search

TiledTensor / TiledCUDA

TiledCUDA is a highly efficient kernel template library designed to elevate CUDA C’s level of abstraction for processing tiles.
MIT License
157 stars 10 forks source link

Re-design and Re-implement the swizzled shared memory layout. #104

Closed haruhi55 closed 3 months ago

haruhi55 commented 3 months ago

Shared memory swizzling is strictly performed within a 16x16 BaseTile. These PRs gradually refactor the implementations for shared memory swizzling.

https://github.com/TiledTensor/TiledCUDA/pull/110 https://github.com/TiledTensor/TiledCUDA/pull/115