Function dense_gemm_splitk 

pub fn dense_gemm_splitk(
    gpu: &dyn GpuBackend,
    partial_kernel: KernelHandle,
    reduce_kernel: KernelHandle,
    input: DevicePtr,
    weight: &DenseWeight,
    output: DevicePtr,
    workspace: DevicePtr,
    m: u32,
    n: u32,
    k: u32,
    k_splits: u32,
    stream: u64,
) -> Result<()>

Split-K GEMM: partial products over K_splits chunks, then reduce. Uses FP32 workspace of size K_splits * M * N * 4 bytes.