import sys
import re
import string
-import traceback
+import inspect, traceback
# get type names
from types import *
from m5.util.grammar import Grammar
+debug=False
+
###################
# Utility functions
s = 'if 1:\n' + s
return s
-# Error handler. Just call exit. 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).
-def error(lineno, string, print_traceback = False):
- spaces = ""
- for (filename, line) in fileNameStack[0:-1]:
- print spaces + "In file included from " + filename + ":"
- spaces += " "
- # Print a Python stack backtrace if requested.
- if (print_traceback):
- traceback.print_exc()
- if lineno != 0:
- line_str = "%d:" % lineno
- else:
- line_str = ""
- sys.exit(spaces + "%s:%s %s" % (fileNameStack[-1][0], line_str, string))
+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.
labelRE = re.compile(r'(?<!%)%\(([^\)]+)\)[sd]')
class Template(object):
- def __init__(self, t):
+ def __init__(self, parser, t):
+ self.parser = parser
self.template = t
def subst(self, d):
# Protect non-Python-dict substitutions (e.g. if there's a printf
# in the templated C++ code)
- template = protect_non_subst_percents(self.template)
+ template = self.parser.protectNonSubstPercents(self.template)
# CPU-model-specific substitutions are handled later (in GenCode).
- template = protect_cpu_symbols(template)
+ 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 = parser.templateMap.copy()
+ myDict = self.parser.templateMap.copy()
if isinstance(d, InstObjParams):
# If we're dealing with an InstObjParams object, we need
snippetLabels = [l for l in labelRE.findall(template)
if d.snippets.has_key(l)]
- snippets = dict([(s, mungeSnippet(d.snippets[s]))
+ snippets = dict([(s, self.parser.mungeSnippet(d.snippets[s]))
for s in snippetLabels])
myDict.update(snippets)
# operands explicitly (like Mem)
compositeCode += ' ' + template
- operands = SubOperandList(compositeCode, d.operands)
+ 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'
is_src = lambda op: op.is_src
is_dest = lambda op: op.is_dest
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')
- myDict['op_wb'] = operands.concatAttrStrings('op_wb')
-
- if d.operands.memOperand:
- myDict['mem_acc_size'] = d.operands.memOperand.mem_acc_size
- myDict['mem_acc_type'] = d.operands.memOperand.mem_acc_type
+ 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.
# CPU-specific term gets interpolated into another template or into
# an output block.
def __str__(self):
- return expand_cpu_symbols_to_string(self.template)
+ return self.parser.expandCpuSymbolsToString(self.template)
################
# Format object.
# a defineInst() method that generates the code for an instruction
# definition.
-exportContextSymbols = ('InstObjParams', 'makeList', 're', 'string')
-
-exportContext = {}
-
-def updateExportContext():
- exportContext.update(exportDict(*exportContextSymbols))
- exportContext.update(parser.templateMap)
-
-def exportDict(*symNames):
- return dict([(s, eval(s)) for s in symNames])
-
-
class Format(object):
def __init__(self, id, params, code):
- # constructor: just save away arguments
self.id = id
self.params = params
label = 'def format ' + id
exec c
self.func = defInst
- def defineInst(self, name, args, lineno):
- context = {}
- updateExportContext()
- context.update(exportContext)
+ 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 })
+ 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(**vars)
+ return GenCode(parser, **vars)
# Special null format to catch an implicit-format instruction
# definition outside of any format block.
def __init__(self):
self.defaultInst = ''
- def defineInst(self, name, args, lineno):
+ def defineInst(self, parser, name, args, lineno):
error(lineno,
'instruction definition "%s" with no active format!' % name)
-# This dictionary maps format name strings to Format objects.
-formatMap = {}
-
-# Define a new format
-def defFormat(id, params, code, lineno):
- # make sure we haven't already defined this one
- if formatMap.get(id, None) != None:
- error(lineno, 'format %s redefined.' % id)
- # create new object and store in global map
- formatMap[id] = Format(id, params, code)
-
-#####################################################################
-#
-# 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 = {}
- for cpu in cpu_models:
- result[cpu.name] = t % cpu.strings
- return result
-
-# *If* the template has CPU-specific references, return a single
-# string containing a copy of the template for each CPU model with the
-# corresponding values substituted in. If the template has no
-# CPU-specific references, it is returned unmodified.
-def expand_cpu_symbols_to_string(template):
- if template.find('%(CPU_') != -1:
- return reduce(lambda x,y: x+y,
- expand_cpu_symbols_to_dict(template).values())
- else:
- return template
-
-# Protect CPU-specific references by doubling the corresponding '%'s
-# (in preparation for substituting a different set of references into
-# the template).
-def protect_cpu_symbols(template):
- return re.sub(r'%(?=\(CPU_)', '%%', template)
-
-# Protect any non-dict-substitution '%'s in a format string
-# (i.e. those not followed by '(')
-def protect_non_subst_percents(s):
- return re.sub(r'%(?!\()', '%%', s)
-
###############
# GenCode class
#
# 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,
+ def __init__(self, parser,
header_output = '', decoder_output = '', exec_output = '',
decode_block = '', has_decode_default = False):
- self.header_output = expand_cpu_symbols_to_string(header_output)
- self.decoder_output = expand_cpu_symbols_to_string(decoder_output)
+ 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 = expand_cpu_symbols_to_dict(exec_output)
- self.decode_block = expand_cpu_symbols_to_string(decode_block)
+ 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 cpu_models:
+ for cpu in self.parser.cpuModels:
n = cpu.name
exec_output[n] = self.exec_output[n] + other.exec_output[n]
- return GenCode(self.header_output + other.header_output,
+ 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.header_output = pre + self.header_output
self.decoder_output = pre + self.decoder_output
self.decode_block = pre + self.decode_block
- for cpu in cpu_models:
+ 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:'
else:
return [ arg ]
-# Generate operandTypeMap from the user's 'def operand_types'
-# statement.
-def buildOperandTypeMap(user_dict, lineno):
- global operandTypeMap
- operandTypeMap = {}
- for (ext, (desc, size)) in user_dict.iteritems():
- if desc == 'signed int':
- ctype = 'int%d_t' % size
- is_signed = 1
- elif desc == 'unsigned int':
- ctype = 'uint%d_t' % size
- is_signed = 0
- elif desc == 'float':
- is_signed = 1 # shouldn't really matter
- if size == 32:
- ctype = 'float'
- elif size == 64:
- ctype = 'double'
- elif desc == 'twin64 int':
- is_signed = 0
- ctype = 'Twin64_t'
- elif desc == 'twin32 int':
- is_signed = 0
- ctype = 'Twin32_t'
- if ctype == '':
- error(lineno, 'Unrecognized type description "%s" in user_dict')
- operandTypeMap[ext] = (size, ctype, is_signed)
-
class Operand(object):
'''Base class for operand descriptors. An instance of this class
(or actually a class derived from this one) represents a specific
type (e.g., "32-bit integer register").'''
def buildReadCode(self, func = None):
- code = self.read_code % {"name": self.base_name,
- "func": func,
- "op_idx": self.src_reg_idx,
- "reg_idx": self.reg_spec,
- "size": self.size,
- "ctype": self.ctype}
- if self.size != self.dflt_size:
- return '%s = bits(%s, %d, 0);\n' % \
- (self.base_name, code, self.size-1)
- else:
- return '%s = %s;\n' % \
- (self.base_name, code)
+ 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):
- if (self.size != self.dflt_size and self.is_signed):
- final_val = 'sext<%d>(%s)' % (self.size, self.base_name)
- else:
- final_val = self.base_name
- code = self.write_code % {"name": self.base_name,
- "func": func,
- "op_idx": self.dest_reg_idx,
- "reg_idx": self.reg_spec,
- "size": self.size,
- "ctype": self.ctype,
- "final_val": final_val}
+ 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, final_val, code)
+ }''' % (self.dflt_ctype, self.base_name, code)
- def __init__(self, full_name, ext, is_src, is_dest):
+ def __init__(self, parser, full_name, ext, is_src, is_dest):
self.full_name = full_name
self.ext = ext
self.is_src = is_src
# extension, if one was explicitly provided, or the default.
if ext:
self.eff_ext = ext
- else:
+ elif hasattr(self, 'dflt_ext'):
self.eff_ext = self.dflt_ext
- (self.size, self.ctype, self.is_signed) = operandTypeMap[self.eff_ext]
-
- # note that mem_acc_size is undefined for non-mem operands...
- # template must be careful not to use it if it doesn't apply.
- if self.isMem():
- self.mem_acc_size = self.makeAccSize()
- if self.ctype in ['Twin32_t', 'Twin64_t']:
- self.mem_acc_type = 'Twin'
- else:
- self.mem_acc_type = 'uint'
+ 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
def isControlReg(self):
return 0
+ def isPCState(self):
+ return 0
+
+ def isPCPart(self):
+ return self.isPCState() and self.reg_spec
+
def getFlags(self):
# note the empty slice '[:]' gives us a copy of self.flags[0]
# instead of a reference to it
def makeRead(self):
if (self.ctype == 'float' or self.ctype == 'double'):
- error(0, 'Attempt to read integer register as FP')
+ error('Attempt to read integer register as FP')
if self.read_code != None:
return self.buildReadCode('readIntRegOperand')
- if (self.size == self.dflt_size):
- return '%s = xc->readIntRegOperand(this, %d);\n' % \
- (self.base_name, self.src_reg_idx)
- elif (self.size > self.dflt_size):
- int_reg_val = 'xc->readIntRegOperand(this, %d)' % \
- (self.src_reg_idx)
- if (self.is_signed):
- int_reg_val = 'sext<%d>(%s)' % (self.dflt_size, int_reg_val)
- return '%s = %s;\n' % (self.base_name, int_reg_val)
- else:
- return '%s = bits(xc->readIntRegOperand(this, %d), %d, 0);\n' % \
- (self.base_name, self.src_reg_idx, self.size-1)
+ int_reg_val = 'xc->readIntRegOperand(this, %d)' % self.src_reg_idx
+ return '%s = %s;\n' % (self.base_name, int_reg_val)
def makeWrite(self):
if (self.ctype == 'float' or self.ctype == 'double'):
- error(0, 'Attempt to write integer register as FP')
+ error('Attempt to write integer register as FP')
if self.write_code != None:
return self.buildWriteCode('setIntRegOperand')
- if (self.size != self.dflt_size and self.is_signed):
- final_val = 'sext<%d>(%s)' % (self.size, self.base_name)
- else:
- final_val = self.base_name
wb = '''
{
%s final_val = %s;
xc->setIntRegOperand(this, %d, final_val);\n
if (traceData) { traceData->setData(final_val); }
- }''' % (self.dflt_ctype, final_val, self.dest_reg_idx)
+ }''' % (self.ctype, self.base_name, self.dest_reg_idx)
return wb
class FloatRegOperand(Operand):
func = 'readFloatRegOperand'
else:
func = 'readFloatRegOperandBits'
- if (self.size != self.dflt_size):
- bit_select = 1
- base = 'xc->%s(this, %d)' % (func, self.src_reg_idx)
if self.read_code != None:
return self.buildReadCode(func)
- if bit_select:
- return '%s = bits(%s, %d, 0);\n' % \
- (self.base_name, base, self.size-1)
- else:
- return '%s = %s;\n' % (self.base_name, base)
+ return '%s = xc->%s(this, %d);\n' % \
+ (self.base_name, func, self.src_reg_idx)
def makeWrite(self):
- final_val = self.base_name
- final_ctype = self.ctype
if (self.ctype == 'float' or self.ctype == 'double'):
func = 'setFloatRegOperand'
- elif (self.ctype == 'uint32_t' or self.ctype == 'uint64_t'):
- func = 'setFloatRegOperandBits'
else:
func = 'setFloatRegOperandBits'
- final_ctype = 'uint%d_t' % self.dflt_size
- if (self.size != self.dflt_size and self.is_signed):
- final_val = 'sext<%d>(%s)' % (self.size, self.base_name)
if self.write_code != None:
return self.buildWriteCode(func)
wb = '''
%s final_val = %s;
xc->%s(this, %d, final_val);\n
if (traceData) { traceData->setData(final_val); }
- }''' % (final_ctype, final_val, func, self.dest_reg_idx)
+ }''' % (self.ctype, self.base_name, func, self.dest_reg_idx)
return wb
class ControlRegOperand(Operand):
def makeRead(self):
bit_select = 0
if (self.ctype == 'float' or self.ctype == 'double'):
- error(0, 'Attempt to read control register as FP')
+ error('Attempt to read control register as FP')
if self.read_code != None:
return self.buildReadCode('readMiscRegOperand')
- base = 'xc->readMiscRegOperand(this, %s)' % self.src_reg_idx
- if self.size == self.dflt_size:
- return '%s = %s;\n' % (self.base_name, base)
- else:
- return '%s = bits(%s, %d, 0);\n' % \
- (self.base_name, base, self.size-1)
+ return '%s = xc->readMiscRegOperand(this, %s);\n' % \
+ (self.base_name, self.src_reg_idx)
def makeWrite(self):
if (self.ctype == 'float' or self.ctype == 'double'):
- error(0, 'Attempt to write control register as FP')
+ error('Attempt to write control register as FP')
if self.write_code != None:
return self.buildWriteCode('setMiscRegOperand')
wb = 'xc->setMiscRegOperand(this, %s, %s);\n' % \
# Note that initializations in the declarations are solely
# to avoid 'uninitialized variable' errors from the compiler.
# Declare memory data variable.
- if self.ctype in ['Twin32_t','Twin64_t']:
- return "%s %s; %s.a = 0; %s.b = 0;\n" % \
- (self.ctype, self.base_name, self.base_name, self.base_name)
return '%s %s = 0;\n' % (self.ctype, self.base_name)
def makeRead(self):
return self.buildWriteCode()
return ''
- # Return the memory access size *in bits*, suitable for
- # forming a type via "uint%d_t". Divide by 8 if you want bytes.
- def makeAccSize(self):
- return self.size
-
-class PCOperand(Operand):
- def makeConstructor(self):
- return ''
-
- def makeRead(self):
- return '%s = xc->readPC();\n' % self.base_name
-
- def makeWrite(self):
- return 'xc->setPC(%s);\n' % self.base_name
-
-class UPCOperand(Operand):
- def makeConstructor(self):
- return ''
-
- def makeRead(self):
- if self.read_code != None:
- return self.buildReadCode('readMicroPC')
- return '%s = xc->readMicroPC();\n' % self.base_name
-
- def makeWrite(self):
- if self.write_code != None:
- return self.buildWriteCode('setMicroPC')
- return 'xc->setMicroPC(%s);\n' % self.base_name
-
-class NUPCOperand(Operand):
+class PCStateOperand(Operand):
def makeConstructor(self):
return ''
def makeRead(self):
- if self.read_code != None:
- return self.buildReadCode('readNextMicroPC')
- return '%s = xc->readNextMicroPC();\n' % self.base_name
-
- def makeWrite(self):
- if self.write_code != None:
- return self.buildWriteCode('setNextMicroPC')
- return 'xc->setNextMicroPC(%s);\n' % self.base_name
-
-class NPCOperand(Operand):
- def makeConstructor(self):
- return ''
-
- def makeRead(self):
- if self.read_code != None:
- return self.buildReadCode('readNextPC')
- return '%s = xc->readNextPC();\n' % self.base_name
+ 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):
- if self.write_code != None:
- return self.buildWriteCode('setNextPC')
- return 'xc->setNextPC(%s);\n' % self.base_name
-
-class NNPCOperand(Operand):
- def makeConstructor(self):
- return ''
+ 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 makeRead(self):
- if self.read_code != None:
- return self.buildReadCode('readNextNPC')
- return '%s = xc->readNextNPC();\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 makeWrite(self):
- if self.write_code != None:
- return self.buildWriteCode('setNextNPC')
- return 'xc->setNextNPC(%s);\n' % self.base_name
-
-def buildOperandNameMap(user_dict, lineno):
- global operandNameMap
- operandNameMap = {}
- for (op_name, val) in user_dict.iteritems():
- (base_cls_name, dflt_ext, reg_spec, flags, sort_pri) = val[:5]
- if len(val) > 5:
- read_code = val[5]
- else:
- read_code = None
- if len(val) > 6:
- write_code = val[6]
- else:
- write_code = None
- if len(val) > 7:
- error(lineno,
- 'error: too many attributes for operand "%s"' %
- base_cls_name)
-
- (dflt_size, dflt_ctype, dflt_is_signed) = operandTypeMap[dflt_ext]
- # Canonical flag structure is a triple of lists, where each list
- # indicates the set of flags implied by this operand always, when
- # used as a source, and when used as a dest, respectively.
- # For simplicity this can be initialized using a variety of fairly
- # obvious shortcuts; we convert these to canonical form here.
- if not flags:
- # no flags specified (e.g., 'None')
- flags = ( [], [], [] )
- elif isinstance(flags, str):
- # a single flag: assumed to be unconditional
- flags = ( [ flags ], [], [] )
- elif isinstance(flags, list):
- # a list of flags: also assumed to be unconditional
- flags = ( flags, [], [] )
- elif isinstance(flags, tuple):
- # it's a tuple: it should be a triple,
- # but each item could be a single string or a list
- (uncond_flags, src_flags, dest_flags) = flags
- flags = (makeList(uncond_flags),
- makeList(src_flags), makeList(dest_flags))
- # Accumulate attributes of new operand class in tmp_dict
- tmp_dict = {}
- for attr in ('dflt_ext', 'reg_spec', 'flags', 'sort_pri',
- 'dflt_size', 'dflt_ctype', 'dflt_is_signed',
- 'read_code', 'write_code'):
- tmp_dict[attr] = eval(attr)
- tmp_dict['base_name'] = op_name
- # New class name will be e.g. "IntReg_Ra"
- cls_name = base_cls_name + '_' + op_name
- # Evaluate string arg to get class object. Note that the
- # actual base class for "IntReg" is "IntRegOperand", i.e. we
- # have to append "Operand".
- try:
- base_cls = eval(base_cls_name + 'Operand')
- except NameError:
- error(lineno,
- 'error: unknown operand base class "%s"' % base_cls_name)
- # The following statement creates a new class called
- # <cls_name> as a subclass of <base_cls> with the attributes
- # in tmp_dict, just as if we evaluated a class declaration.
- operandNameMap[op_name] = type(cls_name, (base_cls,), tmp_dict)
-
- # Define operand variables.
- operands = user_dict.keys()
-
- operandsREString = (r'''
- (?<![\w\.]) # neg. lookbehind assertion: prevent partial matches
- ((%s)(?:\.(\w+))?) # match: operand with optional '.' then suffix
- (?![\w\.]) # neg. lookahead assertion: prevent partial matches
- '''
- % string.join(operands, '|'))
-
- global operandsRE
- operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE)
-
- # Same as operandsREString, but extension is mandatory, and only two
- # groups are returned (base and ext, not full name as above).
- # Used for subtituting '_' for '.' to make C++ identifiers.
- operandsWithExtREString = (r'(?<![\w\.])(%s)\.(\w+)(?![\w\.])'
- % string.join(operands, '|'))
-
- global operandsWithExtRE
- operandsWithExtRE = re.compile(operandsWithExtREString, re.MULTILINE)
-
-maxInstSrcRegs = 0
-maxInstDestRegs = 0
+ 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, code):
+ def __init__(self, parser, code):
self.items = []
self.bases = {}
- # delete comments so we don't match on reg specifiers inside
- code = commentRE.sub('', code)
+ # 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 = operandsRE.search(code, next_pos)
+ match = parser.operandsRE.search(code, next_pos)
if not match:
# no more matches: we're done
break
op_desc = self.find_base(op_base)
if op_desc:
if op_desc.ext != op_ext:
- error(0, 'Inconsistent extensions for operand %s' % \
+ 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 = operandNameMap[op_base](op_full, op_ext,
- is_src, is_dest)
+ 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.numIntDestRegs += 1
elif op_desc.isMem():
if self.memOperand:
- error(0, "Code block has more than one memory operand.")
+ error("Code block has more than one memory operand.")
self.memOperand = op_desc
- global maxInstSrcRegs
- global maxInstDestRegs
- if maxInstSrcRegs < self.numSrcRegs:
- maxInstSrcRegs = self.numSrcRegs
- if maxInstDestRegs < self.numDestRegs:
- maxInstDestRegs = self.numDestRegs
+ if parser.maxInstSrcRegs < self.numSrcRegs:
+ parser.maxInstSrcRegs = self.numSrcRegs
+ if parser.maxInstDestRegs < self.numDestRegs:
+ parser.maxInstDestRegs = self.numDestRegs
# now make a final pass to finalize op_desc fields that may depend
# on the register enumeration
for op_desc in self.items:
class SubOperandList(OperandList):
'''Find all the operands in the given code block. Returns an operand
descriptor list (instance of class OperandList).'''
- def __init__(self, code, master_list):
+ def __init__(self, parser, code, master_list):
self.items = []
self.bases = {}
- # delete comments so we don't match on reg specifiers inside
- code = commentRE.sub('', code)
+ # 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 = operandsRE.search(code, next_pos)
+ match = parser.operandsRE.search(code, next_pos)
if not match:
# no more matches: we're done
break
# find this op in the master list
op_desc = master_list.find_base(op_base)
if not op_desc:
- error(0, 'Found operand %s which is not in the master list!' \
- ' This is an internal error' % \
- op_base)
+ 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)
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
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(0, "Code block has more than one memory operand.")
+ error("Code block has more than one memory operand.")
self.memOperand = op_desc
+# 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'//.*\n')
+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)
-# Munge operand names in code string to make legal C++ variable names.
-# This means getting rid of the type extension if any.
-# (Will match base_name attribute of Operand object.)
-def substMungedOpNames(code):
- return operandsWithExtRE.sub(r'\1', code)
-
-# Fix up code snippets for final substitution in templates.
-def mungeSnippet(s):
- if isinstance(s, str):
- return substMungedOpNames(substBitOps(s))
- else:
- return s
-
def makeFlagConstructor(flag_list):
if len(flag_list) == 0:
return ''
opClassRE = re.compile(r'.*Op|No_OpClass')
class InstObjParams(object):
- def __init__(self, mnem, class_name, base_class = '',
+ def __init__(self, parser, mnem, class_name, base_class = '',
snippets = {}, opt_args = []):
self.mnemonic = mnem
self.class_name = class_name
compositeCode = ' '.join(map(str, snippets.values()))
self.snippets = snippets
- self.operands = OperandList(compositeCode)
+ self.operands = OperandList(parser, compositeCode)
self.constructor = self.operands.concatAttrStrings('constructor')
self.constructor += \
'\n\t_numSrcRegs = %d;' % self.operands.numSrcRegs
elif opClassRE.match(oa):
self.op_class = oa
else:
- error(0, 'InstObjParams: optional arg "%s" not recognized '
+ error('InstObjParams: optional arg "%s" not recognized '
'as StaticInst::Flag or OpClass.' % oa)
# add flag initialization to contructor here to include
def top(self):
return self[-1]
-# The global format stack.
-formatStack = Stack(NoFormat())
-
-# The global default case stack.
-defaultStack = Stack(None)
-
-# Global 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.
-fileNameStack = Stack()
-
-
#######################
#
# Output file template
'''
class ISAParser(Grammar):
- def __init__(self, *args, **kwargs):
- super(ISAParser, self).__init__(*args, **kwargs)
+ 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
+
#####################################################################
#
# Lexer
try:
t.value = int(t.value,0)
except ValueError:
- error(t.lexer.lineno, 'Integer value "%s" too large' % t.value)
+ error(t, 'Integer value "%s" too large' % t.value)
t.value = 0
return t
return t
def t_NEWFILE(self, t):
- r'^\#\#newfile\s+"[\w/.-]*"'
- fileNameStack.push((t.value[11:-1], t.lexer.lineno))
+ 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) = fileNameStack.pop()
+ (old_filename, t.lexer.lineno) = self.fileNameStack.pop()
#
# The functions t_NEWLINE, t_ignore, and t_error are
# Error handler
def t_error(self, t):
- error(t.lexer.lineno, "illegal character '%s'" % t.value[0])
+ error(t, "illegal character '%s'" % t.value[0])
t.skip(1)
#####################################################################
# wrap the decode block as a function definition
t[4].wrap_decode_block('''
StaticInstPtr
-%(isa_name)s::decodeInst(%(isa_name)s::ExtMachInst machInst)
+%(isa_name)s::Decoder::decodeInst(%(isa_name)s::ExtMachInst machInst)
{
using namespace %(namespace)s;
''' % vars(), '}')
# def and/or output statements.
def p_opt_defs_and_outputs_0(self, t):
'opt_defs_and_outputs : empty'
- t[0] = GenCode()
+ t[0] = GenCode(self)
def p_opt_defs_and_outputs_1(self, t):
'opt_defs_and_outputs : defs_and_outputs'
# get handled in GenCode) by doubling them first so that the
# format operation will reduce them back to single '%'s.
def process_output(self, s):
- s = protect_non_subst_percents(s)
+ s = self.protectNonSubstPercents(s)
# protects cpu-specific symbols too
- s = protect_cpu_symbols(s)
+ s = self.protectCpuSymbols(s)
return substBitOps(s % self.templateMap)
def p_output_header(self, t):
'output_header : OUTPUT HEADER CODELIT SEMI'
- t[0] = GenCode(header_output = self.process_output(t[3]))
+ t[0] = GenCode(self, header_output = self.process_output(t[3]))
def p_output_decoder(self, t):
'output_decoder : OUTPUT DECODER CODELIT SEMI'
- t[0] = GenCode(decoder_output = self.process_output(t[3]))
+ t[0] = GenCode(self, decoder_output = self.process_output(t[3]))
def p_output_exec(self, t):
'output_exec : OUTPUT EXEC CODELIT SEMI'
- t[0] = GenCode(exec_output = self.process_output(t[3]))
+ t[0] = GenCode(self, exec_output = self.process_output(t[3]))
# global let blocks 'let {{...}}' (Python code blocks) are
# executed directly when seen. Note that these execute in a
# from polluting this script's namespace.
def p_global_let(self, t):
'global_let : LET CODELIT SEMI'
- updateExportContext()
- exportContext["header_output"] = ''
- exportContext["decoder_output"] = ''
- exportContext["exec_output"] = ''
- exportContext["decode_block"] = ''
+ self.updateExportContext()
+ self.exportContext["header_output"] = ''
+ self.exportContext["decoder_output"] = ''
+ self.exportContext["exec_output"] = ''
+ self.exportContext["decode_block"] = ''
try:
- exec fixPythonIndentation(t[2]) in exportContext
+ exec fixPythonIndentation(t[2]) in self.exportContext
except Exception, exc:
- error(t.lexer.lineno,
- 'error: %s in global let block "%s".' % (exc, t[2]))
- t[0] = GenCode(header_output = exportContext["header_output"],
- decoder_output = exportContext["decoder_output"],
- exec_output = exportContext["exec_output"],
- decode_block = exportContext["decode_block"])
+ if debug:
+ raise
+ error(t, 'error: %s in global let block "%s".' % (exc, t[2]))
+ t[0] = 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"])
# 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:
- user_dict = eval('{' + t[3] + '}')
+ self.operandTypeMap = eval('{' + t[3] + '}')
except Exception, exc:
- error(t.lexer.lineno,
+ if debug:
+ raise
+ error(t,
'error: %s in def operand_types block "%s".' % (exc, t[3]))
- buildOperandTypeMap(user_dict, t.lexer.lineno)
- t[0] = GenCode() # contributes nothing to the output C++ file
+ t[0] = GenCode(self) # contributes nothing to the output C++ file
# 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 globals().has_key('operandTypeMap'):
- error(t.lexer.lineno,
- 'error: operand types must be defined before operands')
+ if not hasattr(self, 'operandTypeMap'):
+ error(t, 'error: operand types must be defined before operands')
try:
- user_dict = eval('{' + t[3] + '}', exportContext)
+ user_dict = eval('{' + t[3] + '}', self.exportContext)
except Exception, exc:
- error(t.lexer.lineno,
- 'error: %s in def operands block "%s".' % (exc, t[3]))
- buildOperandNameMap(user_dict, t.lexer.lineno)
- t[0] = GenCode() # contributes nothing to the output C++ file
+ if debug:
+ raise
+ error(t, 'error: %s in def operands block "%s".' % (exc, t[3]))
+ self.buildOperandNameMap(user_dict, t.lexer.lineno)
+ t[0] = GenCode(self) # contributes nothing to the output C++ file
# A bitfield definition looks like:
# 'def [signed] bitfield <ID> [<first>:<last>]'
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)
- t[0] = GenCode(header_output = hash_define)
+ t[0] = GenCode(self, header_output=hash_define)
# alternate form for single bit: 'def [signed] bitfield <ID> [<bit>]'
def p_def_bitfield_1(self, t):
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)
- t[0] = GenCode(header_output = hash_define)
+ t[0] = GenCode(self, header_output=hash_define)
# 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.lexer.lineno,
- 'error: structure bitfields are always unsigned.')
+ error(t, '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)
+ t[0] = GenCode(self, header_output=hash_define)
def p_id_with_dot_0(self, t):
'id_with_dot : ID'
def p_def_template(self, t):
'def_template : DEF TEMPLATE ID CODELIT SEMI'
- self.templateMap[t[3]] = Template(t[4])
- t[0] = GenCode()
+ self.templateMap[t[3]] = Template(self, t[4])
+ t[0] = GenCode(self)
# 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])
- defFormat(id, params, code, t.lexer.lineno)
- t[0] = GenCode()
+ self.defFormat(id, params, code, t.lexer.lineno)
+ t[0] = GenCode(self)
# The formal parameter list for an instruction format is a
# possibly empty list of comma-separated parameters. Positional
#
def p_decode_block(self, t):
'decode_block : DECODE ID opt_default LBRACE decode_stmt_list RBRACE'
- default_defaults = defaultStack.pop()
+ default_defaults = self.defaultStack.pop()
codeObj = t[5]
# use the "default defaults" only if there was no explicit
# default statement in decode_stmt_list
'opt_default : empty'
# no default specified: reuse the one currently at the top of
# the stack
- defaultStack.push(defaultStack.top())
+ self.defaultStack.push(self.defaultStack.top())
# no meaningful value returned
t[0] = None
# push the new default
codeObj = t[2]
codeObj.wrap_decode_block('\ndefault:\n', 'break;\n')
- defaultStack.push(codeObj)
+ self.defaultStack.push(codeObj)
# no meaningful value returned
t[0] = None
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.lexer.lineno, 'Two default cases in decode block')
+ error(t, 'Two default cases in decode block')
t[0] = t[1] + t[2]
#
# the code generated by the other statements.
def p_decode_stmt_cpp(self, t):
'decode_stmt : CPPDIRECTIVE'
- t[0] = GenCode(t[1], t[1], t[1], t[1])
+ 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
# 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()
+ self.formatStack.pop()
t[0] = t[4]
# This rule exists so we can set the current format (& push the
def p_push_format_id(self, t):
'push_format_id : ID'
try:
- formatStack.push(formatMap[t[1]])
+ self.formatStack.push(self.formatMap[t[1]])
t[0] = ('', '// format %s' % t[1])
except KeyError:
- error(t.lexer.lineno,
- 'instruction format "%s" not defined.' % t[1])
+ error(t, '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_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 = formatStack.top()
- codeObj = currentFormat.defineInst(t[1], t[3], t.lexer.lineno)
+ 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)
def p_inst_1(self, t):
'inst : ID DBLCOLON ID LPAREN arg_list RPAREN'
try:
- format = formatMap[t[1]]
+ format = self.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)
+ error(t, '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
# t.value)
def p_error(self, t):
if t:
- error(t.lexer.lineno, "syntax error at '%s'" % t.value)
+ error(t, "syntax error at '%s'" % t.value)
else:
- error(0, "unknown syntax error", True)
+ error("unknown syntax error")
# END OF GRAMMAR RULES
-# Now build the parser.
-parser = ISAParser()
-
-# Update the output file only if the new contents are different from
-# the current contents. Minimizes the files that need to be rebuilt
-# after minor changes.
-def update_if_needed(file, contents):
- update = False
- if os.access(file, os.R_OK):
- f = open(file, 'r')
- old_contents = f.read()
- f.close()
- if contents != old_contents:
- print 'Updating', file
- os.remove(file) # in case it's write-protected
- update = True
+ def updateExportContext(self):
+
+ # create a continuation that allows us to grab the current parser
+ def wrapInstObjParams(*args):
+ return InstObjParams(self, *args)
+ self.exportContext['InstObjParams'] = wrapInstObjParams
+ self.exportContext.update(self.templateMap)
+
+ def defFormat(self, id, params, code, lineno):
+ '''Define a new format'''
+
+ # make sure we haven't already defined this one
+ if id in self.formatMap:
+ error(lineno, 'format %s redefined.' % id)
+
+ # create new object and store in global map
+ self.formatMap[id] = Format(id, params, code)
+
+ def expandCpuSymbolsToDict(self, template):
+ '''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.'''
+
+ # Protect '%'s that don't go with CPU-specific terms
+ t = re.sub(r'%(?!\(CPU_)', '%%', template)
+ result = {}
+ for cpu in self.cpuModels:
+ result[cpu.name] = t % cpu.strings
+ return result
+
+ def expandCpuSymbolsToString(self, template):
+ '''*If* the template has CPU-specific references, return a
+ single string containing a copy of the template for each CPU
+ model with the corresponding values substituted in. If the
+ template has no CPU-specific references, it is returned
+ unmodified.'''
+
+ if template.find('%(CPU_') != -1:
+ return reduce(lambda x,y: x+y,
+ self.expandCpuSymbolsToDict(template).values())
else:
- print 'File', file, 'is unchanged'
- else:
- print 'Generating', file
- update = True
- if update:
- f = open(file, 'w')
- f.write(contents)
- f.close()
-
-# This regular expression matches '##include' directives
-includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[\w/.-]*)".*$',
- re.MULTILINE)
-
-# Function to replace a matched '##include' directive with the
-# contents of the specified file (with nested ##includes replaced
-# recursively). 'matchobj' is an re match object (from a match of
-# includeRE) and 'dirname' is the directory relative to which the file
-# path should be resolved.
-def replace_include(matchobj, dirname):
- fname = matchobj.group('filename')
- full_fname = os.path.normpath(os.path.join(dirname, fname))
- contents = '##newfile "%s"\n%s\n##endfile\n' % \
- (full_fname, read_and_flatten(full_fname))
- return contents
-
-# Read a file and recursively flatten nested '##include' files.
-def read_and_flatten(filename):
- current_dir = os.path.dirname(filename)
- try:
- contents = open(filename).read()
- except IOError:
- error(0, 'Error including file "%s"' % filename)
- fileNameStack.push((filename, 0))
- # Find any includes and include them
- contents = includeRE.sub(lambda m: replace_include(m, current_dir),
- contents)
- fileNameStack.pop()
- return contents
+ return template
-#
-# Read in and parse the ISA description.
-#
-def parse_isa_desc(isa_desc_file, output_dir):
- # Read file and (recursively) all included files into a string.
- # PLY requires that the input be in a single string so we have to
- # do this up front.
- isa_desc = read_and_flatten(isa_desc_file)
-
- # Initialize filename stack with outer file.
- fileNameStack.push((isa_desc_file, 0))
-
- # Parse it.
- (isa_name, namespace, global_code, namespace_code) = parser.parse(isa_desc)
-
- # grab the last three path components of isa_desc_file to put in
- # the output
- filename = '/'.join(isa_desc_file.split('/')[-3:])
-
- # generate decoder.hh
- includes = '#include "base/bitfield.hh" // for bitfield support'
- global_output = global_code.header_output
- namespace_output = namespace_code.header_output
- decode_function = ''
- update_if_needed(output_dir + '/decoder.hh', file_template % vars())
-
- # generate decoder.cc
- includes = '#include "decoder.hh"'
- global_output = global_code.decoder_output
- namespace_output = namespace_code.decoder_output
- # namespace_output += namespace_code.decode_block
- decode_function = namespace_code.decode_block
- update_if_needed(output_dir + '/decoder.cc', file_template % vars())
-
- # generate per-cpu exec files
- for cpu in cpu_models:
- includes = '#include "decoder.hh"\n'
- includes += cpu.includes
- global_output = global_code.exec_output[cpu.name]
- namespace_output = namespace_code.exec_output[cpu.name]
+ def protectCpuSymbols(self, template):
+ '''Protect CPU-specific references by doubling the
+ corresponding '%'s (in preparation for substituting a different
+ set of references into the template).'''
+
+ return re.sub(r'%(?=\(CPU_)', '%%', template)
+
+ def protectNonSubstPercents(self, s):
+ '''Protect any non-dict-substitution '%'s in a format string
+ (i.e. those not followed by '(')'''
+
+ return re.sub(r'%(?!\()', '%%', s)
+
+ def buildOperandNameMap(self, user_dict, lineno):
+ operand_name = {}
+ for op_name, val in user_dict.iteritems():
+ base_cls_name, dflt_ext, reg_spec, flags, sort_pri = val[:5]
+ if len(val) > 5:
+ read_code = val[5]
+ else:
+ read_code = None
+ if len(val) > 6:
+ write_code = val[6]
+ else:
+ write_code = None
+ if len(val) > 7:
+ error(lineno,
+ 'error: too many attributes for operand "%s"' %
+ base_cls_name)
+
+ # Canonical flag structure is a triple of lists, where each list
+ # indicates the set of flags implied by this operand always, when
+ # used as a source, and when used as a dest, respectively.
+ # For simplicity this can be initialized using a variety of fairly
+ # obvious shortcuts; we convert these to canonical form here.
+ if not flags:
+ # no flags specified (e.g., 'None')
+ flags = ( [], [], [] )
+ elif isinstance(flags, str):
+ # a single flag: assumed to be unconditional
+ flags = ( [ flags ], [], [] )
+ elif isinstance(flags, list):
+ # a list of flags: also assumed to be unconditional
+ flags = ( flags, [], [] )
+ elif isinstance(flags, tuple):
+ # it's a tuple: it should be a triple,
+ # but each item could be a single string or a list
+ (uncond_flags, src_flags, dest_flags) = flags
+ flags = (makeList(uncond_flags),
+ makeList(src_flags), makeList(dest_flags))
+ # Accumulate attributes of new operand class in tmp_dict
+ tmp_dict = {}
+ attrList = ['reg_spec', 'flags', 'sort_pri',
+ 'read_code', 'write_code']
+ if dflt_ext:
+ dflt_ctype = self.operandTypeMap[dflt_ext]
+ attrList.extend(['dflt_ctype', 'dflt_ext'])
+ for attr in attrList:
+ tmp_dict[attr] = eval(attr)
+ tmp_dict['base_name'] = op_name
+ # New class name will be e.g. "IntReg_Ra"
+ cls_name = base_cls_name + '_' + op_name
+ # Evaluate string arg to get class object. Note that the
+ # actual base class for "IntReg" is "IntRegOperand", i.e. we
+ # have to append "Operand".
+ try:
+ base_cls = eval(base_cls_name + 'Operand')
+ except NameError:
+ error(lineno,
+ 'error: unknown operand base class "%s"' % base_cls_name)
+ # The following statement creates a new class called
+ # <cls_name> as a subclass of <base_cls> with the attributes
+ # in tmp_dict, just as if we evaluated a class declaration.
+ operand_name[op_name] = type(cls_name, (base_cls,), tmp_dict)
+
+ self.operandNameMap = operand_name
+
+ # Define operand variables.
+ operands = user_dict.keys()
+ extensions = self.operandTypeMap.keys()
+
+ operandsREString = r'''
+ (?<!\w) # neg. lookbehind assertion: prevent partial matches
+ ((%s)(?:_(%s))?) # match: operand with optional '_' then suffix
+ (?!\w) # neg. lookahead assertion: prevent partial matches
+ ''' % (string.join(operands, '|'), string.join(extensions, '|'))
+
+ self.operandsRE = re.compile(operandsREString, re.MULTILINE|re.VERBOSE)
+
+ # Same as operandsREString, but extension is mandatory, and only two
+ # groups are returned (base and ext, not full name as above).
+ # Used for subtituting '_' for '.' to make C++ identifiers.
+ operandsWithExtREString = r'(?<!\w)(%s)_(%s)(?!\w)' \
+ % (string.join(operands, '|'), string.join(extensions, '|'))
+
+ self.operandsWithExtRE = \
+ re.compile(operandsWithExtREString, re.MULTILINE)
+
+ def substMungedOpNames(self, code):
+ '''Munge operand names in code string to make legal C++
+ variable names. This means getting rid of the type extension
+ if any. Will match base_name attribute of Operand object.)'''
+ return self.operandsWithExtRE.sub(r'\1', code)
+
+ def mungeSnippet(self, s):
+ '''Fix up code snippets for final substitution in templates.'''
+ if isinstance(s, str):
+ return self.substMungedOpNames(substBitOps(s))
+ else:
+ return s
+
+ def update_if_needed(self, file, contents):
+ '''Update the output file only if the new contents are
+ different from the current contents. Minimizes the files that
+ need to be rebuilt after minor changes.'''
+
+ file = os.path.join(self.output_dir, file)
+ update = False
+ if os.access(file, os.R_OK):
+ f = open(file, 'r')
+ old_contents = f.read()
+ f.close()
+ if contents != old_contents:
+ os.remove(file) # in case it's write-protected
+ update = True
+ else:
+ print 'File', file, 'is unchanged'
+ else:
+ update = True
+ if update:
+ f = open(file, 'w')
+ f.write(contents)
+ f.close()
+
+ # This regular expression matches '##include' directives
+ includeRE = re.compile(r'^\s*##include\s+"(?P<filename>[^"]*)".*$',
+ re.MULTILINE)
+
+ def replace_include(self, matchobj, dirname):
+ """Function to replace a matched '##include' directive with the
+ contents of the specified file (with nested ##includes
+ replaced recursively). 'matchobj' is an re match object
+ (from a match of includeRE) and 'dirname' is the directory
+ relative to which the file path should be resolved."""
+
+ fname = matchobj.group('filename')
+ full_fname = os.path.normpath(os.path.join(dirname, fname))
+ contents = '##newfile "%s"\n%s\n##endfile\n' % \
+ (full_fname, self.read_and_flatten(full_fname))
+ return contents
+
+ def read_and_flatten(self, filename):
+ """Read a file and recursively flatten nested '##include' files."""
+
+ current_dir = os.path.dirname(filename)
+ try:
+ contents = open(filename).read()
+ except IOError:
+ error('Error including file "%s"' % filename)
+
+ self.fileNameStack.push((filename, 0))
+
+ # Find any includes and include them
+ def replace(matchobj):
+ return self.replace_include(matchobj, current_dir)
+ contents = self.includeRE.sub(replace, contents)
+
+ self.fileNameStack.pop()
+ return contents
+
+ def _parse_isa_desc(self, isa_desc_file):
+ '''Read in and parse the ISA description.'''
+
+ # Read file and (recursively) all included files into a string.
+ # PLY requires that the input be in a single string so we have to
+ # do this up front.
+ isa_desc = self.read_and_flatten(isa_desc_file)
+
+ # Initialize filename stack with outer file.
+ self.fileNameStack.push((isa_desc_file, 0))
+
+ # Parse it.
+ (isa_name, namespace, global_code, namespace_code) = \
+ self.parse_string(isa_desc)
+
+ # grab the last three path components of isa_desc_file to put in
+ # the output
+ filename = '/'.join(isa_desc_file.split('/')[-3:])
+
+ # generate decoder.hh
+ includes = '#include "base/bitfield.hh" // for bitfield support'
+ global_output = global_code.header_output
+ namespace_output = namespace_code.header_output
decode_function = ''
- update_if_needed(output_dir + '/' + cpu.filename,
- file_template % vars())
-
- # The variable names here are hacky, but this will creat local variables
- # which will be referenced in vars() which have the value of the globals.
- global maxInstSrcRegs
- MaxInstSrcRegs = maxInstSrcRegs
- global maxInstDestRegs
- MaxInstDestRegs = maxInstDestRegs
- # max_inst_regs.hh
- update_if_needed(output_dir + '/max_inst_regs.hh', \
- max_inst_regs_template % vars())
-
-# global list of CpuModel objects (see cpu_models.py)
-cpu_models = []
+ self.update_if_needed('decoder.hh', file_template % vars())
+
+ # generate decoder.cc
+ includes = '#include "decoder.hh"'
+ global_output = global_code.decoder_output
+ namespace_output = namespace_code.decoder_output
+ # namespace_output += namespace_code.decode_block
+ decode_function = namespace_code.decode_block
+ self.update_if_needed('decoder.cc', file_template % vars())
+
+ # generate per-cpu exec files
+ for cpu in self.cpuModels:
+ includes = '#include "decoder.hh"\n'
+ includes += cpu.includes
+ global_output = global_code.exec_output[cpu.name]
+ namespace_output = namespace_code.exec_output[cpu.name]
+ decode_function = ''
+ self.update_if_needed(cpu.filename, file_template % vars())
+
+ # The variable names here are hacky, but this will creat local
+ # variables which will be referenced in vars() which have the
+ # value of the globals.
+ MaxInstSrcRegs = self.maxInstSrcRegs
+ MaxInstDestRegs = self.maxInstDestRegs
+ # max_inst_regs.hh
+ self.update_if_needed('max_inst_regs.hh',
+ max_inst_regs_template % vars())
+
+ def parse_isa_desc(self, *args, **kwargs):
+ try:
+ self._parse_isa_desc(*args, **kwargs)
+ except ISAParserError, e:
+ e.exit(self.fileNameStack)
# Called as script: get args from command line.
# Args are: <path to cpu_models.py> <isa desc file> <output dir> <cpu models>
if __name__ == '__main__':
execfile(sys.argv[1]) # read in CpuModel definitions
cpu_models = [CpuModel.dict[cpu] for cpu in sys.argv[4:]]
- parse_isa_desc(sys.argv[2], sys.argv[3])
+ ISAParser(sys.argv[3], cpu_models).parse_isa_desc(sys.argv[2])
projects / gem5.git / blobdiff