isa_parser.py revision 3735:86a7cf4dcc11
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', 'CodeBlock',
812                        'makeList', 're', 'string')
813
814exportContext = {}
815
816def updateExportContext():
817    exportContext.update(exportDict(*exportContextSymbols))
818    exportContext.update(templateMap)
819
820def exportDict(*symNames):
821    return dict([(s, eval(s)) for s in symNames])
822
823
824class Format:
825    def __init__(self, id, params, code):
826        # constructor: just save away arguments
827        self.id = id
828        self.params = params
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')
837        exec c
838        self.func = defInst
839
840    def defineInst(self, name, args, lineno):
841        context = {}
842        updateExportContext()
843        context.update(exportContext)
844        context.update({ 'name': name, 'Name': string.capitalize(name) })
845        try:
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'):
852                del vars[k]
853        return GenCode(**vars)
854
855# Special null format to catch an implicit-format instruction
856# definition outside of any format block.
857class NoFormat:
858    def __init__(self):
859        self.defaultInst = ''
860
861    def defineInst(self, name, args, lineno):
862        error(lineno,
863              'instruction definition "%s" with no active format!' % name)
864
865# This dictionary maps format name strings to Format objects.
866formatMap = {}
867
868# Define a new format
869def 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)
875
876
877##############
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).
882
883class Stack(list):
884    def __init__(self, *items):
885        list.__init__(self, items)
886
887    def push(self, item):
888        self.append(item);
889
890    def top(self):
891        return self[-1]
892
893# The global format stack.
894formatStack = Stack(NoFormat())
895
896# The global default case stack.
897defaultStack = Stack( None )
898
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
902# it was included.
903fileNameStack = Stack()
904
905###################
906# Utility functions
907
908#
909# Indent every line in string 's' by two spaces
910# (except preprocessor directives).
911# Used to make nested code blocks look pretty.
912#
913def indent(s):
914    return re.sub(r'(?m)^(?!#)', '  ', s)
915
916#
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.
920#
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).
929#
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.
932
933def 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])):
937        s = 'if 1:\n' + s
938    return s
939
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).
944def error(lineno, string, print_traceback = False):
945    spaces = ""
946    for (filename, line) in fileNameStack[0:-1]:
947        print spaces + "In file included from " + filename + ":"
948        spaces += "  "
949    # Print a Python stack backtrace if requested.
950    if (print_traceback):
951        traceback.print_exc()
952    if lineno != 0:
953        line_str = "%d:" % lineno
954    else:
955        line_str = ""
956    sys.exit(spaces + "%s:%s %s" % (fileNameStack[-1][0], line_str, string))
957
958
959#####################################################################
960#
961#                      Bitfield Operator Support
962#
963#####################################################################
964
965bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>')
966
967bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>')
968bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>')
969
970def 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)
980    while match:
981        exprEnd = match.start()
982        here = exprEnd - 1
983        nestLevel = 1
984        while nestLevel > 0:
985            if code[here] == '(':
986                nestLevel -= 1
987            elif code[here] == ')':
988                nestLevel += 1
989            here -= 1
990            if here < 0:
991                sys.exit("Didn't find '('!")
992        exprStart = here+1
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)
997    return code
998
999
1000####################
1001# Template objects.
1002#
1003# Template objects are format strings that allow substitution from
1004# the attribute spaces of other objects (e.g. InstObjParams instances).
1005
1006class Template:
1007    def __init__(self, t):
1008        self.template = t
1009
1010    def subst(self, d):
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):
1016            myDict.update(d)
1017        # if the argument is an object, we use its attribute map.
1018        elif hasattr(d, '__dict__'):
1019            myDict.update(d.__dict__)
1020        else:
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
1028
1029    # Convert to string.  This handles the case when a template with a
1030    # CPU-specific term gets interpolated into another template or into
1031    # an output block.
1032    def __str__(self):
1033        return expand_cpu_symbols_to_string(self.template)
1034
1035#####################################################################
1036#
1037#                             Code Parser
1038#
1039# The remaining code is the support for automatically extracting
1040# instruction characteristics from pseudocode.
1041#
1042#####################################################################
1043
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.
1047def makeList(arg):
1048    if isinstance(arg, list):
1049        return arg
1050    elif isinstance(arg, tuple):
1051        return list(arg)
1052    elif not arg:
1053        return []
1054    else:
1055        return [ arg ]
1056
1057# Generate operandTypeMap from the user's 'def operand_types'
1058# statement.
1059def buildOperandTypeMap(userDict, lineno):
1060    global operandTypeMap
1061    operandTypeMap = {}
1062    for (ext, (desc, size)) in userDict.iteritems():
1063        if desc == 'signed int':
1064            ctype = 'int%d_t' % size
1065            is_signed = 1
1066        elif desc == 'unsigned int':
1067            ctype = 'uint%d_t' % size
1068            is_signed = 0
1069        elif desc == 'float':
1070            is_signed = 1	# shouldn't really matter
1071            if size == 32:
1072                ctype = 'float'
1073            elif size == 64:
1074                ctype = 'double'
1075        if ctype == '':
1076            error(lineno, 'Unrecognized type description "%s" in userDict')
1077        operandTypeMap[ext] = (size, ctype, is_signed)
1078
1079#
1080#
1081#
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").
1087#
1088class Operand(object):
1089    def __init__(self, full_name, ext, is_src, is_dest):
1090        self.full_name = full_name
1091        self.ext = ext
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.
1096        if ext:
1097            self.eff_ext = ext
1098        else:
1099            self.eff_ext = self.dflt_ext
1100
1101        (self.size, self.ctype, self.is_signed) = operandTypeMap[self.eff_ext]
1102
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.
1105        if self.isMem():
1106            self.mem_acc_size = self.makeAccSize()
1107            self.mem_acc_type = self.ctype
1108
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__().
1113    def finalize(self):
1114        self.flags = self.getFlags()
1115        self.constructor = self.makeConstructor()
1116        self.op_decl = self.makeDecl()
1117
1118        if self.is_src:
1119            self.op_rd = self.makeRead()
1120            self.op_src_decl = self.makeDecl()
1121        else:
1122            self.op_rd = ''
1123            self.op_src_decl = ''
1124
1125        if self.is_dest:
1126            self.op_wb = self.makeWrite()
1127            self.op_dest_decl = self.makeDecl()
1128        else:
1129            self.op_wb = ''
1130            self.op_dest_decl = ''
1131
1132    def isMem(self):
1133        return 0
1134
1135    def isReg(self):
1136        return 0
1137
1138    def isFloatReg(self):
1139        return 0
1140
1141    def isIntReg(self):
1142        return 0
1143
1144    def isControlReg(self):
1145        return 0
1146
1147    def getFlags(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][:]
1151        if self.is_src:
1152            my_flags += self.flags[1]
1153        if self.is_dest:
1154            my_flags += self.flags[2]
1155        return my_flags
1156
1157    def makeDecl(self):
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';
1161
1162class IntRegOperand(Operand):
1163    def isReg(self):
1164        return 1
1165
1166    def isIntReg(self):
1167        return 1
1168
1169    def makeConstructor(self):
1170        c = ''
1171        if self.is_src:
1172            c += '\n\t_srcRegIdx[%d] = %s;' % \
1173                 (self.src_reg_idx, self.reg_spec)
1174        if self.is_dest:
1175            c += '\n\t_destRegIdx[%d] = %s;' % \
1176                 (self.dest_reg_idx, self.reg_spec)
1177        return c
1178
1179    def makeRead(self):
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->readIntRegOperand(this, %d);\n' % \
1184                   (self.base_name, self.src_reg_idx)
1185        elif (self.size > self.dflt_size):
1186            int_reg_val = 'xc->readIntRegOperand(this, %d)' % \
1187                          (self.src_reg_idx)
1188            if (self.is_signed):
1189                int_reg_val = 'sext<%d>(%s)' % (self.dflt_size, int_reg_val)
1190            return '%s = %s;\n' % (self.base_name, int_reg_val)
1191        else:
1192            return '%s = bits(xc->readIntRegOperand(this, %d), %d, 0);\n' % \
1193                   (self.base_name, self.src_reg_idx, self.size-1)
1194
1195    def makeWrite(self):
1196        if (self.ctype == 'float' or self.ctype == 'double'):
1197            error(0, 'Attempt to write integer register as FP')
1198        if (self.size != self.dflt_size and self.is_signed):
1199            final_val = 'sext<%d>(%s)' % (self.size, self.base_name)
1200        else:
1201            final_val = self.base_name
1202        wb = '''
1203        {
1204            %s final_val = %s;
1205            xc->setIntRegOperand(this, %d, final_val);\n
1206            if (traceData) { traceData->setData(final_val); }
1207        }''' % (self.dflt_ctype, final_val, self.dest_reg_idx)
1208        return wb
1209
1210class FloatRegOperand(Operand):
1211    def isReg(self):
1212        return 1
1213
1214    def isFloatReg(self):
1215        return 1
1216
1217    def makeConstructor(self):
1218        c = ''
1219        if self.is_src:
1220            c += '\n\t_srcRegIdx[%d] = %s + FP_Base_DepTag;' % \
1221                 (self.src_reg_idx, self.reg_spec)
1222        if self.is_dest:
1223            c += '\n\t_destRegIdx[%d] = %s + FP_Base_DepTag;' % \
1224                 (self.dest_reg_idx, self.reg_spec)
1225        return c
1226
1227    def makeRead(self):
1228        bit_select = 0
1229        width = 0;
1230        if (self.ctype == 'float'):
1231            func = 'readFloatRegOperand'
1232            width = 32;
1233        elif (self.ctype == 'double'):
1234            func = 'readFloatRegOperand'
1235            width = 64;
1236        else:
1237            func = 'readFloatRegOperandBits'
1238            if (self.ctype == 'uint32_t'):
1239                width = 32;
1240            elif (self.ctype == 'uint64_t'):
1241                width = 64;
1242            if (self.size != self.dflt_size):
1243                bit_select = 1
1244        if width:
1245            base = 'xc->%s(this, %d, %d)' % \
1246                   (func, self.src_reg_idx, width)
1247        else:
1248            base = 'xc->%s(this, %d)' % \
1249                   (func, self.src_reg_idx)
1250        if bit_select:
1251            return '%s = bits(%s, %d, 0);\n' % \
1252                   (self.base_name, base, self.size-1)
1253        else:
1254            return '%s = %s;\n' % (self.base_name, base)
1255
1256    def makeWrite(self):
1257        final_val = self.base_name
1258        final_ctype = self.ctype
1259        widthSpecifier = ''
1260        width = 0
1261        if (self.ctype == 'float'):
1262            width = 32
1263            func = 'setFloatRegOperand'
1264        elif (self.ctype == 'double'):
1265            width = 64
1266            func = 'setFloatRegOperand'
1267        elif (self.ctype == 'uint32_t'):
1268            func = 'setFloatRegOperandBits'
1269            width = 32
1270        elif (self.ctype == 'uint64_t'):
1271            func = 'setFloatRegOperandBits'
1272            width = 64
1273        else:
1274            func = 'setFloatRegOperandBits'
1275            final_ctype = 'uint%d_t' % self.dflt_size
1276            if (self.size != self.dflt_size and self.is_signed):
1277                final_val = 'sext<%d>(%s)' % (self.size, self.base_name)
1278        if width:
1279            widthSpecifier = ', %d' % width
1280        wb = '''
1281        {
1282            %s final_val = %s;
1283            xc->%s(this, %d, final_val%s);\n
1284            if (traceData) { traceData->setData(final_val); }
1285        }''' % (final_ctype, final_val, func, self.dest_reg_idx,
1286                widthSpecifier)
1287        return wb
1288
1289class ControlRegOperand(Operand):
1290    def isReg(self):
1291        return 1
1292
1293    def isControlReg(self):
1294        return 1
1295
1296    def makeConstructor(self):
1297        c = ''
1298        if self.is_src:
1299            c += '\n\t_srcRegIdx[%d] = %s;' % \
1300                 (self.src_reg_idx, self.reg_spec)
1301        if self.is_dest:
1302            c += '\n\t_destRegIdx[%d] = %s;' % \
1303                 (self.dest_reg_idx, self.reg_spec)
1304        return c
1305
1306    def makeRead(self):
1307        bit_select = 0
1308        if (self.ctype == 'float' or self.ctype == 'double'):
1309            error(0, 'Attempt to read control register as FP')
1310        base = 'xc->readMiscReg(%s)' % self.reg_spec
1311        if self.size == self.dflt_size:
1312            return '%s = %s;\n' % (self.base_name, base)
1313        else:
1314            return '%s = bits(%s, %d, 0);\n' % \
1315                   (self.base_name, base, self.size-1)
1316
1317    def makeWrite(self):
1318        if (self.ctype == 'float' or self.ctype == 'double'):
1319            error(0, 'Attempt to write control register as FP')
1320        wb = 'xc->setMiscRegWithEffect(%s, %s);\n' % (self.reg_spec, self.base_name)
1321        wb += 'if (traceData) { traceData->setData(%s); }' % \
1322              self.base_name
1323        return wb
1324
1325class MemOperand(Operand):
1326    def isMem(self):
1327        return 1
1328
1329    def makeConstructor(self):
1330        return ''
1331
1332    def makeDecl(self):
1333        # Note that initializations in the declarations are solely
1334        # to avoid 'uninitialized variable' errors from the compiler.
1335        # Declare memory data variable.
1336        c = '%s %s = 0;\n' % (self.ctype, self.base_name)
1337        return c
1338
1339    def makeRead(self):
1340        return ''
1341
1342    def makeWrite(self):
1343        return ''
1344
1345    # Return the memory access size *in bits*, suitable for
1346    # forming a type via "uint%d_t".  Divide by 8 if you want bytes.
1347    def makeAccSize(self):
1348        return self.size
1349
1350
1351class NPCOperand(Operand):
1352    def makeConstructor(self):
1353        return ''
1354
1355    def makeRead(self):
1356        return '%s = xc->readNextPC();\n' % self.base_name
1357
1358    def makeWrite(self):
1359        return 'xc->setNextPC(%s);\n' % self.base_name
1360
1361class NNPCOperand(Operand):
1362    def makeConstructor(self):
1363        return ''
1364
1365    def makeRead(self):
1366        return '%s = xc->readNextNPC();\n' % self.base_name
1367
1368    def makeWrite(self):
1369        return 'xc->setNextNPC(%s);\n' % self.base_name
1370
1371def buildOperandNameMap(userDict, lineno):
1372    global operandNameMap
1373    operandNameMap = {}
1374    for (op_name, val) in userDict.iteritems():
1375        (base_cls_name, dflt_ext, reg_spec, flags, sort_pri) = val
1376        (dflt_size, dflt_ctype, dflt_is_signed) = operandTypeMap[dflt_ext]
1377        # Canonical flag structure is a triple of lists, where each list
1378        # indicates the set of flags implied by this operand always, when
1379        # used as a source, and when used as a dest, respectively.
1380        # For simplicity this can be initialized using a variety of fairly
1381        # obvious shortcuts; we convert these to canonical form here.
1382        if not flags:
1383            # no flags specified (e.g., 'None')
1384            flags = ( [], [], [] )
1385        elif isinstance(flags, str):
1386            # a single flag: assumed to be unconditional
1387            flags = ( [ flags ], [], [] )
1388        elif isinstance(flags, list):
1389            # a list of flags: also assumed to be unconditional
1390            flags = ( flags, [], [] )
1391        elif isinstance(flags, tuple):
1392            # it's a tuple: it should be a triple,
1393            # but each item could be a single string or a list
1394            (uncond_flags, src_flags, dest_flags) = flags
1395            flags = (makeList(uncond_flags),
1396                     makeList(src_flags), makeList(dest_flags))
1397        # Accumulate attributes of new operand class in tmp_dict
1398        tmp_dict = {}
1399        for attr in ('dflt_ext', 'reg_spec', 'flags', 'sort_pri',
1400                     'dflt_size', 'dflt_ctype', 'dflt_is_signed'):
1401            tmp_dict[attr] = eval(attr)
1402        tmp_dict['base_name'] = op_name
1403        # New class name will be e.g. "IntReg_Ra"
1404        cls_name = base_cls_name + '_' + op_name
1405        # Evaluate string arg to get class object.  Note that the
1406        # actual base class for "IntReg" is "IntRegOperand", i.e. we
1407        # have to append "Operand".
1408        try:
1409            base_cls = eval(base_cls_name + 'Operand')
1410        except NameError:
1411            error(lineno,
1412                  'error: unknown operand base class "%s"' % base_cls_name)
1413        # The following statement creates a new class called
1414        # <cls_name> as a subclass of <base_cls> with the attributes
1415        # in tmp_dict, just as if we evaluated a class declaration.
1416        operandNameMap[op_name] = type(cls_name, (base_cls,), tmp_dict)
1417
1418    # Define operand variables.
1419    operands = userDict.keys()
1420
1421    operandsREString = (r'''
1422    (?<![\w\.])	     # neg. lookbehind assertion: prevent partial matches
1423    ((%s)(?:\.(\w+))?)   # match: operand with optional '.' then suffix
1424    (?![\w\.])	     # neg. lookahead assertion: prevent partial matches
1425    '''
1426                        % string.join(operands, '|'))
1427
1428    global operandsRE
1429    operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE)
1430
1431    # Same as operandsREString, but extension is mandatory, and only two
1432    # groups are returned (base and ext, not full name as above).
1433    # Used for subtituting '_' for '.' to make C++ identifiers.
1434    operandsWithExtREString = (r'(?<![\w\.])(%s)\.(\w+)(?![\w\.])'
1435                               % string.join(operands, '|'))
1436
1437    global operandsWithExtRE
1438    operandsWithExtRE = re.compile(operandsWithExtREString, re.MULTILINE)
1439
1440
1441class OperandList:
1442
1443    # Find all the operands in the given code block.  Returns an operand
1444    # descriptor list (instance of class OperandList).
1445    def __init__(self, code):
1446        self.items = []
1447        self.bases = {}
1448        # delete comments so we don't match on reg specifiers inside
1449        code = commentRE.sub('', code)
1450        # search for operands
1451        next_pos = 0
1452        while 1:
1453            match = operandsRE.search(code, next_pos)
1454            if not match:
1455                # no more matches: we're done
1456                break
1457            op = match.groups()
1458            # regexp groups are operand full name, base, and extension
1459            (op_full, op_base, op_ext) = op
1460            # if the token following the operand is an assignment, this is
1461            # a destination (LHS), else it's a source (RHS)
1462            is_dest = (assignRE.match(code, match.end()) != None)
1463            is_src = not is_dest
1464            # see if we've already seen this one
1465            op_desc = self.find_base(op_base)
1466            if op_desc:
1467                if op_desc.ext != op_ext:
1468                    error(0, 'Inconsistent extensions for operand %s' % \
1469                          op_base)
1470                op_desc.is_src = op_desc.is_src or is_src
1471                op_desc.is_dest = op_desc.is_dest or is_dest
1472            else:
1473                # new operand: create new descriptor
1474                op_desc = operandNameMap[op_base](op_full, op_ext,
1475                                                  is_src, is_dest)
1476                self.append(op_desc)
1477            # start next search after end of current match
1478            next_pos = match.end()
1479        self.sort()
1480        # enumerate source & dest register operands... used in building
1481        # constructor later
1482        self.numSrcRegs = 0
1483        self.numDestRegs = 0
1484        self.numFPDestRegs = 0
1485        self.numIntDestRegs = 0
1486        self.memOperand = None
1487        for op_desc in self.items:
1488            if op_desc.isReg():
1489                if op_desc.is_src:
1490                    op_desc.src_reg_idx = self.numSrcRegs
1491                    self.numSrcRegs += 1
1492                if op_desc.is_dest:
1493                    op_desc.dest_reg_idx = self.numDestRegs
1494                    self.numDestRegs += 1
1495                    if op_desc.isFloatReg():
1496                        self.numFPDestRegs += 1
1497                    elif op_desc.isIntReg():
1498                        self.numIntDestRegs += 1
1499            elif op_desc.isMem():
1500                if self.memOperand:
1501                    error(0, "Code block has more than one memory operand.")
1502                self.memOperand = op_desc
1503        # now make a final pass to finalize op_desc fields that may depend
1504        # on the register enumeration
1505        for op_desc in self.items:
1506            op_desc.finalize()
1507
1508    def __len__(self):
1509        return len(self.items)
1510
1511    def __getitem__(self, index):
1512        return self.items[index]
1513
1514    def append(self, op_desc):
1515        self.items.append(op_desc)
1516        self.bases[op_desc.base_name] = op_desc
1517
1518    def find_base(self, base_name):
1519        # like self.bases[base_name], but returns None if not found
1520        # (rather than raising exception)
1521        return self.bases.get(base_name)
1522
1523    # internal helper function for concat[Some]Attr{Strings|Lists}
1524    def __internalConcatAttrs(self, attr_name, filter, result):
1525        for op_desc in self.items:
1526            if filter(op_desc):
1527                result += getattr(op_desc, attr_name)
1528        return result
1529
1530    # return a single string that is the concatenation of the (string)
1531    # values of the specified attribute for all operands
1532    def concatAttrStrings(self, attr_name):
1533        return self.__internalConcatAttrs(attr_name, lambda x: 1, '')
1534
1535    # like concatAttrStrings, but only include the values for the operands
1536    # for which the provided filter function returns true
1537    def concatSomeAttrStrings(self, filter, attr_name):
1538        return self.__internalConcatAttrs(attr_name, filter, '')
1539
1540    # return a single list that is the concatenation of the (list)
1541    # values of the specified attribute for all operands
1542    def concatAttrLists(self, attr_name):
1543        return self.__internalConcatAttrs(attr_name, lambda x: 1, [])
1544
1545    # like concatAttrLists, but only include the values for the operands
1546    # for which the provided filter function returns true
1547    def concatSomeAttrLists(self, filter, attr_name):
1548        return self.__internalConcatAttrs(attr_name, filter, [])
1549
1550    def sort(self):
1551        self.items.sort(lambda a, b: a.sort_pri - b.sort_pri)
1552
1553# Regular expression object to match C++ comments
1554# (used in findOperands())
1555commentRE = re.compile(r'//.*\n')
1556
1557# Regular expression object to match assignment statements
1558# (used in findOperands())
1559assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE)
1560
1561# Munge operand names in code string to make legal C++ variable names.
1562# This means getting rid of the type extension if any.
1563# (Will match base_name attribute of Operand object.)
1564def substMungedOpNames(code):
1565    return operandsWithExtRE.sub(r'\1', code)
1566
1567def joinLists(t):
1568    return map(string.join, t)
1569
1570def makeFlagConstructor(flag_list):
1571    if len(flag_list) == 0:
1572        return ''
1573    # filter out repeated flags
1574    flag_list.sort()
1575    i = 1
1576    while i < len(flag_list):
1577        if flag_list[i] == flag_list[i-1]:
1578            del flag_list[i]
1579        else:
1580            i += 1
1581    pre = '\n\tflags['
1582    post = '] = true;'
1583    code = pre + string.join(flag_list, post + pre) + post
1584    return code
1585
1586class CodeBlock:
1587    def __init__(self, code):
1588        self.orig_code = code
1589        self.operands = OperandList(code)
1590        self.code = substMungedOpNames(substBitOps(code))
1591        self.constructor = self.operands.concatAttrStrings('constructor')
1592        self.constructor += \
1593                 '\n\t_numSrcRegs = %d;' % self.operands.numSrcRegs
1594        self.constructor += \
1595                 '\n\t_numDestRegs = %d;' % self.operands.numDestRegs
1596        self.constructor += \
1597                 '\n\t_numFPDestRegs = %d;' % self.operands.numFPDestRegs
1598        self.constructor += \
1599                 '\n\t_numIntDestRegs = %d;' % self.operands.numIntDestRegs
1600
1601        self.op_decl = self.operands.concatAttrStrings('op_decl')
1602
1603        is_src = lambda op: op.is_src
1604        is_dest = lambda op: op.is_dest
1605
1606        self.op_src_decl = \
1607                  self.operands.concatSomeAttrStrings(is_src, 'op_src_decl')
1608        self.op_dest_decl = \
1609                  self.operands.concatSomeAttrStrings(is_dest, 'op_dest_decl')
1610
1611        self.op_rd = self.operands.concatAttrStrings('op_rd')
1612        self.op_wb = self.operands.concatAttrStrings('op_wb')
1613
1614        self.flags = self.operands.concatAttrLists('flags')
1615
1616        if self.operands.memOperand:
1617            self.mem_acc_size = self.operands.memOperand.mem_acc_size
1618            self.mem_acc_type = self.operands.memOperand.mem_acc_type
1619
1620        # Make a basic guess on the operand class (function unit type).
1621        # These are good enough for most cases, and will be overridden
1622        # later otherwise.
1623        if 'IsStore' in self.flags:
1624            self.op_class = 'MemWriteOp'
1625        elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags:
1626            self.op_class = 'MemReadOp'
1627        elif 'IsFloating' in self.flags:
1628            self.op_class = 'FloatAddOp'
1629        else:
1630            self.op_class = 'IntAluOp'
1631
1632# Assume all instruction flags are of the form 'IsFoo'
1633instFlagRE = re.compile(r'Is.*')
1634
1635# OpClass constants end in 'Op' except No_OpClass
1636opClassRE = re.compile(r'.*Op|No_OpClass')
1637
1638class InstObjParams:
1639    def __init__(self, mnem, class_name, base_class = '',
1640                 code = None, opt_args = [], extras = {}):
1641        self.mnemonic = mnem
1642        self.class_name = class_name
1643        self.base_class = base_class
1644        if code:
1645            #If the user already made a CodeBlock, pick the parts from it
1646            if isinstance(code, CodeBlock):
1647                origCode = code.orig_code
1648                codeBlock = code
1649            else:
1650                origCode = code
1651                codeBlock = CodeBlock(code)
1652            stringExtras = {}
1653            otherExtras = {}
1654            for (k, v) in extras.items():
1655                if type(v) == str:
1656                    stringExtras[k] = v
1657                else:
1658                    otherExtras[k] = v
1659            compositeCode = "\n".join([origCode] + stringExtras.values())
1660            # compositeCode = '\n'.join([origCode] +
1661            #	    [pair[1] for pair in extras])
1662            compositeBlock = CodeBlock(compositeCode)
1663            for code_attr in compositeBlock.__dict__.keys():
1664                setattr(self, code_attr, getattr(compositeBlock, code_attr))
1665            for (key, snippet) in stringExtras.items():
1666                setattr(self, key, CodeBlock(snippet).code)
1667            for (key, item) in otherExtras.items():
1668                setattr(self, key, item)
1669            self.code = codeBlock.code
1670            self.orig_code = origCode
1671        else:
1672            self.constructor = ''
1673            self.flags = []
1674        # Optional arguments are assumed to be either StaticInst flags
1675        # or an OpClass value.  To avoid having to import a complete
1676        # list of these values to match against, we do it ad-hoc
1677        # with regexps.
1678        for oa in opt_args:
1679            if instFlagRE.match(oa):
1680                self.flags.append(oa)
1681            elif opClassRE.match(oa):
1682                self.op_class = oa
1683            else:
1684                error(0, 'InstObjParams: optional arg "%s" not recognized '
1685                      'as StaticInst::Flag or OpClass.' % oa)
1686
1687        # add flag initialization to contructor here to include
1688        # any flags added via opt_args
1689        self.constructor += makeFlagConstructor(self.flags)
1690
1691        # if 'IsFloating' is set, add call to the FP enable check
1692        # function (which should be provided by isa_desc via a declare)
1693        if 'IsFloating' in self.flags:
1694            self.fp_enable_check = 'fault = checkFpEnableFault(xc);'
1695        else:
1696            self.fp_enable_check = ''
1697
1698#######################
1699#
1700# Output file template
1701#
1702
1703file_template = '''
1704/*
1705 * DO NOT EDIT THIS FILE!!!
1706 *
1707 * It was automatically generated from the ISA description in %(filename)s
1708 */
1709
1710%(includes)s
1711
1712%(global_output)s
1713
1714namespace %(namespace)s {
1715
1716%(namespace_output)s
1717
1718} // namespace %(namespace)s
1719
1720%(decode_function)s
1721'''
1722
1723
1724# Update the output file only if the new contents are different from
1725# the current contents.  Minimizes the files that need to be rebuilt
1726# after minor changes.
1727def update_if_needed(file, contents):
1728    update = False
1729    if os.access(file, os.R_OK):
1730        f = open(file, 'r')
1731        old_contents = f.read()
1732        f.close()
1733        if contents != old_contents:
1734            print 'Updating', file
1735            os.remove(file) # in case it's write-protected
1736            update = True
1737        else:
1738            print 'File', file, 'is unchanged'
1739    else:
1740        print 'Generating', file
1741        update = True
1742    if update:
1743        f = open(file, 'w')
1744        f.write(contents)
1745        f.close()
1746
1747# This regular expression matches '##include' directives
1748includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[\w/.-]*)".*$',
1749                       re.MULTILINE)
1750
1751# Function to replace a matched '##include' directive with the
1752# contents of the specified file (with nested ##includes replaced
1753# recursively).  'matchobj' is an re match object (from a match of
1754# includeRE) and 'dirname' is the directory relative to which the file
1755# path should be resolved.
1756def replace_include(matchobj, dirname):
1757    fname = matchobj.group('filename')
1758    full_fname = os.path.normpath(os.path.join(dirname, fname))
1759    contents = '##newfile "%s"\n%s\n##endfile\n' % \
1760               (full_fname, read_and_flatten(full_fname))
1761    return contents
1762
1763# Read a file and recursively flatten nested '##include' files.
1764def read_and_flatten(filename):
1765    current_dir = os.path.dirname(filename)
1766    try:
1767        contents = open(filename).read()
1768    except IOError:
1769        error(0, 'Error including file "%s"' % filename)
1770    fileNameStack.push((filename, 0))
1771    # Find any includes and include them
1772    contents = includeRE.sub(lambda m: replace_include(m, current_dir),
1773                             contents)
1774    fileNameStack.pop()
1775    return contents
1776
1777#
1778# Read in and parse the ISA description.
1779#
1780def parse_isa_desc(isa_desc_file, output_dir):
1781    # Read file and (recursively) all included files into a string.
1782    # PLY requires that the input be in a single string so we have to
1783    # do this up front.
1784    isa_desc = read_and_flatten(isa_desc_file)
1785
1786    # Initialize filename stack with outer file.
1787    fileNameStack.push((isa_desc_file, 0))
1788
1789    # Parse it.
1790    (isa_name, namespace, global_code, namespace_code) = yacc.parse(isa_desc)
1791
1792    # grab the last three path components of isa_desc_file to put in
1793    # the output
1794    filename = '/'.join(isa_desc_file.split('/')[-3:])
1795
1796    # generate decoder.hh
1797    includes = '#include "base/bitfield.hh" // for bitfield support'
1798    global_output = global_code.header_output
1799    namespace_output = namespace_code.header_output
1800    decode_function = ''
1801    update_if_needed(output_dir + '/decoder.hh', file_template % vars())
1802
1803    # generate decoder.cc
1804    includes = '#include "decoder.hh"'
1805    global_output = global_code.decoder_output
1806    namespace_output = namespace_code.decoder_output
1807    # namespace_output += namespace_code.decode_block
1808    decode_function = namespace_code.decode_block
1809    update_if_needed(output_dir + '/decoder.cc', file_template % vars())
1810
1811    # generate per-cpu exec files
1812    for cpu in cpu_models:
1813        includes = '#include "decoder.hh"\n'
1814        includes += cpu.includes
1815        global_output = global_code.exec_output[cpu.name]
1816        namespace_output = namespace_code.exec_output[cpu.name]
1817        decode_function = ''
1818        update_if_needed(output_dir + '/' + cpu.filename,
1819                          file_template % vars())
1820
1821# global list of CpuModel objects (see cpu_models.py)
1822cpu_models = []
1823
1824# Called as script: get args from command line.
1825# Args are: <path to cpu_models.py> <isa desc file> <output dir> <cpu models>
1826if __name__ == '__main__':
1827    execfile(sys.argv[1])  # read in CpuModel definitions
1828    cpu_models = [CpuModel.dict[cpu] for cpu in sys.argv[4:]]
1829    parse_isa_desc(sys.argv[2], sys.argv[3])
1830