Skip to main content

gpu_batchnorm_forward

ferrotorch_gpu::kernels

Function gpu_batchnorm_forward 

Source 
pub fn gpu_batchnorm_forward(
    _input: &CudaBuffer<f32>,
    _weight: &CudaBuffer<f32>,
    _bias: &CudaBuffer<f32>,
    _running_mean: &mut CudaBuffer<f32>,
    _running_var: &mut CudaBuffer<f32>,
    _channels: usize,
    _spatial: usize,
    _eps: f32,
    _momentum: f32,
    _training: bool,
    device: &GpuDevice,
) -> GpuResult<(CudaBuffer<f32>, CudaBuffer<f32>, CudaBuffer<f32>)>
Expand description

BatchNorm2d forward on GPU (placeholder — kernel pass-1 indexing needs refinement). Currently validates the kernel compiles and falls back to returning an error so callers use the CPU path.