+ # This tricky setup (immediately above) allows us to just write
+ # (e.g.) "split('exec')" in the Python code and the split #ifdef's
+ # will automatically be added to the exec_output variable. The inner
+ # Python execution environment doesn't know about the split points,
+ # so we carefully inject and wrap a closure that can retrieve the
+ # next split's #define from the parser and add it to the current
+ # emission-in-progress.
+ try:
+ exec split_setup+fixPythonIndentation(t[2]) in self.exportContext
+ except Exception, exc:
+ if debug:
+ raise
+ error(t.lineno(1), 'In global let block: %s' % exc)
+ GenCode(self,
+ header_output=self.exportContext["header_output"],
+ decoder_output=self.exportContext["decoder_output"],
+ exec_output=self.exportContext["exec_output"],
+ decode_block=self.exportContext["decode_block"]).emit()
+
+ # Define the mapping from operand type extensions to C++ types and
+ # bit widths (stored in operandTypeMap).
+ def p_def_operand_types(self, t):
+ 'def_operand_types : DEF OPERAND_TYPES CODELIT SEMI'
+ try:
+ self.operandTypeMap = eval('{' + t[3] + '}')
+ except Exception, exc:
+ if debug:
+ raise
+ error(t.lineno(1),
+ 'In def operand_types: %s' % exc)
+
+ # Define the mapping from operand names to operand classes and
+ # other traits. Stored in operandNameMap.
+ def p_def_operands(self, t):
+ 'def_operands : DEF OPERANDS CODELIT SEMI'
+ if not hasattr(self, 'operandTypeMap'):
+ error(t.lineno(1),
+ 'error: operand types must be defined before operands')
+ try:
+ user_dict = eval('{' + t[3] + '}', self.exportContext)
+ except Exception, exc:
+ if debug:
+ raise
+ error(t.lineno(1), 'In def operands: %s' % exc)
+ self.buildOperandNameMap(user_dict, t.lexer.lineno)
+
+ # A bitfield definition looks like:
+ # 'def [signed] bitfield <ID> [<first>:<last>]'
+ # This generates a preprocessor macro in the output file.
+ def p_def_bitfield_0(self, t):
+ 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT COLON INTLIT GREATER SEMI'
+ expr = 'bits(machInst, %2d, %2d)' % (t[6], t[8])
+ if (t[2] == 'signed'):
+ expr = 'sext<%d>(%s)' % (t[6] - t[8] + 1, expr)
+ hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr)
+ GenCode(self, header_output=hash_define).emit()
+
+ # alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]'
+ def p_def_bitfield_1(self, t):
+ 'def_bitfield : DEF opt_signed BITFIELD ID LESS INTLIT GREATER SEMI'
+ expr = 'bits(machInst, %2d, %2d)' % (t[6], t[6])
+ if (t[2] == 'signed'):
+ expr = 'sext<%d>(%s)' % (1, expr)
+ hash_define = '#undef %s\n#define %s\t%s\n' % (t[4], t[4], expr)
+ GenCode(self, header_output=hash_define).emit()
+
+ # alternate form for structure member: 'def bitfield <ID> <ID>'
+ def p_def_bitfield_struct(self, t):
+ 'def_bitfield_struct : DEF opt_signed BITFIELD ID id_with_dot SEMI'
+ if (t[2] != ''):
+ error(t.lineno(1),
+ '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)
+ GenCode(self, header_output=hash_define).emit()
+
+ def p_id_with_dot_0(self, t):
+ 'id_with_dot : ID'
+ t[0] = t[1]
+
+ def p_id_with_dot_1(self, t):
+ 'id_with_dot : ID DOT id_with_dot'
+ t[0] = t[1] + t[2] + t[3]
+
+ def p_opt_signed_0(self, t):
+ 'opt_signed : SIGNED'
+ t[0] = t[1]
+
+ def p_opt_signed_1(self, t):
+ 'opt_signed : empty'
+ t[0] = ''
+
+ def p_def_template(self, t):
+ 'def_template : DEF TEMPLATE ID CODELIT SEMI'
+ if t[3] in self.templateMap:
+ print "warning: template %s already defined" % t[3]
+ self.templateMap[t[3]] = Template(self, t[4])
+
+ # An instruction format definition looks like
+ # "def format <fmt>(<params>) {{...}};"
+ def p_def_format(self, t):
+ 'def_format : DEF FORMAT ID LPAREN param_list RPAREN CODELIT SEMI'
+ (id, params, code) = (t[3], t[5], t[7])
+ self.defFormat(id, params, code, t.lexer.lineno)
+
+ # 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(self, t):
+ 'param_list : positional_param_list COMMA nonpositional_param_list'
+ t[0] = t[1] + t[3]
+
+ def p_param_list_1(self, t):
+ '''param_list : positional_param_list
+ | nonpositional_param_list'''
+ t[0] = t[1]
+
+ def p_positional_param_list_0(self, t):
+ 'positional_param_list : empty'
+ t[0] = []
+
+ def p_positional_param_list_1(self, t):
+ 'positional_param_list : ID'
+ t[0] = [t[1]]
+
+ def p_positional_param_list_2(self, t):
+ 'positional_param_list : positional_param_list COMMA ID'
+ t[0] = t[1] + [t[3]]
+
+ def p_nonpositional_param_list_0(self, t):
+ 'nonpositional_param_list : keyword_param_list COMMA excess_args_param'
+ t[0] = t[1] + t[3]
+
+ def p_nonpositional_param_list_1(self, t):
+ '''nonpositional_param_list : keyword_param_list
+ | excess_args_param'''
+ t[0] = t[1]
+
+ def p_keyword_param_list_0(self, t):
+ 'keyword_param_list : keyword_param'
+ t[0] = [t[1]]
+
+ def p_keyword_param_list_1(self, t):
+ 'keyword_param_list : keyword_param_list COMMA keyword_param'
+ t[0] = t[1] + [t[3]]
+
+ def p_keyword_param(self, t):
+ 'keyword_param : ID EQUALS expr'
+ t[0] = t[1] + ' = ' + t[3].__repr__()
+
+ def p_excess_args_param(self, 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_top_level_decode_block(self, t):
+ 'top_level_decode_block : decode_block'
+ codeObj = t[1]
+ codeObj.wrap_decode_block('''
+StaticInstPtr
+%(isa_name)s::Decoder::decodeInst(%(isa_name)s::ExtMachInst machInst)
+{
+ using namespace %(namespace)s;
+''' % self, '}')
+
+ codeObj.emit()
+
+ def p_decode_block(self, t):
+ 'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE'
+ default_defaults = self.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(self, t):
+ 'opt_default : empty'
+ # no default specified: reuse the one currently at the top of
+ # the stack
+ self.defaultStack.push(self.defaultStack.top())
+ # no meaningful value returned
+ t[0] = None
+
+ def p_opt_default_1(self, t):
+ 'opt_default : DEFAULT inst'
+ # push the new default
+ codeObj = t[2]
+ codeObj.wrap_decode_block('\ndefault:\n', 'break;\n')
+ self.defaultStack.push(codeObj)
+ # no meaningful value returned
+ t[0] = None
+
+ def p_decode_stmt_list_0(self, t):
+ 'decode_stmt_list : decode_stmt'
+ t[0] = t[1]
+
+ def p_decode_stmt_list_1(self, t):
+ 'decode_stmt_list : decode_stmt decode_stmt_list'
+ if (t[1].has_decode_default and t[2].has_decode_default):
+ error(t.lineno(1), '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(self, t):
+ 'decode_stmt : CPPDIRECTIVE'
+ t[0] = GenCode(self, 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(self, 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.
+ self.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(self, t):
+ 'push_format_id : ID'
+ try:
+ self.formatStack.push(self.formatMap[t[1]])
+ t[0] = ('', '// format %s' % t[1])
+ except KeyError:
+ error(t.lineno(1), 'instruction format "%s" not defined.' % t[1])
+
+ # Nested decode block: if the value of the current field matches
+ # the specified constant(s), do a nested decode on some other field.
+ def p_decode_stmt_decode(self, t):
+ 'decode_stmt : case_list COLON decode_block'
+ case_list = 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' % ''.join(case_list))
+ codeObj.has_decode_default = (case_list == ['default:'])
+ t[0] = codeObj
+
+ # Instruction definition (finally!).
+ def p_decode_stmt_inst(self, t):
+ 'decode_stmt : case_list COLON inst SEMI'
+ case_list = t[1]
+ codeObj = t[3]
+ codeObj.wrap_decode_block('\n%s' % ''.join(case_list), 'break;\n')
+ codeObj.has_decode_default = (case_list == ['default:'])
+ t[0] = codeObj
+
+ # The constant list for a decode case label must be non-empty, and must
+ # either be the keyword 'default', or made up of one or more
+ # comma-separated integer literals or strings which evaluate to
+ # constants when compiled as C++.
+ def p_case_list_0(self, t):
+ 'case_list : DEFAULT'
+ t[0] = ['default:']
+
+ def prep_int_lit_case_label(self, lit):
+ if lit >= 2**32:
+ return 'case ULL(%#x): ' % lit
+ else:
+ return 'case %#x: ' % lit
+
+ def prep_str_lit_case_label(self, lit):
+ return 'case %s: ' % lit
+
+ def p_case_list_1(self, t):
+ 'case_list : INTLIT'
+ t[0] = [self.prep_int_lit_case_label(t[1])]
+
+ def p_case_list_2(self, t):
+ 'case_list : STRLIT'
+ t[0] = [self.prep_str_lit_case_label(t[1])]
+
+ def p_case_list_3(self, t):
+ 'case_list : case_list COMMA INTLIT'
+ t[0] = t[1]
+ t[0].append(self.prep_int_lit_case_label(t[3]))
+
+ def p_case_list_4(self, t):
+ 'case_list : case_list COMMA STRLIT'
+ t[0] = t[1]
+ t[0].append(self.prep_str_lit_case_label(t[3]))
+
+ # Define an instruction using the current instruction format
+ # (specified by an enclosing format block).
+ # "<mnemonic>(<args>)"
+ def p_inst_0(self, t):
+ 'inst : ID LPAREN arg_list RPAREN'
+ # Pass the ID and arg list to the current format class to deal with.
+ currentFormat = self.formatStack.top()
+ codeObj = currentFormat.defineInst(self, 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(self, t):
+ 'inst : ID DBLCOLON ID LPAREN arg_list RPAREN'
+ try:
+ format = self.formatMap[t[1]]
+ except KeyError:
+ error(t.lineno(1), 'instruction format "%s" not defined.' % t[1])
+
+ codeObj = format.defineInst(self, 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(self, t):
+ 'arg_list : positional_arg_list COMMA keyword_arg_list'
+ t[0] = ( t[1], t[3] )
+
+ def p_arg_list_1(self, t):
+ 'arg_list : positional_arg_list'
+ t[0] = ( t[1], {} )
+
+ def p_arg_list_2(self, t):
+ 'arg_list : keyword_arg_list'
+ t[0] = ( [], t[1] )
+
+ def p_positional_arg_list_0(self, t):
+ 'positional_arg_list : empty'
+ t[0] = []
+
+ def p_positional_arg_list_1(self, t):
+ 'positional_arg_list : expr'
+ t[0] = [t[1]]
+
+ def p_positional_arg_list_2(self, t):
+ 'positional_arg_list : positional_arg_list COMMA expr'
+ t[0] = t[1] + [t[3]]
+
+ def p_keyword_arg_list_0(self, t):
+ 'keyword_arg_list : keyword_arg'
+ t[0] = t[1]
+
+ def p_keyword_arg_list_1(self, t):
+ 'keyword_arg_list : keyword_arg_list COMMA keyword_arg'
+ t[0] = t[1]
+ t[0].update(t[3])
+
+ def p_keyword_arg(self, 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(self, t):
+ '''expr : ID
+ | INTLIT
+ | STRLIT
+ | CODELIT'''
+ t[0] = t[1]
+
+ def p_expr_1(self, t):
+ '''expr : LBRACKET list_expr RBRACKET'''
+ t[0] = t[2]
+
+ def p_list_expr_0(self, t):
+ 'list_expr : expr'
+ t[0] = [t[1]]
+
+ def p_list_expr_1(self, t):
+ 'list_expr : list_expr COMMA expr'
+ t[0] = t[1] + [t[3]]
+
+ def p_list_expr_2(self, t):
+ 'list_expr : empty'
+ t[0] = []
+
+ #
+ # Empty production... use in other rules for readability.
+ #
+ def p_empty(self, 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(self, t):
+ if t:
+ error(t.lexer.lineno, "syntax error at '%s'" % t.value)
+ else:
+ error("unknown syntax error")