Delayed Profiling

When using the msprof CLI to collect profile data, you can use the --delay and --duration options described in this section to configure the collection duration and delay.

Dynamic Profiling is not supported in delayed profiling scenarios.

Availability

Atlas 200/500 A2 Inference Product

Atlas Inference Series Product

Atlas Training Series Product

Atlas A2 Training Series Product/Atlas 800I A2 Inference Product

Atlas A3 Training Series Product

Precautions

  • Ensure that an AI task can run properly in the operating environment.
  • Ensure that operations in Before You Start have been completed.

The Python call stack, PyTorch or MindSpore framework layer data cannot be profiled. You can use the framework APIs to profile such data.

Command Example

Log in as the running user to the environment where the CANN Toolkit package and ops operator package are installed, and run the following command to collect profile data:

msprof --delay=3 --duration=3 /home/projects/MyApp/out/main

Delayed profiling is supported only when AI task runtime profiling is enabled, and it requires the passing of a user application. This option cannot be configured together with --dynamic.

Options

Table 1 Options

Option

Description

Required/Optional

--delay

Profiling delay. The value range is [1, 4294967295], in seconds. The default value is 0. If the configured time exceeds the execution time of an AI task, the collection will not be started during the execution of the AI job.

Optional

--duration

Profiling duration. The value range is [1, 4294967295], in seconds. By default, this option is not set. That is, the profiling automatically starts and stops with the task execution. If the --delay option is set, the duration starts from the time when the delay ends.

Optional