Add some different parameters. The main change is that the writeback count is now limited so that it doesn't overflow the buffer.

src/cpu/o3/alpha_cpu_builder.cc:
src/cpu/o3/alpha_params.hh:
    Add in dispatchWidth, wbWidth, wbDepth parameters.  wbDepth is the number of cycles of wbWidth instructions that can be buffered.
src/cpu/o3/iew.hh:
    Include separate parameter for dispatch width.
    Also limit the number of outstanding writebacks so the writeback buffer isn't overflowed.  The IQ must make sure with the IEW stage that it can issue instructions prior to issuing.
src/cpu/o3/iew_impl.hh:
    Include separate parameter for dispatch width.
    Also limit the number of outstanding writebacks so the writeback buffer isn't overflowed.
src/cpu/o3/inst_queue_impl.hh:
    IQ needs to check with the IEW to make sure it can issue instructions, and increments the IEW wb counter each time there is an outstanding instruction that will writeback.
src/cpu/o3/lsq_unit_impl.hh:
    Be sure to decrement the writeback counter if there's a squashed load that returned.
src/python/m5/objects/AlphaO3CPU.py:
    Change the parameters to include dispatch width, writeback width, and writeback depth.

--HG--
extra : convert_revision : 31c8cc495273e3c481b79055562fc40f71291fc4

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 b1e141ff4..4f5dd0465 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(DerivAlphaO3CPU)
                "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(DerivAlphaO3CPU)
     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 f0732733e..78246e353 100644
--- a/src/cpu/o3/alpha_params.hh
+++ b/src/cpu/o3/alpha_params.hh
@@ -104,7 +104,10 @@ class AlphaSimpleParams : public BaseO3CPU::Params
     unsigned commitToIEWDelay;
     unsigned renameToIEWDelay;
     unsigned issueToExecuteDelay;
+    unsigned dispatchWidth;
     unsigned issueWidth;
+    unsigned wbWidth;
+    unsigned wbDepth;
     FUPool *fuPool;
 
     //
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 1ef1b2cff..61c04cc2b 100644
--- a/src/cpu/o3/inst_queue_impl.hh
+++ b/src/cpu/o3/inst_queue_impl.hh
@@ -686,6 +686,7 @@ InstructionQueue<Impl>::scheduleReadyInsts()
     int total_issued = 0;
 
     while (total_issued < totalWidth &&
+           iewStage->canIssue() &&
            order_it != order_end_it) {
         OpClass op_class = (*order_it).queueType;
 
@@ -783,6 +784,7 @@ 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;