kernel_tuner
kernel_tuner copied to clipboard
KernelTuner
Kernel Tuner
Trying to use this library on Windows 11, Python 3.10, I receive: ``` Traceback (most recent call last): File "C:\Users\\kernel_tuner\examples\cuda\matmul.py", line 10, in import kernel_tuner File "C:\Users\\kernel_tuner\kernel_tuner\__init__.py", line 4, in...
These lines in cost_func: https://github.com/benvanwerkhoven/kernel_tuner/blob/master/kernel_tuner/strategies/minimize.py#L121-L130 Come after the config has been stored in the cache, which means the times computed here do not end up in the cachefile. Also, brute_force...
Using the PythonKernel in a test case gave some strange issues, and replacing these with calls to run_kernel solved the problem. So there is a bug somewhere in the PythonKernel...
It appears there are some HPC systems where users that are allowed to install software using conda, but not pip. It would be great if people could install Kernel Tuner...
We are planning to develop a HIP backend for Kernel Tuner, to be able to tune kernels written for HIP just as easily as CUDA kernels. One issue with creating...
I wanted to announce that soon we will be moving this repository from my private GitHub account to a new GitHub organization created specifically for Kernel Tuner: https://github.com/KernelTuner/ We have...
Nvidia's official Python bindings for CUDA: https://github.com/NVIDIA/cuda-python are in the end the most future-proof Python bindings for Kernel Tuner to use. @isazi already started working on a backend for Kernel...
The following code throws an exception. ```python import kernel_tuner source = """ __global__ void foo() { } """ problem_size = 100 args = [] tune_params = dict( block_size_x=[1, 64], block_size_y=[1,...
For the BAT benchmark suite we would like to use the cupy backend, but most benchmarks use texture memory and this isn't supported yet in Kernel Tuner's cupy backend.
Change the python logging code. Each python file now logs using its own logger instance with the filename (from __name__) as name. This should solve some issues with logging not...
