google / decoder-benchmarks-for-rust Goto Github PK

The HTML output from criterion is very detailed, but doesn't give a nice overview. It also concentrates on performance differences since the last build, which we don't care about here.

Create a script that generates the table used in the README from the results of a benchmark run.

Currently the test images are all resized versions of the same image. Our results will be more robust if we used a wider variety of images.

libjpeg-turbo mentions a couple of test images and also links to imagecompression.info. The latter corpus does seem focused on whether image compression is lossy or not, rather than performance, but I'm unsure if that matters.

jpeg-decoder's benchmark images are here.

We're able to compile and run the benchmarks using GitHub actions, but we probably need to investigate whether or not whatever shared infrastructure GitHub actions uses is isolated enough from other traffic to generate solid benchmarks.