CK team works closely with Meta [AITemplate](???to.be.added???) team ([Bing Xu](https://github.com/antinucleon), Ying Zhang, etc). Most of the lucrative graph optimization opportunities in ML models were identified by AITemplate team, and we also co-designed many high performance fused kernels for AMD GPUs. Without this collaboration, CK would not reach its current potential.
## Acknowledgement
CK team works closely with Meta [AITemplate](???to.be.added???) team ([Bing Xu](https://github.com/antinucleon), [Hao Lu](https://github.com/hlu1), [Ying Zhang](https://github.com/ipiszy), etc). Most of the lucrative graph optimization opportunities in ML models were identified by AITemplate team, and we also co-designed many high performance fused kernels for AMD GPUs. Without this collaboration, CK would not reach its current potential.
Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for Machine Learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
CK utilizes two concepts to achieve performance portabilatity and code maintainbility:
CK utilizes two concepts to achieve performance portability and code maintainability:
* A tile-based programming model
* A tile-based programming model
* Algorithm complexity reduction for complex ML operators, using innovative technique we call "Tensor Coordinate Transformation".
* Algorithm complexity reduction for complex ML operators, using innovative technique we call "Tensor Coordinate Transformation".
...
@@ -11,7 +11,7 @@ CK utilizes two concepts to achieve performance portabilatity and code maintainb
...
@@ -11,7 +11,7 @@ CK utilizes two concepts to achieve performance portabilatity and code maintainb
## Code Structure
## Code Structure
Current CK library are structured into 4 layers:
Current CK library are structured into 4 layers:
* "Templated Tile Operators"
* "Templated Tile Operators" layer
* "Templated Kernel and Invoker" layer
* "Templated Kernel and Invoker" layer
* "Instantiated Kernel and Invoker" layer
* "Instantiated Kernel and Invoker" layer
* "Client API" layer
* "Client API" layer
...
@@ -90,7 +90,7 @@ Instructions for using CK as a pre-built kernel library are under [client_exampl
...
@@ -90,7 +90,7 @@ Instructions for using CK as a pre-built kernel library are under [client_exampl
### Kernel Timing and Verification
### Kernel Timing and Verification
CK's own kernel timer will warn up kernel once, and then run it multiple times
CK's own kernel timer will warn up kernel once, and then run it multiple times
to get average kernel time. For some kernels that use atomic add, this will cause
to get average kernel time. For some kernels that use atomic add, this will cause
output buffer to be accumulated multiple times, causing verfication failure.
output buffer to be accumulated multiple times, causing verification failure.
To work around it, do not use CK's own timer and do verification at the same time.
To work around it, do not use CK's own timer and do verification at the same time.
CK's own timer and verification in each example and ckProfiler can be enabled or
CK's own timer and verification in each example and ckProfiler can be enabled or
