mem: per-thread cache occupancy and per-block ages

This patch enables tracking of cache occupancy per thread along with
ages (in buckets) per cache blocks.  Cache occupancy stats are
recalculated on each stat dump.

4 files changed, 14 insertions, 0 deletions

diff --git a/src/cpu/base_dyn_inst.hh b/src/cpu/base_dyn_inst.hh
index b7b076820..f12a89bbd 100644
--- a/src/cpu/base_dyn_inst.hh
+++ b/src/cpu/base_dyn_inst.hh
@@ -890,6 +890,8 @@ BaseDynInst<Impl>::readMem(Addr addr, uint8_t *data,
         req = new Request(asid, addr, size, flags, masterId(), this->pc.instAddr(),
                           thread->contextId(), threadNumber);
 
+        req->taskId(cpu->taskId());
+
         // Only split the request if the ISA supports unaligned accesses.
         if (TheISA::HasUnalignedMemAcc) {
             splitRequest(req, sreqLow, sreqHigh);
@@ -953,6 +955,8 @@ BaseDynInst<Impl>::writeMem(uint8_t *data, unsigned size,
         req = new Request(asid, addr, size, flags, masterId(), this->pc.instAddr(),
                           thread->contextId(), threadNumber);
 
+        req->taskId(cpu->taskId());
+
         // Only split the request if the ISA supports unaligned accesses.
         if (TheISA::HasUnalignedMemAcc) {
             splitRequest(req, sreqLow, sreqHigh);
diff --git a/src/cpu/o3/fetch_impl.hh b/src/cpu/o3/fetch_impl.hh
index 5b04c2a25..a81125da6 100644
--- a/src/cpu/o3/fetch_impl.hh
+++ b/src/cpu/o3/fetch_impl.hh
@@ -604,6 +604,8 @@ DefaultFetch<Impl>::fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc)
                     Request::INST_FETCH, cpu->instMasterId(), pc,
                     cpu->thread[tid]->contextId(), tid);
 
+    mem_req->taskId(cpu->taskId());
+
     memReq[tid] = mem_req;
 
     // Initiate translation of the icache block
diff --git a/src/cpu/simple/atomic.cc b/src/cpu/simple/atomic.cc
index 13c4b9bd3..617e845a5 100644
--- a/src/cpu/simple/atomic.cc
+++ b/src/cpu/simple/atomic.cc
@@ -301,6 +301,7 @@ AtomicSimpleCPU::readMem(Addr addr, uint8_t * data,
 
     dcache_latency = 0;
 
+    req->taskId(taskId());
     while (1) {
         req->setVirt(0, addr, size, flags, dataMasterId(), thread->pcState().instAddr());
 
@@ -387,6 +388,7 @@ AtomicSimpleCPU::writeMem(uint8_t *data, unsigned size,
 
     dcache_latency = 0;
 
+    req->taskId(taskId());
     while(1) {
         req->setVirt(0, addr, size, flags, dataMasterId(), thread->pcState().instAddr());
 
@@ -492,6 +494,7 @@ AtomicSimpleCPU::tick()
         bool needToFetch = !isRomMicroPC(pcState.microPC()) &&
                            !curMacroStaticInst;
         if (needToFetch) {
+            ifetch_req.taskId(taskId());
             setupFetchRequest(&ifetch_req);
             fault = thread->itb->translateAtomic(&ifetch_req, tc,
                                                  BaseTLB::Execute);
diff --git a/src/cpu/simple/timing.cc b/src/cpu/simple/timing.cc
index 9253d8005..7996a6ddd 100644
--- a/src/cpu/simple/timing.cc
+++ b/src/cpu/simple/timing.cc
@@ -415,6 +415,8 @@ TimingSimpleCPU::readMem(Addr addr, uint8_t *data,
     RequestPtr req  = new Request(asid, addr, size,
                                   flags, dataMasterId(), pc, _cpuId, tid);
 
+    req->taskId(taskId());
+
     Addr split_addr = roundDown(addr + size - 1, block_size);
     assert(split_addr <= addr || split_addr - addr < block_size);
 
@@ -484,6 +486,8 @@ TimingSimpleCPU::writeMem(uint8_t *data, unsigned size,
     RequestPtr req = new Request(asid, addr, size,
                                  flags, dataMasterId(), pc, _cpuId, tid);
 
+    req->taskId(taskId());
+
     Addr split_addr = roundDown(addr + size - 1, block_size);
     assert(split_addr <= addr || split_addr - addr < block_size);
 
@@ -561,6 +565,7 @@ TimingSimpleCPU::fetch()
     if (needToFetch) {
         _status = BaseSimpleCPU::Running;
         Request *ifetch_req = new Request();
+        ifetch_req->taskId(taskId());
         ifetch_req->setThreadContext(_cpuId, /* thread ID */ 0);
         setupFetchRequest(ifetch_req);
         DPRINTF(SimpleCPU, "Translating address %#x\n", ifetch_req->getVaddr());