Low-Level-Programming

TL;DR: New speedrun here. Details below. I generally experiment in PyTorch on my local GPUs. Recently, I received some TPU credits from the TPU Research Cloud (TRC) program to do some research, which gave me a good reason to finally learn JAX. Near about all I knew was that you could often replace numpy with jax.numpy and get code to run on accelerators, and that JAX code is supposed to be functional. This was also the first time I wrote code for TPUs. With this, I set out on a small side project: porting the modded nanoGPT speedrun to pure JAX with the goal of achieving the best possible performance. ...

Disclaimer: Please verify that whichever of the following solutions you choose to use indeed increases the stack limit for the execution of the solution, since different OSes/versions can behave differently. On Linux (and probably MacOS), running ulimit -s unlimited in every instance of the shell you use to execute your program will work just fine (assuming you don’t have a hard limit). On some compilers, passing -Wl,--stack=268435456 to the compiler options should give you a 256MB stack. ...

This post was initially meant to request updates to the C compiler on Codeforces (given the large number of posts complaining about “mysterious” issues in their submissions in C). While writing it, I realized that the chances of the said updates would be higher if I mentioned a few reasons why people would like to code in C instead of C++ (some of the reasons are not completely serious, as should be evident from the context). ...

Introduction A while ago ToxicPie9 and I made and posted this meme: To my surprise, many people are quite confused about it. In fact, I realized there are a lot of misconceptions about pragmas. Most C++ users on Codeforces put a few lines of pragmas at the start of every submission. However, I believe many of them don’t fully understand what they’re doing or how pragmas work; the only thing people seem to think is that “add pragmas -> code go brr”. ...