isa_parser.py revision 4662:96fb82f4f3d5
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# Gabe Black 29# Korey Sewell 30 31import os 32import sys 33import re 34import string 35import traceback 36# get type names 37from types import * 38 39# Prepend the directory where the PLY lex & yacc modules are found 40# to the search path. Assumes we're compiling in a subdirectory 41# of 'build' in the current tree. 42sys.path[0:0] = [os.environ['M5_PLY']] 43 44from ply import lex 45from ply import yacc 46 47##################################################################### 48# 49# Lexer 50# 51# The PLY lexer module takes two things as input: 52# - A list of token names (the string list 'tokens') 53# - A regular expression describing a match for each token. The 54# regexp for token FOO can be provided in two ways: 55# - as a string variable named t_FOO 56# - as the doc string for a function named t_FOO. In this case, 57# the function is also executed, allowing an action to be 58# associated with each token match. 59# 60##################################################################### 61 62# Reserved words. These are listed separately as they are matched 63# using the same regexp as generic IDs, but distinguished in the 64# t_ID() function. The PLY documentation suggests this approach. 65reserved = ( 66 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT', 67 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS', 68 'OUTPUT', 'SIGNED', 'TEMPLATE' 69 ) 70 71# List of tokens. The lex module requires this. 72tokens = reserved + ( 73 # identifier 74 'ID', 75 76 # integer literal 77 'INTLIT', 78 79 # string literal 80 'STRLIT', 81 82 # code literal 83 'CODELIT', 84 85 # ( ) [ ] { } < > , ; . : :: * 86 'LPAREN', 'RPAREN', 87 'LBRACKET', 'RBRACKET', 88 'LBRACE', 'RBRACE', 89 'LESS', 'GREATER', 'EQUALS', 90 'COMMA', 'SEMI', 'DOT', 'COLON', 'DBLCOLON', 91 'ASTERISK', 92 93 # C preprocessor directives 94 'CPPDIRECTIVE' 95 96# The following are matched but never returned. commented out to 97# suppress PLY warning 98 # newfile directive 99# 'NEWFILE', 100 101 # endfile directive 102# 'ENDFILE' 103) 104 105# Regular expressions for token matching 106t_LPAREN = r'\(' 107t_RPAREN = r'\)' 108t_LBRACKET = r'\[' 109t_RBRACKET = r'\]' 110t_LBRACE = r'\{' 111t_RBRACE = r'\}' 112t_LESS = r'\<' 113t_GREATER = r'\>' 114t_EQUALS = r'=' 115t_COMMA = r',' 116t_SEMI = r';' 117t_DOT = r'\.' 118t_COLON = r':' 119t_DBLCOLON = r'::' 120t_ASTERISK = r'\*' 121 122# Identifiers and reserved words 123reserved_map = { } 124for r in reserved: 125 reserved_map[r.lower()] = r 126 127def t_ID(t): 128 r'[A-Za-z_]\w*' 129 t.type = reserved_map.get(t.value,'ID') 130 return t 131 132# Integer literal 133def t_INTLIT(t): 134 r'(0x[\da-fA-F]+)|\d+' 135 try: 136 t.value = int(t.value,0) 137 except ValueError: 138 error(t.lineno, 'Integer value "%s" too large' % t.value) 139 t.value = 0 140 return t 141 142# String literal. Note that these use only single quotes, and 143# can span multiple lines. 144def t_STRLIT(t): 145 r"(?m)'([^'])+'" 146 # strip off quotes 147 t.value = t.value[1:-1] 148 t.lineno += t.value.count('\n') 149 return t 150 151 152# "Code literal"... like a string literal, but delimiters are 153# '{{' and '}}' so they get formatted nicely under emacs c-mode 154def t_CODELIT(t): 155 r"(?m)\{\{([^\}]|}(?!\}))+\}\}" 156 # strip off {{ & }} 157 t.value = t.value[2:-2] 158 t.lineno += t.value.count('\n') 159 return t 160 161def t_CPPDIRECTIVE(t): 162 r'^\#[^\#].*\n' 163 t.lineno += t.value.count('\n') 164 return t 165 166def t_NEWFILE(t): 167 r'^\#\#newfile\s+"[\w/.-]*"' 168 fileNameStack.push((t.value[11:-1], t.lineno)) 169 t.lineno = 0 170 171def t_ENDFILE(t): 172 r'^\#\#endfile' 173 (old_filename, t.lineno) = fileNameStack.pop() 174 175# 176# The functions t_NEWLINE, t_ignore, and t_error are 177# special for the lex module. 178# 179 180# Newlines 181def t_NEWLINE(t): 182 r'\n+' 183 t.lineno += t.value.count('\n') 184 185# Comments 186def t_comment(t): 187 r'//.*' 188 189# Completely ignored characters 190t_ignore = ' \t\x0c' 191 192# Error handler 193def t_error(t): 194 error(t.lineno, "illegal character '%s'" % t.value[0]) 195 t.skip(1) 196 197# Build the lexer 198lexer = lex.lex() 199 200##################################################################### 201# 202# Parser 203# 204# Every function whose name starts with 'p_' defines a grammar rule. 205# The rule is encoded in the function's doc string, while the 206# function body provides the action taken when the rule is matched. 207# The argument to each function is a list of the values of the 208# rule's symbols: t[0] for the LHS, and t[1..n] for the symbols 209# on the RHS. For tokens, the value is copied from the t.value 210# attribute provided by the lexer. For non-terminals, the value 211# is assigned by the producing rule; i.e., the job of the grammar 212# rule function is to set the value for the non-terminal on the LHS 213# (by assigning to t[0]). 214##################################################################### 215 216# The LHS of the first grammar rule is used as the start symbol 217# (in this case, 'specification'). Note that this rule enforces 218# that there will be exactly one namespace declaration, with 0 or more 219# global defs/decls before and after it. The defs & decls before 220# the namespace decl will be outside the namespace; those after 221# will be inside. The decoder function is always inside the namespace. 222def p_specification(t): 223 'specification : opt_defs_and_outputs name_decl opt_defs_and_outputs decode_block' 224 global_code = t[1] 225 isa_name = t[2] 226 namespace = isa_name + "Inst" 227 # wrap the decode block as a function definition 228 t[4].wrap_decode_block(''' 229StaticInstPtr 230%(isa_name)s::decodeInst(%(isa_name)s::ExtMachInst machInst) 231{ 232 using namespace %(namespace)s; 233''' % vars(), '}') 234 # both the latter output blocks and the decode block are in the namespace 235 namespace_code = t[3] + t[4] 236 # pass it all back to the caller of yacc.parse() 237 t[0] = (isa_name, namespace, global_code, namespace_code) 238 239# ISA name declaration looks like "namespace <foo>;" 240def p_name_decl(t): 241 'name_decl : NAMESPACE ID SEMI' 242 t[0] = t[2] 243 244# 'opt_defs_and_outputs' is a possibly empty sequence of 245# def and/or output statements. 246def p_opt_defs_and_outputs_0(t): 247 'opt_defs_and_outputs : empty' 248 t[0] = GenCode() 249 250def p_opt_defs_and_outputs_1(t): 251 'opt_defs_and_outputs : defs_and_outputs' 252 t[0] = t[1] 253 254def p_defs_and_outputs_0(t): 255 'defs_and_outputs : def_or_output' 256 t[0] = t[1] 257 258def p_defs_and_outputs_1(t): 259 'defs_and_outputs : defs_and_outputs def_or_output' 260 t[0] = t[1] + t[2] 261 262# The list of possible definition/output statements. 263def p_def_or_output(t): 264 '''def_or_output : def_format 265 | def_bitfield 266 | def_bitfield_struct 267 | def_template 268 | def_operand_types 269 | def_operands 270 | output_header 271 | output_decoder 272 | output_exec 273 | global_let''' 274 t[0] = t[1] 275 276# Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied 277# directly to the appropriate output section. 278 279 280# Protect any non-dict-substitution '%'s in a format string 281# (i.e. those not followed by '(') 282def protect_non_subst_percents(s): 283 return re.sub(r'%(?!\()', '%%', s) 284 285# Massage output block by substituting in template definitions and bit 286# operators. We handle '%'s embedded in the string that don't 287# indicate template substitutions (or CPU-specific symbols, which get 288# handled in GenCode) by doubling them first so that the format 289# operation will reduce them back to single '%'s. 290def process_output(s): 291 s = protect_non_subst_percents(s) 292 # protects cpu-specific symbols too 293 s = protect_cpu_symbols(s) 294 return substBitOps(s % templateMap) 295 296def p_output_header(t): 297 'output_header : OUTPUT HEADER CODELIT SEMI' 298 t[0] = GenCode(header_output = process_output(t[3])) 299 300def p_output_decoder(t): 301 'output_decoder : OUTPUT DECODER CODELIT SEMI' 302 t[0] = GenCode(decoder_output = process_output(t[3])) 303 304def p_output_exec(t): 305 'output_exec : OUTPUT EXEC CODELIT SEMI' 306 t[0] = GenCode(exec_output = process_output(t[3])) 307 308# global let blocks 'let {{...}}' (Python code blocks) are executed 309# directly when seen. Note that these execute in a special variable 310# context 'exportContext' to prevent the code from polluting this 311# script's namespace. 312def p_global_let(t): 313 'global_let : LET CODELIT SEMI' 314 updateExportContext() 315 exportContext["header_output"] = '' 316 exportContext["decoder_output"] = '' 317 exportContext["exec_output"] = '' 318 exportContext["decode_block"] = '' 319 try: 320 exec fixPythonIndentation(t[2]) in exportContext 321 except Exception, exc: 322 error(t.lineno(1), 323 'error: %s in global let block "%s".' % (exc, t[2])) 324 t[0] = GenCode(header_output = exportContext["header_output"], 325 decoder_output = exportContext["decoder_output"], 326 exec_output = exportContext["exec_output"], 327 decode_block = exportContext["decode_block"]) 328 329# Define the mapping from operand type extensions to C++ types and bit 330# widths (stored in operandTypeMap). 331def p_def_operand_types(t): 332 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI' 333 try: 334 userDict = eval('{' + t[3] + '}') 335 except Exception, exc: 336 error(t.lineno(1), 337 'error: %s in def operand_types block "%s".' % (exc, t[3])) 338 buildOperandTypeMap(userDict, t.lineno(1)) 339 t[0] = GenCode() # contributes nothing to the output C++ file 340 341# Define the mapping from operand names to operand classes and other 342# traits. Stored in operandNameMap. 343def p_def_operands(t): 344 'def_operands : DEF OPERANDS CODELIT SEMI' 345 if not globals().has_key('operandTypeMap'): 346 error(t.lineno(1), 347 'error: operand types must be defined before operands') 348 try: 349 userDict = eval('{' + t[3] + '}') 350 except Exception, exc: 351 error(t.lineno(1), 352 'error: %s in def operands block "%s".' % (exc, t[3])) 353 buildOperandNameMap(userDict, t.lineno(1)) 354 t[0] = GenCode() # contributes nothing to the output C++ file 355 356# A bitfield definition looks like: 357# 'def [signed] bitfield <ID> [<first>:<last>]' 358# This generates a preprocessor macro in the output file. 359def p_def_bitfield_0(t): 360 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI' 361 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8]) 362 if (t[2] == 'signed'): 363 expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr) 364 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 365 t[0] = GenCode(header_output = hash_define) 366 367# alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]' 368def p_def_bitfield_1(t): 369 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI' 370 expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6]) 371 if (t[2] == 'signed'): 372 expr = 'sext<%d>(%s)' % (1, expr) 373 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 374 t[0] = GenCode(header_output = hash_define) 375 376# alternate form for structure member: 'def bitfield <ID> <ID>' 377def p_def_bitfield_struct(t): 378 'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI' 379 if (t[2] != ''): 380 error(t.lineno(1), 'error: structure bitfields are always unsigned.') 381 expr = 'machInst.%s' % t[5] 382 hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr) 383 t[0] = GenCode(header_output = hash_define) 384 385def p_id_with_dot_0(t): 386 'id_with_dot : ID' 387 t[0] = t[1] 388 389def p_id_with_dot_1(t): 390 'id_with_dot : ID DOT id_with_dot' 391 t[0] = t[1] + t[2] + t[3] 392 393def p_opt_signed_0(t): 394 'opt_signed : SIGNED' 395 t[0] = t[1] 396 397def p_opt_signed_1(t): 398 'opt_signed : empty' 399 t[0] = '' 400 401# Global map variable to hold templates 402templateMap = {} 403 404def p_def_template(t): 405 'def_template : DEF TEMPLATE ID CODELIT SEMI' 406 templateMap[t[3]] = Template(t[4]) 407 t[0] = GenCode() 408 409# An instruction format definition looks like 410# "def format <fmt>(<params>) {{...}};" 411def p_def_format(t): 412 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI' 413 (id, params, code) = (t[3], t[5], t[7]) 414 defFormat(id, params, code, t.lineno(1)) 415 t[0] = GenCode() 416 417# The formal parameter list for an instruction format is a possibly 418# empty list of comma-separated parameters. Positional (standard, 419# non-keyword) parameters must come first, followed by keyword 420# parameters, followed by a '*foo' parameter that gets excess 421# positional arguments (as in Python). Each of these three parameter 422# categories is optional. 423# 424# Note that we do not support the '**foo' parameter for collecting 425# otherwise undefined keyword args. Otherwise the parameter list is 426# (I believe) identical to what is supported in Python. 427# 428# The param list generates a tuple, where the first element is a list of 429# the positional params and the second element is a dict containing the 430# keyword params. 431def p_param_list_0(t): 432 'param_list : positional_param_list COMMA nonpositional_param_list' 433 t[0] = t[1] + t[3] 434 435def p_param_list_1(t): 436 '''param_list : positional_param_list 437 | nonpositional_param_list''' 438 t[0] = t[1] 439 440def p_positional_param_list_0(t): 441 'positional_param_list : empty' 442 t[0] = [] 443 444def p_positional_param_list_1(t): 445 'positional_param_list : ID' 446 t[0] = [t[1]] 447 448def p_positional_param_list_2(t): 449 'positional_param_list : positional_param_list COMMA ID' 450 t[0] = t[1] + [t[3]] 451 452def p_nonpositional_param_list_0(t): 453 'nonpositional_param_list : keyword_param_list COMMA excess_args_param' 454 t[0] = t[1] + t[3] 455 456def p_nonpositional_param_list_1(t): 457 '''nonpositional_param_list : keyword_param_list 458 | excess_args_param''' 459 t[0] = t[1] 460 461def p_keyword_param_list_0(t): 462 'keyword_param_list : keyword_param' 463 t[0] = [t[1]] 464 465def p_keyword_param_list_1(t): 466 'keyword_param_list : keyword_param_list COMMA keyword_param' 467 t[0] = t[1] + [t[3]] 468 469def p_keyword_param(t): 470 'keyword_param : ID EQUALS expr' 471 t[0] = t[1] + ' = ' + t[3].__repr__() 472 473def p_excess_args_param(t): 474 'excess_args_param : ASTERISK ID' 475 # Just concatenate them: '*ID'. Wrap in list to be consistent 476 # with positional_param_list and keyword_param_list. 477 t[0] = [t[1] + t[2]] 478 479# End of format definition-related rules. 480############## 481 482# 483# A decode block looks like: 484# decode <field1> [, <field2>]* [default <inst>] { ... } 485# 486def p_decode_block(t): 487 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE' 488 default_defaults = defaultStack.pop() 489 codeObj = t[5] 490 # use the "default defaults" only if there was no explicit 491 # default statement in decode_stmt_list 492 if not codeObj.has_decode_default: 493 codeObj += default_defaults 494 codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n') 495 t[0] = codeObj 496 497# The opt_default statement serves only to push the "default defaults" 498# onto defaultStack. This value will be used by nested decode blocks, 499# and used and popped off when the current decode_block is processed 500# (in p_decode_block() above). 501def p_opt_default_0(t): 502 'opt_default : empty' 503 # no default specified: reuse the one currently at the top of the stack 504 defaultStack.push(defaultStack.top()) 505 # no meaningful value returned 506 t[0] = None 507 508def p_opt_default_1(t): 509 'opt_default : DEFAULT inst' 510 # push the new default 511 codeObj = t[2] 512 codeObj.wrap_decode_block('\ndefault:\n', 'break;\n') 513 defaultStack.push(codeObj) 514 # no meaningful value returned 515 t[0] = None 516 517def p_decode_stmt_list_0(t): 518 'decode_stmt_list : decode_stmt' 519 t[0] = t[1] 520 521def p_decode_stmt_list_1(t): 522 'decode_stmt_list : decode_stmt decode_stmt_list' 523 if (t[1].has_decode_default and t[2].has_decode_default): 524 error(t.lineno(1), 'Two default cases in decode block') 525 t[0] = t[1] + t[2] 526 527# 528# Decode statement rules 529# 530# There are four types of statements allowed in a decode block: 531# 1. Format blocks 'format <foo> { ... }' 532# 2. Nested decode blocks 533# 3. Instruction definitions. 534# 4. C preprocessor directives. 535 536 537# Preprocessor directives found in a decode statement list are passed 538# through to the output, replicated to all of the output code 539# streams. This works well for ifdefs, so we can ifdef out both the 540# declarations and the decode cases generated by an instruction 541# definition. Handling them as part of the grammar makes it easy to 542# keep them in the right place with respect to the code generated by 543# the other statements. 544def p_decode_stmt_cpp(t): 545 'decode_stmt : CPPDIRECTIVE' 546 t[0] = GenCode(t[1], t[1], t[1], t[1]) 547 548# A format block 'format <foo> { ... }' sets the default instruction 549# format used to handle instruction definitions inside the block. 550# This format can be overridden by using an explicit format on the 551# instruction definition or with a nested format block. 552def p_decode_stmt_format(t): 553 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE' 554 # The format will be pushed on the stack when 'push_format_id' is 555 # processed (see below). Once the parser has recognized the full 556 # production (though the right brace), we're done with the format, 557 # so now we can pop it. 558 formatStack.pop() 559 t[0] = t[4] 560 561# This rule exists so we can set the current format (& push the stack) 562# when we recognize the format name part of the format block. 563def p_push_format_id(t): 564 'push_format_id : ID' 565 try: 566 formatStack.push(formatMap[t[1]]) 567 t[0] = ('', '// format %s' % t[1]) 568 except KeyError: 569 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 570 571# Nested decode block: if the value of the current field matches the 572# specified constant, do a nested decode on some other field. 573def p_decode_stmt_decode(t): 574 'decode_stmt : case_label COLON decode_block' 575 label = t[1] 576 codeObj = t[3] 577 # just wrap the decoding code from the block as a case in the 578 # outer switch statement. 579 codeObj.wrap_decode_block('\n%s:\n' % label) 580 codeObj.has_decode_default = (label == 'default') 581 t[0] = codeObj 582 583# Instruction definition (finally!). 584def p_decode_stmt_inst(t): 585 'decode_stmt : case_label COLON inst SEMI' 586 label = t[1] 587 codeObj = t[3] 588 codeObj.wrap_decode_block('\n%s:' % label, 'break;\n') 589 codeObj.has_decode_default = (label == 'default') 590 t[0] = codeObj 591 592# The case label is either a list of one or more constants or 'default' 593def p_case_label_0(t): 594 'case_label : intlit_list' 595 t[0] = ': '.join(map(lambda a: 'case %#x' % a, t[1])) 596 597def p_case_label_1(t): 598 'case_label : DEFAULT' 599 t[0] = 'default' 600 601# 602# The constant list for a decode case label must be non-empty, but may have 603# one or more comma-separated integer literals in it. 604# 605def p_intlit_list_0(t): 606 'intlit_list : INTLIT' 607 t[0] = [t[1]] 608 609def p_intlit_list_1(t): 610 'intlit_list : intlit_list COMMA INTLIT' 611 t[0] = t[1] 612 t[0].append(t[3]) 613 614# Define an instruction using the current instruction format (specified 615# by an enclosing format block). 616# "<mnemonic>(<args>)" 617def p_inst_0(t): 618 'inst : ID LPAREN arg_list RPAREN' 619 # Pass the ID and arg list to the current format class to deal with. 620 currentFormat = formatStack.top() 621 codeObj = currentFormat.defineInst(t[1], t[3], t.lineno(1)) 622 args = ','.join(map(str, t[3])) 623 args = re.sub('(?m)^', '//', args) 624 args = re.sub('^//', '', args) 625 comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args) 626 codeObj.prepend_all(comment) 627 t[0] = codeObj 628 629# Define an instruction using an explicitly specified format: 630# "<fmt>::<mnemonic>(<args>)" 631def p_inst_1(t): 632 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN' 633 try: 634 format = formatMap[t[1]] 635 except KeyError: 636 error(t.lineno(1), 'instruction format "%s" not defined.' % t[1]) 637 codeObj = format.defineInst(t[3], t[5], t.lineno(1)) 638 comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5]) 639 codeObj.prepend_all(comment) 640 t[0] = codeObj 641 642# The arg list generates a tuple, where the first element is a list of 643# the positional args and the second element is a dict containing the 644# keyword args. 645def p_arg_list_0(t): 646 'arg_list : positional_arg_list COMMA keyword_arg_list' 647 t[0] = ( t[1], t[3] ) 648 649def p_arg_list_1(t): 650 'arg_list : positional_arg_list' 651 t[0] = ( t[1], {} ) 652 653def p_arg_list_2(t): 654 'arg_list : keyword_arg_list' 655 t[0] = ( [], t[1] ) 656 657def p_positional_arg_list_0(t): 658 'positional_arg_list : empty' 659 t[0] = [] 660 661def p_positional_arg_list_1(t): 662 'positional_arg_list : expr' 663 t[0] = [t[1]] 664 665def p_positional_arg_list_2(t): 666 'positional_arg_list : positional_arg_list COMMA expr' 667 t[0] = t[1] + [t[3]] 668 669def p_keyword_arg_list_0(t): 670 'keyword_arg_list : keyword_arg' 671 t[0] = t[1] 672 673def p_keyword_arg_list_1(t): 674 'keyword_arg_list : keyword_arg_list COMMA keyword_arg' 675 t[0] = t[1] 676 t[0].update(t[3]) 677 678def p_keyword_arg(t): 679 'keyword_arg : ID EQUALS expr' 680 t[0] = { t[1] : t[3] } 681 682# 683# Basic expressions. These constitute the argument values of 684# "function calls" (i.e. instruction definitions in the decode block) 685# and default values for formal parameters of format functions. 686# 687# Right now, these are either strings, integers, or (recursively) 688# lists of exprs (using Python square-bracket list syntax). Note that 689# bare identifiers are trated as string constants here (since there 690# isn't really a variable namespace to refer to). 691# 692def p_expr_0(t): 693 '''expr : ID 694 | INTLIT 695 | STRLIT 696 | CODELIT''' 697 t[0] = t[1] 698 699def p_expr_1(t): 700 '''expr : LBRACKET list_expr RBRACKET''' 701 t[0] = t[2] 702 703def p_list_expr_0(t): 704 'list_expr : expr' 705 t[0] = [t[1]] 706 707def p_list_expr_1(t): 708 'list_expr : list_expr COMMA expr' 709 t[0] = t[1] + [t[3]] 710 711def p_list_expr_2(t): 712 'list_expr : empty' 713 t[0] = [] 714 715# 716# Empty production... use in other rules for readability. 717# 718def p_empty(t): 719 'empty :' 720 pass 721 722# Parse error handler. Note that the argument here is the offending 723# *token*, not a grammar symbol (hence the need to use t.value) 724def p_error(t): 725 if t: 726 error(t.lineno, "syntax error at '%s'" % t.value) 727 else: 728 error(0, "unknown syntax error", True) 729 730# END OF GRAMMAR RULES 731# 732# Now build the parser. 733parser = yacc.yacc() 734 735 736##################################################################### 737# 738# Support Classes 739# 740##################################################################### 741 742# Expand template with CPU-specific references into a dictionary with 743# an entry for each CPU model name. The entry key is the model name 744# and the corresponding value is the template with the CPU-specific 745# refs substituted for that model. 746def expand_cpu_symbols_to_dict(template): 747 # Protect '%'s that don't go with CPU-specific terms 748 t = re.sub(r'%(?!\(CPU_)', '%%', template) 749 result = {} 750 for cpu in cpu_models: 751 result[cpu.name] = t % cpu.strings 752 return result 753 754# *If* the template has CPU-specific references, return a single 755# string containing a copy of the template for each CPU model with the 756# corresponding values substituted in. If the template has no 757# CPU-specific references, it is returned unmodified. 758def expand_cpu_symbols_to_string(template): 759 if template.find('%(CPU_') != -1: 760 return reduce(lambda x,y: x+y, 761 expand_cpu_symbols_to_dict(template).values()) 762 else: 763 return template 764 765# Protect CPU-specific references by doubling the corresponding '%'s 766# (in preparation for substituting a different set of references into 767# the template). 768def protect_cpu_symbols(template): 769 return re.sub(r'%(?=\(CPU_)', '%%', template) 770 771############### 772# GenCode class 773# 774# The GenCode class encapsulates generated code destined for various 775# output files. The header_output and decoder_output attributes are 776# strings containing code destined for decoder.hh and decoder.cc 777# respectively. The decode_block attribute contains code to be 778# incorporated in the decode function itself (that will also end up in 779# decoder.cc). The exec_output attribute is a dictionary with a key 780# for each CPU model name; the value associated with a particular key 781# is the string of code for that CPU model's exec.cc file. The 782# has_decode_default attribute is used in the decode block to allow 783# explicit default clauses to override default default clauses. 784 785class GenCode: 786 # Constructor. At this point we substitute out all CPU-specific 787 # symbols. For the exec output, these go into the per-model 788 # dictionary. For all other output types they get collapsed into 789 # a single string. 790 def __init__(self, 791 header_output = '', decoder_output = '', exec_output = '', 792 decode_block = '', has_decode_default = False): 793 self.header_output = expand_cpu_symbols_to_string(header_output) 794 self.decoder_output = expand_cpu_symbols_to_string(decoder_output) 795 if isinstance(exec_output, dict): 796 self.exec_output = exec_output 797 elif isinstance(exec_output, str): 798 # If the exec_output arg is a single string, we replicate 799 # it for each of the CPU models, substituting and 800 # %(CPU_foo)s params appropriately. 801 self.exec_output = expand_cpu_symbols_to_dict(exec_output) 802 self.decode_block = expand_cpu_symbols_to_string(decode_block) 803 self.has_decode_default = has_decode_default 804 805 # Override '+' operator: generate a new GenCode object that 806 # concatenates all the individual strings in the operands. 807 def __add__(self, other): 808 exec_output = {} 809 for cpu in cpu_models: 810 n = cpu.name 811 exec_output[n] = self.exec_output[n] + other.exec_output[n] 812 return GenCode(self.header_output + other.header_output, 813 self.decoder_output + other.decoder_output, 814 exec_output, 815 self.decode_block + other.decode_block, 816 self.has_decode_default or other.has_decode_default) 817 818 # Prepend a string (typically a comment) to all the strings. 819 def prepend_all(self, pre): 820 self.header_output = pre + self.header_output 821 self.decoder_output = pre + self.decoder_output 822 self.decode_block = pre + self.decode_block 823 for cpu in cpu_models: 824 self.exec_output[cpu.name] = pre + self.exec_output[cpu.name] 825 826 # Wrap the decode block in a pair of strings (e.g., 'case foo:' 827 # and 'break;'). Used to build the big nested switch statement. 828 def wrap_decode_block(self, pre, post = ''): 829 self.decode_block = pre + indent(self.decode_block) + post 830 831################ 832# Format object. 833# 834# A format object encapsulates an instruction format. It must provide 835# a defineInst() method that generates the code for an instruction 836# definition. 837 838exportContextSymbols = ('InstObjParams', 'makeList', 're', 'string') 839 840exportContext = {} 841 842def updateExportContext(): 843 exportContext.update(exportDict(*exportContextSymbols)) 844 exportContext.update(templateMap) 845 846def exportDict(*symNames): 847 return dict([(s, eval(s)) for s in symNames]) 848 849 850class Format: 851 def __init__(self, id, params, code): 852 # constructor: just save away arguments 853 self.id = id 854 self.params = params 855 label = 'def format ' + id 856 self.user_code = compile(fixPythonIndentation(code), label, 'exec') 857 param_list = string.join(params, ", ") 858 f = '''def defInst(_code, _context, %s): 859 my_locals = vars().copy() 860 exec _code in _context, my_locals 861 return my_locals\n''' % param_list 862 c = compile(f, label + ' wrapper', 'exec') 863 exec c 864 self.func = defInst 865 866 def defineInst(self, name, args, lineno): 867 context = {} 868 updateExportContext() 869 context.update(exportContext) 870 if len(name): 871 Name = name[0].upper() 872 if len(name) > 1: 873 Name += name[1:] 874 context.update({ 'name': name, 'Name': Name }) 875 try: 876 vars = self.func(self.user_code, context, *args[0], **args[1]) 877 except Exception, exc: 878 error(lineno, 'error defining "%s": %s.' % (name, exc)) 879 for k in vars.keys(): 880 if k not in ('header_output', 'decoder_output', 881 'exec_output', 'decode_block'): 882 del vars[k] 883 return GenCode(**vars) 884 885# Special null format to catch an implicit-format instruction 886# definition outside of any format block. 887class NoFormat: 888 def __init__(self): 889 self.defaultInst = '' 890 891 def defineInst(self, name, args, lineno): 892 error(lineno, 893 'instruction definition "%s" with no active format!' % name) 894 895# This dictionary maps format name strings to Format objects. 896formatMap = {} 897 898# Define a new format 899def defFormat(id, params, code, lineno): 900 # make sure we haven't already defined this one 901 if formatMap.get(id, None) != None: 902 error(lineno, 'format %s redefined.' % id) 903 # create new object and store in global map 904 formatMap[id] = Format(id, params, code) 905 906 907############## 908# Stack: a simple stack object. Used for both formats (formatStack) 909# and default cases (defaultStack). Simply wraps a list to give more 910# stack-like syntax and enable initialization with an argument list 911# (as opposed to an argument that's a list). 912 913class Stack(list): 914 def __init__(self, *items): 915 list.__init__(self, items) 916 917 def push(self, item): 918 self.append(item); 919 920 def top(self): 921 return self[-1] 922 923# The global format stack. 924formatStack = Stack(NoFormat()) 925 926# The global default case stack. 927defaultStack = Stack( None ) 928 929# Global stack that tracks current file and line number. 930# Each element is a tuple (filename, lineno) that records the 931# *current* filename and the line number in the *previous* file where 932# it was included. 933fileNameStack = Stack() 934 935################### 936# Utility functions 937 938# 939# Indent every line in string 's' by two spaces 940# (except preprocessor directives). 941# Used to make nested code blocks look pretty. 942# 943def indent(s): 944 return re.sub(r'(?m)^(?!#)', ' ', s) 945 946# 947# Munge a somewhat arbitrarily formatted piece of Python code 948# (e.g. from a format 'let' block) into something whose indentation 949# will get by the Python parser. 950# 951# The two keys here are that Python will give a syntax error if 952# there's any whitespace at the beginning of the first line, and that 953# all lines at the same lexical nesting level must have identical 954# indentation. Unfortunately the way code literals work, an entire 955# let block tends to have some initial indentation. Rather than 956# trying to figure out what that is and strip it off, we prepend 'if 957# 1:' to make the let code the nested block inside the if (and have 958# the parser automatically deal with the indentation for us). 959# 960# We don't want to do this if (1) the code block is empty or (2) the 961# first line of the block doesn't have any whitespace at the front. 962 963def fixPythonIndentation(s): 964 # get rid of blank lines first 965 s = re.sub(r'(?m)^\s*\n', '', s); 966 if (s != '' and re.match(r'[ \t]', s[0])): 967 s = 'if 1:\n' + s 968 return s 969 970# Error handler. Just call exit. Output formatted to work under 971# Emacs compile-mode. Optional 'print_traceback' arg, if set to True, 972# prints a Python stack backtrace too (can be handy when trying to 973# debug the parser itself). 974def error(lineno, string, print_traceback = False): 975 spaces = "" 976 for (filename, line) in fileNameStack[0:-1]: 977 print spaces + "In file included from " + filename + ":" 978 spaces += " " 979 # Print a Python stack backtrace if requested. 980 if (print_traceback): 981 traceback.print_exc() 982 if lineno != 0: 983 line_str = "%d:" % lineno 984 else: 985 line_str = "" 986 sys.exit(spaces + "%s:%s %s" % (fileNameStack[-1][0], line_str, string)) 987 988 989##################################################################### 990# 991# Bitfield Operator Support 992# 993##################################################################### 994 995bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>') 996 997bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>') 998bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>') 999 1000def substBitOps(code): 1001 # first convert single-bit selectors to two-index form 1002 # i.e., <n> --> <n:n> 1003 code = bitOp1ArgRE.sub(r'<\1:\1>', code) 1004 # simple case: selector applied to ID (name) 1005 # i.e., foo<a:b> --> bits(foo, a, b) 1006 code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code) 1007 # if selector is applied to expression (ending in ')'), 1008 # we need to search backward for matching '(' 1009 match = bitOpExprRE.search(code) 1010 while match: 1011 exprEnd = match.start() 1012 here = exprEnd - 1 1013 nestLevel = 1 1014 while nestLevel > 0: 1015 if code[here] == '(': 1016 nestLevel -= 1 1017 elif code[here] == ')': 1018 nestLevel += 1 1019 here -= 1 1020 if here < 0: 1021 sys.exit("Didn't find '('!") 1022 exprStart = here+1 1023 newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1], 1024 match.group(1), match.group(2)) 1025 code = code[:exprStart] + newExpr + code[match.end():] 1026 match = bitOpExprRE.search(code) 1027 return code 1028 1029 1030#################### 1031# Template objects. 1032# 1033# Template objects are format strings that allow substitution from 1034# the attribute spaces of other objects (e.g. InstObjParams instances). 1035 1036labelRE = re.compile(r'(?<!%)%\(([^\)]+)\)[sd]') 1037 1038class Template: 1039 def __init__(self, t): 1040 self.template = t 1041 1042 def subst(self, d): 1043 myDict = None 1044 1045 # Protect non-Python-dict substitutions (e.g. if there's a printf 1046 # in the templated C++ code) 1047 template = protect_non_subst_percents(self.template) 1048 # CPU-model-specific substitutions are handled later (in GenCode). 1049 template = protect_cpu_symbols(template) 1050 1051 # Build a dict ('myDict') to use for the template substitution. 1052 # Start with the template namespace. Make a copy since we're 1053 # going to modify it. 1054 myDict = templateMap.copy() 1055 1056 if isinstance(d, InstObjParams): 1057 # If we're dealing with an InstObjParams object, we need 1058 # to be a little more sophisticated. The instruction-wide 1059 # parameters are already formed, but the parameters which 1060 # are only function wide still need to be generated. 1061 compositeCode = '' 1062 1063 myDict.update(d.__dict__) 1064 # The "operands" and "snippets" attributes of the InstObjParams 1065 # objects are for internal use and not substitution. 1066 del myDict['operands'] 1067 del myDict['snippets'] 1068 1069 snippetLabels = [l for l in labelRE.findall(template) 1070 if d.snippets.has_key(l)] 1071 1072 snippets = dict([(s, mungeSnippet(d.snippets[s])) 1073 for s in snippetLabels]) 1074 1075 myDict.update(snippets) 1076 1077 compositeCode = ' '.join(map(str, snippets.values())) 1078 1079 # Add in template itself in case it references any 1080 # operands explicitly (like Mem) 1081 compositeCode += ' ' + template 1082 1083 operands = SubOperandList(compositeCode, d.operands) 1084 1085 myDict['op_decl'] = operands.concatAttrStrings('op_decl') 1086 1087 is_src = lambda op: op.is_src 1088 is_dest = lambda op: op.is_dest 1089 1090 myDict['op_src_decl'] = \ 1091 operands.concatSomeAttrStrings(is_src, 'op_src_decl') 1092 myDict['op_dest_decl'] = \ 1093 operands.concatSomeAttrStrings(is_dest, 'op_dest_decl') 1094 1095 myDict['op_rd'] = operands.concatAttrStrings('op_rd') 1096 myDict['op_wb'] = operands.concatAttrStrings('op_wb') 1097 1098 if d.operands.memOperand: 1099 myDict['mem_acc_size'] = d.operands.memOperand.mem_acc_size 1100 myDict['mem_acc_type'] = d.operands.memOperand.mem_acc_type 1101 1102 elif isinstance(d, dict): 1103 # if the argument is a dictionary, we just use it. 1104 myDict.update(d) 1105 elif hasattr(d, '__dict__'): 1106 # if the argument is an object, we use its attribute map. 1107 myDict.update(d.__dict__) 1108 else: 1109 raise TypeError, "Template.subst() arg must be or have dictionary" 1110 return template % myDict 1111 1112 # Convert to string. This handles the case when a template with a 1113 # CPU-specific term gets interpolated into another template or into 1114 # an output block. 1115 def __str__(self): 1116 return expand_cpu_symbols_to_string(self.template) 1117 1118##################################################################### 1119# 1120# Code Parser 1121# 1122# The remaining code is the support for automatically extracting 1123# instruction characteristics from pseudocode. 1124# 1125##################################################################### 1126 1127# Force the argument to be a list. Useful for flags, where a caller 1128# can specify a singleton flag or a list of flags. Also usful for 1129# converting tuples to lists so they can be modified. 1130def makeList(arg): 1131 if isinstance(arg, list): 1132 return arg 1133 elif isinstance(arg, tuple): 1134 return list(arg) 1135 elif not arg: 1136 return [] 1137 else: 1138 return [ arg ] 1139 1140# Generate operandTypeMap from the user's 'def operand_types' 1141# statement. 1142def buildOperandTypeMap(userDict, lineno): 1143 global operandTypeMap 1144 operandTypeMap = {} 1145 for (ext, (desc, size)) in userDict.iteritems(): 1146 if desc == 'signed int': 1147 ctype = 'int%d_t' % size 1148 is_signed = 1 1149 elif desc == 'unsigned int': 1150 ctype = 'uint%d_t' % size 1151 is_signed = 0 1152 elif desc == 'float': 1153 is_signed = 1 # shouldn't really matter 1154 if size == 32: 1155 ctype = 'float' 1156 elif size == 64: 1157 ctype = 'double' 1158 elif desc == 'twin64 int': 1159 is_signed = 0 1160 ctype = 'Twin64_t' 1161 elif desc == 'twin32 int': 1162 is_signed = 0 1163 ctype = 'Twin32_t' 1164 if ctype == '': 1165 error(lineno, 'Unrecognized type description "%s" in userDict') 1166 operandTypeMap[ext] = (size, ctype, is_signed) 1167 1168# 1169# 1170# 1171# Base class for operand descriptors. An instance of this class (or 1172# actually a class derived from this one) represents a specific 1173# operand for a code block (e.g, "Rc.sq" as a dest). Intermediate 1174# derived classes encapsulates the traits of a particular operand type 1175# (e.g., "32-bit integer register"). 1176# 1177class Operand(object): 1178 def __init__(self, full_name, ext, is_src, is_dest): 1179 self.full_name = full_name 1180 self.ext = ext 1181 self.is_src = is_src 1182 self.is_dest = is_dest 1183 # The 'effective extension' (eff_ext) is either the actual 1184 # extension, if one was explicitly provided, or the default. 1185 if ext: 1186 self.eff_ext = ext 1187 else: 1188 self.eff_ext = self.dflt_ext 1189 1190 (self.size, self.ctype, self.is_signed) = operandTypeMap[self.eff_ext] 1191 1192 # note that mem_acc_size is undefined for non-mem operands... 1193 # template must be careful not to use it if it doesn't apply. 1194 if self.isMem(): 1195 self.mem_acc_size = self.makeAccSize() 1196 if self.ctype in ['Twin32_t', 'Twin64_t']: 1197 self.mem_acc_type = 'Twin' 1198 else: 1199 self.mem_acc_type = 'uint' 1200 1201 # Finalize additional fields (primarily code fields). This step 1202 # is done separately since some of these fields may depend on the 1203 # register index enumeration that hasn't been performed yet at the 1204 # time of __init__(). 1205 def finalize(self): 1206 self.flags = self.getFlags() 1207 self.constructor = self.makeConstructor() 1208 self.op_decl = self.makeDecl() 1209 1210 if self.is_src: 1211 self.op_rd = self.makeRead() 1212 self.op_src_decl = self.makeDecl() 1213 else: 1214 self.op_rd = '' 1215 self.op_src_decl = '' 1216 1217 if self.is_dest: 1218 self.op_wb = self.makeWrite() 1219 self.op_dest_decl = self.makeDecl() 1220 else: 1221 self.op_wb = '' 1222 self.op_dest_decl = '' 1223 1224 def isMem(self): 1225 return 0 1226 1227 def isReg(self): 1228 return 0 1229 1230 def isFloatReg(self): 1231 return 0 1232 1233 def isIntReg(self): 1234 return 0 1235 1236 def isControlReg(self): 1237 return 0 1238 1239 def getFlags(self): 1240 # note the empty slice '[:]' gives us a copy of self.flags[0] 1241 # instead of a reference to it 1242 my_flags = self.flags[0][:] 1243 if self.is_src: 1244 my_flags += self.flags[1] 1245 if self.is_dest: 1246 my_flags += self.flags[2] 1247 return my_flags 1248 1249 def makeDecl(self): 1250 # Note that initializations in the declarations are solely 1251 # to avoid 'uninitialized variable' errors from the compiler. 1252 return self.ctype + ' ' + self.base_name + ' = 0;\n'; 1253 1254class IntRegOperand(Operand): 1255 def isReg(self): 1256 return 1 1257 1258 def isIntReg(self): 1259 return 1 1260 1261 def makeConstructor(self): 1262 c = '' 1263 if self.is_src: 1264 c += '\n\t_srcRegIdx[%d] = %s;' % \ 1265 (self.src_reg_idx, self.reg_spec) 1266 if self.is_dest: 1267 c += '\n\t_destRegIdx[%d] = %s;' % \ 1268 (self.dest_reg_idx, self.reg_spec) 1269 return c 1270 1271 def makeRead(self): 1272 if (self.ctype == 'float' or self.ctype == 'double'): 1273 error(0, 'Attempt to read integer register as FP') 1274 if (self.size == self.dflt_size): 1275 return '%s = xc->readIntRegOperand(this, %d);\n' % \ 1276 (self.base_name, self.src_reg_idx) 1277 elif (self.size > self.dflt_size): 1278 int_reg_val = 'xc->readIntRegOperand(this, %d)' % \ 1279 (self.src_reg_idx) 1280 if (self.is_signed): 1281 int_reg_val = 'sext<%d>(%s)' % (self.dflt_size, int_reg_val) 1282 return '%s = %s;\n' % (self.base_name, int_reg_val) 1283 else: 1284 return '%s = bits(xc->readIntRegOperand(this, %d), %d, 0);\n' % \ 1285 (self.base_name, self.src_reg_idx, self.size-1) 1286 1287 def makeWrite(self): 1288 if (self.ctype == 'float' or self.ctype == 'double'): 1289 error(0, 'Attempt to write integer register as FP') 1290 if (self.size != self.dflt_size and self.is_signed): 1291 final_val = 'sext<%d>(%s)' % (self.size, self.base_name) 1292 else: 1293 final_val = self.base_name 1294 wb = ''' 1295 { 1296 %s final_val = %s; 1297 xc->setIntRegOperand(this, %d, final_val);\n 1298 if (traceData) { traceData->setData(final_val); } 1299 }''' % (self.dflt_ctype, final_val, self.dest_reg_idx) 1300 return wb 1301 1302class FloatRegOperand(Operand): 1303 def isReg(self): 1304 return 1 1305 1306 def isFloatReg(self): 1307 return 1 1308 1309 def makeConstructor(self): 1310 c = '' 1311 if self.is_src: 1312 c += '\n\t_srcRegIdx[%d] = %s + FP_Base_DepTag;' % \ 1313 (self.src_reg_idx, self.reg_spec) 1314 if self.is_dest: 1315 c += '\n\t_destRegIdx[%d] = %s + FP_Base_DepTag;' % \ 1316 (self.dest_reg_idx, self.reg_spec) 1317 return c 1318 1319 def makeRead(self): 1320 bit_select = 0 1321 width = 0; 1322 if (self.ctype == 'float'): 1323 func = 'readFloatRegOperand' 1324 width = 32; 1325 elif (self.ctype == 'double'): 1326 func = 'readFloatRegOperand' 1327 width = 64; 1328 else: 1329 func = 'readFloatRegOperandBits' 1330 if (self.ctype == 'uint32_t'): 1331 width = 32; 1332 elif (self.ctype == 'uint64_t'): 1333 width = 64; 1334 if (self.size != self.dflt_size): 1335 bit_select = 1 1336 if width: 1337 base = 'xc->%s(this, %d, %d)' % \ 1338 (func, self.src_reg_idx, width) 1339 else: 1340 base = 'xc->%s(this, %d)' % \ 1341 (func, self.src_reg_idx) 1342 if bit_select: 1343 return '%s = bits(%s, %d, 0);\n' % \ 1344 (self.base_name, base, self.size-1) 1345 else: 1346 return '%s = %s;\n' % (self.base_name, base) 1347 1348 def makeWrite(self): 1349 final_val = self.base_name 1350 final_ctype = self.ctype 1351 widthSpecifier = '' 1352 width = 0 1353 if (self.ctype == 'float'): 1354 width = 32 1355 func = 'setFloatRegOperand' 1356 elif (self.ctype == 'double'): 1357 width = 64 1358 func = 'setFloatRegOperand' 1359 elif (self.ctype == 'uint32_t'): 1360 func = 'setFloatRegOperandBits' 1361 width = 32 1362 elif (self.ctype == 'uint64_t'): 1363 func = 'setFloatRegOperandBits' 1364 width = 64 1365 else: 1366 func = 'setFloatRegOperandBits' 1367 final_ctype = 'uint%d_t' % self.dflt_size 1368 if (self.size != self.dflt_size and self.is_signed): 1369 final_val = 'sext<%d>(%s)' % (self.size, self.base_name) 1370 if width: 1371 widthSpecifier = ', %d' % width 1372 wb = ''' 1373 { 1374 %s final_val = %s; 1375 xc->%s(this, %d, final_val%s);\n 1376 if (traceData) { traceData->setData(final_val); } 1377 }''' % (final_ctype, final_val, func, self.dest_reg_idx, 1378 widthSpecifier) 1379 return wb 1380 1381class ControlRegOperand(Operand): 1382 def isReg(self): 1383 return 1 1384 1385 def isControlReg(self): 1386 return 1 1387 1388 def makeConstructor(self): 1389 c = '' 1390 if self.is_src: 1391 c += '\n\t_srcRegIdx[%d] = %s + Ctrl_Base_DepTag;' % \ 1392 (self.src_reg_idx, self.reg_spec) 1393 if self.is_dest: 1394 c += '\n\t_destRegIdx[%d] = %s + Ctrl_Base_DepTag;' % \ 1395 (self.dest_reg_idx, self.reg_spec) 1396 return c 1397 1398 def makeRead(self): 1399 bit_select = 0 1400 if (self.ctype == 'float' or self.ctype == 'double'): 1401 error(0, 'Attempt to read control register as FP') 1402 base = 'xc->readMiscRegOperand(this, %s)' % self.src_reg_idx 1403 if self.size == self.dflt_size: 1404 return '%s = %s;\n' % (self.base_name, base) 1405 else: 1406 return '%s = bits(%s, %d, 0);\n' % \ 1407 (self.base_name, base, self.size-1) 1408 1409 def makeWrite(self): 1410 if (self.ctype == 'float' or self.ctype == 'double'): 1411 error(0, 'Attempt to write control register as FP') 1412 wb = 'xc->setMiscRegOperand(this, %s, %s);\n' % \ 1413 (self.dest_reg_idx, self.base_name) 1414 1415class ControlBitfieldOperand(ControlRegOperand): 1416 def makeRead(self): 1417 bit_select = 0 1418 if (self.ctype == 'float' or self.ctype == 'double'): 1419 error(0, 'Attempt to read control register as FP') 1420 base = 'xc->readMiscReg(%s)' % self.reg_spec 1421 name = self.base_name 1422 return '%s = bits(%s, %s_HI, %s_LO);' % \ 1423 (name, base, name, name) 1424 1425 def makeWrite(self): 1426 if (self.ctype == 'float' or self.ctype == 'double'): 1427 error(0, 'Attempt to write control register as FP') 1428 base = 'xc->readMiscReg(%s)' % self.reg_spec 1429 name = self.base_name 1430 wb_val = 'insertBits(%s, %s_HI, %s_LO, %s)' % \ 1431 (base, name, name, self.base_name) 1432 wb = 'xc->setMiscRegOperand(this, %s, %s );\n' % (self.dest_reg_idx, wb_val) 1433 wb += 'if (traceData) { traceData->setData(%s); }' % \ 1434 self.base_name 1435 return wb 1436 1437class MemOperand(Operand): 1438 def isMem(self): 1439 return 1 1440 1441 def makeConstructor(self): 1442 return '' 1443 1444 def makeDecl(self): 1445 # Note that initializations in the declarations are solely 1446 # to avoid 'uninitialized variable' errors from the compiler. 1447 # Declare memory data variable. 1448 if self.ctype in ['Twin32_t','Twin64_t']: 1449 return "%s %s; %s.a = 0; %s.b = 0;\n" % (self.ctype, self.base_name, 1450 self.base_name, self.base_name) 1451 c = '%s %s = 0;\n' % (self.ctype, self.base_name) 1452 return c 1453 1454 def makeRead(self): 1455 return '' 1456 1457 def makeWrite(self): 1458 return '' 1459 1460 # Return the memory access size *in bits*, suitable for 1461 # forming a type via "uint%d_t". Divide by 8 if you want bytes. 1462 def makeAccSize(self): 1463 return self.size 1464 1465 1466class NPCOperand(Operand): 1467 def makeConstructor(self): 1468 return '' 1469 1470 def makeRead(self): 1471 return '%s = xc->readNextPC();\n' % self.base_name 1472 1473 def makeWrite(self): 1474 return 'xc->setNextPC(%s);\n' % self.base_name 1475 1476class NNPCOperand(Operand): 1477 def makeConstructor(self): 1478 return '' 1479 1480 def makeRead(self): 1481 return '%s = xc->readNextNPC();\n' % self.base_name 1482 1483 def makeWrite(self): 1484 return 'xc->setNextNPC(%s);\n' % self.base_name 1485 1486def buildOperandNameMap(userDict, lineno): 1487 global operandNameMap 1488 operandNameMap = {} 1489 for (op_name, val) in userDict.iteritems(): 1490 (base_cls_name, dflt_ext, reg_spec, flags, sort_pri) = val 1491 (dflt_size, dflt_ctype, dflt_is_signed) = operandTypeMap[dflt_ext] 1492 # Canonical flag structure is a triple of lists, where each list 1493 # indicates the set of flags implied by this operand always, when 1494 # used as a source, and when used as a dest, respectively. 1495 # For simplicity this can be initialized using a variety of fairly 1496 # obvious shortcuts; we convert these to canonical form here. 1497 if not flags: 1498 # no flags specified (e.g., 'None') 1499 flags = ( [], [], [] ) 1500 elif isinstance(flags, str): 1501 # a single flag: assumed to be unconditional 1502 flags = ( [ flags ], [], [] ) 1503 elif isinstance(flags, list): 1504 # a list of flags: also assumed to be unconditional 1505 flags = ( flags, [], [] ) 1506 elif isinstance(flags, tuple): 1507 # it's a tuple: it should be a triple, 1508 # but each item could be a single string or a list 1509 (uncond_flags, src_flags, dest_flags) = flags 1510 flags = (makeList(uncond_flags), 1511 makeList(src_flags), makeList(dest_flags)) 1512 # Accumulate attributes of new operand class in tmp_dict 1513 tmp_dict = {} 1514 for attr in ('dflt_ext', 'reg_spec', 'flags', 'sort_pri', 1515 'dflt_size', 'dflt_ctype', 'dflt_is_signed'): 1516 tmp_dict[attr] = eval(attr) 1517 tmp_dict['base_name'] = op_name 1518 # New class name will be e.g. "IntReg_Ra" 1519 cls_name = base_cls_name + '_' + op_name 1520 # Evaluate string arg to get class object. Note that the 1521 # actual base class for "IntReg" is "IntRegOperand", i.e. we 1522 # have to append "Operand". 1523 try: 1524 base_cls = eval(base_cls_name + 'Operand') 1525 except NameError: 1526 error(lineno, 1527 'error: unknown operand base class "%s"' % base_cls_name) 1528 # The following statement creates a new class called 1529 # <cls_name> as a subclass of <base_cls> with the attributes 1530 # in tmp_dict, just as if we evaluated a class declaration. 1531 operandNameMap[op_name] = type(cls_name, (base_cls,), tmp_dict) 1532 1533 # Define operand variables. 1534 operands = userDict.keys() 1535 1536 operandsREString = (r''' 1537 (?<![\w\.]) # neg. lookbehind assertion: prevent partial matches 1538 ((%s)(?:\.(\w+))?) # match: operand with optional '.' then suffix 1539 (?![\w\.]) # neg. lookahead assertion: prevent partial matches 1540 ''' 1541 % string.join(operands, '|')) 1542 1543 global operandsRE 1544 operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE) 1545 1546 # Same as operandsREString, but extension is mandatory, and only two 1547 # groups are returned (base and ext, not full name as above). 1548 # Used for subtituting '_' for '.' to make C++ identifiers. 1549 operandsWithExtREString = (r'(?<![\w\.])(%s)\.(\w+)(?![\w\.])' 1550 % string.join(operands, '|')) 1551 1552 global operandsWithExtRE 1553 operandsWithExtRE = re.compile(operandsWithExtREString, re.MULTILINE) 1554 1555 1556class OperandList: 1557 1558 # Find all the operands in the given code block. Returns an operand 1559 # descriptor list (instance of class OperandList). 1560 def __init__(self, code): 1561 self.items = [] 1562 self.bases = {} 1563 # delete comments so we don't match on reg specifiers inside 1564 code = commentRE.sub('', code) 1565 # search for operands 1566 next_pos = 0 1567 while 1: 1568 match = operandsRE.search(code, next_pos) 1569 if not match: 1570 # no more matches: we're done 1571 break 1572 op = match.groups() 1573 # regexp groups are operand full name, base, and extension 1574 (op_full, op_base, op_ext) = op 1575 # if the token following the operand is an assignment, this is 1576 # a destination (LHS), else it's a source (RHS) 1577 is_dest = (assignRE.match(code, match.end()) != None) 1578 is_src = not is_dest 1579 # see if we've already seen this one 1580 op_desc = self.find_base(op_base) 1581 if op_desc: 1582 if op_desc.ext != op_ext: 1583 error(0, 'Inconsistent extensions for operand %s' % \ 1584 op_base) 1585 op_desc.is_src = op_desc.is_src or is_src 1586 op_desc.is_dest = op_desc.is_dest or is_dest 1587 else: 1588 # new operand: create new descriptor 1589 op_desc = operandNameMap[op_base](op_full, op_ext, 1590 is_src, is_dest) 1591 self.append(op_desc) 1592 # start next search after end of current match 1593 next_pos = match.end() 1594 self.sort() 1595 # enumerate source & dest register operands... used in building 1596 # constructor later 1597 self.numSrcRegs = 0 1598 self.numDestRegs = 0 1599 self.numFPDestRegs = 0 1600 self.numIntDestRegs = 0 1601 self.memOperand = None 1602 for op_desc in self.items: 1603 if op_desc.isReg(): 1604 if op_desc.is_src: 1605 op_desc.src_reg_idx = self.numSrcRegs 1606 self.numSrcRegs += 1 1607 if op_desc.is_dest: 1608 op_desc.dest_reg_idx = self.numDestRegs 1609 self.numDestRegs += 1 1610 if op_desc.isFloatReg(): 1611 self.numFPDestRegs += 1 1612 elif op_desc.isIntReg(): 1613 self.numIntDestRegs += 1 1614 elif op_desc.isMem(): 1615 if self.memOperand: 1616 error(0, "Code block has more than one memory operand.") 1617 self.memOperand = op_desc 1618 # now make a final pass to finalize op_desc fields that may depend 1619 # on the register enumeration 1620 for op_desc in self.items: 1621 op_desc.finalize() 1622 1623 def __len__(self): 1624 return len(self.items) 1625 1626 def __getitem__(self, index): 1627 return self.items[index] 1628 1629 def append(self, op_desc): 1630 self.items.append(op_desc) 1631 self.bases[op_desc.base_name] = op_desc 1632 1633 def find_base(self, base_name): 1634 # like self.bases[base_name], but returns None if not found 1635 # (rather than raising exception) 1636 return self.bases.get(base_name) 1637 1638 # internal helper function for concat[Some]Attr{Strings|Lists} 1639 def __internalConcatAttrs(self, attr_name, filter, result): 1640 for op_desc in self.items: 1641 if filter(op_desc): 1642 result += getattr(op_desc, attr_name) 1643 return result 1644 1645 # return a single string that is the concatenation of the (string) 1646 # values of the specified attribute for all operands 1647 def concatAttrStrings(self, attr_name): 1648 return self.__internalConcatAttrs(attr_name, lambda x: 1, '') 1649 1650 # like concatAttrStrings, but only include the values for the operands 1651 # for which the provided filter function returns true 1652 def concatSomeAttrStrings(self, filter, attr_name): 1653 return self.__internalConcatAttrs(attr_name, filter, '') 1654 1655 # return a single list that is the concatenation of the (list) 1656 # values of the specified attribute for all operands 1657 def concatAttrLists(self, attr_name): 1658 return self.__internalConcatAttrs(attr_name, lambda x: 1, []) 1659 1660 # like concatAttrLists, but only include the values for the operands 1661 # for which the provided filter function returns true 1662 def concatSomeAttrLists(self, filter, attr_name): 1663 return self.__internalConcatAttrs(attr_name, filter, []) 1664 1665 def sort(self): 1666 self.items.sort(lambda a, b: a.sort_pri - b.sort_pri) 1667 1668class SubOperandList(OperandList): 1669 1670 # Find all the operands in the given code block. Returns an operand 1671 # descriptor list (instance of class OperandList). 1672 def __init__(self, code, master_list): 1673 self.items = [] 1674 self.bases = {} 1675 # delete comments so we don't match on reg specifiers inside 1676 code = commentRE.sub('', code) 1677 # search for operands 1678 next_pos = 0 1679 while 1: 1680 match = operandsRE.search(code, next_pos) 1681 if not match: 1682 # no more matches: we're done 1683 break 1684 op = match.groups() 1685 # regexp groups are operand full name, base, and extension 1686 (op_full, op_base, op_ext) = op 1687 # find this op in the master list 1688 op_desc = master_list.find_base(op_base) 1689 if not op_desc: 1690 error(0, 'Found operand %s which is not in the master list!' \ 1691 ' This is an internal error' % \ 1692 op_base) 1693 else: 1694 # See if we've already found this operand 1695 op_desc = self.find_base(op_base) 1696 if not op_desc: 1697 # if not, add a reference to it to this sub list 1698 self.append(master_list.bases[op_base]) 1699 1700 # start next search after end of current match 1701 next_pos = match.end() 1702 self.sort() 1703 self.memOperand = None 1704 for op_desc in self.items: 1705 if op_desc.isMem(): 1706 if self.memOperand: 1707 error(0, "Code block has more than one memory operand.") 1708 self.memOperand = op_desc 1709 1710# Regular expression object to match C++ comments 1711# (used in findOperands()) 1712commentRE = re.compile(r'//.*\n') 1713 1714# Regular expression object to match assignment statements 1715# (used in findOperands()) 1716assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE) 1717 1718# Munge operand names in code string to make legal C++ variable names. 1719# This means getting rid of the type extension if any. 1720# (Will match base_name attribute of Operand object.) 1721def substMungedOpNames(code): 1722 return operandsWithExtRE.sub(r'\1', code) 1723 1724# Fix up code snippets for final substitution in templates. 1725def mungeSnippet(s): 1726 if isinstance(s, str): 1727 return substMungedOpNames(substBitOps(s)) 1728 else: 1729 return s 1730 1731def makeFlagConstructor(flag_list): 1732 if len(flag_list) == 0: 1733 return '' 1734 # filter out repeated flags 1735 flag_list.sort() 1736 i = 1 1737 while i < len(flag_list): 1738 if flag_list[i] == flag_list[i-1]: 1739 del flag_list[i] 1740 else: 1741 i += 1 1742 pre = '\n\tflags[' 1743 post = '] = true;' 1744 code = pre + string.join(flag_list, post + pre) + post 1745 return code 1746 1747# Assume all instruction flags are of the form 'IsFoo' 1748instFlagRE = re.compile(r'Is.*') 1749 1750# OpClass constants end in 'Op' except No_OpClass 1751opClassRE = re.compile(r'.*Op|No_OpClass') 1752 1753class InstObjParams: 1754 def __init__(self, mnem, class_name, base_class = '', 1755 snippets = {}, opt_args = []): 1756 self.mnemonic = mnem 1757 self.class_name = class_name 1758 self.base_class = base_class 1759 if not isinstance(snippets, dict): 1760 snippets = {'code' : snippets} 1761 compositeCode = ' '.join(map(str, snippets.values())) 1762 self.snippets = snippets 1763 1764 self.operands = OperandList(compositeCode) 1765 self.constructor = self.operands.concatAttrStrings('constructor') 1766 self.constructor += \ 1767 '\n\t_numSrcRegs = %d;' % self.operands.numSrcRegs 1768 self.constructor += \ 1769 '\n\t_numDestRegs = %d;' % self.operands.numDestRegs 1770 self.constructor += \ 1771 '\n\t_numFPDestRegs = %d;' % self.operands.numFPDestRegs 1772 self.constructor += \ 1773 '\n\t_numIntDestRegs = %d;' % self.operands.numIntDestRegs 1774 self.flags = self.operands.concatAttrLists('flags') 1775 1776 # Make a basic guess on the operand class (function unit type). 1777 # These are good enough for most cases, and can be overridden 1778 # later otherwise. 1779 if 'IsStore' in self.flags: 1780 self.op_class = 'MemWriteOp' 1781 elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags: 1782 self.op_class = 'MemReadOp' 1783 elif 'IsFloating' in self.flags: 1784 self.op_class = 'FloatAddOp' 1785 else: 1786 self.op_class = 'IntAluOp' 1787 1788 # Optional arguments are assumed to be either StaticInst flags 1789 # or an OpClass value. To avoid having to import a complete 1790 # list of these values to match against, we do it ad-hoc 1791 # with regexps. 1792 for oa in opt_args: 1793 if instFlagRE.match(oa): 1794 self.flags.append(oa) 1795 elif opClassRE.match(oa): 1796 self.op_class = oa 1797 else: 1798 error(0, 'InstObjParams: optional arg "%s" not recognized ' 1799 'as StaticInst::Flag or OpClass.' % oa) 1800 1801 # add flag initialization to contructor here to include 1802 # any flags added via opt_args 1803 self.constructor += makeFlagConstructor(self.flags) 1804 1805 # if 'IsFloating' is set, add call to the FP enable check 1806 # function (which should be provided by isa_desc via a declare) 1807 if 'IsFloating' in self.flags: 1808 self.fp_enable_check = 'fault = checkFpEnableFault(xc);' 1809 else: 1810 self.fp_enable_check = '' 1811 1812####################### 1813# 1814# Output file template 1815# 1816 1817file_template = ''' 1818/* 1819 * DO NOT EDIT THIS FILE!!! 1820 * 1821 * It was automatically generated from the ISA description in %(filename)s 1822 */ 1823 1824%(includes)s 1825 1826%(global_output)s 1827 1828namespace %(namespace)s { 1829 1830%(namespace_output)s 1831 1832} // namespace %(namespace)s 1833 1834%(decode_function)s 1835''' 1836 1837 1838# Update the output file only if the new contents are different from 1839# the current contents. Minimizes the files that need to be rebuilt 1840# after minor changes. 1841def update_if_needed(file, contents): 1842 update = False 1843 if os.access(file, os.R_OK): 1844 f = open(file, 'r') 1845 old_contents = f.read() 1846 f.close() 1847 if contents != old_contents: 1848 print 'Updating', file 1849 os.remove(file) # in case it's write-protected 1850 update = True 1851 else: 1852 print 'File', file, 'is unchanged' 1853 else: 1854 print 'Generating', file 1855 update = True 1856 if update: 1857 f = open(file, 'w') 1858 f.write(contents) 1859 f.close() 1860 1861# This regular expression matches '##include' directives 1862includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[\w/.-]*)".*$', 1863 re.MULTILINE) 1864 1865# Function to replace a matched '##include' directive with the 1866# contents of the specified file (with nested ##includes replaced 1867# recursively). 'matchobj' is an re match object (from a match of 1868# includeRE) and 'dirname' is the directory relative to which the file 1869# path should be resolved. 1870def replace_include(matchobj, dirname): 1871 fname = matchobj.group('filename') 1872 full_fname = os.path.normpath(os.path.join(dirname, fname)) 1873 contents = '##newfile "%s"\n%s\n##endfile\n' % \ 1874 (full_fname, read_and_flatten(full_fname)) 1875 return contents 1876 1877# Read a file and recursively flatten nested '##include' files. 1878def read_and_flatten(filename): 1879 current_dir = os.path.dirname(filename) 1880 try: 1881 contents = open(filename).read() 1882 except IOError: 1883 error(0, 'Error including file "%s"' % filename) 1884 fileNameStack.push((filename, 0)) 1885 # Find any includes and include them 1886 contents = includeRE.sub(lambda m: replace_include(m, current_dir), 1887 contents) 1888 fileNameStack.pop() 1889 return contents 1890 1891# 1892# Read in and parse the ISA description. 1893# 1894def parse_isa_desc(isa_desc_file, output_dir): 1895 # Read file and (recursively) all included files into a string. 1896 # PLY requires that the input be in a single string so we have to 1897 # do this up front. 1898 isa_desc = read_and_flatten(isa_desc_file) 1899 1900 # Initialize filename stack with outer file. 1901 fileNameStack.push((isa_desc_file, 0)) 1902 1903 # Parse it. 1904 (isa_name, namespace, global_code, namespace_code) = \ 1905 parser.parse(isa_desc, lexer=lexer) 1906 1907 # grab the last three path components of isa_desc_file to put in 1908 # the output 1909 filename = '/'.join(isa_desc_file.split('/')[-3:]) 1910 1911 # generate decoder.hh 1912 includes = '#include "base/bitfield.hh" // for bitfield support' 1913 global_output = global_code.header_output 1914 namespace_output = namespace_code.header_output 1915 decode_function = '' 1916 update_if_needed(output_dir + '/decoder.hh', file_template % vars()) 1917 1918 # generate decoder.cc 1919 includes = '#include "decoder.hh"' 1920 global_output = global_code.decoder_output 1921 namespace_output = namespace_code.decoder_output 1922 # namespace_output += namespace_code.decode_block 1923 decode_function = namespace_code.decode_block 1924 update_if_needed(output_dir + '/decoder.cc', file_template % vars()) 1925 1926 # generate per-cpu exec files 1927 for cpu in cpu_models: 1928 includes = '#include "decoder.hh"\n' 1929 includes += cpu.includes 1930 global_output = global_code.exec_output[cpu.name] 1931 namespace_output = namespace_code.exec_output[cpu.name] 1932 decode_function = '' 1933 update_if_needed(output_dir + '/' + cpu.filename, 1934 file_template % vars()) 1935 1936# global list of CpuModel objects (see cpu_models.py) 1937cpu_models = [] 1938 1939# Called as script: get args from command line. 1940# Args are: <path to cpu_models.py> <isa desc file> <output dir> <cpu models> 1941if __name__ == '__main__': 1942 execfile(sys.argv[1]) # read in CpuModel definitions 1943 cpu_models = [CpuModel.dict[cpu] for cpu in sys.argv[4:]] 1944 parse_isa_desc(sys.argv[2], sys.argv[3]) 1945