pub fn w8a16_gemm_pipelined(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
input: DevicePtr,
weight: DevicePtr,
block_scale: DevicePtr,
output: DevicePtr,
m: u32,
n: u32,
k: u32,
stream: u64,
) -> Result<()>Expand description
W8A16 GEMM pipelined (M>1): bit-identical (cosine=1.0) faster rewrite of
w8a16_gemm — same args, same numerics, ~4.6× faster on GB10/sm_121.
Fix-A occupancy + cp.async pipelined kernel: 128×32 tile (M×N), 256-thread
block (8 warps). Geometry mirrors the validated w8a16_microtest
"w8a16_gemm_pipelined" arm (PM_M_TILE=128, PM_N_TILE=32).
Grid: (ceil(N/32), ceil(M/128), 1) Block: (256, 1, 1)