summaryrefslogtreecommitdiff
path: root/src/mem/ruby
diff options
context:
space:
mode:
Diffstat (limited to 'src/mem/ruby')
-rw-r--r--src/mem/ruby/profiler/Profiler.cc99
-rw-r--r--src/mem/ruby/profiler/Profiler.hh24
-rw-r--r--src/mem/ruby/system/Sequencer.cc49
-rw-r--r--src/mem/ruby/system/Sequencer.hh19
4 files changed, 187 insertions, 4 deletions
diff --git a/src/mem/ruby/profiler/Profiler.cc b/src/mem/ruby/profiler/Profiler.cc
index 753fdd230..a42d919f7 100644
--- a/src/mem/ruby/profiler/Profiler.cc
+++ b/src/mem/ruby/profiler/Profiler.cc
@@ -286,6 +286,35 @@ Profiler::printStats(ostream& out, bool short_stats)
}
}
+ out << "miss_latency_wCC_issue_to_initial_request: "
+ << m_wCCIssueToInitialRequestHistogram << endl;
+ out << "miss_latency_wCC_initial_forward_request: "
+ << m_wCCInitialRequestToForwardRequestHistogram << endl;
+ out << "miss_latency_wCC_forward_to_first_response: "
+ << m_wCCForwardRequestToFirstResponseHistogram << endl;
+ out << "miss_latency_wCC_first_response_to_completion: "
+ << m_wCCFirstResponseToCompleteHistogram << endl;
+ out << "imcomplete_wCC_Times: " << m_wCCIncompleteTimes << endl;
+ out << "miss_latency_dir_issue_to_initial_request: "
+ << m_dirIssueToInitialRequestHistogram << endl;
+ out << "miss_latency_dir_initial_forward_request: "
+ << m_dirInitialRequestToForwardRequestHistogram << endl;
+ out << "miss_latency_dir_forward_to_first_response: "
+ << m_dirForwardRequestToFirstResponseHistogram << endl;
+ out << "miss_latency_dir_first_response_to_completion: "
+ << m_dirFirstResponseToCompleteHistogram << endl;
+ out << "imcomplete_dir_Times: " << m_dirIncompleteTimes << endl;
+
+ for (int i = 0; i < m_missMachLatencyHistograms.size(); i++) {
+ for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
+ if (m_missMachLatencyHistograms[i][j].size() > 0) {
+ out << "miss_latency_" << RubyRequestType(i)
+ << "_" << GenericMachineType(j) << ": "
+ << m_missMachLatencyHistograms[i][j] << endl;
+ }
+ }
+ }
+
out << endl;
out << "All Non-Zero Cycle SW Prefetch Requests" << endl;
@@ -454,7 +483,24 @@ Profiler::clearStats()
for (int i = 0; i < m_machLatencyHistograms.size(); i++) {
m_machLatencyHistograms[i].clear(200);
}
+ m_missMachLatencyHistograms.resize(RubyRequestType_NUM);
+ for (int i = 0; i < m_missLatencyHistograms.size(); i++) {
+ m_missMachLatencyHistograms[i].resize(GenericMachineType_NUM+1);
+ for (int j = 0; j < m_missMachLatencyHistograms[i].size(); j++) {
+ m_missMachLatencyHistograms[i][j].clear(200);
+ }
+ }
m_allMissLatencyHistogram.clear(200);
+ m_wCCIssueToInitialRequestHistogram.clear(200);
+ m_wCCInitialRequestToForwardRequestHistogram.clear(200);
+ m_wCCForwardRequestToFirstResponseHistogram.clear(200);
+ m_wCCFirstResponseToCompleteHistogram.clear(200);
+ m_wCCIncompleteTimes = 0;
+ m_dirIssueToInitialRequestHistogram.clear(200);
+ m_dirInitialRequestToForwardRequestHistogram.clear(200);
+ m_dirForwardRequestToFirstResponseHistogram.clear(200);
+ m_dirFirstResponseToCompleteHistogram.clear(200);
+ m_dirIncompleteTimes = 0;
m_SWPrefetchLatencyHistograms.resize(CacheRequestType_NUM);
for (int i = 0; i < m_SWPrefetchLatencyHistograms.size(); i++) {
@@ -581,6 +627,59 @@ Profiler::missLatency(Time cycles,
m_allMissLatencyHistogram.add(cycles);
m_missLatencyHistograms[type].add(cycles);
m_machLatencyHistograms[respondingMach].add(cycles);
+ m_missMachLatencyHistograms[type][respondingMach].add(cycles);
+}
+
+void
+Profiler::missLatencyWcc(Time issuedTime,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime,
+ Time completionTime)
+{
+ if ((issuedTime <= initialRequestTime) &&
+ (initialRequestTime <= forwardRequestTime) &&
+ (forwardRequestTime <= firstResponseTime) &&
+ (firstResponseTime <= completionTime)) {
+ m_wCCIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
+
+ m_wCCInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
+ initialRequestTime);
+
+ m_wCCForwardRequestToFirstResponseHistogram.add(firstResponseTime -
+ forwardRequestTime);
+
+ m_wCCFirstResponseToCompleteHistogram.add(completionTime -
+ firstResponseTime);
+ } else {
+ m_wCCIncompleteTimes++;
+ }
+}
+
+void
+Profiler::missLatencyDir(Time issuedTime,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime,
+ Time completionTime)
+{
+ if ((issuedTime <= initialRequestTime) &&
+ (initialRequestTime <= forwardRequestTime) &&
+ (forwardRequestTime <= firstResponseTime) &&
+ (firstResponseTime <= completionTime)) {
+ m_dirIssueToInitialRequestHistogram.add(initialRequestTime - issuedTime);
+
+ m_dirInitialRequestToForwardRequestHistogram.add(forwardRequestTime -
+ initialRequestTime);
+
+ m_dirForwardRequestToFirstResponseHistogram.add(firstResponseTime -
+ forwardRequestTime);
+
+ m_dirFirstResponseToCompleteHistogram.add(completionTime -
+ firstResponseTime);
+ } else {
+ m_dirIncompleteTimes++;
+ }
}
// non-zero cycle prefetch request
diff --git a/src/mem/ruby/profiler/Profiler.hh b/src/mem/ruby/profiler/Profiler.hh
index de9834f05..3a11f0596 100644
--- a/src/mem/ruby/profiler/Profiler.hh
+++ b/src/mem/ruby/profiler/Profiler.hh
@@ -138,6 +138,18 @@ class Profiler : public SimObject, public Consumer
RubyRequestType type,
const GenericMachineType respondingMach);
+ void missLatencyWcc(Time issuedTime,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime,
+ Time completionTime);
+
+ void missLatencyDir(Time issuedTime,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime,
+ Time completionTime);
+
void swPrefetchLatency(Time t,
CacheRequestType type,
const GenericMachineType respondingMach);
@@ -200,6 +212,18 @@ class Profiler : public SimObject, public Consumer
std::vector<Histogram> m_missLatencyHistograms;
std::vector<Histogram> m_machLatencyHistograms;
+ std::vector< std::vector<Histogram> > m_missMachLatencyHistograms;
+ Histogram m_wCCIssueToInitialRequestHistogram;
+ Histogram m_wCCInitialRequestToForwardRequestHistogram;
+ Histogram m_wCCForwardRequestToFirstResponseHistogram;
+ Histogram m_wCCFirstResponseToCompleteHistogram;
+ int64 m_wCCIncompleteTimes;
+ Histogram m_dirIssueToInitialRequestHistogram;
+ Histogram m_dirInitialRequestToForwardRequestHistogram;
+ Histogram m_dirForwardRequestToFirstResponseHistogram;
+ Histogram m_dirFirstResponseToCompleteHistogram;
+ int64 m_dirIncompleteTimes;
+
Histogram m_allMissLatencyHistogram;
Histogram m_allSWPrefetchLatencyHistogram;
diff --git a/src/mem/ruby/system/Sequencer.cc b/src/mem/ruby/system/Sequencer.cc
index 175bb6721..74b6355e8 100644
--- a/src/mem/ruby/system/Sequencer.cc
+++ b/src/mem/ruby/system/Sequencer.cc
@@ -357,6 +357,17 @@ Sequencer::writeCallback(const Address& address,
GenericMachineType mach,
DataBlock& data)
{
+ writeCallback(address, mach, data, 0, 0, 0);
+}
+
+void
+Sequencer::writeCallback(const Address& address,
+ GenericMachineType mach,
+ DataBlock& data,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime)
+{
assert(address == line_address(address));
assert(m_writeRequestTable.count(line_address(address)));
@@ -385,7 +396,8 @@ Sequencer::writeCallback(const Address& address,
m_controller->unblock(address);
}
- hitCallback(request, mach, data, success);
+ hitCallback(request, mach, data, success,
+ initialRequestTime, forwardRequestTime, firstResponseTime);
}
void
@@ -399,6 +411,17 @@ Sequencer::readCallback(const Address& address,
GenericMachineType mach,
DataBlock& data)
{
+ readCallback(address, mach, data, 0, 0, 0);
+}
+
+void
+Sequencer::readCallback(const Address& address,
+ GenericMachineType mach,
+ DataBlock& data,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime)
+{
assert(address == line_address(address));
assert(m_readRequestTable.count(line_address(address)));
@@ -413,14 +436,18 @@ Sequencer::readCallback(const Address& address,
(request->ruby_request.type == RubyRequestType_RMW_Read) ||
(request->ruby_request.type == RubyRequestType_IFETCH));
- hitCallback(request, mach, data, true);
+ hitCallback(request, mach, data, true,
+ initialRequestTime, forwardRequestTime, firstResponseTime);
}
void
Sequencer::hitCallback(SequencerRequest* srequest,
GenericMachineType mach,
DataBlock& data,
- bool success)
+ bool success,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime)
{
const RubyRequest & ruby_request = srequest->ruby_request;
Address request_address(ruby_request.paddr);
@@ -445,6 +472,22 @@ Sequencer::hitCallback(SequencerRequest* srequest,
if (miss_latency != 0) {
g_system_ptr->getProfiler()->missLatency(miss_latency, type, mach);
+ if (mach == GenericMachineType_L1Cache_wCC) {
+ g_system_ptr->getProfiler()->missLatencyWcc(issued_time,
+ initialRequestTime,
+ forwardRequestTime,
+ firstResponseTime,
+ g_eventQueue_ptr->getTime());
+ }
+
+ if (mach == GenericMachineType_Directory) {
+ g_system_ptr->getProfiler()->missLatencyDir(issued_time,
+ initialRequestTime,
+ forwardRequestTime,
+ firstResponseTime,
+ g_eventQueue_ptr->getTime());
+ }
+
if (Debug::getProtocolTrace()) {
if (success) {
g_system_ptr->getProfiler()->
diff --git a/src/mem/ruby/system/Sequencer.hh b/src/mem/ruby/system/Sequencer.hh
index fecfc9a1a..4ab85dac8 100644
--- a/src/mem/ruby/system/Sequencer.hh
+++ b/src/mem/ruby/system/Sequencer.hh
@@ -80,12 +80,26 @@ class Sequencer : public RubyPort, public Consumer
GenericMachineType mach,
DataBlock& data);
+ void writeCallback(const Address& address,
+ GenericMachineType mach,
+ DataBlock& data,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime);
+
void readCallback(const Address& address, DataBlock& data);
void readCallback(const Address& address,
GenericMachineType mach,
DataBlock& data);
+ void readCallback(const Address& address,
+ GenericMachineType mach,
+ DataBlock& data,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime);
+
RequestStatus makeRequest(const RubyRequest & request);
RequestStatus getRequestStatus(const RubyRequest& request);
bool empty() const;
@@ -106,7 +120,10 @@ class Sequencer : public RubyPort, public Consumer
void hitCallback(SequencerRequest* request,
GenericMachineType mach,
DataBlock& data,
- bool success);
+ bool success,
+ Time initialRequestTime,
+ Time forwardRequestTime,
+ Time firstResponseTime);
bool insertRequest(SequencerRequest* request);