Merge remote-tracking branch 'github/master' into i5036-aarch64-scatt…

…er-gather-multivector

Change-Id: Ia157d828be53d49cb2af574e72ac5c7d7bf6a775

api/samples/bbsize.c

@@ -61,7 +61,7 @@ static void
 event_exit(void);
 static dr_emit_flags_t
 event_bb_analysis(void *drcontext, void *tag, instrlist_t *bb, bool for_trace,
-                  bool translating, OUT void **user_data);
+                  bool translating, DR_PARAM_OUT void **user_data);
 
 DR_EXPORT void
 dr_client_main(client_id_t id, int argc, const char *argv[])
@@ -110,7 +110,7 @@ event_exit(void)
 
 static dr_emit_flags_t
 event_bb_analysis(void *drcontext, void *tag, instrlist_t *bb, bool for_trace,
-                  bool translating, OUT void **user_data)
+                  bool translating, DR_PARAM_OUT void **user_data)
 {
     int cur_size = 0;
 

api/samples/callstack.cpp

@@ -86,7 +86,7 @@ print_qualified_function_name(app_pc pc)
 }
 
 static void
-wrap_pre(void *wrapcxt, OUT void **user_data)
+wrap_pre(void *wrapcxt, DR_PARAM_OUT void **user_data)
 {
     dr_fprintf(STDERR, "%s called from:\n", trace_function.get_value().c_str());
     // Get the context.  The pc field is set by drwrap to the wrapped function

api/samples/div.c

@@ -106,7 +106,7 @@ callback(app_pc addr, uint divisor)
 
 /* If instr is unsigned division, return true and set *opnd to divisor. */
 static bool
-instr_is_div(instr_t *instr, OUT opnd_t *opnd)
+instr_is_div(instr_t *instr, DR_PARAM_OUT opnd_t *opnd)
 {
     int opc = instr_get_opcode(instr);
 #if defined(X86)

api/samples/hot_bbcount.c

@@ -129,7 +129,7 @@ set_up_bb_dups(void *drbbdup_ctx, void *drcontext, void *tag, instrlist_t *bb,
 
 static void
 analyse_orig_bb(void *drcontext, void *tag, instrlist_t *bb, void *user_data,
-                IN void **orig_analysis_data)
+                DR_PARAM_IN void **orig_analysis_data)
 {
     /* Extract bb_pc and store it as analysis data. */
     app_pc *bb_pc = dr_thread_alloc(drcontext, sizeof(app_pc));

api/samples/memtrace_simple.c

@@ -121,10 +121,6 @@ static int tls_idx;
 
 #define MINSERT instrlist_meta_preinsert
 
-#ifdef AARCH64
-static bool reported_sg_warning = false;
-#endif
-
 static void
 memtrace(void *drcontext)
 {
@@ -320,43 +316,13 @@ event_app_instruction(void *drcontext, void *tag, instrlist_t *bb, instr_t *wher
     for (i = 0; i < instr_num_srcs(instr_operands); i++) {
         const opnd_t src = instr_get_src(instr_operands, i);
         if (opnd_is_memory_reference(src)) {
-#ifdef AARCH64
-            /* TODO i#5036: Memory references involving SVE registers are not
-             * supported yet. To be implemented as part of scatter/gather work.
-             */
-            if (opnd_is_base_disp(src) &&
-                (reg_is_z(opnd_get_base(src)) || reg_is_z(opnd_get_index(src)))) {
-                if (!reported_sg_warning) {
-                    dr_fprintf(STDERR,
-                               "WARNING: Scatter/gather is not supported, results will "
-                               "be inaccurate\n");
-                    reported_sg_warning = true;
-                }
-                continue;
-            }
-#endif
             instrument_mem(drcontext, bb, where, src, false);
         }
     }
 
     for (i = 0; i < instr_num_dsts(instr_operands); i++) {
         const opnd_t dst = instr_get_dst(instr_operands, i);
         if (opnd_is_memory_reference(dst)) {
-#ifdef AARCH64
-            /* TODO i#5036: Memory references involving SVE registers are not
-             * supported yet. To be implemented as part of scatter/gather work.
-             */
-            if (opnd_is_base_disp(dst) &&
-                (reg_is_z(opnd_get_base(dst)) || reg_is_z(opnd_get_index(dst)))) {
-                if (!reported_sg_warning) {
-                    dr_fprintf(STDERR,
-                               "WARNING: Scatter/gather is not supported, results will "
-                               "be inaccurate\n");
-                    reported_sg_warning = true;
-                }
-                continue;
-            }
-#endif
             instrument_mem(drcontext, bb, where, dst, true);
         }
     }

api/samples/memval_simple.c

@@ -104,10 +104,6 @@ static int tls_idx;
 static drx_buf_t *write_buffer;
 static drx_buf_t *trace_buffer;
 
-#ifdef AARCH64
-static bool reported_sg_warning = false;
-#endif
-
 /* Requires that hex_buf be at least as long as 2*memref->size + 1. */
 static char *
 write_hexdump(char *hex_buf, byte *write_base, mem_ref_t *mem_ref)
@@ -332,23 +328,6 @@ handle_post_write(void *drcontext, instrlist_t *ilist, instr_t *where, reg_id_t
                 DR_ASSERT_MSG(false, "Found inst with multiple memory destinations");
                 break;
             }
-
-#ifdef AARCH64
-            /* TODO i#5036: Memory references involving SVE registers are not
-             * supported yet. To be implemented as part of scatter/gather work.
-             */
-            if (opnd_is_base_disp(dst) &&
-                (reg_is_z(opnd_get_base(dst)) || reg_is_z(opnd_get_index(dst)))) {
-                if (!reported_sg_warning) {
-                    dr_fprintf(STDERR,
-                               "WARNING: Scatter/gather is not supported, results "
-                               "will be inaccurate\n");
-                    reported_sg_warning = true;
-                }
-                continue;
-            }
-#endif
-
             seen_memref = true;
             instrument_post_write(drcontext, ilist, where, dst, prev_instr, reg_addr);
         }
@@ -404,21 +383,6 @@ event_app_instruction(void *drcontext, void *tag, instrlist_t *bb, instr_t *wher
                     DR_ASSERT_MSG(false, "Found inst with multiple memory destinations");
                     break;
                 }
-#ifdef AARCH64
-                /* TODO i#5036: Memory references involving SVE registers are not
-                 * supported yet. To be implemented as part of scatter/gather work.
-                 */
-                if (opnd_is_base_disp(dst) &&
-                    (reg_is_z(opnd_get_base(dst)) || reg_is_z(opnd_get_index(dst)))) {
-                    if (!reported_sg_warning) {
-                        dr_fprintf(STDERR,
-                                   "WARNING: Scatter/gather is not supported, results "
-                                   "will be inaccurate\n");
-                        reported_sg_warning = true;
-                    }
-                    continue;
-                }
-#endif
                 data->reg_addr = instrument_pre_write(
                     drcontext, bb, where, data->last_opcode, instr_operands, dst);
                 seen_memref = true;

api/samples/opcode_count.cpp

@@ -111,7 +111,7 @@ event_opcode_instruction(void *drcontext, void *tag, instrlist_t *bb, instr_t *i
 
 static dr_emit_flags_t
 event_bb_analysis(void *drcontext, void *tag, instrlist_t *bb, bool for_trace,
-                  bool translating, OUT void **user_data)
+                  bool translating, DR_PARAM_OUT void **user_data)
 {
     intptr_t bb_size = (intptr_t)drx_instrlist_app_size(bb);
     *user_data = (void *)bb_size;

api/samples/ssljack.c

@@ -51,9 +51,9 @@
 
 /* OpenSSL and GnuTLS - shared wrappers */
 static void
-wrap_pre_SSL_write(void *wrapcxt, OUT void **user_data);
+wrap_pre_SSL_write(void *wrapcxt, DR_PARAM_OUT void **user_data);
 static void
-wrap_pre_SSL_read(void *wrapcxt, OUT void **user_data);
+wrap_pre_SSL_read(void *wrapcxt, DR_PARAM_OUT void **user_data);
 static void
 wrap_post_SSL_read(void *wrapcxt, void *user_data);
 
@@ -128,7 +128,7 @@ dr_init(client_id_t id)
 }
 
 static void
-wrap_pre_SSL_write(void *wrapcxt, OUT void **user_data)
+wrap_pre_SSL_write(void *wrapcxt, DR_PARAM_OUT void **user_data)
 {
     /* int SSL_write(SSL *ssl, const void *buf, int num);
      *
@@ -161,7 +161,7 @@ wrap_pre_SSL_write(void *wrapcxt, OUT void **user_data)
 }
 
 static void
-wrap_pre_SSL_read(void *wrapcxt, OUT void **user_data)
+wrap_pre_SSL_read(void *wrapcxt, DR_PARAM_OUT void **user_data)
 {
     struct SSL_read_data *sd;
 

api/samples/statecmp.c

@@ -65,7 +65,7 @@ error_callback(const char *msg, void *tag)
 
 static dr_emit_flags_t
 event_analysis(void *drcontext, void *tag, instrlist_t *bb, bool for_trace,
-               bool translating, OUT void **user_data)
+               bool translating, DR_PARAM_OUT void **user_data)
 {
 
     bool *side_effect_free = (bool *)dr_thread_alloc(drcontext, sizeof(bool));

api/samples/wrap.c

@@ -59,7 +59,7 @@
 static void
 event_exit(void);
 static void
-wrap_pre(void *wrapcxt, OUT void **user_data);
+wrap_pre(void *wrapcxt, DR_PARAM_OUT void **user_data);
 static void
 wrap_post(void *wrapcxt, void *user_data);
 
@@ -141,7 +141,7 @@ event_exit(void)
 }
 
 static void
-wrap_pre(void *wrapcxt, OUT void **user_data)
+wrap_pre(void *wrapcxt, DR_PARAM_OUT void **user_data)
 {
     /* malloc(size) or HeapAlloc(heap, flags, size) */
     size_t sz = (size_t)drwrap_get_arg(wrapcxt, IF_WINDOWS_ELSE(2, 0));

clients/drcachesim/CMakeLists.txt

@@ -748,6 +748,9 @@ add_subdirectory(tools/external)
 # We build larger executables here.  All tests are added in suite/tests/ except unit tests.
 # Be sure to give the targets qualified test names ("tool.drcache*...").
 
+# XXX: Try to add a macro add_drcachesim_test() to share common pieces
+# of these executables.
+
 if (BUILD_TESTS)
   add_executable(tool.reuse_distance.unit_tests tests/reuse_distance_test.cpp)
   target_link_libraries(tool.reuse_distance.unit_tests drmemtrace_reuse_distance
@@ -907,6 +910,16 @@ if (BUILD_TESTS)
     set_tests_properties(tool.drcachesim.invariant_checker_test PROPERTIES
       TIMEOUT ${test_seconds})
 
+    add_executable(tool.drcachesim.schedule_stats_test tests/schedule_stats_test.cpp)
+    configure_DynamoRIO_standalone(tool.drcachesim.schedule_stats_test)
+    add_win32_flags(tool.drcachesim.schedule_stats_test)
+    target_link_libraries(tool.drcachesim.schedule_stats_test drmemtrace_schedule_stats
+        drmemtrace_static drmemtrace_analyzer test_helpers)
+    add_test(NAME tool.drcachesim.schedule_stats_test
+             COMMAND tool.drcachesim.schedule_stats_test)
+    set_tests_properties(tool.drcachesim.schedule_stats_test PROPERTIES
+      TIMEOUT ${test_seconds})
+
     add_executable(tool.drcacheoff.view_test tests/view_test.cpp reader/file_reader.cpp)
     configure_DynamoRIO_standalone(tool.drcacheoff.view_test)
     add_win32_flags(tool.drcacheoff.view_test)

clients/drcachesim/analyzer.cpp

@@ -492,6 +492,7 @@ analyzer_tmpl_t<RecordType, ReaderType>::process_shard_exit(
 {
     VPRINT(this, 1, "Worker %d finished trace shard %s\n", worker->index,
            worker->stream->get_stream_name().c_str());
+    worker->shard_data[shard_index].exited = true;
     if (interval_microseconds_ != 0 &&
         !process_interval(worker->shard_data[shard_index].cur_interval_index,
                           worker->shard_data[shard_index].cur_interval_init_instr_count,
@@ -560,6 +561,7 @@ analyzer_tmpl_t<RecordType, ReaderType>::process_tasks(analyzer_worker_data_t *w
                     tools_[i]->parallel_shard_init_stream(
                         shard_index, user_worker_data[i], worker->stream);
             }
+            worker->shard_data[shard_index].shard_index = shard_index;
         }
         memref_tid_t tid;
         if (worker->shard_data[shard_index].shard_id == 0) {
@@ -597,6 +599,12 @@ analyzer_tmpl_t<RecordType, ReaderType>::process_tasks(analyzer_worker_data_t *w
         if (!process_shard_exit(worker, worker->index))
             return;
     }
+    for (const auto &keyval : worker->shard_data) {
+        if (!keyval.second.exited) {
+            if (!process_shard_exit(worker, keyval.second.shard_index))
+                return;
+        }
+    }
     for (int i = 0; i < num_tools_; ++i) {
         const std::string error = tools_[i]->parallel_worker_exit(user_worker_data[i]);
         if (!error.empty()) {

clients/drcachesim/analyzer.h

@@ -168,10 +168,12 @@ template <typename RecordType, typename ReaderType> class analyzer_tmpl_t {
 
         uint64_t cur_interval_index;
         uint64_t cur_interval_init_instr_count;
-        // Identifier for the shard. Currently, shards map only to threads, so this is
-        // the thread id.
+        // Identifier for the shard (thread or core id).
         int64_t shard_id;
+        // Ordinal for the shard.
+        int shard_index = 0;
         std::vector<analyzer_tool_shard_data_t> tool_data;
+        bool exited = false;
 
     private:
         // Delete copy constructor and assignment operator to avoid overhead of

clients/drcachesim/analyzer_multi.cpp

@@ -257,6 +257,8 @@ analyzer_multi_t::init_dynamic_schedule()
     sched_ops.quantum_duration = op_sched_quantum.get_value();
     if (op_sched_time.get_value())
         sched_ops.quantum_unit = scheduler_t::QUANTUM_TIME;
+    sched_ops.syscall_switch_threshold = op_sched_syscall_switch_us.get_value();
+    sched_ops.blocking_switch_threshold = op_sched_blocking_switch_us.get_value();
 #ifdef HAS_ZIP
     if (!op_record_file.get_value().empty()) {
         record_schedule_zip_.reset(new zipfile_ostream_t(op_record_file.get_value()));

clients/drcachesim/common/named_pipe.h

@@ -56,11 +56,11 @@ typedef int ssize_t;
 #    endif
 #endif
 
-#ifndef OUT
-#    define OUT // nothing
+#ifndef DR_PARAM_OUT
+#    define DR_PARAM_OUT // nothing
 #endif
-#ifndef IN
-#    define IN // nothing
+#ifndef DR_PARAM_IN
+#    define DR_PARAM_IN // nothing
 #endif
 
 // Usage is as follows:
@@ -98,12 +98,12 @@ class named_pipe_t {
     // Returns < 0 on EOF or an error.
     // On success (or partial read) returns number of bytes read.
     ssize_t
-    read(void *buf OUT, size_t sz);
+    read(void *buf DR_PARAM_OUT, size_t sz);
 
     // Returns < 0 on an error.
     // On success (or partial write) returns number of bytes written.
     ssize_t
-    write(const void *buf IN, size_t sz);
+    write(const void *buf DR_PARAM_IN, size_t sz);
 
 #ifdef UNIX
     // On UNIX, rather than calling open_for_{read,write}, The caller

clients/drcachesim/common/named_pipe_unix.cpp

@@ -192,7 +192,7 @@ named_pipe_t::set_fd(int fd)
 }
 
 ssize_t
-named_pipe_t::read(void *buf OUT, size_t sz)
+named_pipe_t::read(void *buf DR_PARAM_OUT, size_t sz)
 {
     int res = -1;
     while (true) {
@@ -211,7 +211,7 @@ named_pipe_t::read(void *buf OUT, size_t sz)
 }
 
 ssize_t
-named_pipe_t::write(const void *buf IN, size_t sz)
+named_pipe_t::write(const void *buf DR_PARAM_IN, size_t sz)
 {
     int res = -1;
     while (true) {

clients/drcachesim/common/named_pipe_win.cpp

@@ -145,7 +145,7 @@ named_pipe_t::maximize_buffer()
 }
 
 ssize_t
-named_pipe_t::read(void *buf OUT, size_t sz)
+named_pipe_t::read(void *buf DR_PARAM_OUT, size_t sz)
 {
     DWORD actual;
     BOOL res = ReadFile(fd_, buf, (DWORD)sz, &actual, NULL);
@@ -156,7 +156,7 @@ named_pipe_t::read(void *buf OUT, size_t sz)
 }
 
 ssize_t
-named_pipe_t::write(const void *buf IN, size_t sz)
+named_pipe_t::write(const void *buf DR_PARAM_IN, size_t sz)
 {
     DWORD actual;
     BOOL res = WriteFile(fd_, buf, (DWORD)sz, &actual, NULL);

clients/drcachesim/common/options.cpp

@@ -838,6 +838,19 @@ droption_t<bool> op_sched_order_time(DROPTION_SCOPE_ALL, "sched_order_time", tru
                                      "Applies to -core_sharded and -core_serial. "
                                      "Whether to honor recorded timestamps for ordering");
 
+droption_t<uint64_t> op_sched_syscall_switch_us(
+    DROPTION_SCOPE_ALL, "sched_syscall_switch_us", 500,
+    "Minimum latency to consider any syscall as incurring a context switch.",
+    "Minimum latency in timestamp units (us) to consider any syscall as incurring "
+    "a context switch.  Applies to -core_sharded and -core_serial. ");
+
+droption_t<uint64_t> op_sched_blocking_switch_us(
+    DROPTION_SCOPE_ALL, "sched_blocking_switch_us", 100,
+    "Minimum latency to consider a maybe-blocking syscall as incurring a context switch.",
+    "Minimum latency in timestamp units (us) to consider any syscall that is marked as "
+    "maybe-blocking to incur a context switch. Applies to -core_sharded and "
+    "-core_serial. ");
+
 #ifdef HAS_ZIP
 droption_t<std::string> op_record_file(DROPTION_SCOPE_FRONTEND, "record_file", "",
                                        "Path for storing record of schedule",

clients/drcachesim/common/options.h

@@ -191,6 +191,8 @@ extern dynamorio::droption::droption_t<bool> op_core_serial;
 extern dynamorio::droption::droption_t<int64_t> op_sched_quantum;
 extern dynamorio::droption::droption_t<bool> op_sched_time;
 extern dynamorio::droption::droption_t<bool> op_sched_order_time;
+extern dynamorio::droption::droption_t<uint64_t> op_sched_syscall_switch_us;
+extern dynamorio::droption::droption_t<uint64_t> op_sched_blocking_switch_us;
 #ifdef HAS_ZIP
 extern dynamorio::droption::droption_t<std::string> op_record_file;
 extern dynamorio::droption::droption_t<std::string> op_replay_file;