how to only pad the start when performing 1D convolution

isaaclee2313 · February 19, 2019, 2:36am

The https://docs.nvidia.com/deeplearning/sdk/cudnn-developer-guide/index.html#cudnnSetConvolutionNdDescriptor documentation for

cudnnSetConvolutionNdDescriptor

states that padding is applied to both the start and the end. Is there a way to only apply padding at the start?

Can I do this manually? ( i.e. if I need padding of 511 zeroes, I would just append that many at the start of the input tensor by hand in the cudaMalloc-ed input array )

Topic		Replies	Views
Question about the "SAME" convolution in cuDNN cuDNN	0	459	March 11, 2019
How to add padding to a convolution layer on tensorrt. TensorRT	3	1360	March 22, 2019
cuDNN for 3D convolution cuDNN cudnn	1	1831	October 24, 2023
[cuDNN] Adding a bias to convolution using cudnnAddTensor4d correctly GPU-Accelerated Libraries	0	1855	December 18, 2014
How use the cudnn graph API for do a convolution cuDNN cudnn	2	49	February 15, 2025
How to set convolution math type in cuDNN 9.2.0? cuDNN cudnn	0	104	June 16, 2024
Best practice of cuDNN implementation cuDNN	1	499	February 15, 2021
Cudnn backend api for fused op cuDNN cudnn	8	2148	September 13, 2021
Problem with cudnnAddTensor() function GPU-Accelerated Libraries	3	2600	September 3, 2015
Convolution with cuDNN CUDA Programming and Performance	7	1989	January 10, 2018

how to only pad the start when performing 1D convolution

Related topics