aclmdlInitDump

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product/Atlas A3 inference product	√
Atlas A2 training product/Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	√
Atlas inference product	√
Atlas training product	√

Description

Initializes dump.

This API needs to be used with other APIs to implement the following functions:

Record dump data to files.
aclmdlInitDump, aclmdlSetDump, aclmdlFinalizeDump work together to record dump data to files. These APIs can be called for multiple times in a single process to obtain dump data of different dump configurations. The specific scenarios are exemplified as follows:
- To execute two different models, you need to set dump configurations differently. The API call sequence is as follows: aclInit --> aclmdlInitDump --> aclmdlSetDump --> model 1 loading --> model 1 execution --> aclmdlFinalizeDump --> model 1 unloading --> aclmdlInitDump --> aclmdlSetDump --> model 2 loading --> model 2 execution --> aclmdlFinalizeDump --> model 2 unloading --> execution of other tasks --> aclFinalize
- To execute the same model twice, you only need to perform the dump operation for the first execution. The API call sequence is as follows: aclInit --> aclmdlInitDump --> aclmdlSetDump --> model loading --> model execution --> aclmdlFinalizeDump --> model unloading --> model loading --> model execution --> execution of other tasks --> aclFinalize
Do not record dump data to files and directly obtain dump data by using a callback function.
aclmdlInitDump, acldumpRegCallback, acldumpUnregCallback, and aclmdlFinalizeDump work together to obtain dump data through a callback function. The callback function registered through acldumpRegCallback needs to be implemented by users, and the implementation logic of the callback function must include the acquisition of the dump data and data length. The specific scenarios are exemplified as follows:
- Execute a model and obtain dump data through callback.
  aclInit --> acldumpRegCallback --> aclmdlInitDump --> model loading --> model execution --> aclmdlFinalizeDump --> acldumpUnregCallback --> model unloading --> aclFinalize
- Execute two different models and obtain dump data through callback. In this context, as long as the callback function is not deregistered by calling acldumpUnregCallback, you can obtain the dump data of both models through the callback function.
  aclInit --> acldumpRegCallback --> aclmdlInitDump --> model 1 loading --> model 1 execution -->--> model 2 loading --> model 2 execution --> aclmdlFinalizeDump --> model unloading --> acldumpUnregCallback --> aclFinalize

Prototype

aclError aclmdlInitDump()

Parameters

None

Returns

0 on success; otherwise, failure. For details, see aclError.

Restrictions

If information about the model, single-operator, and overflow/underflow operator dump configurations has been configured by the aclInit call, a failure message is returned when aclmdlInitDump is called.
aclmdlInitDump must be called after the aclInit call and before the model loading API call.

Reference

Currently, the aclInit API is also provided. During initialization, the dump configuration is passed as a JSON configuration file to dump the app data at runtime. In this mode, aclInit can be called only once in a process. To change the dump configuration, modify the JSON configuration file.

Parent topic: Dump Configuration