Code makes the world a better place

Performance Analysis with perf on Linux (Beginner)

  • August 29, 2020
  • Liu, An-Chi 劉安齊

Introduction

With a profiler, we can learn more about how software runs—such as memory usage, CPU cycles, cache misses, I/O time, and more. This information is extremely helpful for locating performance bottlenecks. The ultimate goal of performance analysis is to find what slows the program down and maximize performance.

This post introduces perf on Linux. Using a simple program example, I will demonstrate how to analyze a program with perf, and you will see that using a profiling tool can make it much easier to identify the root cause. This post references Gabriel Krisman Bertaz’s Performance analysis in Linux。

Continue reading

Converting Pthreads to JavaScript with Emscripten and Performance Analysis

  • July 7, 2020
  • Liu, An-Chi 劉安齊

Introduction

Emscripten is a tool that compiles C/C++ to WebAssembly. Under the hood it goes through LLVM. It supports compiling Pthreads: it turns them into JavaScript Web Workers plus WebAssembly. It can even translate OpenGL to WebGL, allowing programs to run in a browser with performance close to native.

The focus of this post is exactly that: converting Pthreads into Web Workers + WebAssembly. I will take an example program and try the conversion in practice. Finding a good benchmark program is not easy, so I wrote a small parallel Pthread program to compute π as the test case.

I will first walk through how to use Emscripten to convert Pthreads to JS. While following the official documentation, I ran into several pitfalls; I also record them here so you don’t fall into the same holes. Then I will compare performance for (1) native C code, (2) Emscripten-generated JS/WASM, and (3) a Web Worker implementation written directly in JavaScript.

Continue reading
  • Page 1 of 1