pub fn mamba2_ssd_bmm(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
b_proj: DevicePtr,
c_proj: DevicePtr,
cb: DevicePtr,
seq_len: u32,
nchunks: u32,
n_groups: u32,
state_size: u32,
batch_size: u32,
bc_stride: u32,
stream: u64,
) -> Result<()>Expand description
K2: CB[c][g][t][s] = C_t . B_s (raw, fp32).