pub fn residual_add(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
residual: DevicePtr,
src: DevicePtr,
num_elements: u32,
stream: u64,
) -> Result<()>Expand description
BF16 residual add: residual[i] += src[i] (in-place).
Kernel: bf16_residual_add(residual, src, n)
Grid: (ceil(n/256), 1, 1) Block: (256, 1, 1)