Minor changes plus updates to O3.

cpu/base.cc:
    Have output message regardless of build.
cpu/checker/cpu_builder.cc:
cpu/checker/o3_cpu_builder.cc:
    Be sure to include all parameters.
cpu/o3/cpu.cc:
    IEW also needs to switch out.
cpu/o3/iew_impl.hh:
    Handle stores with faults properly.
cpu/o3/inst_queue_impl.hh:
    Switch out properly, handle squashing properly.
cpu/o3/lsq_unit_impl.hh:
    Minor fixes.
cpu/o3/mem_dep_unit_impl.hh:
    Make sure mem dep unit is switched out properly.
cpu/o3/rename_impl.hh:
    Switch out fix.

--HG--
extra : convert_revision : b94deb83f724225c01166c84a1b3fdd3543cbe9a

9 files changed, 61 insertions, 63 deletions

diff --git a/cpu/base.cc b/cpu/base.cc
index 044fafca9..d4ba8c812 100644
--- a/cpu/base.cc
+++ b/cpu/base.cc
@@ -60,15 +60,19 @@ int maxThreadsPerCPU = 1;
 void
 CPUProgressEvent::process()
 {
-#ifndef NDEBUG
     Counter temp = cpu->totalInstructions();
+#ifndef NDEBUG
     double ipc = double(temp - lastNumInst) / (interval / cpu->cycles(1));
+
     DPRINTFN("%s progress event, instructions committed: %lli, IPC: %0.8d\n",
              cpu->name(), temp - lastNumInst, ipc);
     ipc = 0.0;
+#else
+    cprintf("%lli: %s progress event, instructions committed: %lli\n",
+            curTick, cpu->name(), temp - lastNumInst);
+#endif
     lastNumInst = temp;
     schedule(curTick + interval);
-#endif
 }
 
 const char *
diff --git a/cpu/checker/cpu_builder.cc b/cpu/checker/cpu_builder.cc
index ec36ae09f..d68dcdcd9 100644
--- a/cpu/checker/cpu_builder.cc
+++ b/cpu/checker/cpu_builder.cc
@@ -144,6 +144,7 @@ CREATE_SIM_OBJECT(OzoneChecker)
     temp = max_insts_all_threads;
     temp = max_loads_any_thread;
     temp = max_loads_all_threads;
+    temp = stats_reset_inst;
     Tick temp2 = progress_interval;
     temp2++;
     params->progress_interval = 0;
diff --git a/cpu/checker/o3_cpu_builder.cc b/cpu/checker/o3_cpu_builder.cc
index 496cca779..079217b0c 100644
--- a/cpu/checker/o3_cpu_builder.cc
+++ b/cpu/checker/o3_cpu_builder.cc
@@ -58,6 +58,7 @@ BEGIN_DECLARE_SIM_OBJECT_PARAMS(O3Checker)
     Param<Counter> max_insts_all_threads;
     Param<Counter> max_loads_any_thread;
     Param<Counter> max_loads_all_threads;
+    Param<Counter> stats_reset_inst;
     Param<Tick> progress_interval;
 
 #if FULL_SYSTEM
@@ -92,6 +93,8 @@ BEGIN_INIT_SIM_OBJECT_PARAMS(O3Checker)
                "terminate when any thread reaches this load count"),
     INIT_PARAM(max_loads_all_threads,
                "terminate when all threads have reached this load count"),
+    INIT_PARAM(stats_reset_inst,
+               "blah"),
     INIT_PARAM_DFLT(progress_interval, "CPU Progress Interval", 0),
 
 #if FULL_SYSTEM
@@ -127,6 +130,7 @@ CREATE_SIM_OBJECT(O3Checker)
     params->max_insts_all_threads = 0;
     params->max_loads_any_thread = 0;
     params->max_loads_all_threads = 0;
+    params->stats_reset_inst = 0;
     params->exitOnError = exitOnError;
     params->updateOnError = updateOnError;
     params->deferRegistration = defer_registration;
@@ -140,7 +144,9 @@ CREATE_SIM_OBJECT(O3Checker)
     temp = max_insts_all_threads;
     temp = max_loads_any_thread;
     temp = max_loads_all_threads;
+    temp = stats_reset_inst;
     Tick temp2 = progress_interval;
+    params->progress_interval = 0;
     temp2++;
     BaseMem *cache = icache;
     cache = dcache;
diff --git a/cpu/o3/cpu.cc b/cpu/o3/cpu.cc
index 88de6c746..21cd1c599 100644
--- a/cpu/o3/cpu.cc
+++ b/cpu/o3/cpu.cc
@@ -697,6 +697,7 @@ FullO3CPU<Impl>::signalSwitched()
     if (++switchCount == NumStages) {
         fetch.doSwitchOut();
         rename.doSwitchOut();
+        iew.doSwitchOut();
         commit.doSwitchOut();
         instList.clear();
 
diff --git a/cpu/o3/iew_impl.hh b/cpu/o3/iew_impl.hh
index 102be4f8d..33fd0f6b9 100644
--- a/cpu/o3/iew_impl.hh
+++ b/cpu/o3/iew_impl.hh
@@ -431,6 +431,8 @@ DefaultIEW<Impl>::doSwitchOut()
 {
     // Clear any state.
     switchedOut = true;
+    assert(insts[0].empty());
+    assert(skidBuffer[0].empty());
 
     instQueue.switchOut();
     ldstQueue.switchOut();
@@ -1281,13 +1283,23 @@ DefaultIEW<Impl>::executeInsts()
                 // event adds the instruction to the queue to commit
                 fault = ldstQueue.executeLoad(inst);
             } else if (inst->isStore()) {
-                ldstQueue.executeStore(inst);
+                fault = ldstQueue.executeStore(inst);
 
                 // If the store had a fault then it may not have a mem req
-                if (inst->req && !(inst->req->flags & LOCKED)) {
+                if (!inst->isStoreConditional() && fault == NoFault) {
                     inst->setExecuted();
 
                     instToCommit(inst);
+                } else if (fault != NoFault) {
+                    // If the instruction faulted, then we need to send it along to commit
+                    // without the instruction completing.
+
+                    // Send this instruction to commit, also make sure iew stage
+                    // realizes there is activity.
+                    inst->setExecuted();
+
+                    instToCommit(inst);
+                    activityThisCycle();
                 }
 
                 // Store conditionals will mark themselves as
@@ -1408,7 +1420,7 @@ DefaultIEW<Impl>::writebackInsts()
         // E.g. Uncached loads have not actually executed when they
         // are first sent to commit.  Instead commit must tell the LSQ
         // when it's ready to execute the uncached load.
-        if (!inst->isSquashed() && inst->isExecuted()) {
+        if (!inst->isSquashed() && inst->isExecuted() && inst->getFault() == NoFault) {
             int dependents = instQueue.wakeDependents(inst);
 
             for (int i = 0; i < inst->numDestRegs(); i++) {
diff --git a/cpu/o3/inst_queue_impl.hh b/cpu/o3/inst_queue_impl.hh
index b6b06ca77..0a17cae5c 100644
--- a/cpu/o3/inst_queue_impl.hh
+++ b/cpu/o3/inst_queue_impl.hh
@@ -386,8 +386,16 @@ template <class Impl>
 void
 InstructionQueue<Impl>::switchOut()
 {
+/*
+    if (!instList[0].empty() || (numEntries != freeEntries) ||
+        !readyInsts[0].empty() || !nonSpecInsts.empty() || !listOrder.empty()) {
+        dumpInsts();
+//        assert(0);
+    }
+*/
     resetState();
     dependGraph.reset();
+    instsToExecute.clear();
     switchedOut = true;
     for (int i = 0; i < numThreads; ++i) {
         memDepUnit[i].switchOut();
@@ -643,9 +651,12 @@ template <class Impl>
 void
 InstructionQueue<Impl>::processFUCompletion(DynInstPtr &inst, int fu_idx)
 {
+    DPRINTF(IQ, "Processing FU completion [sn:%lli]\n", inst->seqNum);
     // 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()) {
+        DPRINTF(IQ, "FU completion not processed, IQ is switched out [sn:%lli]\n",
+                inst->seqNum);
         return;
     }
 
@@ -1033,6 +1044,10 @@ InstructionQueue<Impl>::doSquash(unsigned tid)
             (squashed_inst->isMemRef() &&
              !squashed_inst->memOpDone)) {
 
+            DPRINTF(IQ, "[tid:%i]: Instruction [sn:%lli] PC %#x "
+                    "squashed.\n",
+                    tid, squashed_inst->seqNum, squashed_inst->readPC());
+
             // Remove the instruction from the dependency list.
             if (!squashed_inst->isNonSpeculative() &&
                 !squashed_inst->isStoreConditional() &&
@@ -1063,7 +1078,7 @@ InstructionQueue<Impl>::doSquash(unsigned tid)
 
                     ++iqSquashedOperandsExamined;
                 }
-            } else {
+            } else if (!squashed_inst->isStoreConditional() || !squashed_inst->isCompleted()) {
                 NonSpecMapIt ns_inst_it =
                     nonSpecInsts.find(squashed_inst->seqNum);
                 assert(ns_inst_it != nonSpecInsts.end());
@@ -1090,10 +1105,6 @@ InstructionQueue<Impl>::doSquash(unsigned tid)
             count[squashed_inst->threadNumber]--;
 
             ++freeEntries;
-
-            DPRINTF(IQ, "[tid:%i]: Instruction [sn:%lli] PC %#x "
-                    "squashed.\n",
-                    tid, squashed_inst->seqNum, squashed_inst->readPC());
         }
 
         instList[tid].erase(squash_it--);
diff --git a/cpu/o3/lsq_unit_impl.hh b/cpu/o3/lsq_unit_impl.hh
index 7086c381e..f75a41cfe 100644
--- a/cpu/o3/lsq_unit_impl.hh
+++ b/cpu/o3/lsq_unit_impl.hh
@@ -198,62 +198,12 @@ void
 LSQUnit<Impl>::switchOut()
 {
     switchedOut = true;
-    for (int i = 0; i < loadQueue.size(); ++i)
+    for (int i = 0; i < loadQueue.size(); ++i) {
+        assert(!loadQueue[i]);
         loadQueue[i] = NULL;
+    }
 
     assert(storesToWB == 0);
-
-    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>
@@ -439,6 +389,11 @@ LSQUnit<Impl>::executeLoad(DynInstPtr &inst)
     if (load_fault != NoFault) {
         // Send this instruction to commit, also make sure iew stage
         // realizes there is activity.
+        // Mark it as executed unless it is an uncached load that
+        // needs to hit the head of commit.
+        if (!(inst->req->flags & UNCACHEABLE) || inst->isAtCommit()) {
+            inst->setExecuted();
+        }
         iewStage->instToCommit(inst);
         iewStage->activityThisCycle();
     }
diff --git a/cpu/o3/mem_dep_unit_impl.hh b/cpu/o3/mem_dep_unit_impl.hh
index bfe694bd8..a2d04ece9 100644
--- a/cpu/o3/mem_dep_unit_impl.hh
+++ b/cpu/o3/mem_dep_unit_impl.hh
@@ -107,6 +107,9 @@ template <class MemDepPred, class Impl>
 void
 MemDepUnit<MemDepPred, Impl>::switchOut()
 {
+    assert(instList[0].empty());
+    assert(instsToReplay.empty());
+    assert(memDepHash.empty());
     // Clear any state.
     for (int i = 0; i < Impl::MaxThreads; ++i) {
         instList[i].clear();
diff --git a/cpu/o3/rename_impl.hh b/cpu/o3/rename_impl.hh
index 49627e3d4..a41e8d016 100644
--- a/cpu/o3/rename_impl.hh
+++ b/cpu/o3/rename_impl.hh
@@ -864,6 +864,11 @@ DefaultRename<Impl>::doSquash(unsigned tid)
         // Put the renamed physical register back on the free list.
         freeList->addReg(hb_it->newPhysReg);
 
+        // Be sure to mark its register as ready if it's a misc register.
+        if (hb_it->newPhysReg >= maxPhysicalRegs) {
+            scoreboard->setReg(hb_it->newPhysReg);
+        }
+
         historyBuffer[tid].erase(hb_it++);
 
         ++renameUndoneMaps;