The code in this repository generates animations visualizing algorithms for cache-optimized matrix multiplication. Generated animations are available on YouTube. One frame from the animation is shown below.
To generate the animations, simply run make (legacy, only available at 379bf8).
Dependencies:
- python3-cairo (
pip install pycairo) - ffmpeg (
choco install ffmpegin windows)
How to run the modified version:
- The default settings only perform a dry run and print the cache hit rate
- Use
-t pdf(fast) or-t mp4(slow) to generate the visualization output - Use
--L1 0to utilize a one-level cache instead of a two-level cache
$ python matrix_mul.py --help
usage: matrix_mul.py [-h] [--matrix-size SIZE] [--transpose] [--cache-line SIZE] [--L1 SIZE]
[--L2 SIZE] [--func {,vc,v0,v1,v2,v3,v3_5}] [--block1 SIZE] [--block2 SIZE]
[--no-memory] [--title TITLE] [--subtitle SUBTITLE] [--output PATH]
[--type {dry,pdf,mp4}] [--max-frame NUM] [--no-auto-correct-extension]
[--framerate FRAMERATE]
Visualization of cache-optimized matrix multiplication
options:
-h, --help show this help message and exit
--matrix-size SIZE n of the n-by-n matrix (default: 16)
--transpose Transpose matrix B (default: False)
--cache-line SIZE number of elements for each cache line, must be power of 2 (default: 4)
--L1 SIZE number of cache lines in L1 cache (default: 4)
--L2 SIZE number of cache lines in L2 cache (default: 16)
--func {,vc,v0,v1,v2,v3,v3_5}
Computing function (default: vc)
--block1 SIZE Inner block size (default: 16)
--block2 SIZE Outer block size (default: 8)
--no-memory Do not draw memory (default: False)
--title TITLE
--subtitle SUBTITLE
--output PATH, -o PATH
Output file path (default: matrix_mul.pdf)
--type {dry,pdf,mp4}, -t {dry,pdf,mp4}
The type of the output file (default: dry)
--max-frame NUM The max number of frames to be generated (default: None)
--no-auto-correct-extension
Do not correct the extension automatically (default: False)
--framerate FRAMERATE
mp4 framerate (default: 24)