diff options
Diffstat (limited to 'src/cpu')
36 files changed, 726 insertions, 657 deletions
diff --git a/src/cpu/base_dyn_inst.hh b/src/cpu/base_dyn_inst.hh index 9037c96df..515cd0836 100644 --- a/src/cpu/base_dyn_inst.hh +++ b/src/cpu/base_dyn_inst.hh @@ -39,6 +39,7 @@ #include "base/fast_alloc.hh" #include "base/trace.hh" #include "config/full_system.hh" +#include "cpu/o3/comm.hh" #include "cpu/exetrace.hh" #include "cpu/inst_seq.hh" #include "cpu/op_class.hh" @@ -62,10 +63,6 @@ class BaseDynInst : public FastAlloc, public RefCounted typedef typename Impl::CPUType ImplCPU; typedef typename ImplCPU::ImplState ImplState; - // Binary machine instruction type. - typedef TheISA::MachInst MachInst; - // Extended machine instruction type - typedef TheISA::ExtMachInst ExtMachInst; // Logical register index type. typedef TheISA::RegIndex RegIndex; // Integer register type. @@ -212,6 +209,7 @@ class BaseDynInst : public FastAlloc, public RefCounted /** PC of this instruction. */ Addr PC; + protected: /** Next non-speculative PC. It is not filled in at fetch, but rather * once the target of the branch is truly known (either decode or * execute). @@ -224,6 +222,14 @@ class BaseDynInst : public FastAlloc, public RefCounted /** Predicted next PC. */ Addr predPC; + /** Predicted next NPC. */ + Addr predNPC; + + /** If this is a branch that was predicted taken */ + bool predTaken; + + public: + /** Count of total number of dynamic instructions. */ static int instcount; @@ -236,16 +242,116 @@ class BaseDynInst : public FastAlloc, public RefCounted */ bool _readySrcRegIdx[MaxInstSrcRegs]; + protected: + /** Flattened register index of the destination registers of this + * instruction. + */ + TheISA::RegIndex _flatDestRegIdx[TheISA::MaxInstDestRegs]; + + /** Flattened register index of the source registers of this + * instruction. + */ + TheISA::RegIndex _flatSrcRegIdx[TheISA::MaxInstSrcRegs]; + + /** Physical register index of the destination registers of this + * instruction. + */ + PhysRegIndex _destRegIdx[TheISA::MaxInstDestRegs]; + + /** Physical register index of the source registers of this + * instruction. + */ + PhysRegIndex _srcRegIdx[TheISA::MaxInstSrcRegs]; + + /** Physical register index of the previous producers of the + * architected destinations. + */ + PhysRegIndex _prevDestRegIdx[TheISA::MaxInstDestRegs]; + public: + + /** Returns the physical register index of the i'th destination + * register. + */ + PhysRegIndex renamedDestRegIdx(int idx) const + { + return _destRegIdx[idx]; + } + + /** Returns the physical register index of the i'th source register. */ + PhysRegIndex renamedSrcRegIdx(int idx) const + { + return _srcRegIdx[idx]; + } + + /** Returns the flattened register index of the i'th destination + * register. + */ + TheISA::RegIndex flattenedDestRegIdx(int idx) const + { + return _flatDestRegIdx[idx]; + } + + /** Returns the flattened register index of the i'th source register */ + TheISA::RegIndex flattenedSrcRegIdx(int idx) const + { + return _flatSrcRegIdx[idx]; + } + + /** Returns the physical register index of the previous physical register + * that remapped to the same logical register index. + */ + PhysRegIndex prevDestRegIdx(int idx) const + { + return _prevDestRegIdx[idx]; + } + + /** Renames a destination register to a physical register. Also records + * the previous physical register that the logical register mapped to. + */ + void renameDestReg(int idx, + PhysRegIndex renamed_dest, + PhysRegIndex previous_rename) + { + _destRegIdx[idx] = renamed_dest; + _prevDestRegIdx[idx] = previous_rename; + } + + /** Renames a source logical register to the physical register which + * has/will produce that logical register's result. + * @todo: add in whether or not the source register is ready. + */ + void renameSrcReg(int idx, PhysRegIndex renamed_src) + { + _srcRegIdx[idx] = renamed_src; + } + + /** Flattens a source architectural register index into a logical index. + */ + void flattenSrcReg(int idx, TheISA::RegIndex flattened_src) + { + _flatSrcRegIdx[idx] = flattened_src; + } + + /** Flattens a destination architectural register index into a logical + * index. + */ + void flattenDestReg(int idx, TheISA::RegIndex flattened_dest) + { + _flatDestRegIdx[idx] = flattened_dest; + } + /** BaseDynInst constructor given a binary instruction. * @param inst The binary instruction. * @param PC The PC of the instruction. * @param pred_PC The predicted next PC. + * @param pred_NPC The predicted next NPC. * @param seq_num The sequence number of the instruction. * @param cpu Pointer to the instruction's CPU. */ - BaseDynInst(ExtMachInst inst, Addr PC, Addr pred_PC, InstSeqNum seq_num, - ImplCPU *cpu); + BaseDynInst(TheISA::ExtMachInst inst, Addr PC, Addr NPC, + Addr pred_PC, Addr pred_NPC, + InstSeqNum seq_num, ImplCPU *cpu); /** BaseDynInst constructor given a StaticInst pointer. * @param _staticInst The StaticInst for this BaseDynInst. @@ -287,29 +393,46 @@ class BaseDynInst : public FastAlloc, public RefCounted /** Returns the next NPC. This could be the speculative next NPC if it is * called prior to the actual branch target being calculated. */ - Addr readNextNPC() { return nextNPC; } + Addr readNextNPC() + { +#if ISA_HAS_DELAY_SLOT + return nextNPC; +#else + return nextPC + sizeof(TheISA::MachInst); +#endif + } /** Set the predicted target of this current instruction. */ - void setPredTarg(Addr predicted_PC) { predPC = predicted_PC; } + void setPredTarg(Addr predicted_PC, Addr predicted_NPC) + { + predPC = predicted_PC; + predNPC = predicted_NPC; + } + + /** Returns the predicted PC immediately after the branch. */ + Addr readPredPC() { return predPC; } - /** Returns the predicted target of the branch. */ - Addr readPredTarg() { return predPC; } + /** Returns the predicted PC two instructions after the branch */ + Addr readPredNPC() { return predNPC; } /** Returns whether the instruction was predicted taken or not. */ - bool predTaken() -#if ISA_HAS_DELAY_SLOT - { return predPC != (nextPC + sizeof(MachInst)); } -#else - { return predPC != (PC + sizeof(MachInst)); } -#endif + bool readPredTaken() + { + return predTaken; + } + + void setPredTaken(bool predicted_taken) + { + predTaken = predicted_taken; + } /** Returns whether the instruction mispredicted. */ bool mispredicted() -#if ISA_HAS_DELAY_SLOT - { return predPC != nextNPC; } -#else - { return predPC != nextPC; } -#endif + { + return readPredPC() != readNextPC() || + readPredNPC() != readNextNPC(); + } + // // Instruction types. Forward checks to StaticInst object. // diff --git a/src/cpu/base_dyn_inst_impl.hh b/src/cpu/base_dyn_inst_impl.hh index 2f6859de2..c3d71e428 100644 --- a/src/cpu/base_dyn_inst_impl.hh +++ b/src/cpu/base_dyn_inst_impl.hh @@ -62,17 +62,20 @@ my_hash_t thishash; #endif template <class Impl> -BaseDynInst<Impl>::BaseDynInst(TheISA::ExtMachInst machInst, Addr inst_PC, - Addr pred_PC, InstSeqNum seq_num, - ImplCPU *cpu) +BaseDynInst<Impl>::BaseDynInst(TheISA::ExtMachInst machInst, + Addr inst_PC, Addr inst_NPC, + Addr pred_PC, Addr pred_NPC, + InstSeqNum seq_num, ImplCPU *cpu) : staticInst(machInst), traceData(NULL), cpu(cpu) { seqNum = seq_num; PC = inst_PC; - nextPC = PC + sizeof(TheISA::MachInst); + nextPC = inst_NPC; nextNPC = nextPC + sizeof(TheISA::MachInst); predPC = pred_PC; + predNPC = pred_NPC; + predTaken = false; initVars(); } diff --git a/src/cpu/exetrace.cc b/src/cpu/exetrace.cc index 9ea90681c..bfd701271 100644 --- a/src/cpu/exetrace.cc +++ b/src/cpu/exetrace.cc @@ -147,13 +147,15 @@ Trace::InstRecord::dump(ostream &outs) outs << hex; outs << "PC = " << thread->readNextPC(); outs << " NPC = " << thread->readNextNPC(); - newVal = thread->readMiscReg(SparcISA::MISCREG_CCR); + newVal = thread->readIntReg(SparcISA::NumIntArchRegs + 2); + //newVal = thread->readMiscReg(SparcISA::MISCREG_CCR); if(newVal != ccr) { outs << " CCR = " << newVal; ccr = newVal; } - newVal = thread->readMiscReg(SparcISA::MISCREG_Y); + newVal = thread->readIntReg(SparcISA::NumIntArchRegs + 1); + //newVal = thread->readMiscReg(SparcISA::MISCREG_Y); if(newVal != y) { outs << " Y = " << newVal; @@ -403,9 +405,13 @@ Trace::InstRecord::dump(ostream &outs) diffHtba = true; if(shared_data->pstate != thread->readMiscReg(MISCREG_PSTATE)) diffPstate = true; - if(shared_data->y != thread->readMiscReg(MISCREG_Y)) + //if(shared_data->y != thread->readMiscReg(MISCREG_Y)) + if(shared_data->y != + thread->readIntReg(NumIntArchRegs + 1)) diffY = true; - if(shared_data->ccr != thread->readMiscReg(MISCREG_CCR)) + //if(shared_data->ccr != thread->readMiscReg(MISCREG_CCR)) + if(shared_data->ccr != + thread->readIntReg(NumIntArchRegs + 2)) diffCcr = true; if(shared_data->gl != thread->readMiscReg(MISCREG_GL)) diffGl = true; @@ -415,14 +421,22 @@ Trace::InstRecord::dump(ostream &outs) diffPil = true; if(shared_data->cwp != thread->readMiscReg(MISCREG_CWP)) diffCwp = true; - if(shared_data->cansave != thread->readMiscReg(MISCREG_CANSAVE)) + //if(shared_data->cansave != thread->readMiscReg(MISCREG_CANSAVE)) + if(shared_data->cansave != + thread->readIntReg(NumIntArchRegs + 3)) diffCansave = true; + //if(shared_data->canrestore != + // thread->readMiscReg(MISCREG_CANRESTORE)) if(shared_data->canrestore != - thread->readMiscReg(MISCREG_CANRESTORE)) + thread->readMiscReg(NumIntArchRegs + 4)) diffCanrestore = true; - if(shared_data->otherwin != thread->readMiscReg(MISCREG_OTHERWIN)) + //if(shared_data->otherwin != thread->readMiscReg(MISCREG_OTHERWIN)) + if(shared_data->otherwin != + thread->readIntReg(NumIntArchRegs + 5)) diffOtherwin = true; - if(shared_data->cleanwin != thread->readMiscReg(MISCREG_CLEANWIN)) + //if(shared_data->cleanwin != thread->readMiscReg(MISCREG_CLEANWIN)) + if(shared_data->cleanwin != + thread->readMiscReg(NumIntArchRegs + 6)) diffCleanwin = true; for (int i = 0; i < 64; i++) { @@ -538,10 +552,12 @@ Trace::InstRecord::dump(ostream &outs) thread->readMiscReg(MISCREG_PSTATE), shared_data->pstate); printRegPair(outs, "Y", - thread->readMiscReg(MISCREG_Y), + //thread->readMiscReg(MISCREG_Y), + thread->readMiscReg(NumIntArchRegs + 1), shared_data->y); printRegPair(outs, "Ccr", - thread->readMiscReg(MISCREG_CCR), + //thread->readMiscReg(MISCREG_CCR), + thread->readMiscReg(NumIntArchRegs + 2), shared_data->ccr); printRegPair(outs, "Tl", thread->readMiscReg(MISCREG_TL), @@ -559,16 +575,20 @@ Trace::InstRecord::dump(ostream &outs) thread->readMiscReg(MISCREG_CWP), shared_data->cwp); printRegPair(outs, "Cansave", - thread->readMiscReg(MISCREG_CANSAVE), + //thread->readMiscReg(MISCREG_CANSAVE), + thread->readIntReg(NumIntArchRegs + 3), shared_data->cansave); printRegPair(outs, "Canrestore", - thread->readMiscReg(MISCREG_CANRESTORE), + //thread->readMiscReg(MISCREG_CANRESTORE), + thread->readIntReg(NumIntArchRegs + 4), shared_data->canrestore); printRegPair(outs, "Otherwin", - thread->readMiscReg(MISCREG_OTHERWIN), + //thread->readMiscReg(MISCREG_OTHERWIN), + thread->readIntReg(NumIntArchRegs + 5), shared_data->otherwin); printRegPair(outs, "Cleanwin", - thread->readMiscReg(MISCREG_CLEANWIN), + //thread->readMiscReg(MISCREG_CLEANWIN), + thread->readIntReg(NumIntArchRegs + 6), shared_data->cleanwin); outs << endl; for (int i = 1; i <= MaxTL; i++) { diff --git a/src/cpu/o3/alpha/cpu.hh b/src/cpu/o3/alpha/cpu.hh index 0078db69f..4a2086296 100644 --- a/src/cpu/o3/alpha/cpu.hh +++ b/src/cpu/o3/alpha/cpu.hh @@ -37,12 +37,6 @@ #include "cpu/o3/cpu.hh" #include "sim/byteswap.hh" -namespace TheISA -{ - class ITB; - class DTB; -} - class EndQuiesceEvent; namespace Kernel { class Statistics; @@ -61,14 +55,6 @@ class TranslatingPort; template <class Impl> class AlphaO3CPU : public FullO3CPU<Impl> { - protected: - typedef TheISA::IntReg IntReg; - typedef TheISA::FloatReg FloatReg; - typedef TheISA::FloatRegBits FloatRegBits; - typedef TheISA::MiscReg MiscReg; - typedef TheISA::RegFile RegFile; - typedef TheISA::MiscRegFile MiscRegFile; - public: typedef O3ThreadState<Impl> ImplState; typedef O3ThreadState<Impl> Thread; @@ -77,13 +63,6 @@ class AlphaO3CPU : public FullO3CPU<Impl> /** Constructs an AlphaO3CPU with the given parameters. */ AlphaO3CPU(Params *params); -#if FULL_SYSTEM - /** ITB pointer. */ - AlphaISA::ITB *itb; - /** DTB pointer. */ - AlphaISA::DTB *dtb; -#endif - /** Registers statistics. */ void regStats(); @@ -91,19 +70,19 @@ class AlphaO3CPU : public FullO3CPU<Impl> /** Translates instruction requestion. */ Fault translateInstReq(RequestPtr &req, Thread *thread) { - return itb->translate(req, thread->getTC()); + return this->itb->translate(req, thread->getTC()); } /** Translates data read request. */ Fault translateDataReadReq(RequestPtr &req, Thread *thread) { - return dtb->translate(req, thread->getTC(), false); + return this->dtb->translate(req, thread->getTC(), false); } /** Translates data write request. */ Fault translateDataWriteReq(RequestPtr &req, Thread *thread) { - return dtb->translate(req, thread->getTC(), true); + return this->dtb->translate(req, thread->getTC(), true); } #else @@ -127,20 +106,22 @@ class AlphaO3CPU : public FullO3CPU<Impl> #endif /** Reads a miscellaneous register. */ - MiscReg readMiscReg(int misc_reg, unsigned tid); + TheISA::MiscReg readMiscReg(int misc_reg, unsigned tid); /** Reads a misc. register, including any side effects the read * might have as defined by the architecture. */ - MiscReg readMiscRegWithEffect(int misc_reg, unsigned tid); + TheISA::MiscReg readMiscRegWithEffect(int misc_reg, unsigned tid); /** Sets a miscellaneous register. */ - void setMiscReg(int misc_reg, const MiscReg &val, unsigned tid); + void setMiscReg(int misc_reg, const TheISA::MiscReg &val, + unsigned tid); /** Sets a misc. register, including any side effects the write * might have as defined by the architecture. */ - void setMiscRegWithEffect(int misc_reg, const MiscReg &val, unsigned tid); + void setMiscRegWithEffect(int misc_reg, const TheISA::MiscReg &val, + unsigned tid); /** Initiates a squash of all in-flight instructions for a given * thread. The source of the squash is an external update of @@ -175,10 +156,10 @@ class AlphaO3CPU : public FullO3CPU<Impl> */ void syscall(int64_t callnum, int tid); /** Gets a syscall argument. */ - IntReg getSyscallArg(int i, int tid); + TheISA::IntReg getSyscallArg(int i, int tid); /** Used to shift args for indirect syscall. */ - void setSyscallArg(int i, IntReg val, int tid); + void setSyscallArg(int i, TheISA::IntReg val, int tid); /** Sets the return value of a syscall. */ void setSyscallReturn(SyscallReturn return_value, int tid); diff --git a/src/cpu/o3/alpha/cpu_impl.hh b/src/cpu/o3/alpha/cpu_impl.hh index 980e70fdd..41f149963 100644 --- a/src/cpu/o3/alpha/cpu_impl.hh +++ b/src/cpu/o3/alpha/cpu_impl.hh @@ -55,12 +55,7 @@ #endif template <class Impl> -AlphaO3CPU<Impl>::AlphaO3CPU(Params *params) -#if FULL_SYSTEM - : FullO3CPU<Impl>(params), itb(params->itb), dtb(params->dtb) -#else - : FullO3CPU<Impl>(params) -#endif +AlphaO3CPU<Impl>::AlphaO3CPU(Params *params) : FullO3CPU<Impl>(params) { DPRINTF(O3CPU, "Creating AlphaO3CPU object.\n"); @@ -173,15 +168,16 @@ AlphaO3CPU<Impl>::readMiscRegWithEffect(int misc_reg, unsigned tid) template <class Impl> void -AlphaO3CPU<Impl>::setMiscReg(int misc_reg, const MiscReg &val, unsigned tid) +AlphaO3CPU<Impl>::setMiscReg(int misc_reg, const TheISA::MiscReg &val, + unsigned tid) { this->regFile.setMiscReg(misc_reg, val, tid); } template <class Impl> void -AlphaO3CPU<Impl>::setMiscRegWithEffect(int misc_reg, const MiscReg &val, - unsigned tid) +AlphaO3CPU<Impl>::setMiscRegWithEffect(int misc_reg, + const TheISA::MiscReg &val, unsigned tid) { this->regFile.setMiscRegWithEffect(misc_reg, val, tid); } @@ -312,7 +308,7 @@ AlphaO3CPU<Impl>::getSyscallArg(int i, int tid) template <class Impl> void -AlphaO3CPU<Impl>::setSyscallArg(int i, IntReg val, int tid) +AlphaO3CPU<Impl>::setSyscallArg(int i, TheISA::IntReg val, int tid) { this->setArchIntReg(AlphaISA::ArgumentReg0 + i, val, tid); } @@ -321,17 +317,6 @@ template <class Impl> void AlphaO3CPU<Impl>::setSyscallReturn(SyscallReturn return_value, int tid) { - // check for error condition. Alpha syscall convention is to - // indicate success/failure in reg a3 (r19) and put the - // return value itself in the standard return value reg (v0). - if (return_value.successful()) { - // no error - this->setArchIntReg(TheISA::SyscallSuccessReg, 0, tid); - this->setArchIntReg(TheISA::ReturnValueReg, return_value.value(), tid); - } else { - // got an error, return details - this->setArchIntReg(TheISA::SyscallSuccessReg, (IntReg) -1, tid); - this->setArchIntReg(TheISA::ReturnValueReg, -return_value.value(), tid); - } + TheISA::setSyscallReturn(return_value, this->tcBase(tid)); } #endif diff --git a/src/cpu/o3/alpha/dyn_inst.hh b/src/cpu/o3/alpha/dyn_inst.hh index 49cc5a201..603a1b52d 100644 --- a/src/cpu/o3/alpha/dyn_inst.hh +++ b/src/cpu/o3/alpha/dyn_inst.hh @@ -73,8 +73,9 @@ class AlphaDynInst : public BaseDynInst<Impl> public: /** BaseDynInst constructor given a binary instruction. */ - AlphaDynInst(ExtMachInst inst, Addr PC, Addr Pred_PC, InstSeqNum seq_num, - O3CPU *cpu); + AlphaDynInst(ExtMachInst inst, Addr PC, Addr NPC, + Addr Pred_PC, Addr Pred_NPC, + InstSeqNum seq_num, O3CPU *cpu); /** BaseDynInst constructor given a static inst pointer. */ AlphaDynInst(StaticInstPtr &_staticInst); @@ -123,6 +124,44 @@ class AlphaDynInst : public BaseDynInst<Impl> this->threadNumber); } + /** Reads a miscellaneous register. */ + TheISA::MiscReg readMiscRegOperand(const StaticInst *si, int idx) + { + return this->cpu->readMiscReg( + si->srcRegIdx(idx) - TheISA::Ctrl_Base_DepTag, + this->threadNumber); + } + + /** Reads a misc. register, including any side-effects the read + * might have as defined by the architecture. + */ + TheISA::MiscReg readMiscRegOperandWithEffect(const StaticInst *si, int idx) + { + return this->cpu->readMiscRegWithEffect( + si->srcRegIdx(idx) - TheISA::Ctrl_Base_DepTag, + this->threadNumber); + } + + /** Sets a misc. register. */ + void setMiscRegOperand(const StaticInst * si, int idx, const MiscReg &val) + { + this->instResult.integer = val; + return this->cpu->setMiscReg( + si->destRegIdx(idx) - TheISA::Ctrl_Base_DepTag, + val, this->threadNumber); + } + + /** Sets a misc. register, including any side-effects the write + * might have as defined by the architecture. + */ + void setMiscRegOperandWithEffect(const StaticInst *si, int idx, + const MiscReg &val) + { + return this->cpu->setMiscRegWithEffect( + si->destRegIdx(idx) - TheISA::Ctrl_Base_DepTag, + val, this->threadNumber); + } + #if FULL_SYSTEM /** Calls hardware return from error interrupt. */ Fault hwrei(); @@ -134,22 +173,6 @@ class AlphaDynInst : public BaseDynInst<Impl> void syscall(int64_t callnum); #endif - private: - /** Physical register index of the destination registers of this - * instruction. - */ - PhysRegIndex _destRegIdx[MaxInstDestRegs]; - - /** Physical register index of the source registers of this - * instruction. - */ - PhysRegIndex _srcRegIdx[MaxInstSrcRegs]; - - /** Physical register index of the previous producers of the - * architected destinations. - */ - PhysRegIndex _prevDestRegIdx[MaxInstDestRegs]; - public: // The register accessor methods provide the index of the @@ -165,28 +188,28 @@ class AlphaDynInst : public BaseDynInst<Impl> uint64_t readIntRegOperand(const StaticInst *si, int idx) { - return this->cpu->readIntReg(_srcRegIdx[idx]); + return this->cpu->readIntReg(this->_srcRegIdx[idx]); } FloatReg readFloatRegOperand(const StaticInst *si, int idx, int width) { - return this->cpu->readFloatReg(_srcRegIdx[idx], width); + return this->cpu->readFloatReg(this->_srcRegIdx[idx], width); } FloatReg readFloatRegOperand(const StaticInst *si, int idx) { - return this->cpu->readFloatReg(_srcRegIdx[idx]); + return this->cpu->readFloatReg(this->_srcRegIdx[idx]); } FloatRegBits readFloatRegOperandBits(const StaticInst *si, int idx, int width) { - return this->cpu->readFloatRegBits(_srcRegIdx[idx], width); + return this->cpu->readFloatRegBits(this->_srcRegIdx[idx], width); } FloatRegBits readFloatRegOperandBits(const StaticInst *si, int idx) { - return this->cpu->readFloatRegBits(_srcRegIdx[idx]); + return this->cpu->readFloatRegBits(this->_srcRegIdx[idx]); } /** @todo: Make results into arrays so they can handle multiple dest @@ -194,79 +217,37 @@ class AlphaDynInst : public BaseDynInst<Impl> */ void setIntRegOperand(const StaticInst *si, int idx, uint64_t val) { - this->cpu->setIntReg(_destRegIdx[idx], val); + this->cpu->setIntReg(this->_destRegIdx[idx], val); BaseDynInst<Impl>::setIntRegOperand(si, idx, val); } void setFloatRegOperand(const StaticInst *si, int idx, FloatReg val, int width) { - this->cpu->setFloatReg(_destRegIdx[idx], val, width); + this->cpu->setFloatReg(this->_destRegIdx[idx], val, width); BaseDynInst<Impl>::setFloatRegOperand(si, idx, val, width); } void setFloatRegOperand(const StaticInst *si, int idx, FloatReg val) { - this->cpu->setFloatReg(_destRegIdx[idx], val); + this->cpu->setFloatReg(this->_destRegIdx[idx], val); BaseDynInst<Impl>::setFloatRegOperand(si, idx, val); } void setFloatRegOperandBits(const StaticInst *si, int idx, FloatRegBits val, int width) { - this->cpu->setFloatRegBits(_destRegIdx[idx], val, width); + this->cpu->setFloatRegBits(this->_destRegIdx[idx], val, width); BaseDynInst<Impl>::setFloatRegOperandBits(si, idx, val); } void setFloatRegOperandBits(const StaticInst *si, int idx, FloatRegBits val) { - this->cpu->setFloatRegBits(_destRegIdx[idx], val); + this->cpu->setFloatRegBits(this->_destRegIdx[idx], val); BaseDynInst<Impl>::setFloatRegOperandBits(si, idx, val); } - /** Returns the physical register index of the i'th destination - * register. - */ - PhysRegIndex renamedDestRegIdx(int idx) const - { - return _destRegIdx[idx]; - } - - /** Returns the physical register index of the i'th source register. */ - PhysRegIndex renamedSrcRegIdx(int idx) const - { - return _srcRegIdx[idx]; - } - - /** Returns the physical register index of the previous physical register - * that remapped to the same logical register index. - */ - PhysRegIndex prevDestRegIdx(int idx) const - { - return _prevDestRegIdx[idx]; - } - - /** Renames a destination register to a physical register. Also records - * the previous physical register that the logical register mapped to. - */ - void renameDestReg(int idx, - PhysRegIndex renamed_dest, - PhysRegIndex previous_rename) - { - _destRegIdx[idx] = renamed_dest; - _prevDestRegIdx[idx] = previous_rename; - } - - /** Renames a source logical register to the physical register which - * has/will produce that logical register's result. - * @todo: add in whether or not the source register is ready. - */ - void renameSrcReg(int idx, PhysRegIndex renamed_src) - { - _srcRegIdx[idx] = renamed_src; - } - public: /** Calculates EA part of a memory instruction. Currently unused, * though it may be useful in the future if we want to split diff --git a/src/cpu/o3/alpha/dyn_inst_impl.hh b/src/cpu/o3/alpha/dyn_inst_impl.hh index 6fc548a85..50cdec408 100644 --- a/src/cpu/o3/alpha/dyn_inst_impl.hh +++ b/src/cpu/o3/alpha/dyn_inst_impl.hh @@ -31,9 +31,10 @@ #include "cpu/o3/alpha/dyn_inst.hh" template <class Impl> -AlphaDynInst<Impl>::AlphaDynInst(ExtMachInst inst, Addr PC, Addr Pred_PC, +AlphaDynInst<Impl>::AlphaDynInst(ExtMachInst inst, Addr PC, Addr NPC, + Addr Pred_PC, Addr Pred_NPC, InstSeqNum seq_num, O3CPU *cpu) - : BaseDynInst<Impl>(inst, PC, Pred_PC, seq_num, cpu) + : BaseDynInst<Impl>(inst, PC, NPC, Pred_PC, Pred_NPC, seq_num, cpu) { initVars(); } @@ -53,11 +54,11 @@ AlphaDynInst<Impl>::initVars() // as the normal register entries. It will allow the IQ to work // without any modifications. for (int i = 0; i < this->staticInst->numDestRegs(); i++) { - _destRegIdx[i] = this->staticInst->destRegIdx(i); + this->_destRegIdx[i] = this->staticInst->destRegIdx(i); } for (int i = 0; i < this->staticInst->numSrcRegs(); i++) { - _srcRegIdx[i] = this->staticInst->srcRegIdx(i); + this->_srcRegIdx[i] = this->staticInst->srcRegIdx(i); this->_readySrcRegIdx[i] = 0; } } diff --git a/src/cpu/o3/alpha/thread_context.hh b/src/cpu/o3/alpha/thread_context.hh index bcecb7087..6d61501ac 100644 --- a/src/cpu/o3/alpha/thread_context.hh +++ b/src/cpu/o3/alpha/thread_context.hh @@ -36,12 +36,6 @@ class AlphaTC : public O3ThreadContext<Impl> { public: #if FULL_SYSTEM - /** Returns a pointer to the ITB. */ - virtual AlphaISA::ITB *getITBPtr() { return this->cpu->itb; } - - /** Returns a pointer to the DTB. */ - virtual AlphaISA::DTB *getDTBPtr() { return this->cpu->dtb; } - /** Returns pointer to the quiesce event. */ virtual EndQuiesceEvent *getQuiesceEvent() { @@ -51,8 +45,7 @@ class AlphaTC : public O3ThreadContext<Impl> virtual uint64_t readNextNPC() { - panic("Alpha has no NextNPC!"); - return 0; + return this->readNextPC() + sizeof(TheISA::MachInst); } virtual void setNextNPC(uint64_t val) diff --git a/src/cpu/o3/bpred_unit_impl.hh b/src/cpu/o3/bpred_unit_impl.hh index 477c8e4cb..84c50b4da 100644 --- a/src/cpu/o3/bpred_unit_impl.hh +++ b/src/cpu/o3/bpred_unit_impl.hh @@ -149,7 +149,7 @@ BPredUnit<Impl>::predict(DynInstPtr &inst, Addr &PC, unsigned tid) using TheISA::MachInst; bool pred_taken = false; - Addr target; + Addr target = PC; ++lookups; @@ -233,14 +233,7 @@ BPredUnit<Impl>::predict(DynInstPtr &inst, Addr &PC, unsigned tid) } } - if (pred_taken) { - // Set the PC and the instruction's predicted target. - PC = target; - inst->setPredTarg(target); - } else { - PC = PC + sizeof(MachInst); - inst->setPredTarg(PC); - } + PC = target; predHist[tid].push_front(predict_record); diff --git a/src/cpu/o3/comm.hh b/src/cpu/o3/comm.hh index aa58fc20e..d96919007 100644 --- a/src/cpu/o3/comm.hh +++ b/src/cpu/o3/comm.hh @@ -87,9 +87,10 @@ struct DefaultIEWDefaultCommit { bool squash[Impl::MaxThreads]; bool branchMispredict[Impl::MaxThreads]; bool branchTaken[Impl::MaxThreads]; - bool condDelaySlotBranch[Impl::MaxThreads]; + bool squashDelaySlot[Impl::MaxThreads]; uint64_t mispredPC[Impl::MaxThreads]; uint64_t nextPC[Impl::MaxThreads]; + uint64_t nextNPC[Impl::MaxThreads]; InstSeqNum squashedSeqNum[Impl::MaxThreads]; bool includeSquashInst[Impl::MaxThreads]; @@ -121,6 +122,7 @@ struct TimeBufStruct { bool branchTaken; uint64_t mispredPC; uint64_t nextPC; + uint64_t nextNPC; unsigned branchCount; }; @@ -160,6 +162,7 @@ struct TimeBufStruct { bool branchTaken; uint64_t mispredPC; uint64_t nextPC; + uint64_t nextNPC; // Represents the instruction that has either been retired or // squashed. Similar to having a single bus that broadcasts the diff --git a/src/cpu/o3/commit_impl.hh b/src/cpu/o3/commit_impl.hh index 483c2f71b..f1457922c 100644 --- a/src/cpu/o3/commit_impl.hh +++ b/src/cpu/o3/commit_impl.hh @@ -519,6 +519,7 @@ DefaultCommit<Impl>::squashAll(unsigned tid) toIEW->commitInfo[tid].branchMispredict = false; toIEW->commitInfo[tid].nextPC = PC[tid]; + toIEW->commitInfo[tid].nextNPC = nextPC[tid]; } template <class Impl> @@ -734,27 +735,15 @@ DefaultCommit<Impl>::commit() InstSeqNum squashed_inst = fromIEW->squashedSeqNum[tid]; #if ISA_HAS_DELAY_SLOT - InstSeqNum bdelay_done_seq_num; - bool squash_bdelay_slot; - - if (fromIEW->branchMispredict[tid]) { - if (fromIEW->branchTaken[tid] && - fromIEW->condDelaySlotBranch[tid]) { - DPRINTF(Commit, "[tid:%i]: Cond. delay slot branch" - "mispredicted as taken. Squashing after previous " - "inst, [sn:%i]\n", - tid, squashed_inst); - bdelay_done_seq_num = squashed_inst; - squash_bdelay_slot = true; - } else { - DPRINTF(Commit, "[tid:%i]: Branch Mispredict. Squashing " - "after delay slot [sn:%i]\n", tid, squashed_inst+1); - bdelay_done_seq_num = squashed_inst + 1; - squash_bdelay_slot = false; - } - } else { - bdelay_done_seq_num = squashed_inst; - squash_bdelay_slot = true; + InstSeqNum bdelay_done_seq_num = squashed_inst; + bool squash_bdelay_slot = fromIEW->squashDelaySlot[tid]; + bool branchMispredict = fromIEW->branchMispredict[tid]; + + // Squashing/not squashing the branch delay slot only makes + // sense when you're squashing from a branch, ie from a branch + // mispredict. + if (branchMispredict && !squash_bdelay_slot) { + bdelay_done_seq_num++; } #endif @@ -793,6 +782,7 @@ DefaultCommit<Impl>::commit() fromIEW->branchTaken[tid]; toIEW->commitInfo[tid].nextPC = fromIEW->nextPC[tid]; + toIEW->commitInfo[tid].nextNPC = fromIEW->nextNPC[tid]; toIEW->commitInfo[tid].mispredPC = fromIEW->mispredPC[tid]; @@ -1122,7 +1112,7 @@ DefaultCommit<Impl>::commitHead(DynInstPtr &head_inst, unsigned inst_num) // Update the commit rename map for (int i = 0; i < head_inst->numDestRegs(); i++) { - renameMap[tid]->setEntry(head_inst->destRegIdx(i), + renameMap[tid]->setEntry(head_inst->flattenedDestRegIdx(i), head_inst->renamedDestRegIdx(i)); } diff --git a/src/cpu/o3/cpu.cc b/src/cpu/o3/cpu.cc index 18cc87c0b..66c75a12d 100644 --- a/src/cpu/o3/cpu.cc +++ b/src/cpu/o3/cpu.cc @@ -150,6 +150,10 @@ FullO3CPU<Impl>::DeallocateContextEvent::description() template <class Impl> FullO3CPU<Impl>::FullO3CPU(Params *params) : BaseO3CPU(params), +#if FULL_SYSTEM + itb(params->itb), + dtb(params->dtb), +#endif tickEvent(this), removeInstsThisCycle(false), fetch(params), @@ -657,9 +661,7 @@ FullO3CPU<Impl>::insertThread(unsigned tid) //Set PC/NPC/NNPC setPC(src_tc->readPC(), tid); setNextPC(src_tc->readNextPC(), tid); -#if ISA_HAS_DELAY_SLOT setNextNPC(src_tc->readNextNPC(), tid); -#endif src_tc->setStatus(ThreadContext::Active); @@ -698,7 +700,7 @@ FullO3CPU<Impl>::removeThread(unsigned tid) // Squash Throughout Pipeline InstSeqNum squash_seq_num = commit.rob->readHeadInst(tid)->seqNum; - fetch.squash(0, squash_seq_num, true, tid); + fetch.squash(0, sizeof(TheISA::MachInst), squash_seq_num, true, tid); decode.squash(tid); rename.squash(squash_seq_num, tid); iew.squash(tid); diff --git a/src/cpu/o3/cpu.hh b/src/cpu/o3/cpu.hh index 2bf9cb23b..d217a3e85 100644 --- a/src/cpu/o3/cpu.hh +++ b/src/cpu/o3/cpu.hh @@ -91,9 +91,6 @@ template <class Impl> class FullO3CPU : public BaseO3CPU { public: - typedef TheISA::FloatReg FloatReg; - typedef TheISA::FloatRegBits FloatRegBits; - // Typedefs from the Impl here. typedef typename Impl::CPUPol CPUPolicy; typedef typename Impl::Params Params; @@ -114,6 +111,11 @@ class FullO3CPU : public BaseO3CPU SwitchedOut }; +#if FULL_SYSTEM + TheISA::ITB * itb; + TheISA::DTB * dtb; +#endif + /** Overall CPU status. */ Status _status; @@ -382,23 +384,23 @@ class FullO3CPU : public BaseO3CPU /** Register accessors. Index refers to the physical register index. */ uint64_t readIntReg(int reg_idx); - FloatReg readFloatReg(int reg_idx); + TheISA::FloatReg readFloatReg(int reg_idx); - FloatReg readFloatReg(int reg_idx, int width); + TheISA::FloatReg readFloatReg(int reg_idx, int width); - FloatRegBits readFloatRegBits(int reg_idx); + TheISA::FloatRegBits readFloatRegBits(int reg_idx); - FloatRegBits readFloatRegBits(int reg_idx, int width); + TheISA::FloatRegBits readFloatRegBits(int reg_idx, int width); void setIntReg(int reg_idx, uint64_t val); - void setFloatReg(int reg_idx, FloatReg val); + void setFloatReg(int reg_idx, TheISA::FloatReg val); - void setFloatReg(int reg_idx, FloatReg val, int width); + void setFloatReg(int reg_idx, TheISA::FloatReg val, int width); - void setFloatRegBits(int reg_idx, FloatRegBits val); + void setFloatRegBits(int reg_idx, TheISA::FloatRegBits val); - void setFloatRegBits(int reg_idx, FloatRegBits val, int width); + void setFloatRegBits(int reg_idx, TheISA::FloatRegBits val, int width); uint64_t readArchIntReg(int reg_idx, unsigned tid); diff --git a/src/cpu/o3/decode_impl.hh b/src/cpu/o3/decode_impl.hh index 26ed40c67..79a0bfdbf 100644 --- a/src/cpu/o3/decode_impl.hh +++ b/src/cpu/o3/decode_impl.hh @@ -282,6 +282,10 @@ DefaultDecode<Impl>::squash(DynInstPtr &inst, unsigned tid) toFetch->decodeInfo[tid].doneSeqNum = inst->seqNum; toFetch->decodeInfo[tid].squash = true; toFetch->decodeInfo[tid].nextPC = inst->branchTarget(); + ///FIXME There needs to be a way to set the nextPC and nextNPC + ///explicitly for ISAs with delay slots. + toFetch->decodeInfo[tid].nextNPC = + inst->branchTarget() + sizeof(TheISA::MachInst); #if ISA_HAS_DELAY_SLOT toFetch->decodeInfo[tid].branchTaken = inst->readNextNPC() != (inst->readNextPC() + sizeof(TheISA::MachInst)); @@ -743,9 +747,9 @@ DefaultDecode<Impl>::decodeInsts(unsigned tid) // Ensure that if it was predicted as a branch, it really is a // branch. - if (inst->predTaken() && !inst->isControl()) { - DPRINTF(Decode, "PredPC : %#x != NextPC: %#x\n",inst->predPC, - inst->nextPC + 4); + if (inst->readPredTaken() && !inst->isControl()) { + DPRINTF(Decode, "PredPC : %#x != NextPC: %#x\n", + inst->readPredPC(), inst->readNextPC() + 4); panic("Instruction predicted as a branch!"); @@ -762,26 +766,29 @@ DefaultDecode<Impl>::decodeInsts(unsigned tid) if (inst->isDirectCtrl() && inst->isUncondCtrl()) { ++decodeBranchResolved; - if (inst->branchTarget() != inst->readPredTarg()) { + if (inst->branchTarget() != inst->readPredPC()) { ++decodeBranchMispred; // Might want to set some sort of boolean and just do // a check at the end #if !ISA_HAS_DELAY_SLOT squash(inst, inst->threadNumber); - inst->setPredTarg(inst->branchTarget()); + Addr target = inst->branchTarget(); + inst->setPredTarg(target, target + sizeof(TheISA::MachInst)); break; #else // If mispredicted as taken, then ignore delay slot // instruction... else keep delay slot and squash // after it is sent to rename - if (inst->predTaken() && inst->isCondDelaySlot()) { + if (inst->readPredTaken() && inst->isCondDelaySlot()) { DPRINTF(Decode, "[tid:%i]: Conditional delay slot inst." "[sn:%i] PC %#x mispredicted as taken.\n", tid, inst->seqNum, inst->PC); bdelayDoneSeqNum[tid] = inst->seqNum; squash(inst, inst->threadNumber); - inst->setPredTarg(inst->branchTarget()); + Addr target = inst->branchTarget(); + inst->setPredTarg(target, + target + sizeof(TheISA::MachInst)); break; } else { DPRINTF(Decode, "[tid:%i]: Misprediction detected at " @@ -800,7 +807,9 @@ DefaultDecode<Impl>::decodeInsts(unsigned tid) if (squashAfterDelaySlot[tid]) { assert(!inst->isSquashed()); squash(squashInst[tid], squashInst[tid]->threadNumber); - squashInst[tid]->setPredTarg(squashInst[tid]->branchTarget()); + Addr target = squashInst[tid]->branchTarget(); + squashInst[tid]->setPredTarg(target, + target + sizeof(TheISA::MachInst)); assert(!inst->isSquashed()); break; } diff --git a/src/cpu/o3/fetch.hh b/src/cpu/o3/fetch.hh index e880e14e4..8347ed775 100644 --- a/src/cpu/o3/fetch.hh +++ b/src/cpu/o3/fetch.hh @@ -239,13 +239,13 @@ class DefaultFetch bool fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid); /** Squashes a specific thread and resets the PC. */ - inline void doSquash(const Addr &new_PC, unsigned tid); + inline void doSquash(const Addr &new_PC, const Addr &new_NPC, unsigned tid); /** Squashes a specific thread and resets the PC. Also tells the CPU to * remove any instructions between fetch and decode that should be sqaushed. */ - void squashFromDecode(const Addr &new_PC, const InstSeqNum &seq_num, - unsigned tid); + void squashFromDecode(const Addr &new_PC, const Addr &new_NPC, + const InstSeqNum &seq_num, unsigned tid); /** Checks if a thread is stalled. */ bool checkStall(unsigned tid) const; @@ -259,7 +259,8 @@ class DefaultFetch * remove any instructions that are not in the ROB. The source of this * squash should be the commit stage. */ - void squash(const Addr &new_PC, const InstSeqNum &seq_num, + void squash(const Addr &new_PC, const Addr &new_NPC, + const InstSeqNum &seq_num, bool squash_delay_slot, unsigned tid); /** Ticks the fetch stage, processing all inputs signals and fetching @@ -361,19 +362,6 @@ class DefaultFetch /** Tracks how many instructions has been fetched this cycle. */ int numInst; - /** Tracks delay slot information for threads in ISAs which use - * delay slots; - */ - struct DelaySlotInfo { - InstSeqNum delaySlotSeqNum; - InstSeqNum branchSeqNum; - int numInsts; - Addr targetAddr; - bool targetReady; - }; - - DelaySlotInfo delaySlotInfo[Impl::MaxThreads]; - /** Source of possible stalls. */ struct Stalls { bool decode; diff --git a/src/cpu/o3/fetch_impl.hh b/src/cpu/o3/fetch_impl.hh index fe320fa79..90d3868a5 100644 --- a/src/cpu/o3/fetch_impl.hh +++ b/src/cpu/o3/fetch_impl.hh @@ -319,9 +319,7 @@ DefaultFetch<Impl>::initStage() for (int tid = 0; tid < numThreads; tid++) { PC[tid] = cpu->readPC(tid); nextPC[tid] = cpu->readNextPC(tid); -#if ISA_HAS_DELAY_SLOT nextNPC[tid] = cpu->readNextNPC(tid); -#endif } // Size of cache block. @@ -343,11 +341,6 @@ DefaultFetch<Impl>::initStage() cacheDataPC[tid] = 0; cacheDataValid[tid] = false; - delaySlotInfo[tid].branchSeqNum = -1; - delaySlotInfo[tid].numInsts = 0; - delaySlotInfo[tid].targetAddr = 0; - delaySlotInfo[tid].targetReady = false; - stalls[tid].decode = false; stalls[tid].rename = false; stalls[tid].iew = false; @@ -441,10 +434,8 @@ DefaultFetch<Impl>::takeOverFrom() nextPC[i] = cpu->readNextPC(i); #if ISA_HAS_DELAY_SLOT nextNPC[i] = cpu->readNextNPC(i); - delaySlotInfo[i].branchSeqNum = -1; - delaySlotInfo[i].numInsts = 0; - delaySlotInfo[i].targetAddr = 0; - delaySlotInfo[i].targetReady = false; +#else + nextNPC[i] = nextPC[i] + sizeof(TheISA::MachInst); #endif fetchStatus[i] = Running; } @@ -503,54 +494,41 @@ DefaultFetch<Impl>::lookupAndUpdateNextPC(DynInstPtr &inst, Addr &next_PC, bool predict_taken; if (!inst->isControl()) { -#if ISA_HAS_DELAY_SLOT - Addr cur_PC = next_PC; - next_PC = cur_PC + instSize; //next_NPC; - next_NPC = cur_PC + (2 * instSize);//next_NPC + instSize; - inst->setPredTarg(next_NPC); -#else - next_PC = next_PC + instSize; - inst->setPredTarg(next_PC); -#endif + next_PC = next_NPC; + next_NPC = next_NPC + instSize; + inst->setPredTarg(next_PC, next_NPC); + inst->setPredTaken(false); return false; } int tid = inst->threadNumber; -#if ISA_HAS_DELAY_SLOT Addr pred_PC = next_PC; predict_taken = branchPred.predict(inst, pred_PC, tid); - if (predict_taken) { - DPRINTF(Fetch, "[tid:%i]: Branch predicted to be true.\n", tid); +/* if (predict_taken) { + DPRINTF(Fetch, "[tid:%i]: Branch predicted to be taken to %#x.\n", + tid, pred_PC); } else { - DPRINTF(Fetch, "[tid:%i]: Branch predicted to be false.\n", tid); - } + DPRINTF(Fetch, "[tid:%i]: Branch predicted to be not taken.\n", tid); + }*/ - if (predict_taken) { - next_PC = next_NPC; +#if ISA_HAS_DELAY_SLOT + next_PC = next_NPC; + if (predict_taken) next_NPC = pred_PC; - - // Update delay slot info - ++delaySlotInfo[tid].numInsts; - delaySlotInfo[tid].targetAddr = pred_PC; - DPRINTF(Fetch, "[tid:%i]: %i delay slot inst(s) to process.\n", tid, - delaySlotInfo[tid].numInsts); - } else { // !predict_taken - if (inst->isCondDelaySlot()) { - next_PC = pred_PC; - // The delay slot is skipped here if there is on - // prediction - } else { - next_PC = next_NPC; - // No need to declare a delay slot here since - // there is no for the pred. target to jump - } - - next_NPC = next_NPC + instSize; - } + else + next_NPC += instSize; #else - predict_taken = branchPred.predict(inst, next_PC, tid); + if (predict_taken) + next_PC = pred_PC; + else + next_PC += instSize; + next_NPC = next_PC + instSize; #endif +/* DPRINTF(Fetch, "[tid:%i]: Branch predicted to go to %#x and then %#x.\n", + tid, next_PC, next_NPC);*/ + inst->setPredTarg(next_PC, next_NPC); + inst->setPredTaken(predict_taken); ++fetchedBranches; @@ -671,14 +649,15 @@ DefaultFetch<Impl>::fetchCacheLine(Addr fetch_PC, Fault &ret_fault, unsigned tid template <class Impl> inline void -DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid) +DefaultFetch<Impl>::doSquash(const Addr &new_PC, + const Addr &new_NPC, unsigned tid) { - DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x.\n", - tid, new_PC); + DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %#x, NPC to: %#x.\n", + tid, new_PC, new_NPC); PC[tid] = new_PC; - nextPC[tid] = new_PC + instSize; - nextNPC[tid] = new_PC + (2 * instSize); + nextPC[tid] = new_NPC; + nextNPC[tid] = new_NPC + instSize; // Clear the icache miss if it's outstanding. if (fetchStatus[tid] == IcacheWaitResponse) { @@ -704,21 +683,13 @@ DefaultFetch<Impl>::doSquash(const Addr &new_PC, unsigned tid) template<class Impl> void -DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, +DefaultFetch<Impl>::squashFromDecode(const Addr &new_PC, const Addr &new_NPC, const InstSeqNum &seq_num, unsigned tid) { DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n",tid); - doSquash(new_PC, tid); - -#if ISA_HAS_DELAY_SLOT - if (seq_num <= delaySlotInfo[tid].branchSeqNum) { - delaySlotInfo[tid].numInsts = 0; - delaySlotInfo[tid].targetAddr = 0; - delaySlotInfo[tid].targetReady = false; - } -#endif + doSquash(new_PC, new_NPC, tid); // Tell the CPU to remove any instructions that are in flight between // fetch and decode. @@ -793,20 +764,15 @@ DefaultFetch<Impl>::updateFetchStatus() template <class Impl> void -DefaultFetch<Impl>::squash(const Addr &new_PC, const InstSeqNum &seq_num, +DefaultFetch<Impl>::squash(const Addr &new_PC, const Addr &new_NPC, + const InstSeqNum &seq_num, bool squash_delay_slot, unsigned tid) { DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n",tid); - doSquash(new_PC, tid); + doSquash(new_PC, new_NPC, tid); #if ISA_HAS_DELAY_SLOT - if (seq_num <= delaySlotInfo[tid].branchSeqNum) { - delaySlotInfo[tid].numInsts = 0; - delaySlotInfo[tid].targetAddr = 0; - delaySlotInfo[tid].targetReady = false; - } - // Tell the CPU to remove any instructions that are not in the ROB. cpu->removeInstsNotInROB(tid, squash_delay_slot, seq_num); #else @@ -929,6 +895,7 @@ DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) #endif // In any case, squash. squash(fromCommit->commitInfo[tid].nextPC, + fromCommit->commitInfo[tid].nextNPC, doneSeqNum, fromCommit->commitInfo[tid].squashDelaySlot, tid); @@ -984,8 +951,12 @@ DefaultFetch<Impl>::checkSignalsAndUpdate(unsigned tid) #else InstSeqNum doneSeqNum = fromDecode->decodeInfo[tid].doneSeqNum; #endif + DPRINTF(Fetch, "Squashing from decode with PC = %#x, NPC = %#x\n", + fromDecode->decodeInfo[tid].nextPC, + fromDecode->decodeInfo[tid].nextNPC); // Squash unless we're already squashing squashFromDecode(fromDecode->decodeInfo[tid].nextPC, + fromDecode->decodeInfo[tid].nextNPC, doneSeqNum, tid); @@ -1042,6 +1013,8 @@ DefaultFetch<Impl>::fetch(bool &status_change) // The current PC. Addr &fetch_PC = PC[tid]; + Addr &fetch_NPC = nextPC[tid]; + // Fault code for memory access. Fault fault = NoFault; @@ -1098,7 +1071,8 @@ DefaultFetch<Impl>::fetch(bool &status_change) } Addr next_PC = fetch_PC; - Addr next_NPC = next_PC + instSize; + Addr next_NPC = fetch_NPC; + InstSeqNum inst_seq; MachInst inst; ExtMachInst ext_inst; @@ -1117,15 +1091,22 @@ DefaultFetch<Impl>::fetch(bool &status_change) // ended this fetch block. bool predicted_branch = false; - // Need to keep track of whether or not a delay slot - // instruction has been fetched - for (; offset < cacheBlkSize && numInst < fetchWidth && - (!predicted_branch || delaySlotInfo[tid].numInsts > 0); + !predicted_branch; ++numInst) { + // If we're branching after this instruction, quite fetching + // from the same block then. + predicted_branch = + (fetch_PC + sizeof(TheISA::MachInst) != fetch_NPC); + if (predicted_branch) { + DPRINTF(Fetch, "Branch detected with PC = %#x, NPC = %#x\n", + fetch_PC, fetch_NPC); + } + + // Get a sequence number. inst_seq = cpu->getAndIncrementInstSeq(); @@ -1145,8 +1126,9 @@ DefaultFetch<Impl>::fetch(bool &status_change) #endif // Create a new DynInst from the instruction fetched. - DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, - next_PC, + DynInstPtr instruction = new DynInst(ext_inst, + fetch_PC, fetch_NPC, + next_PC, next_NPC, inst_seq, cpu); instruction->setTid(tid); @@ -1158,6 +1140,8 @@ DefaultFetch<Impl>::fetch(bool &status_change) "[sn:%lli]\n", tid, instruction->readPC(), inst_seq); + //DPRINTF(Fetch, "[tid:%i]: MachInst is %#x\n", tid, ext_inst); + DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", tid, instruction->staticInst->disassemble(fetch_PC)); @@ -1166,8 +1150,8 @@ DefaultFetch<Impl>::fetch(bool &status_change) instruction->staticInst, instruction->readPC()); - predicted_branch = lookupAndUpdateNextPC(instruction, next_PC, - next_NPC); + lookupAndUpdateNextPC(instruction, next_PC, next_NPC); + predicted_branch |= (next_PC != fetch_NPC); // Add instruction to the CPU's list of instructions. instruction->setInstListIt(cpu->addInst(instruction)); @@ -1183,6 +1167,7 @@ DefaultFetch<Impl>::fetch(bool &status_change) // Move to the next instruction, unless we have a branch. fetch_PC = next_PC; + fetch_NPC = next_NPC; if (instruction->isQuiesce()) { DPRINTF(Fetch, "Quiesce instruction encountered, halting fetch!", @@ -1194,29 +1179,6 @@ DefaultFetch<Impl>::fetch(bool &status_change) } offset += instSize; - -#if ISA_HAS_DELAY_SLOT - if (predicted_branch) { - delaySlotInfo[tid].branchSeqNum = inst_seq; - - DPRINTF(Fetch, "[tid:%i]: Delay slot branch set to [sn:%i]\n", - tid, inst_seq); - continue; - } else if (delaySlotInfo[tid].numInsts > 0) { - --delaySlotInfo[tid].numInsts; - - // It's OK to set PC to target of branch - if (delaySlotInfo[tid].numInsts == 0) { - delaySlotInfo[tid].targetReady = true; - - // Break the looping condition - predicted_branch = true; - } - - DPRINTF(Fetch, "[tid:%i]: %i delay slot inst(s) left to" - " process.\n", tid, delaySlotInfo[tid].numInsts); - } -#endif } if (offset >= cacheBlkSize) { @@ -1225,7 +1187,7 @@ DefaultFetch<Impl>::fetch(bool &status_change) } else if (numInst >= fetchWidth) { DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth " "for this cycle.\n", tid); - } else if (predicted_branch && delaySlotInfo[tid].numInsts <= 0) { + } else if (predicted_branch) { DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch " "instruction encountered.\n", tid); } @@ -1238,26 +1200,13 @@ DefaultFetch<Impl>::fetch(bool &status_change) // Now that fetching is completed, update the PC to signify what the next // cycle will be. if (fault == NoFault) { + PC[tid] = next_PC; + nextPC[tid] = next_NPC; + nextNPC[tid] = next_NPC + instSize; #if ISA_HAS_DELAY_SLOT - if (delaySlotInfo[tid].targetReady && - delaySlotInfo[tid].numInsts == 0) { - // Set PC to target - PC[tid] = delaySlotInfo[tid].targetAddr; //next_PC - nextPC[tid] = next_PC + instSize; //next_NPC - nextNPC[tid] = next_PC + (2 * instSize); - - delaySlotInfo[tid].targetReady = false; - } else { - PC[tid] = next_PC; - nextPC[tid] = next_NPC; - nextNPC[tid] = next_NPC + instSize; - } - DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n", tid, PC[tid]); #else - DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n",tid, next_PC); - PC[tid] = next_PC; - nextPC[tid] = next_PC + instSize; + DPRINTF(Fetch, "[tid:%i]: Setting PC to %08p.\n", tid, next_PC); #endif } else { // We shouldn't be in an icache miss and also have a fault (an ITB @@ -1277,10 +1226,11 @@ DefaultFetch<Impl>::fetch(bool &status_change) ext_inst = TheISA::NoopMachInst; // Create a new DynInst from the dummy nop. - DynInstPtr instruction = new DynInst(ext_inst, fetch_PC, - next_PC, + DynInstPtr instruction = new DynInst(ext_inst, + fetch_PC, fetch_NPC, + next_PC, next_NPC, inst_seq, cpu); - instruction->setPredTarg(next_PC + instSize); + instruction->setPredTarg(next_PC, next_NPC); instruction->setTid(tid); instruction->setASID(tid); diff --git a/src/cpu/o3/iew_impl.hh b/src/cpu/o3/iew_impl.hh index a8962f2f7..f24eaf2c4 100644 --- a/src/cpu/o3/iew_impl.hh +++ b/src/cpu/o3/iew_impl.hh @@ -480,23 +480,37 @@ DefaultIEW<Impl>::squashDueToBranch(DynInstPtr &inst, unsigned tid) toCommit->mispredPC[tid] = inst->readPC(); toCommit->branchMispredict[tid] = true; + int instSize = sizeof(TheISA::MachInst); #if ISA_HAS_DELAY_SLOT - bool branch_taken = inst->readNextNPC() != - (inst->readNextPC() + sizeof(TheISA::MachInst)); + bool branch_taken = + !(inst->readNextPC() + instSize == inst->readNextNPC() && + (inst->readNextPC() == inst->readPC() + instSize || + inst->readNextPC() == inst->readPC() + 2 * instSize)); + DPRINTF(Sparc, "Branch taken = %s [sn:%i]\n", + branch_taken ? "true": "false", inst->seqNum); toCommit->branchTaken[tid] = branch_taken; - toCommit->condDelaySlotBranch[tid] = inst->isCondDelaySlot(); - - if (inst->isCondDelaySlot() && branch_taken) { + bool squashDelaySlot = true; +// (inst->readNextPC() != inst->readPC() + sizeof(TheISA::MachInst)); + DPRINTF(Sparc, "Squash delay slot = %s [sn:%i]\n", + squashDelaySlot ? "true": "false", inst->seqNum); + toCommit->squashDelaySlot[tid] = squashDelaySlot; + //If we're squashing the delay slot, we need to pick back up at NextPC. + //Otherwise, NextPC isn't being squashed, so we should pick back up at + //NextNPC. + if (squashDelaySlot) { toCommit->nextPC[tid] = inst->readNextPC(); + toCommit->nextNPC[tid] = inst->readNextNPC(); } else { toCommit->nextPC[tid] = inst->readNextNPC(); + toCommit->nextNPC[tid] = inst->readNextNPC() + instSize; } #else toCommit->branchTaken[tid] = inst->readNextPC() != (inst->readPC() + sizeof(TheISA::MachInst)); toCommit->nextPC[tid] = inst->readNextPC(); + toCommit->nextNPC[tid] = inst->readNextPC() + instSize; #endif toCommit->includeSquashInst[tid] = false; @@ -514,6 +528,11 @@ DefaultIEW<Impl>::squashDueToMemOrder(DynInstPtr &inst, unsigned tid) toCommit->squash[tid] = true; toCommit->squashedSeqNum[tid] = inst->seqNum; toCommit->nextPC[tid] = inst->readNextPC(); +#if ISA_HAS_DELAY_SLOT + toCommit->nextNPC[tid] = inst->readNextNPC(); +#else + toCommit->nextNPC[tid] = inst->readNextPC() + sizeof(TheISA::MachInst); +#endif toCommit->branchMispredict[tid] = false; toCommit->includeSquashInst[tid] = false; @@ -531,6 +550,11 @@ DefaultIEW<Impl>::squashDueToMemBlocked(DynInstPtr &inst, unsigned tid) toCommit->squash[tid] = true; toCommit->squashedSeqNum[tid] = inst->seqNum; toCommit->nextPC[tid] = inst->readPC(); +#if ISA_HAS_DELAY_SLOT + toCommit->nextNPC[tid] = inst->readNextPC(); +#else + toCommit->nextNPC[tid] = inst->readPC() + sizeof(TheISA::MachInst); +#endif toCommit->branchMispredict[tid] = false; // Must include the broadcasted SN in the squash. @@ -1341,17 +1365,15 @@ DefaultIEW<Impl>::executeInsts() fetchRedirect[tid] = true; DPRINTF(IEW, "Execute: Branch mispredict detected.\n"); -#if ISA_HAS_DELAY_SLOT - DPRINTF(IEW, "Execute: Redirecting fetch to PC: %#x.\n", - inst->nextNPC); -#else - DPRINTF(IEW, "Execute: Redirecting fetch to PC: %#x.\n", - inst->nextPC); -#endif + DPRINTF(IEW, "Predicted target was %#x, %#x.\n", + inst->readPredPC(), inst->readPredNPC()); + DPRINTF(IEW, "Execute: Redirecting fetch to PC: %#x," + " NPC: %#x.\n", inst->readNextPC(), + inst->readNextNPC()); // If incorrect, then signal the ROB that it must be squashed. squashDueToBranch(inst, tid); - if (inst->predTaken()) { + if (inst->readPredTaken()) { predictedTakenIncorrect++; } else { predictedNotTakenIncorrect++; @@ -1421,7 +1443,7 @@ DefaultIEW<Impl>::writebackInsts() // mark scoreboard that this instruction is finally complete. // Either have IEW have direct access to scoreboard, or have this // as part of backwards communication. - for (int inst_num = 0; inst_num < issueWidth && + for (int inst_num = 0; inst_num < wbWidth && toCommit->insts[inst_num]; inst_num++) { DynInstPtr inst = toCommit->insts[inst_num]; int tid = inst->threadNumber; diff --git a/src/cpu/o3/lsq_unit.hh b/src/cpu/o3/lsq_unit.hh index 14f9d5031..9c7eb7780 100644 --- a/src/cpu/o3/lsq_unit.hh +++ b/src/cpu/o3/lsq_unit.hh @@ -559,6 +559,12 @@ LSQUnit<Impl>::read(Request *req, T &data, int load_idx) // Cast this to type T? data = storeQueue[store_idx].data >> shift_amt; + // When the data comes from the store queue entry, it's in host + // order. When it gets sent to the load, it needs to be in guest + // order so when the load converts it again, it ends up back + // in host order like the inst expects. + data = TheISA::htog(data); + assert(!load_inst->memData); load_inst->memData = new uint8_t[64]; diff --git a/src/cpu/o3/lsq_unit_impl.hh b/src/cpu/o3/lsq_unit_impl.hh index 3b84d3411..ebd9301f6 100644 --- a/src/cpu/o3/lsq_unit_impl.hh +++ b/src/cpu/o3/lsq_unit_impl.hh @@ -597,8 +597,21 @@ LSQUnit<Impl>::writebackStores() assert(!inst->memData); inst->memData = new uint8_t[64]; - memcpy(inst->memData, (uint8_t *)&storeQueue[storeWBIdx].data, - req->getSize()); + + TheISA::IntReg convertedData = + TheISA::htog(storeQueue[storeWBIdx].data); + + //FIXME This is a hack to get SPARC working. It, along with endianness + //in the memory system in general, need to be straightened out more + //formally. The problem is that the data's endianness is swapped when + //it's in the 64 bit data field in the store queue. The data that you + //want won't start at the beginning of the field anymore unless it was + //a 64 bit access. + memcpy(inst->memData, + (uint8_t *)&convertedData + + (TheISA::ByteOrderDiffers ? + (sizeof(TheISA::IntReg) - req->getSize()) : 0), + req->getSize()); PacketPtr data_pkt = new Packet(req, Packet::WriteReq, Packet::Broadcast); data_pkt->dataStatic(inst->memData); @@ -612,7 +625,7 @@ LSQUnit<Impl>::writebackStores() DPRINTF(LSQUnit, "D-Cache: Writing back store idx:%i PC:%#x " "to Addr:%#x, data:%#x [sn:%lli]\n", storeWBIdx, inst->readPC(), - req->getPaddr(), *(inst->memData), + req->getPaddr(), (int)*(inst->memData), inst->seqNum); // @todo: Remove this SC hack once the memory system handles it. diff --git a/src/cpu/o3/mips/cpu_impl.hh b/src/cpu/o3/mips/cpu_impl.hh index 08e9ba483..e7dbd3aba 100644 --- a/src/cpu/o3/mips/cpu_impl.hh +++ b/src/cpu/o3/mips/cpu_impl.hh @@ -220,16 +220,6 @@ template <class Impl> void MipsO3CPU<Impl>::setSyscallReturn(SyscallReturn return_value, int tid) { - // check for error condition. - if (return_value.successful()) { - // no error - this->setArchIntReg(TheISA::SyscallSuccessReg, 0, tid); - this->setArchIntReg(TheISA::ReturnValueReg, return_value.value(), tid); - } else { - // got an error, return details - this->setArchIntReg(TheISA::SyscallSuccessReg, - (TheISA::IntReg) -1, tid); - this->setArchIntReg(TheISA::ReturnValueReg, -return_value.value(), tid); - } + TheISA::setSyscallReturn(return_value, this->tcBase(tid)); } #endif diff --git a/src/cpu/o3/mips/dyn_inst.hh b/src/cpu/o3/mips/dyn_inst.hh index 833371e00..f53530908 100755 --- a/src/cpu/o3/mips/dyn_inst.hh +++ b/src/cpu/o3/mips/dyn_inst.hh @@ -54,10 +54,6 @@ class MipsDynInst : public BaseDynInst<Impl> /** Typedef for the CPU. */ typedef typename Impl::O3CPU O3CPU; - /** Binary machine instruction type. */ - typedef TheISA::MachInst MachInst; - /** Extended machine instruction type. */ - typedef TheISA::ExtMachInst ExtMachInst; /** Logical register index type. */ typedef TheISA::RegIndex RegIndex; /** Integer register index type. */ @@ -74,8 +70,10 @@ class MipsDynInst : public BaseDynInst<Impl> public: /** BaseDynInst constructor given a binary instruction. */ - MipsDynInst(ExtMachInst inst, Addr PC, Addr Pred_PC, InstSeqNum seq_num, - O3CPU *cpu); + MipsDynInst(ExtMachInst inst, + Addr PC, Addr NPC, + Addr Pred_PC, Addr Pred_NPC, + InstSeqNum seq_num, O3CPU *cpu); /** BaseDynInst constructor given a static inst pointer. */ MipsDynInst(StaticInstPtr &_staticInst); @@ -127,22 +125,6 @@ class MipsDynInst : public BaseDynInst<Impl> /** Calls a syscall. */ void syscall(int64_t callnum); - private: - /** Physical register index of the destination registers of this - * instruction. - */ - PhysRegIndex _destRegIdx[MaxInstDestRegs]; - - /** Physical register index of the source registers of this - * instruction. - */ - PhysRegIndex _srcRegIdx[MaxInstSrcRegs]; - - /** Physical register index of the previous producers of the - * architected destinations. - */ - PhysRegIndex _prevDestRegIdx[MaxInstDestRegs]; - public: // The register accessor methods provide the index of the @@ -158,28 +140,28 @@ class MipsDynInst : public BaseDynInst<Impl> uint64_t readIntRegOperand(const StaticInst *si, int idx) { - return this->cpu->readIntReg(_srcRegIdx[idx]); + return this->cpu->readIntReg(this->_srcRegIdx[idx]); } FloatReg readFloatRegOperand(const StaticInst *si, int idx, int width) { - return this->cpu->readFloatReg(_srcRegIdx[idx], width); + return this->cpu->readFloatReg(this->_srcRegIdx[idx], width); } FloatReg readFloatRegOperand(const StaticInst *si, int idx) { - return this->cpu->readFloatReg(_srcRegIdx[idx]); + return this->cpu->readFloatReg(this->_srcRegIdx[idx]); } FloatRegBits readFloatRegOperandBits(const StaticInst *si, int idx, int width) { - return this->cpu->readFloatRegBits(_srcRegIdx[idx], width); + return this->cpu->readFloatRegBits(this->_srcRegIdx[idx], width); } FloatRegBits readFloatRegOperandBits(const StaticInst *si, int idx) { - return this->cpu->readFloatRegBits(_srcRegIdx[idx]); + return this->cpu->readFloatRegBits(this->_srcRegIdx[idx]); } /** @todo: Make results into arrays so they can handle multiple dest @@ -187,79 +169,37 @@ class MipsDynInst : public BaseDynInst<Impl> */ void setIntRegOperand(const StaticInst *si, int idx, uint64_t val) { - this->cpu->setIntReg(_destRegIdx[idx], val); + this->cpu->setIntReg(this->_destRegIdx[idx], val); BaseDynInst<Impl>::setIntRegOperand(si, idx, val); } void setFloatRegOperand(const StaticInst *si, int idx, FloatReg val, int width) { - this->cpu->setFloatReg(_destRegIdx[idx], val, width); + this->cpu->setFloatReg(this->_destRegIdx[idx], val, width); BaseDynInst<Impl>::setFloatRegOperand(si, idx, val, width); } void setFloatRegOperand(const StaticInst *si, int idx, FloatReg val) { - this->cpu->setFloatReg(_destRegIdx[idx], val); + this->cpu->setFloatReg(this->_destRegIdx[idx], val); BaseDynInst<Impl>::setFloatRegOperand(si, idx, val); } void setFloatRegOperandBits(const StaticInst *si, int idx, FloatRegBits val, int width) { - this->cpu->setFloatRegBits(_destRegIdx[idx], val, width); + this->cpu->setFloatRegBits(this->_destRegIdx[idx], val, width); BaseDynInst<Impl>::setFloatRegOperandBits(si, idx, val); } void setFloatRegOperandBits(const StaticInst *si, int idx, FloatRegBits val) { - this->cpu->setFloatRegBits(_destRegIdx[idx], val); + this->cpu->setFloatRegBits(this->_destRegIdx[idx], val); BaseDynInst<Impl>::setFloatRegOperandBits(si, idx, val); } - /** Returns the physical register index of the i'th destination - * register. - */ - PhysRegIndex renamedDestRegIdx(int idx) const - { - return _destRegIdx[idx]; - } - - /** Returns the physical register index of the i'th source register. */ - PhysRegIndex renamedSrcRegIdx(int idx) const - { - return _srcRegIdx[idx]; - } - - /** Returns the physical register index of the previous physical register - * that remapped to the same logical register index. - */ - PhysRegIndex prevDestRegIdx(int idx) const - { - return _prevDestRegIdx[idx]; - } - - /** Renames a destination register to a physical register. Also records - * the previous physical register that the logical register mapped to. - */ - void renameDestReg(int idx, - PhysRegIndex renamed_dest, - PhysRegIndex previous_rename) - { - _destRegIdx[idx] = renamed_dest; - _prevDestRegIdx[idx] = previous_rename; - } - - /** Renames a source logical register to the physical register which - * has/will produce that logical register's result. - * @todo: add in whether or not the source register is ready. - */ - void renameSrcReg(int idx, PhysRegIndex renamed_src) - { - _srcRegIdx[idx] = renamed_src; - } - public: /** Calculates EA part of a memory instruction. Currently unused, * though it may be useful in the future if we want to split diff --git a/src/cpu/o3/mips/dyn_inst_impl.hh b/src/cpu/o3/mips/dyn_inst_impl.hh index 5bc01b9b3..c0f9ae771 100755 --- a/src/cpu/o3/mips/dyn_inst_impl.hh +++ b/src/cpu/o3/mips/dyn_inst_impl.hh @@ -31,9 +31,11 @@ #include "cpu/o3/mips/dyn_inst.hh" template <class Impl> -MipsDynInst<Impl>::MipsDynInst(ExtMachInst inst, Addr PC, Addr Pred_PC, - InstSeqNum seq_num, O3CPU *cpu) - : BaseDynInst<Impl>(inst, PC, Pred_PC, seq_num, cpu) +MipsDynInst<Impl>::MipsDynInst(ExtMachInst inst, + Addr PC, Addr NPC, + Addr Pred_PC, Addr Pred_NPC, + InstSeqNum seq_num, O3CPU *cpu) + : BaseDynInst<Impl>(inst, PC, NPC, Pred_PC, Pred_NPC, seq_num, cpu) { initVars(); } @@ -53,11 +55,11 @@ MipsDynInst<Impl>::initVars() // as the normal register entries. It will allow the IQ to work // without any modifications. for (int i = 0; i < this->staticInst->numDestRegs(); i++) { - _destRegIdx[i] = this->staticInst->destRegIdx(i); + this->_destRegIdx[i] = this->staticInst->destRegIdx(i); } for (int i = 0; i < this->staticInst->numSrcRegs(); i++) { - _srcRegIdx[i] = this->staticInst->srcRegIdx(i); + this->_srcRegIdx[i] = this->staticInst->srcRegIdx(i); this->_readySrcRegIdx[i] = 0; } } diff --git a/src/cpu/o3/rename.hh b/src/cpu/o3/rename.hh index 177b9cb87..6b4628f92 100644 --- a/src/cpu/o3/rename.hh +++ b/src/cpu/o3/rename.hh @@ -411,6 +411,14 @@ class DefaultRename /** Whether or not rename needs to block this cycle. */ bool blockThisCycle; + /** Whether or not rename needs to resume a serialize instruction + * after squashing. */ + bool resumeSerialize; + + /** Whether or not rename needs to resume clearing out the skidbuffer + * after squashing. */ + bool resumeUnblocking; + /** The number of threads active in rename. */ unsigned numThreads; diff --git a/src/cpu/o3/rename_impl.hh b/src/cpu/o3/rename_impl.hh index 3a8e503a0..e303f1cee 100644 --- a/src/cpu/o3/rename_impl.hh +++ b/src/cpu/o3/rename_impl.hh @@ -31,6 +31,8 @@ #include <list> +#include "arch/isa_traits.hh" +#include "arch/regfile.hh" #include "config/full_system.hh" #include "cpu/o3/rename.hh" @@ -41,6 +43,8 @@ DefaultRename<Impl>::DefaultRename(Params *params) commitToRenameDelay(params->commitToRenameDelay), renameWidth(params->renameWidth), commitWidth(params->commitWidth), + resumeSerialize(false), + resumeUnblocking(false), numThreads(params->numberOfThreads), maxPhysicalRegs(params->numPhysIntRegs + params->numPhysFloatRegs) { @@ -332,12 +336,22 @@ DefaultRename<Impl>::squash(const InstSeqNum &squash_seq_num, unsigned tid) // If it still needs to block, the blocking should happen the next // cycle and there should be space to hold everything due to the squash. if (renameStatus[tid] == Blocked || - renameStatus[tid] == Unblocking || - renameStatus[tid] == SerializeStall) { - + renameStatus[tid] == Unblocking) { toDecode->renameUnblock[tid] = 1; + resumeSerialize = false; serializeInst[tid] = NULL; + } else if (renameStatus[tid] == SerializeStall) { + if (serializeInst[tid]->seqNum <= squash_seq_num) { + DPRINTF(Rename, "Rename will resume serializing after squash\n"); + resumeSerialize = true; + assert(serializeInst[tid]); + } else { + resumeSerialize = false; + toDecode->renameUnblock[tid] = 1; + + serializeInst[tid] = NULL; + } } // Set the status to Squashing. @@ -392,6 +406,9 @@ DefaultRename<Impl>::squash(const InstSeqNum &squash_seq_num, unsigned tid) } slist_it++; } + resumeUnblocking = (skidBuffer[tid].size() != 0); + DPRINTF(Rename, "Resume unblocking set to %s\n", + resumeUnblocking ? "true" : "false"); #else skidBuffer[tid].clear(); #endif @@ -476,6 +493,20 @@ DefaultRename<Impl>::rename(bool &status_change, unsigned tid) ++renameSquashCycles; } else if (renameStatus[tid] == SerializeStall) { ++renameSerializeStallCycles; + // If we are currently in SerializeStall and resumeSerialize + // was set, then that means that we are resuming serializing + // this cycle. Tell the previous stages to block. + if (resumeSerialize) { + resumeSerialize = false; + block(tid); + toDecode->renameUnblock[tid] = false; + } + } else if (renameStatus[tid] == Unblocking) { + if (resumeUnblocking) { + block(tid); + resumeUnblocking = false; + toDecode->renameUnblock[tid] = false; + } } if (renameStatus[tid] == Running || @@ -741,7 +772,17 @@ DefaultRename<Impl>::skidInsert(unsigned tid) } if (skidBuffer[tid].size() > skidBufferMax) + { + typename InstQueue::iterator it; + warn("Skidbuffer contents:\n"); + for(it = skidBuffer[tid].begin(); it != skidBuffer[tid].end(); it++) + { + warn("[tid:%u]: %s [sn:%i].\n", tid, + (*it)->staticInst->disassemble(inst->readPC()), + (*it)->seqNum); + } panic("Skidbuffer Exceeded Max Size"); + } } template <class Impl> @@ -830,7 +871,10 @@ DefaultRename<Impl>::block(unsigned tid) // Only signal backwards to block if the previous stages do not think // rename is already blocked. if (renameStatus[tid] != Blocked) { - if (renameStatus[tid] != Unblocking) { + // If resumeUnblocking is set, we unblocked during the squash, + // but now we're have unblocking status. We need to tell earlier + // stages to block. + if (resumeUnblocking || renameStatus[tid] != Unblocking) { toDecode->renameBlock[tid] = true; toDecode->renameUnblock[tid] = false; wroteToTimeBuffer = true; @@ -963,13 +1007,19 @@ DefaultRename<Impl>::renameSrcRegs(DynInstPtr &inst,unsigned tid) // Will need to mark dependencies though. for (int src_idx = 0; src_idx < num_src_regs; src_idx++) { RegIndex src_reg = inst->srcRegIdx(src_idx); + RegIndex flat_src_reg = src_reg; + if (src_reg < TheISA::FP_Base_DepTag) { + flat_src_reg = TheISA::flattenIntIndex(inst->tcBase(), src_reg); + DPRINTF(Rename, "Flattening index %d to %d.\n", (int)src_reg, (int)flat_src_reg); + } + inst->flattenSrcReg(src_idx, flat_src_reg); // Look up the source registers to get the phys. register they've // been renamed to, and set the sources to those registers. - PhysRegIndex renamed_reg = renameMap[tid]->lookup(src_reg); + PhysRegIndex renamed_reg = renameMap[tid]->lookup(flat_src_reg); DPRINTF(Rename, "[tid:%u]: Looking up arch reg %i, got " - "physical reg %i.\n", tid, (int)src_reg, + "physical reg %i.\n", tid, (int)flat_src_reg, (int)renamed_reg); inst->renameSrcReg(src_idx, renamed_reg); @@ -996,20 +1046,27 @@ DefaultRename<Impl>::renameDestRegs(DynInstPtr &inst,unsigned tid) // Rename the destination registers. for (int dest_idx = 0; dest_idx < num_dest_regs; dest_idx++) { RegIndex dest_reg = inst->destRegIdx(dest_idx); + RegIndex flat_dest_reg = dest_reg; + if (dest_reg < TheISA::FP_Base_DepTag) { + flat_dest_reg = TheISA::flattenIntIndex(inst->tcBase(), dest_reg); + DPRINTF(Rename, "Flattening index %d to %d.\n", (int)dest_reg, (int)flat_dest_reg); + } + + inst->flattenDestReg(dest_idx, flat_dest_reg); // Get the physical register that the destination will be // renamed to. - rename_result = renameMap[tid]->rename(dest_reg); + rename_result = renameMap[tid]->rename(flat_dest_reg); //Mark Scoreboard entry as not ready scoreboard->unsetReg(rename_result.first); DPRINTF(Rename, "[tid:%u]: Renaming arch reg %i to physical " - "reg %i.\n", tid, (int)dest_reg, + "reg %i.\n", tid, (int)flat_dest_reg, (int)rename_result.first); // Record the rename information so that a history can be kept. - RenameHistory hb_entry(inst->seqNum, dest_reg, + RenameHistory hb_entry(inst->seqNum, flat_dest_reg, rename_result.first, rename_result.second); @@ -1233,12 +1290,24 @@ DefaultRename<Impl>::checkSignalsAndUpdate(unsigned tid) if (renameStatus[tid] == Squashing) { // Switch status to running if rename isn't being told to block or // squash this cycle. - DPRINTF(Rename, "[tid:%u]: Done squashing, switching to running.\n", - tid); + if (resumeSerialize) { + DPRINTF(Rename, "[tid:%u]: Done squashing, switching to serialize.\n", + tid); - renameStatus[tid] = Running; + renameStatus[tid] = SerializeStall; + return true; + } else if (resumeUnblocking) { + DPRINTF(Rename, "[tid:%u]: Done squashing, switching to unblocking.\n", + tid); + renameStatus[tid] = Unblocking; + return true; + } else { + DPRINTF(Rename, "[tid:%u]: Done squashing, switching to running.\n", + tid); - return false; + renameStatus[tid] = Running; + return false; + } } if (renameStatus[tid] == SerializeStall) { diff --git a/src/cpu/o3/sparc/cpu.hh b/src/cpu/o3/sparc/cpu.hh index c4df79832..08ebd2710 100644 --- a/src/cpu/o3/sparc/cpu.hh +++ b/src/cpu/o3/sparc/cpu.hh @@ -37,12 +37,6 @@ #include "cpu/o3/cpu.hh" #include "sim/byteswap.hh" -namespace TheISA -{ - class ITB; - class DTB; -} - class EndQuiesceEvent; namespace Kernel { class Statistics; @@ -61,14 +55,6 @@ class TranslatingPort; template <class Impl> class SparcO3CPU : public FullO3CPU<Impl> { - protected: - typedef TheISA::IntReg IntReg; - typedef TheISA::FloatReg FloatReg; - typedef TheISA::FloatRegBits FloatRegBits; - typedef TheISA::MiscReg MiscReg; - typedef TheISA::RegFile RegFile; - typedef TheISA::MiscRegFile MiscRegFile; - public: typedef O3ThreadState<Impl> ImplState; typedef O3ThreadState<Impl> Thread; @@ -77,13 +63,6 @@ class SparcO3CPU : public FullO3CPU<Impl> /** Constructs an AlphaO3CPU with the given parameters. */ SparcO3CPU(Params *params); -#if FULL_SYSTEM - /** ITB pointer. */ - SparcISA::ITB *itb; - /** DTB pointer. */ - SparcISA::DTB *dtb; -#endif - /** Registers statistics. */ void regStats(); @@ -91,19 +70,19 @@ class SparcO3CPU : public FullO3CPU<Impl> /** Translates instruction requestion. */ Fault translateInstReq(RequestPtr &req, Thread *thread) { - return itb->translate(req, thread->getTC()); + return this->itb->translate(req, thread->getTC()); } /** Translates data read request. */ Fault translateDataReadReq(RequestPtr &req, Thread *thread) { - return dtb->translate(req, thread->getTC(), false); + return this->dtb->translate(req, thread->getTC(), false); } /** Translates data write request. */ Fault translateDataWriteReq(RequestPtr &req, Thread *thread) { - return dtb->translate(req, thread->getTC(), true); + return this->dtb->translate(req, thread->getTC(), true); } #else @@ -127,20 +106,21 @@ class SparcO3CPU : public FullO3CPU<Impl> #endif /** Reads a miscellaneous register. */ - MiscReg readMiscReg(int misc_reg, unsigned tid); + TheISA::MiscReg readMiscReg(int misc_reg, unsigned tid); /** Reads a misc. register, including any side effects the read * might have as defined by the architecture. */ - MiscReg readMiscRegWithEffect(int misc_reg, unsigned tid); + TheISA::MiscReg readMiscRegWithEffect(int misc_reg, unsigned tid); /** Sets a miscellaneous register. */ - void setMiscReg(int misc_reg, const MiscReg &val, unsigned tid); + void setMiscReg(int misc_reg, const TheISA::MiscReg &val, unsigned tid); /** Sets a misc. register, including any side effects the write * might have as defined by the architecture. */ - void setMiscRegWithEffect(int misc_reg, const MiscReg &val, unsigned tid); + void setMiscRegWithEffect(int misc_reg, const TheISA::MiscReg &val, + unsigned tid); /** Initiates a squash of all in-flight instructions for a given * thread. The source of the squash is an external update of @@ -175,10 +155,10 @@ class SparcO3CPU : public FullO3CPU<Impl> */ void syscall(int64_t callnum, int tid); /** Gets a syscall argument. */ - IntReg getSyscallArg(int i, int tid); + TheISA::IntReg getSyscallArg(int i, int tid); /** Used to shift args for indirect syscall. */ - void setSyscallArg(int i, IntReg val, int tid); + void setSyscallArg(int i, TheISA::IntReg val, int tid); /** Sets the return value of a syscall. */ void setSyscallReturn(SyscallReturn return_value, int tid); @@ -204,4 +184,4 @@ class SparcO3CPU : public FullO3CPU<Impl> bool lockFlag; }; -#endif // __CPU_O3_ALPHA_CPU_HH__ +#endif // __CPU_O3_SPARC_CPU_HH__ diff --git a/src/cpu/o3/sparc/cpu_builder.cc b/src/cpu/o3/sparc/cpu_builder.cc index 81f419ee0..3cac89bad 100644 --- a/src/cpu/o3/sparc/cpu_builder.cc +++ b/src/cpu/o3/sparc/cpu_builder.cc @@ -55,8 +55,8 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(DerivO3CPU) #if FULL_SYSTEM SimObjectParam<System *> system; Param<int> cpu_id; - SimObjectParam<AlphaISA::ITB *> itb; - SimObjectParam<AlphaISA::DTB *> dtb; + SimObjectParam<SparcISA::ITB *> itb; + SimObjectParam<SparcISA::DTB *> dtb; Param<Tick> profile; Param<bool> do_quiesce; diff --git a/src/cpu/o3/sparc/cpu_impl.hh b/src/cpu/o3/sparc/cpu_impl.hh index 66bf7d1c0..c039a8fec 100644 --- a/src/cpu/o3/sparc/cpu_impl.hh +++ b/src/cpu/o3/sparc/cpu_impl.hh @@ -55,12 +55,7 @@ #endif template <class Impl> -SparcO3CPU<Impl>::SparcO3CPU(Params *params) -#if FULL_SYSTEM - : FullO3CPU<Impl>(params), itb(params->itb), dtb(params->dtb) -#else - : FullO3CPU<Impl>(params) -#endif +SparcO3CPU<Impl>::SparcO3CPU(Params *params) : FullO3CPU<Impl>(params) { DPRINTF(O3CPU, "Creating SparcO3CPU object.\n"); @@ -172,15 +167,16 @@ SparcO3CPU<Impl>::readMiscRegWithEffect(int misc_reg, unsigned tid) template <class Impl> void -SparcO3CPU<Impl>::setMiscReg(int misc_reg, const MiscReg &val, unsigned tid) +SparcO3CPU<Impl>::setMiscReg(int misc_reg, + const SparcISA::MiscReg &val, unsigned tid) { this->regFile.setMiscReg(misc_reg, val, tid); } template <class Impl> void -SparcO3CPU<Impl>::setMiscRegWithEffect(int misc_reg, const MiscReg &val, - unsigned tid) +SparcO3CPU<Impl>::setMiscRegWithEffect(int misc_reg, + const SparcISA::MiscReg &val, unsigned tid) { this->regFile.setMiscRegWithEffect(misc_reg, val, tid); } @@ -284,35 +280,24 @@ template <class Impl> TheISA::IntReg SparcO3CPU<Impl>::getSyscallArg(int i, int tid) { - return this->readArchIntReg(SparcISA::ArgumentReg0 + i, tid); + TheISA::IntReg idx = TheISA::flattenIntIndex(this->tcBase(tid), + SparcISA::ArgumentReg0 + i); + return this->readArchIntReg(idx, tid); } template <class Impl> void -SparcO3CPU<Impl>::setSyscallArg(int i, IntReg val, int tid) +SparcO3CPU<Impl>::setSyscallArg(int i, TheISA::IntReg val, int tid) { - this->setArchIntReg(SparcISA::ArgumentReg0 + i, val, tid); + TheISA::IntReg idx = TheISA::flattenIntIndex(this->tcBase(tid), + SparcISA::ArgumentReg0 + i); + this->setArchIntReg(idx, val, tid); } template <class Impl> void SparcO3CPU<Impl>::setSyscallReturn(SyscallReturn return_value, int tid) { - // check for error condition. SPARC syscall convention is to - // indicate success/failure in reg the carry bit of the ccr - // and put the return value itself in the standard return value reg (). - if (return_value.successful()) { - // no error, clear XCC.C - this->setMiscReg(SparcISA::MISCREG_CCR, - this->readMiscReg(SparcISA::MISCREG_CCR, tid) & 0xEE, tid); - this->setArchIntReg(SparcISA::ReturnValueReg, - return_value.value(), tid); - } else { - // got an error, set XCC.C - this->setMiscReg(SparcISA::MISCREG_CCR, - this->readMiscReg(SparcISA::MISCREG_CCR, tid) | 0x11, tid); - this->setArchIntReg(SparcISA::ReturnValueReg, - return_value.value(), tid); - } + TheISA::setSyscallReturn(return_value, this->tcBase(tid)); } #endif diff --git a/src/cpu/o3/sparc/dyn_inst.hh b/src/cpu/o3/sparc/dyn_inst.hh index f8d6bb63f..e95ae2fd5 100644 --- a/src/cpu/o3/sparc/dyn_inst.hh +++ b/src/cpu/o3/sparc/dyn_inst.hh @@ -32,6 +32,7 @@ #define __CPU_O3_SPARC_DYN_INST_HH__ #include "arch/sparc/isa_traits.hh" +#include "arch/sparc/types.hh" #include "cpu/base_dyn_inst.hh" #include "cpu/inst_seq.hh" #include "cpu/o3/sparc/cpu.hh" @@ -55,8 +56,8 @@ class SparcDynInst : public BaseDynInst<Impl> public: /** BaseDynInst constructor given a binary instruction. */ - SparcDynInst(TheISA::ExtMachInst inst, Addr PC, - Addr Pred_PC, InstSeqNum seq_num, O3CPU *cpu); + SparcDynInst(TheISA::ExtMachInst inst, Addr PC, Addr NPC, + Addr Pred_PC, Addr Pred_NPC, InstSeqNum seq_num, O3CPU *cpu); /** BaseDynInst constructor given a static inst pointer. */ SparcDynInst(StaticInstPtr &_staticInst); @@ -105,6 +106,45 @@ class SparcDynInst : public BaseDynInst<Impl> this->threadNumber); } + /** Reads a miscellaneous register. */ + TheISA::MiscReg readMiscRegOperand(const StaticInst *si, int idx) + { + return this->cpu->readMiscReg( + si->srcRegIdx(idx) - TheISA::Ctrl_Base_DepTag, + this->threadNumber); + } + + /** Reads a misc. register, including any side-effects the read + * might have as defined by the architecture. + */ + TheISA::MiscReg readMiscRegOperandWithEffect(const StaticInst *si, int idx) + { + return this->cpu->readMiscRegWithEffect( + si->srcRegIdx(idx) - TheISA::Ctrl_Base_DepTag, + this->threadNumber); + } + + /** Sets a misc. register. */ + void setMiscRegOperand(const StaticInst * si, + int idx, const TheISA::MiscReg &val) + { + this->instResult.integer = val; + return this->cpu->setMiscReg( + si->destRegIdx(idx) - TheISA::Ctrl_Base_DepTag, + val, this->threadNumber); + } + + /** Sets a misc. register, including any side-effects the write + * might have as defined by the architecture. + */ + void setMiscRegOperandWithEffect( + const StaticInst *si, int idx, const TheISA::MiscReg &val) + { + return this->cpu->setMiscRegWithEffect( + si->destRegIdx(idx) - TheISA::Ctrl_Base_DepTag, + val, this->threadNumber); + } + #if FULL_SYSTEM /** Calls hardware return from error interrupt. */ Fault hwrei(); @@ -116,22 +156,6 @@ class SparcDynInst : public BaseDynInst<Impl> void syscall(int64_t callnum); #endif - private: - /** Physical register index of the destination registers of this - * instruction. - */ - PhysRegIndex _destRegIdx[TheISA::MaxInstDestRegs]; - - /** Physical register index of the source registers of this - * instruction. - */ - PhysRegIndex _srcRegIdx[TheISA::MaxInstSrcRegs]; - - /** Physical register index of the previous producers of the - * architected destinations. - */ - PhysRegIndex _prevDestRegIdx[TheISA::MaxInstDestRegs]; - public: // The register accessor methods provide the index of the @@ -145,108 +169,70 @@ class SparcDynInst : public BaseDynInst<Impl> // storage (which is pretty hard to imagine they would have reason // to do). - uint64_t readIntReg(const StaticInst *si, int idx) + uint64_t readIntRegOperand(const StaticInst *si, int idx) { - return this->cpu->readIntReg(_srcRegIdx[idx]); + uint64_t val = this->cpu->readIntReg(this->_srcRegIdx[idx]); + DPRINTF(Sparc, "Reading int reg %d (%d, %d) as %x\n", (int)this->_flatSrcRegIdx[idx], (int)this->_srcRegIdx[idx], idx, val); + return this->cpu->readIntReg(this->_srcRegIdx[idx]); } - TheISA::FloatReg readFloatReg(const StaticInst *si, int idx, int width) + TheISA::FloatReg readFloatRegOperand(const StaticInst *si, + int idx, int width) { - return this->cpu->readFloatReg(_srcRegIdx[idx], width); + return this->cpu->readFloatReg(this->_srcRegIdx[idx], width); } - TheISA::FloatReg readFloatReg(const StaticInst *si, int idx) + TheISA::FloatReg readFloatRegOperand(const StaticInst *si, int idx) { - return this->cpu->readFloatReg(_srcRegIdx[idx]); + return this->cpu->readFloatReg(this->_srcRegIdx[idx]); } - TheISA::FloatRegBits readFloatRegBits(const StaticInst *si, + TheISA::FloatRegBits readFloatRegOperandBits(const StaticInst *si, int idx, int width) { - return this->cpu->readFloatRegBits(_srcRegIdx[idx], width); + return this->cpu->readFloatRegBits(this->_srcRegIdx[idx], width); } - TheISA::FloatRegBits readFloatRegBits(const StaticInst *si, int idx) + TheISA::FloatRegBits readFloatRegOperandBits(const StaticInst *si, int idx) { - return this->cpu->readFloatRegBits(_srcRegIdx[idx]); + return this->cpu->readFloatRegBits(this->_srcRegIdx[idx]); } /** @todo: Make results into arrays so they can handle multiple dest * registers. */ - void setIntReg(const StaticInst *si, int idx, uint64_t val) + void setIntRegOperand(const StaticInst *si, int idx, uint64_t val) { - this->cpu->setIntReg(_destRegIdx[idx], val); - BaseDynInst<Impl>::setIntReg(si, idx, val); + DPRINTF(Sparc, "Setting int reg %d (%d, %d) to %x\n", (int)this->_flatDestRegIdx[idx], (int)this->_destRegIdx[idx], idx, val); + this->cpu->setIntReg(this->_destRegIdx[idx], val); + BaseDynInst<Impl>::setIntRegOperand(si, idx, val); } - void setFloatReg(const StaticInst *si, int idx, + void setFloatRegOperand(const StaticInst *si, int idx, TheISA::FloatReg val, int width) { - this->cpu->setFloatReg(_destRegIdx[idx], val, width); - BaseDynInst<Impl>::setFloatReg(si, idx, val, width); + this->cpu->setFloatReg(this->_destRegIdx[idx], val, width); + BaseDynInst<Impl>::setFloatRegOperand(si, idx, val, width); } - void setFloatReg(const StaticInst *si, int idx, TheISA::FloatReg val) + void setFloatRegOperand(const StaticInst *si, int idx, TheISA::FloatReg val) { - this->cpu->setFloatReg(_destRegIdx[idx], val); - BaseDynInst<Impl>::setFloatReg(si, idx, val); + this->cpu->setFloatReg(this->_destRegIdx[idx], val); + BaseDynInst<Impl>::setFloatRegOperand(si, idx, val); } - void setFloatRegBits(const StaticInst *si, int idx, + void setFloatRegOperandBits(const StaticInst *si, int idx, TheISA::FloatRegBits val, int width) { - this->cpu->setFloatRegBits(_destRegIdx[idx], val, width); - BaseDynInst<Impl>::setFloatRegBits(si, idx, val); + this->cpu->setFloatRegBits(this->_destRegIdx[idx], val, width); + BaseDynInst<Impl>::setFloatRegOperandBits(si, idx, val); } - void setFloatRegBits(const StaticInst *si, + void setFloatRegOperandBits(const StaticInst *si, int idx, TheISA::FloatRegBits val) { - this->cpu->setFloatRegBits(_destRegIdx[idx], val); - BaseDynInst<Impl>::setFloatRegBits(si, idx, val); - } - - /** Returns the physical register index of the i'th destination - * register. - */ - PhysRegIndex renamedDestRegIdx(int idx) const - { - return _destRegIdx[idx]; - } - - /** Returns the physical register index of the i'th source register. */ - PhysRegIndex renamedSrcRegIdx(int idx) const - { - return _srcRegIdx[idx]; - } - - /** Returns the physical register index of the previous physical register - * that remapped to the same logical register index. - */ - PhysRegIndex prevDestRegIdx(int idx) const - { - return _prevDestRegIdx[idx]; - } - - /** Renames a destination register to a physical register. Also records - * the previous physical register that the logical register mapped to. - */ - void renameDestReg(int idx, - PhysRegIndex renamed_dest, - PhysRegIndex previous_rename) - { - _destRegIdx[idx] = renamed_dest; - _prevDestRegIdx[idx] = previous_rename; - } - - /** Renames a source logical register to the physical register which - * has/will produce that logical register's result. - * @todo: add in whether or not the source register is ready. - */ - void renameSrcReg(int idx, PhysRegIndex renamed_src) - { - _srcRegIdx[idx] = renamed_src; + this->cpu->setFloatRegBits(this->_destRegIdx[idx], val); + BaseDynInst<Impl>::setFloatRegOperandBits(si, idx, val); } public: diff --git a/src/cpu/o3/sparc/dyn_inst_impl.hh b/src/cpu/o3/sparc/dyn_inst_impl.hh index 210daace2..c4d30b6f4 100644 --- a/src/cpu/o3/sparc/dyn_inst_impl.hh +++ b/src/cpu/o3/sparc/dyn_inst_impl.hh @@ -31,9 +31,10 @@ #include "cpu/o3/sparc/dyn_inst.hh" template <class Impl> -SparcDynInst<Impl>::SparcDynInst(TheISA::ExtMachInst inst, Addr PC, - Addr Pred_PC, InstSeqNum seq_num, O3CPU *cpu) - : BaseDynInst<Impl>(inst, PC, Pred_PC, seq_num, cpu) +SparcDynInst<Impl>::SparcDynInst(TheISA::ExtMachInst inst, + Addr PC, Addr NPC, Addr Pred_PC, Addr Pred_NPC, + InstSeqNum seq_num, O3CPU *cpu) + : BaseDynInst<Impl>(inst, PC, NPC, Pred_PC, Pred_NPC, seq_num, cpu) { initVars(); } @@ -53,11 +54,11 @@ SparcDynInst<Impl>::initVars() // as the normal register entries. It will allow the IQ to work // without any modifications. for (int i = 0; i < this->staticInst->numDestRegs(); i++) { - _destRegIdx[i] = this->staticInst->destRegIdx(i); + this->_destRegIdx[i] = this->staticInst->destRegIdx(i); } for (int i = 0; i < this->staticInst->numSrcRegs(); i++) { - _srcRegIdx[i] = this->staticInst->srcRegIdx(i); + this->_srcRegIdx[i] = this->staticInst->srcRegIdx(i); this->_readySrcRegIdx[i] = 0; } } @@ -126,7 +127,8 @@ template <class Impl> bool SparcDynInst<Impl>::simPalCheck(int palFunc) { - return this->cpu->simPalCheck(palFunc, this->threadNumber); + panic("simPalCheck called, but there's no PAL in SPARC!\n"); + return false; } #else template <class Impl> diff --git a/src/cpu/o3/sparc/thread_context.hh b/src/cpu/o3/sparc/thread_context.hh index 69d1e2d04..7497959e4 100644 --- a/src/cpu/o3/sparc/thread_context.hh +++ b/src/cpu/o3/sparc/thread_context.hh @@ -36,12 +36,6 @@ class SparcTC : public O3ThreadContext<Impl> { public: #if FULL_SYSTEM - /** Returns a pointer to the ITB. */ - virtual SparcISA::ITB *getITBPtr() { return this->cpu->itb; } - - /** Returns a pointer to the DTB. */ - virtual SparcISA::DTB *getDTBPtr() { return this->cpu->dtb; } - /** Returns pointer to the quiesce event. */ virtual EndQuiesceEvent *getQuiesceEvent() { @@ -62,7 +56,7 @@ class SparcTC : public O3ThreadContext<Impl> virtual void changeRegFileContext(TheISA::RegContextParam param, TheISA::RegContextVal val) { - panic("This doesn't make sense!\n"); + //XXX Ignore this for now. This -really- needs to get fixed. } diff --git a/src/cpu/o3/thread_context.hh b/src/cpu/o3/thread_context.hh index 390569c3d..4987d6eb4 100755 --- a/src/cpu/o3/thread_context.hh +++ b/src/cpu/o3/thread_context.hh @@ -66,6 +66,14 @@ class O3ThreadContext : public ThreadContext /** Pointer to the thread state that this TC corrseponds to. */ O3ThreadState<Impl> *thread; +#if FULL_SYSTEM + /** Returns a pointer to the ITB. */ + TheISA::ITB *getITBPtr() { return cpu->itb; } + + /** Returns a pointer to the DTB. */ + TheISA::DTB *getDTBPtr() { return cpu->dtb; } +#endif + /** Returns a pointer to this CPU. */ virtual BaseCPU *getCpuPtr() { return cpu; } diff --git a/src/cpu/o3/thread_context_impl.hh b/src/cpu/o3/thread_context_impl.hh index afebf294f..af98fa1f5 100755 --- a/src/cpu/o3/thread_context_impl.hh +++ b/src/cpu/o3/thread_context_impl.hh @@ -29,6 +29,7 @@ * Korey Sewell */ +#include "arch/regfile.hh" #include "cpu/o3/thread_context.hh" #include "cpu/quiesce_event.hh" @@ -305,6 +306,7 @@ template <class Impl> uint64_t O3ThreadContext<Impl>::readIntReg(int reg_idx) { + reg_idx = TheISA::flattenIntIndex(this, reg_idx); return cpu->readArchIntReg(reg_idx, thread->readTid()); } @@ -349,6 +351,7 @@ template <class Impl> void O3ThreadContext<Impl>::setIntReg(int reg_idx, uint64_t val) { + reg_idx = TheISA::flattenIntIndex(this, reg_idx); cpu->setArchIntReg(reg_idx, val, thread->readTid()); // Squash if we're not already in a state update mode. diff --git a/src/cpu/ozone/cpu.hh b/src/cpu/ozone/cpu.hh index 0da446c9c..baea7a546 100644 --- a/src/cpu/ozone/cpu.hh +++ b/src/cpu/ozone/cpu.hh @@ -219,11 +219,19 @@ class OzoneCPU : public BaseCPU uint64_t readNextNPC() { - return 0; +#if ISA_HAS_DELAY_SLOT + panic("Ozone needs to support nextNPC"); +#else + return thread->nextPC + sizeof(TheISA::MachInst); +#endif } void setNextNPC(uint64_t val) - { } + { +#if ISA_HAS_DELAY_SLOT + panic("Ozone needs to support nextNPC"); +#endif + } public: // ISA stuff: diff --git a/src/cpu/simple/base.hh b/src/cpu/simple/base.hh index 31fd00977..c4853b916 100644 --- a/src/cpu/simple/base.hh +++ b/src/cpu/simple/base.hh @@ -54,15 +54,15 @@ namespace TheISA } class MemObject; -class RemoteGDB; -class GDBListener; - #else class Process; #endif // FULL_SYSTEM +class RemoteGDB; +class GDBListener; + class ThreadContext; class Checkpoint; @@ -304,6 +304,31 @@ class BaseSimpleCPU : public BaseCPU return thread->setMiscRegWithEffect(misc_reg, val); } + MiscReg readMiscRegOperand(const StaticInst *si, int idx) + { + int reg_idx = si->srcRegIdx(idx) - TheISA::Ctrl_Base_DepTag; + return thread->readMiscReg(reg_idx); + } + + MiscReg readMiscRegOperandWithEffect(const StaticInst *si, int idx) + { + int reg_idx = si->srcRegIdx(idx) - TheISA::Ctrl_Base_DepTag; + return thread->readMiscRegWithEffect(reg_idx); + } + + void setMiscRegOperand(const StaticInst *si, int idx, const MiscReg &val) + { + int reg_idx = si->destRegIdx(idx) - TheISA::Ctrl_Base_DepTag; + return thread->setMiscReg(reg_idx, val); + } + + void setMiscRegOperandWithEffect( + const StaticInst *si, int idx, const MiscReg &val) + { + int reg_idx = si->destRegIdx(idx) - TheISA::Ctrl_Base_DepTag; + return thread->setMiscRegWithEffect(reg_idx, val); + } + #if FULL_SYSTEM Fault hwrei() { return thread->hwrei(); } void ev5_trap(Fault fault) { fault->invoke(tc); } diff --git a/src/cpu/simple_thread.hh b/src/cpu/simple_thread.hh index 10bbe292c..f2f79c070 100644 --- a/src/cpu/simple_thread.hh +++ b/src/cpu/simple_thread.hh @@ -33,6 +33,8 @@ #define __CPU_SIMPLE_THREAD_HH__ #include "arch/isa_traits.hh" +#include "arch/regfile.hh" +#include "arch/syscallreturn.hh" #include "config/full_system.hh" #include "cpu/thread_context.hh" #include "cpu/thread_state.hh" @@ -250,7 +252,7 @@ class SimpleThread : public ThreadState // uint64_t readIntReg(int reg_idx) { - return regs.readIntReg(reg_idx); + return regs.readIntReg(TheISA::flattenIntIndex(getTC(), reg_idx)); } FloatReg readFloatReg(int reg_idx, int width) @@ -275,7 +277,7 @@ class SimpleThread : public ThreadState void setIntReg(int reg_idx, uint64_t val) { - regs.setIntReg(reg_idx, val); + regs.setIntReg(TheISA::flattenIntIndex(getTC(), reg_idx), val); } void setFloatReg(int reg_idx, FloatReg val, int width) @@ -376,18 +378,20 @@ class SimpleThread : public ThreadState #if !FULL_SYSTEM TheISA::IntReg getSyscallArg(int i) { - return regs.readIntReg(TheISA::ArgumentReg0 + i); + return regs.readIntReg(TheISA::flattenIntIndex(getTC(), + TheISA::ArgumentReg0 + i)); } // used to shift args for indirect syscall void setSyscallArg(int i, TheISA::IntReg val) { - regs.setIntReg(TheISA::ArgumentReg0 + i, val); + regs.setIntReg(TheISA::flattenIntIndex(getTC(), + TheISA::ArgumentReg0 + i), val); } void setSyscallReturn(SyscallReturn return_value) { - TheISA::setSyscallReturn(return_value, ®s); + TheISA::setSyscallReturn(return_value, getTC()); } void syscall(int64_t callnum) diff --git a/src/cpu/thread_context.hh b/src/cpu/thread_context.hh index bb9cc9e16..16e491fd3 100644 --- a/src/cpu/thread_context.hh +++ b/src/cpu/thread_context.hh @@ -32,13 +32,13 @@ #define __CPU_THREAD_CONTEXT_HH__ #include "arch/regfile.hh" -#include "arch/syscallreturn.hh" #include "arch/types.hh" #include "config/full_system.hh" #include "mem/request.hh" #include "sim/faults.hh" #include "sim/host.hh" #include "sim/serialize.hh" +#include "sim/syscallreturn.hh" #include "sim/byteswap.hh" // @todo: Figure out a more architecture independent way to obtain the ITB and |