | output_decoder 271 | output_exec 272 | global_let''' 273 t[0] = t[1] 274 275# Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied 276# directly to the appropriate output section. 277 278 279# Protect any non-dict-substitution '%'s in a format string 280# (i.e. those not followed by '(') 281def protect_non_subst_percents(s): 282 return re.sub(r'%(?!\()', '%%', s) 283 284# Massage output block by substituting in template definitions and bit 285# operators. We handle '%'s embedded in the string that don't 286# indicate template substitutions (or CPU-specific symbols, which get 287# handled in GenCode) by doubling them first so that the format 288# operation will reduce them back to single '%'s. 289def process_output(s): 290 s = protect_non_subst_percents(s) 291 # protects cpu-specific symbols too 292 s = protect_cpu_symbols(s) 293 return substBitOps(s % templateMap) 294 295def p_output_header(t): 296 'output_header : OUTPUT HEADER CODELIT SEMI' 297 t[0] = GenCode(header_output = process_output(t[3])) 298 299def p_output_decoder(t): 300 'output_decoder : OUTPUT DECODER CODELIT SEMI' 301 t[0] = GenCode(decoder_output = process_output(t[3])) 302 303def p_output_exec(t): 304 'output_exec : OUTPUT EXEC CODELIT SEMI' 305 t[0] = GenCode(exec_output = process_output(t[3])) 306 307# global let blocks 'let {{...}}' (Python code blocks) are executed 308# directly when seen. Note that these execute in a special variable 309# context 'exportContext' to prevent the code from polluting this 310# script's namespace. 311def p_global_let(t): 312 'global_let : LET CODELIT SEMI' 313 updateExportContext() 314 exportContext["header_output"] = '' 315 exportContext["decoder_output"] = '' 316 exportContext["exec_output"] = '' 317 exportContext["decode_block"] = '' 318 try: 319 exec fixPythonIndentation(t[2]) in exportContext 320 except Exception, exc: 321 error(t.lineno(1), 322 'error: %s in global let block "%s".' % (exc, t[2])) 323 t[0] = GenCode(header_output = exportContext["header_output"], 324 decoder_output = exportContext["decoder_output"], 325 exec_output = exportContext["exec_output"], 326 decode_block = exportContext["decode_block"]) 327 328# Define the mapping from operand type extensions to C++ types and bit 329# widths (stored in operandTypeMap). 330def p_def_operand_types(t): 331 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI' 332 try: 333 userDict = eval('{' + t[3] + '}') 334 except Exception, exc: 335 error(t.lineno(1), 336 'error: %s in def operand_types block "%s".' % (exc, t[3])) 337 buildOperandTypeMap(userDict, t.lineno(1)) 338 t[0] = GenCode() # contributes nothing to the output C++ file 339 340# Define the mapping from operand names to operand classes and other 341# traits. Stored in operandNameMap. 342def p_def_operands(t): 343 'def_operands : DEF OPERANDS CODELIT SEMI' 344 if not globals().has_key('operandTypeMap'): 345 error(t.lineno(1), 346 'error: operand types must be defined before operands') 347 try: 348 userDict = eval('{' + t[3] + '}') 349 except Exception, exc: 350 error(t.lineno(1), 351 'error: %s in def operands block "%s".' % (exc, t[3])) 352 buildOperandNameMap(userDict, t.lineno(1)) 353 t[0] = GenCode() # contributes nothing to the output C++ file 354 355# A bitfield definition looks like: 356# 'def [signed] bitfield <ID> [<first>:<last>]' 357# This generates a preprocessor macro in the output file. 358def p_def_bitfield_0(t): 359 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI' 360 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8]) 361 if (t[2] == 'signed'): 362 expr = 'sext<%d>(%s)' % (t[6] - t[8] + 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 single bit: 'def [signed] bitfield <ID> [<bit>]' 367def p_def_bitfield_1(t): 368 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI' 369 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6]) 370 if (t[2] == 'signed'): 371 expr = 'sext<%d>(%s)' % (1, expr) 372 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 373 t[0] = GenCode(header_output = hash_define) 374 375# alternate form for structure member: 'def bitfield <ID> <ID>' 376def p_def_bitfield_struct(t): 377 'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI' 378 if (t[2] != ''): 379 error(t.lineno(1), 'error: structure bitfields are always unsigned.') 380 expr = 'machInst.%s' % t[5] 381 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 382 t[0] = GenCode(header_output = hash_define) 383 384def p_id_with_dot_0(t): 385 'id_with_dot : ID' 386 t[0] = t[1] 387 388def p_id_with_dot_1(t): 389 'id_with_dot : ID DOT id_with_dot' 390 t[0] = t[1] + t[2] + t[3] 391 392def p_opt_signed_0(t): 393 'opt_signed : SIGNED' 394 t[0] = t[1] 395 396def p_opt_signed_1(t): 397 'opt_signed : empty' 398 t[0] = '' 399 400# Global map variable to hold templates 401templateMap = {} 402 403def p_def_template(t): 404 'def_template : DEF TEMPLATE ID CODELIT SEMI' 405 templateMap[t[3]] = Template(t[4]) 406 t[0] = GenCode() 407 408# An instruction format definition looks like 409# "def format <fmt>(<params>) {{...}};" 410def p_def_format(t): 411 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI' 412 (id, params, code) = (t[3], t[5], t[7]) 413 defFormat(id, params, code, t.lineno(1)) 414 t[0] = GenCode() 415 416# The formal parameter list for an instruction format is a possibly 417# empty list of comma-separated parameters. Positional (standard, 418# non-keyword) parameters must come first, followed by keyword 419# parameters, followed by a '*foo' parameter that gets excess 420# positional arguments (as in Python). Each of these three parameter 421# categories is optional. 422# 423# Note that we do not support the '**foo' parameter for collecting 424# otherwise undefined keyword args. Otherwise the parameter list is 425# (I believe) identical to what is supported in Python. 426# 427# The param list generates a tuple, where the first element is a list of 428# the positional params and the second element is a dict containing the 429# keyword params. 430def p_param_list_0(t): 431 'param_list : positional_param_list COMMA nonpositional_param_list' 432 t[0] = t[1] + t[3] 433 434def p_param_list_1(t): 435 '''param_list : positional_param_list 436 | nonpositional_param_list''' 437 t[0] = t[1] 438 439def p_positional_param_list_0(t): 440 'positional_param_list : empty' 441 t[0] = [] 442 443def p_positional_param_list_1(t): 444 'positional_param_list : ID' 445 t[0] = [t[1]] 446 447def p_positional_param_list_2(t): 448 'positional_param_list : positional_param_list COMMA ID' 449 t[0] = t[1] + [t[3]] 450 451def p_nonpositional_param_list_0(t): 452 'nonpositional_param_list : keyword_param_list COMMA excess_args_param' 453 t[0] = t[1] + t[3] 454 455def p_nonpositional_param_list_1(t): 456 '''nonpositional_param_list : keyword_param_list 457 | excess_args_param''' 458 t[0] = t[1] 459 460def p_keyword_param_list_0(t): 461 'keyword_param_list : keyword_param' 462 t[0] = [t[1]] 463 464def p_keyword_param_list_1(t): 465 'keyword_param_list : keyword_param_list COMMA keyword_param' 466 t[0] = t[1] + [t[3]] 467 468def p_keyword_param(t): 469 'keyword_param : ID EQUALS expr' 470 t[0] = t[1] + ' = ' + t[3].__repr__() 471 472def p_excess_args_param(t): 473 'excess_args_param : ASTERISK ID' 474 # Just concatenate them: '*ID'. Wrap in list to be consistent 475 # with positional_param_list and keyword_param_list. 476 t[0] = [t[1] + t[2]] 477 478# End of format definition-related rules. 479############## 480 481# 482# A decode block looks like: 483# decode <field1> [, <field2>]* [default <inst>] { ... } 484# 485def p_decode_block(t): 486 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE' 487 default_defaults = defaultStack.pop() 488 codeObj = t[5] 489 # use the "default defaults" only if there was no explicit 490 # default statement in decode_stmt_list 491 if not codeObj.has_decode_default: 492 codeObj += default_defaults 493 codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n') 494 t[0] = codeObj 495 496# The opt_default statement serves only to push the "default defaults" 497# onto defaultStack. This value will be used by nested decode blocks, 498# and used and popped off when the current decode_block is processed 499# (in p_decode_block() above). 500def p_opt_default_0(t): 501 'opt_default : empty' 502 # no default specified: reuse the one currently at the top of the stack 503 defaultStack.push(defaultStack.top()) 504 # no meaningful value returned 505 t[0] = None 506 507def p_opt_default_1(t): 508 'opt_default : DEFAULT inst' 509 # push the new default 510 codeObj = t[2] 511 codeObj.wrap_decode_block('\ndefault:\n', 'break;\n') 512 defaultStack.push(codeObj) 513 # no meaningful value returned 514 t[0] = None 515 516def p_decode_stmt_list_0(t): 517 'decode_stmt_list : decode_stmt' 518 t[0] = t[1] 519 520def p_decode_stmt_list_1(t): 521 'decode_stmt_list : decode_stmt decode_stmt_list' 522 if (t[1].has_decode_default and t[2].has_decode_default): 523 error(t.lineno(1), 'Two default cases in decode block') 524 t[0] = t[1] + t[2] 525 526# 527# Decode statement rules 528# 529# There are four types of statements allowed in a decode block: 530# 1. Format blocks 'format <foo> { ... }' 531# 2. Nested decode blocks 532# 3. Instruction definitions. 533# 4. C preprocessor directives. 534 535 536# Preprocessor directives found in a decode statement list are passed 537# through to the output, replicated to all of the output code 538# streams. This works well for ifdefs, so we can ifdef out both the 539# declarations and the decode cases generated by an instruction 540# definition. Handling them as part of the grammar makes it easy to 541# keep them in the right place with respect to the code generated by 542# the other statements. 543def p_decode_stmt_cpp(t): 544 'decode_stmt : CPPDIRECTIVE' 545 t[0] = GenCode(t[1], t[1], t[1], t[1]) 546 547# A format block 'format <foo> { ... }' sets the default instruction 548# format used to handle instruction definitions inside the block. 549# This format can be overridden by using an explicit format on the 550# instruction definition or with a nested format block. 551def p_decode_stmt_format(t): 552 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE' 553 # The format will be pushed on the stack when 'push_format_id' is 554 # processed (see below). Once the parser has recognized the full 555 # production (though the right brace), we're done with the format, 556 # so now we can pop it. 557 formatStack.pop() 558 t[0] = t[4] 559 560# This rule exists so we can set the current format (& push the stack) 561# when we recognize the format name part of the format block. 562def p_push_format_id(t): 563 'push_format_id : ID' 564 try: 565 formatStack.push(formatMap[t[1]]) 566 t[0] = ('', '// format %s' % t[1]) 567 except KeyError: 568 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 569 570# Nested decode block: if the value of the current field matches the 571# specified constant, do a nested decode on some other field. 572def p_decode_stmt_decode(t): 573 'decode_stmt : case_label COLON decode_block' 574 label = t[1] 575 codeObj = t[3] 576 # just wrap the decoding code from the block as a case in the 577 # outer switch statement. 578 codeObj.wrap_decode_block('\n%s:\n' % label) 579 codeObj.has_decode_default = (label == 'default') 580 t[0] = codeObj 581 582# Instruction definition (finally!). 583def p_decode_stmt_inst(t): 584 'decode_stmt : case_label COLON inst SEMI' 585 label = t[1] 586 codeObj = t[3] 587 codeObj.wrap_decode_block('\n%s:' % label, 'break;\n') 588 codeObj.has_decode_default = (label == 'default') 589 t[0] = codeObj 590 591# The case label is either a list of one or more constants or 'default' 592def p_case_label_0(t): 593 'case_label : intlit_list' 594 t[0] = ': '.join(map(lambda a: 'case %#x' % a, t[1])) 595 596def p_case_label_1(t): 597 'case_label : DEFAULT' 598 t[0] = 'default' 599 600# 601# The constant list for a decode case label must be non-empty, but may have 602# one or more comma-separated integer literals in it. 603# 604def p_intlit_list_0(t): 605 'intlit_list : INTLIT' 606 t[0] = [t[1]] 607 608def p_intlit_list_1(t): 609 'intlit_list : intlit_list COMMA INTLIT' 610 t[0] = t[1] 611 t[0].append(t[3]) 612 613# Define an instruction using the current instruction format (specified 614# by an enclosing format block). 615# "<mnemonic>(<args>)" 616def p_inst_0(t): 617 'inst : ID LPAREN arg_list RPAREN' 618 # Pass the ID and arg list to the current format class to deal with. 619 currentFormat = formatStack.top() 620 codeObj = currentFormat.defineInst(t[1], t[3], t.lineno(1)) 621 args = ','.join(map(str, t[3])) 622 args = re.sub('(?m)^', '//', args) 623 args = re.sub('^//', '', args) 624 comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args) 625 codeObj.prepend_all(comment) 626 t[0] = codeObj 627 628# Define an instruction using an explicitly specified format: 629# "<fmt>::<mnemonic>(<args>)" 630def p_inst_1(t): 631 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN' 632 try: 633 format = formatMap[t[1]] 634 except KeyError: 635 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 636 codeObj = format.defineInst(t[3], t[5], t.lineno(1)) 637 comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5]) 638 codeObj.prepend_all(comment) 639 t[0] = codeObj 640 641# The arg list generates a tuple, where the first element is a list of 642# the positional args and the second element is a dict containing the 643# keyword args. 644def p_arg_list_0(t): 645 'arg_list : positional_arg_list COMMA keyword_arg_list' 646 t[0] = ( t[1], t[3] ) 647 648def p_arg_list_1(t): 649 'arg_list : positional_arg_list' 650 t[0] = ( t[1], {} ) 651 652def p_arg_list_2(t): 653 'arg_list : keyword_arg_list' 654 t[0] = ( [], t[1] ) 655 656def p_positional_arg_list_0(t): 657 'positional_arg_list : empty' 658 t[0] = [] 659 660def p_positional_arg_list_1(t): 661 'positional_arg_list : expr' 662 t[0] = [t[1]] 663 664def p_positional_arg_list_2(t): 665 'positional_arg_list : positional_arg_list COMMA expr' 666 t[0] = t[1] + [t[3]] 667 668def p_keyword_arg_list_0(t): 669 'keyword_arg_list : keyword_arg' 670 t[0] = t[1] 671 672def p_keyword_arg_list_1(t): 673 'keyword_arg_list : keyword_arg_list COMMA keyword_arg' 674 t[0] = t[1] 675 t[0].update(t[3]) 676 677def p_keyword_arg(t): 678 'keyword_arg : ID EQUALS expr' 679 t[0] = { t[1] : t[3] } 680 681# 682# Basic expressions. These constitute the argument values of 683# "function calls" (i.e. instruction definitions in the decode block) 684# and default values for formal parameters of format functions. 685# 686# Right now, these are either strings, integers, or (recursively) 687# lists of exprs (using Python square-bracket list syntax). Note that 688# bare identifiers are trated as string constants here (since there 689# isn't really a variable namespace to refer to). 690# 691def p_expr_0(t): 692 '''expr : ID 693 | INTLIT 694 | STRLIT 695 | CODELIT''' 696 t[0] = t[1] 697 698def p_expr_1(t): 699 '''expr : LBRACKET list_expr RBRACKET''' 700 t[0] = t[2] 701 702def p_list_expr_0(t): 703 'list_expr : expr' 704 t[0] = [t[1]] 705 706def p_list_expr_1(t): 707 'list_expr : list_expr COMMA expr' 708 t[0] = t[1] + [t[3]] 709 710def p_list_expr_2(t): 711 'list_expr : empty' 712 t[0] = [] 713 714# 715# Empty production... use in other rules for readability. 716# 717def p_empty(t): 718 'empty :' 719 pass 720 721# Parse error handler. Note that the argument here is the offending 722# *token*, not a grammar symbol (hence the need to use t.value) 723def p_error(t): 724 if t: 725 error(t.lineno, "syntax error at '%s'" % t.value) 726 else: 727 error(0, "unknown syntax error", True) 728 729# END OF GRAMMAR RULES 730# 731# Now build the parser. 732parser = yacc.yacc() 733 734 735##################################################################### 736# 737# Support Classes 738# 739##################################################################### 740 741# Expand template with CPU-specific references into a dictionary with 742# an entry for each CPU model name. The entry key is the model name 743# and the corresponding value is the template with the CPU-specific 744# refs substituted for that model. 745def expand_cpu_symbols_to_dict(template): 746 # Protect '%'s that don't go with CPU-specific terms 747 t = re.sub(r'%(?!\(CPU_)', '%%', template) 748 result = {} 749 for cpu in cpu_models: 750 result[cpu.name] = t % cpu.strings 751 return result 752 753# *If* the template has CPU-specific references, return a single 754# string containing a copy of the template for each CPU model with the 755# corresponding values substituted in. If the template has no 756# CPU-specific references, it is returned unmodified. 757def expand_cpu_symbols_to_string(template): 758 if template.find('%(CPU_') != -1: 759 return reduce(lambda x,y: x+y, 760 expand_cpu_symbols_to_dict(template).values()) 761 else: 762 return template 763 764# Protect CPU-specific references by doubling the corresponding '%'s 765# (in preparation for substituting a different set of references into 766# the template). 767def protect_cpu_symbols(template): 768 return re.sub(r'%(?=\(CPU_)', '%%', template) 769 770############### 771# GenCode class 772# 773# The GenCode class encapsulates generated code destined for various 774# output files. The header_output and decoder_output attributes are 775# strings containing code destined for decoder.hh and decoder.cc 776# respectively. The decode_block attribute contains code to be 777# incorporated in the decode function itself (that will also end up in 778# decoder.cc). The exec_output attribute is a dictionary with a key 779# for each CPU model name; the value associated with a particular key 780# is the string of code for that CPU model's exec.cc file. The 781# has_decode_default attribute is used in the decode block to allow 782# explicit default clauses to override default default clauses. 783 784class GenCode: 785 # Constructor. At this point we substitute out all CPU-specific 786 # symbols. For the exec output, these go into the per-model 787 # dictionary. For all other output types they get collapsed into 788 # a single string. 789 def __init__(self, 790 header_output = '', decoder_output = '', exec_output = '', 791 decode_block = '', has_decode_default = False): 792 self.header_output = expand_cpu_symbols_to_string(header_output) 793 self.decoder_output = expand_cpu_symbols_to_string(decoder_output) 794 if isinstance(exec_output, dict): 795 self.exec_output = exec_output 796 elif isinstance(exec_output, str): 797 # If the exec_output arg is a single string, we replicate 798 # it for each of the CPU models, substituting and 799 # %(CPU_foo)s params appropriately. 800 self.exec_output = expand_cpu_symbols_to_dict(exec_output) 801 self.decode_block = expand_cpu_symbols_to_string(decode_block) 802 self.has_decode_default = has_decode_default 803 804 # Override '+' operator: generate a new GenCode object that 805 # concatenates all the individual strings in the operands. 806 def __add__(self, other): 807 exec_output = {} 808 for cpu in cpu_models: 809 n = cpu.name 810 exec_output[n] = self.exec_output[n] + other.exec_output[n] 811 return GenCode(self.header_output + other.header_output, 812 self.decoder_output + other.decoder_output, 813 exec_output, 814 self.decode_block + other.decode_block, 815 self.has_decode_default or other.has_decode_default) 816 817 # Prepend a string (typically a comment) to all the strings. 818 def prepend_all(self, pre): 819 self.header_output = pre + self.header_output 820 self.decoder_output = pre + self.decoder_output 821 self.decode_block = pre + self.decode_block 822 for cpu in cpu_models: 823 self.exec_output[cpu.name] = pre + self.exec_output[cpu.name] 824 825 # Wrap the decode block in a pair of strings (e.g., 'case foo:' 826 # and 'break;'). Used to build the big nested switch statement. 827 def wrap_decode_block(self, pre, post = ''): 828 self.decode_block = pre + indent(self.decode_block) + post 829 830################ 831# Format object. 832# 833# A format object encapsulates an instruction format. It must provide 834# a defineInst() method that generates the code for an instruction 835# definition. 836 837exportContextSymbols = ('InstObjParams', 'makeList', 're', 'string') 838 839exportContext = {} 840 841def updateExportContext(): 842 exportContext.update(exportDict(*exportContextSymbols)) 843 exportContext.update(templateMap) 844 845def exportDict(*symNames): 846 return dict([(s, eval(s)) for s in symNames]) 847 848 849class Format: 850 def __init__(self, id, params, code): 851 # constructor: just save away arguments 852 self.id = id 853 self.params = params 854 label = 'def format ' + id 855 self.user_code = compile(fixPythonIndentation(code), label, 'exec') 856 param_list = string.join(params, ", ") 857 f = '''def defInst(_code, _context, %s): 858 my_locals = vars().copy() 859 exec _code in _context, my_locals 860 return my_locals\n''' % param_list 861 c = compile(f, label + ' wrapper', 'exec') 862 exec c 863 self.func = defInst 864 865 def defineInst(self, name, args, lineno): 866 context = {} 867 updateExportContext() 868 context.update(exportContext) 869 if len(name): 870 Name = name[0].upper() 871 if len(name) > 1: 872 Name += name[1:] 873 context.update({ 'name': name, 'Name': Name }) 874 try: 875 vars = self.func(self.user_code, context, *args[0], **args[1]) 876 except Exception, exc: 877 error(lineno, 'error defining "%s": %s.' % (name, exc)) 878 for k in vars.keys(): 879 if k not in ('header_output', 'decoder_output', 880 'exec_output', 'decode_block'): 881 del vars[k] 882 return GenCode(**vars) 883 884# Special null format to catch an implicit-format instruction 885# definition outside of any format block. 886class NoFormat: 887 def __init__(self): 888 self.defaultInst = '' 889 890 def defineInst(self, name, args, lineno): 891 error(lineno, 892 'instruction definition "%s" with no active format!' % name) 893 894# This dictionary maps format name strings to Format objects. 895formatMap = {} 896 897# Define a new format 898def defFormat(id, params, code, lineno): 899 # make sure we haven't already defined this one 900 if formatMap.get(id, None) != None: 901 error(lineno, 'format %s redefined.' % id) 902 # create new object and store in global map 903 formatMap[id] = Format(id, params, code) 904 905 906############## 907# Stack: a simple stack object. Used for both formats (formatStack) 908# and default cases (defaultStack). Simply wraps a list to give more 909# stack-like syntax and enable initialization with an argument list 910# (as opposed to an argument that's a list). 911 912class Stack(list): 913 def __init__(self, *items): 914 list.__init__(self, items) 915 916 def push(self, item): 917 self.append(item); 918 919 def top(self): 920 return self[-1] 921 922# The global format stack. 923formatStack = Stack(NoFormat()) 924 925# The global default case stack. 926defaultStack = Stack( None ) 927 928# Global stack that tracks current file and line number. 929# Each element is a tuple (filename, lineno) that records the 930# *current* filename and the line number in the *previous* file where 931# it was included. 932fileNameStack = Stack() 933 934################### 935# Utility functions 936 937# 938# Indent every line in string 's' by two spaces 939# (except preprocessor directives). 940# Used to make nested code blocks look pretty. 941# 942def indent(s): 943 return re.sub(r'(?m)^(?!#)', ' ', s) 944 945# 946# Munge a somewhat arbitrarily formatted piece of Python code 947# (e.g. from a format 'let' block) into something whose indentation 948# will get by the Python parser. 949# 950# The two keys here are that Python will give a syntax error if 951# there's any whitespace at the beginning of the first line, and that 952# all lines at the same lexical nesting level must have identical 953# indentation. Unfortunately the way code literals work, an entire 954# let block tends to have some initial indentation. Rather than 955# trying to figure out what that is and strip it off, we prepend 'if 956# 1:' to make the let code the nested block inside the if (and have 957# the parser automatically deal with the indentation for us). 958# 959# We don't want to do this if (1) the code block is empty or (2) the 960# first line of the block doesn't have any whitespace at the front. 961 962def fixPythonIndentation(s): 963 # get rid of blank lines first 964 s = re.sub(r'(?m)^\s*\n', '', s); 965 if (s != '' and re.match(r'[ \t]', s[0])): 966 s = 'if 1:\n' + s 967 return s 968 969# Error handler. Just call exit. Output formatted to work under 970# Emacs compile-mode. Optional 'print_traceback' arg, if set to True, 971# prints a Python stack backtrace too (can be handy when trying to 972# debug the parser itself). 973def error(lineno, string, print_traceback = False): 974 spaces = "" 975 for (filename, line) in fileNameStack[0:-1]: 976 print spaces + "In file included from " + filename + ":" 977 spaces += " " 978 # Print a Python stack backtrace if requested. 979 if (print_traceback): 980 traceback.print_exc() 981 if lineno != 0: 982 line_str = "%d:" % lineno 983 else: 984 line_str = "" 985 sys.exit(spaces + "%s:%s %s" % (fileNameStack[-1][0], line_str, string)) 986 987 988##################################################################### 989# 990# Bitfield Operator Support 991# 992##################################################################### 993 994bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>') 995 996bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>') 997bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>') 998 999def substBitOps(code): 1000 # first convert single-bit selectors to two-index form 1001 # i.e., <n> --> <n:n> 1002 code = bitOp1ArgRE.sub(r'<\1:\1>', code) 1003 # simple case: selector applied to ID (name) 1004 # i.e., foo<a:b> --> bits(foo, a, b) 1005 code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code) 1006 # if selector is applied to expression (ending in ')'), 1007 # we need to search backward for matching '(' 1008 match = bitOpExprRE.search(code) 1009 while match: 1010 exprEnd = match.start() 1011 here = exprEnd - 1 1012 nestLevel = 1 1013 while nestLevel > 0: 1014 if code[here] == '(': 1015 nestLevel -= 1 1016 elif code[here] == ')': 1017 nestLevel += 1 1018 here -= 1 1019 if here < 0: 1020 sys.exit("Didn't find '('!") 1021 exprStart = here+1 1022 newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1], 1023 match.group(1), match.group(2)) 1024 code = code[:exprStart] + newExpr + code[match.end():] 1025 match = bitOpExprRE.search(code) 1026 return code 1027 1028 1029#################### 1030# Template objects. 1031# 1032# Template objects are format strings that allow substitution from 1033# the attribute spaces of other objects (e.g. InstObjParams instances). 1034 1035labelRE = re.compile(r'(?<!%)%\(([^\)]+)\)[sd]') 1036 1037class Template: 1038 def __init__(self, t): 1039 self.template = t 1040 1041 def subst(self, d): 1042 myDict = None 1043 1044 # Protect non-Python-dict substitutions (e.g. if there's a printf 1045 # in the templated C++ code) 1046 template = protect_non_subst_percents(self.template) 1047 # CPU-model-specific substitutions are handled later (in GenCode). 1048 template = protect_cpu_symbols(template) 1049 1050 # Build a dict ('myDict') to use for the template substitution. 1051 # Start with the template namespace. Make a copy since we're 1052 # going to modify it. 1053 myDict = templateMap.copy() 1054 1055 if isinstance(d, InstObjParams): 1056 # If we're dealing with an InstObjParams object, we need 1057 # to be a little more sophisticated. The instruction-wide 1058 # parameters are already formed, but the parameters which 1059 # are only function wide still need to be generated. 1060 compositeCode = '' 1061 1062 myDict.update(d.__dict__) 1063 # The "operands" and "snippets" attributes of the InstObjParams 1064 # objects are for internal use and not substitution. 1065 del myDict['operands'] 1066 del myDict['snippets'] 1067 1068 snippetLabels = [l for l in labelRE.findall(template) 1069 if d.snippets.has_key(l)] 1070 1071 snippets = dict([(s, mungeSnippet(d.snippets[s])) 1072 for s in snippetLabels]) 1073 1074 myDict.update(snippets) 1075 1076 compositeCode = ' '.join(map(str, snippets.values())) 1077 1078 # Add in template itself in case it references any 1079 # operands explicitly (like Mem) 1080 compositeCode += ' ' + template 1081 1082 operands = SubOperandList(compositeCode, d.operands) 1083 1084 myDict['op_decl'] = operands.concatAttrStrings('op_decl') 1085 1086 is_src = lambda op: op.is_src 1087 is_dest = lambda op: op.is_dest 1088 1089 myDict['op_src_decl'] = \ 1090 operands.concatSomeAttrStrings(is_src, 'op_src_decl') 1091 myDict['op_dest_decl'] = \ 1092 operands.concatSomeAttrStrings(is_dest, 'op_dest_decl') 1093 1094 myDict['op_rd'] = operands.concatAttrStrings('op_rd') 1095 myDict['op_wb'] = operands.concatAttrStrings('op_wb') 1096 1097 if d.operands.memOperand: 1098 myDict['mem_acc_size'] = d.operands.memOperand.mem_acc_size 1099 myDict['mem_acc_type'] = d.operands.memOperand.mem_acc_type 1100 1101 elif isinstance(d, dict): 1102 # if the argument is a dictionary, we just use it. 1103 myDict.update(d) 1104 elif hasattr(d, '__dict__'): 1105 # if the argument is an object, we use its attribute map. 1106 myDict.update(d.__dict__) 1107 else: 1108 raise TypeError, "Template.subst() arg must be or have dictionary" 1109 return template % myDict 1110 1111 # Convert to string. This handles the case when a template with a 1112 # CPU-specific term gets interpolated into another template or into 1113 # an output block. 1114 def __str__(self): 1115 return expand_cpu_symbols_to_string(self.template) 1116 1117##################################################################### 1118# 1119# Code Parser 1120# 1121# The remaining code is the support for automatically extracting 1122# instruction characteristics from pseudocode. 1123# 1124##################################################################### 1125 1126# Force the argument to be a list. Useful for flags, where a caller 1127# can specify a singleton flag or a list of flags. Also usful for 1128# converting tuples to lists so they can be modified. 1129def makeList(arg): 1130 if isinstance(arg, list): 1131 return arg 1132 elif isinstance(arg, tuple): 1133 return list(arg) 1134 elif not arg: 1135 return [] 1136 else: 1137 return [ arg ] 1138 1139# Generate operandTypeMap from the user's 'def operand_types' 1140# statement. 1141def buildOperandTypeMap(userDict, lineno): 1142 global operandTypeMap 1143 operandTypeMap = {} 1144 for (ext, (desc, size)) in userDict.iteritems(): 1145 if desc == 'signed int': 1146 ctype = 'int%d_t' % size 1147 is_signed = 1 1148 elif desc == 'unsigned int': 1149 ctype = 'uint%d_t' % size 1150 is_signed = 0 1151 elif desc == 'float': 1152 is_signed = 1 # shouldn't really matter 1153 if size == 32: 1154 ctype = 'float' 1155 elif size == 64: 1156 ctype = 'double' 1157 elif desc == 'twin64 int': 1158 is_signed = 0 1159 ctype = 'Twin64_t' 1160 elif desc == 'twin32 int': 1161 is_signed = 0 1162 ctype = 'Twin32_t' 1163 if ctype == '': 1164 error(lineno, 'Unrecognized type description "%s" in userDict') 1165 operandTypeMap[ext] = (size, ctype, is_signed) 1166 1167# 1168# 1169# 1170# Base class for operand descriptors. An instance of this class (or 1171# actually a class derived from this one) represents a specific 1172# operand for a code block (e.g, "Rc.sq" as a dest). Intermediate 1173# derived classes encapsulates the traits of a particular operand type 1174# (e.g., "32-bit integer register"). 1175# 1176class Operand(object): 1177 def __init__(self, full_name, ext, is_src, is_dest): 1178 self.full_name = full_name 1179 self.ext = ext 1180 self.is_src = is_src 1181 self.is_dest = is_dest 1182 # The 'effective extension' (eff_ext) is either the actual 1183 # extension, if one was explicitly provided, or the default. 1184 if ext: 1185 self.eff_ext = ext 1186 else: 1187 self.eff_ext = self.dflt_ext 1188 1189 (self.size, self.ctype, self.is_signed) = operandTypeMap[self.eff_ext] 1190 1191 # note that mem_acc_size is undefined for non-mem operands... 1192 # template must be careful not to use it if it doesn't apply. 1193 if self.isMem(): 1194 self.mem_acc_size = self.makeAccSize() 1195 if self.ctype in ['Twin32_t', 'Twin64_t']: 1196 self.mem_acc_type = 'Twin' 1197 else: 1198 self.mem_acc_type = 'uint' 1199 1200 # Finalize additional fields (primarily code fields). This step 1201 # is done separately since some of these fields may depend on the 1202 # register index enumeration that hasn't been performed yet at the 1203 # time of __init__(). 1204 def finalize(self): 1205 self.flags = self.getFlags() 1206 self.constructor = self.makeConstructor() 1207 self.op_decl = self.makeDecl() 1208 1209 if self.is_src: 1210 self.op_rd = self.makeRead() 1211 self.op_src_decl = self.makeDecl() 1212 else: 1213 self.op_rd = '' 1214 self.op_src_decl = '' 1215 1216 if self.is_dest: 1217 self.op_wb = self.makeWrite() 1218 self.op_dest_decl = self.makeDecl() 1219 else: 1220 self.op_wb = '' 1221 self.op_dest_decl = '' 1222 1223 def isMem(self): 1224 return 0 1225 1226 def isReg(self): 1227 return 0 1228 1229 def isFloatReg(self): 1230 return 0 1231 1232 def isIntReg(self): 1233 return 0 1234 1235 def isControlReg(self): 1236 return 0 1237 1238 def getFlags(self): 1239 # note the empty slice '[:]' gives us a copy of self.flags[0] 1240 # instead of a reference to it 1241 my_flags = self.flags[0][:] 1242 if self.is_src: 1243 my_flags += self.flags[1] 1244 if self.is_dest: 1245 my_flags += self.flags[2] 1246 return my_flags 1247 1248 def makeDecl(self): 1249 # Note that initializations in the declarations are solely 1250 # to avoid 'uninitialized variable' errors from the compiler. 1251 return self.ctype + ' ' + self.base_name + ' = 0;\n'; 1252 1253class IntRegOperand(Operand): 1254 def isReg(self): 1255 return 1 1256 1257 def isIntReg(self): 1258 return 1 1259 1260 def makeConstructor(self): 1261 c = '' 1262 if self.is_src: 1263 c += '\n\t_srcRegIdx[%d] = %s;' % \ 1264 (self.src_reg_idx, self.reg_spec) 1265 if self.is_dest: 1266 c += '\n\t_destRegIdx[%d] = %s;' % \ 1267 (self.dest_reg_idx, self.reg_spec) 1268 return c 1269 1270 def makeRead(self): 1271 if (self.ctype == 'float' or self.ctype == 'double'): 1272 error(0, 'Attempt to read integer register as FP') 1273 if (self.size == self.dflt_size): 1274 return '%s = xc->readIntRegOperand(this, %d);\n' % \ 1275 (self.base_name, self.src_reg_idx) 1276 elif (self.size > self.dflt_size): 1277 int_reg_val = 'xc->readIntRegOperand(this, %d)' % \ 1278 (self.src_reg_idx) 1279 if (self.is_signed): 1280 int_reg_val = 'sext<%d>(%s)' % (self.dflt_size, int_reg_val) 1281 return '%s = %s;\n' % (self.base_name, int_reg_val) 1282 else: 1283 return '%s = bits(xc->readIntRegOperand(this, %d), %d, 0);\n' % \ 1284 (self.base_name, self.src_reg_idx, self.size-1) 1285 1286 def makeWrite(self): 1287 if (self.ctype == 'float' or self.ctype == 'double'): 1288 error(0, 'Attempt to write integer register as FP') 1289 if (self.size != self.dflt_size and self.is_signed): 1290 final_val = 'sext<%d>(%s)' % (self.size, self.base_name) 1291 else: 1292 final_val = self.base_name 1293 wb = ''' 1294 { 1295 %s final_val = %s; 1296 xc->setIntRegOperand(this, %d, final_val);\n 1297 if (traceData) { traceData->setData(final_val); } 1298 }''' % (self.dflt_ctype, final_val, self.dest_reg_idx) 1299 return wb 1300 1301class FloatRegOperand(Operand): 1302 def isReg(self): 1303 return 1 1304 1305 def isFloatReg(self): 1306 return 1 1307 1308 def makeConstructor(self): 1309 c = '' 1310 if self.is_src: 1311 c += '\n\t_srcRegIdx[%d] = %s + FP_Base_DepTag;' % \ 1312 (self.src_reg_idx, self.reg_spec) 1313 if self.is_dest: 1314 c += '\n\t_destRegIdx[%d] = %s + FP_Base_DepTag;' % \ 1315 (self.dest_reg_idx, self.reg_spec) 1316 return c 1317 1318 def makeRead(self): 1319 bit_select = 0 1320 width = 0; 1321 if (self.ctype == 'float'): 1322 func = 'readFloatRegOperand' 1323 width = 32; 1324 elif (self.ctype == 'double'): 1325 func = 'readFloatRegOperand' 1326 width = 64; 1327 else: 1328 func = 'readFloatRegOperandBits' 1329 if (self.ctype == 'uint32_t'): 1330 width = 32; 1331 elif (self.ctype == 'uint64_t'): 1332 width = 64; 1333 if (self.size != self.dflt_size): 1334 bit_select = 1 1335 if width: 1336 base = 'xc->%s(this, %d, %d)' % \ 1337 (func, self.src_reg_idx, width) 1338 else: 1339 base = 'xc->%s(this, %d)' % \ 1340 (func, self.src_reg_idx) 1341 if bit_select: 1342 return '%s = bits(%s, %d, 0);\n' % \ 1343 (self.base_name, base, self.size-1) 1344 else: 1345 return '%s = %s;\n' % (self.base_name, base) 1346 1347 def makeWrite(self): 1348 final_val = self.base_name 1349 final_ctype = self.ctype 1350 widthSpecifier = '' 1351 width = 0 1352 if (self.ctype == 'float'): 1353 width = 32 1354 func = 'setFloatRegOperand' 1355 elif (self.ctype == 'double'): 1356 width = 64 1357 func = 'setFloatRegOperand' 1358 elif (self.ctype == 'uint32_t'): 1359 func = 'setFloatRegOperandBits' 1360 width = 32 1361 elif (self.ctype == 'uint64_t'): 1362 func = 'setFloatRegOperandBits' 1363 width = 64 1364 else: 1365 func = 'setFloatRegOperandBits' 1366 final_ctype = 'uint%d_t' % self.dflt_size 1367 if (self.size != self.dflt_size and self.is_signed): 1368 final_val = 'sext<%d>(%s)' % (self.size, self.base_name) 1369 if width: 1370 widthSpecifier = ', %d' % width 1371 wb = ''' 1372 { 1373 %s final_val = %s; 1374 xc->%s(this, %d, final_val%s);\n 1375 if (traceData) { traceData->setData(final_val); } 1376 }''' % (final_ctype, final_val, func, self.dest_reg_idx, 1377 widthSpecifier) 1378 return wb 1379 1380class ControlRegOperand(Operand): 1381 def isReg(self): 1382 return 1 1383 1384 def isControlReg(self): 1385 return 1 1386 1387 def makeConstructor(self): 1388 c = '' 1389 if self.is_src: 1390 c += '\n\t_srcRegIdx[%d] = %s + Ctrl_Base_DepTag;' % \ 1391 (self.src_reg_idx, self.reg_spec) 1392 if self.is_dest: 1393 c += '\n\t_destRegIdx[%d] = %s + Ctrl_Base_DepTag;' % \ 1394 (self.dest_reg_idx, self.reg_spec) 1395 return c 1396 1397 def makeRead(self): 1398 bit_select = 0 1399 if (self.ctype == 'float' or self.ctype == 'double'): 1400 error(0, 'Attempt to read control register as FP') 1401 base = 'xc->readMiscRegOperand(this, %s)' % self.src_reg_idx 1402 if self.size == self.dflt_size: 1403 return '%s = %s;\n' % (self.base_name, base) 1404 else: 1405 return '%s = bits(%s, %d, 0);\n' % \ 1406 (self.base_name, base, self.size-1) 1407 1408 def makeWrite(self): 1409 if (self.ctype == 'float' or self.ctype == 'double'): 1410 error(0, 'Attempt to write control register as FP') 1411 wb = 'xc->setMiscRegOperand(this, %s, %s);\n' % \ 1412 (self.dest_reg_idx, self.base_name) 1413 wb += 'if (traceData) { traceData->setData(%s); }' % \ 1414 self.base_name 1415 return wb 1416 1417class MemOperand(Operand): 1418 def isMem(self): 1419 return 1 1420 1421 def makeConstructor(self): 1422 return '' 1423 1424 def makeDecl(self): 1425 # Note that initializations in the declarations are solely 1426 # to avoid 'uninitialized variable' errors from the compiler. 1427 # Declare memory data variable. 1428 if self.ctype in ['Twin32_t','Twin64_t']: 1429 return "%s %s; %s.a = 0; %s.b = 0;\n" % (self.ctype, self.base_name, 1430 self.base_name, self.base_name) 1431 c = '%s %s = 0;\n' % (self.ctype, self.base_name) 1432 return c 1433 1434 def makeRead(self): 1435 return '' 1436 1437 def makeWrite(self): 1438 return '' 1439 1440 # Return the memory access size *in bits*, suitable for 1441 # forming a type via "uint%d_t". Divide by 8 if you want bytes. 1442 def makeAccSize(self): 1443 return self.size 1444 1445 1446class NPCOperand(Operand): 1447 def makeConstructor(self): 1448 return '' 1449 1450 def makeRead(self): 1451 return '%s = xc->readNextPC();\n' % self.base_name 1452 1453 def makeWrite(self): 1454 return 'xc->setNextPC(%s);\n' % self.base_name 1455 1456class NNPCOperand(Operand): 1457 def makeConstructor(self): 1458 return '' 1459 1460 def makeRead(self): 1461 return '%s = xc->readNextNPC();\n' % self.base_name 1462 1463 def makeWrite(self): 1464 return 'xc->setNextNPC(%s);\n' % self.base_name 1465 1466def buildOperandNameMap(userDict, lineno): 1467 global operandNameMap 1468 operandNameMap = {} 1469 for (op_name, val) in userDict.iteritems(): 1470 (base_cls_name, dflt_ext, reg_spec, flags, sort_pri) = val 1471 (dflt_size, dflt_ctype, dflt_is_signed) = operandTypeMap[dflt_ext] 1472 # Canonical flag structure is a triple of lists, where each list 1473 # indicates the set of flags implied by this operand always, when 1474 # used as a source, and when used as a dest, respectively. 1475 # For simplicity this can be initialized using a variety of fairly 1476 # obvious shortcuts; we convert these to canonical form here. 1477 if not flags: 1478 # no flags specified (e.g., 'None') 1479 flags = ( [], [], [] ) 1480 elif isinstance(flags, str): 1481 # a single flag: assumed to be unconditional 1482 flags = ( [ flags ], [], [] ) 1483 elif isinstance(flags, list): 1484 # a list of flags: also assumed to be unconditional 1485 flags = ( flags, [], [] ) 1486 elif isinstance(flags, tuple): 1487 # it's a tuple: it should be a triple, 1488 # but each item could be a single string or a list 1489 (uncond_flags, src_flags, dest_flags) = flags 1490 flags = (makeList(uncond_flags), 1491 makeList(src_flags), makeList(dest_flags)) 1492 # Accumulate attributes of new operand class in tmp_dict 1493 tmp_dict = {} 1494 for attr in ('dflt_ext', 'reg_spec', 'flags', 'sort_pri', 1495 'dflt_size', 'dflt_ctype', 'dflt_is_signed'): 1496 tmp_dict[attr] = eval(attr) 1497 tmp_dict['base_name'] = op_name 1498 # New class name will be e.g. "IntReg_Ra" 1499 cls_name = base_cls_name + '_' + op_name 1500 # Evaluate string arg to get class object. Note that the 1501 # actual base class for "IntReg" is "IntRegOperand", i.e. we 1502 # have to append "Operand". 1503 try: 1504 base_cls = eval(base_cls_name + 'Operand') 1505 except NameError: 1506 error(lineno, 1507 'error: unknown operand base class "%s"' % base_cls_name) 1508 # The following statement creates a new class called 1509 # <cls_name> as a subclass of <base_cls> with the attributes 1510 # in tmp_dict, just as if we evaluated a class declaration. 1511 operandNameMap[op_name] = type(cls_name, (base_cls,), tmp_dict) 1512 1513 # Define operand variables. 1514 operands = userDict.keys() 1515 1516 operandsREString = (r''' 1517 (?<![\w\.]) # neg. lookbehind assertion: prevent partial matches 1518 ((%s)(?:\.(\w+))?) # match: operand with optional '.' then suffix 1519 (?![\w\.]) # neg. lookahead assertion: prevent partial matches 1520 ''' 1521 % string.join(operands, '|')) 1522 1523 global operandsRE 1524 operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE) 1525 1526 # Same as operandsREString, but extension is mandatory, and only two 1527 # groups are returned (base and ext, not full name as above). 1528 # Used for subtituting '_' for '.' to make C++ identifiers. 1529 operandsWithExtREString = (r'(?<![\w\.])(%s)\.(\w+)(?![\w\.])' 1530 % string.join(operands, '|')) 1531 1532 global operandsWithExtRE 1533 operandsWithExtRE = re.compile(operandsWithExtREString, re.MULTILINE) 1534 1535 1536class OperandList: 1537 1538 # Find all the operands in the given code block. Returns an operand 1539 # descriptor list (instance of class OperandList). 1540 def __init__(self, code): 1541 self.items = [] 1542 self.bases = {} 1543 # delete comments so we don't match on reg specifiers inside 1544 code = commentRE.sub('', code) 1545 # search for operands 1546 next_pos = 0 1547 while 1: 1548 match = operandsRE.search(code, next_pos) 1549 if not match: 1550 # no more matches: we're done 1551 break 1552 op = match.groups() 1553 # regexp groups are operand full name, base, and extension 1554 (op_full, op_base, op_ext) = op 1555 # if the token following the operand is an assignment, this is 1556 # a destination (LHS), else it's a source (RHS) 1557 is_dest = (assignRE.match(code, match.end()) != None) 1558 is_src = not is_dest 1559 # see if we've already seen this one 1560 op_desc = self.find_base(op_base) 1561 if op_desc: 1562 if op_desc.ext != op_ext: 1563 error(0, 'Inconsistent extensions for operand %s' % \ 1564 op_base) 1565 op_desc.is_src = op_desc.is_src or is_src 1566 op_desc.is_dest = op_desc.is_dest or is_dest 1567 else: 1568 # new operand: create new descriptor 1569 op_desc = operandNameMap[op_base](op_full, op_ext, 1570 is_src, is_dest) 1571 self.append(op_desc) 1572 # start next search after end of current match 1573 next_pos = match.end() 1574 self.sort() 1575 # enumerate source & dest register operands... used in building 1576 # constructor later 1577 self.numSrcRegs = 0 1578 self.numDestRegs = 0 1579 self.numFPDestRegs = 0 1580 self.numIntDestRegs = 0 1581 self.memOperand = None 1582 for op_desc in self.items: 1583 if op_desc.isReg(): 1584 if op_desc.is_src: 1585 op_desc.src_reg_idx = self.numSrcRegs 1586 self.numSrcRegs += 1 1587 if op_desc.is_dest: 1588 op_desc.dest_reg_idx = self.numDestRegs 1589 self.numDestRegs += 1 1590 if op_desc.isFloatReg(): 1591 self.numFPDestRegs += 1 1592 elif op_desc.isIntReg(): 1593 self.numIntDestRegs += 1 1594 elif op_desc.isMem(): 1595 if self.memOperand: 1596 error(0, "Code block has more than one memory operand.") 1597 self.memOperand = op_desc 1598 # now make a final pass to finalize op_desc fields that may depend 1599 # on the register enumeration 1600 for op_desc in self.items: 1601 op_desc.finalize() 1602 1603 def __len__(self): 1604 return len(self.items) 1605 1606 def __getitem__(self, index): 1607 return self.items[index] 1608 1609 def append(self, op_desc): 1610 self.items.append(op_desc) 1611 self.bases[op_desc.base_name] = op_desc 1612 1613 def find_base(self, base_name): 1614 # like self.bases[base_name], but returns None if not found 1615 # (rather than raising exception) 1616 return self.bases.get(base_name) 1617 1618 # internal helper function for concat[Some]Attr{Strings|Lists} 1619 def __internalConcatAttrs(self, attr_name, filter, result): 1620 for op_desc in self.items: 1621 if filter(op_desc): 1622 result += getattr(op_desc, attr_name) 1623 return result 1624 1625 # return a single string that is the concatenation of the (string) 1626 # values of the specified attribute for all operands 1627 def concatAttrStrings(self, attr_name): 1628 return self.__internalConcatAttrs(attr_name, lambda x: 1, '') 1629 1630 # like concatAttrStrings, but only include the values for the operands 1631 # for which the provided filter function returns true 1632 def concatSomeAttrStrings(self, filter, attr_name): 1633 return self.__internalConcatAttrs(attr_name, filter, '') 1634 1635 # return a single list that is the concatenation of the (list) 1636 # values of the specified attribute for all operands 1637 def concatAttrLists(self, attr_name): 1638 return self.__internalConcatAttrs(attr_name, lambda x: 1, []) 1639 1640 # like concatAttrLists, but only include the values for the operands 1641 # for which the provided filter function returns true 1642 def concatSomeAttrLists(self, filter, attr_name): 1643 return self.__internalConcatAttrs(attr_name, filter, []) 1644 1645 def sort(self): 1646 self.items.sort(lambda a, b: a.sort_pri - b.sort_pri) 1647 1648class SubOperandList(OperandList): 1649 1650 # Find all the operands in the given code block. Returns an operand 1651 # descriptor list (instance of class OperandList). 1652 def __init__(self, code, master_list): 1653 self.items = [] 1654 self.bases = {} 1655 # delete comments so we don't match on reg specifiers inside 1656 code = commentRE.sub('', code) 1657 # search for operands 1658 next_pos = 0 1659 while 1: 1660 match = operandsRE.search(code, next_pos) 1661 if not match: 1662 # no more matches: we're done 1663 break 1664 op = match.groups() 1665 # regexp groups are operand full name, base, and extension 1666 (op_full, op_base, op_ext) = op 1667 # find this op in the master list 1668 op_desc = master_list.find_base(op_base) 1669 if not op_desc: 1670 error(0, 'Found operand %s which is not in the master list!' \ 1671 ' This is an internal error' % \ 1672 op_base) 1673 else: 1674 # See if we've already found this operand 1675 op_desc = self.find_base(op_base) 1676 if not op_desc: 1677 # if not, add a reference to it to this sub list 1678 self.append(master_list.bases[op_base]) 1679 1680 # start next search after end of current match 1681 next_pos = match.end() 1682 self.sort() 1683 self.memOperand = None 1684 for op_desc in self.items: 1685 if op_desc.isMem(): 1686 if self.memOperand: 1687 error(0, "Code block has more than one memory operand.") 1688 self.memOperand = op_desc 1689 1690# Regular expression object to match C++ comments 1691# (used in findOperands()) 1692commentRE = re.compile(r'//.*\n') 1693 1694# Regular expression object to match assignment statements 1695# (used in findOperands()) 1696assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE) 1697 1698# Munge operand names in code string to make legal C++ variable names. 1699# This means getting rid of the type extension if any. 1700# (Will match base_name attribute of Operand object.) 1701def substMungedOpNames(code): 1702 return operandsWithExtRE.sub(r'\1', code) 1703 1704# Fix up code snippets for final substitution in templates. 1705def mungeSnippet(s): 1706 if isinstance(s, str): 1707 return substMungedOpNames(substBitOps(s)) 1708 else: 1709 return s 1710 1711def makeFlagConstructor(flag_list): 1712 if len(flag_list) == 0: 1713 return '' 1714 # filter out repeated flags 1715 flag_list.sort() 1716 i = 1 1717 while i < len(flag_list): 1718 if flag_list[i] == flag_list[i-1]: 1719 del flag_list[i] 1720 else: 1721 i += 1 1722 pre = '\n\tflags[' 1723 post = '] = true;' 1724 code = pre + string.join(flag_list, post + pre) + post 1725 return code 1726 1727# Assume all instruction flags are of the form 'IsFoo' 1728instFlagRE = re.compile(r'Is.*') 1729 1730# OpClass constants end in 'Op' except No_OpClass 1731opClassRE = re.compile(r'.*Op|No_OpClass') 1732 1733class InstObjParams: 1734 def __init__(self, mnem, class_name, base_class = '', 1735 snippets = {}, opt_args = []): 1736 self.mnemonic = mnem 1737 self.class_name = class_name 1738 self.base_class = base_class 1739 if not isinstance(snippets, dict): 1740 snippets = {'code' : snippets} 1741 compositeCode = ' '.join(map(str, snippets.values())) 1742 self.snippets = snippets 1743 1744 self.operands = OperandList(compositeCode) 1745 self.constructor = self.operands.concatAttrStrings('constructor') 1746 self.constructor += \ 1747 '\n\t_numSrcRegs = %d;' % self.operands.numSrcRegs 1748 self.constructor += \ 1749 '\n\t_numDestRegs = %d;' % self.operands.numDestRegs 1750 self.constructor += \ 1751 '\n\t_numFPDestRegs = %d;' % self.operands.numFPDestRegs 1752 self.constructor += \ 1753 '\n\t_numIntDestRegs = %d;' % self.operands.numIntDestRegs 1754 self.flags = self.operands.concatAttrLists('flags') 1755 1756 # Make a basic guess on the operand class (function unit type). 1757 # These are good enough for most cases, and can be overridden 1758 # later otherwise. 1759 if 'IsStore' in self.flags: 1760 self.op_class = 'MemWriteOp' 1761 elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags: 1762 self.op_class = 'MemReadOp' 1763 elif 'IsFloating' in self.flags: 1764 self.op_class = 'FloatAddOp' 1765 else: 1766 self.op_class = 'IntAluOp' 1767 1768 # Optional arguments are assumed to be either StaticInst flags 1769 # or an OpClass value. To avoid having to import a complete 1770 # list of these values to match against, we do it ad-hoc 1771 # with regexps. 1772 for oa in opt_args: 1773 if instFlagRE.match(oa): 1774 self.flags.append(oa) 1775 elif opClassRE.match(oa): 1776 self.op_class = oa 1777 else: 1778 error(0, 'InstObjParams: optional arg "%s" not recognized ' 1779 'as StaticInst::Flag or OpClass.' % oa) 1780 1781 # add flag initialization to contructor here to include 1782 # any flags added via opt_args 1783 self.constructor += makeFlagConstructor(self.flags) 1784 1785 # if 'IsFloating' is set, add call to the FP enable check 1786 # function (which should be provided by isa_desc via a declare) 1787 if 'IsFloating' in self.flags: 1788 self.fp_enable_check = 'fault = checkFpEnableFault(xc);' 1789 else: 1790 self.fp_enable_check = '' 1791 1792####################### 1793# 1794# Output file template 1795# 1796 1797file_template = ''' 1798/* 1799 * DO NOT EDIT THIS FILE!!! 1800 * 1801 * It was automatically generated from the ISA description in %(filename)s 1802 */ 1803 1804%(includes)s 1805 1806%(global_output)s 1807 1808namespace %(namespace)s { 1809 1810%(namespace_output)s 1811 1812} // namespace %(namespace)s 1813 1814%(decode_function)s 1815''' 1816 1817 1818# Update the output file only if the new contents are different from 1819# the current contents. Minimizes the files that need to be rebuilt 1820# after minor changes. 1821def update_if_needed(file, contents): 1822 update = False 1823 if os.access(file, os.R_OK): 1824 f = open(file, 'r') 1825 old_contents = f.read() 1826 f.close() 1827 if contents != old_contents: 1828 print 'Updating', file 1829 os.remove(file) # in case it's write-protected 1830 update = True 1831 else: 1832 print 'File', file, 'is unchanged' 1833 else: 1834 print 'Generating', file 1835 update = True 1836 if update: 1837 f = open(file, 'w') 1838 f.write(contents) 1839 f.close() 1840 1841# This regular expression matches '##include' directives 1842includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[\w/.-]*)".*$', 1843 re.MULTILINE) 1844 1845# Function to replace a matched '##include' directive with the 1846# contents of the specified file (with nested ##includes replaced 1847# recursively). 'matchobj' is an re match object (from a match of 1848# includeRE) and 'dirname' is the directory relative to which the file 1849# path should be resolved. 1850def replace_include(matchobj, dirname): 1851 fname = matchobj.group('filename') 1852 full_fname = os.path.normpath(os.path.join(dirname, fname)) 1853 contents = '##newfile "%s"\n%s\n##endfile\n' % \ 1854 (full_fname, read_and_flatten(full_fname)) 1855 return contents 1856 1857# Read a file and recursively flatten nested '##include' files. 1858def read_and_flatten(filename): 1859 current_dir = os.path.dirname(filename) 1860 try: 1861 contents = open(filename).read() 1862 except IOError: 1863 error(0, 'Error including file "%s"' % filename) 1864 fileNameStack.push((filename, 0)) 1865 # Find any includes and include them 1866 contents = includeRE.sub(lambda m: replace_include(m, current_dir), 1867 contents) 1868 fileNameStack.pop() 1869 return contents 1870 1871# 1872# Read in and parse the ISA description. 1873# 1874def parse_isa_desc(isa_desc_file, output_dir): 1875 # Read file and (recursively) all included files into a string. 1876 # PLY requires that the input be in a single string so we have to 1877 # do this up front. 1878 isa_desc = read_and_flatten(isa_desc_file) 1879 1880 # Initialize filename stack with outer file. 1881 fileNameStack.push((isa_desc_file, 0)) 1882 1883 # Parse it. 1884 (isa_name, namespace, global_code, namespace_code) = \ 1885 parser.parse(isa_desc, lexer=lexer) 1886 1887 # grab the last three path components of isa_desc_file to put in 1888 # the output 1889 filename = '/'.join(isa_desc_file.split('/')[-3:]) 1890 1891 # generate decoder.hh 1892 includes = '#include "base/bitfield.hh" // for bitfield support' 1893 global_output = global_code.header_output 1894 namespace_output = namespace_code.header_output 1895 decode_function = '' 1896 update_if_needed(output_dir + '/decoder.hh', file_template % vars()) 1897 1898 # generate decoder.cc 1899 includes = '#include "decoder.hh"' 1900 global_output = global_code.decoder_output 1901 namespace_output = namespace_code.decoder_output 1902 # namespace_output += namespace_code.decode_block 1903 decode_function = namespace_code.decode_block 1904 update_if_needed(output_dir + '/decoder.cc', file_template % vars()) 1905 1906 # generate per-cpu exec files 1907 for cpu in cpu_models: 1908 includes = '#include "decoder.hh"\n' 1909 includes += cpu.includes 1910 global_output = global_code.exec_output[cpu.name] 1911 namespace_output = namespace_code.exec_output[cpu.name] 1912 decode_function = '' 1913 update_if_needed(output_dir + '/' + cpu.filename, 1914 file_template % vars()) 1915 1916# global list of CpuModel objects (see cpu_models.py) 1917cpu_models = [] 1918 1919# Called as script: get args from command line. 1920# Args are: <path to cpu_models.py> <isa desc file> <output dir> <cpu models> 1921if __name__ == '__main__': 1922 execfile(sys.argv[1]) # read in CpuModel definitions 1923 cpu_models = [CpuModel.dict[cpu] for cpu in sys.argv[4:]] 1924 parse_isa_desc(sys.argv[2], sys.argv[3]) 1925