#include "arch/isa_traits.hh"
#include "base/statistics.hh"
#include "cpu/exec_context.hh"
#include "cpu/exetrace.hh"
#include "cpu/ozone/front_end.hh"
#include "mem/mem_interface.hh"
#include "sim/byte_swap.hh"

using namespace TheISA;

template <class Impl>
FrontEnd<Impl>::FrontEnd(Params *params)
    : branchPred(params),
      cacheCompletionEvent(this),
      icacheInterface(params->icacheInterface),
      instBufferSize(0),
      maxInstBufferSize(params->maxInstBufferSize),
      width(params->frontEndWidth),
      freeRegs(params->numPhysicalRegs),
      numPhysRegs(params->numPhysicalRegs),
      serializeNext(false)
{
    status = Idle;

    // Setup branch predictor.

    // Setup Memory Request
    memReq = new MemReq();
    memReq->asid = 0;
    memReq->data = new uint8_t[64];

    // Size of cache block.
    cacheBlkSize = icacheInterface ? icacheInterface->getBlockSize() : 64;

    assert(isPowerOf2(cacheBlkSize));

    // Create mask to get rid of offset bits.
    cacheBlkMask = (cacheBlkSize - 1);

    // Create space to store a cache line.
    cacheData = new uint8_t[cacheBlkSize];

    fetchCacheLineNextCycle = true;

    cacheBlkValid = false;

#if !FULL_SYSTEM
    pTable = params->pTable;
#endif
}

template <class Impl>
std::string
FrontEnd<Impl>::name() const
{
    return cpu->name() + ".frontend";
}

template <class Impl>
void
FrontEnd<Impl>::setCommBuffer(TimeBuffer<CommStruct> *_comm)
{
    comm = _comm;
    // @todo: Hardcoded for now.  Allow this to be set by a latency.
    fromCommit = comm->getWire(-1);
}

template <class Impl>
void
FrontEnd<Impl>::setXC(ExecContext *xc_ptr)
{
    xc = xc_ptr;
    memReq->xc = xc;
}

template <class Impl>
void
FrontEnd<Impl>::regStats()
{
    icacheStallCycles
        .name(name() + ".icacheStallCycles")
        .desc("Number of cycles fetch is stalled on an Icache miss")
        .prereq(icacheStallCycles);

    fetchedInsts
        .name(name() + ".fetchedInsts")
        .desc("Number of instructions fetch has processed")
        .prereq(fetchedInsts);

    fetchedBranches
        .name(name() + ".fetchedBranches")
        .desc("Number of fetched branches")
        .prereq(fetchedBranches);

    predictedBranches
        .name(name() + ".predictedBranches")
        .desc("Number of branches that fetch has predicted taken")
        .prereq(predictedBranches);

    fetchCycles
        .name(name() + ".fetchCycles")
        .desc("Number of cycles fetch has run and was not squashing or"
              " blocked")
        .prereq(fetchCycles);

    fetchIdleCycles
        .name(name() + ".fetchIdleCycles")
        .desc("Number of cycles fetch was idle")
        .prereq(fetchIdleCycles);

    fetchSquashCycles
        .name(name() + ".fetchSquashCycles")
        .desc("Number of cycles fetch has spent squashing")
        .prereq(fetchSquashCycles);

    fetchBlockedCycles
        .name(name() + ".fetchBlockedCycles")
        .desc("Number of cycles fetch has spent blocked")
        .prereq(fetchBlockedCycles);

    fetchedCacheLines
        .name(name() + ".fetchedCacheLines")
        .desc("Number of cache lines fetched")
        .prereq(fetchedCacheLines);

    fetchNisnDist
        .init(/* base value */ 0,
              /* last value */ width,
              /* bucket size */ 1)
        .name(name() + ".rateDist")
        .desc("Number of instructions fetched each cycle (Total)")
        .flags(Stats::pdf);

    idleRate
        .name(name() + ".idleRate")
        .desc("Percent of cycles fetch was idle")
        .prereq(idleRate);
    idleRate = fetchIdleCycles * 100 / cpu->numCycles;

    branchRate
        .name(name() + ".branchRate")
        .desc("Number of branch fetches per cycle")
        .flags(Stats::total);
    branchRate = fetchedBranches / cpu->numCycles;

    fetchRate
        .name(name() + ".rate")
        .desc("Number of inst fetches per cycle")
        .flags(Stats::total);
    fetchRate = fetchedInsts / cpu->numCycles;

    IFQCount
        .name(name() + ".IFQ:count")
        .desc("cumulative IFQ occupancy")
        ;

    IFQFcount
        .name(name() + ".IFQ:fullCount")
        .desc("cumulative IFQ full count")
        .flags(Stats::total)
        ;

    IFQOccupancy
        .name(name() + ".IFQ:occupancy")
        .desc("avg IFQ occupancy (inst's)")
        ;
    IFQOccupancy = IFQCount / cpu->numCycles;

    IFQLatency
        .name(name() + ".IFQ:latency")
        .desc("avg IFQ occupant latency (cycle's)")
        .flags(Stats::total)
        ;

    IFQFullRate
        .name(name() + ".IFQ:fullRate")
        .desc("fraction of time (cycles) IFQ was full")
        .flags(Stats::total);
        ;
    IFQFullRate = IFQFcount * Stats::constant(100) / cpu->numCycles;

    dispatchCountStat
        .name(name() + ".DIS:count")
        .desc("cumulative count of dispatched insts")
        .flags(Stats::total)
        ;

    dispatchedSerializing
        .name(name() + ".DIS:serializingInsts")
        .desc("count of serializing insts dispatched")
        .flags(Stats::total)
        ;

    dispatchedTempSerializing
        .name(name() + ".DIS:tempSerializingInsts")
        .desc("count of temporary serializing insts dispatched")
        .flags(Stats::total)
        ;

    dispatchSerializeStallCycles
        .name(name() + ".DIS:serializeStallCycles")
        .desc("count of cycles dispatch stalled for serializing inst")
        .flags(Stats::total)
        ;

    dispatchRate
        .name(name() + ".DIS:rate")
        .desc("dispatched insts per cycle")
        .flags(Stats::total)
        ;
    dispatchRate = dispatchCountStat / cpu->numCycles;

    regIntFull
        .name(name() + ".REG:int:full")
        .desc("number of cycles where there were no INT registers")
        ;

    regFpFull
        .name(name() + ".REG:fp:full")
        .desc("number of cycles where there were no FP registers")
        ;
    IFQLatency = IFQOccupancy / dispatchRate;

    branchPred.regStats();
}

template <class Impl>
void
FrontEnd<Impl>::tick()
{
    // @todo: Maybe I want to just have direct communication...
    if (fromCommit->doneSeqNum) {
        branchPred.update(fromCommit->doneSeqNum, 0);
    }

    IFQCount += instBufferSize;
    IFQFcount += instBufferSize == maxInstBufferSize;

    // Fetch cache line
    if (status == IcacheMissComplete) {
        cacheBlkValid = true;

        status = Running;
        if (barrierInst)
            status = SerializeBlocked;
        if (freeRegs <= 0)
            status = RenameBlocked;
        checkBE();
    } else if (status == IcacheMissStall) {
        DPRINTF(FE, "Still in Icache miss stall.\n");
        icacheStallCycles++;
        return;
    }

    if (status == RenameBlocked || status == SerializeBlocked ||
        status == BEBlocked) {
        // This might cause the front end to run even though it
        // shouldn't, but this should only be a problem for one cycle.
        // Also will cause a one cycle bubble between changing state
        // and restarting.
        DPRINTF(FE, "In blocked status.\n");

        fetchBlockedCycles++;

        if (status == SerializeBlocked) {
            dispatchSerializeStallCycles++;
        }
        updateStatus();
        return;
    } else if (status != IcacheMissComplete) {
        if (fetchCacheLineNextCycle) {
            Fault fault = fetchCacheLine();
            if (fault != NoFault) {
                handleFault(fault);
                return;
            }
            fetchCacheLineNextCycle = false;
        }
        // If miss, stall until it returns.
        if (status == IcacheMissStall) {
            // Tell CPU to not tick me for now.
            return;
        }
    }

    fetchCycles++;

    int num_inst = 0;

    // Otherwise loop and process instructions.
    // One way to hack infinite width is to set width and maxInstBufferSize
    // both really high.  Inelegant, but probably will work.
    while (num_inst < width &&
           instBufferSize < maxInstBufferSize) {
        // Get instruction from cache line.
        DynInstPtr inst = getInstFromCacheline();

        if (!inst) {
            // PC is no longer in the cache line, end fetch.
            // Might want to check this at the end of the cycle so that
            // there's no cycle lost to checking for a new cache line.
            DPRINTF(FE, "Need to get new cache line\n");
            fetchCacheLineNextCycle = true;
            break;
        }

        // if (generalizeFetch) {
        processInst(inst);

        if (status == SerializeBlocked) {
            break;
        }

        // Possibly push into a time buffer that estimates the front end
        // latency
        instBuffer.push_back(inst);
        ++instBufferSize;
        ++num_inst;
        // } else {
        // fetch(num_inst);
        // decode(num_inst);
        // rename(num_inst);
        // }

        if (inst->predTaken()) {
            // Start over with tick?
            break;
        } else if (freeRegs <= 0) {
            DPRINTF(FE, "Ran out of free registers to rename to!\n");
            status = RenameBlocked;
            break;
        } else if (serializeNext) {
            break;
        }
    }

    fetchNisnDist.sample(num_inst);
    checkBE();

    DPRINTF(FE, "Num insts processed: %i, Inst Buffer size: %i, Free "
            "Regs %i\n", num_inst, instBufferSize, freeRegs);
}

template <class Impl>
Fault
FrontEnd<Impl>::fetchCacheLine()
{
    // Read a cache line, based on the current PC.
#if FULL_SYSTEM
    // Flag to say whether or not address is physical addr.
    unsigned flags = cpu->inPalMode() ? PHYSICAL : 0;
#else
    unsigned flags = 0;
#endif // FULL_SYSTEM
    Fault fault = NoFault;

    // Align the fetch PC so it's at the start of a cache block.
    Addr fetch_PC = icacheBlockAlignPC(PC);

    DPRINTF(FE, "Fetching cache line starting at %#x.\n", fetch_PC);

    // Setup the memReq to do a read of the first isntruction's address.
    // Set the appropriate read size and flags as well.
    memReq->cmd = Read;
    memReq->reset(fetch_PC, cacheBlkSize, flags);

    // Translate the instruction request.
    fault = cpu->translateInstReq(memReq);

    // In the case of faults, the fetch stage may need to stall and wait
    // on what caused the fetch (ITB or Icache miss).
//    assert(fault == NoFault);

    // Now do the timing access to see whether or not the instruction
    // exists within the cache.
    if (icacheInterface && fault == NoFault) {
        memReq->completionEvent = NULL;

        memReq->time = curTick;

        MemAccessResult res = icacheInterface->access(memReq);

        // If the cache missed then schedule an event to wake
        // up this stage once the cache miss completes.
        if (icacheInterface->doEvents() && res != MA_HIT) {
            memReq->completionEvent = new ICacheCompletionEvent(this);

            status = IcacheMissStall;

            cacheBlkValid = false;

            DPRINTF(FE, "Cache miss.\n");
        }  else {
            DPRINTF(FE, "Cache hit.\n");

            cacheBlkValid = true;

            memcpy(cacheData, memReq->data, memReq->size);
        }
    }

    // Note that this will set the cache block PC a bit earlier than it should
    // be set.
    cacheBlkPC = fetch_PC;

    ++fetchedCacheLines;

    DPRINTF(FE, "Done fetching cache line.\n");

    return fault;
}

template <class Impl>
void
FrontEnd<Impl>::processInst(DynInstPtr &inst)
{
    if (processBarriers(inst)) {
        return;
    }

    Addr inst_PC = inst->readPC();

//    BPredInfo bp_info = branchPred.lookup(inst_PC);
    if (!inst->isControl()) {
        inst->setPredTarg(inst->readNextPC());
    } else {
        fetchedBranches++;
        if (branchPred.predict(inst, inst_PC, inst->threadNumber)) {
            predictedBranches++;
        }
    }

    Addr next_PC = inst->readPredTarg();

    DPRINTF(FE, "[sn:%lli] Predicted and processed inst PC %#x, next PC "
            "%#x\n", inst->seqNum, inst_PC, next_PC);

//    inst->setNextPC(next_PC);
//    inst->setBPredInfo(bp_info);

    // Not sure where I should set this
    PC = next_PC;

    renameInst(inst);
}

template <class Impl>
bool
FrontEnd<Impl>::processBarriers(DynInstPtr &inst)
{
    if (serializeNext) {
        inst->setSerializeBefore();
        serializeNext = false;
    } else if (!inst->isSerializing()) {
        return false;
    }

    if (inst->isSerializeBefore() && !inst->isSerializeHandled()) {
        DPRINTF(FE, "Serialize before instruction encountered.\n");

        if (!inst->isTempSerializeBefore()) {
            dispatchedSerializing++;
            inst->setSerializeHandled();
        } else {
            dispatchedTempSerializing++;
        }

        // Change status over to BarrierStall so that other stages know
        // what this is blocked on.
        status = SerializeBlocked;

        barrierInst = inst;
        return true;
    } else if (inst->isSerializeAfter() && !inst->isSerializeHandled()) {
        DPRINTF(FE, "Serialize after instruction encountered.\n");

        inst->setSerializeHandled();

        dispatchedSerializing++;

        serializeNext = true;
        return false;
    }
    return false;
}

template <class Impl>
void
FrontEnd<Impl>::handleFault(Fault &fault)
{
    DPRINTF(FE, "Fault at fetch, telling commit\n");
    backEnd->fetchFault(fault);
    // We're blocked on the back end until it handles this fault.
    status = BEBlocked;
}

template <class Impl>
void
FrontEnd<Impl>::squash(const InstSeqNum &squash_num, const Addr &next_PC,
                       const bool is_branch, const bool branch_taken)
{
    DPRINTF(FE, "Squashing from [sn:%lli], setting PC to %#x\n",
            squash_num, next_PC);

    while (!instBuffer.empty() &&
           instBuffer.back()->seqNum > squash_num) {
        DynInstPtr inst = instBuffer.back();

        DPRINTF(FE, "Squashing instruction [sn:%lli] PC %#x\n",
                inst->seqNum, inst->readPC());

        inst->clearDependents();

        instBuffer.pop_back();
        --instBufferSize;

        // Fix up branch predictor if necessary.
//        branchPred.undo(inst->getBPredInfo());

        freeRegs+= inst->numDestRegs();
    }

    // Copy over rename table from the back end.
    renameTable.copyFrom(backEnd->renameTable);

    PC = next_PC;

    // Update BP with proper information.
    if (is_branch) {
        branchPred.squash(squash_num, next_PC, branch_taken, 0);
    } else {
        branchPred.squash(squash_num, 0);
    }

    // Clear the icache miss if it's outstanding.
    if (status == IcacheMissStall && icacheInterface) {
        DPRINTF(FE, "Squashing outstanding Icache miss.\n");
        icacheInterface->squash(0);
    }

    if (status == SerializeBlocked) {
        assert(barrierInst->seqNum > squash_num);
        barrierInst = NULL;
    }

    // Unless this squash originated from the front end, we're probably
    // in running mode now.
    // Actually might want to make this latency dependent.
    status = Running;
    fetchCacheLineNextCycle = true;
}

template <class Impl>
typename Impl::DynInstPtr
FrontEnd<Impl>::getInst()
{
    if (instBufferSize == 0) {
        return NULL;
    }

    DynInstPtr inst = instBuffer.front();

    instBuffer.pop_front();

    --instBufferSize;

    dispatchCountStat++;

    return inst;
}

template <class Impl>
void
FrontEnd<Impl>::processCacheCompletion()
{
    DPRINTF(FE, "Processing cache completion\n");

    // Do something here.
    if (status != IcacheMissStall) {
        DPRINTF(FE, "Previous fetch was squashed.\n");
        return;
    }

    status = IcacheMissComplete;

/*    if (checkStall(tid)) {
        fetchStatus[tid] = Blocked;
    } else {
        fetchStatus[tid] = IcacheMissComplete;
    }
*/
    memcpy(cacheData, memReq->data, memReq->size);

    // Reset the completion event to NULL.
    memReq->completionEvent = NULL;
}

template <class Impl>
void
FrontEnd<Impl>::addFreeRegs(int num_freed)
{
    if (status == RenameBlocked && freeRegs + num_freed > 0) {
        status = Running;
    }

    freeRegs+= num_freed;

    assert(freeRegs <= numPhysRegs);
}

template <class Impl>
bool
FrontEnd<Impl>::updateStatus()
{
//    bool rename_block = freeRegs <= 0;
    bool serialize_block = !backEnd->robEmpty() || instBufferSize;
    bool be_block = cpu->decoupledFrontEnd ? false : backEnd->isBlocked();
    bool ret_val = false;
/*
  // Should already be handled through addFreeRegs function
    if (status == RenameBlocked && !rename_block) {
        status = Running;
        ret_val = true;
    }
*/

    if (status == SerializeBlocked && !serialize_block) {
        status = SerializeComplete;
        ret_val = true;
    }

    if (status == BEBlocked && !be_block) {
        if (barrierInst) {
            status = SerializeBlocked;
        } else {
            status = Running;
        }
        ret_val = true;
    }
    return ret_val;
}

template <class Impl>
void
FrontEnd<Impl>::checkBE()
{
    bool be_block = cpu->decoupledFrontEnd ? false : backEnd->isBlocked();
    if (be_block) {
        if (status == Running || status == Idle) {
            status = BEBlocked;
        }
    }
}

template <class Impl>
typename Impl::DynInstPtr
FrontEnd<Impl>::getInstFromCacheline()
{
    if (status == SerializeComplete) {
        DynInstPtr inst = barrierInst;
        status = Running;
        barrierInst = NULL;
        return inst;
    }

    InstSeqNum inst_seq;
    MachInst inst;
    // @todo: Fix this magic number used here to handle word offset (and
    // getting rid of PAL bit)
    unsigned offset = (PC & cacheBlkMask) & ~3;

    // PC of inst is not in this cache block
    if (PC >= (cacheBlkPC + cacheBlkSize) || PC < cacheBlkPC || !cacheBlkValid) {
//        DPRINTF(OoOCPU, "OoOCPU: PC is not in this cache block\n");
//        DPRINTF(OoOCPU, "OoOCPU: PC: %#x, cacheBlkPC: %#x, cacheBlkValid: %i",
//                PC, cacheBlkPC, cacheBlkValid);
//        panic("Instruction not in cache line or cache line invalid!");
        return NULL;
    }

    //////////////////////////
    // Fetch one instruction
    //////////////////////////

    // Get a sequence number.
    inst_seq = getAndIncrementInstSeq();

    // Make sure this is a valid index.
    assert(offset <= cacheBlkSize - sizeof(MachInst));

    // Get the instruction from the array of the cache line.
    inst = htog(*reinterpret_cast<MachInst *>(&cacheData[offset]));

    ExtMachInst decode_inst = TheISA::makeExtMI(inst, PC);

    // Create a new DynInst from the instruction fetched.
    DynInstPtr instruction = new DynInst(decode_inst, PC, PC+sizeof(MachInst),
                                         inst_seq, cpu);

    instruction->setState(thread);

    DPRINTF(FE, "Instruction [sn:%lli] created, with PC %#x\n%s\n",
            inst_seq, instruction->readPC(),
            instruction->staticInst->disassemble(PC));

    instruction->traceData =
        Trace::getInstRecord(curTick, xc, cpu,
                             instruction->staticInst,
                             instruction->readPC(), 0);

    // Increment stat of fetched instructions.
    ++fetchedInsts;

    return instruction;
}

template <class Impl>
void
FrontEnd<Impl>::renameInst(DynInstPtr &inst)
{
    DynInstPtr src_inst = NULL;
    int num_src_regs = inst->numSrcRegs();
    if (num_src_regs == 0) {
        inst->setCanIssue();
    } else {
        for (int i = 0; i < num_src_regs; ++i) {
            src_inst = renameTable[inst->srcRegIdx(i)];

            inst->setSrcInst(src_inst, i);

            DPRINTF(FE, "[sn:%lli]: Src reg %i is inst [sn:%lli]\n",
                    inst->seqNum, (int)inst->srcRegIdx(i), src_inst->seqNum);

            if (src_inst->isCompleted()) {
                DPRINTF(FE, "Reg ready.\n");
                inst->markSrcRegReady(i);
            } else {
                DPRINTF(FE, "Adding to dependent list.\n");
                src_inst->addDependent(inst);
            }
        }
    }

    for (int i = 0; i < inst->numDestRegs(); ++i) {
        RegIndex idx = inst->destRegIdx(i);

        DPRINTF(FE, "Dest reg %i is now inst [sn:%lli], was previously "
                "[sn:%lli]\n",
                (int)inst->destRegIdx(i), inst->seqNum,
                renameTable[idx]->seqNum);

        inst->setPrevDestInst(renameTable[idx], i);

        renameTable[idx] = inst;
        --freeRegs;
    }
}

template <class Impl>
void
FrontEnd<Impl>::dumpInsts()
{
    cprintf("instBuffer size: %i\n", instBuffer.size());

    InstBuffIt buff_it = instBuffer.begin();

    for (int num = 0; buff_it != instBuffer.end(); num++) {
        cprintf("Instruction:%i\nPC:%#x\n[tid:%i]\n[sn:%lli]\nIssued:%i\n"
                "Squashed:%i\n\n",
                num, (*buff_it)->readPC(), (*buff_it)->threadNumber,
                (*buff_it)->seqNum, (*buff_it)->isIssued(),
                (*buff_it)->isSquashed());
        buff_it++;
    }

}

template <class Impl>
FrontEnd<Impl>::ICacheCompletionEvent::ICacheCompletionEvent(FrontEnd *fe)
    : Event(&mainEventQueue, Delayed_Writeback_Pri), frontEnd(fe)
{
    this->setFlags(Event::AutoDelete);
}

template <class Impl>
void
FrontEnd<Impl>::ICacheCompletionEvent::process()
{
    frontEnd->processCacheCompletion();
}

template <class Impl>
const char *
FrontEnd<Impl>::ICacheCompletionEvent::description()
{
    return "ICache completion event";
}