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, 151 insertions, 24 deletions

diff --git a/src/cpu/o3/decode.hh b/src/cpu/o3/decode.hh
index 5b9a0f822..3035b3387 100644
--- a/src/cpu/o3/decode.hh
+++ b/src/cpu/o3/decode.hh
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2004-2005 The Regents of The University of Michigan
+ * Copyright (c) 2004-2006 The Regents of The University of Michigan
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -26,16 +26,23 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#ifndef __CPU_O3_CPU_SIMPLE_DECODE_HH__
-#define __CPU_O3_CPU_SIMPLE_DECODE_HH__
+#ifndef __CPU_O3_DECODE_HH__
+#define __CPU_O3_DECODE_HH__
 
 #include <queue>
 
 #include "base/statistics.hh"
 #include "base/timebuf.hh"
 
+/**
+ * DefaultDecode class handles both single threaded and SMT
+ * decode. Its width is specified by the parameters; each cycles it
+ * tries to decode that many instructions. Because instructions are
+ * actually decoded when the StaticInst is created, this stage does
+ * not do much other than check any PC-relative branches.
+ */
 template<class Impl>
-class SimpleDecode
+class DefaultDecode
 {
   private:
     // Typedefs from the Impl.
@@ -50,49 +57,129 @@ class SimpleDecode
     typedef typename CPUPol::TimeStruct TimeStruct;
 
   public:
-    // The only time decode will become blocked is if dispatch becomes
-    // blocked, which means IQ or ROB is probably full.
-    enum Status {
+    /** Overall decode stage status. Used to determine if the CPU can
+     * deschedule itself due to a lack of activity.
+     */
+    enum DecodeStatus {
+        Active,
+        Inactive
+    };
+
+    /** Individual thread status. */
+    enum ThreadStatus {
         Running,
         Idle,
+        StartSquash,
         Squashing,
         Blocked,
         Unblocking
     };
 
   private:
-    // May eventually need statuses on a per thread basis.
-    Status _status;
+    /** Decode status. */
+    DecodeStatus _status;
+
+    /** Per-thread status. */
+    ThreadStatus decodeStatus[Impl::MaxThreads];
 
   public:
-    SimpleDecode(Params &params);
+    /** DefaultDecode constructor. */
+    DefaultDecode(Params *params);
 
+    /** Returns the name of decode. */
+    std::string name() const;
+
+    /** Registers statistics. */
     void regStats();
 
+    /** Sets CPU pointer. */
     void setCPU(FullCPU *cpu_ptr);
 
+    /** Sets the main backwards communication time buffer pointer. */
     void setTimeBuffer(TimeBuffer<TimeStruct> *tb_ptr);
 
+    /** Sets pointer to time buffer used to communicate to the next stage. */
     void setDecodeQueue(TimeBuffer<DecodeStruct> *dq_ptr);
 
+    /** Sets pointer to time buffer coming from fetch. */
     void setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr);
 
+    /** Sets pointer to list of active threads. */
+    void setActiveThreads(std::list<unsigned> *at_ptr);
+
+    void switchOut();
+
+    void takeOverFrom();
+    /** Ticks decode, processing all input signals and decoding as many
+     * instructions as possible.
+     */
     void tick();
 
-    void decode();
+    /** Determines what to do based on decode's current status.
+     * @param status_change decode() sets this variable if there was a status
+     * change (ie switching from from blocking to unblocking).
+     * @param tid Thread id to decode instructions from.
+     */
+    void decode(bool &status_change, unsigned tid);
+
+    /** Processes instructions from fetch and passes them on to rename.
+     * Decoding of instructions actually happens when they are created in
+     * fetch, so this function mostly checks if PC-relative branches are
+     * correct.
+     */
+    void decodeInsts(unsigned tid);
 
   private:
+    /** Inserts a thread's instructions into the skid buffer, to be decoded
+     * once decode unblocks.
+     */
+    void skidInsert(unsigned tid);
+
+    /** Returns if all of the skid buffers are empty. */
+    bool skidsEmpty();
+
+    /** Updates overall decode status based on all of the threads' statuses. */
+    void updateStatus();
+
+    /** Separates instructions from fetch into individual lists of instructions
+     * sorted by thread.
+     */
+    void sortInsts();
+
+    /** Reads all stall signals from the backwards communication timebuffer. */
+    void readStallSignals(unsigned tid);
+
+    /** Checks all input signals and updates decode's status appropriately. */
+    bool checkSignalsAndUpdate(unsigned tid);
+
+    /** Checks all stall signals, and returns if any are true. */
+    bool checkStall(unsigned tid) const;
+
+    /** Returns if there any instructions from fetch on this cycle. */
     inline bool fetchInstsValid();
 
-    void block();
+    /** Switches decode to blocking, and signals back that decode has
+     * become blocked.
+     * @return Returns true if there is a status change.
+     */
+    bool block(unsigned tid);
 
-    inline void unblock();
+    /** Switches decode to unblocking if the skid buffer is empty, and
+     * signals back that decode has unblocked.
+     * @return Returns true if there is a status change.
+     */
+    bool unblock(unsigned tid);
 
-    void squash(DynInstPtr &inst);
+    /** Squashes if there is a PC-relative branch that was predicted
+     * incorrectly. Sends squash information back to fetch.
+     */
+    void squash(DynInstPtr &inst, unsigned tid);
 
   public:
-    // Might want to make squash a friend function.
-    void squash();
+    /** Squashes due to commit signalling a squash. Changes status to
+     * squashing and clears block/unblock signals as needed.
+     */
+    unsigned squash(unsigned tid);
 
   private:
     // Interfaces to objects outside of decode.
@@ -127,10 +214,27 @@ class SimpleDecode
     /** Wire to get fetch's output from fetch queue. */
     typename TimeBuffer<FetchStruct>::wire fromFetch;
 
+    /** Queue of all instructions coming from fetch this cycle. */
+    std::queue<DynInstPtr> insts[Impl::MaxThreads];
+
     /** Skid buffer between fetch and decode. */
-    std::queue<FetchStruct> skidBuffer;
+    std::queue<DynInstPtr> skidBuffer[Impl::MaxThreads];
+
+    /** Variable that tracks if decode has written to the time buffer this
+     * cycle. Used to tell CPU if there is activity this cycle.
+     */
+    bool wroteToTimeBuffer;
+
+    /** Source of possible stalls. */
+    struct Stalls {
+        bool rename;
+        bool iew;
+        bool commit;
+    };
+
+    /** Tracks which stages are telling decode to stall. */
+    Stalls stalls[Impl::MaxThreads];
 
-    //Consider making these unsigned to avoid any confusion.
     /** Rename to decode delay, in ticks. */
     unsigned renameToDecodeDelay;
 
@@ -146,20 +250,43 @@ class SimpleDecode
     /** The width of decode, in instructions. */
     unsigned decodeWidth;
 
-    /** The instruction that decode is currently on.  It needs to have
-     *  persistent state so that when a stall occurs in the middle of a
-     *  group of instructions, it can restart at the proper instruction.
-     */
-    unsigned numInst;
+    /** Index of instructions being sent to rename. */
+    unsigned toRenameIndex;
+
+    /** number of Active Threads*/
+    unsigned numThreads;
 
+    /** List of active thread ids */
+    std::list<unsigned> *activeThreads;
+
+    /** Number of branches in flight. */
+    unsigned branchCount[Impl::MaxThreads];
+
+    /** Maximum size of the skid buffer. */
+    unsigned skidBufferMax;
+
+    /** Stat for total number of idle cycles. */
     Stats::Scalar<> decodeIdleCycles;
+    /** Stat for total number of blocked cycles. */
     Stats::Scalar<> decodeBlockedCycles;
+    /** Stat for total number of normal running cycles. */
+    Stats::Scalar<> decodeRunCycles;
+    /** Stat for total number of unblocking cycles. */
     Stats::Scalar<> decodeUnblockCycles;
+    /** Stat for total number of squashing cycles. */
     Stats::Scalar<> decodeSquashCycles;
+    /** Stat for number of times a branch is resolved at decode. */
+    Stats::Scalar<> decodeBranchResolved;
+    /** Stat for number of times a branch mispredict is detected. */
     Stats::Scalar<> decodeBranchMispred;
+    /** Stat for number of times decode detected a non-control instruction
+     * incorrectly predicted as a branch.
+     */
     Stats::Scalar<> decodeControlMispred;
+    /** Stat for total number of decoded instructions. */
     Stats::Scalar<> decodeDecodedInsts;
+    /** Stat for total number of squashed instructions. */
     Stats::Scalar<> decodeSquashedInsts;
 };
 
-#endif // __CPU_O3_CPU_SIMPLE_DECODE_HH__
+#endif // __CPU_O3_DECODE_HH__