isa_parser.py revision 2632:1bb2f91485ea
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
27import os
28import sys
29import re
30import string
31import traceback
32# get type names
33from types import *
34
35# Prepend the directory where the PLY lex & yacc modules are found
36# to the search path.  Assumes we're compiling in a subdirectory
37# of 'build' in the current tree.
38sys.path[0:0] = [os.environ['M5_PLY']]
39
40import lex
41import yacc
42
43#####################################################################
44#
45#                                Lexer
46#
47# The PLY lexer module takes two things as input:
48# - A list of token names (the string list 'tokens')
49# - A regular expression describing a match for each token.  The
50#   regexp for token FOO can be provided in two ways:
51#   - as a string variable named t_FOO
52#   - as the doc string for a function named t_FOO.  In this case,
53#     the function is also executed, allowing an action to be
54#     associated with each token match.
55#
56#####################################################################
57
58# Reserved words.  These are listed separately as they are matched
59# using the same regexp as generic IDs, but distinguished in the
60# t_ID() function.  The PLY documentation suggests this approach.
61reserved = (
62    'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT',
63    'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS',
64    'OUTPUT', 'SIGNED', 'TEMPLATE'
65    )
66
67# List of tokens.  The lex module requires this.
68tokens = reserved + (
69    # identifier
70    'ID',
71
72    # integer literal
73    'INTLIT',
74
75    # string literal
76    'STRLIT',
77
78    # code literal
79    'CODELIT',
80
81    # ( ) [ ] { } < > , ; : :: *
82    'LPAREN', 'RPAREN',
83    'LBRACKET', 'RBRACKET',
84    'LBRACE', 'RBRACE',
85    'LESS', 'GREATER', 'EQUALS',
86    'COMMA', 'SEMI', 'COLON', 'DBLCOLON',
87    'ASTERISK',
88
89    # C preprocessor directives
90    'CPPDIRECTIVE'
91
92# The following are matched but never returned. commented out to
93# suppress PLY warning
94    # newfile directive
95#    'NEWFILE',
96
97    # endfile directive
98#    'ENDFILE'
99)
100
101# Regular expressions for token matching
102t_LPAREN           = r'\('
103t_RPAREN           = r'\)'
104t_LBRACKET         = r'\['
105t_RBRACKET         = r'\]'
106t_LBRACE           = r'\{'
107t_RBRACE           = r'\}'
108t_LESS             = r'\<'
109t_GREATER          = r'\>'
110t_EQUALS           = r'='
111t_COMMA            = r','
112t_SEMI             = r';'
113t_COLON            = r':'
114t_DBLCOLON         = r'::'
115t_ASTERISK	   = r'\*'
116
117# Identifiers and reserved words
118reserved_map = { }
119for r in reserved:
120    reserved_map[r.lower()] = r
121
122def t_ID(t):
123    r'[A-Za-z_]\w*'
124    t.type = reserved_map.get(t.value,'ID')
125    return t
126
127# Integer literal
128def t_INTLIT(t):
129    r'(0x[\da-fA-F]+)|\d+'
130    try:
131        t.value = int(t.value,0)
132    except ValueError:
133        error(t.lineno, 'Integer value "%s" too large' % t.value)
134        t.value = 0
135    return t
136
137# String literal.  Note that these use only single quotes, and
138# can span multiple lines.
139def t_STRLIT(t):
140    r"(?m)'([^'])+'"
141    # strip off quotes
142    t.value = t.value[1:-1]
143    t.lineno += t.value.count('\n')
144    return t
145
146
147# "Code literal"... like a string literal, but delimiters are
148# '{{' and '}}' so they get formatted nicely under emacs c-mode
149def t_CODELIT(t):
150    r"(?m)\{\{([^\}]|}(?!\}))+\}\}"
151    # strip off {{ & }}
152    t.value = t.value[2:-2]
153    t.lineno += t.value.count('\n')
154    return t
155
156def t_CPPDIRECTIVE(t):
157    r'^\#[^\#].*\n'
158    t.lineno += t.value.count('\n')
159    return t
160
161def t_NEWFILE(t):
162    r'^\#\#newfile\s+"[\w/.-]*"'
163    fileNameStack.push((t.value[11:-1], t.lineno))
164    t.lineno = 0
165
166def t_ENDFILE(t):
167    r'^\#\#endfile'
168    (old_filename, t.lineno) = fileNameStack.pop()
169
170#
171# The functions t_NEWLINE, t_ignore, and t_error are
172# special for the lex module.
173#
174
175# Newlines
176def t_NEWLINE(t):
177    r'\n+'
178    t.lineno += t.value.count('\n')
179
180# Comments
181def t_comment(t):
182    r'//.*'
183
184# Completely ignored characters
185t_ignore           = ' \t\x0c'
186
187# Error handler
188def t_error(t):
189    error(t.lineno, "illegal character '%s'" % t.value[0])
190    t.skip(1)
191
192# Build the lexer
193lex.lex()
194
195#####################################################################
196#
197#                                Parser
198#
199# Every function whose name starts with 'p_' defines a grammar rule.
200# The rule is encoded in the function's doc string, while the
201# function body provides the action taken when the rule is matched.
202# The argument to each function is a list of the values of the
203# rule's symbols: t[0] for the LHS, and t[1..n] for the symbols
204# on the RHS.  For tokens, the value is copied from the t.value
205# attribute provided by the lexer.  For non-terminals, the value
206# is assigned by the producing rule; i.e., the job of the grammar
207# rule function is to set the value for the non-terminal on the LHS
208# (by assigning to t[0]).
209#####################################################################
210
211# The LHS of the first grammar rule is used as the start symbol
212# (in this case, 'specification').  Note that this rule enforces
213# that there will be exactly one namespace declaration, with 0 or more
214# global defs/decls before and after it.  The defs & decls before
215# the namespace decl will be outside the namespace; those after
216# will be inside.  The decoder function is always inside the namespace.
217def p_specification(t):
218    'specification : opt_defs_and_outputs name_decl opt_defs_and_outputs decode_block'
219    global_code = t[1]
220    isa_name = t[2]
221    namespace = isa_name + "Inst"
222    # wrap the decode block as a function definition
223    t[4].wrap_decode_block('''
224StaticInstPtr
225%(isa_name)s::decodeInst(%(isa_name)s::ExtMachInst machInst)
226{
227    using namespace %(namespace)s;
228''' % vars(), '}')
229    # both the latter output blocks and the decode block are in the namespace
230    namespace_code = t[3] + t[4]
231    # pass it all back to the caller of yacc.parse()
232    t[0] = (isa_name, namespace, global_code, namespace_code)
233
234# ISA name declaration looks like "namespace <foo>;"
235def p_name_decl(t):
236    'name_decl : NAMESPACE ID SEMI'
237    t[0] = t[2]
238
239# 'opt_defs_and_outputs' is a possibly empty sequence of
240# def and/or output statements.
241def p_opt_defs_and_outputs_0(t):
242    'opt_defs_and_outputs : empty'
243    t[0] = GenCode()
244
245def p_opt_defs_and_outputs_1(t):
246    'opt_defs_and_outputs : defs_and_outputs'
247    t[0] = t[1]
248
249def p_defs_and_outputs_0(t):
250    'defs_and_outputs : def_or_output'
251    t[0] = t[1]
252
253def p_defs_and_outputs_1(t):
254    'defs_and_outputs : defs_and_outputs def_or_output'
255    t[0] = t[1] + t[2]
256
257# The list of possible definition/output statements.
258def p_def_or_output(t):
259    '''def_or_output : def_format
260                     | def_bitfield
261                     | def_template
262                     | def_operand_types
263                     | def_operands
264                     | output_header
265                     | output_decoder
266                     | output_exec
267                     | global_let'''
268    t[0] = t[1]
269
270# Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied
271# directly to the appropriate output section.
272
273
274# Protect any non-dict-substitution '%'s in a format string
275# (i.e. those not followed by '(')
276def protect_non_subst_percents(s):
277    return re.sub(r'%(?!\()', '%%', s)
278
279# Massage output block by substituting in template definitions and bit
280# operators.  We handle '%'s embedded in the string that don't
281# indicate template substitutions (or CPU-specific symbols, which get
282# handled in GenCode) by doubling them first so that the format
283# operation will reduce them back to single '%'s.
284def process_output(s):
285    s = protect_non_subst_percents(s)
286    # protects cpu-specific symbols too
287    s = protect_cpu_symbols(s)
288    return substBitOps(s % templateMap)
289
290def p_output_header(t):
291    'output_header : OUTPUT HEADER CODELIT SEMI'
292    t[0] = GenCode(header_output = process_output(t[3]))
293
294def p_output_decoder(t):
295    'output_decoder : OUTPUT DECODER CODELIT SEMI'
296    t[0] = GenCode(decoder_output = process_output(t[3]))
297
298def p_output_exec(t):
299    'output_exec : OUTPUT EXEC CODELIT SEMI'
300    t[0] = GenCode(exec_output = process_output(t[3]))
301
302# global let blocks 'let {{...}}' (Python code blocks) are executed
303# directly when seen.  Note that these execute in a special variable
304# context 'exportContext' to prevent the code from polluting this
305# script's namespace.
306def p_global_let(t):
307    'global_let : LET CODELIT SEMI'
308    updateExportContext()
309    try:
310        exec fixPythonIndentation(t[2]) in exportContext
311    except Exception, exc:
312        error(t.lineno(1),
313              'error: %s in global let block "%s".' % (exc, t[2]))
314    t[0] = GenCode() # contributes nothing to the output C++ file
315
316# Define the mapping from operand type extensions to C++ types and bit
317# widths (stored in operandTypeMap).
318def p_def_operand_types(t):
319    'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI'
320    try:
321        userDict = eval('{' + t[3] + '}')
322    except Exception, exc:
323        error(t.lineno(1),
324              'error: %s in def operand_types block "%s".' % (exc, t[3]))
325    buildOperandTypeMap(userDict, t.lineno(1))
326    t[0] = GenCode() # contributes nothing to the output C++ file
327
328# Define the mapping from operand names to operand classes and other
329# traits.  Stored in operandNameMap.
330def p_def_operands(t):
331    'def_operands : DEF OPERANDS CODELIT SEMI'
332    if not globals().has_key('operandTypeMap'):
333        error(t.lineno(1),
334              'error: operand types must be defined before operands')
335    try:
336        userDict = eval('{' + t[3] + '}')
337    except Exception, exc:
338        error(t.lineno(1),
339              'error: %s in def operands block "%s".' % (exc, t[3]))
340    buildOperandNameMap(userDict, t.lineno(1))
341    t[0] = GenCode() # contributes nothing to the output C++ file
342
343# A bitfield definition looks like:
344# 'def [signed] bitfield <ID> [<first>:<last>]'
345# This generates a preprocessor macro in the output file.
346def p_def_bitfield_0(t):
347    'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI'
348    expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8])
349    if (t[2] == 'signed'):
350        expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr)
351    hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr)
352    t[0] = GenCode(header_output = hash_define)
353
354# alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]'
355def p_def_bitfield_1(t):
356    'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI'
357    expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6])
358    if (t[2] == 'signed'):
359        expr = 'sext<%d>(%s)' % (1, expr)
360    hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr)
361    t[0] = GenCode(header_output = hash_define)
362
363def p_opt_signed_0(t):
364    'opt_signed : SIGNED'
365    t[0] = t[1]
366
367def p_opt_signed_1(t):
368    'opt_signed : empty'
369    t[0] = ''
370
371# Global map variable to hold templates
372templateMap = {}
373
374def p_def_template(t):
375    'def_template : DEF TEMPLATE ID CODELIT SEMI'
376    templateMap[t[3]] = Template(t[4])
377    t[0] = GenCode()
378
379# An instruction format definition looks like
380# "def format <fmt>(<params>) {{...}};"
381def p_def_format(t):
382    'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI'
383    (id, params, code) = (t[3], t[5], t[7])
384    defFormat(id, params, code, t.lineno(1))
385    t[0] = GenCode()
386
387# The formal parameter list for an instruction format is a possibly
388# empty list of comma-separated parameters.  Positional (standard,
389# non-keyword) parameters must come first, followed by keyword
390# parameters, followed by a '*foo' parameter that gets excess
391# positional arguments (as in Python).  Each of these three parameter
392# categories is optional.
393#
394# Note that we do not support the '**foo' parameter for collecting
395# otherwise undefined keyword args.  Otherwise the parameter list is
396# (I believe) identical to what is supported in Python.
397#
398# The param list generates a tuple, where the first element is a list of
399# the positional params and the second element is a dict containing the
400# keyword params.
401def p_param_list_0(t):
402    'param_list : positional_param_list COMMA nonpositional_param_list'
403    t[0] = t[1] + t[3]
404
405def p_param_list_1(t):
406    '''param_list : positional_param_list
407                  | nonpositional_param_list'''
408    t[0] = t[1]
409
410def p_positional_param_list_0(t):
411    'positional_param_list : empty'
412    t[0] = []
413
414def p_positional_param_list_1(t):
415    'positional_param_list : ID'
416    t[0] = [t[1]]
417
418def p_positional_param_list_2(t):
419    'positional_param_list : positional_param_list COMMA ID'
420    t[0] = t[1] + [t[3]]
421
422def p_nonpositional_param_list_0(t):
423    'nonpositional_param_list : keyword_param_list COMMA excess_args_param'
424    t[0] = t[1] + t[3]
425
426def p_nonpositional_param_list_1(t):
427    '''nonpositional_param_list : keyword_param_list
428                                | excess_args_param'''
429    t[0] = t[1]
430
431def p_keyword_param_list_0(t):
432    'keyword_param_list : keyword_param'
433    t[0] = [t[1]]
434
435def p_keyword_param_list_1(t):
436    'keyword_param_list : keyword_param_list COMMA keyword_param'
437    t[0] = t[1] + [t[3]]
438
439def p_keyword_param(t):
440    'keyword_param : ID EQUALS expr'
441    t[0] = t[1] + ' = ' + t[3].__repr__()
442
443def p_excess_args_param(t):
444    'excess_args_param : ASTERISK ID'
445    # Just concatenate them: '*ID'.  Wrap in list to be consistent
446    # with positional_param_list and keyword_param_list.
447    t[0] = [t[1] + t[2]]
448
449# End of format definition-related rules.
450##############
451
452#
453# A decode block looks like:
454#	decode <field1> [, <field2>]* [default <inst>] { ... }
455#
456def p_decode_block(t):
457    'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE'
458    default_defaults = defaultStack.pop()
459    codeObj = t[5]
460    # use the "default defaults" only if there was no explicit
461    # default statement in decode_stmt_list
462    if not codeObj.has_decode_default:
463        codeObj += default_defaults
464    codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n')
465    t[0] = codeObj
466
467# The opt_default statement serves only to push the "default defaults"
468# onto defaultStack.  This value will be used by nested decode blocks,
469# and used and popped off when the current decode_block is processed
470# (in p_decode_block() above).
471def p_opt_default_0(t):
472    'opt_default : empty'
473    # no default specified: reuse the one currently at the top of the stack
474    defaultStack.push(defaultStack.top())
475    # no meaningful value returned
476    t[0] = None
477
478def p_opt_default_1(t):
479    'opt_default : DEFAULT inst'
480    # push the new default
481    codeObj = t[2]
482    codeObj.wrap_decode_block('\ndefault:\n', 'break;\n')
483    defaultStack.push(codeObj)
484    # no meaningful value returned
485    t[0] = None
486
487def p_decode_stmt_list_0(t):
488    'decode_stmt_list : decode_stmt'
489    t[0] = t[1]
490
491def p_decode_stmt_list_1(t):
492    'decode_stmt_list : decode_stmt decode_stmt_list'
493    if (t[1].has_decode_default and t[2].has_decode_default):
494        error(t.lineno(1), 'Two default cases in decode block')
495    t[0] = t[1] + t[2]
496
497#
498# Decode statement rules
499#
500# There are four types of statements allowed in a decode block:
501# 1. Format blocks 'format <foo> { ... }'
502# 2. Nested decode blocks
503# 3. Instruction definitions.
504# 4. C preprocessor directives.
505
506
507# Preprocessor directives found in a decode statement list are passed
508# through to the output, replicated to all of the output code
509# streams.  This works well for ifdefs, so we can ifdef out both the
510# declarations and the decode cases generated by an instruction
511# definition.  Handling them as part of the grammar makes it easy to
512# keep them in the right place with respect to the code generated by
513# the other statements.
514def p_decode_stmt_cpp(t):
515    'decode_stmt : CPPDIRECTIVE'
516    t[0] = GenCode(t[1], t[1], t[1], t[1])
517
518# A format block 'format <foo> { ... }' sets the default instruction
519# format used to handle instruction definitions inside the block.
520# This format can be overridden by using an explicit format on the
521# instruction definition or with a nested format block.
522def p_decode_stmt_format(t):
523    'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE'
524    # The format will be pushed on the stack when 'push_format_id' is
525    # processed (see below).  Once the parser has recognized the full
526    # production (though the right brace), we're done with the format,
527    # so now we can pop it.
528    formatStack.pop()
529    t[0] = t[4]
530
531# This rule exists so we can set the current format (& push the stack)
532# when we recognize the format name part of the format block.
533def p_push_format_id(t):
534    'push_format_id : ID'
535    try:
536        formatStack.push(formatMap[t[1]])
537        t[0] = ('', '// format %s' % t[1])
538    except KeyError:
539        error(t.lineno(1), 'instruction format "%s" not defined.' % t[1])
540
541# Nested decode block: if the value of the current field matches the
542# specified constant, do a nested decode on some other field.
543def p_decode_stmt_decode(t):
544    'decode_stmt : case_label COLON decode_block'
545    label = t[1]
546    codeObj = t[3]
547    # just wrap the decoding code from the block as a case in the
548    # outer switch statement.
549    codeObj.wrap_decode_block('\n%s:\n' % label)
550    codeObj.has_decode_default = (label == 'default')
551    t[0] = codeObj
552
553# Instruction definition (finally!).
554def p_decode_stmt_inst(t):
555    'decode_stmt : case_label COLON inst SEMI'
556    label = t[1]
557    codeObj = t[3]
558    codeObj.wrap_decode_block('\n%s:' % label, 'break;\n')
559    codeObj.has_decode_default = (label == 'default')
560    t[0] = codeObj
561
562# The case label is either a list of one or more constants or 'default'
563def p_case_label_0(t):
564    'case_label : intlit_list'
565    t[0] = ': '.join(map(lambda a: 'case %#x' % a, t[1]))
566
567def p_case_label_1(t):
568    'case_label : DEFAULT'
569    t[0] = 'default'
570
571#
572# The constant list for a decode case label must be non-empty, but may have
573# one or more comma-separated integer literals in it.
574#
575def p_intlit_list_0(t):
576    'intlit_list : INTLIT'
577    t[0] = [t[1]]
578
579def p_intlit_list_1(t):
580    'intlit_list : intlit_list COMMA INTLIT'
581    t[0] = t[1]
582    t[0].append(t[3])
583
584# Define an instruction using the current instruction format (specified
585# by an enclosing format block).
586# "<mnemonic>(<args>)"
587def p_inst_0(t):
588    'inst : ID LPAREN arg_list RPAREN'
589    # Pass the ID and arg list to the current format class to deal with.
590    currentFormat = formatStack.top()
591    codeObj = currentFormat.defineInst(t[1], t[3], t.lineno(1))
592    args = ','.join(map(str, t[3]))
593    args = re.sub('(?m)^', '//', args)
594    args = re.sub('^//', '', args)
595    comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args)
596    codeObj.prepend_all(comment)
597    t[0] = codeObj
598
599# Define an instruction using an explicitly specified format:
600# "<fmt>::<mnemonic>(<args>)"
601def p_inst_1(t):
602    'inst : ID DBLCOLON ID LPAREN arg_list RPAREN'
603    try:
604        format = formatMap[t[1]]
605    except KeyError:
606        error(t.lineno(1), 'instruction format "%s" not defined.' % t[1])
607    codeObj = format.defineInst(t[3], t[5], t.lineno(1))
608    comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5])
609    codeObj.prepend_all(comment)
610    t[0] = codeObj
611
612# The arg list generates a tuple, where the first element is a list of
613# the positional args and the second element is a dict containing the
614# keyword args.
615def p_arg_list_0(t):
616    'arg_list : positional_arg_list COMMA keyword_arg_list'
617    t[0] = ( t[1], t[3] )
618
619def p_arg_list_1(t):
620    'arg_list : positional_arg_list'
621    t[0] = ( t[1], {} )
622
623def p_arg_list_2(t):
624    'arg_list : keyword_arg_list'
625    t[0] = ( [], t[1] )
626
627def p_positional_arg_list_0(t):
628    'positional_arg_list : empty'
629    t[0] = []
630
631def p_positional_arg_list_1(t):
632    'positional_arg_list : expr'
633    t[0] = [t[1]]
634
635def p_positional_arg_list_2(t):
636    'positional_arg_list : positional_arg_list COMMA expr'
637    t[0] = t[1] + [t[3]]
638
639def p_keyword_arg_list_0(t):
640    'keyword_arg_list : keyword_arg'
641    t[0] = t[1]
642
643def p_keyword_arg_list_1(t):
644    'keyword_arg_list : keyword_arg_list COMMA keyword_arg'
645    t[0] = t[1]
646    t[0].update(t[3])
647
648def p_keyword_arg(t):
649    'keyword_arg : ID EQUALS expr'
650    t[0] = { t[1] : t[3] }
651
652#
653# Basic expressions.  These constitute the argument values of
654# "function calls" (i.e. instruction definitions in the decode block)
655# and default values for formal parameters of format functions.
656#
657# Right now, these are either strings, integers, or (recursively)
658# lists of exprs (using Python square-bracket list syntax).  Note that
659# bare identifiers are trated as string constants here (since there
660# isn't really a variable namespace to refer to).
661#
662def p_expr_0(t):
663    '''expr : ID
664            | INTLIT
665            | STRLIT
666            | CODELIT'''
667    t[0] = t[1]
668
669def p_expr_1(t):
670    '''expr : LBRACKET list_expr RBRACKET'''
671    t[0] = t[2]
672
673def p_list_expr_0(t):
674    'list_expr : expr'
675    t[0] = [t[1]]
676
677def p_list_expr_1(t):
678    'list_expr : list_expr COMMA expr'
679    t[0] = t[1] + [t[3]]
680
681def p_list_expr_2(t):
682    'list_expr : empty'
683    t[0] = []
684
685#
686# Empty production... use in other rules for readability.
687#
688def p_empty(t):
689    'empty :'
690    pass
691
692# Parse error handler.  Note that the argument here is the offending
693# *token*, not a grammar symbol (hence the need to use t.value)
694def p_error(t):
695    if t:
696        error(t.lineno, "syntax error at '%s'" % t.value)
697    else:
698        error(0, "unknown syntax error", True)
699
700# END OF GRAMMAR RULES
701#
702# Now build the parser.
703yacc.yacc()
704
705
706#####################################################################
707#
708#                           Support Classes
709#
710#####################################################################
711
712# Expand template with CPU-specific references into a dictionary with
713# an entry for each CPU model name.  The entry key is the model name
714# and the corresponding value is the template with the CPU-specific
715# refs substituted for that model.
716def expand_cpu_symbols_to_dict(template):
717    # Protect '%'s that don't go with CPU-specific terms
718    t = re.sub(r'%(?!\(CPU_)', '%%', template)
719    result = {}
720    for cpu in cpu_models:
721        result[cpu.name] = t % cpu.strings
722    return result
723
724# *If* the template has CPU-specific references, return a single
725# string containing a copy of the template for each CPU model with the
726# corresponding values substituted in.  If the template has no
727# CPU-specific references, it is returned unmodified.
728def expand_cpu_symbols_to_string(template):
729    if template.find('%(CPU_') != -1:
730        return reduce(lambda x,y: x+y,
731                      expand_cpu_symbols_to_dict(template).values())
732    else:
733        return template
734
735# Protect CPU-specific references by doubling the corresponding '%'s
736# (in preparation for substituting a different set of references into
737# the template).
738def protect_cpu_symbols(template):
739    return re.sub(r'%(?=\(CPU_)', '%%', template)
740
741###############
742# GenCode class
743#
744# The GenCode class encapsulates generated code destined for various
745# output files.  The header_output and decoder_output attributes are
746# strings containing code destined for decoder.hh and decoder.cc
747# respectively.  The decode_block attribute contains code to be
748# incorporated in the decode function itself (that will also end up in
749# decoder.cc).  The exec_output attribute is a dictionary with a key
750# for each CPU model name; the value associated with a particular key
751# is the string of code for that CPU model's exec.cc file.  The
752# has_decode_default attribute is used in the decode block to allow
753# explicit default clauses to override default default clauses.
754
755class GenCode:
756    # Constructor.  At this point we substitute out all CPU-specific
757    # symbols.  For the exec output, these go into the per-model
758    # dictionary.  For all other output types they get collapsed into
759    # a single string.
760    def __init__(self,
761                 header_output = '', decoder_output = '', exec_output = '',
762                 decode_block = '', has_decode_default = False):
763        self.header_output = expand_cpu_symbols_to_string(header_output)
764        self.decoder_output = expand_cpu_symbols_to_string(decoder_output)
765        if isinstance(exec_output, dict):
766            self.exec_output = exec_output
767        elif isinstance(exec_output, str):
768            # If the exec_output arg is a single string, we replicate
769            # it for each of the CPU models, substituting and
770            # %(CPU_foo)s params appropriately.
771            self.exec_output = expand_cpu_symbols_to_dict(exec_output)
772        self.decode_block = expand_cpu_symbols_to_string(decode_block)
773        self.has_decode_default = has_decode_default
774
775    # Override '+' operator: generate a new GenCode object that
776    # concatenates all the individual strings in the operands.
777    def __add__(self, other):
778        exec_output = {}
779        for cpu in cpu_models:
780            n = cpu.name
781            exec_output[n] = self.exec_output[n] + other.exec_output[n]
782        return GenCode(self.header_output + other.header_output,
783                       self.decoder_output + other.decoder_output,
784                       exec_output,
785                       self.decode_block + other.decode_block,
786                       self.has_decode_default or other.has_decode_default)
787
788    # Prepend a string (typically a comment) to all the strings.
789    def prepend_all(self, pre):
790        self.header_output = pre + self.header_output
791        self.decoder_output  = pre + self.decoder_output
792        self.decode_block = pre + self.decode_block
793        for cpu in cpu_models:
794            self.exec_output[cpu.name] = pre + self.exec_output[cpu.name]
795
796    # Wrap the decode block in a pair of strings (e.g., 'case foo:'
797    # and 'break;').  Used to build the big nested switch statement.
798    def wrap_decode_block(self, pre, post = ''):
799        self.decode_block = pre + indent(self.decode_block) + post
800
801################
802# Format object.
803#
804# A format object encapsulates an instruction format.  It must provide
805# a defineInst() method that generates the code for an instruction
806# definition.
807
808exportContextSymbols = ('InstObjParams', 'CodeBlock',
809                        'makeList', 're', 'string')
810
811exportContext = {}
812
813def updateExportContext():
814    exportContext.update(exportDict(*exportContextSymbols))
815    exportContext.update(templateMap)
816
817def exportDict(*symNames):
818    return dict([(s, eval(s)) for s in symNames])
819
820
821class Format:
822    def __init__(self, id, params, code):
823        # constructor: just save away arguments
824        self.id = id
825        self.params = params
826        label = 'def format ' + id
827        self.user_code = compile(fixPythonIndentation(code), label, 'exec')
828        param_list = string.join(params, ", ")
829        f = '''def defInst(_code, _context, %s):
830                my_locals = vars().copy()
831                exec _code in _context, my_locals
832                return my_locals\n''' % param_list
833        c = compile(f, label + ' wrapper', 'exec')
834        exec c
835        self.func = defInst
836
837    def defineInst(self, name, args, lineno):
838        context = {}
839        updateExportContext()
840        context.update(exportContext)
841        context.update({ 'name': name, 'Name': string.capitalize(name) })
842        try:
843            vars = self.func(self.user_code, context, *args[0], **args[1])
844        except Exception, exc:
845            error(lineno, 'error defining "%s": %s.' % (name, exc))
846        for k in vars.keys():
847            if k not in ('header_output', 'decoder_output',
848                         'exec_output', 'decode_block'):
849                del vars[k]
850        return GenCode(**vars)
851
852# Special null format to catch an implicit-format instruction
853# definition outside of any format block.
854class NoFormat:
855    def __init__(self):
856        self.defaultInst = ''
857
858    def defineInst(self, name, args, lineno):
859        error(lineno,
860              'instruction definition "%s" with no active format!' % name)
861
862# This dictionary maps format name strings to Format objects.
863formatMap = {}
864
865# Define a new format
866def defFormat(id, params, code, lineno):
867    # make sure we haven't already defined this one
868    if formatMap.get(id, None) != None:
869        error(lineno, 'format %s redefined.' % id)
870    # create new object and store in global map
871    formatMap[id] = Format(id, params, code)
872
873
874##############
875# Stack: a simple stack object.  Used for both formats (formatStack)
876# and default cases (defaultStack).  Simply wraps a list to give more
877# stack-like syntax and enable initialization with an argument list
878# (as opposed to an argument that's a list).
879
880class Stack(list):
881    def __init__(self, *items):
882        list.__init__(self, items)
883
884    def push(self, item):
885        self.append(item);
886
887    def top(self):
888        return self[-1]
889
890# The global format stack.
891formatStack = Stack(NoFormat())
892
893# The global default case stack.
894defaultStack = Stack( None )
895
896# Global stack that tracks current file and line number.
897# Each element is a tuple (filename, lineno) that records the
898# *current* filename and the line number in the *previous* file where
899# it was included.
900fileNameStack = Stack()
901
902###################
903# Utility functions
904
905#
906# Indent every line in string 's' by two spaces
907# (except preprocessor directives).
908# Used to make nested code blocks look pretty.
909#
910def indent(s):
911    return re.sub(r'(?m)^(?!#)', '  ', s)
912
913#
914# Munge a somewhat arbitrarily formatted piece of Python code
915# (e.g. from a format 'let' block) into something whose indentation
916# will get by the Python parser.
917#
918# The two keys here are that Python will give a syntax error if
919# there's any whitespace at the beginning of the first line, and that
920# all lines at the same lexical nesting level must have identical
921# indentation.  Unfortunately the way code literals work, an entire
922# let block tends to have some initial indentation.  Rather than
923# trying to figure out what that is and strip it off, we prepend 'if
924# 1:' to make the let code the nested block inside the if (and have
925# the parser automatically deal with the indentation for us).
926#
927# We don't want to do this if (1) the code block is empty or (2) the
928# first line of the block doesn't have any whitespace at the front.
929
930def fixPythonIndentation(s):
931    # get rid of blank lines first
932    s = re.sub(r'(?m)^\s*\n', '', s);
933    if (s != '' and re.match(r'[ \t]', s[0])):
934        s = 'if 1:\n' + s
935    return s
936
937# Error handler.  Just call exit.  Output formatted to work under
938# Emacs compile-mode.  Optional 'print_traceback' arg, if set to True,
939# prints a Python stack backtrace too (can be handy when trying to
940# debug the parser itself).
941def error(lineno, string, print_traceback = False):
942    spaces = ""
943    for (filename, line) in fileNameStack[0:-1]:
944        print spaces + "In file included from " + filename + ":"
945        spaces += "  "
946    # Print a Python stack backtrace if requested.
947    if (print_traceback):
948        traceback.print_exc()
949    if lineno != 0:
950        line_str = "%d:" % lineno
951    else:
952        line_str = ""
953    sys.exit(spaces + "%s:%s %s" % (fileNameStack[-1][0], line_str, string))
954
955
956#####################################################################
957#
958#                      Bitfield Operator Support
959#
960#####################################################################
961
962bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>')
963
964bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>')
965bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>')
966
967def substBitOps(code):
968    # first convert single-bit selectors to two-index form
969    # i.e., <n> --> <n:n>
970    code = bitOp1ArgRE.sub(r'<\1:\1>', code)
971    # simple case: selector applied to ID (name)
972    # i.e., foo<a:b> --> bits(foo, a, b)
973    code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code)
974    # if selector is applied to expression (ending in ')'),
975    # we need to search backward for matching '('
976    match = bitOpExprRE.search(code)
977    while match:
978        exprEnd = match.start()
979        here = exprEnd - 1
980        nestLevel = 1
981        while nestLevel > 0:
982            if code[here] == '(':
983                nestLevel -= 1
984            elif code[here] == ')':
985                nestLevel += 1
986            here -= 1
987            if here < 0:
988                sys.exit("Didn't find '('!")
989        exprStart = here+1
990        newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1],
991                                         match.group(1), match.group(2))
992        code = code[:exprStart] + newExpr + code[match.end():]
993        match = bitOpExprRE.search(code)
994    return code
995
996
997####################
998# Template objects.
999#
1000# Template objects are format strings that allow substitution from
1001# the attribute spaces of other objects (e.g. InstObjParams instances).
1002
1003class Template:
1004    def __init__(self, t):
1005        self.template = t
1006
1007    def subst(self, d):
1008        # Start with the template namespace.  Make a copy since we're
1009        # going to modify it.
1010        myDict = templateMap.copy()
1011        # if the argument is a dictionary, we just use it.
1012        if isinstance(d, dict):
1013            myDict.update(d)
1014        # if the argument is an object, we use its attribute map.
1015        elif hasattr(d, '__dict__'):
1016            myDict.update(d.__dict__)
1017        else:
1018            raise TypeError, "Template.subst() arg must be or have dictionary"
1019        # Protect non-Python-dict substitutions (e.g. if there's a printf
1020        # in the templated C++ code)
1021        template = protect_non_subst_percents(self.template)
1022        # CPU-model-specific substitutions are handled later (in GenCode).
1023        template = protect_cpu_symbols(template)
1024        return template % myDict
1025
1026    # Convert to string.  This handles the case when a template with a
1027    # CPU-specific term gets interpolated into another template or into
1028    # an output block.
1029    def __str__(self):
1030        return expand_cpu_symbols_to_string(self.template)
1031
1032#####################################################################
1033#
1034#                             Code Parser
1035#
1036# The remaining code is the support for automatically extracting
1037# instruction characteristics from pseudocode.
1038#
1039#####################################################################
1040
1041# Force the argument to be a list.  Useful for flags, where a caller
1042# can specify a singleton flag or a list of flags.  Also usful for
1043# converting tuples to lists so they can be modified.
1044def makeList(arg):
1045    if isinstance(arg, list):
1046        return arg
1047    elif isinstance(arg, tuple):
1048        return list(arg)
1049    elif not arg:
1050        return []
1051    else:
1052        return [ arg ]
1053
1054# Generate operandTypeMap from the user's 'def operand_types'
1055# statement.
1056def buildOperandTypeMap(userDict, lineno):
1057    global operandTypeMap
1058    operandTypeMap = {}
1059    for (ext, (desc, size)) in userDict.iteritems():
1060        if desc == 'signed int':
1061            ctype = 'int%d_t' % size
1062            is_signed = 1
1063        elif desc == 'unsigned int':
1064            ctype = 'uint%d_t' % size
1065            is_signed = 0
1066        elif desc == 'float':
1067            is_signed = 1	# shouldn't really matter
1068            if size == 32:
1069                ctype = 'float'
1070            elif size == 64:
1071                ctype = 'double'
1072        if ctype == '':
1073            error(lineno, 'Unrecognized type description "%s" in userDict')
1074        operandTypeMap[ext] = (size, ctype, is_signed)
1075
1076#
1077#
1078#
1079# Base class for operand descriptors.  An instance of this class (or
1080# actually a class derived from this one) represents a specific
1081# operand for a code block (e.g, "Rc.sq" as a dest). Intermediate
1082# derived classes encapsulates the traits of a particular operand type
1083# (e.g., "32-bit integer register").
1084#
1085class Operand(object):
1086    def __init__(self, full_name, ext, is_src, is_dest):
1087        self.full_name = full_name
1088        self.ext = ext
1089        self.is_src = is_src
1090        self.is_dest = is_dest
1091        # The 'effective extension' (eff_ext) is either the actual
1092        # extension, if one was explicitly provided, or the default.
1093        if ext:
1094            self.eff_ext = ext
1095        else:
1096            self.eff_ext = self.dflt_ext
1097
1098        (self.size, self.ctype, self.is_signed) = operandTypeMap[self.eff_ext]
1099
1100        # note that mem_acc_size is undefined for non-mem operands...
1101        # template must be careful not to use it if it doesn't apply.
1102        if self.isMem():
1103            self.mem_acc_size = self.makeAccSize()
1104            self.mem_acc_type = self.ctype
1105
1106    # Finalize additional fields (primarily code fields).  This step
1107    # is done separately since some of these fields may depend on the
1108    # register index enumeration that hasn't been performed yet at the
1109    # time of __init__().
1110    def finalize(self):
1111        self.flags = self.getFlags()
1112        self.constructor = self.makeConstructor()
1113        self.op_decl = self.makeDecl()
1114
1115        if self.is_src:
1116            self.op_rd = self.makeRead()
1117            self.op_src_decl = self.makeDecl()
1118        else:
1119            self.op_rd = ''
1120            self.op_src_decl = ''
1121
1122        if self.is_dest:
1123            self.op_wb = self.makeWrite()
1124            self.op_dest_decl = self.makeDecl()
1125        else:
1126            self.op_wb = ''
1127            self.op_dest_decl = ''
1128
1129    def isMem(self):
1130        return 0
1131
1132    def isReg(self):
1133        return 0
1134
1135    def isFloatReg(self):
1136        return 0
1137
1138    def isIntReg(self):
1139        return 0
1140
1141    def isControlReg(self):
1142        return 0
1143
1144    def getFlags(self):
1145        # note the empty slice '[:]' gives us a copy of self.flags[0]
1146        # instead of a reference to it
1147        my_flags = self.flags[0][:]
1148        if self.is_src:
1149            my_flags += self.flags[1]
1150        if self.is_dest:
1151            my_flags += self.flags[2]
1152        return my_flags
1153
1154    def makeDecl(self):
1155        # Note that initializations in the declarations are solely
1156        # to avoid 'uninitialized variable' errors from the compiler.
1157        return self.ctype + ' ' + self.base_name + ' = 0;\n';
1158
1159class IntRegOperand(Operand):
1160    def isReg(self):
1161        return 1
1162
1163    def isIntReg(self):
1164        return 1
1165
1166    def makeConstructor(self):
1167        c = ''
1168        if self.is_src:
1169            c += '\n\t_srcRegIdx[%d] = %s;' % \
1170                 (self.src_reg_idx, self.reg_spec)
1171        if self.is_dest:
1172            c += '\n\t_destRegIdx[%d] = %s;' % \
1173                 (self.dest_reg_idx, self.reg_spec)
1174        return c
1175
1176    def makeRead(self):
1177        if (self.ctype == 'float' or self.ctype == 'double'):
1178            error(0, 'Attempt to read integer register as FP')
1179        if (self.size == self.dflt_size):
1180            return '%s = xc->readIntReg(this, %d);\n' % \
1181                   (self.base_name, self.src_reg_idx)
1182        else:
1183            return '%s = bits(xc->readIntReg(this, %d), %d, 0);\n' % \
1184                   (self.base_name, self.src_reg_idx, self.size-1)
1185
1186    def makeWrite(self):
1187        if (self.ctype == 'float' or self.ctype == 'double'):
1188            error(0, 'Attempt to write integer register as FP')
1189        if (self.size != self.dflt_size and self.is_signed):
1190            final_val = 'sext<%d>(%s)' % (self.size, self.base_name)
1191        else:
1192            final_val = self.base_name
1193        wb = '''
1194        {
1195            %s final_val = %s;
1196            xc->setIntReg(this, %d, final_val);\n
1197            if (traceData) { traceData->setData(final_val); }
1198        }''' % (self.dflt_ctype, final_val, self.dest_reg_idx)
1199        return wb
1200
1201class FloatRegOperand(Operand):
1202    def isReg(self):
1203        return 1
1204
1205    def isFloatReg(self):
1206        return 1
1207
1208    def makeConstructor(self):
1209        c = ''
1210        if self.is_src:
1211            c += '\n\t_srcRegIdx[%d] = %s + FP_Base_DepTag;' % \
1212                 (self.src_reg_idx, self.reg_spec)
1213        if self.is_dest:
1214            c += '\n\t_destRegIdx[%d] = %s + FP_Base_DepTag;' % \
1215                 (self.dest_reg_idx, self.reg_spec)
1216        return c
1217
1218    def makeRead(self):
1219        bit_select = 0
1220        width = 0;
1221        if (self.ctype == 'float'):
1222            func = 'readFloatReg'
1223            width = 32;
1224        elif (self.ctype == 'double'):
1225            func = 'readFloatReg'
1226            width = 64;
1227        else:
1228            func = 'readFloatRegBits'
1229            if (self.ctype == 'uint32_t'):
1230                width = 32;
1231            elif (self.ctype == 'uint64_t'):
1232                width = 64;
1233            if (self.size != self.dflt_size):
1234                bit_select = 1
1235        if width:
1236            base = 'xc->%s(this, %d, %d)' % \
1237                   (func, self.src_reg_idx, width)
1238        else:
1239            base = 'xc->%s(this, %d)' % \
1240                   (func, self.src_reg_idx)
1241        if bit_select:
1242            return '%s = bits(%s, %d, 0);\n' % \
1243                   (self.base_name, base, self.size-1)
1244        else:
1245            return '%s = %s;\n' % (self.base_name, base)
1246
1247    def makeWrite(self):
1248        final_val = self.base_name
1249        final_ctype = self.ctype
1250        widthSpecifier = ''
1251        width = 0
1252        if (self.ctype == 'float'):
1253            width = 32
1254            func = 'setFloatReg'
1255        elif (self.ctype == 'double'):
1256            width = 64
1257            func = 'setFloatReg'
1258        elif (self.ctype == 'uint32_t'):
1259            func = 'setFloatRegBits'
1260            width = 32
1261        elif (self.ctype == 'uint64_t'):
1262            func = 'setFloatRegBits'
1263            width = 64
1264        else:
1265            func = 'setFloatRegBits'
1266            final_ctype = 'uint%d_t' % self.dflt_size
1267            if (self.size != self.dflt_size and self.is_signed):
1268                final_val = 'sext<%d>(%s)' % (self.size, self.base_name)
1269        if width:
1270            widthSpecifier = ', %d' % width
1271        wb = '''
1272        {
1273            %s final_val = %s;
1274            xc->%s(this, %d, final_val%s);\n
1275            if (traceData) { traceData->setData(final_val); }
1276        }''' % (final_ctype, final_val, func, self.dest_reg_idx,
1277                widthSpecifier)
1278        return wb
1279
1280class ControlRegOperand(Operand):
1281    def isReg(self):
1282        return 1
1283
1284    def isControlReg(self):
1285        return 1
1286
1287    def makeConstructor(self):
1288        c = ''
1289        if self.is_src:
1290            c += '\n\t_srcRegIdx[%d] = %s;' % \
1291                 (self.src_reg_idx, self.reg_spec)
1292        if self.is_dest:
1293            c += '\n\t_destRegIdx[%d] = %s;' % \
1294                 (self.dest_reg_idx, self.reg_spec)
1295        return c
1296
1297    def makeRead(self):
1298        bit_select = 0
1299        if (self.ctype == 'float' or self.ctype == 'double'):
1300            error(0, 'Attempt to read control register as FP')
1301        base = 'xc->readMiscReg(%s)' % self.reg_spec
1302        if self.size == self.dflt_size:
1303            return '%s = %s;\n' % (self.base_name, base)
1304        else:
1305            return '%s = bits(%s, %d, 0);\n' % \
1306                   (self.base_name, base, self.size-1)
1307
1308    def makeWrite(self):
1309        if (self.ctype == 'float' or self.ctype == 'double'):
1310            error(0, 'Attempt to write control register as FP')
1311        wb = 'xc->setMiscReg(%s, %s);\n' % (self.reg_spec, self.base_name)
1312        wb += 'if (traceData) { traceData->setData(%s); }' % \
1313              self.base_name
1314        return wb
1315
1316class MemOperand(Operand):
1317    def isMem(self):
1318        return 1
1319
1320    def makeConstructor(self):
1321        return ''
1322
1323    def makeDecl(self):
1324        # Note that initializations in the declarations are solely
1325        # to avoid 'uninitialized variable' errors from the compiler.
1326        # Declare memory data variable.
1327        c = '%s %s = 0;\n' % (self.ctype, self.base_name)
1328        return c
1329
1330    def makeRead(self):
1331        return ''
1332
1333    def makeWrite(self):
1334        return ''
1335
1336    # Return the memory access size *in bits*, suitable for
1337    # forming a type via "uint%d_t".  Divide by 8 if you want bytes.
1338    def makeAccSize(self):
1339        return self.size
1340
1341
1342class NPCOperand(Operand):
1343    def makeConstructor(self):
1344        return ''
1345
1346    def makeRead(self):
1347        return '%s = xc->readNextPC();\n' % self.base_name
1348
1349    def makeWrite(self):
1350        return 'xc->setNextPC(%s);\n' % self.base_name
1351
1352class NNPCOperand(Operand):
1353    def makeConstructor(self):
1354        return ''
1355
1356    def makeRead(self):
1357        return '%s = xc->readNextNPC();\n' % self.base_name
1358
1359    def makeWrite(self):
1360        return 'xc->setNextNPC(%s);\n' % self.base_name
1361
1362def buildOperandNameMap(userDict, lineno):
1363    global operandNameMap
1364    operandNameMap = {}
1365    for (op_name, val) in userDict.iteritems():
1366        (base_cls_name, dflt_ext, reg_spec, flags, sort_pri) = val
1367        (dflt_size, dflt_ctype, dflt_is_signed) = operandTypeMap[dflt_ext]
1368        # Canonical flag structure is a triple of lists, where each list
1369        # indicates the set of flags implied by this operand always, when
1370        # used as a source, and when used as a dest, respectively.
1371        # For simplicity this can be initialized using a variety of fairly
1372        # obvious shortcuts; we convert these to canonical form here.
1373        if not flags:
1374            # no flags specified (e.g., 'None')
1375            flags = ( [], [], [] )
1376        elif isinstance(flags, str):
1377            # a single flag: assumed to be unconditional
1378            flags = ( [ flags ], [], [] )
1379        elif isinstance(flags, list):
1380            # a list of flags: also assumed to be unconditional
1381            flags = ( flags, [], [] )
1382        elif isinstance(flags, tuple):
1383            # it's a tuple: it should be a triple,
1384            # but each item could be a single string or a list
1385            (uncond_flags, src_flags, dest_flags) = flags
1386            flags = (makeList(uncond_flags),
1387                     makeList(src_flags), makeList(dest_flags))
1388        # Accumulate attributes of new operand class in tmp_dict
1389        tmp_dict = {}
1390        for attr in ('dflt_ext', 'reg_spec', 'flags', 'sort_pri',
1391                     'dflt_size', 'dflt_ctype', 'dflt_is_signed'):
1392            tmp_dict[attr] = eval(attr)
1393        tmp_dict['base_name'] = op_name
1394        # New class name will be e.g. "IntReg_Ra"
1395        cls_name = base_cls_name + '_' + op_name
1396        # Evaluate string arg to get class object.  Note that the
1397        # actual base class for "IntReg" is "IntRegOperand", i.e. we
1398        # have to append "Operand".
1399        try:
1400            base_cls = eval(base_cls_name + 'Operand')
1401        except NameError:
1402            error(lineno,
1403                  'error: unknown operand base class "%s"' % base_cls_name)
1404        # The following statement creates a new class called
1405        # <cls_name> as a subclass of <base_cls> with the attributes
1406        # in tmp_dict, just as if we evaluated a class declaration.
1407        operandNameMap[op_name] = type(cls_name, (base_cls,), tmp_dict)
1408
1409    # Define operand variables.
1410    operands = userDict.keys()
1411
1412    operandsREString = (r'''
1413    (?<![\w\.])	     # neg. lookbehind assertion: prevent partial matches
1414    ((%s)(?:\.(\w+))?)   # match: operand with optional '.' then suffix
1415    (?![\w\.])	     # neg. lookahead assertion: prevent partial matches
1416    '''
1417                        % string.join(operands, '|'))
1418
1419    global operandsRE
1420    operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE)
1421
1422    # Same as operandsREString, but extension is mandatory, and only two
1423    # groups are returned (base and ext, not full name as above).
1424    # Used for subtituting '_' for '.' to make C++ identifiers.
1425    operandsWithExtREString = (r'(?<![\w\.])(%s)\.(\w+)(?![\w\.])'
1426                               % string.join(operands, '|'))
1427
1428    global operandsWithExtRE
1429    operandsWithExtRE = re.compile(operandsWithExtREString, re.MULTILINE)
1430
1431
1432class OperandList:
1433
1434    # Find all the operands in the given code block.  Returns an operand
1435    # descriptor list (instance of class OperandList).
1436    def __init__(self, code):
1437        self.items = []
1438        self.bases = {}
1439        # delete comments so we don't match on reg specifiers inside
1440        code = commentRE.sub('', code)
1441        # search for operands
1442        next_pos = 0
1443        while 1:
1444            match = operandsRE.search(code, next_pos)
1445            if not match:
1446                # no more matches: we're done
1447                break
1448            op = match.groups()
1449            # regexp groups are operand full name, base, and extension
1450            (op_full, op_base, op_ext) = op
1451            # if the token following the operand is an assignment, this is
1452            # a destination (LHS), else it's a source (RHS)
1453            is_dest = (assignRE.match(code, match.end()) != None)
1454            is_src = not is_dest
1455            # see if we've already seen this one
1456            op_desc = self.find_base(op_base)
1457            if op_desc:
1458                if op_desc.ext != op_ext:
1459                    error(0, 'Inconsistent extensions for operand %s' % \
1460                          op_base)
1461                op_desc.is_src = op_desc.is_src or is_src
1462                op_desc.is_dest = op_desc.is_dest or is_dest
1463            else:
1464                # new operand: create new descriptor
1465                op_desc = operandNameMap[op_base](op_full, op_ext,
1466                                                  is_src, is_dest)
1467                self.append(op_desc)
1468            # start next search after end of current match
1469            next_pos = match.end()
1470        self.sort()
1471        # enumerate source & dest register operands... used in building
1472        # constructor later
1473        self.numSrcRegs = 0
1474        self.numDestRegs = 0
1475        self.numFPDestRegs = 0
1476        self.numIntDestRegs = 0
1477        self.memOperand = None
1478        for op_desc in self.items:
1479            if op_desc.isReg():
1480                if op_desc.is_src:
1481                    op_desc.src_reg_idx = self.numSrcRegs
1482                    self.numSrcRegs += 1
1483                if op_desc.is_dest:
1484                    op_desc.dest_reg_idx = self.numDestRegs
1485                    self.numDestRegs += 1
1486                    if op_desc.isFloatReg():
1487                        self.numFPDestRegs += 1
1488                    elif op_desc.isIntReg():
1489                        self.numIntDestRegs += 1
1490            elif op_desc.isMem():
1491                if self.memOperand:
1492                    error(0, "Code block has more than one memory operand.")
1493                self.memOperand = op_desc
1494        # now make a final pass to finalize op_desc fields that may depend
1495        # on the register enumeration
1496        for op_desc in self.items:
1497            op_desc.finalize()
1498
1499    def __len__(self):
1500        return len(self.items)
1501
1502    def __getitem__(self, index):
1503        return self.items[index]
1504
1505    def append(self, op_desc):
1506        self.items.append(op_desc)
1507        self.bases[op_desc.base_name] = op_desc
1508
1509    def find_base(self, base_name):
1510        # like self.bases[base_name], but returns None if not found
1511        # (rather than raising exception)
1512        return self.bases.get(base_name)
1513
1514    # internal helper function for concat[Some]Attr{Strings|Lists}
1515    def __internalConcatAttrs(self, attr_name, filter, result):
1516        for op_desc in self.items:
1517            if filter(op_desc):
1518                result += getattr(op_desc, attr_name)
1519        return result
1520
1521    # return a single string that is the concatenation of the (string)
1522    # values of the specified attribute for all operands
1523    def concatAttrStrings(self, attr_name):
1524        return self.__internalConcatAttrs(attr_name, lambda x: 1, '')
1525
1526    # like concatAttrStrings, but only include the values for the operands
1527    # for which the provided filter function returns true
1528    def concatSomeAttrStrings(self, filter, attr_name):
1529        return self.__internalConcatAttrs(attr_name, filter, '')
1530
1531    # return a single list that is the concatenation of the (list)
1532    # values of the specified attribute for all operands
1533    def concatAttrLists(self, attr_name):
1534        return self.__internalConcatAttrs(attr_name, lambda x: 1, [])
1535
1536    # like concatAttrLists, but only include the values for the operands
1537    # for which the provided filter function returns true
1538    def concatSomeAttrLists(self, filter, attr_name):
1539        return self.__internalConcatAttrs(attr_name, filter, [])
1540
1541    def sort(self):
1542        self.items.sort(lambda a, b: a.sort_pri - b.sort_pri)
1543
1544# Regular expression object to match C++ comments
1545# (used in findOperands())
1546commentRE = re.compile(r'//.*\n')
1547
1548# Regular expression object to match assignment statements
1549# (used in findOperands())
1550assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE)
1551
1552# Munge operand names in code string to make legal C++ variable names.
1553# This means getting rid of the type extension if any.
1554# (Will match base_name attribute of Operand object.)
1555def substMungedOpNames(code):
1556    return operandsWithExtRE.sub(r'\1', code)
1557
1558def joinLists(t):
1559    return map(string.join, t)
1560
1561def makeFlagConstructor(flag_list):
1562    if len(flag_list) == 0:
1563        return ''
1564    # filter out repeated flags
1565    flag_list.sort()
1566    i = 1
1567    while i < len(flag_list):
1568        if flag_list[i] == flag_list[i-1]:
1569            del flag_list[i]
1570        else:
1571            i += 1
1572    pre = '\n\tflags['
1573    post = '] = true;'
1574    code = pre + string.join(flag_list, post + pre) + post
1575    return code
1576
1577class CodeBlock:
1578    def __init__(self, code):
1579        self.orig_code = code
1580        self.operands = OperandList(code)
1581        self.code = substMungedOpNames(substBitOps(code))
1582        self.constructor = self.operands.concatAttrStrings('constructor')
1583        self.constructor += \
1584                 '\n\t_numSrcRegs = %d;' % self.operands.numSrcRegs
1585        self.constructor += \
1586                 '\n\t_numDestRegs = %d;' % self.operands.numDestRegs
1587        self.constructor += \
1588                 '\n\t_numFPDestRegs = %d;' % self.operands.numFPDestRegs
1589        self.constructor += \
1590                 '\n\t_numIntDestRegs = %d;' % self.operands.numIntDestRegs
1591
1592        self.op_decl = self.operands.concatAttrStrings('op_decl')
1593
1594        is_src = lambda op: op.is_src
1595        is_dest = lambda op: op.is_dest
1596
1597        self.op_src_decl = \
1598                  self.operands.concatSomeAttrStrings(is_src, 'op_src_decl')
1599        self.op_dest_decl = \
1600                  self.operands.concatSomeAttrStrings(is_dest, 'op_dest_decl')
1601
1602        self.op_rd = self.operands.concatAttrStrings('op_rd')
1603        self.op_wb = self.operands.concatAttrStrings('op_wb')
1604
1605        self.flags = self.operands.concatAttrLists('flags')
1606
1607        if self.operands.memOperand:
1608            self.mem_acc_size = self.operands.memOperand.mem_acc_size
1609            self.mem_acc_type = self.operands.memOperand.mem_acc_type
1610
1611        # Make a basic guess on the operand class (function unit type).
1612        # These are good enough for most cases, and will be overridden
1613        # later otherwise.
1614        if 'IsStore' in self.flags:
1615            self.op_class = 'MemWriteOp'
1616        elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags:
1617            self.op_class = 'MemReadOp'
1618        elif 'IsFloating' in self.flags:
1619            self.op_class = 'FloatAddOp'
1620        else:
1621            self.op_class = 'IntAluOp'
1622
1623# Assume all instruction flags are of the form 'IsFoo'
1624instFlagRE = re.compile(r'Is.*')
1625
1626# OpClass constants end in 'Op' except No_OpClass
1627opClassRE = re.compile(r'.*Op|No_OpClass')
1628
1629class InstObjParams:
1630    def __init__(self, mnem, class_name, base_class = '',
1631                 code = None, opt_args = [], *extras):
1632        self.mnemonic = mnem
1633        self.class_name = class_name
1634        self.base_class = base_class
1635        if code:
1636            #If the user already made a CodeBlock, pick the parts from it
1637            if isinstance(code, CodeBlock):
1638                origCode = code.orig_code
1639                codeBlock = code
1640            else:
1641                origCode = code
1642                codeBlock = CodeBlock(code)
1643            compositeCode = '\n'.join([origCode] +
1644                    [pair[1] for pair in extras])
1645            compositeBlock = CodeBlock(compositeCode)
1646            for code_attr in compositeBlock.__dict__.keys():
1647                setattr(self, code_attr, getattr(compositeBlock, code_attr))
1648            for (key, snippet) in extras:
1649                setattr(self, key, CodeBlock(snippet).code)
1650            self.code = codeBlock.code
1651            self.orig_code = origCode
1652        else:
1653            self.constructor = ''
1654            self.flags = []
1655        # Optional arguments are assumed to be either StaticInst flags
1656        # or an OpClass value.  To avoid having to import a complete
1657        # list of these values to match against, we do it ad-hoc
1658        # with regexps.
1659        for oa in opt_args:
1660            if instFlagRE.match(oa):
1661                self.flags.append(oa)
1662            elif opClassRE.match(oa):
1663                self.op_class = oa
1664            else:
1665                error(0, 'InstObjParams: optional arg "%s" not recognized '
1666                      'as StaticInst::Flag or OpClass.' % oa)
1667
1668        # add flag initialization to contructor here to include
1669        # any flags added via opt_args
1670        self.constructor += makeFlagConstructor(self.flags)
1671
1672        # if 'IsFloating' is set, add call to the FP enable check
1673        # function (which should be provided by isa_desc via a declare)
1674        if 'IsFloating' in self.flags:
1675            self.fp_enable_check = 'fault = checkFpEnableFault(xc);'
1676        else:
1677            self.fp_enable_check = ''
1678
1679#######################
1680#
1681# Output file template
1682#
1683
1684file_template = '''
1685/*
1686 * DO NOT EDIT THIS FILE!!!
1687 *
1688 * It was automatically generated from the ISA description in %(filename)s
1689 */
1690
1691%(includes)s
1692
1693%(global_output)s
1694
1695namespace %(namespace)s {
1696
1697%(namespace_output)s
1698
1699} // namespace %(namespace)s
1700
1701%(decode_function)s
1702'''
1703
1704
1705# Update the output file only if the new contents are different from
1706# the current contents.  Minimizes the files that need to be rebuilt
1707# after minor changes.
1708def update_if_needed(file, contents):
1709    update = False
1710    if os.access(file, os.R_OK):
1711        f = open(file, 'r')
1712        old_contents = f.read()
1713        f.close()
1714        if contents != old_contents:
1715            print 'Updating', file
1716            os.remove(file) # in case it's write-protected
1717            update = True
1718        else:
1719            print 'File', file, 'is unchanged'
1720    else:
1721        print 'Generating', file
1722        update = True
1723    if update:
1724        f = open(file, 'w')
1725        f.write(contents)
1726        f.close()
1727
1728# This regular expression matches '##include' directives
1729includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[\w/.-]*)".*$',
1730                       re.MULTILINE)
1731
1732# Function to replace a matched '##include' directive with the
1733# contents of the specified file (with nested ##includes replaced
1734# recursively).  'matchobj' is an re match object (from a match of
1735# includeRE) and 'dirname' is the directory relative to which the file
1736# path should be resolved.
1737def replace_include(matchobj, dirname):
1738    fname = matchobj.group('filename')
1739    full_fname = os.path.normpath(os.path.join(dirname, fname))
1740    contents = '##newfile "%s"\n%s\n##endfile\n' % \
1741               (full_fname, read_and_flatten(full_fname))
1742    return contents
1743
1744# Read a file and recursively flatten nested '##include' files.
1745def read_and_flatten(filename):
1746    current_dir = os.path.dirname(filename)
1747    try:
1748        contents = open(filename).read()
1749    except IOError:
1750        error(0, 'Error including file "%s"' % filename)
1751    fileNameStack.push((filename, 0))
1752    # Find any includes and include them
1753    contents = includeRE.sub(lambda m: replace_include(m, current_dir),
1754                             contents)
1755    fileNameStack.pop()
1756    return contents
1757
1758#
1759# Read in and parse the ISA description.
1760#
1761def parse_isa_desc(isa_desc_file, output_dir):
1762    # Read file and (recursively) all included files into a string.
1763    # PLY requires that the input be in a single string so we have to
1764    # do this up front.
1765    isa_desc = read_and_flatten(isa_desc_file)
1766
1767    # Initialize filename stack with outer file.
1768    fileNameStack.push((isa_desc_file, 0))
1769
1770    # Parse it.
1771    (isa_name, namespace, global_code, namespace_code) = yacc.parse(isa_desc)
1772
1773    # grab the last three path components of isa_desc_file to put in
1774    # the output
1775    filename = '/'.join(isa_desc_file.split('/')[-3:])
1776
1777    # generate decoder.hh
1778    includes = '#include "base/bitfield.hh" // for bitfield support'
1779    global_output = global_code.header_output
1780    namespace_output = namespace_code.header_output
1781    decode_function = ''
1782    update_if_needed(output_dir + '/decoder.hh', file_template % vars())
1783
1784    # generate decoder.cc
1785    includes = '#include "decoder.hh"'
1786    global_output = global_code.decoder_output
1787    namespace_output = namespace_code.decoder_output
1788    # namespace_output += namespace_code.decode_block
1789    decode_function = namespace_code.decode_block
1790    update_if_needed(output_dir + '/decoder.cc', file_template % vars())
1791
1792    # generate per-cpu exec files
1793    for cpu in cpu_models:
1794        includes = '#include "decoder.hh"\n'
1795        includes += cpu.includes
1796        global_output = global_code.exec_output[cpu.name]
1797        namespace_output = namespace_code.exec_output[cpu.name]
1798        decode_function = ''
1799        update_if_needed(output_dir + '/' + cpu.filename,
1800                          file_template % vars())
1801
1802# global list of CpuModel objects (see cpu_models.py)
1803cpu_models = []
1804
1805# Called as script: get args from command line.
1806# Args are: <path to cpu_models.py> <isa desc file> <output dir> <cpu models>
1807if __name__ == '__main__':
1808    execfile(sys.argv[1])  # read in CpuModel definitions
1809    cpu_models = [CpuModel.dict[cpu] for cpu in sys.argv[4:]]
1810    parse_isa_desc(sys.argv[2], sys.argv[3])
1811