https://icyveins7.github.io/tags/atomic/Recent content in Atomic on icyveins7's blogHugoen-usThis work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.Sun, 16 Mar 2025 20:00:00 +0800https://icyveins7.github.io/posts/2025/03/atomicminfloat-overloading-integer-only-atomics-for-floating-point-numbers-in-cuda/Sun, 16 Mar 2025 20:00:00 +0800https://icyveins7.github.io/posts/2025/03/atomicminfloat-overloading-integer-only-atomics-for-floating-point-numbers-in-cuda/<blockquote> <p>Convincing you (and myself) that with some minor edits, we can still use atomicMin for floats in CUDA..</p> </blockquote> <p>CUDA has a set of atomic functions for <em>safely</em> updating the same memory address with different threads. There&rsquo;s a whole list of them in the programming guide <a href="https://docs.nvidia.com/cuda/cuda-c-programming-guide/#atomic-functions">here</a>.</p> <p>However, for <code>atomicMin</code> (and also <code>atomicMax</code>), there isn&rsquo;t an overload that works with <code>float</code>s (or <code>double</code>s , but who uses those in CUDA anyway..). For this discussion we&rsquo;ll just focus on <code>atomicMin</code>, but all the points we discuss can be inverted to explain <code>atomicMax</code>.</p>