mirror of
https://github.com/andreinechaev/nvcc4jupyter.git
synced 2026-06-13 18:50:47 +05:30
Ștefan-Cosmin Ciocan
887c809d07
* Use NVIDIA Nsight Compute CLI profiler * Add profile and profiler-args options to argument parser. * Add missing comma to profiler-args option. * Use profile args in version 1 of the plugin * Change profiler-args option to take all remaining arguments * Change profiler_args type from string to list of strings * Add profile option to version 2 of the plugin * Add profiler usage instructions
NVCC Plugin for Jupyter notebook
V2 is available
V2 brings support of multiple source and header files.
Usage
- Load Extension
%load_ext nvcc_plugin
- Mark a cell to be treated as cuda cell
%%cuda --name example.cu --compile falseNOTE: The cell must contain either code or comments to be run successfully. It accepts 2 arguments.
-n|--name- which is the name of either CUDA source or Header The name parameter must have extension.cuor.hSecond argument-c|--compile; default value isfalse. The argument is a flag to specify if the cell will be compiled and run right away or not. It might be usefull if you're playing in themainfunction
- To compile and run all CUDA files you need to run
%%cuda_run
# This line just to bypass an exeption and can contain any text
- To profile your CUDA kernels using NVIDIA Nsight Compute CLI profiler you need to run
%%cu --profile
- You can add options to the profiler. Keep in mind that any argument after "--profiler-args" will be considered as a profiler argument. For example, to select which sections to collect metrics for you need to run
%%cu --profile --profiler-args --section SpeedOfLight --section MemoryWorkloadAnalysis --section Occupancy