w4a16_gemv_batchm

spark_model::layers::ops

Function w4a16_gemv_batchm 

Source 
pub fn w4a16_gemv_batchm(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    input: DevicePtr,
    weight: &QuantizedWeight,
    output: DevicePtr,
    m: u32,
    n: u32,
    k: u32,
    stream: u64,
) -> Result<()>
Expand description

W4A16 batched GEMV (M<=MAX_M) — the NVFP4 sibling of w8a16_gemv_batch4/16.

Reads the NVFP4 weight matrix ONCE and computes m outputs (one per seq), amortizing the weight read across the batch. kernel is w4a16_gemv_batch4 (M<=4), w4a16_gemv_batch8 (M<=8, chain verify) or w4a16_gemv_batch16 (M<=16). A:[m,K] BF16, C:[m,N] BF16.

Kernel: w4a16_gemv_batch4/8/16(A, B_packed, B_scale, scale2, C, M, N, K) Grid: (ceil(N/4), 1, 1) Block: (256, 1, 1)