O3 CPU now handles being used with the sampler.

cpu/o3/2bit_local_pred.cc:
cpu/o3/2bit_local_pred.hh:
cpu/o3/bpred_unit.hh:
cpu/o3/bpred_unit_impl.hh:
cpu/o3/btb.cc:
cpu/o3/btb.hh:
cpu/o3/commit.hh:
cpu/o3/commit_impl.hh:
cpu/o3/cpu.cc:
cpu/o3/cpu.hh:
cpu/o3/decode.hh:
cpu/o3/decode_impl.hh:
cpu/o3/fetch.hh:
cpu/o3/fetch_impl.hh:
cpu/o3/fu_pool.cc:
cpu/o3/fu_pool.hh:
cpu/o3/iew.hh:
cpu/o3/iew_impl.hh:
cpu/o3/inst_queue.hh:
cpu/o3/inst_queue_impl.hh:
cpu/o3/lsq.hh:
cpu/o3/lsq_impl.hh:
cpu/o3/lsq_unit.hh:
cpu/o3/lsq_unit_impl.hh:
cpu/o3/mem_dep_unit.hh:
cpu/o3/mem_dep_unit_impl.hh:
cpu/o3/ras.cc:
cpu/o3/ras.hh:
cpu/o3/rename.hh:
cpu/o3/rename_impl.hh:
cpu/o3/rob.hh:
cpu/o3/rob_impl.hh:
cpu/o3/sat_counter.cc:
cpu/o3/sat_counter.hh:
cpu/o3/thread_state.hh:
    Handle switching out and taking over.  Needs to be able to reset all state.
cpu/o3/alpha_cpu_impl.hh:
    Handle taking over from another XC.

--HG--
extra : convert_revision : b936e826f0f8a18319bfa940ff35097b4192b449

36 files changed, 786 insertions, 152 deletions

diff --git a/cpu/o3/2bit_local_pred.cc b/cpu/o3/2bit_local_pred.cc
index 458fbd663..eab98531d 100644
--- a/cpu/o3/2bit_local_pred.cc
+++ b/cpu/o3/2bit_local_pred.cc
@@ -67,6 +67,14 @@ DefaultBP::DefaultBP(unsigned _localPredictorSize,
             instShiftAmt);
 }
 
+void
+DefaultBP::reset()
+{
+    for (int i = 0; i < localPredictorSets; ++i) {
+        localCtrs[i].reset();
+    }
+}
+
 bool
 DefaultBP::lookup(Addr &branch_addr)
 {
diff --git a/cpu/o3/2bit_local_pred.hh b/cpu/o3/2bit_local_pred.hh
index 38d3f4842..0dfe53819 100644
--- a/cpu/o3/2bit_local_pred.hh
+++ b/cpu/o3/2bit_local_pred.hh
@@ -62,6 +62,8 @@ class DefaultBP
      */
     void update(Addr &branch_addr, bool taken);
 
+    void reset();
+
   private:
 
     /**
diff --git a/cpu/o3/alpha_cpu_impl.hh b/cpu/o3/alpha_cpu_impl.hh
index 86f7d9f28..7a2d5d2b9 100644
--- a/cpu/o3/alpha_cpu_impl.hh
+++ b/cpu/o3/alpha_cpu_impl.hh
@@ -151,6 +151,26 @@ template <class Impl>
 void
 AlphaFullCPU<Impl>::AlphaXC::takeOverFrom(ExecContext *old_context)
 {
+    // some things should already be set up
+    assert(getMemPtr() == old_context->getMemPtr());
+#if FULL_SYSTEM
+    assert(getSystemPtr() == old_context->getSystemPtr());
+#else
+    assert(getProcessPtr() == old_context->getProcessPtr());
+#endif
+
+    // copy over functional state
+    setStatus(old_context->status());
+    copyArchRegs(old_context);
+    setCpuId(old_context->readCpuId());
+#if !FULL_SYSTEM
+    thread->funcExeInst = old_context->readFuncExeInst();
+#endif
+
+    old_context->setStatus(ExecContext::Unallocated);
+
+    thread->inSyscall = false;
+    thread->trapPending = false;
 }
 
 template <class Impl>
diff --git a/cpu/o3/bpred_unit.hh b/cpu/o3/bpred_unit.hh
index 67c300989..ee7ffc183 100644
--- a/cpu/o3/bpred_unit.hh
+++ b/cpu/o3/bpred_unit.hh
@@ -67,6 +67,10 @@ class TwobitBPredUnit
      */
     void regStats();
 
+    void switchOut();
+
+    void takeOverFrom();
+
     /**
      * Predicts whether or not the instruction is a taken branch, and the
      * target of the branch if it is taken.
diff --git a/cpu/o3/bpred_unit_impl.hh b/cpu/o3/bpred_unit_impl.hh
index f79b67b6c..872c0c62e 100644
--- a/cpu/o3/bpred_unit_impl.hh
+++ b/cpu/o3/bpred_unit_impl.hh
@@ -95,6 +95,26 @@ TwobitBPredUnit<Impl>::regStats()
 }
 
 template <class Impl>
+void
+TwobitBPredUnit<Impl>::switchOut()
+{
+    for (int i = 0; i < Impl::MaxThreads; ++i) {
+        predHist[i].clear();
+    }
+}
+
+template <class Impl>
+void
+TwobitBPredUnit<Impl>::takeOverFrom()
+{
+    for (int i = 0; i < Impl::MaxThreads; ++i)
+        RAS[i].reset();
+
+    BP.reset();
+    BTB.reset();
+}
+
+template <class Impl>
 bool
 TwobitBPredUnit<Impl>::predict(DynInstPtr &inst, Addr &PC, unsigned tid)
 {
@@ -297,5 +317,6 @@ TwobitBPredUnit<Impl>::squash(const InstSeqNum &squashed_sn,
         BP.update(pred_hist.front().PC, actually_taken);
 
         BTB.update(pred_hist.front().PC, corr_target, tid);
+        pred_hist.pop_front();
     }
 }
diff --git a/cpu/o3/btb.cc b/cpu/o3/btb.cc
index e084142d7..e5f69043a 100644
--- a/cpu/o3/btb.cc
+++ b/cpu/o3/btb.cc
@@ -58,6 +58,14 @@ DefaultBTB::DefaultBTB(unsigned _numEntries,
     tagShiftAmt = instShiftAmt + floorLog2(numEntries);
 }
 
+void
+DefaultBTB::reset()
+{
+    for (int i = 0; i < numEntries; ++i) {
+        btb[i].valid = false;
+    }
+}
+
 inline
 unsigned
 DefaultBTB::getIndex(const Addr &inst_PC)
diff --git a/cpu/o3/btb.hh b/cpu/o3/btb.hh
index aaa9945f7..b9ff42573 100644
--- a/cpu/o3/btb.hh
+++ b/cpu/o3/btb.hh
@@ -65,6 +65,8 @@ class DefaultBTB
     DefaultBTB(unsigned numEntries, unsigned tagBits,
                unsigned instShiftAmt);
 
+    void reset();
+
     /** Looks up an address in the BTB. Must call valid() first on the address.
      *  @param inst_PC The address of the branch to look up.
      *  @param tid The thread id.
diff --git a/cpu/o3/commit.hh b/cpu/o3/commit.hh
index f374b8fb7..028bd5295 100644
--- a/cpu/o3/commit.hh
+++ b/cpu/o3/commit.hh
@@ -175,6 +175,10 @@ class DefaultCommit
     /** Initializes stage by sending back the number of free entries. */
     void initStage();
 
+    void switchOut();
+
+    void takeOverFrom();
+
     /** Ticks the commit stage, which tries to commit instructions. */
     void tick();
 
@@ -351,6 +355,8 @@ class DefaultCommit
     /** Number of Active Threads */
     unsigned numThreads;
 
+    bool switchedOut;
+
     Tick trapLatency;
 
     Tick fetchTrapLatency;
diff --git a/cpu/o3/commit_impl.hh b/cpu/o3/commit_impl.hh
index 157e688c7..7834460e2 100644
--- a/cpu/o3/commit_impl.hh
+++ b/cpu/o3/commit_impl.hh
@@ -54,6 +54,7 @@ template <class Impl>
 void
 DefaultCommit<Impl>::TrapEvent::process()
 {
+    // This will get reset if it was switched out.
     commit->trapSquash[tid] = true;
 }
 
@@ -75,7 +76,8 @@ DefaultCommit<Impl>::DefaultCommit(Params *params)
       renameWidth(params->renameWidth),
       iewWidth(params->executeWidth),
       commitWidth(params->commitWidth),
-      numThreads(params->numberOfThreads)
+      numThreads(params->numberOfThreads),
+      switchedOut(false)
 {
     _status = Active;
     _nextStatus = Inactive;
@@ -254,6 +256,9 @@ DefaultCommit<Impl>::setCPU(FullCPU *cpu_ptr)
     // Commit must broadcast the number of free entries it has at the start of
     // the simulation, so it starts as active.
     cpu->activateStage(FullCPU::CommitIdx);
+
+    trapLatency = cpu->cycles(6);
+    fetchTrapLatency = cpu->cycles(12);
 }
 
 template <class Impl>
@@ -362,6 +367,29 @@ DefaultCommit<Impl>::initStage()
 
 template <class Impl>
 void
+DefaultCommit<Impl>::switchOut()
+{
+    rob->switchOut();
+}
+
+template <class Impl>
+void
+DefaultCommit<Impl>::takeOverFrom()
+{
+    _status = Active;
+    _nextStatus = Inactive;
+    for (int i=0; i < numThreads; i++) {
+        commitStatus[i] = Idle;
+        changedROBNumEntries[i] = false;
+        trapSquash[i] = false;
+        xcSquash[i] = false;
+    }
+    squashCounter = 0;
+    rob->takeOverFrom();
+}
+
+template <class Impl>
+void
 DefaultCommit<Impl>::updateStatus()
 {
     if (commitStatus[0] == TrapPending ||
@@ -719,8 +747,9 @@ DefaultCommit<Impl>::commit()
     while (threads != (*activeThreads).end()) {
         unsigned tid = *threads++;
 
-        if (fromFetch->fetchFault) {
+        if (fromFetch->fetchFault && commitStatus[0] != TrapPending) {
             // Record the fault.  Wait until it's empty in the ROB.  Then handle the trap.
+            // Ignore it if there's already a trap pending as fetch will be redirected.
             fetchFault = fromFetch->fetchFault;
             fetchFaultSN = fromFetch->fetchFaultSN;
             fetchFaultTick = curTick + fetchTrapLatency;
@@ -975,6 +1004,7 @@ DefaultCommit<Impl>::commitInsts()
                 }
 
                 PC[tid] = nextPC[tid];
+                nextPC[tid] = nextPC[tid] + sizeof(TheISA::MachInst);
 #if FULL_SYSTEM
                 int count = 0;
                 Addr oldpc;
@@ -1002,6 +1032,10 @@ DefaultCommit<Impl>::commitInsts()
 
     DPRINTF(CommitRate, "%i\n", num_committed);
     numCommittedDist.sample(num_committed);
+
+    if (num_committed == commitWidth) {
+        commit_eligible[0]++;
+    }
 }
 
 template <class Impl>
diff --git a/cpu/o3/cpu.cc b/cpu/o3/cpu.cc
index ac8c4236e..fc8372026 100644
--- a/cpu/o3/cpu.cc
+++ b/cpu/o3/cpu.cc
@@ -124,6 +124,7 @@ FullO3CPU<Impl>::FullO3CPU(Params *params)
       mem(params->mem),
 #else
 //      pTable(params->pTable),
+      mem(params->workload[0]->getMemory()),
 #endif // FULL_SYSTEM
 
       icacheInterface(params->icacheInterface),
@@ -176,9 +177,9 @@ FullO3CPU<Impl>::FullO3CPU(Params *params)
     numThreads = number_of_threads;
 
 #if !FULL_SYSTEM
-    int activeThreads = params->workload.size();
+    int active_threads = params->workload.size();
 #else
-    int activeThreads = 1;
+    int active_threads = 1;
 #endif
 
     assert(params->numPhysIntRegs   >= numThreads * TheISA::NumIntRegs);
@@ -192,7 +193,7 @@ FullO3CPU<Impl>::FullO3CPU(Params *params)
     PhysRegIndex freg_idx = params->numPhysIntRegs; //Index to 1 after int regs
 
     for (int tid=0; tid < numThreads; tid++) {
-        bool bindRegs = (tid <= activeThreads - 1);
+        bool bindRegs = (tid <= active_threads - 1);
 
         commitRenameMap[tid].init(TheISA::NumIntRegs,
                                   params->numPhysIntRegs,
@@ -357,7 +358,7 @@ FullO3CPU<Impl>::tick()
     }
 
     if (activityCount && !tickEvent.scheduled()) {
-        tickEvent.schedule(curTick + 1);
+        tickEvent.schedule(curTick + cycles(1));
     }
 
 #if !FULL_SYSTEM
@@ -370,8 +371,8 @@ template <class Impl>
 void
 FullO3CPU<Impl>::init()
 {
-    if (deferRegistration) {
-        return;
+    if (!deferRegistration) {
+        registerExecContexts();
     }
 
     // Set inSyscall so that the CPU doesn't squash when initially
@@ -379,7 +380,6 @@ FullO3CPU<Impl>::init()
     for (int i = 0; i < number_of_threads; ++i)
         thread[i]->inSyscall = true;
 
-    registerExecContexts();
 
     // Need to do a copy of the xc->regs into the CPU's regfile so
     // that it can start properly.
@@ -388,7 +388,7 @@ FullO3CPU<Impl>::init()
         // Need to do a copy of the xc->regs into the CPU's regfile so
         // that it can start properly.
 #if FULL_SYSTEM
-        ExecContext *src_xc = system->execContexts[tid];
+        ExecContext *src_xc = execContexts[tid];
 #else
         ExecContext *src_xc = thread[tid]->getXCProxy();
 #endif
@@ -584,7 +584,7 @@ FullO3CPU<Impl>::activateContext(int tid, int delay)
         activeThreads.push_back(tid);
     }
 
-    assert(_status == Idle);
+    assert(_status == Idle || _status == SwitchedOut);
 
     scheduleTickEvent(delay);
 
@@ -658,21 +658,64 @@ FullO3CPU<Impl>::haltContext(int tid)
 
 template <class Impl>
 void
-FullO3CPU<Impl>::switchOut()
+FullO3CPU<Impl>::switchOut(Sampler *sampler)
 {
-    panic("FullO3CPU does not have a switch out function.\n");
+//    panic("FullO3CPU does not have a switch out function.\n");
+    fetch.switchOut();
+    decode.switchOut();
+    rename.switchOut();
+    iew.switchOut();
+    commit.switchOut();
+    if (tickEvent.scheduled())
+        tickEvent.squash();
+    sampler->signalSwitched();
+    _status = SwitchedOut;
 }
 
 template <class Impl>
 void
 FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU)
 {
+    for (int i = 0; i < 6; ++i) {
+        timeBuffer.advance();
+        fetchQueue.advance();
+        decodeQueue.advance();
+        renameQueue.advance();
+        iewQueue.advance();
+        activityBuffer.advance();
+    }
+
+    activityCount = 0;
+    bzero(&stageActive, sizeof(stageActive));
+
     BaseCPU::takeOverFrom(oldCPU);
 
+    fetch.takeOverFrom();
+    decode.takeOverFrom();
+    rename.takeOverFrom();
+    iew.takeOverFrom();
+    commit.takeOverFrom();
+
     assert(!tickEvent.scheduled());
 
+    // @todo: Figure out how to properly select the tid to put onto the active threads list.
+    int tid = 0;
+
+    list<unsigned>::iterator isActive = find(
+        activeThreads.begin(), activeThreads.end(), tid);
+
+    if (isActive == activeThreads.end()) {
+        //May Need to Re-code this if the delay variable is the
+        //delay needed for thread to activate
+        DPRINTF(FullCPU, "Adding Thread %i to active threads list\n",
+                tid);
+
+        activeThreads.push_back(tid);
+    }
+
     // Set all status's to active, schedule the
     // CPU's tick event.
+    // @todo: Fix up statuses so this is handled properly
     for (int i = 0; i < execContexts.size(); ++i) {
         ExecContext *xc = execContexts[i];
         if (xc->status() == ExecContext::Active && _status != Running) {
@@ -680,6 +723,8 @@ FullO3CPU<Impl>::takeOverFrom(BaseCPU *oldCPU)
             tickEvent.schedule(curTick);
         }
     }
+    if (!tickEvent.scheduled())
+        tickEvent.schedule(curTick);
 }
 
 template <class Impl>
@@ -758,7 +803,8 @@ template <class Impl>
 float
 FullO3CPU<Impl>::readArchFloatRegSingle(int reg_idx, unsigned tid)
 {
-    PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
+    int idx = reg_idx + TheISA::FP_Base_DepTag;
+    PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
 
     return regFile.readFloatRegSingle(phys_reg);
 }
@@ -767,7 +813,8 @@ template <class Impl>
 double
 FullO3CPU<Impl>::readArchFloatRegDouble(int reg_idx, unsigned tid)
 {
-    PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
+    int idx = reg_idx + TheISA::FP_Base_DepTag;
+    PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
 
     return regFile.readFloatRegDouble(phys_reg);
 }
@@ -776,7 +823,8 @@ template <class Impl>
 uint64_t
 FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, unsigned tid)
 {
-    PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
+    int idx = reg_idx + TheISA::FP_Base_DepTag;
+    PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
 
     return regFile.readFloatRegInt(phys_reg);
 }
diff --git a/cpu/o3/cpu.hh b/cpu/o3/cpu.hh
index 91eaf9d6f..621ddf541 100644
--- a/cpu/o3/cpu.hh
+++ b/cpu/o3/cpu.hh
@@ -82,7 +82,8 @@ class FullO3CPU : public BaseFullCPU
         Running,
         Idle,
         Halted,
-        Blocked
+        Blocked,
+        SwitchedOut
     };
 
     /** Overall CPU status. */
@@ -112,9 +113,9 @@ class FullO3CPU : public BaseFullCPU
     void scheduleTickEvent(int delay)
     {
         if (tickEvent.squashed())
-            tickEvent.reschedule(curTick + delay);
+            tickEvent.reschedule(curTick + cycles(delay));
         else if (!tickEvent.scheduled())
-            tickEvent.schedule(curTick + delay);
+            tickEvent.schedule(curTick + cycles(delay));
     }
 
     /** Unschedule tick event, regardless of its current state. */
@@ -196,7 +197,7 @@ class FullO3CPU : public BaseFullCPU
     /** Switches out this CPU.
      *  @todo: Implement this.
      */
-    void switchOut();
+    void switchOut(Sampler *sampler);
 
     /** Takes over from another CPU.
      *  @todo: Implement this.
diff --git a/cpu/o3/decode.hh b/cpu/o3/decode.hh
index 279ff556e..3f3f68247 100644
--- a/cpu/o3/decode.hh
+++ b/cpu/o3/decode.hh
@@ -107,6 +107,9 @@ class DefaultDecode
     /** Sets pointer to list of active threads. */
     void setActiveThreads(std::list<unsigned> *at_ptr);
 
+    void switchOut();
+
+    void takeOverFrom();
     /** Ticks decode, processing all input signals and decoding as many
      * instructions as possible.
      */
@@ -272,6 +275,8 @@ class DefaultDecode
     Stats::Scalar<> decodeUnblockCycles;
     /** Stat for total number of squashing cycles. */
     Stats::Scalar<> decodeSquashCycles;
+    /** Stat for number of times a branch is resolved at decode. */
+    Stats::Scalar<> decodeBranchResolved;
     /** Stat for number of times a branch mispredict is detected. */
     Stats::Scalar<> decodeBranchMispred;
     /** Stat for number of times decode detected a non-control instruction
diff --git a/cpu/o3/decode_impl.hh b/cpu/o3/decode_impl.hh
index f1aea27b4..caa97067b 100644
--- a/cpu/o3/decode_impl.hh
+++ b/cpu/o3/decode_impl.hh
@@ -66,40 +66,44 @@ void
 DefaultDecode<Impl>::regStats()
 {
     decodeIdleCycles
-        .name(name() + ".decodeIdleCycles")
+        .name(name() + ".DECODE:IdleCycles")
         .desc("Number of cycles decode is idle")
         .prereq(decodeIdleCycles);
     decodeBlockedCycles
-        .name(name() + ".decodeBlockedCycles")
+        .name(name() + ".DECODE:BlockedCycles")
         .desc("Number of cycles decode is blocked")
         .prereq(decodeBlockedCycles);
     decodeRunCycles
-        .name(name() + ".decodeRunCycles")
+        .name(name() + ".DECODE:RunCycles")
         .desc("Number of cycles decode is running")
         .prereq(decodeRunCycles);
     decodeUnblockCycles
-        .name(name() + ".decodeUnblockCycles")
+        .name(name() + ".DECODE:UnblockCycles")
         .desc("Number of cycles decode is unblocking")
         .prereq(decodeUnblockCycles);
     decodeSquashCycles
-        .name(name() + ".decodeSquashCycles")
+        .name(name() + ".DECODE:SquashCycles")
         .desc("Number of cycles decode is squashing")
         .prereq(decodeSquashCycles);
+    decodeBranchResolved
+        .name(name() + ".DECODE:BranchResolved")
+        .desc("Number of times decode resolved a branch")
+        .prereq(decodeBranchResolved);
     decodeBranchMispred
-        .name(name() + ".decodeBranchMispred")
+        .name(name() + ".DECODE:BranchMispred")
         .desc("Number of times decode detected a branch misprediction")
         .prereq(decodeBranchMispred);
     decodeControlMispred
-        .name(name() + ".decodeControlMispred")
+        .name(name() + ".DECODE:ControlMispred")
         .desc("Number of times decode detected an instruction incorrectly"
               " predicted as a control")
         .prereq(decodeControlMispred);
     decodeDecodedInsts
-        .name(name() + ".decodeDecodedInsts")
+        .name(name() + ".DECODE:DecodedInsts")
         .desc("Number of instructions handled by decode")
         .prereq(decodeDecodedInsts);
     decodeSquashedInsts
-        .name(name() + ".decodeSquashedInsts")
+        .name(name() + ".DECODE:SquashedInsts")
         .desc("Number of squashed instructions handled by decode")
         .prereq(decodeSquashedInsts);
 }
@@ -158,6 +162,33 @@ DefaultDecode<Impl>::setActiveThreads(list<unsigned> *at_ptr)
     activeThreads = at_ptr;
 }
 
+template <class Impl>
+void
+DefaultDecode<Impl>::switchOut()
+{
+}
+
+template <class Impl>
+void
+DefaultDecode<Impl>::takeOverFrom()
+{
+    _status = Inactive;
+
+    for (int i = 0; i < numThreads; ++i) {
+        decodeStatus[i] = Idle;
+
+        stalls[i].rename = false;
+        stalls[i].iew = false;
+        stalls[i].commit = false;
+        while (!insts[i].empty())
+            insts[i].pop();
+        while (!skidBuffer[i].empty())
+            skidBuffer[i].pop();
+        branchCount[i] = 0;
+    }
+    wroteToTimeBuffer = false;
+}
+
 template<class Impl>
 bool
 DefaultDecode<Impl>::checkStall(unsigned tid) const
@@ -680,6 +711,7 @@ DefaultDecode<Impl>::decodeInsts(unsigned tid)
 
         // Go ahead and compute any PC-relative branches.
         if (inst->isDirectCtrl() && inst->isUncondCtrl()) {
+            ++decodeBranchResolved;
             inst->setNextPC(inst->branchTarget());
 
             if (inst->mispredicted()) {
diff --git a/cpu/o3/fetch.hh b/cpu/o3/fetch.hh
index f0b15cb86..6074831c6 100644
--- a/cpu/o3/fetch.hh
+++ b/cpu/o3/fetch.hh
@@ -35,6 +35,8 @@
 #include "mem/mem_interface.hh"
 #include "sim/eventq.hh"
 
+class Sampler;
+
 /**
  * DefaultFetch class handles both single threaded and SMT fetch. Its width is
  * specified by the parameters; each cycle it tries to fetch that many
@@ -81,6 +83,7 @@ class DefaultFetch
         Fetching,
         TrapPending,
         QuiescePending,
+        SwitchOut,
         IcacheMissStall,
         IcacheMissComplete
     };
@@ -160,6 +163,12 @@ class DefaultFetch
     /** Processes cache completion event. */
     void processCacheCompletion(MemReqPtr &req);
 
+    void switchOut();
+
+    void takeOverFrom();
+
+    bool isSwitchedOut() { return switchedOut; }
+
     void wakeFromQuiesce();
 
   private:
@@ -360,6 +369,8 @@ class DefaultFetch
 
     bool interruptPending;
 
+    bool switchedOut;
+
 #if !FULL_SYSTEM
     /** Page table pointer. */
 //    PageTable *pTable;
@@ -382,6 +393,8 @@ class DefaultFetch
      */
     Stats::Scalar<> fetchIdleCycles;
     Stats::Scalar<> fetchBlockedCycles;
+
+    Stats::Scalar<> fetchMiscStallCycles;
     /** Stat for total number of fetched cache lines. */
     Stats::Scalar<> fetchedCacheLines;
 
diff --git a/cpu/o3/fetch_impl.hh b/cpu/o3/fetch_impl.hh
index 563a767df..92f923c65 100644
--- a/cpu/o3/fetch_impl.hh
+++ b/cpu/o3/fetch_impl.hh
@@ -169,53 +169,59 @@ void
 DefaultFetch<Impl>::regStats()
 {
     icacheStallCycles
-        .name(name() + ".icacheStallCycles")
+        .name(name() + ".FETCH:icacheStallCycles")
         .desc("Number of cycles fetch is stalled on an Icache miss")
         .prereq(icacheStallCycles);
 
     fetchedInsts
-        .name(name() + ".fetchedInsts")
+        .name(name() + ".FETCH:Insts")
         .desc("Number of instructions fetch has processed")
         .prereq(fetchedInsts);
 
     fetchedBranches
-        .name(name() + ".fetchedBranches")
+        .name(name() + ".FETCH:Branches")
         .desc("Number of branches that fetch encountered")
         .prereq(fetchedBranches);
 
     predictedBranches
-        .name(name() + ".predictedBranches")
+        .name(name() + ".FETCH:predictedBranches")
         .desc("Number of branches that fetch has predicted taken")
         .prereq(predictedBranches);
 
     fetchCycles
-        .name(name() + ".fetchCycles")
+        .name(name() + ".FETCH:Cycles")
         .desc("Number of cycles fetch has run and was not squashing or"
               " blocked")
         .prereq(fetchCycles);
 
     fetchSquashCycles
-        .name(name() + ".fetchSquashCycles")
+        .name(name() + ".FETCH:SquashCycles")
         .desc("Number of cycles fetch has spent squashing")
         .prereq(fetchSquashCycles);
 
     fetchIdleCycles
-        .name(name() + ".fetchIdleCycles")
+        .name(name() + ".FETCH:IdleCycles")
         .desc("Number of cycles fetch was idle")
         .prereq(fetchIdleCycles);
 
     fetchBlockedCycles
-        .name(name() + ".fetchBlockedCycles")
+        .name(name() + ".FETCH:BlockedCycles")
         .desc("Number of cycles fetch has spent blocked")
         .prereq(fetchBlockedCycles);
 
     fetchedCacheLines
-        .name(name() + ".fetchedCacheLines")
+        .name(name() + ".FETCH:CacheLines")
         .desc("Number of cache lines fetched")
         .prereq(fetchedCacheLines);
 
+    fetchMiscStallCycles
+        .name(name() + ".FETCH:MiscStallCycles")
+        .desc("Number of cycles fetch has spent waiting on interrupts, or "
+              "bad addresses, or out of MSHRs")
+        .prereq(fetchMiscStallCycles);
+
     fetchIcacheSquashes
-        .name(name() + ".fetchIcacheSquashes")
+        .name(name() + ".FETCH:IcacheSquashes")
         .desc("Number of outstanding Icache misses that were squashed")
         .prereq(fetchIcacheSquashes);
 
@@ -223,24 +229,24 @@ DefaultFetch<Impl>::regStats()
         .init(/* base value */ 0,
               /* last value */ fetchWidth,
               /* bucket size */ 1)
-        .name(name() + ".rateDist")
+        .name(name() + ".FETCH:rateDist")
         .desc("Number of instructions fetched each cycle (Total)")
         .flags(Stats::pdf);
 
     idleRate
-        .name(name() + ".idleRate")
+        .name(name() + ".FETCH:idleRate")
         .desc("Percent of cycles fetch was idle")
         .prereq(idleRate);
     idleRate = fetchIdleCycles * 100 / cpu->numCycles;
 
     branchRate
-        .name(name() + ".branchRate")
+        .name(name() + ".FETCH:branchRate")
         .desc("Number of branch fetches per cycle")
         .flags(Stats::total);
     branchRate = predictedBranches / cpu->numCycles;
 
     fetchRate
-        .name(name() + ".rate")
+        .name(name() + ".FETCH:rate")
         .desc("Number of inst fetches per cycle")
         .flags(Stats::total);
     fetchRate = fetchedInsts / cpu->numCycles;
@@ -332,7 +338,8 @@ DefaultFetch<Impl>::processCacheCompletion(MemReqPtr &req)
     // Can keep track of how many cache accesses go unused due to
     // misspeculation here.
     if (fetchStatus[tid] != IcacheMissStall ||
-        req != memReq[tid]) {
+        req != memReq[tid] ||
+        isSwitchedOut()) {
         ++fetchIcacheSquashes;
         return;
     }
@@ -362,6 +369,35 @@ DefaultFetch<Impl>::processCacheCompletion(MemReqPtr &req)
 
 template <class Impl>
 void
+DefaultFetch<Impl>::switchOut()
+{
+    switchedOut = true;
+    branchPred.switchOut();
+}
+
+template <class Impl>
+void
+DefaultFetch<Impl>::takeOverFrom()
+{
+    // Reset all state
+    for (int i = 0; i < Impl::MaxThreads; ++i) {
+        stalls[i].decode = 0;
+        stalls[i].rename = 0;
+        stalls[i].iew = 0;
+        stalls[i].commit = 0;
+        PC[i] = cpu->readPC(i);
+        nextPC[i] = cpu->readNextPC(i);
+        fetchStatus[i] = Running;
+    }
+    numInst = 0;
+    wroteToTimeBuffer = false;
+    _status = Inactive;
+    switchedOut = false;
+    branchPred.takeOverFrom();
+}
+
+template <class Impl>
+void
 DefaultFetch<Impl>::wakeFromQuiesce()
 {
     DPRINTF(Fetch, "Waking up from quiesce\n");
@@ -902,8 +938,10 @@ DefaultFetch<Impl>::fetch(bool &status_change)
                 tid, fetch_PC);
 
         bool fetch_success = fetchCacheLine(fetch_PC, fault, tid);
-        if (!fetch_success)
+        if (!fetch_success) {
+            ++fetchMiscStallCycles;
             return;
+        }
     } else {
         if (fetchStatus[tid] == Idle) {
             ++fetchIdleCycles;
diff --git a/cpu/o3/fu_pool.cc b/cpu/o3/fu_pool.cc
index 9b6ac15d9..cb7a15061 100644
--- a/cpu/o3/fu_pool.cc
+++ b/cpu/o3/fu_pool.cc
@@ -242,6 +242,20 @@ FUPool::dump()
     }
 }
 
+void
+FUPool::switchOut()
+{
+}
+
+void
+FUPool::takeOverFrom()
+{
+    for (int i = 0; i < numFU; i++) {
+        unitBusy[i] = false;
+    }
+    unitsToBeFreed.clear();
+}
+
 //
 
 ////////////////////////////////////////////////////////////////////////////
diff --git a/cpu/o3/fu_pool.hh b/cpu/o3/fu_pool.hh
index d7b7acadb..7df5ad5f3 100644
--- a/cpu/o3/fu_pool.hh
+++ b/cpu/o3/fu_pool.hh
@@ -154,6 +154,9 @@ class FUPool : public SimObject
     unsigned getIssueLatency(OpClass capability) {
         return maxIssueLatencies[capability];
     }
+
+    void switchOut();
+    void takeOverFrom();
 };
 
 #endif // __CPU_O3_FU_POOL_HH__
diff --git a/cpu/o3/iew.hh b/cpu/o3/iew.hh
index 58cd68b21..ae0ba6a21 100644
--- a/cpu/o3/iew.hh
+++ b/cpu/o3/iew.hh
@@ -157,6 +157,12 @@ class DefaultIEW
     /** Sets pointer to the scoreboard. */
     void setScoreboard(Scoreboard *sb_ptr);
 
+    void switchOut();
+
+    void takeOverFrom();
+
+    bool isSwitchedOut() { return switchedOut; }
+
     /** Sets page table pointer within LSQ. */
 //    void setPageTable(PageTable *pt_ptr);
 
@@ -420,6 +426,8 @@ class DefaultIEW
     /** Maximum size of the skid buffer. */
     unsigned skidBufferMax;
 
+    bool switchedOut;
+
     /** Stat for total number of idle cycles. */
     Stats::Scalar<> iewIdleCycles;
     /** Stat for total number of squashing cycles. */
diff --git a/cpu/o3/iew_impl.hh b/cpu/o3/iew_impl.hh
index 2ae2e1361..42d83ee72 100644
--- a/cpu/o3/iew_impl.hh
+++ b/cpu/o3/iew_impl.hh
@@ -55,13 +55,13 @@ DefaultIEW<Impl>::LdWritebackEvent::process()
 
     //iewStage->ldstQueue.removeMSHR(inst->threadNumber,inst->seqNum);
 
-    iewStage->wakeCPU();
-
-    if (inst->isSquashed()) {
+    if (inst->isSquashed() || iewStage->isSwitchedOut()) {
         inst = NULL;
         return;
     }
 
+    iewStage->wakeCPU();
+
     if (!inst->isExecuted()) {
         inst->setExecuted();
 
@@ -101,7 +101,8 @@ DefaultIEW<Impl>::DefaultIEW(Params *params)
       issueReadWidth(params->issueWidth),
       issueWidth(params->issueWidth),
       executeWidth(params->executeWidth),
-      numThreads(params->numberOfThreads)
+      numThreads(params->numberOfThreads),
+      switchedOut(false)
 {
     DPRINTF(IEW, "executeIntWidth: %i.\n", params->executeIntWidth);
     _status = Active;
@@ -436,6 +437,53 @@ DefaultIEW<Impl>::setPageTable(PageTable *pt_ptr)
 }
 #endif
 
+template <class Impl>
+void
+DefaultIEW<Impl>::switchOut()
+{
+    switchedOut = true;
+    instQueue.switchOut();
+    ldstQueue.switchOut();
+    fuPool->switchOut();
+
+    for (int i = 0; i < numThreads; i++) {
+        while (!insts[i].empty())
+            insts[i].pop();
+        while (!skidBuffer[i].empty())
+            skidBuffer[i].pop();
+    }
+}
+
+template <class Impl>
+void
+DefaultIEW<Impl>::takeOverFrom()
+{
+    _status = Active;
+    exeStatus = Running;
+    wbStatus = Idle;
+    switchedOut = false;
+
+    instQueue.takeOverFrom();
+    ldstQueue.takeOverFrom();
+    fuPool->takeOverFrom();
+
+    initStage();
+    cpu->activityThisCycle();
+
+    for (int i=0; i < numThreads; i++) {
+        dispatchStatus[i] = Running;
+        stalls[i].commit = false;
+        fetchRedirect[i] = false;
+    }
+
+    updateLSQNextCycle = false;
+
+    // @todo: Fix hardcoded number
+    for (int i = 0; i < 6; ++i) {
+        issueToExecQueue.advance();
+    }
+}
+
 template<class Impl>
 void
 DefaultIEW<Impl>::squash(unsigned tid)
diff --git a/cpu/o3/inst_queue.hh b/cpu/o3/inst_queue.hh
index 06d9937f2..982294b4f 100644
--- a/cpu/o3/inst_queue.hh
+++ b/cpu/o3/inst_queue.hh
@@ -112,6 +112,10 @@ class InstructionQueue
     /** Registers statistics. */
     void regStats();
 
+    void resetState();
+
+    void resetDependencyGraph();
+
     /** Sets CPU pointer. */
     void setCPU(FullCPU *_cpu) { cpu = _cpu; }
 
@@ -127,6 +131,12 @@ class InstructionQueue
     /** Sets the global time buffer. */
     void setTimeBuffer(TimeBuffer<TimeStruct> *tb_ptr);
 
+    void switchOut();
+
+    void takeOverFrom();
+
+    bool isSwitchedOut() { return switchedOut; }
+
     /** Number of entries needed for given amount of threads. */
     int entryAmount(int num_threads);
 
@@ -385,6 +395,8 @@ class InstructionQueue
      */
     unsigned commitToIEWDelay;
 
+    bool switchedOut;
+
     //////////////////////////////////
     // Variables needed for squashing
     //////////////////////////////////
@@ -507,7 +519,7 @@ class InstructionQueue
     Stats::Scalar<> iqSquashedNonSpecRemoved;
 
     Stats::VectorDistribution<> queue_res_dist;
-    Stats::Vector<> n_issued_dist;
+    Stats::Distribution<> n_issued_dist;
     Stats::VectorDistribution<> issue_delay_dist;
 
     Stats::Vector<> stat_fu_busy;
diff --git a/cpu/o3/inst_queue_impl.hh b/cpu/o3/inst_queue_impl.hh
index 804bc2472..0d9cc09f3 100644
--- a/cpu/o3/inst_queue_impl.hh
+++ b/cpu/o3/inst_queue_impl.hh
@@ -82,15 +82,9 @@ InstructionQueue<Impl>::InstructionQueue(Params *params)
 {
     assert(fuPool);
 
-    numThreads = params->numberOfThreads;
+    switchedOut = false;
 
-    //Initialize thread IQ counts
-    for (int i = 0; i <numThreads; i++) {
-        count[i] = 0;
-    }
-
-    // Initialize the number of free IQ entries.
-    freeEntries = numEntries;
+    numThreads = params->numberOfThreads;
 
     // Set the number of physical registers as the number of int + float
     numPhysRegs = numPhysIntRegs + numPhysFloatRegs;
@@ -101,37 +95,24 @@ InstructionQueue<Impl>::InstructionQueue(Params *params)
     //dependency graph.
     dependGraph = new DependencyEntry[numPhysRegs];
 
-    // Resize the register scoreboard.
-    regScoreboard.resize(numPhysRegs);
-
-    //Initialize Mem Dependence Units
-    for (int i = 0; i < numThreads; i++) {
-        memDepUnit[i].init(params,i);
-        memDepUnit[i].setIQ(this);
-    }
-
     // Initialize all the head pointers to point to NULL, and all the
     // entries as unready.
-    // Note that in actuality, the registers corresponding to the logical
-    // registers start off as ready.  However this doesn't matter for the
-    // IQ as the instruction should have been correctly told if those
-    // registers are ready in rename.  Thus it can all be initialized as
-    // unready.
     for (int i = 0; i < numPhysRegs; ++i) {
         dependGraph[i].next = NULL;
         dependGraph[i].inst = NULL;
-        regScoreboard[i] = false;
     }
 
-    for (int i = 0; i < numThreads; ++i) {
-        squashedSeqNum[i] = 0;
-    }
+    // Resize the register scoreboard.
+    regScoreboard.resize(numPhysRegs);
 
-    for (int i = 0; i < Num_OpClasses; ++i) {
-        queueOnList[i] = false;
-        readyIt[i] = listOrder.end();
+    //Initialize Mem Dependence Units
+    for (int i = 0; i < numThreads; i++) {
+        memDepUnit[i].init(params,i);
+        memDepUnit[i].setIQ(this);
     }
 
+    resetState();
+
     string policy = params->smtIQPolicy;
 
     //Convert string to lowercase
@@ -184,30 +165,7 @@ InstructionQueue<Impl>::InstructionQueue(Params *params)
 template <class Impl>
 InstructionQueue<Impl>::~InstructionQueue()
 {
-    // Clear the dependency graph
-    DependencyEntry *curr;
-    DependencyEntry *prev;
-
-    for (int i = 0; i < numPhysRegs; ++i) {
-        curr = dependGraph[i].next;
-
-        while (curr) {
-            DependencyEntry::mem_alloc_counter--;
-
-            prev = curr;
-            curr = prev->next;
-            prev->inst = NULL;
-
-            delete prev;
-        }
-
-        if (dependGraph[i].inst) {
-            dependGraph[i].inst = NULL;
-        }
-
-        dependGraph[i].next = NULL;
-    }
-
+    resetDependencyGraph();
     assert(DependencyEntry::mem_alloc_counter == 0);
 
     delete [] dependGraph;
@@ -307,10 +265,10 @@ InstructionQueue<Impl>::regStats()
         queue_res_dist.subname(i, opClassStrings[i]);
     }
     n_issued_dist
-        .init(totalWidth + 1)
+        .init(0,totalWidth,1)
         .name(name() + ".ISSUE:issued_per_cycle")
         .desc("Number of insts issued each cycle")
-        .flags(total | pdf | dist)
+        .flags(pdf)
         ;
 /*
     dist_unissued
@@ -402,6 +360,71 @@ InstructionQueue<Impl>::regStats()
 
 template <class Impl>
 void
+InstructionQueue<Impl>::resetState()
+{
+    //Initialize thread IQ counts
+    for (int i = 0; i <numThreads; i++) {
+        count[i] = 0;
+        instList[i].clear();
+    }
+
+    // Initialize the number of free IQ entries.
+    freeEntries = numEntries;
+
+    // Note that in actuality, the registers corresponding to the logical
+    // registers start off as ready.  However this doesn't matter for the
+    // IQ as the instruction should have been correctly told if those
+    // registers are ready in rename.  Thus it can all be initialized as
+    // unready.
+    for (int i = 0; i < numPhysRegs; ++i) {
+        regScoreboard[i] = false;
+    }
+
+    for (int i = 0; i < numThreads; ++i) {
+        squashedSeqNum[i] = 0;
+    }
+
+    for (int i = 0; i < Num_OpClasses; ++i) {
+        while (!readyInsts[i].empty())
+            readyInsts[i].pop();
+        queueOnList[i] = false;
+        readyIt[i] = listOrder.end();
+    }
+    nonSpecInsts.clear();
+    listOrder.clear();
+}
+
+template <class Impl>
+void
+InstructionQueue<Impl>::resetDependencyGraph()
+{
+    // Clear the dependency graph
+    DependencyEntry *curr;
+    DependencyEntry *prev;
+
+    for (int i = 0; i < numPhysRegs; ++i) {
+        curr = dependGraph[i].next;
+
+        while (curr) {
+            DependencyEntry::mem_alloc_counter--;
+
+            prev = curr;
+            curr = prev->next;
+            prev->inst = NULL;
+
+            delete prev;
+        }
+
+        if (dependGraph[i].inst) {
+            dependGraph[i].inst = NULL;
+        }
+
+        dependGraph[i].next = NULL;
+    }
+}
+
+template <class Impl>
+void
 InstructionQueue<Impl>::setActiveThreads(list<unsigned> *at_ptr)
 {
     DPRINTF(IQ, "Setting active threads list pointer.\n");
@@ -427,6 +450,25 @@ InstructionQueue<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *tb_ptr)
 }
 
 template <class Impl>
+void
+InstructionQueue<Impl>::switchOut()
+{
+    resetState();
+    resetDependencyGraph();
+    switchedOut = true;
+    for (int i = 0; i < numThreads; ++i) {
+        memDepUnit[i].switchOut();
+    }
+}
+
+template <class Impl>
+void
+InstructionQueue<Impl>::takeOverFrom()
+{
+    switchedOut = false;
+}
+
+template <class Impl>
 int
 InstructionQueue<Impl>::entryAmount(int num_threads)
 {
@@ -685,6 +727,10 @@ InstructionQueue<Impl>::processFUCompletion(DynInstPtr &inst, int fu_idx)
 {
     // The CPU could have been sleeping until this op completed (*extremely*
     // long latency op).  Wake it if it was.  This may be overkill.
+    if (isSwitchedOut()) {
+        return;
+    }
+
     iewStage->wakeCPU();
 
     fuPool->freeUnit(fu_idx);
@@ -816,7 +862,7 @@ InstructionQueue<Impl>::scheduleReadyInsts()
                     FUCompletion *execution = new FUCompletion(issuing_inst,
                                                                idx, this);
 
-                    execution->schedule(curTick + issue_latency - 1);
+                    execution->schedule(curTick + cpu->cycles(issue_latency - 1));
                 } else {
                     i2e_info->insts[exec_queue_slot++] = issuing_inst;
                     i2e_info->size++;
@@ -862,6 +908,8 @@ InstructionQueue<Impl>::scheduleReadyInsts()
         }
     }
 
+    n_issued_dist.sample(total_issued);
+
     if (total_issued) {
         cpu->activityThisCycle();
     } else {
diff --git a/cpu/o3/lsq.hh b/cpu/o3/lsq.hh
index c59b5f13b..d5f893e57 100644
--- a/cpu/o3/lsq.hh
+++ b/cpu/o3/lsq.hh
@@ -71,6 +71,9 @@ class LSQ {
     /** Sets the page table pointer. */
 //    void setPageTable(PageTable *pt_ptr);
 
+    void switchOut();
+    void takeOverFrom();
+
     /** Number of entries needed for the given amount of threads.*/
     int entryAmount(int num_threads);
     void removeEntries(unsigned tid);
@@ -271,15 +274,6 @@ class LSQ {
     /** Max SQ Size - Used to Enforce Sharing Policies. */
     unsigned maxSQEntries;
 
-    /** Global Load Count. */
-    int loads;
-
-    /** Global Store Count */
-    int stores;
-
-    /** Global Store To WB Count */
-    int storesToWB;
-
     /** Number of Threads. */
     unsigned numThreads;
 };
diff --git a/cpu/o3/lsq_impl.hh b/cpu/o3/lsq_impl.hh
index 523517869..c43c19619 100644
--- a/cpu/o3/lsq_impl.hh
+++ b/cpu/o3/lsq_impl.hh
@@ -33,7 +33,6 @@ using namespace std;
 template <class Impl>
 LSQ<Impl>::LSQ(Params *params)
     : LQEntries(params->LQEntries), SQEntries(params->SQEntries),
-      loads(0), stores(0), storesToWB(0),
       numThreads(params->numberOfThreads)
 {
     DPRINTF(LSQ, "Creating LSQ object.\n");
@@ -144,6 +143,24 @@ LSQ<Impl>::setPageTable(PageTable *pt_ptr)
 #endif
 
 template <class Impl>
+void
+LSQ<Impl>::switchOut()
+{
+    for (int tid = 0; tid < numThreads; tid++) {
+        thread[tid].switchOut();
+    }
+}
+
+template <class Impl>
+void
+LSQ<Impl>::takeOverFrom()
+{
+    for (int tid = 0; tid < numThreads; tid++) {
+        thread[tid].takeOverFrom();
+    }
+}
+
+template <class Impl>
 int
 LSQ<Impl>::entryAmount(int num_threads)
 {
diff --git a/cpu/o3/lsq_unit.hh b/cpu/o3/lsq_unit.hh
index ba8b1d2e2..d17efe96a 100644
--- a/cpu/o3/lsq_unit.hh
+++ b/cpu/o3/lsq_unit.hh
@@ -38,6 +38,7 @@
 #include "cpu/inst_seq.hh"
 #include "mem/mem_interface.hh"
 //#include "mem/page_table.hh"
+#include "sim/debug.hh"
 #include "sim/sim_object.hh"
 #include "arch/faults.hh"
 
@@ -110,6 +111,12 @@ class LSQUnit {
     /** Sets the page table pointer. */
 //    void setPageTable(PageTable *pt_ptr);
 
+    void switchOut();
+
+    void takeOverFrom();
+
+    bool isSwitchedOut() { return switchedOut; }
+
     /** Ticks the LSQ unit, which in this case only resets the number of
      * used cache ports.
      * @todo: Move the number of used ports up to the LSQ level so it can
@@ -278,20 +285,20 @@ class LSQUnit {
         /** Whether or not the store is completed. */
         bool completed;
     };
-
+/*
     enum Status {
         Running,
         Idle,
         DcacheMissStall,
         DcacheMissSwitch
     };
-
+*/
   private:
     /** The LSQUnit thread id. */
     unsigned lsqID;
 
     /** The status of the LSQ unit. */
-    Status _status;
+//    Status _status;
 
     /** The store queue. */
     std::vector<SQEntry> storeQueue;
@@ -335,6 +342,8 @@ class LSQUnit {
     /** The number of used cache ports in this cycle. */
     int usedPorts;
 
+    bool switchedOut;
+
     //list<InstSeqNum> mshrSeqNums;
 
      //Stats::Scalar<> dcacheStallCycles;
@@ -373,7 +382,25 @@ class LSQUnit {
     // Will also need how many read/write ports the Dcache has.  Or keep track
     // of that in stage that is one level up, and only call executeLoad/Store
     // the appropriate number of times.
+/*
+    // total number of loads forwaded from LSQ stores
+    Stats::Vector<> lsq_forw_loads;
+
+    // total number of loads ignored due to invalid addresses
+    Stats::Vector<> inv_addr_loads;
+
+    // total number of software prefetches ignored due to invalid addresses
+    Stats::Vector<> inv_addr_swpfs;
+
+    // total non-speculative bogus addresses seen (debug var)
+    Counter sim_invalid_addrs;
+    Stats::Vector<> fu_busy;  //cumulative fu busy
 
+    // ready loads blocked due to memory disambiguation
+    Stats::Vector<> lsq_blocked_loads;
+
+    Stats::Scalar<> lsqInversion;
+*/
   public:
     /** Executes the load at the given index. */
     template <class T>
@@ -590,7 +617,12 @@ LSQUnit<Impl>::read(MemReqPtr &req, T &data, int load_idx)
         }
         DPRINTF(LSQUnit, "Doing timing access for inst PC %#x\n",
                 loadQueue[load_idx]->readPC());
-
+/*
+        Addr debug_addr = ULL(0xfffffc0000be81a8);
+        if (req->vaddr == debug_addr) {
+            debug_break();
+        }
+*/
         assert(!req->completionEvent);
         req->completionEvent =
             new typename IEW::LdWritebackEvent(loadQueue[load_idx], iewStage);
@@ -608,7 +640,7 @@ LSQUnit<Impl>::read(MemReqPtr &req, T &data, int load_idx)
 
             lastDcacheStall = curTick;
 
-            _status = DcacheMissStall;
+//            _status = DcacheMissStall;
 
         } else {
             DPRINTF(Activity, "Activity: ld accessing mem hit [sn:%lli]\n",
@@ -694,7 +726,12 @@ LSQUnit<Impl>::write(MemReqPtr &req, T &data, int store_idx)
     storeQueue[store_idx].req = req;
     storeQueue[store_idx].size = sizeof(T);
     storeQueue[store_idx].data = data;
-
+/*
+    Addr debug_addr = ULL(0xfffffc0000be81a8);
+    if (req->vaddr == debug_addr) {
+        debug_break();
+    }
+*/
     // This function only writes the data to the store queue, so no fault
     // can happen here.
     return NoFault;
diff --git a/cpu/o3/lsq_unit_impl.hh b/cpu/o3/lsq_unit_impl.hh
index d9a118b0e..c5ce34c70 100644
--- a/cpu/o3/lsq_unit_impl.hh
+++ b/cpu/o3/lsq_unit_impl.hh
@@ -50,6 +50,9 @@ LSQUnit<Impl>::StoreCompletionEvent::process()
 
     //lsqPtr->removeMSHR(lsqPtr->storeQueue[storeIdx].inst->seqNum);
 
+    if (lsqPtr->isSwitchedOut())
+        return;
+
     lsqPtr->cpu->wakeCPU();
     if (wbEvent)
         wbEvent->process();
@@ -78,6 +81,8 @@ LSQUnit<Impl>::init(Params *params, unsigned maxLQEntries,
 {
     DPRINTF(LSQUnit, "Creating LSQUnit%i object.\n",id);
 
+    switchedOut = false;
+
     lsqID = id;
 
     LQEntries = maxLQEntries;
@@ -140,6 +145,89 @@ LSQUnit<Impl>::setPageTable(PageTable *pt_ptr)
 
 template<class Impl>
 void
+LSQUnit<Impl>::switchOut()
+{
+    switchedOut = true;
+    for (int i = 0; i < loadQueue.size(); ++i)
+        loadQueue[i] = NULL;
+
+    while (storesToWB > 0 &&
+           storeWBIdx != storeTail &&
+           storeQueue[storeWBIdx].inst &&
+           storeQueue[storeWBIdx].canWB) {
+
+        if (storeQueue[storeWBIdx].size == 0 ||
+            storeQueue[storeWBIdx].inst->isDataPrefetch() ||
+            storeQueue[storeWBIdx].committed ||
+            storeQueue[storeWBIdx].req->flags & LOCKED) {
+            incrStIdx(storeWBIdx);
+
+            continue;
+        }
+
+        assert(storeQueue[storeWBIdx].req);
+        assert(!storeQueue[storeWBIdx].committed);
+
+        MemReqPtr req = storeQueue[storeWBIdx].req;
+        storeQueue[storeWBIdx].committed = true;
+
+        req->cmd = Write;
+        req->completionEvent = NULL;
+        req->time = curTick;
+        assert(!req->data);
+        req->data = new uint8_t[64];
+        memcpy(req->data, (uint8_t *)&storeQueue[storeWBIdx].data, req->size);
+
+        DPRINTF(LSQUnit, "D-Cache: Writing back store idx:%i PC:%#x "
+                "to Addr:%#x, data:%#x [sn:%lli]\n",
+                storeWBIdx,storeQueue[storeWBIdx].inst->readPC(),
+                req->paddr, *(req->data),
+                storeQueue[storeWBIdx].inst->seqNum);
+
+        switch(storeQueue[storeWBIdx].size) {
+          case 1:
+            cpu->write(req, (uint8_t &)storeQueue[storeWBIdx].data);
+            break;
+          case 2:
+            cpu->write(req, (uint16_t &)storeQueue[storeWBIdx].data);
+            break;
+          case 4:
+            cpu->write(req, (uint32_t &)storeQueue[storeWBIdx].data);
+            break;
+          case 8:
+            cpu->write(req, (uint64_t &)storeQueue[storeWBIdx].data);
+            break;
+          default:
+            panic("Unexpected store size!\n");
+        }
+        incrStIdx(storeWBIdx);
+    }
+}
+
+template<class Impl>
+void
+LSQUnit<Impl>::takeOverFrom()
+{
+    switchedOut = false;
+    loads = stores = storesToWB = 0;
+
+    loadHead = loadTail = 0;
+
+    storeHead = storeWBIdx = storeTail = 0;
+
+    usedPorts = 0;
+
+    loadFaultInst = storeFaultInst = memDepViolator = NULL;
+
+    blockedLoadSeqNum = 0;
+
+    stalled = false;
+    isLoadBlocked = false;
+    loadBlockedHandled = false;
+}
+
+template<class Impl>
+void
 LSQUnit<Impl>::resizeLQ(unsigned size)
 {
     assert( size >= LQEntries);
@@ -647,7 +735,7 @@ LSQUnit<Impl>::writebackStores()
 
                 lastDcacheStall = curTick;
 
-                _status = DcacheMissStall;
+//                _status = DcacheMissStall;
 
                 //mshrSeqNums.push_back(storeQueue[storeWBIdx].inst->seqNum);
 
diff --git a/cpu/o3/mem_dep_unit.hh b/cpu/o3/mem_dep_unit.hh
index 32ce9f768..141e0fdc4 100644
--- a/cpu/o3/mem_dep_unit.hh
+++ b/cpu/o3/mem_dep_unit.hh
@@ -84,6 +84,10 @@ class MemDepUnit {
     /** Registers statistics. */
     void regStats();
 
+    void switchOut();
+
+    void takeOverFrom();
+
     /** Sets the pointer to the IQ. */
     void setIQ(InstructionQueue<Impl> *iq_ptr);
 
diff --git a/cpu/o3/mem_dep_unit_impl.hh b/cpu/o3/mem_dep_unit_impl.hh
index 771a0505e..05a33685d 100644
--- a/cpu/o3/mem_dep_unit_impl.hh
+++ b/cpu/o3/mem_dep_unit_impl.hh
@@ -103,6 +103,26 @@ MemDepUnit<MemDepPred, Impl>::regStats()
 
 template <class MemDepPred, class Impl>
 void
+MemDepUnit<MemDepPred, Impl>::switchOut()
+{
+    for (int i = 0; i < Impl::MaxThreads; ++i) {
+        instList[i].clear();
+    }
+    instsToReplay.clear();
+    memDepHash.clear();
+}
+
+template <class MemDepPred, class Impl>
+void
+MemDepUnit<MemDepPred, Impl>::takeOverFrom()
+{
+    loadBarrier = storeBarrier = false;
+    loadBarrierSN = storeBarrierSN = 0;
+    depPred.clear();
+}
+
+template <class MemDepPred, class Impl>
+void
 MemDepUnit<MemDepPred, Impl>::setIQ(InstructionQueue<Impl> *iq_ptr)
 {
     iqPtr = iq_ptr;
diff --git a/cpu/o3/ras.cc b/cpu/o3/ras.cc
index 5e7ef38ae..0b3ea4918 100644
--- a/cpu/o3/ras.cc
+++ b/cpu/o3/ras.cc
@@ -42,6 +42,15 @@ ReturnAddrStack::init(unsigned _numEntries)
 }
 
 void
+ReturnAddrStack::reset()
+{
+    usedEntries = 0;
+    tos = 0;
+    for (int i = 0; i < numEntries; ++i)
+        addrStack[i] = 0;
+}
+
+void
 ReturnAddrStack::push(const Addr &return_addr)
 {
     incrTos();
diff --git a/cpu/o3/ras.hh b/cpu/o3/ras.hh
index 5aa4fc05f..27e7c2df4 100644
--- a/cpu/o3/ras.hh
+++ b/cpu/o3/ras.hh
@@ -47,6 +47,8 @@ class ReturnAddrStack
      */
     void init(unsigned numEntries);
 
+    void reset();
+
     /** Returns the top address on the RAS. */
     Addr top()
     { return addrStack[tos]; }
diff --git a/cpu/o3/rename.hh b/cpu/o3/rename.hh
index c6f8f97aa..4c5c46356 100644
--- a/cpu/o3/rename.hh
+++ b/cpu/o3/rename.hh
@@ -153,6 +153,10 @@ class DefaultRename
     /** Sets pointer to the scoreboard. */
     void setScoreboard(Scoreboard *_scoreboard);
 
+    void switchOut();
+
+    void takeOverFrom();
+
     /** Squashes all instructions in a thread. */
     void squash(unsigned tid);
 
@@ -448,6 +452,7 @@ class DefaultRename
     Stats::Scalar<> renameUndoneMaps;
     Stats::Scalar<> renamedSerializing;
     Stats::Scalar<> renamedTempSerializing;
+    Stats::Scalar<> renameSkidInsts;
 };
 
 #endif // __CPU_O3_RENAME_HH__
diff --git a/cpu/o3/rename_impl.hh b/cpu/o3/rename_impl.hh
index e29211921..d41058deb 100644
--- a/cpu/o3/rename_impl.hh
+++ b/cpu/o3/rename_impl.hh
@@ -151,6 +151,11 @@ DefaultRename<Impl>::regStats()
         .desc("count of temporary serializing insts renamed")
         .flags(Stats::total)
         ;
+    renameSkidInsts
+        .name(name() + ".RENAME:skidInsts")
+        .desc("count of insts added to the skid buffer")
+        .flags(Stats::total)
+        ;
 }
 
 template <class Impl>
@@ -213,8 +218,8 @@ DefaultRename<Impl>::initStage()
 
     // Clear these pointers so they are not accidentally used in
     // non-initialization code.
-    iew_ptr = NULL;
-    commit_ptr = NULL;
+//    iew_ptr = NULL;
+//    commit_ptr = NULL;
 }
 
 template<class Impl>
@@ -255,6 +260,55 @@ DefaultRename<Impl>::setScoreboard(Scoreboard *_scoreboard)
 
 template <class Impl>
 void
+DefaultRename<Impl>::switchOut()
+{
+    for (int i = 0; i < numThreads; i++) {
+        typename list<RenameHistory>::iterator hb_it = historyBuffer[i].begin();
+
+        while (!historyBuffer[i].empty()) {
+            assert(hb_it != historyBuffer[i].end());
+
+            DPRINTF(Rename, "[tid:%u]: Removing history entry with sequence "
+                    "number %i.\n", i, (*hb_it).instSeqNum);
+
+            // Tell the rename map to set the architected register to the
+            // previous physical register that it was renamed to.
+            renameMap[i]->setEntry(hb_it->archReg, hb_it->prevPhysReg);
+
+            // Put the renamed physical register back on the free list.
+            freeList->addReg(hb_it->newPhysReg);
+
+            historyBuffer[i].erase(hb_it++);
+        }
+        insts[i].clear();
+        skidBuffer[i].clear();
+    }
+}
+
+template <class Impl>
+void
+DefaultRename<Impl>::takeOverFrom()
+{
+    _status = Inactive;
+    initStage();
+
+    for (int i=0; i< numThreads; i++) {
+        renameStatus[i] = Idle;
+
+        stalls[i].iew = false;
+        stalls[i].commit = false;
+        serializeInst[i] = NULL;
+
+        instsInProgress[i] = 0;
+
+        emptyROB[i] = true;
+
+        serializeOnNextInst[i] = false;
+    }
+}
+
+template <class Impl>
+void
 DefaultRename<Impl>::squash(unsigned tid)
 {
     DPRINTF(Rename, "[tid:%u]: Squashing instructions.\n",tid);
@@ -393,7 +447,7 @@ DefaultRename<Impl>::rename(bool &status_change, unsigned tid)
     } else if (renameStatus[tid] == Unblocking) {
         renameInsts(tid);
 
-        ++renameUnblockCycles;
+//        ++renameUnblockCycles;
 
         if (validInsts()) {
             // Add the current inputs to the skid buffer so they can be
@@ -564,6 +618,8 @@ DefaultRename<Impl>::renameInsts(unsigned tid)
         } else if (inst->isSerializeAfter() && !inst->isSerializeHandled()) {
             DPRINTF(Rename, "Serialize after instruction encountered.\n");
 
+            renamedSerializing++;
+
             inst->setSerializeHandled();
 
             serializeAfter(insts_to_rename, tid);
@@ -594,13 +650,12 @@ DefaultRename<Impl>::renameInsts(unsigned tid)
         // Increment which instruction we're on.
         ++toIEWIndex;
 
-        ++renameRenamedInsts;
-
         // Decrement how many instructions are available.
         --insts_available;
     }
 
     instsInProgress[tid] += renamed_insts;
+    renameRenamedInsts += renamed_insts;
 
     // If we wrote to the time buffer, record this.
     if (toIEWIndex) {
@@ -635,6 +690,8 @@ DefaultRename<Impl>::skidInsert(unsigned tid)
         DPRINTF(Rename, "[tid:%u]: Inserting [sn:%lli] PC:%#x into Rename "
                 "skidBuffer\n", tid, inst->seqNum, inst->readPC());
 
+        ++renameSkidInsts;
+
         skidBuffer[tid].push_back(inst);
     }
 
diff --git a/cpu/o3/rob.hh b/cpu/o3/rob.hh
index 48199915f..0748850ea 100644
--- a/cpu/o3/rob.hh
+++ b/cpu/o3/rob.hh
@@ -97,6 +97,10 @@ class ROB
      */
     void setActiveThreads(std::list<unsigned>* at_ptr);
 
+    void switchOut();
+
+    void takeOverFrom();
+
     /** Function to insert an instruction into the ROB. Note that whatever
      *  calls this function must ensure that there is enough space within the
      *  ROB for the new instruction.
diff --git a/cpu/o3/rob_impl.hh b/cpu/o3/rob_impl.hh
index 96d907cda..02a4bfbee 100644
--- a/cpu/o3/rob_impl.hh
+++ b/cpu/o3/rob_impl.hh
@@ -121,6 +121,31 @@ ROB<Impl>::setActiveThreads(list<unsigned> *at_ptr)
     activeThreads = at_ptr;
 }
 
+template <class Impl>
+void
+ROB<Impl>::switchOut()
+{
+    for (int tid = 0; tid < numThreads; tid++) {
+        instList[tid].clear();
+    }
+}
+
+template <class Impl>
+void
+ROB<Impl>::takeOverFrom()
+{
+    for (int tid=0; tid  < numThreads; tid++) {
+        doneSquashing[tid] = true;
+        threadEntries[tid] = 0;
+        squashIt[tid] = instList[tid].end();
+    }
+    numInstsInROB = 0;
+
+    // Initialize the "universal" ROB head & tail point to invalid
+    // pointers
+    head = instList[0].end();
+    tail = instList[0].end();
+}
 
 template <class Impl>
 void
diff --git a/cpu/o3/sat_counter.cc b/cpu/o3/sat_counter.cc
index a6e131483..b481b4ad2 100644
--- a/cpu/o3/sat_counter.cc
+++ b/cpu/o3/sat_counter.cc
@@ -30,17 +30,17 @@
 #include "cpu/o3/sat_counter.hh"
 
 SatCounter::SatCounter()
-    : maxVal(0), counter(0)
+    : initialVal(0), counter(0)
 {
 }
 
 SatCounter::SatCounter(unsigned bits)
-    : maxVal((1 << bits) - 1), counter(0)
+    : initialVal(0), maxVal((1 << bits) - 1), counter(0)
 {
 }
 
-SatCounter::SatCounter(unsigned bits, unsigned initial_val)
-    : maxVal((1 << bits) - 1), counter(initial_val)
+SatCounter::SatCounter(unsigned bits, uint8_t initial_val)
+    : initialVal(initialVal), maxVal((1 << bits) - 1), counter(initial_val)
 {
     // Check to make sure initial value doesn't exceed the max counter value.
     if (initial_val > maxVal) {
@@ -53,19 +53,3 @@ SatCounter::setBits(unsigned bits)
 {
     maxVal = (1 << bits) - 1;
 }
-
-void
-SatCounter::increment()
-{
-    if (counter < maxVal) {
-        ++counter;
-    }
-}
-
-void
-SatCounter::decrement()
-{
-    if (counter > 0) {
-        --counter;
-    }
-}
diff --git a/cpu/o3/sat_counter.hh b/cpu/o3/sat_counter.hh
index 952f1f86d..1d20a8a8f 100644
--- a/cpu/o3/sat_counter.hh
+++ b/cpu/o3/sat_counter.hh
@@ -57,22 +57,34 @@ class SatCounter
      * @param bits How many bits the counter will have.
      * @param initial_val Starting value for each counter.
      */
-    SatCounter(unsigned bits, unsigned initial_val);
+    SatCounter(unsigned bits, uint8_t initial_val);
 
     /**
      * Sets the number of bits.
      */
     void setBits(unsigned bits);
 
+    void reset() { counter = initialVal; }
+
     /**
      * Increments the counter's current value.
      */
-    void increment();
+    void increment()
+    {
+        if (counter < maxVal) {
+            ++counter;
+        }
+    }
 
     /**
      * Decrements the counter's current value.
      */
-    void decrement();
+    void decrement()
+    {
+        if (counter > 0) {
+            --counter;
+        }
+    }
 
     /**
      * Read the counter's value.
@@ -81,6 +93,7 @@ class SatCounter
     { return counter; }
 
   private:
+    uint8_t initialVal;
     uint8_t maxVal;
     uint8_t counter;
 };
diff --git a/cpu/o3/thread_state.hh b/cpu/o3/thread_state.hh
index 846f44176..17719bdeb 100644
--- a/cpu/o3/thread_state.hh
+++ b/cpu/o3/thread_state.hh
@@ -60,7 +60,7 @@ struct O3ThreadState : public ThreadState {
     { }
 #else
     O3ThreadState(FullCPU *_cpu, int _thread_num, Process *_process, int _asid)
-        : ThreadState(-1, _thread_num, NULL, _process, _asid),
+        : ThreadState(-1, _thread_num, _process->getMemory(), _process, _asid),
           cpu(_cpu), inSyscall(0), trapPending(0)
     { }