From 1aa9698fa00e8ffce9b8d3c90b3bd76c3c9e950e Mon Sep 17 00:00:00 2001 From: Gabe Black Date: Sun, 13 Feb 2011 17:42:05 -0800 Subject: X86: Define fault objects to carry debug messages. These faults can panic/warn/warn_once, etc., instead of instructions doing that themselves directly. That way, instructions can be speculatively executed, and only if they're actually going to commit will their fault be invoked and the panic, etc., happen. --- src/arch/x86/isa/includes.isa | 1 + src/arch/x86/isa/microops/debug.isa | 67 +++++++++++++++++-------------------- 2 files changed, 31 insertions(+), 37 deletions(-) (limited to 'src/arch/x86/isa') diff --git a/src/arch/x86/isa/includes.isa b/src/arch/x86/isa/includes.isa index 58b1fbc62..32708043e 100644 --- a/src/arch/x86/isa/includes.isa +++ b/src/arch/x86/isa/includes.isa @@ -53,6 +53,7 @@ output header {{ #include #include +#include "arch/generic/debugfaults.hh" #include "arch/x86/emulenv.hh" #include "arch/x86/insts/macroop.hh" #include "arch/x86/insts/microfpop.hh" diff --git a/src/arch/x86/isa/microops/debug.isa b/src/arch/x86/isa/microops/debug.isa index 4b2ecdd5a..c2735565d 100644 --- a/src/arch/x86/isa/microops/debug.isa +++ b/src/arch/x86/isa/microops/debug.isa @@ -45,16 +45,29 @@ output header {{ class MicroDebugBase : public X86ISA::X86MicroopBase { protected: + typedef GenericISA::M5DebugFault::DebugFunc DebugFunc; + DebugFunc func; std::string message; uint8_t cc; public: - MicroDebugBase(ExtMachInst _machInst, const char * mnem, + MicroDebugBase(ExtMachInst machInst, const char * mnem, const char * instMnem, uint64_t setFlags, - std::string _message, uint8_t _cc); + DebugFunc _func, std::string _message, uint8_t _cc) : + X86MicroopBase(machInst, mnem, instMnem, setFlags, No_OpClass), + func(_func), message(_message), cc(_cc) + {} - std::string generateDisassembly(Addr pc, - const SymbolTable *symtab) const; + std::string + generateDisassembly(Addr pc, const SymbolTable *symtab) const + { + std::stringstream response; + + printMnemonic(response, instMnem, mnemonic); + response << "\"" << message << "\""; + + return response.str(); + } }; }}; @@ -70,53 +83,31 @@ def template MicroDebugDeclare {{ }}; def template MicroDebugExecute {{ - Fault %(class_name)s::execute(%(CPU_exec_context)s *xc, + Fault + %(class_name)s::execute(%(CPU_exec_context)s *xc, Trace::InstRecord *traceData) const { %(op_decl)s %(op_rd)s if (%(cond_test)s) { - %(func)s("%s\n", message); + return new GenericISA::M5DebugFault(func, message); + } else { + return NoFault; } - return NoFault; } }}; -output decoder {{ - inline MicroDebugBase::MicroDebugBase( - ExtMachInst machInst, const char * mnem, const char * instMnem, - uint64_t setFlags, std::string _message, uint8_t _cc) : - X86MicroopBase(machInst, mnem, instMnem, - setFlags, No_OpClass), - message(_message), cc(_cc) - { - } -}}; - def template MicroDebugConstructor {{ inline %(class_name)s::%(class_name)s( ExtMachInst machInst, const char * instMnem, uint64_t setFlags, std::string _message, uint8_t _cc) : %(base_class)s(machInst, "%(func)s", instMnem, - setFlags, _message, _cc) + setFlags, %(func_num)s, _message, _cc) { %(constructor)s; } }}; -output decoder {{ - std::string MicroDebugBase::generateDisassembly(Addr pc, - const SymbolTable *symtab) const - { - std::stringstream response; - - printMnemonic(response, instMnem, mnemonic); - response << "\"" << message << "\""; - - return response.str(); - } -}}; - let {{ class MicroDebug(X86Microop): def __init__(self, message, flags=None): @@ -142,13 +133,14 @@ let {{ header_output = "" decoder_output = "" - def buildDebugMicro(func): + def buildDebugMicro(func, func_num): global exec_output, header_output, decoder_output iop = InstObjParams(func, "Micro%sFlags" % func.capitalize(), "MicroDebugBase", {"code": "", "func": func, + "func_num": "GenericISA::M5DebugFault::%s" % func_num, "cond_test": "checkCondition(ccFlagBits, cc)"}) exec_output += MicroDebugExecute.subst(iop) header_output += MicroDebugDeclare.subst(iop) @@ -158,6 +150,7 @@ let {{ "MicroDebugBase", {"code": "", "func": func, + "func_num": "GenericISA::M5DebugFault::%s" % func_num, "cond_test": "true"}) exec_output += MicroDebugExecute.subst(iop) header_output += MicroDebugDeclare.subst(iop) @@ -169,8 +162,8 @@ let {{ global microopClasses microopClasses[func] = MicroDebugChild - buildDebugMicro("panic") - buildDebugMicro("fatal") - buildDebugMicro("warn") - buildDebugMicro("warn_once") + buildDebugMicro("panic", "PanicFunc") + buildDebugMicro("fatal", "FatalFunc") + buildDebugMicro("warn", "WarnFunc") + buildDebugMicro("warn_once", "WarnOnceFunc") }}; -- cgit v1.2.3 From 4e1adf85f77edf761466af3568576d3f9134a14c Mon Sep 17 00:00:00 2001 From: Gabe Black Date: Sun, 13 Feb 2011 17:44:24 -0800 Subject: X86: Don't read in dest regs if all bits are replaced. In x86, 32 and 64 bit writes to registers in which registers appear to be 32 or 64 bits wide overwrite all bits of the destination register. This change removes false dependencies in these cases where the previous value of a register doesn't need to be read to write a new value. New versions of most microops are created that have a "Big" suffix which simply overwrite their destination, and the right version to use is selected during microop allocation based on the selected data size. This does not change the performance of the O3 CPU model significantly, I assume because there are other false dependencies from the condition code bits in the flags register. --- src/arch/x86/isa/microops/ldstop.isa | 80 +++++++++-- src/arch/x86/isa/microops/limmop.isa | 27 ++-- src/arch/x86/isa/microops/regop.isa | 269 ++++++++++++++++++++++++----------- 3 files changed, 267 insertions(+), 109 deletions(-) (limited to 'src/arch/x86/isa') diff --git a/src/arch/x86/isa/microops/ldstop.isa b/src/arch/x86/isa/microops/ldstop.isa index 216a74c6c..cd649d644 100644 --- a/src/arch/x86/isa/microops/ldstop.isa +++ b/src/arch/x86/isa/microops/ldstop.isa @@ -301,6 +301,46 @@ let {{ "dataSize" : self.dataSize, "addressSize" : self.addressSize, "memFlags" : self.memFlags} return allocator + + class BigLdStOp(X86Microop): + def __init__(self, data, segment, addr, disp, + dataSize, addressSize, baseFlags, atCPL0, prefetch): + self.data = data + [self.scale, self.index, self.base] = addr + self.disp = disp + self.segment = segment + self.dataSize = dataSize + self.addressSize = addressSize + self.memFlags = baseFlags + if atCPL0: + self.memFlags += " | (CPL0FlagBit << FlagShift)" + if prefetch: + self.memFlags += " | Request::PREFETCH" + self.memFlags += " | (machInst.legacy.addr ? " + \ + "(AddrSizeFlagBit << FlagShift) : 0)" + + def getAllocator(self, microFlags): + allocString = ''' + (%(dataSize)s >= 4) ? + (StaticInstPtr)(new %(class_name)sBig(machInst, + macrocodeBlock, %(flags)s, %(scale)s, %(index)s, + %(base)s, %(disp)s, %(segment)s, %(data)s, + %(dataSize)s, %(addressSize)s, %(memFlags)s)) : + (StaticInstPtr)(new %(class_name)s(machInst, + macrocodeBlock, %(flags)s, %(scale)s, %(index)s, + %(base)s, %(disp)s, %(segment)s, %(data)s, + %(dataSize)s, %(addressSize)s, %(memFlags)s)) + ''' + allocator = allocString % { + "class_name" : self.className, + "flags" : self.microFlagsText(microFlags), + "scale" : self.scale, "index" : self.index, + "base" : self.base, + "disp" : self.disp, + "segment" : self.segment, "data" : self.data, + "dataSize" : self.dataSize, "addressSize" : self.addressSize, + "memFlags" : self.memFlags} + return allocator }}; let {{ @@ -315,7 +355,8 @@ let {{ EA = bits(SegBase + scale * Index + Base + disp, addressSize * 8 - 1, 0); ''' - def defineMicroLoadOp(mnemonic, code, mem_flags="0"): + def defineMicroLoadOp(mnemonic, code, bigCode='', + mem_flags="0", big=True): global header_output global decoder_output global exec_output @@ -324,16 +365,22 @@ let {{ name = mnemonic.lower() # Build up the all register version of this micro op - iop = InstObjParams(name, Name, 'X86ISA::LdStOp', - {"code": code, - "ea_code": calculateEA}) - header_output += MicroLdStOpDeclare.subst(iop) - decoder_output += MicroLdStOpConstructor.subst(iop) - exec_output += MicroLoadExecute.subst(iop) - exec_output += MicroLoadInitiateAcc.subst(iop) - exec_output += MicroLoadCompleteAcc.subst(iop) - - class LoadOp(LdStOp): + iops = [InstObjParams(name, Name, 'X86ISA::LdStOp', + {"code": code, "ea_code": calculateEA})] + if big: + iops += [InstObjParams(name, Name + "Big", 'X86ISA::LdStOp', + {"code": bigCode, "ea_code": calculateEA})] + for iop in iops: + header_output += MicroLdStOpDeclare.subst(iop) + decoder_output += MicroLdStOpConstructor.subst(iop) + exec_output += MicroLoadExecute.subst(iop) + exec_output += MicroLoadInitiateAcc.subst(iop) + exec_output += MicroLoadCompleteAcc.subst(iop) + + base = LdStOp + if big: + base = BigLdStOp + class LoadOp(base): def __init__(self, data, segment, addr, disp = 0, dataSize="env.dataSize", addressSize="env.addressSize", @@ -346,12 +393,15 @@ let {{ microopClasses[name] = LoadOp - defineMicroLoadOp('Ld', 'Data = merge(Data, Mem, dataSize);') + defineMicroLoadOp('Ld', 'Data = merge(Data, Mem, dataSize);', + 'Data = Mem & mask(dataSize * 8);') defineMicroLoadOp('Ldst', 'Data = merge(Data, Mem, dataSize);', - '(StoreCheck << FlagShift)') + 'Data = Mem & mask(dataSize * 8);', + '(StoreCheck << FlagShift)') defineMicroLoadOp('Ldstl', 'Data = merge(Data, Mem, dataSize);', - '(StoreCheck << FlagShift) | Request::LOCKED') - defineMicroLoadOp('Ldfp', 'FpData.uqw = Mem;') + 'Data = Mem & mask(dataSize * 8);', + '(StoreCheck << FlagShift) | Request::LOCKED') + defineMicroLoadOp('Ldfp', 'FpData.uqw = Mem;', big = False) def defineMicroStoreOp(mnemonic, code, \ postCode="", completeCode="", mem_flags="0"): diff --git a/src/arch/x86/isa/microops/limmop.isa b/src/arch/x86/isa/microops/limmop.isa index 2871d5a89..ac78b090d 100644 --- a/src/arch/x86/isa/microops/limmop.isa +++ b/src/arch/x86/isa/microops/limmop.isa @@ -114,8 +114,16 @@ let {{ self.dataSize = dataSize def getAllocator(self, microFlags): - allocator = '''new %(class_name)s(machInst, macrocodeBlock, - %(flags)s, %(dest)s, %(imm)s, %(dataSize)s)''' % { + allocString = ''' + (%(dataSize)s >= 4) ? + (StaticInstPtr)(new %(class_name)sBig(machInst, + macrocodeBlock, %(flags)s, %(dest)s, %(imm)s, + %(dataSize)s)) : + (StaticInstPtr)(new %(class_name)s(machInst, + macrocodeBlock, %(flags)s, %(dest)s, %(imm)s, + %(dataSize)s)) + ''' + allocator = allocString % { "class_name" : self.className, "mnemonic" : self.mnemonic, "flags" : self.microFlagsText(microFlags), @@ -152,12 +160,15 @@ let {{ let {{ # Build up the all register version of this micro op - iop = InstObjParams("limm", "Limm", 'X86MicroopBase', - {"code" : "DestReg = merge(DestReg, imm, dataSize);"}) - header_output += MicroLimmOpDeclare.subst(iop) - decoder_output += MicroLimmOpConstructor.subst(iop) - decoder_output += MicroLimmOpDisassembly.subst(iop) - exec_output += MicroLimmOpExecute.subst(iop) + iops = [InstObjParams("limm", "Limm", 'X86MicroopBase', + {"code" : "DestReg = merge(DestReg, imm, dataSize);"}), + InstObjParams("limm", "LimmBig", 'X86MicroopBase', + {"code" : "DestReg = imm & mask(dataSize * 8);"})] + for iop in iops: + header_output += MicroLimmOpDeclare.subst(iop) + decoder_output += MicroLimmOpConstructor.subst(iop) + decoder_output += MicroLimmOpDisassembly.subst(iop) + exec_output += MicroLimmOpExecute.subst(iop) iop = InstObjParams("lfpimm", "Lfpimm", 'X86MicroopBase', {"code" : "FpDestReg.uqw = imm"}) diff --git a/src/arch/x86/isa/microops/regop.isa b/src/arch/x86/isa/microops/regop.isa index ccfcb3a69..158bfdd59 100644 --- a/src/arch/x86/isa/microops/regop.isa +++ b/src/arch/x86/isa/microops/regop.isa @@ -224,8 +224,8 @@ let {{ MicroRegOpExecute) class RegOpMeta(type): - def buildCppClasses(self, name, Name, suffix, \ - code, flag_code, cond_check, else_code, cond_control_flag_init): + def buildCppClasses(self, name, Name, suffix, code, big_code, \ + flag_code, cond_check, else_code, cond_control_flag_init): # Globals to stick the output in global header_output @@ -235,11 +235,13 @@ let {{ # Stick all the code together so it can be searched at once allCode = "|".join((code, flag_code, cond_check, else_code, cond_control_flag_init)) + allBigCode = "|".join((big_code, flag_code, cond_check, else_code, + cond_control_flag_init)) # If op2 is used anywhere, make register and immediate versions # of this code. matcher = re.compile("(?s?)op2(?P\\.\\w+)?") - match = matcher.search(allCode) + match = matcher.search(allCode + allBigCode) if match: typeQual = "" if match.group("typeQual"): @@ -247,6 +249,7 @@ let {{ src2_name = "%spsrc2%s" % (match.group("prefix"), typeQual) self.buildCppClasses(name, Name, suffix, matcher.sub(src2_name, code), + matcher.sub(src2_name, big_code), matcher.sub(src2_name, flag_code), matcher.sub(src2_name, cond_check), matcher.sub(src2_name, else_code), @@ -254,6 +257,7 @@ let {{ imm_name = "%simm8" % match.group("prefix") self.buildCppClasses(name + "i", Name, suffix + "Imm", matcher.sub(imm_name, code), + matcher.sub(imm_name, big_code), matcher.sub(imm_name, flag_code), matcher.sub(imm_name, cond_check), matcher.sub(imm_name, else_code), @@ -264,27 +268,32 @@ let {{ # a version without it and fix up this version to use it. if flag_code != "" or cond_check != "true": self.buildCppClasses(name, Name, suffix, - code, "", "true", else_code, "") + code, big_code, "", "true", else_code, "") suffix = "Flags" + suffix # If psrc1 or psrc2 is used, we need to actually insert code to # compute it. - matcher = re.compile("(?= 4) ? + (StaticInstPtr)(new %(class_name)sBig(machInst, + macrocodeBlock, %(flags)s, %(src1)s, %(op2)s, + %(dest)s, %(dataSize)s, %(ext)s)) : + (StaticInstPtr)(new %(class_name)s(machInst, + macrocodeBlock, %(flags)s, %(src1)s, %(op2)s, + %(dest)s, %(dataSize)s, %(ext)s)) + ''' + allocator = allocString % { + "class_name" : className, + "flags" : self.microFlagsText(microFlags), + "src1" : self.src1, "op2" : self.op2, + "dest" : self.dest, + "dataSize" : self.dataSize, + "ext" : self.ext} + return allocator + else: + className = self.className + if self.mnemonic == self.base_mnemonic + 'i': + className += "Imm" + allocator = '''new %(class_name)s(machInst, macrocodeBlock, + %(flags)s, %(src1)s, %(op2)s, %(dest)s, + %(dataSize)s, %(ext)s)''' % { + "class_name" : className, + "flags" : self.microFlagsText(microFlags), + "src1" : self.src1, "op2" : self.op2, + "dest" : self.dest, + "dataSize" : self.dataSize, + "ext" : self.ext} + return allocator class LogicRegOp(RegOp): abstract = True @@ -429,30 +472,43 @@ let {{ class Add(FlagRegOp): code = 'DestReg = merge(DestReg, psrc1 + op2, dataSize);' + big_code = 'DestReg = (psrc1 + op2) & mask(dataSize * 8);' class Or(LogicRegOp): code = 'DestReg = merge(DestReg, psrc1 | op2, dataSize);' + big_code = 'DestReg = (psrc1 | op2) & mask(dataSize * 8);' class Adc(FlagRegOp): code = ''' CCFlagBits flags = ccFlagBits; DestReg = merge(DestReg, psrc1 + op2 + flags.cf, dataSize); ''' + big_code = ''' + CCFlagBits flags = ccFlagBits; + DestReg = (psrc1 + op2 + flags.cf) & mask(dataSize * 8); + ''' class Sbb(SubRegOp): code = ''' CCFlagBits flags = ccFlagBits; DestReg = merge(DestReg, psrc1 - op2 - flags.cf, dataSize); ''' + big_code = ''' + CCFlagBits flags = ccFlagBits; + DestReg = (psrc1 - op2 - flags.cf) & mask(dataSize * 8); + ''' class And(LogicRegOp): code = 'DestReg = merge(DestReg, psrc1 & op2, dataSize)' + big_code = 'DestReg = (psrc1 & op2) & mask(dataSize * 8)' class Sub(SubRegOp): code = 'DestReg = merge(DestReg, psrc1 - op2, dataSize)' + big_code = 'DestReg = (psrc1 - op2) & mask(dataSize * 8)' class Xor(LogicRegOp): code = 'DestReg = merge(DestReg, psrc1 ^ op2, dataSize)' + big_code = 'DestReg = (psrc1 ^ op2) & mask(dataSize * 8)' class Mul1s(WrRegOp): code = ''' @@ -505,6 +561,7 @@ let {{ class Mulel(RdRegOp): code = 'DestReg = merge(SrcReg1, ProdLow, dataSize);' + big_code = 'DestReg = ProdLow & mask(dataSize * 8);' class Muleh(RdRegOp): def __init__(self, dest, src1=None, flags=None, dataSize="env.dataSize"): @@ -513,6 +570,7 @@ let {{ super(RdRegOp, self).__init__(dest, src1, \ "InstRegIndex(NUM_INTREGS)", flags, dataSize) code = 'DestReg = merge(SrcReg1, ProdHi, dataSize);' + big_code = 'DestReg = ProdHi & mask(dataSize * 8);' # One or two bit divide class Div1(WrRegOp): @@ -540,7 +598,7 @@ let {{ # Step divide class Div2(RegOp): - code = ''' + divCode = ''' uint64_t dividend = Remainder; uint64_t divisor = Divisor; uint64_t quotient = Quotient; @@ -587,11 +645,13 @@ let {{ } } //Keep track of how many bits there are still to pull in. - DestReg = merge(DestReg, remaining, dataSize); + %s //Record the final results Remainder = remainder; Quotient = quotient; ''' + code = divCode % "DestReg = merge(DestReg, remaining, dataSize);" + big_code = divCode % "DestReg = remaining & mask(dataSize * 8);" flag_code = ''' if (remaining == 0) ccFlagBits = ccFlagBits | (ext & EZFBit); @@ -601,9 +661,11 @@ let {{ class Divq(RdRegOp): code = 'DestReg = merge(SrcReg1, Quotient, dataSize);' + big_code = 'DestReg = Quotient & mask(dataSize * 8);' class Divr(RdRegOp): code = 'DestReg = merge(SrcReg1, Remainder, dataSize);' + big_code = 'DestReg = Remainder & mask(dataSize * 8);' class Mov(CondRegOp): code = 'DestReg = merge(SrcReg1, op2, dataSize)' @@ -616,6 +678,10 @@ let {{ uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); DestReg = merge(DestReg, psrc1 << shiftAmt, dataSize); ''' + big_code = ''' + uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); + DestReg = (psrc1 << shiftAmt) & mask(dataSize * 8); + ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { @@ -641,14 +707,19 @@ let {{ ''' class Srl(RegOp): + # Because what happens to the bits shift -in- on a right shift + # is not defined in the C/C++ standard, we have to mask them out + # to be sure they're zero. code = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); - // Because what happens to the bits shift -in- on a right shift - // is not defined in the C/C++ standard, we have to mask them out - // to be sure they're zero. uint64_t logicalMask = mask(dataSize * 8 - shiftAmt); DestReg = merge(DestReg, (psrc1 >> shiftAmt) & logicalMask, dataSize); ''' + big_code = ''' + uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); + uint64_t logicalMask = mask(dataSize * 8 - shiftAmt); + DestReg = (psrc1 >> shiftAmt) & logicalMask; + ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { @@ -671,15 +742,21 @@ let {{ ''' class Sra(RegOp): + # Because what happens to the bits shift -in- on a right shift + # is not defined in the C/C++ standard, we have to sign extend + # them manually to be sure. code = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); - // Because what happens to the bits shift -in- on a right shift - // is not defined in the C/C++ standard, we have to sign extend - // them manually to be sure. uint64_t arithMask = (shiftAmt == 0) ? 0 : -bits(psrc1, dataSize * 8 - 1) << (dataSize * 8 - shiftAmt); DestReg = merge(DestReg, (psrc1 >> shiftAmt) | arithMask, dataSize); ''' + big_code = ''' + uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); + uint64_t arithMask = (shiftAmt == 0) ? 0 : + -bits(psrc1, dataSize * 8 - 1) << (dataSize * 8 - shiftAmt); + DestReg = ((psrc1 >> shiftAmt) | arithMask) & mask(dataSize * 8); + ''' flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { @@ -704,13 +781,11 @@ let {{ uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); uint8_t realShiftAmt = shiftAmt % (dataSize * 8); - if(realShiftAmt) - { + if (realShiftAmt) { uint64_t top = psrc1 << (dataSize * 8 - realShiftAmt); uint64_t bottom = bits(psrc1, dataSize * 8, realShiftAmt); DestReg = merge(DestReg, top | bottom, dataSize); - } - else + } else DestReg = merge(DestReg, DestReg, dataSize); ''' flag_code = ''' @@ -739,16 +814,14 @@ let {{ uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); uint8_t realShiftAmt = shiftAmt % (dataSize * 8 + 1); - if(realShiftAmt) - { + if (realShiftAmt) { CCFlagBits flags = ccFlagBits; uint64_t top = flags.cf << (dataSize * 8 - realShiftAmt); if (realShiftAmt > 1) top |= psrc1 << (dataSize * 8 - realShiftAmt + 1); uint64_t bottom = bits(psrc1, dataSize * 8 - 1, realShiftAmt); DestReg = merge(DestReg, top | bottom, dataSize); - } - else + } else DestReg = merge(DestReg, DestReg, dataSize); ''' flag_code = ''' @@ -780,14 +853,12 @@ let {{ uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); uint8_t realShiftAmt = shiftAmt % (dataSize * 8); - if(realShiftAmt) - { + if (realShiftAmt) { uint64_t top = psrc1 << realShiftAmt; uint64_t bottom = bits(psrc1, dataSize * 8 - 1, dataSize * 8 - realShiftAmt); DestReg = merge(DestReg, top | bottom, dataSize); - } - else + } else DestReg = merge(DestReg, DestReg, dataSize); ''' flag_code = ''' @@ -816,8 +887,7 @@ let {{ uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); uint8_t realShiftAmt = shiftAmt % (dataSize * 8 + 1); - if(realShiftAmt) - { + if (realShiftAmt) { CCFlagBits flags = ccFlagBits; uint64_t top = psrc1 << realShiftAmt; uint64_t bottom = flags.cf << (realShiftAmt - 1); @@ -826,8 +896,7 @@ let {{ bits(psrc1, dataSize * 8 - 1, dataSize * 8 - realShiftAmt + 1); DestReg = merge(DestReg, top | bottom, dataSize); - } - else + } else DestReg = merge(DestReg, DestReg, dataSize); ''' flag_code = ''' @@ -853,10 +922,10 @@ let {{ ''' class Sld(RegOp): - code = ''' + sldCode = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); uint8_t dataBits = dataSize * 8; - uint8_t realShiftAmt = shiftAmt % (2 * dataBits); + uint8_t realShiftAmt = shiftAmt %% (2 * dataBits); uint64_t result; if (realShiftAmt == 0) { result = psrc1; @@ -867,8 +936,10 @@ let {{ result = (DoubleBits << (realShiftAmt - dataBits)) | (psrc1 >> (2 * dataBits - realShiftAmt)); } - DestReg = merge(DestReg, result, dataSize); + %s ''' + code = sldCode % "DestReg = merge(DestReg, result, dataSize);" + big_code = sldCode % "DestReg = result & mask(dataSize * 8);" flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { @@ -899,10 +970,10 @@ let {{ ''' class Srd(RegOp): - code = ''' + srdCode = ''' uint8_t shiftAmt = (op2 & ((dataSize == 8) ? mask(6) : mask(5))); uint8_t dataBits = dataSize * 8; - uint8_t realShiftAmt = shiftAmt % (2 * dataBits); + uint8_t realShiftAmt = shiftAmt %% (2 * dataBits); uint64_t result; if (realShiftAmt == 0) { result = psrc1; @@ -919,8 +990,10 @@ let {{ logicalMask) | (psrc1 << (2 * dataBits - realShiftAmt)); } - DestReg = merge(DestReg, result, dataSize); + %s ''' + code = srdCode % "DestReg = merge(DestReg, result, dataSize);" + big_code = srdCode % "DestReg = result & mask(dataSize * 8);" flag_code = ''' // If the shift amount is zero, no flags should be modified. if (shiftAmt) { @@ -986,6 +1059,12 @@ let {{ ccFlagBits = (flag == 0) ? (ccFlagBits | EZFBit) : (ccFlagBits & ~EZFBit); ''' + big_code = ''' + int flag = bits(ccFlagBits, imm8); + DestReg = flag & mask(dataSize * 8); + ccFlagBits = (flag == 0) ? (ccFlagBits | EZFBit) : + (ccFlagBits & ~EZFBit); + ''' def __init__(self, dest, imm, flags=None, \ dataSize="env.dataSize"): super(Ruflag, self).__init__(dest, \ @@ -1000,6 +1079,14 @@ let {{ ccFlagBits = (flag == 0) ? (ccFlagBits | EZFBit) : (ccFlagBits & ~EZFBit); ''' + big_code = ''' + MiscReg flagMask = 0x3F7FDD5; + MiscReg flags = (nccFlagBits | ccFlagBits) & flagMask; + int flag = bits(flags, imm8); + DestReg = flag & mask(dataSize * 8); + ccFlagBits = (flag == 0) ? (ccFlagBits | EZFBit) : + (ccFlagBits & ~EZFBit); + ''' def __init__(self, dest, imm, flags=None, \ dataSize="env.dataSize"): super(Rflag, self).__init__(dest, \ @@ -1015,6 +1102,15 @@ let {{ val = sign_bit ? (val | ~maskVal) : (val & maskVal); DestReg = merge(DestReg, val, dataSize); ''' + big_code = ''' + IntReg val = psrc1; + // Mask the bit position so that it wraps. + int bitPos = op2 & (dataSize * 8 - 1); + int sign_bit = bits(val, bitPos, bitPos); + uint64_t maskVal = mask(bitPos+1); + val = sign_bit ? (val | ~maskVal) : (val & maskVal); + DestReg = val & mask(dataSize * 8); + ''' flag_code = ''' if (!sign_bit) ccFlagBits = ccFlagBits & @@ -1026,12 +1122,13 @@ let {{ class Zext(RegOp): code = 'DestReg = merge(DestReg, bits(psrc1, op2, 0), dataSize);' + big_code = 'DestReg = bits(psrc1, op2, 0) & mask(dataSize * 8);' class Rddr(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): super(Rddr, self).__init__(dest, \ src1, "InstRegIndex(NUM_INTREGS)", flags, dataSize) - code = ''' + rdrCode = ''' CR4 cr4 = CR4Op; DR7 dr7 = DR7Op; if ((cr4.de == 1 && (src1 == 4 || src1 == 5)) || src1 >= 8) { @@ -1039,9 +1136,11 @@ let {{ } else if (dr7.gd) { fault = new DebugException(); } else { - DestReg = merge(DestReg, DebugSrc1, dataSize); + %s } ''' + code = rdrCode % "DestReg = merge(DestReg, DebugSrc1, dataSize);" + big_code = rdrCode % "DestReg = DebugSrc1 & mask(dataSize * 8);" class Wrdr(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): @@ -1066,13 +1165,15 @@ let {{ def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): super(Rdcr, self).__init__(dest, \ src1, "InstRegIndex(NUM_INTREGS)", flags, dataSize) - code = ''' + rdcrCode = ''' if (src1 == 1 || (src1 > 4 && src1 < 8) || (src1 > 8)) { fault = new InvalidOpcode(); } else { - DestReg = merge(DestReg, ControlSrc1, dataSize); + %s } ''' + code = rdcrCode % "DestReg = merge(DestReg, ControlSrc1, dataSize);" + big_code = rdcrCode % "DestReg = ControlSrc1 & mask(dataSize * 8);" class Wrcr(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): @@ -1154,24 +1255,20 @@ let {{ ''' class Rdbase(SegOp): - code = ''' - DestReg = merge(DestReg, SegBaseSrc1, dataSize); - ''' + code = 'DestReg = merge(DestReg, SegBaseSrc1, dataSize);' + big_code = 'DestReg = SegBaseSrc1 & mask(dataSize * 8);' class Rdlimit(SegOp): - code = ''' - DestReg = merge(DestReg, SegLimitSrc1, dataSize); - ''' + code = 'DestReg = merge(DestReg, SegLimitSrc1, dataSize);' + big_code = 'DestReg = SegLimitSrc1 & mask(dataSize * 8);' class RdAttr(SegOp): - code = ''' - DestReg = merge(DestReg, SegAttrSrc1, dataSize); - ''' + code = 'DestReg = merge(DestReg, SegAttrSrc1, dataSize);' + big_code = 'DestReg = SegAttrSrc1 & mask(dataSize * 8);' class Rdsel(SegOp): - code = ''' - DestReg = merge(DestReg, SegSelSrc1, dataSize); - ''' + code = 'DestReg = merge(DestReg, SegSelSrc1, dataSize);' + big_code = 'DestReg = SegSelSrc1 & mask(dataSize * 8);' class Rdval(RegOp): def __init__(self, dest, src1, flags=None, dataSize="env.dataSize"): -- cgit v1.2.3 From bce2be525d0ff05af914515bc2d9e9ac9d66a123 Mon Sep 17 00:00:00 2001 From: Gabe Black Date: Sun, 13 Feb 2011 17:45:12 -0800 Subject: X86: Put the result used for flags in an intermediate variable. Using the destination register directly causes the ISA parser to treat it as a source even if none of the original bits are used. --- src/arch/x86/isa/includes.isa | 1 + src/arch/x86/isa/microops/regop.isa | 38 ++++++++++++++++++++----------------- 2 files changed, 22 insertions(+), 17 deletions(-) (limited to 'src/arch/x86/isa') diff --git a/src/arch/x86/isa/includes.isa b/src/arch/x86/isa/includes.isa index 32708043e..674e69e98 100644 --- a/src/arch/x86/isa/includes.isa +++ b/src/arch/x86/isa/includes.isa @@ -114,6 +114,7 @@ output exec {{ #include "arch/x86/regs/misc.hh" #include "arch/x86/tlb.hh" #include "base/bigint.hh" +#include "base/compiler.hh" #include "base/condcodes.hh" #include "cpu/base.hh" #include "cpu/exetrace.hh" diff --git a/src/arch/x86/isa/microops/regop.isa b/src/arch/x86/isa/microops/regop.isa index 158bfdd59..e2a51c127 100644 --- a/src/arch/x86/isa/microops/regop.isa +++ b/src/arch/x86/isa/microops/regop.isa @@ -51,6 +51,8 @@ def template MicroRegOpExecute {{ %(op_decl)s; %(op_rd)s; + IntReg result M5_VAR_USED; + if(%(cond_check)s) { %(code)s; @@ -79,6 +81,8 @@ def template MicroRegOpImmExecute {{ %(op_decl)s; %(op_rd)s; + IntReg result M5_VAR_USED; + if(%(cond_check)s) { %(code)s; @@ -434,7 +438,7 @@ let {{ flag_code = ''' //Don't have genFlags handle the OF or CF bits uint64_t mask = CFBit | ECFBit | OFBit; - ccFlagBits = genFlags(ccFlagBits, ext & ~mask, DestReg, psrc1, op2); + ccFlagBits = genFlags(ccFlagBits, ext & ~mask, result, psrc1, op2); //If a logic microop wants to set these, it wants to set them to 0. ccFlagBits &= ~(CFBit & ext); ccFlagBits &= ~(ECFBit & ext); @@ -444,12 +448,12 @@ let {{ class FlagRegOp(RegOp): abstract = True flag_code = \ - "ccFlagBits = genFlags(ccFlagBits, ext, DestReg, psrc1, op2);" + "ccFlagBits = genFlags(ccFlagBits, ext, result, psrc1, op2);" class SubRegOp(RegOp): abstract = True flag_code = \ - "ccFlagBits = genFlags(ccFlagBits, ext, DestReg, psrc1, ~op2, true);" + "ccFlagBits = genFlags(ccFlagBits, ext, result, psrc1, ~op2, true);" class CondRegOp(RegOp): abstract = True @@ -471,44 +475,44 @@ let {{ src1, src2, flags, dataSize) class Add(FlagRegOp): - code = 'DestReg = merge(DestReg, psrc1 + op2, dataSize);' - big_code = 'DestReg = (psrc1 + op2) & mask(dataSize * 8);' + code = 'DestReg = merge(DestReg, result = (psrc1 + op2), dataSize);' + big_code = 'DestReg = result = (psrc1 + op2) & mask(dataSize * 8);' class Or(LogicRegOp): - code = 'DestReg = merge(DestReg, psrc1 | op2, dataSize);' - big_code = 'DestReg = (psrc1 | op2) & mask(dataSize * 8);' + code = 'DestReg = merge(DestReg, result = (psrc1 | op2), dataSize);' + big_code = 'DestReg = result = (psrc1 | op2) & mask(dataSize * 8);' class Adc(FlagRegOp): code = ''' CCFlagBits flags = ccFlagBits; - DestReg = merge(DestReg, psrc1 + op2 + flags.cf, dataSize); + DestReg = merge(DestReg, result = (psrc1 + op2 + flags.cf), dataSize); ''' big_code = ''' CCFlagBits flags = ccFlagBits; - DestReg = (psrc1 + op2 + flags.cf) & mask(dataSize * 8); + DestReg = result = (psrc1 + op2 + flags.cf) & mask(dataSize * 8); ''' class Sbb(SubRegOp): code = ''' CCFlagBits flags = ccFlagBits; - DestReg = merge(DestReg, psrc1 - op2 - flags.cf, dataSize); + DestReg = merge(DestReg, result = (psrc1 - op2 - flags.cf), dataSize); ''' big_code = ''' CCFlagBits flags = ccFlagBits; - DestReg = (psrc1 - op2 - flags.cf) & mask(dataSize * 8); + DestReg = result = (psrc1 - op2 - flags.cf) & mask(dataSize * 8); ''' class And(LogicRegOp): - code = 'DestReg = merge(DestReg, psrc1 & op2, dataSize)' - big_code = 'DestReg = (psrc1 & op2) & mask(dataSize * 8)' + code = 'DestReg = merge(DestReg, result = (psrc1 & op2), dataSize)' + big_code = 'DestReg = result = (psrc1 & op2) & mask(dataSize * 8)' class Sub(SubRegOp): - code = 'DestReg = merge(DestReg, psrc1 - op2, dataSize)' - big_code = 'DestReg = (psrc1 - op2) & mask(dataSize * 8)' + code = 'DestReg = merge(DestReg, result = (psrc1 - op2), dataSize)' + big_code = 'DestReg = result = (psrc1 - op2) & mask(dataSize * 8)' class Xor(LogicRegOp): - code = 'DestReg = merge(DestReg, psrc1 ^ op2, dataSize)' - big_code = 'DestReg = (psrc1 ^ op2) & mask(dataSize * 8)' + code = 'DestReg = merge(DestReg, result = (psrc1 ^ op2), dataSize)' + big_code = 'DestReg = result = (psrc1 ^ op2) & mask(dataSize * 8)' class Mul1s(WrRegOp): code = ''' -- cgit v1.2.3 From fde8b5c3876ad431b193989ab64c802d1cec1ed0 Mon Sep 17 00:00:00 2001 From: Gabe Black Date: Tue, 15 Feb 2011 15:58:16 -0800 Subject: X86: Get rid of "inline" on the MicroPanic constructor in decoder.cc. This was making certain versions of gcc omit the function from the object file which would break the build. --- src/arch/x86/isa/microops/debug.isa | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'src/arch/x86/isa') diff --git a/src/arch/x86/isa/microops/debug.isa b/src/arch/x86/isa/microops/debug.isa index c2735565d..220c1af97 100644 --- a/src/arch/x86/isa/microops/debug.isa +++ b/src/arch/x86/isa/microops/debug.isa @@ -98,7 +98,7 @@ def template MicroDebugExecute {{ }}; def template MicroDebugConstructor {{ - inline %(class_name)s::%(class_name)s( + %(class_name)s::%(class_name)s( ExtMachInst machInst, const char * instMnem, uint64_t setFlags, std::string _message, uint8_t _cc) : %(base_class)s(machInst, "%(func)s", instMnem, -- cgit v1.2.3