Merge branch 'master' into i6678-rename-ci-aarchxx-cross-workflow

clients/drcachesim/common/utils.h

@@ -57,6 +57,8 @@ namespace drmemtrace {
 
 // XXX: DR should export this
 #define INVALID_THREAD_ID 0
+// We avoid collisions with DR's INVALID_PROCESS_ID by using our own name.
+#define INVALID_PID -1
 
 // XXX: perhaps we should use a C++-ish stream approach instead
 // This cannot be named ERROR as that conflicts with Windows headers.

clients/drcachesim/reader/record_file_reader.h

@@ -1,5 +1,5 @@
 /* **********************************************************
- * Copyright (c) 2022-2023 Google, Inc.  All rights reserved.
+ * Copyright (c) 2022-2024 Google, Inc.  All rights reserved.
  * **********************************************************/
 
 /*
@@ -126,7 +126,7 @@ class record_reader_t : public std::iterator<std::input_iterator_tag, trace_entr
     virtual ~record_reader_t()
     {
     }
-    bool
+    virtual bool
     init()
     {
         if (!open_input_file())

clients/drcachesim/scheduler/scheduler.cpp

@@ -207,6 +207,17 @@ scheduler_tmpl_t<memref_t, reader_t>::record_type_has_tid(memref_t record,
     return true;
 }
 
+template <>
+bool
+scheduler_tmpl_t<memref_t, reader_t>::record_type_has_pid(memref_t record,
+                                                          memref_pid_t &pid)
+{
+    if (record.marker.pid == INVALID_PID)
+        return false;
+    pid = record.marker.pid;
+    return true;
+}
+
 template <>
 void
 scheduler_tmpl_t<memref_t, reader_t>::record_type_set_tid(memref_t &record,
@@ -222,6 +233,23 @@ scheduler_tmpl_t<memref_t, reader_t>::record_type_is_instr(memref_t record)
     return type_is_instr(record.instr.type);
 }
 
+template <>
+bool
+scheduler_tmpl_t<memref_t, reader_t>::record_type_is_encoding(memref_t record)
+{
+    // There are no separate memref_t encoding records: encoding info is
+    // inside instruction records.
+    return false;
+}
+
+template <>
+bool
+scheduler_tmpl_t<memref_t, reader_t>::record_type_is_instr_boundary(memref_t record,
+                                                                    memref_t prev_record)
+{
+    return record_type_is_instr(record);
+}
+
 template <>
 bool
 scheduler_tmpl_t<memref_t, reader_t>::record_type_is_marker(memref_t record,
@@ -302,6 +330,13 @@ scheduler_tmpl_t<memref_t, reader_t>::print_record(const memref_t &record)
     fprintf(stderr, "\n");
 }
 
+template <>
+void
+scheduler_tmpl_t<memref_t, reader_t>::insert_switch_tid_pid(input_info_t &info)
+{
+    // We do nothing, as every record has a tid from the separate inputs.
+}
+
 /******************************************************************************
  * Specializations for scheduler_tmpl_t<record_reader_t>, aka record_scheduler_t.
  */
@@ -343,6 +378,17 @@ scheduler_tmpl_t<trace_entry_t, record_reader_t>::record_type_has_tid(
     return true;
 }
 
+template <>
+bool
+scheduler_tmpl_t<trace_entry_t, record_reader_t>::record_type_has_pid(
+    trace_entry_t record, memref_pid_t &pid)
+{
+    if (record.type != TRACE_TYPE_PID)
+        return false;
+    pid = static_cast<memref_pid_t>(record.addr);
+    return true;
+}
+
 template <>
 void
 scheduler_tmpl_t<trace_entry_t, record_reader_t>::record_type_set_tid(
@@ -361,6 +407,34 @@ scheduler_tmpl_t<trace_entry_t, record_reader_t>::record_type_is_instr(
     return type_is_instr(static_cast<trace_type_t>(record.type));
 }
 
+template <>
+bool
+scheduler_tmpl_t<trace_entry_t, record_reader_t>::record_type_is_encoding(
+    trace_entry_t record)
+{
+    return static_cast<trace_type_t>(record.type) == TRACE_TYPE_ENCODING;
+}
+
+template <>
+bool
+scheduler_tmpl_t<trace_entry_t, record_reader_t>::record_type_is_instr_boundary(
+    trace_entry_t record, trace_entry_t prev_record)
+{
+    // Don't advance past encodings and split them from their associated instr.
+    return (record_type_is_instr(record) || record_type_is_encoding(record)) &&
+        !record_type_is_encoding(prev_record);
+}
+
+template <>
+typename scheduler_tmpl_t<trace_entry_t, record_reader_t>::stream_status_t
+scheduler_tmpl_t<trace_entry_t, record_reader_t>::unread_last_record(
+    output_ordinal_t output, trace_entry_t &record, input_info_t *&input)
+{
+    // See the general unread_last_record() below: we don't support this as
+    // we can't provide the prev-prev record for record_type_is_instr_boundary().
+    return STATUS_NOT_IMPLEMENTED;
+}
+
 template <>
 bool
 scheduler_tmpl_t<trace_entry_t, record_reader_t>::record_type_is_marker(
@@ -437,6 +511,27 @@ scheduler_tmpl_t<trace_entry_t, record_reader_t>::print_record(
             record.addr);
 }
 
+template <>
+void
+scheduler_tmpl_t<trace_entry_t, record_reader_t>::insert_switch_tid_pid(
+    input_info_t &input)
+{
+    // We need explicit tid,pid records so reader_t will see the new context.
+    // We insert at the front, so we have reverse order.
+    trace_entry_t pid;
+    pid.type = TRACE_TYPE_PID;
+    pid.size = 0;
+    pid.addr = static_cast<addr_t>(input.pid);
+
+    trace_entry_t tid;
+    tid.type = TRACE_TYPE_THREAD;
+    tid.size = 0;
+    tid.addr = static_cast<addr_t>(input.tid);
+
+    input.queue.push_front(pid);
+    input.queue.push_front(tid);
+}
+
 /***************************************************************************
  * Scheduled stream.
  */
@@ -1470,7 +1565,8 @@ scheduler_tmpl_t<RecordType, ReaderType>::get_tid(output_ordinal_t output)
     int index = outputs_[output].cur_input;
     if (index < 0)
         return -1;
-    if (inputs_[index].is_combined_stream())
+    if (inputs_[index].is_combined_stream() ||
+        TESTANY(OFFLINE_FILE_TYPE_CORE_SHARDED, inputs_[index].reader->get_filetype()))
         return inputs_[index].last_record_tid;
     return inputs_[index].tid;
 }
@@ -2035,6 +2131,10 @@ scheduler_tmpl_t<RecordType, ReaderType>::set_cur_input(output_ordinal_t output,
         outputs_[output].stream->filetype_ = inputs_[input].reader->get_filetype();
     }
 
+    if (inputs_[input].pid != INVALID_PID) {
+        insert_switch_tid_pid(inputs_[input]);
+    }
+
     if (!switch_sequence_.empty() &&
         outputs_[output].stream->get_instruction_ordinal() > 0) {
         sched_type_t::switch_type_t switch_type = SWITCH_INVALID;
@@ -2583,7 +2683,8 @@ scheduler_tmpl_t<RecordType, ReaderType>::next_record(output_ordinal_t output,
                     } else {
                         input->switch_to_input = it->second;
                     }
-                } else if (record_type_is_instr(record)) {
+                } else if (record_type_is_instr_boundary(record,
+                                                         outputs_[output].last_record)) {
                     if (syscall_incurs_switch(input, blocked_time)) {
                         // Model as blocking and should switch to a different input.
                         need_new_input = true;
@@ -2641,7 +2742,8 @@ scheduler_tmpl_t<RecordType, ReaderType>::next_record(output_ordinal_t output,
                 }
             }
             if (options_.quantum_unit == QUANTUM_INSTRUCTIONS &&
-                record_type_is_instr(record) && !outputs_[output].in_kernel_code) {
+                record_type_is_instr_boundary(record, outputs_[output].last_record) &&
+                !outputs_[output].in_kernel_code) {
                 ++input->instrs_in_quantum;
                 if (input->instrs_in_quantum > options_.quantum_duration) {
                     // We again prefer to switch to another input even if the current
@@ -2669,7 +2771,7 @@ scheduler_tmpl_t<RecordType, ReaderType>::next_record(output_ordinal_t output,
                     // We only switch on instruction boundaries.  We could possibly switch
                     // in between (e.g., scatter/gather long sequence of reads/writes) by
                     // setting input->switching_pre_instruction.
-                    record_type_is_instr(record)) {
+                    record_type_is_instr_boundary(record, outputs_[output].last_record)) {
                     VPRINT(this, 4,
                            "next_record[%d]: hit end of time quantum after %" PRIu64 "\n",
                            output, input->time_spent_in_quantum);
@@ -2715,7 +2817,8 @@ scheduler_tmpl_t<RecordType, ReaderType>::next_record(output_ordinal_t output,
                        prev_input, outputs_[output].cur_input);
                 if (!preempt) {
                     if (options_.quantum_unit == QUANTUM_INSTRUCTIONS &&
-                        record_type_is_instr(record)) {
+                        record_type_is_instr_boundary(record,
+                                                      outputs_[output].last_record)) {
                         --inputs_[prev_input].instrs_in_quantum;
                     } else if (options_.quantum_unit == QUANTUM_TIME) {
                         inputs_[prev_input].time_spent_in_quantum -=
@@ -2764,6 +2867,7 @@ scheduler_tmpl_t<RecordType, ReaderType>::next_record(output_ordinal_t output,
 
     outputs_[output].last_record = record;
     record_type_has_tid(record, input->last_record_tid);
+    record_type_has_pid(record, input->pid);
     return sched_type_t::STATUS_OK;
 }
 
@@ -2784,6 +2888,9 @@ scheduler_tmpl_t<RecordType, ReaderType>::unread_last_record(output_ordinal_t ou
     VPRINT(this, 4, "next_record[%d]: unreading last record, from %d\n", output,
            input->index);
     input->queue.push_back(outinfo.last_record);
+    // XXX: This should be record_type_is_instr_boundary() but we don't have the pre-prev
+    // record.  For now we don't support unread_last_record() for record_reader_t,
+    // enforced in a specialization of unread_last_record().
     if (options_.quantum_unit == QUANTUM_INSTRUCTIONS && record_type_is_instr(record))
         --input->instrs_in_quantum;
     outinfo.last_record = create_invalid_record();

clients/drcachesim/scheduler/scheduler.h

@@ -1132,6 +1132,8 @@ template <typename RecordType, typename ReaderType> class scheduler_tmpl_t {
             : lock(new std::mutex)
         {
         }
+        // Returns whether the stream mixes threads (online analysis mode) yet
+        // wants to treat them as separate shards (so not core-sharded-on-disk).
         bool
         is_combined_stream()
         {
@@ -1152,6 +1154,8 @@ template <typename RecordType, typename ReaderType> class scheduler_tmpl_t {
         int workload = -1;
         // If left invalid, this is a combined stream (online analysis mode).
         memref_tid_t tid = INVALID_THREAD_ID;
+        memref_pid_t pid = INVALID_PID;
+        // Used for combined streams.
         memref_tid_t last_record_tid = INVALID_THREAD_ID;
         // If non-empty these records should be returned before incrementing the reader.
         // This is used for read-ahead and inserting synthetic records.
@@ -1437,6 +1441,10 @@ template <typename RecordType, typename ReaderType> class scheduler_tmpl_t {
     bool
     record_type_has_tid(RecordType record, memref_tid_t &tid);
 
+    // If the given record has a process id field, returns true and the value.
+    bool
+    record_type_has_pid(RecordType record, memref_pid_t &pid);
+
     // For trace_entry_t, only sets the tid for record types that have it.
     void
     record_type_set_tid(RecordType &record, memref_tid_t tid);
@@ -1456,6 +1464,12 @@ template <typename RecordType, typename ReaderType> class scheduler_tmpl_t {
     bool
     record_type_is_invalid(RecordType record);
 
+    bool
+    record_type_is_encoding(RecordType record);
+
+    bool
+    record_type_is_instr_boundary(RecordType record, RecordType prev_record);
+
     // Creates the marker we insert between regions of interest.
     RecordType
     create_region_separator_marker(memref_tid_t tid, uintptr_t value);
@@ -1467,6 +1481,11 @@ template <typename RecordType, typename ReaderType> class scheduler_tmpl_t {
     RecordType
     create_invalid_record();
 
+    // If necessary, inserts context switch info on the incoming pid+tid.
+    // The lock for 'input' is held by the caller.
+    void
+    insert_switch_tid_pid(input_info_t &input);
+
     // Used for diagnostics: prints record fields to stderr.
     void
     print_record(const RecordType &record);

clients/drcachesim/tests/core_on_disk.templatex

@@ -3,27 +3,19 @@ Total counts:
 .*
            8 total threads
 .*
-Core [0-5] counts:
+Core [0-3] counts:
 .*
            2 threads
 .*
-Core [0-5] counts:
+Core [0-3] counts:
 .*
            2 threads
 .*
-Core [0-5] counts:
+Core [0-3] counts:
 .*
            2 threads
 .*
-Core [0-5] counts:
+Core [0-3] counts:
 .*
-           1 threads
-.*
-Core [0-5] counts:
-.*
-           1 threads
-.*
-Core [0-5] counts:
-.*
-           1 threads
+           2 threads
 .*

clients/drcachesim/tests/core_on_disk_schedule.templatex

@@ -0,0 +1,5 @@
+.*
+Core #0 schedule: (FJ_|EK_|GI|CHC_C__)
+Core #1 schedule: (FJ_|EK_|GI|CHC_C__)
+Core #2 schedule: (FJ_|EK_|GI|CHC_C__)
+Core #3 schedule: (FJ_|EK_|GI|CHC_C__)