Merge branch 'Rust-GPU:main' into main

Author: niklebedenko

.github/workflows/ci_linux.yml

@@ -28,9 +28,18 @@ jobs:
           - name: Ubuntu-24.04 / CUDA-12.8.1 / ARM64
             image: "ghcr.io/rust-gpu/rust-cuda-ubuntu24-cuda12:latest"
             runner: ubuntu-24.04-arm
+          - name: Ubuntu-24.04 / CUDA-13.0.2 / x86_64
+            image: "ghcr.io/rust-gpu/rust-cuda-ubuntu24-cuda13:latest"
+            runner: ubuntu-latest
+          - name: Ubuntu-24.04 / CUDA-13.0.2 / ARM64
+            image: "ghcr.io/rust-gpu/rust-cuda-ubuntu24-cuda13:latest"
+            runner: ubuntu-24.04-arm
           - name: RockyLinux-9 / CUDA-12.8.1 / x86_64
             image: "ghcr.io/rust-gpu/rust-cuda-rockylinux9-cuda12:latest"
             runner: ubuntu-latest
+          - name: RockyLinux-9 / CUDA-13.0.2 / x86_64
+            image: "ghcr.io/rust-gpu/rust-cuda-rockylinux9-cuda13:latest"
+            runner: ubuntu-latest
 
     steps:
       - name: Free up space

crates/cudnn-sys/build/cudnn_sdk.rs

@@ -62,7 +62,15 @@ impl CudnnSdk {
         let cudnn_include_dir = env::var_os("CUDNN_INCLUDE_DIR");
 
         #[cfg(not(target_os = "windows"))]
-        const CUDNN_DEFAULT_PATHS: &[&str] = &["/usr/include", "/usr/local/include"];
+        const CUDNN_DEFAULT_PATHS: &[&str] = &[
+            "/usr/include",
+            "/usr/local/include",
+            // CUDA 13 seems to have moved the headers into arch-specific directories.
+            "/usr/include/x86_64-linux-gnu",
+            "/usr/include/aarch64-linux-gnu",
+            "/usr/local/include/x86_64-linux-gnu",
+            "/usr/local/include/aarch64-linux-gnu",
+        ];
         #[cfg(target_os = "windows")]
         const CUDNN_DEFAULT_PATHS: &[&str] = &[
             "C:/Program Files/NVIDIA/CUDNN/v9.x/include",

examples/README.md

@@ -1,23 +1,36 @@
 # CUDA + Rust examples
 
-The examples in here showcase both the GPU side and the CPU side of writing a tool which uses the GPU.
+The examples in here showcase both the GPU side and the CPU side of writing a tool which uses the
+GPU.
 
 ## Available Examples
 
 ### [vecadd](vecadd)
+
 A simple vector addition example demonstrating basic CUDA kernel usage.
 
 ### [gemm](gemm)
+
 General Matrix Multiplication (GEMM) implementation showing more complex CUDA operations.
 
+### [i128_demo](i128_demo)
+
+A simple demo that uses 128-bit integer operations.
+
 ### [sha2_crates_io](sha2_crates_io)
-Demonstrates using an existing Rust crate ([`sha2`](https://crates.io/crates/sha2) from crates.io) on both CPU and GPU without modification. Shows that the same cryptographic hashing code can run on CUDA, producing identical results to the CPU implementation.
 
-### [Interactive Path Tracer](path_tracer)
-A very simple interactive Path Tracer inspired by [Ray Tracing In One Weekend](https://raytracing.github.io/books/RayTracingInOneWeekend.html)
-which runs on CPU or GPU, with the additional option of running OptiX denoising.
+Demonstrates using an existing Rust crate ([`sha2`](https://crates.io/crates/sha2) from crates.io)
+on both CPU and GPU without modification. Shows that the same cryptographic hashing code can run on
+CUDA, producing identical results to the CPU implementation.
+
+### [Interactive Path Tracer](../crates/optix/examples/path_tracer/)
+
+A very simple interactive Path Tracer inspired by [Ray Tracing In One
+Weekend](https://raytracing.github.io/books/RayTracingInOneWeekend.html) which runs on CPU or GPU,
+with the additional option of running OptiX denoising.
 
 ![Path Tracer](assets/path_tracer.png)
 
-The Path Tracer uses cuda_builder to compile the core path tracer for the GPU and GPU (hardware raytracing), and uses the core path tracer as a normal crate
-for CPU rendering and sharing structures.
+The Path Tracer uses `cuda_builder` to compile the core path tracer for the GPU and GPU (hardware
+raytracing), and uses the core path tracer as a normal crate for CPU rendering and sharing
+structures.