Merge branch 'devel' into fabric_mem_rebased

argonne-lcf · Oct 1, 2024 · 0a175ba · 0a175ba
2 parents 02d92a2 + 41248b2
commit 0a175ba
Show file tree

Hide file tree

Showing 2 changed files with 17 additions and 6 deletions.
diff --git a/xprof/xprof.rb.in b/xprof/xprof.rb.in
@@ -385,11 +385,16 @@ def env_tracers
   end
 
   # Sample
-  if sampling? 
+  #   Currently the same `so` does the tracing, and the sampling
+  #   This mean that is the local rank is not part of the `traced-ranks`
+  #   No sampling will be performed
+  if sampling?
     LOGGER.debug('Sampling Enabled')
     h['LTTNG_UST_SAMPLING'] = 1
     h['LTTNG_UST_SAMPLING_ENERGY'] = 1
-    h['ZES_ENABLE_SYSMAN'] = 0  if OPTIONS[:'backend-names'].include?('ze')
+    # The current only reliable way to use zes api
+    #   is to call zesInit and set ZES_ENABLE_SYSMAN to 0
+    h['ZES_ENABLE_SYSMAN'] = 0 if OPTIONS[:'backend-names'].include?('ze')
   end
 
   backends = [] unless need_backend
@@ -522,7 +527,12 @@ end
 
 def enable_events_mpi(channel_name, tracing_mode: 'default', profiling: true)
   lttng_enable = "lttng enable-event --userspace --session=#{lttng_session_uuid} --channel=#{channel_name}"
-  exec("#{lttng_enable} lttng_ust_mpi:*")
+  case tracing_mode
+  when 'full'
+    exec("#{lttng_enable} lttng_ust_mpi:*")
+  when 'default'
+    exec("#{lttng_enable} lttng_ust_mpi:* -x lttng_ust_mpi:MPI_WTime*")
+  end
   exec("#{lttng_enable} lttng_ust_mpi_type:*")
 end
 

diff --git a/ze/tracer_ze_helpers.include.c b/ze/tracer_ze_helpers.include.c
@@ -1036,6 +1036,8 @@ static int initializeHandles() {
   _sampling_hSubDevices = (ze_device_handle_t***) calloc(_sampling_driverCount, sizeof(ze_device_handle_t**));
   for (uint32_t driverIdx = 0; driverIdx < _sampling_driverCount; driverIdx++) {
     _sampling_deviceCount[driverIdx] = 0;
+  // Query device count
+  for (uint32_t driverIdx = 0; driverIdx < _sampling_driverCount; driverIdx++) {
     res = ZES_DEVICE_GET_PTR(_sampling_hDrivers[driverIdx], &_sampling_deviceCount[driverIdx], NULL);
     if (res != ZE_RESULT_SUCCESS || _sampling_deviceCount[driverIdx] == 0) {
       fprintf(stderr, "ERROR: No device found!\n");
@@ -1053,7 +1055,6 @@ static int initializeHandles() {
     _sampling_subDeviceCount[driverIdx] = (uint32_t*) calloc(_sampling_deviceCount[driverIdx], sizeof(uint32_t));
     _sampling_hSubDevices[driverIdx] = (ze_device_handle_t**) calloc(_sampling_deviceCount[driverIdx], sizeof(ze_device_handle_t*));
     for (uint32_t deviceIdx = 0; deviceIdx < _sampling_deviceCount[driverIdx]; deviceIdx++) {
-     // zes_device_ext_properties_t deviceProps = {0};
       zes_device_properties_t deviceProps = {0};
       deviceProps.stype = ZES_STRUCTURE_TYPE_DEVICE_PROPERTIES;
       deviceProps.pNext = NULL;
@@ -1065,9 +1066,9 @@ static int initializeHandles() {
                       &deviceProps );
 
       _sampling_subDeviceCount[driverIdx][deviceIdx] = 0;
-      res = ZE_DEVICE_GET_SUB_DEVICES_PTR((ze_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], &_sampling_subDeviceCount[driverIdx][deviceIdx], NULL);
+      res = ZE_DEVICE_GET_SUB_DEVICES_PTR((ze_device_handle_t)_sampling_hDevices[driverIdx][deviceIdx], &_sampling_subDeviceCount[driverIdx][deviceIdx], NULL)
       if (res != ZE_RESULT_SUCCESS) {
-        _ZE_ERROR_MSG("ZE_DEVICE_GET_SUB_DEVICES_PTR", res);
+        _ZE_ERROR_MSG("ZES_DEVICE_GET_PROPERTIES_PTR", res);
         _sampling_subDeviceCount[driverIdx][deviceIdx] = 0;
       }
       if (_sampling_subDeviceCount[driverIdx][deviceIdx] > 0) {