isa_parser.py revision 3954:d689b611d9dc
1# Copyright (c) 2003-2005 The Regents of The University of Michigan
2# All rights reserved.
3#
4# Redistribution and use in source and binary forms, with or without
5# modification, are permitted provided that the following conditions are
6# met: redistributions of source code must retain the above copyright
7# notice, this list of conditions and the following disclaimer;
8# redistributions in binary form must reproduce the above copyright
9# notice, this list of conditions and the following disclaimer in the
10# documentation and/or other materials provided with the distribution;
11# neither the name of the copyright holders nor the names of its
12# contributors may be used to endorse or promote products derived from
13# this software without specific prior written permission.
14#
15# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
18# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
19# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
20# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
21# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
25# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26#
27# Authors: Steve Reinhardt
28#          Korey Sewell
29
30import os
31import sys
32import re
33import string
34import traceback
35# get type names
36from types import *
37
38# Prepend the directory where the PLY lex & yacc modules are found
39# to the search path.  Assumes we're compiling in a subdirectory
40# of 'build' in the current tree.
41sys.path[0:0] = [os.environ['M5_PLY']]
42
43import lex
44import yacc
45
46#####################################################################
47#
48#                                Lexer
49#
50# The PLY lexer module takes two things as input:
51# - A list of token names (the string list 'tokens')
52# - A regular expression describing a match for each token.  The
53#   regexp for token FOO can be provided in two ways:
54#   - as a string variable named t_FOO
55#   - as the doc string for a function named t_FOO.  In this case,
56#     the function is also executed, allowing an action to be
57#     associated with each token match.
58#
59#####################################################################
60
61# Reserved words.  These are listed separately as they are matched
62# using the same regexp as generic IDs, but distinguished in the
63# t_ID() function.  The PLY documentation suggests this approach.
64reserved = (
65    'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT',
66    'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS',
67    'OUTPUT', 'SIGNED', 'TEMPLATE'
68    )
69
70# List of tokens.  The lex module requires this.
71tokens = reserved + (
72    # identifier
73    'ID',
74
75    # integer literal
76    'INTLIT',
77
78    # string literal
79    'STRLIT',
80
81    # code literal
82    'CODELIT',
83
84    # ( ) [ ] { } < > , ; : :: *
85    'LPAREN', 'RPAREN',
86    'LBRACKET', 'RBRACKET',
87    'LBRACE', 'RBRACE',
88    'LESS', 'GREATER', 'EQUALS',
89    'COMMA', 'SEMI', 'COLON', 'DBLCOLON',
90    'ASTERISK',
91
92    # C preprocessor directives
93    'CPPDIRECTIVE'
94
95# The following are matched but never returned. commented out to
96# suppress PLY warning
97    # newfile directive
98#    'NEWFILE',
99
100    # endfile directive
101#    'ENDFILE'
102)
103
104# Regular expressions for token matching
105t_LPAREN           = r'\('
106t_RPAREN           = r'\)'
107t_LBRACKET         = r'\['
108t_RBRACKET         = r'\]'
109t_LBRACE           = r'\{'
110t_RBRACE           = r'\}'
111t_LESS             = r'\<'
112t_GREATER          = r'\>'
113t_EQUALS           = r'='
114t_COMMA            = r','
115t_SEMI             = r';'
116t_COLON            = r':'
117t_DBLCOLON         = r'::'
118t_ASTERISK	   = r'\*'
119
120# Identifiers and reserved words
121reserved_map = { }
122for r in reserved:
123    reserved_map[r.lower()] = r
124
125def t_ID(t):
126    r'[A-Za-z_]\w*'
127    t.type = reserved_map.get(t.value,'ID')
128    return t
129
130# Integer literal
131def t_INTLIT(t):
132    r'(0x[\da-fA-F]+)|\d+'
133    try:
134        t.value = int(t.value,0)
135    except ValueError:
136        error(t.lineno, 'Integer value "%s" too large' % t.value)
137        t.value = 0
138    return t
139
140# String literal.  Note that these use only single quotes, and
141# can span multiple lines.
142def t_STRLIT(t):
143    r"(?m)'([^'])+'"
144    # strip off quotes
145    t.value = t.value[1:-1]
146    t.lineno += t.value.count('\n')
147    return t
148
149
150# "Code literal"... like a string literal, but delimiters are
151# '{{' and '}}' so they get formatted nicely under emacs c-mode
152def t_CODELIT(t):
153    r"(?m)\{\{([^\}]|}(?!\}))+\}\}"
154    # strip off {{ & }}
155    t.value = t.value[2:-2]
156    t.lineno += t.value.count('\n')
157    return t
158
159def t_CPPDIRECTIVE(t):
160    r'^\#[^\#].*\n'
161    t.lineno += t.value.count('\n')
162    return t
163
164def t_NEWFILE(t):
165    r'^\#\#newfile\s+"[\w/.-]*"'
166    fileNameStack.push((t.value[11:-1], t.lineno))
167    t.lineno = 0
168
169def t_ENDFILE(t):
170    r'^\#\#endfile'
171    (old_filename, t.lineno) = fileNameStack.pop()
172
173#
174# The functions t_NEWLINE, t_ignore, and t_error are
175# special for the lex module.
176#
177
178# Newlines
179def t_NEWLINE(t):
180    r'\n+'
181    t.lineno += t.value.count('\n')
182
183# Comments
184def t_comment(t):
185    r'//.*'
186
187# Completely ignored characters
188t_ignore           = ' \t\x0c'
189
190# Error handler
191def t_error(t):
192    error(t.lineno, "illegal character '%s'" % t.value[0])
193    t.skip(1)
194
195# Build the lexer
196lex.lex()
197
198#####################################################################
199#
200#                                Parser
201#
202# Every function whose name starts with 'p_' defines a grammar rule.
203# The rule is encoded in the function's doc string, while the
204# function body provides the action taken when the rule is matched.
205# The argument to each function is a list of the values of the
206# rule's symbols: t[0] for the LHS, and t[1..n] for the symbols
207# on the RHS.  For tokens, the value is copied from the t.value
208# attribute provided by the lexer.  For non-terminals, the value
209# is assigned by the producing rule; i.e., the job of the grammar
210# rule function is to set the value for the non-terminal on the LHS
211# (by assigning to t[0]).
212#####################################################################
213
214# The LHS of the first grammar rule is used as the start symbol
215# (in this case, 'specification').  Note that this rule enforces
216# that there will be exactly one namespace declaration, with 0 or more
217# global defs/decls before and after it.  The defs & decls before
218# the namespace decl will be outside the namespace; those after
219# will be inside.  The decoder function is always inside the namespace.
220def p_specification(t):
221    'specification : opt_defs_and_outputs name_decl opt_defs_and_outputs decode_block'
222    global_code = t[1]
223    isa_name = t[2]
224    namespace = isa_name + "Inst"
225    # wrap the decode block as a function definition
226    t[4].wrap_decode_block('''
227StaticInstPtr
228%(isa_name)s::decodeInst(%(isa_name)s::ExtMachInst machInst)
229{
230    using namespace %(namespace)s;
231''' % vars(), '}')
232    # both the latter output blocks and the decode block are in the namespace
233    namespace_code = t[3] + t[4]
234    # pass it all back to the caller of yacc.parse()
235    t[0] = (isa_name, namespace, global_code, namespace_code)
236
237# ISA name declaration looks like "namespace <foo>;"
238def p_name_decl(t):
239    'name_decl : NAMESPACE ID SEMI'
240    t[0] = t[2]
241
242# 'opt_defs_and_outputs' is a possibly empty sequence of
243# def and/or output statements.
244def p_opt_defs_and_outputs_0(t):
245    'opt_defs_and_outputs : empty'
246    t[0] = GenCode()
247
248def p_opt_defs_and_outputs_1(t):
249    'opt_defs_and_outputs : defs_and_outputs'
250    t[0] = t[1]
251
252def p_defs_and_outputs_0(t):
253    'defs_and_outputs : def_or_output'
254    t[0] = t[1]
255
256def p_defs_and_outputs_1(t):
257    'defs_and_outputs : defs_and_outputs def_or_output'
258    t[0] = t[1] + t[2]
259
260# The list of possible definition/output statements.
261def p_def_or_output(t):
262    '''def_or_output : def_format
263                     | def_bitfield
264                     | def_template
265                     | def_operand_types
266                     | def_operands
267                     | output_header
268                     | output_decoder
269                     | output_exec
270                     | global_let'''
271    t[0] = t[1]
272
273# Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied
274# directly to the appropriate output section.
275
276
277# Protect any non-dict-substitution '%'s in a format string
278# (i.e. those not followed by '(')
279def protect_non_subst_percents(s):
280    return re.sub(r'%(?!\()', '%%', s)
281
282# Massage output block by substituting in template definitions and bit
283# operators.  We handle '%'s embedded in the string that don't
284# indicate template substitutions (or CPU-specific symbols, which get
285# handled in GenCode) by doubling them first so that the format
286# operation will reduce them back to single '%'s.
287def process_output(s):
288    s = protect_non_subst_percents(s)
289    # protects cpu-specific symbols too
290    s = protect_cpu_symbols(s)
291    return substBitOps(s % templateMap)
292
293def p_output_header(t):
294    'output_header : OUTPUT HEADER CODELIT SEMI'
295    t[0] = GenCode(header_output = process_output(t[3]))
296
297def p_output_decoder(t):
298    'output_decoder : OUTPUT DECODER CODELIT SEMI'
299    t[0] = GenCode(decoder_output = process_output(t[3]))
300
301def p_output_exec(t):
302    'output_exec : OUTPUT EXEC CODELIT SEMI'
303    t[0] = GenCode(exec_output = process_output(t[3]))
304
305# global let blocks 'let {{...}}' (Python code blocks) are executed
306# directly when seen.  Note that these execute in a special variable
307# context 'exportContext' to prevent the code from polluting this
308# script's namespace.
309def p_global_let(t):
310    'global_let : LET CODELIT SEMI'
311    updateExportContext()
312    try:
313        exec fixPythonIndentation(t[2]) in exportContext
314    except Exception, exc:
315        error(t.lineno(1),
316              'error: %s in global let block "%s".' % (exc, t[2]))
317    t[0] = GenCode() # contributes nothing to the output C++ file
318
319# Define the mapping from operand type extensions to C++ types and bit
320# widths (stored in operandTypeMap).
321def p_def_operand_types(t):
322    'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI'
323    try:
324        userDict = eval('{' + t[3] + '}')
325    except Exception, exc:
326        error(t.lineno(1),
327              'error: %s in def operand_types block "%s".' % (exc, t[3]))
328    buildOperandTypeMap(userDict, t.lineno(1))
329    t[0] = GenCode() # contributes nothing to the output C++ file
330
331# Define the mapping from operand names to operand classes and other
332# traits.  Stored in operandNameMap.
333def p_def_operands(t):
334    'def_operands : DEF OPERANDS CODELIT SEMI'
335    if not globals().has_key('operandTypeMap'):
336        error(t.lineno(1),
337              'error: operand types must be defined before operands')
338    try:
339        userDict = eval('{' + t[3] + '}')
340    except Exception, exc:
341        error(t.lineno(1),
342              'error: %s in def operands block "%s".' % (exc, t[3]))
343    buildOperandNameMap(userDict, t.lineno(1))
344    t[0] = GenCode() # contributes nothing to the output C++ file
345
346# A bitfield definition looks like:
347# 'def [signed] bitfield <ID> [<first>:<last>]'
348# This generates a preprocessor macro in the output file.
349def p_def_bitfield_0(t):
350    'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI'
351    expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8])
352    if (t[2] == 'signed'):
353        expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr)
354    hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr)
355    t[0] = GenCode(header_output = hash_define)
356
357# alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]'
358def p_def_bitfield_1(t):
359    'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI'
360    expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6])
361    if (t[2] == 'signed'):
362        expr = 'sext<%d>(%s)' % (1, expr)
363    hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr)
364    t[0] = GenCode(header_output = hash_define)
365
366def p_opt_signed_0(t):
367    'opt_signed : SIGNED'
368    t[0] = t[1]
369
370def p_opt_signed_1(t):
371    'opt_signed : empty'
372    t[0] = ''
373
374# Global map variable to hold templates
375templateMap = {}
376
377def p_def_template(t):
378    'def_template : DEF TEMPLATE ID CODELIT SEMI'
379    templateMap[t[3]] = Template(t[4])
380    t[0] = GenCode()
381
382# An instruction format definition looks like
383# "def format <fmt>(<params>) {{...}};"
384def p_def_format(t):
385    'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI'
386    (id, params, code) = (t[3], t[5], t[7])
387    defFormat(id, params, code, t.lineno(1))
388    t[0] = GenCode()
389
390# The formal parameter list for an instruction format is a possibly
391# empty list of comma-separated parameters.  Positional (standard,
392# non-keyword) parameters must come first, followed by keyword
393# parameters, followed by a '*foo' parameter that gets excess
394# positional arguments (as in Python).  Each of these three parameter
395# categories is optional.
396#
397# Note that we do not support the '**foo' parameter for collecting
398# otherwise undefined keyword args.  Otherwise the parameter list is
399# (I believe) identical to what is supported in Python.
400#
401# The param list generates a tuple, where the first element is a list of
402# the positional params and the second element is a dict containing the
403# keyword params.
404def p_param_list_0(t):
405    'param_list : positional_param_list COMMA nonpositional_param_list'
406    t[0] = t[1] + t[3]
407
408def p_param_list_1(t):
409    '''param_list : positional_param_list
410                  | nonpositional_param_list'''
411    t[0] = t[1]
412
413def p_positional_param_list_0(t):
414    'positional_param_list : empty'
415    t[0] = []
416
417def p_positional_param_list_1(t):
418    'positional_param_list : ID'
419    t[0] = [t[1]]
420
421def p_positional_param_list_2(t):
422    'positional_param_list : positional_param_list COMMA ID'
423    t[0] = t[1] + [t[3]]
424
425def p_nonpositional_param_list_0(t):
426    'nonpositional_param_list : keyword_param_list COMMA excess_args_param'
427    t[0] = t[1] + t[3]
428
429def p_nonpositional_param_list_1(t):
430    '''nonpositional_param_list : keyword_param_list
431                                | excess_args_param'''
432    t[0] = t[1]
433
434def p_keyword_param_list_0(t):
435    'keyword_param_list : keyword_param'
436    t[0] = [t[1]]
437
438def p_keyword_param_list_1(t):
439    'keyword_param_list : keyword_param_list COMMA keyword_param'
440    t[0] = t[1] + [t[3]]
441
442def p_keyword_param(t):
443    'keyword_param : ID EQUALS expr'
444    t[0] = t[1] + ' = ' + t[3].__repr__()
445
446def p_excess_args_param(t):
447    'excess_args_param : ASTERISK ID'
448    # Just concatenate them: '*ID'.  Wrap in list to be consistent
449    # with positional_param_list and keyword_param_list.
450    t[0] = [t[1] + t[2]]
451
452# End of format definition-related rules.
453##############
454
455#
456# A decode block looks like:
457#	decode <field1> [, <field2>]* [default <inst>] { ... }
458#
459def p_decode_block(t):
460    'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE'
461    default_defaults = defaultStack.pop()
462    codeObj = t[5]
463    # use the "default defaults" only if there was no explicit
464    # default statement in decode_stmt_list
465    if not codeObj.has_decode_default:
466        codeObj += default_defaults
467    codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n')
468    t[0] = codeObj
469
470# The opt_default statement serves only to push the "default defaults"
471# onto defaultStack.  This value will be used by nested decode blocks,
472# and used and popped off when the current decode_block is processed
473# (in p_decode_block() above).
474def p_opt_default_0(t):
475    'opt_default : empty'
476    # no default specified: reuse the one currently at the top of the stack
477    defaultStack.push(defaultStack.top())
478    # no meaningful value returned
479    t[0] = None
480
481def p_opt_default_1(t):
482    'opt_default : DEFAULT inst'
483    # push the new default
484    codeObj = t[2]
485    codeObj.wrap_decode_block('\ndefault:\n', 'break;\n')
486    defaultStack.push(codeObj)
487    # no meaningful value returned
488    t[0] = None
489
490def p_decode_stmt_list_0(t):
491    'decode_stmt_list : decode_stmt'
492    t[0] = t[1]
493
494def p_decode_stmt_list_1(t):
495    'decode_stmt_list : decode_stmt decode_stmt_list'
496    if (t[1].has_decode_default and t[2].has_decode_default):
497        error(t.lineno(1), 'Two default cases in decode block')
498    t[0] = t[1] + t[2]
499
500#
501# Decode statement rules
502#
503# There are four types of statements allowed in a decode block:
504# 1. Format blocks 'format <foo> { ... }'
505# 2. Nested decode blocks
506# 3. Instruction definitions.
507# 4. C preprocessor directives.
508
509
510# Preprocessor directives found in a decode statement list are passed
511# through to the output, replicated to all of the output code
512# streams.  This works well for ifdefs, so we can ifdef out both the
513# declarations and the decode cases generated by an instruction
514# definition.  Handling them as part of the grammar makes it easy to
515# keep them in the right place with respect to the code generated by
516# the other statements.
517def p_decode_stmt_cpp(t):
518    'decode_stmt : CPPDIRECTIVE'
519    t[0] = GenCode(t[1], t[1], t[1], t[1])
520
521# A format block 'format <foo> { ... }' sets the default instruction
522# format used to handle instruction definitions inside the block.
523# This format can be overridden by using an explicit format on the
524# instruction definition or with a nested format block.
525def p_decode_stmt_format(t):
526    'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE'
527    # The format will be pushed on the stack when 'push_format_id' is
528    # processed (see below).  Once the parser has recognized the full
529    # production (though the right brace), we're done with the format,
530    # so now we can pop it.
531    formatStack.pop()
532    t[0] = t[4]
533
534# This rule exists so we can set the current format (& push the stack)
535# when we recognize the format name part of the format block.
536def p_push_format_id(t):
537    'push_format_id : ID'
538    try:
539        formatStack.push(formatMap[t[1]])
540        t[0] = ('', '// format %s' % t[1])
541    except KeyError:
542        error(t.lineno(1), 'instruction format "%s" not defined.' % t[1])
543
544# Nested decode block: if the value of the current field matches the
545# specified constant, do a nested decode on some other field.
546def p_decode_stmt_decode(t):
547    'decode_stmt : case_label COLON decode_block'
548    label = t[1]
549    codeObj = t[3]
550    # just wrap the decoding code from the block as a case in the
551    # outer switch statement.
552    codeObj.wrap_decode_block('\n%s:\n' % label)
553    codeObj.has_decode_default = (label == 'default')
554    t[0] = codeObj
555
556# Instruction definition (finally!).
557def p_decode_stmt_inst(t):
558    'decode_stmt : case_label COLON inst SEMI'
559    label = t[1]
560    codeObj = t[3]
561    codeObj.wrap_decode_block('\n%s:' % label, 'break;\n')
562    codeObj.has_decode_default = (label == 'default')
563    t[0] = codeObj
564
565# The case label is either a list of one or more constants or 'default'
566def p_case_label_0(t):
567    'case_label : intlit_list'
568    t[0] = ': '.join(map(lambda a: 'case %#x' % a, t[1]))
569
570def p_case_label_1(t):
571    'case_label : DEFAULT'
572    t[0] = 'default'
573
574#
575# The constant list for a decode case label must be non-empty, but may have
576# one or more comma-separated integer literals in it.
577#
578def p_intlit_list_0(t):
579    'intlit_list : INTLIT'
580    t[0] = [t[1]]
581
582def p_intlit_list_1(t):
583    'intlit_list : intlit_list COMMA INTLIT'
584    t[0] = t[1]
585    t[0].append(t[3])
586
587# Define an instruction using the current instruction format (specified
588# by an enclosing format block).
589# "<mnemonic>(<args>)"
590def p_inst_0(t):
591    'inst : ID LPAREN arg_list RPAREN'
592    # Pass the ID and arg list to the current format class to deal with.
593    currentFormat = formatStack.top()
594    codeObj = currentFormat.defineInst(t[1], t[3], t.lineno(1))
595    args = ','.join(map(str, t[3]))
596    args = re.sub('(?m)^', '//', args)
597    args = re.sub('^//', '', args)
598    comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args)
599    codeObj.prepend_all(comment)
600    t[0] = codeObj
601
602# Define an instruction using an explicitly specified format:
603# "<fmt>::<mnemonic>(<args>)"
604def p_inst_1(t):
605    'inst : ID DBLCOLON ID LPAREN arg_list RPAREN'
606    try:
607        format = formatMap[t[1]]
608    except KeyError:
609        error(t.lineno(1), 'instruction format "%s" not defined.' % t[1])
610    codeObj = format.defineInst(t[3], t[5], t.lineno(1))
611    comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5])
612    codeObj.prepend_all(comment)
613    t[0] = codeObj
614
615# The arg list generates a tuple, where the first element is a list of
616# the positional args and the second element is a dict containing the
617# keyword args.
618def p_arg_list_0(t):
619    'arg_list : positional_arg_list COMMA keyword_arg_list'
620    t[0] = ( t[1], t[3] )
621
622def p_arg_list_1(t):
623    'arg_list : positional_arg_list'
624    t[0] = ( t[1], {} )
625
626def p_arg_list_2(t):
627    'arg_list : keyword_arg_list'
628    t[0] = ( [], t[1] )
629
630def p_positional_arg_list_0(t):
631    'positional_arg_list : empty'
632    t[0] = []
633
634def p_positional_arg_list_1(t):
635    'positional_arg_list : expr'
636    t[0] = [t[1]]
637
638def p_positional_arg_list_2(t):
639    'positional_arg_list : positional_arg_list COMMA expr'
640    t[0] = t[1] + [t[3]]
641
642def p_keyword_arg_list_0(t):
643    'keyword_arg_list : keyword_arg'
644    t[0] = t[1]
645
646def p_keyword_arg_list_1(t):
647    'keyword_arg_list : keyword_arg_list COMMA keyword_arg'
648    t[0] = t[1]
649    t[0].update(t[3])
650
651def p_keyword_arg(t):
652    'keyword_arg : ID EQUALS expr'
653    t[0] = { t[1] : t[3] }
654
655#
656# Basic expressions.  These constitute the argument values of
657# "function calls" (i.e. instruction definitions in the decode block)
658# and default values for formal parameters of format functions.
659#
660# Right now, these are either strings, integers, or (recursively)
661# lists of exprs (using Python square-bracket list syntax).  Note that
662# bare identifiers are trated as string constants here (since there
663# isn't really a variable namespace to refer to).
664#
665def p_expr_0(t):
666    '''expr : ID
667            | INTLIT
668            | STRLIT
669            | CODELIT'''
670    t[0] = t[1]
671
672def p_expr_1(t):
673    '''expr : LBRACKET list_expr RBRACKET'''
674    t[0] = t[2]
675
676def p_list_expr_0(t):
677    'list_expr : expr'
678    t[0] = [t[1]]
679
680def p_list_expr_1(t):
681    'list_expr : list_expr COMMA expr'
682    t[0] = t[1] + [t[3]]
683
684def p_list_expr_2(t):
685    'list_expr : empty'
686    t[0] = []
687
688#
689# Empty production... use in other rules for readability.
690#
691def p_empty(t):
692    'empty :'
693    pass
694
695# Parse error handler.  Note that the argument here is the offending
696# *token*, not a grammar symbol (hence the need to use t.value)
697def p_error(t):
698    if t:
699        error(t.lineno, "syntax error at '%s'" % t.value)
700    else:
701        error(0, "unknown syntax error", True)
702
703# END OF GRAMMAR RULES
704#
705# Now build the parser.
706yacc.yacc()
707
708
709#####################################################################
710#
711#                           Support Classes
712#
713#####################################################################
714
715# Expand template with CPU-specific references into a dictionary with
716# an entry for each CPU model name.  The entry key is the model name
717# and the corresponding value is the template with the CPU-specific
718# refs substituted for that model.
719def expand_cpu_symbols_to_dict(template):
720    # Protect '%'s that don't go with CPU-specific terms
721    t = re.sub(r'%(?!\(CPU_)', '%%', template)
722    result = {}
723    for cpu in cpu_models:
724        result[cpu.name] = t % cpu.strings
725    return result
726
727# *If* the template has CPU-specific references, return a single
728# string containing a copy of the template for each CPU model with the
729# corresponding values substituted in.  If the template has no
730# CPU-specific references, it is returned unmodified.
731def expand_cpu_symbols_to_string(template):
732    if template.find('%(CPU_') != -1:
733        return reduce(lambda x,y: x+y,
734                      expand_cpu_symbols_to_dict(template).values())
735    else:
736        return template
737
738# Protect CPU-specific references by doubling the corresponding '%'s
739# (in preparation for substituting a different set of references into
740# the template).
741def protect_cpu_symbols(template):
742    return re.sub(r'%(?=\(CPU_)', '%%', template)
743
744###############
745# GenCode class
746#
747# The GenCode class encapsulates generated code destined for various
748# output files.  The header_output and decoder_output attributes are
749# strings containing code destined for decoder.hh and decoder.cc
750# respectively.  The decode_block attribute contains code to be
751# incorporated in the decode function itself (that will also end up in
752# decoder.cc).  The exec_output attribute is a dictionary with a key
753# for each CPU model name; the value associated with a particular key
754# is the string of code for that CPU model's exec.cc file.  The
755# has_decode_default attribute is used in the decode block to allow
756# explicit default clauses to override default default clauses.
757
758class GenCode:
759    # Constructor.  At this point we substitute out all CPU-specific
760    # symbols.  For the exec output, these go into the per-model
761    # dictionary.  For all other output types they get collapsed into
762    # a single string.
763    def __init__(self,
764                 header_output = '', decoder_output = '', exec_output = '',
765                 decode_block = '', has_decode_default = False):
766        self.header_output = expand_cpu_symbols_to_string(header_output)
767        self.decoder_output = expand_cpu_symbols_to_string(decoder_output)
768        if isinstance(exec_output, dict):
769            self.exec_output = exec_output
770        elif isinstance(exec_output, str):
771            # If the exec_output arg is a single string, we replicate
772            # it for each of the CPU models, substituting and
773            # %(CPU_foo)s params appropriately.
774            self.exec_output = expand_cpu_symbols_to_dict(exec_output)
775        self.decode_block = expand_cpu_symbols_to_string(decode_block)
776        self.has_decode_default = has_decode_default
777
778    # Override '+' operator: generate a new GenCode object that
779    # concatenates all the individual strings in the operands.
780    def __add__(self, other):
781        exec_output = {}
782        for cpu in cpu_models:
783            n = cpu.name
784            exec_output[n] = self.exec_output[n] + other.exec_output[n]
785        return GenCode(self.header_output + other.header_output,
786                       self.decoder_output + other.decoder_output,
787                       exec_output,
788                       self.decode_block + other.decode_block,
789                       self.has_decode_default or other.has_decode_default)
790
791    # Prepend a string (typically a comment) to all the strings.
792    def prepend_all(self, pre):
793        self.header_output = pre + self.header_output
794        self.decoder_output  = pre + self.decoder_output
795        self.decode_block = pre + self.decode_block
796        for cpu in cpu_models:
797            self.exec_output[cpu.name] = pre + self.exec_output[cpu.name]
798
799    # Wrap the decode block in a pair of strings (e.g., 'case foo:'
800    # and 'break;').  Used to build the big nested switch statement.
801    def wrap_decode_block(self, pre, post = ''):
802        self.decode_block = pre + indent(self.decode_block) + post
803
804################
805# Format object.
806#
807# A format object encapsulates an instruction format.  It must provide
808# a defineInst() method that generates the code for an instruction
809# definition.
810
811exportContextSymbols = ('InstObjParams', 'makeList', 're', 'string')
812
813exportContext = {}
814
815def updateExportContext():
816    exportContext.update(exportDict(*exportContextSymbols))
817    exportContext.update(templateMap)
818
819def exportDict(*symNames):
820    return dict([(s, eval(s)) for s in symNames])
821
822
823class Format:
824    def __init__(self, id, params, code):
825        # constructor: just save away arguments
826        self.id = id
827        self.params = params
828        label = 'def format ' + id
829        self.user_code = compile(fixPythonIndentation(code), label, 'exec')
830        param_list = string.join(params, ", ")
831        f = '''def defInst(_code, _context, %s):
832                my_locals = vars().copy()
833                exec _code in _context, my_locals
834                return my_locals\n''' % param_list
835        c = compile(f, label + ' wrapper', 'exec')
836        exec c
837        self.func = defInst
838
839    def defineInst(self, name, args, lineno):
840        context = {}
841        updateExportContext()
842        context.update(exportContext)
843        context.update({ 'name': name, 'Name': string.capitalize(name) })
844        try:
845            vars = self.func(self.user_code, context, *args[0], **args[1])
846        except Exception, exc:
847            error(lineno, 'error defining "%s": %s.' % (name, exc))
848        for k in vars.keys():
849            if k not in ('header_output', 'decoder_output',
850                         'exec_output', 'decode_block'):
851                del vars[k]
852        return GenCode(**vars)
853
854# Special null format to catch an implicit-format instruction
855# definition outside of any format block.
856class NoFormat:
857    def __init__(self):
858        self.defaultInst = ''
859
860    def defineInst(self, name, args, lineno):
861        error(lineno,
862              'instruction definition "%s" with no active format!' % name)
863
864# This dictionary maps format name strings to Format objects.
865formatMap = {}
866
867# Define a new format
868def defFormat(id, params, code, lineno):
869    # make sure we haven't already defined this one
870    if formatMap.get(id, None) != None:
871        error(lineno, 'format %s redefined.' % id)
872    # create new object and store in global map
873    formatMap[id] = Format(id, params, code)
874
875
876##############
877# Stack: a simple stack object.  Used for both formats (formatStack)
878# and default cases (defaultStack).  Simply wraps a list to give more
879# stack-like syntax and enable initialization with an argument list
880# (as opposed to an argument that's a list).
881
882class Stack(list):
883    def __init__(self, *items):
884        list.__init__(self, items)
885
886    def push(self, item):
887        self.append(item);
888
889    def top(self):
890        return self[-1]
891
892# The global format stack.
893formatStack = Stack(NoFormat())
894
895# The global default case stack.
896defaultStack = Stack( None )
897
898# Global stack that tracks current file and line number.
899# Each element is a tuple (filename, lineno) that records the
900# *current* filename and the line number in the *previous* file where
901# it was included.
902fileNameStack = Stack()
903
904###################
905# Utility functions
906
907#
908# Indent every line in string 's' by two spaces
909# (except preprocessor directives).
910# Used to make nested code blocks look pretty.
911#
912def indent(s):
913    return re.sub(r'(?m)^(?!#)', '  ', s)
914
915#
916# Munge a somewhat arbitrarily formatted piece of Python code
917# (e.g. from a format 'let' block) into something whose indentation
918# will get by the Python parser.
919#
920# The two keys here are that Python will give a syntax error if
921# there's any whitespace at the beginning of the first line, and that
922# all lines at the same lexical nesting level must have identical
923# indentation.  Unfortunately the way code literals work, an entire
924# let block tends to have some initial indentation.  Rather than
925# trying to figure out what that is and strip it off, we prepend 'if
926# 1:' to make the let code the nested block inside the if (and have
927# the parser automatically deal with the indentation for us).
928#
929# We don't want to do this if (1) the code block is empty or (2) the
930# first line of the block doesn't have any whitespace at the front.
931
932def fixPythonIndentation(s):
933    # get rid of blank lines first
934    s = re.sub(r'(?m)^\s*\n', '', s);
935    if (s != '' and re.match(r'[ \t]', s[0])):
936        s = 'if 1:\n' + s
937    return s
938
939# Error handler.  Just call exit.  Output formatted to work under
940# Emacs compile-mode.  Optional 'print_traceback' arg, if set to True,
941# prints a Python stack backtrace too (can be handy when trying to
942# debug the parser itself).
943def error(lineno, string, print_traceback = False):
944    spaces = ""
945    for (filename, line) in fileNameStack[0:-1]:
946        print spaces + "In file included from " + filename + ":"
947        spaces += "  "
948    # Print a Python stack backtrace if requested.
949    if (print_traceback):
950        traceback.print_exc()
951    if lineno != 0:
952        line_str = "%d:" % lineno
953    else:
954        line_str = ""
955    sys.exit(spaces + "%s:%s %s" % (fileNameStack[-1][0], line_str, string))
956
957
958#####################################################################
959#
960#                      Bitfield Operator Support
961#
962#####################################################################
963
964bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>')
965
966bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>')
967bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>')
968
969def substBitOps(code):
970    # first convert single-bit selectors to two-index form
971    # i.e., <n> --> <n:n>
972    code = bitOp1ArgRE.sub(r'<\1:\1>', code)
973    # simple case: selector applied to ID (name)
974    # i.e., foo<a:b> --> bits(foo, a, b)
975    code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code)
976    # if selector is applied to expression (ending in ')'),
977    # we need to search backward for matching '('
978    match = bitOpExprRE.search(code)
979    while match:
980        exprEnd = match.start()
981        here = exprEnd - 1
982        nestLevel = 1
983        while nestLevel > 0:
984            if code[here] == '(':
985                nestLevel -= 1
986            elif code[here] == ')':
987                nestLevel += 1
988            here -= 1
989            if here < 0:
990                sys.exit("Didn't find '('!")
991        exprStart = here+1
992        newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1],
993                                         match.group(1), match.group(2))
994        code = code[:exprStart] + newExpr + code[match.end():]
995        match = bitOpExprRE.search(code)
996    return code
997
998
999####################
1000# Template objects.
1001#
1002# Template objects are format strings that allow substitution from
1003# the attribute spaces of other objects (e.g. InstObjParams instances).
1004
1005labelRE = re.compile(r'[^%]%\(([^\)]+)\)[sd]')
1006
1007class Template:
1008    def __init__(self, t):
1009        self.template = t
1010
1011    def subst(self, d):
1012        myDict = None
1013
1014        # Protect non-Python-dict substitutions (e.g. if there's a printf
1015        # in the templated C++ code)
1016        template = protect_non_subst_percents(self.template)
1017        # CPU-model-specific substitutions are handled later (in GenCode).
1018        template = protect_cpu_symbols(template)
1019
1020        # Build a dict ('myDict') to use for the template substitution.
1021        # Start with the template namespace.  Make a copy since we're
1022        # going to modify it.
1023        myDict = templateMap.copy()
1024
1025        if isinstance(d, InstObjParams):
1026            # If we're dealing with an InstObjParams object, we need
1027            # to be a little more sophisticated.  The instruction-wide
1028            # parameters are already formed, but the parameters which
1029            # are only function wide still need to be generated.
1030            compositeCode = ''
1031
1032            myDict.update(d.__dict__)
1033            # The "operands" and "snippets" attributes of the InstObjParams
1034            # objects are for internal use and not substitution.
1035            del myDict['operands']
1036            del myDict['snippets']
1037
1038            snippetLabels = [l for l in labelRE.findall(template)
1039                             if d.snippets.has_key(l)]
1040
1041            snippets = dict([(s, mungeSnippet(d.snippets[s]))
1042                             for s in snippetLabels])
1043
1044            myDict.update(snippets)
1045
1046            compositeCode = ' '.join(map(str, snippets.values()))
1047
1048            # Add in template itself in case it references any
1049            # operands explicitly (like Mem)
1050            compositeCode += ' ' + template
1051
1052            operands = SubOperandList(compositeCode, d.operands)
1053
1054            myDict['op_decl'] = operands.concatAttrStrings('op_decl')
1055
1056            is_src = lambda op: op.is_src
1057            is_dest = lambda op: op.is_dest
1058
1059            myDict['op_src_decl'] = \
1060                      operands.concatSomeAttrStrings(is_src, 'op_src_decl')
1061            myDict['op_dest_decl'] = \
1062                      operands.concatSomeAttrStrings(is_dest, 'op_dest_decl')
1063
1064            myDict['op_rd'] = operands.concatAttrStrings('op_rd')
1065            myDict['op_wb'] = operands.concatAttrStrings('op_wb')
1066
1067            if d.operands.memOperand:
1068                myDict['mem_acc_size'] = d.operands.memOperand.mem_acc_size
1069                myDict['mem_acc_type'] = d.operands.memOperand.mem_acc_type
1070
1071        elif isinstance(d, dict):
1072            # if the argument is a dictionary, we just use it.
1073            myDict.update(d)
1074        elif hasattr(d, '__dict__'):
1075            # if the argument is an object, we use its attribute map.
1076            myDict.update(d.__dict__)
1077        else:
1078            raise TypeError, "Template.subst() arg must be or have dictionary"
1079        return template % myDict
1080
1081    # Convert to string.  This handles the case when a template with a
1082    # CPU-specific term gets interpolated into another template or into
1083    # an output block.
1084    def __str__(self):
1085        return expand_cpu_symbols_to_string(self.template)
1086
1087#####################################################################
1088#
1089#                             Code Parser
1090#
1091# The remaining code is the support for automatically extracting
1092# instruction characteristics from pseudocode.
1093#
1094#####################################################################
1095
1096# Force the argument to be a list.  Useful for flags, where a caller
1097# can specify a singleton flag or a list of flags.  Also usful for
1098# converting tuples to lists so they can be modified.
1099def makeList(arg):
1100    if isinstance(arg, list):
1101        return arg
1102    elif isinstance(arg, tuple):
1103        return list(arg)
1104    elif not arg:
1105        return []
1106    else:
1107        return [ arg ]
1108
1109# Generate operandTypeMap from the user's 'def operand_types'
1110# statement.
1111def buildOperandTypeMap(userDict, lineno):
1112    global operandTypeMap
1113    operandTypeMap = {}
1114    for (ext, (desc, size)) in userDict.iteritems():
1115        if desc == 'signed int':
1116            ctype = 'int%d_t' % size
1117            is_signed = 1
1118        elif desc == 'unsigned int':
1119            ctype = 'uint%d_t' % size
1120            is_signed = 0
1121        elif desc == 'float':
1122            is_signed = 1	# shouldn't really matter
1123            if size == 32:
1124                ctype = 'float'
1125            elif size == 64:
1126                ctype = 'double'
1127        if ctype == '':
1128            error(lineno, 'Unrecognized type description "%s" in userDict')
1129        operandTypeMap[ext] = (size, ctype, is_signed)
1130
1131#
1132#
1133#
1134# Base class for operand descriptors.  An instance of this class (or
1135# actually a class derived from this one) represents a specific
1136# operand for a code block (e.g, "Rc.sq" as a dest). Intermediate
1137# derived classes encapsulates the traits of a particular operand type
1138# (e.g., "32-bit integer register").
1139#
1140class Operand(object):
1141    def __init__(self, full_name, ext, is_src, is_dest):
1142        self.full_name = full_name
1143        self.ext = ext
1144        self.is_src = is_src
1145        self.is_dest = is_dest
1146        # The 'effective extension' (eff_ext) is either the actual
1147        # extension, if one was explicitly provided, or the default.
1148        if ext:
1149            self.eff_ext = ext
1150        else:
1151            self.eff_ext = self.dflt_ext
1152
1153        (self.size, self.ctype, self.is_signed) = operandTypeMap[self.eff_ext]
1154
1155        # note that mem_acc_size is undefined for non-mem operands...
1156        # template must be careful not to use it if it doesn't apply.
1157        if self.isMem():
1158            self.mem_acc_size = self.makeAccSize()
1159            self.mem_acc_type = self.ctype
1160
1161    # Finalize additional fields (primarily code fields).  This step
1162    # is done separately since some of these fields may depend on the
1163    # register index enumeration that hasn't been performed yet at the
1164    # time of __init__().
1165    def finalize(self):
1166        self.flags = self.getFlags()
1167        self.constructor = self.makeConstructor()
1168        self.op_decl = self.makeDecl()
1169
1170        if self.is_src:
1171            self.op_rd = self.makeRead()
1172            self.op_src_decl = self.makeDecl()
1173        else:
1174            self.op_rd = ''
1175            self.op_src_decl = ''
1176
1177        if self.is_dest:
1178            self.op_wb = self.makeWrite()
1179            self.op_dest_decl = self.makeDecl()
1180        else:
1181            self.op_wb = ''
1182            self.op_dest_decl = ''
1183
1184    def isMem(self):
1185        return 0
1186
1187    def isReg(self):
1188        return 0
1189
1190    def isFloatReg(self):
1191        return 0
1192
1193    def isIntReg(self):
1194        return 0
1195
1196    def isControlReg(self):
1197        return 0
1198
1199    def getFlags(self):
1200        # note the empty slice '[:]' gives us a copy of self.flags[0]
1201        # instead of a reference to it
1202        my_flags = self.flags[0][:]
1203        if self.is_src:
1204            my_flags += self.flags[1]
1205        if self.is_dest:
1206            my_flags += self.flags[2]
1207        return my_flags
1208
1209    def makeDecl(self):
1210        # Note that initializations in the declarations are solely
1211        # to avoid 'uninitialized variable' errors from the compiler.
1212        return self.ctype + ' ' + self.base_name + ' = 0;\n';
1213
1214class IntRegOperand(Operand):
1215    def isReg(self):
1216        return 1
1217
1218    def isIntReg(self):
1219        return 1
1220
1221    def makeConstructor(self):
1222        c = ''
1223        if self.is_src:
1224            c += '\n\t_srcRegIdx[%d] = %s;' % \
1225                 (self.src_reg_idx, self.reg_spec)
1226        if self.is_dest:
1227            c += '\n\t_destRegIdx[%d] = %s;' % \
1228                 (self.dest_reg_idx, self.reg_spec)
1229        return c
1230
1231    def makeRead(self):
1232        if (self.ctype == 'float' or self.ctype == 'double'):
1233            error(0, 'Attempt to read integer register as FP')
1234        if (self.size == self.dflt_size):
1235            return '%s = xc->readIntRegOperand(this, %d);\n' % \
1236                   (self.base_name, self.src_reg_idx)
1237        elif (self.size > self.dflt_size):
1238            int_reg_val = 'xc->readIntRegOperand(this, %d)' % \
1239                          (self.src_reg_idx)
1240            if (self.is_signed):
1241                int_reg_val = 'sext<%d>(%s)' % (self.dflt_size, int_reg_val)
1242            return '%s = %s;\n' % (self.base_name, int_reg_val)
1243        else:
1244            return '%s = bits(xc->readIntRegOperand(this, %d), %d, 0);\n' % \
1245                   (self.base_name, self.src_reg_idx, self.size-1)
1246
1247    def makeWrite(self):
1248        if (self.ctype == 'float' or self.ctype == 'double'):
1249            error(0, 'Attempt to write integer register as FP')
1250        if (self.size != self.dflt_size and self.is_signed):
1251            final_val = 'sext<%d>(%s)' % (self.size, self.base_name)
1252        else:
1253            final_val = self.base_name
1254        wb = '''
1255        {
1256            %s final_val = %s;
1257            xc->setIntRegOperand(this, %d, final_val);\n
1258            if (traceData) { traceData->setData(final_val); }
1259        }''' % (self.dflt_ctype, final_val, self.dest_reg_idx)
1260        return wb
1261
1262class FloatRegOperand(Operand):
1263    def isReg(self):
1264        return 1
1265
1266    def isFloatReg(self):
1267        return 1
1268
1269    def makeConstructor(self):
1270        c = ''
1271        if self.is_src:
1272            c += '\n\t_srcRegIdx[%d] = %s + FP_Base_DepTag;' % \
1273                 (self.src_reg_idx, self.reg_spec)
1274        if self.is_dest:
1275            c += '\n\t_destRegIdx[%d] = %s + FP_Base_DepTag;' % \
1276                 (self.dest_reg_idx, self.reg_spec)
1277        return c
1278
1279    def makeRead(self):
1280        bit_select = 0
1281        width = 0;
1282        if (self.ctype == 'float'):
1283            func = 'readFloatRegOperand'
1284            width = 32;
1285        elif (self.ctype == 'double'):
1286            func = 'readFloatRegOperand'
1287            width = 64;
1288        else:
1289            func = 'readFloatRegOperandBits'
1290            if (self.ctype == 'uint32_t'):
1291                width = 32;
1292            elif (self.ctype == 'uint64_t'):
1293                width = 64;
1294            if (self.size != self.dflt_size):
1295                bit_select = 1
1296        if width:
1297            base = 'xc->%s(this, %d, %d)' % \
1298                   (func, self.src_reg_idx, width)
1299        else:
1300            base = 'xc->%s(this, %d)' % \
1301                   (func, self.src_reg_idx)
1302        if bit_select:
1303            return '%s = bits(%s, %d, 0);\n' % \
1304                   (self.base_name, base, self.size-1)
1305        else:
1306            return '%s = %s;\n' % (self.base_name, base)
1307
1308    def makeWrite(self):
1309        final_val = self.base_name
1310        final_ctype = self.ctype
1311        widthSpecifier = ''
1312        width = 0
1313        if (self.ctype == 'float'):
1314            width = 32
1315            func = 'setFloatRegOperand'
1316        elif (self.ctype == 'double'):
1317            width = 64
1318            func = 'setFloatRegOperand'
1319        elif (self.ctype == 'uint32_t'):
1320            func = 'setFloatRegOperandBits'
1321            width = 32
1322        elif (self.ctype == 'uint64_t'):
1323            func = 'setFloatRegOperandBits'
1324            width = 64
1325        else:
1326            func = 'setFloatRegOperandBits'
1327            final_ctype = 'uint%d_t' % self.dflt_size
1328            if (self.size != self.dflt_size and self.is_signed):
1329                final_val = 'sext<%d>(%s)' % (self.size, self.base_name)
1330        if width:
1331            widthSpecifier = ', %d' % width
1332        wb = '''
1333        {
1334            %s final_val = %s;
1335            xc->%s(this, %d, final_val%s);\n
1336            if (traceData) { traceData->setData(final_val); }
1337        }''' % (final_ctype, final_val, func, self.dest_reg_idx,
1338                widthSpecifier)
1339        return wb
1340
1341class ControlRegOperand(Operand):
1342    def isReg(self):
1343        return 1
1344
1345    def isControlReg(self):
1346        return 1
1347
1348    def makeConstructor(self):
1349        c = ''
1350        if self.is_src:
1351            c += '\n\t_srcRegIdx[%d] = %s + Ctrl_Base_DepTag;' % \
1352                 (self.src_reg_idx, self.reg_spec)
1353        if self.is_dest:
1354            c += '\n\t_destRegIdx[%d] = %s + Ctrl_Base_DepTag;' % \
1355                 (self.dest_reg_idx, self.reg_spec)
1356        return c
1357
1358    def makeRead(self):
1359        bit_select = 0
1360        if (self.ctype == 'float' or self.ctype == 'double'):
1361            error(0, 'Attempt to read control register as FP')
1362        base = 'xc->readMiscRegOperandWithEffect(this, %s)' % self.src_reg_idx
1363        if self.size == self.dflt_size:
1364            return '%s = %s;\n' % (self.base_name, base)
1365        else:
1366            return '%s = bits(%s, %d, 0);\n' % \
1367                   (self.base_name, base, self.size-1)
1368
1369    def makeWrite(self):
1370        if (self.ctype == 'float' or self.ctype == 'double'):
1371            error(0, 'Attempt to write control register as FP')
1372        wb = 'xc->setMiscRegOperandWithEffect(this, %s, %s);\n' % \
1373             (self.dest_reg_idx, self.base_name)
1374        wb += 'if (traceData) { traceData->setData(%s); }' % \
1375              self.base_name
1376        return wb
1377
1378class MemOperand(Operand):
1379    def isMem(self):
1380        return 1
1381
1382    def makeConstructor(self):
1383        return ''
1384
1385    def makeDecl(self):
1386        # Note that initializations in the declarations are solely
1387        # to avoid 'uninitialized variable' errors from the compiler.
1388        # Declare memory data variable.
1389        c = '%s %s = 0;\n' % (self.ctype, self.base_name)
1390        return c
1391
1392    def makeRead(self):
1393        return ''
1394
1395    def makeWrite(self):
1396        return ''
1397
1398    # Return the memory access size *in bits*, suitable for
1399    # forming a type via "uint%d_t".  Divide by 8 if you want bytes.
1400    def makeAccSize(self):
1401        return self.size
1402
1403
1404class NPCOperand(Operand):
1405    def makeConstructor(self):
1406        return ''
1407
1408    def makeRead(self):
1409        return '%s = xc->readNextPC();\n' % self.base_name
1410
1411    def makeWrite(self):
1412        return 'xc->setNextPC(%s);\n' % self.base_name
1413
1414class NNPCOperand(Operand):
1415    def makeConstructor(self):
1416        return ''
1417
1418    def makeRead(self):
1419        return '%s = xc->readNextNPC();\n' % self.base_name
1420
1421    def makeWrite(self):
1422        return 'xc->setNextNPC(%s);\n' % self.base_name
1423
1424def buildOperandNameMap(userDict, lineno):
1425    global operandNameMap
1426    operandNameMap = {}
1427    for (op_name, val) in userDict.iteritems():
1428        (base_cls_name, dflt_ext, reg_spec, flags, sort_pri) = val
1429        (dflt_size, dflt_ctype, dflt_is_signed) = operandTypeMap[dflt_ext]
1430        # Canonical flag structure is a triple of lists, where each list
1431        # indicates the set of flags implied by this operand always, when
1432        # used as a source, and when used as a dest, respectively.
1433        # For simplicity this can be initialized using a variety of fairly
1434        # obvious shortcuts; we convert these to canonical form here.
1435        if not flags:
1436            # no flags specified (e.g., 'None')
1437            flags = ( [], [], [] )
1438        elif isinstance(flags, str):
1439            # a single flag: assumed to be unconditional
1440            flags = ( [ flags ], [], [] )
1441        elif isinstance(flags, list):
1442            # a list of flags: also assumed to be unconditional
1443            flags = ( flags, [], [] )
1444        elif isinstance(flags, tuple):
1445            # it's a tuple: it should be a triple,
1446            # but each item could be a single string or a list
1447            (uncond_flags, src_flags, dest_flags) = flags
1448            flags = (makeList(uncond_flags),
1449                     makeList(src_flags), makeList(dest_flags))
1450        # Accumulate attributes of new operand class in tmp_dict
1451        tmp_dict = {}
1452        for attr in ('dflt_ext', 'reg_spec', 'flags', 'sort_pri',
1453                     'dflt_size', 'dflt_ctype', 'dflt_is_signed'):
1454            tmp_dict[attr] = eval(attr)
1455        tmp_dict['base_name'] = op_name
1456        # New class name will be e.g. "IntReg_Ra"
1457        cls_name = base_cls_name + '_' + op_name
1458        # Evaluate string arg to get class object.  Note that the
1459        # actual base class for "IntReg" is "IntRegOperand", i.e. we
1460        # have to append "Operand".
1461        try:
1462            base_cls = eval(base_cls_name + 'Operand')
1463        except NameError:
1464            error(lineno,
1465                  'error: unknown operand base class "%s"' % base_cls_name)
1466        # The following statement creates a new class called
1467        # <cls_name> as a subclass of <base_cls> with the attributes
1468        # in tmp_dict, just as if we evaluated a class declaration.
1469        operandNameMap[op_name] = type(cls_name, (base_cls,), tmp_dict)
1470
1471    # Define operand variables.
1472    operands = userDict.keys()
1473
1474    operandsREString = (r'''
1475    (?<![\w\.])	     # neg. lookbehind assertion: prevent partial matches
1476    ((%s)(?:\.(\w+))?)   # match: operand with optional '.' then suffix
1477    (?![\w\.])	     # neg. lookahead assertion: prevent partial matches
1478    '''
1479                        % string.join(operands, '|'))
1480
1481    global operandsRE
1482    operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE)
1483
1484    # Same as operandsREString, but extension is mandatory, and only two
1485    # groups are returned (base and ext, not full name as above).
1486    # Used for subtituting '_' for '.' to make C++ identifiers.
1487    operandsWithExtREString = (r'(?<![\w\.])(%s)\.(\w+)(?![\w\.])'
1488                               % string.join(operands, '|'))
1489
1490    global operandsWithExtRE
1491    operandsWithExtRE = re.compile(operandsWithExtREString, re.MULTILINE)
1492
1493
1494class OperandList:
1495
1496    # Find all the operands in the given code block.  Returns an operand
1497    # descriptor list (instance of class OperandList).
1498    def __init__(self, code):
1499        self.items = []
1500        self.bases = {}
1501        # delete comments so we don't match on reg specifiers inside
1502        code = commentRE.sub('', code)
1503        # search for operands
1504        next_pos = 0
1505        while 1:
1506            match = operandsRE.search(code, next_pos)
1507            if not match:
1508                # no more matches: we're done
1509                break
1510            op = match.groups()
1511            # regexp groups are operand full name, base, and extension
1512            (op_full, op_base, op_ext) = op
1513            # if the token following the operand is an assignment, this is
1514            # a destination (LHS), else it's a source (RHS)
1515            is_dest = (assignRE.match(code, match.end()) != None)
1516            is_src = not is_dest
1517            # see if we've already seen this one
1518            op_desc = self.find_base(op_base)
1519            if op_desc:
1520                if op_desc.ext != op_ext:
1521                    error(0, 'Inconsistent extensions for operand %s' % \
1522                          op_base)
1523                op_desc.is_src = op_desc.is_src or is_src
1524                op_desc.is_dest = op_desc.is_dest or is_dest
1525            else:
1526                # new operand: create new descriptor
1527                op_desc = operandNameMap[op_base](op_full, op_ext,
1528                                                  is_src, is_dest)
1529                self.append(op_desc)
1530            # start next search after end of current match
1531            next_pos = match.end()
1532        self.sort()
1533        # enumerate source & dest register operands... used in building
1534        # constructor later
1535        self.numSrcRegs = 0
1536        self.numDestRegs = 0
1537        self.numFPDestRegs = 0
1538        self.numIntDestRegs = 0
1539        self.memOperand = None
1540        for op_desc in self.items:
1541            if op_desc.isReg():
1542                if op_desc.is_src:
1543                    op_desc.src_reg_idx = self.numSrcRegs
1544                    self.numSrcRegs += 1
1545                if op_desc.is_dest:
1546                    op_desc.dest_reg_idx = self.numDestRegs
1547                    self.numDestRegs += 1
1548                    if op_desc.isFloatReg():
1549                        self.numFPDestRegs += 1
1550                    elif op_desc.isIntReg():
1551                        self.numIntDestRegs += 1
1552            elif op_desc.isMem():
1553                if self.memOperand:
1554                    error(0, "Code block has more than one memory operand.")
1555                self.memOperand = op_desc
1556        # now make a final pass to finalize op_desc fields that may depend
1557        # on the register enumeration
1558        for op_desc in self.items:
1559            op_desc.finalize()
1560
1561    def __len__(self):
1562        return len(self.items)
1563
1564    def __getitem__(self, index):
1565        return self.items[index]
1566
1567    def append(self, op_desc):
1568        self.items.append(op_desc)
1569        self.bases[op_desc.base_name] = op_desc
1570
1571    def find_base(self, base_name):
1572        # like self.bases[base_name], but returns None if not found
1573        # (rather than raising exception)
1574        return self.bases.get(base_name)
1575
1576    # internal helper function for concat[Some]Attr{Strings|Lists}
1577    def __internalConcatAttrs(self, attr_name, filter, result):
1578        for op_desc in self.items:
1579            if filter(op_desc):
1580                result += getattr(op_desc, attr_name)
1581        return result
1582
1583    # return a single string that is the concatenation of the (string)
1584    # values of the specified attribute for all operands
1585    def concatAttrStrings(self, attr_name):
1586        return self.__internalConcatAttrs(attr_name, lambda x: 1, '')
1587
1588    # like concatAttrStrings, but only include the values for the operands
1589    # for which the provided filter function returns true
1590    def concatSomeAttrStrings(self, filter, attr_name):
1591        return self.__internalConcatAttrs(attr_name, filter, '')
1592
1593    # return a single list that is the concatenation of the (list)
1594    # values of the specified attribute for all operands
1595    def concatAttrLists(self, attr_name):
1596        return self.__internalConcatAttrs(attr_name, lambda x: 1, [])
1597
1598    # like concatAttrLists, but only include the values for the operands
1599    # for which the provided filter function returns true
1600    def concatSomeAttrLists(self, filter, attr_name):
1601        return self.__internalConcatAttrs(attr_name, filter, [])
1602
1603    def sort(self):
1604        self.items.sort(lambda a, b: a.sort_pri - b.sort_pri)
1605
1606class SubOperandList(OperandList):
1607
1608    # Find all the operands in the given code block.  Returns an operand
1609    # descriptor list (instance of class OperandList).
1610    def __init__(self, code, master_list):
1611        self.items = []
1612        self.bases = {}
1613        # delete comments so we don't match on reg specifiers inside
1614        code = commentRE.sub('', code)
1615        # search for operands
1616        next_pos = 0
1617        while 1:
1618            match = operandsRE.search(code, next_pos)
1619            if not match:
1620                # no more matches: we're done
1621                break
1622            op = match.groups()
1623            # regexp groups are operand full name, base, and extension
1624            (op_full, op_base, op_ext) = op
1625            # find this op in the master list
1626            op_desc = master_list.find_base(op_base)
1627            if not op_desc:
1628                error(0, 'Found operand %s which is not in the master list!' \
1629                        ' This is an internal error' % \
1630                          op_base)
1631            else:
1632                # See if we've already found this operand
1633                op_desc = self.find_base(op_base)
1634                if not op_desc:
1635                    # if not, add a reference to it to this sub list
1636                    self.append(master_list.bases[op_base])
1637
1638            # start next search after end of current match
1639            next_pos = match.end()
1640        self.sort()
1641        self.memOperand = None
1642        for op_desc in self.items:
1643            if op_desc.isMem():
1644                if self.memOperand:
1645                    error(0, "Code block has more than one memory operand.")
1646                self.memOperand = op_desc
1647
1648# Regular expression object to match C++ comments
1649# (used in findOperands())
1650commentRE = re.compile(r'//.*\n')
1651
1652# Regular expression object to match assignment statements
1653# (used in findOperands())
1654assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE)
1655
1656# Munge operand names in code string to make legal C++ variable names.
1657# This means getting rid of the type extension if any.
1658# (Will match base_name attribute of Operand object.)
1659def substMungedOpNames(code):
1660    return operandsWithExtRE.sub(r'\1', code)
1661
1662# Fix up code snippets for final substitution in templates.
1663def mungeSnippet(s):
1664    if isinstance(s, str):
1665        return substMungedOpNames(substBitOps(s))
1666    else:
1667        return s
1668
1669def makeFlagConstructor(flag_list):
1670    if len(flag_list) == 0:
1671        return ''
1672    # filter out repeated flags
1673    flag_list.sort()
1674    i = 1
1675    while i < len(flag_list):
1676        if flag_list[i] == flag_list[i-1]:
1677            del flag_list[i]
1678        else:
1679            i += 1
1680    pre = '\n\tflags['
1681    post = '] = true;'
1682    code = pre + string.join(flag_list, post + pre) + post
1683    return code
1684
1685# Assume all instruction flags are of the form 'IsFoo'
1686instFlagRE = re.compile(r'Is.*')
1687
1688# OpClass constants end in 'Op' except No_OpClass
1689opClassRE = re.compile(r'.*Op|No_OpClass')
1690
1691class InstObjParams:
1692    def __init__(self, mnem, class_name, base_class = '',
1693                 snippets = {}, opt_args = []):
1694        self.mnemonic = mnem
1695        self.class_name = class_name
1696        self.base_class = base_class
1697        if not isinstance(snippets, dict):
1698            snippets = {'code' : snippets}
1699        compositeCode = ' '.join(map(str, snippets.values()))
1700        self.snippets = snippets
1701
1702        self.operands = OperandList(compositeCode)
1703        self.constructor = self.operands.concatAttrStrings('constructor')
1704        self.constructor += \
1705                 '\n\t_numSrcRegs = %d;' % self.operands.numSrcRegs
1706        self.constructor += \
1707                 '\n\t_numDestRegs = %d;' % self.operands.numDestRegs
1708        self.constructor += \
1709                 '\n\t_numFPDestRegs = %d;' % self.operands.numFPDestRegs
1710        self.constructor += \
1711                 '\n\t_numIntDestRegs = %d;' % self.operands.numIntDestRegs
1712        self.flags = self.operands.concatAttrLists('flags')
1713
1714        # Make a basic guess on the operand class (function unit type).
1715        # These are good enough for most cases, and can be overridden
1716        # later otherwise.
1717        if 'IsStore' in self.flags:
1718            self.op_class = 'MemWriteOp'
1719        elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags:
1720            self.op_class = 'MemReadOp'
1721        elif 'IsFloating' in self.flags:
1722            self.op_class = 'FloatAddOp'
1723        else:
1724            self.op_class = 'IntAluOp'
1725
1726        # Optional arguments are assumed to be either StaticInst flags
1727        # or an OpClass value.  To avoid having to import a complete
1728        # list of these values to match against, we do it ad-hoc
1729        # with regexps.
1730        for oa in opt_args:
1731            if instFlagRE.match(oa):
1732                self.flags.append(oa)
1733            elif opClassRE.match(oa):
1734                self.op_class = oa
1735            else:
1736                error(0, 'InstObjParams: optional arg "%s" not recognized '
1737                      'as StaticInst::Flag or OpClass.' % oa)
1738
1739        # add flag initialization to contructor here to include
1740        # any flags added via opt_args
1741        self.constructor += makeFlagConstructor(self.flags)
1742
1743        # if 'IsFloating' is set, add call to the FP enable check
1744        # function (which should be provided by isa_desc via a declare)
1745        if 'IsFloating' in self.flags:
1746            self.fp_enable_check = 'fault = checkFpEnableFault(xc);'
1747        else:
1748            self.fp_enable_check = ''
1749
1750#######################
1751#
1752# Output file template
1753#
1754
1755file_template = '''
1756/*
1757 * DO NOT EDIT THIS FILE!!!
1758 *
1759 * It was automatically generated from the ISA description in %(filename)s
1760 */
1761
1762%(includes)s
1763
1764%(global_output)s
1765
1766namespace %(namespace)s {
1767
1768%(namespace_output)s
1769
1770} // namespace %(namespace)s
1771
1772%(decode_function)s
1773'''
1774
1775
1776# Update the output file only if the new contents are different from
1777# the current contents.  Minimizes the files that need to be rebuilt
1778# after minor changes.
1779def update_if_needed(file, contents):
1780    update = False
1781    if os.access(file, os.R_OK):
1782        f = open(file, 'r')
1783        old_contents = f.read()
1784        f.close()
1785        if contents != old_contents:
1786            print 'Updating', file
1787            os.remove(file) # in case it's write-protected
1788            update = True
1789        else:
1790            print 'File', file, 'is unchanged'
1791    else:
1792        print 'Generating', file
1793        update = True
1794    if update:
1795        f = open(file, 'w')
1796        f.write(contents)
1797        f.close()
1798
1799# This regular expression matches '##include' directives
1800includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[\w/.-]*)".*$',
1801                       re.MULTILINE)
1802
1803# Function to replace a matched '##include' directive with the
1804# contents of the specified file (with nested ##includes replaced
1805# recursively).  'matchobj' is an re match object (from a match of
1806# includeRE) and 'dirname' is the directory relative to which the file
1807# path should be resolved.
1808def replace_include(matchobj, dirname):
1809    fname = matchobj.group('filename')
1810    full_fname = os.path.normpath(os.path.join(dirname, fname))
1811    contents = '##newfile "%s"\n%s\n##endfile\n' % \
1812               (full_fname, read_and_flatten(full_fname))
1813    return contents
1814
1815# Read a file and recursively flatten nested '##include' files.
1816def read_and_flatten(filename):
1817    current_dir = os.path.dirname(filename)
1818    try:
1819        contents = open(filename).read()
1820    except IOError:
1821        error(0, 'Error including file "%s"' % filename)
1822    fileNameStack.push((filename, 0))
1823    # Find any includes and include them
1824    contents = includeRE.sub(lambda m: replace_include(m, current_dir),
1825                             contents)
1826    fileNameStack.pop()
1827    return contents
1828
1829#
1830# Read in and parse the ISA description.
1831#
1832def parse_isa_desc(isa_desc_file, output_dir):
1833    # Read file and (recursively) all included files into a string.
1834    # PLY requires that the input be in a single string so we have to
1835    # do this up front.
1836    isa_desc = read_and_flatten(isa_desc_file)
1837
1838    # Initialize filename stack with outer file.
1839    fileNameStack.push((isa_desc_file, 0))
1840
1841    # Parse it.
1842    (isa_name, namespace, global_code, namespace_code) = yacc.parse(isa_desc)
1843
1844    # grab the last three path components of isa_desc_file to put in
1845    # the output
1846    filename = '/'.join(isa_desc_file.split('/')[-3:])
1847
1848    # generate decoder.hh
1849    includes = '#include "base/bitfield.hh" // for bitfield support'
1850    global_output = global_code.header_output
1851    namespace_output = namespace_code.header_output
1852    decode_function = ''
1853    update_if_needed(output_dir + '/decoder.hh', file_template % vars())
1854
1855    # generate decoder.cc
1856    includes = '#include "decoder.hh"'
1857    global_output = global_code.decoder_output
1858    namespace_output = namespace_code.decoder_output
1859    # namespace_output += namespace_code.decode_block
1860    decode_function = namespace_code.decode_block
1861    update_if_needed(output_dir + '/decoder.cc', file_template % vars())
1862
1863    # generate per-cpu exec files
1864    for cpu in cpu_models:
1865        includes = '#include "decoder.hh"\n'
1866        includes += cpu.includes
1867        global_output = global_code.exec_output[cpu.name]
1868        namespace_output = namespace_code.exec_output[cpu.name]
1869        decode_function = ''
1870        update_if_needed(output_dir + '/' + cpu.filename,
1871                          file_template % vars())
1872
1873# global list of CpuModel objects (see cpu_models.py)
1874cpu_models = []
1875
1876# Called as script: get args from command line.
1877# Args are: <path to cpu_models.py> <isa desc file> <output dir> <cpu models>
1878if __name__ == '__main__':
1879    execfile(sys.argv[1])  # read in CpuModel definitions
1880    cpu_models = [CpuModel.dict[cpu] for cpu in sys.argv[4:]]
1881    parse_isa_desc(sys.argv[2], sys.argv[3])
1882