pub fn transpose_block_scale(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
src: DevicePtr,
dst: DevicePtr,
n_blocks: u32,
k_blocks: u32,
stream: u64,
) -> Result<()>Expand description
Transpose block scales: [N/128, K/128] → [K/128, N/128].