Yet another merge with the main repository.

--HG--
rename : tests/long/10.linux-boot/ref/x86/linux/pc-o3-timing/config.ini => tests/long/fs/10.linux-boot/ref/x86/linux/pc-o3-timing/config.ini
rename : tests/long/10.linux-boot/ref/x86/linux/pc-o3-timing/simout => tests/long/fs/10.linux-boot/ref/x86/linux/pc-o3-timing/simout
rename : tests/long/10.linux-boot/ref/x86/linux/pc-o3-timing/stats.txt => tests/long/fs/10.linux-boot/ref/x86/linux/pc-o3-timing/stats.txt
rename : tests/long/10.linux-boot/ref/x86/linux/pc-o3-timing/system.pc.com_1.terminal => tests/long/fs/10.linux-boot/ref/x86/linux/pc-o3-timing/system.pc.com_1.terminal
rename : tests/long/00.gzip/ref/x86/linux/o3-timing/config.ini => tests/long/se/00.gzip/ref/x86/linux/o3-timing/config.ini
rename : tests/long/00.gzip/ref/x86/linux/o3-timing/simout => tests/long/se/00.gzip/ref/x86/linux/o3-timing/simout
rename : tests/long/00.gzip/ref/x86/linux/o3-timing/stats.txt => tests/long/se/00.gzip/ref/x86/linux/o3-timing/stats.txt
rename : tests/long/10.mcf/ref/x86/linux/o3-timing/config.ini => tests/long/se/10.mcf/ref/x86/linux/o3-timing/config.ini
rename : tests/long/10.mcf/ref/x86/linux/o3-timing/simout => tests/long/se/10.mcf/ref/x86/linux/o3-timing/simout
rename : tests/long/10.mcf/ref/x86/linux/o3-timing/stats.txt => tests/long/se/10.mcf/ref/x86/linux/o3-timing/stats.txt
rename : tests/long/20.parser/ref/x86/linux/o3-timing/config.ini => tests/long/se/20.parser/ref/x86/linux/o3-timing/config.ini
rename : tests/long/20.parser/ref/x86/linux/o3-timing/simout => tests/long/se/20.parser/ref/x86/linux/o3-timing/simout
rename : tests/long/20.parser/ref/x86/linux/o3-timing/stats.txt => tests/long/se/20.parser/ref/x86/linux/o3-timing/stats.txt
rename : tests/long/70.twolf/ref/x86/linux/o3-timing/config.ini => tests/long/se/70.twolf/ref/x86/linux/o3-timing/config.ini
rename : tests/long/70.twolf/ref/x86/linux/o3-timing/simout => tests/long/se/70.twolf/ref/x86/linux/o3-timing/simout
rename : tests/long/70.twolf/ref/x86/linux/o3-timing/stats.txt => tests/long/se/70.twolf/ref/x86/linux/o3-timing/stats.txt
rename : tests/quick/00.hello/ref/x86/linux/o3-timing/config.ini => tests/quick/se/00.hello/ref/x86/linux/o3-timing/config.ini
rename : tests/quick/00.hello/ref/x86/linux/o3-timing/simout => tests/quick/se/00.hello/ref/x86/linux/o3-timing/simout
rename : tests/quick/00.hello/ref/x86/linux/o3-timing/stats.txt => tests/quick/se/00.hello/ref/x86/linux/o3-timing/stats.txt

src/cpu/o3/O3CPU.py

@@ -139,3 +139,5 @@ class DerivO3CPU(BaseCPU):
     smtROBThreshold = Param.Int(100, "SMT ROB Threshold Sharing Parameter")
     smtCommitPolicy = Param.String('RoundRobin', "SMT Commit Policy")
 
+    needsTSO = Param.Bool(buildEnv['TARGET_ISA'] == 'x86',
+                          "Enable TSO Memory model")

src/cpu/o3/lsq_unit.hh

@@ -452,6 +452,9 @@ class LSQUnit {
     /** Has the blocked load been handled. */
     bool loadBlockedHandled;
 
+    /** Whether or not a store is in flight. */
+    bool storeInFlight;
+
     /** The sequence number of the blocked load. */
     InstSeqNum blockedLoadSeqNum;
 
@@ -465,6 +468,9 @@ class LSQUnit {
     /** The packet that is pending free cache ports. */
     PacketPtr pendingPkt;
 
+    /** Flag for memory model. */
+    bool needsTSO;
+
     // 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.

src/cpu/o3/lsq_unit_impl.hh

@@ -138,7 +138,7 @@ template <class Impl>
 LSQUnit<Impl>::LSQUnit()
     : loads(0), stores(0), storesToWB(0), cacheBlockMask(0), stalled(false),
       isStoreBlocked(false), isLoadBlocked(false),
-      loadBlockedHandled(false), hasPendingPkt(false)
+      loadBlockedHandled(false), storeInFlight(false), hasPendingPkt(false)
 {
 }
 
@@ -182,6 +182,7 @@ LSQUnit<Impl>::init(O3CPU *cpu_ptr, IEW *iew_ptr, DerivO3CPUParams *params,
     memDepViolator = NULL;
 
     blockedLoadSeqNum = 0;
+    needsTSO = params->needsTSO;
 }
 
 template<class Impl>
@@ -770,6 +771,7 @@ LSQUnit<Impl>::writebackStores()
            storeWBIdx != storeTail &&
            storeQueue[storeWBIdx].inst &&
            storeQueue[storeWBIdx].canWB &&
+           ((!needsTSO) || (!storeInFlight)) &&
            usedPorts < cachePorts) {
 
         if (isStoreBlocked || lsq->cacheBlocked()) {
@@ -1090,6 +1092,10 @@ LSQUnit<Impl>::storePostSend(PacketPtr pkt)
 #endif
     }
 
+    if (needsTSO) {
+        storeInFlight = true;
+    }
+
     incrStIdx(storeWBIdx);
 }
 
@@ -1163,6 +1169,10 @@ LSQUnit<Impl>::completeStore(int store_idx)
 
     storeQueue[store_idx].inst->setCompleted();
 
+    if (needsTSO) {
+        storeInFlight = false;
+    }
+
     // Tell the checker we've completed this instruction.  Some stores
     // may get reported twice to the checker, but the checker can
     // handle that case.