-
KLARAPTOR is a compile-time tool for CUDA programs which constructs and evaluates a rational program encoding the MWP-CWP performance model to optimize the program’s performance. The tool dynamically chooses kernel launch parameters (thread block configurations) which optimize its performance for data and hardware parameters of CUDA kernels.
-
KLARAPTOR is built in the C programming language making use of (1) NVIDIA Nsight Compute CLI (ncu) to measure low-level metrics, (2) LAPACK for linear algebra, (3) BPAS (Basic Polynomial Algebra Subprograms) for efficient rational functions, and (4) LLVM Pass Framework to connect the constructed rational program and the CUDA program.
-
NVIDIA cuda 11.7 or later must be already intalled, also, variable CUDA_PATH should be part of your ~/.bashrc.
-
KLARAPTOR has only been tested on Ubuntu 20.04 for NVIDIA GPUs of the following compute capability:
- 7.5
-
After installation, you might need to reload nvidia modules. The easiest way is to reboot.
-
Authors
- Alexander Brandt abrandt5@uwo.ca
- Taabish Jeshani tjeshan@uwo.ca
- Davood Mohajerani mohajerani.d@gmail.com
- Marc Moreno Maza moreno@csd.uwo.ca
- Linxiao Wang lwang739@uwo.ca
-
Caveats: Will update with paper link once published.
-
For further install instructions please contact:
- Taabish Jeshani tjeshan@uwo.ca
-
Notifications
You must be signed in to change notification settings - Fork 1
KLARAPTOR: A Tool for Dynamically Finding Optimal Kernel Launch Parameters Targeting CUDA Programs
License
orcca-uwo/KLARAPTOR
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
About
KLARAPTOR: A Tool for Dynamically Finding Optimal Kernel Launch Parameters Targeting CUDA Programs
Topics
Resources
License
Stars
Watchers
Forks
Releases
No releases published
Packages 0
No packages published