/src/arch/isa_parser.py

Large files files are truncated, but you can click here to view the full file

# Copyright (c) 2003-2005 The Regents of The University of Michigan
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met: redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer;
# redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution;
# neither the name of the copyright holders nor the names of its
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Authors: Steve Reinhardt

import os
import sys
import re
import string
import inspect, traceback
# get type names
from types import *

from m5.util.grammar import Grammar

debug=False

###################
# Utility functions

#
# Indent every line in string 's' by two spaces
# (except preprocessor directives).
# Used to make nested code blocks look pretty.
#
def indent(s):
    return re.sub(r'(?m)^(?!#)', '  ', s)

#
# Munge a somewhat arbitrarily formatted piece of Python code
# (e.g. from a format 'let' block) into something whose indentation
# will get by the Python parser.
#
# The two keys here are that Python will give a syntax error if
# there's any whitespace at the beginning of the first line, and that
# all lines at the same lexical nesting level must have identical
# indentation.  Unfortunately the way code literals work, an entire
# let block tends to have some initial indentation.  Rather than
# trying to figure out what that is and strip it off, we prepend 'if
# 1:' to make the let code the nested block inside the if (and have
# the parser automatically deal with the indentation for us).
#
# We don't want to do this if (1) the code block is empty or (2) the
# first line of the block doesn't have any whitespace at the front.

def fixPythonIndentation(s):
    # get rid of blank lines first
    s = re.sub(r'(?m)^\s*\n', '', s);
    if (s != '' and re.match(r'[ \t]', s[0])):
        s = 'if 1:\n' + s
    return s

class ISAParserError(Exception):
    """Error handler for parser errors"""
    def __init__(self, first, second=None):
        if second is None:
            self.lineno = 0
            self.string = first
        else:
            if hasattr(first, 'lexer'):
                first = first.lexer.lineno
            self.lineno = first
            self.string = second

    def display(self, filename_stack, print_traceback=debug):
        # Output formatted to work under Emacs compile-mode.  Optional
        # 'print_traceback' arg, if set to True, prints a Python stack
        # backtrace too (can be handy when trying to debug the parser
        # itself).

        spaces = ""
        for (filename, line) in filename_stack[:-1]:
            print "%sIn file included from %s:" % (spaces, filename)
            spaces += "  "

        # Print a Python stack backtrace if requested.
        if print_traceback or not self.lineno:
            traceback.print_exc()

        line_str = "%s:" % (filename_stack[-1][0], )
        if self.lineno:
            line_str += "%d:" % (self.lineno, )

        return "%s%s %s" % (spaces, line_str, self.string)

    def exit(self, filename_stack, print_traceback=debug):
        # Just call exit.

        sys.exit(self.display(filename_stack, print_traceback))

def error(*args):
    raise ISAParserError(*args)

####################
# Template objects.
#
# Template objects are format strings that allow substitution from
# the attribute spaces of other objects (e.g. InstObjParams instances).

labelRE = re.compile(r'(?<!%)%\(([^\)]+)\)[sd]')

class Template(object):
    def __init__(self, parser, t):
        self.parser = parser
        self.template = t

    def subst(self, d):
        myDict = None

        # Protect non-Python-dict substitutions (e.g. if there's a printf
        # in the templated C++ code)
        template = self.parser.protectNonSubstPercents(self.template)
        # CPU-model-specific substitutions are handled later (in GenCode).
        template = self.parser.protectCpuSymbols(template)

        # Build a dict ('myDict') to use for the template substitution.
        # Start with the template namespace.  Make a copy since we're
        # going to modify it.
        myDict = self.parser.templateMap.copy()

        if isinstance(d, InstObjParams):
            # If we're dealing with an InstObjParams object, we need
            # to be a little more sophisticated.  The instruction-wide
            # parameters are already formed, but the parameters which
            # are only function wide still need to be generated.
            compositeCode = ''

            myDict.update(d.__dict__)
            # The "operands" and "snippets" attributes of the InstObjParams
            # objects are for internal use and not substitution.
            del myDict['operands']
            del myDict['snippets']

            snippetLabels = [l for l in labelRE.findall(template)
                             if d.snippets.has_key(l)]

            snippets = dict([(s, self.parser.mungeSnippet(d.snippets[s]))
                             for s in snippetLabels])

            myDict.update(snippets)

            compositeCode = ' '.join(map(str, snippets.values()))

            # Add in template itself in case it references any
            # operands explicitly (like Mem)
            compositeCode += ' ' + template

            operands = SubOperandList(self.parser, compositeCode, d.operands)

            myDict['op_decl'] = operands.concatAttrStrings('op_decl')
            if operands.readPC or operands.setPC:
                myDict['op_decl'] += 'TheISA::PCState __parserAutoPCState;\n'

            # In case there are predicated register reads and write, declare
            # the variables for register indicies. It is being assumed that
            # all the operands in the OperandList are also in the
            # SubOperandList and in the same order. Otherwise, it is
            # expected that predication would not be used for the operands.
            if operands.predRead:
                myDict['op_decl'] += 'uint8_t _sourceIndex = 0;\n'
            if operands.predWrite:
                myDict['op_decl'] += 'uint8_t M5_VAR_USED _destIndex = 0;\n'

            is_src = lambda op: op.is_src
            is_dest = lambda op: op.is_dest

            myDict['op_src_decl'] = \
                      operands.concatSomeAttrStrings(is_src, 'op_src_decl')
            myDict['op_dest_decl'] = \
                      operands.concatSomeAttrStrings(is_dest, 'op_dest_decl')
            if operands.readPC:
                myDict['op_src_decl'] += \
                    'TheISA::PCState __parserAutoPCState;\n'
            if operands.setPC:
                myDict['op_dest_decl'] += \
                    'TheISA::PCState __parserAutoPCState;\n'

            myDict['op_rd'] = operands.concatAttrStrings('op_rd')
            if operands.readPC:
                myDict['op_rd'] = '__parserAutoPCState = xc->pcState();\n' + \
                                  myDict['op_rd']

            # Compose the op_wb string. If we're going to write back the
            # PC state because we changed some of its elements, we'll need to
            # do that as early as possible. That allows later uncoordinated
            # modifications to the PC to layer appropriately.
            reordered = list(operands.items)
            reordered.reverse()
            op_wb_str = ''
            pcWbStr = 'xc->pcState(__parserAutoPCState);\n'
            for op_desc in reordered:
                if op_desc.isPCPart() and op_desc.is_dest:
                    op_wb_str = op_desc.op_wb + pcWbStr + op_wb_str
                    pcWbStr = ''
                else:
                    op_wb_str = op_desc.op_wb + op_wb_str
            myDict['op_wb'] = op_wb_str

        elif isinstance(d, dict):
            # if the argument is a dictionary, we just use it.
            myDict.update(d)
        elif hasattr(d, '__dict__'):
            # if the argument is an object, we use its attribute map.
            myDict.update(d.__dict__)
        else:
            raise TypeError, "Template.subst() arg must be or have dictionary"
        return template % myDict

    # Convert to string.  This handles the case when a template with a
    # CPU-specific term gets interpolated into another template or into
    # an output block.
    def __str__(self):
        return self.parser.expandCpuSymbolsToString(self.template)

################
# Format object.
#
# A format object encapsulates an instruction format.  It must provide
# a defineInst() method that generates the code for an instruction
# definition.

class Format(object):
    def __init__(self, id, params, code):
        self.id = id
        self.params = params
        label = 'def format ' + id
        self.user_code = compile(fixPythonIndentation(code), label, 'exec')
        param_list = string.join(params, ", ")
        f = '''def defInst(_code, _context, %s):
                my_locals = vars().copy()
                exec _code in _context, my_locals
                return my_locals\n''' % param_list
        c = compile(f, label + ' wrapper', 'exec')
        exec c
        self.func = defInst

    def defineInst(self, parser, name, args, lineno):
        parser.updateExportContext()
        context = parser.exportContext.copy()
        if len(name):
            Name = name[0].upper()
            if len(name) > 1:
                Name += name[1:]
        context.update({ 'name' : name, 'Name' : Name })
        try:
            vars = self.func(self.user_code, context, *args[0], **args[1])
        except Exception, exc:
            if debug:
                raise
            error(lineno, 'error defining "%s": %s.' % (name, exc))
        for k in vars.keys():
            if k not in ('header_output', 'decoder_output',
                         'exec_output', 'decode_block'):
                del vars[k]
        return GenCode(parser, **vars)

# Special null format to catch an implicit-format instruction
# definition outside of any format block.
class NoFormat(object):
    def __init__(self):
        self.defaultInst = ''

    def defineInst(self, parser, name, args, lineno):
        error(lineno,
              'instruction definition "%s" with no active format!' % name)

###############
# GenCode class
#
# The GenCode class encapsulates generated code destined for various
# output files.  The header_output and decoder_output attributes are
# strings containing code destined for decoder.hh and decoder.cc
# respectively.  The decode_block attribute contains code to be
# incorporated in the decode function itself (that will also end up in
# decoder.cc).  The exec_output attribute is a dictionary with a key
# for each CPU model name; the value associated with a particular key
# is the string of code for that CPU model's exec.cc file.  The
# has_decode_default attribute is used in the decode block to allow
# explicit default clauses to override default default clauses.

class GenCode(object):
    # Constructor.  At this point we substitute out all CPU-specific
    # symbols.  For the exec output, these go into the per-model
    # dictionary.  For all other output types they get collapsed into
    # a single string.
    def __init__(self, parser,
                 header_output = '', decoder_output = '', exec_output = '',
                 decode_block = '', has_decode_default = False):
        self.parser = parser
        self.header_output = parser.expandCpuSymbolsToString(header_output)
        self.decoder_output = parser.expandCpuSymbolsToString(decoder_output)
        if isinstance(exec_output, dict):
            self.exec_output = exec_output
        elif isinstance(exec_output, str):
            # If the exec_output arg is a single string, we replicate
            # it for each of the CPU models, substituting and
            # %(CPU_foo)s params appropriately.
            self.exec_output = parser.expandCpuSymbolsToDict(exec_output)
        self.decode_block = parser.expandCpuSymbolsToString(decode_block)
        self.has_decode_default = has_decode_default

    # Override '+' operator: generate a new GenCode object that
    # concatenates all the individual strings in the operands.
    def __add__(self, other):
        exec_output = {}
        for cpu in self.parser.cpuModels:
            n = cpu.name
            exec_output[n] = self.exec_output[n] + other.exec_output[n]
        return GenCode(self.parser,
                       self.header_output + other.header_output,
                       self.decoder_output + other.decoder_output,
                       exec_output,
                       self.decode_block + other.decode_block,
                       self.has_decode_default or other.has_decode_default)

    # Prepend a string (typically a comment) to all the strings.
    def prepend_all(self, pre):
        self.header_output = pre + self.header_output
        self.decoder_output  = pre + self.decoder_output
        self.decode_block = pre + self.decode_block
        for cpu in self.parser.cpuModels:
            self.exec_output[cpu.name] = pre + self.exec_output[cpu.name]

    # Wrap the decode block in a pair of strings (e.g., 'case foo:'
    # and 'break;').  Used to build the big nested switch statement.
    def wrap_decode_block(self, pre, post = ''):
        self.decode_block = pre + indent(self.decode_block) + post

#####################################################################
#
#                      Bitfield Operator Support
#
#####################################################################

bitOp1ArgRE = re.compile(r'<\s*(\w+)\s*:\s*>')

bitOpWordRE = re.compile(r'(?<![\w\.])([\w\.]+)<\s*(\w+)\s*:\s*(\w+)\s*>')
bitOpExprRE = re.compile(r'\)<\s*(\w+)\s*:\s*(\w+)\s*>')

def substBitOps(code):
    # first convert single-bit selectors to two-index form
    # i.e., <n> --> <n:n>
    code = bitOp1ArgRE.sub(r'<\1:\1>', code)
    # simple case: selector applied to ID (name)
    # i.e., foo<a:b> --> bits(foo, a, b)
    code = bitOpWordRE.sub(r'bits(\1, \2, \3)', code)
    # if selector is applied to expression (ending in ')'),
    # we need to search backward for matching '('
    match = bitOpExprRE.search(code)
    while match:
        exprEnd = match.start()
        here = exprEnd - 1
        nestLevel = 1
        while nestLevel > 0:
            if code[here] == '(':
                nestLevel -= 1
            elif code[here] == ')':
                nestLevel += 1
            here -= 1
            if here < 0:
                sys.exit("Didn't find '('!")
        exprStart = here+1
        newExpr = r'bits(%s, %s, %s)' % (code[exprStart:exprEnd+1],
                                         match.group(1), match.group(2))
        code = code[:exprStart] + newExpr + code[match.end():]
        match = bitOpExprRE.search(code)
    return code


#####################################################################
#
#                             Code Parser
#
# The remaining code is the support for automatically extracting
# instruction characteristics from pseudocode.
#
#####################################################################

# Force the argument to be a list.  Useful for flags, where a caller
# can specify a singleton flag or a list of flags.  Also usful for
# converting tuples to lists so they can be modified.
def makeList(arg):
    if isinstance(arg, list):
        return arg
    elif isinstance(arg, tuple):
        return list(arg)
    elif not arg:
        return []
    else:
        return [ arg ]

class Operand(object):
    '''Base class for operand descriptors.  An instance of this class
    (or actually a class derived from this one) represents a specific
    operand for a code block (e.g, "Rc.sq" as a dest). Intermediate
    derived classes encapsulates the traits of a particular operand
    type (e.g., "32-bit integer register").'''

    def buildReadCode(self, func = None):
        subst_dict = {"name": self.base_name,
                      "func": func,
                      "reg_idx": self.reg_spec,
                      "ctype": self.ctype}
        if hasattr(self, 'src_reg_idx'):
            subst_dict['op_idx'] = self.src_reg_idx
        code = self.read_code % subst_dict
        return '%s = %s;\n' % (self.base_name, code)

    def buildWriteCode(self, func = None):
        subst_dict = {"name": self.base_name,
                      "func": func,
                      "reg_idx": self.reg_spec,
                      "ctype": self.ctype,
                      "final_val": self.base_name}
        if hasattr(self, 'dest_reg_idx'):
            subst_dict['op_idx'] = self.dest_reg_idx
        code = self.write_code % subst_dict
        return '''
        {
            %s final_val = %s;
            %s;
            if (traceData) { traceData->setData(final_val); }
        }''' % (self.dflt_ctype, self.base_name, code)

    def __init__(self, parser, full_name, ext, is_src, is_dest):
        self.full_name = full_name
        self.ext = ext
        self.is_src = is_src
        self.is_dest = is_dest
        # The 'effective extension' (eff_ext) is either the actual
        # extension, if one was explicitly provided, or the default.
        if ext:
            self.eff_ext = ext
        elif hasattr(self, 'dflt_ext'):
            self.eff_ext = self.dflt_ext

        if hasattr(self, 'eff_ext'):
            self.ctype = parser.operandTypeMap[self.eff_ext]

    # Finalize additional fields (primarily code fields).  This step
    # is done separately since some of these fields may depend on the
    # register index enumeration that hasn't been performed yet at the
    # time of __init__(). The register index enumeration is affected
    # by predicated register reads/writes. Hence, we forward the flags
    # that indicate whether or not predication is in use.
    def finalize(self, predRead, predWrite):
        self.flags = self.getFlags()
        self.constructor = self.makeConstructor(predRead, predWrite)
        self.op_decl = self.makeDecl()

        if self.is_src:
            self.op_rd = self.makeRead(predRead)
            self.op_src_decl = self.makeDecl()
        else:
            self.op_rd = ''
            self.op_src_decl = ''

        if self.is_dest:
            self.op_wb = self.makeWrite(predWrite)
            self.op_dest_decl = self.makeDecl()
        else:
            self.op_wb = ''
            self.op_dest_decl = ''

    def isMem(self):
        return 0

    def isReg(self):
        return 0

    def isFloatReg(self):
        return 0

    def isIntReg(self):
        return 0

    def isControlReg(self):
        return 0

    def isPCState(self):
        return 0

    def isPCPart(self):
        return self.isPCState() and self.reg_spec

    def hasReadPred(self):
        return self.read_predicate != None

    def hasWritePred(self):
        return self.write_predicate != None

    def getFlags(self):
        # note the empty slice '[:]' gives us a copy of self.flags[0]
        # instead of a reference to it
        my_flags = self.flags[0][:]
        if self.is_src:
            my_flags += self.flags[1]
        if self.is_dest:
            my_flags += self.flags[2]
        return my_flags

    def makeDecl(self):
        # Note that initializations in the declarations are solely
        # to avoid 'uninitialized variable' errors from the compiler.
        return self.ctype + ' ' + self.base_name + ' = 0;\n';

class IntRegOperand(Operand):
    def isReg(self):
        return 1

    def isIntReg(self):
        return 1

    def makeConstructor(self, predRead, predWrite):
        c_src = ''
        c_dest = ''

        if self.is_src:
            c_src = '\n\t_srcRegIdx[_numSrcRegs++] = %s;' % (self.reg_spec)
            if self.hasReadPred():
                c_src = '\n\tif (%s) {%s\n\t}' % \
                        (self.read_predicate, c_src)

        if self.is_dest:
            c_dest = '\n\t_destRegIdx[_numDestRegs++] = %s;' % \
                    (self.reg_spec)
            c_dest += '\n\t_numIntDestRegs++;'
            if self.hasWritePred():
                c_dest = '\n\tif (%s) {%s\n\t}' % \
                         (self.write_predicate, c_dest)

        return c_src + c_dest

    def makeRead(self, predRead):
        if (self.ctype == 'float' or self.ctype == 'double'):
            error('Attempt to read integer register as FP')
        if self.read_code != None:
            return self.buildReadCode('readIntRegOperand')

        int_reg_val = ''
        if predRead:
            int_reg_val = 'xc->readIntRegOperand(this, _sourceIndex++)'
            if self.hasReadPred():
                int_reg_val = '(%s) ? %s : 0' % \
                              (self.read_predicate, int_reg_val)
        else:
            int_reg_val = 'xc->readIntRegOperand(this, %d)' % self.src_reg_idx

        return '%s = %s;\n' % (self.base_name, int_reg_val)

    def makeWrite(self, predWrite):
        if (self.ctype == 'float' or self.ctype == 'double'):
            error('Attempt to write integer register as FP')
        if self.write_code != None:
            return self.buildWriteCode('setIntRegOperand')

        if predWrite:
            wp = 'true'
            if self.hasWritePred():
                wp = self.write_predicate

            wcond = 'if (%s)' % (wp)
            windex = '_destIndex++'
        else:
            wcond = ''
            windex = '%d' % self.dest_reg_idx

        wb = '''
        %s
        {
            %s final_val = %s;
            xc->setIntRegOperand(this, %s, final_val);\n
            if (traceData) { traceData->setData(final_val); }
        }''' % (wcond, self.ctype, self.base_name, windex)

        return wb

class FloatRegOperand(Operand):
    def isReg(self):
        return 1

    def isFloatReg(self):
        return 1

    def makeConstructor(self, predRead, predWrite):
        c_src = ''
        c_dest = ''

        if self.is_src:
            c_src = '\n\t_srcRegIdx[_numSrcRegs++] = %s + FP_Base_DepTag;' % \
                    (self.reg_spec)

        if self.is_dest:
            c_dest = \
              '\n\t_destRegIdx[_numDestRegs++] = %s + FP_Base_DepTag;' % \
              (self.reg_spec)
            c_dest += '\n\t_numFPDestRegs++;'

        return c_src + c_dest

    def makeRead(self, predRead):
        bit_select = 0
        if (self.ctype == 'float' or self.ctype == 'double'):
            func = 'readFloatRegOperand'
        else:
            func = 'readFloatRegOperandBits'
        if self.read_code != None:
            return self.buildReadCode(func)

        if predRead:
            rindex = '_sourceIndex++'
        else:
            rindex = '%d' % self.src_reg_idx

        return '%s = xc->%s(this, %s);\n' % \
            (self.base_name, func, rindex)

    def makeWrite(self, predWrite):
        if (self.ctype == 'float' or self.ctype == 'double'):
            func = 'setFloatRegOperand'
        else:
            func = 'setFloatRegOperandBits'
        if self.write_code != None:
            return self.buildWriteCode(func)

        if predWrite:
            wp = '_destIndex++'
        else:
            wp = '%d' % self.dest_reg_idx
        wp = 'xc->%s(this, %s, final_val);' % (func, wp)

        wb = '''
        {
            %s final_val = %s;
            %s\n
            if (traceData) { traceData->setData(final_val); }
        }''' % (self.ctype, self.base_name, wp)
        return wb

class ControlRegOperand(Operand):
    def isReg(self):
        return 1

    def isControlReg(self):
        return 1

    def makeConstructor(self, predRead, predWrite):
        c_src = ''
        c_dest = ''

        if self.is_src:
            c_src = \
              '\n\t_srcRegIdx[_numSrcRegs++] = %s + Ctrl_Base_DepTag;' % \
              (self.reg_spec)

        if self.is_dest:
            c_dest = \
              '\n\t_destRegIdx[_numDestRegs++] = %s + Ctrl_Base_DepTag;' % \
              (self.reg_spec)

        return c_src + c_dest

    def makeRead(self, predRead):
        bit_select = 0
        if (self.ctype == 'float' or self.ctype == 'double'):
            error('Attempt to read control register as FP')
        if self.read_code != None:
            return self.buildReadCode('readMiscRegOperand')

        if predRead:
            rindex = '_sourceIndex++'
        else:
            rindex = '%d' % self.src_reg_idx

        return '%s = xc->readMiscRegOperand(this, %s);\n' % \
            (self.base_name, rindex)

    def makeWrite(self, predWrite):
        if (self.ctype == 'float' or self.ctype == 'double'):
            error('Attempt to write control register as FP')
        if self.write_code != None:
            return self.buildWriteCode('setMiscRegOperand')

        if predWrite:
            windex = '_destIndex++'
        else:
            windex = '%d' % self.dest_reg_idx

        wb = 'xc->setMiscRegOperand(this, %s, %s);\n' % \
             (windex, self.base_name)
        wb += 'if (traceData) { traceData->setData(%s); }' % \
              self.base_name

        return wb

class MemOperand(Operand):
    def isMem(self):
        return 1

    def makeConstructor(self, predRead, predWrite):
        return ''

    def makeDecl(self):
        # Note that initializations in the declarations are solely
        # to avoid 'uninitialized variable' errors from the compiler.
        # Declare memory data variable.
        return '%s %s = 0;\n' % (self.ctype, self.base_name)

    def makeRead(self, predRead):
        if self.read_code != None:
            return self.buildReadCode()
        return ''

    def makeWrite(self, predWrite):
        if self.write_code != None:
            return self.buildWriteCode()
        return ''

class PCStateOperand(Operand):
    def makeConstructor(self, predRead, predWrite):
        return ''

    def makeRead(self, predRead):
        if self.reg_spec:
            # A component of the PC state.
            return '%s = __parserAutoPCState.%s();\n' % \
                (self.base_name, self.reg_spec)
        else:
            # The whole PC state itself.
            return '%s = xc->pcState();\n' % self.base_name

    def makeWrite(self, predWrite):
        if self.reg_spec:
            # A component of the PC state.
            return '__parserAutoPCState.%s(%s);\n' % \
                (self.reg_spec, self.base_name)
        else:
            # The whole PC state itself.
            return 'xc->pcState(%s);\n' % self.base_name

    def makeDecl(self):
        ctype = 'TheISA::PCState'
        if self.isPCPart():
            ctype = self.ctype
        return "%s %s;\n" % (ctype, self.base_name)

    def isPCState(self):
        return 1

class OperandList(object):
    '''Find all the operands in the given code block.  Returns an operand
    descriptor list (instance of class OperandList).'''
    def __init__(self, parser, code):
        self.items = []
        self.bases = {}
        # delete strings and comments so we don't match on operands inside
        for regEx in (stringRE, commentRE):
            code = regEx.sub('', code)
        # search for operands
        next_pos = 0
        while 1:
            match = parser.operandsRE.search(code, next_pos)
            if not match:
                # no more matches: we're done
                break
            op = match.groups()
            # regexp groups are operand full name, base, and extension
            (op_full, op_base, op_ext) = op
            # if the token following the operand is an assignment, this is
            # a destination (LHS), else it's a source (RHS)
            is_dest = (assignRE.match(code, match.end()) != None)
            is_src = not is_dest
            # see if we've already seen this one
            op_desc = self.find_base(op_base)
            if op_desc:
                if op_desc.ext != op_ext:
                    error('Inconsistent extensions for operand %s' % \
                          op_base)
                op_desc.is_src = op_desc.is_src or is_src
                op_desc.is_dest = op_desc.is_dest or is_dest
            else:
                # new operand: create new descriptor
                op_desc = parser.operandNameMap[op_base](parser,
                    op_full, op_ext, is_src, is_dest)
                self.append(op_desc)
            # start next search after end of current match
            next_pos = match.end()
        self.sort()
        # enumerate source & dest register operands... used in building
        # constructor later
        self.numSrcRegs = 0
        self.numDestRegs = 0
        self.numFPDestRegs = 0
        self.numIntDestRegs = 0
        self.numMiscDestRegs = 0
        self.memOperand = None

        # Flags to keep track if one or more operands are to be read/written
        # conditionally.
        self.predRead = False
        self.predWrite = False

        for op_desc in self.items:
            if op_desc.isReg():
                if op_desc.is_src:
                    op_desc.src_reg_idx = self.numSrcRegs
                    self.numSrcRegs += 1
                if op_desc.is_dest:
                    op_desc.dest_reg_idx = self.numDestRegs
                    self.numDestRegs += 1
                    if op_desc.isFloatReg():
                        self.numFPDestRegs += 1
                    elif op_desc.isIntReg():
                        self.numIntDestRegs += 1
                    elif op_desc.isControlReg():
                        self.numMiscDestRegs += 1
            elif op_desc.isMem():
                if self.memOperand:
                    error("Code block has more than one memory operand.")
                self.memOperand = op_desc

            # Check if this operand has read/write predication. If true, then
            # the microop will dynamically index source/dest registers.
            self.predRead = self.predRead or op_desc.hasReadPred()
            self.predWrite = self.predWrite or op_desc.hasWritePred()

        if parser.maxInstSrcRegs < self.numSrcRegs:
            parser.maxInstSrcRegs = self.numSrcRegs
        if parser.maxInstDestRegs < self.numDestRegs:
            parser.maxInstDestRegs = self.numDestRegs
        if parser.maxMiscDestRegs < self.numMiscDestRegs:
            parser.maxMiscDestRegs = self.numMiscDestRegs

        # now make a final pass to finalize op_desc fields that may depend
        # on the register enumeration
        for op_desc in self.items:
            op_desc.finalize(self.predRead, self.predWrite)

    def __len__(self):
        return len(self.items)

    def __getitem__(self, index):
        return self.items[index]

    def append(self, op_desc):
        self.items.append(op_desc)
        self.bases[op_desc.base_name] = op_desc

    def find_base(self, base_name):
        # like self.bases[base_name], but returns None if not found
        # (rather than raising exception)
        return self.bases.get(base_name)

    # internal helper function for concat[Some]Attr{Strings|Lists}
    def __internalConcatAttrs(self, attr_name, filter, result):
        for op_desc in self.items:
            if filter(op_desc):
                result += getattr(op_desc, attr_name)
        return result

    # return a single string that is the concatenation of the (string)
    # values of the specified attribute for all operands
    def concatAttrStrings(self, attr_name):
        return self.__internalConcatAttrs(attr_name, lambda x: 1, '')

    # like concatAttrStrings, but only include the values for the operands
    # for which the provided filter function returns true
    def concatSomeAttrStrings(self, filter, attr_name):
        return self.__internalConcatAttrs(attr_name, filter, '')

    # return a single list that is the concatenation of the (list)
    # values of the specified attribute for all operands
    def concatAttrLists(self, attr_name):
        return self.__internalConcatAttrs(attr_name, lambda x: 1, [])

    # like concatAttrLists, but only include the values for the operands
    # for which the provided filter function returns true
    def concatSomeAttrLists(self, filter, attr_name):
        return self.__internalConcatAttrs(attr_name, filter, [])

    def sort(self):
        self.items.sort(lambda a, b: a.sort_pri - b.sort_pri)

class SubOperandList(OperandList):
    '''Find all the operands in the given code block.  Returns an operand
    descriptor list (instance of class OperandList).'''
    def __init__(self, parser, code, master_list):
        self.items = []
        self.bases = {}
        # delete strings and comments so we don't match on operands inside
        for regEx in (stringRE, commentRE):
            code = regEx.sub('', code)
        # search for operands
        next_pos = 0
        while 1:
            match = parser.operandsRE.search(code, next_pos)
            if not match:
                # no more matches: we're done
                break
            op = match.groups()
            # regexp groups are operand full name, base, and extension
            (op_full, op_base, op_ext) = op
            # find this op in the master list
            op_desc = master_list.find_base(op_base)
            if not op_desc:
                error('Found operand %s which is not in the master list!' \
                      ' This is an internal error' % op_base)
            else:
                # See if we've already found this operand
                op_desc = self.find_base(op_base)
                if not op_desc:
                    # if not, add a reference to it to this sub list
                    self.append(master_list.bases[op_base])

            # start next search after end of current match
            next_pos = match.end()
        self.sort()
        self.memOperand = None
        # Whether the whole PC needs to be read so parts of it can be accessed
        self.readPC = False
        # Whether the whole PC needs to be written after parts of it were
        # changed
        self.setPC = False
        # Whether this instruction manipulates the whole PC or parts of it.
        # Mixing the two is a bad idea and flagged as an error.
        self.pcPart = None

        # Flags to keep track if one or more operands are to be read/written
        # conditionally.
        self.predRead = False
        self.predWrite = False

        for op_desc in self.items:
            if op_desc.isPCPart():
                self.readPC = True
                if op_desc.is_dest:
                    self.setPC = True

            if op_desc.isPCState():
                if self.pcPart is not None:
                    if self.pcPart and not op_desc.isPCPart() or \
                            not self.pcPart and op_desc.isPCPart():
                        error("Mixed whole and partial PC state operands.")
                self.pcPart = op_desc.isPCPart()

            if op_desc.isMem():
                if self.memOperand:
                    error("Code block has more than one memory operand.")
                self.memOperand = op_desc

            # Check if this operand has read/write predication. If true, then
            # the microop will dynamically index source/dest registers.
            self.predRead = self.predRead or op_desc.hasReadPred()
            self.predWrite = self.predWrite or op_desc.hasWritePred()

# Regular expression object to match C++ strings
stringRE = re.compile(r'"([^"\\]|\\.)*"')

# Regular expression object to match C++ comments
# (used in findOperands())
commentRE = re.compile(r'(^)?[^\S\n]*/(?:\*(.*?)\*/[^\S\n]*|/[^\n]*)($)?',
        re.DOTALL | re.MULTILINE)

# Regular expression object to match assignment statements
# (used in findOperands())
assignRE = re.compile(r'\s*=(?!=)', re.MULTILINE)

def makeFlagConstructor(flag_list):
    if len(flag_list) == 0:
        return ''
    # filter out repeated flags
    flag_list.sort()
    i = 1
    while i < len(flag_list):
        if flag_list[i] == flag_list[i-1]:
            del flag_list[i]
        else:
            i += 1
    pre = '\n\tflags['
    post = '] = true;'
    code = pre + string.join(flag_list, post + pre) + post
    return code

# Assume all instruction flags are of the form 'IsFoo'
instFlagRE = re.compile(r'Is.*')

# OpClass constants end in 'Op' except No_OpClass
opClassRE = re.compile(r'.*Op|No_OpClass')

class InstObjParams(object):
    def __init__(self, parser, mnem, class_name, base_class = '',
                 snippets = {}, opt_args = []):
        self.mnemonic = mnem
        self.class_name = class_name
        self.base_class = base_class
        if not isinstance(snippets, dict):
            snippets = {'code' : snippets}
        compositeCode = ' '.join(map(str, snippets.values()))
        self.snippets = snippets

        self.operands = OperandList(parser, compositeCode)

        # The header of the constructor declares the variables to be used
        # in the body of the constructor.
        header = ''
        header += '\n\t_numSrcRegs = 0;'
        header += '\n\t_numDestRegs = 0;'
        header += '\n\t_numFPDestRegs = 0;'
        header += '\n\t_numIntDestRegs = 0;'

        self.constructor = header + \
                           self.operands.concatAttrStrings('constructor')

        self.flags = self.operands.concatAttrLists('flags')

        # Make a basic guess on the operand class (function unit type).
        # These are good enough for most cases, and can be overridden
        # later otherwise.
        if 'IsStore' in self.flags:
            self.op_class = 'MemWriteOp'
        elif 'IsLoad' in self.flags or 'IsPrefetch' in self.flags:
            self.op_class = 'MemReadOp'
        elif 'IsFloating' in self.flags:
            self.op_class = 'FloatAddOp'
        else:
            self.op_class = 'IntAluOp'

        # Optional arguments are assumed to be either StaticInst flags
        # or an OpClass value.  To avoid having to import a complete
        # list of these values to match against, we do it ad-hoc
        # with regexps.
        for oa in opt_args:
            if instFlagRE.match(oa):
                self.flags.append(oa)
            elif opClassRE.match(oa):
                self.op_class = oa
            else:
                error('InstObjParams: optional arg "%s" not recognized '
                      'as StaticInst::Flag or OpClass.' % oa)

        # add flag initialization to contructor here to include
        # any flags added via opt_args
        self.constructor += makeFlagConstructor(self.flags)

        # if 'IsFloating' is set, add call to the FP enable check
        # function (which should be provided by isa_desc via a declare)
        if 'IsFloating' in self.flags:
            self.fp_enable_check = 'fault = checkFpEnableFault(xc);'
        else:
            self.fp_enable_check = ''

##############
# Stack: a simple stack object.  Used for both formats (formatStack)
# and default cases (defaultStack).  Simply wraps a list to give more
# stack-like syntax and enable initialization with an argument list
# (as opposed to an argument that's a list).

class Stack(list):
    def __init__(self, *items):
        list.__init__(self, items)

    def push(self, item):
        self.append(item);

    def top(self):
        return self[-1]

#######################
#
# Output file template
#

file_template = '''
/*
 * DO NOT EDIT THIS FILE!!!
 *
 * It was automatically generated from the ISA description in %(filename)s
 */

%(includes)s

%(global_output)s

namespace %(namespace)s {

%(namespace_output)s

} // namespace %(namespace)s

%(decode_function)s
'''

max_inst_regs_template = '''
/*
 * DO NOT EDIT THIS FILE!!!
 *
 * It was automatically generated from the ISA description in %(filename)s
 */

namespace %(namespace)s {

    const int MaxInstSrcRegs = %(MaxInstSrcRegs)d;
    const int MaxInstDestRegs = %(MaxInstDestRegs)d;
    const int MaxMiscDestRegs = %(MaxMiscDestRegs)d;

} // namespace %(namespace)s

'''

class ISAParser(Grammar):
    def __init__(self, output_dir, cpu_models):
        super(ISAParser, self).__init__()
        self.output_dir = output_dir

        self.cpuModels = cpu_models

        # variable to hold templates
        self.templateMap = {}

        # This dictionary maps format name strings to Format objects.
        self.formatMap = {}

        # The format stack.
        self.formatStack = Stack(NoFormat())

        # The default case stack.
        self.defaultStack = Stack(None)

        # Stack that tracks current file and line number.  Each
        # element is a tuple (filename, lineno) that records the
        # *current* filename and the line number in the *previous*
        # file where it was included.
        self.fileNameStack = Stack()

        symbols = ('makeList', 're', 'string')
        self.exportContext = dict([(s, eval(s)) for s in symbols])

        self.maxInstSrcRegs = 0
        self.maxInstDestRegs = 0
        self.maxMiscDestRegs = 0

    #####################################################################
    #
    #                                Lexer
    #
    # The PLY lexer module takes two things as input:
    # - A list of token names (the string list 'tokens')
    # - A regular expression describing a match for each token.  The
    #   regexp for token FOO can be provided in two ways:
    #   - as a string variable named t_FOO
    #   - as the doc string for a function named t_FOO.  In this case,
    #     the function is also executed, allowing an action to be
    #     associated with each token match.
    #
    #####################################################################

    # Reserved words.  These are listed separately as they are matched
    # using the same regexp as generic IDs, but distinguished in the
    # t_ID() function.  The PLY documentation suggests this approach.
    reserved = (
        'BITFIELD', 'DECODE', 'DECODER', 'DEFAULT', 'DEF', 'EXEC', 'FORMAT',
        'HEADER', 'LET', 'NAMESPACE', 'OPERAND_TYPES', 'OPERANDS',
        'OUTPUT', 'SIGNED', 'TEMPLATE'
        )

    # List of tokens.  The lex module requires this.
    tokens = reserved + (
        # identifier
        'ID',

        # integer literal
        'INTLIT',

        # string literal
        'STRLIT',

        # code literal
        'CODELIT',

        # ( ) [ ] { } < > , ; . : :: *
        'LPAREN', 'RPAREN',
        'LBRACKET', 'RBRACKET',
        'LBRACE', 'RBRACE',
        'LESS', 'GREATER', 'EQUALS',
        'COMMA', 'SEMI', 'DOT', 'COLON', 'DBLCOLON',
        'ASTERISK',

        # C preprocessor directives
        'CPPDIRECTIVE'

    # The following are matched but never returned. commented out to
    # suppress PLY warning
        # newfile directive
    #    'NEWFILE',

        # endfile directive
    #    'ENDFILE'
    )

    # Regular expressions for token matching
    t_LPAREN           = r'\('
    t_RPAREN           = r'\)'
    t_LBRACKET         = r'\['
    t_RBRACKET         = r'\]'
    t_LBRACE           = r'\{'
    t_RBRACE           = r'\}'
    t_LESS             = r'\<'
    t_GREATER          = r'\>'
    t_EQUALS           = r'='
    t_COMMA            = r','
    t_SEMI             = r';'
    t_DOT              = r'\.'
    t_COLON            = r':'
    t_DBLCOLON         = r'::'
    t_ASTERISK         = r'\*'

    # Identifiers and reserved words
    reserved_map = { }
    for r in reserved:
        reserved_map[r.lower()] = r

    def t_ID(self, t):
        r'[A-Za-z_]\w*'
        t.type = self.reserved_map.get(t.value, 'ID')
        return t

    # Integer literal
    def t_INTLIT(self, t):
        r'-?(0x[\da-fA-F]+)|\d+'
        try:
            t.value = int(t.value,0)
        except ValueError:
            error(t, 'Integer value "%s" too large' % t.value)
            t.value = 0
        return t

    # String literal.  Note that these use only single quotes, and
    # can span multiple lines.
    def t_STRLIT(self, t):
        r"(?m)'([^'])+'"
        # strip off quotes
        t.value = t.value[1:-1]
        t.lexer.lineno += t.value.count('\n')
        return t


    # "Code literal"... like a string literal, but delimiters are
    # '{{' and '}}' so they get formatted nicely under emacs c-mode
    def t_CODELIT(self, t):
        r"(?m)\{\{([^\}]|}(?!\}))+\}\}"
        # strip off {{ & }}
        t.value = t.value[2:-2]
        t.lexer.lineno += t.value.count('\n')
        return t

    def t_CPPDIRECTIVE(self, t):
        r'^\#[^\#].*\n'
        t.lexer.lineno += t.value.count('\n')
        return t

    def t_NEWFILE(self, t):
        r'^\#\#newfile\s+"[^"]*"'
        self.fileNameStack.push((t.value[11:-1], t.lexer.lineno))
        t.lexer.lineno = 0

    def t_ENDFILE(self, t):
        r'^\#\#endfile'
        (old_filename, t.lexer.lineno) = self.fileNameStack.pop()

    #
    # The functions t_NEWLINE, t_ignore, and t_error are
    # special for the lex module.
    #

    # Newlines
    def t_NEWLINE(self, t):
        r'\n+'
        t.lexer.lineno += t.value.count('\n')

    # Comments
    def t_comment(self, t):
        r'//.*'

    # Completely ignored characters
    t_ignore = ' \t\x0c'

    # Error handler
    def t_error(self, t):
        error(t, "illegal character '%s'" % t.value[0])
        t.skip(1)

    #####################################################################
    #
    #                                Parser
    #
    # Every function whose name starts with 'p_' defines a grammar
    # rule.  The rule is encoded in the function's doc string, while
    # the function body provides the action taken when the rule is
    # matched.  The argument to each function is a list of the values
    # of the rule's symbols: t[0] for the LHS, and t[1..n] for the
    # symbols on the RHS.  For tokens, the value is copied from the
    # t.value attribute provided by the lexer.  For non-terminals, the
    # value is assigned by the producing rule; i.e., the job of the
    # grammar rule function is to set the value for the non-terminal
    # on the LHS (by assigning to t[0]).
    #####################################################################

    # The LHS of the first grammar rule is used as the start symbol
    # (in this case, 'specification').  Note that this rule enforces
    # that there will be exactly one namespace declaration, with 0 or
    # more global defs/decls before and after it.  The defs & decls
    # before the namespace decl will be outside the namespace; those
    # after will be inside.  The decoder function is always inside the
    # namespace.
    def p_specification(self, t):
        'specification : opt_defs_and_outputs name_decl opt_defs_and_outputs decode_block'
        global_code = t[1]
        isa_name = t[2]
        namespace = isa_name + "Inst"
        # wrap the decode block as a function definition
        t[4].wrap_decode_block('''
StaticInstPtr
%(isa_name)s::Decoder::decodeInst(%(isa_name)s::ExtMachInst machInst)
{
    using namespace %(namespace)s;
''' % vars(), '}')
        # both the latter output blocks and the decode block are in
        # the namespace
        namespace_code = t[3] + t[4]
        # pass it all back to the caller of yacc.parse()
        t[0] = (isa_name, namespace, global_code, namespace_code)

    # ISA name declaration looks like "namespace <foo>;"
    def p_name_decl(self, t):
        'name_decl : NAMESPACE ID SEMI'
        t[0] = t[2]

    # 'opt_defs_and_outputs' is a possibly empty sequence of
    # def and/or output statements.
    def p_opt_defs_and_outputs_0(self, t):
        'opt_defs_and_outputs : empty'
        t[0] = GenCode(self)

    def p_opt_defs_and_outputs_1(self, t):
        'opt_defs_and_outputs : defs_and_outputs'
        t[0] = t[1]

    def p_defs_and_outputs_0(self, t):
        'defs_and_outputs : def_or_output'
        t[0] = t[1]

    def p_defs_and_outputs_1(self, t):
        'defs_and_outputs : defs_and_outputs def_or_output'
        t[0] = t[1] + t[2]

    # The list of possible definition/output statements.
    def p_def_or_output(self, t):
        '''def_or_output : def_format
                         | def_bitfield
                         | def_bitfield_struct
                         | def_template
                         | def_operand_types
                         | def_operands
                         | output_header
                         | output_decoder
                         | output_exec
                         | global_let'''
        t[0] = t[1]

    # Output blocks 'output <foo> {{...}}' (C++ code blocks) are copied
    # directly to the appropriate output section.

    # Massage output block by substituting in template definitions and
    # bit operators.  We handle '%'s embedded in the string that don't
    # indicate template substitutions (or CPU-specific symbols, which
    # get handled in GenCode) by doubling them first so that the
    # format operation will reduce them back to single '%'s.
   …