Function fp8_fp8_gemm_m128_mfast 

pub fn fp8_fp8_gemm_m128_mfast(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    input: DevicePtr,
    b_fp8: DevicePtr,
    output: DevicePtr,
    m: u32,
    n: u32,
    k: u32,
    stream: u64,
) -> Result<()>

fp8_fp8_gemm_t_m128_mfast: FP8 A x FP8 B, 128-row M tile, m on the fast axis. A must already be E4M3 (see bf16_to_fp8); the MMA consumed E4M3 either way, so pre-casting A is numerically identical to the BF16-A kernel.