profiler init add aclrtSetDevice

mindspore/ccsrc/profiler/device/ascend/ascend_profiling.cc

@@ -28,6 +28,7 @@
 #include "plugin/device/ascend/hal/device/profiling/profiling_reporter.h"
 #include "kernel/kernel.h"
 #include "backend/common/session/kernel_graph.h"
+#include "acl/acl_rt.h"
 
 using mindspore::device::ascend::ProfilingManager;
 using mindspore::device::ascend::ProfilingReporter;
@@ -69,6 +70,11 @@ void AscendProfiler::InitProfiling(const std::string &profiling_path, uint32_t d
   profile_data_path_ = profiling_path;
   device_id_ = device_id;
 
+  aclError ret = aclrtSetDevice(static_cast<int32_t>(device_id));
+  if (ret != ACL_ERROR_NONE) {
+    MS_LOG(EXCEPTION) << "Device " << device_id << " call aclrtSetDevice failed, ret[" << static_cast<int>(ret) << "]";
+  }
+
   // Init ErrorManager instance in order to get error msg reported by Ascend.
   if (ErrorManager::GetInstance().Init() != 0) {
     MS_LOG(WARNING) << "[Internal Error] Failed to init ErrorManager class.";

mindspore/python/mindspore/profiler/parser/aicpu_data_parser.py

@@ -175,8 +175,10 @@ class DataPreProcessParser:
             ai_cpu = AiCpuStruct(*ai_cpu_data)
             if ai_cpu.task_id < self._task_id_threshold:
                 node_type_name = f'{ai_cpu.stream_id}_{ai_cpu.task_id}'
-                if self._op_task_dict:
+                if self._op_task_dict and node_type_name in self._op_task_dict:
                     node_type_name = self._op_task_dict[node_type_name].split('/')[-1]
+                else:
+                    logger.warning("[profiler] the op name of %s cannot be found.", node_type_name)
                 exe_time = (float(ai_cpu.run_end) - float(ai_cpu.run_start)) / self._ms_unit
                 total_time = ai_cpu.total_time / self._ms_unit
                 result_list.append([serial_number, node_type_name, total_time, ai_cpu.dispatch_time / self._ms_unit,

tests/st/profiler/test_profiler.py

@@ -177,10 +177,10 @@ class TestProfiler:
         if ds_train.get_dataset_size() == 0:
             raise ValueError("Please check dataset size > 0 and batch_size <= dataset size")
 
-        lenet = LeNet5()
         profiler = Profiler(profile_memory=profile_memory, output_path='data')
         profiler_name = os.listdir(os.path.join(os.getcwd(), 'data'))[0]
         self.profiler_path = os.path.join(os.getcwd(), f'data/{profiler_name}/')
+        lenet = LeNet5()
         loss = nn.SoftmaxCrossEntropyWithLogits(sparse=True, reduction="mean")
         optim = Momentum(lenet.trainable_params(), learning_rate=0.1, momentum=0.9)
         model = Model(lenet, loss_fn=loss, optimizer=optim, metrics={'acc': Accuracy()})

tests/ut/cpp/stub/runtime/acl_rt.cc

@@ -62,4 +62,17 @@ ACL_FUNC_VISIBILITY aclError aclrtMemcpy(void *dst, size_t destMax, const void *
 ACL_FUNC_VISIBILITY aclError aclrtMemcpyAsync(void *dst, size_t destMax, const void *src, size_t count,
                                               aclrtMemcpyKind kind, aclrtStream stream) {
   return ACL_ERROR_NONE;
-}
+}
+
+/**
+ * @ingroup AscendCL
+ * @brief Specify the device used for computing in the current process, and implicitly create a default context
+ *
+ * @param deviceId [IN]    the device id of the resource used by this process
+ *
+ * @retval ACL_SUCCESS The function is successfully executed.
+ * @retval OtherValues Failure
+ */
+ACL_FUNC_VISIBILITY aclError aclrtSetDevice(int32_t deviceId) {
+  return ACL_ERROR_NONE;
+}