Mojo module
conv_tile_loader
Tile loader for SM100 convolution with hardware im2col TMA.
This module provides TileLoaderTMAIm2col, which uses TMA's im2col addressing mode to perform implicit GEMM convolution without explicit im2col buffers. The TMA descriptor encodes convolution geometry and transforms coordinates on-the-fly during memory loads.
Structs
-
TileLoaderTMAIm2col: TMA tile loader using hardware im2col for implicit GEMM convolution.
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!