What is the Composable Kernel library

What is the Composable Kernel library#

Methodology#

The Composable Kernel (CK) library provides a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs and CPUs, through general purpose kernel languages like HIP C++.

CK utilizes two concepts to achieve performance portability and code maintainability:

  • A tile-based programming model

  • Algorithm complexity reduction for complex ML operators using an innovative technique called “Tensor Coordinate Transformation”.

CK Components

Code Structure#

The CK library is structured into 4 layers:

  • “Templated Tile Operators” layer

  • “Templated Kernel and Invoker” layer

  • “Instantiated Kernel and Invoker” layer

  • “Client API” layer

It also includes a simple wrapper component used to perform tensor transform operations more easily and with fewer lines of code.

CK Layers
Contents