rustyverse/rustytorch
2
0
Fork 0
Code Issues Pull Requests Actions 46 Packages Projects Releases Wiki Activity

rtx-cfd legacy GPU: kernel outputs were device-buffer CLONES (every write lost) — the buffers are borrowed mutably now; the Poisson kernels keep boundary Dirichlet values; tests: the advection pulse marched 50 steps, the Jacobi budget 20k, the reduction reference in f64
CI / Python Bindings (maturin) (ubuntu-latest) (push) Blocked by required conditions
Details
CI / Build (macos-latest) (push) Waiting to run
Details
CI / Test (macos-latest) (push) Blocked by required conditions
Details
CI / Test (ubuntu-latest) (push) Blocked by required conditions
Details
CI / Python Bindings (maturin) (macos-latest) (push) Blocked by required conditions
Details
CI / WASM Build + Size Check (push) Blocked by required conditions
Details
CI / Distributed Training Tests (push) Blocked by required conditions
Details
CI / CI Success (push) Blocked by required conditions
Details
CI / Build CPU-Only (Explicit) (push) Failing after 5s
Details
Documentation / Build User Guide (push) Successful in 5s
Details
Documentation / Build API Documentation (push) Failing after 5s
Details
CI / Format Check (push) Failing after 10s
Details
CI / Build (ubuntu-latest) (push) Failing after 1m22s
Details
CI / Clippy Check (push) Failing after 1m38s
Details
Performance Benchmarks / Run Benchmarks (push) Successful in 2m50s
Details

Browse Source
This commit is contained in:
Omar Sobh
2026-09-16 07:28:42 -05:00
parent f3f6ac21fb
commit 0de750bad1
8 changed files with 58 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/specialized/rtx-cfd/src/kernels/cuda/poisson.cu
+2 -4
View File
@@ -25,9 +25,9 @@ extern "C" __global__ void poisson_jacobi_2d(
int idx = j * nx + i;
// Boundary conditions (Dirichlet - zero on boundaries for now)
// Boundary cells keep the Dirichlet values they hold.
if (i == 0 || i == nx - 1 || j == 0 || j == ny - 1) {
phi_new[idx] = 0.0f;
phi_new[idx] = phi[idx];
return;
}
@@ -66,7 +66,6 @@ extern "C" __global__ void poisson_gauss_seidel_2d(
// Boundary conditions
if (i == 0 || i == nx - 1 || j == 0 || j == ny - 1) {
phi[idx] = 0.0f;
return;
}
@@ -105,7 +104,6 @@ extern "C" __global__ void poisson_sor_2d(
// Boundary conditions
if (i == 0 || i == nx - 1 || j == 0 || j == ny - 1) {
phi[idx] = 0.0f;
return;
}
crates/specialized/rtx-cfd/src/solvers/incompressible/piso_gpu.rs
+6 -6
View File
@@ -147,7 +147,7 @@ impl PisoGpuSolver {
fn copy_to_gpu(&self, flow_field: &FlowField) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
// Convert nalgebra matrices to flat vectors
@@ -171,7 +171,7 @@ impl PisoGpuSolver {
fn copy_from_gpu(&self, flow_field: &mut FlowField) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
// Copy from GPU
@@ -271,7 +271,7 @@ impl PisoGpuSolver {
let (nx, ny) = {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
(buffers.nx, buffers.ny)
};
@@ -313,7 +313,7 @@ impl PisoGpuSolver {
let residual_norm = {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
self.matrix_ops_kernel.vector_norm(&buffers.mass_source)?
};
@@ -464,8 +464,8 @@ impl PisoGpuSolver {
.stream()
.launch_builder(&func)
.arg(&buffers.pressure_correction)
.arg(&mut buffers.u_correction.clone())
.arg(&mut buffers.v_correction.clone())
.arg(&mut buffers.u_correction)
.arg(&mut buffers.v_correction)
.arg(&correction_factor)
.arg(&dx)
.arg(&dy)
crates/specialized/rtx-cfd/src/solvers/incompressible/simple_gpu.rs
+3 -3
View File
@@ -136,7 +136,7 @@ impl SimpleGpuSolver {
fn copy_to_gpu(&self, flow_field: &FlowField) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
// Convert nalgebra matrices to flat vectors
@@ -163,7 +163,7 @@ impl SimpleGpuSolver {
fn copy_from_gpu(&self, flow_field: &mut FlowField) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
// Copy from GPU
@@ -300,7 +300,7 @@ impl SimpleGpuSolver {
async fn gpu_pressure_update(&self, pressure_relaxation: f32) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
// p = p + α_p * p' (pressure update with relaxation)
crates/specialized/rtx-cfd/src/solvers/lbm/d2q9_gpu.rs
+19 -19
View File
@@ -256,7 +256,7 @@ impl D2Q9GpuSolver {
fn gpu_collision_step(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
// First compute equilibrium distributions
@@ -284,7 +284,7 @@ impl D2Q9GpuSolver {
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f.clone())
.arg(&mut buffers.f)
.arg(&buffers.f_eq)
.arg(&(self.omega as f32))
.arg(&(buffers.nx as i32))
@@ -303,7 +303,7 @@ impl D2Q9GpuSolver {
pub fn gpu_streaming_step(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -327,7 +327,7 @@ impl D2Q9GpuSolver {
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f_temp.clone())
.arg(&mut buffers.f_temp)
.arg(&buffers.f)
.arg(&(buffers.nx as i32))
.arg(&(buffers.ny as i32))
@@ -347,7 +347,7 @@ impl D2Q9GpuSolver {
fn gpu_compute_macroscopic(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -371,9 +371,9 @@ impl D2Q9GpuSolver {
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.density.clone())
.arg(&mut buffers.velocity_x.clone())
.arg(&mut buffers.velocity_y.clone())
.arg(&mut buffers.density)
.arg(&mut buffers.velocity_x)
.arg(&mut buffers.velocity_y)
.arg(&buffers.f)
.arg(&(buffers.nx as i32))
.arg(&(buffers.ny as i32))
@@ -394,7 +394,7 @@ impl D2Q9GpuSolver {
fn gpu_compute_equilibrium(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -418,7 +418,7 @@ impl D2Q9GpuSolver {
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f_eq.clone())
.arg(&mut buffers.f_eq)
.arg(&buffers.density)
.arg(&buffers.velocity_x)
.arg(&buffers.velocity_y)
@@ -438,7 +438,7 @@ impl D2Q9GpuSolver {
pub fn gpu_apply_bounce_back_boundaries(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -462,7 +462,7 @@ impl D2Q9GpuSolver {
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f.clone())
.arg(&mut buffers.f)
.arg(&(buffers.nx as i32))
.arg(&(buffers.ny as i32))
.launch(config)
@@ -492,7 +492,7 @@ impl D2Q9GpuSolver {
) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -516,10 +516,10 @@ impl D2Q9GpuSolver {
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f.clone())
.arg(&mut buffers.density.clone())
.arg(&mut buffers.velocity_x.clone())
.arg(&mut buffers.velocity_y.clone())
.arg(&mut buffers.f)
.arg(&mut buffers.density)
.arg(&mut buffers.velocity_x)
.arg(&mut buffers.velocity_y)
.arg(&(density as f32))
.arg(&(velocity.x as f32))
.arg(&(velocity.y as f32))
@@ -539,7 +539,7 @@ impl D2Q9GpuSolver {
pub fn get_macroscopic_at(&self, i: usize, j: usize) -> CfdResult<(f64, Vector2<f64>)> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let idx = j * buffers.nx + i;
@@ -585,7 +585,7 @@ impl D2Q9GpuSolver {
pub fn update_flow_field(&self, flow_field: &mut FlowField) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let (nx, ny, _, _) = flow_field.grid_info();
crates/specialized/rtx-cfd/src/solvers/lbm/d3q19_gpu.rs
+18 -18
View File
@@ -365,7 +365,7 @@ __global__ void d3q19_bounce_back_boundaries(
pub fn gpu_collision_step(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let omega = 1.0 / self.params.tau;
@@ -399,7 +399,7 @@ __global__ void d3q19_bounce_back_boundaries(
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f.clone())
.arg(&mut buffers.f)
.arg(&buffers.f_eq)
.arg(&(omega as f32))
.arg(&(buffers.nx as i32))
@@ -419,7 +419,7 @@ __global__ void d3q19_bounce_back_boundaries(
pub fn gpu_streaming_step(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -444,7 +444,7 @@ __global__ void d3q19_bounce_back_boundaries(
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f_temp.clone())
.arg(&mut buffers.f_temp)
.arg(&buffers.f)
.arg(&(buffers.nx as i32))
.arg(&(buffers.ny as i32))
@@ -465,7 +465,7 @@ __global__ void d3q19_bounce_back_boundaries(
fn gpu_extract_macroscopic_variables(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -495,10 +495,10 @@ __global__ void d3q19_bounce_back_boundaries(
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.density.clone())
.arg(&mut buffers.velocity_x.clone())
.arg(&mut buffers.velocity_y.clone())
.arg(&mut buffers.velocity_z.clone())
.arg(&mut buffers.density)
.arg(&mut buffers.velocity_x)
.arg(&mut buffers.velocity_y)
.arg(&mut buffers.velocity_z)
.arg(&buffers.f)
.arg(&(buffers.nx as i32))
.arg(&(buffers.ny as i32))
@@ -520,7 +520,7 @@ __global__ void d3q19_bounce_back_boundaries(
fn gpu_compute_equilibrium(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -545,7 +545,7 @@ __global__ void d3q19_bounce_back_boundaries(
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f_eq.clone())
.arg(&mut buffers.f_eq)
.arg(&buffers.density)
.arg(&buffers.velocity_x)
.arg(&buffers.velocity_y)
@@ -567,7 +567,7 @@ __global__ void d3q19_bounce_back_boundaries(
pub fn gpu_apply_bounce_back_boundaries(&self) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -591,7 +591,7 @@ __global__ void d3q19_bounce_back_boundaries(
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f.clone())
.arg(&mut buffers.f)
.arg(&(buffers.nx as i32))
.arg(&(buffers.ny as i32))
.arg(&(buffers.nz as i32))
@@ -621,7 +621,7 @@ __global__ void d3q19_bounce_back_boundaries(
) -> CfdResult<()> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let module = self
@@ -648,7 +648,7 @@ __global__ void d3q19_bounce_back_boundaries(
self.kernel_manager
.stream()
.launch_builder(&func)
.arg(&mut buffers.f.clone())
.arg(&mut buffers.f)
.arg(&(density as f32))
.arg(&(velocity.x as f32))
.arg(&(velocity.y as f32))
@@ -675,7 +675,7 @@ __global__ void d3q19_bounce_back_boundaries(
) -> CfdResult<MacroscopicVariables3D> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
// Extract macroscopic variables first
@@ -726,7 +726,7 @@ __global__ void d3q19_bounce_back_boundaries(
pub fn gpu_total_mass(&self) -> CfdResult<f64> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
self.gpu_extract_macroscopic_variables()?;
@@ -739,7 +739,7 @@ __global__ void d3q19_bounce_back_boundaries(
pub fn gpu_kinetic_energy(&self) -> CfdResult<f64> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
self.gpu_extract_macroscopic_variables()?;
crates/specialized/rtx-cfd/src/turbulence/k_epsilon_gpu.rs
+1 -1
View File
@@ -429,7 +429,7 @@ impl KEpsilonGpuModel {
pub fn get_eddy_viscosity(&self) -> CfdResult<Vec<f32>> {
let buffers = self
.gpu_buffers
.as_ref()
.as_mut()
.ok_or_else(|| CfdError::gpu_error("GPU buffers not initialized"))?;
let n = buffers.nx * buffers.ny * buffers.nz;