isa_parser.py

来自「M5,一个功能强大的多处理器系统模拟器.很多针对处理器架构,性能的研究都使用它作」· Python 代码 · 共 1,867 行 · 第 1/5 页

# alternate form for structure member: 'def bitfield <ID> <ID>'
def p_def_bitfield_struct(t):
    'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI'
    if (t[2] != ''):
        error(t.lexer.lineno, 'error: structure bitfields are always unsigned.')
    expr = 'machInst.%s' % t[5]
    hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr)
    t[0] = GenCode(header_output = hash_define)

def p_id_with_dot_0(t):
    'id_with_dot : ID'
    t[0] = t[1]

def p_id_with_dot_1(t):
    'id_with_dot : ID DOT id_with_dot'
    t[0] = t[1] + t[2] + t[3]

def p_opt_signed_0(t):
    'opt_signed : SIGNED'
    t[0] = t[1]

def p_opt_signed_1(t):
    'opt_signed : empty'
    t[0] = ''

# Global map variable to hold templates
templateMap = {}

def p_def_template(t):
    'def_template : DEF TEMPLATE ID CODELIT SEMI'
    templateMap[t[3]] = Template(t[4])
    t[0] = GenCode()

# An instruction format definition looks like
# "def format <fmt>(<params>) {{...}};"
def p_def_format(t):
    'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI'
    (id, params, code) = (t[3], t[5], t[7])
    defFormat(id, params, code, t.lexer.lineno)
    t[0] = GenCode()

# The formal parameter list for an instruction format is a possibly
# empty list of comma-separated parameters.  Positional (standard,
# non-keyword) parameters must come first, followed by keyword
# parameters, followed by a '*foo' parameter that gets excess
# positional arguments (as in Python).  Each of these three parameter
# categories is optional.
#
# Note that we do not support the '**foo' parameter for collecting
# otherwise undefined keyword args.  Otherwise the parameter list is
# (I believe) identical to what is supported in Python.
#
# The param list generates a tuple, where the first element is a list of
# the positional params and the second element is a dict containing the
# keyword params.
def p_param_list_0(t):
    'param_list : positional_param_list COMMA nonpositional_param_list'
    t[0] = t[1] + t[3]

def p_param_list_1(t):
    '''param_list : positional_param_list
                  | nonpositional_param_list'''
    t[0] = t[1]

def p_positional_param_list_0(t):
    'positional_param_list : empty'
    t[0] = []

def p_positional_param_list_1(t):
    'positional_param_list : ID'
    t[0] = [t[1]]

def p_positional_param_list_2(t):
    'positional_param_list : positional_param_list COMMA ID'
    t[0] = t[1] + [t[3]]

def p_nonpositional_param_list_0(t):
    'nonpositional_param_list : keyword_param_list COMMA excess_args_param'
    t[0] = t[1] + t[3]

def p_nonpositional_param_list_1(t):
    '''nonpositional_param_list : keyword_param_list
                                | excess_args_param'''
    t[0] = t[1]

def p_keyword_param_list_0(t):
    'keyword_param_list : keyword_param'
    t[0] = [t[1]]

def p_keyword_param_list_1(t):
    'keyword_param_list : keyword_param_list COMMA keyword_param'
    t[0] = t[1] + [t[3]]

def p_keyword_param(t):
    'keyword_param : ID EQUALS expr'
    t[0] = t[1] + ' = ' + t[3].__repr__()

def p_excess_args_param(t):
    'excess_args_param : ASTERISK ID'
    # Just concatenate them: '*ID'.  Wrap in list to be consistent
    # with positional_param_list and keyword_param_list.
    t[0] = [t[1] + t[2]]

# End of format definition-related rules.
##############

#
# A decode block looks like:
#	decode <field1> [, <field2>]* [default <inst>] { ... }
#
def p_decode_block(t):
    'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE'
    default_defaults = defaultStack.pop()
    codeObj = t[5]
    # use the "default defaults" only if there was no explicit
    # default statement in decode_stmt_list
    if not codeObj.has_decode_default:
        codeObj += default_defaults
    codeObj.wrap_decode_block('switch (%s) {\n' % t[2], '}\n')
    t[0] = codeObj

# The opt_default statement serves only to push the "default defaults"
# onto defaultStack.  This value will be used by nested decode blocks,
# and used and popped off when the current decode_block is processed
# (in p_decode_block() above).
def p_opt_default_0(t):
    'opt_default : empty'
    # no default specified: reuse the one currently at the top of the stack
    defaultStack.push(defaultStack.top())
    # no meaningful value returned
    t[0] = None

def p_opt_default_1(t):
    'opt_default : DEFAULT inst'
    # push the new default
    codeObj = t[2]
    codeObj.wrap_decode_block('\ndefault:\n', 'break;\n')
    defaultStack.push(codeObj)
    # no meaningful value returned
    t[0] = None

def p_decode_stmt_list_0(t):
    'decode_stmt_list : decode_stmt'
    t[0] = t[1]

def p_decode_stmt_list_1(t):
    'decode_stmt_list : decode_stmt decode_stmt_list'
    if (t[1].has_decode_default and t[2].has_decode_default):
        error(t.lexer.lineno, 'Two default cases in decode block')
    t[0] = t[1] + t[2]

#
# Decode statement rules
#
# There are four types of statements allowed in a decode block:
# 1. Format blocks 'format <foo> { ... }'
# 2. Nested decode blocks
# 3. Instruction definitions.
# 4. C preprocessor directives.


# Preprocessor directives found in a decode statement list are passed
# through to the output, replicated to all of the output code
# streams.  This works well for ifdefs, so we can ifdef out both the
# declarations and the decode cases generated by an instruction
# definition.  Handling them as part of the grammar makes it easy to
# keep them in the right place with respect to the code generated by
# the other statements.
def p_decode_stmt_cpp(t):
    'decode_stmt : CPPDIRECTIVE'
    t[0] = GenCode(t[1], t[1], t[1], t[1])

# A format block 'format <foo> { ... }' sets the default instruction
# format used to handle instruction definitions inside the block.
# This format can be overridden by using an explicit format on the
# instruction definition or with a nested format block.
def p_decode_stmt_format(t):
    'decode_stmt : FORMAT push_format_id LBRACE decode_stmt_list RBRACE'
    # The format will be pushed on the stack when 'push_format_id' is
    # processed (see below).  Once the parser has recognized the full
    # production (though the right brace), we're done with the format,
    # so now we can pop it.
    formatStack.pop()
    t[0] = t[4]

# This rule exists so we can set the current format (& push the stack)
# when we recognize the format name part of the format block.
def p_push_format_id(t):
    'push_format_id : ID'
    try:
        formatStack.push(formatMap[t[1]])
        t[0] = ('', '// format %s' % t[1])
    except KeyError:
        error(t.lexer.lineno, 'instruction format "%s" not defined.' % t[1])

# Nested decode block: if the value of the current field matches the
# specified constant, do a nested decode on some other field.
def p_decode_stmt_decode(t):
    'decode_stmt : case_label COLON decode_block'
    label = t[1]
    codeObj = t[3]
    # just wrap the decoding code from the block as a case in the
    # outer switch statement.
    codeObj.wrap_decode_block('\n%s:\n' % label)
    codeObj.has_decode_default = (label == 'default')
    t[0] = codeObj

# Instruction definition (finally!).
def p_decode_stmt_inst(t):
    'decode_stmt : case_label COLON inst SEMI'
    label = t[1]
    codeObj = t[3]
    codeObj.wrap_decode_block('\n%s:' % label, 'break;\n')
    codeObj.has_decode_default = (label == 'default')
    t[0] = codeObj

# The case label is either a list of one or more constants or 'default'
def p_case_label_0(t):
    'case_label : intlit_list'
    t[0] = ': '.join(map(lambda a: 'case %#x' % a, t[1]))

def p_case_label_1(t):
    'case_label : DEFAULT'
    t[0] = 'default'

#
# The constant list for a decode case label must be non-empty, but may have
# one or more comma-separated integer literals in it.
#
def p_intlit_list_0(t):
    'intlit_list : INTLIT'
    t[0] = [t[1]]

def p_intlit_list_1(t):
    'intlit_list : intlit_list COMMA INTLIT'
    t[0] = t[1]
    t[0].append(t[3])

# Define an instruction using the current instruction format (specified
# by an enclosing format block).
# "<mnemonic>(<args>)"
def p_inst_0(t):
    'inst : ID LPAREN arg_list RPAREN'
    # Pass the ID and arg list to the current format class to deal with.
    currentFormat = formatStack.top()
    codeObj = currentFormat.defineInst(t[1], t[3], t.lexer.lineno)
    args = ','.join(map(str, t[3]))
    args = re.sub('(?m)^', '//', args)
    args = re.sub('^//', '', args)
    comment = '\n// %s::%s(%s)\n' % (currentFormat.id, t[1], args)
    codeObj.prepend_all(comment)
    t[0] = codeObj

# Define an instruction using an explicitly specified format:
# "<fmt>::<mnemonic>(<args>)"
def p_inst_1(t):
    'inst : ID DBLCOLON ID LPAREN arg_list RPAREN'
    try:
        format = formatMap[t[1]]
    except KeyError:
        error(t.lexer.lineno, 'instruction format "%s" not defined.' % t[1])
    codeObj = format.defineInst(t[3], t[5], t.lexer.lineno)
    comment = '\n// %s::%s(%s)\n' % (t[1], t[3], t[5])
    codeObj.prepend_all(comment)
    t[0] = codeObj

# The arg list generates a tuple, where the first element is a list of
# the positional args and the second element is a dict containing the
# keyword args.
def p_arg_list_0(t):
    'arg_list : positional_arg_list COMMA keyword_arg_list'
    t[0] = ( t[1], t[3] )

def p_arg_list_1(t):
    'arg_list : positional_arg_list'
    t[0] = ( t[1], {} )

def p_arg_list_2(t):
    'arg_list : keyword_arg_list'
    t[0] = ( [], t[1] )

def p_positional_arg_list_0(t):
    'positional_arg_list : empty'
    t[0] = []

def p_positional_arg_list_1(t):
    'positional_arg_list : expr'
    t[0] = [t[1]]

def p_positional_arg_list_2(t):
    'positional_arg_list : positional_arg_list COMMA expr'
    t[0] = t[1] + [t[3]]

def p_keyword_arg_list_0(t):
    'keyword_arg_list : keyword_arg'
    t[0] = t[1]

def p_keyword_arg_list_1(t):
    'keyword_arg_list : keyword_arg_list COMMA keyword_arg'
    t[0] = t[1]
    t[0].update(t[3])

def p_keyword_arg(t):
    'keyword_arg : ID EQUALS expr'
    t[0] = { t[1] : t[3] }

#
# Basic expressions.  These constitute the argument values of
# "function calls" (i.e. instruction definitions in the decode block)
# and default values for formal parameters of format functions.
#
# Right now, these are either strings, integers, or (recursively)
# lists of exprs (using Python square-bracket list syntax).  Note that
# bare identifiers are trated as string constants here (since there
# isn't really a variable namespace to refer to).
#
def p_expr_0(t):
    '''expr : ID
            | INTLIT
            | STRLIT
            | CODELIT'''
    t[0] = t[1]

def p_expr_1(t):
    '''expr : LBRACKET list_expr RBRACKET'''
    t[0] = t[2]

def p_list_expr_0(t):
    'list_expr : expr'
    t[0] = [t[1]]

def p_list_expr_1(t):
    'list_expr : list_expr COMMA expr'
    t[0] = t[1] + [t[3]]

def p_list_expr_2(t):
    'list_expr : empty'
    t[0] = []

#
# Empty production... use in other rules for readability.
#
def p_empty(t):
    'empty :'
    pass

# Parse error handler.  Note that the argument here is the offending
# *token*, not a grammar symbol (hence the need to use t.value)
def p_error(t):
    if t:
        error(t.lexer.lineno, "syntax error at '%s'" % t.value)
    else:
        error(0, "unknown syntax error", True)

# END OF GRAMMAR RULES
#
# Now build the parser.
parser = yacc.yacc()


#####################################################################
#
#                           Support Classes
#
#####################################################################

# Expand template with CPU-specific references into a dictionary with
# an entry for each CPU model name.  The entry key is the model name
# and the corresponding value is the template with the CPU-specific
# refs substituted for that model.
def expand_cpu_symbols_to_dict(template):
    # Protect '%'s that don't go with CPU-specific terms
    t = re.sub(r'%(?!\(CPU_)', '%%', template)
    result = {}