rtx-cfd legacy GPU tests: the current CudaKernelManager API (Arc-held manager, allocate-on-copy, copy_from_device returning the host vector), MatrixOpsKernel::tridiagonal_matvec added over the existing kernel, PoissonKernel::solve_jacobi_2d kept as the older name; CfdConfig literals take ..Default

crates/specialized/rtx-cfd/src/kernels/cuda_simple.rs

@@ -601,6 +601,23 @@ impl PoissonKernel {
         })
     }
 
+    /// [`Self::solve_2d`] under its older name.
+    #[allow(clippy::too_many_arguments)]
+    pub fn solve_jacobi_2d(
+        &self,
+        phi: &mut CudaSlice<f32>,
+        source: &CudaSlice<f32>,
+        nx: usize,
+        ny: usize,
+        dx: f32,
+        dy: f32,
+        max_iterations: usize,
+        tolerance: f32,
+    ) -> CfdResult<usize> {
+        self.solve_2d(phi, source, nx, ny, dx, dy, max_iterations, tolerance)
+    }
+
+    #[allow(clippy::too_many_arguments)]
     pub fn solve_2d(
         &self,
         phi: &mut CudaSlice<f32>,
@@ -757,6 +774,39 @@ impl MatrixOpsKernel {
         Ok(result)
     }
 
+    /// `y = T x` for the tridiagonal matrix with `diagonal` (n) and
+    /// `off_diagonal` (n − 1) — the `tridiagonal_matvec` kernel.
+    pub fn tridiagonal_matvec(
+        &self,
+        diagonal: &CudaSlice<f32>,
+        off_diagonal: &CudaSlice<f32>,
+        x: &CudaSlice<f32>,
+        y: &mut CudaSlice<f32>,
+    ) -> CfdResult<()> {
+        let n = x.len();
+        let func = self
+            .module
+            .load_function("tridiagonal_matvec")
+            .map_err(|e| CfdError::gpu_error(&format!("Failed to get kernel: {}", e)))?;
+        let config = LaunchConfig {
+            grid_dim: ((n as u32 + 255) / 256, 1, 1),
+            block_dim: (256, 1, 1),
+            shared_mem_bytes: 0,
+        };
+        unsafe {
+            self.stream
+                .launch_builder(&func)
+                .arg(diagonal)
+                .arg(off_diagonal)
+                .arg(x)
+                .arg(y)
+                .arg(&(n as i32))
+                .launch(config)
+                .map_err(|e| CfdError::gpu_error(&format!("Kernel launch failed: {}", e)))?;
+        }
+        Ok(())
+    }
+
     /// Compute L2 norm of vector
     pub fn vector_norm(&self, x: &CudaSlice<f32>) -> CfdResult<f32> {
         let n = x.len();