Merge ktlim@zizzer:/bk/m5

into  zamp.eecs.umich.edu:/z/ktlim2/clean/newmem

SConstruct:
src/SConscript:
src/arch/SConscript:
src/arch/alpha/faults.cc:
src/arch/alpha/tlb.cc:
src/base/traceflags.py:
src/cpu/SConscript:
src/cpu/base.cc:
src/cpu/base.hh:
src/cpu/base_dyn_inst.cc:
src/cpu/cpu_exec_context.cc:
src/cpu/cpu_exec_context.hh:
src/cpu/exec_context.hh:
src/cpu/o3/alpha_cpu.hh:
src/cpu/o3/alpha_cpu_impl.hh:
src/cpu/o3/alpha_dyn_inst.hh:
src/cpu/o3/cpu.cc:
src/cpu/o3/cpu.hh:
src/cpu/o3/regfile.hh:
src/cpu/ozone/cpu.hh:
src/cpu/simple/base.cc:
src/cpu/base_dyn_inst.hh:
src/cpu/o3/2bit_local_pred.cc:
src/cpu/o3/2bit_local_pred.hh:
src/cpu/o3/alpha_cpu.cc:
src/cpu/o3/alpha_cpu_builder.cc:
src/cpu/o3/alpha_dyn_inst.cc:
src/cpu/o3/alpha_dyn_inst_impl.hh:
src/cpu/o3/alpha_impl.hh:
src/cpu/o3/alpha_params.hh:
src/cpu/o3/bpred_unit.cc:
src/cpu/o3/bpred_unit.hh:
src/cpu/o3/bpred_unit_impl.hh:
src/cpu/o3/btb.cc:
src/cpu/o3/btb.hh:
src/cpu/o3/comm.hh:
src/cpu/o3/commit.cc:
src/cpu/o3/commit.hh:
src/cpu/o3/commit_impl.hh:
src/cpu/o3/cpu_policy.hh:
src/cpu/o3/decode.cc:
src/cpu/o3/decode.hh:
src/cpu/o3/decode_impl.hh:
src/cpu/o3/fetch.cc:
src/cpu/o3/fetch.hh:
src/cpu/o3/fetch_impl.hh:
src/cpu/o3/free_list.cc:
src/cpu/o3/free_list.hh:
src/cpu/o3/iew.cc:
src/cpu/o3/iew.hh:
src/cpu/o3/iew_impl.hh:
src/cpu/o3/inst_queue.cc:
src/cpu/o3/inst_queue.hh:
src/cpu/o3/inst_queue_impl.hh:
src/cpu/o3/mem_dep_unit.cc:
src/cpu/o3/mem_dep_unit.hh:
src/cpu/o3/mem_dep_unit_impl.hh:
src/cpu/o3/ras.cc:
src/cpu/o3/ras.hh:
src/cpu/o3/rename.cc:
src/cpu/o3/rename.hh:
src/cpu/o3/rename_impl.hh:
src/cpu/o3/rename_map.cc:
src/cpu/o3/rename_map.hh:
src/cpu/o3/rob.cc:
src/cpu/o3/rob.hh:
src/cpu/o3/rob_impl.hh:
src/cpu/o3/sat_counter.cc:
src/cpu/o3/sat_counter.hh:
src/cpu/o3/store_set.cc:
src/cpu/o3/store_set.hh:
src/cpu/o3/tournament_pred.cc:
src/cpu/o3/tournament_pred.hh:
    Hand merges.

--HG--
rename : build/SConstruct => SConstruct
rename : SConscript => src/SConscript
rename : arch/alpha/ev5.cc => src/arch/alpha/ev5.cc
rename : arch/alpha/isa/decoder.isa => src/arch/alpha/isa/decoder.isa
rename : arch/alpha/isa/pal.isa => src/arch/alpha/isa/pal.isa
rename : base/traceflags.py => src/base/traceflags.py
rename : cpu/SConscript => src/cpu/SConscript
rename : cpu/base.cc => src/cpu/base.cc
rename : cpu/base.hh => src/cpu/base.hh
rename : cpu/base_dyn_inst.cc => src/cpu/base_dyn_inst.cc
rename : cpu/base_dyn_inst.hh => src/cpu/base_dyn_inst.hh
rename : cpu/cpu_exec_context.cc => src/cpu/cpu_exec_context.cc
rename : cpu/cpu_exec_context.hh => src/cpu/cpu_exec_context.hh
rename : cpu/cpu_models.py => src/cpu/cpu_models.py
rename : cpu/exec_context.hh => src/cpu/exec_context.hh
rename : cpu/exetrace.cc => src/cpu/exetrace.cc
rename : cpu/exetrace.hh => src/cpu/exetrace.hh
rename : cpu/inst_seq.hh => src/cpu/inst_seq.hh
rename : cpu/o3/2bit_local_pred.cc => src/cpu/o3/2bit_local_pred.cc
rename : cpu/o3/2bit_local_pred.hh => src/cpu/o3/2bit_local_pred.hh
rename : cpu/o3/alpha_cpu.hh => src/cpu/o3/alpha_cpu.hh
rename : cpu/o3/alpha_cpu_builder.cc => src/cpu/o3/alpha_cpu_builder.cc
rename : cpu/o3/alpha_cpu_impl.hh => src/cpu/o3/alpha_cpu_impl.hh
rename : cpu/o3/alpha_dyn_inst.hh => src/cpu/o3/alpha_dyn_inst.hh
rename : cpu/o3/alpha_dyn_inst_impl.hh => src/cpu/o3/alpha_dyn_inst_impl.hh
rename : cpu/o3/alpha_impl.hh => src/cpu/o3/alpha_impl.hh
rename : cpu/o3/alpha_params.hh => src/cpu/o3/alpha_params.hh
rename : cpu/o3/bpred_unit.cc => src/cpu/o3/bpred_unit.cc
rename : cpu/o3/bpred_unit.hh => src/cpu/o3/bpred_unit.hh
rename : cpu/o3/bpred_unit_impl.hh => src/cpu/o3/bpred_unit_impl.hh
rename : cpu/o3/btb.cc => src/cpu/o3/btb.cc
rename : cpu/o3/btb.hh => src/cpu/o3/btb.hh
rename : cpu/o3/comm.hh => src/cpu/o3/comm.hh
rename : cpu/o3/commit.cc => src/cpu/o3/commit.cc
rename : cpu/o3/commit.hh => src/cpu/o3/commit.hh
rename : cpu/o3/commit_impl.hh => src/cpu/o3/commit_impl.hh
rename : cpu/o3/cpu.cc => src/cpu/o3/cpu.cc
rename : cpu/o3/cpu.hh => src/cpu/o3/cpu.hh
rename : cpu/o3/cpu_policy.hh => src/cpu/o3/cpu_policy.hh
rename : cpu/o3/decode.cc => src/cpu/o3/decode.cc
rename : cpu/o3/decode.hh => src/cpu/o3/decode.hh
rename : cpu/o3/decode_impl.hh => src/cpu/o3/decode_impl.hh
rename : cpu/o3/fetch.cc => src/cpu/o3/fetch.cc
rename : cpu/o3/fetch.hh => src/cpu/o3/fetch.hh
rename : cpu/o3/fetch_impl.hh => src/cpu/o3/fetch_impl.hh
rename : cpu/o3/free_list.cc => src/cpu/o3/free_list.cc
rename : cpu/o3/free_list.hh => src/cpu/o3/free_list.hh
rename : cpu/o3/iew.cc => src/cpu/o3/iew.cc
rename : cpu/o3/iew.hh => src/cpu/o3/iew.hh
rename : cpu/o3/iew_impl.hh => src/cpu/o3/iew_impl.hh
rename : cpu/o3/inst_queue.cc => src/cpu/o3/inst_queue.cc
rename : cpu/o3/inst_queue.hh => src/cpu/o3/inst_queue.hh
rename : cpu/o3/inst_queue_impl.hh => src/cpu/o3/inst_queue_impl.hh
rename : cpu/o3/mem_dep_unit.cc => src/cpu/o3/mem_dep_unit.cc
rename : cpu/o3/mem_dep_unit.hh => src/cpu/o3/mem_dep_unit.hh
rename : cpu/o3/mem_dep_unit_impl.hh => src/cpu/o3/mem_dep_unit_impl.hh
rename : cpu/o3/ras.cc => src/cpu/o3/ras.cc
rename : cpu/o3/ras.hh => src/cpu/o3/ras.hh
rename : cpu/o3/regfile.hh => src/cpu/o3/regfile.hh
rename : cpu/o3/rename.cc => src/cpu/o3/rename.cc
rename : cpu/o3/rename.hh => src/cpu/o3/rename.hh
rename : cpu/o3/rename_impl.hh => src/cpu/o3/rename_impl.hh
rename : cpu/o3/rename_map.cc => src/cpu/o3/rename_map.cc
rename : cpu/o3/rename_map.hh => src/cpu/o3/rename_map.hh
rename : cpu/o3/rob.hh => src/cpu/o3/rob.hh
rename : cpu/o3/rob_impl.hh => src/cpu/o3/rob_impl.hh
rename : cpu/o3/sat_counter.hh => src/cpu/o3/sat_counter.hh
rename : cpu/o3/store_set.cc => src/cpu/o3/store_set.cc
rename : cpu/o3/store_set.hh => src/cpu/o3/store_set.hh
rename : cpu/o3/tournament_pred.cc => src/cpu/o3/tournament_pred.cc
rename : cpu/o3/tournament_pred.hh => src/cpu/o3/tournament_pred.hh
rename : cpu/ozone/cpu.cc => src/cpu/ozone/cpu.cc
rename : cpu/ozone/cpu.hh => src/cpu/ozone/cpu.hh
rename : cpu/ozone/cpu_impl.hh => src/cpu/ozone/cpu_impl.hh
rename : cpu/static_inst.hh => src/cpu/static_inst.hh
rename : kern/system_events.cc => src/kern/system_events.cc
rename : kern/tru64/tru64.hh => src/kern/tru64/tru64.hh
rename : python/m5/objects/AlphaFullCPU.py => src/python/m5/objects/AlphaFullCPU.py
rename : sim/pseudo_inst.cc => src/sim/pseudo_inst.cc
extra : convert_revision : ff351fc0e3a7c0f23e59fdbec33d8209eb9280be

1 files changed, 127 insertions, 34 deletions

diff --git a/src/cpu/o3/bpred_unit.hh b/src/cpu/o3/bpred_unit.hh
index 2725684f7..b7814b2e9 100644
--- a/src/cpu/o3/bpred_unit.hh
+++ b/src/cpu/o3/bpred_unit.hh
@@ -26,8 +26,8 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#ifndef __BPRED_UNIT_HH__
-#define __BPRED_UNIT_HH__
+#ifndef __CPU_O3_BPRED_UNIT_HH__
+#define __CPU_O3_BPRED_UNIT_HH__
 
 // For Addr type.
 #include "arch/isa_traits.hh"
@@ -35,20 +35,15 @@
 #include "cpu/inst_seq.hh"
 
 #include "cpu/o3/2bit_local_pred.hh"
-#include "cpu/o3/tournament_pred.hh"
 #include "cpu/o3/btb.hh"
 #include "cpu/o3/ras.hh"
+#include "cpu/o3/tournament_pred.hh"
 
 #include <list>
 
 /**
  * Basically a wrapper class to hold both the branch predictor
- * and the BTB.  Right now I'm unsure of the implementation; it would
- * be nicer to have something closer to the CPUPolicy or the Impl where
- * this is just typedefs, but it forces the upper level stages to be
- * aware of the constructors of the BP and the BTB.  The nicer thing
- * to do is have this templated on the Impl, accept the usual Params
- * object, and be able to call the constructors on the BP and BTB.
+ * and the BTB.
  */
 template<class Impl>
 class TwobitBPredUnit
@@ -57,77 +52,175 @@ class TwobitBPredUnit
     typedef typename Impl::Params Params;
     typedef typename Impl::DynInstPtr DynInstPtr;
 
-    TwobitBPredUnit(Params &params);
+    /**
+     * @param params The params object, that has the size of the BP and BTB.
+     */
+    TwobitBPredUnit(Params *params);
 
+    /**
+     * Registers statistics.
+     */
     void regStats();
 
-    bool predict(DynInstPtr &inst, Addr &PC);
-
-    void update(const InstSeqNum &done_sn);
-
-    void squash(const InstSeqNum &squashed_sn);
-
+    void switchOut();
+
+    void takeOverFrom();
+
+    /**
+     * Predicts whether or not the instruction is a taken branch, and the
+     * target of the branch if it is taken.
+     * @param inst The branch instruction.
+     * @param PC The predicted PC is passed back through this parameter.
+     * @param tid The thread id.
+     * @return Returns if the branch is taken or not.
+     */
+    bool predict(DynInstPtr &inst, Addr &PC, unsigned tid);
+
+    /**
+     * Tells the branch predictor to commit any updates until the given
+     * sequence number.
+     * @param done_sn The sequence number to commit any older updates up until.
+     * @param tid The thread id.
+     */
+    void update(const InstSeqNum &done_sn, unsigned tid);
+
+    /**
+     * Squashes all outstanding updates until a given sequence number.
+     * @param squashed_sn The sequence number to squash any younger updates up
+     * until.
+     * @param tid The thread id.
+     */
+    void squash(const InstSeqNum &squashed_sn, unsigned tid);
+
+    /**
+     * Squashes all outstanding updates until a given sequence number, and
+     * corrects that sn's update with the proper address and taken/not taken.
+     * @param squashed_sn The sequence number to squash any younger updates up
+     * until.
+     * @param corr_target The correct branch target.
+     * @param actually_taken The correct branch direction.
+     * @param tid The thread id.
+     */
     void squash(const InstSeqNum &squashed_sn, const Addr &corr_target,
-                bool actually_taken);
+                bool actually_taken, unsigned tid);
 
+    /**
+     * Looks up a given PC in the BP to see if it is taken or not taken.
+     * @param inst_PC The PC to look up.
+     * @return Whether the branch is taken or not taken.
+     */
     bool BPLookup(Addr &inst_PC)
     { return BP.lookup(inst_PC); }
 
+    /**
+     * Looks up a given PC in the BTB to see if a matching entry exists.
+     * @param inst_PC The PC to look up.
+     * @return Whether the BTB contains the given PC.
+     */
     bool BTBValid(Addr &inst_PC)
-    { return BTB.valid(inst_PC); }
+    { return BTB.valid(inst_PC, 0); }
 
+    /**
+     * Looks up a given PC in the BTB to get the predicted target.
+     * @param inst_PC The PC to look up.
+     * @return The address of the target of the branch.
+     */
     Addr BTBLookup(Addr &inst_PC)
-    { return BTB.lookup(inst_PC); }
-
-    // Will want to include global history.
+    { return BTB.lookup(inst_PC, 0); }
+
+    /**
+     * Updates the BP with taken/not taken information.
+     * @param inst_PC The branch's PC that will be updated.
+     * @param taken Whether the branch was taken or not taken.
+     * @todo Make this update flexible enough to handle a global predictor.
+     */
     void BPUpdate(Addr &inst_PC, bool taken)
     { BP.update(inst_PC, taken); }
 
+    /**
+     * Updates the BTB with the target of a branch.
+     * @param inst_PC The branch's PC that will be updated.
+     * @param target_PC The branch's target that will be added to the BTB.
+     */
     void BTBUpdate(Addr &inst_PC, Addr &target_PC)
-    { BTB.update(inst_PC, target_PC); }
+    { BTB.update(inst_PC, target_PC,0); }
 
   private:
     struct PredictorHistory {
+        /**
+         * Makes a predictor history struct that contains a sequence number,
+         * the PC of its instruction, and whether or not it was predicted
+         * taken.
+         */
         PredictorHistory(const InstSeqNum &seq_num, const Addr &inst_PC,
-                         const bool pred_taken)
-            : seqNum(seq_num), PC(inst_PC), predTaken(pred_taken),
-              globalHistory(0), usedRAS(0), wasCall(0), RASIndex(0),
-              RASTarget(0)
+                         const bool pred_taken, const unsigned _tid)
+            : seqNum(seq_num), PC(inst_PC), RASTarget(0), globalHistory(0),
+              RASIndex(0), tid(_tid), predTaken(pred_taken), usedRAS(0),
+              wasCall(0)
         { }
 
+        /** The sequence number for the predictor history entry. */
         InstSeqNum seqNum;
 
+        /** The PC associated with the sequence number. */
         Addr PC;
 
-        bool predTaken;
+        /** The RAS target (only valid if a return). */
+        Addr RASTarget;
 
+        /** The global history at the time this entry was created. */
         unsigned globalHistory;
 
-        bool usedRAS;
+        /** The RAS index of the instruction (only valid if a call). */
+        unsigned RASIndex;
 
-        bool wasCall;
+        /** The thread id. */
+        unsigned tid;
 
-        unsigned RASIndex;
+        /** Whether or not it was predicted taken. */
+        bool predTaken;
 
-        Addr RASTarget;
+        /** Whether or not the RAS was used. */
+        bool usedRAS;
+
+        /** Whether or not the instruction was a call. */
+        bool wasCall;
     };
 
-    std::list<PredictorHistory> predHist;
+    typedef std::list<PredictorHistory> History;
+
+    /**
+     * The per-thread predictor history. This is used to update the predictor
+     * as instructions are committed, or restore it to the proper state after
+     * a squash.
+     */
+    History predHist[Impl::MaxThreads];
 
+    /** The branch predictor. */
     DefaultBP BP;
 
+    /** The BTB. */
     DefaultBTB BTB;
 
-    ReturnAddrStack RAS;
+    /** The per-thread return address stack. */
+    ReturnAddrStack RAS[Impl::MaxThreads];
 
+    /** Stat for number of BP lookups. */
     Stats::Scalar<> lookups;
+    /** Stat for number of conditional branches predicted. */
     Stats::Scalar<> condPredicted;
+    /** Stat for number of conditional branches predicted incorrectly. */
     Stats::Scalar<> condIncorrect;
+    /** Stat for number of BTB lookups. */
     Stats::Scalar<> BTBLookups;
+    /** Stat for number of BTB hits. */
     Stats::Scalar<> BTBHits;
+    /** Stat for number of times the BTB is correct. */
     Stats::Scalar<> BTBCorrect;
+    /** Stat for number of times the RAS is used to get a target. */
     Stats::Scalar<> usedRAS;
+    /** Stat for number of times the RAS is incorrect. */
     Stats::Scalar<> RASIncorrect;
 };
 
-#endif // __BPRED_UNIT_HH__
+#endif // __CPU_O3_BPRED_UNIT_HH__