mcpat: Adds McPAT performance counters

Updated patches from Rick Strong's set that modify performance counters for
McPAT

11 files changed, 211 insertions, 2 deletions

diff --git a/src/cpu/o3/commit.hh b/src/cpu/o3/commit.hh
index 659b0ad5f..047e29f5d 100644
--- a/src/cpu/o3/commit.hh
+++ b/src/cpu/o3/commit.hh
@@ -473,6 +473,12 @@ class DefaultCommit
     Stats::Vector statComMembars;
     /** Total number of committed branches. */
     Stats::Vector statComBranches;
+    /** Total number of floating point instructions */
+    Stats::Vector statComFloating;
+    /** Total number of integer instructions */
+    Stats::Vector statComInteger;
+    /** Total number of function calls */
+    Stats::Vector statComFunctionCalls;
 
     /** Number of cycles where the commit bandwidth limit is reached. */
     Stats::Scalar commitEligibleSamples;
diff --git a/src/cpu/o3/commit_impl.hh b/src/cpu/o3/commit_impl.hh
index a49e1497e..50c08e162 100644
--- a/src/cpu/o3/commit_impl.hh
+++ b/src/cpu/o3/commit_impl.hh
@@ -230,6 +230,27 @@ DefaultCommit<Impl>::regStats()
         .flags(total)
         ;
 
+    statComFloating
+        .init(cpu->numThreads)
+        .name(name() + ".COM:fp_insts")
+        .desc("Number of committed floating point instructions.")
+        .flags(total)
+        ;
+
+    statComInteger
+        .init(cpu->numThreads)
+        .name(name()+".COM:int_insts")
+        .desc("Number of committed integer instructions.")
+        .flags(total)
+        ;
+
+    statComFunctionCalls
+        .init(cpu->numThreads)
+        .name(name()+".COM:function_calls")
+        .desc("Number of function calls committed.")
+        .flags(total)
+        ;
+
     commitEligible
         .init(cpu->numThreads)
         .name(name() + ".COM:bw_limited")
@@ -1321,6 +1342,19 @@ DefaultCommit<Impl>::updateComInstStats(DynInstPtr &inst)
     if (inst->isMemBarrier()) {
         statComMembars[tid]++;
     }
+
+    // Integer Instruction
+    if (inst->isInteger())
+        statComInteger[tid]++;
+
+    // Floating Point Instruction
+    if (inst->isFloating())
+        statComFloating[tid]++;
+
+    // Function Calls
+    if (inst->isCall())
+        statComFunctionCalls[tid]++;
+
 }
 
 ////////////////////////////////////////
diff --git a/src/cpu/o3/cpu.cc b/src/cpu/o3/cpu.cc
index 9becc6601..2d3bc3f72 100644
--- a/src/cpu/o3/cpu.cc
+++ b/src/cpu/o3/cpu.cc
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2004-2006 The Regents of The University of Michigan
+ * Copyright (c) 2011 Regents of the University of California
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -27,6 +28,7 @@
  *
  * Authors: Kevin Lim
  *          Korey Sewell
+ *          Rick Strong
  */
 
 #include "config/full_system.hh"
@@ -480,6 +482,37 @@ FullO3CPU<Impl>::regStats()
     this->rename.regStats();
     this->iew.regStats();
     this->commit.regStats();
+    this->rob.regStats();
+
+    intRegfileReads
+        .name(name() + ".int_regfile_reads")
+        .desc("number of integer regfile reads")
+        .prereq(intRegfileReads);
+
+    intRegfileWrites
+        .name(name() + ".int_regfile_writes")
+        .desc("number of integer regfile writes")
+        .prereq(intRegfileWrites);
+
+    fpRegfileReads
+        .name(name() + ".fp_regfile_reads")
+        .desc("number of floating regfile reads")
+        .prereq(fpRegfileReads);
+
+    fpRegfileWrites
+        .name(name() + ".fp_regfile_writes")
+        .desc("number of floating regfile writes")
+        .prereq(fpRegfileWrites);
+
+    miscRegfileReads
+        .name(name() + ".misc_regfile_reads")
+        .desc("number of misc regfile reads")
+        .prereq(miscRegfileReads);
+
+    miscRegfileWrites
+        .name(name() + ".misc_regfile_writes")
+        .desc("number of misc regfile writes")
+        .prereq(miscRegfileWrites);
 }
 
 template <class Impl>
@@ -1184,6 +1217,7 @@ template <class Impl>
 TheISA::MiscReg
 FullO3CPU<Impl>::readMiscReg(int misc_reg, ThreadID tid)
 {
+    miscRegfileReads++;
     return this->isa[tid].readMiscReg(misc_reg, tcBase(tid));
 }
 
@@ -1200,6 +1234,7 @@ void
 FullO3CPU<Impl>::setMiscReg(int misc_reg,
         const TheISA::MiscReg &val, ThreadID tid)
 {
+    miscRegfileWrites++;
     this->isa[tid].setMiscReg(misc_reg, val, tcBase(tid));
 }
 
@@ -1207,6 +1242,7 @@ template <class Impl>
 uint64_t
 FullO3CPU<Impl>::readIntReg(int reg_idx)
 {
+    intRegfileReads++;
     return regFile.readIntReg(reg_idx);
 }
 
@@ -1214,6 +1250,7 @@ template <class Impl>
 FloatReg
 FullO3CPU<Impl>::readFloatReg(int reg_idx)
 {
+    fpRegfileReads++;
     return regFile.readFloatReg(reg_idx);
 }
 
@@ -1221,6 +1258,7 @@ template <class Impl>
 FloatRegBits
 FullO3CPU<Impl>::readFloatRegBits(int reg_idx)
 {
+    fpRegfileReads++;
     return regFile.readFloatRegBits(reg_idx);
 }
 
@@ -1228,6 +1266,7 @@ template <class Impl>
 void
 FullO3CPU<Impl>::setIntReg(int reg_idx, uint64_t val)
 {
+    intRegfileWrites++;
     regFile.setIntReg(reg_idx, val);
 }
 
@@ -1235,6 +1274,7 @@ template <class Impl>
 void
 FullO3CPU<Impl>::setFloatReg(int reg_idx, FloatReg val)
 {
+    fpRegfileWrites++;
     regFile.setFloatReg(reg_idx, val);
 }
 
@@ -1242,6 +1282,7 @@ template <class Impl>
 void
 FullO3CPU<Impl>::setFloatRegBits(int reg_idx, FloatRegBits val)
 {
+    fpRegfileWrites++;
     regFile.setFloatRegBits(reg_idx, val);
 }
 
@@ -1249,6 +1290,7 @@ template <class Impl>
 uint64_t
 FullO3CPU<Impl>::readArchIntReg(int reg_idx, ThreadID tid)
 {
+    intRegfileReads++;
     PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
 
     return regFile.readIntReg(phys_reg);
@@ -1258,6 +1300,7 @@ template <class Impl>
 float
 FullO3CPU<Impl>::readArchFloatReg(int reg_idx, ThreadID tid)
 {
+    fpRegfileReads++;
     int idx = reg_idx + TheISA::NumIntRegs;
     PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
 
@@ -1268,6 +1311,7 @@ template <class Impl>
 uint64_t
 FullO3CPU<Impl>::readArchFloatRegInt(int reg_idx, ThreadID tid)
 {
+    fpRegfileReads++;
     int idx = reg_idx + TheISA::NumIntRegs;
     PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
 
@@ -1278,6 +1322,7 @@ template <class Impl>
 void
 FullO3CPU<Impl>::setArchIntReg(int reg_idx, uint64_t val, ThreadID tid)
 {
+    intRegfileWrites++;
     PhysRegIndex phys_reg = commitRenameMap[tid].lookup(reg_idx);
 
     regFile.setIntReg(phys_reg, val);
@@ -1287,6 +1332,7 @@ template <class Impl>
 void
 FullO3CPU<Impl>::setArchFloatReg(int reg_idx, float val, ThreadID tid)
 {
+    fpRegfileWrites++;
     int idx = reg_idx + TheISA::NumIntRegs;
     PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
 
@@ -1297,6 +1343,7 @@ template <class Impl>
 void
 FullO3CPU<Impl>::setArchFloatRegInt(int reg_idx, uint64_t val, ThreadID tid)
 {
+    fpRegfileWrites++;
     int idx = reg_idx + TheISA::NumIntRegs;
     PhysRegIndex phys_reg = commitRenameMap[tid].lookup(idx);
 
@@ -1364,9 +1411,10 @@ FullO3CPU<Impl>::instDone(ThreadID tid)
     thread[tid]->numInsts++;
     committedInsts[tid]++;
     totalCommittedInsts++;
-
+    system->totalNumInsts++;
     // Check for instruction-count-based events.
     comInstEventQueue[tid]->serviceEvents(thread[tid]->numInst);
+    system->instEventQueue.serviceEvents(system->totalNumInsts);
 }
 
 template <class Impl>
diff --git a/src/cpu/o3/cpu.hh b/src/cpu/o3/cpu.hh
index e3d13c840..69289996b 100644
--- a/src/cpu/o3/cpu.hh
+++ b/src/cpu/o3/cpu.hh
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2004-2005 The Regents of The University of Michigan
+ * Copyright (c) 2011 Regents of the University of California
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -27,6 +28,7 @@
  *
  * Authors: Kevin Lim
  *          Korey Sewell
+ *          Rick Strong
  */
 
 #ifndef __CPU_O3_CPU_HH__
@@ -726,6 +728,16 @@ class FullO3CPU : public BaseO3CPU
     Stats::Formula ipc;
     /** Stat for the total IPC. */
     Stats::Formula totalIpc;
+
+    //number of integer register file accesses
+    Stats::Scalar intRegfileReads;
+    Stats::Scalar intRegfileWrites;
+    //number of float register file accesses
+    Stats::Scalar fpRegfileReads;
+    Stats::Scalar fpRegfileWrites;
+    //number of misc
+    Stats::Scalar miscRegfileReads;
+    Stats::Scalar miscRegfileWrites;
 };
 
 #endif // __CPU_O3_CPU_HH__
diff --git a/src/cpu/o3/iew_impl.hh b/src/cpu/o3/iew_impl.hh
index ce58868ba..3f3761ff3 100644
--- a/src/cpu/o3/iew_impl.hh
+++ b/src/cpu/o3/iew_impl.hh
@@ -698,6 +698,7 @@ DefaultIEW<Impl>::updateStatus()
     // If there are no ready instructions waiting to be scheduled by the IQ,
     // and there's no stores waiting to write back, and dispatch is not
     // unblocking, then there is no internal activity for the IEW stage.
+    instQueue.intInstQueueReads++;
     if (_status == Active && !instQueue.hasReadyInsts() &&
         !ldstQueue.willWB() && !any_unblocking) {
         DPRINTF(IEW, "IEW switching to idle\n");
diff --git a/src/cpu/o3/inst_queue.hh b/src/cpu/o3/inst_queue.hh
index 56124d60f..be936e204 100644
--- a/src/cpu/o3/inst_queue.hh
+++ b/src/cpu/o3/inst_queue.hh
@@ -497,6 +497,16 @@ class InstructionQueue
     Stats::Vector fuBusy;
     /** Number of times the FU was busy per instruction issued. */
     Stats::Formula fuBusyRate;
+   public:
+    Stats::Scalar intInstQueueReads;
+    Stats::Scalar intInstQueueWrites;
+    Stats::Scalar intInstQueueWakeupAccesses;
+    Stats::Scalar fpInstQueueReads;
+    Stats::Scalar fpInstQueueWrites;
+    Stats::Scalar fpInstQueueWakeupQccesses;
+
+    Stats::Scalar intAluAccesses;
+    Stats::Scalar fpAluAccesses;
 };
 
 #endif //__CPU_O3_INST_QUEUE_HH__
diff --git a/src/cpu/o3/inst_queue_impl.hh b/src/cpu/o3/inst_queue_impl.hh
index ce408dfd0..91cb2f0c8 100644
--- a/src/cpu/o3/inst_queue_impl.hh
+++ b/src/cpu/o3/inst_queue_impl.hh
@@ -320,6 +320,47 @@ InstructionQueue<Impl>::regStats()
         // Tell mem dependence unit to reg stats as well.
         memDepUnit[tid].regStats();
     }
+
+    intInstQueueReads
+        .name(name() + ".int_inst_queue_reads")
+        .desc("Number of integer instruction queue reads")
+        .flags(total);
+
+    intInstQueueWrites
+        .name(name() + ".int_inst_queue_writes")
+        .desc("Number of integer instruction queue writes")
+        .flags(total);
+
+    intInstQueueWakeupAccesses
+        .name(name() + ".int_inst_queue_wakeup_accesses")
+        .desc("Number of integer instruction queue wakeup accesses")
+        .flags(total);
+
+    fpInstQueueReads
+        .name(name() + ".fp_inst_queue_reads")
+        .desc("Number of floating instruction queue reads")
+        .flags(total);
+
+    fpInstQueueWrites
+        .name(name() + ".fp_inst_queue_writes")
+        .desc("Number of floating instruction queue writes")
+        .flags(total);
+
+    fpInstQueueWakeupQccesses
+        .name(name() + ".fp_inst_queue_wakeup_accesses")
+        .desc("Number of floating instruction queue wakeup accesses")
+        .flags(total);
+
+    intAluAccesses
+        .name(name() + ".int_alu_accesses")
+        .desc("Number of integer alu accesses")
+        .flags(total);
+
+    fpAluAccesses
+        .name(name() + ".fp_alu_accesses")
+        .desc("Number of floating point alu accesses")
+        .flags(total);
+
 }
 
 template <class Impl>
@@ -501,6 +542,7 @@ template <class Impl>
 void
 InstructionQueue<Impl>::insert(DynInstPtr &new_inst)
 {
+    new_inst->isFloating() ? fpInstQueueWrites++ : intInstQueueWrites++;
     // Make sure the instruction is valid
     assert(new_inst);
 
@@ -542,6 +584,7 @@ InstructionQueue<Impl>::insertNonSpec(DynInstPtr &new_inst)
 {
     // @todo: Clean up this code; can do it by setting inst as unable
     // to issue, then calling normal insert on the inst.
+    new_inst->isFloating() ? fpInstQueueWrites++ : intInstQueueWrites++;
 
     assert(new_inst);
 
@@ -592,6 +635,11 @@ InstructionQueue<Impl>::getInstToExecute()
     assert(!instsToExecute.empty());
     DynInstPtr inst = instsToExecute.front();
     instsToExecute.pop_front();
+    if (inst->isFloating()){
+        fpInstQueueReads++;
+    } else {
+        intInstQueueReads++;
+    }
     return inst;
 }
 
@@ -706,6 +754,8 @@ InstructionQueue<Impl>::scheduleReadyInsts()
 
         DynInstPtr issuing_inst = readyInsts[op_class].top();
 
+        issuing_inst->isFloating() ? fpInstQueueReads++ : intInstQueueReads++;
+
         assert(issuing_inst->seqNum == (*order_it).oldestInst);
 
         if (issuing_inst->isSquashed()) {
@@ -731,7 +781,7 @@ InstructionQueue<Impl>::scheduleReadyInsts()
 
         if (op_class != No_OpClass) {
             idx = fuPool->getUnit(op_class);
-
+            issuing_inst->isFloating() ? fpAluAccesses++ : intAluAccesses++;
             if (idx > -1) {
                 op_latency = fuPool->getOpLatency(op_class);
             }
@@ -867,6 +917,13 @@ InstructionQueue<Impl>::wakeDependents(DynInstPtr &completed_inst)
 {
     int dependents = 0;
 
+    // The instruction queue here takes care of both floating and int ops
+    if (completed_inst->isFloating()) {
+        fpInstQueueWakeupQccesses++;
+    } else {
+        intInstQueueWakeupAccesses++;
+    }
+
     DPRINTF(IQ, "Waking dependents of completed instruction.\n");
 
     assert(!completed_inst->isSquashed());
@@ -997,6 +1054,7 @@ void
 InstructionQueue<Impl>::violation(DynInstPtr &store,
                                   DynInstPtr &faulting_load)
 {
+    intInstQueueWrites++;
     memDepUnit[store->threadNumber].violation(store, faulting_load);
 }
 
@@ -1037,6 +1095,7 @@ InstructionQueue<Impl>::doSquash(ThreadID tid)
            (*squash_it)->seqNum > squashedSeqNum[tid]) {
 
         DynInstPtr squashed_inst = (*squash_it);
+        squashed_inst->isFloating() ? fpInstQueueWrites++ : intInstQueueWrites++;
 
         // Only handle the instruction if it actually is in the IQ and
         // hasn't already been squashed in the IQ.
diff --git a/src/cpu/o3/rename.hh b/src/cpu/o3/rename.hh
index 4598a8d7b..901283111 100644
--- a/src/cpu/o3/rename.hh
+++ b/src/cpu/o3/rename.hh
@@ -470,6 +470,8 @@ class DefaultRename
     Stats::Scalar renameRenamedOperands;
     /** Stat for total number of source register rename lookups. */
     Stats::Scalar renameRenameLookups;
+    Stats::Scalar intRenameLookups;
+    Stats::Scalar fpRenameLookups;
     /** Stat for total number of committed renaming mappings. */
     Stats::Scalar renameCommittedMaps;
     /** Stat for total number of mappings that were undone due to a squash. */
diff --git a/src/cpu/o3/rename_impl.hh b/src/cpu/o3/rename_impl.hh
index ac2421dc7..1f34b7255 100644
--- a/src/cpu/o3/rename_impl.hh
+++ b/src/cpu/o3/rename_impl.hh
@@ -179,6 +179,14 @@ DefaultRename<Impl>::regStats()
         .desc("count of insts added to the skid buffer")
         .flags(Stats::total)
         ;
+    intRenameLookups
+        .name(name() + ".RENAME:int_rename_lookups")
+        .desc("Number of integer rename lookups")
+        .prereq(intRenameLookups);
+    fpRenameLookups
+        .name(name() + ".RENAME:fp_rename_lookups")
+        .desc("Number of floating rename lookups")
+        .prereq(fpRenameLookups);
 }
 
 template <class Impl>
@@ -1012,6 +1020,7 @@ DefaultRename<Impl>::renameSrcRegs(DynInstPtr &inst, ThreadID tid)
         }
 
         ++renameRenameLookups;
+        inst->isFloating() ? fpRenameLookups++ : intRenameLookups++;
     }
 }
 
diff --git a/src/cpu/o3/rob.hh b/src/cpu/o3/rob.hh
index bdea07d1a..510c8c5dc 100644
--- a/src/cpu/o3/rob.hh
+++ b/src/cpu/o3/rob.hh
@@ -253,6 +253,9 @@ class ROB
      */
     int countInsts(ThreadID tid);
 
+    /** Registers statistics. */
+    void regStats();
+
   private:
     /** Pointer to the CPU. */
     O3CPU *cpu;
@@ -312,6 +315,11 @@ class ROB
 
     /** Number of active threads. */
     ThreadID numThreads;
+
+    // The number of rob_reads
+    Stats::Scalar robReads;
+    // The number of rob_writes
+    Stats::Scalar robWrites;
 };
 
 #endif //__CPU_O3_ROB_HH__
diff --git a/src/cpu/o3/rob_impl.hh b/src/cpu/o3/rob_impl.hh
index 37f1c5504..d9d1daded 100644
--- a/src/cpu/o3/rob_impl.hh
+++ b/src/cpu/o3/rob_impl.hh
@@ -204,6 +204,8 @@ ROB<Impl>::insertInst(DynInstPtr &inst)
 {
     assert(inst);
 
+    robWrites++;
+
     DPRINTF(ROB, "Adding inst PC %s to the ROB.\n", inst->pcState());
 
     assert(numInstsInROB != numEntries);
@@ -237,6 +239,8 @@ template <class Impl>
 void
 ROB<Impl>::retireHead(ThreadID tid)
 {
+    robWrites++;
+
     assert(numInstsInROB > 0);
 
     // Get the head ROB instruction.
@@ -271,6 +275,7 @@ template <class Impl>
 bool
 ROB<Impl>::isHeadReady(ThreadID tid)
 {
+    robReads++;
     if (threadEntries[tid] != 0) {
         return instList[tid].front()->readyToCommit();
     }
@@ -315,6 +320,7 @@ template <class Impl>
 void
 ROB<Impl>::doSquash(ThreadID tid)
 {
+    robWrites++;
     DPRINTF(ROB, "[tid:%u]: Squashing instructions until [sn:%i].\n",
             tid, squashedSeqNum[tid]);
 
@@ -523,3 +529,17 @@ ROB<Impl>::readTailInst(ThreadID tid)
     return *tail_thread;
 }
 
+template <class Impl>
+void
+ROB<Impl>::regStats()
+{
+    using namespace Stats;
+    robReads
+        .name(name() + ".rob_reads")
+        .desc("The number of ROB reads");
+
+    robWrites
+        .name(name() + ".rob_writes")
+        .desc("The number of ROB writes");
+}
+