pub type ContiguousBatchGemv = fn(&dyn GpuBackend, KernelHandle, DevicePtr, DevicePtr, DevicePtr, DevicePtr, u32, u32, u32, u64) -> Result<()>;Expand description
The shared shape of w8a16_gemv_batch4 / w8a16_gemv_batch16 (contiguous
A and C), so a caller that picks its MAX_M tier by row count can hold the
wrapper and the handle as one pair instead of duplicating the call site.
The _strided pair’s sibling alias lives with its own callers.