m5: added work completed monitoring support

9 files changed, 204 insertions, 1 deletions

diff --git a/src/arch/x86/isa/decoder/two_byte_opcodes.isa b/src/arch/x86/isa/decoder/two_byte_opcodes.isa
index 1dc32122e..11cd6eeb1 100644
--- a/src/arch/x86/isa/decoder/two_byte_opcodes.isa
+++ b/src/arch/x86/isa/decoder/two_byte_opcodes.isa
@@ -216,6 +216,12 @@
                         0x59: m5reserved5({{
                             warn("M5 reserved opcode 5 ignored.\n");
                         }}, IsNonSpeculative);
+                        0x5a: m5_work_begin({{
+                            PseudoInst::workbegin(xc->tcBase(), Rdi, Rsi);
+                        }}, IsNonSpeculative);
+                        0x5b: m5_work_end({{
+                            PseudoInst::workend(xc->tcBase(), Rdi, Rsi);
+                        }}, IsNonSpeculative);
                         default: Inst::UD2();
                     }
                 }
diff --git a/src/cpu/base.cc b/src/cpu/base.cc
index b7decaec1..17ae65793 100644
--- a/src/cpu/base.cc
+++ b/src/cpu/base.cc
@@ -255,6 +255,16 @@ BaseCPU::regStats()
         .desc("number of cpu cycles simulated")
         ;
 
+    numWorkItemsStarted
+        .name(name() + ".numWorkItemsStarted")
+        .desc("number of work items this cpu started")
+        ;
+
+    numWorkItemsCompleted
+        .name(name() + ".numWorkItemsCompleted")
+        .desc("number of work items this cpu completed")
+        ;
+
     int size = threadContexts.size();
     if (size > 1) {
         for (int i = 0; i < size; ++i) {
diff --git a/src/cpu/base.hh b/src/cpu/base.hh
index bea15aa08..a56f3db24 100644
--- a/src/cpu/base.hh
+++ b/src/cpu/base.hh
@@ -104,6 +104,8 @@ class BaseCPU : public MemObject
     inline Tick ticks(int numCycles) const { return clock * numCycles; }
     inline Tick curCycle() const { return curTick() / clock; }
     inline Tick tickToCycles(Tick val) const { return val / clock; }
+    inline void workItemBegin() { numWorkItemsStarted++; }
+    inline void workItemEnd() { numWorkItemsCompleted++; }
     // @todo remove me after debugging with legion done
     Tick instCount() { return instCnt; }
 
@@ -317,6 +319,8 @@ class BaseCPU : public MemObject
   public:
     // Number of CPU cycles simulated
     Stats::Scalar numCycles;
+    Stats::Scalar numWorkItemsStarted;
+    Stats::Scalar numWorkItemsCompleted;
 };
 
 #endif // __CPU_BASE_HH__
diff --git a/src/sim/SConscript b/src/sim/SConscript
index f959870cd..5c22e328e 100644
--- a/src/sim/SConscript
+++ b/src/sim/SConscript
@@ -78,3 +78,4 @@ TraceFlag('TLB')
 TraceFlag('Thread')
 TraceFlag('Timer')
 TraceFlag('VtoPhys')
+TraceFlag('WorkItems')
diff --git a/src/sim/System.py b/src/sim/System.py
index e2d5b279b..fd707c353 100644
--- a/src/sim/System.py
+++ b/src/sim/System.py
@@ -47,6 +47,20 @@ class System(SimObject):
     physmem = Param.PhysicalMemory(Parent.any, "physical memory")
     mem_mode = Param.MemoryMode('atomic', "The mode the memory system is in")
 
+    work_item_id = Param.Int(-1, "specific work item id")
+    work_begin_cpu_id_exit = Param.Int(-1,
+        "work started on specific id, now exit simulation")
+    work_begin_ckpt_count = Param.Counter(0,
+        "create checkpoint when work items begin count value is reached")
+    work_begin_exit_count = Param.Counter(0,
+        "exit simulation when work items begin count value is reached")
+    work_end_ckpt_count = Param.Counter(0,
+        "create checkpoint when work items end count value is reached")
+    work_end_exit_count = Param.Counter(0,
+        "exit simulation when work items end count value is reached")
+    work_cpus_ckpt_count = Param.Counter(0,
+        "create checkpoint when active cpu count value is reached")
+
     if buildEnv['FULL_SYSTEM']:
         abstract = True
         boot_cpu_frequency = Param.Frequency(Self.cpu[0].clock.frequency,
diff --git a/src/sim/pseudo_inst.cc b/src/sim/pseudo_inst.cc
index 9ee473014..f3b10f6d2 100644
--- a/src/sim/pseudo_inst.cc
+++ b/src/sim/pseudo_inst.cc
@@ -328,4 +328,121 @@ switchcpu(ThreadContext *tc)
     exitSimLoop("switchcpu");
 }
 
+//
+// This function is executed when annotated work items begin.  Depending on 
+// what the user specified at the command line, the simulation may exit and/or
+// take a checkpoint when a certain work item begins.
+//
+void
+workbegin(ThreadContext *tc, uint64_t workid, uint64_t threadid)
+{
+    tc->getCpuPtr()->workItemBegin();
+    System *sys = tc->getSystemPtr();
+
+    DPRINTF(WorkItems, "Work Begin workid: %d, threadid %d\n", workid, 
+            threadid);
+
+    //
+    // If specified, determine if this is the specific work item the user
+    // identified
+    //
+    if (sys->params()->work_item_id == -1 || 
+        sys->params()->work_item_id == workid) {
+
+        uint64_t systemWorkBeginCount = sys->incWorkItemsBegin();
+        int cpuId = tc->getCpuPtr()->cpuId();
+
+        if (sys->params()->work_cpus_ckpt_count != 0 &&
+            sys->markWorkItem(cpuId) >= sys->params()->work_cpus_ckpt_count) {
+            //
+            // If active cpus equals checkpoint count, create checkpoint
+            //
+            Event *event = new SimLoopExitEvent("checkpoint", 0);
+            mainEventQueue.schedule(event, curTick());
+        }
+
+        if (systemWorkBeginCount == sys->params()->work_begin_ckpt_count) {
+            //
+            // Note: the string specified as the cause of the exit event must
+            // exactly equal "checkpoint" inorder to create a checkpoint
+            //
+            Event *event = new SimLoopExitEvent("checkpoint", 0);
+            mainEventQueue.schedule(event, curTick());
+        }
+
+        if (systemWorkBeginCount == sys->params()->work_begin_exit_count) {
+            //
+            // If a certain number of work items started, exit simulation
+            //
+            Event *event = new SimLoopExitEvent("work started count reach", 0);
+            mainEventQueue.schedule(event, curTick());
+        }
+
+        if (tc->getCpuPtr()->cpuId() == sys->params()->work_begin_cpu_id_exit) {
+            //
+            // If work started on the specific cpu id specified, exit simulation
+            //
+            Event *event = new SimLoopExitEvent("work started on specific cpu",
+                                                0);
+
+            mainEventQueue.schedule(event, curTick() + 1);
+        }
+    }
+}
+
+//
+// This function is executed when annotated work items end.  Depending on 
+// what the user specified at the command line, the simulation may exit and/or
+// take a checkpoint when a certain work item ends.
+//
+void
+workend(ThreadContext *tc, uint64_t workid, uint64_t threadid)
+{
+    tc->getCpuPtr()->workItemEnd();
+    System *sys = tc->getSystemPtr();
+
+    DPRINTF(WorkItems, "Work End workid: %d, threadid %d\n", workid, threadid);
+
+    //
+    // If specified, determine if this is the specific work item the user
+    // identified
+    //
+    if (sys->params()->work_item_id == -1 || 
+        sys->params()->work_item_id == workid) {
+
+        uint64_t systemWorkEndCount = sys->incWorkItemsEnd();
+        int cpuId = tc->getCpuPtr()->cpuId();
+
+        if (sys->params()->work_cpus_ckpt_count != 0 &&
+            sys->markWorkItem(cpuId) >= sys->params()->work_cpus_ckpt_count) {
+            //
+            // If active cpus equals checkpoint count, create checkpoint
+            //
+            Event *event = new SimLoopExitEvent("checkpoint", 0);
+            mainEventQueue.schedule(event, curTick());
+        }
+
+        if (sys->params()->work_end_ckpt_count != 0 &&
+            systemWorkEndCount == sys->params()->work_end_ckpt_count) {
+            //
+            // If total work items completed equals checkpoint count, create
+            // checkpoint
+            //
+            Event *event = new SimLoopExitEvent("checkpoint", 0);
+            mainEventQueue.schedule(event, curTick());
+        }
+
+        if (sys->params()->work_end_exit_count != 0 &&
+            systemWorkEndCount == sys->params()->work_end_exit_count) {
+            //
+            // If total work items completed equals exit count, exit simulation
+            //
+            Event *event = new SimLoopExitEvent("work items exit count reached",
+                                                0);
+
+            mainEventQueue.schedule(event, curTick());
+        }
+    }
+}
+
 } // namespace PseudoInst
diff --git a/src/sim/pseudo_inst.hh b/src/sim/pseudo_inst.hh
index 2f9671155..296b1556b 100644
--- a/src/sim/pseudo_inst.hh
+++ b/src/sim/pseudo_inst.hh
@@ -63,5 +63,7 @@ void dumpresetstats(ThreadContext *tc, Tick delay, Tick period);
 void m5checkpoint(ThreadContext *tc, Tick delay, Tick period);
 void debugbreak(ThreadContext *tc);
 void switchcpu(ThreadContext *tc);
+void workbegin(ThreadContext *tc, uint64_t workid, uint64_t threadid);
+void workend(ThreadContext *tc, uint64_t workid, uint64_t threadid);
 
 } // namespace PseudoInst
diff --git a/src/sim/system.cc b/src/sim/system.cc
index 68b02272e..e9fd727f1 100644
--- a/src/sim/system.cc
+++ b/src/sim/system.cc
@@ -72,7 +72,10 @@ System::System(Params *p)
       pagePtr(0),
       nextPID(0),
 #endif
-      memoryMode(p->mem_mode), _params(p),
+      memoryMode(p->mem_mode),
+      workItemsBegin(0),
+      workItemsEnd(0),
+      _params(p),
       totalNumInsts(0), 
       instEventQueue("system instruction-based event queue")
 {
@@ -144,6 +147,8 @@ System::System(Params *p)
 
     // increment the number of running systms
     numSystemsRunning++;
+
+    activeCpus.clear();
 }
 
 System::~System()
@@ -218,6 +223,8 @@ System::registerThreadContext(ThreadContext *tc, int assigned)
         remoteGDB[id] = rgdb;
     }
 
+    activeCpus.push_back(false);
+
     return id;
 }
 
diff --git a/src/sim/system.hh b/src/sim/system.hh
index 6c4f3e9ed..0be16247f 100644
--- a/src/sim/system.hh
+++ b/src/sim/system.hh
@@ -161,6 +161,48 @@ class System : public SimObject
 
   protected:
     Enums::MemoryMode memoryMode;
+    uint64_t workItemsBegin;
+    uint64_t workItemsEnd;
+    std::vector<bool> activeCpus;
+
+  public:
+    /**
+     * Called by pseudo_inst to track the number of work items started by this
+     * system.
+     */
+    uint64_t 
+    incWorkItemsBegin()
+    {
+        return ++workItemsBegin;
+    }
+
+    /**
+     * Called by pseudo_inst to track the number of work items completed by
+     * this system.
+     */
+    uint64_t 
+    incWorkItemsEnd()
+    {
+        return ++workItemsEnd;
+    }
+
+    /**
+     * Called by pseudo_inst to mark the cpus actively executing work items.
+     * Returns the total number of cpus that have executed work item begin or
+     * ends.
+     */
+    int 
+    markWorkItem(int index)
+    {
+        int count = 0;
+        assert(index < activeCpus.size());
+        activeCpus[index] = true;
+        for (std::vector<bool>::iterator i = activeCpus.begin(); 
+             i < activeCpus.end(); i++) {
+            if (*i) count++;
+        }
+        return count;
+    }
 
 #if FULL_SYSTEM
     /**