Exporting Profiling Results

You can print the profiling results on the screen, or use a script to save them in CSV format by taking the following steps:

1. Log in to the server as the HwHiAiUser user created during installation.
2. Go to the /home/HwHiAiUser/Ascend/ascend-toolkit/latest/toolkit/tools/profiler/profiler_tool/analysis/interface directory.
3. Export CSV files.

   The following is the command syntax of basic options. For details about others, see Scripts and Interfaces.

   python3.7.5 get_msvp_info.pyc --save_file --project=$output_data --deviceid=$device_id --data_type=$export_data_type

   For example, you can run the following command to export the Runtime API calls:

   python3.7.5 get_msvp_info.pyc --save_file --project=/home/HwHiAiUser/tools/out --deviceid=0 --data_type=runtime_api

   The options in the preceding command are described as follows:

   • --project: path for storing exported CSV files of the profiling results.
   • --deviceid: device ID.
   • --data_type: profiling results to export. For details, see Table 7-4.

   --deviceid needs to be set to a specific device ID when the profiling results in the path specified by --project are associated to multiple devices.

   In the exported ge_basic_{device_id}.csv, ai_core_op_summary_{device_id}.csv and op_counter_{device_id}.csv files, values in the Model Name column may be empty.

   Table 7-4 Profiling results and target CSV file names

   Profiling Result		data_type	Target CSV File
   Runtime API calls	-	--data_type=runtime_api	runtime_api_{device_id}.csv
   Task Scheduler	-	--data_type=task_scheduler	task_scheduler_{device_id}.csv
   AI Core	Task-based AI Core	--data_type=ai_core_pmu_events	ai_core_pmu_events_{device_id}.csv
   	Sample-based AI Core	--data_type=ai_core_pmu_events	ai_core_pmu_events_{device_id}.csv
   Operator data provided by the GE component	GE task and graph	--data_type=ge_basic	ge_basic_{device_id}.csv
   	Information of GE loading a model	--data_type=ge_model_load	ge_model_load_{device_id}.csv
   	Time taken by GE to load a model	--data_type=ge_model_time	ge_model_time_{device_id}.csv
   CPU PMU events and hotspot functions	Ctrl CPU PMU (Performance Monitoring Unit) events	--data_type=control_cpu_pmu_events	control_cpu_pmu_events_{device_id}.csv
   	Top 5 Ctrl CPU hotspot functions	--data_type=control_cpu_top_functions	control_cpu_top_functions_{device_id}.csv
   	AI CPU PMU events	--data_type=ai_cpu_pmu_events	ai_cpu_pmu_events_{device_id}.csv
   	Top 5 AI CPU hotspot functions	--data_type=ai_cpu_top_functions	ai_cpu_top_functions_{device_id}.csv
   	TS CPU PMU events	--data_type=ts_cpu_pmu_events	ts_cpu_pmu_events_{device_id}.csv
   Peripheral	NIC	--data_type=nic	nic_{device_id}.csv
   	DVPP	--data_type=dvpp	dvpp_{device_id}.csv
   Hardware	LLC bandwidth	--data_type=llc	llc_{device_id}.csv
   	LLC capacity (AI CPU)	--data_type=llc_aicpu	llc_aicpu_{device_id}.csv
   	LLC capacity (Ctrl CPU)	--data_type=llc_ctrlcpu	llc_ctrlcpu_{device_id}.csv
   	Read/write bandwidth (DDR)	--data_type=ddr	ddr_{device_id}.csv
   CPU utilization	System AI CPU and Ctrl CPU	--data_type=sys_cpu_usage	sys_cpu_usage_{device_id}.csv
   	Process CPU (top 50 data records)	--data_type=process_cpu_usage	process_cpu_usage_{device_id}.csv
   Memory	System Memory Summary	--data_type=sys_mem	sys_mem_{device_id}.csv
   	Process Memory Info	--data_type=process_mem	process_mem_{device_id}.csv
   AscendCL	-	--data_type=acl	acl_{device_id}.csv
   AI Core Operator Summary	-	--data_type=ai_core_op_summary	ai_core_op_summary_{device_id}.csv
   AI Core Operator Statistics	-	--data_type=op_counter	op_counter_{device_id}.csv
   Top-down information	-	--data_type=top_down	top_down_{device_id}.csv
   All profiling results	-	--data_type=all	all_data_{device_id}.csv