pub fn conv1d_fwd(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
input: DevicePtr,
weight: &DenseWeight,
output: DevicePtr,
batch: u32,
dim: u32,
seq_len: u32,
d_conv: u32,
stream: u64,
) -> Result<()>Expand description
Full-sequence causal depthwise conv1d + SiLU activation.
Kernel: causal_conv1d_fwd(input, weight, bias, output, batch, dim, seq_len, d_conv)
Grid: (dim, batch, 1) Block: (min(seq_len, 1024), 1, 1)
Input: [batch, dim, seq_len] BF16 (channel-first) Weight: [dim, d_conv] BF16 Output: [batch, dim, seq_len] BF16