Questions regarding keywords in tensorRT log file

I was going over the tensorRT log while building engine and came upon these following keywords.
caskConvolution
CudaConvolution
CudadepthwiseConvolution
FusedConvActConvolution
LegacySASSConvolution

They were shown like this.
[TensorRT] VERBOSE: --------------- Timing Runner: FeatureExtractor/MobilenetV2/Conv/Conv2D + FeatureExtractor/MobilenetV2/Conv/Relu6 (LegacySASSConvolution)

What exactly do they mean? What are the differences between them?

Hi @joseph398,
These are part of Profiling.
During the build phase, all possible tactics are tried and timed. Profiling this portion of the execution will not show any meaningful performance measurements and will include all possible kernels, not the ones actually selected for inference.
The below links might be helpful in understanding the same.
https://docs.nvidia.com/deeplearning/tensorrt/best-practices/index.html#nvprof
https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#deter-repro

Thanks!

1 Like