Main Content

Auto tuning

Code generation that uses cuDNN library with auto tuning enabled

Since R2021a

Model Configuration Pane: Code Generation / Interface

Description

The Auto tuning parameter specifies whether the code generator applies auto tuning for the cuDNN library. Auto tuning increases performance for larger networks such as SegNet and ResNet.

Dependencies

  • This parameter requires a GPU Coder™ license.

  • To enable this parameter, set the Language parameter to C++ and the Target library parameter to cuDNN.

Settings

On (default) | Off
On

Enables auto tuning for the cuDNN library. Enabling auto tuning allows the library to find the fastest convolution algorithms.

Off

Disables auto tuning for the cuDNN library.

Recommended Settings

ApplicationSetting
Debugging 
Traceability 
Efficiency 
Safety precaution 

Programmatic Use

Parameter: DLAutoTuning
Value: 'on' | 'off'
Default: 'on'

Version History

Introduced in R2021a