Add CLI option -sched_max_cores and an end-to-end analyzer test runni…

…ng threadsig

clients/drcachesim/analyzer.cpp

@@ -230,7 +230,7 @@ template <typename RecordType, typename ReaderType>
 bool
 analyzer_tmpl_t<RecordType, ReaderType>::init_scheduler(
     const std::string &trace_path, const std::set<memref_tid_t> &only_threads,
-    const std::set<int> &only_shards, int verbosity,
+    const std::set<int> &only_shards, int output_limit, int verbosity,
     typename sched_type_t::scheduler_options_t options)
 {
     verbosity_ = verbosity;
@@ -249,6 +249,7 @@ analyzer_tmpl_t<RecordType, ReaderType>::init_scheduler(
     typename sched_type_t::input_workload_t workload(trace_path, regions);
     workload.only_threads = only_threads;
     workload.only_shards = only_shards;
+    workload.output_limit = output_limit;
     if (regions.empty() && skip_to_timestamp_ > 0) {
         workload.times_of_interest.emplace_back(skip_to_timestamp_, 0);
     }
@@ -383,7 +384,7 @@ analyzer_tmpl_t<RecordType, ReaderType>::analyzer_tmpl_t(
     // The scheduler will call reader_t::init() for each input file.  We assume
     // that won't block (analyzer_multi_t separates out IPC readers).
     typename sched_type_t::scheduler_options_t sched_ops;
-    if (!init_scheduler(trace_path, {}, {}, verbosity, std::move(sched_ops))) {
+    if (!init_scheduler(trace_path, {}, {}, 0, verbosity, std::move(sched_ops))) {
         success_ = false;
         error_string_ = "Failed to create scheduler";
         return;

clients/drcachesim/analyzer.h

@@ -221,7 +221,7 @@ template <typename RecordType, typename ReaderType> class analyzer_tmpl_t {
     init_scheduler(const std::string &trace_path,
                    // To include all threads/shards, use empty sets.
                    const std::set<memref_tid_t> &only_threads,
-                   const std::set<int> &only_shards, int verbosity,
+                   const std::set<int> &only_shards, int output_limit, int verbosity,
                    typename sched_type_t::scheduler_options_t options);
 
     // For core-sharded, worker_count_ must be set prior to calling this; for parallel

clients/drcachesim/analyzer_multi.cpp

@@ -549,7 +549,8 @@ analyzer_multi_tmpl_t<RecordType, ReaderType>::analyzer_multi_tmpl_t()
             return;
         }
         if (!this->init_scheduler(tracedir, only_threads, only_shards,
-                                  op_verbose.get_value(), std::move(sched_ops))) {
+                                  op_sched_max_cores.get_value(), op_verbose.get_value(),
+                                  std::move(sched_ops))) {
             this->success_ = false;
             return;
         }
@@ -573,7 +574,8 @@ analyzer_multi_tmpl_t<RecordType, ReaderType>::analyzer_multi_tmpl_t()
         }
     } else {
         // Legacy file.
-        if (!this->init_scheduler(op_infile.get_value(), {}, {}, op_verbose.get_value(),
+        if (!this->init_scheduler(op_infile.get_value(), {}, {},
+                                  op_sched_max_cores.get_value(), op_verbose.get_value(),
                                   std::move(sched_ops))) {
             this->success_ = false;
             return;

clients/drcachesim/common/options.cpp

@@ -1061,6 +1061,12 @@ droption_t<double> op_sched_exit_if_fraction_inputs_left(
     "count is not considered (as it is not available), use discretion when raising "
     "this value on uneven inputs.");
 
+droption_t<int> op_sched_max_cores(
+    DROPTION_SCOPE_ALL, "sched_max_cores", 0,
+    "Limit scheduling to this many peak live cores",
+    "If non-zero, only this many live cores can be scheduled at any one time.  "
+    "Other cores will remain idle.");
+
 // Schedule_stats options.
 droption_t<uint64_t>
     op_schedule_stats_print_every(DROPTION_SCOPE_ALL, "schedule_stats_print_every",

clients/drcachesim/common/options.h

@@ -222,6 +222,7 @@ extern dynamorio::droption::droption_t<uint64_t> op_sched_migration_threshold_us
 extern dynamorio::droption::droption_t<uint64_t> op_sched_rebalance_period_us;
 extern dynamorio::droption::droption_t<double> op_sched_time_units_per_us;
 extern dynamorio::droption::droption_t<double> op_sched_exit_if_fraction_inputs_left;
+extern dynamorio::droption::droption_t<int> op_sched_max_cores;
 extern dynamorio::droption::droption_t<uint64_t> op_schedule_stats_print_every;
 extern dynamorio::droption::droption_t<std::string> op_syscall_template_file;
 extern dynamorio::droption::droption_t<uint64_t> op_filter_stop_timestamp;

suite/tests/CMakeLists.txt

@@ -3936,6 +3936,11 @@ if (BUILD_CLIENTS)
           "")
         set(tool.schedule_stats_nopreempt_rawtemp ON) # no preprocessor
 
+        torunonly_simtool(schedule_stats_maxcores ${ci_shared_app}
+          "-indir ${thread_trace_dir} -tool schedule_stats -core_sharded -sched_max_cores 2"
+          "")
+        set(tool.schedule_stats_nopreempt_rawtemp ON) # no preprocessor
+
         torunonly_simtool(core_serial ${ci_shared_app}
           "-indir ${thread_trace_dir} -tool schedule_stats:basic_counts -core_serial"
           "")