mem: track per-request latencies and access depths in the cache hierarchy

Add some values and methods to the request object to track the translation
and access latency for a request and which level of the cache hierarchy responded
to the request.

6 files changed, 55 insertions, 3 deletions

diff --git a/src/cpu/o3/fetch_impl.hh b/src/cpu/o3/fetch_impl.hh
index b35dd80f3..b121ba707 100644
--- a/src/cpu/o3/fetch_impl.hh
+++ b/src/cpu/o3/fetch_impl.hh
@@ -400,6 +400,7 @@ DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt)
         fetchStatus[tid] = IcacheAccessComplete;
     }
 
+    pkt->req->setAccessLatency();
     // Reset the mem req to NULL.
     delete pkt->req;
     delete pkt;
diff --git a/src/cpu/o3/lsq_unit_impl.hh b/src/cpu/o3/lsq_unit_impl.hh
index 77b67ac69..ade076995 100644
--- a/src/cpu/o3/lsq_unit_impl.hh
+++ b/src/cpu/o3/lsq_unit_impl.hh
@@ -129,6 +129,8 @@ LSQUnit<Impl>::completeDataAccess(PacketPtr pkt)
         delete state->mainPkt->req;
         delete state->mainPkt;
     }
+
+    pkt->req->setAccessLatency();
     delete state;
     delete pkt->req;
     delete pkt;
diff --git a/src/cpu/simple/timing.cc b/src/cpu/simple/timing.cc
index 744bf8397..9253d8005 100644
--- a/src/cpu/simple/timing.cc
+++ b/src/cpu/simple/timing.cc
@@ -646,7 +646,6 @@ TimingSimpleCPU::completeIfetch(PacketPtr pkt)
 
     // received a response from the icache: execute the received
     // instruction
-
     assert(!pkt || !pkt->isError());
     assert(_status == IcacheWaitResponse);
 
@@ -655,6 +654,10 @@ TimingSimpleCPU::completeIfetch(PacketPtr pkt)
     numCycles += curCycle() - previousCycle;
     previousCycle = curCycle();
 
+    if (pkt)
+        pkt->req->setAccessLatency();
+
+
     preExecute();
     if (curStaticInst && curStaticInst->isMemRef()) {
         // load or store: just send to dcache
@@ -749,6 +752,7 @@ TimingSimpleCPU::completeDataAccess(PacketPtr pkt)
     assert(_status == DcacheWaitResponse || _status == DTBWaitResponse ||
            pkt->req->getFlags().isSet(Request::NO_ACCESS));
 
+    pkt->req->setAccessLatency();
     numCycles += curCycle() - previousCycle;
     previousCycle = curCycle();
 
diff --git a/src/cpu/translation.hh b/src/cpu/translation.hh
index 90fffa03d..c05cc86a0 100644
--- a/src/cpu/translation.hh
+++ b/src/cpu/translation.hh
@@ -256,6 +256,7 @@ class DataTranslation : public BaseTLB::Translation
         assert(mode == state->mode);
         if (state->finish(fault, index)) {
             xc->finishTranslation(state);
+            req->setTranslateLatency();
         }
         delete this;
     }
diff --git a/src/mem/cache/base.hh b/src/mem/cache/base.hh
index dce40e915..50ba396b2 100644
--- a/src/mem/cache/base.hh
+++ b/src/mem/cache/base.hh
@@ -568,7 +568,7 @@ class BaseCache : public MemObject
     {
         assert(pkt->req->masterId() < system->maxMasters());
         misses[pkt->cmdToIndex()][pkt->req->masterId()]++;
-
+        pkt->req->incAccessDepth();
         if (missCount) {
             --missCount;
             if (missCount == 0)
diff --git a/src/mem/request.hh b/src/mem/request.hh
index a0ff50910..54b671645 100644
--- a/src/mem/request.hh
+++ b/src/mem/request.hh
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2012 ARM Limited
+ * Copyright (c) 2012-2013 ARM Limited
  * All rights reserved
  *
  * The license below extends only to copyright in the software and shall
@@ -244,6 +244,7 @@ class Request
      *  default constructor.)
      */
     Request()
+        : translateDelta(0), accessDelta(0), depth(0)
     {}
 
     /**
@@ -304,6 +305,9 @@ class Request
         _flags.set(flags);
         privateFlags.clear(~STICKY_PRIVATE_FLAGS);
         privateFlags.set(VALID_PADDR|VALID_SIZE);
+        depth = 0;
+        accessDelta = 0;
+        //translateDelta = 0;
     }
 
     void
@@ -331,6 +335,9 @@ class Request
         _flags.set(flags);
         privateFlags.clear(~STICKY_PRIVATE_FLAGS);
         privateFlags.set(VALID_VADDR|VALID_SIZE|VALID_PC);
+        depth = 0;
+        accessDelta = 0;
+        translateDelta = 0;
     }
 
     /**
@@ -382,6 +389,23 @@ class Request
     }
 
     /**
+     * Time for the TLB/table walker to successfully translate this request.
+     */
+    Tick translateDelta;
+
+    /**
+     * Access latency to complete this memory transaction not including
+     * translation time.
+     */
+    Tick accessDelta;
+
+    /**
+     * Level of the cache hierachy where this request was responded to
+     * (e.g. 0 = L1; 1 = L2).
+     */
+    int depth;
+
+    /**
      *  Accessor for size.
      */
     bool
@@ -535,6 +559,26 @@ class Request
         return _pc;
     }
 
+    /**
+     * Increment/Get the depth at which this request is responded to.
+     * This currently happens when the request misses in any cache level.
+     */
+    void incAccessDepth() { depth++; }
+    int getAccessDepth() const { return depth; }
+
+    /**
+     * Set/Get the time taken for this request to be successfully translated.
+     */
+    void setTranslateLatency() { translateDelta = curTick() - _time; }
+    Tick getTranslateLatency() const { return translateDelta; }
+
+    /**
+     * Set/Get the time taken to complete this request's access, not including
+     *  the time to successfully translate the request.
+     */
+    void setAccessLatency() { accessDelta = curTick() - _time - translateDelta; }
+    Tick getAccessLatency() const { return accessDelta; }
+
     /** Accessor functions for flags.  Note that these are for testing
        only; setting flags should be done via setFlags(). */
     bool isUncacheable() const { return _flags.isSet(UNCACHEABLE); }