https://github.com/willcrichton/corrset-benchmark

A repository to test different performance optimizations in a sparse matrix computation.
https://github.com/willcrichton/corrset-benchmark

Last synced: 2 months ago
JSON representation

A repository to test different performance optimizations in a sparse matrix computation.

• Host: GitHub
• URL: https://github.com/willcrichton/corrset-benchmark
• Owner: willcrichton
• Created: 2023-10-06T05:04:46.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2023-11-03T16:32:33.000Z (about 1 year ago)
• Last Synced: 2023-11-03T17:35:14.802Z (about 1 year ago)
• Language: Jupyter Notebook
• Homepage: https://willcrichton.net/notes/k-corrset/
• Size: 7.35 MB
• Stars: 37
• Watchers: 2
• Forks: 8
• Open Issues: 2
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# k-CorrSet Benchmark

A repository to test different performance optimizations in a sparse matrix computation. Read the post explaining the benchmark: 

## Running the Benchmark

To run the benchmark, first generate the synthetic data:

```

mkdir data

cargo run --release --bin gen-data -- 60000 200 0.2 > data/data-large.json

```

Then run the benchmark:

```

cargo bench

```

## Running One Configuration

To run a single configuration, you can use the `top` binary. It takes at least two arguments: the name of the outer loop (like `2_batched`) and the name of the inner loop (like `6_alloc`).

Check out the [`outer_names`](https://github.com/willcrichton/corrset-benchmark/blob/main/src/outer/mod.rs) and [`inner_names`](https://github.com/willcrichton/corrset-benchmark/blob/main/src/inner/mod.rs) functions to see a list of all the names. You can also optionally provide a value of `k` (default 5) and a name for the dataset (default `"large"`).

To get a quick-and-dirty ETA for the full computation, you can run with a progress bar enabled:

```

cargo run --release --bin top --features progress -- 2_batched 6_alloc

```

To get a profile for the configuration, install the [samply](https://github.com/mstange/samply/) tool. Then run:

```

cargo build --release --bin top

samply record ./target/release/top 2_batched 6_alloc

```

Let it run for ~30s, then hit Ctrl+C and the profile should open up in your browser.