diff --git a/src/arch/isa_parser.py b/src/arch/isa_parser.py index a0d671da1a..4c8d0706d9 100755 --- a/src/arch/isa_parser.py +++ b/src/arch/isa_parser.py @@ -866,7 +866,11 @@ class Format: context = {} updateExportContext() context.update(exportContext) - context.update({ 'name': name, 'Name': string.capitalize(name) }) + 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: @@ -1028,7 +1032,7 @@ def substBitOps(code): # Template objects are format strings that allow substitution from # the attribute spaces of other objects (e.g. InstObjParams instances). -labelRE = re.compile(r'[^%]%\(([^\)]+)\)[sd]') +labelRE = re.compile(r'(?%d)(?=[^0-9]|$)" % opNum) - tag = opTags[opNum] - # Build up a name for this instructions class using the argument - # types. Each variation will get its own name this way. - postfix = '_' + tag + postfix - tagParser = re.compile(r"(?P[A-Z][A-Z]*)(?P[a-z][a-z]*)|(r(?P[A-Za-z0-9][A-Za-z0-9]*))") - tagMatch = tagParser.search(tag) - if tagMatch == None: - raise Exception, "Problem parsing operand tag %s" % tag - reg = tagMatch.group("tagReg") - tagType = tagMatch.group("tagType") - tagSize = tagMatch.group("tagSize") - if reg: - #Figure out what to do with fixed register operands - if reg in ("Ax", "Bx", "Cx", "Dx"): - code = opRe.sub("{INTREG_R%s}" % reg.upper(), code) - elif reg == "Al": - # We need a way to specify register width - code = opRe.sub("{INTREG_RAX}", code) - else: - print "Didn't know how to encode fixed register %s!" % reg - elif tagType == None or tagSize == None: - raise Exception, "Problem parsing operand tag: %s" % tag - elif tagType == "C" or tagType == "D" or tagType == "G" or \ - tagType == "P" or tagType == "S" or \ - tagType == "T" or tagType == "V": - # Use the "reg" field of the ModRM byte to select the register - code = opRe.sub("{(uint8_t)MODRM_REG}", code) - elif tagType == "E" or tagType == "Q" or tagType == "W": - # This might refer to memory or to a register. We need to - # divide it up farther. - regCode = opRe.sub("{(uint8_t)MODRM_RM}", code) - regTags = copy.copy(opTags) - regTags.pop(-1) - # This needs to refer to memory, but we'll fill in the details - # later. It needs to take into account unaligned memory - # addresses. - memCode = opRe.sub("0", code) - memTags = copy.copy(opTags) - memTags.pop(-1) - return doMultiOp(name, Name, doCompOps, "MODRM_MOD", - {"3" : (regCode, regTags, postfix)}, - (memCode, memTags, postfix)) - elif tagType == "I" or tagType == "J": - # Substitute in an immediate - code = opRe.sub("{IMMEDIATE}", code) - elif tagType == "M": - # This needs to refer to memory, but we'll fill in the details - # later. It needs to take into account unaligned memory - # addresses. - code = opRe.sub("0", code) - elif tagType == "PR" or tagType == "R" or tagType == "VR": - # There should probably be a check here to verify that mod - # is equal to 11b - code = opRe.sub("{(uint8_t)MODRM_RM}", code) - else: - raise Exception, "Unrecognized tag %s." % tag - opTags.pop(-1) - - # At this point, we've built up "code" to have all the necessary extra - # instructions needed to implement whatever types of operands were - # specified. Now we'll assemble it it into a microOp sequence. - ops = assembleMicro(code) - - # Build a macroop to contain the sequence of microops we've - # constructed. The decode block will be used to fill in our - # inner decode structure, and the rest will be concatenated and - # passed back. - return genInst(name, Name + postfix, ops) -}}; - -def format TaggedOp(code, tagSet) {{ +def format Inst(*opTypeSet) {{ (header_output, decoder_output, decode_block, - exec_output) = doCompOps(name, Name, code, tagSet, '') + exec_output) = doInst(name, Name, list(opTypeSet)) }}; -def format MultiOp(code, switchVal, opTags, *opt_flags) {{ +def format MultiInst(switchVal, *opTypeSets) {{ switcher = {} - for (count, tagSet) in zip(xrange(len(opTags) - 1), opTags): - switcher[count] = (code, tagSet, '') + for (count, opTypeSet) in zip(xrange(len(opTypeSets)), opTypeSets): + switcher[count] = (opTypeSet,) (header_output, decoder_output, decode_block, - exec_output) = doMultiOp(name, Name, doCompOps, switchVal, switcher) + exec_output) = doSplitDecode(name, Name, doInst, switchVal, switcher) }}; diff --git a/src/arch/x86/isa/main.isa b/src/arch/x86/isa/main.isa index fe1d4e5150..cc3a9bee4d 100644 --- a/src/arch/x86/isa/main.isa +++ b/src/arch/x86/isa/main.isa @@ -84,6 +84,9 @@ namespace X86ISA; //Include the base class for x86 instructions, and some support code ##include "base.isa" +//Include the instruction definitions +##include "insts/insts.isa" + //Include the definitions for the instruction formats ##include "formats/formats.isa" diff --git a/src/arch/x86/isa/microasm.isa b/src/arch/x86/isa/microasm.isa index 711ebf6672..b94b55aab5 100644 --- a/src/arch/x86/isa/microasm.isa +++ b/src/arch/x86/isa/microasm.isa @@ -57,11 +57,153 @@ //////////////////////////////////////////////////////////////////// // -// Code to "assemble" microcode sequences +// Code to "specialize" a microcode sequence to use a particular +// variety of operands // let {{ - class MicroOpStatement: + # This builds either a regular or macro op to implement the sequence of + # ops we give it. + def genInst(name, Name, ops): + # If we can implement this instruction with exactly one microop, just + # use that directly. + newStmnt = '' + if len(ops) == 1: + decode_block = "return (X86StaticInst *)(%s);" % \ + ops[0].getAllocator() + return ('', '', decode_block, '') + else: + # Build a macroop to contain the sequence of microops we've + # been given. + return genMacroOp(name, Name, ops) +}}; + +let {{ + # This code builds up a decode block which decodes based on switchval. + # vals is a dict which matches case values with what should be decoded to. + # builder is called on the exploded contents of "vals" values to generate + # whatever code should be used. + def doSplitDecode(name, Name, builder, switchVal, vals, default = None): + header_output = '' + decoder_output = '' + decode_block = 'switch(%s) {\n' % switchVal + exec_output = '' + for (val, todo) in vals.items(): + (new_header_output, + new_decoder_output, + new_decode_block, + new_exec_output) = builder(name, Name, *todo) + header_output += new_header_output + decoder_output += new_decoder_output + decode_block += '\tcase %s: %s\n' % (val, new_decode_block) + exec_output += new_exec_output + if default: + (new_header_output, + new_decoder_output, + new_decode_block, + new_exec_output) = builder(name, Name, *default) + header_output += new_header_output + decoder_output += new_decoder_output + decode_block += '\tdefault: %s\n' % new_decode_block + exec_output += new_exec_output + decode_block += '}\n' + return (header_output, decoder_output, decode_block, exec_output) +}}; + +let {{ + class OpType(object): + parser = re.compile(r"(?P[A-Z][A-Z]*)(?P[a-z][a-z]*)|(r(?P[A-Za-z0-9][A-Za-z0-9]*))") + def __init__(self, opTypeString): + match = OpType.parser.search(opTypeString) + if match == None: + raise Exception, "Problem parsing operand type %s" % opTypeString + self.reg = match.group("reg") + self.tag = match.group("tag") + self.size = match.group("size") +}}; + +let {{ + + # This function specializes the given piece of code to use a particular + # set of argument types described by "opTypes". These are "implemented" + # in reverse order. + def specializeInst(name, Name, code, opTypes): + opNum = len(opTypes) - 1 + while len(opTypes): + # print "Building a composite op with tags", opTypes + # print "And code", code + opNum = len(opTypes) - 1 + # A regular expression to find the operand placeholders we're + # interested in. + opRe = re.compile("\\^(?P%d)(?=[^0-9]|$)" % opNum) + + # Parse the operand type strign we're working with + opType = OpType(opTypes[opNum]) + + if opType.reg: + #Figure out what to do with fixed register operands + if opType.reg in ("Ax", "Bx", "Cx", "Dx"): + code = opRe.sub("%%{INTREG_R%s}" % opType.reg.upper(), code) + elif opType.reg == "Al": + # We need a way to specify register width + code = opRe.sub("%{INTREG_RAX}", code) + else: + print "Didn't know how to encode fixed register %s!" % opType.reg + elif opType.tag == None or opType.size == None: + raise Exception, "Problem parsing operand tag: %s" % opType.tag + elif opType.tag in ("C", "D", "G", "P", "S", "T", "V"): + # Use the "reg" field of the ModRM byte to select the register + code = opRe.sub("%{(uint8_t)MODRM_REG}", code) + elif opType.tag in ("E", "Q", "W"): + # This might refer to memory or to a register. We need to + # divide it up farther. + regCode = opRe.sub("%{(uint8_t)MODRM_RM}", code) + regTypes = copy.copy(opTypes) + regTypes.pop(-1) + # This needs to refer to memory, but we'll fill in the details + # later. It needs to take into account unaligned memory + # addresses. + memCode = opRe.sub("%0", code) + memTypes = copy.copy(opTypes) + memTypes.pop(-1) + return doSplitDecode(name, Name, specializeInst, "MODRM_MOD", + {"3" : (regCode, regTypes)}, (memCode, memTypes)) + elif opType.tag in ("I", "J"): + # Immediates are already in the instruction, so don't leave in + # those parameters + code = opRe.sub("${IMMEDIATE}", code) + elif opType.tag == "M": + # This needs to refer to memory, but we'll fill in the details + # later. It needs to take into account unaligned memory + # addresses. + code = opRe.sub("%0", code) + elif opType.tag in ("PR", "R", "VR"): + # There should probably be a check here to verify that mod + # is equal to 11b + code = opRe.sub("%{(uint8_t)MODRM_RM}", code) + else: + raise Exception, "Unrecognized tag %s." % opType.tag + opTypes.pop(-1) + + # At this point, we've built up "code" to have all the necessary extra + # instructions needed to implement whatever types of operands were + # specified. Now we'll assemble it it into a microOp sequence. + ops = assembleMicro(code) + + # Build a macroop to contain the sequence of microops we've + # constructed. The decode block will be used to fill in our + # inner decode structure, and the rest will be concatenated and + # passed back. + return genInst(name, Name, ops) +}}; + +//////////////////////////////////////////////////////////////////// +// +// The microcode assembler +// + +let {{ + class MicroOpStatement(object): def __init__(self): self.className = '' self.label = '' @@ -80,16 +222,24 @@ let {{ def getAllocator(self, *microFlags): args = '' + signature = "<" + emptySig = True for arg in self.args: - if arg.has_key("operandConst"): - args += ", %s" % arg["operandConst"] - elif arg.has_key("operandCode"): - args += ", %s" % arg["operandCode"] + if not emptySig: + signature += ", " + emptySig = False + if arg.has_key("operandImm"): + args += ", %s" % arg["operandImm"] + signature += ImmOpType + elif arg.has_key("operandReg"): + args += ", %s" % arg["operandReg"] + signature += RegOpType elif arg.has_key("operandLabel"): raise Exception, "Found a label while creating allocator string." else: raise Exception, "Unrecognized operand type." - return 'new %s(machInst%s%s)' % (self.className, self.microFlagsText(microFlags), args) + signature += ">" + return 'new %s%s(machInst%s%s)' % (self.className, signature, self.microFlagsText(microFlags), args) }}; let {{ @@ -101,7 +251,9 @@ let {{ labels[op.label] = count micropc += 1 return labels +}}; +let{{ def assembleMicro(code): # This function takes in a block of microcode assembly and returns # a python list of objects which describe it. @@ -115,7 +267,7 @@ let {{ # time. Each expression expects the thing it's looking for to be at # the beginning of the line, so the previous component is stripped # before continuing. - labelRe = re.compile(r'^[ \t]*(?P