i#6471 sched idle: Add replay fallback and fix

Adds a fallback to return EOF when all outputs are past the last
record in replay mode, to handle a schedule where not every input
reaches EOF or has regions of interest.  Adds a unit test for this
case, which hangs without the fix.

Fixes a record bug which seems to cause the final segment entry for an
input to have a stop instruction ordinal prior to the input's real
endpoint.  This is what caused hangs in actual usage of PR #6472 and
led to adding the fallback just described (which is useful on its
own).  On selecting a new input which turned out to be at EOF, the
original code called close_schedule_segment() on *prev_input* for some
reason.  That replaces the default -1 sentinel (read to EOF) with the
current ordinal.  The code then tries again for a new input, but if it
ends up finding the prior one it would keep executing that without a
new schedule segment entry.  If that's the last input on that core,
close_schedule_segment() is not called (the default is relied upon).
I observed missing sentinels for one input in some runs and this is my
theory as to what happened; those cases are gone now with this fix.
Unfortunately it is difficult to create a unit test for this, but
testing on the original hangs (without the fallback) show it fixed.

Issue: #6471

clients/drcachesim/scheduler/scheduler.cpp

@@ -642,6 +642,8 @@ scheduler_tmpl_t<RecordType, ReaderType>::set_initial_schedule(
     std::unordered_map<int, std::vector<int>> &workload2inputs)
 {
     if (options_.mapping == MAP_AS_PREVIOUSLY) {
+        live_replay_output_count_.store(static_cast<int>(outputs_.size()),
+                                        std::memory_order_release);
         if (options_.schedule_replay_istream == nullptr ||
             options_.schedule_record_ostream != nullptr)
             return STATUS_ERROR_INVALID_PARAMETER;
@@ -1646,8 +1648,13 @@ scheduler_tmpl_t<RecordType, ReaderType>::pick_next_input_as_previously(
     output_ordinal_t output, input_ordinal_t &index)
 {
     if (outputs_[output].record_index + 1 >=
-        static_cast<int>(outputs_[output].record.size()))
+        static_cast<int>(outputs_[output].record.size())) {
+        if (!outputs_[output].at_eof) {
+            outputs_[output].at_eof = true;
+            live_replay_output_count_.fetch_add(-1, std::memory_order_release);
+        }
         return eof_or_idle(output);
+    }
     const schedule_record_t &segment =
         outputs_[output].record[outputs_[output].record_index + 1];
     index = segment.key.input;
@@ -1895,14 +1902,12 @@ scheduler_tmpl_t<RecordType, ReaderType>::pick_next_input(output_ordinal_t outpu
         if (inputs_[index].at_eof ||
             *inputs_[index].reader == *inputs_[index].reader_end) {
             VPRINT(this, 2, "next_record[%d]: input #%d at eof\n", output, index);
-            if (options_.schedule_record_ostream != nullptr &&
-                prev_index != INVALID_INPUT_ORDINAL)
-                close_schedule_segment(output, inputs_[prev_index]);
             if (!inputs_[index].at_eof)
                 mark_input_eof(inputs_[index]);
+            if (options_.schedule_record_ostream != nullptr)
+                close_schedule_segment(output, inputs_[index]);
             index = INVALID_INPUT_ORDINAL;
             // Loop and pick next thread.
-            prev_index = INVALID_INPUT_ORDINAL;
             continue;
         }
         break;
@@ -2276,17 +2281,26 @@ template <typename RecordType, typename ReaderType>
 void
 scheduler_tmpl_t<RecordType, ReaderType>::mark_input_eof(input_info_t &input)
 {
+    if (input.at_eof)
+        return;
     input.at_eof = true;
     assert(live_input_count_.load(std::memory_order_acquire) > 0);
     live_input_count_.fetch_add(-1, std::memory_order_release);
+    VPRINT(this, 2, "input %d at eof; %d live inputs left\n", input.index,
+           live_input_count_.load(std::memory_order_acquire));
 }
 
 template <typename RecordType, typename ReaderType>
 typename scheduler_tmpl_t<RecordType, ReaderType>::stream_status_t
 scheduler_tmpl_t<RecordType, ReaderType>::eof_or_idle(output_ordinal_t output)
 {
     if (options_.mapping == MAP_TO_CONSISTENT_OUTPUT ||
-        live_input_count_.load(std::memory_order_acquire) == 0) {
+        live_input_count_.load(std::memory_order_acquire) == 0 ||
+        // While a full schedule recorded should have each input hit either its
+        // EOF or ROI end, we have a fallback to avoid hangs for possible recorded
+        // schedules that end an input early deliberately without an ROI.
+        (options_.mapping == MAP_AS_PREVIOUSLY &&
+         live_replay_output_count_.load(std::memory_order_acquire) == 0)) {
         return sched_type_t::STATUS_EOF;
     } else {
         outputs_[output].waiting = true;

clients/drcachesim/scheduler/scheduler.h

@@ -1094,6 +1094,8 @@ template <typename RecordType, typename ReaderType> class scheduler_tmpl_t {
         uint64_t cur_time = 0;
         // Used for MAP_TO_RECORDED_OUTPUT get_output_cpuid().
         int64_t as_traced_cpuid = -1;
+        // Used for MAP_AS_PREVIOUSLY with live_replay_output_count_.
+        bool at_eof = false;
     };
 
     // Called just once at initialization time to set the initial input-to-output
@@ -1346,6 +1348,8 @@ template <typename RecordType, typename ReaderType> class scheduler_tmpl_t {
     uint64_t ready_counter_ = 0;
     // Count of inputs not yet at eof.
     std::atomic<int> live_input_count_;
+    // In replay mode, count of outputs not yet at the end of the replay sequence.
+    std::atomic<int> live_replay_output_count_;
     // Map from workload,tid pair to input.
     struct workload_tid_t {
         workload_tid_t(int wl, memref_tid_t tid)

clients/drcachesim/tests/scheduler_unit_tests.cpp

@@ -2328,6 +2328,116 @@ test_replay_timestamps()
 #endif // HAS_ZIP
 }
 
+#ifdef HAS_ZIP
+// We subclass scheduler_t to access its record struct and functions.
+class test_noeof_scheduler_t : public scheduler_t {
+public:
+    void
+    write_test_schedule(std::string record_fname)
+    {
+        // We duplicate test_scheduler_t but we have one input ending early before
+        // eof.
+        scheduler_t scheduler;
+        std::vector<schedule_record_t> sched0;
+        sched0.emplace_back(scheduler_t::schedule_record_t::VERSION, 0, 0, 0, 0);
+        sched0.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 0, 0, 4, 11);
+        // There is a huge time gap here.
+        sched0.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 2, 7,
+                            0xffffffffffffffffUL, 91);
+        sched0.emplace_back(scheduler_t::schedule_record_t::FOOTER, 0, 0, 0, 0);
+        std::vector<schedule_record_t> sched1;
+        sched1.emplace_back(scheduler_t::schedule_record_t::VERSION, 0, 0, 0, 0);
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 1, 0, 4, 10);
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 2, 0, 4, 20);
+        // Input 2 advances early so core 0 is no longer waiting on it but only
+        // the timestamp.
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 2, 4, 7, 60);
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 3, 0, 4, 30);
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 0, 4, 7, 40);
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 1, 4, 7, 50);
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 3, 4, 7, 70);
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 0, 7,
+                            0xffffffffffffffffUL, 80);
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 1, 7,
+                            0xffffffffffffffffUL, 90);
+        // Input 3 never reaches EOF (end is exclusive).
+        sched1.emplace_back(scheduler_t::schedule_record_t::DEFAULT, 3, 7, 9, 110);
+        sched1.emplace_back(scheduler_t::schedule_record_t::FOOTER, 0, 0, 0, 0);
+        zipfile_ostream_t outfile(record_fname);
+        std::string err = outfile.open_new_component(recorded_schedule_component_name(0));
+        assert(err.empty());
+        if (!outfile.write(reinterpret_cast<char *>(sched0.data()),
+                           sched0.size() * sizeof(sched0[0])))
+            assert(false);
+        err = outfile.open_new_component(recorded_schedule_component_name(1));
+        assert(err.empty());
+        if (!outfile.write(reinterpret_cast<char *>(sched1.data()),
+                           sched1.size() * sizeof(sched1[0])))
+            assert(false);
+    }
+};
+#endif
+
+static void
+test_replay_noeof()
+{
+#ifdef HAS_ZIP
+    std::cerr << "\n----------------\nTesting replay with no eof\n";
+    static constexpr int NUM_INPUTS = 4;
+    static constexpr int NUM_OUTPUTS = 2;
+    static constexpr int NUM_INSTRS = 9;
+    static constexpr memref_tid_t TID_BASE = 100;
+    std::vector<trace_entry_t> inputs[NUM_INPUTS];
+    for (int i = 0; i < NUM_INPUTS; i++) {
+        memref_tid_t tid = TID_BASE + i;
+        inputs[i].push_back(make_thread(tid));
+        inputs[i].push_back(make_pid(1));
+        // We need a timestamp so the scheduler will find one for initial
+        // input processing.  We do not try to duplicate the timestamp
+        // sequences in the stored file and just use a dummy timestamp here.
+        inputs[i].push_back(make_timestamp(10 + i));
+        for (int j = 0; j < NUM_INSTRS; j++)
+            inputs[i].push_back(make_instr(42 + j * 4));
+        inputs[i].push_back(make_exit(tid));
+    }
+
+    // Create a record file with timestamps requiring waiting.
+    // We cooperate with the test_noeof_scheduler_t class which constructs this schedule:
+    static const char *const CORE0_SCHED_STRING = ".AAA-------------------------CCC.__";
+    static const char *const CORE1_SCHED_STRING = ".BBB.CCCCCC.DDDAAABBBDDDAAA.BBB.DD";
+    std::string record_fname = "tmp_test_replay_noeof_timestamp.zip";
+    test_noeof_scheduler_t test_scheduler;
+    test_scheduler.write_test_schedule(record_fname);
+
+    // Replay the recorded schedule.
+    std::vector<scheduler_t::input_workload_t> sched_inputs;
+    for (int i = 0; i < NUM_INPUTS; i++) {
+        memref_tid_t tid = TID_BASE + i;
+        std::vector<scheduler_t::input_reader_t> readers;
+        readers.emplace_back(std::unique_ptr<mock_reader_t>(new mock_reader_t(inputs[i])),
+                             std::unique_ptr<mock_reader_t>(new mock_reader_t()), tid);
+        sched_inputs.emplace_back(std::move(readers));
+    }
+    scheduler_t::scheduler_options_t sched_ops(scheduler_t::MAP_AS_PREVIOUSLY,
+                                               scheduler_t::DEPENDENCY_TIMESTAMPS,
+                                               scheduler_t::SCHEDULER_DEFAULTS,
+                                               /*verbosity=*/4);
+    zipfile_istream_t infile(record_fname);
+    sched_ops.schedule_replay_istream = &infile;
+    scheduler_t scheduler;
+    if (scheduler.init(sched_inputs, NUM_OUTPUTS, sched_ops) !=
+        scheduler_t::STATUS_SUCCESS)
+        assert(false);
+    std::vector<std::string> sched_as_string =
+        run_lockstep_simulation(scheduler, NUM_OUTPUTS, TID_BASE);
+    for (int i = 0; i < NUM_OUTPUTS; i++) {
+        std::cerr << "cpu #" << i << " schedule: " << sched_as_string[i] << "\n";
+    }
+    assert(sched_as_string[0] == CORE0_SCHED_STRING);
+    assert(sched_as_string[1] == CORE1_SCHED_STRING);
+#endif // HAS_ZIP
+}
+
 static void
 test_replay_skip()
 {
@@ -3200,6 +3310,7 @@ test_main(int argc, const char *argv[])
     test_replay();
     test_replay_multi_threaded(argv[1]);
     test_replay_timestamps();
+    test_replay_noeof();
     test_replay_skip();
     test_replay_limit();
     test_replay_as_traced_from_file(argv[1]);