Mojo module
conv_config
Configuration types for SM100 structured convolution kernels.
This module provides:
- Conv2dProblemShape: Defines the convolution problem geometry
- Conv2dConfig: Kernel configuration (tile sizes, pipeline depths, etc.)
The convolution is mapped to GEMM as implicit im2col:
- M = N_batch * H_out * W_out (output spatial)
- N = K_filters (output channels)
- K = C_in * R * S (input channels * filter spatial)
Structs
-
Conv2dConfig: Configuration for SM100 Conv2D kernel. -
Conv2dProblemShape: Defines 2D convolution problem geometry.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!