Run some unit tests on CI.

Most tests are out of reach because they require an NVIDIA GPU, but a
few can run. (Ones in `gpu_rand` and `cuda_std` need minor fixes.) Might
as well enable them.

Author: nnethercote

.github/workflows/ci_linux.yml

@@ -171,6 +171,18 @@ jobs:
               --exclude "cudnn*"
           '
 
+      # Very limited testing because we can only run tests that don't rely on having a CUDA GPU.
+      - name: Test
+        run: |
+          docker exec "$CONTAINER_NAME" bash -lc 'set -euo pipefail
+            export RUSTFLAGS=-Dwarnings
+            cargo test \
+              -p cuda_std \
+              -p gpu_rand \
+              -p nvvm \
+              -p rustc_codegen_nvvm
+          '
+
       - name: Check documentation
         run: |
           docker exec "$CONTAINER_NAME" bash -lc 'set -euo pipefail

.github/workflows/ci_windows.yml

@@ -96,9 +96,9 @@ jobs:
           RUSTFLAGS: -Dwarnings
         run: cargo clippy --workspace --exclude "optix*" --exclude "path-tracer" --exclude "denoiser" --exclude "vecadd*" --exclude "gemm*" --exclude "ex0*" --exclude "cudnn*" --exclude "sha2*"
 
-      # Don't currently test because many tests rely on the system having a CUDA GPU
-      # - name: Test
-      #   run: cargo test --workspace
+      # Very limited testing because we can only run tests that don't rely on having a CUDA GPU.
+      - name: Test
+        run: cargo test -p cuda_std -p gpu_rand -p nvvm -p rustc_codegen_nvvm
 
       - name: Check documentation
         env:

crates/cuda_std/src/thread.rs

@@ -42,16 +42,24 @@
 //! ## Computing global indices (examples)
 //!
 //! 1D global thread index:
-//! ```rust
-//! use cuda_std::thread;
-//! let gx = thread::block_idx_x() * thread::block_dim_x() + thread::thread_idx_x();
+//! ```no_run
+//! # use cuda_std::kernel;
+//! ##[kernel]
+//! pub unsafe fn f1d() {
+//!     use cuda_std::thread;
+//!     let gx = thread::block_idx_x() * thread::block_dim_x() + thread::thread_idx_x();
+//! }
 //! ```
 //!
 //! 2D global coordinates (x, y):
 //! ```rust
-//! use cuda_std::thread;
-//! let x = thread::block_idx_x() * thread::block_dim_x() + thread::thread_idx_x();
-//! let y = thread::block_idx_y() * thread::block_dim_y() + thread::thread_idx_y();
+//! # use cuda_std::kernel;
+//! ##[kernel]
+//! pub unsafe fn f2d() {
+//!     use cuda_std::thread;
+//!     let x = thread::block_idx_x() * thread::block_dim_x() + thread::thread_idx_x();
+//!     let y = thread::block_idx_y() * thread::block_dim_y() + thread::thread_idx_y();
+//! }
 //! ```
 //!
 //! Note: Hardware limits for block dimensions, grid dimensions, and total threads per block

crates/gpu_rand/src/xoroshiro/xoroshiro128plus.rs

@@ -27,8 +27,8 @@ impl Xoroshiro128Plus {
     /// parallel computations.
     ///
     /// ```
-    /// use rand_xoshiro::rand_core::SeedableRng;
-    /// use rand_xoshiro::Xoroshiro128Plus;
+    /// use gpu_rand::xoroshiro::rand_core::SeedableRng;
+    /// use gpu_rand::xoroshiro::Xoroshiro128Plus;
     ///
     /// let rng1 = Xoroshiro128Plus::seed_from_u64(0);
     /// let mut rng2 = rng1.clone();

crates/gpu_rand/src/xoroshiro/xoroshiro128plusplus.rs

@@ -26,8 +26,8 @@ impl Xoroshiro128PlusPlus {
     /// parallel computations.
     ///
     /// ```
-    /// use rand_xoshiro::rand_core::SeedableRng;
-    /// use rand_xoshiro::Xoroshiro128PlusPlus;
+    /// use gpu_rand::xoroshiro::rand_core::SeedableRng;
+    /// use gpu_rand::xoroshiro::Xoroshiro128PlusPlus;
     ///
     /// let rng1 = Xoroshiro128PlusPlus::seed_from_u64(0);
     /// let mut rng2 = rng1.clone();

crates/gpu_rand/src/xoroshiro/xoroshiro128starstar.rs

@@ -26,8 +26,8 @@ impl Xoroshiro128StarStar {
     /// parallel computations.
     ///
     /// ```
-    /// use rand_xoshiro::rand_core::SeedableRng;
-    /// use rand_xoshiro::Xoroshiro128StarStar;
+    /// use gpu_rand::xoroshiro::rand_core::SeedableRng;
+    /// use gpu_rand::xoroshiro::Xoroshiro128StarStar;
     ///
     /// let rng1 = Xoroshiro128StarStar::seed_from_u64(0);
     /// let mut rng2 = rng1.clone();

crates/gpu_rand/src/xoroshiro/xoshiro128plus.rs

@@ -25,8 +25,8 @@ impl Xoshiro128Plus {
     /// parallel computations.
     ///
     /// ```
-    /// use rand_xoshiro::rand_core::SeedableRng;
-    /// use rand_xoshiro::Xoroshiro128StarStar;
+    /// use gpu_rand::xoroshiro::rand_core::SeedableRng;
+    /// use gpu_rand::xoroshiro::Xoroshiro128StarStar;
     ///
     /// let rng1 = Xoroshiro128StarStar::seed_from_u64(0);
     /// let mut rng2 = rng1.clone();

crates/gpu_rand/src/xoroshiro/xoshiro128plusplus.rs

@@ -24,8 +24,8 @@ impl Xoshiro128PlusPlus {
     /// parallel computations.
     ///
     /// ```
-    /// use rand_xoshiro::rand_core::SeedableRng;
-    /// use rand_xoshiro::Xoroshiro128PlusPlus;
+    /// use gpu_rand::xoroshiro::rand_core::SeedableRng;
+    /// use gpu_rand::xoroshiro::Xoroshiro128PlusPlus;
     ///
     /// let rng1 = Xoroshiro128PlusPlus::seed_from_u64(0);
     /// let mut rng2 = rng1.clone();

crates/gpu_rand/src/xoroshiro/xoshiro128starstar.rs

@@ -24,8 +24,8 @@ impl Xoshiro128StarStar {
     /// parallel computations.
     ///
     /// ```
-    /// use rand_xoshiro::rand_core::SeedableRng;
-    /// use rand_xoshiro::Xoroshiro128StarStar;
+    /// use gpu_rand::xoroshiro::rand_core::SeedableRng;
+    /// use gpu_rand::xoroshiro::Xoroshiro128StarStar;
     ///
     /// let rng1 = Xoroshiro128StarStar::seed_from_u64(0);
     /// let mut rng2 = rng1.clone();

crates/gpu_rand/src/xoroshiro/xoshiro256plus.rs

@@ -25,8 +25,8 @@ impl Xoshiro256Plus {
     /// parallel computations.
     ///
     /// ```
-    /// use rand_xoshiro::rand_core::SeedableRng;
-    /// use rand_xoshiro::Xoshiro256Plus;
+    /// use gpu_rand::xoroshiro::rand_core::SeedableRng;
+    /// use gpu_rand::xoroshiro::Xoshiro256Plus;
     ///
     /// let rng1 = Xoshiro256Plus::seed_from_u64(0);
     /// let mut rng2 = rng1.clone();