-
Porting your ML model as an onnxruntime WebAssembly application (pseudo-tutorial)
I ported a keras ML model to a fully isolated inferring application in WebAssembly. It runs completely in the browser. To do so, I use Emscripten, the onnxruntime and a few lines of Rust code.
-
wasmtime, IntelPIN and a lot of traces
I use IntelPIN and wasmtime to trace only Wasm instructions for a wasmtime execution.
-
Obfuscation experiment for some cryptominers.
We use wasm-mutate to generate a 18877 variants from a seed cryptominer. The generated "population" has a particular property, all behave equally. When a Wasm variant is executed, it then returns the same result as the original program.
-
How to deploy hand made Wasm code in Fastly Compute@Edge.
How to deploy your first Fastly's Compute@Edge service ? How to deploy custom Wasm to it ? ... and a little bit of metaprogramming with Rust.
-
Fuzzing Survey.
Fuzzing is an automated technique for software testing. I will try to put in this blog entry some of the basics ideas about fuzzing, as well as pros and cons of some fuzzers and techniques to face specific issues.
-
A non intrusive JS code analysis.
Making a function coverage tool for JS using Graal-JS source code modification.
-
Java in Flames. Profiler results
We show the results and analysis of integrating linux-perf-tool to our perturbation instrumentation. The use of java flames graph provide a better visual performance. Also, We explain our instrumentation workflow.
-
Graal VM and Truffle. Proof of concept with Tiger language
Graal VM group some languages in one polyglot compiler that runs in Java Virtual Machine. This work has proved the performance in that VM of a custom academic language called Tiger using the Truffle API.