We’ve developed a static analysis tool that takes an MLIR affine loop nest as input and produces its reuse interval distribution and miss ratio curve — all without profiling or executing the program.
Demo sites are available at:
- 🌐 [US East] https://affine.zhuyi.fan
- 🌐 [EU] https://affine2.zhuyi.fan
The source code is available here:
- Core analysis: https://github.com/SchrodingerZhu/autolala
- Supporting library: https://github.com/Roc-Locality/barvinok-rs
The web interface is a frontend for the CLI tool found in the autolala repository.
Note: I built the website in just two days, so bugs are to be expected. The servers are hosted on low-power machines with limited resources (30s timeout, 64MB RSS), so large inputs may be interrupted.
Feel free to try it out — we’d love to hear your feedback!
Demo sites are available at:
- 🌐 [US East] https://affine.zhuyi.fan
- 🌐 [EU] https://affine2.zhuyi.fan
The source code is available here:
- Core analysis: https://github.com/SchrodingerZhu/autolala
- Supporting library: https://github.com/Roc-Locality/barvinok-rs
The web interface is a frontend for the CLI tool found in the autolala repository.
Note: I built the website in just two days, so bugs are to be expected. The servers are hosted on low-power machines with limited resources (30s timeout, 64MB RSS), so large inputs may be interrupted.
Feel free to try it out — we’d love to hear your feedback!