←back to thread

The Weird Concept of Branchless Programming

(sanixdk.xyz)
170 points judicious | 2 comments | 28 Sep 25 16:40 UTC | HN request time: 0.415s | source
Show context
Animats ◴[28 Sep 25 21:03 UTC] No.45407980[source]▶
This is just cutesy on CPUs, but is a big part of GPU programming.
replies(1): >>45409699 #
1. itemize123 ◴[29 Sep 25 01:49 UTC] No.45409699[source]▶
off topic, what are good resources to dive into gpu programming (for someone mostly in the cpu world)
replies(1): >>45409882 #
2. Animats ◴[29 Sep 25 02:30 UTC] No.45409882[source]▶
Here's a short CUDA demo from NVidia, of adding two arrays of a million numbers each, elementwise. The line that actually does the add is

     add<<<1, 1>>>(N, x, y);
All N adds are conceptually done in parallel, with no side effects. In practice, hundreds or thousands of adds are done simultaneously, depending on the available hardware.

This is true branchless programming.

[1] https://developer.nvidia.com/blog/even-easier-introduction-c...