1 # Copyright (c) 2003-2005 The Regents of The University of Michigan
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.
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.
27 # Authors: Steve Reinhardt
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.
41 sys
.path
[0:0] = [os
.environ
['M5_PLY']]
46 #####################################################################
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.
59 #####################################################################
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.
65 'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT',
66 'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS',
67 'OUTPUT', 'SIGNED', 'TEMPLATE'
70 # List of tokens. The lex module requires this.
84 # ( ) [ ] { } < > , ; : :: *
86 'LBRACKET', 'RBRACKET',
88 'LESS', 'GREATER', 'EQUALS',
89 'COMMA', 'SEMI', 'COLON', 'DBLCOLON',
92 # C preprocessor directives
95 # The following are matched but never returned. commented out to
96 # suppress PLY warning
104 # Regular expressions for token matching
120 # Identifiers and reserved words
123 reserved_map
[r
.lower()] = r
127 t
.type = reserved_map
.get(t
.value
,'ID')
132 r
'(0x[\da-fA-F]+)|\d+'
134 t
.value
= int(t
.value
,0)
136 error(t
.lineno
, 'Integer value "%s" too large' % t
.value
)
140 # String literal. Note that these use only single quotes, and
141 # can span multiple lines.
145 t
.value
= t
.value
[1:-1]
146 t
.lineno
+= t
.value
.count('\n')
150 # "Code literal"... like a string literal, but delimiters are
151 # '{{' and '}}' so they get formatted nicely under emacs c-mode
153 r
"(?m)\{\{([^\}]|}(?!\}))+\}\}"
155 t
.value
= t
.value
[2:-2]
156 t
.lineno
+= t
.value
.count('\n')
159 def t_CPPDIRECTIVE(t
):
161 t
.lineno
+= t
.value
.count('\n')
165 r
'^\#\#newfile\s+"[\w/.-]*"'
166 fileNameStack
.push((t
.value
[11:-1], t
.lineno
))
171 (old_filename
, t
.lineno
) = fileNameStack
.pop()
174 # The functions t_NEWLINE, t_ignore, and t_error are
175 # special for the lex module.
181 t
.lineno
+= t
.value
.count('\n')
187 # Completely ignored characters
192 error(t
.lineno
, "illegal character '%s'" % t
.value
[0])
198 #####################################################################
202 # Every function whose name starts with 'p_' defines a grammar rule.
203 # The rule is encoded in the function's doc string, while the
204 # function body provides the action taken when the rule is matched.
205 # The argument to each function is a list of the values of the
206 # rule's symbols: t[0] for the LHS, and t[1..n] for the symbols
207 # on the RHS. For tokens, the value is copied from the t.value
208 # attribute provided by the lexer. For non-terminals, the value
209 # is assigned by the producing rule; i.e., the job of the grammar
210 # rule function is to set the value for the non-terminal on the LHS
211 # (by assigning to t[0]).
212 #####################################################################
214 # The LHS of the first grammar rule is used as the start symbol
215 # (in this case, 'specification'). Note that this rule enforces
216 # that there will be exactly one namespace declaration, with 0 or more
217 # global defs/decls before and after it. The defs & decls before
218 # the namespace decl will be outside the namespace; those after
219 # will be inside. The decoder function is always inside the namespace.
220 def p_specification(t
):
221 'specification : opt_defs_and_outputs name_decl opt_defs_and_outputs decode_block'
224 namespace
= isa_name
+ "Inst"
225 # wrap the decode block as a function definition
226 t
[4].wrap_decode_block('''
228 %(isa_name)s::decodeInst(%(isa_name)s::ExtMachInst machInst)
230 using namespace %(namespace)s;
232 # both the latter output blocks and the decode block are in the namespace
233 namespace_code
= t
[3] + t
[4]
234 # pass it all back to the caller of yacc.parse()
235 t
[0] = (isa_name
, namespace
, global_code
, namespace_code
)
237 # ISA name declaration looks like "namespace <foo>;"
239 'name_decl : NAMESPACE ID SEMI'
242 # 'opt_defs_and_outputs' is a possibly empty sequence of
243 # def and/or output statements.
244 def p_opt_defs_and_outputs_0(t
):
245 'opt_defs_and_outputs : empty'
248 def p_opt_defs_and_outputs_1(t
):
249 'opt_defs_and_outputs : defs_and_outputs'
252 def p_defs_and_outputs_0(t
):
253 'defs_and_outputs : def_or_output'
256 def p_defs_and_outputs_1(t
):
257 'defs_and_outputs : defs_and_outputs def_or_output'
260 # The list of possible definition/output statements.
261 def p_def_or_output(t
):
262 '''def_or_output : def_format
273 # Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied
274 # directly to the appropriate output section.
277 # Protect any non-dict-substitution '%'s in a format string
278 # (i.e. those not followed by '(')
279 def protect_non_subst_percents(s
):
280 return re
.sub(r
'%(?!\()', '%%', s
)
282 # Massage output block by substituting in template definitions and bit
283 # operators. We handle '%'s embedded in the string that don't
284 # indicate template substitutions (or CPU-specific symbols, which get
285 # handled in GenCode) by doubling them first so that the format
286 # operation will reduce them back to single '%'s.
287 def process_output(s
):
288 s
= protect_non_subst_percents(s
)
289 # protects cpu-specific symbols too
290 s
= protect_cpu_symbols(s
)
291 return substBitOps(s
% templateMap
)
293 def p_output_header(t
):
294 'output_header : OUTPUT HEADER CODELIT SEMI'
295 t
[0] = GenCode(header_output
= process_output(t
[3]))
297 def p_output_decoder(t
):
298 'output_decoder : OUTPUT DECODER CODELIT SEMI'
299 t
[0] = GenCode(decoder_output
= process_output(t
[3]))
301 def p_output_exec(t
):
302 'output_exec : OUTPUT EXEC CODELIT SEMI'
303 t
[0] = GenCode(exec_output
= process_output(t
[3]))
305 # global let blocks 'let {{...}}' (Python code blocks) are executed
306 # directly when seen. Note that these execute in a special variable
307 # context 'exportContext' to prevent the code from polluting this
308 # script's namespace.
310 'global_let : LET CODELIT SEMI'
311 updateExportContext()
313 exec fixPythonIndentation(t
[2]) in exportContext
314 except Exception, exc
:
316 'error: %s in global let block "%s".' % (exc
, t
[2]))
317 t
[0] = GenCode() # contributes nothing to the output C++ file
319 # Define the mapping from operand type extensions to C++ types and bit
320 # widths (stored in operandTypeMap).
321 def p_def_operand_types(t
):
322 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI'
324 userDict
= eval('{' + t
[3] + '}')
325 except Exception, exc
:
327 'error: %s in def operand_types block "%s".' % (exc
, t
[3]))
328 buildOperandTypeMap(userDict
, t
.lineno(1))
329 t
[0] = GenCode() # contributes nothing to the output C++ file
331 # Define the mapping from operand names to operand classes and other
332 # traits. Stored in operandNameMap.
333 def p_def_operands(t
):
334 'def_operands : DEF OPERANDS CODELIT SEMI'
335 if not globals().has_key('operandTypeMap'):
337 'error: operand types must be defined before operands')
339 userDict
= eval('{' + t
[3] + '}')
340 except Exception, exc
:
342 'error: %s in def operands block "%s".' % (exc
, t
[3]))
343 buildOperandNameMap(userDict
, t
.lineno(1))
344 t
[0] = GenCode() # contributes nothing to the output C++ file
346 # A bitfield definition looks like:
347 # 'def [signed] bitfield <ID> [<first>:<last>]'
348 # This generates a preprocessor macro in the output file.
349 def p_def_bitfield_0(t
):
350 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI'
351 expr
= 'bits(machInst, %2d, %2d)' % (t
[6], t
[8])
352 if (t
[2] == 'signed'):
353 expr
= 'sext<%d>(%s)' % (t
[6] - t
[8] + 1, expr
)
354 hash_define
= '#undef %s\n#define %s\t%s\n' % (t
[4], t
[4], expr
)
355 t
[0] = GenCode(header_output
= hash_define
)
357 # alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]'
358 def p_def_bitfield_1(t
):
359 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI'
360 expr
= 'bits(machInst, %2d, %2d)' % (t
[6], t
[6])
361 if (t
[2] == 'signed'):
362 expr
= 'sext<%d>(%s)' % (1, expr
)
363 hash_define
= '#undef %s\n#define %s\t%s\n' % (t
[4], t
[4], expr
)
364 t
[0] = GenCode(header_output
= hash_define
)
366 def p_opt_signed_0(t
):
367 'opt_signed : SIGNED'
370 def p_opt_signed_1(t
):
374 # Global map variable to hold templates
377 def p_def_template(t
):
378 'def_template : DEF TEMPLATE ID CODELIT SEMI'
379 templateMap
[t
[3]] = Template(t
[4])
382 # An instruction format definition looks like
383 # "def format <fmt>(<params>) {{...}};"
385 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI'
386 (id, params
, code
) = (t
[3], t
[5], t
[7])
387 defFormat(id, params
, code
, t
.lineno(1))
390 # The formal parameter list for an instruction format is a possibly
391 # empty list of comma-separated parameters. Positional (standard,
392 # non-keyword) parameters must come first, followed by keyword
393 # parameters, followed by a '*foo' parameter that gets excess
394 # positional arguments (as in Python). Each of these three parameter
395 # categories is optional.
397 # Note that we do not support the '**foo' parameter for collecting
398 # otherwise undefined keyword args. Otherwise the parameter list is
399 # (I believe) identical to what is supported in Python.
401 # The param list generates a tuple, where the first element is a list of
402 # the positional params and the second element is a dict containing the
404 def p_param_list_0(t
):
405 'param_list : positional_param_list COMMA nonpositional_param_list'
408 def p_param_list_1(t
):
409 '''param_list : positional_param_list
410 | nonpositional_param_list'''
413 def p_positional_param_list_0(t
):
414 'positional_param_list : empty'
417 def p_positional_param_list_1(t
):
418 'positional_param_list : ID'
421 def p_positional_param_list_2(t
):
422 'positional_param_list : positional_param_list COMMA ID'
425 def p_nonpositional_param_list_0(t
):
426 'nonpositional_param_list : keyword_param_list COMMA excess_args_param'
429 def p_nonpositional_param_list_1(t
):
430 '''nonpositional_param_list : keyword_param_list
431 | excess_args_param'''
434 def p_keyword_param_list_0(t
):
435 'keyword_param_list : keyword_param'
438 def p_keyword_param_list_1(t
):
439 'keyword_param_list : keyword_param_list COMMA keyword_param'
442 def p_keyword_param(t
):
443 'keyword_param : ID EQUALS expr'
444 t
[0] = t
[1] + ' = ' + t
[3].__repr
__()
446 def p_excess_args_param(t
):
447 'excess_args_param : ASTERISK ID'
448 # Just concatenate them: '*ID'. Wrap in list to be consistent
449 # with positional_param_list and keyword_param_list.
452 # End of format definition-related rules.
456 # A decode block looks like:
457 # decode <field1> [, <field2>]* [default <inst>] { ... }
459 def p_decode_block(t
):
460 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE'
461 default_defaults
= defaultStack
.pop()
463 # use the "default defaults" only if there was no explicit
464 # default statement in decode_stmt_list
465 if not codeObj
.has_decode_default
:
466 codeObj
+= default_defaults
467 codeObj
.wrap_decode_block('switch (%s) {\n' % t
[2], '}\n')
470 # The opt_default statement serves only to push the "default defaults"
471 # onto defaultStack. This value will be used by nested decode blocks,
472 # and used and popped off when the current decode_block is processed
473 # (in p_decode_block() above).
474 def p_opt_default_0(t
):
475 'opt_default : empty'
476 # no default specified: reuse the one currently at the top of the stack
477 defaultStack
.push(defaultStack
.top())
478 # no meaningful value returned
481 def p_opt_default_1(t
):
482 'opt_default : DEFAULT inst'
483 # push the new default
485 codeObj
.wrap_decode_block('\ndefault:\n', 'break;\n')
486 defaultStack
.push(codeObj
)
487 # no meaningful value returned
490 def p_decode_stmt_list_0(t
):
491 'decode_stmt_list : decode_stmt'
494 def p_decode_stmt_list_1(t
):
495 'decode_stmt_list : decode_stmt decode_stmt_list'
496 if (t
[1].has_decode_default
and t
[2].has_decode_default
):
497 error(t
.lineno(1), 'Two default cases in decode block')
501 # Decode statement rules
503 # There are four types of statements allowed in a decode block:
504 # 1. Format blocks 'format <foo> { ... }'
505 # 2. Nested decode blocks
506 # 3. Instruction definitions.
507 # 4. C preprocessor directives.
510 # Preprocessor directives found in a decode statement list are passed
511 # through to the output, replicated to all of the output code
512 # streams. This works well for ifdefs, so we can ifdef out both the
513 # declarations and the decode cases generated by an instruction
514 # definition. Handling them as part of the grammar makes it easy to
515 # keep them in the right place with respect to the code generated by
516 # the other statements.
517 def p_decode_stmt_cpp(t
):
518 'decode_stmt : CPPDIRECTIVE'
519 t
[0] = GenCode(t
[1], t
[1], t
[1], t
[1])
521 # A format block 'format <foo> { ... }' sets the default instruction
522 # format used to handle instruction definitions inside the block.
523 # This format can be overridden by using an explicit format on the
524 # instruction definition or with a nested format block.
525 def p_decode_stmt_format(t
):
526 'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE'
527 # The format will be pushed on the stack when 'push_format_id' is
528 # processed (see below). Once the parser has recognized the full
529 # production (though the right brace), we're done with the format,
530 # so now we can pop it.
534 # This rule exists so we can set the current format (& push the stack)
535 # when we recognize the format name part of the format block.
536 def p_push_format_id(t
):
537 'push_format_id : ID'
539 formatStack
.push(formatMap
[t
[1]])
540 t
[0] = ('', '// format %s' % t
[1])
542 error(t
.lineno(1), 'instruction format "%s" not defined.' % t
[1])
544 # Nested decode block: if the value of the current field matches the
545 # specified constant, do a nested decode on some other field.
546 def p_decode_stmt_decode(t
):
547 'decode_stmt : case_label COLON decode_block'
550 # just wrap the decoding code from the block as a case in the
551 # outer switch statement.
552 codeObj
.wrap_decode_block('\n%s:\n' % label
)
553 codeObj
.has_decode_default
= (label
== 'default')
556 # Instruction definition (finally!).
557 def p_decode_stmt_inst(t
):
558 'decode_stmt : case_label COLON inst SEMI'
561 codeObj
.wrap_decode_block('\n%s:' % label
, 'break;\n')
562 codeObj
.has_decode_default
= (label
== 'default')
565 # The case label is either a list of one or more constants or 'default'
566 def p_case_label_0(t
):
567 'case_label : intlit_list'
568 t
[0] = ': '.join(map(lambda a
: 'case %#x' % a
, t
[1]))
570 def p_case_label_1(t
):
571 'case_label : DEFAULT'
575 # The constant list for a decode case label must be non-empty, but may have
576 # one or more comma-separated integer literals in it.
578 def p_intlit_list_0(t
):
579 'intlit_list : INTLIT'
582 def p_intlit_list_1(t
):
583 'intlit_list : intlit_list COMMA INTLIT'
587 # Define an instruction using the current instruction format (specified
588 # by an enclosing format block).
589 # "<mnemonic>(<args>)"
591 'inst : ID LPAREN arg_list RPAREN'
592 # Pass the ID and arg list to the current format class to deal with.
593 currentFormat
= formatStack
.top()
594 codeObj
= currentFormat
.defineInst(t
[1], t
[3], t
.lineno(1))
595 args
= ','.join(map(str, t
[3]))
596 args
= re
.sub('(?m)^', '//', args
)
597 args
= re
.sub('^//', '', args
)
598 comment
= '\n// %s::%s(%s)\n' % (currentFormat
.id, t
[1], args
)
599 codeObj
.prepend_all(comment
)
602 # Define an instruction using an explicitly specified format:
603 # "<fmt>::<mnemonic>(<args>)"
605 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN'
607 format
= formatMap
[t
[1]]
609 error(t
.lineno(1), 'instruction format "%s" not defined.' % t
[1])
610 codeObj
= format
.defineInst(t
[3], t
[5], t
.lineno(1))
611 comment
= '\n// %s::%s(%s)\n' % (t
[1], t
[3], t
[5])
612 codeObj
.prepend_all(comment
)
615 # The arg list generates a tuple, where the first element is a list of
616 # the positional args and the second element is a dict containing the
619 'arg_list : positional_arg_list COMMA keyword_arg_list'
620 t
[0] = ( t
[1], t
[3] )
623 'arg_list : positional_arg_list'
627 'arg_list : keyword_arg_list'
630 def p_positional_arg_list_0(t
):
631 'positional_arg_list : empty'
634 def p_positional_arg_list_1(t
):
635 'positional_arg_list : expr'
638 def p_positional_arg_list_2(t
):
639 'positional_arg_list : positional_arg_list COMMA expr'
642 def p_keyword_arg_list_0(t
):
643 'keyword_arg_list : keyword_arg'
646 def p_keyword_arg_list_1(t
):
647 'keyword_arg_list : keyword_arg_list COMMA keyword_arg'
651 def p_keyword_arg(t
):
652 'keyword_arg : ID EQUALS expr'
653 t
[0] = { t
[1] : t
[3] }
656 # Basic expressions. These constitute the argument values of
657 # "function calls" (i.e. instruction definitions in the decode block)
658 # and default values for formal parameters of format functions.
660 # Right now, these are either strings, integers, or (recursively)
661 # lists of exprs (using Python square-bracket list syntax). Note that
662 # bare identifiers are trated as string constants here (since there
663 # isn't really a variable namespace to refer to).
673 '''expr : LBRACKET list_expr RBRACKET'''
676 def p_list_expr_0(t
):
680 def p_list_expr_1(t
):
681 'list_expr : list_expr COMMA expr'
684 def p_list_expr_2(t
):
689 # Empty production... use in other rules for readability.
695 # Parse error handler. Note that the argument here is the offending
696 # *token*, not a grammar symbol (hence the need to use t.value)
699 error(t
.lineno
, "syntax error at '%s'" % t
.value
)
701 error(0, "unknown syntax error", True)
703 # END OF GRAMMAR RULES
705 # Now build the parser.
709 #####################################################################
713 #####################################################################
715 # Expand template with CPU-specific references into a dictionary with
716 # an entry for each CPU model name. The entry key is the model name
717 # and the corresponding value is the template with the CPU-specific
718 # refs substituted for that model.
719 def expand_cpu_symbols_to_dict(template
):
720 # Protect '%'s that don't go with CPU-specific terms
721 t
= re
.sub(r
'%(?!\(CPU_)', '%%', template
)
723 for cpu
in cpu_models
:
724 result
[cpu
.name
] = t
% cpu
.strings
727 # *If* the template has CPU-specific references, return a single
728 # string containing a copy of the template for each CPU model with the
729 # corresponding values substituted in. If the template has no
730 # CPU-specific references, it is returned unmodified.
731 def expand_cpu_symbols_to_string(template
):
732 if template
.find('%(CPU_') != -1:
733 return reduce(lambda x
,y
: x
+y
,
734 expand_cpu_symbols_to_dict(template
).values())
738 # Protect CPU-specific references by doubling the corresponding '%'s
739 # (in preparation for substituting a different set of references into
741 def protect_cpu_symbols(template
):
742 return re
.sub(r
'%(?=\(CPU_)', '%%', template
)
747 # The GenCode class encapsulates generated code destined for various
748 # output files. The header_output and decoder_output attributes are
749 # strings containing code destined for decoder.hh and decoder.cc
750 # respectively. The decode_block attribute contains code to be
751 # incorporated in the decode function itself (that will also end up in
752 # decoder.cc). The exec_output attribute is a dictionary with a key
753 # for each CPU model name; the value associated with a particular key
754 # is the string of code for that CPU model's exec.cc file. The
755 # has_decode_default attribute is used in the decode block to allow
756 # explicit default clauses to override default default clauses.
759 # Constructor. At this point we substitute out all CPU-specific
760 # symbols. For the exec output, these go into the per-model
761 # dictionary. For all other output types they get collapsed into
764 header_output
= '', decoder_output
= '', exec_output
= '',
765 decode_block
= '', has_decode_default
= False):
766 self
.header_output
= expand_cpu_symbols_to_string(header_output
)
767 self
.decoder_output
= expand_cpu_symbols_to_string(decoder_output
)
768 if isinstance(exec_output
, dict):
769 self
.exec_output
= exec_output
770 elif isinstance(exec_output
, str):
771 # If the exec_output arg is a single string, we replicate
772 # it for each of the CPU models, substituting and
773 # %(CPU_foo)s params appropriately.
774 self
.exec_output
= expand_cpu_symbols_to_dict(exec_output
)
775 self
.decode_block
= expand_cpu_symbols_to_string(decode_block
)
776 self
.has_decode_default
= has_decode_default
778 # Override '+' operator: generate a new GenCode object that
779 # concatenates all the individual strings in the operands.
780 def __add__(self
, other
):
782 for cpu
in cpu_models
:
784 exec_output
[n
] = self
.exec_output
[n
] + other
.exec_output
[n
]
785 return GenCode(self
.header_output
+ other
.header_output
,
786 self
.decoder_output
+ other
.decoder_output
,
788 self
.decode_block
+ other
.decode_block
,
789 self
.has_decode_default
or other
.has_decode_default
)
791 # Prepend a string (typically a comment) to all the strings.
792 def prepend_all(self
, pre
):
793 self
.header_output
= pre
+ self
.header_output
794 self
.decoder_output
= pre
+ self
.decoder_output
795 self
.decode_block
= pre
+ self
.decode_block
796 for cpu
in cpu_models
:
797 self
.exec_output
[cpu
.name
] = pre
+ self
.exec_output
[cpu
.name
]
799 # Wrap the decode block in a pair of strings (e.g., 'case foo:'
800 # and 'break;'). Used to build the big nested switch statement.
801 def wrap_decode_block(self
, pre
, post
= ''):
802 self
.decode_block
= pre
+ indent(self
.decode_block
) + post
807 # A format object encapsulates an instruction format. It must provide
808 # a defineInst() method that generates the code for an instruction
811 exportContextSymbols
= ('InstObjParams', 'CodeBlock',
812 'makeList', 're', 'string')
816 def updateExportContext():
817 exportContext
.update(exportDict(*exportContextSymbols
))
818 exportContext
.update(templateMap
)
820 def exportDict(*symNames
):
821 return dict([(s
, eval(s
)) for s
in symNames
])
825 def __init__(self
, id, params
, code
):
826 # constructor: just save away arguments
829 label
= 'def format ' + id
830 self
.user_code
= compile(fixPythonIndentation(code
), label
, 'exec')
831 param_list
= string
.join(params
, ", ")
832 f
= '''def defInst(_code, _context, %s):
833 my_locals = vars().copy()
834 exec _code in _context, my_locals
835 return my_locals\n''' % param_list
836 c
= compile(f
, label
+ ' wrapper', 'exec')
840 def defineInst(self
, name
, args
, lineno
):
842 updateExportContext()
843 context
.update(exportContext
)
844 context
.update({ 'name': name
, 'Name': string
.capitalize(name
) })
846 vars = self
.func(self
.user_code
, context
, *args
[0], **args
[1])
847 except Exception, exc
:
848 error(lineno
, 'error defining "%s": %s.' % (name
, exc
))
849 for k
in vars.keys():
850 if k
not in ('header_output', 'decoder_output',
851 'exec_output', 'decode_block'):
853 return GenCode(**vars)
855 # Special null format to catch an implicit-format instruction
856 # definition outside of any format block.
859 self
.defaultInst
= ''
861 def defineInst(self
, name
, args
, lineno
):
863 'instruction definition "%s" with no active format!' % name
)
865 # This dictionary maps format name strings to Format objects.
868 # Define a new format
869 def defFormat(id, params
, code
, lineno
):
870 # make sure we haven't already defined this one
871 if formatMap
.get(id, None) != None:
872 error(lineno
, 'format %s redefined.' % id)
873 # create new object and store in global map
874 formatMap
[id] = Format(id, params
, code
)
878 # Stack: a simple stack object. Used for both formats (formatStack)
879 # and default cases (defaultStack). Simply wraps a list to give more
880 # stack-like syntax and enable initialization with an argument list
881 # (as opposed to an argument that's a list).
884 def __init__(self
, *items
):
885 list.__init
__(self
, items
)
887 def push(self
, item
):
893 # The global format stack.
894 formatStack
= Stack(NoFormat())
896 # The global default case stack.
897 defaultStack
= Stack( None )
899 # Global stack that tracks current file and line number.
900 # Each element is a tuple (filename, lineno) that records the
901 # *current* filename and the line number in the *previous* file where
903 fileNameStack
= Stack()
909 # Indent every line in string 's' by two spaces
910 # (except preprocessor directives).
911 # Used to make nested code blocks look pretty.
914 return re
.sub(r
'(?m)^(?!#)', ' ', s
)
917 # Munge a somewhat arbitrarily formatted piece of Python code
918 # (e.g. from a format 'let' block) into something whose indentation
919 # will get by the Python parser.
921 # The two keys here are that Python will give a syntax error if
922 # there's any whitespace at the beginning of the first line, and that
923 # all lines at the same lexical nesting level must have identical
924 # indentation. Unfortunately the way code literals work, an entire
925 # let block tends to have some initial indentation. Rather than
926 # trying to figure out what that is and strip it off, we prepend 'if
927 # 1:' to make the let code the nested block inside the if (and have
928 # the parser automatically deal with the indentation for us).
930 # We don't want to do this if (1) the code block is empty or (2) the
931 # first line of the block doesn't have any whitespace at the front.
933 def fixPythonIndentation(s
):
934 # get rid of blank lines first
935 s
= re
.sub(r
'(?m)^\s*\n', '', s
);
936 if (s
!= '' and re
.match(r
'[ \t]', s
[0])):
940 # Error handler. Just call exit. Output formatted to work under
941 # Emacs compile-mode. Optional 'print_traceback' arg, if set to True,
942 # prints a Python stack backtrace too (can be handy when trying to
943 # debug the parser itself).
944 def error(lineno
, string
, print_traceback
= False):
946 for (filename
, line
) in fileNameStack
[0:-1]:
947 print spaces
+ "In file included from " + filename
+ ":"
949 # Print a Python stack backtrace if requested.
950 if (print_traceback
):
951 traceback
.print_exc()
953 line_str
= "%d:" % lineno
956 sys
.exit(spaces
+ "%s:%s %s" % (fileNameStack
[-1][0], line_str
, string
))
959 #####################################################################
961 # Bitfield Operator Support
963 #####################################################################
965 bitOp1ArgRE
= re
.compile(r
'<\s*(\w+)\s*:\s*>')
967 bitOpWordRE
= re
.compile(r
'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>')
968 bitOpExprRE
= re
.compile(r
'\)<\s*(\w+)\s*:\s*(\w+)\s*>')
970 def substBitOps(code
):
971 # first convert single-bit selectors to two-index form
972 # i.e., <n> --> <n:n>
973 code
= bitOp1ArgRE
.sub(r
'<\1:\1>', code
)
974 # simple case: selector applied to ID (name)
975 # i.e., foo<a:b> --> bits(foo, a, b)
976 code
= bitOpWordRE
.sub(r
'bits(\1, \2, \3)', code
)
977 # if selector is applied to expression (ending in ')'),
978 # we need to search backward for matching '('
979 match
= bitOpExprRE
.search(code
)
981 exprEnd
= match
.start()
985 if code
[here
] == '(':
987 elif code
[here
] == ')':
991 sys
.exit("Didn't find '('!")
993 newExpr
= r
'bits(%s, %s, %s)' % (code
[exprStart
:exprEnd
+1],
994 match
.group(1), match
.group(2))
995 code
= code
[:exprStart
] + newExpr
+ code
[match
.end():]
996 match
= bitOpExprRE
.search(code
)
1000 ####################
1003 # Template objects are format strings that allow substitution from
1004 # the attribute spaces of other objects (e.g. InstObjParams instances).
1007 def __init__(self
, t
):
1011 # Start with the template namespace. Make a copy since we're
1012 # going to modify it.
1013 myDict
= templateMap
.copy()
1014 # if the argument is a dictionary, we just use it.
1015 if isinstance(d
, dict):
1017 # if the argument is an object, we use its attribute map.
1018 elif hasattr(d
, '__dict__'):
1019 myDict
.update(d
.__dict
__)
1021 raise TypeError, "Template.subst() arg must be or have dictionary"
1022 # Protect non-Python-dict substitutions (e.g. if there's a printf
1023 # in the templated C++ code)
1024 template
= protect_non_subst_percents(self
.template
)
1025 # CPU-model-specific substitutions are handled later (in GenCode).
1026 template
= protect_cpu_symbols(template
)
1027 return template
% myDict
1029 # Convert to string. This handles the case when a template with a
1030 # CPU-specific term gets interpolated into another template or into
1033 return expand_cpu_symbols_to_string(self
.template
)
1035 #####################################################################
1039 # The remaining code is the support for automatically extracting
1040 # instruction characteristics from pseudocode.
1042 #####################################################################
1044 # Force the argument to be a list. Useful for flags, where a caller
1045 # can specify a singleton flag or a list of flags. Also usful for
1046 # converting tuples to lists so they can be modified.
1048 if isinstance(arg
, list):
1050 elif isinstance(arg
, tuple):
1057 # Generate operandTypeMap from the user's 'def operand_types'
1059 def buildOperandTypeMap(userDict
, lineno
):
1060 global operandTypeMap
1062 for (ext
, (desc
, size
)) in userDict
.iteritems():
1063 if desc
== 'signed int':
1064 ctype
= 'int%d_t' % size
1066 elif desc
== 'unsigned int':
1067 ctype
= 'uint%d_t' % size
1069 elif desc
== 'float':
1070 is_signed
= 1 # shouldn't really matter
1076 error(lineno
, 'Unrecognized type description "%s" in userDict')
1077 operandTypeMap
[ext
] = (size
, ctype
, is_signed
)
1082 # Base class for operand descriptors. An instance of this class (or
1083 # actually a class derived from this one) represents a specific
1084 # operand for a code block (e.g, "Rc.sq" as a dest). Intermediate
1085 # derived classes encapsulates the traits of a particular operand type
1086 # (e.g., "32-bit integer register").
1088 class Operand(object):
1089 def __init__(self
, full_name
, ext
, is_src
, is_dest
):
1090 self
.full_name
= full_name
1092 self
.is_src
= is_src
1093 self
.is_dest
= is_dest
1094 # The 'effective extension' (eff_ext) is either the actual
1095 # extension, if one was explicitly provided, or the default.
1099 self
.eff_ext
= self
.dflt_ext
1101 (self
.size
, self
.ctype
, self
.is_signed
) = operandTypeMap
[self
.eff_ext
]
1103 # note that mem_acc_size is undefined for non-mem operands...
1104 # template must be careful not to use it if it doesn't apply.
1106 self
.mem_acc_size
= self
.makeAccSize()
1107 self
.mem_acc_type
= self
.ctype
1109 # Finalize additional fields (primarily code fields). This step
1110 # is done separately since some of these fields may depend on the
1111 # register index enumeration that hasn't been performed yet at the
1112 # time of __init__().
1114 self
.flags
= self
.getFlags()
1115 self
.constructor
= self
.makeConstructor()
1116 self
.op_decl
= self
.makeDecl()
1119 self
.op_rd
= self
.makeRead()
1120 self
.op_src_decl
= self
.makeDecl()
1123 self
.op_src_decl
= ''
1126 self
.op_wb
= self
.makeWrite()
1127 self
.op_dest_decl
= self
.makeDecl()
1130 self
.op_dest_decl
= ''
1138 def isFloatReg(self
):
1144 def isControlReg(self
):
1148 # note the empty slice '[:]' gives us a copy of self.flags[0]
1149 # instead of a reference to it
1150 my_flags
= self
.flags
[0][:]
1152 my_flags
+= self
.flags
[1]
1154 my_flags
+= self
.flags
[2]
1158 # Note that initializations in the declarations are solely
1159 # to avoid 'uninitialized variable' errors from the compiler.
1160 return self
.ctype
+ ' ' + self
.base_name
+ ' = 0;\n';
1162 class IntRegOperand(Operand
):
1169 def makeConstructor(self
):
1172 c
+= '\n\t_srcRegIdx[%d] = %s;' % \
1173 (self
.src_reg_idx
, self
.reg_spec
)
1175 c
+= '\n\t_destRegIdx[%d] = %s;' % \
1176 (self
.dest_reg_idx
, self
.reg_spec
)
1180 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
1181 error(0, 'Attempt to read integer register as FP')
1182 if (self
.size
== self
.dflt_size
):
1183 return '%s = xc->readIntReg(this, %d);\n' % \
1184 (self
.base_name
, self
.src_reg_idx
)
1185 elif (self
.size
> self
.dflt_size
):
1186 int_reg_val
= 'xc->readIntReg(this, %d)' % (self
.src_reg_idx
)
1187 if (self
.is_signed
):
1188 int_reg_val
= 'sext<%d>(%s)' % (self
.dflt_size
, int_reg_val
)
1189 return '%s = %s;\n' % (self
.base_name
, int_reg_val
)
1191 return '%s = bits(xc->readIntReg(this, %d), %d, 0);\n' % \
1192 (self
.base_name
, self
.src_reg_idx
, self
.size
-1)
1194 def makeWrite(self
):
1195 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
1196 error(0, 'Attempt to write integer register as FP')
1197 if (self
.size
!= self
.dflt_size
and self
.is_signed
):
1198 final_val
= 'sext<%d>(%s)' % (self
.size
, self
.base_name
)
1200 final_val
= self
.base_name
1204 xc->setIntReg(this, %d, final_val);\n
1205 if (traceData) { traceData->setData(final_val); }
1206 }''' % (self
.dflt_ctype
, final_val
, self
.dest_reg_idx
)
1209 class FloatRegOperand(Operand
):
1213 def isFloatReg(self
):
1216 def makeConstructor(self
):
1219 c
+= '\n\t_srcRegIdx[%d] = %s + FP_Base_DepTag;' % \
1220 (self
.src_reg_idx
, self
.reg_spec
)
1222 c
+= '\n\t_destRegIdx[%d] = %s + FP_Base_DepTag;' % \
1223 (self
.dest_reg_idx
, self
.reg_spec
)
1229 if (self
.ctype
== 'float'):
1230 func
= 'readFloatReg'
1232 elif (self
.ctype
== 'double'):
1233 func
= 'readFloatReg'
1236 func
= 'readFloatRegBits'
1237 if (self
.ctype
== 'uint32_t'):
1239 elif (self
.ctype
== 'uint64_t'):
1241 if (self
.size
!= self
.dflt_size
):
1244 base
= 'xc->%s(this, %d, %d)' % \
1245 (func
, self
.src_reg_idx
, width
)
1247 base
= 'xc->%s(this, %d)' % \
1248 (func
, self
.src_reg_idx
)
1250 return '%s = bits(%s, %d, 0);\n' % \
1251 (self
.base_name
, base
, self
.size
-1)
1253 return '%s = %s;\n' % (self
.base_name
, base
)
1255 def makeWrite(self
):
1256 final_val
= self
.base_name
1257 final_ctype
= self
.ctype
1260 if (self
.ctype
== 'float'):
1262 func
= 'setFloatReg'
1263 elif (self
.ctype
== 'double'):
1265 func
= 'setFloatReg'
1266 elif (self
.ctype
== 'uint32_t'):
1267 func
= 'setFloatRegBits'
1269 elif (self
.ctype
== 'uint64_t'):
1270 func
= 'setFloatRegBits'
1273 func
= 'setFloatRegBits'
1274 final_ctype
= 'uint%d_t' % self
.dflt_size
1275 if (self
.size
!= self
.dflt_size
and self
.is_signed
):
1276 final_val
= 'sext<%d>(%s)' % (self
.size
, self
.base_name
)
1278 widthSpecifier
= ', %d' % width
1282 xc->%s(this, %d, final_val%s);\n
1283 if (traceData) { traceData->setData(final_val); }
1284 }''' % (final_ctype
, final_val
, func
, self
.dest_reg_idx
,
1288 class ControlRegOperand(Operand
):
1292 def isControlReg(self
):
1295 def makeConstructor(self
):
1298 c
+= '\n\t_srcRegIdx[%d] = %s;' % \
1299 (self
.src_reg_idx
, self
.reg_spec
)
1301 c
+= '\n\t_destRegIdx[%d] = %s;' % \
1302 (self
.dest_reg_idx
, self
.reg_spec
)
1307 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
1308 error(0, 'Attempt to read control register as FP')
1309 base
= 'xc->readMiscReg(%s)' % self
.reg_spec
1310 if self
.size
== self
.dflt_size
:
1311 return '%s = %s;\n' % (self
.base_name
, base
)
1313 return '%s = bits(%s, %d, 0);\n' % \
1314 (self
.base_name
, base
, self
.size
-1)
1316 def makeWrite(self
):
1317 if (self
.ctype
== 'float' or self
.ctype
== 'double'):
1318 error(0, 'Attempt to write control register as FP')
1319 wb
= 'xc->setMiscReg(%s, %s);\n' % (self
.reg_spec
, self
.base_name
)
1320 wb
+= 'if (traceData) { traceData->setData(%s); }' % \
1324 class MemOperand(Operand
):
1328 def makeConstructor(self
):
1332 # Note that initializations in the declarations are solely
1333 # to avoid 'uninitialized variable' errors from the compiler.
1334 # Declare memory data variable.
1335 c
= '%s %s = 0;\n' % (self
.ctype
, self
.base_name
)
1341 def makeWrite(self
):
1344 # Return the memory access size *in bits*, suitable for
1345 # forming a type via "uint%d_t". Divide by 8 if you want bytes.
1346 def makeAccSize(self
):
1350 class NPCOperand(Operand
):
1351 def makeConstructor(self
):
1355 return '%s = xc->readNextPC();\n' % self
.base_name
1357 def makeWrite(self
):
1358 return 'xc->setNextPC(%s);\n' % self
.base_name
1360 class NNPCOperand(Operand
):
1361 def makeConstructor(self
):
1365 return '%s = xc->readNextNPC();\n' % self
.base_name
1367 def makeWrite(self
):
1368 return 'xc->setNextNPC(%s);\n' % self
.base_name
1370 def buildOperandNameMap(userDict
, lineno
):
1371 global operandNameMap
1373 for (op_name
, val
) in userDict
.iteritems():
1374 (base_cls_name
, dflt_ext
, reg_spec
, flags
, sort_pri
) = val
1375 (dflt_size
, dflt_ctype
, dflt_is_signed
) = operandTypeMap
[dflt_ext
]
1376 # Canonical flag structure is a triple of lists, where each list
1377 # indicates the set of flags implied by this operand always, when
1378 # used as a source, and when used as a dest, respectively.
1379 # For simplicity this can be initialized using a variety of fairly
1380 # obvious shortcuts; we convert these to canonical form here.
1382 # no flags specified (e.g., 'None')
1383 flags
= ( [], [], [] )
1384 elif isinstance(flags
, str):
1385 # a single flag: assumed to be unconditional
1386 flags
= ( [ flags
], [], [] )
1387 elif isinstance(flags
, list):
1388 # a list of flags: also assumed to be unconditional
1389 flags
= ( flags
, [], [] )
1390 elif isinstance(flags
, tuple):
1391 # it's a tuple: it should be a triple,
1392 # but each item could be a single string or a list
1393 (uncond_flags
, src_flags
, dest_flags
) = flags
1394 flags
= (makeList(uncond_flags
),
1395 makeList(src_flags
), makeList(dest_flags
))
1396 # Accumulate attributes of new operand class in tmp_dict
1398 for attr
in ('dflt_ext', 'reg_spec', 'flags', 'sort_pri',
1399 'dflt_size', 'dflt_ctype', 'dflt_is_signed'):
1400 tmp_dict
[attr
] = eval(attr
)
1401 tmp_dict
['base_name'] = op_name
1402 # New class name will be e.g. "IntReg_Ra"
1403 cls_name
= base_cls_name
+ '_' + op_name
1404 # Evaluate string arg to get class object. Note that the
1405 # actual base class for "IntReg" is "IntRegOperand", i.e. we
1406 # have to append "Operand".
1408 base_cls
= eval(base_cls_name
+ 'Operand')
1411 'error: unknown operand base class "%s"' % base_cls_name
)
1412 # The following statement creates a new class called
1413 # <cls_name> as a subclass of <base_cls> with the attributes
1414 # in tmp_dict, just as if we evaluated a class declaration.
1415 operandNameMap
[op_name
] = type(cls_name
, (base_cls
,), tmp_dict
)
1417 # Define operand variables.
1418 operands
= userDict
.keys()
1420 operandsREString
= (r
'''
1421 (?<![\w\.]) # neg. lookbehind assertion: prevent partial matches
1422 ((%s)(?:\.(\w+))?) # match: operand with optional '.' then suffix
1423 (?![\w\.]) # neg. lookahead assertion: prevent partial matches
1425 % string
.join(operands
, '|'))
1428 operandsRE
= re
.compile(operandsREString
, re
.MULTILINE|re
.VERBOSE
)
1430 # Same as operandsREString, but extension is mandatory, and only two
1431 # groups are returned (base and ext, not full name as above).
1432 # Used for subtituting '_' for '.' to make C++ identifiers.
1433 operandsWithExtREString
= (r
'(?<![\w\.])(%s)\.(\w+)(?![\w\.])'
1434 % string
.join(operands
, '|'))
1436 global operandsWithExtRE
1437 operandsWithExtRE
= re
.compile(operandsWithExtREString
, re
.MULTILINE
)
1442 # Find all the operands in the given code block. Returns an operand
1443 # descriptor list (instance of class OperandList).
1444 def __init__(self
, code
):
1447 # delete comments so we don't match on reg specifiers inside
1448 code
= commentRE
.sub('', code
)
1449 # search for operands
1452 match
= operandsRE
.search(code
, next_pos
)
1454 # no more matches: we're done
1457 # regexp groups are operand full name, base, and extension
1458 (op_full
, op_base
, op_ext
) = op
1459 # if the token following the operand is an assignment, this is
1460 # a destination (LHS), else it's a source (RHS)
1461 is_dest
= (assignRE
.match(code
, match
.end()) != None)
1462 is_src
= not is_dest
1463 # see if we've already seen this one
1464 op_desc
= self
.find_base(op_base
)
1466 if op_desc
.ext
!= op_ext
:
1467 error(0, 'Inconsistent extensions for operand %s' % \
1469 op_desc
.is_src
= op_desc
.is_src
or is_src
1470 op_desc
.is_dest
= op_desc
.is_dest
or is_dest
1472 # new operand: create new descriptor
1473 op_desc
= operandNameMap
[op_base
](op_full
, op_ext
,
1475 self
.append(op_desc
)
1476 # start next search after end of current match
1477 next_pos
= match
.end()
1479 # enumerate source & dest register operands... used in building
1482 self
.numDestRegs
= 0
1483 self
.numFPDestRegs
= 0
1484 self
.numIntDestRegs
= 0
1485 self
.memOperand
= None
1486 for op_desc
in self
.items
:
1489 op_desc
.src_reg_idx
= self
.numSrcRegs
1490 self
.numSrcRegs
+= 1
1492 op_desc
.dest_reg_idx
= self
.numDestRegs
1493 self
.numDestRegs
+= 1
1494 if op_desc
.isFloatReg():
1495 self
.numFPDestRegs
+= 1
1496 elif op_desc
.isIntReg():
1497 self
.numIntDestRegs
+= 1
1498 elif op_desc
.isMem():
1500 error(0, "Code block has more than one memory operand.")
1501 self
.memOperand
= op_desc
1502 # now make a final pass to finalize op_desc fields that may depend
1503 # on the register enumeration
1504 for op_desc
in self
.items
:
1508 return len(self
.items
)
1510 def __getitem__(self
, index
):
1511 return self
.items
[index
]
1513 def append(self
, op_desc
):
1514 self
.items
.append(op_desc
)
1515 self
.bases
[op_desc
.base_name
] = op_desc
1517 def find_base(self
, base_name
):
1518 # like self.bases[base_name], but returns None if not found
1519 # (rather than raising exception)
1520 return self
.bases
.get(base_name
)
1522 # internal helper function for concat[Some]Attr{Strings|Lists}
1523 def __internalConcatAttrs(self
, attr_name
, filter, result
):
1524 for op_desc
in self
.items
:
1526 result
+= getattr(op_desc
, attr_name
)
1529 # return a single string that is the concatenation of the (string)
1530 # values of the specified attribute for all operands
1531 def concatAttrStrings(self
, attr_name
):
1532 return self
.__internalConcatAttrs
(attr_name
, lambda x
: 1, '')
1534 # like concatAttrStrings, but only include the values for the operands
1535 # for which the provided filter function returns true
1536 def concatSomeAttrStrings(self
, filter, attr_name
):
1537 return self
.__internalConcatAttrs
(attr_name
, filter, '')
1539 # return a single list that is the concatenation of the (list)
1540 # values of the specified attribute for all operands
1541 def concatAttrLists(self
, attr_name
):
1542 return self
.__internalConcatAttrs
(attr_name
, lambda x
: 1, [])
1544 # like concatAttrLists, but only include the values for the operands
1545 # for which the provided filter function returns true
1546 def concatSomeAttrLists(self
, filter, attr_name
):
1547 return self
.__internalConcatAttrs
(attr_name
, filter, [])
1550 self
.items
.sort(lambda a
, b
: a
.sort_pri
- b
.sort_pri
)
1552 # Regular expression object to match C++ comments
1553 # (used in findOperands())
1554 commentRE
= re
.compile(r
'//.*\n')
1556 # Regular expression object to match assignment statements
1557 # (used in findOperands())
1558 assignRE
= re
.compile(r
'\s*=(?!=)', re
.MULTILINE
)
1560 # Munge operand names in code string to make legal C++ variable names.
1561 # This means getting rid of the type extension if any.
1562 # (Will match base_name attribute of Operand object.)
1563 def substMungedOpNames(code
):
1564 return operandsWithExtRE
.sub(r
'\1', code
)
1567 return map(string
.join
, t
)
1569 def makeFlagConstructor(flag_list
):
1570 if len(flag_list
) == 0:
1572 # filter out repeated flags
1575 while i
< len(flag_list
):
1576 if flag_list
[i
] == flag_list
[i
-1]:
1582 code
= pre
+ string
.join(flag_list
, post
+ pre
) + post
1586 def __init__(self
, code
):
1587 self
.orig_code
= code
1588 self
.operands
= OperandList(code
)
1589 self
.code
= substMungedOpNames(substBitOps(code
))
1590 self
.constructor
= self
.operands
.concatAttrStrings('constructor')
1591 self
.constructor
+= \
1592 '\n\t_numSrcRegs = %d;' % self
.operands
.numSrcRegs
1593 self
.constructor
+= \
1594 '\n\t_numDestRegs = %d;' % self
.operands
.numDestRegs
1595 self
.constructor
+= \
1596 '\n\t_numFPDestRegs = %d;' % self
.operands
.numFPDestRegs
1597 self
.constructor
+= \
1598 '\n\t_numIntDestRegs = %d;' % self
.operands
.numIntDestRegs
1600 self
.op_decl
= self
.operands
.concatAttrStrings('op_decl')
1602 is_src
= lambda op
: op
.is_src
1603 is_dest
= lambda op
: op
.is_dest
1605 self
.op_src_decl
= \
1606 self
.operands
.concatSomeAttrStrings(is_src
, 'op_src_decl')
1607 self
.op_dest_decl
= \
1608 self
.operands
.concatSomeAttrStrings(is_dest
, 'op_dest_decl')
1610 self
.op_rd
= self
.operands
.concatAttrStrings('op_rd')
1611 self
.op_wb
= self
.operands
.concatAttrStrings('op_wb')
1613 self
.flags
= self
.operands
.concatAttrLists('flags')
1615 if self
.operands
.memOperand
:
1616 self
.mem_acc_size
= self
.operands
.memOperand
.mem_acc_size
1617 self
.mem_acc_type
= self
.operands
.memOperand
.mem_acc_type
1619 # Make a basic guess on the operand class (function unit type).
1620 # These are good enough for most cases, and will be overridden
1622 if 'IsStore' in self
.flags
:
1623 self
.op_class
= 'MemWriteOp'
1624 elif 'IsLoad' in self
.flags
or 'IsPrefetch' in self
.flags
:
1625 self
.op_class
= 'MemReadOp'
1626 elif 'IsFloating' in self
.flags
:
1627 self
.op_class
= 'FloatAddOp'
1629 self
.op_class
= 'IntAluOp'
1631 # Assume all instruction flags are of the form 'IsFoo'
1632 instFlagRE
= re
.compile(r
'Is.*')
1634 # OpClass constants end in 'Op' except No_OpClass
1635 opClassRE
= re
.compile(r
'.*Op|No_OpClass')
1637 class InstObjParams
:
1638 def __init__(self
, mnem
, class_name
, base_class
= '',
1639 code
= None, opt_args
= [], *extras
):
1640 self
.mnemonic
= mnem
1641 self
.class_name
= class_name
1642 self
.base_class
= base_class
1644 #If the user already made a CodeBlock, pick the parts from it
1645 if isinstance(code
, CodeBlock
):
1646 origCode
= code
.orig_code
1650 codeBlock
= CodeBlock(code
)
1651 compositeCode
= '\n'.join([origCode
] +
1652 [pair
[1] for pair
in extras
])
1653 compositeBlock
= CodeBlock(compositeCode
)
1654 for code_attr
in compositeBlock
.__dict
__.keys():
1655 setattr(self
, code_attr
, getattr(compositeBlock
, code_attr
))
1656 for (key
, snippet
) in extras
:
1657 setattr(self
, key
, CodeBlock(snippet
).code
)
1658 self
.code
= codeBlock
.code
1659 self
.orig_code
= origCode
1661 self
.constructor
= ''
1663 # Optional arguments are assumed to be either StaticInst flags
1664 # or an OpClass value. To avoid having to import a complete
1665 # list of these values to match against, we do it ad-hoc
1668 if instFlagRE
.match(oa
):
1669 self
.flags
.append(oa
)
1670 elif opClassRE
.match(oa
):
1673 error(0, 'InstObjParams: optional arg "%s" not recognized '
1674 'as StaticInst::Flag or OpClass.' % oa
)
1676 # add flag initialization to contructor here to include
1677 # any flags added via opt_args
1678 self
.constructor
+= makeFlagConstructor(self
.flags
)
1680 # if 'IsFloating' is set, add call to the FP enable check
1681 # function (which should be provided by isa_desc via a declare)
1682 if 'IsFloating' in self
.flags
:
1683 self
.fp_enable_check
= 'fault = checkFpEnableFault(xc);'
1685 self
.fp_enable_check
= ''
1687 #######################
1689 # Output file template
1694 * DO NOT EDIT THIS FILE!!!
1696 * It was automatically generated from the ISA description in %(filename)s
1703 namespace %(namespace)s {
1705 %(namespace_output)s
1707 } // namespace %(namespace)s
1713 # Update the output file only if the new contents are different from
1714 # the current contents. Minimizes the files that need to be rebuilt
1715 # after minor changes.
1716 def update_if_needed(file, contents
):
1718 if os
.access(file, os
.R_OK
):
1720 old_contents
= f
.read()
1722 if contents
!= old_contents
:
1723 print 'Updating', file
1724 os
.remove(file) # in case it's write-protected
1727 print 'File', file, 'is unchanged'
1729 print 'Generating', file
1736 # This regular expression matches '##include' directives
1737 includeRE
= re
.compile(r
'^\s*##include\s+"(?P<filename>[\w/.-]*)".*$',
1740 # Function to replace a matched '##include' directive with the
1741 # contents of the specified file (with nested ##includes replaced
1742 # recursively). 'matchobj' is an re match object (from a match of
1743 # includeRE) and 'dirname' is the directory relative to which the file
1744 # path should be resolved.
1745 def replace_include(matchobj
, dirname
):
1746 fname
= matchobj
.group('filename')
1747 full_fname
= os
.path
.normpath(os
.path
.join(dirname
, fname
))
1748 contents
= '##newfile "%s"\n%s\n##endfile\n' % \
1749 (full_fname
, read_and_flatten(full_fname
))
1752 # Read a file and recursively flatten nested '##include' files.
1753 def read_and_flatten(filename
):
1754 current_dir
= os
.path
.dirname(filename
)
1756 contents
= open(filename
).read()
1758 error(0, 'Error including file "%s"' % filename
)
1759 fileNameStack
.push((filename
, 0))
1760 # Find any includes and include them
1761 contents
= includeRE
.sub(lambda m
: replace_include(m
, current_dir
),
1767 # Read in and parse the ISA description.
1769 def parse_isa_desc(isa_desc_file
, output_dir
):
1770 # Read file and (recursively) all included files into a string.
1771 # PLY requires that the input be in a single string so we have to
1773 isa_desc
= read_and_flatten(isa_desc_file
)
1775 # Initialize filename stack with outer file.
1776 fileNameStack
.push((isa_desc_file
, 0))
1779 (isa_name
, namespace
, global_code
, namespace_code
) = yacc
.parse(isa_desc
)
1781 # grab the last three path components of isa_desc_file to put in
1783 filename
= '/'.join(isa_desc_file
.split('/')[-3:])
1785 # generate decoder.hh
1786 includes
= '#include "base/bitfield.hh" // for bitfield support'
1787 global_output
= global_code
.header_output
1788 namespace_output
= namespace_code
.header_output
1789 decode_function
= ''
1790 update_if_needed(output_dir
+ '/decoder.hh', file_template
% vars())
1792 # generate decoder.cc
1793 includes
= '#include "decoder.hh"'
1794 global_output
= global_code
.decoder_output
1795 namespace_output
= namespace_code
.decoder_output
1796 # namespace_output += namespace_code.decode_block
1797 decode_function
= namespace_code
.decode_block
1798 update_if_needed(output_dir
+ '/decoder.cc', file_template
% vars())
1800 # generate per-cpu exec files
1801 for cpu
in cpu_models
:
1802 includes
= '#include "decoder.hh"\n'
1803 includes
+= cpu
.includes
1804 global_output
= global_code
.exec_output
[cpu
.name
]
1805 namespace_output
= namespace_code
.exec_output
[cpu
.name
]
1806 decode_function
= ''
1807 update_if_needed(output_dir
+ '/' + cpu
.filename
,
1808 file_template
% vars())
1810 # global list of CpuModel objects (see cpu_models.py)
1813 # Called as script: get args from command line.
1814 # Args are: <path to cpu_models.py> <isa desc file> <output dir> <cpu models>
1815 if __name__
== '__main__':
1816 execfile(sys
.argv
[1]) # read in CpuModel definitions
1817 cpu_models
= [CpuModel
.dict[cpu
] for cpu
in sys
.argv
[4:]]
1818 parse_isa_desc(sys
.argv
[2], sys
.argv
[3])