6 files changed, 80 insertions, 9 deletions

diff --git a/src/cpu/o3/alpha/cpu_builder.cc b/src/cpu/o3/alpha/cpu_builder.cc
index 490305cbf..d61eee4b1 100644
--- a/src/cpu/o3/alpha/cpu_builder.cc
+++ b/src/cpu/o3/alpha/cpu_builder.cc
@@ -91,7 +91,10 @@ Param<unsigned> renameWidth;
 Param<unsigned> commitToIEWDelay;
 Param<unsigned> renameToIEWDelay;
 Param<unsigned> issueToExecuteDelay;
+Param<unsigned> dispatchWidth;
 Param<unsigned> issueWidth;
+Param<unsigned> wbWidth;
+Param<unsigned> wbDepth;
 SimObjectParam<FUPool *> fuPool;
 
 Param<unsigned> iewToCommitDelay;
@@ -207,7 +210,10 @@ BEGIN_INIT_SIM_OBJECT_PARAMS(DerivO3CPU)
                "Issue/Execute/Writeback delay"),
     INIT_PARAM(issueToExecuteDelay, "Issue to execute delay (internal"
                "to the IEW stage)"),
+    INIT_PARAM(dispatchWidth, "Dispatch width"),
     INIT_PARAM(issueWidth, "Issue width"),
+    INIT_PARAM(wbWidth, "Writeback width"),
+    INIT_PARAM(wbDepth, "Writeback depth (number of cycles it can buffer)"),
     INIT_PARAM_DFLT(fuPool, "Functional unit pool", NULL),
 
     INIT_PARAM(iewToCommitDelay, "Issue/Execute/Writeback to commit "
@@ -333,7 +339,10 @@ CREATE_SIM_OBJECT(DerivO3CPU)
     params->commitToIEWDelay = commitToIEWDelay;
     params->renameToIEWDelay = renameToIEWDelay;
     params->issueToExecuteDelay = issueToExecuteDelay;
+    params->dispatchWidth = dispatchWidth;
     params->issueWidth = issueWidth;
+    params->wbWidth = wbWidth;
+    params->wbDepth = wbDepth;
     params->fuPool = fuPool;
 
     params->iewToCommitDelay = iewToCommitDelay;
diff --git a/src/cpu/o3/alpha/params.hh b/src/cpu/o3/alpha/params.hh
index b1f2a487d..8f7364dd0 100644
--- a/src/cpu/o3/alpha/params.hh
+++ b/src/cpu/o3/alpha/params.hh
@@ -64,6 +64,9 @@ class AlphaSimpleParams : public O3Params
     BaseCPU *checker;
 
     unsigned decodeToFetchDelay;
+    unsigned dispatchWidth;
+    unsigned wbWidth;
+    unsigned wbDepth;
 };
 
 #endif // __CPU_O3_ALPHA_PARAMS_HH__
diff --git a/src/cpu/o3/iew.hh b/src/cpu/o3/iew.hh
index 2af68d8fc..9627609c2 100644
--- a/src/cpu/o3/iew.hh
+++ b/src/cpu/o3/iew.hh
@@ -204,6 +204,45 @@ class DefaultIEW
     /** Returns if the LSQ has any stores to writeback. */
     bool hasStoresToWB() { return ldstQueue.hasStoresToWB(); }
 
+    void incrWb(InstSeqNum &sn)
+    {
+        if (++wbOutstanding == wbMax)
+            ableToIssue = false;
+        DPRINTF(IEW, "wbOutstanding: %i\n", wbOutstanding);
+#if DEBUG
+        wbList.insert(sn);
+#endif
+    }
+
+    void decrWb(InstSeqNum &sn)
+    {
+        if (wbOutstanding-- == wbMax)
+            ableToIssue = true;
+        DPRINTF(IEW, "wbOutstanding: %i\n", wbOutstanding);
+#if DEBUG
+        assert(wbList.find(sn) != wbList.end());
+        wbList.erase(sn);
+#endif
+    }
+
+#if DEBUG
+    std::set<InstSeqNum> wbList;
+
+    void dumpWb()
+    {
+        std::set<InstSeqNum>::iterator wb_it = wbList.begin();
+        while (wb_it != wbList.end()) {
+            cprintf("[sn:%lli]\n",
+                    (*wb_it));
+            wb_it++;
+        }
+    }
+#endif
+
+    bool canIssue() { return ableToIssue; }
+
+    bool ableToIssue;
+
   private:
     /** Sends commit proper information for a squash due to a branch
      * mispredict.
@@ -384,11 +423,8 @@ class DefaultIEW
      */
     unsigned issueToExecuteDelay;
 
-    /** Width of issue's read path, in instructions.  The read path is both
-     *  the skid buffer and the rename instruction queue.
-     *  Note to self: is this really different than issueWidth?
-     */
-    unsigned issueReadWidth;
+    /** Width of dispatch, in instructions. */
+    unsigned dispatchWidth;
 
     /** Width of issue, in instructions. */
     unsigned issueWidth;
@@ -403,6 +439,17 @@ class DefaultIEW
      */
     unsigned wbCycle;
 
+    /** Number of instructions in flight that will writeback. */
+    unsigned wbOutstanding;
+
+    /** Writeback width. */
+    unsigned wbWidth;
+
+    /** Writeback width * writeback depth, where writeback depth is
+     * the number of cycles of writing back instructions that can be
+     * buffered. */
+    unsigned wbMax;
+
     /** Number of active threads. */
     unsigned numThreads;
 
diff --git a/src/cpu/o3/iew_impl.hh b/src/cpu/o3/iew_impl.hh
index 8e6fd46a1..118038b65 100644
--- a/src/cpu/o3/iew_impl.hh
+++ b/src/cpu/o3/iew_impl.hh
@@ -50,8 +50,10 @@ DefaultIEW<Impl>::DefaultIEW(Params *params)
       commitToIEWDelay(params->commitToIEWDelay),
       renameToIEWDelay(params->renameToIEWDelay),
       issueToExecuteDelay(params->issueToExecuteDelay),
-      issueReadWidth(params->issueWidth),
+      dispatchWidth(params->dispatchWidth),
       issueWidth(params->issueWidth),
+      wbOutstanding(0),
+      wbWidth(params->wbWidth),
       numThreads(params->numberOfThreads),
       switchedOut(false)
 {
@@ -74,8 +76,12 @@ DefaultIEW<Impl>::DefaultIEW(Params *params)
         fetchRedirect[i] = false;
     }
 
+    wbMax = wbWidth * params->wbDepth;
+
     updateLSQNextCycle = false;
 
+    ableToIssue = true;
+
     skidBufferMax = (3 * (renameToIEWDelay * params->renameWidth)) + issueWidth;
 }
 
@@ -559,12 +565,12 @@ DefaultIEW<Impl>::instToCommit(DynInstPtr &inst)
     // free slot.
     while ((*iewQueue)[wbCycle].insts[wbNumInst]) {
         ++wbNumInst;
-        if (wbNumInst == issueWidth) {
+        if (wbNumInst == wbWidth) {
             ++wbCycle;
             wbNumInst = 0;
         }
 
-        assert(wbCycle < 5);
+        assert((wbCycle * wbWidth + wbNumInst) < wbMax);
     }
 
     // Add finished instruction to queue to commit.
@@ -937,7 +943,7 @@ DefaultIEW<Impl>::dispatchInsts(unsigned tid)
     // Loop through the instructions, putting them in the instruction
     // queue.
     for ( ; dis_num_inst < insts_to_add &&
-              dis_num_inst < issueReadWidth;
+              dis_num_inst < dispatchWidth;
           ++dis_num_inst)
     {
         inst = insts_to_dispatch.front();
@@ -1189,6 +1195,7 @@ DefaultIEW<Impl>::executeInsts()
 
             ++iewExecSquashedInsts;
 
+            decrWb(inst->seqNum);
             continue;
         }
 
@@ -1351,6 +1358,8 @@ DefaultIEW<Impl>::writebackInsts()
             }
             writebackCount[tid]++;
         }
+
+        decrWb(inst->seqNum);
     }
 }
 
diff --git a/src/cpu/o3/inst_queue_impl.hh b/src/cpu/o3/inst_queue_impl.hh
index b99bd0900..36e0842be 100644
--- a/src/cpu/o3/inst_queue_impl.hh
+++ b/src/cpu/o3/inst_queue_impl.hh
@@ -687,6 +687,7 @@ InstructionQueue<Impl>::scheduleReadyInsts()
     int total_issued = 0;
 
     while (total_issued < totalWidth &&
+           iewStage->canIssue() &&
            order_it != order_end_it) {
         OpClass op_class = (*order_it).queueType;
 
@@ -784,6 +785,7 @@ InstructionQueue<Impl>::scheduleReadyInsts()
 
             listOrder.erase(order_it++);
             statIssuedInstType[tid][op_class]++;
+            iewStage->incrWb(issuing_inst->seqNum);
         } else {
             statFuBusy[op_class]++;
             fuBusy[tid]++;
diff --git a/src/cpu/o3/lsq_unit_impl.hh b/src/cpu/o3/lsq_unit_impl.hh
index 714acb2ef..bb3da7eec 100644
--- a/src/cpu/o3/lsq_unit_impl.hh
+++ b/src/cpu/o3/lsq_unit_impl.hh
@@ -77,6 +77,7 @@ LSQUnit<Impl>::completeDataAccess(PacketPtr pkt)
     //iewStage->ldstQueue.removeMSHR(inst->threadNumber,inst->seqNum);
 
     if (isSwitchedOut() || inst->isSquashed()) {
+        iewStage->decrWb(inst->seqNum);
         delete state;
         delete pkt;
         return;